Linux Audio

Check our new training course

Loading...
v4.6
   1/*
   2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   3 * Written by Alex Tomas <alex@clusterfs.com>
   4 *
   5 * Architecture independence:
   6 *   Copyright (c) 2005, Bull S.A.
   7 *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License version 2 as
  11 * published by the Free Software Foundation.
  12 *
  13 * This program is distributed in the hope that it will be useful,
  14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16 * GNU General Public License for more details.
  17 *
  18 * You should have received a copy of the GNU General Public License
  19 * along with this program; if not, write to the Free Software
  20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  21 */
  22
  23/*
  24 * Extents support for EXT4
  25 *
  26 * TODO:
  27 *   - ext4*_error() should be used in some situations
  28 *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  29 *   - smart tree reduction
  30 */
  31
  32#include <linux/fs.h>
  33#include <linux/time.h>
  34#include <linux/jbd2.h>
  35#include <linux/highuid.h>
  36#include <linux/pagemap.h>
  37#include <linux/quotaops.h>
  38#include <linux/string.h>
  39#include <linux/slab.h>
  40#include <asm/uaccess.h>
  41#include <linux/fiemap.h>
  42#include <linux/backing-dev.h>
  43#include "ext4_jbd2.h"
  44#include "ext4_extents.h"
  45#include "xattr.h"
  46
  47#include <trace/events/ext4.h>
  48
  49/*
  50 * used by extent splitting.
  51 */
  52#define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
  53					due to ENOSPC */
  54#define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
  55#define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
  56
  57#define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
  58#define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
  59
  60static __le32 ext4_extent_block_csum(struct inode *inode,
  61				     struct ext4_extent_header *eh)
  62{
  63	struct ext4_inode_info *ei = EXT4_I(inode);
  64	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  65	__u32 csum;
  66
  67	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  68			   EXT4_EXTENT_TAIL_OFFSET(eh));
  69	return cpu_to_le32(csum);
  70}
  71
  72static int ext4_extent_block_csum_verify(struct inode *inode,
  73					 struct ext4_extent_header *eh)
  74{
  75	struct ext4_extent_tail *et;
  76
  77	if (!ext4_has_metadata_csum(inode->i_sb))
  78		return 1;
  79
  80	et = find_ext4_extent_tail(eh);
  81	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  82		return 0;
  83	return 1;
  84}
  85
  86static void ext4_extent_block_csum_set(struct inode *inode,
  87				       struct ext4_extent_header *eh)
  88{
  89	struct ext4_extent_tail *et;
  90
  91	if (!ext4_has_metadata_csum(inode->i_sb))
  92		return;
  93
  94	et = find_ext4_extent_tail(eh);
  95	et->et_checksum = ext4_extent_block_csum(inode, eh);
  96}
  97
  98static int ext4_split_extent(handle_t *handle,
  99				struct inode *inode,
 100				struct ext4_ext_path **ppath,
 101				struct ext4_map_blocks *map,
 102				int split_flag,
 103				int flags);
 104
 105static int ext4_split_extent_at(handle_t *handle,
 106			     struct inode *inode,
 107			     struct ext4_ext_path **ppath,
 108			     ext4_lblk_t split,
 109			     int split_flag,
 110			     int flags);
 111
 112static int ext4_find_delayed_extent(struct inode *inode,
 113				    struct extent_status *newes);
 114
 115static int ext4_ext_truncate_extend_restart(handle_t *handle,
 116					    struct inode *inode,
 117					    int needed)
 118{
 119	int err;
 120
 121	if (!ext4_handle_valid(handle))
 122		return 0;
 123	if (handle->h_buffer_credits > needed)
 124		return 0;
 125	err = ext4_journal_extend(handle, needed);
 
 
 
 
 
 126	if (err <= 0)
 127		return err;
 128	err = ext4_truncate_restart_trans(handle, inode, needed);
 129	if (err == 0)
 130		err = -EAGAIN;
 131
 132	return err;
 133}
 134
 135/*
 136 * could return:
 137 *  - EROFS
 138 *  - ENOMEM
 139 */
 140static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 141				struct ext4_ext_path *path)
 142{
 143	if (path->p_bh) {
 144		/* path points to block */
 145		BUFFER_TRACE(path->p_bh, "get_write_access");
 146		return ext4_journal_get_write_access(handle, path->p_bh);
 147	}
 148	/* path points to leaf/index in inode body */
 149	/* we use in-core data, no need to protect them */
 150	return 0;
 151}
 152
 153/*
 154 * could return:
 155 *  - EROFS
 156 *  - ENOMEM
 157 *  - EIO
 158 */
 159int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
 160		     struct inode *inode, struct ext4_ext_path *path)
 161{
 162	int err;
 163
 164	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 165	if (path->p_bh) {
 166		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 167		/* path points to block */
 168		err = __ext4_handle_dirty_metadata(where, line, handle,
 169						   inode, path->p_bh);
 170	} else {
 171		/* path points to leaf/index in inode body */
 172		err = ext4_mark_inode_dirty(handle, inode);
 173	}
 174	return err;
 175}
 176
 177static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 178			      struct ext4_ext_path *path,
 179			      ext4_lblk_t block)
 180{
 181	if (path) {
 182		int depth = path->p_depth;
 183		struct ext4_extent *ex;
 184
 185		/*
 186		 * Try to predict block placement assuming that we are
 187		 * filling in a file which will eventually be
 188		 * non-sparse --- i.e., in the case of libbfd writing
 189		 * an ELF object sections out-of-order but in a way
 190		 * the eventually results in a contiguous object or
 191		 * executable file, or some database extending a table
 192		 * space file.  However, this is actually somewhat
 193		 * non-ideal if we are writing a sparse file such as
 194		 * qemu or KVM writing a raw image file that is going
 195		 * to stay fairly sparse, since it will end up
 196		 * fragmenting the file system's free space.  Maybe we
 197		 * should have some hueristics or some way to allow
 198		 * userspace to pass a hint to file system,
 199		 * especially if the latter case turns out to be
 200		 * common.
 201		 */
 202		ex = path[depth].p_ext;
 203		if (ex) {
 204			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 205			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 206
 207			if (block > ext_block)
 208				return ext_pblk + (block - ext_block);
 209			else
 210				return ext_pblk - (ext_block - block);
 211		}
 212
 213		/* it looks like index is empty;
 214		 * try to find starting block from index itself */
 215		if (path[depth].p_bh)
 216			return path[depth].p_bh->b_blocknr;
 217	}
 218
 219	/* OK. use inode's group */
 220	return ext4_inode_to_goal_block(inode);
 221}
 222
 223/*
 224 * Allocation for a meta data block
 225 */
 226static ext4_fsblk_t
 227ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 228			struct ext4_ext_path *path,
 229			struct ext4_extent *ex, int *err, unsigned int flags)
 230{
 231	ext4_fsblk_t goal, newblock;
 232
 233	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 234	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 235					NULL, err);
 236	return newblock;
 237}
 238
 239static inline int ext4_ext_space_block(struct inode *inode, int check)
 240{
 241	int size;
 242
 243	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 244			/ sizeof(struct ext4_extent);
 245#ifdef AGGRESSIVE_TEST
 246	if (!check && size > 6)
 247		size = 6;
 248#endif
 249	return size;
 250}
 251
 252static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 253{
 254	int size;
 255
 256	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 257			/ sizeof(struct ext4_extent_idx);
 258#ifdef AGGRESSIVE_TEST
 259	if (!check && size > 5)
 260		size = 5;
 261#endif
 262	return size;
 263}
 264
 265static inline int ext4_ext_space_root(struct inode *inode, int check)
 266{
 267	int size;
 268
 269	size = sizeof(EXT4_I(inode)->i_data);
 270	size -= sizeof(struct ext4_extent_header);
 271	size /= sizeof(struct ext4_extent);
 272#ifdef AGGRESSIVE_TEST
 273	if (!check && size > 3)
 274		size = 3;
 275#endif
 276	return size;
 277}
 278
 279static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 280{
 281	int size;
 282
 283	size = sizeof(EXT4_I(inode)->i_data);
 284	size -= sizeof(struct ext4_extent_header);
 285	size /= sizeof(struct ext4_extent_idx);
 286#ifdef AGGRESSIVE_TEST
 287	if (!check && size > 4)
 288		size = 4;
 289#endif
 290	return size;
 291}
 292
 293static inline int
 294ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 295			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 296			   int nofail)
 297{
 298	struct ext4_ext_path *path = *ppath;
 299	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 300
 301	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 302			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 303			EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
 304			(nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
 305}
 306
 307/*
 308 * Calculate the number of metadata blocks needed
 309 * to allocate @blocks
 310 * Worse case is one block per extent
 311 */
 312int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 313{
 314	struct ext4_inode_info *ei = EXT4_I(inode);
 315	int idxs;
 316
 317	idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 318		/ sizeof(struct ext4_extent_idx));
 319
 320	/*
 321	 * If the new delayed allocation block is contiguous with the
 322	 * previous da block, it can share index blocks with the
 323	 * previous block, so we only need to allocate a new index
 324	 * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 325	 * an additional index block, and at ldxs**3 blocks, yet
 326	 * another index blocks.
 327	 */
 328	if (ei->i_da_metadata_calc_len &&
 329	    ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 330		int num = 0;
 331
 332		if ((ei->i_da_metadata_calc_len % idxs) == 0)
 333			num++;
 334		if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 335			num++;
 336		if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 337			num++;
 338			ei->i_da_metadata_calc_len = 0;
 339		} else
 340			ei->i_da_metadata_calc_len++;
 341		ei->i_da_metadata_calc_last_lblock++;
 342		return num;
 343	}
 344
 345	/*
 346	 * In the worst case we need a new set of index blocks at
 347	 * every level of the inode's extent tree.
 348	 */
 349	ei->i_da_metadata_calc_len = 1;
 350	ei->i_da_metadata_calc_last_lblock = lblock;
 351	return ext_depth(inode) + 1;
 352}
 353
 354static int
 355ext4_ext_max_entries(struct inode *inode, int depth)
 356{
 357	int max;
 358
 359	if (depth == ext_depth(inode)) {
 360		if (depth == 0)
 361			max = ext4_ext_space_root(inode, 1);
 362		else
 363			max = ext4_ext_space_root_idx(inode, 1);
 364	} else {
 365		if (depth == 0)
 366			max = ext4_ext_space_block(inode, 1);
 367		else
 368			max = ext4_ext_space_block_idx(inode, 1);
 369	}
 370
 371	return max;
 372}
 373
 374static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 375{
 376	ext4_fsblk_t block = ext4_ext_pblock(ext);
 377	int len = ext4_ext_get_actual_len(ext);
 378	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 379	ext4_lblk_t last = lblock + len - 1;
 380
 381	if (len == 0 || lblock > last)
 
 
 
 
 
 382		return 0;
 383	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 384}
 385
 386static int ext4_valid_extent_idx(struct inode *inode,
 387				struct ext4_extent_idx *ext_idx)
 388{
 389	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 390
 391	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 392}
 393
 394static int ext4_valid_extent_entries(struct inode *inode,
 395				struct ext4_extent_header *eh,
 396				int depth)
 397{
 398	unsigned short entries;
 399	if (eh->eh_entries == 0)
 400		return 1;
 401
 402	entries = le16_to_cpu(eh->eh_entries);
 403
 404	if (depth == 0) {
 405		/* leaf entries */
 406		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 407		struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 408		ext4_fsblk_t pblock = 0;
 409		ext4_lblk_t lblock = 0;
 410		ext4_lblk_t prev = 0;
 411		int len = 0;
 412		while (entries) {
 413			if (!ext4_valid_extent(inode, ext))
 414				return 0;
 415
 416			/* Check for overlapping extents */
 417			lblock = le32_to_cpu(ext->ee_block);
 418			len = ext4_ext_get_actual_len(ext);
 419			if ((lblock <= prev) && prev) {
 420				pblock = ext4_ext_pblock(ext);
 421				es->s_last_error_block = cpu_to_le64(pblock);
 422				return 0;
 423			}
 424			ext++;
 425			entries--;
 426			prev = lblock + len - 1;
 427		}
 428	} else {
 429		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 430		while (entries) {
 431			if (!ext4_valid_extent_idx(inode, ext_idx))
 432				return 0;
 433			ext_idx++;
 434			entries--;
 435		}
 436	}
 437	return 1;
 438}
 439
 440static int __ext4_ext_check(const char *function, unsigned int line,
 441			    struct inode *inode, struct ext4_extent_header *eh,
 442			    int depth, ext4_fsblk_t pblk)
 443{
 444	const char *error_msg;
 445	int max = 0, err = -EFSCORRUPTED;
 446
 447	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 448		error_msg = "invalid magic";
 449		goto corrupted;
 450	}
 451	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 452		error_msg = "unexpected eh_depth";
 453		goto corrupted;
 454	}
 455	if (unlikely(eh->eh_max == 0)) {
 456		error_msg = "invalid eh_max";
 457		goto corrupted;
 458	}
 459	max = ext4_ext_max_entries(inode, depth);
 460	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 461		error_msg = "too large eh_max";
 462		goto corrupted;
 463	}
 464	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 465		error_msg = "invalid eh_entries";
 466		goto corrupted;
 467	}
 468	if (!ext4_valid_extent_entries(inode, eh, depth)) {
 469		error_msg = "invalid extent entries";
 470		goto corrupted;
 471	}
 
 
 
 
 472	/* Verify checksum on non-root extent tree nodes */
 473	if (ext_depth(inode) != depth &&
 474	    !ext4_extent_block_csum_verify(inode, eh)) {
 475		error_msg = "extent tree corrupted";
 476		err = -EFSBADCRC;
 477		goto corrupted;
 478	}
 479	return 0;
 480
 481corrupted:
 482	ext4_error_inode(inode, function, line, 0,
 483			 "pblk %llu bad header/extent: %s - magic %x, "
 484			 "entries %u, max %u(%u), depth %u(%u)",
 485			 (unsigned long long) pblk, error_msg,
 486			 le16_to_cpu(eh->eh_magic),
 487			 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 488			 max, le16_to_cpu(eh->eh_depth), depth);
 489	return err;
 490}
 491
 492#define ext4_ext_check(inode, eh, depth, pblk)			\
 493	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
 494
 495int ext4_ext_check_inode(struct inode *inode)
 496{
 497	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 498}
 499
 500static struct buffer_head *
 501__read_extent_tree_block(const char *function, unsigned int line,
 502			 struct inode *inode, ext4_fsblk_t pblk, int depth,
 503			 int flags)
 504{
 505	struct buffer_head		*bh;
 506	int				err;
 507
 508	bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
 509	if (unlikely(!bh))
 510		return ERR_PTR(-ENOMEM);
 511
 512	if (!bh_uptodate_or_lock(bh)) {
 513		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 514		err = bh_submit_read(bh);
 515		if (err < 0)
 516			goto errout;
 517	}
 518	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 519		return bh;
 520	err = __ext4_ext_check(function, line, inode,
 521			       ext_block_hdr(bh), depth, pblk);
 522	if (err)
 523		goto errout;
 524	set_buffer_verified(bh);
 525	/*
 526	 * If this is a leaf block, cache all of its entries
 527	 */
 528	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 529		struct ext4_extent_header *eh = ext_block_hdr(bh);
 530		struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 531		ext4_lblk_t prev = 0;
 532		int i;
 533
 534		for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 535			unsigned int status = EXTENT_STATUS_WRITTEN;
 536			ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 537			int len = ext4_ext_get_actual_len(ex);
 538
 539			if (prev && (prev != lblk))
 540				ext4_es_cache_extent(inode, prev,
 541						     lblk - prev, ~0,
 542						     EXTENT_STATUS_HOLE);
 543
 544			if (ext4_ext_is_unwritten(ex))
 545				status = EXTENT_STATUS_UNWRITTEN;
 546			ext4_es_cache_extent(inode, lblk, len,
 547					     ext4_ext_pblock(ex), status);
 548			prev = lblk + len;
 549		}
 550	}
 551	return bh;
 552errout:
 553	put_bh(bh);
 554	return ERR_PTR(err);
 555
 556}
 557
 558#define read_extent_tree_block(inode, pblk, depth, flags)		\
 559	__read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
 560				 (depth), (flags))
 561
 562/*
 563 * This function is called to cache a file's extent information in the
 564 * extent status tree
 565 */
 566int ext4_ext_precache(struct inode *inode)
 567{
 568	struct ext4_inode_info *ei = EXT4_I(inode);
 569	struct ext4_ext_path *path = NULL;
 570	struct buffer_head *bh;
 571	int i = 0, depth, ret = 0;
 572
 573	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 574		return 0;	/* not an extent-mapped inode */
 575
 576	down_read(&ei->i_data_sem);
 577	depth = ext_depth(inode);
 578
 579	path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
 580		       GFP_NOFS);
 581	if (path == NULL) {
 582		up_read(&ei->i_data_sem);
 583		return -ENOMEM;
 584	}
 585
 586	/* Don't cache anything if there are no external extent blocks */
 587	if (depth == 0)
 588		goto out;
 589	path[0].p_hdr = ext_inode_hdr(inode);
 590	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 591	if (ret)
 592		goto out;
 593	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 594	while (i >= 0) {
 595		/*
 596		 * If this is a leaf block or we've reached the end of
 597		 * the index block, go up
 598		 */
 599		if ((i == depth) ||
 600		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 601			brelse(path[i].p_bh);
 602			path[i].p_bh = NULL;
 603			i--;
 604			continue;
 605		}
 606		bh = read_extent_tree_block(inode,
 607					    ext4_idx_pblock(path[i].p_idx++),
 608					    depth - i - 1,
 609					    EXT4_EX_FORCE_CACHE);
 610		if (IS_ERR(bh)) {
 611			ret = PTR_ERR(bh);
 612			break;
 613		}
 614		i++;
 615		path[i].p_bh = bh;
 616		path[i].p_hdr = ext_block_hdr(bh);
 617		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 618	}
 619	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 620out:
 621	up_read(&ei->i_data_sem);
 622	ext4_ext_drop_refs(path);
 623	kfree(path);
 624	return ret;
 625}
 626
 627#ifdef EXT_DEBUG
 628static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 629{
 630	int k, l = path->p_depth;
 631
 632	ext_debug("path:");
 633	for (k = 0; k <= l; k++, path++) {
 634		if (path->p_idx) {
 635		  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 636			    ext4_idx_pblock(path->p_idx));
 637		} else if (path->p_ext) {
 638			ext_debug("  %d:[%d]%d:%llu ",
 639				  le32_to_cpu(path->p_ext->ee_block),
 640				  ext4_ext_is_unwritten(path->p_ext),
 641				  ext4_ext_get_actual_len(path->p_ext),
 642				  ext4_ext_pblock(path->p_ext));
 643		} else
 644			ext_debug("  []");
 645	}
 646	ext_debug("\n");
 647}
 648
 649static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 650{
 651	int depth = ext_depth(inode);
 652	struct ext4_extent_header *eh;
 653	struct ext4_extent *ex;
 654	int i;
 655
 656	if (!path)
 657		return;
 658
 659	eh = path[depth].p_hdr;
 660	ex = EXT_FIRST_EXTENT(eh);
 661
 662	ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 663
 664	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 665		ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 666			  ext4_ext_is_unwritten(ex),
 667			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 668	}
 669	ext_debug("\n");
 670}
 671
 672static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 673			ext4_fsblk_t newblock, int level)
 674{
 675	int depth = ext_depth(inode);
 676	struct ext4_extent *ex;
 677
 678	if (depth != level) {
 679		struct ext4_extent_idx *idx;
 680		idx = path[level].p_idx;
 681		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 682			ext_debug("%d: move %d:%llu in new index %llu\n", level,
 683					le32_to_cpu(idx->ei_block),
 684					ext4_idx_pblock(idx),
 685					newblock);
 686			idx++;
 687		}
 688
 689		return;
 690	}
 691
 692	ex = path[depth].p_ext;
 693	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 694		ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 695				le32_to_cpu(ex->ee_block),
 696				ext4_ext_pblock(ex),
 697				ext4_ext_is_unwritten(ex),
 698				ext4_ext_get_actual_len(ex),
 699				newblock);
 700		ex++;
 701	}
 702}
 703
 704#else
 705#define ext4_ext_show_path(inode, path)
 706#define ext4_ext_show_leaf(inode, path)
 707#define ext4_ext_show_move(inode, path, newblock, level)
 708#endif
 709
 710void ext4_ext_drop_refs(struct ext4_ext_path *path)
 711{
 712	int depth, i;
 713
 714	if (!path)
 715		return;
 716	depth = path->p_depth;
 717	for (i = 0; i <= depth; i++, path++)
 718		if (path->p_bh) {
 719			brelse(path->p_bh);
 720			path->p_bh = NULL;
 721		}
 722}
 723
 724/*
 725 * ext4_ext_binsearch_idx:
 726 * binary search for the closest index of the given block
 727 * the header must be checked before calling this
 728 */
 729static void
 730ext4_ext_binsearch_idx(struct inode *inode,
 731			struct ext4_ext_path *path, ext4_lblk_t block)
 732{
 733	struct ext4_extent_header *eh = path->p_hdr;
 734	struct ext4_extent_idx *r, *l, *m;
 735
 736
 737	ext_debug("binsearch for %u(idx):  ", block);
 738
 739	l = EXT_FIRST_INDEX(eh) + 1;
 740	r = EXT_LAST_INDEX(eh);
 741	while (l <= r) {
 742		m = l + (r - l) / 2;
 743		if (block < le32_to_cpu(m->ei_block))
 744			r = m - 1;
 745		else
 746			l = m + 1;
 747		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 748				m, le32_to_cpu(m->ei_block),
 749				r, le32_to_cpu(r->ei_block));
 750	}
 751
 752	path->p_idx = l - 1;
 753	ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 754		  ext4_idx_pblock(path->p_idx));
 755
 756#ifdef CHECK_BINSEARCH
 757	{
 758		struct ext4_extent_idx *chix, *ix;
 759		int k;
 760
 761		chix = ix = EXT_FIRST_INDEX(eh);
 762		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 763		  if (k != 0 &&
 764		      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 765				printk(KERN_DEBUG "k=%d, ix=0x%p, "
 766				       "first=0x%p\n", k,
 767				       ix, EXT_FIRST_INDEX(eh));
 768				printk(KERN_DEBUG "%u <= %u\n",
 769				       le32_to_cpu(ix->ei_block),
 770				       le32_to_cpu(ix[-1].ei_block));
 771			}
 772			BUG_ON(k && le32_to_cpu(ix->ei_block)
 773					   <= le32_to_cpu(ix[-1].ei_block));
 774			if (block < le32_to_cpu(ix->ei_block))
 775				break;
 776			chix = ix;
 777		}
 778		BUG_ON(chix != path->p_idx);
 779	}
 780#endif
 781
 782}
 783
 784/*
 785 * ext4_ext_binsearch:
 786 * binary search for closest extent of the given block
 787 * the header must be checked before calling this
 788 */
 789static void
 790ext4_ext_binsearch(struct inode *inode,
 791		struct ext4_ext_path *path, ext4_lblk_t block)
 792{
 793	struct ext4_extent_header *eh = path->p_hdr;
 794	struct ext4_extent *r, *l, *m;
 795
 796	if (eh->eh_entries == 0) {
 797		/*
 798		 * this leaf is empty:
 799		 * we get such a leaf in split/add case
 800		 */
 801		return;
 802	}
 803
 804	ext_debug("binsearch for %u:  ", block);
 805
 806	l = EXT_FIRST_EXTENT(eh) + 1;
 807	r = EXT_LAST_EXTENT(eh);
 808
 809	while (l <= r) {
 810		m = l + (r - l) / 2;
 811		if (block < le32_to_cpu(m->ee_block))
 812			r = m - 1;
 813		else
 814			l = m + 1;
 815		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 816				m, le32_to_cpu(m->ee_block),
 817				r, le32_to_cpu(r->ee_block));
 818	}
 819
 820	path->p_ext = l - 1;
 821	ext_debug("  -> %d:%llu:[%d]%d ",
 822			le32_to_cpu(path->p_ext->ee_block),
 823			ext4_ext_pblock(path->p_ext),
 824			ext4_ext_is_unwritten(path->p_ext),
 825			ext4_ext_get_actual_len(path->p_ext));
 826
 827#ifdef CHECK_BINSEARCH
 828	{
 829		struct ext4_extent *chex, *ex;
 830		int k;
 831
 832		chex = ex = EXT_FIRST_EXTENT(eh);
 833		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 834			BUG_ON(k && le32_to_cpu(ex->ee_block)
 835					  <= le32_to_cpu(ex[-1].ee_block));
 836			if (block < le32_to_cpu(ex->ee_block))
 837				break;
 838			chex = ex;
 839		}
 840		BUG_ON(chex != path->p_ext);
 841	}
 842#endif
 843
 844}
 845
 846int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 847{
 848	struct ext4_extent_header *eh;
 849
 850	eh = ext_inode_hdr(inode);
 851	eh->eh_depth = 0;
 852	eh->eh_entries = 0;
 853	eh->eh_magic = EXT4_EXT_MAGIC;
 854	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 855	ext4_mark_inode_dirty(handle, inode);
 856	return 0;
 857}
 858
 859struct ext4_ext_path *
 860ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 861		 struct ext4_ext_path **orig_path, int flags)
 862{
 863	struct ext4_extent_header *eh;
 864	struct buffer_head *bh;
 865	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 866	short int depth, i, ppos = 0;
 867	int ret;
 868
 869	eh = ext_inode_hdr(inode);
 870	depth = ext_depth(inode);
 871
 872	if (path) {
 873		ext4_ext_drop_refs(path);
 874		if (depth > path[0].p_maxdepth) {
 875			kfree(path);
 876			*orig_path = path = NULL;
 877		}
 878	}
 879	if (!path) {
 880		/* account possible depth increase */
 881		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 882				GFP_NOFS);
 883		if (unlikely(!path))
 884			return ERR_PTR(-ENOMEM);
 885		path[0].p_maxdepth = depth + 1;
 886	}
 887	path[0].p_hdr = eh;
 888	path[0].p_bh = NULL;
 889
 890	i = depth;
 891	/* walk through the tree */
 892	while (i) {
 893		ext_debug("depth %d: num %d, max %d\n",
 894			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 895
 896		ext4_ext_binsearch_idx(inode, path + ppos, block);
 897		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 898		path[ppos].p_depth = i;
 899		path[ppos].p_ext = NULL;
 900
 901		bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
 902					    flags);
 903		if (IS_ERR(bh)) {
 904			ret = PTR_ERR(bh);
 905			goto err;
 906		}
 907
 908		eh = ext_block_hdr(bh);
 909		ppos++;
 910		if (unlikely(ppos > depth)) {
 911			put_bh(bh);
 912			EXT4_ERROR_INODE(inode,
 913					 "ppos %d > depth %d", ppos, depth);
 914			ret = -EFSCORRUPTED;
 915			goto err;
 916		}
 917		path[ppos].p_bh = bh;
 918		path[ppos].p_hdr = eh;
 919	}
 920
 921	path[ppos].p_depth = i;
 922	path[ppos].p_ext = NULL;
 923	path[ppos].p_idx = NULL;
 924
 925	/* find extent */
 926	ext4_ext_binsearch(inode, path + ppos, block);
 927	/* if not an empty leaf */
 928	if (path[ppos].p_ext)
 929		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 930
 931	ext4_ext_show_path(inode, path);
 932
 933	return path;
 934
 935err:
 936	ext4_ext_drop_refs(path);
 937	kfree(path);
 938	if (orig_path)
 939		*orig_path = NULL;
 940	return ERR_PTR(ret);
 941}
 942
 943/*
 944 * ext4_ext_insert_index:
 945 * insert new index [@logical;@ptr] into the block at @curp;
 946 * check where to insert: before @curp or after @curp
 947 */
 948static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 949				 struct ext4_ext_path *curp,
 950				 int logical, ext4_fsblk_t ptr)
 951{
 952	struct ext4_extent_idx *ix;
 953	int len, err;
 954
 955	err = ext4_ext_get_access(handle, inode, curp);
 956	if (err)
 957		return err;
 958
 959	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 960		EXT4_ERROR_INODE(inode,
 961				 "logical %d == ei_block %d!",
 962				 logical, le32_to_cpu(curp->p_idx->ei_block));
 963		return -EFSCORRUPTED;
 964	}
 965
 966	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 967			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
 968		EXT4_ERROR_INODE(inode,
 969				 "eh_entries %d >= eh_max %d!",
 970				 le16_to_cpu(curp->p_hdr->eh_entries),
 971				 le16_to_cpu(curp->p_hdr->eh_max));
 972		return -EFSCORRUPTED;
 973	}
 974
 975	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 976		/* insert after */
 977		ext_debug("insert new index %d after: %llu\n", logical, ptr);
 978		ix = curp->p_idx + 1;
 979	} else {
 980		/* insert before */
 981		ext_debug("insert new index %d before: %llu\n", logical, ptr);
 982		ix = curp->p_idx;
 983	}
 984
 985	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 986	BUG_ON(len < 0);
 987	if (len > 0) {
 988		ext_debug("insert new index %d: "
 989				"move %d indices from 0x%p to 0x%p\n",
 990				logical, len, ix, ix + 1);
 991		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 992	}
 993
 994	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 995		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 996		return -EFSCORRUPTED;
 997	}
 998
 999	ix->ei_block = cpu_to_le32(logical);
1000	ext4_idx_store_pblock(ix, ptr);
1001	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1002
1003	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1004		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1005		return -EFSCORRUPTED;
1006	}
1007
1008	err = ext4_ext_dirty(handle, inode, curp);
1009	ext4_std_error(inode->i_sb, err);
1010
1011	return err;
1012}
1013
1014/*
1015 * ext4_ext_split:
1016 * inserts new subtree into the path, using free index entry
1017 * at depth @at:
1018 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1019 * - makes decision where to split
1020 * - moves remaining extents and index entries (right to the split point)
1021 *   into the newly allocated blocks
1022 * - initializes subtree
1023 */
1024static int ext4_ext_split(handle_t *handle, struct inode *inode,
1025			  unsigned int flags,
1026			  struct ext4_ext_path *path,
1027			  struct ext4_extent *newext, int at)
1028{
1029	struct buffer_head *bh = NULL;
1030	int depth = ext_depth(inode);
1031	struct ext4_extent_header *neh;
1032	struct ext4_extent_idx *fidx;
1033	int i = at, k, m, a;
1034	ext4_fsblk_t newblock, oldblock;
1035	__le32 border;
1036	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1037	int err = 0;
1038
1039	/* make decision: where to split? */
1040	/* FIXME: now decision is simplest: at current extent */
1041
1042	/* if current leaf will be split, then we should use
1043	 * border from split point */
1044	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1045		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1046		return -EFSCORRUPTED;
1047	}
1048	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1049		border = path[depth].p_ext[1].ee_block;
1050		ext_debug("leaf will be split."
1051				" next leaf starts at %d\n",
1052				  le32_to_cpu(border));
1053	} else {
1054		border = newext->ee_block;
1055		ext_debug("leaf will be added."
1056				" next leaf starts at %d\n",
1057				le32_to_cpu(border));
1058	}
1059
1060	/*
1061	 * If error occurs, then we break processing
1062	 * and mark filesystem read-only. index won't
1063	 * be inserted and tree will be in consistent
1064	 * state. Next mount will repair buffers too.
1065	 */
1066
1067	/*
1068	 * Get array to track all allocated blocks.
1069	 * We need this to handle errors and free blocks
1070	 * upon them.
1071	 */
1072	ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1073	if (!ablocks)
1074		return -ENOMEM;
1075
1076	/* allocate all needed blocks */
1077	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1078	for (a = 0; a < depth - at; a++) {
1079		newblock = ext4_ext_new_meta_block(handle, inode, path,
1080						   newext, &err, flags);
1081		if (newblock == 0)
1082			goto cleanup;
1083		ablocks[a] = newblock;
1084	}
1085
1086	/* initialize new leaf */
1087	newblock = ablocks[--a];
1088	if (unlikely(newblock == 0)) {
1089		EXT4_ERROR_INODE(inode, "newblock == 0!");
1090		err = -EFSCORRUPTED;
1091		goto cleanup;
1092	}
1093	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1094	if (unlikely(!bh)) {
1095		err = -ENOMEM;
1096		goto cleanup;
1097	}
1098	lock_buffer(bh);
1099
1100	err = ext4_journal_get_create_access(handle, bh);
1101	if (err)
1102		goto cleanup;
1103
1104	neh = ext_block_hdr(bh);
1105	neh->eh_entries = 0;
1106	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1107	neh->eh_magic = EXT4_EXT_MAGIC;
1108	neh->eh_depth = 0;
1109
1110	/* move remainder of path[depth] to the new leaf */
1111	if (unlikely(path[depth].p_hdr->eh_entries !=
1112		     path[depth].p_hdr->eh_max)) {
1113		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1114				 path[depth].p_hdr->eh_entries,
1115				 path[depth].p_hdr->eh_max);
1116		err = -EFSCORRUPTED;
1117		goto cleanup;
1118	}
1119	/* start copy from next extent */
1120	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1121	ext4_ext_show_move(inode, path, newblock, depth);
1122	if (m) {
1123		struct ext4_extent *ex;
1124		ex = EXT_FIRST_EXTENT(neh);
1125		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1126		le16_add_cpu(&neh->eh_entries, m);
1127	}
1128
1129	ext4_extent_block_csum_set(inode, neh);
1130	set_buffer_uptodate(bh);
1131	unlock_buffer(bh);
1132
1133	err = ext4_handle_dirty_metadata(handle, inode, bh);
1134	if (err)
1135		goto cleanup;
1136	brelse(bh);
1137	bh = NULL;
1138
1139	/* correct old leaf */
1140	if (m) {
1141		err = ext4_ext_get_access(handle, inode, path + depth);
1142		if (err)
1143			goto cleanup;
1144		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1145		err = ext4_ext_dirty(handle, inode, path + depth);
1146		if (err)
1147			goto cleanup;
1148
1149	}
1150
1151	/* create intermediate indexes */
1152	k = depth - at - 1;
1153	if (unlikely(k < 0)) {
1154		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1155		err = -EFSCORRUPTED;
1156		goto cleanup;
1157	}
1158	if (k)
1159		ext_debug("create %d intermediate indices\n", k);
1160	/* insert new index into current index block */
1161	/* current depth stored in i var */
1162	i = depth - 1;
1163	while (k--) {
1164		oldblock = newblock;
1165		newblock = ablocks[--a];
1166		bh = sb_getblk(inode->i_sb, newblock);
1167		if (unlikely(!bh)) {
1168			err = -ENOMEM;
1169			goto cleanup;
1170		}
1171		lock_buffer(bh);
1172
1173		err = ext4_journal_get_create_access(handle, bh);
1174		if (err)
1175			goto cleanup;
1176
1177		neh = ext_block_hdr(bh);
1178		neh->eh_entries = cpu_to_le16(1);
1179		neh->eh_magic = EXT4_EXT_MAGIC;
1180		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1181		neh->eh_depth = cpu_to_le16(depth - i);
1182		fidx = EXT_FIRST_INDEX(neh);
1183		fidx->ei_block = border;
1184		ext4_idx_store_pblock(fidx, oldblock);
1185
1186		ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1187				i, newblock, le32_to_cpu(border), oldblock);
1188
1189		/* move remainder of path[i] to the new index block */
1190		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1191					EXT_LAST_INDEX(path[i].p_hdr))) {
1192			EXT4_ERROR_INODE(inode,
1193					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1194					 le32_to_cpu(path[i].p_ext->ee_block));
1195			err = -EFSCORRUPTED;
1196			goto cleanup;
1197		}
1198		/* start copy indexes */
1199		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1200		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1201				EXT_MAX_INDEX(path[i].p_hdr));
1202		ext4_ext_show_move(inode, path, newblock, i);
1203		if (m) {
1204			memmove(++fidx, path[i].p_idx,
1205				sizeof(struct ext4_extent_idx) * m);
1206			le16_add_cpu(&neh->eh_entries, m);
1207		}
1208		ext4_extent_block_csum_set(inode, neh);
1209		set_buffer_uptodate(bh);
1210		unlock_buffer(bh);
1211
1212		err = ext4_handle_dirty_metadata(handle, inode, bh);
1213		if (err)
1214			goto cleanup;
1215		brelse(bh);
1216		bh = NULL;
1217
1218		/* correct old index */
1219		if (m) {
1220			err = ext4_ext_get_access(handle, inode, path + i);
1221			if (err)
1222				goto cleanup;
1223			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1224			err = ext4_ext_dirty(handle, inode, path + i);
1225			if (err)
1226				goto cleanup;
1227		}
1228
1229		i--;
1230	}
1231
1232	/* insert new index */
1233	err = ext4_ext_insert_index(handle, inode, path + at,
1234				    le32_to_cpu(border), newblock);
1235
1236cleanup:
1237	if (bh) {
1238		if (buffer_locked(bh))
1239			unlock_buffer(bh);
1240		brelse(bh);
1241	}
1242
1243	if (err) {
1244		/* free all allocated blocks in error case */
1245		for (i = 0; i < depth; i++) {
1246			if (!ablocks[i])
1247				continue;
1248			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1249					 EXT4_FREE_BLOCKS_METADATA);
1250		}
1251	}
1252	kfree(ablocks);
1253
1254	return err;
1255}
1256
1257/*
1258 * ext4_ext_grow_indepth:
1259 * implements tree growing procedure:
1260 * - allocates new block
1261 * - moves top-level data (index block or leaf) into the new block
1262 * - initializes new top-level, creating index that points to the
1263 *   just created block
1264 */
1265static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1266				 unsigned int flags)
1267{
1268	struct ext4_extent_header *neh;
1269	struct buffer_head *bh;
1270	ext4_fsblk_t newblock, goal = 0;
1271	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1272	int err = 0;
1273
1274	/* Try to prepend new index to old one */
1275	if (ext_depth(inode))
1276		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1277	if (goal > le32_to_cpu(es->s_first_data_block)) {
1278		flags |= EXT4_MB_HINT_TRY_GOAL;
1279		goal--;
1280	} else
1281		goal = ext4_inode_to_goal_block(inode);
1282	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1283					NULL, &err);
1284	if (newblock == 0)
1285		return err;
1286
1287	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1288	if (unlikely(!bh))
1289		return -ENOMEM;
1290	lock_buffer(bh);
1291
1292	err = ext4_journal_get_create_access(handle, bh);
1293	if (err) {
1294		unlock_buffer(bh);
1295		goto out;
1296	}
1297
1298	/* move top-level index/leaf into new block */
1299	memmove(bh->b_data, EXT4_I(inode)->i_data,
1300		sizeof(EXT4_I(inode)->i_data));
1301
1302	/* set size of new block */
1303	neh = ext_block_hdr(bh);
1304	/* old root could have indexes or leaves
1305	 * so calculate e_max right way */
1306	if (ext_depth(inode))
1307		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1308	else
1309		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1310	neh->eh_magic = EXT4_EXT_MAGIC;
1311	ext4_extent_block_csum_set(inode, neh);
1312	set_buffer_uptodate(bh);
1313	unlock_buffer(bh);
1314
1315	err = ext4_handle_dirty_metadata(handle, inode, bh);
1316	if (err)
1317		goto out;
1318
1319	/* Update top-level index: num,max,pointer */
1320	neh = ext_inode_hdr(inode);
1321	neh->eh_entries = cpu_to_le16(1);
1322	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1323	if (neh->eh_depth == 0) {
1324		/* Root extent block becomes index block */
1325		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1326		EXT_FIRST_INDEX(neh)->ei_block =
1327			EXT_FIRST_EXTENT(neh)->ee_block;
1328	}
1329	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1330		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1331		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1332		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1333
1334	le16_add_cpu(&neh->eh_depth, 1);
1335	ext4_mark_inode_dirty(handle, inode);
1336out:
1337	brelse(bh);
1338
1339	return err;
1340}
1341
1342/*
1343 * ext4_ext_create_new_leaf:
1344 * finds empty index and adds new leaf.
1345 * if no free index is found, then it requests in-depth growing.
1346 */
1347static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1348				    unsigned int mb_flags,
1349				    unsigned int gb_flags,
1350				    struct ext4_ext_path **ppath,
1351				    struct ext4_extent *newext)
1352{
1353	struct ext4_ext_path *path = *ppath;
1354	struct ext4_ext_path *curp;
1355	int depth, i, err = 0;
1356
1357repeat:
1358	i = depth = ext_depth(inode);
1359
1360	/* walk up to the tree and look for free index entry */
1361	curp = path + depth;
1362	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1363		i--;
1364		curp--;
1365	}
1366
1367	/* we use already allocated block for index block,
1368	 * so subsequent data blocks should be contiguous */
1369	if (EXT_HAS_FREE_INDEX(curp)) {
1370		/* if we found index with free entry, then use that
1371		 * entry: create all needed subtree and add new leaf */
1372		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1373		if (err)
1374			goto out;
1375
1376		/* refill path */
1377		path = ext4_find_extent(inode,
1378				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1379				    ppath, gb_flags);
1380		if (IS_ERR(path))
1381			err = PTR_ERR(path);
1382	} else {
1383		/* tree is full, time to grow in depth */
1384		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1385		if (err)
1386			goto out;
1387
1388		/* refill path */
1389		path = ext4_find_extent(inode,
1390				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1391				    ppath, gb_flags);
1392		if (IS_ERR(path)) {
1393			err = PTR_ERR(path);
1394			goto out;
1395		}
1396
1397		/*
1398		 * only first (depth 0 -> 1) produces free space;
1399		 * in all other cases we have to split the grown tree
1400		 */
1401		depth = ext_depth(inode);
1402		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1403			/* now we need to split */
1404			goto repeat;
1405		}
1406	}
1407
1408out:
1409	return err;
1410}
1411
1412/*
1413 * search the closest allocated block to the left for *logical
1414 * and returns it at @logical + it's physical address at @phys
1415 * if *logical is the smallest allocated block, the function
1416 * returns 0 at @phys
1417 * return value contains 0 (success) or error code
1418 */
1419static int ext4_ext_search_left(struct inode *inode,
1420				struct ext4_ext_path *path,
1421				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1422{
1423	struct ext4_extent_idx *ix;
1424	struct ext4_extent *ex;
1425	int depth, ee_len;
1426
1427	if (unlikely(path == NULL)) {
1428		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1429		return -EFSCORRUPTED;
1430	}
1431	depth = path->p_depth;
1432	*phys = 0;
1433
1434	if (depth == 0 && path->p_ext == NULL)
1435		return 0;
1436
1437	/* usually extent in the path covers blocks smaller
1438	 * then *logical, but it can be that extent is the
1439	 * first one in the file */
1440
1441	ex = path[depth].p_ext;
1442	ee_len = ext4_ext_get_actual_len(ex);
1443	if (*logical < le32_to_cpu(ex->ee_block)) {
1444		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1445			EXT4_ERROR_INODE(inode,
1446					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1447					 *logical, le32_to_cpu(ex->ee_block));
1448			return -EFSCORRUPTED;
1449		}
1450		while (--depth >= 0) {
1451			ix = path[depth].p_idx;
1452			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1453				EXT4_ERROR_INODE(inode,
1454				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1455				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1456				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1457		le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1458				  depth);
1459				return -EFSCORRUPTED;
1460			}
1461		}
1462		return 0;
1463	}
1464
1465	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1466		EXT4_ERROR_INODE(inode,
1467				 "logical %d < ee_block %d + ee_len %d!",
1468				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1469		return -EFSCORRUPTED;
1470	}
1471
1472	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1473	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1474	return 0;
1475}
1476
1477/*
1478 * search the closest allocated block to the right for *logical
1479 * and returns it at @logical + it's physical address at @phys
1480 * if *logical is the largest allocated block, the function
1481 * returns 0 at @phys
1482 * return value contains 0 (success) or error code
1483 */
1484static int ext4_ext_search_right(struct inode *inode,
1485				 struct ext4_ext_path *path,
1486				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1487				 struct ext4_extent **ret_ex)
1488{
1489	struct buffer_head *bh = NULL;
1490	struct ext4_extent_header *eh;
1491	struct ext4_extent_idx *ix;
1492	struct ext4_extent *ex;
1493	ext4_fsblk_t block;
1494	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1495	int ee_len;
1496
1497	if (unlikely(path == NULL)) {
1498		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1499		return -EFSCORRUPTED;
1500	}
1501	depth = path->p_depth;
1502	*phys = 0;
1503
1504	if (depth == 0 && path->p_ext == NULL)
1505		return 0;
1506
1507	/* usually extent in the path covers blocks smaller
1508	 * then *logical, but it can be that extent is the
1509	 * first one in the file */
1510
1511	ex = path[depth].p_ext;
1512	ee_len = ext4_ext_get_actual_len(ex);
1513	if (*logical < le32_to_cpu(ex->ee_block)) {
1514		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1515			EXT4_ERROR_INODE(inode,
1516					 "first_extent(path[%d].p_hdr) != ex",
1517					 depth);
1518			return -EFSCORRUPTED;
1519		}
1520		while (--depth >= 0) {
1521			ix = path[depth].p_idx;
1522			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1523				EXT4_ERROR_INODE(inode,
1524						 "ix != EXT_FIRST_INDEX *logical %d!",
1525						 *logical);
1526				return -EFSCORRUPTED;
1527			}
1528		}
1529		goto found_extent;
1530	}
1531
1532	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1533		EXT4_ERROR_INODE(inode,
1534				 "logical %d < ee_block %d + ee_len %d!",
1535				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1536		return -EFSCORRUPTED;
1537	}
1538
1539	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1540		/* next allocated block in this leaf */
1541		ex++;
1542		goto found_extent;
1543	}
1544
1545	/* go up and search for index to the right */
1546	while (--depth >= 0) {
1547		ix = path[depth].p_idx;
1548		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1549			goto got_index;
1550	}
1551
1552	/* we've gone up to the root and found no index to the right */
1553	return 0;
1554
1555got_index:
1556	/* we've found index to the right, let's
1557	 * follow it and find the closest allocated
1558	 * block to the right */
1559	ix++;
1560	block = ext4_idx_pblock(ix);
1561	while (++depth < path->p_depth) {
1562		/* subtract from p_depth to get proper eh_depth */
1563		bh = read_extent_tree_block(inode, block,
1564					    path->p_depth - depth, 0);
1565		if (IS_ERR(bh))
1566			return PTR_ERR(bh);
1567		eh = ext_block_hdr(bh);
1568		ix = EXT_FIRST_INDEX(eh);
1569		block = ext4_idx_pblock(ix);
1570		put_bh(bh);
1571	}
1572
1573	bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1574	if (IS_ERR(bh))
1575		return PTR_ERR(bh);
1576	eh = ext_block_hdr(bh);
1577	ex = EXT_FIRST_EXTENT(eh);
1578found_extent:
1579	*logical = le32_to_cpu(ex->ee_block);
1580	*phys = ext4_ext_pblock(ex);
1581	*ret_ex = ex;
1582	if (bh)
1583		put_bh(bh);
1584	return 0;
1585}
1586
1587/*
1588 * ext4_ext_next_allocated_block:
1589 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1590 * NOTE: it considers block number from index entry as
1591 * allocated block. Thus, index entries have to be consistent
1592 * with leaves.
1593 */
1594ext4_lblk_t
1595ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1596{
1597	int depth;
1598
1599	BUG_ON(path == NULL);
1600	depth = path->p_depth;
1601
1602	if (depth == 0 && path->p_ext == NULL)
1603		return EXT_MAX_BLOCKS;
1604
1605	while (depth >= 0) {
1606		if (depth == path->p_depth) {
1607			/* leaf */
1608			if (path[depth].p_ext &&
1609				path[depth].p_ext !=
1610					EXT_LAST_EXTENT(path[depth].p_hdr))
1611			  return le32_to_cpu(path[depth].p_ext[1].ee_block);
1612		} else {
1613			/* index */
1614			if (path[depth].p_idx !=
1615					EXT_LAST_INDEX(path[depth].p_hdr))
1616			  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1617		}
1618		depth--;
1619	}
1620
1621	return EXT_MAX_BLOCKS;
1622}
1623
1624/*
1625 * ext4_ext_next_leaf_block:
1626 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1627 */
1628static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1629{
1630	int depth;
1631
1632	BUG_ON(path == NULL);
1633	depth = path->p_depth;
1634
1635	/* zero-tree has no leaf blocks at all */
1636	if (depth == 0)
1637		return EXT_MAX_BLOCKS;
1638
1639	/* go to index block */
1640	depth--;
1641
1642	while (depth >= 0) {
1643		if (path[depth].p_idx !=
1644				EXT_LAST_INDEX(path[depth].p_hdr))
1645			return (ext4_lblk_t)
1646				le32_to_cpu(path[depth].p_idx[1].ei_block);
1647		depth--;
1648	}
1649
1650	return EXT_MAX_BLOCKS;
1651}
1652
1653/*
1654 * ext4_ext_correct_indexes:
1655 * if leaf gets modified and modified extent is first in the leaf,
1656 * then we have to correct all indexes above.
1657 * TODO: do we need to correct tree in all cases?
1658 */
1659static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1660				struct ext4_ext_path *path)
1661{
1662	struct ext4_extent_header *eh;
1663	int depth = ext_depth(inode);
1664	struct ext4_extent *ex;
1665	__le32 border;
1666	int k, err = 0;
1667
1668	eh = path[depth].p_hdr;
1669	ex = path[depth].p_ext;
1670
1671	if (unlikely(ex == NULL || eh == NULL)) {
1672		EXT4_ERROR_INODE(inode,
1673				 "ex %p == NULL or eh %p == NULL", ex, eh);
1674		return -EFSCORRUPTED;
1675	}
1676
1677	if (depth == 0) {
1678		/* there is no tree at all */
1679		return 0;
1680	}
1681
1682	if (ex != EXT_FIRST_EXTENT(eh)) {
1683		/* we correct tree if first leaf got modified only */
1684		return 0;
1685	}
1686
1687	/*
1688	 * TODO: we need correction if border is smaller than current one
1689	 */
1690	k = depth - 1;
1691	border = path[depth].p_ext->ee_block;
1692	err = ext4_ext_get_access(handle, inode, path + k);
1693	if (err)
1694		return err;
1695	path[k].p_idx->ei_block = border;
1696	err = ext4_ext_dirty(handle, inode, path + k);
1697	if (err)
1698		return err;
1699
1700	while (k--) {
1701		/* change all left-side indexes */
1702		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1703			break;
1704		err = ext4_ext_get_access(handle, inode, path + k);
1705		if (err)
1706			break;
1707		path[k].p_idx->ei_block = border;
1708		err = ext4_ext_dirty(handle, inode, path + k);
1709		if (err)
1710			break;
1711	}
1712
1713	return err;
1714}
1715
1716int
1717ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1718				struct ext4_extent *ex2)
1719{
1720	unsigned short ext1_ee_len, ext2_ee_len;
1721
1722	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1723		return 0;
1724
1725	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1726	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1727
1728	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1729			le32_to_cpu(ex2->ee_block))
1730		return 0;
1731
1732	/*
1733	 * To allow future support for preallocated extents to be added
1734	 * as an RO_COMPAT feature, refuse to merge to extents if
1735	 * this can result in the top bit of ee_len being set.
1736	 */
1737	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1738		return 0;
1739	/*
1740	 * The check for IO to unwritten extent is somewhat racy as we
1741	 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1742	 * dropping i_data_sem. But reserved blocks should save us in that
1743	 * case.
1744	 */
1745	if (ext4_ext_is_unwritten(ex1) &&
1746	    (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1747	     atomic_read(&EXT4_I(inode)->i_unwritten) ||
1748	     (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1749		return 0;
1750#ifdef AGGRESSIVE_TEST
1751	if (ext1_ee_len >= 4)
1752		return 0;
1753#endif
1754
1755	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1756		return 1;
1757	return 0;
1758}
1759
1760/*
1761 * This function tries to merge the "ex" extent to the next extent in the tree.
1762 * It always tries to merge towards right. If you want to merge towards
1763 * left, pass "ex - 1" as argument instead of "ex".
1764 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1765 * 1 if they got merged.
1766 */
1767static int ext4_ext_try_to_merge_right(struct inode *inode,
1768				 struct ext4_ext_path *path,
1769				 struct ext4_extent *ex)
1770{
1771	struct ext4_extent_header *eh;
1772	unsigned int depth, len;
1773	int merge_done = 0, unwritten;
1774
1775	depth = ext_depth(inode);
1776	BUG_ON(path[depth].p_hdr == NULL);
1777	eh = path[depth].p_hdr;
1778
1779	while (ex < EXT_LAST_EXTENT(eh)) {
1780		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1781			break;
1782		/* merge with next extent! */
1783		unwritten = ext4_ext_is_unwritten(ex);
1784		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1785				+ ext4_ext_get_actual_len(ex + 1));
1786		if (unwritten)
1787			ext4_ext_mark_unwritten(ex);
1788
1789		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1790			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1791				* sizeof(struct ext4_extent);
1792			memmove(ex + 1, ex + 2, len);
1793		}
1794		le16_add_cpu(&eh->eh_entries, -1);
1795		merge_done = 1;
1796		WARN_ON(eh->eh_entries == 0);
1797		if (!eh->eh_entries)
1798			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1799	}
1800
1801	return merge_done;
1802}
1803
1804/*
1805 * This function does a very simple check to see if we can collapse
1806 * an extent tree with a single extent tree leaf block into the inode.
1807 */
1808static void ext4_ext_try_to_merge_up(handle_t *handle,
1809				     struct inode *inode,
1810				     struct ext4_ext_path *path)
1811{
1812	size_t s;
1813	unsigned max_root = ext4_ext_space_root(inode, 0);
1814	ext4_fsblk_t blk;
1815
1816	if ((path[0].p_depth != 1) ||
1817	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1818	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1819		return;
1820
1821	/*
1822	 * We need to modify the block allocation bitmap and the block
1823	 * group descriptor to release the extent tree block.  If we
1824	 * can't get the journal credits, give up.
1825	 */
1826	if (ext4_journal_extend(handle, 2))
1827		return;
1828
1829	/*
1830	 * Copy the extent data up to the inode
1831	 */
1832	blk = ext4_idx_pblock(path[0].p_idx);
1833	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1834		sizeof(struct ext4_extent_idx);
1835	s += sizeof(struct ext4_extent_header);
1836
1837	path[1].p_maxdepth = path[0].p_maxdepth;
1838	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1839	path[0].p_depth = 0;
1840	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1841		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1842	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1843
1844	brelse(path[1].p_bh);
1845	ext4_free_blocks(handle, inode, NULL, blk, 1,
1846			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1847}
1848
1849/*
1850 * This function tries to merge the @ex extent to neighbours in the tree.
1851 * return 1 if merge left else 0.
1852 */
1853static void ext4_ext_try_to_merge(handle_t *handle,
1854				  struct inode *inode,
1855				  struct ext4_ext_path *path,
1856				  struct ext4_extent *ex) {
1857	struct ext4_extent_header *eh;
1858	unsigned int depth;
1859	int merge_done = 0;
1860
1861	depth = ext_depth(inode);
1862	BUG_ON(path[depth].p_hdr == NULL);
1863	eh = path[depth].p_hdr;
1864
1865	if (ex > EXT_FIRST_EXTENT(eh))
1866		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1867
1868	if (!merge_done)
1869		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1870
1871	ext4_ext_try_to_merge_up(handle, inode, path);
1872}
1873
1874/*
1875 * check if a portion of the "newext" extent overlaps with an
1876 * existing extent.
1877 *
1878 * If there is an overlap discovered, it updates the length of the newext
1879 * such that there will be no overlap, and then returns 1.
1880 * If there is no overlap found, it returns 0.
1881 */
1882static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1883					   struct inode *inode,
1884					   struct ext4_extent *newext,
1885					   struct ext4_ext_path *path)
1886{
1887	ext4_lblk_t b1, b2;
1888	unsigned int depth, len1;
1889	unsigned int ret = 0;
1890
1891	b1 = le32_to_cpu(newext->ee_block);
1892	len1 = ext4_ext_get_actual_len(newext);
1893	depth = ext_depth(inode);
1894	if (!path[depth].p_ext)
1895		goto out;
1896	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1897
1898	/*
1899	 * get the next allocated block if the extent in the path
1900	 * is before the requested block(s)
1901	 */
1902	if (b2 < b1) {
1903		b2 = ext4_ext_next_allocated_block(path);
1904		if (b2 == EXT_MAX_BLOCKS)
1905			goto out;
1906		b2 = EXT4_LBLK_CMASK(sbi, b2);
1907	}
1908
1909	/* check for wrap through zero on extent logical start block*/
1910	if (b1 + len1 < b1) {
1911		len1 = EXT_MAX_BLOCKS - b1;
1912		newext->ee_len = cpu_to_le16(len1);
1913		ret = 1;
1914	}
1915
1916	/* check for overlap */
1917	if (b1 + len1 > b2) {
1918		newext->ee_len = cpu_to_le16(b2 - b1);
1919		ret = 1;
1920	}
1921out:
1922	return ret;
1923}
1924
1925/*
1926 * ext4_ext_insert_extent:
1927 * tries to merge requsted extent into the existing extent or
1928 * inserts requested extent as new one into the tree,
1929 * creating new leaf in the no-space case.
1930 */
1931int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1932				struct ext4_ext_path **ppath,
1933				struct ext4_extent *newext, int gb_flags)
1934{
1935	struct ext4_ext_path *path = *ppath;
1936	struct ext4_extent_header *eh;
1937	struct ext4_extent *ex, *fex;
1938	struct ext4_extent *nearex; /* nearest extent */
1939	struct ext4_ext_path *npath = NULL;
1940	int depth, len, err;
1941	ext4_lblk_t next;
1942	int mb_flags = 0, unwritten;
1943
1944	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1945		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1946	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1947		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1948		return -EFSCORRUPTED;
1949	}
1950	depth = ext_depth(inode);
1951	ex = path[depth].p_ext;
1952	eh = path[depth].p_hdr;
1953	if (unlikely(path[depth].p_hdr == NULL)) {
1954		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1955		return -EFSCORRUPTED;
1956	}
1957
1958	/* try to insert block into found extent and return */
1959	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1960
1961		/*
1962		 * Try to see whether we should rather test the extent on
1963		 * right from ex, or from the left of ex. This is because
1964		 * ext4_find_extent() can return either extent on the
1965		 * left, or on the right from the searched position. This
1966		 * will make merging more effective.
1967		 */
1968		if (ex < EXT_LAST_EXTENT(eh) &&
1969		    (le32_to_cpu(ex->ee_block) +
1970		    ext4_ext_get_actual_len(ex) <
1971		    le32_to_cpu(newext->ee_block))) {
1972			ex += 1;
1973			goto prepend;
1974		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1975			   (le32_to_cpu(newext->ee_block) +
1976			   ext4_ext_get_actual_len(newext) <
1977			   le32_to_cpu(ex->ee_block)))
1978			ex -= 1;
1979
1980		/* Try to append newex to the ex */
1981		if (ext4_can_extents_be_merged(inode, ex, newext)) {
1982			ext_debug("append [%d]%d block to %u:[%d]%d"
1983				  "(from %llu)\n",
1984				  ext4_ext_is_unwritten(newext),
1985				  ext4_ext_get_actual_len(newext),
1986				  le32_to_cpu(ex->ee_block),
1987				  ext4_ext_is_unwritten(ex),
1988				  ext4_ext_get_actual_len(ex),
1989				  ext4_ext_pblock(ex));
1990			err = ext4_ext_get_access(handle, inode,
1991						  path + depth);
1992			if (err)
1993				return err;
1994			unwritten = ext4_ext_is_unwritten(ex);
1995			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1996					+ ext4_ext_get_actual_len(newext));
1997			if (unwritten)
1998				ext4_ext_mark_unwritten(ex);
1999			eh = path[depth].p_hdr;
2000			nearex = ex;
2001			goto merge;
2002		}
2003
2004prepend:
2005		/* Try to prepend newex to the ex */
2006		if (ext4_can_extents_be_merged(inode, newext, ex)) {
2007			ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2008				  "(from %llu)\n",
2009				  le32_to_cpu(newext->ee_block),
2010				  ext4_ext_is_unwritten(newext),
2011				  ext4_ext_get_actual_len(newext),
2012				  le32_to_cpu(ex->ee_block),
2013				  ext4_ext_is_unwritten(ex),
2014				  ext4_ext_get_actual_len(ex),
2015				  ext4_ext_pblock(ex));
2016			err = ext4_ext_get_access(handle, inode,
2017						  path + depth);
2018			if (err)
2019				return err;
2020
2021			unwritten = ext4_ext_is_unwritten(ex);
2022			ex->ee_block = newext->ee_block;
2023			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2024			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2025					+ ext4_ext_get_actual_len(newext));
2026			if (unwritten)
2027				ext4_ext_mark_unwritten(ex);
2028			eh = path[depth].p_hdr;
2029			nearex = ex;
2030			goto merge;
2031		}
2032	}
2033
2034	depth = ext_depth(inode);
2035	eh = path[depth].p_hdr;
2036	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2037		goto has_space;
2038
2039	/* probably next leaf has space for us? */
2040	fex = EXT_LAST_EXTENT(eh);
2041	next = EXT_MAX_BLOCKS;
2042	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2043		next = ext4_ext_next_leaf_block(path);
2044	if (next != EXT_MAX_BLOCKS) {
2045		ext_debug("next leaf block - %u\n", next);
2046		BUG_ON(npath != NULL);
2047		npath = ext4_find_extent(inode, next, NULL, 0);
2048		if (IS_ERR(npath))
2049			return PTR_ERR(npath);
2050		BUG_ON(npath->p_depth != path->p_depth);
2051		eh = npath[depth].p_hdr;
2052		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2053			ext_debug("next leaf isn't full(%d)\n",
2054				  le16_to_cpu(eh->eh_entries));
2055			path = npath;
2056			goto has_space;
2057		}
2058		ext_debug("next leaf has no free space(%d,%d)\n",
2059			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2060	}
2061
2062	/*
2063	 * There is no free space in the found leaf.
2064	 * We're gonna add a new leaf in the tree.
2065	 */
2066	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2067		mb_flags |= EXT4_MB_USE_RESERVED;
2068	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2069				       ppath, newext);
2070	if (err)
2071		goto cleanup;
2072	depth = ext_depth(inode);
2073	eh = path[depth].p_hdr;
2074
2075has_space:
2076	nearex = path[depth].p_ext;
2077
2078	err = ext4_ext_get_access(handle, inode, path + depth);
2079	if (err)
2080		goto cleanup;
2081
2082	if (!nearex) {
2083		/* there is no extent in this leaf, create first one */
2084		ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2085				le32_to_cpu(newext->ee_block),
2086				ext4_ext_pblock(newext),
2087				ext4_ext_is_unwritten(newext),
2088				ext4_ext_get_actual_len(newext));
2089		nearex = EXT_FIRST_EXTENT(eh);
2090	} else {
2091		if (le32_to_cpu(newext->ee_block)
2092			   > le32_to_cpu(nearex->ee_block)) {
2093			/* Insert after */
2094			ext_debug("insert %u:%llu:[%d]%d before: "
2095					"nearest %p\n",
2096					le32_to_cpu(newext->ee_block),
2097					ext4_ext_pblock(newext),
2098					ext4_ext_is_unwritten(newext),
2099					ext4_ext_get_actual_len(newext),
2100					nearex);
2101			nearex++;
2102		} else {
2103			/* Insert before */
2104			BUG_ON(newext->ee_block == nearex->ee_block);
2105			ext_debug("insert %u:%llu:[%d]%d after: "
2106					"nearest %p\n",
2107					le32_to_cpu(newext->ee_block),
2108					ext4_ext_pblock(newext),
2109					ext4_ext_is_unwritten(newext),
2110					ext4_ext_get_actual_len(newext),
2111					nearex);
2112		}
2113		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2114		if (len > 0) {
2115			ext_debug("insert %u:%llu:[%d]%d: "
2116					"move %d extents from 0x%p to 0x%p\n",
2117					le32_to_cpu(newext->ee_block),
2118					ext4_ext_pblock(newext),
2119					ext4_ext_is_unwritten(newext),
2120					ext4_ext_get_actual_len(newext),
2121					len, nearex, nearex + 1);
2122			memmove(nearex + 1, nearex,
2123				len * sizeof(struct ext4_extent));
2124		}
2125	}
2126
2127	le16_add_cpu(&eh->eh_entries, 1);
2128	path[depth].p_ext = nearex;
2129	nearex->ee_block = newext->ee_block;
2130	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2131	nearex->ee_len = newext->ee_len;
2132
2133merge:
2134	/* try to merge extents */
2135	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2136		ext4_ext_try_to_merge(handle, inode, path, nearex);
2137
2138
2139	/* time to correct all indexes above */
2140	err = ext4_ext_correct_indexes(handle, inode, path);
2141	if (err)
2142		goto cleanup;
2143
2144	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2145
2146cleanup:
2147	ext4_ext_drop_refs(npath);
2148	kfree(npath);
2149	return err;
2150}
2151
2152static int ext4_fill_fiemap_extents(struct inode *inode,
2153				    ext4_lblk_t block, ext4_lblk_t num,
2154				    struct fiemap_extent_info *fieinfo)
2155{
2156	struct ext4_ext_path *path = NULL;
2157	struct ext4_extent *ex;
2158	struct extent_status es;
2159	ext4_lblk_t next, next_del, start = 0, end = 0;
2160	ext4_lblk_t last = block + num;
2161	int exists, depth = 0, err = 0;
2162	unsigned int flags = 0;
2163	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2164
2165	while (block < last && block != EXT_MAX_BLOCKS) {
2166		num = last - block;
2167		/* find extent for this block */
2168		down_read(&EXT4_I(inode)->i_data_sem);
2169
2170		path = ext4_find_extent(inode, block, &path, 0);
2171		if (IS_ERR(path)) {
2172			up_read(&EXT4_I(inode)->i_data_sem);
2173			err = PTR_ERR(path);
2174			path = NULL;
2175			break;
2176		}
2177
2178		depth = ext_depth(inode);
2179		if (unlikely(path[depth].p_hdr == NULL)) {
2180			up_read(&EXT4_I(inode)->i_data_sem);
2181			EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2182			err = -EFSCORRUPTED;
2183			break;
2184		}
2185		ex = path[depth].p_ext;
2186		next = ext4_ext_next_allocated_block(path);
2187
2188		flags = 0;
2189		exists = 0;
2190		if (!ex) {
2191			/* there is no extent yet, so try to allocate
2192			 * all requested space */
2193			start = block;
2194			end = block + num;
2195		} else if (le32_to_cpu(ex->ee_block) > block) {
2196			/* need to allocate space before found extent */
2197			start = block;
2198			end = le32_to_cpu(ex->ee_block);
2199			if (block + num < end)
2200				end = block + num;
2201		} else if (block >= le32_to_cpu(ex->ee_block)
2202					+ ext4_ext_get_actual_len(ex)) {
2203			/* need to allocate space after found extent */
2204			start = block;
2205			end = block + num;
2206			if (end >= next)
2207				end = next;
2208		} else if (block >= le32_to_cpu(ex->ee_block)) {
2209			/*
2210			 * some part of requested space is covered
2211			 * by found extent
2212			 */
2213			start = block;
2214			end = le32_to_cpu(ex->ee_block)
2215				+ ext4_ext_get_actual_len(ex);
2216			if (block + num < end)
2217				end = block + num;
2218			exists = 1;
2219		} else {
2220			BUG();
2221		}
2222		BUG_ON(end <= start);
2223
2224		if (!exists) {
2225			es.es_lblk = start;
2226			es.es_len = end - start;
2227			es.es_pblk = 0;
2228		} else {
2229			es.es_lblk = le32_to_cpu(ex->ee_block);
2230			es.es_len = ext4_ext_get_actual_len(ex);
2231			es.es_pblk = ext4_ext_pblock(ex);
2232			if (ext4_ext_is_unwritten(ex))
2233				flags |= FIEMAP_EXTENT_UNWRITTEN;
2234		}
2235
2236		/*
2237		 * Find delayed extent and update es accordingly. We call
2238		 * it even in !exists case to find out whether es is the
2239		 * last existing extent or not.
2240		 */
2241		next_del = ext4_find_delayed_extent(inode, &es);
2242		if (!exists && next_del) {
2243			exists = 1;
2244			flags |= (FIEMAP_EXTENT_DELALLOC |
2245				  FIEMAP_EXTENT_UNKNOWN);
2246		}
2247		up_read(&EXT4_I(inode)->i_data_sem);
2248
2249		if (unlikely(es.es_len == 0)) {
2250			EXT4_ERROR_INODE(inode, "es.es_len == 0");
2251			err = -EFSCORRUPTED;
2252			break;
2253		}
2254
2255		/*
2256		 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2257		 * we need to check next == EXT_MAX_BLOCKS because it is
2258		 * possible that an extent is with unwritten and delayed
2259		 * status due to when an extent is delayed allocated and
2260		 * is allocated by fallocate status tree will track both of
2261		 * them in a extent.
2262		 *
2263		 * So we could return a unwritten and delayed extent, and
2264		 * its block is equal to 'next'.
2265		 */
2266		if (next == next_del && next == EXT_MAX_BLOCKS) {
2267			flags |= FIEMAP_EXTENT_LAST;
2268			if (unlikely(next_del != EXT_MAX_BLOCKS ||
2269				     next != EXT_MAX_BLOCKS)) {
2270				EXT4_ERROR_INODE(inode,
2271						 "next extent == %u, next "
2272						 "delalloc extent = %u",
2273						 next, next_del);
2274				err = -EFSCORRUPTED;
2275				break;
2276			}
2277		}
2278
2279		if (exists) {
2280			err = fiemap_fill_next_extent(fieinfo,
2281				(__u64)es.es_lblk << blksize_bits,
2282				(__u64)es.es_pblk << blksize_bits,
2283				(__u64)es.es_len << blksize_bits,
2284				flags);
2285			if (err < 0)
2286				break;
2287			if (err == 1) {
2288				err = 0;
2289				break;
2290			}
2291		}
2292
2293		block = es.es_lblk + es.es_len;
2294	}
2295
2296	ext4_ext_drop_refs(path);
2297	kfree(path);
2298	return err;
2299}
2300
2301/*
2302 * ext4_ext_determine_hole - determine hole around given block
2303 * @inode:	inode we lookup in
2304 * @path:	path in extent tree to @lblk
2305 * @lblk:	pointer to logical block around which we want to determine hole
2306 *
2307 * Determine hole length (and start if easily possible) around given logical
2308 * block. We don't try too hard to find the beginning of the hole but @path
2309 * actually points to extent before @lblk, we provide it.
2310 *
2311 * The function returns the length of a hole starting at @lblk. We update @lblk
2312 * to the beginning of the hole if we managed to find it.
2313 */
2314static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2315					   struct ext4_ext_path *path,
2316					   ext4_lblk_t *lblk)
2317{
2318	int depth = ext_depth(inode);
2319	struct ext4_extent *ex;
2320	ext4_lblk_t len;
2321
2322	ex = path[depth].p_ext;
2323	if (ex == NULL) {
2324		/* there is no extent yet, so gap is [0;-] */
2325		*lblk = 0;
2326		len = EXT_MAX_BLOCKS;
2327	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2328		len = le32_to_cpu(ex->ee_block) - *lblk;
2329	} else if (*lblk >= le32_to_cpu(ex->ee_block)
2330			+ ext4_ext_get_actual_len(ex)) {
2331		ext4_lblk_t next;
2332
2333		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2334		next = ext4_ext_next_allocated_block(path);
2335		BUG_ON(next == *lblk);
2336		len = next - *lblk;
2337	} else {
2338		BUG();
2339	}
2340	return len;
2341}
2342
2343/*
2344 * ext4_ext_put_gap_in_cache:
2345 * calculate boundaries of the gap that the requested block fits into
2346 * and cache this gap
2347 */
2348static void
2349ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2350			  ext4_lblk_t hole_len)
2351{
2352	struct extent_status es;
2353
2354	ext4_es_find_delayed_extent_range(inode, hole_start,
2355					  hole_start + hole_len - 1, &es);
2356	if (es.es_len) {
2357		/* There's delayed extent containing lblock? */
2358		if (es.es_lblk <= hole_start)
2359			return;
2360		hole_len = min(es.es_lblk - hole_start, hole_len);
2361	}
2362	ext_debug(" -> %u:%u\n", hole_start, hole_len);
2363	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2364			      EXTENT_STATUS_HOLE);
2365}
2366
2367/*
2368 * ext4_ext_rm_idx:
2369 * removes index from the index block.
2370 */
2371static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2372			struct ext4_ext_path *path, int depth)
2373{
2374	int err;
2375	ext4_fsblk_t leaf;
2376
2377	/* free index block */
2378	depth--;
2379	path = path + depth;
2380	leaf = ext4_idx_pblock(path->p_idx);
2381	if (unlikely(path->p_hdr->eh_entries == 0)) {
2382		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2383		return -EFSCORRUPTED;
2384	}
2385	err = ext4_ext_get_access(handle, inode, path);
2386	if (err)
2387		return err;
2388
2389	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2390		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2391		len *= sizeof(struct ext4_extent_idx);
2392		memmove(path->p_idx, path->p_idx + 1, len);
2393	}
2394
2395	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2396	err = ext4_ext_dirty(handle, inode, path);
2397	if (err)
2398		return err;
2399	ext_debug("index is empty, remove it, free block %llu\n", leaf);
2400	trace_ext4_ext_rm_idx(inode, leaf);
2401
2402	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2403			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2404
2405	while (--depth >= 0) {
2406		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2407			break;
2408		path--;
2409		err = ext4_ext_get_access(handle, inode, path);
2410		if (err)
2411			break;
2412		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2413		err = ext4_ext_dirty(handle, inode, path);
2414		if (err)
2415			break;
2416	}
2417	return err;
2418}
2419
2420/*
2421 * ext4_ext_calc_credits_for_single_extent:
2422 * This routine returns max. credits that needed to insert an extent
2423 * to the extent tree.
2424 * When pass the actual path, the caller should calculate credits
2425 * under i_data_sem.
2426 */
2427int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2428						struct ext4_ext_path *path)
2429{
2430	if (path) {
2431		int depth = ext_depth(inode);
2432		int ret = 0;
2433
2434		/* probably there is space in leaf? */
2435		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2436				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2437
2438			/*
2439			 *  There are some space in the leaf tree, no
2440			 *  need to account for leaf block credit
2441			 *
2442			 *  bitmaps and block group descriptor blocks
2443			 *  and other metadata blocks still need to be
2444			 *  accounted.
2445			 */
2446			/* 1 bitmap, 1 block group descriptor */
2447			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2448			return ret;
2449		}
2450	}
2451
2452	return ext4_chunk_trans_blocks(inode, nrblocks);
2453}
2454
2455/*
2456 * How many index/leaf blocks need to change/allocate to add @extents extents?
2457 *
2458 * If we add a single extent, then in the worse case, each tree level
2459 * index/leaf need to be changed in case of the tree split.
2460 *
2461 * If more extents are inserted, they could cause the whole tree split more
2462 * than once, but this is really rare.
2463 */
2464int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2465{
2466	int index;
2467	int depth;
2468
2469	/* If we are converting the inline data, only one is needed here. */
2470	if (ext4_has_inline_data(inode))
2471		return 1;
2472
2473	depth = ext_depth(inode);
2474
2475	if (extents <= 1)
2476		index = depth * 2;
2477	else
2478		index = depth * 3;
2479
2480	return index;
2481}
2482
2483static inline int get_default_free_blocks_flags(struct inode *inode)
2484{
2485	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2486		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2487	else if (ext4_should_journal_data(inode))
2488		return EXT4_FREE_BLOCKS_FORGET;
2489	return 0;
2490}
2491
2492static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2493			      struct ext4_extent *ex,
2494			      long long *partial_cluster,
2495			      ext4_lblk_t from, ext4_lblk_t to)
2496{
2497	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2498	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2499	ext4_fsblk_t pblk;
2500	int flags = get_default_free_blocks_flags(inode);
2501
2502	/*
2503	 * For bigalloc file systems, we never free a partial cluster
2504	 * at the beginning of the extent.  Instead, we make a note
2505	 * that we tried freeing the cluster, and check to see if we
2506	 * need to free it on a subsequent call to ext4_remove_blocks,
2507	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2508	 */
2509	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2510
2511	trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2512	/*
2513	 * If we have a partial cluster, and it's different from the
2514	 * cluster of the last block, we need to explicitly free the
2515	 * partial cluster here.
2516	 */
2517	pblk = ext4_ext_pblock(ex) + ee_len - 1;
2518	if (*partial_cluster > 0 &&
2519	    *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2520		ext4_free_blocks(handle, inode, NULL,
2521				 EXT4_C2B(sbi, *partial_cluster),
2522				 sbi->s_cluster_ratio, flags);
2523		*partial_cluster = 0;
2524	}
2525
2526#ifdef EXTENTS_STATS
2527	{
2528		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2529		spin_lock(&sbi->s_ext_stats_lock);
2530		sbi->s_ext_blocks += ee_len;
2531		sbi->s_ext_extents++;
2532		if (ee_len < sbi->s_ext_min)
2533			sbi->s_ext_min = ee_len;
2534		if (ee_len > sbi->s_ext_max)
2535			sbi->s_ext_max = ee_len;
2536		if (ext_depth(inode) > sbi->s_depth_max)
2537			sbi->s_depth_max = ext_depth(inode);
2538		spin_unlock(&sbi->s_ext_stats_lock);
2539	}
2540#endif
2541	if (from >= le32_to_cpu(ex->ee_block)
2542	    && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2543		/* tail removal */
2544		ext4_lblk_t num;
2545		long long first_cluster;
2546
2547		num = le32_to_cpu(ex->ee_block) + ee_len - from;
2548		pblk = ext4_ext_pblock(ex) + ee_len - num;
2549		/*
2550		 * Usually we want to free partial cluster at the end of the
2551		 * extent, except for the situation when the cluster is still
2552		 * used by any other extent (partial_cluster is negative).
2553		 */
2554		if (*partial_cluster < 0 &&
2555		    *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2556			flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2557
2558		ext_debug("free last %u blocks starting %llu partial %lld\n",
2559			  num, pblk, *partial_cluster);
2560		ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2561		/*
2562		 * If the block range to be freed didn't start at the
2563		 * beginning of a cluster, and we removed the entire
2564		 * extent and the cluster is not used by any other extent,
2565		 * save the partial cluster here, since we might need to
2566		 * delete if we determine that the truncate or punch hole
2567		 * operation has removed all of the blocks in the cluster.
2568		 * If that cluster is used by another extent, preserve its
2569		 * negative value so it isn't freed later on.
2570		 *
2571		 * If the whole extent wasn't freed, we've reached the
2572		 * start of the truncated/punched region and have finished
2573		 * removing blocks.  If there's a partial cluster here it's
2574		 * shared with the remainder of the extent and is no longer
2575		 * a candidate for removal.
2576		 */
2577		if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2578			first_cluster = (long long) EXT4_B2C(sbi, pblk);
2579			if (first_cluster != -*partial_cluster)
2580				*partial_cluster = first_cluster;
2581		} else {
2582			*partial_cluster = 0;
2583		}
2584	} else
2585		ext4_error(sbi->s_sb, "strange request: removal(2) "
2586			   "%u-%u from %u:%u\n",
2587			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2588	return 0;
2589}
2590
2591
2592/*
2593 * ext4_ext_rm_leaf() Removes the extents associated with the
2594 * blocks appearing between "start" and "end".  Both "start"
2595 * and "end" must appear in the same extent or EIO is returned.
2596 *
2597 * @handle: The journal handle
2598 * @inode:  The files inode
2599 * @path:   The path to the leaf
2600 * @partial_cluster: The cluster which we'll have to free if all extents
2601 *                   has been released from it.  However, if this value is
2602 *                   negative, it's a cluster just to the right of the
2603 *                   punched region and it must not be freed.
2604 * @start:  The first block to remove
2605 * @end:   The last block to remove
2606 */
2607static int
2608ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2609		 struct ext4_ext_path *path,
2610		 long long *partial_cluster,
2611		 ext4_lblk_t start, ext4_lblk_t end)
2612{
2613	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2614	int err = 0, correct_index = 0;
2615	int depth = ext_depth(inode), credits;
2616	struct ext4_extent_header *eh;
2617	ext4_lblk_t a, b;
2618	unsigned num;
2619	ext4_lblk_t ex_ee_block;
2620	unsigned short ex_ee_len;
2621	unsigned unwritten = 0;
2622	struct ext4_extent *ex;
2623	ext4_fsblk_t pblk;
2624
2625	/* the header must be checked already in ext4_ext_remove_space() */
2626	ext_debug("truncate since %u in leaf to %u\n", start, end);
2627	if (!path[depth].p_hdr)
2628		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2629	eh = path[depth].p_hdr;
2630	if (unlikely(path[depth].p_hdr == NULL)) {
2631		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2632		return -EFSCORRUPTED;
2633	}
2634	/* find where to start removing */
2635	ex = path[depth].p_ext;
2636	if (!ex)
2637		ex = EXT_LAST_EXTENT(eh);
2638
2639	ex_ee_block = le32_to_cpu(ex->ee_block);
2640	ex_ee_len = ext4_ext_get_actual_len(ex);
2641
2642	trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2643
2644	while (ex >= EXT_FIRST_EXTENT(eh) &&
2645			ex_ee_block + ex_ee_len > start) {
2646
2647		if (ext4_ext_is_unwritten(ex))
2648			unwritten = 1;
2649		else
2650			unwritten = 0;
2651
2652		ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2653			  unwritten, ex_ee_len);
2654		path[depth].p_ext = ex;
2655
2656		a = ex_ee_block > start ? ex_ee_block : start;
2657		b = ex_ee_block+ex_ee_len - 1 < end ?
2658			ex_ee_block+ex_ee_len - 1 : end;
2659
2660		ext_debug("  border %u:%u\n", a, b);
2661
2662		/* If this extent is beyond the end of the hole, skip it */
2663		if (end < ex_ee_block) {
2664			/*
2665			 * We're going to skip this extent and move to another,
2666			 * so note that its first cluster is in use to avoid
2667			 * freeing it when removing blocks.  Eventually, the
2668			 * right edge of the truncated/punched region will
2669			 * be just to the left.
2670			 */
2671			if (sbi->s_cluster_ratio > 1) {
2672				pblk = ext4_ext_pblock(ex);
2673				*partial_cluster =
2674					-(long long) EXT4_B2C(sbi, pblk);
2675			}
2676			ex--;
2677			ex_ee_block = le32_to_cpu(ex->ee_block);
2678			ex_ee_len = ext4_ext_get_actual_len(ex);
2679			continue;
2680		} else if (b != ex_ee_block + ex_ee_len - 1) {
2681			EXT4_ERROR_INODE(inode,
2682					 "can not handle truncate %u:%u "
2683					 "on extent %u:%u",
2684					 start, end, ex_ee_block,
2685					 ex_ee_block + ex_ee_len - 1);
2686			err = -EFSCORRUPTED;
2687			goto out;
2688		} else if (a != ex_ee_block) {
2689			/* remove tail of the extent */
2690			num = a - ex_ee_block;
2691		} else {
2692			/* remove whole extent: excellent! */
2693			num = 0;
2694		}
2695		/*
2696		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2697		 * descriptor) for each block group; assume two block
2698		 * groups plus ex_ee_len/blocks_per_block_group for
2699		 * the worst case
2700		 */
2701		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2702		if (ex == EXT_FIRST_EXTENT(eh)) {
2703			correct_index = 1;
2704			credits += (ext_depth(inode)) + 1;
2705		}
2706		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2707
2708		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2709		if (err)
2710			goto out;
2711
2712		err = ext4_ext_get_access(handle, inode, path + depth);
2713		if (err)
2714			goto out;
2715
2716		err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2717					 a, b);
2718		if (err)
2719			goto out;
2720
2721		if (num == 0)
2722			/* this extent is removed; mark slot entirely unused */
2723			ext4_ext_store_pblock(ex, 0);
2724
2725		ex->ee_len = cpu_to_le16(num);
2726		/*
2727		 * Do not mark unwritten if all the blocks in the
2728		 * extent have been removed.
2729		 */
2730		if (unwritten && num)
2731			ext4_ext_mark_unwritten(ex);
2732		/*
2733		 * If the extent was completely released,
2734		 * we need to remove it from the leaf
2735		 */
2736		if (num == 0) {
2737			if (end != EXT_MAX_BLOCKS - 1) {
2738				/*
2739				 * For hole punching, we need to scoot all the
2740				 * extents up when an extent is removed so that
2741				 * we dont have blank extents in the middle
2742				 */
2743				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2744					sizeof(struct ext4_extent));
2745
2746				/* Now get rid of the one at the end */
2747				memset(EXT_LAST_EXTENT(eh), 0,
2748					sizeof(struct ext4_extent));
2749			}
2750			le16_add_cpu(&eh->eh_entries, -1);
2751		}
2752
2753		err = ext4_ext_dirty(handle, inode, path + depth);
2754		if (err)
2755			goto out;
2756
2757		ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2758				ext4_ext_pblock(ex));
2759		ex--;
2760		ex_ee_block = le32_to_cpu(ex->ee_block);
2761		ex_ee_len = ext4_ext_get_actual_len(ex);
2762	}
2763
2764	if (correct_index && eh->eh_entries)
2765		err = ext4_ext_correct_indexes(handle, inode, path);
2766
2767	/*
2768	 * If there's a partial cluster and at least one extent remains in
2769	 * the leaf, free the partial cluster if it isn't shared with the
2770	 * current extent.  If it is shared with the current extent
2771	 * we zero partial_cluster because we've reached the start of the
2772	 * truncated/punched region and we're done removing blocks.
2773	 */
2774	if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2775		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2776		if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2777			ext4_free_blocks(handle, inode, NULL,
2778					 EXT4_C2B(sbi, *partial_cluster),
2779					 sbi->s_cluster_ratio,
2780					 get_default_free_blocks_flags(inode));
2781		}
2782		*partial_cluster = 0;
2783	}
2784
2785	/* if this leaf is free, then we should
2786	 * remove it from index block above */
2787	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2788		err = ext4_ext_rm_idx(handle, inode, path, depth);
2789
2790out:
2791	return err;
2792}
2793
2794/*
2795 * ext4_ext_more_to_rm:
2796 * returns 1 if current index has to be freed (even partial)
2797 */
2798static int
2799ext4_ext_more_to_rm(struct ext4_ext_path *path)
2800{
2801	BUG_ON(path->p_idx == NULL);
2802
2803	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2804		return 0;
2805
2806	/*
2807	 * if truncate on deeper level happened, it wasn't partial,
2808	 * so we have to consider current index for truncation
2809	 */
2810	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2811		return 0;
2812	return 1;
2813}
2814
2815int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2816			  ext4_lblk_t end)
2817{
2818	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2819	int depth = ext_depth(inode);
2820	struct ext4_ext_path *path = NULL;
2821	long long partial_cluster = 0;
2822	handle_t *handle;
2823	int i = 0, err = 0;
2824
2825	ext_debug("truncate since %u to %u\n", start, end);
2826
2827	/* probably first extent we're gonna free will be last in block */
2828	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2829	if (IS_ERR(handle))
2830		return PTR_ERR(handle);
2831
2832again:
2833	trace_ext4_ext_remove_space(inode, start, end, depth);
2834
2835	/*
2836	 * Check if we are removing extents inside the extent tree. If that
2837	 * is the case, we are going to punch a hole inside the extent tree
2838	 * so we have to check whether we need to split the extent covering
2839	 * the last block to remove so we can easily remove the part of it
2840	 * in ext4_ext_rm_leaf().
2841	 */
2842	if (end < EXT_MAX_BLOCKS - 1) {
2843		struct ext4_extent *ex;
2844		ext4_lblk_t ee_block, ex_end, lblk;
2845		ext4_fsblk_t pblk;
2846
2847		/* find extent for or closest extent to this block */
2848		path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2849		if (IS_ERR(path)) {
2850			ext4_journal_stop(handle);
2851			return PTR_ERR(path);
2852		}
2853		depth = ext_depth(inode);
2854		/* Leaf not may not exist only if inode has no blocks at all */
2855		ex = path[depth].p_ext;
2856		if (!ex) {
2857			if (depth) {
2858				EXT4_ERROR_INODE(inode,
2859						 "path[%d].p_hdr == NULL",
2860						 depth);
2861				err = -EFSCORRUPTED;
2862			}
2863			goto out;
2864		}
2865
2866		ee_block = le32_to_cpu(ex->ee_block);
2867		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2868
2869		/*
2870		 * See if the last block is inside the extent, if so split
2871		 * the extent at 'end' block so we can easily remove the
2872		 * tail of the first part of the split extent in
2873		 * ext4_ext_rm_leaf().
2874		 */
2875		if (end >= ee_block && end < ex_end) {
2876
2877			/*
2878			 * If we're going to split the extent, note that
2879			 * the cluster containing the block after 'end' is
2880			 * in use to avoid freeing it when removing blocks.
2881			 */
2882			if (sbi->s_cluster_ratio > 1) {
2883				pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2884				partial_cluster =
2885					-(long long) EXT4_B2C(sbi, pblk);
2886			}
2887
2888			/*
2889			 * Split the extent in two so that 'end' is the last
2890			 * block in the first new extent. Also we should not
2891			 * fail removing space due to ENOSPC so try to use
2892			 * reserved block if that happens.
2893			 */
2894			err = ext4_force_split_extent_at(handle, inode, &path,
2895							 end + 1, 1);
2896			if (err < 0)
2897				goto out;
2898
2899		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2900			/*
2901			 * If there's an extent to the right its first cluster
2902			 * contains the immediate right boundary of the
2903			 * truncated/punched region.  Set partial_cluster to
2904			 * its negative value so it won't be freed if shared
2905			 * with the current extent.  The end < ee_block case
2906			 * is handled in ext4_ext_rm_leaf().
2907			 */
2908			lblk = ex_end + 1;
2909			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2910						    &ex);
2911			if (err)
2912				goto out;
2913			if (pblk)
2914				partial_cluster =
2915					-(long long) EXT4_B2C(sbi, pblk);
2916		}
2917	}
2918	/*
2919	 * We start scanning from right side, freeing all the blocks
2920	 * after i_size and walking into the tree depth-wise.
2921	 */
2922	depth = ext_depth(inode);
2923	if (path) {
2924		int k = i = depth;
2925		while (--k > 0)
2926			path[k].p_block =
2927				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2928	} else {
2929		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2930			       GFP_NOFS);
2931		if (path == NULL) {
2932			ext4_journal_stop(handle);
2933			return -ENOMEM;
2934		}
2935		path[0].p_maxdepth = path[0].p_depth = depth;
2936		path[0].p_hdr = ext_inode_hdr(inode);
2937		i = 0;
2938
2939		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2940			err = -EFSCORRUPTED;
2941			goto out;
2942		}
2943	}
2944	err = 0;
2945
2946	while (i >= 0 && err == 0) {
2947		if (i == depth) {
2948			/* this is leaf block */
2949			err = ext4_ext_rm_leaf(handle, inode, path,
2950					       &partial_cluster, start,
2951					       end);
2952			/* root level has p_bh == NULL, brelse() eats this */
2953			brelse(path[i].p_bh);
2954			path[i].p_bh = NULL;
2955			i--;
2956			continue;
2957		}
2958
2959		/* this is index block */
2960		if (!path[i].p_hdr) {
2961			ext_debug("initialize header\n");
2962			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2963		}
2964
2965		if (!path[i].p_idx) {
2966			/* this level hasn't been touched yet */
2967			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2968			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2969			ext_debug("init index ptr: hdr 0x%p, num %d\n",
2970				  path[i].p_hdr,
2971				  le16_to_cpu(path[i].p_hdr->eh_entries));
2972		} else {
2973			/* we were already here, see at next index */
2974			path[i].p_idx--;
2975		}
2976
2977		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2978				i, EXT_FIRST_INDEX(path[i].p_hdr),
2979				path[i].p_idx);
2980		if (ext4_ext_more_to_rm(path + i)) {
2981			struct buffer_head *bh;
2982			/* go to the next level */
2983			ext_debug("move to level %d (block %llu)\n",
2984				  i + 1, ext4_idx_pblock(path[i].p_idx));
2985			memset(path + i + 1, 0, sizeof(*path));
2986			bh = read_extent_tree_block(inode,
2987				ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2988				EXT4_EX_NOCACHE);
2989			if (IS_ERR(bh)) {
2990				/* should we reset i_size? */
2991				err = PTR_ERR(bh);
2992				break;
2993			}
2994			/* Yield here to deal with large extent trees.
2995			 * Should be a no-op if we did IO above. */
2996			cond_resched();
2997			if (WARN_ON(i + 1 > depth)) {
2998				err = -EFSCORRUPTED;
2999				break;
3000			}
3001			path[i + 1].p_bh = bh;
3002
3003			/* save actual number of indexes since this
3004			 * number is changed at the next iteration */
3005			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3006			i++;
3007		} else {
3008			/* we finished processing this index, go up */
3009			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3010				/* index is empty, remove it;
3011				 * handle must be already prepared by the
3012				 * truncatei_leaf() */
3013				err = ext4_ext_rm_idx(handle, inode, path, i);
3014			}
3015			/* root level has p_bh == NULL, brelse() eats this */
3016			brelse(path[i].p_bh);
3017			path[i].p_bh = NULL;
3018			i--;
3019			ext_debug("return to level %d\n", i);
3020		}
3021	}
3022
3023	trace_ext4_ext_remove_space_done(inode, start, end, depth,
3024			partial_cluster, path->p_hdr->eh_entries);
3025
3026	/*
3027	 * If we still have something in the partial cluster and we have removed
3028	 * even the first extent, then we should free the blocks in the partial
3029	 * cluster as well.  (This code will only run when there are no leaves
3030	 * to the immediate left of the truncated/punched region.)
3031	 */
3032	if (partial_cluster > 0 && err == 0) {
3033		/* don't zero partial_cluster since it's not used afterwards */
3034		ext4_free_blocks(handle, inode, NULL,
3035				 EXT4_C2B(sbi, partial_cluster),
3036				 sbi->s_cluster_ratio,
3037				 get_default_free_blocks_flags(inode));
3038	}
3039
3040	/* TODO: flexible tree reduction should be here */
3041	if (path->p_hdr->eh_entries == 0) {
3042		/*
3043		 * truncate to zero freed all the tree,
3044		 * so we need to correct eh_depth
3045		 */
3046		err = ext4_ext_get_access(handle, inode, path);
3047		if (err == 0) {
3048			ext_inode_hdr(inode)->eh_depth = 0;
3049			ext_inode_hdr(inode)->eh_max =
3050				cpu_to_le16(ext4_ext_space_root(inode, 0));
3051			err = ext4_ext_dirty(handle, inode, path);
3052		}
3053	}
3054out:
3055	ext4_ext_drop_refs(path);
3056	kfree(path);
3057	path = NULL;
3058	if (err == -EAGAIN)
3059		goto again;
3060	ext4_journal_stop(handle);
3061
3062	return err;
3063}
3064
3065/*
3066 * called at mount time
3067 */
3068void ext4_ext_init(struct super_block *sb)
3069{
3070	/*
3071	 * possible initialization would be here
3072	 */
3073
3074	if (ext4_has_feature_extents(sb)) {
3075#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3076		printk(KERN_INFO "EXT4-fs: file extents enabled"
3077#ifdef AGGRESSIVE_TEST
3078		       ", aggressive tests"
3079#endif
3080#ifdef CHECK_BINSEARCH
3081		       ", check binsearch"
3082#endif
3083#ifdef EXTENTS_STATS
3084		       ", stats"
3085#endif
3086		       "\n");
3087#endif
3088#ifdef EXTENTS_STATS
3089		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3090		EXT4_SB(sb)->s_ext_min = 1 << 30;
3091		EXT4_SB(sb)->s_ext_max = 0;
3092#endif
3093	}
3094}
3095
3096/*
3097 * called at umount time
3098 */
3099void ext4_ext_release(struct super_block *sb)
3100{
3101	if (!ext4_has_feature_extents(sb))
3102		return;
3103
3104#ifdef EXTENTS_STATS
3105	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3106		struct ext4_sb_info *sbi = EXT4_SB(sb);
3107		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3108			sbi->s_ext_blocks, sbi->s_ext_extents,
3109			sbi->s_ext_blocks / sbi->s_ext_extents);
3110		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3111			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3112	}
3113#endif
3114}
3115
3116static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3117{
3118	ext4_lblk_t  ee_block;
3119	ext4_fsblk_t ee_pblock;
3120	unsigned int ee_len;
3121
3122	ee_block  = le32_to_cpu(ex->ee_block);
3123	ee_len    = ext4_ext_get_actual_len(ex);
3124	ee_pblock = ext4_ext_pblock(ex);
3125
3126	if (ee_len == 0)
3127		return 0;
3128
3129	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3130				     EXTENT_STATUS_WRITTEN);
3131}
3132
3133/* FIXME!! we need to try to merge to left or right after zero-out  */
3134static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3135{
3136	ext4_fsblk_t ee_pblock;
3137	unsigned int ee_len;
3138
3139	ee_len    = ext4_ext_get_actual_len(ex);
3140	ee_pblock = ext4_ext_pblock(ex);
3141	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3142				  ee_len);
3143}
3144
3145/*
3146 * ext4_split_extent_at() splits an extent at given block.
3147 *
3148 * @handle: the journal handle
3149 * @inode: the file inode
3150 * @path: the path to the extent
3151 * @split: the logical block where the extent is splitted.
3152 * @split_flags: indicates if the extent could be zeroout if split fails, and
3153 *		 the states(init or unwritten) of new extents.
3154 * @flags: flags used to insert new extent to extent tree.
3155 *
3156 *
3157 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3158 * of which are deterimined by split_flag.
3159 *
3160 * There are two cases:
3161 *  a> the extent are splitted into two extent.
3162 *  b> split is not needed, and just mark the extent.
3163 *
3164 * return 0 on success.
3165 */
3166static int ext4_split_extent_at(handle_t *handle,
3167			     struct inode *inode,
3168			     struct ext4_ext_path **ppath,
3169			     ext4_lblk_t split,
3170			     int split_flag,
3171			     int flags)
3172{
3173	struct ext4_ext_path *path = *ppath;
3174	ext4_fsblk_t newblock;
3175	ext4_lblk_t ee_block;
3176	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3177	struct ext4_extent *ex2 = NULL;
3178	unsigned int ee_len, depth;
3179	int err = 0;
3180
3181	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3182	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3183
3184	ext_debug("ext4_split_extents_at: inode %lu, logical"
3185		"block %llu\n", inode->i_ino, (unsigned long long)split);
3186
3187	ext4_ext_show_leaf(inode, path);
3188
3189	depth = ext_depth(inode);
3190	ex = path[depth].p_ext;
3191	ee_block = le32_to_cpu(ex->ee_block);
3192	ee_len = ext4_ext_get_actual_len(ex);
3193	newblock = split - ee_block + ext4_ext_pblock(ex);
3194
3195	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3196	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3197	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3198			     EXT4_EXT_MARK_UNWRIT1 |
3199			     EXT4_EXT_MARK_UNWRIT2));
3200
3201	err = ext4_ext_get_access(handle, inode, path + depth);
3202	if (err)
3203		goto out;
3204
3205	if (split == ee_block) {
3206		/*
3207		 * case b: block @split is the block that the extent begins with
3208		 * then we just change the state of the extent, and splitting
3209		 * is not needed.
3210		 */
3211		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3212			ext4_ext_mark_unwritten(ex);
3213		else
3214			ext4_ext_mark_initialized(ex);
3215
3216		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3217			ext4_ext_try_to_merge(handle, inode, path, ex);
3218
3219		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3220		goto out;
3221	}
3222
3223	/* case a */
3224	memcpy(&orig_ex, ex, sizeof(orig_ex));
3225	ex->ee_len = cpu_to_le16(split - ee_block);
3226	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3227		ext4_ext_mark_unwritten(ex);
3228
3229	/*
3230	 * path may lead to new leaf, not to original leaf any more
3231	 * after ext4_ext_insert_extent() returns,
3232	 */
3233	err = ext4_ext_dirty(handle, inode, path + depth);
3234	if (err)
3235		goto fix_extent_len;
3236
3237	ex2 = &newex;
3238	ex2->ee_block = cpu_to_le32(split);
3239	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3240	ext4_ext_store_pblock(ex2, newblock);
3241	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3242		ext4_ext_mark_unwritten(ex2);
3243
3244	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3245	if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3246		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3247			if (split_flag & EXT4_EXT_DATA_VALID1) {
3248				err = ext4_ext_zeroout(inode, ex2);
3249				zero_ex.ee_block = ex2->ee_block;
3250				zero_ex.ee_len = cpu_to_le16(
3251						ext4_ext_get_actual_len(ex2));
3252				ext4_ext_store_pblock(&zero_ex,
3253						      ext4_ext_pblock(ex2));
3254			} else {
3255				err = ext4_ext_zeroout(inode, ex);
3256				zero_ex.ee_block = ex->ee_block;
3257				zero_ex.ee_len = cpu_to_le16(
3258						ext4_ext_get_actual_len(ex));
3259				ext4_ext_store_pblock(&zero_ex,
3260						      ext4_ext_pblock(ex));
3261			}
3262		} else {
3263			err = ext4_ext_zeroout(inode, &orig_ex);
3264			zero_ex.ee_block = orig_ex.ee_block;
3265			zero_ex.ee_len = cpu_to_le16(
3266						ext4_ext_get_actual_len(&orig_ex));
3267			ext4_ext_store_pblock(&zero_ex,
3268					      ext4_ext_pblock(&orig_ex));
3269		}
3270
3271		if (err)
3272			goto fix_extent_len;
3273		/* update the extent length and mark as initialized */
3274		ex->ee_len = cpu_to_le16(ee_len);
3275		ext4_ext_try_to_merge(handle, inode, path, ex);
3276		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3277		if (err)
3278			goto fix_extent_len;
3279
3280		/* update extent status tree */
3281		err = ext4_zeroout_es(inode, &zero_ex);
3282
3283		goto out;
3284	} else if (err)
3285		goto fix_extent_len;
3286
3287out:
3288	ext4_ext_show_leaf(inode, path);
3289	return err;
3290
3291fix_extent_len:
3292	ex->ee_len = orig_ex.ee_len;
3293	ext4_ext_dirty(handle, inode, path + path->p_depth);
3294	return err;
3295}
3296
3297/*
3298 * ext4_split_extents() splits an extent and mark extent which is covered
3299 * by @map as split_flags indicates
3300 *
3301 * It may result in splitting the extent into multiple extents (up to three)
3302 * There are three possibilities:
3303 *   a> There is no split required
3304 *   b> Splits in two extents: Split is happening at either end of the extent
3305 *   c> Splits in three extents: Somone is splitting in middle of the extent
3306 *
3307 */
3308static int ext4_split_extent(handle_t *handle,
3309			      struct inode *inode,
3310			      struct ext4_ext_path **ppath,
3311			      struct ext4_map_blocks *map,
3312			      int split_flag,
3313			      int flags)
3314{
3315	struct ext4_ext_path *path = *ppath;
3316	ext4_lblk_t ee_block;
3317	struct ext4_extent *ex;
3318	unsigned int ee_len, depth;
3319	int err = 0;
3320	int unwritten;
3321	int split_flag1, flags1;
3322	int allocated = map->m_len;
3323
3324	depth = ext_depth(inode);
3325	ex = path[depth].p_ext;
3326	ee_block = le32_to_cpu(ex->ee_block);
3327	ee_len = ext4_ext_get_actual_len(ex);
3328	unwritten = ext4_ext_is_unwritten(ex);
3329
3330	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3331		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3332		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3333		if (unwritten)
3334			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3335				       EXT4_EXT_MARK_UNWRIT2;
3336		if (split_flag & EXT4_EXT_DATA_VALID2)
3337			split_flag1 |= EXT4_EXT_DATA_VALID1;
3338		err = ext4_split_extent_at(handle, inode, ppath,
3339				map->m_lblk + map->m_len, split_flag1, flags1);
3340		if (err)
3341			goto out;
3342	} else {
3343		allocated = ee_len - (map->m_lblk - ee_block);
3344	}
3345	/*
3346	 * Update path is required because previous ext4_split_extent_at() may
3347	 * result in split of original leaf or extent zeroout.
3348	 */
3349	path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3350	if (IS_ERR(path))
3351		return PTR_ERR(path);
3352	depth = ext_depth(inode);
3353	ex = path[depth].p_ext;
3354	if (!ex) {
3355		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3356				 (unsigned long) map->m_lblk);
3357		return -EFSCORRUPTED;
3358	}
3359	unwritten = ext4_ext_is_unwritten(ex);
3360	split_flag1 = 0;
3361
3362	if (map->m_lblk >= ee_block) {
3363		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3364		if (unwritten) {
3365			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3366			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3367						     EXT4_EXT_MARK_UNWRIT2);
3368		}
3369		err = ext4_split_extent_at(handle, inode, ppath,
3370				map->m_lblk, split_flag1, flags);
3371		if (err)
3372			goto out;
3373	}
3374
3375	ext4_ext_show_leaf(inode, path);
3376out:
3377	return err ? err : allocated;
3378}
3379
3380/*
3381 * This function is called by ext4_ext_map_blocks() if someone tries to write
3382 * to an unwritten extent. It may result in splitting the unwritten
3383 * extent into multiple extents (up to three - one initialized and two
3384 * unwritten).
3385 * There are three possibilities:
3386 *   a> There is no split required: Entire extent should be initialized
3387 *   b> Splits in two extents: Write is happening at either end of the extent
3388 *   c> Splits in three extents: Somone is writing in middle of the extent
3389 *
3390 * Pre-conditions:
3391 *  - The extent pointed to by 'path' is unwritten.
3392 *  - The extent pointed to by 'path' contains a superset
3393 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3394 *
3395 * Post-conditions on success:
3396 *  - the returned value is the number of blocks beyond map->l_lblk
3397 *    that are allocated and initialized.
3398 *    It is guaranteed to be >= map->m_len.
3399 */
3400static int ext4_ext_convert_to_initialized(handle_t *handle,
3401					   struct inode *inode,
3402					   struct ext4_map_blocks *map,
3403					   struct ext4_ext_path **ppath,
3404					   int flags)
3405{
3406	struct ext4_ext_path *path = *ppath;
3407	struct ext4_sb_info *sbi;
3408	struct ext4_extent_header *eh;
3409	struct ext4_map_blocks split_map;
3410	struct ext4_extent zero_ex;
3411	struct ext4_extent *ex, *abut_ex;
3412	ext4_lblk_t ee_block, eof_block;
3413	unsigned int ee_len, depth, map_len = map->m_len;
3414	int allocated = 0, max_zeroout = 0;
3415	int err = 0;
3416	int split_flag = 0;
3417
3418	ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3419		"block %llu, max_blocks %u\n", inode->i_ino,
3420		(unsigned long long)map->m_lblk, map_len);
3421
3422	sbi = EXT4_SB(inode->i_sb);
3423	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3424		inode->i_sb->s_blocksize_bits;
3425	if (eof_block < map->m_lblk + map_len)
3426		eof_block = map->m_lblk + map_len;
3427
3428	depth = ext_depth(inode);
3429	eh = path[depth].p_hdr;
3430	ex = path[depth].p_ext;
3431	ee_block = le32_to_cpu(ex->ee_block);
3432	ee_len = ext4_ext_get_actual_len(ex);
3433	zero_ex.ee_len = 0;
3434
3435	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3436
3437	/* Pre-conditions */
3438	BUG_ON(!ext4_ext_is_unwritten(ex));
3439	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3440
3441	/*
3442	 * Attempt to transfer newly initialized blocks from the currently
3443	 * unwritten extent to its neighbor. This is much cheaper
3444	 * than an insertion followed by a merge as those involve costly
3445	 * memmove() calls. Transferring to the left is the common case in
3446	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3447	 * followed by append writes.
3448	 *
3449	 * Limitations of the current logic:
3450	 *  - L1: we do not deal with writes covering the whole extent.
3451	 *    This would require removing the extent if the transfer
3452	 *    is possible.
3453	 *  - L2: we only attempt to merge with an extent stored in the
3454	 *    same extent tree node.
3455	 */
3456	if ((map->m_lblk == ee_block) &&
3457		/* See if we can merge left */
3458		(map_len < ee_len) &&		/*L1*/
3459		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3460		ext4_lblk_t prev_lblk;
3461		ext4_fsblk_t prev_pblk, ee_pblk;
3462		unsigned int prev_len;
3463
3464		abut_ex = ex - 1;
3465		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3466		prev_len = ext4_ext_get_actual_len(abut_ex);
3467		prev_pblk = ext4_ext_pblock(abut_ex);
3468		ee_pblk = ext4_ext_pblock(ex);
3469
3470		/*
3471		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3472		 * upon those conditions:
3473		 * - C1: abut_ex is initialized,
3474		 * - C2: abut_ex is logically abutting ex,
3475		 * - C3: abut_ex is physically abutting ex,
3476		 * - C4: abut_ex can receive the additional blocks without
3477		 *   overflowing the (initialized) length limit.
3478		 */
3479		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3480			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3481			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3482			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3483			err = ext4_ext_get_access(handle, inode, path + depth);
3484			if (err)
3485				goto out;
3486
3487			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3488				map, ex, abut_ex);
3489
3490			/* Shift the start of ex by 'map_len' blocks */
3491			ex->ee_block = cpu_to_le32(ee_block + map_len);
3492			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3493			ex->ee_len = cpu_to_le16(ee_len - map_len);
3494			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3495
3496			/* Extend abut_ex by 'map_len' blocks */
3497			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3498
3499			/* Result: number of initialized blocks past m_lblk */
3500			allocated = map_len;
3501		}
3502	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3503		   (map_len < ee_len) &&	/*L1*/
3504		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3505		/* See if we can merge right */
3506		ext4_lblk_t next_lblk;
3507		ext4_fsblk_t next_pblk, ee_pblk;
3508		unsigned int next_len;
3509
3510		abut_ex = ex + 1;
3511		next_lblk = le32_to_cpu(abut_ex->ee_block);
3512		next_len = ext4_ext_get_actual_len(abut_ex);
3513		next_pblk = ext4_ext_pblock(abut_ex);
3514		ee_pblk = ext4_ext_pblock(ex);
3515
3516		/*
3517		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3518		 * upon those conditions:
3519		 * - C1: abut_ex is initialized,
3520		 * - C2: abut_ex is logically abutting ex,
3521		 * - C3: abut_ex is physically abutting ex,
3522		 * - C4: abut_ex can receive the additional blocks without
3523		 *   overflowing the (initialized) length limit.
3524		 */
3525		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3526		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3527		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3528		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3529			err = ext4_ext_get_access(handle, inode, path + depth);
3530			if (err)
3531				goto out;
3532
3533			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3534				map, ex, abut_ex);
3535
3536			/* Shift the start of abut_ex by 'map_len' blocks */
3537			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3538			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3539			ex->ee_len = cpu_to_le16(ee_len - map_len);
3540			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3541
3542			/* Extend abut_ex by 'map_len' blocks */
3543			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3544
3545			/* Result: number of initialized blocks past m_lblk */
3546			allocated = map_len;
3547		}
3548	}
3549	if (allocated) {
3550		/* Mark the block containing both extents as dirty */
3551		ext4_ext_dirty(handle, inode, path + depth);
3552
3553		/* Update path to point to the right extent */
3554		path[depth].p_ext = abut_ex;
3555		goto out;
3556	} else
3557		allocated = ee_len - (map->m_lblk - ee_block);
3558
3559	WARN_ON(map->m_lblk < ee_block);
3560	/*
3561	 * It is safe to convert extent to initialized via explicit
3562	 * zeroout only if extent is fully inside i_size or new_size.
3563	 */
3564	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3565
3566	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3567		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3568			(inode->i_sb->s_blocksize_bits - 10);
3569
3570	if (ext4_encrypted_inode(inode))
3571		max_zeroout = 0;
3572
3573	/* If extent is less than s_max_zeroout_kb, zeroout directly */
3574	if (max_zeroout && (ee_len <= max_zeroout)) {
3575		err = ext4_ext_zeroout(inode, ex);
3576		if (err)
3577			goto out;
3578		zero_ex.ee_block = ex->ee_block;
3579		zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3580		ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3581
3582		err = ext4_ext_get_access(handle, inode, path + depth);
3583		if (err)
3584			goto out;
3585		ext4_ext_mark_initialized(ex);
3586		ext4_ext_try_to_merge(handle, inode, path, ex);
3587		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3588		goto out;
3589	}
3590
3591	/*
3592	 * four cases:
3593	 * 1. split the extent into three extents.
3594	 * 2. split the extent into two extents, zeroout the first half.
3595	 * 3. split the extent into two extents, zeroout the second half.
3596	 * 4. split the extent into two extents with out zeroout.
3597	 */
3598	split_map.m_lblk = map->m_lblk;
3599	split_map.m_len = map->m_len;
3600
3601	if (max_zeroout && (allocated > map->m_len)) {
3602		if (allocated <= max_zeroout) {
3603			/* case 3 */
3604			zero_ex.ee_block =
3605					 cpu_to_le32(map->m_lblk);
3606			zero_ex.ee_len = cpu_to_le16(allocated);
3607			ext4_ext_store_pblock(&zero_ex,
3608				ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3609			err = ext4_ext_zeroout(inode, &zero_ex);
3610			if (err)
3611				goto out;
3612			split_map.m_lblk = map->m_lblk;
3613			split_map.m_len = allocated;
3614		} else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3615			/* case 2 */
3616			if (map->m_lblk != ee_block) {
3617				zero_ex.ee_block = ex->ee_block;
3618				zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3619							ee_block);
3620				ext4_ext_store_pblock(&zero_ex,
3621						      ext4_ext_pblock(ex));
3622				err = ext4_ext_zeroout(inode, &zero_ex);
3623				if (err)
3624					goto out;
3625			}
3626
3627			split_map.m_lblk = ee_block;
3628			split_map.m_len = map->m_lblk - ee_block + map->m_len;
3629			allocated = map->m_len;
3630		}
3631	}
3632
3633	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3634				flags);
3635	if (err > 0)
3636		err = 0;
3637out:
3638	/* If we have gotten a failure, don't zero out status tree */
3639	if (!err)
3640		err = ext4_zeroout_es(inode, &zero_ex);
3641	return err ? err : allocated;
3642}
3643
3644/*
3645 * This function is called by ext4_ext_map_blocks() from
3646 * ext4_get_blocks_dio_write() when DIO to write
3647 * to an unwritten extent.
3648 *
3649 * Writing to an unwritten extent may result in splitting the unwritten
3650 * extent into multiple initialized/unwritten extents (up to three)
3651 * There are three possibilities:
3652 *   a> There is no split required: Entire extent should be unwritten
3653 *   b> Splits in two extents: Write is happening at either end of the extent
3654 *   c> Splits in three extents: Somone is writing in middle of the extent
3655 *
3656 * This works the same way in the case of initialized -> unwritten conversion.
3657 *
3658 * One of more index blocks maybe needed if the extent tree grow after
3659 * the unwritten extent split. To prevent ENOSPC occur at the IO
3660 * complete, we need to split the unwritten extent before DIO submit
3661 * the IO. The unwritten extent called at this time will be split
3662 * into three unwritten extent(at most). After IO complete, the part
3663 * being filled will be convert to initialized by the end_io callback function
3664 * via ext4_convert_unwritten_extents().
3665 *
3666 * Returns the size of unwritten extent to be written on success.
3667 */
3668static int ext4_split_convert_extents(handle_t *handle,
3669					struct inode *inode,
3670					struct ext4_map_blocks *map,
3671					struct ext4_ext_path **ppath,
3672					int flags)
3673{
3674	struct ext4_ext_path *path = *ppath;
3675	ext4_lblk_t eof_block;
3676	ext4_lblk_t ee_block;
3677	struct ext4_extent *ex;
3678	unsigned int ee_len;
3679	int split_flag = 0, depth;
3680
3681	ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3682		  __func__, inode->i_ino,
3683		  (unsigned long long)map->m_lblk, map->m_len);
3684
3685	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3686		inode->i_sb->s_blocksize_bits;
3687	if (eof_block < map->m_lblk + map->m_len)
3688		eof_block = map->m_lblk + map->m_len;
3689	/*
3690	 * It is safe to convert extent to initialized via explicit
3691	 * zeroout only if extent is fully insde i_size or new_size.
3692	 */
3693	depth = ext_depth(inode);
3694	ex = path[depth].p_ext;
3695	ee_block = le32_to_cpu(ex->ee_block);
3696	ee_len = ext4_ext_get_actual_len(ex);
3697
3698	/* Convert to unwritten */
3699	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3700		split_flag |= EXT4_EXT_DATA_VALID1;
3701	/* Convert to initialized */
3702	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3703		split_flag |= ee_block + ee_len <= eof_block ?
3704			      EXT4_EXT_MAY_ZEROOUT : 0;
3705		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3706	}
3707	flags |= EXT4_GET_BLOCKS_PRE_IO;
3708	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3709}
3710
3711static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3712						struct inode *inode,
3713						struct ext4_map_blocks *map,
3714						struct ext4_ext_path **ppath)
3715{
3716	struct ext4_ext_path *path = *ppath;
3717	struct ext4_extent *ex;
3718	ext4_lblk_t ee_block;
3719	unsigned int ee_len;
3720	int depth;
3721	int err = 0;
3722
3723	depth = ext_depth(inode);
3724	ex = path[depth].p_ext;
3725	ee_block = le32_to_cpu(ex->ee_block);
3726	ee_len = ext4_ext_get_actual_len(ex);
3727
3728	ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3729		"block %llu, max_blocks %u\n", inode->i_ino,
3730		  (unsigned long long)ee_block, ee_len);
3731
3732	/* If extent is larger than requested it is a clear sign that we still
3733	 * have some extent state machine issues left. So extent_split is still
3734	 * required.
3735	 * TODO: Once all related issues will be fixed this situation should be
3736	 * illegal.
3737	 */
3738	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3739#ifdef EXT4_DEBUG
3740		ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3741			     " len %u; IO logical block %llu, len %u\n",
3742			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3743			     (unsigned long long)map->m_lblk, map->m_len);
3744#endif
3745		err = ext4_split_convert_extents(handle, inode, map, ppath,
3746						 EXT4_GET_BLOCKS_CONVERT);
3747		if (err < 0)
3748			return err;
3749		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3750		if (IS_ERR(path))
3751			return PTR_ERR(path);
3752		depth = ext_depth(inode);
3753		ex = path[depth].p_ext;
3754	}
3755
3756	err = ext4_ext_get_access(handle, inode, path + depth);
3757	if (err)
3758		goto out;
3759	/* first mark the extent as initialized */
3760	ext4_ext_mark_initialized(ex);
3761
3762	/* note: ext4_ext_correct_indexes() isn't needed here because
3763	 * borders are not changed
3764	 */
3765	ext4_ext_try_to_merge(handle, inode, path, ex);
3766
3767	/* Mark modified extent as dirty */
3768	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3769out:
3770	ext4_ext_show_leaf(inode, path);
3771	return err;
3772}
3773
3774static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3775			sector_t block, int count)
3776{
3777	int i;
3778	for (i = 0; i < count; i++)
3779                unmap_underlying_metadata(bdev, block + i);
3780}
3781
3782/*
3783 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3784 */
3785static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3786			      ext4_lblk_t lblk,
3787			      struct ext4_ext_path *path,
3788			      unsigned int len)
3789{
3790	int i, depth;
3791	struct ext4_extent_header *eh;
3792	struct ext4_extent *last_ex;
3793
3794	if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3795		return 0;
3796
3797	depth = ext_depth(inode);
3798	eh = path[depth].p_hdr;
3799
3800	/*
3801	 * We're going to remove EOFBLOCKS_FL entirely in future so we
3802	 * do not care for this case anymore. Simply remove the flag
3803	 * if there are no extents.
3804	 */
3805	if (unlikely(!eh->eh_entries))
3806		goto out;
3807	last_ex = EXT_LAST_EXTENT(eh);
3808	/*
3809	 * We should clear the EOFBLOCKS_FL flag if we are writing the
3810	 * last block in the last extent in the file.  We test this by
3811	 * first checking to see if the caller to
3812	 * ext4_ext_get_blocks() was interested in the last block (or
3813	 * a block beyond the last block) in the current extent.  If
3814	 * this turns out to be false, we can bail out from this
3815	 * function immediately.
3816	 */
3817	if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3818	    ext4_ext_get_actual_len(last_ex))
3819		return 0;
3820	/*
3821	 * If the caller does appear to be planning to write at or
3822	 * beyond the end of the current extent, we then test to see
3823	 * if the current extent is the last extent in the file, by
3824	 * checking to make sure it was reached via the rightmost node
3825	 * at each level of the tree.
3826	 */
3827	for (i = depth-1; i >= 0; i--)
3828		if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3829			return 0;
3830out:
3831	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3832	return ext4_mark_inode_dirty(handle, inode);
3833}
3834
3835/**
3836 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3837 *
3838 * Return 1 if there is a delalloc block in the range, otherwise 0.
3839 */
3840int ext4_find_delalloc_range(struct inode *inode,
3841			     ext4_lblk_t lblk_start,
3842			     ext4_lblk_t lblk_end)
3843{
3844	struct extent_status es;
3845
3846	ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3847	if (es.es_len == 0)
3848		return 0; /* there is no delay extent in this tree */
3849	else if (es.es_lblk <= lblk_start &&
3850		 lblk_start < es.es_lblk + es.es_len)
3851		return 1;
3852	else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3853		return 1;
3854	else
3855		return 0;
3856}
3857
3858int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3859{
3860	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3861	ext4_lblk_t lblk_start, lblk_end;
3862	lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3863	lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3864
3865	return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3866}
3867
3868/**
3869 * Determines how many complete clusters (out of those specified by the 'map')
3870 * are under delalloc and were reserved quota for.
3871 * This function is called when we are writing out the blocks that were
3872 * originally written with their allocation delayed, but then the space was
3873 * allocated using fallocate() before the delayed allocation could be resolved.
3874 * The cases to look for are:
3875 * ('=' indicated delayed allocated blocks
3876 *  '-' indicates non-delayed allocated blocks)
3877 * (a) partial clusters towards beginning and/or end outside of allocated range
3878 *     are not delalloc'ed.
3879 *	Ex:
3880 *	|----c---=|====c====|====c====|===-c----|
3881 *	         |++++++ allocated ++++++|
3882 *	==> 4 complete clusters in above example
3883 *
3884 * (b) partial cluster (outside of allocated range) towards either end is
3885 *     marked for delayed allocation. In this case, we will exclude that
3886 *     cluster.
3887 *	Ex:
3888 *	|----====c========|========c========|
3889 *	     |++++++ allocated ++++++|
3890 *	==> 1 complete clusters in above example
3891 *
3892 *	Ex:
3893 *	|================c================|
3894 *            |++++++ allocated ++++++|
3895 *	==> 0 complete clusters in above example
3896 *
3897 * The ext4_da_update_reserve_space will be called only if we
3898 * determine here that there were some "entire" clusters that span
3899 * this 'allocated' range.
3900 * In the non-bigalloc case, this function will just end up returning num_blks
3901 * without ever calling ext4_find_delalloc_range.
3902 */
3903static unsigned int
3904get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3905			   unsigned int num_blks)
3906{
3907	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3908	ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3909	ext4_lblk_t lblk_from, lblk_to, c_offset;
3910	unsigned int allocated_clusters = 0;
3911
3912	alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3913	alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3914
3915	/* max possible clusters for this allocation */
3916	allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3917
3918	trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3919
3920	/* Check towards left side */
3921	c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3922	if (c_offset) {
3923		lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3924		lblk_to = lblk_from + c_offset - 1;
3925
3926		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3927			allocated_clusters--;
3928	}
3929
3930	/* Now check towards right. */
3931	c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3932	if (allocated_clusters && c_offset) {
3933		lblk_from = lblk_start + num_blks;
3934		lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3935
3936		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3937			allocated_clusters--;
3938	}
3939
3940	return allocated_clusters;
3941}
3942
3943static int
3944convert_initialized_extent(handle_t *handle, struct inode *inode,
3945			   struct ext4_map_blocks *map,
3946			   struct ext4_ext_path **ppath,
3947			   unsigned int allocated)
3948{
3949	struct ext4_ext_path *path = *ppath;
3950	struct ext4_extent *ex;
3951	ext4_lblk_t ee_block;
3952	unsigned int ee_len;
3953	int depth;
3954	int err = 0;
3955
3956	/*
3957	 * Make sure that the extent is no bigger than we support with
3958	 * unwritten extent
3959	 */
3960	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3961		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3962
3963	depth = ext_depth(inode);
3964	ex = path[depth].p_ext;
3965	ee_block = le32_to_cpu(ex->ee_block);
3966	ee_len = ext4_ext_get_actual_len(ex);
3967
3968	ext_debug("%s: inode %lu, logical"
3969		"block %llu, max_blocks %u\n", __func__, inode->i_ino,
3970		  (unsigned long long)ee_block, ee_len);
3971
3972	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3973		err = ext4_split_convert_extents(handle, inode, map, ppath,
3974				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3975		if (err < 0)
3976			return err;
3977		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3978		if (IS_ERR(path))
3979			return PTR_ERR(path);
3980		depth = ext_depth(inode);
3981		ex = path[depth].p_ext;
3982		if (!ex) {
3983			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3984					 (unsigned long) map->m_lblk);
3985			return -EFSCORRUPTED;
3986		}
3987	}
3988
3989	err = ext4_ext_get_access(handle, inode, path + depth);
3990	if (err)
3991		return err;
3992	/* first mark the extent as unwritten */
3993	ext4_ext_mark_unwritten(ex);
3994
3995	/* note: ext4_ext_correct_indexes() isn't needed here because
3996	 * borders are not changed
3997	 */
3998	ext4_ext_try_to_merge(handle, inode, path, ex);
3999
4000	/* Mark modified extent as dirty */
4001	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4002	if (err)
4003		return err;
4004	ext4_ext_show_leaf(inode, path);
4005
4006	ext4_update_inode_fsync_trans(handle, inode, 1);
4007	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4008	if (err)
4009		return err;
4010	map->m_flags |= EXT4_MAP_UNWRITTEN;
4011	if (allocated > map->m_len)
4012		allocated = map->m_len;
4013	map->m_len = allocated;
4014	return allocated;
4015}
4016
4017static int
4018ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4019			struct ext4_map_blocks *map,
4020			struct ext4_ext_path **ppath, int flags,
4021			unsigned int allocated, ext4_fsblk_t newblock)
4022{
4023	struct ext4_ext_path *path = *ppath;
4024	int ret = 0;
4025	int err = 0;
4026
4027	ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4028		  "block %llu, max_blocks %u, flags %x, allocated %u\n",
4029		  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4030		  flags, allocated);
4031	ext4_ext_show_leaf(inode, path);
4032
4033	/*
4034	 * When writing into unwritten space, we should not fail to
4035	 * allocate metadata blocks for the new extent block if needed.
4036	 */
4037	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4038
4039	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4040						    allocated, newblock);
4041
4042	/* get_block() before submit the IO, split the extent */
4043	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4044		ret = ext4_split_convert_extents(handle, inode, map, ppath,
4045					 flags | EXT4_GET_BLOCKS_CONVERT);
4046		if (ret <= 0)
4047			goto out;
4048		map->m_flags |= EXT4_MAP_UNWRITTEN;
4049		goto out;
4050	}
4051	/* IO end_io complete, convert the filled extent to written */
4052	if (flags & EXT4_GET_BLOCKS_CONVERT) {
4053		if (flags & EXT4_GET_BLOCKS_ZERO) {
4054			if (allocated > map->m_len)
4055				allocated = map->m_len;
4056			err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4057						 allocated);
4058			if (err < 0)
4059				goto out2;
4060		}
4061		ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4062							   ppath);
4063		if (ret >= 0) {
4064			ext4_update_inode_fsync_trans(handle, inode, 1);
4065			err = check_eofblocks_fl(handle, inode, map->m_lblk,
4066						 path, map->m_len);
4067		} else
4068			err = ret;
4069		map->m_flags |= EXT4_MAP_MAPPED;
4070		map->m_pblk = newblock;
4071		if (allocated > map->m_len)
4072			allocated = map->m_len;
4073		map->m_len = allocated;
4074		goto out2;
4075	}
4076	/* buffered IO case */
4077	/*
4078	 * repeat fallocate creation request
4079	 * we already have an unwritten extent
4080	 */
4081	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4082		map->m_flags |= EXT4_MAP_UNWRITTEN;
4083		goto map_out;
4084	}
4085
4086	/* buffered READ or buffered write_begin() lookup */
4087	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4088		/*
4089		 * We have blocks reserved already.  We
4090		 * return allocated blocks so that delalloc
4091		 * won't do block reservation for us.  But
4092		 * the buffer head will be unmapped so that
4093		 * a read from the block returns 0s.
4094		 */
4095		map->m_flags |= EXT4_MAP_UNWRITTEN;
4096		goto out1;
4097	}
4098
4099	/* buffered write, writepage time, convert*/
4100	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4101	if (ret >= 0)
4102		ext4_update_inode_fsync_trans(handle, inode, 1);
4103out:
4104	if (ret <= 0) {
4105		err = ret;
4106		goto out2;
4107	} else
4108		allocated = ret;
4109	map->m_flags |= EXT4_MAP_NEW;
4110	/*
4111	 * if we allocated more blocks than requested
4112	 * we need to make sure we unmap the extra block
4113	 * allocated. The actual needed block will get
4114	 * unmapped later when we find the buffer_head marked
4115	 * new.
4116	 */
4117	if (allocated > map->m_len) {
4118		unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4119					newblock + map->m_len,
4120					allocated - map->m_len);
4121		allocated = map->m_len;
4122	}
4123	map->m_len = allocated;
4124
4125	/*
4126	 * If we have done fallocate with the offset that is already
4127	 * delayed allocated, we would have block reservation
4128	 * and quota reservation done in the delayed write path.
4129	 * But fallocate would have already updated quota and block
4130	 * count for this offset. So cancel these reservation
4131	 */
4132	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4133		unsigned int reserved_clusters;
4134		reserved_clusters = get_reserved_cluster_alloc(inode,
4135				map->m_lblk, map->m_len);
4136		if (reserved_clusters)
4137			ext4_da_update_reserve_space(inode,
4138						     reserved_clusters,
4139						     0);
4140	}
4141
4142map_out:
4143	map->m_flags |= EXT4_MAP_MAPPED;
4144	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4145		err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4146					 map->m_len);
4147		if (err < 0)
4148			goto out2;
4149	}
4150out1:
4151	if (allocated > map->m_len)
4152		allocated = map->m_len;
4153	ext4_ext_show_leaf(inode, path);
4154	map->m_pblk = newblock;
4155	map->m_len = allocated;
4156out2:
4157	return err ? err : allocated;
4158}
4159
4160/*
4161 * get_implied_cluster_alloc - check to see if the requested
4162 * allocation (in the map structure) overlaps with a cluster already
4163 * allocated in an extent.
4164 *	@sb	The filesystem superblock structure
4165 *	@map	The requested lblk->pblk mapping
4166 *	@ex	The extent structure which might contain an implied
4167 *			cluster allocation
4168 *
4169 * This function is called by ext4_ext_map_blocks() after we failed to
4170 * find blocks that were already in the inode's extent tree.  Hence,
4171 * we know that the beginning of the requested region cannot overlap
4172 * the extent from the inode's extent tree.  There are three cases we
4173 * want to catch.  The first is this case:
4174 *
4175 *		 |--- cluster # N--|
4176 *    |--- extent ---|	|---- requested region ---|
4177 *			|==========|
4178 *
4179 * The second case that we need to test for is this one:
4180 *
4181 *   |--------- cluster # N ----------------|
4182 *	   |--- requested region --|   |------- extent ----|
4183 *	   |=======================|
4184 *
4185 * The third case is when the requested region lies between two extents
4186 * within the same cluster:
4187 *          |------------- cluster # N-------------|
4188 * |----- ex -----|                  |---- ex_right ----|
4189 *                  |------ requested region ------|
4190 *                  |================|
4191 *
4192 * In each of the above cases, we need to set the map->m_pblk and
4193 * map->m_len so it corresponds to the return the extent labelled as
4194 * "|====|" from cluster #N, since it is already in use for data in
4195 * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
4196 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4197 * as a new "allocated" block region.  Otherwise, we will return 0 and
4198 * ext4_ext_map_blocks() will then allocate one or more new clusters
4199 * by calling ext4_mb_new_blocks().
4200 */
4201static int get_implied_cluster_alloc(struct super_block *sb,
4202				     struct ext4_map_blocks *map,
4203				     struct ext4_extent *ex,
4204				     struct ext4_ext_path *path)
4205{
4206	struct ext4_sb_info *sbi = EXT4_SB(sb);
4207	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4208	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4209	ext4_lblk_t rr_cluster_start;
4210	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4211	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4212	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4213
4214	/* The extent passed in that we are trying to match */
4215	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4216	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4217
4218	/* The requested region passed into ext4_map_blocks() */
4219	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4220
4221	if ((rr_cluster_start == ex_cluster_end) ||
4222	    (rr_cluster_start == ex_cluster_start)) {
4223		if (rr_cluster_start == ex_cluster_end)
4224			ee_start += ee_len - 1;
4225		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4226		map->m_len = min(map->m_len,
4227				 (unsigned) sbi->s_cluster_ratio - c_offset);
4228		/*
4229		 * Check for and handle this case:
4230		 *
4231		 *   |--------- cluster # N-------------|
4232		 *		       |------- extent ----|
4233		 *	   |--- requested region ---|
4234		 *	   |===========|
4235		 */
4236
4237		if (map->m_lblk < ee_block)
4238			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4239
4240		/*
4241		 * Check for the case where there is already another allocated
4242		 * block to the right of 'ex' but before the end of the cluster.
4243		 *
4244		 *          |------------- cluster # N-------------|
4245		 * |----- ex -----|                  |---- ex_right ----|
4246		 *                  |------ requested region ------|
4247		 *                  |================|
4248		 */
4249		if (map->m_lblk > ee_block) {
4250			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4251			map->m_len = min(map->m_len, next - map->m_lblk);
4252		}
4253
4254		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4255		return 1;
4256	}
4257
4258	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4259	return 0;
4260}
4261
4262
4263/*
4264 * Block allocation/map/preallocation routine for extents based files
4265 *
4266 *
4267 * Need to be called with
4268 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4269 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4270 *
4271 * return > 0, number of of blocks already mapped/allocated
4272 *          if create == 0 and these are pre-allocated blocks
4273 *          	buffer head is unmapped
4274 *          otherwise blocks are mapped
4275 *
4276 * return = 0, if plain look up failed (blocks have not been allocated)
4277 *          buffer head is unmapped
4278 *
4279 * return < 0, error case.
4280 */
4281int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4282			struct ext4_map_blocks *map, int flags)
4283{
4284	struct ext4_ext_path *path = NULL;
4285	struct ext4_extent newex, *ex, *ex2;
4286	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4287	ext4_fsblk_t newblock = 0;
4288	int free_on_err = 0, err = 0, depth, ret;
4289	unsigned int allocated = 0, offset = 0;
4290	unsigned int allocated_clusters = 0;
4291	struct ext4_allocation_request ar;
4292	ext4_lblk_t cluster_offset;
4293	bool map_from_cluster = false;
4294
4295	ext_debug("blocks %u/%u requested for inode %lu\n",
4296		  map->m_lblk, map->m_len, inode->i_ino);
4297	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4298
4299	/* find extent for this block */
4300	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4301	if (IS_ERR(path)) {
4302		err = PTR_ERR(path);
4303		path = NULL;
4304		goto out2;
4305	}
4306
4307	depth = ext_depth(inode);
4308
4309	/*
4310	 * consistent leaf must not be empty;
4311	 * this situation is possible, though, _during_ tree modification;
4312	 * this is why assert can't be put in ext4_find_extent()
4313	 */
4314	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4315		EXT4_ERROR_INODE(inode, "bad extent address "
4316				 "lblock: %lu, depth: %d pblock %lld",
4317				 (unsigned long) map->m_lblk, depth,
4318				 path[depth].p_block);
4319		err = -EFSCORRUPTED;
4320		goto out2;
4321	}
4322
4323	ex = path[depth].p_ext;
4324	if (ex) {
4325		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4326		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4327		unsigned short ee_len;
4328
4329
4330		/*
4331		 * unwritten extents are treated as holes, except that
4332		 * we split out initialized portions during a write.
4333		 */
4334		ee_len = ext4_ext_get_actual_len(ex);
4335
4336		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4337
4338		/* if found extent covers block, simply return it */
4339		if (in_range(map->m_lblk, ee_block, ee_len)) {
4340			newblock = map->m_lblk - ee_block + ee_start;
4341			/* number of remaining blocks in the extent */
4342			allocated = ee_len - (map->m_lblk - ee_block);
4343			ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4344				  ee_block, ee_len, newblock);
4345
4346			/*
4347			 * If the extent is initialized check whether the
4348			 * caller wants to convert it to unwritten.
4349			 */
4350			if ((!ext4_ext_is_unwritten(ex)) &&
4351			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4352				allocated = convert_initialized_extent(
4353						handle, inode, map, &path,
4354						allocated);
4355				goto out2;
4356			} else if (!ext4_ext_is_unwritten(ex))
4357				goto out;
4358
4359			ret = ext4_ext_handle_unwritten_extents(
4360				handle, inode, map, &path, flags,
4361				allocated, newblock);
4362			if (ret < 0)
4363				err = ret;
4364			else
4365				allocated = ret;
4366			goto out2;
4367		}
4368	}
4369
4370	/*
4371	 * requested block isn't allocated yet;
4372	 * we couldn't try to create block if create flag is zero
4373	 */
4374	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4375		ext4_lblk_t hole_start, hole_len;
4376
4377		hole_start = map->m_lblk;
4378		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4379		/*
4380		 * put just found gap into cache to speed up
4381		 * subsequent requests
4382		 */
4383		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4384
4385		/* Update hole_len to reflect hole size after map->m_lblk */
4386		if (hole_start != map->m_lblk)
4387			hole_len -= map->m_lblk - hole_start;
4388		map->m_pblk = 0;
4389		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4390
4391		goto out2;
4392	}
4393
4394	/*
4395	 * Okay, we need to do block allocation.
4396	 */
4397	newex.ee_block = cpu_to_le32(map->m_lblk);
4398	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4399
4400	/*
4401	 * If we are doing bigalloc, check to see if the extent returned
4402	 * by ext4_find_extent() implies a cluster we can use.
4403	 */
4404	if (cluster_offset && ex &&
4405	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4406		ar.len = allocated = map->m_len;
4407		newblock = map->m_pblk;
4408		map_from_cluster = true;
4409		goto got_allocated_blocks;
4410	}
4411
4412	/* find neighbour allocated blocks */
4413	ar.lleft = map->m_lblk;
4414	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4415	if (err)
4416		goto out2;
4417	ar.lright = map->m_lblk;
4418	ex2 = NULL;
4419	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4420	if (err)
4421		goto out2;
4422
4423	/* Check if the extent after searching to the right implies a
4424	 * cluster we can use. */
4425	if ((sbi->s_cluster_ratio > 1) && ex2 &&
4426	    get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4427		ar.len = allocated = map->m_len;
4428		newblock = map->m_pblk;
4429		map_from_cluster = true;
4430		goto got_allocated_blocks;
4431	}
4432
4433	/*
4434	 * See if request is beyond maximum number of blocks we can have in
4435	 * a single extent. For an initialized extent this limit is
4436	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4437	 * EXT_UNWRITTEN_MAX_LEN.
4438	 */
4439	if (map->m_len > EXT_INIT_MAX_LEN &&
4440	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4441		map->m_len = EXT_INIT_MAX_LEN;
4442	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4443		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4444		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4445
4446	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4447	newex.ee_len = cpu_to_le16(map->m_len);
4448	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4449	if (err)
4450		allocated = ext4_ext_get_actual_len(&newex);
4451	else
4452		allocated = map->m_len;
4453
4454	/* allocate new block */
4455	ar.inode = inode;
4456	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4457	ar.logical = map->m_lblk;
4458	/*
4459	 * We calculate the offset from the beginning of the cluster
4460	 * for the logical block number, since when we allocate a
4461	 * physical cluster, the physical block should start at the
4462	 * same offset from the beginning of the cluster.  This is
4463	 * needed so that future calls to get_implied_cluster_alloc()
4464	 * work correctly.
4465	 */
4466	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4467	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4468	ar.goal -= offset;
4469	ar.logical -= offset;
4470	if (S_ISREG(inode->i_mode))
4471		ar.flags = EXT4_MB_HINT_DATA;
4472	else
4473		/* disable in-core preallocation for non-regular files */
4474		ar.flags = 0;
4475	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4476		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4477	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4478		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4479	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4480		ar.flags |= EXT4_MB_USE_RESERVED;
4481	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4482	if (!newblock)
4483		goto out2;
4484	ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4485		  ar.goal, newblock, allocated);
4486	free_on_err = 1;
4487	allocated_clusters = ar.len;
4488	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4489	if (ar.len > allocated)
4490		ar.len = allocated;
4491
4492got_allocated_blocks:
4493	/* try to insert new extent into found leaf and return */
4494	ext4_ext_store_pblock(&newex, newblock + offset);
4495	newex.ee_len = cpu_to_le16(ar.len);
4496	/* Mark unwritten */
4497	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4498		ext4_ext_mark_unwritten(&newex);
4499		map->m_flags |= EXT4_MAP_UNWRITTEN;
4500	}
4501
4502	err = 0;
4503	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4504		err = check_eofblocks_fl(handle, inode, map->m_lblk,
4505					 path, ar.len);
4506	if (!err)
4507		err = ext4_ext_insert_extent(handle, inode, &path,
4508					     &newex, flags);
4509
4510	if (err && free_on_err) {
4511		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4512			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4513		/* free data blocks we just allocated */
4514		/* not a good idea to call discard here directly,
4515		 * but otherwise we'd need to call it every free() */
4516		ext4_discard_preallocations(inode);
4517		ext4_free_blocks(handle, inode, NULL, newblock,
4518				 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4519		goto out2;
4520	}
4521
4522	/* previous routine could use block we allocated */
4523	newblock = ext4_ext_pblock(&newex);
4524	allocated = ext4_ext_get_actual_len(&newex);
4525	if (allocated > map->m_len)
4526		allocated = map->m_len;
4527	map->m_flags |= EXT4_MAP_NEW;
4528
4529	/*
4530	 * Update reserved blocks/metadata blocks after successful
4531	 * block allocation which had been deferred till now.
4532	 */
4533	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4534		unsigned int reserved_clusters;
4535		/*
4536		 * Check how many clusters we had reserved this allocated range
4537		 */
4538		reserved_clusters = get_reserved_cluster_alloc(inode,
4539						map->m_lblk, allocated);
4540		if (!map_from_cluster) {
4541			BUG_ON(allocated_clusters < reserved_clusters);
4542			if (reserved_clusters < allocated_clusters) {
4543				struct ext4_inode_info *ei = EXT4_I(inode);
4544				int reservation = allocated_clusters -
4545						  reserved_clusters;
4546				/*
4547				 * It seems we claimed few clusters outside of
4548				 * the range of this allocation. We should give
4549				 * it back to the reservation pool. This can
4550				 * happen in the following case:
4551				 *
4552				 * * Suppose s_cluster_ratio is 4 (i.e., each
4553				 *   cluster has 4 blocks. Thus, the clusters
4554				 *   are [0-3],[4-7],[8-11]...
4555				 * * First comes delayed allocation write for
4556				 *   logical blocks 10 & 11. Since there were no
4557				 *   previous delayed allocated blocks in the
4558				 *   range [8-11], we would reserve 1 cluster
4559				 *   for this write.
4560				 * * Next comes write for logical blocks 3 to 8.
4561				 *   In this case, we will reserve 2 clusters
4562				 *   (for [0-3] and [4-7]; and not for [8-11] as
4563				 *   that range has a delayed allocated blocks.
4564				 *   Thus total reserved clusters now becomes 3.
4565				 * * Now, during the delayed allocation writeout
4566				 *   time, we will first write blocks [3-8] and
4567				 *   allocate 3 clusters for writing these
4568				 *   blocks. Also, we would claim all these
4569				 *   three clusters above.
4570				 * * Now when we come here to writeout the
4571				 *   blocks [10-11], we would expect to claim
4572				 *   the reservation of 1 cluster we had made
4573				 *   (and we would claim it since there are no
4574				 *   more delayed allocated blocks in the range
4575				 *   [8-11]. But our reserved cluster count had
4576				 *   already gone to 0.
4577				 *
4578				 *   Thus, at the step 4 above when we determine
4579				 *   that there are still some unwritten delayed
4580				 *   allocated blocks outside of our current
4581				 *   block range, we should increment the
4582				 *   reserved clusters count so that when the
4583				 *   remaining blocks finally gets written, we
4584				 *   could claim them.
4585				 */
4586				dquot_reserve_block(inode,
4587						EXT4_C2B(sbi, reservation));
4588				spin_lock(&ei->i_block_reservation_lock);
4589				ei->i_reserved_data_blocks += reservation;
4590				spin_unlock(&ei->i_block_reservation_lock);
4591			}
4592			/*
4593			 * We will claim quota for all newly allocated blocks.
4594			 * We're updating the reserved space *after* the
4595			 * correction above so we do not accidentally free
4596			 * all the metadata reservation because we might
4597			 * actually need it later on.
4598			 */
4599			ext4_da_update_reserve_space(inode, allocated_clusters,
4600							1);
4601		}
4602	}
4603
4604	/*
4605	 * Cache the extent and update transaction to commit on fdatasync only
4606	 * when it is _not_ an unwritten extent.
4607	 */
4608	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4609		ext4_update_inode_fsync_trans(handle, inode, 1);
4610	else
4611		ext4_update_inode_fsync_trans(handle, inode, 0);
4612out:
4613	if (allocated > map->m_len)
4614		allocated = map->m_len;
4615	ext4_ext_show_leaf(inode, path);
4616	map->m_flags |= EXT4_MAP_MAPPED;
4617	map->m_pblk = newblock;
4618	map->m_len = allocated;
4619out2:
4620	ext4_ext_drop_refs(path);
4621	kfree(path);
4622
4623	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4624				       err ? err : allocated);
4625	return err ? err : allocated;
4626}
4627
4628void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4629{
4630	struct super_block *sb = inode->i_sb;
4631	ext4_lblk_t last_block;
4632	int err = 0;
4633
4634	/*
4635	 * TODO: optimization is possible here.
4636	 * Probably we need not scan at all,
4637	 * because page truncation is enough.
4638	 */
4639
4640	/* we have to know where to truncate from in crash case */
4641	EXT4_I(inode)->i_disksize = inode->i_size;
4642	ext4_mark_inode_dirty(handle, inode);
 
 
4643
4644	last_block = (inode->i_size + sb->s_blocksize - 1)
4645			>> EXT4_BLOCK_SIZE_BITS(sb);
4646retry:
4647	err = ext4_es_remove_extent(inode, last_block,
4648				    EXT_MAX_BLOCKS - last_block);
4649	if (err == -ENOMEM) {
4650		cond_resched();
4651		congestion_wait(BLK_RW_ASYNC, HZ/50);
4652		goto retry;
4653	}
4654	if (err) {
4655		ext4_std_error(inode->i_sb, err);
4656		return;
4657	}
4658	err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4659	ext4_std_error(inode->i_sb, err);
4660}
4661
4662static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4663				  ext4_lblk_t len, loff_t new_size,
4664				  int flags, int mode)
4665{
4666	struct inode *inode = file_inode(file);
4667	handle_t *handle;
4668	int ret = 0;
4669	int ret2 = 0;
4670	int retries = 0;
4671	int depth = 0;
4672	struct ext4_map_blocks map;
4673	unsigned int credits;
4674	loff_t epos;
4675
 
4676	map.m_lblk = offset;
4677	map.m_len = len;
4678	/*
4679	 * Don't normalize the request if it can fit in one extent so
4680	 * that it doesn't get unnecessarily split into multiple
4681	 * extents.
4682	 */
4683	if (len <= EXT_UNWRITTEN_MAX_LEN)
4684		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4685
4686	/*
4687	 * credits to insert 1 extent into extent tree
4688	 */
4689	credits = ext4_chunk_trans_blocks(inode, len);
4690	/*
4691	 * We can only call ext_depth() on extent based inodes
4692	 */
4693	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4694		depth = ext_depth(inode);
4695	else
4696		depth = -1;
4697
4698retry:
4699	while (ret >= 0 && len) {
4700		/*
4701		 * Recalculate credits when extent tree depth changes.
4702		 */
4703		if (depth >= 0 && depth != ext_depth(inode)) {
4704			credits = ext4_chunk_trans_blocks(inode, len);
4705			depth = ext_depth(inode);
4706		}
4707
4708		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4709					    credits);
4710		if (IS_ERR(handle)) {
4711			ret = PTR_ERR(handle);
4712			break;
4713		}
4714		ret = ext4_map_blocks(handle, inode, &map, flags);
4715		if (ret <= 0) {
4716			ext4_debug("inode #%lu: block %u: len %u: "
4717				   "ext4_ext_map_blocks returned %d",
4718				   inode->i_ino, map.m_lblk,
4719				   map.m_len, ret);
4720			ext4_mark_inode_dirty(handle, inode);
4721			ret2 = ext4_journal_stop(handle);
4722			break;
4723		}
4724		map.m_lblk += ret;
4725		map.m_len = len = len - ret;
4726		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4727		inode->i_ctime = ext4_current_time(inode);
4728		if (new_size) {
4729			if (epos > new_size)
4730				epos = new_size;
4731			if (ext4_update_inode_size(inode, epos) & 0x1)
4732				inode->i_mtime = inode->i_ctime;
4733		} else {
4734			if (epos > inode->i_size)
4735				ext4_set_inode_flag(inode,
4736						    EXT4_INODE_EOFBLOCKS);
4737		}
4738		ext4_mark_inode_dirty(handle, inode);
4739		ret2 = ext4_journal_stop(handle);
4740		if (ret2)
4741			break;
4742	}
4743	if (ret == -ENOSPC &&
4744			ext4_should_retry_alloc(inode->i_sb, &retries)) {
4745		ret = 0;
4746		goto retry;
4747	}
4748
4749	return ret > 0 ? ret2 : ret;
4750}
4751
4752static long ext4_zero_range(struct file *file, loff_t offset,
4753			    loff_t len, int mode)
4754{
4755	struct inode *inode = file_inode(file);
4756	handle_t *handle = NULL;
4757	unsigned int max_blocks;
4758	loff_t new_size = 0;
4759	int ret = 0;
4760	int flags;
4761	int credits;
4762	int partial_begin, partial_end;
4763	loff_t start, end;
4764	ext4_lblk_t lblk;
4765	unsigned int blkbits = inode->i_blkbits;
4766
4767	trace_ext4_zero_range(inode, offset, len, mode);
4768
4769	if (!S_ISREG(inode->i_mode))
4770		return -EINVAL;
4771
4772	/* Call ext4_force_commit to flush all data in case of data=journal. */
4773	if (ext4_should_journal_data(inode)) {
4774		ret = ext4_force_commit(inode->i_sb);
4775		if (ret)
4776			return ret;
4777	}
4778
4779	/*
4780	 * Round up offset. This is not fallocate, we neet to zero out
4781	 * blocks, so convert interior block aligned part of the range to
4782	 * unwritten and possibly manually zero out unaligned parts of the
4783	 * range.
4784	 */
4785	start = round_up(offset, 1 << blkbits);
4786	end = round_down((offset + len), 1 << blkbits);
4787
4788	if (start < offset || end > offset + len)
4789		return -EINVAL;
4790	partial_begin = offset & ((1 << blkbits) - 1);
4791	partial_end = (offset + len) & ((1 << blkbits) - 1);
4792
4793	lblk = start >> blkbits;
4794	max_blocks = (end >> blkbits);
4795	if (max_blocks < lblk)
4796		max_blocks = 0;
4797	else
4798		max_blocks -= lblk;
4799
4800	inode_lock(inode);
4801
4802	/*
4803	 * Indirect files do not support unwritten extnets
4804	 */
4805	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4806		ret = -EOPNOTSUPP;
4807		goto out_mutex;
4808	}
4809
4810	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811	     offset + len > i_size_read(inode)) {
4812		new_size = offset + len;
4813		ret = inode_newsize_ok(inode, new_size);
4814		if (ret)
4815			goto out_mutex;
4816	}
4817
4818	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4819	if (mode & FALLOC_FL_KEEP_SIZE)
4820		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4821
4822	/* Wait all existing dio workers, newcomers will block on i_mutex */
4823	ext4_inode_block_unlocked_dio(inode);
4824	inode_dio_wait(inode);
4825
4826	/* Preallocate the range including the unaligned edges */
4827	if (partial_begin || partial_end) {
4828		ret = ext4_alloc_file_blocks(file,
4829				round_down(offset, 1 << blkbits) >> blkbits,
4830				(round_up((offset + len), 1 << blkbits) -
4831				 round_down(offset, 1 << blkbits)) >> blkbits,
4832				new_size, flags, mode);
4833		if (ret)
4834			goto out_dio;
4835
4836	}
4837
4838	/* Zero range excluding the unaligned edges */
4839	if (max_blocks > 0) {
4840		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4841			  EXT4_EX_NOCACHE);
4842
4843		/*
4844		 * Prevent page faults from reinstantiating pages we have
4845		 * released from page cache.
4846		 */
4847		down_write(&EXT4_I(inode)->i_mmap_sem);
4848		ret = ext4_update_disksize_before_punch(inode, offset, len);
4849		if (ret) {
4850			up_write(&EXT4_I(inode)->i_mmap_sem);
4851			goto out_dio;
4852		}
4853		/* Now release the pages and zero block aligned part of pages */
4854		truncate_pagecache_range(inode, start, end - 1);
4855		inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4856
4857		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4858					     flags, mode);
4859		up_write(&EXT4_I(inode)->i_mmap_sem);
4860		if (ret)
4861			goto out_dio;
4862	}
4863	if (!partial_begin && !partial_end)
4864		goto out_dio;
4865
4866	/*
4867	 * In worst case we have to writeout two nonadjacent unwritten
4868	 * blocks and update the inode
4869	 */
4870	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4871	if (ext4_should_journal_data(inode))
4872		credits += 2;
4873	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4874	if (IS_ERR(handle)) {
4875		ret = PTR_ERR(handle);
4876		ext4_std_error(inode->i_sb, ret);
4877		goto out_dio;
4878	}
4879
4880	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4881	if (new_size) {
4882		ext4_update_inode_size(inode, new_size);
4883	} else {
4884		/*
4885		* Mark that we allocate beyond EOF so the subsequent truncate
4886		* can proceed even if the new size is the same as i_size.
4887		*/
4888		if ((offset + len) > i_size_read(inode))
4889			ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4890	}
4891	ext4_mark_inode_dirty(handle, inode);
4892
4893	/* Zero out partial block at the edges of the range */
4894	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4895
4896	if (file->f_flags & O_SYNC)
4897		ext4_handle_sync(handle);
4898
4899	ext4_journal_stop(handle);
4900out_dio:
4901	ext4_inode_resume_unlocked_dio(inode);
4902out_mutex:
4903	inode_unlock(inode);
4904	return ret;
4905}
4906
4907/*
4908 * preallocate space for a file. This implements ext4's fallocate file
4909 * operation, which gets called from sys_fallocate system call.
4910 * For block-mapped files, posix_fallocate should fall back to the method
4911 * of writing zeroes to the required new blocks (the same behavior which is
4912 * expected for file systems which do not support fallocate() system call).
4913 */
4914long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4915{
4916	struct inode *inode = file_inode(file);
4917	loff_t new_size = 0;
4918	unsigned int max_blocks;
4919	int ret = 0;
4920	int flags;
4921	ext4_lblk_t lblk;
4922	unsigned int blkbits = inode->i_blkbits;
4923
4924	/*
4925	 * Encrypted inodes can't handle collapse range or insert
4926	 * range since we would need to re-encrypt blocks with a
4927	 * different IV or XTS tweak (which are based on the logical
4928	 * block number).
4929	 *
4930	 * XXX It's not clear why zero range isn't working, but we'll
4931	 * leave it disabled for encrypted inodes for now.  This is a
4932	 * bug we should fix....
4933	 */
4934	if (ext4_encrypted_inode(inode) &&
4935	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4936		     FALLOC_FL_ZERO_RANGE)))
4937		return -EOPNOTSUPP;
4938
4939	/* Return error if mode is not supported */
4940	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4941		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4942		     FALLOC_FL_INSERT_RANGE))
4943		return -EOPNOTSUPP;
4944
4945	if (mode & FALLOC_FL_PUNCH_HOLE)
4946		return ext4_punch_hole(inode, offset, len);
4947
4948	ret = ext4_convert_inline_data(inode);
4949	if (ret)
4950		return ret;
4951
4952	if (mode & FALLOC_FL_COLLAPSE_RANGE)
4953		return ext4_collapse_range(inode, offset, len);
4954
4955	if (mode & FALLOC_FL_INSERT_RANGE)
4956		return ext4_insert_range(inode, offset, len);
4957
4958	if (mode & FALLOC_FL_ZERO_RANGE)
4959		return ext4_zero_range(file, offset, len, mode);
4960
4961	trace_ext4_fallocate_enter(inode, offset, len, mode);
4962	lblk = offset >> blkbits;
4963	/*
4964	 * We can't just convert len to max_blocks because
4965	 * If blocksize = 4096 offset = 3072 and len = 2048
4966	 */
4967	max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4968		- lblk;
4969
 
4970	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4971	if (mode & FALLOC_FL_KEEP_SIZE)
4972		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4973
4974	inode_lock(inode);
4975
4976	/*
4977	 * We only support preallocation for extent-based files only
4978	 */
4979	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4980		ret = -EOPNOTSUPP;
4981		goto out;
4982	}
4983
4984	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4985	     offset + len > i_size_read(inode)) {
4986		new_size = offset + len;
4987		ret = inode_newsize_ok(inode, new_size);
4988		if (ret)
4989			goto out;
4990	}
4991
4992	/* Wait all existing dio workers, newcomers will block on i_mutex */
4993	ext4_inode_block_unlocked_dio(inode);
4994	inode_dio_wait(inode);
4995
4996	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4997				     flags, mode);
4998	ext4_inode_resume_unlocked_dio(inode);
4999	if (ret)
5000		goto out;
5001
5002	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
5003		ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5004						EXT4_I(inode)->i_sync_tid);
5005	}
5006out:
5007	inode_unlock(inode);
5008	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
5009	return ret;
5010}
5011
5012/*
5013 * This function convert a range of blocks to written extents
5014 * The caller of this function will pass the start offset and the size.
5015 * all unwritten extents within this range will be converted to
5016 * written extents.
5017 *
5018 * This function is called from the direct IO end io call back
5019 * function, to convert the fallocated extents after IO is completed.
5020 * Returns 0 on success.
5021 */
5022int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5023				   loff_t offset, ssize_t len)
5024{
5025	unsigned int max_blocks;
5026	int ret = 0;
5027	int ret2 = 0;
5028	struct ext4_map_blocks map;
5029	unsigned int credits, blkbits = inode->i_blkbits;
5030
5031	map.m_lblk = offset >> blkbits;
5032	/*
5033	 * We can't just convert len to max_blocks because
5034	 * If blocksize = 4096 offset = 3072 and len = 2048
5035	 */
5036	max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5037		      map.m_lblk);
5038	/*
5039	 * This is somewhat ugly but the idea is clear: When transaction is
5040	 * reserved, everything goes into it. Otherwise we rather start several
5041	 * smaller transactions for conversion of each extent separately.
5042	 */
5043	if (handle) {
5044		handle = ext4_journal_start_reserved(handle,
5045						     EXT4_HT_EXT_CONVERT);
5046		if (IS_ERR(handle))
5047			return PTR_ERR(handle);
5048		credits = 0;
5049	} else {
5050		/*
5051		 * credits to insert 1 extent into extent tree
5052		 */
5053		credits = ext4_chunk_trans_blocks(inode, max_blocks);
5054	}
5055	while (ret >= 0 && ret < max_blocks) {
5056		map.m_lblk += ret;
5057		map.m_len = (max_blocks -= ret);
5058		if (credits) {
5059			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5060						    credits);
5061			if (IS_ERR(handle)) {
5062				ret = PTR_ERR(handle);
5063				break;
5064			}
5065		}
5066		ret = ext4_map_blocks(handle, inode, &map,
5067				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5068		if (ret <= 0)
5069			ext4_warning(inode->i_sb,
5070				     "inode #%lu: block %u: len %u: "
5071				     "ext4_ext_map_blocks returned %d",
5072				     inode->i_ino, map.m_lblk,
5073				     map.m_len, ret);
5074		ext4_mark_inode_dirty(handle, inode);
5075		if (credits)
5076			ret2 = ext4_journal_stop(handle);
5077		if (ret <= 0 || ret2)
5078			break;
5079	}
5080	if (!credits)
5081		ret2 = ext4_journal_stop(handle);
5082	return ret > 0 ? ret2 : ret;
5083}
5084
5085/*
5086 * If newes is not existing extent (newes->ec_pblk equals zero) find
5087 * delayed extent at start of newes and update newes accordingly and
5088 * return start of the next delayed extent.
5089 *
5090 * If newes is existing extent (newes->ec_pblk is not equal zero)
5091 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5092 * extent found. Leave newes unmodified.
5093 */
5094static int ext4_find_delayed_extent(struct inode *inode,
5095				    struct extent_status *newes)
5096{
5097	struct extent_status es;
5098	ext4_lblk_t block, next_del;
5099
5100	if (newes->es_pblk == 0) {
5101		ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5102				newes->es_lblk + newes->es_len - 1, &es);
5103
5104		/*
5105		 * No extent in extent-tree contains block @newes->es_pblk,
5106		 * then the block may stay in 1)a hole or 2)delayed-extent.
5107		 */
5108		if (es.es_len == 0)
5109			/* A hole found. */
5110			return 0;
5111
5112		if (es.es_lblk > newes->es_lblk) {
5113			/* A hole found. */
5114			newes->es_len = min(es.es_lblk - newes->es_lblk,
5115					    newes->es_len);
5116			return 0;
5117		}
5118
5119		newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5120	}
5121
5122	block = newes->es_lblk + newes->es_len;
5123	ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5124	if (es.es_len == 0)
5125		next_del = EXT_MAX_BLOCKS;
5126	else
5127		next_del = es.es_lblk;
5128
5129	return next_del;
5130}
5131/* fiemap flags we can handle specified here */
5132#define EXT4_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5133
5134static int ext4_xattr_fiemap(struct inode *inode,
5135				struct fiemap_extent_info *fieinfo)
5136{
5137	__u64 physical = 0;
5138	__u64 length;
5139	__u32 flags = FIEMAP_EXTENT_LAST;
5140	int blockbits = inode->i_sb->s_blocksize_bits;
5141	int error = 0;
5142
5143	/* in-inode? */
5144	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5145		struct ext4_iloc iloc;
5146		int offset;	/* offset of xattr in inode */
5147
5148		error = ext4_get_inode_loc(inode, &iloc);
5149		if (error)
5150			return error;
5151		physical = (__u64)iloc.bh->b_blocknr << blockbits;
5152		offset = EXT4_GOOD_OLD_INODE_SIZE +
5153				EXT4_I(inode)->i_extra_isize;
5154		physical += offset;
5155		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5156		flags |= FIEMAP_EXTENT_DATA_INLINE;
5157		brelse(iloc.bh);
5158	} else { /* external block */
5159		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5160		length = inode->i_sb->s_blocksize;
5161	}
5162
5163	if (physical)
5164		error = fiemap_fill_next_extent(fieinfo, 0, physical,
5165						length, flags);
5166	return (error < 0 ? error : 0);
5167}
5168
5169int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5170		__u64 start, __u64 len)
5171{
5172	ext4_lblk_t start_blk;
5173	int error = 0;
5174
5175	if (ext4_has_inline_data(inode)) {
5176		int has_inline = 1;
5177
5178		error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5179						start, len);
5180
5181		if (has_inline)
5182			return error;
5183	}
5184
5185	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5186		error = ext4_ext_precache(inode);
5187		if (error)
5188			return error;
5189	}
5190
5191	/* fallback to generic here if not in extents fmt */
5192	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5193		return generic_block_fiemap(inode, fieinfo, start, len,
5194			ext4_get_block);
5195
5196	if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5197		return -EBADR;
5198
5199	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5200		error = ext4_xattr_fiemap(inode, fieinfo);
5201	} else {
5202		ext4_lblk_t len_blks;
5203		__u64 last_blk;
5204
5205		start_blk = start >> inode->i_sb->s_blocksize_bits;
5206		last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5207		if (last_blk >= EXT_MAX_BLOCKS)
5208			last_blk = EXT_MAX_BLOCKS-1;
5209		len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5210
5211		/*
5212		 * Walk the extent tree gathering extent information
5213		 * and pushing extents back to the user.
5214		 */
5215		error = ext4_fill_fiemap_extents(inode, start_blk,
5216						 len_blks, fieinfo);
5217	}
5218	return error;
5219}
5220
5221/*
5222 * ext4_access_path:
5223 * Function to access the path buffer for marking it dirty.
5224 * It also checks if there are sufficient credits left in the journal handle
5225 * to update path.
5226 */
5227static int
5228ext4_access_path(handle_t *handle, struct inode *inode,
5229		struct ext4_ext_path *path)
5230{
5231	int credits, err;
5232
5233	if (!ext4_handle_valid(handle))
5234		return 0;
5235
5236	/*
5237	 * Check if need to extend journal credits
5238	 * 3 for leaf, sb, and inode plus 2 (bmap and group
5239	 * descriptor) for each block group; assume two block
5240	 * groups
5241	 */
5242	if (handle->h_buffer_credits < 7) {
5243		credits = ext4_writepage_trans_blocks(inode);
5244		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5245		/* EAGAIN is success */
5246		if (err && err != -EAGAIN)
5247			return err;
5248	}
5249
5250	err = ext4_ext_get_access(handle, inode, path);
5251	return err;
5252}
5253
5254/*
5255 * ext4_ext_shift_path_extents:
5256 * Shift the extents of a path structure lying between path[depth].p_ext
5257 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5258 * if it is right shift or left shift operation.
5259 */
5260static int
5261ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5262			    struct inode *inode, handle_t *handle,
5263			    enum SHIFT_DIRECTION SHIFT)
5264{
5265	int depth, err = 0;
5266	struct ext4_extent *ex_start, *ex_last;
5267	bool update = 0;
5268	depth = path->p_depth;
5269
5270	while (depth >= 0) {
5271		if (depth == path->p_depth) {
5272			ex_start = path[depth].p_ext;
5273			if (!ex_start)
5274				return -EFSCORRUPTED;
5275
5276			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5277
5278			err = ext4_access_path(handle, inode, path + depth);
5279			if (err)
5280				goto out;
5281
5282			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5283				update = 1;
5284
5285			while (ex_start <= ex_last) {
5286				if (SHIFT == SHIFT_LEFT) {
5287					le32_add_cpu(&ex_start->ee_block,
5288						-shift);
5289					/* Try to merge to the left. */
5290					if ((ex_start >
5291					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5292					    &&
5293					    ext4_ext_try_to_merge_right(inode,
5294					    path, ex_start - 1))
5295						ex_last--;
5296					else
5297						ex_start++;
5298				} else {
5299					le32_add_cpu(&ex_last->ee_block, shift);
5300					ext4_ext_try_to_merge_right(inode, path,
5301						ex_last);
5302					ex_last--;
5303				}
5304			}
5305			err = ext4_ext_dirty(handle, inode, path + depth);
5306			if (err)
5307				goto out;
5308
5309			if (--depth < 0 || !update)
5310				break;
5311		}
5312
5313		/* Update index too */
5314		err = ext4_access_path(handle, inode, path + depth);
5315		if (err)
5316			goto out;
5317
5318		if (SHIFT == SHIFT_LEFT)
5319			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5320		else
5321			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5322		err = ext4_ext_dirty(handle, inode, path + depth);
5323		if (err)
5324			goto out;
5325
5326		/* we are done if current index is not a starting index */
5327		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5328			break;
5329
5330		depth--;
5331	}
5332
5333out:
5334	return err;
5335}
5336
5337/*
5338 * ext4_ext_shift_extents:
5339 * All the extents which lies in the range from @start to the last allocated
5340 * block for the @inode are shifted either towards left or right (depending
5341 * upon @SHIFT) by @shift blocks.
5342 * On success, 0 is returned, error otherwise.
5343 */
5344static int
5345ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5346		       ext4_lblk_t start, ext4_lblk_t shift,
5347		       enum SHIFT_DIRECTION SHIFT)
5348{
5349	struct ext4_ext_path *path;
5350	int ret = 0, depth;
5351	struct ext4_extent *extent;
5352	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5353
5354	/* Let path point to the last extent */
5355	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
 
5356	if (IS_ERR(path))
5357		return PTR_ERR(path);
5358
5359	depth = path->p_depth;
5360	extent = path[depth].p_ext;
5361	if (!extent)
5362		goto out;
5363
5364	stop = le32_to_cpu(extent->ee_block) +
5365			ext4_ext_get_actual_len(extent);
5366
5367       /*
5368	 * In case of left shift, Don't start shifting extents until we make
5369	 * sure the hole is big enough to accommodate the shift.
5370	*/
5371	if (SHIFT == SHIFT_LEFT) {
5372		path = ext4_find_extent(inode, start - 1, &path, 0);
 
5373		if (IS_ERR(path))
5374			return PTR_ERR(path);
5375		depth = path->p_depth;
5376		extent =  path[depth].p_ext;
5377		if (extent) {
5378			ex_start = le32_to_cpu(extent->ee_block);
5379			ex_end = le32_to_cpu(extent->ee_block) +
5380				ext4_ext_get_actual_len(extent);
5381		} else {
5382			ex_start = 0;
5383			ex_end = 0;
5384		}
5385
5386		if ((start == ex_start && shift > ex_start) ||
5387		    (shift > start - ex_end)) {
5388			ext4_ext_drop_refs(path);
5389			kfree(path);
5390			return -EINVAL;
5391		}
5392	}
5393
5394	/*
5395	 * In case of left shift, iterator points to start and it is increased
5396	 * till we reach stop. In case of right shift, iterator points to stop
5397	 * and it is decreased till we reach start.
5398	 */
5399	if (SHIFT == SHIFT_LEFT)
5400		iterator = &start;
5401	else
5402		iterator = &stop;
5403
5404	/* Its safe to start updating extents */
5405	while (start < stop) {
5406		path = ext4_find_extent(inode, *iterator, &path, 0);
 
 
 
 
 
5407		if (IS_ERR(path))
5408			return PTR_ERR(path);
5409		depth = path->p_depth;
5410		extent = path[depth].p_ext;
5411		if (!extent) {
5412			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5413					 (unsigned long) *iterator);
5414			return -EFSCORRUPTED;
5415		}
5416		if (SHIFT == SHIFT_LEFT && *iterator >
5417		    le32_to_cpu(extent->ee_block)) {
5418			/* Hole, move to the next extent */
5419			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5420				path[depth].p_ext++;
5421			} else {
5422				*iterator = ext4_ext_next_allocated_block(path);
5423				continue;
5424			}
5425		}
5426
5427		if (SHIFT == SHIFT_LEFT) {
5428			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5429			*iterator = le32_to_cpu(extent->ee_block) +
5430					ext4_ext_get_actual_len(extent);
5431		} else {
5432			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5433			*iterator =  le32_to_cpu(extent->ee_block) > 0 ?
5434				le32_to_cpu(extent->ee_block) - 1 : 0;
 
 
 
5435			/* Update path extent in case we need to stop */
5436			while (le32_to_cpu(extent->ee_block) < start)
5437				extent++;
5438			path[depth].p_ext = extent;
5439		}
5440		ret = ext4_ext_shift_path_extents(path, shift, inode,
5441				handle, SHIFT);
5442		if (ret)
5443			break;
5444	}
5445out:
5446	ext4_ext_drop_refs(path);
5447	kfree(path);
5448	return ret;
5449}
5450
5451/*
5452 * ext4_collapse_range:
5453 * This implements the fallocate's collapse range functionality for ext4
5454 * Returns: 0 and non-zero on error.
5455 */
5456int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5457{
5458	struct super_block *sb = inode->i_sb;
5459	ext4_lblk_t punch_start, punch_stop;
5460	handle_t *handle;
5461	unsigned int credits;
5462	loff_t new_size, ioffset;
5463	int ret;
5464
5465	/*
5466	 * We need to test this early because xfstests assumes that a
5467	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5468	 * system does not support collapse range.
5469	 */
5470	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5471		return -EOPNOTSUPP;
5472
5473	/* Collapse range works only on fs block size aligned offsets. */
5474	if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5475	    len & (EXT4_CLUSTER_SIZE(sb) - 1))
5476		return -EINVAL;
5477
5478	if (!S_ISREG(inode->i_mode))
5479		return -EINVAL;
5480
5481	trace_ext4_collapse_range(inode, offset, len);
5482
5483	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5484	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5485
5486	/* Call ext4_force_commit to flush all data in case of data=journal. */
5487	if (ext4_should_journal_data(inode)) {
5488		ret = ext4_force_commit(inode->i_sb);
5489		if (ret)
5490			return ret;
5491	}
5492
5493	inode_lock(inode);
5494	/*
5495	 * There is no need to overlap collapse range with EOF, in which case
5496	 * it is effectively a truncate operation
5497	 */
5498	if (offset + len >= i_size_read(inode)) {
5499		ret = -EINVAL;
5500		goto out_mutex;
5501	}
5502
5503	/* Currently just for extent based files */
5504	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5505		ret = -EOPNOTSUPP;
5506		goto out_mutex;
5507	}
5508
5509	/* Wait for existing dio to complete */
5510	ext4_inode_block_unlocked_dio(inode);
5511	inode_dio_wait(inode);
5512
5513	/*
5514	 * Prevent page faults from reinstantiating pages we have released from
5515	 * page cache.
5516	 */
5517	down_write(&EXT4_I(inode)->i_mmap_sem);
5518	/*
5519	 * Need to round down offset to be aligned with page size boundary
5520	 * for page size > block size.
5521	 */
5522	ioffset = round_down(offset, PAGE_SIZE);
5523	/*
5524	 * Write tail of the last page before removed range since it will get
5525	 * removed from the page cache below.
5526	 */
5527	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5528	if (ret)
5529		goto out_mmap;
5530	/*
5531	 * Write data that will be shifted to preserve them when discarding
5532	 * page cache below. We are also protected from pages becoming dirty
5533	 * by i_mmap_sem.
5534	 */
5535	ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5536					   LLONG_MAX);
5537	if (ret)
5538		goto out_mmap;
5539	truncate_pagecache(inode, ioffset);
5540
5541	credits = ext4_writepage_trans_blocks(inode);
5542	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5543	if (IS_ERR(handle)) {
5544		ret = PTR_ERR(handle);
5545		goto out_mmap;
5546	}
5547
5548	down_write(&EXT4_I(inode)->i_data_sem);
5549	ext4_discard_preallocations(inode);
5550
5551	ret = ext4_es_remove_extent(inode, punch_start,
5552				    EXT_MAX_BLOCKS - punch_start);
5553	if (ret) {
5554		up_write(&EXT4_I(inode)->i_data_sem);
5555		goto out_stop;
5556	}
5557
5558	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5559	if (ret) {
5560		up_write(&EXT4_I(inode)->i_data_sem);
5561		goto out_stop;
5562	}
5563	ext4_discard_preallocations(inode);
5564
5565	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5566				     punch_stop - punch_start, SHIFT_LEFT);
5567	if (ret) {
5568		up_write(&EXT4_I(inode)->i_data_sem);
5569		goto out_stop;
5570	}
5571
5572	new_size = i_size_read(inode) - len;
5573	i_size_write(inode, new_size);
5574	EXT4_I(inode)->i_disksize = new_size;
5575
5576	up_write(&EXT4_I(inode)->i_data_sem);
5577	if (IS_SYNC(inode))
5578		ext4_handle_sync(handle);
5579	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5580	ext4_mark_inode_dirty(handle, inode);
5581
5582out_stop:
5583	ext4_journal_stop(handle);
5584out_mmap:
5585	up_write(&EXT4_I(inode)->i_mmap_sem);
5586	ext4_inode_resume_unlocked_dio(inode);
5587out_mutex:
5588	inode_unlock(inode);
5589	return ret;
5590}
5591
5592/*
5593 * ext4_insert_range:
5594 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5595 * The data blocks starting from @offset to the EOF are shifted by @len
5596 * towards right to create a hole in the @inode. Inode size is increased
5597 * by len bytes.
5598 * Returns 0 on success, error otherwise.
5599 */
5600int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5601{
5602	struct super_block *sb = inode->i_sb;
5603	handle_t *handle;
5604	struct ext4_ext_path *path;
5605	struct ext4_extent *extent;
5606	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5607	unsigned int credits, ee_len;
5608	int ret = 0, depth, split_flag = 0;
5609	loff_t ioffset;
5610
5611	/*
5612	 * We need to test this early because xfstests assumes that an
5613	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5614	 * system does not support insert range.
5615	 */
5616	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5617		return -EOPNOTSUPP;
5618
5619	/* Insert range works only on fs block size aligned offsets. */
5620	if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5621			len & (EXT4_CLUSTER_SIZE(sb) - 1))
5622		return -EINVAL;
5623
5624	if (!S_ISREG(inode->i_mode))
5625		return -EOPNOTSUPP;
5626
5627	trace_ext4_insert_range(inode, offset, len);
5628
5629	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5630	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5631
5632	/* Call ext4_force_commit to flush all data in case of data=journal */
5633	if (ext4_should_journal_data(inode)) {
5634		ret = ext4_force_commit(inode->i_sb);
5635		if (ret)
5636			return ret;
5637	}
5638
5639	inode_lock(inode);
5640	/* Currently just for extent based files */
5641	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5642		ret = -EOPNOTSUPP;
5643		goto out_mutex;
5644	}
5645
5646	/* Check for wrap through zero */
5647	if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5648		ret = -EFBIG;
5649		goto out_mutex;
5650	}
5651
5652	/* Offset should be less than i_size */
5653	if (offset >= i_size_read(inode)) {
5654		ret = -EINVAL;
5655		goto out_mutex;
5656	}
5657
5658	/* Wait for existing dio to complete */
5659	ext4_inode_block_unlocked_dio(inode);
5660	inode_dio_wait(inode);
5661
5662	/*
5663	 * Prevent page faults from reinstantiating pages we have released from
5664	 * page cache.
5665	 */
5666	down_write(&EXT4_I(inode)->i_mmap_sem);
5667	/*
5668	 * Need to round down to align start offset to page size boundary
5669	 * for page size > block size.
5670	 */
5671	ioffset = round_down(offset, PAGE_SIZE);
5672	/* Write out all dirty pages */
5673	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5674			LLONG_MAX);
5675	if (ret)
5676		goto out_mmap;
5677	truncate_pagecache(inode, ioffset);
5678
5679	credits = ext4_writepage_trans_blocks(inode);
5680	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5681	if (IS_ERR(handle)) {
5682		ret = PTR_ERR(handle);
5683		goto out_mmap;
5684	}
5685
5686	/* Expand file to avoid data loss if there is error while shifting */
5687	inode->i_size += len;
5688	EXT4_I(inode)->i_disksize += len;
5689	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5690	ret = ext4_mark_inode_dirty(handle, inode);
5691	if (ret)
5692		goto out_stop;
5693
5694	down_write(&EXT4_I(inode)->i_data_sem);
5695	ext4_discard_preallocations(inode);
5696
5697	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5698	if (IS_ERR(path)) {
5699		up_write(&EXT4_I(inode)->i_data_sem);
5700		goto out_stop;
5701	}
5702
5703	depth = ext_depth(inode);
5704	extent = path[depth].p_ext;
5705	if (extent) {
5706		ee_start_lblk = le32_to_cpu(extent->ee_block);
5707		ee_len = ext4_ext_get_actual_len(extent);
5708
5709		/*
5710		 * If offset_lblk is not the starting block of extent, split
5711		 * the extent @offset_lblk
5712		 */
5713		if ((offset_lblk > ee_start_lblk) &&
5714				(offset_lblk < (ee_start_lblk + ee_len))) {
5715			if (ext4_ext_is_unwritten(extent))
5716				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5717					EXT4_EXT_MARK_UNWRIT2;
5718			ret = ext4_split_extent_at(handle, inode, &path,
5719					offset_lblk, split_flag,
5720					EXT4_EX_NOCACHE |
5721					EXT4_GET_BLOCKS_PRE_IO |
5722					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5723		}
5724
5725		ext4_ext_drop_refs(path);
5726		kfree(path);
5727		if (ret < 0) {
5728			up_write(&EXT4_I(inode)->i_data_sem);
5729			goto out_stop;
5730		}
 
 
 
5731	}
5732
5733	ret = ext4_es_remove_extent(inode, offset_lblk,
5734			EXT_MAX_BLOCKS - offset_lblk);
5735	if (ret) {
5736		up_write(&EXT4_I(inode)->i_data_sem);
5737		goto out_stop;
5738	}
5739
5740	/*
5741	 * if offset_lblk lies in a hole which is at start of file, use
5742	 * ee_start_lblk to shift extents
5743	 */
5744	ret = ext4_ext_shift_extents(inode, handle,
5745		ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5746		len_lblk, SHIFT_RIGHT);
5747
5748	up_write(&EXT4_I(inode)->i_data_sem);
5749	if (IS_SYNC(inode))
5750		ext4_handle_sync(handle);
5751
5752out_stop:
5753	ext4_journal_stop(handle);
5754out_mmap:
5755	up_write(&EXT4_I(inode)->i_mmap_sem);
5756	ext4_inode_resume_unlocked_dio(inode);
5757out_mutex:
5758	inode_unlock(inode);
5759	return ret;
5760}
5761
5762/**
5763 * ext4_swap_extents - Swap extents between two inodes
5764 *
5765 * @inode1:	First inode
5766 * @inode2:	Second inode
5767 * @lblk1:	Start block for first inode
5768 * @lblk2:	Start block for second inode
5769 * @count:	Number of blocks to swap
5770 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5771 * @erp:	Pointer to save error value
5772 *
5773 * This helper routine does exactly what is promise "swap extents". All other
5774 * stuff such as page-cache locking consistency, bh mapping consistency or
5775 * extent's data copying must be performed by caller.
5776 * Locking:
5777 * 		i_mutex is held for both inodes
5778 * 		i_data_sem is locked for write for both inodes
5779 * Assumptions:
5780 *		All pages from requested range are locked for both inodes
5781 */
5782int
5783ext4_swap_extents(handle_t *handle, struct inode *inode1,
5784		     struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5785		  ext4_lblk_t count, int unwritten, int *erp)
5786{
5787	struct ext4_ext_path *path1 = NULL;
5788	struct ext4_ext_path *path2 = NULL;
5789	int replaced_count = 0;
5790
5791	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5792	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5793	BUG_ON(!inode_is_locked(inode1));
5794	BUG_ON(!inode_is_locked(inode2));
5795
5796	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5797	if (unlikely(*erp))
5798		return 0;
5799	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5800	if (unlikely(*erp))
5801		return 0;
5802
5803	while (count) {
5804		struct ext4_extent *ex1, *ex2, tmp_ex;
5805		ext4_lblk_t e1_blk, e2_blk;
5806		int e1_len, e2_len, len;
5807		int split = 0;
5808
5809		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5810		if (IS_ERR(path1)) {
5811			*erp = PTR_ERR(path1);
5812			path1 = NULL;
5813		finish:
5814			count = 0;
5815			goto repeat;
5816		}
5817		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5818		if (IS_ERR(path2)) {
5819			*erp = PTR_ERR(path2);
5820			path2 = NULL;
5821			goto finish;
5822		}
5823		ex1 = path1[path1->p_depth].p_ext;
5824		ex2 = path2[path2->p_depth].p_ext;
5825		/* Do we have somthing to swap ? */
5826		if (unlikely(!ex2 || !ex1))
5827			goto finish;
5828
5829		e1_blk = le32_to_cpu(ex1->ee_block);
5830		e2_blk = le32_to_cpu(ex2->ee_block);
5831		e1_len = ext4_ext_get_actual_len(ex1);
5832		e2_len = ext4_ext_get_actual_len(ex2);
5833
5834		/* Hole handling */
5835		if (!in_range(lblk1, e1_blk, e1_len) ||
5836		    !in_range(lblk2, e2_blk, e2_len)) {
5837			ext4_lblk_t next1, next2;
5838
5839			/* if hole after extent, then go to next extent */
5840			next1 = ext4_ext_next_allocated_block(path1);
5841			next2 = ext4_ext_next_allocated_block(path2);
5842			/* If hole before extent, then shift to that extent */
5843			if (e1_blk > lblk1)
5844				next1 = e1_blk;
5845			if (e2_blk > lblk2)
5846				next2 = e1_blk;
5847			/* Do we have something to swap */
5848			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5849				goto finish;
5850			/* Move to the rightest boundary */
5851			len = next1 - lblk1;
5852			if (len < next2 - lblk2)
5853				len = next2 - lblk2;
5854			if (len > count)
5855				len = count;
5856			lblk1 += len;
5857			lblk2 += len;
5858			count -= len;
5859			goto repeat;
5860		}
5861
5862		/* Prepare left boundary */
5863		if (e1_blk < lblk1) {
5864			split = 1;
5865			*erp = ext4_force_split_extent_at(handle, inode1,
5866						&path1, lblk1, 0);
5867			if (unlikely(*erp))
5868				goto finish;
5869		}
5870		if (e2_blk < lblk2) {
5871			split = 1;
5872			*erp = ext4_force_split_extent_at(handle, inode2,
5873						&path2,  lblk2, 0);
5874			if (unlikely(*erp))
5875				goto finish;
5876		}
5877		/* ext4_split_extent_at() may result in leaf extent split,
5878		 * path must to be revalidated. */
5879		if (split)
5880			goto repeat;
5881
5882		/* Prepare right boundary */
5883		len = count;
5884		if (len > e1_blk + e1_len - lblk1)
5885			len = e1_blk + e1_len - lblk1;
5886		if (len > e2_blk + e2_len - lblk2)
5887			len = e2_blk + e2_len - lblk2;
5888
5889		if (len != e1_len) {
5890			split = 1;
5891			*erp = ext4_force_split_extent_at(handle, inode1,
5892						&path1, lblk1 + len, 0);
5893			if (unlikely(*erp))
5894				goto finish;
5895		}
5896		if (len != e2_len) {
5897			split = 1;
5898			*erp = ext4_force_split_extent_at(handle, inode2,
5899						&path2, lblk2 + len, 0);
5900			if (*erp)
5901				goto finish;
5902		}
5903		/* ext4_split_extent_at() may result in leaf extent split,
5904		 * path must to be revalidated. */
5905		if (split)
5906			goto repeat;
5907
5908		BUG_ON(e2_len != e1_len);
5909		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5910		if (unlikely(*erp))
5911			goto finish;
5912		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5913		if (unlikely(*erp))
5914			goto finish;
5915
5916		/* Both extents are fully inside boundaries. Swap it now */
5917		tmp_ex = *ex1;
5918		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5919		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5920		ex1->ee_len = cpu_to_le16(e2_len);
5921		ex2->ee_len = cpu_to_le16(e1_len);
5922		if (unwritten)
5923			ext4_ext_mark_unwritten(ex2);
5924		if (ext4_ext_is_unwritten(&tmp_ex))
5925			ext4_ext_mark_unwritten(ex1);
5926
5927		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5928		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5929		*erp = ext4_ext_dirty(handle, inode2, path2 +
5930				      path2->p_depth);
5931		if (unlikely(*erp))
5932			goto finish;
5933		*erp = ext4_ext_dirty(handle, inode1, path1 +
5934				      path1->p_depth);
5935		/*
5936		 * Looks scarry ah..? second inode already points to new blocks,
5937		 * and it was successfully dirtied. But luckily error may happen
5938		 * only due to journal error, so full transaction will be
5939		 * aborted anyway.
5940		 */
5941		if (unlikely(*erp))
5942			goto finish;
5943		lblk1 += len;
5944		lblk2 += len;
5945		replaced_count += len;
5946		count -= len;
5947
5948	repeat:
5949		ext4_ext_drop_refs(path1);
5950		kfree(path1);
5951		ext4_ext_drop_refs(path2);
5952		kfree(path2);
5953		path1 = path2 = NULL;
5954	}
5955	return replaced_count;
5956}
v4.10.11
   1/*
   2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   3 * Written by Alex Tomas <alex@clusterfs.com>
   4 *
   5 * Architecture independence:
   6 *   Copyright (c) 2005, Bull S.A.
   7 *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License version 2 as
  11 * published by the Free Software Foundation.
  12 *
  13 * This program is distributed in the hope that it will be useful,
  14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16 * GNU General Public License for more details.
  17 *
  18 * You should have received a copy of the GNU General Public License
  19 * along with this program; if not, write to the Free Software
  20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  21 */
  22
  23/*
  24 * Extents support for EXT4
  25 *
  26 * TODO:
  27 *   - ext4*_error() should be used in some situations
  28 *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  29 *   - smart tree reduction
  30 */
  31
  32#include <linux/fs.h>
  33#include <linux/time.h>
  34#include <linux/jbd2.h>
  35#include <linux/highuid.h>
  36#include <linux/pagemap.h>
  37#include <linux/quotaops.h>
  38#include <linux/string.h>
  39#include <linux/slab.h>
  40#include <linux/uaccess.h>
  41#include <linux/fiemap.h>
  42#include <linux/backing-dev.h>
  43#include "ext4_jbd2.h"
  44#include "ext4_extents.h"
  45#include "xattr.h"
  46
  47#include <trace/events/ext4.h>
  48
  49/*
  50 * used by extent splitting.
  51 */
  52#define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
  53					due to ENOSPC */
  54#define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
  55#define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
  56
  57#define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
  58#define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
  59
  60static __le32 ext4_extent_block_csum(struct inode *inode,
  61				     struct ext4_extent_header *eh)
  62{
  63	struct ext4_inode_info *ei = EXT4_I(inode);
  64	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  65	__u32 csum;
  66
  67	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  68			   EXT4_EXTENT_TAIL_OFFSET(eh));
  69	return cpu_to_le32(csum);
  70}
  71
  72static int ext4_extent_block_csum_verify(struct inode *inode,
  73					 struct ext4_extent_header *eh)
  74{
  75	struct ext4_extent_tail *et;
  76
  77	if (!ext4_has_metadata_csum(inode->i_sb))
  78		return 1;
  79
  80	et = find_ext4_extent_tail(eh);
  81	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  82		return 0;
  83	return 1;
  84}
  85
  86static void ext4_extent_block_csum_set(struct inode *inode,
  87				       struct ext4_extent_header *eh)
  88{
  89	struct ext4_extent_tail *et;
  90
  91	if (!ext4_has_metadata_csum(inode->i_sb))
  92		return;
  93
  94	et = find_ext4_extent_tail(eh);
  95	et->et_checksum = ext4_extent_block_csum(inode, eh);
  96}
  97
  98static int ext4_split_extent(handle_t *handle,
  99				struct inode *inode,
 100				struct ext4_ext_path **ppath,
 101				struct ext4_map_blocks *map,
 102				int split_flag,
 103				int flags);
 104
 105static int ext4_split_extent_at(handle_t *handle,
 106			     struct inode *inode,
 107			     struct ext4_ext_path **ppath,
 108			     ext4_lblk_t split,
 109			     int split_flag,
 110			     int flags);
 111
 112static int ext4_find_delayed_extent(struct inode *inode,
 113				    struct extent_status *newes);
 114
 115static int ext4_ext_truncate_extend_restart(handle_t *handle,
 116					    struct inode *inode,
 117					    int needed)
 118{
 119	int err;
 120
 121	if (!ext4_handle_valid(handle))
 122		return 0;
 123	if (handle->h_buffer_credits >= needed)
 124		return 0;
 125	/*
 126	 * If we need to extend the journal get a few extra blocks
 127	 * while we're at it for efficiency's sake.
 128	 */
 129	needed += 3;
 130	err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
 131	if (err <= 0)
 132		return err;
 133	err = ext4_truncate_restart_trans(handle, inode, needed);
 134	if (err == 0)
 135		err = -EAGAIN;
 136
 137	return err;
 138}
 139
 140/*
 141 * could return:
 142 *  - EROFS
 143 *  - ENOMEM
 144 */
 145static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 146				struct ext4_ext_path *path)
 147{
 148	if (path->p_bh) {
 149		/* path points to block */
 150		BUFFER_TRACE(path->p_bh, "get_write_access");
 151		return ext4_journal_get_write_access(handle, path->p_bh);
 152	}
 153	/* path points to leaf/index in inode body */
 154	/* we use in-core data, no need to protect them */
 155	return 0;
 156}
 157
 158/*
 159 * could return:
 160 *  - EROFS
 161 *  - ENOMEM
 162 *  - EIO
 163 */
 164int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
 165		     struct inode *inode, struct ext4_ext_path *path)
 166{
 167	int err;
 168
 169	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 170	if (path->p_bh) {
 171		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 172		/* path points to block */
 173		err = __ext4_handle_dirty_metadata(where, line, handle,
 174						   inode, path->p_bh);
 175	} else {
 176		/* path points to leaf/index in inode body */
 177		err = ext4_mark_inode_dirty(handle, inode);
 178	}
 179	return err;
 180}
 181
 182static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 183			      struct ext4_ext_path *path,
 184			      ext4_lblk_t block)
 185{
 186	if (path) {
 187		int depth = path->p_depth;
 188		struct ext4_extent *ex;
 189
 190		/*
 191		 * Try to predict block placement assuming that we are
 192		 * filling in a file which will eventually be
 193		 * non-sparse --- i.e., in the case of libbfd writing
 194		 * an ELF object sections out-of-order but in a way
 195		 * the eventually results in a contiguous object or
 196		 * executable file, or some database extending a table
 197		 * space file.  However, this is actually somewhat
 198		 * non-ideal if we are writing a sparse file such as
 199		 * qemu or KVM writing a raw image file that is going
 200		 * to stay fairly sparse, since it will end up
 201		 * fragmenting the file system's free space.  Maybe we
 202		 * should have some hueristics or some way to allow
 203		 * userspace to pass a hint to file system,
 204		 * especially if the latter case turns out to be
 205		 * common.
 206		 */
 207		ex = path[depth].p_ext;
 208		if (ex) {
 209			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 210			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 211
 212			if (block > ext_block)
 213				return ext_pblk + (block - ext_block);
 214			else
 215				return ext_pblk - (ext_block - block);
 216		}
 217
 218		/* it looks like index is empty;
 219		 * try to find starting block from index itself */
 220		if (path[depth].p_bh)
 221			return path[depth].p_bh->b_blocknr;
 222	}
 223
 224	/* OK. use inode's group */
 225	return ext4_inode_to_goal_block(inode);
 226}
 227
 228/*
 229 * Allocation for a meta data block
 230 */
 231static ext4_fsblk_t
 232ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 233			struct ext4_ext_path *path,
 234			struct ext4_extent *ex, int *err, unsigned int flags)
 235{
 236	ext4_fsblk_t goal, newblock;
 237
 238	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 239	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 240					NULL, err);
 241	return newblock;
 242}
 243
 244static inline int ext4_ext_space_block(struct inode *inode, int check)
 245{
 246	int size;
 247
 248	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 249			/ sizeof(struct ext4_extent);
 250#ifdef AGGRESSIVE_TEST
 251	if (!check && size > 6)
 252		size = 6;
 253#endif
 254	return size;
 255}
 256
 257static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 258{
 259	int size;
 260
 261	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 262			/ sizeof(struct ext4_extent_idx);
 263#ifdef AGGRESSIVE_TEST
 264	if (!check && size > 5)
 265		size = 5;
 266#endif
 267	return size;
 268}
 269
 270static inline int ext4_ext_space_root(struct inode *inode, int check)
 271{
 272	int size;
 273
 274	size = sizeof(EXT4_I(inode)->i_data);
 275	size -= sizeof(struct ext4_extent_header);
 276	size /= sizeof(struct ext4_extent);
 277#ifdef AGGRESSIVE_TEST
 278	if (!check && size > 3)
 279		size = 3;
 280#endif
 281	return size;
 282}
 283
 284static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 285{
 286	int size;
 287
 288	size = sizeof(EXT4_I(inode)->i_data);
 289	size -= sizeof(struct ext4_extent_header);
 290	size /= sizeof(struct ext4_extent_idx);
 291#ifdef AGGRESSIVE_TEST
 292	if (!check && size > 4)
 293		size = 4;
 294#endif
 295	return size;
 296}
 297
 298static inline int
 299ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 300			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 301			   int nofail)
 302{
 303	struct ext4_ext_path *path = *ppath;
 304	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 305
 306	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 307			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 308			EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
 309			(nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
 310}
 311
 312/*
 313 * Calculate the number of metadata blocks needed
 314 * to allocate @blocks
 315 * Worse case is one block per extent
 316 */
 317int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 318{
 319	struct ext4_inode_info *ei = EXT4_I(inode);
 320	int idxs;
 321
 322	idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 323		/ sizeof(struct ext4_extent_idx));
 324
 325	/*
 326	 * If the new delayed allocation block is contiguous with the
 327	 * previous da block, it can share index blocks with the
 328	 * previous block, so we only need to allocate a new index
 329	 * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 330	 * an additional index block, and at ldxs**3 blocks, yet
 331	 * another index blocks.
 332	 */
 333	if (ei->i_da_metadata_calc_len &&
 334	    ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 335		int num = 0;
 336
 337		if ((ei->i_da_metadata_calc_len % idxs) == 0)
 338			num++;
 339		if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 340			num++;
 341		if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 342			num++;
 343			ei->i_da_metadata_calc_len = 0;
 344		} else
 345			ei->i_da_metadata_calc_len++;
 346		ei->i_da_metadata_calc_last_lblock++;
 347		return num;
 348	}
 349
 350	/*
 351	 * In the worst case we need a new set of index blocks at
 352	 * every level of the inode's extent tree.
 353	 */
 354	ei->i_da_metadata_calc_len = 1;
 355	ei->i_da_metadata_calc_last_lblock = lblock;
 356	return ext_depth(inode) + 1;
 357}
 358
 359static int
 360ext4_ext_max_entries(struct inode *inode, int depth)
 361{
 362	int max;
 363
 364	if (depth == ext_depth(inode)) {
 365		if (depth == 0)
 366			max = ext4_ext_space_root(inode, 1);
 367		else
 368			max = ext4_ext_space_root_idx(inode, 1);
 369	} else {
 370		if (depth == 0)
 371			max = ext4_ext_space_block(inode, 1);
 372		else
 373			max = ext4_ext_space_block_idx(inode, 1);
 374	}
 375
 376	return max;
 377}
 378
 379static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 380{
 381	ext4_fsblk_t block = ext4_ext_pblock(ext);
 382	int len = ext4_ext_get_actual_len(ext);
 383	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 
 384
 385	/*
 386	 * We allow neither:
 387	 *  - zero length
 388	 *  - overflow/wrap-around
 389	 */
 390	if (lblock + len <= lblock)
 391		return 0;
 392	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 393}
 394
 395static int ext4_valid_extent_idx(struct inode *inode,
 396				struct ext4_extent_idx *ext_idx)
 397{
 398	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 399
 400	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 401}
 402
 403static int ext4_valid_extent_entries(struct inode *inode,
 404				struct ext4_extent_header *eh,
 405				int depth)
 406{
 407	unsigned short entries;
 408	if (eh->eh_entries == 0)
 409		return 1;
 410
 411	entries = le16_to_cpu(eh->eh_entries);
 412
 413	if (depth == 0) {
 414		/* leaf entries */
 415		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 416		struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 417		ext4_fsblk_t pblock = 0;
 418		ext4_lblk_t lblock = 0;
 419		ext4_lblk_t prev = 0;
 420		int len = 0;
 421		while (entries) {
 422			if (!ext4_valid_extent(inode, ext))
 423				return 0;
 424
 425			/* Check for overlapping extents */
 426			lblock = le32_to_cpu(ext->ee_block);
 427			len = ext4_ext_get_actual_len(ext);
 428			if ((lblock <= prev) && prev) {
 429				pblock = ext4_ext_pblock(ext);
 430				es->s_last_error_block = cpu_to_le64(pblock);
 431				return 0;
 432			}
 433			ext++;
 434			entries--;
 435			prev = lblock + len - 1;
 436		}
 437	} else {
 438		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 439		while (entries) {
 440			if (!ext4_valid_extent_idx(inode, ext_idx))
 441				return 0;
 442			ext_idx++;
 443			entries--;
 444		}
 445	}
 446	return 1;
 447}
 448
 449static int __ext4_ext_check(const char *function, unsigned int line,
 450			    struct inode *inode, struct ext4_extent_header *eh,
 451			    int depth, ext4_fsblk_t pblk)
 452{
 453	const char *error_msg;
 454	int max = 0, err = -EFSCORRUPTED;
 455
 456	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 457		error_msg = "invalid magic";
 458		goto corrupted;
 459	}
 460	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 461		error_msg = "unexpected eh_depth";
 462		goto corrupted;
 463	}
 464	if (unlikely(eh->eh_max == 0)) {
 465		error_msg = "invalid eh_max";
 466		goto corrupted;
 467	}
 468	max = ext4_ext_max_entries(inode, depth);
 469	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 470		error_msg = "too large eh_max";
 471		goto corrupted;
 472	}
 473	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 474		error_msg = "invalid eh_entries";
 475		goto corrupted;
 476	}
 477	if (!ext4_valid_extent_entries(inode, eh, depth)) {
 478		error_msg = "invalid extent entries";
 479		goto corrupted;
 480	}
 481	if (unlikely(depth > 32)) {
 482		error_msg = "too large eh_depth";
 483		goto corrupted;
 484	}
 485	/* Verify checksum on non-root extent tree nodes */
 486	if (ext_depth(inode) != depth &&
 487	    !ext4_extent_block_csum_verify(inode, eh)) {
 488		error_msg = "extent tree corrupted";
 489		err = -EFSBADCRC;
 490		goto corrupted;
 491	}
 492	return 0;
 493
 494corrupted:
 495	ext4_error_inode(inode, function, line, 0,
 496			 "pblk %llu bad header/extent: %s - magic %x, "
 497			 "entries %u, max %u(%u), depth %u(%u)",
 498			 (unsigned long long) pblk, error_msg,
 499			 le16_to_cpu(eh->eh_magic),
 500			 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 501			 max, le16_to_cpu(eh->eh_depth), depth);
 502	return err;
 503}
 504
 505#define ext4_ext_check(inode, eh, depth, pblk)			\
 506	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
 507
 508int ext4_ext_check_inode(struct inode *inode)
 509{
 510	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 511}
 512
 513static struct buffer_head *
 514__read_extent_tree_block(const char *function, unsigned int line,
 515			 struct inode *inode, ext4_fsblk_t pblk, int depth,
 516			 int flags)
 517{
 518	struct buffer_head		*bh;
 519	int				err;
 520
 521	bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
 522	if (unlikely(!bh))
 523		return ERR_PTR(-ENOMEM);
 524
 525	if (!bh_uptodate_or_lock(bh)) {
 526		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 527		err = bh_submit_read(bh);
 528		if (err < 0)
 529			goto errout;
 530	}
 531	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 532		return bh;
 533	err = __ext4_ext_check(function, line, inode,
 534			       ext_block_hdr(bh), depth, pblk);
 535	if (err)
 536		goto errout;
 537	set_buffer_verified(bh);
 538	/*
 539	 * If this is a leaf block, cache all of its entries
 540	 */
 541	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 542		struct ext4_extent_header *eh = ext_block_hdr(bh);
 543		struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 544		ext4_lblk_t prev = 0;
 545		int i;
 546
 547		for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 548			unsigned int status = EXTENT_STATUS_WRITTEN;
 549			ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 550			int len = ext4_ext_get_actual_len(ex);
 551
 552			if (prev && (prev != lblk))
 553				ext4_es_cache_extent(inode, prev,
 554						     lblk - prev, ~0,
 555						     EXTENT_STATUS_HOLE);
 556
 557			if (ext4_ext_is_unwritten(ex))
 558				status = EXTENT_STATUS_UNWRITTEN;
 559			ext4_es_cache_extent(inode, lblk, len,
 560					     ext4_ext_pblock(ex), status);
 561			prev = lblk + len;
 562		}
 563	}
 564	return bh;
 565errout:
 566	put_bh(bh);
 567	return ERR_PTR(err);
 568
 569}
 570
 571#define read_extent_tree_block(inode, pblk, depth, flags)		\
 572	__read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
 573				 (depth), (flags))
 574
 575/*
 576 * This function is called to cache a file's extent information in the
 577 * extent status tree
 578 */
 579int ext4_ext_precache(struct inode *inode)
 580{
 581	struct ext4_inode_info *ei = EXT4_I(inode);
 582	struct ext4_ext_path *path = NULL;
 583	struct buffer_head *bh;
 584	int i = 0, depth, ret = 0;
 585
 586	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 587		return 0;	/* not an extent-mapped inode */
 588
 589	down_read(&ei->i_data_sem);
 590	depth = ext_depth(inode);
 591
 592	path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
 593		       GFP_NOFS);
 594	if (path == NULL) {
 595		up_read(&ei->i_data_sem);
 596		return -ENOMEM;
 597	}
 598
 599	/* Don't cache anything if there are no external extent blocks */
 600	if (depth == 0)
 601		goto out;
 602	path[0].p_hdr = ext_inode_hdr(inode);
 603	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 604	if (ret)
 605		goto out;
 606	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 607	while (i >= 0) {
 608		/*
 609		 * If this is a leaf block or we've reached the end of
 610		 * the index block, go up
 611		 */
 612		if ((i == depth) ||
 613		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 614			brelse(path[i].p_bh);
 615			path[i].p_bh = NULL;
 616			i--;
 617			continue;
 618		}
 619		bh = read_extent_tree_block(inode,
 620					    ext4_idx_pblock(path[i].p_idx++),
 621					    depth - i - 1,
 622					    EXT4_EX_FORCE_CACHE);
 623		if (IS_ERR(bh)) {
 624			ret = PTR_ERR(bh);
 625			break;
 626		}
 627		i++;
 628		path[i].p_bh = bh;
 629		path[i].p_hdr = ext_block_hdr(bh);
 630		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 631	}
 632	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 633out:
 634	up_read(&ei->i_data_sem);
 635	ext4_ext_drop_refs(path);
 636	kfree(path);
 637	return ret;
 638}
 639
 640#ifdef EXT_DEBUG
 641static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 642{
 643	int k, l = path->p_depth;
 644
 645	ext_debug("path:");
 646	for (k = 0; k <= l; k++, path++) {
 647		if (path->p_idx) {
 648		  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 649			    ext4_idx_pblock(path->p_idx));
 650		} else if (path->p_ext) {
 651			ext_debug("  %d:[%d]%d:%llu ",
 652				  le32_to_cpu(path->p_ext->ee_block),
 653				  ext4_ext_is_unwritten(path->p_ext),
 654				  ext4_ext_get_actual_len(path->p_ext),
 655				  ext4_ext_pblock(path->p_ext));
 656		} else
 657			ext_debug("  []");
 658	}
 659	ext_debug("\n");
 660}
 661
 662static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 663{
 664	int depth = ext_depth(inode);
 665	struct ext4_extent_header *eh;
 666	struct ext4_extent *ex;
 667	int i;
 668
 669	if (!path)
 670		return;
 671
 672	eh = path[depth].p_hdr;
 673	ex = EXT_FIRST_EXTENT(eh);
 674
 675	ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 676
 677	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 678		ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 679			  ext4_ext_is_unwritten(ex),
 680			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 681	}
 682	ext_debug("\n");
 683}
 684
 685static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 686			ext4_fsblk_t newblock, int level)
 687{
 688	int depth = ext_depth(inode);
 689	struct ext4_extent *ex;
 690
 691	if (depth != level) {
 692		struct ext4_extent_idx *idx;
 693		idx = path[level].p_idx;
 694		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 695			ext_debug("%d: move %d:%llu in new index %llu\n", level,
 696					le32_to_cpu(idx->ei_block),
 697					ext4_idx_pblock(idx),
 698					newblock);
 699			idx++;
 700		}
 701
 702		return;
 703	}
 704
 705	ex = path[depth].p_ext;
 706	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 707		ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 708				le32_to_cpu(ex->ee_block),
 709				ext4_ext_pblock(ex),
 710				ext4_ext_is_unwritten(ex),
 711				ext4_ext_get_actual_len(ex),
 712				newblock);
 713		ex++;
 714	}
 715}
 716
 717#else
 718#define ext4_ext_show_path(inode, path)
 719#define ext4_ext_show_leaf(inode, path)
 720#define ext4_ext_show_move(inode, path, newblock, level)
 721#endif
 722
 723void ext4_ext_drop_refs(struct ext4_ext_path *path)
 724{
 725	int depth, i;
 726
 727	if (!path)
 728		return;
 729	depth = path->p_depth;
 730	for (i = 0; i <= depth; i++, path++)
 731		if (path->p_bh) {
 732			brelse(path->p_bh);
 733			path->p_bh = NULL;
 734		}
 735}
 736
 737/*
 738 * ext4_ext_binsearch_idx:
 739 * binary search for the closest index of the given block
 740 * the header must be checked before calling this
 741 */
 742static void
 743ext4_ext_binsearch_idx(struct inode *inode,
 744			struct ext4_ext_path *path, ext4_lblk_t block)
 745{
 746	struct ext4_extent_header *eh = path->p_hdr;
 747	struct ext4_extent_idx *r, *l, *m;
 748
 749
 750	ext_debug("binsearch for %u(idx):  ", block);
 751
 752	l = EXT_FIRST_INDEX(eh) + 1;
 753	r = EXT_LAST_INDEX(eh);
 754	while (l <= r) {
 755		m = l + (r - l) / 2;
 756		if (block < le32_to_cpu(m->ei_block))
 757			r = m - 1;
 758		else
 759			l = m + 1;
 760		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 761				m, le32_to_cpu(m->ei_block),
 762				r, le32_to_cpu(r->ei_block));
 763	}
 764
 765	path->p_idx = l - 1;
 766	ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 767		  ext4_idx_pblock(path->p_idx));
 768
 769#ifdef CHECK_BINSEARCH
 770	{
 771		struct ext4_extent_idx *chix, *ix;
 772		int k;
 773
 774		chix = ix = EXT_FIRST_INDEX(eh);
 775		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 776		  if (k != 0 &&
 777		      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 778				printk(KERN_DEBUG "k=%d, ix=0x%p, "
 779				       "first=0x%p\n", k,
 780				       ix, EXT_FIRST_INDEX(eh));
 781				printk(KERN_DEBUG "%u <= %u\n",
 782				       le32_to_cpu(ix->ei_block),
 783				       le32_to_cpu(ix[-1].ei_block));
 784			}
 785			BUG_ON(k && le32_to_cpu(ix->ei_block)
 786					   <= le32_to_cpu(ix[-1].ei_block));
 787			if (block < le32_to_cpu(ix->ei_block))
 788				break;
 789			chix = ix;
 790		}
 791		BUG_ON(chix != path->p_idx);
 792	}
 793#endif
 794
 795}
 796
 797/*
 798 * ext4_ext_binsearch:
 799 * binary search for closest extent of the given block
 800 * the header must be checked before calling this
 801 */
 802static void
 803ext4_ext_binsearch(struct inode *inode,
 804		struct ext4_ext_path *path, ext4_lblk_t block)
 805{
 806	struct ext4_extent_header *eh = path->p_hdr;
 807	struct ext4_extent *r, *l, *m;
 808
 809	if (eh->eh_entries == 0) {
 810		/*
 811		 * this leaf is empty:
 812		 * we get such a leaf in split/add case
 813		 */
 814		return;
 815	}
 816
 817	ext_debug("binsearch for %u:  ", block);
 818
 819	l = EXT_FIRST_EXTENT(eh) + 1;
 820	r = EXT_LAST_EXTENT(eh);
 821
 822	while (l <= r) {
 823		m = l + (r - l) / 2;
 824		if (block < le32_to_cpu(m->ee_block))
 825			r = m - 1;
 826		else
 827			l = m + 1;
 828		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 829				m, le32_to_cpu(m->ee_block),
 830				r, le32_to_cpu(r->ee_block));
 831	}
 832
 833	path->p_ext = l - 1;
 834	ext_debug("  -> %d:%llu:[%d]%d ",
 835			le32_to_cpu(path->p_ext->ee_block),
 836			ext4_ext_pblock(path->p_ext),
 837			ext4_ext_is_unwritten(path->p_ext),
 838			ext4_ext_get_actual_len(path->p_ext));
 839
 840#ifdef CHECK_BINSEARCH
 841	{
 842		struct ext4_extent *chex, *ex;
 843		int k;
 844
 845		chex = ex = EXT_FIRST_EXTENT(eh);
 846		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 847			BUG_ON(k && le32_to_cpu(ex->ee_block)
 848					  <= le32_to_cpu(ex[-1].ee_block));
 849			if (block < le32_to_cpu(ex->ee_block))
 850				break;
 851			chex = ex;
 852		}
 853		BUG_ON(chex != path->p_ext);
 854	}
 855#endif
 856
 857}
 858
 859int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 860{
 861	struct ext4_extent_header *eh;
 862
 863	eh = ext_inode_hdr(inode);
 864	eh->eh_depth = 0;
 865	eh->eh_entries = 0;
 866	eh->eh_magic = EXT4_EXT_MAGIC;
 867	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 868	ext4_mark_inode_dirty(handle, inode);
 869	return 0;
 870}
 871
 872struct ext4_ext_path *
 873ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 874		 struct ext4_ext_path **orig_path, int flags)
 875{
 876	struct ext4_extent_header *eh;
 877	struct buffer_head *bh;
 878	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 879	short int depth, i, ppos = 0;
 880	int ret;
 881
 882	eh = ext_inode_hdr(inode);
 883	depth = ext_depth(inode);
 884
 885	if (path) {
 886		ext4_ext_drop_refs(path);
 887		if (depth > path[0].p_maxdepth) {
 888			kfree(path);
 889			*orig_path = path = NULL;
 890		}
 891	}
 892	if (!path) {
 893		/* account possible depth increase */
 894		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 895				GFP_NOFS);
 896		if (unlikely(!path))
 897			return ERR_PTR(-ENOMEM);
 898		path[0].p_maxdepth = depth + 1;
 899	}
 900	path[0].p_hdr = eh;
 901	path[0].p_bh = NULL;
 902
 903	i = depth;
 904	/* walk through the tree */
 905	while (i) {
 906		ext_debug("depth %d: num %d, max %d\n",
 907			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 908
 909		ext4_ext_binsearch_idx(inode, path + ppos, block);
 910		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 911		path[ppos].p_depth = i;
 912		path[ppos].p_ext = NULL;
 913
 914		bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
 915					    flags);
 916		if (IS_ERR(bh)) {
 917			ret = PTR_ERR(bh);
 918			goto err;
 919		}
 920
 921		eh = ext_block_hdr(bh);
 922		ppos++;
 
 
 
 
 
 
 
 923		path[ppos].p_bh = bh;
 924		path[ppos].p_hdr = eh;
 925	}
 926
 927	path[ppos].p_depth = i;
 928	path[ppos].p_ext = NULL;
 929	path[ppos].p_idx = NULL;
 930
 931	/* find extent */
 932	ext4_ext_binsearch(inode, path + ppos, block);
 933	/* if not an empty leaf */
 934	if (path[ppos].p_ext)
 935		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 936
 937	ext4_ext_show_path(inode, path);
 938
 939	return path;
 940
 941err:
 942	ext4_ext_drop_refs(path);
 943	kfree(path);
 944	if (orig_path)
 945		*orig_path = NULL;
 946	return ERR_PTR(ret);
 947}
 948
 949/*
 950 * ext4_ext_insert_index:
 951 * insert new index [@logical;@ptr] into the block at @curp;
 952 * check where to insert: before @curp or after @curp
 953 */
 954static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 955				 struct ext4_ext_path *curp,
 956				 int logical, ext4_fsblk_t ptr)
 957{
 958	struct ext4_extent_idx *ix;
 959	int len, err;
 960
 961	err = ext4_ext_get_access(handle, inode, curp);
 962	if (err)
 963		return err;
 964
 965	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 966		EXT4_ERROR_INODE(inode,
 967				 "logical %d == ei_block %d!",
 968				 logical, le32_to_cpu(curp->p_idx->ei_block));
 969		return -EFSCORRUPTED;
 970	}
 971
 972	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 973			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
 974		EXT4_ERROR_INODE(inode,
 975				 "eh_entries %d >= eh_max %d!",
 976				 le16_to_cpu(curp->p_hdr->eh_entries),
 977				 le16_to_cpu(curp->p_hdr->eh_max));
 978		return -EFSCORRUPTED;
 979	}
 980
 981	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 982		/* insert after */
 983		ext_debug("insert new index %d after: %llu\n", logical, ptr);
 984		ix = curp->p_idx + 1;
 985	} else {
 986		/* insert before */
 987		ext_debug("insert new index %d before: %llu\n", logical, ptr);
 988		ix = curp->p_idx;
 989	}
 990
 991	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 992	BUG_ON(len < 0);
 993	if (len > 0) {
 994		ext_debug("insert new index %d: "
 995				"move %d indices from 0x%p to 0x%p\n",
 996				logical, len, ix, ix + 1);
 997		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 998	}
 999
1000	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1001		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1002		return -EFSCORRUPTED;
1003	}
1004
1005	ix->ei_block = cpu_to_le32(logical);
1006	ext4_idx_store_pblock(ix, ptr);
1007	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1008
1009	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1010		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1011		return -EFSCORRUPTED;
1012	}
1013
1014	err = ext4_ext_dirty(handle, inode, curp);
1015	ext4_std_error(inode->i_sb, err);
1016
1017	return err;
1018}
1019
1020/*
1021 * ext4_ext_split:
1022 * inserts new subtree into the path, using free index entry
1023 * at depth @at:
1024 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1025 * - makes decision where to split
1026 * - moves remaining extents and index entries (right to the split point)
1027 *   into the newly allocated blocks
1028 * - initializes subtree
1029 */
1030static int ext4_ext_split(handle_t *handle, struct inode *inode,
1031			  unsigned int flags,
1032			  struct ext4_ext_path *path,
1033			  struct ext4_extent *newext, int at)
1034{
1035	struct buffer_head *bh = NULL;
1036	int depth = ext_depth(inode);
1037	struct ext4_extent_header *neh;
1038	struct ext4_extent_idx *fidx;
1039	int i = at, k, m, a;
1040	ext4_fsblk_t newblock, oldblock;
1041	__le32 border;
1042	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1043	int err = 0;
1044
1045	/* make decision: where to split? */
1046	/* FIXME: now decision is simplest: at current extent */
1047
1048	/* if current leaf will be split, then we should use
1049	 * border from split point */
1050	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1051		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1052		return -EFSCORRUPTED;
1053	}
1054	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1055		border = path[depth].p_ext[1].ee_block;
1056		ext_debug("leaf will be split."
1057				" next leaf starts at %d\n",
1058				  le32_to_cpu(border));
1059	} else {
1060		border = newext->ee_block;
1061		ext_debug("leaf will be added."
1062				" next leaf starts at %d\n",
1063				le32_to_cpu(border));
1064	}
1065
1066	/*
1067	 * If error occurs, then we break processing
1068	 * and mark filesystem read-only. index won't
1069	 * be inserted and tree will be in consistent
1070	 * state. Next mount will repair buffers too.
1071	 */
1072
1073	/*
1074	 * Get array to track all allocated blocks.
1075	 * We need this to handle errors and free blocks
1076	 * upon them.
1077	 */
1078	ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1079	if (!ablocks)
1080		return -ENOMEM;
1081
1082	/* allocate all needed blocks */
1083	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1084	for (a = 0; a < depth - at; a++) {
1085		newblock = ext4_ext_new_meta_block(handle, inode, path,
1086						   newext, &err, flags);
1087		if (newblock == 0)
1088			goto cleanup;
1089		ablocks[a] = newblock;
1090	}
1091
1092	/* initialize new leaf */
1093	newblock = ablocks[--a];
1094	if (unlikely(newblock == 0)) {
1095		EXT4_ERROR_INODE(inode, "newblock == 0!");
1096		err = -EFSCORRUPTED;
1097		goto cleanup;
1098	}
1099	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1100	if (unlikely(!bh)) {
1101		err = -ENOMEM;
1102		goto cleanup;
1103	}
1104	lock_buffer(bh);
1105
1106	err = ext4_journal_get_create_access(handle, bh);
1107	if (err)
1108		goto cleanup;
1109
1110	neh = ext_block_hdr(bh);
1111	neh->eh_entries = 0;
1112	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1113	neh->eh_magic = EXT4_EXT_MAGIC;
1114	neh->eh_depth = 0;
1115
1116	/* move remainder of path[depth] to the new leaf */
1117	if (unlikely(path[depth].p_hdr->eh_entries !=
1118		     path[depth].p_hdr->eh_max)) {
1119		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1120				 path[depth].p_hdr->eh_entries,
1121				 path[depth].p_hdr->eh_max);
1122		err = -EFSCORRUPTED;
1123		goto cleanup;
1124	}
1125	/* start copy from next extent */
1126	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1127	ext4_ext_show_move(inode, path, newblock, depth);
1128	if (m) {
1129		struct ext4_extent *ex;
1130		ex = EXT_FIRST_EXTENT(neh);
1131		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1132		le16_add_cpu(&neh->eh_entries, m);
1133	}
1134
1135	ext4_extent_block_csum_set(inode, neh);
1136	set_buffer_uptodate(bh);
1137	unlock_buffer(bh);
1138
1139	err = ext4_handle_dirty_metadata(handle, inode, bh);
1140	if (err)
1141		goto cleanup;
1142	brelse(bh);
1143	bh = NULL;
1144
1145	/* correct old leaf */
1146	if (m) {
1147		err = ext4_ext_get_access(handle, inode, path + depth);
1148		if (err)
1149			goto cleanup;
1150		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1151		err = ext4_ext_dirty(handle, inode, path + depth);
1152		if (err)
1153			goto cleanup;
1154
1155	}
1156
1157	/* create intermediate indexes */
1158	k = depth - at - 1;
1159	if (unlikely(k < 0)) {
1160		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1161		err = -EFSCORRUPTED;
1162		goto cleanup;
1163	}
1164	if (k)
1165		ext_debug("create %d intermediate indices\n", k);
1166	/* insert new index into current index block */
1167	/* current depth stored in i var */
1168	i = depth - 1;
1169	while (k--) {
1170		oldblock = newblock;
1171		newblock = ablocks[--a];
1172		bh = sb_getblk(inode->i_sb, newblock);
1173		if (unlikely(!bh)) {
1174			err = -ENOMEM;
1175			goto cleanup;
1176		}
1177		lock_buffer(bh);
1178
1179		err = ext4_journal_get_create_access(handle, bh);
1180		if (err)
1181			goto cleanup;
1182
1183		neh = ext_block_hdr(bh);
1184		neh->eh_entries = cpu_to_le16(1);
1185		neh->eh_magic = EXT4_EXT_MAGIC;
1186		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1187		neh->eh_depth = cpu_to_le16(depth - i);
1188		fidx = EXT_FIRST_INDEX(neh);
1189		fidx->ei_block = border;
1190		ext4_idx_store_pblock(fidx, oldblock);
1191
1192		ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1193				i, newblock, le32_to_cpu(border), oldblock);
1194
1195		/* move remainder of path[i] to the new index block */
1196		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1197					EXT_LAST_INDEX(path[i].p_hdr))) {
1198			EXT4_ERROR_INODE(inode,
1199					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1200					 le32_to_cpu(path[i].p_ext->ee_block));
1201			err = -EFSCORRUPTED;
1202			goto cleanup;
1203		}
1204		/* start copy indexes */
1205		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1206		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1207				EXT_MAX_INDEX(path[i].p_hdr));
1208		ext4_ext_show_move(inode, path, newblock, i);
1209		if (m) {
1210			memmove(++fidx, path[i].p_idx,
1211				sizeof(struct ext4_extent_idx) * m);
1212			le16_add_cpu(&neh->eh_entries, m);
1213		}
1214		ext4_extent_block_csum_set(inode, neh);
1215		set_buffer_uptodate(bh);
1216		unlock_buffer(bh);
1217
1218		err = ext4_handle_dirty_metadata(handle, inode, bh);
1219		if (err)
1220			goto cleanup;
1221		brelse(bh);
1222		bh = NULL;
1223
1224		/* correct old index */
1225		if (m) {
1226			err = ext4_ext_get_access(handle, inode, path + i);
1227			if (err)
1228				goto cleanup;
1229			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1230			err = ext4_ext_dirty(handle, inode, path + i);
1231			if (err)
1232				goto cleanup;
1233		}
1234
1235		i--;
1236	}
1237
1238	/* insert new index */
1239	err = ext4_ext_insert_index(handle, inode, path + at,
1240				    le32_to_cpu(border), newblock);
1241
1242cleanup:
1243	if (bh) {
1244		if (buffer_locked(bh))
1245			unlock_buffer(bh);
1246		brelse(bh);
1247	}
1248
1249	if (err) {
1250		/* free all allocated blocks in error case */
1251		for (i = 0; i < depth; i++) {
1252			if (!ablocks[i])
1253				continue;
1254			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1255					 EXT4_FREE_BLOCKS_METADATA);
1256		}
1257	}
1258	kfree(ablocks);
1259
1260	return err;
1261}
1262
1263/*
1264 * ext4_ext_grow_indepth:
1265 * implements tree growing procedure:
1266 * - allocates new block
1267 * - moves top-level data (index block or leaf) into the new block
1268 * - initializes new top-level, creating index that points to the
1269 *   just created block
1270 */
1271static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1272				 unsigned int flags)
1273{
1274	struct ext4_extent_header *neh;
1275	struct buffer_head *bh;
1276	ext4_fsblk_t newblock, goal = 0;
1277	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1278	int err = 0;
1279
1280	/* Try to prepend new index to old one */
1281	if (ext_depth(inode))
1282		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1283	if (goal > le32_to_cpu(es->s_first_data_block)) {
1284		flags |= EXT4_MB_HINT_TRY_GOAL;
1285		goal--;
1286	} else
1287		goal = ext4_inode_to_goal_block(inode);
1288	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1289					NULL, &err);
1290	if (newblock == 0)
1291		return err;
1292
1293	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1294	if (unlikely(!bh))
1295		return -ENOMEM;
1296	lock_buffer(bh);
1297
1298	err = ext4_journal_get_create_access(handle, bh);
1299	if (err) {
1300		unlock_buffer(bh);
1301		goto out;
1302	}
1303
1304	/* move top-level index/leaf into new block */
1305	memmove(bh->b_data, EXT4_I(inode)->i_data,
1306		sizeof(EXT4_I(inode)->i_data));
1307
1308	/* set size of new block */
1309	neh = ext_block_hdr(bh);
1310	/* old root could have indexes or leaves
1311	 * so calculate e_max right way */
1312	if (ext_depth(inode))
1313		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1314	else
1315		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1316	neh->eh_magic = EXT4_EXT_MAGIC;
1317	ext4_extent_block_csum_set(inode, neh);
1318	set_buffer_uptodate(bh);
1319	unlock_buffer(bh);
1320
1321	err = ext4_handle_dirty_metadata(handle, inode, bh);
1322	if (err)
1323		goto out;
1324
1325	/* Update top-level index: num,max,pointer */
1326	neh = ext_inode_hdr(inode);
1327	neh->eh_entries = cpu_to_le16(1);
1328	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1329	if (neh->eh_depth == 0) {
1330		/* Root extent block becomes index block */
1331		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1332		EXT_FIRST_INDEX(neh)->ei_block =
1333			EXT_FIRST_EXTENT(neh)->ee_block;
1334	}
1335	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1336		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1337		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1338		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1339
1340	le16_add_cpu(&neh->eh_depth, 1);
1341	ext4_mark_inode_dirty(handle, inode);
1342out:
1343	brelse(bh);
1344
1345	return err;
1346}
1347
1348/*
1349 * ext4_ext_create_new_leaf:
1350 * finds empty index and adds new leaf.
1351 * if no free index is found, then it requests in-depth growing.
1352 */
1353static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1354				    unsigned int mb_flags,
1355				    unsigned int gb_flags,
1356				    struct ext4_ext_path **ppath,
1357				    struct ext4_extent *newext)
1358{
1359	struct ext4_ext_path *path = *ppath;
1360	struct ext4_ext_path *curp;
1361	int depth, i, err = 0;
1362
1363repeat:
1364	i = depth = ext_depth(inode);
1365
1366	/* walk up to the tree and look for free index entry */
1367	curp = path + depth;
1368	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1369		i--;
1370		curp--;
1371	}
1372
1373	/* we use already allocated block for index block,
1374	 * so subsequent data blocks should be contiguous */
1375	if (EXT_HAS_FREE_INDEX(curp)) {
1376		/* if we found index with free entry, then use that
1377		 * entry: create all needed subtree and add new leaf */
1378		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1379		if (err)
1380			goto out;
1381
1382		/* refill path */
1383		path = ext4_find_extent(inode,
1384				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1385				    ppath, gb_flags);
1386		if (IS_ERR(path))
1387			err = PTR_ERR(path);
1388	} else {
1389		/* tree is full, time to grow in depth */
1390		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1391		if (err)
1392			goto out;
1393
1394		/* refill path */
1395		path = ext4_find_extent(inode,
1396				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1397				    ppath, gb_flags);
1398		if (IS_ERR(path)) {
1399			err = PTR_ERR(path);
1400			goto out;
1401		}
1402
1403		/*
1404		 * only first (depth 0 -> 1) produces free space;
1405		 * in all other cases we have to split the grown tree
1406		 */
1407		depth = ext_depth(inode);
1408		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1409			/* now we need to split */
1410			goto repeat;
1411		}
1412	}
1413
1414out:
1415	return err;
1416}
1417
1418/*
1419 * search the closest allocated block to the left for *logical
1420 * and returns it at @logical + it's physical address at @phys
1421 * if *logical is the smallest allocated block, the function
1422 * returns 0 at @phys
1423 * return value contains 0 (success) or error code
1424 */
1425static int ext4_ext_search_left(struct inode *inode,
1426				struct ext4_ext_path *path,
1427				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1428{
1429	struct ext4_extent_idx *ix;
1430	struct ext4_extent *ex;
1431	int depth, ee_len;
1432
1433	if (unlikely(path == NULL)) {
1434		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1435		return -EFSCORRUPTED;
1436	}
1437	depth = path->p_depth;
1438	*phys = 0;
1439
1440	if (depth == 0 && path->p_ext == NULL)
1441		return 0;
1442
1443	/* usually extent in the path covers blocks smaller
1444	 * then *logical, but it can be that extent is the
1445	 * first one in the file */
1446
1447	ex = path[depth].p_ext;
1448	ee_len = ext4_ext_get_actual_len(ex);
1449	if (*logical < le32_to_cpu(ex->ee_block)) {
1450		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1451			EXT4_ERROR_INODE(inode,
1452					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1453					 *logical, le32_to_cpu(ex->ee_block));
1454			return -EFSCORRUPTED;
1455		}
1456		while (--depth >= 0) {
1457			ix = path[depth].p_idx;
1458			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1459				EXT4_ERROR_INODE(inode,
1460				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1461				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1462				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1463		le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1464				  depth);
1465				return -EFSCORRUPTED;
1466			}
1467		}
1468		return 0;
1469	}
1470
1471	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1472		EXT4_ERROR_INODE(inode,
1473				 "logical %d < ee_block %d + ee_len %d!",
1474				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1475		return -EFSCORRUPTED;
1476	}
1477
1478	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1479	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1480	return 0;
1481}
1482
1483/*
1484 * search the closest allocated block to the right for *logical
1485 * and returns it at @logical + it's physical address at @phys
1486 * if *logical is the largest allocated block, the function
1487 * returns 0 at @phys
1488 * return value contains 0 (success) or error code
1489 */
1490static int ext4_ext_search_right(struct inode *inode,
1491				 struct ext4_ext_path *path,
1492				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1493				 struct ext4_extent **ret_ex)
1494{
1495	struct buffer_head *bh = NULL;
1496	struct ext4_extent_header *eh;
1497	struct ext4_extent_idx *ix;
1498	struct ext4_extent *ex;
1499	ext4_fsblk_t block;
1500	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1501	int ee_len;
1502
1503	if (unlikely(path == NULL)) {
1504		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1505		return -EFSCORRUPTED;
1506	}
1507	depth = path->p_depth;
1508	*phys = 0;
1509
1510	if (depth == 0 && path->p_ext == NULL)
1511		return 0;
1512
1513	/* usually extent in the path covers blocks smaller
1514	 * then *logical, but it can be that extent is the
1515	 * first one in the file */
1516
1517	ex = path[depth].p_ext;
1518	ee_len = ext4_ext_get_actual_len(ex);
1519	if (*logical < le32_to_cpu(ex->ee_block)) {
1520		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1521			EXT4_ERROR_INODE(inode,
1522					 "first_extent(path[%d].p_hdr) != ex",
1523					 depth);
1524			return -EFSCORRUPTED;
1525		}
1526		while (--depth >= 0) {
1527			ix = path[depth].p_idx;
1528			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1529				EXT4_ERROR_INODE(inode,
1530						 "ix != EXT_FIRST_INDEX *logical %d!",
1531						 *logical);
1532				return -EFSCORRUPTED;
1533			}
1534		}
1535		goto found_extent;
1536	}
1537
1538	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1539		EXT4_ERROR_INODE(inode,
1540				 "logical %d < ee_block %d + ee_len %d!",
1541				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1542		return -EFSCORRUPTED;
1543	}
1544
1545	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1546		/* next allocated block in this leaf */
1547		ex++;
1548		goto found_extent;
1549	}
1550
1551	/* go up and search for index to the right */
1552	while (--depth >= 0) {
1553		ix = path[depth].p_idx;
1554		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1555			goto got_index;
1556	}
1557
1558	/* we've gone up to the root and found no index to the right */
1559	return 0;
1560
1561got_index:
1562	/* we've found index to the right, let's
1563	 * follow it and find the closest allocated
1564	 * block to the right */
1565	ix++;
1566	block = ext4_idx_pblock(ix);
1567	while (++depth < path->p_depth) {
1568		/* subtract from p_depth to get proper eh_depth */
1569		bh = read_extent_tree_block(inode, block,
1570					    path->p_depth - depth, 0);
1571		if (IS_ERR(bh))
1572			return PTR_ERR(bh);
1573		eh = ext_block_hdr(bh);
1574		ix = EXT_FIRST_INDEX(eh);
1575		block = ext4_idx_pblock(ix);
1576		put_bh(bh);
1577	}
1578
1579	bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1580	if (IS_ERR(bh))
1581		return PTR_ERR(bh);
1582	eh = ext_block_hdr(bh);
1583	ex = EXT_FIRST_EXTENT(eh);
1584found_extent:
1585	*logical = le32_to_cpu(ex->ee_block);
1586	*phys = ext4_ext_pblock(ex);
1587	*ret_ex = ex;
1588	if (bh)
1589		put_bh(bh);
1590	return 0;
1591}
1592
1593/*
1594 * ext4_ext_next_allocated_block:
1595 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1596 * NOTE: it considers block number from index entry as
1597 * allocated block. Thus, index entries have to be consistent
1598 * with leaves.
1599 */
1600ext4_lblk_t
1601ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1602{
1603	int depth;
1604
1605	BUG_ON(path == NULL);
1606	depth = path->p_depth;
1607
1608	if (depth == 0 && path->p_ext == NULL)
1609		return EXT_MAX_BLOCKS;
1610
1611	while (depth >= 0) {
1612		if (depth == path->p_depth) {
1613			/* leaf */
1614			if (path[depth].p_ext &&
1615				path[depth].p_ext !=
1616					EXT_LAST_EXTENT(path[depth].p_hdr))
1617			  return le32_to_cpu(path[depth].p_ext[1].ee_block);
1618		} else {
1619			/* index */
1620			if (path[depth].p_idx !=
1621					EXT_LAST_INDEX(path[depth].p_hdr))
1622			  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1623		}
1624		depth--;
1625	}
1626
1627	return EXT_MAX_BLOCKS;
1628}
1629
1630/*
1631 * ext4_ext_next_leaf_block:
1632 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1633 */
1634static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1635{
1636	int depth;
1637
1638	BUG_ON(path == NULL);
1639	depth = path->p_depth;
1640
1641	/* zero-tree has no leaf blocks at all */
1642	if (depth == 0)
1643		return EXT_MAX_BLOCKS;
1644
1645	/* go to index block */
1646	depth--;
1647
1648	while (depth >= 0) {
1649		if (path[depth].p_idx !=
1650				EXT_LAST_INDEX(path[depth].p_hdr))
1651			return (ext4_lblk_t)
1652				le32_to_cpu(path[depth].p_idx[1].ei_block);
1653		depth--;
1654	}
1655
1656	return EXT_MAX_BLOCKS;
1657}
1658
1659/*
1660 * ext4_ext_correct_indexes:
1661 * if leaf gets modified and modified extent is first in the leaf,
1662 * then we have to correct all indexes above.
1663 * TODO: do we need to correct tree in all cases?
1664 */
1665static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1666				struct ext4_ext_path *path)
1667{
1668	struct ext4_extent_header *eh;
1669	int depth = ext_depth(inode);
1670	struct ext4_extent *ex;
1671	__le32 border;
1672	int k, err = 0;
1673
1674	eh = path[depth].p_hdr;
1675	ex = path[depth].p_ext;
1676
1677	if (unlikely(ex == NULL || eh == NULL)) {
1678		EXT4_ERROR_INODE(inode,
1679				 "ex %p == NULL or eh %p == NULL", ex, eh);
1680		return -EFSCORRUPTED;
1681	}
1682
1683	if (depth == 0) {
1684		/* there is no tree at all */
1685		return 0;
1686	}
1687
1688	if (ex != EXT_FIRST_EXTENT(eh)) {
1689		/* we correct tree if first leaf got modified only */
1690		return 0;
1691	}
1692
1693	/*
1694	 * TODO: we need correction if border is smaller than current one
1695	 */
1696	k = depth - 1;
1697	border = path[depth].p_ext->ee_block;
1698	err = ext4_ext_get_access(handle, inode, path + k);
1699	if (err)
1700		return err;
1701	path[k].p_idx->ei_block = border;
1702	err = ext4_ext_dirty(handle, inode, path + k);
1703	if (err)
1704		return err;
1705
1706	while (k--) {
1707		/* change all left-side indexes */
1708		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1709			break;
1710		err = ext4_ext_get_access(handle, inode, path + k);
1711		if (err)
1712			break;
1713		path[k].p_idx->ei_block = border;
1714		err = ext4_ext_dirty(handle, inode, path + k);
1715		if (err)
1716			break;
1717	}
1718
1719	return err;
1720}
1721
1722int
1723ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1724				struct ext4_extent *ex2)
1725{
1726	unsigned short ext1_ee_len, ext2_ee_len;
1727
1728	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1729		return 0;
1730
1731	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1732	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1733
1734	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1735			le32_to_cpu(ex2->ee_block))
1736		return 0;
1737
1738	/*
1739	 * To allow future support for preallocated extents to be added
1740	 * as an RO_COMPAT feature, refuse to merge to extents if
1741	 * this can result in the top bit of ee_len being set.
1742	 */
1743	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1744		return 0;
1745	/*
1746	 * The check for IO to unwritten extent is somewhat racy as we
1747	 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1748	 * dropping i_data_sem. But reserved blocks should save us in that
1749	 * case.
1750	 */
1751	if (ext4_ext_is_unwritten(ex1) &&
1752	    (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1753	     atomic_read(&EXT4_I(inode)->i_unwritten) ||
1754	     (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1755		return 0;
1756#ifdef AGGRESSIVE_TEST
1757	if (ext1_ee_len >= 4)
1758		return 0;
1759#endif
1760
1761	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1762		return 1;
1763	return 0;
1764}
1765
1766/*
1767 * This function tries to merge the "ex" extent to the next extent in the tree.
1768 * It always tries to merge towards right. If you want to merge towards
1769 * left, pass "ex - 1" as argument instead of "ex".
1770 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1771 * 1 if they got merged.
1772 */
1773static int ext4_ext_try_to_merge_right(struct inode *inode,
1774				 struct ext4_ext_path *path,
1775				 struct ext4_extent *ex)
1776{
1777	struct ext4_extent_header *eh;
1778	unsigned int depth, len;
1779	int merge_done = 0, unwritten;
1780
1781	depth = ext_depth(inode);
1782	BUG_ON(path[depth].p_hdr == NULL);
1783	eh = path[depth].p_hdr;
1784
1785	while (ex < EXT_LAST_EXTENT(eh)) {
1786		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1787			break;
1788		/* merge with next extent! */
1789		unwritten = ext4_ext_is_unwritten(ex);
1790		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1791				+ ext4_ext_get_actual_len(ex + 1));
1792		if (unwritten)
1793			ext4_ext_mark_unwritten(ex);
1794
1795		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1796			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1797				* sizeof(struct ext4_extent);
1798			memmove(ex + 1, ex + 2, len);
1799		}
1800		le16_add_cpu(&eh->eh_entries, -1);
1801		merge_done = 1;
1802		WARN_ON(eh->eh_entries == 0);
1803		if (!eh->eh_entries)
1804			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1805	}
1806
1807	return merge_done;
1808}
1809
1810/*
1811 * This function does a very simple check to see if we can collapse
1812 * an extent tree with a single extent tree leaf block into the inode.
1813 */
1814static void ext4_ext_try_to_merge_up(handle_t *handle,
1815				     struct inode *inode,
1816				     struct ext4_ext_path *path)
1817{
1818	size_t s;
1819	unsigned max_root = ext4_ext_space_root(inode, 0);
1820	ext4_fsblk_t blk;
1821
1822	if ((path[0].p_depth != 1) ||
1823	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1824	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1825		return;
1826
1827	/*
1828	 * We need to modify the block allocation bitmap and the block
1829	 * group descriptor to release the extent tree block.  If we
1830	 * can't get the journal credits, give up.
1831	 */
1832	if (ext4_journal_extend(handle, 2))
1833		return;
1834
1835	/*
1836	 * Copy the extent data up to the inode
1837	 */
1838	blk = ext4_idx_pblock(path[0].p_idx);
1839	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1840		sizeof(struct ext4_extent_idx);
1841	s += sizeof(struct ext4_extent_header);
1842
1843	path[1].p_maxdepth = path[0].p_maxdepth;
1844	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1845	path[0].p_depth = 0;
1846	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1847		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1848	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1849
1850	brelse(path[1].p_bh);
1851	ext4_free_blocks(handle, inode, NULL, blk, 1,
1852			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1853}
1854
1855/*
1856 * This function tries to merge the @ex extent to neighbours in the tree.
1857 * return 1 if merge left else 0.
1858 */
1859static void ext4_ext_try_to_merge(handle_t *handle,
1860				  struct inode *inode,
1861				  struct ext4_ext_path *path,
1862				  struct ext4_extent *ex) {
1863	struct ext4_extent_header *eh;
1864	unsigned int depth;
1865	int merge_done = 0;
1866
1867	depth = ext_depth(inode);
1868	BUG_ON(path[depth].p_hdr == NULL);
1869	eh = path[depth].p_hdr;
1870
1871	if (ex > EXT_FIRST_EXTENT(eh))
1872		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1873
1874	if (!merge_done)
1875		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1876
1877	ext4_ext_try_to_merge_up(handle, inode, path);
1878}
1879
1880/*
1881 * check if a portion of the "newext" extent overlaps with an
1882 * existing extent.
1883 *
1884 * If there is an overlap discovered, it updates the length of the newext
1885 * such that there will be no overlap, and then returns 1.
1886 * If there is no overlap found, it returns 0.
1887 */
1888static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1889					   struct inode *inode,
1890					   struct ext4_extent *newext,
1891					   struct ext4_ext_path *path)
1892{
1893	ext4_lblk_t b1, b2;
1894	unsigned int depth, len1;
1895	unsigned int ret = 0;
1896
1897	b1 = le32_to_cpu(newext->ee_block);
1898	len1 = ext4_ext_get_actual_len(newext);
1899	depth = ext_depth(inode);
1900	if (!path[depth].p_ext)
1901		goto out;
1902	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1903
1904	/*
1905	 * get the next allocated block if the extent in the path
1906	 * is before the requested block(s)
1907	 */
1908	if (b2 < b1) {
1909		b2 = ext4_ext_next_allocated_block(path);
1910		if (b2 == EXT_MAX_BLOCKS)
1911			goto out;
1912		b2 = EXT4_LBLK_CMASK(sbi, b2);
1913	}
1914
1915	/* check for wrap through zero on extent logical start block*/
1916	if (b1 + len1 < b1) {
1917		len1 = EXT_MAX_BLOCKS - b1;
1918		newext->ee_len = cpu_to_le16(len1);
1919		ret = 1;
1920	}
1921
1922	/* check for overlap */
1923	if (b1 + len1 > b2) {
1924		newext->ee_len = cpu_to_le16(b2 - b1);
1925		ret = 1;
1926	}
1927out:
1928	return ret;
1929}
1930
1931/*
1932 * ext4_ext_insert_extent:
1933 * tries to merge requsted extent into the existing extent or
1934 * inserts requested extent as new one into the tree,
1935 * creating new leaf in the no-space case.
1936 */
1937int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1938				struct ext4_ext_path **ppath,
1939				struct ext4_extent *newext, int gb_flags)
1940{
1941	struct ext4_ext_path *path = *ppath;
1942	struct ext4_extent_header *eh;
1943	struct ext4_extent *ex, *fex;
1944	struct ext4_extent *nearex; /* nearest extent */
1945	struct ext4_ext_path *npath = NULL;
1946	int depth, len, err;
1947	ext4_lblk_t next;
1948	int mb_flags = 0, unwritten;
1949
1950	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1951		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1952	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1953		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1954		return -EFSCORRUPTED;
1955	}
1956	depth = ext_depth(inode);
1957	ex = path[depth].p_ext;
1958	eh = path[depth].p_hdr;
1959	if (unlikely(path[depth].p_hdr == NULL)) {
1960		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1961		return -EFSCORRUPTED;
1962	}
1963
1964	/* try to insert block into found extent and return */
1965	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1966
1967		/*
1968		 * Try to see whether we should rather test the extent on
1969		 * right from ex, or from the left of ex. This is because
1970		 * ext4_find_extent() can return either extent on the
1971		 * left, or on the right from the searched position. This
1972		 * will make merging more effective.
1973		 */
1974		if (ex < EXT_LAST_EXTENT(eh) &&
1975		    (le32_to_cpu(ex->ee_block) +
1976		    ext4_ext_get_actual_len(ex) <
1977		    le32_to_cpu(newext->ee_block))) {
1978			ex += 1;
1979			goto prepend;
1980		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1981			   (le32_to_cpu(newext->ee_block) +
1982			   ext4_ext_get_actual_len(newext) <
1983			   le32_to_cpu(ex->ee_block)))
1984			ex -= 1;
1985
1986		/* Try to append newex to the ex */
1987		if (ext4_can_extents_be_merged(inode, ex, newext)) {
1988			ext_debug("append [%d]%d block to %u:[%d]%d"
1989				  "(from %llu)\n",
1990				  ext4_ext_is_unwritten(newext),
1991				  ext4_ext_get_actual_len(newext),
1992				  le32_to_cpu(ex->ee_block),
1993				  ext4_ext_is_unwritten(ex),
1994				  ext4_ext_get_actual_len(ex),
1995				  ext4_ext_pblock(ex));
1996			err = ext4_ext_get_access(handle, inode,
1997						  path + depth);
1998			if (err)
1999				return err;
2000			unwritten = ext4_ext_is_unwritten(ex);
2001			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2002					+ ext4_ext_get_actual_len(newext));
2003			if (unwritten)
2004				ext4_ext_mark_unwritten(ex);
2005			eh = path[depth].p_hdr;
2006			nearex = ex;
2007			goto merge;
2008		}
2009
2010prepend:
2011		/* Try to prepend newex to the ex */
2012		if (ext4_can_extents_be_merged(inode, newext, ex)) {
2013			ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2014				  "(from %llu)\n",
2015				  le32_to_cpu(newext->ee_block),
2016				  ext4_ext_is_unwritten(newext),
2017				  ext4_ext_get_actual_len(newext),
2018				  le32_to_cpu(ex->ee_block),
2019				  ext4_ext_is_unwritten(ex),
2020				  ext4_ext_get_actual_len(ex),
2021				  ext4_ext_pblock(ex));
2022			err = ext4_ext_get_access(handle, inode,
2023						  path + depth);
2024			if (err)
2025				return err;
2026
2027			unwritten = ext4_ext_is_unwritten(ex);
2028			ex->ee_block = newext->ee_block;
2029			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2030			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2031					+ ext4_ext_get_actual_len(newext));
2032			if (unwritten)
2033				ext4_ext_mark_unwritten(ex);
2034			eh = path[depth].p_hdr;
2035			nearex = ex;
2036			goto merge;
2037		}
2038	}
2039
2040	depth = ext_depth(inode);
2041	eh = path[depth].p_hdr;
2042	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2043		goto has_space;
2044
2045	/* probably next leaf has space for us? */
2046	fex = EXT_LAST_EXTENT(eh);
2047	next = EXT_MAX_BLOCKS;
2048	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2049		next = ext4_ext_next_leaf_block(path);
2050	if (next != EXT_MAX_BLOCKS) {
2051		ext_debug("next leaf block - %u\n", next);
2052		BUG_ON(npath != NULL);
2053		npath = ext4_find_extent(inode, next, NULL, 0);
2054		if (IS_ERR(npath))
2055			return PTR_ERR(npath);
2056		BUG_ON(npath->p_depth != path->p_depth);
2057		eh = npath[depth].p_hdr;
2058		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2059			ext_debug("next leaf isn't full(%d)\n",
2060				  le16_to_cpu(eh->eh_entries));
2061			path = npath;
2062			goto has_space;
2063		}
2064		ext_debug("next leaf has no free space(%d,%d)\n",
2065			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2066	}
2067
2068	/*
2069	 * There is no free space in the found leaf.
2070	 * We're gonna add a new leaf in the tree.
2071	 */
2072	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2073		mb_flags |= EXT4_MB_USE_RESERVED;
2074	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2075				       ppath, newext);
2076	if (err)
2077		goto cleanup;
2078	depth = ext_depth(inode);
2079	eh = path[depth].p_hdr;
2080
2081has_space:
2082	nearex = path[depth].p_ext;
2083
2084	err = ext4_ext_get_access(handle, inode, path + depth);
2085	if (err)
2086		goto cleanup;
2087
2088	if (!nearex) {
2089		/* there is no extent in this leaf, create first one */
2090		ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2091				le32_to_cpu(newext->ee_block),
2092				ext4_ext_pblock(newext),
2093				ext4_ext_is_unwritten(newext),
2094				ext4_ext_get_actual_len(newext));
2095		nearex = EXT_FIRST_EXTENT(eh);
2096	} else {
2097		if (le32_to_cpu(newext->ee_block)
2098			   > le32_to_cpu(nearex->ee_block)) {
2099			/* Insert after */
2100			ext_debug("insert %u:%llu:[%d]%d before: "
2101					"nearest %p\n",
2102					le32_to_cpu(newext->ee_block),
2103					ext4_ext_pblock(newext),
2104					ext4_ext_is_unwritten(newext),
2105					ext4_ext_get_actual_len(newext),
2106					nearex);
2107			nearex++;
2108		} else {
2109			/* Insert before */
2110			BUG_ON(newext->ee_block == nearex->ee_block);
2111			ext_debug("insert %u:%llu:[%d]%d after: "
2112					"nearest %p\n",
2113					le32_to_cpu(newext->ee_block),
2114					ext4_ext_pblock(newext),
2115					ext4_ext_is_unwritten(newext),
2116					ext4_ext_get_actual_len(newext),
2117					nearex);
2118		}
2119		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2120		if (len > 0) {
2121			ext_debug("insert %u:%llu:[%d]%d: "
2122					"move %d extents from 0x%p to 0x%p\n",
2123					le32_to_cpu(newext->ee_block),
2124					ext4_ext_pblock(newext),
2125					ext4_ext_is_unwritten(newext),
2126					ext4_ext_get_actual_len(newext),
2127					len, nearex, nearex + 1);
2128			memmove(nearex + 1, nearex,
2129				len * sizeof(struct ext4_extent));
2130		}
2131	}
2132
2133	le16_add_cpu(&eh->eh_entries, 1);
2134	path[depth].p_ext = nearex;
2135	nearex->ee_block = newext->ee_block;
2136	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2137	nearex->ee_len = newext->ee_len;
2138
2139merge:
2140	/* try to merge extents */
2141	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2142		ext4_ext_try_to_merge(handle, inode, path, nearex);
2143
2144
2145	/* time to correct all indexes above */
2146	err = ext4_ext_correct_indexes(handle, inode, path);
2147	if (err)
2148		goto cleanup;
2149
2150	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2151
2152cleanup:
2153	ext4_ext_drop_refs(npath);
2154	kfree(npath);
2155	return err;
2156}
2157
2158static int ext4_fill_fiemap_extents(struct inode *inode,
2159				    ext4_lblk_t block, ext4_lblk_t num,
2160				    struct fiemap_extent_info *fieinfo)
2161{
2162	struct ext4_ext_path *path = NULL;
2163	struct ext4_extent *ex;
2164	struct extent_status es;
2165	ext4_lblk_t next, next_del, start = 0, end = 0;
2166	ext4_lblk_t last = block + num;
2167	int exists, depth = 0, err = 0;
2168	unsigned int flags = 0;
2169	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2170
2171	while (block < last && block != EXT_MAX_BLOCKS) {
2172		num = last - block;
2173		/* find extent for this block */
2174		down_read(&EXT4_I(inode)->i_data_sem);
2175
2176		path = ext4_find_extent(inode, block, &path, 0);
2177		if (IS_ERR(path)) {
2178			up_read(&EXT4_I(inode)->i_data_sem);
2179			err = PTR_ERR(path);
2180			path = NULL;
2181			break;
2182		}
2183
2184		depth = ext_depth(inode);
2185		if (unlikely(path[depth].p_hdr == NULL)) {
2186			up_read(&EXT4_I(inode)->i_data_sem);
2187			EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2188			err = -EFSCORRUPTED;
2189			break;
2190		}
2191		ex = path[depth].p_ext;
2192		next = ext4_ext_next_allocated_block(path);
2193
2194		flags = 0;
2195		exists = 0;
2196		if (!ex) {
2197			/* there is no extent yet, so try to allocate
2198			 * all requested space */
2199			start = block;
2200			end = block + num;
2201		} else if (le32_to_cpu(ex->ee_block) > block) {
2202			/* need to allocate space before found extent */
2203			start = block;
2204			end = le32_to_cpu(ex->ee_block);
2205			if (block + num < end)
2206				end = block + num;
2207		} else if (block >= le32_to_cpu(ex->ee_block)
2208					+ ext4_ext_get_actual_len(ex)) {
2209			/* need to allocate space after found extent */
2210			start = block;
2211			end = block + num;
2212			if (end >= next)
2213				end = next;
2214		} else if (block >= le32_to_cpu(ex->ee_block)) {
2215			/*
2216			 * some part of requested space is covered
2217			 * by found extent
2218			 */
2219			start = block;
2220			end = le32_to_cpu(ex->ee_block)
2221				+ ext4_ext_get_actual_len(ex);
2222			if (block + num < end)
2223				end = block + num;
2224			exists = 1;
2225		} else {
2226			BUG();
2227		}
2228		BUG_ON(end <= start);
2229
2230		if (!exists) {
2231			es.es_lblk = start;
2232			es.es_len = end - start;
2233			es.es_pblk = 0;
2234		} else {
2235			es.es_lblk = le32_to_cpu(ex->ee_block);
2236			es.es_len = ext4_ext_get_actual_len(ex);
2237			es.es_pblk = ext4_ext_pblock(ex);
2238			if (ext4_ext_is_unwritten(ex))
2239				flags |= FIEMAP_EXTENT_UNWRITTEN;
2240		}
2241
2242		/*
2243		 * Find delayed extent and update es accordingly. We call
2244		 * it even in !exists case to find out whether es is the
2245		 * last existing extent or not.
2246		 */
2247		next_del = ext4_find_delayed_extent(inode, &es);
2248		if (!exists && next_del) {
2249			exists = 1;
2250			flags |= (FIEMAP_EXTENT_DELALLOC |
2251				  FIEMAP_EXTENT_UNKNOWN);
2252		}
2253		up_read(&EXT4_I(inode)->i_data_sem);
2254
2255		if (unlikely(es.es_len == 0)) {
2256			EXT4_ERROR_INODE(inode, "es.es_len == 0");
2257			err = -EFSCORRUPTED;
2258			break;
2259		}
2260
2261		/*
2262		 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2263		 * we need to check next == EXT_MAX_BLOCKS because it is
2264		 * possible that an extent is with unwritten and delayed
2265		 * status due to when an extent is delayed allocated and
2266		 * is allocated by fallocate status tree will track both of
2267		 * them in a extent.
2268		 *
2269		 * So we could return a unwritten and delayed extent, and
2270		 * its block is equal to 'next'.
2271		 */
2272		if (next == next_del && next == EXT_MAX_BLOCKS) {
2273			flags |= FIEMAP_EXTENT_LAST;
2274			if (unlikely(next_del != EXT_MAX_BLOCKS ||
2275				     next != EXT_MAX_BLOCKS)) {
2276				EXT4_ERROR_INODE(inode,
2277						 "next extent == %u, next "
2278						 "delalloc extent = %u",
2279						 next, next_del);
2280				err = -EFSCORRUPTED;
2281				break;
2282			}
2283		}
2284
2285		if (exists) {
2286			err = fiemap_fill_next_extent(fieinfo,
2287				(__u64)es.es_lblk << blksize_bits,
2288				(__u64)es.es_pblk << blksize_bits,
2289				(__u64)es.es_len << blksize_bits,
2290				flags);
2291			if (err < 0)
2292				break;
2293			if (err == 1) {
2294				err = 0;
2295				break;
2296			}
2297		}
2298
2299		block = es.es_lblk + es.es_len;
2300	}
2301
2302	ext4_ext_drop_refs(path);
2303	kfree(path);
2304	return err;
2305}
2306
2307/*
2308 * ext4_ext_determine_hole - determine hole around given block
2309 * @inode:	inode we lookup in
2310 * @path:	path in extent tree to @lblk
2311 * @lblk:	pointer to logical block around which we want to determine hole
2312 *
2313 * Determine hole length (and start if easily possible) around given logical
2314 * block. We don't try too hard to find the beginning of the hole but @path
2315 * actually points to extent before @lblk, we provide it.
2316 *
2317 * The function returns the length of a hole starting at @lblk. We update @lblk
2318 * to the beginning of the hole if we managed to find it.
2319 */
2320static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2321					   struct ext4_ext_path *path,
2322					   ext4_lblk_t *lblk)
2323{
2324	int depth = ext_depth(inode);
2325	struct ext4_extent *ex;
2326	ext4_lblk_t len;
2327
2328	ex = path[depth].p_ext;
2329	if (ex == NULL) {
2330		/* there is no extent yet, so gap is [0;-] */
2331		*lblk = 0;
2332		len = EXT_MAX_BLOCKS;
2333	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2334		len = le32_to_cpu(ex->ee_block) - *lblk;
2335	} else if (*lblk >= le32_to_cpu(ex->ee_block)
2336			+ ext4_ext_get_actual_len(ex)) {
2337		ext4_lblk_t next;
2338
2339		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2340		next = ext4_ext_next_allocated_block(path);
2341		BUG_ON(next == *lblk);
2342		len = next - *lblk;
2343	} else {
2344		BUG();
2345	}
2346	return len;
2347}
2348
2349/*
2350 * ext4_ext_put_gap_in_cache:
2351 * calculate boundaries of the gap that the requested block fits into
2352 * and cache this gap
2353 */
2354static void
2355ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2356			  ext4_lblk_t hole_len)
2357{
2358	struct extent_status es;
2359
2360	ext4_es_find_delayed_extent_range(inode, hole_start,
2361					  hole_start + hole_len - 1, &es);
2362	if (es.es_len) {
2363		/* There's delayed extent containing lblock? */
2364		if (es.es_lblk <= hole_start)
2365			return;
2366		hole_len = min(es.es_lblk - hole_start, hole_len);
2367	}
2368	ext_debug(" -> %u:%u\n", hole_start, hole_len);
2369	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2370			      EXTENT_STATUS_HOLE);
2371}
2372
2373/*
2374 * ext4_ext_rm_idx:
2375 * removes index from the index block.
2376 */
2377static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2378			struct ext4_ext_path *path, int depth)
2379{
2380	int err;
2381	ext4_fsblk_t leaf;
2382
2383	/* free index block */
2384	depth--;
2385	path = path + depth;
2386	leaf = ext4_idx_pblock(path->p_idx);
2387	if (unlikely(path->p_hdr->eh_entries == 0)) {
2388		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2389		return -EFSCORRUPTED;
2390	}
2391	err = ext4_ext_get_access(handle, inode, path);
2392	if (err)
2393		return err;
2394
2395	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2396		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2397		len *= sizeof(struct ext4_extent_idx);
2398		memmove(path->p_idx, path->p_idx + 1, len);
2399	}
2400
2401	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2402	err = ext4_ext_dirty(handle, inode, path);
2403	if (err)
2404		return err;
2405	ext_debug("index is empty, remove it, free block %llu\n", leaf);
2406	trace_ext4_ext_rm_idx(inode, leaf);
2407
2408	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2409			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2410
2411	while (--depth >= 0) {
2412		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2413			break;
2414		path--;
2415		err = ext4_ext_get_access(handle, inode, path);
2416		if (err)
2417			break;
2418		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2419		err = ext4_ext_dirty(handle, inode, path);
2420		if (err)
2421			break;
2422	}
2423	return err;
2424}
2425
2426/*
2427 * ext4_ext_calc_credits_for_single_extent:
2428 * This routine returns max. credits that needed to insert an extent
2429 * to the extent tree.
2430 * When pass the actual path, the caller should calculate credits
2431 * under i_data_sem.
2432 */
2433int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2434						struct ext4_ext_path *path)
2435{
2436	if (path) {
2437		int depth = ext_depth(inode);
2438		int ret = 0;
2439
2440		/* probably there is space in leaf? */
2441		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2442				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2443
2444			/*
2445			 *  There are some space in the leaf tree, no
2446			 *  need to account for leaf block credit
2447			 *
2448			 *  bitmaps and block group descriptor blocks
2449			 *  and other metadata blocks still need to be
2450			 *  accounted.
2451			 */
2452			/* 1 bitmap, 1 block group descriptor */
2453			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2454			return ret;
2455		}
2456	}
2457
2458	return ext4_chunk_trans_blocks(inode, nrblocks);
2459}
2460
2461/*
2462 * How many index/leaf blocks need to change/allocate to add @extents extents?
2463 *
2464 * If we add a single extent, then in the worse case, each tree level
2465 * index/leaf need to be changed in case of the tree split.
2466 *
2467 * If more extents are inserted, they could cause the whole tree split more
2468 * than once, but this is really rare.
2469 */
2470int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2471{
2472	int index;
2473	int depth;
2474
2475	/* If we are converting the inline data, only one is needed here. */
2476	if (ext4_has_inline_data(inode))
2477		return 1;
2478
2479	depth = ext_depth(inode);
2480
2481	if (extents <= 1)
2482		index = depth * 2;
2483	else
2484		index = depth * 3;
2485
2486	return index;
2487}
2488
2489static inline int get_default_free_blocks_flags(struct inode *inode)
2490{
2491	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2492		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2493	else if (ext4_should_journal_data(inode))
2494		return EXT4_FREE_BLOCKS_FORGET;
2495	return 0;
2496}
2497
2498static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2499			      struct ext4_extent *ex,
2500			      long long *partial_cluster,
2501			      ext4_lblk_t from, ext4_lblk_t to)
2502{
2503	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2504	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2505	ext4_fsblk_t pblk;
2506	int flags = get_default_free_blocks_flags(inode);
2507
2508	/*
2509	 * For bigalloc file systems, we never free a partial cluster
2510	 * at the beginning of the extent.  Instead, we make a note
2511	 * that we tried freeing the cluster, and check to see if we
2512	 * need to free it on a subsequent call to ext4_remove_blocks,
2513	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2514	 */
2515	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2516
2517	trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2518	/*
2519	 * If we have a partial cluster, and it's different from the
2520	 * cluster of the last block, we need to explicitly free the
2521	 * partial cluster here.
2522	 */
2523	pblk = ext4_ext_pblock(ex) + ee_len - 1;
2524	if (*partial_cluster > 0 &&
2525	    *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2526		ext4_free_blocks(handle, inode, NULL,
2527				 EXT4_C2B(sbi, *partial_cluster),
2528				 sbi->s_cluster_ratio, flags);
2529		*partial_cluster = 0;
2530	}
2531
2532#ifdef EXTENTS_STATS
2533	{
2534		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2535		spin_lock(&sbi->s_ext_stats_lock);
2536		sbi->s_ext_blocks += ee_len;
2537		sbi->s_ext_extents++;
2538		if (ee_len < sbi->s_ext_min)
2539			sbi->s_ext_min = ee_len;
2540		if (ee_len > sbi->s_ext_max)
2541			sbi->s_ext_max = ee_len;
2542		if (ext_depth(inode) > sbi->s_depth_max)
2543			sbi->s_depth_max = ext_depth(inode);
2544		spin_unlock(&sbi->s_ext_stats_lock);
2545	}
2546#endif
2547	if (from >= le32_to_cpu(ex->ee_block)
2548	    && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2549		/* tail removal */
2550		ext4_lblk_t num;
2551		long long first_cluster;
2552
2553		num = le32_to_cpu(ex->ee_block) + ee_len - from;
2554		pblk = ext4_ext_pblock(ex) + ee_len - num;
2555		/*
2556		 * Usually we want to free partial cluster at the end of the
2557		 * extent, except for the situation when the cluster is still
2558		 * used by any other extent (partial_cluster is negative).
2559		 */
2560		if (*partial_cluster < 0 &&
2561		    *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2562			flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2563
2564		ext_debug("free last %u blocks starting %llu partial %lld\n",
2565			  num, pblk, *partial_cluster);
2566		ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2567		/*
2568		 * If the block range to be freed didn't start at the
2569		 * beginning of a cluster, and we removed the entire
2570		 * extent and the cluster is not used by any other extent,
2571		 * save the partial cluster here, since we might need to
2572		 * delete if we determine that the truncate or punch hole
2573		 * operation has removed all of the blocks in the cluster.
2574		 * If that cluster is used by another extent, preserve its
2575		 * negative value so it isn't freed later on.
2576		 *
2577		 * If the whole extent wasn't freed, we've reached the
2578		 * start of the truncated/punched region and have finished
2579		 * removing blocks.  If there's a partial cluster here it's
2580		 * shared with the remainder of the extent and is no longer
2581		 * a candidate for removal.
2582		 */
2583		if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2584			first_cluster = (long long) EXT4_B2C(sbi, pblk);
2585			if (first_cluster != -*partial_cluster)
2586				*partial_cluster = first_cluster;
2587		} else {
2588			*partial_cluster = 0;
2589		}
2590	} else
2591		ext4_error(sbi->s_sb, "strange request: removal(2) "
2592			   "%u-%u from %u:%u",
2593			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2594	return 0;
2595}
2596
2597
2598/*
2599 * ext4_ext_rm_leaf() Removes the extents associated with the
2600 * blocks appearing between "start" and "end".  Both "start"
2601 * and "end" must appear in the same extent or EIO is returned.
2602 *
2603 * @handle: The journal handle
2604 * @inode:  The files inode
2605 * @path:   The path to the leaf
2606 * @partial_cluster: The cluster which we'll have to free if all extents
2607 *                   has been released from it.  However, if this value is
2608 *                   negative, it's a cluster just to the right of the
2609 *                   punched region and it must not be freed.
2610 * @start:  The first block to remove
2611 * @end:   The last block to remove
2612 */
2613static int
2614ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2615		 struct ext4_ext_path *path,
2616		 long long *partial_cluster,
2617		 ext4_lblk_t start, ext4_lblk_t end)
2618{
2619	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2620	int err = 0, correct_index = 0;
2621	int depth = ext_depth(inode), credits;
2622	struct ext4_extent_header *eh;
2623	ext4_lblk_t a, b;
2624	unsigned num;
2625	ext4_lblk_t ex_ee_block;
2626	unsigned short ex_ee_len;
2627	unsigned unwritten = 0;
2628	struct ext4_extent *ex;
2629	ext4_fsblk_t pblk;
2630
2631	/* the header must be checked already in ext4_ext_remove_space() */
2632	ext_debug("truncate since %u in leaf to %u\n", start, end);
2633	if (!path[depth].p_hdr)
2634		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2635	eh = path[depth].p_hdr;
2636	if (unlikely(path[depth].p_hdr == NULL)) {
2637		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2638		return -EFSCORRUPTED;
2639	}
2640	/* find where to start removing */
2641	ex = path[depth].p_ext;
2642	if (!ex)
2643		ex = EXT_LAST_EXTENT(eh);
2644
2645	ex_ee_block = le32_to_cpu(ex->ee_block);
2646	ex_ee_len = ext4_ext_get_actual_len(ex);
2647
2648	trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2649
2650	while (ex >= EXT_FIRST_EXTENT(eh) &&
2651			ex_ee_block + ex_ee_len > start) {
2652
2653		if (ext4_ext_is_unwritten(ex))
2654			unwritten = 1;
2655		else
2656			unwritten = 0;
2657
2658		ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2659			  unwritten, ex_ee_len);
2660		path[depth].p_ext = ex;
2661
2662		a = ex_ee_block > start ? ex_ee_block : start;
2663		b = ex_ee_block+ex_ee_len - 1 < end ?
2664			ex_ee_block+ex_ee_len - 1 : end;
2665
2666		ext_debug("  border %u:%u\n", a, b);
2667
2668		/* If this extent is beyond the end of the hole, skip it */
2669		if (end < ex_ee_block) {
2670			/*
2671			 * We're going to skip this extent and move to another,
2672			 * so note that its first cluster is in use to avoid
2673			 * freeing it when removing blocks.  Eventually, the
2674			 * right edge of the truncated/punched region will
2675			 * be just to the left.
2676			 */
2677			if (sbi->s_cluster_ratio > 1) {
2678				pblk = ext4_ext_pblock(ex);
2679				*partial_cluster =
2680					-(long long) EXT4_B2C(sbi, pblk);
2681			}
2682			ex--;
2683			ex_ee_block = le32_to_cpu(ex->ee_block);
2684			ex_ee_len = ext4_ext_get_actual_len(ex);
2685			continue;
2686		} else if (b != ex_ee_block + ex_ee_len - 1) {
2687			EXT4_ERROR_INODE(inode,
2688					 "can not handle truncate %u:%u "
2689					 "on extent %u:%u",
2690					 start, end, ex_ee_block,
2691					 ex_ee_block + ex_ee_len - 1);
2692			err = -EFSCORRUPTED;
2693			goto out;
2694		} else if (a != ex_ee_block) {
2695			/* remove tail of the extent */
2696			num = a - ex_ee_block;
2697		} else {
2698			/* remove whole extent: excellent! */
2699			num = 0;
2700		}
2701		/*
2702		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2703		 * descriptor) for each block group; assume two block
2704		 * groups plus ex_ee_len/blocks_per_block_group for
2705		 * the worst case
2706		 */
2707		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2708		if (ex == EXT_FIRST_EXTENT(eh)) {
2709			correct_index = 1;
2710			credits += (ext_depth(inode)) + 1;
2711		}
2712		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2713
2714		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2715		if (err)
2716			goto out;
2717
2718		err = ext4_ext_get_access(handle, inode, path + depth);
2719		if (err)
2720			goto out;
2721
2722		err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2723					 a, b);
2724		if (err)
2725			goto out;
2726
2727		if (num == 0)
2728			/* this extent is removed; mark slot entirely unused */
2729			ext4_ext_store_pblock(ex, 0);
2730
2731		ex->ee_len = cpu_to_le16(num);
2732		/*
2733		 * Do not mark unwritten if all the blocks in the
2734		 * extent have been removed.
2735		 */
2736		if (unwritten && num)
2737			ext4_ext_mark_unwritten(ex);
2738		/*
2739		 * If the extent was completely released,
2740		 * we need to remove it from the leaf
2741		 */
2742		if (num == 0) {
2743			if (end != EXT_MAX_BLOCKS - 1) {
2744				/*
2745				 * For hole punching, we need to scoot all the
2746				 * extents up when an extent is removed so that
2747				 * we dont have blank extents in the middle
2748				 */
2749				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2750					sizeof(struct ext4_extent));
2751
2752				/* Now get rid of the one at the end */
2753				memset(EXT_LAST_EXTENT(eh), 0,
2754					sizeof(struct ext4_extent));
2755			}
2756			le16_add_cpu(&eh->eh_entries, -1);
2757		}
2758
2759		err = ext4_ext_dirty(handle, inode, path + depth);
2760		if (err)
2761			goto out;
2762
2763		ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2764				ext4_ext_pblock(ex));
2765		ex--;
2766		ex_ee_block = le32_to_cpu(ex->ee_block);
2767		ex_ee_len = ext4_ext_get_actual_len(ex);
2768	}
2769
2770	if (correct_index && eh->eh_entries)
2771		err = ext4_ext_correct_indexes(handle, inode, path);
2772
2773	/*
2774	 * If there's a partial cluster and at least one extent remains in
2775	 * the leaf, free the partial cluster if it isn't shared with the
2776	 * current extent.  If it is shared with the current extent
2777	 * we zero partial_cluster because we've reached the start of the
2778	 * truncated/punched region and we're done removing blocks.
2779	 */
2780	if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2781		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2782		if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2783			ext4_free_blocks(handle, inode, NULL,
2784					 EXT4_C2B(sbi, *partial_cluster),
2785					 sbi->s_cluster_ratio,
2786					 get_default_free_blocks_flags(inode));
2787		}
2788		*partial_cluster = 0;
2789	}
2790
2791	/* if this leaf is free, then we should
2792	 * remove it from index block above */
2793	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2794		err = ext4_ext_rm_idx(handle, inode, path, depth);
2795
2796out:
2797	return err;
2798}
2799
2800/*
2801 * ext4_ext_more_to_rm:
2802 * returns 1 if current index has to be freed (even partial)
2803 */
2804static int
2805ext4_ext_more_to_rm(struct ext4_ext_path *path)
2806{
2807	BUG_ON(path->p_idx == NULL);
2808
2809	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2810		return 0;
2811
2812	/*
2813	 * if truncate on deeper level happened, it wasn't partial,
2814	 * so we have to consider current index for truncation
2815	 */
2816	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2817		return 0;
2818	return 1;
2819}
2820
2821int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2822			  ext4_lblk_t end)
2823{
2824	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2825	int depth = ext_depth(inode);
2826	struct ext4_ext_path *path = NULL;
2827	long long partial_cluster = 0;
2828	handle_t *handle;
2829	int i = 0, err = 0;
2830
2831	ext_debug("truncate since %u to %u\n", start, end);
2832
2833	/* probably first extent we're gonna free will be last in block */
2834	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2835	if (IS_ERR(handle))
2836		return PTR_ERR(handle);
2837
2838again:
2839	trace_ext4_ext_remove_space(inode, start, end, depth);
2840
2841	/*
2842	 * Check if we are removing extents inside the extent tree. If that
2843	 * is the case, we are going to punch a hole inside the extent tree
2844	 * so we have to check whether we need to split the extent covering
2845	 * the last block to remove so we can easily remove the part of it
2846	 * in ext4_ext_rm_leaf().
2847	 */
2848	if (end < EXT_MAX_BLOCKS - 1) {
2849		struct ext4_extent *ex;
2850		ext4_lblk_t ee_block, ex_end, lblk;
2851		ext4_fsblk_t pblk;
2852
2853		/* find extent for or closest extent to this block */
2854		path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2855		if (IS_ERR(path)) {
2856			ext4_journal_stop(handle);
2857			return PTR_ERR(path);
2858		}
2859		depth = ext_depth(inode);
2860		/* Leaf not may not exist only if inode has no blocks at all */
2861		ex = path[depth].p_ext;
2862		if (!ex) {
2863			if (depth) {
2864				EXT4_ERROR_INODE(inode,
2865						 "path[%d].p_hdr == NULL",
2866						 depth);
2867				err = -EFSCORRUPTED;
2868			}
2869			goto out;
2870		}
2871
2872		ee_block = le32_to_cpu(ex->ee_block);
2873		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2874
2875		/*
2876		 * See if the last block is inside the extent, if so split
2877		 * the extent at 'end' block so we can easily remove the
2878		 * tail of the first part of the split extent in
2879		 * ext4_ext_rm_leaf().
2880		 */
2881		if (end >= ee_block && end < ex_end) {
2882
2883			/*
2884			 * If we're going to split the extent, note that
2885			 * the cluster containing the block after 'end' is
2886			 * in use to avoid freeing it when removing blocks.
2887			 */
2888			if (sbi->s_cluster_ratio > 1) {
2889				pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2890				partial_cluster =
2891					-(long long) EXT4_B2C(sbi, pblk);
2892			}
2893
2894			/*
2895			 * Split the extent in two so that 'end' is the last
2896			 * block in the first new extent. Also we should not
2897			 * fail removing space due to ENOSPC so try to use
2898			 * reserved block if that happens.
2899			 */
2900			err = ext4_force_split_extent_at(handle, inode, &path,
2901							 end + 1, 1);
2902			if (err < 0)
2903				goto out;
2904
2905		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2906			/*
2907			 * If there's an extent to the right its first cluster
2908			 * contains the immediate right boundary of the
2909			 * truncated/punched region.  Set partial_cluster to
2910			 * its negative value so it won't be freed if shared
2911			 * with the current extent.  The end < ee_block case
2912			 * is handled in ext4_ext_rm_leaf().
2913			 */
2914			lblk = ex_end + 1;
2915			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2916						    &ex);
2917			if (err)
2918				goto out;
2919			if (pblk)
2920				partial_cluster =
2921					-(long long) EXT4_B2C(sbi, pblk);
2922		}
2923	}
2924	/*
2925	 * We start scanning from right side, freeing all the blocks
2926	 * after i_size and walking into the tree depth-wise.
2927	 */
2928	depth = ext_depth(inode);
2929	if (path) {
2930		int k = i = depth;
2931		while (--k > 0)
2932			path[k].p_block =
2933				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2934	} else {
2935		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2936			       GFP_NOFS);
2937		if (path == NULL) {
2938			ext4_journal_stop(handle);
2939			return -ENOMEM;
2940		}
2941		path[0].p_maxdepth = path[0].p_depth = depth;
2942		path[0].p_hdr = ext_inode_hdr(inode);
2943		i = 0;
2944
2945		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2946			err = -EFSCORRUPTED;
2947			goto out;
2948		}
2949	}
2950	err = 0;
2951
2952	while (i >= 0 && err == 0) {
2953		if (i == depth) {
2954			/* this is leaf block */
2955			err = ext4_ext_rm_leaf(handle, inode, path,
2956					       &partial_cluster, start,
2957					       end);
2958			/* root level has p_bh == NULL, brelse() eats this */
2959			brelse(path[i].p_bh);
2960			path[i].p_bh = NULL;
2961			i--;
2962			continue;
2963		}
2964
2965		/* this is index block */
2966		if (!path[i].p_hdr) {
2967			ext_debug("initialize header\n");
2968			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2969		}
2970
2971		if (!path[i].p_idx) {
2972			/* this level hasn't been touched yet */
2973			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2974			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2975			ext_debug("init index ptr: hdr 0x%p, num %d\n",
2976				  path[i].p_hdr,
2977				  le16_to_cpu(path[i].p_hdr->eh_entries));
2978		} else {
2979			/* we were already here, see at next index */
2980			path[i].p_idx--;
2981		}
2982
2983		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2984				i, EXT_FIRST_INDEX(path[i].p_hdr),
2985				path[i].p_idx);
2986		if (ext4_ext_more_to_rm(path + i)) {
2987			struct buffer_head *bh;
2988			/* go to the next level */
2989			ext_debug("move to level %d (block %llu)\n",
2990				  i + 1, ext4_idx_pblock(path[i].p_idx));
2991			memset(path + i + 1, 0, sizeof(*path));
2992			bh = read_extent_tree_block(inode,
2993				ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2994				EXT4_EX_NOCACHE);
2995			if (IS_ERR(bh)) {
2996				/* should we reset i_size? */
2997				err = PTR_ERR(bh);
2998				break;
2999			}
3000			/* Yield here to deal with large extent trees.
3001			 * Should be a no-op if we did IO above. */
3002			cond_resched();
3003			if (WARN_ON(i + 1 > depth)) {
3004				err = -EFSCORRUPTED;
3005				break;
3006			}
3007			path[i + 1].p_bh = bh;
3008
3009			/* save actual number of indexes since this
3010			 * number is changed at the next iteration */
3011			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3012			i++;
3013		} else {
3014			/* we finished processing this index, go up */
3015			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3016				/* index is empty, remove it;
3017				 * handle must be already prepared by the
3018				 * truncatei_leaf() */
3019				err = ext4_ext_rm_idx(handle, inode, path, i);
3020			}
3021			/* root level has p_bh == NULL, brelse() eats this */
3022			brelse(path[i].p_bh);
3023			path[i].p_bh = NULL;
3024			i--;
3025			ext_debug("return to level %d\n", i);
3026		}
3027	}
3028
3029	trace_ext4_ext_remove_space_done(inode, start, end, depth,
3030			partial_cluster, path->p_hdr->eh_entries);
3031
3032	/*
3033	 * If we still have something in the partial cluster and we have removed
3034	 * even the first extent, then we should free the blocks in the partial
3035	 * cluster as well.  (This code will only run when there are no leaves
3036	 * to the immediate left of the truncated/punched region.)
3037	 */
3038	if (partial_cluster > 0 && err == 0) {
3039		/* don't zero partial_cluster since it's not used afterwards */
3040		ext4_free_blocks(handle, inode, NULL,
3041				 EXT4_C2B(sbi, partial_cluster),
3042				 sbi->s_cluster_ratio,
3043				 get_default_free_blocks_flags(inode));
3044	}
3045
3046	/* TODO: flexible tree reduction should be here */
3047	if (path->p_hdr->eh_entries == 0) {
3048		/*
3049		 * truncate to zero freed all the tree,
3050		 * so we need to correct eh_depth
3051		 */
3052		err = ext4_ext_get_access(handle, inode, path);
3053		if (err == 0) {
3054			ext_inode_hdr(inode)->eh_depth = 0;
3055			ext_inode_hdr(inode)->eh_max =
3056				cpu_to_le16(ext4_ext_space_root(inode, 0));
3057			err = ext4_ext_dirty(handle, inode, path);
3058		}
3059	}
3060out:
3061	ext4_ext_drop_refs(path);
3062	kfree(path);
3063	path = NULL;
3064	if (err == -EAGAIN)
3065		goto again;
3066	ext4_journal_stop(handle);
3067
3068	return err;
3069}
3070
3071/*
3072 * called at mount time
3073 */
3074void ext4_ext_init(struct super_block *sb)
3075{
3076	/*
3077	 * possible initialization would be here
3078	 */
3079
3080	if (ext4_has_feature_extents(sb)) {
3081#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3082		printk(KERN_INFO "EXT4-fs: file extents enabled"
3083#ifdef AGGRESSIVE_TEST
3084		       ", aggressive tests"
3085#endif
3086#ifdef CHECK_BINSEARCH
3087		       ", check binsearch"
3088#endif
3089#ifdef EXTENTS_STATS
3090		       ", stats"
3091#endif
3092		       "\n");
3093#endif
3094#ifdef EXTENTS_STATS
3095		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3096		EXT4_SB(sb)->s_ext_min = 1 << 30;
3097		EXT4_SB(sb)->s_ext_max = 0;
3098#endif
3099	}
3100}
3101
3102/*
3103 * called at umount time
3104 */
3105void ext4_ext_release(struct super_block *sb)
3106{
3107	if (!ext4_has_feature_extents(sb))
3108		return;
3109
3110#ifdef EXTENTS_STATS
3111	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3112		struct ext4_sb_info *sbi = EXT4_SB(sb);
3113		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3114			sbi->s_ext_blocks, sbi->s_ext_extents,
3115			sbi->s_ext_blocks / sbi->s_ext_extents);
3116		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3117			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3118	}
3119#endif
3120}
3121
3122static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3123{
3124	ext4_lblk_t  ee_block;
3125	ext4_fsblk_t ee_pblock;
3126	unsigned int ee_len;
3127
3128	ee_block  = le32_to_cpu(ex->ee_block);
3129	ee_len    = ext4_ext_get_actual_len(ex);
3130	ee_pblock = ext4_ext_pblock(ex);
3131
3132	if (ee_len == 0)
3133		return 0;
3134
3135	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3136				     EXTENT_STATUS_WRITTEN);
3137}
3138
3139/* FIXME!! we need to try to merge to left or right after zero-out  */
3140static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3141{
3142	ext4_fsblk_t ee_pblock;
3143	unsigned int ee_len;
3144
3145	ee_len    = ext4_ext_get_actual_len(ex);
3146	ee_pblock = ext4_ext_pblock(ex);
3147	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3148				  ee_len);
3149}
3150
3151/*
3152 * ext4_split_extent_at() splits an extent at given block.
3153 *
3154 * @handle: the journal handle
3155 * @inode: the file inode
3156 * @path: the path to the extent
3157 * @split: the logical block where the extent is splitted.
3158 * @split_flags: indicates if the extent could be zeroout if split fails, and
3159 *		 the states(init or unwritten) of new extents.
3160 * @flags: flags used to insert new extent to extent tree.
3161 *
3162 *
3163 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3164 * of which are deterimined by split_flag.
3165 *
3166 * There are two cases:
3167 *  a> the extent are splitted into two extent.
3168 *  b> split is not needed, and just mark the extent.
3169 *
3170 * return 0 on success.
3171 */
3172static int ext4_split_extent_at(handle_t *handle,
3173			     struct inode *inode,
3174			     struct ext4_ext_path **ppath,
3175			     ext4_lblk_t split,
3176			     int split_flag,
3177			     int flags)
3178{
3179	struct ext4_ext_path *path = *ppath;
3180	ext4_fsblk_t newblock;
3181	ext4_lblk_t ee_block;
3182	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3183	struct ext4_extent *ex2 = NULL;
3184	unsigned int ee_len, depth;
3185	int err = 0;
3186
3187	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3188	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3189
3190	ext_debug("ext4_split_extents_at: inode %lu, logical"
3191		"block %llu\n", inode->i_ino, (unsigned long long)split);
3192
3193	ext4_ext_show_leaf(inode, path);
3194
3195	depth = ext_depth(inode);
3196	ex = path[depth].p_ext;
3197	ee_block = le32_to_cpu(ex->ee_block);
3198	ee_len = ext4_ext_get_actual_len(ex);
3199	newblock = split - ee_block + ext4_ext_pblock(ex);
3200
3201	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3202	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3203	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3204			     EXT4_EXT_MARK_UNWRIT1 |
3205			     EXT4_EXT_MARK_UNWRIT2));
3206
3207	err = ext4_ext_get_access(handle, inode, path + depth);
3208	if (err)
3209		goto out;
3210
3211	if (split == ee_block) {
3212		/*
3213		 * case b: block @split is the block that the extent begins with
3214		 * then we just change the state of the extent, and splitting
3215		 * is not needed.
3216		 */
3217		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3218			ext4_ext_mark_unwritten(ex);
3219		else
3220			ext4_ext_mark_initialized(ex);
3221
3222		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3223			ext4_ext_try_to_merge(handle, inode, path, ex);
3224
3225		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3226		goto out;
3227	}
3228
3229	/* case a */
3230	memcpy(&orig_ex, ex, sizeof(orig_ex));
3231	ex->ee_len = cpu_to_le16(split - ee_block);
3232	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3233		ext4_ext_mark_unwritten(ex);
3234
3235	/*
3236	 * path may lead to new leaf, not to original leaf any more
3237	 * after ext4_ext_insert_extent() returns,
3238	 */
3239	err = ext4_ext_dirty(handle, inode, path + depth);
3240	if (err)
3241		goto fix_extent_len;
3242
3243	ex2 = &newex;
3244	ex2->ee_block = cpu_to_le32(split);
3245	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3246	ext4_ext_store_pblock(ex2, newblock);
3247	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3248		ext4_ext_mark_unwritten(ex2);
3249
3250	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3251	if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3252		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3253			if (split_flag & EXT4_EXT_DATA_VALID1) {
3254				err = ext4_ext_zeroout(inode, ex2);
3255				zero_ex.ee_block = ex2->ee_block;
3256				zero_ex.ee_len = cpu_to_le16(
3257						ext4_ext_get_actual_len(ex2));
3258				ext4_ext_store_pblock(&zero_ex,
3259						      ext4_ext_pblock(ex2));
3260			} else {
3261				err = ext4_ext_zeroout(inode, ex);
3262				zero_ex.ee_block = ex->ee_block;
3263				zero_ex.ee_len = cpu_to_le16(
3264						ext4_ext_get_actual_len(ex));
3265				ext4_ext_store_pblock(&zero_ex,
3266						      ext4_ext_pblock(ex));
3267			}
3268		} else {
3269			err = ext4_ext_zeroout(inode, &orig_ex);
3270			zero_ex.ee_block = orig_ex.ee_block;
3271			zero_ex.ee_len = cpu_to_le16(
3272						ext4_ext_get_actual_len(&orig_ex));
3273			ext4_ext_store_pblock(&zero_ex,
3274					      ext4_ext_pblock(&orig_ex));
3275		}
3276
3277		if (err)
3278			goto fix_extent_len;
3279		/* update the extent length and mark as initialized */
3280		ex->ee_len = cpu_to_le16(ee_len);
3281		ext4_ext_try_to_merge(handle, inode, path, ex);
3282		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3283		if (err)
3284			goto fix_extent_len;
3285
3286		/* update extent status tree */
3287		err = ext4_zeroout_es(inode, &zero_ex);
3288
3289		goto out;
3290	} else if (err)
3291		goto fix_extent_len;
3292
3293out:
3294	ext4_ext_show_leaf(inode, path);
3295	return err;
3296
3297fix_extent_len:
3298	ex->ee_len = orig_ex.ee_len;
3299	ext4_ext_dirty(handle, inode, path + path->p_depth);
3300	return err;
3301}
3302
3303/*
3304 * ext4_split_extents() splits an extent and mark extent which is covered
3305 * by @map as split_flags indicates
3306 *
3307 * It may result in splitting the extent into multiple extents (up to three)
3308 * There are three possibilities:
3309 *   a> There is no split required
3310 *   b> Splits in two extents: Split is happening at either end of the extent
3311 *   c> Splits in three extents: Somone is splitting in middle of the extent
3312 *
3313 */
3314static int ext4_split_extent(handle_t *handle,
3315			      struct inode *inode,
3316			      struct ext4_ext_path **ppath,
3317			      struct ext4_map_blocks *map,
3318			      int split_flag,
3319			      int flags)
3320{
3321	struct ext4_ext_path *path = *ppath;
3322	ext4_lblk_t ee_block;
3323	struct ext4_extent *ex;
3324	unsigned int ee_len, depth;
3325	int err = 0;
3326	int unwritten;
3327	int split_flag1, flags1;
3328	int allocated = map->m_len;
3329
3330	depth = ext_depth(inode);
3331	ex = path[depth].p_ext;
3332	ee_block = le32_to_cpu(ex->ee_block);
3333	ee_len = ext4_ext_get_actual_len(ex);
3334	unwritten = ext4_ext_is_unwritten(ex);
3335
3336	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3337		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3338		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3339		if (unwritten)
3340			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3341				       EXT4_EXT_MARK_UNWRIT2;
3342		if (split_flag & EXT4_EXT_DATA_VALID2)
3343			split_flag1 |= EXT4_EXT_DATA_VALID1;
3344		err = ext4_split_extent_at(handle, inode, ppath,
3345				map->m_lblk + map->m_len, split_flag1, flags1);
3346		if (err)
3347			goto out;
3348	} else {
3349		allocated = ee_len - (map->m_lblk - ee_block);
3350	}
3351	/*
3352	 * Update path is required because previous ext4_split_extent_at() may
3353	 * result in split of original leaf or extent zeroout.
3354	 */
3355	path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3356	if (IS_ERR(path))
3357		return PTR_ERR(path);
3358	depth = ext_depth(inode);
3359	ex = path[depth].p_ext;
3360	if (!ex) {
3361		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3362				 (unsigned long) map->m_lblk);
3363		return -EFSCORRUPTED;
3364	}
3365	unwritten = ext4_ext_is_unwritten(ex);
3366	split_flag1 = 0;
3367
3368	if (map->m_lblk >= ee_block) {
3369		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3370		if (unwritten) {
3371			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3372			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3373						     EXT4_EXT_MARK_UNWRIT2);
3374		}
3375		err = ext4_split_extent_at(handle, inode, ppath,
3376				map->m_lblk, split_flag1, flags);
3377		if (err)
3378			goto out;
3379	}
3380
3381	ext4_ext_show_leaf(inode, path);
3382out:
3383	return err ? err : allocated;
3384}
3385
3386/*
3387 * This function is called by ext4_ext_map_blocks() if someone tries to write
3388 * to an unwritten extent. It may result in splitting the unwritten
3389 * extent into multiple extents (up to three - one initialized and two
3390 * unwritten).
3391 * There are three possibilities:
3392 *   a> There is no split required: Entire extent should be initialized
3393 *   b> Splits in two extents: Write is happening at either end of the extent
3394 *   c> Splits in three extents: Somone is writing in middle of the extent
3395 *
3396 * Pre-conditions:
3397 *  - The extent pointed to by 'path' is unwritten.
3398 *  - The extent pointed to by 'path' contains a superset
3399 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3400 *
3401 * Post-conditions on success:
3402 *  - the returned value is the number of blocks beyond map->l_lblk
3403 *    that are allocated and initialized.
3404 *    It is guaranteed to be >= map->m_len.
3405 */
3406static int ext4_ext_convert_to_initialized(handle_t *handle,
3407					   struct inode *inode,
3408					   struct ext4_map_blocks *map,
3409					   struct ext4_ext_path **ppath,
3410					   int flags)
3411{
3412	struct ext4_ext_path *path = *ppath;
3413	struct ext4_sb_info *sbi;
3414	struct ext4_extent_header *eh;
3415	struct ext4_map_blocks split_map;
3416	struct ext4_extent zero_ex;
3417	struct ext4_extent *ex, *abut_ex;
3418	ext4_lblk_t ee_block, eof_block;
3419	unsigned int ee_len, depth, map_len = map->m_len;
3420	int allocated = 0, max_zeroout = 0;
3421	int err = 0;
3422	int split_flag = 0;
3423
3424	ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3425		"block %llu, max_blocks %u\n", inode->i_ino,
3426		(unsigned long long)map->m_lblk, map_len);
3427
3428	sbi = EXT4_SB(inode->i_sb);
3429	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3430		inode->i_sb->s_blocksize_bits;
3431	if (eof_block < map->m_lblk + map_len)
3432		eof_block = map->m_lblk + map_len;
3433
3434	depth = ext_depth(inode);
3435	eh = path[depth].p_hdr;
3436	ex = path[depth].p_ext;
3437	ee_block = le32_to_cpu(ex->ee_block);
3438	ee_len = ext4_ext_get_actual_len(ex);
3439	zero_ex.ee_len = 0;
3440
3441	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3442
3443	/* Pre-conditions */
3444	BUG_ON(!ext4_ext_is_unwritten(ex));
3445	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3446
3447	/*
3448	 * Attempt to transfer newly initialized blocks from the currently
3449	 * unwritten extent to its neighbor. This is much cheaper
3450	 * than an insertion followed by a merge as those involve costly
3451	 * memmove() calls. Transferring to the left is the common case in
3452	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3453	 * followed by append writes.
3454	 *
3455	 * Limitations of the current logic:
3456	 *  - L1: we do not deal with writes covering the whole extent.
3457	 *    This would require removing the extent if the transfer
3458	 *    is possible.
3459	 *  - L2: we only attempt to merge with an extent stored in the
3460	 *    same extent tree node.
3461	 */
3462	if ((map->m_lblk == ee_block) &&
3463		/* See if we can merge left */
3464		(map_len < ee_len) &&		/*L1*/
3465		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3466		ext4_lblk_t prev_lblk;
3467		ext4_fsblk_t prev_pblk, ee_pblk;
3468		unsigned int prev_len;
3469
3470		abut_ex = ex - 1;
3471		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3472		prev_len = ext4_ext_get_actual_len(abut_ex);
3473		prev_pblk = ext4_ext_pblock(abut_ex);
3474		ee_pblk = ext4_ext_pblock(ex);
3475
3476		/*
3477		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3478		 * upon those conditions:
3479		 * - C1: abut_ex is initialized,
3480		 * - C2: abut_ex is logically abutting ex,
3481		 * - C3: abut_ex is physically abutting ex,
3482		 * - C4: abut_ex can receive the additional blocks without
3483		 *   overflowing the (initialized) length limit.
3484		 */
3485		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3486			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3487			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3488			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3489			err = ext4_ext_get_access(handle, inode, path + depth);
3490			if (err)
3491				goto out;
3492
3493			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3494				map, ex, abut_ex);
3495
3496			/* Shift the start of ex by 'map_len' blocks */
3497			ex->ee_block = cpu_to_le32(ee_block + map_len);
3498			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3499			ex->ee_len = cpu_to_le16(ee_len - map_len);
3500			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3501
3502			/* Extend abut_ex by 'map_len' blocks */
3503			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3504
3505			/* Result: number of initialized blocks past m_lblk */
3506			allocated = map_len;
3507		}
3508	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3509		   (map_len < ee_len) &&	/*L1*/
3510		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3511		/* See if we can merge right */
3512		ext4_lblk_t next_lblk;
3513		ext4_fsblk_t next_pblk, ee_pblk;
3514		unsigned int next_len;
3515
3516		abut_ex = ex + 1;
3517		next_lblk = le32_to_cpu(abut_ex->ee_block);
3518		next_len = ext4_ext_get_actual_len(abut_ex);
3519		next_pblk = ext4_ext_pblock(abut_ex);
3520		ee_pblk = ext4_ext_pblock(ex);
3521
3522		/*
3523		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3524		 * upon those conditions:
3525		 * - C1: abut_ex is initialized,
3526		 * - C2: abut_ex is logically abutting ex,
3527		 * - C3: abut_ex is physically abutting ex,
3528		 * - C4: abut_ex can receive the additional blocks without
3529		 *   overflowing the (initialized) length limit.
3530		 */
3531		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3532		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3533		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3534		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3535			err = ext4_ext_get_access(handle, inode, path + depth);
3536			if (err)
3537				goto out;
3538
3539			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3540				map, ex, abut_ex);
3541
3542			/* Shift the start of abut_ex by 'map_len' blocks */
3543			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3544			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3545			ex->ee_len = cpu_to_le16(ee_len - map_len);
3546			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3547
3548			/* Extend abut_ex by 'map_len' blocks */
3549			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3550
3551			/* Result: number of initialized blocks past m_lblk */
3552			allocated = map_len;
3553		}
3554	}
3555	if (allocated) {
3556		/* Mark the block containing both extents as dirty */
3557		ext4_ext_dirty(handle, inode, path + depth);
3558
3559		/* Update path to point to the right extent */
3560		path[depth].p_ext = abut_ex;
3561		goto out;
3562	} else
3563		allocated = ee_len - (map->m_lblk - ee_block);
3564
3565	WARN_ON(map->m_lblk < ee_block);
3566	/*
3567	 * It is safe to convert extent to initialized via explicit
3568	 * zeroout only if extent is fully inside i_size or new_size.
3569	 */
3570	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3571
3572	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3573		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3574			(inode->i_sb->s_blocksize_bits - 10);
3575
3576	if (ext4_encrypted_inode(inode))
3577		max_zeroout = 0;
3578
3579	/* If extent is less than s_max_zeroout_kb, zeroout directly */
3580	if (max_zeroout && (ee_len <= max_zeroout)) {
3581		err = ext4_ext_zeroout(inode, ex);
3582		if (err)
3583			goto out;
3584		zero_ex.ee_block = ex->ee_block;
3585		zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3586		ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3587
3588		err = ext4_ext_get_access(handle, inode, path + depth);
3589		if (err)
3590			goto out;
3591		ext4_ext_mark_initialized(ex);
3592		ext4_ext_try_to_merge(handle, inode, path, ex);
3593		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3594		goto out;
3595	}
3596
3597	/*
3598	 * four cases:
3599	 * 1. split the extent into three extents.
3600	 * 2. split the extent into two extents, zeroout the first half.
3601	 * 3. split the extent into two extents, zeroout the second half.
3602	 * 4. split the extent into two extents with out zeroout.
3603	 */
3604	split_map.m_lblk = map->m_lblk;
3605	split_map.m_len = map->m_len;
3606
3607	if (max_zeroout && (allocated > map->m_len)) {
3608		if (allocated <= max_zeroout) {
3609			/* case 3 */
3610			zero_ex.ee_block =
3611					 cpu_to_le32(map->m_lblk);
3612			zero_ex.ee_len = cpu_to_le16(allocated);
3613			ext4_ext_store_pblock(&zero_ex,
3614				ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3615			err = ext4_ext_zeroout(inode, &zero_ex);
3616			if (err)
3617				goto out;
3618			split_map.m_lblk = map->m_lblk;
3619			split_map.m_len = allocated;
3620		} else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3621			/* case 2 */
3622			if (map->m_lblk != ee_block) {
3623				zero_ex.ee_block = ex->ee_block;
3624				zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3625							ee_block);
3626				ext4_ext_store_pblock(&zero_ex,
3627						      ext4_ext_pblock(ex));
3628				err = ext4_ext_zeroout(inode, &zero_ex);
3629				if (err)
3630					goto out;
3631			}
3632
3633			split_map.m_lblk = ee_block;
3634			split_map.m_len = map->m_lblk - ee_block + map->m_len;
3635			allocated = map->m_len;
3636		}
3637	}
3638
3639	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3640				flags);
3641	if (err > 0)
3642		err = 0;
3643out:
3644	/* If we have gotten a failure, don't zero out status tree */
3645	if (!err)
3646		err = ext4_zeroout_es(inode, &zero_ex);
3647	return err ? err : allocated;
3648}
3649
3650/*
3651 * This function is called by ext4_ext_map_blocks() from
3652 * ext4_get_blocks_dio_write() when DIO to write
3653 * to an unwritten extent.
3654 *
3655 * Writing to an unwritten extent may result in splitting the unwritten
3656 * extent into multiple initialized/unwritten extents (up to three)
3657 * There are three possibilities:
3658 *   a> There is no split required: Entire extent should be unwritten
3659 *   b> Splits in two extents: Write is happening at either end of the extent
3660 *   c> Splits in three extents: Somone is writing in middle of the extent
3661 *
3662 * This works the same way in the case of initialized -> unwritten conversion.
3663 *
3664 * One of more index blocks maybe needed if the extent tree grow after
3665 * the unwritten extent split. To prevent ENOSPC occur at the IO
3666 * complete, we need to split the unwritten extent before DIO submit
3667 * the IO. The unwritten extent called at this time will be split
3668 * into three unwritten extent(at most). After IO complete, the part
3669 * being filled will be convert to initialized by the end_io callback function
3670 * via ext4_convert_unwritten_extents().
3671 *
3672 * Returns the size of unwritten extent to be written on success.
3673 */
3674static int ext4_split_convert_extents(handle_t *handle,
3675					struct inode *inode,
3676					struct ext4_map_blocks *map,
3677					struct ext4_ext_path **ppath,
3678					int flags)
3679{
3680	struct ext4_ext_path *path = *ppath;
3681	ext4_lblk_t eof_block;
3682	ext4_lblk_t ee_block;
3683	struct ext4_extent *ex;
3684	unsigned int ee_len;
3685	int split_flag = 0, depth;
3686
3687	ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3688		  __func__, inode->i_ino,
3689		  (unsigned long long)map->m_lblk, map->m_len);
3690
3691	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3692		inode->i_sb->s_blocksize_bits;
3693	if (eof_block < map->m_lblk + map->m_len)
3694		eof_block = map->m_lblk + map->m_len;
3695	/*
3696	 * It is safe to convert extent to initialized via explicit
3697	 * zeroout only if extent is fully insde i_size or new_size.
3698	 */
3699	depth = ext_depth(inode);
3700	ex = path[depth].p_ext;
3701	ee_block = le32_to_cpu(ex->ee_block);
3702	ee_len = ext4_ext_get_actual_len(ex);
3703
3704	/* Convert to unwritten */
3705	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3706		split_flag |= EXT4_EXT_DATA_VALID1;
3707	/* Convert to initialized */
3708	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3709		split_flag |= ee_block + ee_len <= eof_block ?
3710			      EXT4_EXT_MAY_ZEROOUT : 0;
3711		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3712	}
3713	flags |= EXT4_GET_BLOCKS_PRE_IO;
3714	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3715}
3716
3717static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3718						struct inode *inode,
3719						struct ext4_map_blocks *map,
3720						struct ext4_ext_path **ppath)
3721{
3722	struct ext4_ext_path *path = *ppath;
3723	struct ext4_extent *ex;
3724	ext4_lblk_t ee_block;
3725	unsigned int ee_len;
3726	int depth;
3727	int err = 0;
3728
3729	depth = ext_depth(inode);
3730	ex = path[depth].p_ext;
3731	ee_block = le32_to_cpu(ex->ee_block);
3732	ee_len = ext4_ext_get_actual_len(ex);
3733
3734	ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3735		"block %llu, max_blocks %u\n", inode->i_ino,
3736		  (unsigned long long)ee_block, ee_len);
3737
3738	/* If extent is larger than requested it is a clear sign that we still
3739	 * have some extent state machine issues left. So extent_split is still
3740	 * required.
3741	 * TODO: Once all related issues will be fixed this situation should be
3742	 * illegal.
3743	 */
3744	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3745#ifdef EXT4_DEBUG
3746		ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3747			     " len %u; IO logical block %llu, len %u",
3748			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3749			     (unsigned long long)map->m_lblk, map->m_len);
3750#endif
3751		err = ext4_split_convert_extents(handle, inode, map, ppath,
3752						 EXT4_GET_BLOCKS_CONVERT);
3753		if (err < 0)
3754			return err;
3755		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3756		if (IS_ERR(path))
3757			return PTR_ERR(path);
3758		depth = ext_depth(inode);
3759		ex = path[depth].p_ext;
3760	}
3761
3762	err = ext4_ext_get_access(handle, inode, path + depth);
3763	if (err)
3764		goto out;
3765	/* first mark the extent as initialized */
3766	ext4_ext_mark_initialized(ex);
3767
3768	/* note: ext4_ext_correct_indexes() isn't needed here because
3769	 * borders are not changed
3770	 */
3771	ext4_ext_try_to_merge(handle, inode, path, ex);
3772
3773	/* Mark modified extent as dirty */
3774	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3775out:
3776	ext4_ext_show_leaf(inode, path);
3777	return err;
3778}
3779
 
 
 
 
 
 
 
 
3780/*
3781 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3782 */
3783static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3784			      ext4_lblk_t lblk,
3785			      struct ext4_ext_path *path,
3786			      unsigned int len)
3787{
3788	int i, depth;
3789	struct ext4_extent_header *eh;
3790	struct ext4_extent *last_ex;
3791
3792	if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3793		return 0;
3794
3795	depth = ext_depth(inode);
3796	eh = path[depth].p_hdr;
3797
3798	/*
3799	 * We're going to remove EOFBLOCKS_FL entirely in future so we
3800	 * do not care for this case anymore. Simply remove the flag
3801	 * if there are no extents.
3802	 */
3803	if (unlikely(!eh->eh_entries))
3804		goto out;
3805	last_ex = EXT_LAST_EXTENT(eh);
3806	/*
3807	 * We should clear the EOFBLOCKS_FL flag if we are writing the
3808	 * last block in the last extent in the file.  We test this by
3809	 * first checking to see if the caller to
3810	 * ext4_ext_get_blocks() was interested in the last block (or
3811	 * a block beyond the last block) in the current extent.  If
3812	 * this turns out to be false, we can bail out from this
3813	 * function immediately.
3814	 */
3815	if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3816	    ext4_ext_get_actual_len(last_ex))
3817		return 0;
3818	/*
3819	 * If the caller does appear to be planning to write at or
3820	 * beyond the end of the current extent, we then test to see
3821	 * if the current extent is the last extent in the file, by
3822	 * checking to make sure it was reached via the rightmost node
3823	 * at each level of the tree.
3824	 */
3825	for (i = depth-1; i >= 0; i--)
3826		if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3827			return 0;
3828out:
3829	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3830	return ext4_mark_inode_dirty(handle, inode);
3831}
3832
3833/**
3834 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3835 *
3836 * Return 1 if there is a delalloc block in the range, otherwise 0.
3837 */
3838int ext4_find_delalloc_range(struct inode *inode,
3839			     ext4_lblk_t lblk_start,
3840			     ext4_lblk_t lblk_end)
3841{
3842	struct extent_status es;
3843
3844	ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3845	if (es.es_len == 0)
3846		return 0; /* there is no delay extent in this tree */
3847	else if (es.es_lblk <= lblk_start &&
3848		 lblk_start < es.es_lblk + es.es_len)
3849		return 1;
3850	else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3851		return 1;
3852	else
3853		return 0;
3854}
3855
3856int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3857{
3858	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3859	ext4_lblk_t lblk_start, lblk_end;
3860	lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3861	lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3862
3863	return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3864}
3865
3866/**
3867 * Determines how many complete clusters (out of those specified by the 'map')
3868 * are under delalloc and were reserved quota for.
3869 * This function is called when we are writing out the blocks that were
3870 * originally written with their allocation delayed, but then the space was
3871 * allocated using fallocate() before the delayed allocation could be resolved.
3872 * The cases to look for are:
3873 * ('=' indicated delayed allocated blocks
3874 *  '-' indicates non-delayed allocated blocks)
3875 * (a) partial clusters towards beginning and/or end outside of allocated range
3876 *     are not delalloc'ed.
3877 *	Ex:
3878 *	|----c---=|====c====|====c====|===-c----|
3879 *	         |++++++ allocated ++++++|
3880 *	==> 4 complete clusters in above example
3881 *
3882 * (b) partial cluster (outside of allocated range) towards either end is
3883 *     marked for delayed allocation. In this case, we will exclude that
3884 *     cluster.
3885 *	Ex:
3886 *	|----====c========|========c========|
3887 *	     |++++++ allocated ++++++|
3888 *	==> 1 complete clusters in above example
3889 *
3890 *	Ex:
3891 *	|================c================|
3892 *            |++++++ allocated ++++++|
3893 *	==> 0 complete clusters in above example
3894 *
3895 * The ext4_da_update_reserve_space will be called only if we
3896 * determine here that there were some "entire" clusters that span
3897 * this 'allocated' range.
3898 * In the non-bigalloc case, this function will just end up returning num_blks
3899 * without ever calling ext4_find_delalloc_range.
3900 */
3901static unsigned int
3902get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3903			   unsigned int num_blks)
3904{
3905	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3906	ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3907	ext4_lblk_t lblk_from, lblk_to, c_offset;
3908	unsigned int allocated_clusters = 0;
3909
3910	alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3911	alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3912
3913	/* max possible clusters for this allocation */
3914	allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3915
3916	trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3917
3918	/* Check towards left side */
3919	c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3920	if (c_offset) {
3921		lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3922		lblk_to = lblk_from + c_offset - 1;
3923
3924		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3925			allocated_clusters--;
3926	}
3927
3928	/* Now check towards right. */
3929	c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3930	if (allocated_clusters && c_offset) {
3931		lblk_from = lblk_start + num_blks;
3932		lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3933
3934		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3935			allocated_clusters--;
3936	}
3937
3938	return allocated_clusters;
3939}
3940
3941static int
3942convert_initialized_extent(handle_t *handle, struct inode *inode,
3943			   struct ext4_map_blocks *map,
3944			   struct ext4_ext_path **ppath,
3945			   unsigned int allocated)
3946{
3947	struct ext4_ext_path *path = *ppath;
3948	struct ext4_extent *ex;
3949	ext4_lblk_t ee_block;
3950	unsigned int ee_len;
3951	int depth;
3952	int err = 0;
3953
3954	/*
3955	 * Make sure that the extent is no bigger than we support with
3956	 * unwritten extent
3957	 */
3958	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3959		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3960
3961	depth = ext_depth(inode);
3962	ex = path[depth].p_ext;
3963	ee_block = le32_to_cpu(ex->ee_block);
3964	ee_len = ext4_ext_get_actual_len(ex);
3965
3966	ext_debug("%s: inode %lu, logical"
3967		"block %llu, max_blocks %u\n", __func__, inode->i_ino,
3968		  (unsigned long long)ee_block, ee_len);
3969
3970	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3971		err = ext4_split_convert_extents(handle, inode, map, ppath,
3972				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3973		if (err < 0)
3974			return err;
3975		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3976		if (IS_ERR(path))
3977			return PTR_ERR(path);
3978		depth = ext_depth(inode);
3979		ex = path[depth].p_ext;
3980		if (!ex) {
3981			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3982					 (unsigned long) map->m_lblk);
3983			return -EFSCORRUPTED;
3984		}
3985	}
3986
3987	err = ext4_ext_get_access(handle, inode, path + depth);
3988	if (err)
3989		return err;
3990	/* first mark the extent as unwritten */
3991	ext4_ext_mark_unwritten(ex);
3992
3993	/* note: ext4_ext_correct_indexes() isn't needed here because
3994	 * borders are not changed
3995	 */
3996	ext4_ext_try_to_merge(handle, inode, path, ex);
3997
3998	/* Mark modified extent as dirty */
3999	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4000	if (err)
4001		return err;
4002	ext4_ext_show_leaf(inode, path);
4003
4004	ext4_update_inode_fsync_trans(handle, inode, 1);
4005	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4006	if (err)
4007		return err;
4008	map->m_flags |= EXT4_MAP_UNWRITTEN;
4009	if (allocated > map->m_len)
4010		allocated = map->m_len;
4011	map->m_len = allocated;
4012	return allocated;
4013}
4014
4015static int
4016ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4017			struct ext4_map_blocks *map,
4018			struct ext4_ext_path **ppath, int flags,
4019			unsigned int allocated, ext4_fsblk_t newblock)
4020{
4021	struct ext4_ext_path *path = *ppath;
4022	int ret = 0;
4023	int err = 0;
4024
4025	ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4026		  "block %llu, max_blocks %u, flags %x, allocated %u\n",
4027		  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4028		  flags, allocated);
4029	ext4_ext_show_leaf(inode, path);
4030
4031	/*
4032	 * When writing into unwritten space, we should not fail to
4033	 * allocate metadata blocks for the new extent block if needed.
4034	 */
4035	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4036
4037	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4038						    allocated, newblock);
4039
4040	/* get_block() before submit the IO, split the extent */
4041	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4042		ret = ext4_split_convert_extents(handle, inode, map, ppath,
4043					 flags | EXT4_GET_BLOCKS_CONVERT);
4044		if (ret <= 0)
4045			goto out;
4046		map->m_flags |= EXT4_MAP_UNWRITTEN;
4047		goto out;
4048	}
4049	/* IO end_io complete, convert the filled extent to written */
4050	if (flags & EXT4_GET_BLOCKS_CONVERT) {
4051		if (flags & EXT4_GET_BLOCKS_ZERO) {
4052			if (allocated > map->m_len)
4053				allocated = map->m_len;
4054			err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4055						 allocated);
4056			if (err < 0)
4057				goto out2;
4058		}
4059		ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4060							   ppath);
4061		if (ret >= 0) {
4062			ext4_update_inode_fsync_trans(handle, inode, 1);
4063			err = check_eofblocks_fl(handle, inode, map->m_lblk,
4064						 path, map->m_len);
4065		} else
4066			err = ret;
4067		map->m_flags |= EXT4_MAP_MAPPED;
4068		map->m_pblk = newblock;
4069		if (allocated > map->m_len)
4070			allocated = map->m_len;
4071		map->m_len = allocated;
4072		goto out2;
4073	}
4074	/* buffered IO case */
4075	/*
4076	 * repeat fallocate creation request
4077	 * we already have an unwritten extent
4078	 */
4079	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4080		map->m_flags |= EXT4_MAP_UNWRITTEN;
4081		goto map_out;
4082	}
4083
4084	/* buffered READ or buffered write_begin() lookup */
4085	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4086		/*
4087		 * We have blocks reserved already.  We
4088		 * return allocated blocks so that delalloc
4089		 * won't do block reservation for us.  But
4090		 * the buffer head will be unmapped so that
4091		 * a read from the block returns 0s.
4092		 */
4093		map->m_flags |= EXT4_MAP_UNWRITTEN;
4094		goto out1;
4095	}
4096
4097	/* buffered write, writepage time, convert*/
4098	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4099	if (ret >= 0)
4100		ext4_update_inode_fsync_trans(handle, inode, 1);
4101out:
4102	if (ret <= 0) {
4103		err = ret;
4104		goto out2;
4105	} else
4106		allocated = ret;
4107	map->m_flags |= EXT4_MAP_NEW;
4108	/*
4109	 * if we allocated more blocks than requested
4110	 * we need to make sure we unmap the extra block
4111	 * allocated. The actual needed block will get
4112	 * unmapped later when we find the buffer_head marked
4113	 * new.
4114	 */
4115	if (allocated > map->m_len) {
4116		clean_bdev_aliases(inode->i_sb->s_bdev, newblock + map->m_len,
4117				   allocated - map->m_len);
 
4118		allocated = map->m_len;
4119	}
4120	map->m_len = allocated;
4121
4122	/*
4123	 * If we have done fallocate with the offset that is already
4124	 * delayed allocated, we would have block reservation
4125	 * and quota reservation done in the delayed write path.
4126	 * But fallocate would have already updated quota and block
4127	 * count for this offset. So cancel these reservation
4128	 */
4129	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4130		unsigned int reserved_clusters;
4131		reserved_clusters = get_reserved_cluster_alloc(inode,
4132				map->m_lblk, map->m_len);
4133		if (reserved_clusters)
4134			ext4_da_update_reserve_space(inode,
4135						     reserved_clusters,
4136						     0);
4137	}
4138
4139map_out:
4140	map->m_flags |= EXT4_MAP_MAPPED;
4141	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4142		err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4143					 map->m_len);
4144		if (err < 0)
4145			goto out2;
4146	}
4147out1:
4148	if (allocated > map->m_len)
4149		allocated = map->m_len;
4150	ext4_ext_show_leaf(inode, path);
4151	map->m_pblk = newblock;
4152	map->m_len = allocated;
4153out2:
4154	return err ? err : allocated;
4155}
4156
4157/*
4158 * get_implied_cluster_alloc - check to see if the requested
4159 * allocation (in the map structure) overlaps with a cluster already
4160 * allocated in an extent.
4161 *	@sb	The filesystem superblock structure
4162 *	@map	The requested lblk->pblk mapping
4163 *	@ex	The extent structure which might contain an implied
4164 *			cluster allocation
4165 *
4166 * This function is called by ext4_ext_map_blocks() after we failed to
4167 * find blocks that were already in the inode's extent tree.  Hence,
4168 * we know that the beginning of the requested region cannot overlap
4169 * the extent from the inode's extent tree.  There are three cases we
4170 * want to catch.  The first is this case:
4171 *
4172 *		 |--- cluster # N--|
4173 *    |--- extent ---|	|---- requested region ---|
4174 *			|==========|
4175 *
4176 * The second case that we need to test for is this one:
4177 *
4178 *   |--------- cluster # N ----------------|
4179 *	   |--- requested region --|   |------- extent ----|
4180 *	   |=======================|
4181 *
4182 * The third case is when the requested region lies between two extents
4183 * within the same cluster:
4184 *          |------------- cluster # N-------------|
4185 * |----- ex -----|                  |---- ex_right ----|
4186 *                  |------ requested region ------|
4187 *                  |================|
4188 *
4189 * In each of the above cases, we need to set the map->m_pblk and
4190 * map->m_len so it corresponds to the return the extent labelled as
4191 * "|====|" from cluster #N, since it is already in use for data in
4192 * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
4193 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4194 * as a new "allocated" block region.  Otherwise, we will return 0 and
4195 * ext4_ext_map_blocks() will then allocate one or more new clusters
4196 * by calling ext4_mb_new_blocks().
4197 */
4198static int get_implied_cluster_alloc(struct super_block *sb,
4199				     struct ext4_map_blocks *map,
4200				     struct ext4_extent *ex,
4201				     struct ext4_ext_path *path)
4202{
4203	struct ext4_sb_info *sbi = EXT4_SB(sb);
4204	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4205	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4206	ext4_lblk_t rr_cluster_start;
4207	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4208	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4209	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4210
4211	/* The extent passed in that we are trying to match */
4212	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4213	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4214
4215	/* The requested region passed into ext4_map_blocks() */
4216	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4217
4218	if ((rr_cluster_start == ex_cluster_end) ||
4219	    (rr_cluster_start == ex_cluster_start)) {
4220		if (rr_cluster_start == ex_cluster_end)
4221			ee_start += ee_len - 1;
4222		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4223		map->m_len = min(map->m_len,
4224				 (unsigned) sbi->s_cluster_ratio - c_offset);
4225		/*
4226		 * Check for and handle this case:
4227		 *
4228		 *   |--------- cluster # N-------------|
4229		 *		       |------- extent ----|
4230		 *	   |--- requested region ---|
4231		 *	   |===========|
4232		 */
4233
4234		if (map->m_lblk < ee_block)
4235			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4236
4237		/*
4238		 * Check for the case where there is already another allocated
4239		 * block to the right of 'ex' but before the end of the cluster.
4240		 *
4241		 *          |------------- cluster # N-------------|
4242		 * |----- ex -----|                  |---- ex_right ----|
4243		 *                  |------ requested region ------|
4244		 *                  |================|
4245		 */
4246		if (map->m_lblk > ee_block) {
4247			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4248			map->m_len = min(map->m_len, next - map->m_lblk);
4249		}
4250
4251		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4252		return 1;
4253	}
4254
4255	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4256	return 0;
4257}
4258
4259
4260/*
4261 * Block allocation/map/preallocation routine for extents based files
4262 *
4263 *
4264 * Need to be called with
4265 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4266 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4267 *
4268 * return > 0, number of of blocks already mapped/allocated
4269 *          if create == 0 and these are pre-allocated blocks
4270 *          	buffer head is unmapped
4271 *          otherwise blocks are mapped
4272 *
4273 * return = 0, if plain look up failed (blocks have not been allocated)
4274 *          buffer head is unmapped
4275 *
4276 * return < 0, error case.
4277 */
4278int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4279			struct ext4_map_blocks *map, int flags)
4280{
4281	struct ext4_ext_path *path = NULL;
4282	struct ext4_extent newex, *ex, *ex2;
4283	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4284	ext4_fsblk_t newblock = 0;
4285	int free_on_err = 0, err = 0, depth, ret;
4286	unsigned int allocated = 0, offset = 0;
4287	unsigned int allocated_clusters = 0;
4288	struct ext4_allocation_request ar;
4289	ext4_lblk_t cluster_offset;
4290	bool map_from_cluster = false;
4291
4292	ext_debug("blocks %u/%u requested for inode %lu\n",
4293		  map->m_lblk, map->m_len, inode->i_ino);
4294	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4295
4296	/* find extent for this block */
4297	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4298	if (IS_ERR(path)) {
4299		err = PTR_ERR(path);
4300		path = NULL;
4301		goto out2;
4302	}
4303
4304	depth = ext_depth(inode);
4305
4306	/*
4307	 * consistent leaf must not be empty;
4308	 * this situation is possible, though, _during_ tree modification;
4309	 * this is why assert can't be put in ext4_find_extent()
4310	 */
4311	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4312		EXT4_ERROR_INODE(inode, "bad extent address "
4313				 "lblock: %lu, depth: %d pblock %lld",
4314				 (unsigned long) map->m_lblk, depth,
4315				 path[depth].p_block);
4316		err = -EFSCORRUPTED;
4317		goto out2;
4318	}
4319
4320	ex = path[depth].p_ext;
4321	if (ex) {
4322		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4323		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4324		unsigned short ee_len;
4325
4326
4327		/*
4328		 * unwritten extents are treated as holes, except that
4329		 * we split out initialized portions during a write.
4330		 */
4331		ee_len = ext4_ext_get_actual_len(ex);
4332
4333		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4334
4335		/* if found extent covers block, simply return it */
4336		if (in_range(map->m_lblk, ee_block, ee_len)) {
4337			newblock = map->m_lblk - ee_block + ee_start;
4338			/* number of remaining blocks in the extent */
4339			allocated = ee_len - (map->m_lblk - ee_block);
4340			ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4341				  ee_block, ee_len, newblock);
4342
4343			/*
4344			 * If the extent is initialized check whether the
4345			 * caller wants to convert it to unwritten.
4346			 */
4347			if ((!ext4_ext_is_unwritten(ex)) &&
4348			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4349				allocated = convert_initialized_extent(
4350						handle, inode, map, &path,
4351						allocated);
4352				goto out2;
4353			} else if (!ext4_ext_is_unwritten(ex))
4354				goto out;
4355
4356			ret = ext4_ext_handle_unwritten_extents(
4357				handle, inode, map, &path, flags,
4358				allocated, newblock);
4359			if (ret < 0)
4360				err = ret;
4361			else
4362				allocated = ret;
4363			goto out2;
4364		}
4365	}
4366
4367	/*
4368	 * requested block isn't allocated yet;
4369	 * we couldn't try to create block if create flag is zero
4370	 */
4371	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4372		ext4_lblk_t hole_start, hole_len;
4373
4374		hole_start = map->m_lblk;
4375		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4376		/*
4377		 * put just found gap into cache to speed up
4378		 * subsequent requests
4379		 */
4380		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4381
4382		/* Update hole_len to reflect hole size after map->m_lblk */
4383		if (hole_start != map->m_lblk)
4384			hole_len -= map->m_lblk - hole_start;
4385		map->m_pblk = 0;
4386		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4387
4388		goto out2;
4389	}
4390
4391	/*
4392	 * Okay, we need to do block allocation.
4393	 */
4394	newex.ee_block = cpu_to_le32(map->m_lblk);
4395	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4396
4397	/*
4398	 * If we are doing bigalloc, check to see if the extent returned
4399	 * by ext4_find_extent() implies a cluster we can use.
4400	 */
4401	if (cluster_offset && ex &&
4402	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4403		ar.len = allocated = map->m_len;
4404		newblock = map->m_pblk;
4405		map_from_cluster = true;
4406		goto got_allocated_blocks;
4407	}
4408
4409	/* find neighbour allocated blocks */
4410	ar.lleft = map->m_lblk;
4411	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4412	if (err)
4413		goto out2;
4414	ar.lright = map->m_lblk;
4415	ex2 = NULL;
4416	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4417	if (err)
4418		goto out2;
4419
4420	/* Check if the extent after searching to the right implies a
4421	 * cluster we can use. */
4422	if ((sbi->s_cluster_ratio > 1) && ex2 &&
4423	    get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4424		ar.len = allocated = map->m_len;
4425		newblock = map->m_pblk;
4426		map_from_cluster = true;
4427		goto got_allocated_blocks;
4428	}
4429
4430	/*
4431	 * See if request is beyond maximum number of blocks we can have in
4432	 * a single extent. For an initialized extent this limit is
4433	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4434	 * EXT_UNWRITTEN_MAX_LEN.
4435	 */
4436	if (map->m_len > EXT_INIT_MAX_LEN &&
4437	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4438		map->m_len = EXT_INIT_MAX_LEN;
4439	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4440		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4441		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4442
4443	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4444	newex.ee_len = cpu_to_le16(map->m_len);
4445	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4446	if (err)
4447		allocated = ext4_ext_get_actual_len(&newex);
4448	else
4449		allocated = map->m_len;
4450
4451	/* allocate new block */
4452	ar.inode = inode;
4453	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4454	ar.logical = map->m_lblk;
4455	/*
4456	 * We calculate the offset from the beginning of the cluster
4457	 * for the logical block number, since when we allocate a
4458	 * physical cluster, the physical block should start at the
4459	 * same offset from the beginning of the cluster.  This is
4460	 * needed so that future calls to get_implied_cluster_alloc()
4461	 * work correctly.
4462	 */
4463	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4464	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4465	ar.goal -= offset;
4466	ar.logical -= offset;
4467	if (S_ISREG(inode->i_mode))
4468		ar.flags = EXT4_MB_HINT_DATA;
4469	else
4470		/* disable in-core preallocation for non-regular files */
4471		ar.flags = 0;
4472	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4473		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4474	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4475		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4476	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4477		ar.flags |= EXT4_MB_USE_RESERVED;
4478	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4479	if (!newblock)
4480		goto out2;
4481	ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4482		  ar.goal, newblock, allocated);
4483	free_on_err = 1;
4484	allocated_clusters = ar.len;
4485	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4486	if (ar.len > allocated)
4487		ar.len = allocated;
4488
4489got_allocated_blocks:
4490	/* try to insert new extent into found leaf and return */
4491	ext4_ext_store_pblock(&newex, newblock + offset);
4492	newex.ee_len = cpu_to_le16(ar.len);
4493	/* Mark unwritten */
4494	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4495		ext4_ext_mark_unwritten(&newex);
4496		map->m_flags |= EXT4_MAP_UNWRITTEN;
4497	}
4498
4499	err = 0;
4500	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4501		err = check_eofblocks_fl(handle, inode, map->m_lblk,
4502					 path, ar.len);
4503	if (!err)
4504		err = ext4_ext_insert_extent(handle, inode, &path,
4505					     &newex, flags);
4506
4507	if (err && free_on_err) {
4508		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4509			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4510		/* free data blocks we just allocated */
4511		/* not a good idea to call discard here directly,
4512		 * but otherwise we'd need to call it every free() */
4513		ext4_discard_preallocations(inode);
4514		ext4_free_blocks(handle, inode, NULL, newblock,
4515				 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4516		goto out2;
4517	}
4518
4519	/* previous routine could use block we allocated */
4520	newblock = ext4_ext_pblock(&newex);
4521	allocated = ext4_ext_get_actual_len(&newex);
4522	if (allocated > map->m_len)
4523		allocated = map->m_len;
4524	map->m_flags |= EXT4_MAP_NEW;
4525
4526	/*
4527	 * Update reserved blocks/metadata blocks after successful
4528	 * block allocation which had been deferred till now.
4529	 */
4530	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4531		unsigned int reserved_clusters;
4532		/*
4533		 * Check how many clusters we had reserved this allocated range
4534		 */
4535		reserved_clusters = get_reserved_cluster_alloc(inode,
4536						map->m_lblk, allocated);
4537		if (!map_from_cluster) {
4538			BUG_ON(allocated_clusters < reserved_clusters);
4539			if (reserved_clusters < allocated_clusters) {
4540				struct ext4_inode_info *ei = EXT4_I(inode);
4541				int reservation = allocated_clusters -
4542						  reserved_clusters;
4543				/*
4544				 * It seems we claimed few clusters outside of
4545				 * the range of this allocation. We should give
4546				 * it back to the reservation pool. This can
4547				 * happen in the following case:
4548				 *
4549				 * * Suppose s_cluster_ratio is 4 (i.e., each
4550				 *   cluster has 4 blocks. Thus, the clusters
4551				 *   are [0-3],[4-7],[8-11]...
4552				 * * First comes delayed allocation write for
4553				 *   logical blocks 10 & 11. Since there were no
4554				 *   previous delayed allocated blocks in the
4555				 *   range [8-11], we would reserve 1 cluster
4556				 *   for this write.
4557				 * * Next comes write for logical blocks 3 to 8.
4558				 *   In this case, we will reserve 2 clusters
4559				 *   (for [0-3] and [4-7]; and not for [8-11] as
4560				 *   that range has a delayed allocated blocks.
4561				 *   Thus total reserved clusters now becomes 3.
4562				 * * Now, during the delayed allocation writeout
4563				 *   time, we will first write blocks [3-8] and
4564				 *   allocate 3 clusters for writing these
4565				 *   blocks. Also, we would claim all these
4566				 *   three clusters above.
4567				 * * Now when we come here to writeout the
4568				 *   blocks [10-11], we would expect to claim
4569				 *   the reservation of 1 cluster we had made
4570				 *   (and we would claim it since there are no
4571				 *   more delayed allocated blocks in the range
4572				 *   [8-11]. But our reserved cluster count had
4573				 *   already gone to 0.
4574				 *
4575				 *   Thus, at the step 4 above when we determine
4576				 *   that there are still some unwritten delayed
4577				 *   allocated blocks outside of our current
4578				 *   block range, we should increment the
4579				 *   reserved clusters count so that when the
4580				 *   remaining blocks finally gets written, we
4581				 *   could claim them.
4582				 */
4583				dquot_reserve_block(inode,
4584						EXT4_C2B(sbi, reservation));
4585				spin_lock(&ei->i_block_reservation_lock);
4586				ei->i_reserved_data_blocks += reservation;
4587				spin_unlock(&ei->i_block_reservation_lock);
4588			}
4589			/*
4590			 * We will claim quota for all newly allocated blocks.
4591			 * We're updating the reserved space *after* the
4592			 * correction above so we do not accidentally free
4593			 * all the metadata reservation because we might
4594			 * actually need it later on.
4595			 */
4596			ext4_da_update_reserve_space(inode, allocated_clusters,
4597							1);
4598		}
4599	}
4600
4601	/*
4602	 * Cache the extent and update transaction to commit on fdatasync only
4603	 * when it is _not_ an unwritten extent.
4604	 */
4605	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4606		ext4_update_inode_fsync_trans(handle, inode, 1);
4607	else
4608		ext4_update_inode_fsync_trans(handle, inode, 0);
4609out:
4610	if (allocated > map->m_len)
4611		allocated = map->m_len;
4612	ext4_ext_show_leaf(inode, path);
4613	map->m_flags |= EXT4_MAP_MAPPED;
4614	map->m_pblk = newblock;
4615	map->m_len = allocated;
4616out2:
4617	ext4_ext_drop_refs(path);
4618	kfree(path);
4619
4620	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4621				       err ? err : allocated);
4622	return err ? err : allocated;
4623}
4624
4625int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4626{
4627	struct super_block *sb = inode->i_sb;
4628	ext4_lblk_t last_block;
4629	int err = 0;
4630
4631	/*
4632	 * TODO: optimization is possible here.
4633	 * Probably we need not scan at all,
4634	 * because page truncation is enough.
4635	 */
4636
4637	/* we have to know where to truncate from in crash case */
4638	EXT4_I(inode)->i_disksize = inode->i_size;
4639	err = ext4_mark_inode_dirty(handle, inode);
4640	if (err)
4641		return err;
4642
4643	last_block = (inode->i_size + sb->s_blocksize - 1)
4644			>> EXT4_BLOCK_SIZE_BITS(sb);
4645retry:
4646	err = ext4_es_remove_extent(inode, last_block,
4647				    EXT_MAX_BLOCKS - last_block);
4648	if (err == -ENOMEM) {
4649		cond_resched();
4650		congestion_wait(BLK_RW_ASYNC, HZ/50);
4651		goto retry;
4652	}
4653	if (err)
4654		return err;
4655	return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
 
 
 
4656}
4657
4658static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4659				  ext4_lblk_t len, loff_t new_size,
4660				  int flags, int mode)
4661{
4662	struct inode *inode = file_inode(file);
4663	handle_t *handle;
4664	int ret = 0;
4665	int ret2 = 0;
4666	int retries = 0;
4667	int depth = 0;
4668	struct ext4_map_blocks map;
4669	unsigned int credits;
4670	loff_t epos;
4671
4672	BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4673	map.m_lblk = offset;
4674	map.m_len = len;
4675	/*
4676	 * Don't normalize the request if it can fit in one extent so
4677	 * that it doesn't get unnecessarily split into multiple
4678	 * extents.
4679	 */
4680	if (len <= EXT_UNWRITTEN_MAX_LEN)
4681		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4682
4683	/*
4684	 * credits to insert 1 extent into extent tree
4685	 */
4686	credits = ext4_chunk_trans_blocks(inode, len);
4687	depth = ext_depth(inode);
 
 
 
 
 
 
4688
4689retry:
4690	while (ret >= 0 && len) {
4691		/*
4692		 * Recalculate credits when extent tree depth changes.
4693		 */
4694		if (depth != ext_depth(inode)) {
4695			credits = ext4_chunk_trans_blocks(inode, len);
4696			depth = ext_depth(inode);
4697		}
4698
4699		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4700					    credits);
4701		if (IS_ERR(handle)) {
4702			ret = PTR_ERR(handle);
4703			break;
4704		}
4705		ret = ext4_map_blocks(handle, inode, &map, flags);
4706		if (ret <= 0) {
4707			ext4_debug("inode #%lu: block %u: len %u: "
4708				   "ext4_ext_map_blocks returned %d",
4709				   inode->i_ino, map.m_lblk,
4710				   map.m_len, ret);
4711			ext4_mark_inode_dirty(handle, inode);
4712			ret2 = ext4_journal_stop(handle);
4713			break;
4714		}
4715		map.m_lblk += ret;
4716		map.m_len = len = len - ret;
4717		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4718		inode->i_ctime = current_time(inode);
4719		if (new_size) {
4720			if (epos > new_size)
4721				epos = new_size;
4722			if (ext4_update_inode_size(inode, epos) & 0x1)
4723				inode->i_mtime = inode->i_ctime;
4724		} else {
4725			if (epos > inode->i_size)
4726				ext4_set_inode_flag(inode,
4727						    EXT4_INODE_EOFBLOCKS);
4728		}
4729		ext4_mark_inode_dirty(handle, inode);
4730		ret2 = ext4_journal_stop(handle);
4731		if (ret2)
4732			break;
4733	}
4734	if (ret == -ENOSPC &&
4735			ext4_should_retry_alloc(inode->i_sb, &retries)) {
4736		ret = 0;
4737		goto retry;
4738	}
4739
4740	return ret > 0 ? ret2 : ret;
4741}
4742
4743static long ext4_zero_range(struct file *file, loff_t offset,
4744			    loff_t len, int mode)
4745{
4746	struct inode *inode = file_inode(file);
4747	handle_t *handle = NULL;
4748	unsigned int max_blocks;
4749	loff_t new_size = 0;
4750	int ret = 0;
4751	int flags;
4752	int credits;
4753	int partial_begin, partial_end;
4754	loff_t start, end;
4755	ext4_lblk_t lblk;
4756	unsigned int blkbits = inode->i_blkbits;
4757
4758	trace_ext4_zero_range(inode, offset, len, mode);
4759
4760	if (!S_ISREG(inode->i_mode))
4761		return -EINVAL;
4762
4763	/* Call ext4_force_commit to flush all data in case of data=journal. */
4764	if (ext4_should_journal_data(inode)) {
4765		ret = ext4_force_commit(inode->i_sb);
4766		if (ret)
4767			return ret;
4768	}
4769
4770	/*
4771	 * Round up offset. This is not fallocate, we neet to zero out
4772	 * blocks, so convert interior block aligned part of the range to
4773	 * unwritten and possibly manually zero out unaligned parts of the
4774	 * range.
4775	 */
4776	start = round_up(offset, 1 << blkbits);
4777	end = round_down((offset + len), 1 << blkbits);
4778
4779	if (start < offset || end > offset + len)
4780		return -EINVAL;
4781	partial_begin = offset & ((1 << blkbits) - 1);
4782	partial_end = (offset + len) & ((1 << blkbits) - 1);
4783
4784	lblk = start >> blkbits;
4785	max_blocks = (end >> blkbits);
4786	if (max_blocks < lblk)
4787		max_blocks = 0;
4788	else
4789		max_blocks -= lblk;
4790
4791	inode_lock(inode);
4792
4793	/*
4794	 * Indirect files do not support unwritten extnets
4795	 */
4796	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4797		ret = -EOPNOTSUPP;
4798		goto out_mutex;
4799	}
4800
4801	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4802	     offset + len > i_size_read(inode)) {
4803		new_size = offset + len;
4804		ret = inode_newsize_ok(inode, new_size);
4805		if (ret)
4806			goto out_mutex;
4807	}
4808
4809	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4810	if (mode & FALLOC_FL_KEEP_SIZE)
4811		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4812
4813	/* Wait all existing dio workers, newcomers will block on i_mutex */
4814	ext4_inode_block_unlocked_dio(inode);
4815	inode_dio_wait(inode);
4816
4817	/* Preallocate the range including the unaligned edges */
4818	if (partial_begin || partial_end) {
4819		ret = ext4_alloc_file_blocks(file,
4820				round_down(offset, 1 << blkbits) >> blkbits,
4821				(round_up((offset + len), 1 << blkbits) -
4822				 round_down(offset, 1 << blkbits)) >> blkbits,
4823				new_size, flags, mode);
4824		if (ret)
4825			goto out_dio;
4826
4827	}
4828
4829	/* Zero range excluding the unaligned edges */
4830	if (max_blocks > 0) {
4831		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4832			  EXT4_EX_NOCACHE);
4833
4834		/*
4835		 * Prevent page faults from reinstantiating pages we have
4836		 * released from page cache.
4837		 */
4838		down_write(&EXT4_I(inode)->i_mmap_sem);
4839		ret = ext4_update_disksize_before_punch(inode, offset, len);
4840		if (ret) {
4841			up_write(&EXT4_I(inode)->i_mmap_sem);
4842			goto out_dio;
4843		}
4844		/* Now release the pages and zero block aligned part of pages */
4845		truncate_pagecache_range(inode, start, end - 1);
4846		inode->i_mtime = inode->i_ctime = current_time(inode);
4847
4848		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4849					     flags, mode);
4850		up_write(&EXT4_I(inode)->i_mmap_sem);
4851		if (ret)
4852			goto out_dio;
4853	}
4854	if (!partial_begin && !partial_end)
4855		goto out_dio;
4856
4857	/*
4858	 * In worst case we have to writeout two nonadjacent unwritten
4859	 * blocks and update the inode
4860	 */
4861	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4862	if (ext4_should_journal_data(inode))
4863		credits += 2;
4864	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4865	if (IS_ERR(handle)) {
4866		ret = PTR_ERR(handle);
4867		ext4_std_error(inode->i_sb, ret);
4868		goto out_dio;
4869	}
4870
4871	inode->i_mtime = inode->i_ctime = current_time(inode);
4872	if (new_size) {
4873		ext4_update_inode_size(inode, new_size);
4874	} else {
4875		/*
4876		* Mark that we allocate beyond EOF so the subsequent truncate
4877		* can proceed even if the new size is the same as i_size.
4878		*/
4879		if ((offset + len) > i_size_read(inode))
4880			ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4881	}
4882	ext4_mark_inode_dirty(handle, inode);
4883
4884	/* Zero out partial block at the edges of the range */
4885	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4886
4887	if (file->f_flags & O_SYNC)
4888		ext4_handle_sync(handle);
4889
4890	ext4_journal_stop(handle);
4891out_dio:
4892	ext4_inode_resume_unlocked_dio(inode);
4893out_mutex:
4894	inode_unlock(inode);
4895	return ret;
4896}
4897
4898/*
4899 * preallocate space for a file. This implements ext4's fallocate file
4900 * operation, which gets called from sys_fallocate system call.
4901 * For block-mapped files, posix_fallocate should fall back to the method
4902 * of writing zeroes to the required new blocks (the same behavior which is
4903 * expected for file systems which do not support fallocate() system call).
4904 */
4905long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4906{
4907	struct inode *inode = file_inode(file);
4908	loff_t new_size = 0;
4909	unsigned int max_blocks;
4910	int ret = 0;
4911	int flags;
4912	ext4_lblk_t lblk;
4913	unsigned int blkbits = inode->i_blkbits;
4914
4915	/*
4916	 * Encrypted inodes can't handle collapse range or insert
4917	 * range since we would need to re-encrypt blocks with a
4918	 * different IV or XTS tweak (which are based on the logical
4919	 * block number).
4920	 *
4921	 * XXX It's not clear why zero range isn't working, but we'll
4922	 * leave it disabled for encrypted inodes for now.  This is a
4923	 * bug we should fix....
4924	 */
4925	if (ext4_encrypted_inode(inode) &&
4926	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4927		     FALLOC_FL_ZERO_RANGE)))
4928		return -EOPNOTSUPP;
4929
4930	/* Return error if mode is not supported */
4931	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4932		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4933		     FALLOC_FL_INSERT_RANGE))
4934		return -EOPNOTSUPP;
4935
4936	if (mode & FALLOC_FL_PUNCH_HOLE)
4937		return ext4_punch_hole(inode, offset, len);
4938
4939	ret = ext4_convert_inline_data(inode);
4940	if (ret)
4941		return ret;
4942
4943	if (mode & FALLOC_FL_COLLAPSE_RANGE)
4944		return ext4_collapse_range(inode, offset, len);
4945
4946	if (mode & FALLOC_FL_INSERT_RANGE)
4947		return ext4_insert_range(inode, offset, len);
4948
4949	if (mode & FALLOC_FL_ZERO_RANGE)
4950		return ext4_zero_range(file, offset, len, mode);
4951
4952	trace_ext4_fallocate_enter(inode, offset, len, mode);
4953	lblk = offset >> blkbits;
 
 
 
 
 
 
4954
4955	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4956	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4957	if (mode & FALLOC_FL_KEEP_SIZE)
4958		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4959
4960	inode_lock(inode);
4961
4962	/*
4963	 * We only support preallocation for extent-based files only
4964	 */
4965	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4966		ret = -EOPNOTSUPP;
4967		goto out;
4968	}
4969
4970	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4971	     offset + len > i_size_read(inode)) {
4972		new_size = offset + len;
4973		ret = inode_newsize_ok(inode, new_size);
4974		if (ret)
4975			goto out;
4976	}
4977
4978	/* Wait all existing dio workers, newcomers will block on i_mutex */
4979	ext4_inode_block_unlocked_dio(inode);
4980	inode_dio_wait(inode);
4981
4982	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4983				     flags, mode);
4984	ext4_inode_resume_unlocked_dio(inode);
4985	if (ret)
4986		goto out;
4987
4988	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4989		ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4990						EXT4_I(inode)->i_sync_tid);
4991	}
4992out:
4993	inode_unlock(inode);
4994	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4995	return ret;
4996}
4997
4998/*
4999 * This function convert a range of blocks to written extents
5000 * The caller of this function will pass the start offset and the size.
5001 * all unwritten extents within this range will be converted to
5002 * written extents.
5003 *
5004 * This function is called from the direct IO end io call back
5005 * function, to convert the fallocated extents after IO is completed.
5006 * Returns 0 on success.
5007 */
5008int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5009				   loff_t offset, ssize_t len)
5010{
5011	unsigned int max_blocks;
5012	int ret = 0;
5013	int ret2 = 0;
5014	struct ext4_map_blocks map;
5015	unsigned int credits, blkbits = inode->i_blkbits;
5016
5017	map.m_lblk = offset >> blkbits;
5018	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
5019
 
 
 
 
5020	/*
5021	 * This is somewhat ugly but the idea is clear: When transaction is
5022	 * reserved, everything goes into it. Otherwise we rather start several
5023	 * smaller transactions for conversion of each extent separately.
5024	 */
5025	if (handle) {
5026		handle = ext4_journal_start_reserved(handle,
5027						     EXT4_HT_EXT_CONVERT);
5028		if (IS_ERR(handle))
5029			return PTR_ERR(handle);
5030		credits = 0;
5031	} else {
5032		/*
5033		 * credits to insert 1 extent into extent tree
5034		 */
5035		credits = ext4_chunk_trans_blocks(inode, max_blocks);
5036	}
5037	while (ret >= 0 && ret < max_blocks) {
5038		map.m_lblk += ret;
5039		map.m_len = (max_blocks -= ret);
5040		if (credits) {
5041			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5042						    credits);
5043			if (IS_ERR(handle)) {
5044				ret = PTR_ERR(handle);
5045				break;
5046			}
5047		}
5048		ret = ext4_map_blocks(handle, inode, &map,
5049				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5050		if (ret <= 0)
5051			ext4_warning(inode->i_sb,
5052				     "inode #%lu: block %u: len %u: "
5053				     "ext4_ext_map_blocks returned %d",
5054				     inode->i_ino, map.m_lblk,
5055				     map.m_len, ret);
5056		ext4_mark_inode_dirty(handle, inode);
5057		if (credits)
5058			ret2 = ext4_journal_stop(handle);
5059		if (ret <= 0 || ret2)
5060			break;
5061	}
5062	if (!credits)
5063		ret2 = ext4_journal_stop(handle);
5064	return ret > 0 ? ret2 : ret;
5065}
5066
5067/*
5068 * If newes is not existing extent (newes->ec_pblk equals zero) find
5069 * delayed extent at start of newes and update newes accordingly and
5070 * return start of the next delayed extent.
5071 *
5072 * If newes is existing extent (newes->ec_pblk is not equal zero)
5073 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5074 * extent found. Leave newes unmodified.
5075 */
5076static int ext4_find_delayed_extent(struct inode *inode,
5077				    struct extent_status *newes)
5078{
5079	struct extent_status es;
5080	ext4_lblk_t block, next_del;
5081
5082	if (newes->es_pblk == 0) {
5083		ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5084				newes->es_lblk + newes->es_len - 1, &es);
5085
5086		/*
5087		 * No extent in extent-tree contains block @newes->es_pblk,
5088		 * then the block may stay in 1)a hole or 2)delayed-extent.
5089		 */
5090		if (es.es_len == 0)
5091			/* A hole found. */
5092			return 0;
5093
5094		if (es.es_lblk > newes->es_lblk) {
5095			/* A hole found. */
5096			newes->es_len = min(es.es_lblk - newes->es_lblk,
5097					    newes->es_len);
5098			return 0;
5099		}
5100
5101		newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5102	}
5103
5104	block = newes->es_lblk + newes->es_len;
5105	ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5106	if (es.es_len == 0)
5107		next_del = EXT_MAX_BLOCKS;
5108	else
5109		next_del = es.es_lblk;
5110
5111	return next_del;
5112}
5113/* fiemap flags we can handle specified here */
5114#define EXT4_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5115
5116static int ext4_xattr_fiemap(struct inode *inode,
5117				struct fiemap_extent_info *fieinfo)
5118{
5119	__u64 physical = 0;
5120	__u64 length;
5121	__u32 flags = FIEMAP_EXTENT_LAST;
5122	int blockbits = inode->i_sb->s_blocksize_bits;
5123	int error = 0;
5124
5125	/* in-inode? */
5126	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5127		struct ext4_iloc iloc;
5128		int offset;	/* offset of xattr in inode */
5129
5130		error = ext4_get_inode_loc(inode, &iloc);
5131		if (error)
5132			return error;
5133		physical = (__u64)iloc.bh->b_blocknr << blockbits;
5134		offset = EXT4_GOOD_OLD_INODE_SIZE +
5135				EXT4_I(inode)->i_extra_isize;
5136		physical += offset;
5137		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5138		flags |= FIEMAP_EXTENT_DATA_INLINE;
5139		brelse(iloc.bh);
5140	} else { /* external block */
5141		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5142		length = inode->i_sb->s_blocksize;
5143	}
5144
5145	if (physical)
5146		error = fiemap_fill_next_extent(fieinfo, 0, physical,
5147						length, flags);
5148	return (error < 0 ? error : 0);
5149}
5150
5151int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5152		__u64 start, __u64 len)
5153{
5154	ext4_lblk_t start_blk;
5155	int error = 0;
5156
5157	if (ext4_has_inline_data(inode)) {
5158		int has_inline = 1;
5159
5160		error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5161						start, len);
5162
5163		if (has_inline)
5164			return error;
5165	}
5166
5167	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5168		error = ext4_ext_precache(inode);
5169		if (error)
5170			return error;
5171	}
5172
5173	/* fallback to generic here if not in extents fmt */
5174	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5175		return generic_block_fiemap(inode, fieinfo, start, len,
5176			ext4_get_block);
5177
5178	if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5179		return -EBADR;
5180
5181	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5182		error = ext4_xattr_fiemap(inode, fieinfo);
5183	} else {
5184		ext4_lblk_t len_blks;
5185		__u64 last_blk;
5186
5187		start_blk = start >> inode->i_sb->s_blocksize_bits;
5188		last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5189		if (last_blk >= EXT_MAX_BLOCKS)
5190			last_blk = EXT_MAX_BLOCKS-1;
5191		len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5192
5193		/*
5194		 * Walk the extent tree gathering extent information
5195		 * and pushing extents back to the user.
5196		 */
5197		error = ext4_fill_fiemap_extents(inode, start_blk,
5198						 len_blks, fieinfo);
5199	}
5200	return error;
5201}
5202
5203/*
5204 * ext4_access_path:
5205 * Function to access the path buffer for marking it dirty.
5206 * It also checks if there are sufficient credits left in the journal handle
5207 * to update path.
5208 */
5209static int
5210ext4_access_path(handle_t *handle, struct inode *inode,
5211		struct ext4_ext_path *path)
5212{
5213	int credits, err;
5214
5215	if (!ext4_handle_valid(handle))
5216		return 0;
5217
5218	/*
5219	 * Check if need to extend journal credits
5220	 * 3 for leaf, sb, and inode plus 2 (bmap and group
5221	 * descriptor) for each block group; assume two block
5222	 * groups
5223	 */
5224	if (handle->h_buffer_credits < 7) {
5225		credits = ext4_writepage_trans_blocks(inode);
5226		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5227		/* EAGAIN is success */
5228		if (err && err != -EAGAIN)
5229			return err;
5230	}
5231
5232	err = ext4_ext_get_access(handle, inode, path);
5233	return err;
5234}
5235
5236/*
5237 * ext4_ext_shift_path_extents:
5238 * Shift the extents of a path structure lying between path[depth].p_ext
5239 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5240 * if it is right shift or left shift operation.
5241 */
5242static int
5243ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5244			    struct inode *inode, handle_t *handle,
5245			    enum SHIFT_DIRECTION SHIFT)
5246{
5247	int depth, err = 0;
5248	struct ext4_extent *ex_start, *ex_last;
5249	bool update = 0;
5250	depth = path->p_depth;
5251
5252	while (depth >= 0) {
5253		if (depth == path->p_depth) {
5254			ex_start = path[depth].p_ext;
5255			if (!ex_start)
5256				return -EFSCORRUPTED;
5257
5258			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5259
5260			err = ext4_access_path(handle, inode, path + depth);
5261			if (err)
5262				goto out;
5263
5264			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5265				update = 1;
5266
5267			while (ex_start <= ex_last) {
5268				if (SHIFT == SHIFT_LEFT) {
5269					le32_add_cpu(&ex_start->ee_block,
5270						-shift);
5271					/* Try to merge to the left. */
5272					if ((ex_start >
5273					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5274					    &&
5275					    ext4_ext_try_to_merge_right(inode,
5276					    path, ex_start - 1))
5277						ex_last--;
5278					else
5279						ex_start++;
5280				} else {
5281					le32_add_cpu(&ex_last->ee_block, shift);
5282					ext4_ext_try_to_merge_right(inode, path,
5283						ex_last);
5284					ex_last--;
5285				}
5286			}
5287			err = ext4_ext_dirty(handle, inode, path + depth);
5288			if (err)
5289				goto out;
5290
5291			if (--depth < 0 || !update)
5292				break;
5293		}
5294
5295		/* Update index too */
5296		err = ext4_access_path(handle, inode, path + depth);
5297		if (err)
5298			goto out;
5299
5300		if (SHIFT == SHIFT_LEFT)
5301			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5302		else
5303			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5304		err = ext4_ext_dirty(handle, inode, path + depth);
5305		if (err)
5306			goto out;
5307
5308		/* we are done if current index is not a starting index */
5309		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5310			break;
5311
5312		depth--;
5313	}
5314
5315out:
5316	return err;
5317}
5318
5319/*
5320 * ext4_ext_shift_extents:
5321 * All the extents which lies in the range from @start to the last allocated
5322 * block for the @inode are shifted either towards left or right (depending
5323 * upon @SHIFT) by @shift blocks.
5324 * On success, 0 is returned, error otherwise.
5325 */
5326static int
5327ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5328		       ext4_lblk_t start, ext4_lblk_t shift,
5329		       enum SHIFT_DIRECTION SHIFT)
5330{
5331	struct ext4_ext_path *path;
5332	int ret = 0, depth;
5333	struct ext4_extent *extent;
5334	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5335
5336	/* Let path point to the last extent */
5337	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5338				EXT4_EX_NOCACHE);
5339	if (IS_ERR(path))
5340		return PTR_ERR(path);
5341
5342	depth = path->p_depth;
5343	extent = path[depth].p_ext;
5344	if (!extent)
5345		goto out;
5346
5347	stop = le32_to_cpu(extent->ee_block);
 
5348
5349       /*
5350	 * In case of left shift, Don't start shifting extents until we make
5351	 * sure the hole is big enough to accommodate the shift.
5352	*/
5353	if (SHIFT == SHIFT_LEFT) {
5354		path = ext4_find_extent(inode, start - 1, &path,
5355					EXT4_EX_NOCACHE);
5356		if (IS_ERR(path))
5357			return PTR_ERR(path);
5358		depth = path->p_depth;
5359		extent =  path[depth].p_ext;
5360		if (extent) {
5361			ex_start = le32_to_cpu(extent->ee_block);
5362			ex_end = le32_to_cpu(extent->ee_block) +
5363				ext4_ext_get_actual_len(extent);
5364		} else {
5365			ex_start = 0;
5366			ex_end = 0;
5367		}
5368
5369		if ((start == ex_start && shift > ex_start) ||
5370		    (shift > start - ex_end)) {
5371			ext4_ext_drop_refs(path);
5372			kfree(path);
5373			return -EINVAL;
5374		}
5375	}
5376
5377	/*
5378	 * In case of left shift, iterator points to start and it is increased
5379	 * till we reach stop. In case of right shift, iterator points to stop
5380	 * and it is decreased till we reach start.
5381	 */
5382	if (SHIFT == SHIFT_LEFT)
5383		iterator = &start;
5384	else
5385		iterator = &stop;
5386
5387	/*
5388	 * Its safe to start updating extents.  Start and stop are unsigned, so
5389	 * in case of right shift if extent with 0 block is reached, iterator
5390	 * becomes NULL to indicate the end of the loop.
5391	 */
5392	while (iterator && start <= stop) {
5393		path = ext4_find_extent(inode, *iterator, &path,
5394					EXT4_EX_NOCACHE);
5395		if (IS_ERR(path))
5396			return PTR_ERR(path);
5397		depth = path->p_depth;
5398		extent = path[depth].p_ext;
5399		if (!extent) {
5400			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5401					 (unsigned long) *iterator);
5402			return -EFSCORRUPTED;
5403		}
5404		if (SHIFT == SHIFT_LEFT && *iterator >
5405		    le32_to_cpu(extent->ee_block)) {
5406			/* Hole, move to the next extent */
5407			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5408				path[depth].p_ext++;
5409			} else {
5410				*iterator = ext4_ext_next_allocated_block(path);
5411				continue;
5412			}
5413		}
5414
5415		if (SHIFT == SHIFT_LEFT) {
5416			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5417			*iterator = le32_to_cpu(extent->ee_block) +
5418					ext4_ext_get_actual_len(extent);
5419		} else {
5420			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5421			if (le32_to_cpu(extent->ee_block) > 0)
5422				*iterator = le32_to_cpu(extent->ee_block) - 1;
5423			else
5424				/* Beginning is reached, end of the loop */
5425				iterator = NULL;
5426			/* Update path extent in case we need to stop */
5427			while (le32_to_cpu(extent->ee_block) < start)
5428				extent++;
5429			path[depth].p_ext = extent;
5430		}
5431		ret = ext4_ext_shift_path_extents(path, shift, inode,
5432				handle, SHIFT);
5433		if (ret)
5434			break;
5435	}
5436out:
5437	ext4_ext_drop_refs(path);
5438	kfree(path);
5439	return ret;
5440}
5441
5442/*
5443 * ext4_collapse_range:
5444 * This implements the fallocate's collapse range functionality for ext4
5445 * Returns: 0 and non-zero on error.
5446 */
5447int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5448{
5449	struct super_block *sb = inode->i_sb;
5450	ext4_lblk_t punch_start, punch_stop;
5451	handle_t *handle;
5452	unsigned int credits;
5453	loff_t new_size, ioffset;
5454	int ret;
5455
5456	/*
5457	 * We need to test this early because xfstests assumes that a
5458	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5459	 * system does not support collapse range.
5460	 */
5461	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5462		return -EOPNOTSUPP;
5463
5464	/* Collapse range works only on fs block size aligned offsets. */
5465	if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5466	    len & (EXT4_CLUSTER_SIZE(sb) - 1))
5467		return -EINVAL;
5468
5469	if (!S_ISREG(inode->i_mode))
5470		return -EINVAL;
5471
5472	trace_ext4_collapse_range(inode, offset, len);
5473
5474	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5475	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5476
5477	/* Call ext4_force_commit to flush all data in case of data=journal. */
5478	if (ext4_should_journal_data(inode)) {
5479		ret = ext4_force_commit(inode->i_sb);
5480		if (ret)
5481			return ret;
5482	}
5483
5484	inode_lock(inode);
5485	/*
5486	 * There is no need to overlap collapse range with EOF, in which case
5487	 * it is effectively a truncate operation
5488	 */
5489	if (offset + len >= i_size_read(inode)) {
5490		ret = -EINVAL;
5491		goto out_mutex;
5492	}
5493
5494	/* Currently just for extent based files */
5495	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5496		ret = -EOPNOTSUPP;
5497		goto out_mutex;
5498	}
5499
5500	/* Wait for existing dio to complete */
5501	ext4_inode_block_unlocked_dio(inode);
5502	inode_dio_wait(inode);
5503
5504	/*
5505	 * Prevent page faults from reinstantiating pages we have released from
5506	 * page cache.
5507	 */
5508	down_write(&EXT4_I(inode)->i_mmap_sem);
5509	/*
5510	 * Need to round down offset to be aligned with page size boundary
5511	 * for page size > block size.
5512	 */
5513	ioffset = round_down(offset, PAGE_SIZE);
5514	/*
5515	 * Write tail of the last page before removed range since it will get
5516	 * removed from the page cache below.
5517	 */
5518	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5519	if (ret)
5520		goto out_mmap;
5521	/*
5522	 * Write data that will be shifted to preserve them when discarding
5523	 * page cache below. We are also protected from pages becoming dirty
5524	 * by i_mmap_sem.
5525	 */
5526	ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5527					   LLONG_MAX);
5528	if (ret)
5529		goto out_mmap;
5530	truncate_pagecache(inode, ioffset);
5531
5532	credits = ext4_writepage_trans_blocks(inode);
5533	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5534	if (IS_ERR(handle)) {
5535		ret = PTR_ERR(handle);
5536		goto out_mmap;
5537	}
5538
5539	down_write(&EXT4_I(inode)->i_data_sem);
5540	ext4_discard_preallocations(inode);
5541
5542	ret = ext4_es_remove_extent(inode, punch_start,
5543				    EXT_MAX_BLOCKS - punch_start);
5544	if (ret) {
5545		up_write(&EXT4_I(inode)->i_data_sem);
5546		goto out_stop;
5547	}
5548
5549	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5550	if (ret) {
5551		up_write(&EXT4_I(inode)->i_data_sem);
5552		goto out_stop;
5553	}
5554	ext4_discard_preallocations(inode);
5555
5556	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5557				     punch_stop - punch_start, SHIFT_LEFT);
5558	if (ret) {
5559		up_write(&EXT4_I(inode)->i_data_sem);
5560		goto out_stop;
5561	}
5562
5563	new_size = i_size_read(inode) - len;
5564	i_size_write(inode, new_size);
5565	EXT4_I(inode)->i_disksize = new_size;
5566
5567	up_write(&EXT4_I(inode)->i_data_sem);
5568	if (IS_SYNC(inode))
5569		ext4_handle_sync(handle);
5570	inode->i_mtime = inode->i_ctime = current_time(inode);
5571	ext4_mark_inode_dirty(handle, inode);
5572
5573out_stop:
5574	ext4_journal_stop(handle);
5575out_mmap:
5576	up_write(&EXT4_I(inode)->i_mmap_sem);
5577	ext4_inode_resume_unlocked_dio(inode);
5578out_mutex:
5579	inode_unlock(inode);
5580	return ret;
5581}
5582
5583/*
5584 * ext4_insert_range:
5585 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5586 * The data blocks starting from @offset to the EOF are shifted by @len
5587 * towards right to create a hole in the @inode. Inode size is increased
5588 * by len bytes.
5589 * Returns 0 on success, error otherwise.
5590 */
5591int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5592{
5593	struct super_block *sb = inode->i_sb;
5594	handle_t *handle;
5595	struct ext4_ext_path *path;
5596	struct ext4_extent *extent;
5597	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5598	unsigned int credits, ee_len;
5599	int ret = 0, depth, split_flag = 0;
5600	loff_t ioffset;
5601
5602	/*
5603	 * We need to test this early because xfstests assumes that an
5604	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5605	 * system does not support insert range.
5606	 */
5607	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5608		return -EOPNOTSUPP;
5609
5610	/* Insert range works only on fs block size aligned offsets. */
5611	if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5612			len & (EXT4_CLUSTER_SIZE(sb) - 1))
5613		return -EINVAL;
5614
5615	if (!S_ISREG(inode->i_mode))
5616		return -EOPNOTSUPP;
5617
5618	trace_ext4_insert_range(inode, offset, len);
5619
5620	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5621	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5622
5623	/* Call ext4_force_commit to flush all data in case of data=journal */
5624	if (ext4_should_journal_data(inode)) {
5625		ret = ext4_force_commit(inode->i_sb);
5626		if (ret)
5627			return ret;
5628	}
5629
5630	inode_lock(inode);
5631	/* Currently just for extent based files */
5632	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5633		ret = -EOPNOTSUPP;
5634		goto out_mutex;
5635	}
5636
5637	/* Check for wrap through zero */
5638	if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5639		ret = -EFBIG;
5640		goto out_mutex;
5641	}
5642
5643	/* Offset should be less than i_size */
5644	if (offset >= i_size_read(inode)) {
5645		ret = -EINVAL;
5646		goto out_mutex;
5647	}
5648
5649	/* Wait for existing dio to complete */
5650	ext4_inode_block_unlocked_dio(inode);
5651	inode_dio_wait(inode);
5652
5653	/*
5654	 * Prevent page faults from reinstantiating pages we have released from
5655	 * page cache.
5656	 */
5657	down_write(&EXT4_I(inode)->i_mmap_sem);
5658	/*
5659	 * Need to round down to align start offset to page size boundary
5660	 * for page size > block size.
5661	 */
5662	ioffset = round_down(offset, PAGE_SIZE);
5663	/* Write out all dirty pages */
5664	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5665			LLONG_MAX);
5666	if (ret)
5667		goto out_mmap;
5668	truncate_pagecache(inode, ioffset);
5669
5670	credits = ext4_writepage_trans_blocks(inode);
5671	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5672	if (IS_ERR(handle)) {
5673		ret = PTR_ERR(handle);
5674		goto out_mmap;
5675	}
5676
5677	/* Expand file to avoid data loss if there is error while shifting */
5678	inode->i_size += len;
5679	EXT4_I(inode)->i_disksize += len;
5680	inode->i_mtime = inode->i_ctime = current_time(inode);
5681	ret = ext4_mark_inode_dirty(handle, inode);
5682	if (ret)
5683		goto out_stop;
5684
5685	down_write(&EXT4_I(inode)->i_data_sem);
5686	ext4_discard_preallocations(inode);
5687
5688	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5689	if (IS_ERR(path)) {
5690		up_write(&EXT4_I(inode)->i_data_sem);
5691		goto out_stop;
5692	}
5693
5694	depth = ext_depth(inode);
5695	extent = path[depth].p_ext;
5696	if (extent) {
5697		ee_start_lblk = le32_to_cpu(extent->ee_block);
5698		ee_len = ext4_ext_get_actual_len(extent);
5699
5700		/*
5701		 * If offset_lblk is not the starting block of extent, split
5702		 * the extent @offset_lblk
5703		 */
5704		if ((offset_lblk > ee_start_lblk) &&
5705				(offset_lblk < (ee_start_lblk + ee_len))) {
5706			if (ext4_ext_is_unwritten(extent))
5707				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5708					EXT4_EXT_MARK_UNWRIT2;
5709			ret = ext4_split_extent_at(handle, inode, &path,
5710					offset_lblk, split_flag,
5711					EXT4_EX_NOCACHE |
5712					EXT4_GET_BLOCKS_PRE_IO |
5713					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5714		}
5715
5716		ext4_ext_drop_refs(path);
5717		kfree(path);
5718		if (ret < 0) {
5719			up_write(&EXT4_I(inode)->i_data_sem);
5720			goto out_stop;
5721		}
5722	} else {
5723		ext4_ext_drop_refs(path);
5724		kfree(path);
5725	}
5726
5727	ret = ext4_es_remove_extent(inode, offset_lblk,
5728			EXT_MAX_BLOCKS - offset_lblk);
5729	if (ret) {
5730		up_write(&EXT4_I(inode)->i_data_sem);
5731		goto out_stop;
5732	}
5733
5734	/*
5735	 * if offset_lblk lies in a hole which is at start of file, use
5736	 * ee_start_lblk to shift extents
5737	 */
5738	ret = ext4_ext_shift_extents(inode, handle,
5739		ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5740		len_lblk, SHIFT_RIGHT);
5741
5742	up_write(&EXT4_I(inode)->i_data_sem);
5743	if (IS_SYNC(inode))
5744		ext4_handle_sync(handle);
5745
5746out_stop:
5747	ext4_journal_stop(handle);
5748out_mmap:
5749	up_write(&EXT4_I(inode)->i_mmap_sem);
5750	ext4_inode_resume_unlocked_dio(inode);
5751out_mutex:
5752	inode_unlock(inode);
5753	return ret;
5754}
5755
5756/**
5757 * ext4_swap_extents - Swap extents between two inodes
5758 *
5759 * @inode1:	First inode
5760 * @inode2:	Second inode
5761 * @lblk1:	Start block for first inode
5762 * @lblk2:	Start block for second inode
5763 * @count:	Number of blocks to swap
5764 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5765 * @erp:	Pointer to save error value
5766 *
5767 * This helper routine does exactly what is promise "swap extents". All other
5768 * stuff such as page-cache locking consistency, bh mapping consistency or
5769 * extent's data copying must be performed by caller.
5770 * Locking:
5771 * 		i_mutex is held for both inodes
5772 * 		i_data_sem is locked for write for both inodes
5773 * Assumptions:
5774 *		All pages from requested range are locked for both inodes
5775 */
5776int
5777ext4_swap_extents(handle_t *handle, struct inode *inode1,
5778		     struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5779		  ext4_lblk_t count, int unwritten, int *erp)
5780{
5781	struct ext4_ext_path *path1 = NULL;
5782	struct ext4_ext_path *path2 = NULL;
5783	int replaced_count = 0;
5784
5785	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5786	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5787	BUG_ON(!inode_is_locked(inode1));
5788	BUG_ON(!inode_is_locked(inode2));
5789
5790	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5791	if (unlikely(*erp))
5792		return 0;
5793	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5794	if (unlikely(*erp))
5795		return 0;
5796
5797	while (count) {
5798		struct ext4_extent *ex1, *ex2, tmp_ex;
5799		ext4_lblk_t e1_blk, e2_blk;
5800		int e1_len, e2_len, len;
5801		int split = 0;
5802
5803		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5804		if (IS_ERR(path1)) {
5805			*erp = PTR_ERR(path1);
5806			path1 = NULL;
5807		finish:
5808			count = 0;
5809			goto repeat;
5810		}
5811		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5812		if (IS_ERR(path2)) {
5813			*erp = PTR_ERR(path2);
5814			path2 = NULL;
5815			goto finish;
5816		}
5817		ex1 = path1[path1->p_depth].p_ext;
5818		ex2 = path2[path2->p_depth].p_ext;
5819		/* Do we have somthing to swap ? */
5820		if (unlikely(!ex2 || !ex1))
5821			goto finish;
5822
5823		e1_blk = le32_to_cpu(ex1->ee_block);
5824		e2_blk = le32_to_cpu(ex2->ee_block);
5825		e1_len = ext4_ext_get_actual_len(ex1);
5826		e2_len = ext4_ext_get_actual_len(ex2);
5827
5828		/* Hole handling */
5829		if (!in_range(lblk1, e1_blk, e1_len) ||
5830		    !in_range(lblk2, e2_blk, e2_len)) {
5831			ext4_lblk_t next1, next2;
5832
5833			/* if hole after extent, then go to next extent */
5834			next1 = ext4_ext_next_allocated_block(path1);
5835			next2 = ext4_ext_next_allocated_block(path2);
5836			/* If hole before extent, then shift to that extent */
5837			if (e1_blk > lblk1)
5838				next1 = e1_blk;
5839			if (e2_blk > lblk2)
5840				next2 = e1_blk;
5841			/* Do we have something to swap */
5842			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5843				goto finish;
5844			/* Move to the rightest boundary */
5845			len = next1 - lblk1;
5846			if (len < next2 - lblk2)
5847				len = next2 - lblk2;
5848			if (len > count)
5849				len = count;
5850			lblk1 += len;
5851			lblk2 += len;
5852			count -= len;
5853			goto repeat;
5854		}
5855
5856		/* Prepare left boundary */
5857		if (e1_blk < lblk1) {
5858			split = 1;
5859			*erp = ext4_force_split_extent_at(handle, inode1,
5860						&path1, lblk1, 0);
5861			if (unlikely(*erp))
5862				goto finish;
5863		}
5864		if (e2_blk < lblk2) {
5865			split = 1;
5866			*erp = ext4_force_split_extent_at(handle, inode2,
5867						&path2,  lblk2, 0);
5868			if (unlikely(*erp))
5869				goto finish;
5870		}
5871		/* ext4_split_extent_at() may result in leaf extent split,
5872		 * path must to be revalidated. */
5873		if (split)
5874			goto repeat;
5875
5876		/* Prepare right boundary */
5877		len = count;
5878		if (len > e1_blk + e1_len - lblk1)
5879			len = e1_blk + e1_len - lblk1;
5880		if (len > e2_blk + e2_len - lblk2)
5881			len = e2_blk + e2_len - lblk2;
5882
5883		if (len != e1_len) {
5884			split = 1;
5885			*erp = ext4_force_split_extent_at(handle, inode1,
5886						&path1, lblk1 + len, 0);
5887			if (unlikely(*erp))
5888				goto finish;
5889		}
5890		if (len != e2_len) {
5891			split = 1;
5892			*erp = ext4_force_split_extent_at(handle, inode2,
5893						&path2, lblk2 + len, 0);
5894			if (*erp)
5895				goto finish;
5896		}
5897		/* ext4_split_extent_at() may result in leaf extent split,
5898		 * path must to be revalidated. */
5899		if (split)
5900			goto repeat;
5901
5902		BUG_ON(e2_len != e1_len);
5903		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5904		if (unlikely(*erp))
5905			goto finish;
5906		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5907		if (unlikely(*erp))
5908			goto finish;
5909
5910		/* Both extents are fully inside boundaries. Swap it now */
5911		tmp_ex = *ex1;
5912		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5913		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5914		ex1->ee_len = cpu_to_le16(e2_len);
5915		ex2->ee_len = cpu_to_le16(e1_len);
5916		if (unwritten)
5917			ext4_ext_mark_unwritten(ex2);
5918		if (ext4_ext_is_unwritten(&tmp_ex))
5919			ext4_ext_mark_unwritten(ex1);
5920
5921		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5922		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5923		*erp = ext4_ext_dirty(handle, inode2, path2 +
5924				      path2->p_depth);
5925		if (unlikely(*erp))
5926			goto finish;
5927		*erp = ext4_ext_dirty(handle, inode1, path1 +
5928				      path1->p_depth);
5929		/*
5930		 * Looks scarry ah..? second inode already points to new blocks,
5931		 * and it was successfully dirtied. But luckily error may happen
5932		 * only due to journal error, so full transaction will be
5933		 * aborted anyway.
5934		 */
5935		if (unlikely(*erp))
5936			goto finish;
5937		lblk1 += len;
5938		lblk2 += len;
5939		replaced_count += len;
5940		count -= len;
5941
5942	repeat:
5943		ext4_ext_drop_refs(path1);
5944		kfree(path1);
5945		ext4_ext_drop_refs(path2);
5946		kfree(path2);
5947		path1 = path2 = NULL;
5948	}
5949	return replaced_count;
5950}