Linux Audio

Check our new training course

Linux BSP development engineering services

Need help to port Linux and bootloaders to your hardware?
Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   4 * Written by Alex Tomas <alex@clusterfs.com>
   5 *
   6 * Architecture independence:
   7 *   Copyright (c) 2005, Bull S.A.
   8 *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
 
 
 
 
 
 
 
 
 
 
 
 
 
   9 */
  10
  11/*
  12 * Extents support for EXT4
  13 *
  14 * TODO:
  15 *   - ext4*_error() should be used in some situations
  16 *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  17 *   - smart tree reduction
  18 */
  19
  20#include <linux/fs.h>
  21#include <linux/time.h>
  22#include <linux/jbd2.h>
  23#include <linux/highuid.h>
  24#include <linux/pagemap.h>
  25#include <linux/quotaops.h>
  26#include <linux/string.h>
  27#include <linux/slab.h>
  28#include <linux/uaccess.h>
 
  29#include <linux/fiemap.h>
  30#include <linux/iomap.h>
  31#include <linux/sched/mm.h>
  32#include "ext4_jbd2.h"
  33#include "ext4_extents.h"
  34#include "xattr.h"
  35
  36#include <trace/events/ext4.h>
  37
  38/*
  39 * used by extent splitting.
  40 */
  41#define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
  42					due to ENOSPC */
  43#define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
  44#define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
  45
  46#define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
  47#define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
  48
  49static __le32 ext4_extent_block_csum(struct inode *inode,
  50				     struct ext4_extent_header *eh)
  51{
  52	struct ext4_inode_info *ei = EXT4_I(inode);
  53	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  54	__u32 csum;
  55
  56	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  57			   EXT4_EXTENT_TAIL_OFFSET(eh));
  58	return cpu_to_le32(csum);
  59}
  60
  61static int ext4_extent_block_csum_verify(struct inode *inode,
  62					 struct ext4_extent_header *eh)
  63{
  64	struct ext4_extent_tail *et;
  65
  66	if (!ext4_has_metadata_csum(inode->i_sb))
 
  67		return 1;
  68
  69	et = find_ext4_extent_tail(eh);
  70	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  71		return 0;
  72	return 1;
  73}
  74
  75static void ext4_extent_block_csum_set(struct inode *inode,
  76				       struct ext4_extent_header *eh)
  77{
  78	struct ext4_extent_tail *et;
  79
  80	if (!ext4_has_metadata_csum(inode->i_sb))
 
  81		return;
  82
  83	et = find_ext4_extent_tail(eh);
  84	et->et_checksum = ext4_extent_block_csum(inode, eh);
  85}
  86
 
 
 
 
 
 
 
  87static int ext4_split_extent_at(handle_t *handle,
  88			     struct inode *inode,
  89			     struct ext4_ext_path **ppath,
  90			     ext4_lblk_t split,
  91			     int split_flag,
  92			     int flags);
  93
  94static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
  95{
  96	/*
  97	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
  98	 * moment, get_block can be called only for blocks inside i_size since
  99	 * page cache has been already dropped and writes are blocked by
 100	 * i_rwsem. So we can safely drop the i_data_sem here.
 101	 */
 102	BUG_ON(EXT4_JOURNAL(inode) == NULL);
 103	ext4_discard_preallocations(inode, 0);
 104	up_write(&EXT4_I(inode)->i_data_sem);
 105	*dropped = 1;
 106	return 0;
 107}
 108
 109static void ext4_ext_drop_refs(struct ext4_ext_path *path)
 110{
 111	int depth, i;
 112
 113	if (!path)
 114		return;
 115	depth = path->p_depth;
 116	for (i = 0; i <= depth; i++, path++) {
 117		brelse(path->p_bh);
 118		path->p_bh = NULL;
 119	}
 120}
 121
 122void ext4_free_ext_path(struct ext4_ext_path *path)
 123{
 124	ext4_ext_drop_refs(path);
 125	kfree(path);
 126}
 127
 128/*
 129 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
 130 * transaction with 'restart_cred' credits. The function drops i_data_sem
 131 * when restarting transaction and gets it after transaction is restarted.
 132 *
 133 * The function returns 0 on success, 1 if transaction had to be restarted,
 134 * and < 0 in case of fatal error.
 135 */
 136int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
 137				int check_cred, int restart_cred,
 138				int revoke_cred)
 139{
 140	int ret;
 141	int dropped = 0;
 142
 143	ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
 144		revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
 145	if (dropped)
 146		down_write(&EXT4_I(inode)->i_data_sem);
 147	return ret;
 148}
 149
 150/*
 151 * could return:
 152 *  - EROFS
 153 *  - ENOMEM
 154 */
 155static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 156				struct ext4_ext_path *path)
 157{
 158	int err = 0;
 159
 160	if (path->p_bh) {
 161		/* path points to block */
 162		BUFFER_TRACE(path->p_bh, "get_write_access");
 163		err = ext4_journal_get_write_access(handle, inode->i_sb,
 164						    path->p_bh, EXT4_JTR_NONE);
 165		/*
 166		 * The extent buffer's verified bit will be set again in
 167		 * __ext4_ext_dirty(). We could leave an inconsistent
 168		 * buffer if the extents updating procudure break off du
 169		 * to some error happens, force to check it again.
 170		 */
 171		if (!err)
 172			clear_buffer_verified(path->p_bh);
 173	}
 174	/* path points to leaf/index in inode body */
 175	/* we use in-core data, no need to protect them */
 176	return err;
 177}
 178
 179/*
 180 * could return:
 181 *  - EROFS
 182 *  - ENOMEM
 183 *  - EIO
 184 */
 
 
 185static int __ext4_ext_dirty(const char *where, unsigned int line,
 186			    handle_t *handle, struct inode *inode,
 187			    struct ext4_ext_path *path)
 188{
 189	int err;
 190
 191	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 192	if (path->p_bh) {
 193		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 194		/* path points to block */
 195		err = __ext4_handle_dirty_metadata(where, line, handle,
 196						   inode, path->p_bh);
 197		/* Extents updating done, re-set verified flag */
 198		if (!err)
 199			set_buffer_verified(path->p_bh);
 200	} else {
 201		/* path points to leaf/index in inode body */
 202		err = ext4_mark_inode_dirty(handle, inode);
 203	}
 204	return err;
 205}
 206
 207#define ext4_ext_dirty(handle, inode, path) \
 208		__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 209
 210static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 211			      struct ext4_ext_path *path,
 212			      ext4_lblk_t block)
 213{
 214	if (path) {
 215		int depth = path->p_depth;
 216		struct ext4_extent *ex;
 217
 218		/*
 219		 * Try to predict block placement assuming that we are
 220		 * filling in a file which will eventually be
 221		 * non-sparse --- i.e., in the case of libbfd writing
 222		 * an ELF object sections out-of-order but in a way
 223		 * the eventually results in a contiguous object or
 224		 * executable file, or some database extending a table
 225		 * space file.  However, this is actually somewhat
 226		 * non-ideal if we are writing a sparse file such as
 227		 * qemu or KVM writing a raw image file that is going
 228		 * to stay fairly sparse, since it will end up
 229		 * fragmenting the file system's free space.  Maybe we
 230		 * should have some hueristics or some way to allow
 231		 * userspace to pass a hint to file system,
 232		 * especially if the latter case turns out to be
 233		 * common.
 234		 */
 235		ex = path[depth].p_ext;
 236		if (ex) {
 237			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 238			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 239
 240			if (block > ext_block)
 241				return ext_pblk + (block - ext_block);
 242			else
 243				return ext_pblk - (ext_block - block);
 244		}
 245
 246		/* it looks like index is empty;
 247		 * try to find starting block from index itself */
 248		if (path[depth].p_bh)
 249			return path[depth].p_bh->b_blocknr;
 250	}
 251
 252	/* OK. use inode's group */
 253	return ext4_inode_to_goal_block(inode);
 254}
 255
 256/*
 257 * Allocation for a meta data block
 258 */
 259static ext4_fsblk_t
 260ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 261			struct ext4_ext_path *path,
 262			struct ext4_extent *ex, int *err, unsigned int flags)
 263{
 264	ext4_fsblk_t goal, newblock;
 265
 266	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 267	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 268					NULL, err);
 269	return newblock;
 270}
 271
 272static inline int ext4_ext_space_block(struct inode *inode, int check)
 273{
 274	int size;
 275
 276	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 277			/ sizeof(struct ext4_extent);
 278#ifdef AGGRESSIVE_TEST
 279	if (!check && size > 6)
 280		size = 6;
 281#endif
 282	return size;
 283}
 284
 285static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 286{
 287	int size;
 288
 289	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 290			/ sizeof(struct ext4_extent_idx);
 291#ifdef AGGRESSIVE_TEST
 292	if (!check && size > 5)
 293		size = 5;
 294#endif
 295	return size;
 296}
 297
 298static inline int ext4_ext_space_root(struct inode *inode, int check)
 299{
 300	int size;
 301
 302	size = sizeof(EXT4_I(inode)->i_data);
 303	size -= sizeof(struct ext4_extent_header);
 304	size /= sizeof(struct ext4_extent);
 305#ifdef AGGRESSIVE_TEST
 306	if (!check && size > 3)
 307		size = 3;
 308#endif
 309	return size;
 310}
 311
 312static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 313{
 314	int size;
 315
 316	size = sizeof(EXT4_I(inode)->i_data);
 317	size -= sizeof(struct ext4_extent_header);
 318	size /= sizeof(struct ext4_extent_idx);
 319#ifdef AGGRESSIVE_TEST
 320	if (!check && size > 4)
 321		size = 4;
 322#endif
 323	return size;
 324}
 325
 326static inline int
 327ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 328			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 329			   int nofail)
 330{
 331	struct ext4_ext_path *path = *ppath;
 332	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 333	int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
 334
 335	if (nofail)
 336		flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
 337
 338	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 339			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 340			flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 341}
 342
 343static int
 344ext4_ext_max_entries(struct inode *inode, int depth)
 345{
 346	int max;
 347
 348	if (depth == ext_depth(inode)) {
 349		if (depth == 0)
 350			max = ext4_ext_space_root(inode, 1);
 351		else
 352			max = ext4_ext_space_root_idx(inode, 1);
 353	} else {
 354		if (depth == 0)
 355			max = ext4_ext_space_block(inode, 1);
 356		else
 357			max = ext4_ext_space_block_idx(inode, 1);
 358	}
 359
 360	return max;
 361}
 362
 363static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 364{
 365	ext4_fsblk_t block = ext4_ext_pblock(ext);
 366	int len = ext4_ext_get_actual_len(ext);
 367	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 368
 369	/*
 370	 * We allow neither:
 371	 *  - zero length
 372	 *  - overflow/wrap-around
 373	 */
 374	if (lblock + len <= lblock)
 375		return 0;
 376	return ext4_inode_block_valid(inode, block, len);
 377}
 378
 379static int ext4_valid_extent_idx(struct inode *inode,
 380				struct ext4_extent_idx *ext_idx)
 381{
 382	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 383
 384	return ext4_inode_block_valid(inode, block, 1);
 385}
 386
 387static int ext4_valid_extent_entries(struct inode *inode,
 388				     struct ext4_extent_header *eh,
 389				     ext4_lblk_t lblk, ext4_fsblk_t *pblk,
 390				     int depth)
 391{
 392	unsigned short entries;
 393	ext4_lblk_t lblock = 0;
 394	ext4_lblk_t cur = 0;
 395
 396	if (eh->eh_entries == 0)
 397		return 1;
 398
 399	entries = le16_to_cpu(eh->eh_entries);
 400
 401	if (depth == 0) {
 402		/* leaf entries */
 403		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 404
 405		/*
 406		 * The logical block in the first entry should equal to
 407		 * the number in the index block.
 408		 */
 409		if (depth != ext_depth(inode) &&
 410		    lblk != le32_to_cpu(ext->ee_block))
 411			return 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			if (lblock < cur) {
 419				*pblk = ext4_ext_pblock(ext);
 420				return 0;
 421			}
 422			cur = lblock + ext4_ext_get_actual_len(ext);
 423			ext++;
 424			entries--;
 425		}
 426	} else {
 427		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 428
 429		/*
 430		 * The logical block in the first entry should equal to
 431		 * the number in the parent index block.
 432		 */
 433		if (depth != ext_depth(inode) &&
 434		    lblk != le32_to_cpu(ext_idx->ei_block))
 435			return 0;
 436		while (entries) {
 437			if (!ext4_valid_extent_idx(inode, ext_idx))
 438				return 0;
 439
 440			/* Check for overlapping index extents */
 441			lblock = le32_to_cpu(ext_idx->ei_block);
 442			if (lblock < cur) {
 443				*pblk = ext4_idx_pblock(ext_idx);
 444				return 0;
 445			}
 446			ext_idx++;
 447			entries--;
 448			cur = lblock + 1;
 449		}
 450	}
 451	return 1;
 452}
 453
 454static int __ext4_ext_check(const char *function, unsigned int line,
 455			    struct inode *inode, struct ext4_extent_header *eh,
 456			    int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
 457{
 458	const char *error_msg;
 459	int max = 0, err = -EFSCORRUPTED;
 460
 461	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 462		error_msg = "invalid magic";
 463		goto corrupted;
 464	}
 465	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 466		error_msg = "unexpected eh_depth";
 467		goto corrupted;
 468	}
 469	if (unlikely(eh->eh_max == 0)) {
 470		error_msg = "invalid eh_max";
 471		goto corrupted;
 472	}
 473	max = ext4_ext_max_entries(inode, depth);
 474	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 475		error_msg = "too large eh_max";
 476		goto corrupted;
 477	}
 478	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 479		error_msg = "invalid eh_entries";
 480		goto corrupted;
 481	}
 482	if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
 483		error_msg = "eh_entries is 0 but eh_depth is > 0";
 484		goto corrupted;
 485	}
 486	if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
 487		error_msg = "invalid extent entries";
 488		goto corrupted;
 489	}
 490	if (unlikely(depth > 32)) {
 491		error_msg = "too large eh_depth";
 492		goto corrupted;
 493	}
 494	/* Verify checksum on non-root extent tree nodes */
 495	if (ext_depth(inode) != depth &&
 496	    !ext4_extent_block_csum_verify(inode, eh)) {
 497		error_msg = "extent tree corrupted";
 498		err = -EFSBADCRC;
 499		goto corrupted;
 500	}
 501	return 0;
 502
 503corrupted:
 504	ext4_error_inode_err(inode, function, line, 0, -err,
 505			     "pblk %llu bad header/extent: %s - magic %x, "
 506			     "entries %u, max %u(%u), depth %u(%u)",
 507			     (unsigned long long) pblk, error_msg,
 508			     le16_to_cpu(eh->eh_magic),
 509			     le16_to_cpu(eh->eh_entries),
 510			     le16_to_cpu(eh->eh_max),
 511			     max, le16_to_cpu(eh->eh_depth), depth);
 512	return err;
 513}
 514
 515#define ext4_ext_check(inode, eh, depth, pblk)			\
 516	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
 517
 518int ext4_ext_check_inode(struct inode *inode)
 519{
 520	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 521}
 522
 523static void ext4_cache_extents(struct inode *inode,
 524			       struct ext4_extent_header *eh)
 525{
 526	struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 527	ext4_lblk_t prev = 0;
 528	int i;
 529
 530	for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 531		unsigned int status = EXTENT_STATUS_WRITTEN;
 532		ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 533		int len = ext4_ext_get_actual_len(ex);
 534
 535		if (prev && (prev != lblk))
 536			ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
 537					     EXTENT_STATUS_HOLE);
 538
 539		if (ext4_ext_is_unwritten(ex))
 540			status = EXTENT_STATUS_UNWRITTEN;
 541		ext4_es_cache_extent(inode, lblk, len,
 542				     ext4_ext_pblock(ex), status);
 543		prev = lblk + len;
 544	}
 545}
 546
 547static struct buffer_head *
 548__read_extent_tree_block(const char *function, unsigned int line,
 549			 struct inode *inode, struct ext4_extent_idx *idx,
 550			 int depth, int flags)
 551{
 552	struct buffer_head		*bh;
 553	int				err;
 554	gfp_t				gfp_flags = __GFP_MOVABLE | GFP_NOFS;
 555	ext4_fsblk_t			pblk;
 556
 557	if (flags & EXT4_EX_NOFAIL)
 558		gfp_flags |= __GFP_NOFAIL;
 559
 560	pblk = ext4_idx_pblock(idx);
 561	bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
 562	if (unlikely(!bh))
 563		return ERR_PTR(-ENOMEM);
 564
 565	if (!bh_uptodate_or_lock(bh)) {
 566		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 567		err = ext4_read_bh(bh, 0, NULL);
 568		if (err < 0)
 569			goto errout;
 570	}
 571	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 572		return bh;
 573	err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
 574			       depth, pblk, le32_to_cpu(idx->ei_block));
 575	if (err)
 576		goto errout;
 577	set_buffer_verified(bh);
 578	/*
 579	 * If this is a leaf block, cache all of its entries
 580	 */
 581	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 582		struct ext4_extent_header *eh = ext_block_hdr(bh);
 583		ext4_cache_extents(inode, eh);
 584	}
 585	return bh;
 586errout:
 587	put_bh(bh);
 588	return ERR_PTR(err);
 589
 590}
 591
 592#define read_extent_tree_block(inode, idx, depth, flags)		\
 593	__read_extent_tree_block(__func__, __LINE__, (inode), (idx),	\
 594				 (depth), (flags))
 595
 596/*
 597 * This function is called to cache a file's extent information in the
 598 * extent status tree
 599 */
 600int ext4_ext_precache(struct inode *inode)
 601{
 602	struct ext4_inode_info *ei = EXT4_I(inode);
 603	struct ext4_ext_path *path = NULL;
 604	struct buffer_head *bh;
 605	int i = 0, depth, ret = 0;
 606
 607	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 608		return 0;	/* not an extent-mapped inode */
 609
 610	down_read(&ei->i_data_sem);
 611	depth = ext_depth(inode);
 612
 613	/* Don't cache anything if there are no external extent blocks */
 614	if (!depth) {
 615		up_read(&ei->i_data_sem);
 616		return ret;
 617	}
 618
 619	path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
 620		       GFP_NOFS);
 621	if (path == NULL) {
 622		up_read(&ei->i_data_sem);
 623		return -ENOMEM;
 624	}
 625
 626	path[0].p_hdr = ext_inode_hdr(inode);
 627	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 
 628	if (ret)
 629		goto out;
 630	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 631	while (i >= 0) {
 632		/*
 633		 * If this is a leaf block or we've reached the end of
 634		 * the index block, go up
 635		 */
 636		if ((i == depth) ||
 637		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 638			brelse(path[i].p_bh);
 639			path[i].p_bh = NULL;
 640			i--;
 641			continue;
 642		}
 643		bh = read_extent_tree_block(inode, path[i].p_idx++,
 644					    depth - i - 1,
 645					    EXT4_EX_FORCE_CACHE);
 646		if (IS_ERR(bh)) {
 647			ret = PTR_ERR(bh);
 648			break;
 649		}
 650		i++;
 651		path[i].p_bh = bh;
 652		path[i].p_hdr = ext_block_hdr(bh);
 653		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 654	}
 655	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 656out:
 657	up_read(&ei->i_data_sem);
 658	ext4_free_ext_path(path);
 659	return ret;
 660}
 661
 
 
 
 662#ifdef EXT_DEBUG
 663static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 664{
 665	int k, l = path->p_depth;
 666
 667	ext_debug(inode, "path:");
 668	for (k = 0; k <= l; k++, path++) {
 669		if (path->p_idx) {
 670			ext_debug(inode, "  %d->%llu",
 671				  le32_to_cpu(path->p_idx->ei_block),
 672				  ext4_idx_pblock(path->p_idx));
 673		} else if (path->p_ext) {
 674			ext_debug(inode, "  %d:[%d]%d:%llu ",
 675				  le32_to_cpu(path->p_ext->ee_block),
 676				  ext4_ext_is_unwritten(path->p_ext),
 677				  ext4_ext_get_actual_len(path->p_ext),
 678				  ext4_ext_pblock(path->p_ext));
 679		} else
 680			ext_debug(inode, "  []");
 681	}
 682	ext_debug(inode, "\n");
 683}
 684
 685static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 686{
 687	int depth = ext_depth(inode);
 688	struct ext4_extent_header *eh;
 689	struct ext4_extent *ex;
 690	int i;
 691
 692	if (!path)
 693		return;
 694
 695	eh = path[depth].p_hdr;
 696	ex = EXT_FIRST_EXTENT(eh);
 697
 698	ext_debug(inode, "Displaying leaf extents\n");
 699
 700	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 701		ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 702			  ext4_ext_is_unwritten(ex),
 703			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 704	}
 705	ext_debug(inode, "\n");
 706}
 707
 708static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 709			ext4_fsblk_t newblock, int level)
 710{
 711	int depth = ext_depth(inode);
 712	struct ext4_extent *ex;
 713
 714	if (depth != level) {
 715		struct ext4_extent_idx *idx;
 716		idx = path[level].p_idx;
 717		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 718			ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
 719				  level, le32_to_cpu(idx->ei_block),
 720				  ext4_idx_pblock(idx), newblock);
 
 721			idx++;
 722		}
 723
 724		return;
 725	}
 726
 727	ex = path[depth].p_ext;
 728	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 729		ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
 730				le32_to_cpu(ex->ee_block),
 731				ext4_ext_pblock(ex),
 732				ext4_ext_is_unwritten(ex),
 733				ext4_ext_get_actual_len(ex),
 734				newblock);
 735		ex++;
 736	}
 737}
 738
 739#else
 740#define ext4_ext_show_path(inode, path)
 741#define ext4_ext_show_leaf(inode, path)
 742#define ext4_ext_show_move(inode, path, newblock, level)
 743#endif
 744
 
 
 
 
 
 
 
 
 
 
 
 
 745/*
 746 * ext4_ext_binsearch_idx:
 747 * binary search for the closest index of the given block
 748 * the header must be checked before calling this
 749 */
 750static void
 751ext4_ext_binsearch_idx(struct inode *inode,
 752			struct ext4_ext_path *path, ext4_lblk_t block)
 753{
 754	struct ext4_extent_header *eh = path->p_hdr;
 755	struct ext4_extent_idx *r, *l, *m;
 756
 757
 758	ext_debug(inode, "binsearch for %u(idx):  ", block);
 759
 760	l = EXT_FIRST_INDEX(eh) + 1;
 761	r = EXT_LAST_INDEX(eh);
 762	while (l <= r) {
 763		m = l + (r - l) / 2;
 764		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 765			  le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
 766			  r, le32_to_cpu(r->ei_block));
 767
 768		if (block < le32_to_cpu(m->ei_block))
 769			r = m - 1;
 770		else
 771			l = m + 1;
 
 
 
 772	}
 773
 774	path->p_idx = l - 1;
 775	ext_debug(inode, "  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 776		  ext4_idx_pblock(path->p_idx));
 777
 778#ifdef CHECK_BINSEARCH
 779	{
 780		struct ext4_extent_idx *chix, *ix;
 781		int k;
 782
 783		chix = ix = EXT_FIRST_INDEX(eh);
 784		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 785			if (k != 0 && le32_to_cpu(ix->ei_block) <=
 786			    le32_to_cpu(ix[-1].ei_block)) {
 787				printk(KERN_DEBUG "k=%d, ix=0x%p, "
 788				       "first=0x%p\n", k,
 789				       ix, EXT_FIRST_INDEX(eh));
 790				printk(KERN_DEBUG "%u <= %u\n",
 791				       le32_to_cpu(ix->ei_block),
 792				       le32_to_cpu(ix[-1].ei_block));
 793			}
 794			BUG_ON(k && le32_to_cpu(ix->ei_block)
 795					   <= le32_to_cpu(ix[-1].ei_block));
 796			if (block < le32_to_cpu(ix->ei_block))
 797				break;
 798			chix = ix;
 799		}
 800		BUG_ON(chix != path->p_idx);
 801	}
 802#endif
 803
 804}
 805
 806/*
 807 * ext4_ext_binsearch:
 808 * binary search for closest extent of the given block
 809 * the header must be checked before calling this
 810 */
 811static void
 812ext4_ext_binsearch(struct inode *inode,
 813		struct ext4_ext_path *path, ext4_lblk_t block)
 814{
 815	struct ext4_extent_header *eh = path->p_hdr;
 816	struct ext4_extent *r, *l, *m;
 817
 818	if (eh->eh_entries == 0) {
 819		/*
 820		 * this leaf is empty:
 821		 * we get such a leaf in split/add case
 822		 */
 823		return;
 824	}
 825
 826	ext_debug(inode, "binsearch for %u:  ", block);
 827
 828	l = EXT_FIRST_EXTENT(eh) + 1;
 829	r = EXT_LAST_EXTENT(eh);
 830
 831	while (l <= r) {
 832		m = l + (r - l) / 2;
 833		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 834			  le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
 835			  r, le32_to_cpu(r->ee_block));
 836
 837		if (block < le32_to_cpu(m->ee_block))
 838			r = m - 1;
 839		else
 840			l = m + 1;
 
 
 
 841	}
 842
 843	path->p_ext = l - 1;
 844	ext_debug(inode, "  -> %d:%llu:[%d]%d ",
 845			le32_to_cpu(path->p_ext->ee_block),
 846			ext4_ext_pblock(path->p_ext),
 847			ext4_ext_is_unwritten(path->p_ext),
 848			ext4_ext_get_actual_len(path->p_ext));
 849
 850#ifdef CHECK_BINSEARCH
 851	{
 852		struct ext4_extent *chex, *ex;
 853		int k;
 854
 855		chex = ex = EXT_FIRST_EXTENT(eh);
 856		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 857			BUG_ON(k && le32_to_cpu(ex->ee_block)
 858					  <= le32_to_cpu(ex[-1].ee_block));
 859			if (block < le32_to_cpu(ex->ee_block))
 860				break;
 861			chex = ex;
 862		}
 863		BUG_ON(chex != path->p_ext);
 864	}
 865#endif
 866
 867}
 868
 869void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 870{
 871	struct ext4_extent_header *eh;
 872
 873	eh = ext_inode_hdr(inode);
 874	eh->eh_depth = 0;
 875	eh->eh_entries = 0;
 876	eh->eh_magic = EXT4_EXT_MAGIC;
 877	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 878	eh->eh_generation = 0;
 879	ext4_mark_inode_dirty(handle, inode);
 
 
 880}
 881
 882struct ext4_ext_path *
 883ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 884		 struct ext4_ext_path **orig_path, int flags)
 885{
 886	struct ext4_extent_header *eh;
 887	struct buffer_head *bh;
 888	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 889	short int depth, i, ppos = 0;
 890	int ret;
 891	gfp_t gfp_flags = GFP_NOFS;
 892
 893	if (flags & EXT4_EX_NOFAIL)
 894		gfp_flags |= __GFP_NOFAIL;
 895
 896	eh = ext_inode_hdr(inode);
 897	depth = ext_depth(inode);
 898	if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
 899		EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
 900				 depth);
 901		ret = -EFSCORRUPTED;
 902		goto err;
 903	}
 904
 905	if (path) {
 906		ext4_ext_drop_refs(path);
 907		if (depth > path[0].p_maxdepth) {
 908			kfree(path);
 909			*orig_path = path = NULL;
 910		}
 911	}
 912	if (!path) {
 913		/* account possible depth increase */
 914		path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
 915				gfp_flags);
 916		if (unlikely(!path))
 917			return ERR_PTR(-ENOMEM);
 918		path[0].p_maxdepth = depth + 1;
 919	}
 920	path[0].p_hdr = eh;
 921	path[0].p_bh = NULL;
 922
 923	i = depth;
 924	if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
 925		ext4_cache_extents(inode, eh);
 926	/* walk through the tree */
 927	while (i) {
 928		ext_debug(inode, "depth %d: num %d, max %d\n",
 929			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 930
 931		ext4_ext_binsearch_idx(inode, path + ppos, block);
 932		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 933		path[ppos].p_depth = i;
 934		path[ppos].p_ext = NULL;
 935
 936		bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
 937		if (IS_ERR(bh)) {
 938			ret = PTR_ERR(bh);
 939			goto err;
 
 
 
 
 
 
 
 940		}
 941
 942		eh = ext_block_hdr(bh);
 943		ppos++;
 
 
 
 
 
 
 944		path[ppos].p_bh = bh;
 945		path[ppos].p_hdr = eh;
 
 
 
 
 946	}
 947
 948	path[ppos].p_depth = i;
 949	path[ppos].p_ext = NULL;
 950	path[ppos].p_idx = NULL;
 951
 952	/* find extent */
 953	ext4_ext_binsearch(inode, path + ppos, block);
 954	/* if not an empty leaf */
 955	if (path[ppos].p_ext)
 956		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 957
 958	ext4_ext_show_path(inode, path);
 959
 960	return path;
 961
 962err:
 963	ext4_free_ext_path(path);
 964	if (orig_path)
 965		*orig_path = NULL;
 966	return ERR_PTR(ret);
 967}
 968
 969/*
 970 * ext4_ext_insert_index:
 971 * insert new index [@logical;@ptr] into the block at @curp;
 972 * check where to insert: before @curp or after @curp
 973 */
 974static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 975				 struct ext4_ext_path *curp,
 976				 int logical, ext4_fsblk_t ptr)
 977{
 978	struct ext4_extent_idx *ix;
 979	int len, err;
 980
 981	err = ext4_ext_get_access(handle, inode, curp);
 982	if (err)
 983		return err;
 984
 985	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 986		EXT4_ERROR_INODE(inode,
 987				 "logical %d == ei_block %d!",
 988				 logical, le32_to_cpu(curp->p_idx->ei_block));
 989		return -EFSCORRUPTED;
 990	}
 991
 992	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 993			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
 994		EXT4_ERROR_INODE(inode,
 995				 "eh_entries %d >= eh_max %d!",
 996				 le16_to_cpu(curp->p_hdr->eh_entries),
 997				 le16_to_cpu(curp->p_hdr->eh_max));
 998		return -EFSCORRUPTED;
 999	}
1000
1001	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
1002		/* insert after */
1003		ext_debug(inode, "insert new index %d after: %llu\n",
1004			  logical, ptr);
1005		ix = curp->p_idx + 1;
1006	} else {
1007		/* insert before */
1008		ext_debug(inode, "insert new index %d before: %llu\n",
1009			  logical, ptr);
1010		ix = curp->p_idx;
1011	}
1012
1013	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1014	BUG_ON(len < 0);
1015	if (len > 0) {
1016		ext_debug(inode, "insert new index %d: "
1017				"move %d indices from 0x%p to 0x%p\n",
1018				logical, len, ix, ix + 1);
1019		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1020	}
1021
1022	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1023		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1024		return -EFSCORRUPTED;
1025	}
1026
1027	ix->ei_block = cpu_to_le32(logical);
1028	ext4_idx_store_pblock(ix, ptr);
1029	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1030
1031	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1032		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1033		return -EFSCORRUPTED;
1034	}
1035
1036	err = ext4_ext_dirty(handle, inode, curp);
1037	ext4_std_error(inode->i_sb, err);
1038
1039	return err;
1040}
1041
1042/*
1043 * ext4_ext_split:
1044 * inserts new subtree into the path, using free index entry
1045 * at depth @at:
1046 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1047 * - makes decision where to split
1048 * - moves remaining extents and index entries (right to the split point)
1049 *   into the newly allocated blocks
1050 * - initializes subtree
1051 */
1052static int ext4_ext_split(handle_t *handle, struct inode *inode,
1053			  unsigned int flags,
1054			  struct ext4_ext_path *path,
1055			  struct ext4_extent *newext, int at)
1056{
1057	struct buffer_head *bh = NULL;
1058	int depth = ext_depth(inode);
1059	struct ext4_extent_header *neh;
1060	struct ext4_extent_idx *fidx;
1061	int i = at, k, m, a;
1062	ext4_fsblk_t newblock, oldblock;
1063	__le32 border;
1064	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1065	gfp_t gfp_flags = GFP_NOFS;
1066	int err = 0;
1067	size_t ext_size = 0;
1068
1069	if (flags & EXT4_EX_NOFAIL)
1070		gfp_flags |= __GFP_NOFAIL;
1071
1072	/* make decision: where to split? */
1073	/* FIXME: now decision is simplest: at current extent */
1074
1075	/* if current leaf will be split, then we should use
1076	 * border from split point */
1077	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1078		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1079		return -EFSCORRUPTED;
1080	}
1081	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1082		border = path[depth].p_ext[1].ee_block;
1083		ext_debug(inode, "leaf will be split."
1084				" next leaf starts at %d\n",
1085				  le32_to_cpu(border));
1086	} else {
1087		border = newext->ee_block;
1088		ext_debug(inode, "leaf will be added."
1089				" next leaf starts at %d\n",
1090				le32_to_cpu(border));
1091	}
1092
1093	/*
1094	 * If error occurs, then we break processing
1095	 * and mark filesystem read-only. index won't
1096	 * be inserted and tree will be in consistent
1097	 * state. Next mount will repair buffers too.
1098	 */
1099
1100	/*
1101	 * Get array to track all allocated blocks.
1102	 * We need this to handle errors and free blocks
1103	 * upon them.
1104	 */
1105	ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1106	if (!ablocks)
1107		return -ENOMEM;
1108
1109	/* allocate all needed blocks */
1110	ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1111	for (a = 0; a < depth - at; a++) {
1112		newblock = ext4_ext_new_meta_block(handle, inode, path,
1113						   newext, &err, flags);
1114		if (newblock == 0)
1115			goto cleanup;
1116		ablocks[a] = newblock;
1117	}
1118
1119	/* initialize new leaf */
1120	newblock = ablocks[--a];
1121	if (unlikely(newblock == 0)) {
1122		EXT4_ERROR_INODE(inode, "newblock == 0!");
1123		err = -EFSCORRUPTED;
1124		goto cleanup;
1125	}
1126	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1127	if (unlikely(!bh)) {
1128		err = -ENOMEM;
1129		goto cleanup;
1130	}
1131	lock_buffer(bh);
1132
1133	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1134					     EXT4_JTR_NONE);
1135	if (err)
1136		goto cleanup;
1137
1138	neh = ext_block_hdr(bh);
1139	neh->eh_entries = 0;
1140	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1141	neh->eh_magic = EXT4_EXT_MAGIC;
1142	neh->eh_depth = 0;
1143	neh->eh_generation = 0;
1144
1145	/* move remainder of path[depth] to the new leaf */
1146	if (unlikely(path[depth].p_hdr->eh_entries !=
1147		     path[depth].p_hdr->eh_max)) {
1148		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1149				 path[depth].p_hdr->eh_entries,
1150				 path[depth].p_hdr->eh_max);
1151		err = -EFSCORRUPTED;
1152		goto cleanup;
1153	}
1154	/* start copy from next extent */
1155	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1156	ext4_ext_show_move(inode, path, newblock, depth);
1157	if (m) {
1158		struct ext4_extent *ex;
1159		ex = EXT_FIRST_EXTENT(neh);
1160		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1161		le16_add_cpu(&neh->eh_entries, m);
1162	}
1163
1164	/* zero out unused area in the extent block */
1165	ext_size = sizeof(struct ext4_extent_header) +
1166		sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1167	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1168	ext4_extent_block_csum_set(inode, neh);
1169	set_buffer_uptodate(bh);
1170	unlock_buffer(bh);
1171
1172	err = ext4_handle_dirty_metadata(handle, inode, bh);
1173	if (err)
1174		goto cleanup;
1175	brelse(bh);
1176	bh = NULL;
1177
1178	/* correct old leaf */
1179	if (m) {
1180		err = ext4_ext_get_access(handle, inode, path + depth);
1181		if (err)
1182			goto cleanup;
1183		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1184		err = ext4_ext_dirty(handle, inode, path + depth);
1185		if (err)
1186			goto cleanup;
1187
1188	}
1189
1190	/* create intermediate indexes */
1191	k = depth - at - 1;
1192	if (unlikely(k < 0)) {
1193		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1194		err = -EFSCORRUPTED;
1195		goto cleanup;
1196	}
1197	if (k)
1198		ext_debug(inode, "create %d intermediate indices\n", k);
1199	/* insert new index into current index block */
1200	/* current depth stored in i var */
1201	i = depth - 1;
1202	while (k--) {
1203		oldblock = newblock;
1204		newblock = ablocks[--a];
1205		bh = sb_getblk(inode->i_sb, newblock);
1206		if (unlikely(!bh)) {
1207			err = -ENOMEM;
1208			goto cleanup;
1209		}
1210		lock_buffer(bh);
1211
1212		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1213						     EXT4_JTR_NONE);
1214		if (err)
1215			goto cleanup;
1216
1217		neh = ext_block_hdr(bh);
1218		neh->eh_entries = cpu_to_le16(1);
1219		neh->eh_magic = EXT4_EXT_MAGIC;
1220		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1221		neh->eh_depth = cpu_to_le16(depth - i);
1222		neh->eh_generation = 0;
1223		fidx = EXT_FIRST_INDEX(neh);
1224		fidx->ei_block = border;
1225		ext4_idx_store_pblock(fidx, oldblock);
1226
1227		ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1228				i, newblock, le32_to_cpu(border), oldblock);
1229
1230		/* move remainder of path[i] to the new index block */
1231		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1232					EXT_LAST_INDEX(path[i].p_hdr))) {
1233			EXT4_ERROR_INODE(inode,
1234					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1235					 le32_to_cpu(path[i].p_ext->ee_block));
1236			err = -EFSCORRUPTED;
1237			goto cleanup;
1238		}
1239		/* start copy indexes */
1240		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1241		ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1242				EXT_MAX_INDEX(path[i].p_hdr));
1243		ext4_ext_show_move(inode, path, newblock, i);
1244		if (m) {
1245			memmove(++fidx, path[i].p_idx,
1246				sizeof(struct ext4_extent_idx) * m);
1247			le16_add_cpu(&neh->eh_entries, m);
1248		}
1249		/* zero out unused area in the extent block */
1250		ext_size = sizeof(struct ext4_extent_header) +
1251		   (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1252		memset(bh->b_data + ext_size, 0,
1253			inode->i_sb->s_blocksize - ext_size);
1254		ext4_extent_block_csum_set(inode, neh);
1255		set_buffer_uptodate(bh);
1256		unlock_buffer(bh);
1257
1258		err = ext4_handle_dirty_metadata(handle, inode, bh);
1259		if (err)
1260			goto cleanup;
1261		brelse(bh);
1262		bh = NULL;
1263
1264		/* correct old index */
1265		if (m) {
1266			err = ext4_ext_get_access(handle, inode, path + i);
1267			if (err)
1268				goto cleanup;
1269			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1270			err = ext4_ext_dirty(handle, inode, path + i);
1271			if (err)
1272				goto cleanup;
1273		}
1274
1275		i--;
1276	}
1277
1278	/* insert new index */
1279	err = ext4_ext_insert_index(handle, inode, path + at,
1280				    le32_to_cpu(border), newblock);
1281
1282cleanup:
1283	if (bh) {
1284		if (buffer_locked(bh))
1285			unlock_buffer(bh);
1286		brelse(bh);
1287	}
1288
1289	if (err) {
1290		/* free all allocated blocks in error case */
1291		for (i = 0; i < depth; i++) {
1292			if (!ablocks[i])
1293				continue;
1294			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1295					 EXT4_FREE_BLOCKS_METADATA);
1296		}
1297	}
1298	kfree(ablocks);
1299
1300	return err;
1301}
1302
1303/*
1304 * ext4_ext_grow_indepth:
1305 * implements tree growing procedure:
1306 * - allocates new block
1307 * - moves top-level data (index block or leaf) into the new block
1308 * - initializes new top-level, creating index that points to the
1309 *   just created block
1310 */
1311static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1312				 unsigned int flags)
 
1313{
1314	struct ext4_extent_header *neh;
1315	struct buffer_head *bh;
1316	ext4_fsblk_t newblock, goal = 0;
1317	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1318	int err = 0;
1319	size_t ext_size = 0;
1320
1321	/* Try to prepend new index to old one */
1322	if (ext_depth(inode))
1323		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1324	if (goal > le32_to_cpu(es->s_first_data_block)) {
1325		flags |= EXT4_MB_HINT_TRY_GOAL;
1326		goal--;
1327	} else
1328		goal = ext4_inode_to_goal_block(inode);
1329	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1330					NULL, &err);
1331	if (newblock == 0)
1332		return err;
1333
1334	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1335	if (unlikely(!bh))
1336		return -ENOMEM;
 
 
 
1337	lock_buffer(bh);
1338
1339	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1340					     EXT4_JTR_NONE);
1341	if (err) {
1342		unlock_buffer(bh);
1343		goto out;
1344	}
1345
1346	ext_size = sizeof(EXT4_I(inode)->i_data);
1347	/* move top-level index/leaf into new block */
1348	memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1349	/* zero out unused area in the extent block */
1350	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1351
1352	/* set size of new block */
1353	neh = ext_block_hdr(bh);
1354	/* old root could have indexes or leaves
1355	 * so calculate e_max right way */
1356	if (ext_depth(inode))
1357		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1358	else
1359		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1360	neh->eh_magic = EXT4_EXT_MAGIC;
1361	ext4_extent_block_csum_set(inode, neh);
1362	set_buffer_uptodate(bh);
1363	set_buffer_verified(bh);
1364	unlock_buffer(bh);
1365
1366	err = ext4_handle_dirty_metadata(handle, inode, bh);
1367	if (err)
1368		goto out;
1369
1370	/* Update top-level index: num,max,pointer */
1371	neh = ext_inode_hdr(inode);
1372	neh->eh_entries = cpu_to_le16(1);
1373	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1374	if (neh->eh_depth == 0) {
1375		/* Root extent block becomes index block */
1376		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1377		EXT_FIRST_INDEX(neh)->ei_block =
1378			EXT_FIRST_EXTENT(neh)->ee_block;
1379	}
1380	ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1381		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1382		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1383		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1384
1385	le16_add_cpu(&neh->eh_depth, 1);
1386	err = ext4_mark_inode_dirty(handle, inode);
1387out:
1388	brelse(bh);
1389
1390	return err;
1391}
1392
1393/*
1394 * ext4_ext_create_new_leaf:
1395 * finds empty index and adds new leaf.
1396 * if no free index is found, then it requests in-depth growing.
1397 */
1398static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1399				    unsigned int mb_flags,
1400				    unsigned int gb_flags,
1401				    struct ext4_ext_path **ppath,
1402				    struct ext4_extent *newext)
1403{
1404	struct ext4_ext_path *path = *ppath;
1405	struct ext4_ext_path *curp;
1406	int depth, i, err = 0;
1407
1408repeat:
1409	i = depth = ext_depth(inode);
1410
1411	/* walk up to the tree and look for free index entry */
1412	curp = path + depth;
1413	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1414		i--;
1415		curp--;
1416	}
1417
1418	/* we use already allocated block for index block,
1419	 * so subsequent data blocks should be contiguous */
1420	if (EXT_HAS_FREE_INDEX(curp)) {
1421		/* if we found index with free entry, then use that
1422		 * entry: create all needed subtree and add new leaf */
1423		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1424		if (err)
1425			goto out;
1426
1427		/* refill path */
1428		path = ext4_find_extent(inode,
 
1429				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1430				    ppath, gb_flags);
1431		if (IS_ERR(path))
1432			err = PTR_ERR(path);
1433	} else {
1434		/* tree is full, time to grow in depth */
1435		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1436		if (err)
1437			goto out;
1438
1439		/* refill path */
1440		path = ext4_find_extent(inode,
 
1441				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1442				    ppath, gb_flags);
1443		if (IS_ERR(path)) {
1444			err = PTR_ERR(path);
1445			goto out;
1446		}
1447
1448		/*
1449		 * only first (depth 0 -> 1) produces free space;
1450		 * in all other cases we have to split the grown tree
1451		 */
1452		depth = ext_depth(inode);
1453		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1454			/* now we need to split */
1455			goto repeat;
1456		}
1457	}
1458
1459out:
1460	return err;
1461}
1462
1463/*
1464 * search the closest allocated block to the left for *logical
1465 * and returns it at @logical + it's physical address at @phys
1466 * if *logical is the smallest allocated block, the function
1467 * returns 0 at @phys
1468 * return value contains 0 (success) or error code
1469 */
1470static int ext4_ext_search_left(struct inode *inode,
1471				struct ext4_ext_path *path,
1472				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1473{
1474	struct ext4_extent_idx *ix;
1475	struct ext4_extent *ex;
1476	int depth, ee_len;
1477
1478	if (unlikely(path == NULL)) {
1479		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1480		return -EFSCORRUPTED;
1481	}
1482	depth = path->p_depth;
1483	*phys = 0;
1484
1485	if (depth == 0 && path->p_ext == NULL)
1486		return 0;
1487
1488	/* usually extent in the path covers blocks smaller
1489	 * then *logical, but it can be that extent is the
1490	 * first one in the file */
1491
1492	ex = path[depth].p_ext;
1493	ee_len = ext4_ext_get_actual_len(ex);
1494	if (*logical < le32_to_cpu(ex->ee_block)) {
1495		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1496			EXT4_ERROR_INODE(inode,
1497					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1498					 *logical, le32_to_cpu(ex->ee_block));
1499			return -EFSCORRUPTED;
1500		}
1501		while (--depth >= 0) {
1502			ix = path[depth].p_idx;
1503			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1504				EXT4_ERROR_INODE(inode,
1505				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1506				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1507				  le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block),
 
1508				  depth);
1509				return -EFSCORRUPTED;
1510			}
1511		}
1512		return 0;
1513	}
1514
1515	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1516		EXT4_ERROR_INODE(inode,
1517				 "logical %d < ee_block %d + ee_len %d!",
1518				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1519		return -EFSCORRUPTED;
1520	}
1521
1522	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1523	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1524	return 0;
1525}
1526
1527/*
1528 * Search the closest allocated block to the right for *logical
1529 * and returns it at @logical + it's physical address at @phys.
1530 * If not exists, return 0 and @phys is set to 0. We will return
1531 * 1 which means we found an allocated block and ret_ex is valid.
1532 * Or return a (< 0) error code.
1533 */
1534static int ext4_ext_search_right(struct inode *inode,
1535				 struct ext4_ext_path *path,
1536				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1537				 struct ext4_extent *ret_ex)
1538{
1539	struct buffer_head *bh = NULL;
1540	struct ext4_extent_header *eh;
1541	struct ext4_extent_idx *ix;
1542	struct ext4_extent *ex;
 
1543	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1544	int ee_len;
1545
1546	if (unlikely(path == NULL)) {
1547		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1548		return -EFSCORRUPTED;
1549	}
1550	depth = path->p_depth;
1551	*phys = 0;
1552
1553	if (depth == 0 && path->p_ext == NULL)
1554		return 0;
1555
1556	/* usually extent in the path covers blocks smaller
1557	 * then *logical, but it can be that extent is the
1558	 * first one in the file */
1559
1560	ex = path[depth].p_ext;
1561	ee_len = ext4_ext_get_actual_len(ex);
1562	if (*logical < le32_to_cpu(ex->ee_block)) {
1563		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1564			EXT4_ERROR_INODE(inode,
1565					 "first_extent(path[%d].p_hdr) != ex",
1566					 depth);
1567			return -EFSCORRUPTED;
1568		}
1569		while (--depth >= 0) {
1570			ix = path[depth].p_idx;
1571			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1572				EXT4_ERROR_INODE(inode,
1573						 "ix != EXT_FIRST_INDEX *logical %d!",
1574						 *logical);
1575				return -EFSCORRUPTED;
1576			}
1577		}
1578		goto found_extent;
1579	}
1580
1581	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1582		EXT4_ERROR_INODE(inode,
1583				 "logical %d < ee_block %d + ee_len %d!",
1584				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1585		return -EFSCORRUPTED;
1586	}
1587
1588	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1589		/* next allocated block in this leaf */
1590		ex++;
1591		goto found_extent;
1592	}
1593
1594	/* go up and search for index to the right */
1595	while (--depth >= 0) {
1596		ix = path[depth].p_idx;
1597		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1598			goto got_index;
1599	}
1600
1601	/* we've gone up to the root and found no index to the right */
1602	return 0;
1603
1604got_index:
1605	/* we've found index to the right, let's
1606	 * follow it and find the closest allocated
1607	 * block to the right */
1608	ix++;
 
1609	while (++depth < path->p_depth) {
1610		/* subtract from p_depth to get proper eh_depth */
1611		bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1612		if (IS_ERR(bh))
1613			return PTR_ERR(bh);
1614		eh = ext_block_hdr(bh);
 
 
 
 
 
 
1615		ix = EXT_FIRST_INDEX(eh);
 
1616		put_bh(bh);
1617	}
1618
1619	bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1620	if (IS_ERR(bh))
1621		return PTR_ERR(bh);
1622	eh = ext_block_hdr(bh);
 
 
 
 
1623	ex = EXT_FIRST_EXTENT(eh);
1624found_extent:
1625	*logical = le32_to_cpu(ex->ee_block);
1626	*phys = ext4_ext_pblock(ex);
1627	if (ret_ex)
1628		*ret_ex = *ex;
1629	if (bh)
1630		put_bh(bh);
1631	return 1;
1632}
1633
1634/*
1635 * ext4_ext_next_allocated_block:
1636 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1637 * NOTE: it considers block number from index entry as
1638 * allocated block. Thus, index entries have to be consistent
1639 * with leaves.
1640 */
1641ext4_lblk_t
1642ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1643{
1644	int depth;
1645
1646	BUG_ON(path == NULL);
1647	depth = path->p_depth;
1648
1649	if (depth == 0 && path->p_ext == NULL)
1650		return EXT_MAX_BLOCKS;
1651
1652	while (depth >= 0) {
1653		struct ext4_ext_path *p = &path[depth];
1654
1655		if (depth == path->p_depth) {
1656			/* leaf */
1657			if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1658				return le32_to_cpu(p->p_ext[1].ee_block);
 
 
1659		} else {
1660			/* index */
1661			if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1662				return le32_to_cpu(p->p_idx[1].ei_block);
 
1663		}
1664		depth--;
1665	}
1666
1667	return EXT_MAX_BLOCKS;
1668}
1669
1670/*
1671 * ext4_ext_next_leaf_block:
1672 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1673 */
1674static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1675{
1676	int depth;
1677
1678	BUG_ON(path == NULL);
1679	depth = path->p_depth;
1680
1681	/* zero-tree has no leaf blocks at all */
1682	if (depth == 0)
1683		return EXT_MAX_BLOCKS;
1684
1685	/* go to index block */
1686	depth--;
1687
1688	while (depth >= 0) {
1689		if (path[depth].p_idx !=
1690				EXT_LAST_INDEX(path[depth].p_hdr))
1691			return (ext4_lblk_t)
1692				le32_to_cpu(path[depth].p_idx[1].ei_block);
1693		depth--;
1694	}
1695
1696	return EXT_MAX_BLOCKS;
1697}
1698
1699/*
1700 * ext4_ext_correct_indexes:
1701 * if leaf gets modified and modified extent is first in the leaf,
1702 * then we have to correct all indexes above.
1703 * TODO: do we need to correct tree in all cases?
1704 */
1705static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1706				struct ext4_ext_path *path)
1707{
1708	struct ext4_extent_header *eh;
1709	int depth = ext_depth(inode);
1710	struct ext4_extent *ex;
1711	__le32 border;
1712	int k, err = 0;
1713
1714	eh = path[depth].p_hdr;
1715	ex = path[depth].p_ext;
1716
1717	if (unlikely(ex == NULL || eh == NULL)) {
1718		EXT4_ERROR_INODE(inode,
1719				 "ex %p == NULL or eh %p == NULL", ex, eh);
1720		return -EFSCORRUPTED;
1721	}
1722
1723	if (depth == 0) {
1724		/* there is no tree at all */
1725		return 0;
1726	}
1727
1728	if (ex != EXT_FIRST_EXTENT(eh)) {
1729		/* we correct tree if first leaf got modified only */
1730		return 0;
1731	}
1732
1733	/*
1734	 * TODO: we need correction if border is smaller than current one
1735	 */
1736	k = depth - 1;
1737	border = path[depth].p_ext->ee_block;
1738	err = ext4_ext_get_access(handle, inode, path + k);
1739	if (err)
1740		return err;
1741	path[k].p_idx->ei_block = border;
1742	err = ext4_ext_dirty(handle, inode, path + k);
1743	if (err)
1744		return err;
1745
1746	while (k--) {
1747		/* change all left-side indexes */
1748		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1749			break;
1750		err = ext4_ext_get_access(handle, inode, path + k);
1751		if (err)
1752			break;
1753		path[k].p_idx->ei_block = border;
1754		err = ext4_ext_dirty(handle, inode, path + k);
1755		if (err)
1756			break;
1757	}
1758
1759	return err;
1760}
1761
1762static int ext4_can_extents_be_merged(struct inode *inode,
1763				      struct ext4_extent *ex1,
1764				      struct ext4_extent *ex2)
1765{
1766	unsigned short ext1_ee_len, ext2_ee_len;
1767
1768	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
 
 
 
 
1769		return 0;
1770
 
 
 
 
 
1771	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1772	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1773
1774	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1775			le32_to_cpu(ex2->ee_block))
1776		return 0;
1777
1778	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1779		return 0;
1780
1781	if (ext4_ext_is_unwritten(ex1) &&
1782	    ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
 
1783		return 0;
1784#ifdef AGGRESSIVE_TEST
1785	if (ext1_ee_len >= 4)
1786		return 0;
1787#endif
1788
1789	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1790		return 1;
1791	return 0;
1792}
1793
1794/*
1795 * This function tries to merge the "ex" extent to the next extent in the tree.
1796 * It always tries to merge towards right. If you want to merge towards
1797 * left, pass "ex - 1" as argument instead of "ex".
1798 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1799 * 1 if they got merged.
1800 */
1801static int ext4_ext_try_to_merge_right(struct inode *inode,
1802				 struct ext4_ext_path *path,
1803				 struct ext4_extent *ex)
1804{
1805	struct ext4_extent_header *eh;
1806	unsigned int depth, len;
1807	int merge_done = 0, unwritten;
 
1808
1809	depth = ext_depth(inode);
1810	BUG_ON(path[depth].p_hdr == NULL);
1811	eh = path[depth].p_hdr;
1812
1813	while (ex < EXT_LAST_EXTENT(eh)) {
1814		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1815			break;
1816		/* merge with next extent! */
1817		unwritten = ext4_ext_is_unwritten(ex);
 
1818		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1819				+ ext4_ext_get_actual_len(ex + 1));
1820		if (unwritten)
1821			ext4_ext_mark_unwritten(ex);
1822
1823		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1824			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1825				* sizeof(struct ext4_extent);
1826			memmove(ex + 1, ex + 2, len);
1827		}
1828		le16_add_cpu(&eh->eh_entries, -1);
1829		merge_done = 1;
1830		WARN_ON(eh->eh_entries == 0);
1831		if (!eh->eh_entries)
1832			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1833	}
1834
1835	return merge_done;
1836}
1837
1838/*
1839 * This function does a very simple check to see if we can collapse
1840 * an extent tree with a single extent tree leaf block into the inode.
1841 */
1842static void ext4_ext_try_to_merge_up(handle_t *handle,
1843				     struct inode *inode,
1844				     struct ext4_ext_path *path)
1845{
1846	size_t s;
1847	unsigned max_root = ext4_ext_space_root(inode, 0);
1848	ext4_fsblk_t blk;
1849
1850	if ((path[0].p_depth != 1) ||
1851	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1852	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1853		return;
1854
1855	/*
1856	 * We need to modify the block allocation bitmap and the block
1857	 * group descriptor to release the extent tree block.  If we
1858	 * can't get the journal credits, give up.
1859	 */
1860	if (ext4_journal_extend(handle, 2,
1861			ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1862		return;
1863
1864	/*
1865	 * Copy the extent data up to the inode
1866	 */
1867	blk = ext4_idx_pblock(path[0].p_idx);
1868	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1869		sizeof(struct ext4_extent_idx);
1870	s += sizeof(struct ext4_extent_header);
1871
1872	path[1].p_maxdepth = path[0].p_maxdepth;
1873	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1874	path[0].p_depth = 0;
1875	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1876		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1877	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1878
1879	brelse(path[1].p_bh);
1880	ext4_free_blocks(handle, inode, NULL, blk, 1,
1881			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1882}
1883
1884/*
1885 * This function tries to merge the @ex extent to neighbours in the tree, then
1886 * tries to collapse the extent tree into the inode.
1887 */
1888static void ext4_ext_try_to_merge(handle_t *handle,
1889				  struct inode *inode,
1890				  struct ext4_ext_path *path,
1891				  struct ext4_extent *ex)
1892{
1893	struct ext4_extent_header *eh;
1894	unsigned int depth;
1895	int merge_done = 0;
 
1896
1897	depth = ext_depth(inode);
1898	BUG_ON(path[depth].p_hdr == NULL);
1899	eh = path[depth].p_hdr;
1900
1901	if (ex > EXT_FIRST_EXTENT(eh))
1902		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1903
1904	if (!merge_done)
1905		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1906
1907	ext4_ext_try_to_merge_up(handle, inode, path);
1908}
1909
1910/*
1911 * check if a portion of the "newext" extent overlaps with an
1912 * existing extent.
1913 *
1914 * If there is an overlap discovered, it updates the length of the newext
1915 * such that there will be no overlap, and then returns 1.
1916 * If there is no overlap found, it returns 0.
1917 */
1918static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1919					   struct inode *inode,
1920					   struct ext4_extent *newext,
1921					   struct ext4_ext_path *path)
1922{
1923	ext4_lblk_t b1, b2;
1924	unsigned int depth, len1;
1925	unsigned int ret = 0;
1926
1927	b1 = le32_to_cpu(newext->ee_block);
1928	len1 = ext4_ext_get_actual_len(newext);
1929	depth = ext_depth(inode);
1930	if (!path[depth].p_ext)
1931		goto out;
1932	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
 
1933
1934	/*
1935	 * get the next allocated block if the extent in the path
1936	 * is before the requested block(s)
1937	 */
1938	if (b2 < b1) {
1939		b2 = ext4_ext_next_allocated_block(path);
1940		if (b2 == EXT_MAX_BLOCKS)
1941			goto out;
1942		b2 = EXT4_LBLK_CMASK(sbi, b2);
1943	}
1944
1945	/* check for wrap through zero on extent logical start block*/
1946	if (b1 + len1 < b1) {
1947		len1 = EXT_MAX_BLOCKS - b1;
1948		newext->ee_len = cpu_to_le16(len1);
1949		ret = 1;
1950	}
1951
1952	/* check for overlap */
1953	if (b1 + len1 > b2) {
1954		newext->ee_len = cpu_to_le16(b2 - b1);
1955		ret = 1;
1956	}
1957out:
1958	return ret;
1959}
1960
1961/*
1962 * ext4_ext_insert_extent:
1963 * tries to merge requested extent into the existing extent or
1964 * inserts requested extent as new one into the tree,
1965 * creating new leaf in the no-space case.
1966 */
1967int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1968				struct ext4_ext_path **ppath,
1969				struct ext4_extent *newext, int gb_flags)
1970{
1971	struct ext4_ext_path *path = *ppath;
1972	struct ext4_extent_header *eh;
1973	struct ext4_extent *ex, *fex;
1974	struct ext4_extent *nearex; /* nearest extent */
1975	struct ext4_ext_path *npath = NULL;
1976	int depth, len, err;
1977	ext4_lblk_t next;
1978	int mb_flags = 0, unwritten;
 
1979
1980	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1981		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1982	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1983		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1984		return -EFSCORRUPTED;
1985	}
1986	depth = ext_depth(inode);
1987	ex = path[depth].p_ext;
1988	eh = path[depth].p_hdr;
1989	if (unlikely(path[depth].p_hdr == NULL)) {
1990		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991		return -EFSCORRUPTED;
1992	}
1993
1994	/* try to insert block into found extent and return */
1995	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
 
 
 
 
 
 
 
 
 
 
 
1996
1997		/*
1998		 * Try to see whether we should rather test the extent on
1999		 * right from ex, or from the left of ex. This is because
2000		 * ext4_find_extent() can return either extent on the
2001		 * left, or on the right from the searched position. This
2002		 * will make merging more effective.
2003		 */
2004		if (ex < EXT_LAST_EXTENT(eh) &&
2005		    (le32_to_cpu(ex->ee_block) +
2006		    ext4_ext_get_actual_len(ex) <
2007		    le32_to_cpu(newext->ee_block))) {
2008			ex += 1;
2009			goto prepend;
2010		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2011			   (le32_to_cpu(newext->ee_block) +
2012			   ext4_ext_get_actual_len(newext) <
2013			   le32_to_cpu(ex->ee_block)))
2014			ex -= 1;
2015
2016		/* Try to append newex to the ex */
2017		if (ext4_can_extents_be_merged(inode, ex, newext)) {
2018			ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2019				  "(from %llu)\n",
2020				  ext4_ext_is_unwritten(newext),
2021				  ext4_ext_get_actual_len(newext),
2022				  le32_to_cpu(ex->ee_block),
2023				  ext4_ext_is_unwritten(ex),
2024				  ext4_ext_get_actual_len(ex),
2025				  ext4_ext_pblock(ex));
2026			err = ext4_ext_get_access(handle, inode,
2027						  path + depth);
2028			if (err)
2029				return err;
2030			unwritten = ext4_ext_is_unwritten(ex);
2031			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2032					+ ext4_ext_get_actual_len(newext));
2033			if (unwritten)
2034				ext4_ext_mark_unwritten(ex);
2035			nearex = ex;
2036			goto merge;
2037		}
2038
2039prepend:
2040		/* Try to prepend newex to the ex */
2041		if (ext4_can_extents_be_merged(inode, newext, ex)) {
2042			ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2043				  "(from %llu)\n",
2044				  le32_to_cpu(newext->ee_block),
2045				  ext4_ext_is_unwritten(newext),
2046				  ext4_ext_get_actual_len(newext),
2047				  le32_to_cpu(ex->ee_block),
2048				  ext4_ext_is_unwritten(ex),
2049				  ext4_ext_get_actual_len(ex),
2050				  ext4_ext_pblock(ex));
2051			err = ext4_ext_get_access(handle, inode,
2052						  path + depth);
2053			if (err)
2054				return err;
2055
2056			unwritten = ext4_ext_is_unwritten(ex);
2057			ex->ee_block = newext->ee_block;
2058			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2059			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2060					+ ext4_ext_get_actual_len(newext));
2061			if (unwritten)
2062				ext4_ext_mark_unwritten(ex);
2063			nearex = ex;
2064			goto merge;
2065		}
2066	}
2067
2068	depth = ext_depth(inode);
2069	eh = path[depth].p_hdr;
2070	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2071		goto has_space;
2072
2073	/* probably next leaf has space for us? */
2074	fex = EXT_LAST_EXTENT(eh);
2075	next = EXT_MAX_BLOCKS;
2076	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2077		next = ext4_ext_next_leaf_block(path);
2078	if (next != EXT_MAX_BLOCKS) {
2079		ext_debug(inode, "next leaf block - %u\n", next);
2080		BUG_ON(npath != NULL);
2081		npath = ext4_find_extent(inode, next, NULL, gb_flags);
2082		if (IS_ERR(npath))
2083			return PTR_ERR(npath);
2084		BUG_ON(npath->p_depth != path->p_depth);
2085		eh = npath[depth].p_hdr;
2086		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2087			ext_debug(inode, "next leaf isn't full(%d)\n",
2088				  le16_to_cpu(eh->eh_entries));
2089			path = npath;
2090			goto has_space;
2091		}
2092		ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2093			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2094	}
2095
2096	/*
2097	 * There is no free space in the found leaf.
2098	 * We're gonna add a new leaf in the tree.
2099	 */
2100	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2101		mb_flags |= EXT4_MB_USE_RESERVED;
2102	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2103				       ppath, newext);
2104	if (err)
2105		goto cleanup;
2106	depth = ext_depth(inode);
2107	eh = path[depth].p_hdr;
2108
2109has_space:
2110	nearex = path[depth].p_ext;
2111
2112	err = ext4_ext_get_access(handle, inode, path + depth);
2113	if (err)
2114		goto cleanup;
2115
2116	if (!nearex) {
2117		/* there is no extent in this leaf, create first one */
2118		ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2119				le32_to_cpu(newext->ee_block),
2120				ext4_ext_pblock(newext),
2121				ext4_ext_is_unwritten(newext),
2122				ext4_ext_get_actual_len(newext));
2123		nearex = EXT_FIRST_EXTENT(eh);
2124	} else {
2125		if (le32_to_cpu(newext->ee_block)
2126			   > le32_to_cpu(nearex->ee_block)) {
2127			/* Insert after */
2128			ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2129					"nearest %p\n",
2130					le32_to_cpu(newext->ee_block),
2131					ext4_ext_pblock(newext),
2132					ext4_ext_is_unwritten(newext),
2133					ext4_ext_get_actual_len(newext),
2134					nearex);
2135			nearex++;
2136		} else {
2137			/* Insert before */
2138			BUG_ON(newext->ee_block == nearex->ee_block);
2139			ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2140					"nearest %p\n",
2141					le32_to_cpu(newext->ee_block),
2142					ext4_ext_pblock(newext),
2143					ext4_ext_is_unwritten(newext),
2144					ext4_ext_get_actual_len(newext),
2145					nearex);
2146		}
2147		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2148		if (len > 0) {
2149			ext_debug(inode, "insert %u:%llu:[%d]%d: "
2150					"move %d extents from 0x%p to 0x%p\n",
2151					le32_to_cpu(newext->ee_block),
2152					ext4_ext_pblock(newext),
2153					ext4_ext_is_unwritten(newext),
2154					ext4_ext_get_actual_len(newext),
2155					len, nearex, nearex + 1);
2156			memmove(nearex + 1, nearex,
2157				len * sizeof(struct ext4_extent));
2158		}
2159	}
2160
2161	le16_add_cpu(&eh->eh_entries, 1);
2162	path[depth].p_ext = nearex;
2163	nearex->ee_block = newext->ee_block;
2164	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2165	nearex->ee_len = newext->ee_len;
2166
2167merge:
2168	/* try to merge extents */
2169	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2170		ext4_ext_try_to_merge(handle, inode, path, nearex);
2171
 
2172
2173	/* time to correct all indexes above */
2174	err = ext4_ext_correct_indexes(handle, inode, path);
2175	if (err)
2176		goto cleanup;
2177
2178	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2179
2180cleanup:
2181	ext4_free_ext_path(npath);
 
 
 
 
2182	return err;
2183}
2184
2185static int ext4_fill_es_cache_info(struct inode *inode,
2186				   ext4_lblk_t block, ext4_lblk_t num,
2187				   struct fiemap_extent_info *fieinfo)
2188{
2189	ext4_lblk_t next, end = block + num - 1;
2190	struct extent_status es;
2191	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2192	unsigned int flags;
2193	int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2194
2195	while (block <= end) {
2196		next = 0;
2197		flags = 0;
2198		if (!ext4_es_lookup_extent(inode, block, &next, &es))
2199			break;
2200		if (ext4_es_is_unwritten(&es))
2201			flags |= FIEMAP_EXTENT_UNWRITTEN;
2202		if (ext4_es_is_delayed(&es))
2203			flags |= (FIEMAP_EXTENT_DELALLOC |
2204				  FIEMAP_EXTENT_UNKNOWN);
2205		if (ext4_es_is_hole(&es))
2206			flags |= EXT4_FIEMAP_EXTENT_HOLE;
2207		if (next == 0)
2208			flags |= FIEMAP_EXTENT_LAST;
2209		if (flags & (FIEMAP_EXTENT_DELALLOC|
2210			     EXT4_FIEMAP_EXTENT_HOLE))
2211			es.es_pblk = 0;
2212		else
2213			es.es_pblk = ext4_es_pblock(&es);
2214		err = fiemap_fill_next_extent(fieinfo,
2215				(__u64)es.es_lblk << blksize_bits,
2216				(__u64)es.es_pblk << blksize_bits,
2217				(__u64)es.es_len << blksize_bits,
2218				flags);
2219		if (next == 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2220			break;
2221		block = next;
 
 
 
2222		if (err < 0)
2223			return err;
2224		if (err == 1)
2225			return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
2226	}
2227	return 0;
 
 
 
 
 
 
2228}
2229
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2230
2231/*
2232 * ext4_ext_determine_hole - determine hole around given block
2233 * @inode:	inode we lookup in
2234 * @path:	path in extent tree to @lblk
2235 * @lblk:	pointer to logical block around which we want to determine hole
2236 *
2237 * Determine hole length (and start if easily possible) around given logical
2238 * block. We don't try too hard to find the beginning of the hole but @path
2239 * actually points to extent before @lblk, we provide it.
2240 *
2241 * The function returns the length of a hole starting at @lblk. We update @lblk
2242 * to the beginning of the hole if we managed to find it.
2243 */
2244static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2245					   struct ext4_ext_path *path,
2246					   ext4_lblk_t *lblk)
2247{
2248	int depth = ext_depth(inode);
 
 
2249	struct ext4_extent *ex;
2250	ext4_lblk_t len;
2251
2252	ex = path[depth].p_ext;
2253	if (ex == NULL) {
2254		/* there is no extent yet, so gap is [0;-] */
2255		*lblk = 0;
2256		len = EXT_MAX_BLOCKS;
2257	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2258		len = le32_to_cpu(ex->ee_block) - *lblk;
2259	} else if (*lblk >= le32_to_cpu(ex->ee_block)
 
 
 
 
 
 
2260			+ ext4_ext_get_actual_len(ex)) {
2261		ext4_lblk_t next;
 
 
2262
2263		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2264		next = ext4_ext_next_allocated_block(path);
2265		BUG_ON(next == *lblk);
2266		len = next - *lblk;
 
 
 
 
2267	} else {
 
2268		BUG();
2269	}
2270	return len;
 
 
2271}
2272
2273/*
2274 * ext4_ext_put_gap_in_cache:
2275 * calculate boundaries of the gap that the requested block fits into
2276 * and cache this gap
 
 
 
 
 
 
 
 
 
 
 
2277 */
2278static void
2279ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2280			  ext4_lblk_t hole_len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2281{
2282	struct extent_status es;
 
 
 
 
 
 
 
 
2283
2284	ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2285				  hole_start + hole_len - 1, &es);
2286	if (es.es_len) {
2287		/* There's delayed extent containing lblock? */
2288		if (es.es_lblk <= hole_start)
2289			return;
2290		hole_len = min(es.es_lblk - hole_start, hole_len);
2291	}
2292	ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2293	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2294			      EXTENT_STATUS_HOLE);
2295}
2296
 
2297/*
2298 * ext4_ext_rm_idx:
2299 * removes index from the index block.
2300 */
2301static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2302			struct ext4_ext_path *path, int depth)
2303{
2304	int err;
2305	ext4_fsblk_t leaf;
2306
2307	/* free index block */
2308	depth--;
2309	path = path + depth;
2310	leaf = ext4_idx_pblock(path->p_idx);
2311	if (unlikely(path->p_hdr->eh_entries == 0)) {
2312		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2313		return -EFSCORRUPTED;
2314	}
2315	err = ext4_ext_get_access(handle, inode, path);
2316	if (err)
2317		return err;
2318
2319	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2320		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2321		len *= sizeof(struct ext4_extent_idx);
2322		memmove(path->p_idx, path->p_idx + 1, len);
2323	}
2324
2325	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2326	err = ext4_ext_dirty(handle, inode, path);
2327	if (err)
2328		return err;
2329	ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2330	trace_ext4_ext_rm_idx(inode, leaf);
2331
2332	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2333			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2334
2335	while (--depth >= 0) {
2336		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2337			break;
2338		path--;
2339		err = ext4_ext_get_access(handle, inode, path);
2340		if (err)
2341			break;
2342		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2343		err = ext4_ext_dirty(handle, inode, path);
2344		if (err)
2345			break;
2346	}
2347	return err;
2348}
2349
2350/*
2351 * ext4_ext_calc_credits_for_single_extent:
2352 * This routine returns max. credits that needed to insert an extent
2353 * to the extent tree.
2354 * When pass the actual path, the caller should calculate credits
2355 * under i_data_sem.
2356 */
2357int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2358						struct ext4_ext_path *path)
2359{
2360	if (path) {
2361		int depth = ext_depth(inode);
2362		int ret = 0;
2363
2364		/* probably there is space in leaf? */
2365		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2366				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2367
2368			/*
2369			 *  There are some space in the leaf tree, no
2370			 *  need to account for leaf block credit
2371			 *
2372			 *  bitmaps and block group descriptor blocks
2373			 *  and other metadata blocks still need to be
2374			 *  accounted.
2375			 */
2376			/* 1 bitmap, 1 block group descriptor */
2377			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2378			return ret;
2379		}
2380	}
2381
2382	return ext4_chunk_trans_blocks(inode, nrblocks);
2383}
2384
2385/*
2386 * How many index/leaf blocks need to change/allocate to add @extents extents?
2387 *
2388 * If we add a single extent, then in the worse case, each tree level
2389 * index/leaf need to be changed in case of the tree split.
 
 
2390 *
2391 * If more extents are inserted, they could cause the whole tree split more
2392 * than once, but this is really rare.
2393 */
2394int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2395{
2396	int index;
2397	int depth;
2398
2399	/* If we are converting the inline data, only one is needed here. */
2400	if (ext4_has_inline_data(inode))
2401		return 1;
2402
2403	depth = ext_depth(inode);
2404
2405	if (extents <= 1)
2406		index = depth * 2;
2407	else
2408		index = depth * 3;
2409
2410	return index;
2411}
2412
2413static inline int get_default_free_blocks_flags(struct inode *inode)
2414{
2415	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2416	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2417		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2418	else if (ext4_should_journal_data(inode))
2419		return EXT4_FREE_BLOCKS_FORGET;
2420	return 0;
2421}
2422
2423/*
2424 * ext4_rereserve_cluster - increment the reserved cluster count when
2425 *                          freeing a cluster with a pending reservation
2426 *
2427 * @inode - file containing the cluster
2428 * @lblk - logical block in cluster to be reserved
2429 *
2430 * Increments the reserved cluster count and adjusts quota in a bigalloc
2431 * file system when freeing a partial cluster containing at least one
2432 * delayed and unwritten block.  A partial cluster meeting that
2433 * requirement will have a pending reservation.  If so, the
2434 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2435 * defer reserved and allocated space accounting to a subsequent call
2436 * to this function.
2437 */
2438static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2439{
2440	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2441	struct ext4_inode_info *ei = EXT4_I(inode);
2442
2443	dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2444
2445	spin_lock(&ei->i_block_reservation_lock);
2446	ei->i_reserved_data_blocks++;
2447	percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2448	spin_unlock(&ei->i_block_reservation_lock);
2449
2450	percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2451	ext4_remove_pending(inode, lblk);
2452}
2453
2454static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2455			      struct ext4_extent *ex,
2456			      struct partial_cluster *partial,
2457			      ext4_lblk_t from, ext4_lblk_t to)
2458{
2459	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2460	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2461	ext4_fsblk_t last_pblk, pblk;
2462	ext4_lblk_t num;
2463	int flags;
2464
2465	/* only extent tail removal is allowed */
2466	if (from < le32_to_cpu(ex->ee_block) ||
2467	    to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2468		ext4_error(sbi->s_sb,
2469			   "strange request: removal(2) %u-%u from %u:%u",
2470			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2471		return 0;
2472	}
2473
2474#ifdef EXTENTS_STATS
2475	spin_lock(&sbi->s_ext_stats_lock);
2476	sbi->s_ext_blocks += ee_len;
2477	sbi->s_ext_extents++;
2478	if (ee_len < sbi->s_ext_min)
2479		sbi->s_ext_min = ee_len;
2480	if (ee_len > sbi->s_ext_max)
2481		sbi->s_ext_max = ee_len;
2482	if (ext_depth(inode) > sbi->s_depth_max)
2483		sbi->s_depth_max = ext_depth(inode);
2484	spin_unlock(&sbi->s_ext_stats_lock);
2485#endif
2486
2487	trace_ext4_remove_blocks(inode, ex, from, to, partial);
2488
 
 
2489	/*
2490	 * if we have a partial cluster, and it's different from the
2491	 * cluster of the last block in the extent, we free it
 
 
 
2492	 */
2493	last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2494
2495	if (partial->state != initial &&
2496	    partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2497		if (partial->state == tofree) {
2498			flags = get_default_free_blocks_flags(inode);
2499			if (ext4_is_pending(inode, partial->lblk))
2500				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2501			ext4_free_blocks(handle, inode, NULL,
2502					 EXT4_C2B(sbi, partial->pclu),
2503					 sbi->s_cluster_ratio, flags);
2504			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2505				ext4_rereserve_cluster(inode, partial->lblk);
2506		}
2507		partial->state = initial;
2508	}
2509
2510	num = le32_to_cpu(ex->ee_block) + ee_len - from;
2511	pblk = ext4_ext_pblock(ex) + ee_len - num;
2512
 
2513	/*
2514	 * We free the partial cluster at the end of the extent (if any),
2515	 * unless the cluster is used by another extent (partial_cluster
2516	 * state is nofree).  If a partial cluster exists here, it must be
2517	 * shared with the last block in the extent.
2518	 */
2519	flags = get_default_free_blocks_flags(inode);
2520
2521	/* partial, left end cluster aligned, right end unaligned */
2522	if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2523	    (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2524	    (partial->state != nofree)) {
2525		if (ext4_is_pending(inode, to))
2526			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2527		ext4_free_blocks(handle, inode, NULL,
2528				 EXT4_PBLK_CMASK(sbi, last_pblk),
2529				 sbi->s_cluster_ratio, flags);
2530		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2531			ext4_rereserve_cluster(inode, to);
2532		partial->state = initial;
2533		flags = get_default_free_blocks_flags(inode);
2534	}
2535
2536	flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2537
2538	/*
2539	 * For bigalloc file systems, we never free a partial cluster
2540	 * at the beginning of the extent.  Instead, we check to see if we
2541	 * need to free it on a subsequent call to ext4_remove_blocks,
2542	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2543	 */
2544	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2545	ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2546
2547	/* reset the partial cluster if we've freed past it */
2548	if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2549		partial->state = initial;
2550
2551	/*
2552	 * If we've freed the entire extent but the beginning is not left
2553	 * cluster aligned and is not marked as ineligible for freeing we
2554	 * record the partial cluster at the beginning of the extent.  It
2555	 * wasn't freed by the preceding ext4_free_blocks() call, and we
2556	 * need to look farther to the left to determine if it's to be freed
2557	 * (not shared with another extent). Else, reset the partial
2558	 * cluster - we're either  done freeing or the beginning of the
2559	 * extent is left cluster aligned.
2560	 */
2561	if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2562		if (partial->state == initial) {
2563			partial->pclu = EXT4_B2C(sbi, pblk);
2564			partial->lblk = from;
2565			partial->state = tofree;
2566		}
2567	} else {
2568		partial->state = initial;
 
 
2569	}
2570
2571	return 0;
2572}
2573
 
2574/*
2575 * ext4_ext_rm_leaf() Removes the extents associated with the
2576 * blocks appearing between "start" and "end".  Both "start"
2577 * and "end" must appear in the same extent or EIO is returned.
2578 *
2579 * @handle: The journal handle
2580 * @inode:  The files inode
2581 * @path:   The path to the leaf
2582 * @partial_cluster: The cluster which we'll have to free if all extents
2583 *                   has been released from it.  However, if this value is
2584 *                   negative, it's a cluster just to the right of the
2585 *                   punched region and it must not be freed.
2586 * @start:  The first block to remove
2587 * @end:   The last block to remove
2588 */
2589static int
2590ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2591		 struct ext4_ext_path *path,
2592		 struct partial_cluster *partial,
2593		 ext4_lblk_t start, ext4_lblk_t end)
2594{
2595	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2596	int err = 0, correct_index = 0;
2597	int depth = ext_depth(inode), credits, revoke_credits;
2598	struct ext4_extent_header *eh;
2599	ext4_lblk_t a, b;
2600	unsigned num;
2601	ext4_lblk_t ex_ee_block;
2602	unsigned short ex_ee_len;
2603	unsigned unwritten = 0;
2604	struct ext4_extent *ex;
2605	ext4_fsblk_t pblk;
2606
2607	/* the header must be checked already in ext4_ext_remove_space() */
2608	ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2609	if (!path[depth].p_hdr)
2610		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2611	eh = path[depth].p_hdr;
2612	if (unlikely(path[depth].p_hdr == NULL)) {
2613		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2614		return -EFSCORRUPTED;
2615	}
2616	/* find where to start removing */
2617	ex = path[depth].p_ext;
2618	if (!ex)
2619		ex = EXT_LAST_EXTENT(eh);
2620
2621	ex_ee_block = le32_to_cpu(ex->ee_block);
2622	ex_ee_len = ext4_ext_get_actual_len(ex);
2623
2624	trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2625
2626	while (ex >= EXT_FIRST_EXTENT(eh) &&
2627			ex_ee_block + ex_ee_len > start) {
2628
2629		if (ext4_ext_is_unwritten(ex))
2630			unwritten = 1;
2631		else
2632			unwritten = 0;
2633
2634		ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2635			  unwritten, ex_ee_len);
2636		path[depth].p_ext = ex;
2637
2638		a = max(ex_ee_block, start);
2639		b = min(ex_ee_block + ex_ee_len - 1, end);
 
2640
2641		ext_debug(inode, "  border %u:%u\n", a, b);
2642
2643		/* If this extent is beyond the end of the hole, skip it */
2644		if (end < ex_ee_block) {
2645			/*
2646			 * We're going to skip this extent and move to another,
2647			 * so note that its first cluster is in use to avoid
2648			 * freeing it when removing blocks.  Eventually, the
2649			 * right edge of the truncated/punched region will
2650			 * be just to the left.
2651			 */
2652			if (sbi->s_cluster_ratio > 1) {
2653				pblk = ext4_ext_pblock(ex);
2654				partial->pclu = EXT4_B2C(sbi, pblk);
2655				partial->state = nofree;
2656			}
2657			ex--;
2658			ex_ee_block = le32_to_cpu(ex->ee_block);
2659			ex_ee_len = ext4_ext_get_actual_len(ex);
2660			continue;
2661		} else if (b != ex_ee_block + ex_ee_len - 1) {
2662			EXT4_ERROR_INODE(inode,
2663					 "can not handle truncate %u:%u "
2664					 "on extent %u:%u",
2665					 start, end, ex_ee_block,
2666					 ex_ee_block + ex_ee_len - 1);
2667			err = -EFSCORRUPTED;
2668			goto out;
2669		} else if (a != ex_ee_block) {
2670			/* remove tail of the extent */
2671			num = a - ex_ee_block;
2672		} else {
2673			/* remove whole extent: excellent! */
2674			num = 0;
2675		}
2676		/*
2677		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2678		 * descriptor) for each block group; assume two block
2679		 * groups plus ex_ee_len/blocks_per_block_group for
2680		 * the worst case
2681		 */
2682		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2683		if (ex == EXT_FIRST_EXTENT(eh)) {
2684			correct_index = 1;
2685			credits += (ext_depth(inode)) + 1;
2686		}
2687		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2688		/*
2689		 * We may end up freeing some index blocks and data from the
2690		 * punched range. Note that partial clusters are accounted for
2691		 * by ext4_free_data_revoke_credits().
2692		 */
2693		revoke_credits =
2694			ext4_free_metadata_revoke_credits(inode->i_sb,
2695							  ext_depth(inode)) +
2696			ext4_free_data_revoke_credits(inode, b - a + 1);
2697
2698		err = ext4_datasem_ensure_credits(handle, inode, credits,
2699						  credits, revoke_credits);
2700		if (err) {
2701			if (err > 0)
2702				err = -EAGAIN;
2703			goto out;
2704		}
2705
2706		err = ext4_ext_get_access(handle, inode, path + depth);
2707		if (err)
2708			goto out;
2709
2710		err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
 
2711		if (err)
2712			goto out;
2713
2714		if (num == 0)
2715			/* this extent is removed; mark slot entirely unused */
2716			ext4_ext_store_pblock(ex, 0);
2717
2718		ex->ee_len = cpu_to_le16(num);
2719		/*
2720		 * Do not mark unwritten if all the blocks in the
2721		 * extent have been removed.
2722		 */
2723		if (unwritten && num)
2724			ext4_ext_mark_unwritten(ex);
2725		/*
2726		 * If the extent was completely released,
2727		 * we need to remove it from the leaf
2728		 */
2729		if (num == 0) {
2730			if (end != EXT_MAX_BLOCKS - 1) {
2731				/*
2732				 * For hole punching, we need to scoot all the
2733				 * extents up when an extent is removed so that
2734				 * we dont have blank extents in the middle
2735				 */
2736				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2737					sizeof(struct ext4_extent));
2738
2739				/* Now get rid of the one at the end */
2740				memset(EXT_LAST_EXTENT(eh), 0,
2741					sizeof(struct ext4_extent));
2742			}
2743			le16_add_cpu(&eh->eh_entries, -1);
2744		}
 
2745
2746		err = ext4_ext_dirty(handle, inode, path + depth);
2747		if (err)
2748			goto out;
2749
2750		ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2751				ext4_ext_pblock(ex));
2752		ex--;
2753		ex_ee_block = le32_to_cpu(ex->ee_block);
2754		ex_ee_len = ext4_ext_get_actual_len(ex);
2755	}
2756
2757	if (correct_index && eh->eh_entries)
2758		err = ext4_ext_correct_indexes(handle, inode, path);
2759
2760	/*
2761	 * If there's a partial cluster and at least one extent remains in
2762	 * the leaf, free the partial cluster if it isn't shared with the
2763	 * current extent.  If it is shared with the current extent
2764	 * we reset the partial cluster because we've reached the start of the
2765	 * truncated/punched region and we're done removing blocks.
2766	 */
2767	if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2768		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2769		if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2770			int flags = get_default_free_blocks_flags(inode);
2771
2772			if (ext4_is_pending(inode, partial->lblk))
2773				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2774			ext4_free_blocks(handle, inode, NULL,
2775					 EXT4_C2B(sbi, partial->pclu),
2776					 sbi->s_cluster_ratio, flags);
2777			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2778				ext4_rereserve_cluster(inode, partial->lblk);
2779		}
2780		partial->state = initial;
2781	}
2782
2783	/* if this leaf is free, then we should
2784	 * remove it from index block above */
2785	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2786		err = ext4_ext_rm_idx(handle, inode, path, depth);
2787
2788out:
2789	return err;
2790}
2791
2792/*
2793 * ext4_ext_more_to_rm:
2794 * returns 1 if current index has to be freed (even partial)
2795 */
2796static int
2797ext4_ext_more_to_rm(struct ext4_ext_path *path)
2798{
2799	BUG_ON(path->p_idx == NULL);
2800
2801	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2802		return 0;
2803
2804	/*
2805	 * if truncate on deeper level happened, it wasn't partial,
2806	 * so we have to consider current index for truncation
2807	 */
2808	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2809		return 0;
2810	return 1;
2811}
2812
2813int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2814			  ext4_lblk_t end)
2815{
2816	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2817	int depth = ext_depth(inode);
2818	struct ext4_ext_path *path = NULL;
2819	struct partial_cluster partial;
2820	handle_t *handle;
2821	int i = 0, err = 0;
2822
2823	partial.pclu = 0;
2824	partial.lblk = 0;
2825	partial.state = initial;
2826
2827	ext_debug(inode, "truncate since %u to %u\n", start, end);
2828
2829	/* probably first extent we're gonna free will be last in block */
2830	handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2831			depth + 1,
2832			ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2833	if (IS_ERR(handle))
2834		return PTR_ERR(handle);
2835
2836again:
2837	trace_ext4_ext_remove_space(inode, start, end, depth);
 
 
2838
2839	/*
2840	 * Check if we are removing extents inside the extent tree. If that
2841	 * is the case, we are going to punch a hole inside the extent tree
2842	 * so we have to check whether we need to split the extent covering
2843	 * the last block to remove so we can easily remove the part of it
2844	 * in ext4_ext_rm_leaf().
2845	 */
2846	if (end < EXT_MAX_BLOCKS - 1) {
2847		struct ext4_extent *ex;
2848		ext4_lblk_t ee_block, ex_end, lblk;
2849		ext4_fsblk_t pblk;
2850
2851		/* find extent for or closest extent to this block */
2852		path = ext4_find_extent(inode, end, NULL,
2853					EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2854		if (IS_ERR(path)) {
2855			ext4_journal_stop(handle);
2856			return PTR_ERR(path);
2857		}
2858		depth = ext_depth(inode);
2859		/* Leaf not may not exist only if inode has no blocks at all */
2860		ex = path[depth].p_ext;
2861		if (!ex) {
2862			if (depth) {
2863				EXT4_ERROR_INODE(inode,
2864						 "path[%d].p_hdr == NULL",
2865						 depth);
2866				err = -EFSCORRUPTED;
2867			}
2868			goto out;
2869		}
2870
2871		ee_block = le32_to_cpu(ex->ee_block);
2872		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2873
2874		/*
2875		 * See if the last block is inside the extent, if so split
2876		 * the extent at 'end' block so we can easily remove the
2877		 * tail of the first part of the split extent in
2878		 * ext4_ext_rm_leaf().
2879		 */
2880		if (end >= ee_block && end < ex_end) {
2881
2882			/*
2883			 * If we're going to split the extent, note that
2884			 * the cluster containing the block after 'end' is
2885			 * in use to avoid freeing it when removing blocks.
2886			 */
2887			if (sbi->s_cluster_ratio > 1) {
2888				pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2889				partial.pclu = EXT4_B2C(sbi, pblk);
2890				partial.state = nofree;
2891			}
2892
2893			/*
2894			 * Split the extent in two so that 'end' is the last
2895			 * block in the first new extent. Also we should not
2896			 * fail removing space due to ENOSPC so try to use
2897			 * reserved block if that happens.
2898			 */
2899			err = ext4_force_split_extent_at(handle, inode, &path,
2900							 end + 1, 1);
2901			if (err < 0)
2902				goto out;
2903
2904		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2905			   partial.state == initial) {
2906			/*
2907			 * If we're punching, there's an extent to the right.
2908			 * If the partial cluster hasn't been set, set it to
2909			 * that extent's first cluster and its state to nofree
2910			 * so it won't be freed should it contain blocks to be
2911			 * removed. If it's already set (tofree/nofree), we're
2912			 * retrying and keep the original partial cluster info
2913			 * so a cluster marked tofree as a result of earlier
2914			 * extent removal is not lost.
2915			 */
2916			lblk = ex_end + 1;
2917			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2918						    NULL);
2919			if (err < 0)
2920				goto out;
2921			if (pblk) {
2922				partial.pclu = EXT4_B2C(sbi, pblk);
2923				partial.state = nofree;
2924			}
2925		}
2926	}
 
 
2927	/*
2928	 * We start scanning from right side, freeing all the blocks
2929	 * after i_size and walking into the tree depth-wise.
2930	 */
2931	depth = ext_depth(inode);
2932	if (path) {
2933		int k = i = depth;
2934		while (--k > 0)
2935			path[k].p_block =
2936				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2937	} else {
2938		path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2939			       GFP_NOFS | __GFP_NOFAIL);
2940		if (path == NULL) {
2941			ext4_journal_stop(handle);
2942			return -ENOMEM;
2943		}
2944		path[0].p_maxdepth = path[0].p_depth = depth;
2945		path[0].p_hdr = ext_inode_hdr(inode);
2946		i = 0;
2947
2948		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2949			err = -EFSCORRUPTED;
2950			goto out;
2951		}
2952	}
2953	err = 0;
2954
2955	while (i >= 0 && err == 0) {
2956		if (i == depth) {
2957			/* this is leaf block */
2958			err = ext4_ext_rm_leaf(handle, inode, path,
2959					       &partial, start, end);
 
2960			/* root level has p_bh == NULL, brelse() eats this */
2961			brelse(path[i].p_bh);
2962			path[i].p_bh = NULL;
2963			i--;
2964			continue;
2965		}
2966
2967		/* this is index block */
2968		if (!path[i].p_hdr) {
2969			ext_debug(inode, "initialize header\n");
2970			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2971		}
2972
2973		if (!path[i].p_idx) {
2974			/* this level hasn't been touched yet */
2975			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2976			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2977			ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2978				  path[i].p_hdr,
2979				  le16_to_cpu(path[i].p_hdr->eh_entries));
2980		} else {
2981			/* we were already here, see at next index */
2982			path[i].p_idx--;
2983		}
2984
2985		ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2986				i, EXT_FIRST_INDEX(path[i].p_hdr),
2987				path[i].p_idx);
2988		if (ext4_ext_more_to_rm(path + i)) {
2989			struct buffer_head *bh;
2990			/* go to the next level */
2991			ext_debug(inode, "move to level %d (block %llu)\n",
2992				  i + 1, ext4_idx_pblock(path[i].p_idx));
2993			memset(path + i + 1, 0, sizeof(*path));
2994			bh = read_extent_tree_block(inode, path[i].p_idx,
2995						    depth - i - 1,
2996						    EXT4_EX_NOCACHE);
2997			if (IS_ERR(bh)) {
2998				/* should we reset i_size? */
2999				err = PTR_ERR(bh);
3000				break;
3001			}
3002			/* Yield here to deal with large extent trees.
3003			 * Should be a no-op if we did IO above. */
3004			cond_resched();
3005			if (WARN_ON(i + 1 > depth)) {
3006				err = -EFSCORRUPTED;
 
 
 
 
 
3007				break;
3008			}
3009			path[i + 1].p_bh = bh;
3010
3011			/* save actual number of indexes since this
3012			 * number is changed at the next iteration */
3013			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3014			i++;
3015		} else {
3016			/* we finished processing this index, go up */
3017			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3018				/* index is empty, remove it;
3019				 * handle must be already prepared by the
3020				 * truncatei_leaf() */
3021				err = ext4_ext_rm_idx(handle, inode, path, i);
3022			}
3023			/* root level has p_bh == NULL, brelse() eats this */
3024			brelse(path[i].p_bh);
3025			path[i].p_bh = NULL;
3026			i--;
3027			ext_debug(inode, "return to level %d\n", i);
3028		}
3029	}
3030
3031	trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3032					 path->p_hdr->eh_entries);
3033
3034	/*
3035	 * if there's a partial cluster and we have removed the first extent
3036	 * in the file, then we also free the partial cluster, if any
3037	 */
3038	if (partial.state == tofree && err == 0) {
3039		int flags = get_default_free_blocks_flags(inode);
 
 
3040
3041		if (ext4_is_pending(inode, partial.lblk))
3042			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3043		ext4_free_blocks(handle, inode, NULL,
3044				 EXT4_C2B(sbi, partial.pclu),
3045				 sbi->s_cluster_ratio, flags);
3046		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3047			ext4_rereserve_cluster(inode, partial.lblk);
3048		partial.state = initial;
3049	}
3050
3051	/* TODO: flexible tree reduction should be here */
3052	if (path->p_hdr->eh_entries == 0) {
3053		/*
3054		 * truncate to zero freed all the tree,
3055		 * so we need to correct eh_depth
3056		 */
3057		err = ext4_ext_get_access(handle, inode, path);
3058		if (err == 0) {
3059			ext_inode_hdr(inode)->eh_depth = 0;
3060			ext_inode_hdr(inode)->eh_max =
3061				cpu_to_le16(ext4_ext_space_root(inode, 0));
3062			err = ext4_ext_dirty(handle, inode, path);
3063		}
3064	}
3065out:
3066	ext4_free_ext_path(path);
3067	path = NULL;
3068	if (err == -EAGAIN)
 
3069		goto again;
 
3070	ext4_journal_stop(handle);
3071
3072	return err;
3073}
3074
3075/*
3076 * called at mount time
3077 */
3078void ext4_ext_init(struct super_block *sb)
3079{
3080	/*
3081	 * possible initialization would be here
3082	 */
3083
3084	if (ext4_has_feature_extents(sb)) {
3085#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3086		printk(KERN_INFO "EXT4-fs: file extents enabled"
3087#ifdef AGGRESSIVE_TEST
3088		       ", aggressive tests"
3089#endif
3090#ifdef CHECK_BINSEARCH
3091		       ", check binsearch"
3092#endif
3093#ifdef EXTENTS_STATS
3094		       ", stats"
3095#endif
3096		       "\n");
3097#endif
3098#ifdef EXTENTS_STATS
3099		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3100		EXT4_SB(sb)->s_ext_min = 1 << 30;
3101		EXT4_SB(sb)->s_ext_max = 0;
3102#endif
3103	}
3104}
3105
3106/*
3107 * called at umount time
3108 */
3109void ext4_ext_release(struct super_block *sb)
3110{
3111	if (!ext4_has_feature_extents(sb))
3112		return;
3113
3114#ifdef EXTENTS_STATS
3115	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3116		struct ext4_sb_info *sbi = EXT4_SB(sb);
3117		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3118			sbi->s_ext_blocks, sbi->s_ext_extents,
3119			sbi->s_ext_blocks / sbi->s_ext_extents);
3120		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3121			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3122	}
3123#endif
3124}
3125
3126static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3127{
3128	ext4_lblk_t  ee_block;
3129	ext4_fsblk_t ee_pblock;
3130	unsigned int ee_len;
3131
3132	ee_block  = le32_to_cpu(ex->ee_block);
3133	ee_len    = ext4_ext_get_actual_len(ex);
3134	ee_pblock = ext4_ext_pblock(ex);
3135
3136	if (ee_len == 0)
3137		return 0;
3138
3139	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3140				     EXTENT_STATUS_WRITTEN);
3141}
3142
3143/* FIXME!! we need to try to merge to left or right after zero-out  */
3144static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3145{
3146	ext4_fsblk_t ee_pblock;
3147	unsigned int ee_len;
 
3148
3149	ee_len    = ext4_ext_get_actual_len(ex);
3150	ee_pblock = ext4_ext_pblock(ex);
3151	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3152				  ee_len);
 
 
 
 
3153}
3154
3155/*
3156 * ext4_split_extent_at() splits an extent at given block.
3157 *
3158 * @handle: the journal handle
3159 * @inode: the file inode
3160 * @path: the path to the extent
3161 * @split: the logical block where the extent is splitted.
3162 * @split_flags: indicates if the extent could be zeroout if split fails, and
3163 *		 the states(init or unwritten) of new extents.
3164 * @flags: flags used to insert new extent to extent tree.
3165 *
3166 *
3167 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3168 * of which are determined by split_flag.
3169 *
3170 * There are two cases:
3171 *  a> the extent are splitted into two extent.
3172 *  b> split is not needed, and just mark the extent.
3173 *
3174 * return 0 on success.
3175 */
3176static int ext4_split_extent_at(handle_t *handle,
3177			     struct inode *inode,
3178			     struct ext4_ext_path **ppath,
3179			     ext4_lblk_t split,
3180			     int split_flag,
3181			     int flags)
3182{
3183	struct ext4_ext_path *path = *ppath;
3184	ext4_fsblk_t newblock;
3185	ext4_lblk_t ee_block;
3186	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3187	struct ext4_extent *ex2 = NULL;
3188	unsigned int ee_len, depth;
3189	int err = 0;
3190
3191	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3192	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3193
3194	ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3195
3196	ext4_ext_show_leaf(inode, path);
3197
3198	depth = ext_depth(inode);
3199	ex = path[depth].p_ext;
3200	ee_block = le32_to_cpu(ex->ee_block);
3201	ee_len = ext4_ext_get_actual_len(ex);
3202	newblock = split - ee_block + ext4_ext_pblock(ex);
3203
3204	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3205	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3206	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3207			     EXT4_EXT_MARK_UNWRIT1 |
3208			     EXT4_EXT_MARK_UNWRIT2));
3209
3210	err = ext4_ext_get_access(handle, inode, path + depth);
3211	if (err)
3212		goto out;
3213
3214	if (split == ee_block) {
3215		/*
3216		 * case b: block @split is the block that the extent begins with
3217		 * then we just change the state of the extent, and splitting
3218		 * is not needed.
3219		 */
3220		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3221			ext4_ext_mark_unwritten(ex);
3222		else
3223			ext4_ext_mark_initialized(ex);
3224
3225		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3226			ext4_ext_try_to_merge(handle, inode, path, ex);
3227
3228		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3229		goto out;
3230	}
3231
3232	/* case a */
3233	memcpy(&orig_ex, ex, sizeof(orig_ex));
3234	ex->ee_len = cpu_to_le16(split - ee_block);
3235	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3236		ext4_ext_mark_unwritten(ex);
3237
3238	/*
3239	 * path may lead to new leaf, not to original leaf any more
3240	 * after ext4_ext_insert_extent() returns,
3241	 */
3242	err = ext4_ext_dirty(handle, inode, path + depth);
3243	if (err)
3244		goto fix_extent_len;
3245
3246	ex2 = &newex;
3247	ex2->ee_block = cpu_to_le32(split);
3248	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3249	ext4_ext_store_pblock(ex2, newblock);
3250	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3251		ext4_ext_mark_unwritten(ex2);
3252
3253	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3254	if (err != -ENOSPC && err != -EDQUOT)
 
 
 
 
 
 
 
3255		goto out;
 
 
3256
3257	if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3258		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3259			if (split_flag & EXT4_EXT_DATA_VALID1) {
3260				err = ext4_ext_zeroout(inode, ex2);
3261				zero_ex.ee_block = ex2->ee_block;
3262				zero_ex.ee_len = cpu_to_le16(
3263						ext4_ext_get_actual_len(ex2));
3264				ext4_ext_store_pblock(&zero_ex,
3265						      ext4_ext_pblock(ex2));
3266			} else {
3267				err = ext4_ext_zeroout(inode, ex);
3268				zero_ex.ee_block = ex->ee_block;
3269				zero_ex.ee_len = cpu_to_le16(
3270						ext4_ext_get_actual_len(ex));
3271				ext4_ext_store_pblock(&zero_ex,
3272						      ext4_ext_pblock(ex));
3273			}
3274		} else {
3275			err = ext4_ext_zeroout(inode, &orig_ex);
3276			zero_ex.ee_block = orig_ex.ee_block;
3277			zero_ex.ee_len = cpu_to_le16(
3278						ext4_ext_get_actual_len(&orig_ex));
3279			ext4_ext_store_pblock(&zero_ex,
3280					      ext4_ext_pblock(&orig_ex));
3281		}
3282
3283		if (!err) {
3284			/* update the extent length and mark as initialized */
3285			ex->ee_len = cpu_to_le16(ee_len);
3286			ext4_ext_try_to_merge(handle, inode, path, ex);
3287			err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3288			if (!err)
3289				/* update extent status tree */
3290				err = ext4_zeroout_es(inode, &zero_ex);
3291			/* If we failed at this point, we don't know in which
3292			 * state the extent tree exactly is so don't try to fix
3293			 * length of the original extent as it may do even more
3294			 * damage.
3295			 */
3296			goto out;
3297		}
3298	}
3299
3300fix_extent_len:
3301	ex->ee_len = orig_ex.ee_len;
3302	/*
3303	 * Ignore ext4_ext_dirty return value since we are already in error path
3304	 * and err is a non-zero error code.
3305	 */
3306	ext4_ext_dirty(handle, inode, path + path->p_depth);
3307	return err;
3308out:
3309	ext4_ext_show_leaf(inode, path);
3310	return err;
3311}
3312
3313/*
3314 * ext4_split_extents() splits an extent and mark extent which is covered
3315 * by @map as split_flags indicates
3316 *
3317 * It may result in splitting the extent into multiple extents (up to three)
3318 * There are three possibilities:
3319 *   a> There is no split required
3320 *   b> Splits in two extents: Split is happening at either end of the extent
3321 *   c> Splits in three extents: Somone is splitting in middle of the extent
3322 *
3323 */
3324static int ext4_split_extent(handle_t *handle,
3325			      struct inode *inode,
3326			      struct ext4_ext_path **ppath,
3327			      struct ext4_map_blocks *map,
3328			      int split_flag,
3329			      int flags)
3330{
3331	struct ext4_ext_path *path = *ppath;
3332	ext4_lblk_t ee_block;
3333	struct ext4_extent *ex;
3334	unsigned int ee_len, depth;
3335	int err = 0;
3336	int unwritten;
3337	int split_flag1, flags1;
3338	int allocated = map->m_len;
3339
3340	depth = ext_depth(inode);
3341	ex = path[depth].p_ext;
3342	ee_block = le32_to_cpu(ex->ee_block);
3343	ee_len = ext4_ext_get_actual_len(ex);
3344	unwritten = ext4_ext_is_unwritten(ex);
3345
3346	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3347		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
 
3348		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3349		if (unwritten)
3350			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3351				       EXT4_EXT_MARK_UNWRIT2;
3352		if (split_flag & EXT4_EXT_DATA_VALID2)
3353			split_flag1 |= EXT4_EXT_DATA_VALID1;
3354		err = ext4_split_extent_at(handle, inode, ppath,
3355				map->m_lblk + map->m_len, split_flag1, flags1);
3356		if (err)
3357			goto out;
3358	} else {
3359		allocated = ee_len - (map->m_lblk - ee_block);
3360	}
3361	/*
3362	 * Update path is required because previous ext4_split_extent_at() may
3363	 * result in split of original leaf or extent zeroout.
3364	 */
3365	path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3366	if (IS_ERR(path))
3367		return PTR_ERR(path);
3368	depth = ext_depth(inode);
3369	ex = path[depth].p_ext;
3370	if (!ex) {
3371		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3372				 (unsigned long) map->m_lblk);
3373		return -EFSCORRUPTED;
3374	}
3375	unwritten = ext4_ext_is_unwritten(ex);
3376
3377	if (map->m_lblk >= ee_block) {
3378		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3379		if (unwritten) {
3380			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3381			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3382						     EXT4_EXT_MARK_UNWRIT2);
3383		}
3384		err = ext4_split_extent_at(handle, inode, ppath,
3385				map->m_lblk, split_flag1, flags);
3386		if (err)
3387			goto out;
3388	}
3389
3390	ext4_ext_show_leaf(inode, path);
3391out:
3392	return err ? err : allocated;
3393}
3394
 
3395/*
3396 * This function is called by ext4_ext_map_blocks() if someone tries to write
3397 * to an unwritten extent. It may result in splitting the unwritten
3398 * extent into multiple extents (up to three - one initialized and two
3399 * unwritten).
3400 * There are three possibilities:
3401 *   a> There is no split required: Entire extent should be initialized
3402 *   b> Splits in two extents: Write is happening at either end of the extent
3403 *   c> Splits in three extents: Somone is writing in middle of the extent
3404 *
3405 * Pre-conditions:
3406 *  - The extent pointed to by 'path' is unwritten.
3407 *  - The extent pointed to by 'path' contains a superset
3408 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3409 *
3410 * Post-conditions on success:
3411 *  - the returned value is the number of blocks beyond map->l_lblk
3412 *    that are allocated and initialized.
3413 *    It is guaranteed to be >= map->m_len.
3414 */
3415static int ext4_ext_convert_to_initialized(handle_t *handle,
3416					   struct inode *inode,
3417					   struct ext4_map_blocks *map,
3418					   struct ext4_ext_path **ppath,
3419					   int flags)
3420{
3421	struct ext4_ext_path *path = *ppath;
3422	struct ext4_sb_info *sbi;
3423	struct ext4_extent_header *eh;
3424	struct ext4_map_blocks split_map;
3425	struct ext4_extent zero_ex1, zero_ex2;
3426	struct ext4_extent *ex, *abut_ex;
3427	ext4_lblk_t ee_block, eof_block;
3428	unsigned int ee_len, depth, map_len = map->m_len;
3429	int allocated = 0, max_zeroout = 0;
3430	int err = 0;
3431	int split_flag = EXT4_EXT_DATA_VALID2;
3432
3433	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3434		  (unsigned long long)map->m_lblk, map_len);
 
3435
3436	sbi = EXT4_SB(inode->i_sb);
3437	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3438			>> inode->i_sb->s_blocksize_bits;
3439	if (eof_block < map->m_lblk + map_len)
3440		eof_block = map->m_lblk + map_len;
3441
3442	depth = ext_depth(inode);
3443	eh = path[depth].p_hdr;
3444	ex = path[depth].p_ext;
3445	ee_block = le32_to_cpu(ex->ee_block);
3446	ee_len = ext4_ext_get_actual_len(ex);
3447	zero_ex1.ee_len = 0;
3448	zero_ex2.ee_len = 0;
3449
3450	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3451
3452	/* Pre-conditions */
3453	BUG_ON(!ext4_ext_is_unwritten(ex));
3454	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3455
3456	/*
3457	 * Attempt to transfer newly initialized blocks from the currently
3458	 * unwritten extent to its neighbor. This is much cheaper
3459	 * than an insertion followed by a merge as those involve costly
3460	 * memmove() calls. Transferring to the left is the common case in
3461	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3462	 * followed by append writes.
3463	 *
3464	 * Limitations of the current logic:
3465	 *  - L1: we do not deal with writes covering the whole extent.
 
 
 
3466	 *    This would require removing the extent if the transfer
3467	 *    is possible.
3468	 *  - L2: we only attempt to merge with an extent stored in the
3469	 *    same extent tree node.
3470	 */
3471	if ((map->m_lblk == ee_block) &&
3472		/* See if we can merge left */
3473		(map_len < ee_len) &&		/*L1*/
3474		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3475		ext4_lblk_t prev_lblk;
3476		ext4_fsblk_t prev_pblk, ee_pblk;
3477		unsigned int prev_len;
3478
3479		abut_ex = ex - 1;
3480		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3481		prev_len = ext4_ext_get_actual_len(abut_ex);
3482		prev_pblk = ext4_ext_pblock(abut_ex);
3483		ee_pblk = ext4_ext_pblock(ex);
 
3484
3485		/*
3486		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3487		 * upon those conditions:
3488		 * - C1: abut_ex is initialized,
3489		 * - C2: abut_ex is logically abutting ex,
3490		 * - C3: abut_ex is physically abutting ex,
3491		 * - C4: abut_ex can receive the additional blocks without
3492		 *   overflowing the (initialized) length limit.
3493		 */
3494		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3495			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3496			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3497			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3498			err = ext4_ext_get_access(handle, inode, path + depth);
3499			if (err)
3500				goto out;
3501
3502			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3503				map, ex, abut_ex);
3504
3505			/* Shift the start of ex by 'map_len' blocks */
3506			ex->ee_block = cpu_to_le32(ee_block + map_len);
3507			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3508			ex->ee_len = cpu_to_le16(ee_len - map_len);
3509			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3510
3511			/* Extend abut_ex by 'map_len' blocks */
3512			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3513
3514			/* Result: number of initialized blocks past m_lblk */
3515			allocated = map_len;
3516		}
3517	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3518		   (map_len < ee_len) &&	/*L1*/
3519		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3520		/* See if we can merge right */
3521		ext4_lblk_t next_lblk;
3522		ext4_fsblk_t next_pblk, ee_pblk;
3523		unsigned int next_len;
3524
3525		abut_ex = ex + 1;
3526		next_lblk = le32_to_cpu(abut_ex->ee_block);
3527		next_len = ext4_ext_get_actual_len(abut_ex);
3528		next_pblk = ext4_ext_pblock(abut_ex);
3529		ee_pblk = ext4_ext_pblock(ex);
3530
3531		/*
3532		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3533		 * upon those conditions:
3534		 * - C1: abut_ex is initialized,
3535		 * - C2: abut_ex is logically abutting ex,
3536		 * - C3: abut_ex is physically abutting ex,
3537		 * - C4: abut_ex can receive the additional blocks without
3538		 *   overflowing the (initialized) length limit.
3539		 */
3540		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3541		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3542		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3543		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3544			err = ext4_ext_get_access(handle, inode, path + depth);
3545			if (err)
3546				goto out;
3547
3548			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3549				map, ex, abut_ex);
3550
3551			/* Shift the start of abut_ex by 'map_len' blocks */
3552			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3553			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3554			ex->ee_len = cpu_to_le16(ee_len - map_len);
3555			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3556
3557			/* Extend abut_ex by 'map_len' blocks */
3558			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3559
3560			/* Result: number of initialized blocks past m_lblk */
3561			allocated = map_len;
 
3562		}
3563	}
3564	if (allocated) {
3565		/* Mark the block containing both extents as dirty */
3566		err = ext4_ext_dirty(handle, inode, path + depth);
3567
3568		/* Update path to point to the right extent */
3569		path[depth].p_ext = abut_ex;
3570		goto out;
3571	} else
3572		allocated = ee_len - (map->m_lblk - ee_block);
3573
3574	WARN_ON(map->m_lblk < ee_block);
3575	/*
3576	 * It is safe to convert extent to initialized via explicit
3577	 * zeroout only if extent is fully inside i_size or new_size.
3578	 */
3579	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3580
3581	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3582		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3583			(inode->i_sb->s_blocksize_bits - 10);
 
 
 
 
 
 
 
 
 
 
 
 
3584
3585	/*
3586	 * five cases:
3587	 * 1. split the extent into three extents.
3588	 * 2. split the extent into two extents, zeroout the head of the first
3589	 *    extent.
3590	 * 3. split the extent into two extents, zeroout the tail of the second
3591	 *    extent.
3592	 * 4. split the extent into two extents with out zeroout.
3593	 * 5. no splitting needed, just possibly zeroout the head and / or the
3594	 *    tail of the extent.
3595	 */
3596	split_map.m_lblk = map->m_lblk;
3597	split_map.m_len = map->m_len;
3598
3599	if (max_zeroout && (allocated > split_map.m_len)) {
3600		if (allocated <= max_zeroout) {
3601			/* case 3 or 5 */
3602			zero_ex1.ee_block =
3603				 cpu_to_le32(split_map.m_lblk +
3604					     split_map.m_len);
3605			zero_ex1.ee_len =
3606				cpu_to_le16(allocated - split_map.m_len);
3607			ext4_ext_store_pblock(&zero_ex1,
3608				ext4_ext_pblock(ex) + split_map.m_lblk +
3609				split_map.m_len - ee_block);
3610			err = ext4_ext_zeroout(inode, &zero_ex1);
3611			if (err)
3612				goto fallback;
 
3613			split_map.m_len = allocated;
3614		}
3615		if (split_map.m_lblk - ee_block + split_map.m_len <
3616								max_zeroout) {
3617			/* case 2 or 5 */
3618			if (split_map.m_lblk != ee_block) {
3619				zero_ex2.ee_block = ex->ee_block;
3620				zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3621							ee_block);
3622				ext4_ext_store_pblock(&zero_ex2,
3623						      ext4_ext_pblock(ex));
3624				err = ext4_ext_zeroout(inode, &zero_ex2);
3625				if (err)
3626					goto fallback;
3627			}
3628
3629			split_map.m_len += split_map.m_lblk - ee_block;
3630			split_map.m_lblk = ee_block;
 
3631			allocated = map->m_len;
3632		}
3633	}
3634
3635fallback:
3636	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3637				flags);
3638	if (err > 0)
3639		err = 0;
3640out:
3641	/* If we have gotten a failure, don't zero out status tree */
3642	if (!err) {
3643		err = ext4_zeroout_es(inode, &zero_ex1);
3644		if (!err)
3645			err = ext4_zeroout_es(inode, &zero_ex2);
3646	}
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(inode, "logical block %llu, max_blocks %u\n",
3688		  (unsigned long long)map->m_lblk, map->m_len);
 
3689
3690	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3691			>> inode->i_sb->s_blocksize_bits;
3692	if (eof_block < map->m_lblk + map->m_len)
3693		eof_block = map->m_lblk + map->m_len;
3694	/*
3695	 * It is safe to convert extent to initialized via explicit
3696	 * zeroout only if extent is fully inside i_size or new_size.
3697	 */
3698	depth = ext_depth(inode);
3699	ex = path[depth].p_ext;
3700	ee_block = le32_to_cpu(ex->ee_block);
3701	ee_len = ext4_ext_get_actual_len(ex);
3702
3703	/* Convert to unwritten */
3704	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3705		split_flag |= EXT4_EXT_DATA_VALID1;
3706	/* Convert to initialized */
3707	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3708		split_flag |= ee_block + ee_len <= eof_block ?
3709			      EXT4_EXT_MAY_ZEROOUT : 0;
3710		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3711	}
3712	flags |= EXT4_GET_BLOCKS_PRE_IO;
3713	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3714}
3715
3716static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3717						struct inode *inode,
3718						struct ext4_map_blocks *map,
3719						struct ext4_ext_path **ppath)
3720{
3721	struct ext4_ext_path *path = *ppath;
3722	struct ext4_extent *ex;
3723	ext4_lblk_t ee_block;
3724	unsigned int ee_len;
3725	int depth;
3726	int err = 0;
3727
3728	depth = ext_depth(inode);
3729	ex = path[depth].p_ext;
3730	ee_block = le32_to_cpu(ex->ee_block);
3731	ee_len = ext4_ext_get_actual_len(ex);
3732
3733	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3734		  (unsigned long long)ee_block, ee_len);
3735
3736	/* If extent is larger than requested it is a clear sign that we still
3737	 * have some extent state machine issues left. So extent_split is still
3738	 * required.
3739	 * TODO: Once all related issues will be fixed this situation should be
3740	 * illegal.
3741	 */
3742	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3743#ifdef CONFIG_EXT4_DEBUG
3744		ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3745			     " len %u; IO logical block %llu, len %u",
3746			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3747			     (unsigned long long)map->m_lblk, map->m_len);
3748#endif
3749		err = ext4_split_convert_extents(handle, inode, map, ppath,
3750						 EXT4_GET_BLOCKS_CONVERT);
3751		if (err < 0)
3752			return err;
3753		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3754		if (IS_ERR(path))
3755			return PTR_ERR(path);
3756		depth = ext_depth(inode);
3757		ex = path[depth].p_ext;
3758	}
3759
3760	err = ext4_ext_get_access(handle, inode, path + depth);
3761	if (err)
3762		goto out;
3763	/* first mark the extent as initialized */
3764	ext4_ext_mark_initialized(ex);
3765
3766	/* note: ext4_ext_correct_indexes() isn't needed here because
3767	 * borders are not changed
3768	 */
3769	ext4_ext_try_to_merge(handle, inode, path, ex);
3770
3771	/* Mark modified extent as dirty */
3772	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3773out:
3774	ext4_ext_show_leaf(inode, path);
3775	return err;
3776}
3777
3778static int
3779convert_initialized_extent(handle_t *handle, struct inode *inode,
3780			   struct ext4_map_blocks *map,
3781			   struct ext4_ext_path **ppath,
3782			   unsigned int *allocated)
3783{
3784	struct ext4_ext_path *path = *ppath;
3785	struct ext4_extent *ex;
3786	ext4_lblk_t ee_block;
3787	unsigned int ee_len;
3788	int depth;
3789	int err = 0;
 
 
 
 
 
 
 
 
 
 
3790
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3791	/*
3792	 * Make sure that the extent is no bigger than we support with
3793	 * unwritten extent
 
 
 
3794	 */
3795	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3796		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
 
 
 
 
 
3797
3798	depth = ext_depth(inode);
3799	ex = path[depth].p_ext;
3800	ee_block = le32_to_cpu(ex->ee_block);
3801	ee_len = ext4_ext_get_actual_len(ex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3802
3803	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3804		  (unsigned long long)ee_block, ee_len);
3805
3806	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3807		err = ext4_split_convert_extents(handle, inode, map, ppath,
3808				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3809		if (err < 0)
3810			return err;
3811		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3812		if (IS_ERR(path))
3813			return PTR_ERR(path);
3814		depth = ext_depth(inode);
3815		ex = path[depth].p_ext;
3816		if (!ex) {
3817			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3818					 (unsigned long) map->m_lblk);
3819			return -EFSCORRUPTED;
3820		}
3821	}
3822
3823	err = ext4_ext_get_access(handle, inode, path + depth);
3824	if (err)
3825		return err;
3826	/* first mark the extent as unwritten */
3827	ext4_ext_mark_unwritten(ex);
3828
3829	/* note: ext4_ext_correct_indexes() isn't needed here because
3830	 * borders are not changed
3831	 */
3832	ext4_ext_try_to_merge(handle, inode, path, ex);
3833
3834	/* Mark modified extent as dirty */
3835	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3836	if (err)
3837		return err;
3838	ext4_ext_show_leaf(inode, path);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3839
3840	ext4_update_inode_fsync_trans(handle, inode, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3841
3842	map->m_flags |= EXT4_MAP_UNWRITTEN;
3843	if (*allocated > map->m_len)
3844		*allocated = map->m_len;
3845	map->m_len = *allocated;
3846	return 0;
3847}
3848
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3849static int
3850ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3851			struct ext4_map_blocks *map,
3852			struct ext4_ext_path **ppath, int flags,
3853			unsigned int allocated, ext4_fsblk_t newblock)
3854{
3855	struct ext4_ext_path __maybe_unused *path = *ppath;
3856	int ret = 0;
3857	int err = 0;
 
3858
3859	ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3860		  (unsigned long long)map->m_lblk, map->m_len, flags,
3861		  allocated);
 
3862	ext4_ext_show_leaf(inode, path);
3863
3864	/*
3865	 * When writing into unwritten space, we should not fail to
3866	 * allocate metadata blocks for the new extent block if needed.
3867	 */
3868	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3869
3870	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3871						    allocated, newblock);
3872
3873	/* get_block() before submitting IO, split the extent */
3874	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3875		ret = ext4_split_convert_extents(handle, inode, map, ppath,
3876					 flags | EXT4_GET_BLOCKS_CONVERT);
3877		if (ret < 0) {
3878			err = ret;
3879			goto out2;
3880		}
3881		/*
3882		 * shouldn't get a 0 return when splitting an extent unless
3883		 * m_len is 0 (bug) or extent has been corrupted
 
3884		 */
3885		if (unlikely(ret == 0)) {
3886			EXT4_ERROR_INODE(inode,
3887					 "unexpected ret == 0, m_len = %u",
3888					 map->m_len);
3889			err = -EFSCORRUPTED;
3890			goto out2;
3891		}
3892		map->m_flags |= EXT4_MAP_UNWRITTEN;
3893		goto out;
3894	}
3895	/* IO end_io complete, convert the filled extent to written */
3896	if (flags & EXT4_GET_BLOCKS_CONVERT) {
3897		err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3898							   ppath);
3899		if (err < 0)
3900			goto out2;
3901		ext4_update_inode_fsync_trans(handle, inode, 1);
3902		goto map_out;
 
 
 
3903	}
3904	/* buffered IO cases */
3905	/*
3906	 * repeat fallocate creation request
3907	 * we already have an unwritten extent
3908	 */
3909	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3910		map->m_flags |= EXT4_MAP_UNWRITTEN;
3911		goto map_out;
3912	}
3913
3914	/* buffered READ or buffered write_begin() lookup */
3915	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3916		/*
3917		 * We have blocks reserved already.  We
3918		 * return allocated blocks so that delalloc
3919		 * won't do block reservation for us.  But
3920		 * the buffer head will be unmapped so that
3921		 * a read from the block returns 0s.
3922		 */
3923		map->m_flags |= EXT4_MAP_UNWRITTEN;
3924		goto out1;
3925	}
3926
3927	/*
3928	 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3929	 * For buffered writes, at writepage time, etc.  Convert a
3930	 * discovered unwritten extent to written.
3931	 */
3932	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3933	if (ret < 0) {
3934		err = ret;
3935		goto out2;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3936	}
3937	ext4_update_inode_fsync_trans(handle, inode, 1);
3938	/*
3939	 * shouldn't get a 0 return when converting an unwritten extent
3940	 * unless m_len is 0 (bug) or extent has been corrupted
3941	 */
3942	if (unlikely(ret == 0)) {
3943		EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3944				 map->m_len);
3945		err = -EFSCORRUPTED;
3946		goto out2;
 
 
 
 
 
 
3947	}
3948
3949out:
3950	allocated = ret;
3951	map->m_flags |= EXT4_MAP_NEW;
3952map_out:
3953	map->m_flags |= EXT4_MAP_MAPPED;
 
 
 
 
 
 
3954out1:
3955	map->m_pblk = newblock;
3956	if (allocated > map->m_len)
3957		allocated = map->m_len;
3958	map->m_len = allocated;
3959	ext4_ext_show_leaf(inode, path);
 
 
3960out2:
 
 
 
 
3961	return err ? err : allocated;
3962}
3963
3964/*
3965 * get_implied_cluster_alloc - check to see if the requested
3966 * allocation (in the map structure) overlaps with a cluster already
3967 * allocated in an extent.
3968 *	@sb	The filesystem superblock structure
3969 *	@map	The requested lblk->pblk mapping
3970 *	@ex	The extent structure which might contain an implied
3971 *			cluster allocation
3972 *
3973 * This function is called by ext4_ext_map_blocks() after we failed to
3974 * find blocks that were already in the inode's extent tree.  Hence,
3975 * we know that the beginning of the requested region cannot overlap
3976 * the extent from the inode's extent tree.  There are three cases we
3977 * want to catch.  The first is this case:
3978 *
3979 *		 |--- cluster # N--|
3980 *    |--- extent ---|	|---- requested region ---|
3981 *			|==========|
3982 *
3983 * The second case that we need to test for is this one:
3984 *
3985 *   |--------- cluster # N ----------------|
3986 *	   |--- requested region --|   |------- extent ----|
3987 *	   |=======================|
3988 *
3989 * The third case is when the requested region lies between two extents
3990 * within the same cluster:
3991 *          |------------- cluster # N-------------|
3992 * |----- ex -----|                  |---- ex_right ----|
3993 *                  |------ requested region ------|
3994 *                  |================|
3995 *
3996 * In each of the above cases, we need to set the map->m_pblk and
3997 * map->m_len so it corresponds to the return the extent labelled as
3998 * "|====|" from cluster #N, since it is already in use for data in
3999 * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
4000 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4001 * as a new "allocated" block region.  Otherwise, we will return 0 and
4002 * ext4_ext_map_blocks() will then allocate one or more new clusters
4003 * by calling ext4_mb_new_blocks().
4004 */
4005static int get_implied_cluster_alloc(struct super_block *sb,
4006				     struct ext4_map_blocks *map,
4007				     struct ext4_extent *ex,
4008				     struct ext4_ext_path *path)
4009{
4010	struct ext4_sb_info *sbi = EXT4_SB(sb);
4011	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4012	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4013	ext4_lblk_t rr_cluster_start;
4014	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4015	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4016	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4017
4018	/* The extent passed in that we are trying to match */
4019	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4020	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4021
4022	/* The requested region passed into ext4_map_blocks() */
4023	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4024
4025	if ((rr_cluster_start == ex_cluster_end) ||
4026	    (rr_cluster_start == ex_cluster_start)) {
4027		if (rr_cluster_start == ex_cluster_end)
4028			ee_start += ee_len - 1;
4029		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
 
4030		map->m_len = min(map->m_len,
4031				 (unsigned) sbi->s_cluster_ratio - c_offset);
4032		/*
4033		 * Check for and handle this case:
4034		 *
4035		 *   |--------- cluster # N-------------|
4036		 *		       |------- extent ----|
4037		 *	   |--- requested region ---|
4038		 *	   |===========|
4039		 */
4040
4041		if (map->m_lblk < ee_block)
4042			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4043
4044		/*
4045		 * Check for the case where there is already another allocated
4046		 * block to the right of 'ex' but before the end of the cluster.
4047		 *
4048		 *          |------------- cluster # N-------------|
4049		 * |----- ex -----|                  |---- ex_right ----|
4050		 *                  |------ requested region ------|
4051		 *                  |================|
4052		 */
4053		if (map->m_lblk > ee_block) {
4054			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4055			map->m_len = min(map->m_len, next - map->m_lblk);
4056		}
4057
4058		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4059		return 1;
4060	}
4061
4062	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4063	return 0;
4064}
4065
4066
4067/*
4068 * Block allocation/map/preallocation routine for extents based files
4069 *
4070 *
4071 * Need to be called with
4072 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4073 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4074 *
4075 * return > 0, number of blocks already mapped/allocated
4076 *          if create == 0 and these are pre-allocated blocks
4077 *          	buffer head is unmapped
4078 *          otherwise blocks are mapped
4079 *
4080 * return = 0, if plain look up failed (blocks have not been allocated)
4081 *          buffer head is unmapped
4082 *
4083 * return < 0, error case.
4084 */
4085int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4086			struct ext4_map_blocks *map, int flags)
4087{
4088	struct ext4_ext_path *path = NULL;
4089	struct ext4_extent newex, *ex, ex2;
4090	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4091	ext4_fsblk_t newblock = 0, pblk;
4092	int err = 0, depth, ret;
4093	unsigned int allocated = 0, offset = 0;
4094	unsigned int allocated_clusters = 0;
4095	struct ext4_allocation_request ar;
 
4096	ext4_lblk_t cluster_offset;
4097
4098	ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
 
4099	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4100
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4101	/* find extent for this block */
4102	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4103	if (IS_ERR(path)) {
4104		err = PTR_ERR(path);
4105		path = NULL;
4106		goto out;
4107	}
4108
4109	depth = ext_depth(inode);
4110
4111	/*
4112	 * consistent leaf must not be empty;
4113	 * this situation is possible, though, _during_ tree modification;
4114	 * this is why assert can't be put in ext4_find_extent()
4115	 */
4116	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4117		EXT4_ERROR_INODE(inode, "bad extent address "
4118				 "lblock: %lu, depth: %d pblock %lld",
4119				 (unsigned long) map->m_lblk, depth,
4120				 path[depth].p_block);
4121		err = -EFSCORRUPTED;
4122		goto out;
4123	}
4124
4125	ex = path[depth].p_ext;
4126	if (ex) {
4127		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4128		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4129		unsigned short ee_len;
4130
4131
4132		/*
4133		 * unwritten extents are treated as holes, except that
4134		 * we split out initialized portions during a write.
4135		 */
4136		ee_len = ext4_ext_get_actual_len(ex);
4137
4138		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4139
4140		/* if found extent covers block, simply return it */
4141		if (in_range(map->m_lblk, ee_block, ee_len)) {
4142			newblock = map->m_lblk - ee_block + ee_start;
4143			/* number of remaining blocks in the extent */
4144			allocated = ee_len - (map->m_lblk - ee_block);
4145			ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4146				  map->m_lblk, ee_block, ee_len, newblock);
4147
4148			/*
4149			 * If the extent is initialized check whether the
4150			 * caller wants to convert it to unwritten.
4151			 */
4152			if ((!ext4_ext_is_unwritten(ex)) &&
4153			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4154				err = convert_initialized_extent(handle,
4155					inode, map, &path, &allocated);
4156				goto out;
4157			} else if (!ext4_ext_is_unwritten(ex)) {
4158				map->m_flags |= EXT4_MAP_MAPPED;
4159				map->m_pblk = newblock;
4160				if (allocated > map->m_len)
4161					allocated = map->m_len;
4162				map->m_len = allocated;
4163				ext4_ext_show_leaf(inode, path);
4164				goto out;
4165			}
4166
4167			ret = ext4_ext_handle_unwritten_extents(
4168				handle, inode, map, &path, flags,
4169				allocated, newblock);
4170			if (ret < 0)
4171				err = ret;
4172			else
4173				allocated = ret;
4174			goto out;
4175		}
4176	}
4177
 
 
 
 
4178	/*
4179	 * requested block isn't allocated yet;
4180	 * we couldn't try to create block if create flag is zero
4181	 */
4182	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4183		ext4_lblk_t hole_start, hole_len;
4184
4185		hole_start = map->m_lblk;
4186		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4187		/*
4188		 * put just found gap into cache to speed up
4189		 * subsequent requests
4190		 */
4191		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4192
4193		/* Update hole_len to reflect hole size after map->m_lblk */
4194		if (hole_start != map->m_lblk)
4195			hole_len -= map->m_lblk - hole_start;
4196		map->m_pblk = 0;
4197		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4198
4199		goto out;
4200	}
4201
4202	/*
4203	 * Okay, we need to do block allocation.
4204	 */
 
4205	newex.ee_block = cpu_to_le32(map->m_lblk);
4206	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4207
4208	/*
4209	 * If we are doing bigalloc, check to see if the extent returned
4210	 * by ext4_find_extent() implies a cluster we can use.
4211	 */
4212	if (cluster_offset && ex &&
4213	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4214		ar.len = allocated = map->m_len;
4215		newblock = map->m_pblk;
 
4216		goto got_allocated_blocks;
4217	}
4218
4219	/* find neighbour allocated blocks */
4220	ar.lleft = map->m_lblk;
4221	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4222	if (err)
4223		goto out;
4224	ar.lright = map->m_lblk;
 
4225	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4226	if (err < 0)
4227		goto out;
4228
4229	/* Check if the extent after searching to the right implies a
4230	 * cluster we can use. */
4231	if ((sbi->s_cluster_ratio > 1) && err &&
4232	    get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4233		ar.len = allocated = map->m_len;
4234		newblock = map->m_pblk;
 
4235		goto got_allocated_blocks;
4236	}
4237
4238	/*
4239	 * See if request is beyond maximum number of blocks we can have in
4240	 * a single extent. For an initialized extent this limit is
4241	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4242	 * EXT_UNWRITTEN_MAX_LEN.
4243	 */
4244	if (map->m_len > EXT_INIT_MAX_LEN &&
4245	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4246		map->m_len = EXT_INIT_MAX_LEN;
4247	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4248		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4249		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4250
4251	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4252	newex.ee_len = cpu_to_le16(map->m_len);
4253	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4254	if (err)
4255		allocated = ext4_ext_get_actual_len(&newex);
4256	else
4257		allocated = map->m_len;
4258
4259	/* allocate new block */
4260	ar.inode = inode;
4261	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4262	ar.logical = map->m_lblk;
4263	/*
4264	 * We calculate the offset from the beginning of the cluster
4265	 * for the logical block number, since when we allocate a
4266	 * physical cluster, the physical block should start at the
4267	 * same offset from the beginning of the cluster.  This is
4268	 * needed so that future calls to get_implied_cluster_alloc()
4269	 * work correctly.
4270	 */
4271	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4272	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4273	ar.goal -= offset;
4274	ar.logical -= offset;
4275	if (S_ISREG(inode->i_mode))
4276		ar.flags = EXT4_MB_HINT_DATA;
4277	else
4278		/* disable in-core preallocation for non-regular files */
4279		ar.flags = 0;
4280	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4281		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4282	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4283		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4284	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4285		ar.flags |= EXT4_MB_USE_RESERVED;
4286	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4287	if (!newblock)
4288		goto out;
 
 
 
4289	allocated_clusters = ar.len;
4290	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4291	ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4292		  ar.goal, newblock, ar.len, allocated);
4293	if (ar.len > allocated)
4294		ar.len = allocated;
4295
4296got_allocated_blocks:
4297	/* try to insert new extent into found leaf and return */
4298	pblk = newblock + offset;
4299	ext4_ext_store_pblock(&newex, pblk);
4300	newex.ee_len = cpu_to_le16(ar.len);
4301	/* Mark unwritten */
4302	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4303		ext4_ext_mark_unwritten(&newex);
4304		map->m_flags |= EXT4_MAP_UNWRITTEN;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4305	}
4306
4307	err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4308	if (err) {
4309		if (allocated_clusters) {
4310			int fb_flags = 0;
4311
4312			/*
4313			 * free data blocks we just allocated.
4314			 * not a good idea to call discard here directly,
4315			 * but otherwise we'd need to call it every free().
4316			 */
4317			ext4_discard_preallocations(inode, 0);
4318			if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4319				fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4320			ext4_free_blocks(handle, inode, NULL, newblock,
4321					 EXT4_C2B(sbi, allocated_clusters),
4322					 fb_flags);
4323		}
4324		goto out;
4325	}
4326
 
 
 
 
 
 
 
4327	/*
4328	 * Reduce the reserved cluster count to reflect successful deferred
4329	 * allocation of delayed allocated clusters or direct allocation of
4330	 * clusters discovered to be delayed allocated.  Once allocated, a
4331	 * cluster is not included in the reserved count.
4332	 */
4333	if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4334		if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4335			/*
4336			 * When allocating delayed allocated clusters, simply
4337			 * reduce the reserved cluster count and claim quota
4338			 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4339			ext4_da_update_reserve_space(inode, allocated_clusters,
4340							1);
4341		} else {
4342			ext4_lblk_t lblk, len;
4343			unsigned int n;
4344
4345			/*
4346			 * When allocating non-delayed allocated clusters
4347			 * (from fallocate, filemap, DIO, or clusters
4348			 * allocated when delalloc has been disabled by
4349			 * ext4_nonda_switch), reduce the reserved cluster
4350			 * count by the number of allocated clusters that
4351			 * have previously been delayed allocated.  Quota
4352			 * has been claimed by ext4_mb_new_blocks() above,
4353			 * so release the quota reservations made for any
4354			 * previously delayed allocated clusters.
4355			 */
4356			lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4357			len = allocated_clusters << sbi->s_cluster_bits;
4358			n = ext4_es_delayed_clu(inode, lblk, len);
4359			if (n > 0)
4360				ext4_da_update_reserve_space(inode, (int) n, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4361		}
4362	}
4363
4364	/*
4365	 * Cache the extent and update transaction to commit on fdatasync only
4366	 * when it is _not_ an unwritten extent.
4367	 */
4368	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
 
4369		ext4_update_inode_fsync_trans(handle, inode, 1);
4370	else
4371		ext4_update_inode_fsync_trans(handle, inode, 0);
4372
4373	map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4374	map->m_pblk = pblk;
4375	map->m_len = ar.len;
4376	allocated = map->m_len;
4377	ext4_ext_show_leaf(inode, path);
4378out:
4379	ext4_free_ext_path(path);
 
 
 
 
 
 
 
 
 
 
 
 
 
4380
4381	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4382				       err ? err : allocated);
4383	return err ? err : allocated;
4384}
4385
4386int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4387{
 
4388	struct super_block *sb = inode->i_sb;
4389	ext4_lblk_t last_block;
 
 
4390	int err = 0;
4391
4392	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4393	 * TODO: optimization is possible here.
4394	 * Probably we need not scan at all,
4395	 * because page truncation is enough.
4396	 */
4397
4398	/* we have to know where to truncate from in crash case */
4399	EXT4_I(inode)->i_disksize = inode->i_size;
4400	err = ext4_mark_inode_dirty(handle, inode);
4401	if (err)
4402		return err;
4403
4404	last_block = (inode->i_size + sb->s_blocksize - 1)
4405			>> EXT4_BLOCK_SIZE_BITS(sb);
4406retry:
4407	err = ext4_es_remove_extent(inode, last_block,
4408				    EXT_MAX_BLOCKS - last_block);
4409	if (err == -ENOMEM) {
4410		memalloc_retry_wait(GFP_ATOMIC);
4411		goto retry;
4412	}
4413	if (err)
4414		return err;
4415retry_remove_space:
4416	err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4417	if (err == -ENOMEM) {
4418		memalloc_retry_wait(GFP_ATOMIC);
4419		goto retry_remove_space;
4420	}
4421	return err;
4422}
4423
4424static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4425				  ext4_lblk_t len, loff_t new_size,
4426				  int flags)
4427{
4428	struct inode *inode = file_inode(file);
4429	handle_t *handle;
4430	int ret = 0, ret2 = 0, ret3 = 0;
4431	int retries = 0;
4432	int depth = 0;
4433	struct ext4_map_blocks map;
4434	unsigned int credits;
4435	loff_t epos;
4436
4437	BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4438	map.m_lblk = offset;
4439	map.m_len = len;
4440	/*
4441	 * Don't normalize the request if it can fit in one extent so
4442	 * that it doesn't get unnecessarily split into multiple
4443	 * extents.
4444	 */
4445	if (len <= EXT_UNWRITTEN_MAX_LEN)
4446		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4447
 
 
 
4448	/*
4449	 * credits to insert 1 extent into extent tree
 
 
 
 
4450	 */
4451	credits = ext4_chunk_trans_blocks(inode, len);
4452	depth = ext_depth(inode);
4453
4454retry:
4455	while (len) {
4456		/*
4457		 * Recalculate credits when extent tree depth changes.
4458		 */
4459		if (depth != ext_depth(inode)) {
4460			credits = ext4_chunk_trans_blocks(inode, len);
4461			depth = ext_depth(inode);
4462		}
4463
4464		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4465					    credits);
4466		if (IS_ERR(handle)) {
4467			ret = PTR_ERR(handle);
4468			break;
4469		}
4470		ret = ext4_map_blocks(handle, inode, &map, flags);
4471		if (ret <= 0) {
4472			ext4_debug("inode #%lu: block %u: len %u: "
4473				   "ext4_ext_map_blocks returned %d",
4474				   inode->i_ino, map.m_lblk,
4475				   map.m_len, ret);
4476			ext4_mark_inode_dirty(handle, inode);
4477			ext4_journal_stop(handle);
4478			break;
4479		}
4480		/*
4481		 * allow a full retry cycle for any remaining allocations
4482		 */
4483		retries = 0;
4484		map.m_lblk += ret;
4485		map.m_len = len = len - ret;
4486		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4487		inode->i_ctime = current_time(inode);
4488		if (new_size) {
4489			if (epos > new_size)
4490				epos = new_size;
4491			if (ext4_update_inode_size(inode, epos) & 0x1)
4492				inode->i_mtime = inode->i_ctime;
4493		}
4494		ret2 = ext4_mark_inode_dirty(handle, inode);
4495		ext4_update_inode_fsync_trans(handle, inode, 1);
4496		ret3 = ext4_journal_stop(handle);
4497		ret2 = ret3 ? ret3 : ret2;
4498		if (unlikely(ret2))
4499			break;
4500	}
4501	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4502		goto retry;
4503
4504	return ret > 0 ? ret2 : ret;
 
 
4505}
4506
4507static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4508
4509static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4510
4511static long ext4_zero_range(struct file *file, loff_t offset,
4512			    loff_t len, int mode)
4513{
4514	struct inode *inode = file_inode(file);
4515	struct address_space *mapping = file->f_mapping;
4516	handle_t *handle = NULL;
4517	unsigned int max_blocks;
4518	loff_t new_size = 0;
4519	int ret = 0;
4520	int flags;
4521	int credits;
4522	int partial_begin, partial_end;
4523	loff_t start, end;
4524	ext4_lblk_t lblk;
4525	unsigned int blkbits = inode->i_blkbits;
4526
4527	trace_ext4_zero_range(inode, offset, len, mode);
4528
4529	/* Call ext4_force_commit to flush all data in case of data=journal. */
4530	if (ext4_should_journal_data(inode)) {
4531		ret = ext4_force_commit(inode->i_sb);
4532		if (ret)
4533			return ret;
4534	}
4535
4536	/*
4537	 * Round up offset. This is not fallocate, we need to zero out
4538	 * blocks, so convert interior block aligned part of the range to
4539	 * unwritten and possibly manually zero out unaligned parts of the
4540	 * range.
4541	 */
4542	start = round_up(offset, 1 << blkbits);
4543	end = round_down((offset + len), 1 << blkbits);
4544
4545	if (start < offset || end > offset + len)
4546		return -EINVAL;
4547	partial_begin = offset & ((1 << blkbits) - 1);
4548	partial_end = (offset + len) & ((1 << blkbits) - 1);
4549
4550	lblk = start >> blkbits;
4551	max_blocks = (end >> blkbits);
4552	if (max_blocks < lblk)
4553		max_blocks = 0;
4554	else
4555		max_blocks -= lblk;
4556
4557	inode_lock(inode);
4558
 
 
 
 
 
4559	/*
4560	 * Indirect files do not support unwritten extents
 
4561	 */
4562	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4563		ret = -EOPNOTSUPP;
4564		goto out_mutex;
4565	}
4566
4567	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4568	    (offset + len > inode->i_size ||
4569	     offset + len > EXT4_I(inode)->i_disksize)) {
4570		new_size = offset + len;
4571		ret = inode_newsize_ok(inode, new_size);
4572		if (ret)
4573			goto out_mutex;
4574	}
4575
4576	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4577
4578	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4579	inode_dio_wait(inode);
4580
4581	ret = file_modified(file);
4582	if (ret)
4583		goto out_mutex;
4584
4585	/* Preallocate the range including the unaligned edges */
4586	if (partial_begin || partial_end) {
4587		ret = ext4_alloc_file_blocks(file,
4588				round_down(offset, 1 << blkbits) >> blkbits,
4589				(round_up((offset + len), 1 << blkbits) -
4590				 round_down(offset, 1 << blkbits)) >> blkbits,
4591				new_size, flags);
4592		if (ret)
4593			goto out_mutex;
4594
4595	}
4596
4597	/* Zero range excluding the unaligned edges */
4598	if (max_blocks > 0) {
4599		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4600			  EXT4_EX_NOCACHE);
4601
4602		/*
4603		 * Prevent page faults from reinstantiating pages we have
4604		 * released from page cache.
4605		 */
4606		filemap_invalidate_lock(mapping);
4607
4608		ret = ext4_break_layouts(inode);
4609		if (ret) {
4610			filemap_invalidate_unlock(mapping);
4611			goto out_mutex;
4612		}
4613
4614		ret = ext4_update_disksize_before_punch(inode, offset, len);
4615		if (ret) {
4616			filemap_invalidate_unlock(mapping);
4617			goto out_mutex;
4618		}
4619		/* Now release the pages and zero block aligned part of pages */
4620		truncate_pagecache_range(inode, start, end - 1);
4621		inode->i_mtime = inode->i_ctime = current_time(inode);
4622
4623		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4624					     flags);
4625		filemap_invalidate_unlock(mapping);
4626		if (ret)
4627			goto out_mutex;
4628	}
4629	if (!partial_begin && !partial_end)
4630		goto out_mutex;
4631
4632	/*
4633	 * In worst case we have to writeout two nonadjacent unwritten
4634	 * blocks and update the inode
4635	 */
4636	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4637	if (ext4_should_journal_data(inode))
4638		credits += 2;
4639	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4640	if (IS_ERR(handle)) {
4641		ret = PTR_ERR(handle);
4642		ext4_std_error(inode->i_sb, ret);
4643		goto out_mutex;
4644	}
4645
4646	inode->i_mtime = inode->i_ctime = current_time(inode);
4647	if (new_size)
4648		ext4_update_inode_size(inode, new_size);
4649	ret = ext4_mark_inode_dirty(handle, inode);
4650	if (unlikely(ret))
4651		goto out_handle;
4652	/* Zero out partial block at the edges of the range */
4653	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4654	if (ret >= 0)
4655		ext4_update_inode_fsync_trans(handle, inode, 1);
4656
4657	if (file->f_flags & O_SYNC)
4658		ext4_handle_sync(handle);
4659
4660out_handle:
4661	ext4_journal_stop(handle);
4662out_mutex:
4663	inode_unlock(inode);
4664	return ret;
4665}
4666
4667/*
4668 * preallocate space for a file. This implements ext4's fallocate file
4669 * operation, which gets called from sys_fallocate system call.
4670 * For block-mapped files, posix_fallocate should fall back to the method
4671 * of writing zeroes to the required new blocks (the same behavior which is
4672 * expected for file systems which do not support fallocate() system call).
4673 */
4674long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4675{
4676	struct inode *inode = file_inode(file);
4677	loff_t new_size = 0;
 
4678	unsigned int max_blocks;
4679	int ret = 0;
 
 
4680	int flags;
4681	ext4_lblk_t lblk;
4682	unsigned int blkbits = inode->i_blkbits;
4683
4684	/*
4685	 * Encrypted inodes can't handle collapse range or insert
4686	 * range since we would need to re-encrypt blocks with a
4687	 * different IV or XTS tweak (which are based on the logical
4688	 * block number).
4689	 */
4690	if (IS_ENCRYPTED(inode) &&
4691	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4692		return -EOPNOTSUPP;
4693
4694	/* Return error if mode is not supported */
4695	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4696		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4697		     FALLOC_FL_INSERT_RANGE))
4698		return -EOPNOTSUPP;
4699
4700	inode_lock(inode);
4701	ret = ext4_convert_inline_data(inode);
4702	inode_unlock(inode);
4703	if (ret)
4704		goto exit;
4705
4706	if (mode & FALLOC_FL_PUNCH_HOLE) {
4707		ret = ext4_punch_hole(file, offset, len);
4708		goto exit;
4709	}
4710
4711	if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4712		ret = ext4_collapse_range(file, offset, len);
4713		goto exit;
4714	}
4715
4716	if (mode & FALLOC_FL_INSERT_RANGE) {
4717		ret = ext4_insert_range(file, offset, len);
4718		goto exit;
4719	}
4720
4721	if (mode & FALLOC_FL_ZERO_RANGE) {
4722		ret = ext4_zero_range(file, offset, len, mode);
4723		goto exit;
4724	}
4725	trace_ext4_fallocate_enter(inode, offset, len, mode);
4726	lblk = offset >> blkbits;
4727
4728	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4729	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4730
4731	inode_lock(inode);
4732
4733	/*
4734	 * We only support preallocation for extent-based files only
 
4735	 */
4736	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4737		ret = -EOPNOTSUPP;
4738		goto out;
 
 
 
 
 
 
 
 
 
4739	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4740
4741	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4742	    (offset + len > inode->i_size ||
4743	     offset + len > EXT4_I(inode)->i_disksize)) {
4744		new_size = offset + len;
4745		ret = inode_newsize_ok(inode, new_size);
4746		if (ret)
4747			goto out;
4748	}
4749
4750	/* Wait all existing dio workers, newcomers will block on i_rwsem */
4751	inode_dio_wait(inode);
4752
4753	ret = file_modified(file);
4754	if (ret)
4755		goto out;
4756
4757	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4758	if (ret)
4759		goto out;
4760
4761	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4762		ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4763					EXT4_I(inode)->i_sync_tid);
4764	}
4765out:
4766	inode_unlock(inode);
4767	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4768exit:
4769	return ret;
4770}
4771
4772/*
4773 * This function convert a range of blocks to written extents
4774 * The caller of this function will pass the start offset and the size.
4775 * all unwritten extents within this range will be converted to
4776 * written extents.
4777 *
4778 * This function is called from the direct IO end io call back
4779 * function, to convert the fallocated extents after IO is completed.
4780 * Returns 0 on success.
4781 */
4782int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4783				   loff_t offset, ssize_t len)
4784{
 
4785	unsigned int max_blocks;
4786	int ret = 0, ret2 = 0, ret3 = 0;
 
4787	struct ext4_map_blocks map;
4788	unsigned int blkbits = inode->i_blkbits;
4789	unsigned int credits = 0;
4790
4791	map.m_lblk = offset >> blkbits;
4792	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4793
4794	if (!handle) {
4795		/*
4796		 * credits to insert 1 extent into extent tree
4797		 */
4798		credits = ext4_chunk_trans_blocks(inode, max_blocks);
4799	}
 
 
4800	while (ret >= 0 && ret < max_blocks) {
4801		map.m_lblk += ret;
4802		map.m_len = (max_blocks -= ret);
4803		if (credits) {
4804			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4805						    credits);
4806			if (IS_ERR(handle)) {
4807				ret = PTR_ERR(handle);
4808				break;
4809			}
4810		}
4811		ret = ext4_map_blocks(handle, inode, &map,
4812				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4813		if (ret <= 0)
4814			ext4_warning(inode->i_sb,
4815				     "inode #%lu: block %u: len %u: "
4816				     "ext4_ext_map_blocks returned %d",
4817				     inode->i_ino, map.m_lblk,
4818				     map.m_len, ret);
4819		ret2 = ext4_mark_inode_dirty(handle, inode);
4820		if (credits) {
4821			ret3 = ext4_journal_stop(handle);
4822			if (unlikely(ret3))
4823				ret2 = ret3;
4824		}
4825
4826		if (ret <= 0 || ret2)
4827			break;
4828	}
4829	return ret > 0 ? ret2 : ret;
4830}
4831
4832int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4833{
4834	int ret = 0, err = 0;
4835	struct ext4_io_end_vec *io_end_vec;
4836
4837	/*
4838	 * This is somewhat ugly but the idea is clear: When transaction is
4839	 * reserved, everything goes into it. Otherwise we rather start several
4840	 * smaller transactions for conversion of each extent separately.
4841	 */
4842	if (handle) {
4843		handle = ext4_journal_start_reserved(handle,
4844						     EXT4_HT_EXT_CONVERT);
4845		if (IS_ERR(handle))
4846			return PTR_ERR(handle);
4847	}
4848
4849	list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4850		ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4851						     io_end_vec->offset,
4852						     io_end_vec->size);
4853		if (ret)
4854			break;
4855	}
4856
4857	if (handle)
4858		err = ext4_journal_stop(handle);
4859
4860	return ret < 0 ? ret : err;
4861}
4862
4863static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4864{
4865	__u64 physical = 0;
4866	__u64 length = 0;
4867	int blockbits = inode->i_sb->s_blocksize_bits;
4868	int error = 0;
4869	u16 iomap_type;
4870
4871	/* in-inode? */
4872	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4873		struct ext4_iloc iloc;
4874		int offset;	/* offset of xattr in inode */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4875
4876		error = ext4_get_inode_loc(inode, &iloc);
4877		if (error)
4878			return error;
4879		physical = (__u64)iloc.bh->b_blocknr << blockbits;
4880		offset = EXT4_GOOD_OLD_INODE_SIZE +
4881				EXT4_I(inode)->i_extra_isize;
4882		physical += offset;
4883		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4884		brelse(iloc.bh);
4885		iomap_type = IOMAP_INLINE;
4886	} else if (EXT4_I(inode)->i_file_acl) { /* external block */
4887		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4888		length = inode->i_sb->s_blocksize;
4889		iomap_type = IOMAP_MAPPED;
4890	} else {
4891		/* no in-inode or external block for xattr, so return -ENOENT */
4892		error = -ENOENT;
4893		goto out;
4894	}
4895
4896	iomap->addr = physical;
4897	iomap->offset = 0;
4898	iomap->length = length;
4899	iomap->type = iomap_type;
4900	iomap->flags = 0;
 
 
 
 
 
4901out:
4902	return error;
4903}
4904
4905static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4906				  loff_t length, unsigned flags,
4907				  struct iomap *iomap, struct iomap *srcmap)
4908{
4909	int error;
4910
4911	error = ext4_iomap_xattr_fiemap(inode, iomap);
4912	if (error == 0 && (offset >= iomap->length))
4913		error = -ENOENT;
4914	return error;
4915}
4916
4917static const struct iomap_ops ext4_iomap_xattr_ops = {
4918	.iomap_begin		= ext4_iomap_xattr_begin,
4919};
4920
4921static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4922{
4923	u64 maxbytes;
4924
4925	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4926		maxbytes = inode->i_sb->s_maxbytes;
4927	else
4928		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4929
4930	if (*len == 0)
4931		return -EINVAL;
4932	if (start > maxbytes)
4933		return -EFBIG;
4934
4935	/*
4936	 * Shrink request scope to what the fs can actually handle.
4937	 */
4938	if (*len > maxbytes || (maxbytes - *len) < start)
4939		*len = maxbytes - start;
4940	return 0;
4941}
4942
4943int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4944		u64 start, u64 len)
4945{
4946	int error = 0;
4947
4948	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4949		error = ext4_ext_precache(inode);
4950		if (error)
4951			return error;
4952		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4953	}
4954
4955	/*
4956	 * For bitmap files the maximum size limit could be smaller than
4957	 * s_maxbytes, so check len here manually instead of just relying on the
4958	 * generic check.
4959	 */
4960	error = ext4_fiemap_check_ranges(inode, start, &len);
4961	if (error)
4962		return error;
4963
4964	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4965		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4966		return iomap_fiemap(inode, fieinfo, start, len,
4967				    &ext4_iomap_xattr_ops);
4968	}
4969
4970	return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4971}
4972
4973int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4974		      __u64 start, __u64 len)
4975{
4976	ext4_lblk_t start_blk, len_blks;
4977	__u64 last_blk;
4978	int error = 0;
4979
4980	if (ext4_has_inline_data(inode)) {
4981		int has_inline;
4982
4983		down_read(&EXT4_I(inode)->xattr_sem);
4984		has_inline = ext4_has_inline_data(inode);
4985		up_read(&EXT4_I(inode)->xattr_sem);
4986		if (has_inline)
4987			return 0;
4988	}
4989
4990	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4991		error = ext4_ext_precache(inode);
4992		if (error)
4993			return error;
4994		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4995	}
4996
4997	error = fiemap_prep(inode, fieinfo, start, &len, 0);
4998	if (error)
4999		return error;
5000
5001	error = ext4_fiemap_check_ranges(inode, start, &len);
5002	if (error)
5003		return error;
5004
5005	start_blk = start >> inode->i_sb->s_blocksize_bits;
5006	last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5007	if (last_blk >= EXT_MAX_BLOCKS)
5008		last_blk = EXT_MAX_BLOCKS-1;
5009	len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5010
5011	/*
5012	 * Walk the extent tree gathering extent information
5013	 * and pushing extents back to the user.
5014	 */
5015	return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5016}
 
 
 
5017
5018/*
5019 * ext4_ext_shift_path_extents:
5020 * Shift the extents of a path structure lying between path[depth].p_ext
5021 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5022 * if it is right shift or left shift operation.
5023 */
5024static int
5025ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5026			    struct inode *inode, handle_t *handle,
5027			    enum SHIFT_DIRECTION SHIFT)
5028{
5029	int depth, err = 0;
5030	struct ext4_extent *ex_start, *ex_last;
5031	bool update = false;
5032	int credits, restart_credits;
5033	depth = path->p_depth;
5034
5035	while (depth >= 0) {
5036		if (depth == path->p_depth) {
5037			ex_start = path[depth].p_ext;
5038			if (!ex_start)
5039				return -EFSCORRUPTED;
5040
5041			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5042			/* leaf + sb + inode */
5043			credits = 3;
5044			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5045				update = true;
5046				/* extent tree + sb + inode */
5047				credits = depth + 2;
5048			}
5049
5050			restart_credits = ext4_writepage_trans_blocks(inode);
5051			err = ext4_datasem_ensure_credits(handle, inode, credits,
5052					restart_credits, 0);
5053			if (err) {
5054				if (err > 0)
5055					err = -EAGAIN;
 
 
5056				goto out;
5057			}
 
 
 
 
 
 
 
 
 
5058
5059			err = ext4_ext_get_access(handle, inode, path + depth);
5060			if (err)
5061				goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5062
5063			while (ex_start <= ex_last) {
5064				if (SHIFT == SHIFT_LEFT) {
5065					le32_add_cpu(&ex_start->ee_block,
5066						-shift);
5067					/* Try to merge to the left. */
5068					if ((ex_start >
5069					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5070					    &&
5071					    ext4_ext_try_to_merge_right(inode,
5072					    path, ex_start - 1))
5073						ex_last--;
5074					else
5075						ex_start++;
5076				} else {
5077					le32_add_cpu(&ex_last->ee_block, shift);
5078					ext4_ext_try_to_merge_right(inode, path,
5079						ex_last);
5080					ex_last--;
5081				}
 
 
 
 
 
 
 
 
5082			}
5083			err = ext4_ext_dirty(handle, inode, path + depth);
5084			if (err)
5085				goto out;
5086
5087			if (--depth < 0 || !update)
5088				break;
 
 
 
 
 
 
 
 
5089		}
5090
5091		/* Update index too */
5092		err = ext4_ext_get_access(handle, inode, path + depth);
5093		if (err)
5094			goto out;
5095
5096		if (SHIFT == SHIFT_LEFT)
5097			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5098		else
5099			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5100		err = ext4_ext_dirty(handle, inode, path + depth);
5101		if (err)
5102			goto out;
5103
5104		/* we are done if current index is not a starting index */
5105		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5106			break;
5107
5108		depth--;
5109	}
5110
5111out:
5112	return err;
 
 
 
 
 
5113}
 
 
5114
5115/*
5116 * ext4_ext_shift_extents:
5117 * All the extents which lies in the range from @start to the last allocated
5118 * block for the @inode are shifted either towards left or right (depending
5119 * upon @SHIFT) by @shift blocks.
5120 * On success, 0 is returned, error otherwise.
5121 */
5122static int
5123ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5124		       ext4_lblk_t start, ext4_lblk_t shift,
5125		       enum SHIFT_DIRECTION SHIFT)
5126{
5127	struct ext4_ext_path *path;
5128	int ret = 0, depth;
5129	struct ext4_extent *extent;
5130	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5131	ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5132
5133	/* Let path point to the last extent */
5134	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5135				EXT4_EX_NOCACHE);
5136	if (IS_ERR(path))
5137		return PTR_ERR(path);
5138
5139	depth = path->p_depth;
5140	extent = path[depth].p_ext;
5141	if (!extent)
5142		goto out;
5143
5144	stop = le32_to_cpu(extent->ee_block);
5145
5146       /*
5147	* For left shifts, make sure the hole on the left is big enough to
5148	* accommodate the shift.  For right shifts, make sure the last extent
5149	* won't be shifted beyond EXT_MAX_BLOCKS.
5150	*/
5151	if (SHIFT == SHIFT_LEFT) {
5152		path = ext4_find_extent(inode, start - 1, &path,
5153					EXT4_EX_NOCACHE);
5154		if (IS_ERR(path))
5155			return PTR_ERR(path);
5156		depth = path->p_depth;
5157		extent =  path[depth].p_ext;
5158		if (extent) {
5159			ex_start = le32_to_cpu(extent->ee_block);
5160			ex_end = le32_to_cpu(extent->ee_block) +
5161				ext4_ext_get_actual_len(extent);
5162		} else {
5163			ex_start = 0;
5164			ex_end = 0;
5165		}
5166
5167		if ((start == ex_start && shift > ex_start) ||
5168		    (shift > start - ex_end)) {
5169			ret = -EINVAL;
5170			goto out;
5171		}
5172	} else {
5173		if (shift > EXT_MAX_BLOCKS -
5174		    (stop + ext4_ext_get_actual_len(extent))) {
5175			ret = -EINVAL;
5176			goto out;
5177		}
 
 
5178	}
5179
5180	/*
5181	 * In case of left shift, iterator points to start and it is increased
5182	 * till we reach stop. In case of right shift, iterator points to stop
5183	 * and it is decreased till we reach start.
5184	 */
5185again:
5186	ret = 0;
5187	if (SHIFT == SHIFT_LEFT)
5188		iterator = &start;
5189	else
5190		iterator = &stop;
5191
5192	if (tmp != EXT_MAX_BLOCKS)
5193		*iterator = tmp;
5194
5195	/*
5196	 * Its safe to start updating extents.  Start and stop are unsigned, so
5197	 * in case of right shift if extent with 0 block is reached, iterator
5198	 * becomes NULL to indicate the end of the loop.
5199	 */
5200	while (iterator && start <= stop) {
5201		path = ext4_find_extent(inode, *iterator, &path,
5202					EXT4_EX_NOCACHE);
5203		if (IS_ERR(path))
5204			return PTR_ERR(path);
5205		depth = path->p_depth;
5206		extent = path[depth].p_ext;
5207		if (!extent) {
5208			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5209					 (unsigned long) *iterator);
5210			return -EFSCORRUPTED;
5211		}
5212		if (SHIFT == SHIFT_LEFT && *iterator >
5213		    le32_to_cpu(extent->ee_block)) {
5214			/* Hole, move to the next extent */
5215			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5216				path[depth].p_ext++;
5217			} else {
5218				*iterator = ext4_ext_next_allocated_block(path);
5219				continue;
5220			}
5221		}
5222
5223		tmp = *iterator;
5224		if (SHIFT == SHIFT_LEFT) {
5225			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5226			*iterator = le32_to_cpu(extent->ee_block) +
5227					ext4_ext_get_actual_len(extent);
5228		} else {
5229			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5230			if (le32_to_cpu(extent->ee_block) > start)
5231				*iterator = le32_to_cpu(extent->ee_block) - 1;
5232			else if (le32_to_cpu(extent->ee_block) == start)
5233				iterator = NULL;
5234			else {
5235				extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5236				while (le32_to_cpu(extent->ee_block) >= start)
5237					extent--;
5238
5239				if (extent == EXT_LAST_EXTENT(path[depth].p_hdr))
5240					break;
5241
5242				extent++;
5243				iterator = NULL;
5244			}
5245			path[depth].p_ext = extent;
5246		}
5247		ret = ext4_ext_shift_path_extents(path, shift, inode,
5248				handle, SHIFT);
5249		/* iterator can be NULL which means we should break */
5250		if (ret == -EAGAIN)
5251			goto again;
5252		if (ret)
5253			break;
5254	}
5255out:
5256	ext4_free_ext_path(path);
5257	return ret;
5258}
5259
5260/*
5261 * ext4_collapse_range:
5262 * This implements the fallocate's collapse range functionality for ext4
5263 * Returns: 0 and non-zero on error.
 
 
 
 
 
 
 
5264 */
5265static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5266{
5267	struct inode *inode = file_inode(file);
5268	struct super_block *sb = inode->i_sb;
 
5269	struct address_space *mapping = inode->i_mapping;
5270	ext4_lblk_t punch_start, punch_stop;
5271	handle_t *handle;
5272	unsigned int credits;
5273	loff_t new_size, ioffset;
5274	int ret;
5275
5276	/*
5277	 * We need to test this early because xfstests assumes that a
5278	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5279	 * system does not support collapse range.
5280	 */
5281	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5282		return -EOPNOTSUPP;
5283
5284	/* Collapse range works only on fs cluster size aligned regions. */
5285	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5286		return -EINVAL;
5287
5288	trace_ext4_collapse_range(inode, offset, len);
5289
5290	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5291	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5292
5293	/* Call ext4_force_commit to flush all data in case of data=journal. */
5294	if (ext4_should_journal_data(inode)) {
5295		ret = ext4_force_commit(inode->i_sb);
5296		if (ret)
5297			return ret;
5298	}
5299
5300	inode_lock(inode);
5301	/*
5302	 * There is no need to overlap collapse range with EOF, in which case
5303	 * it is effectively a truncate operation
5304	 */
5305	if (offset + len >= inode->i_size) {
5306		ret = -EINVAL;
5307		goto out_mutex;
5308	}
5309
5310	/* Currently just for extent based files */
5311	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5312		ret = -EOPNOTSUPP;
5313		goto out_mutex;
5314	}
5315
5316	/* Wait for existing dio to complete */
5317	inode_dio_wait(inode);
5318
5319	ret = file_modified(file);
5320	if (ret)
5321		goto out_mutex;
5322
5323	/*
5324	 * Prevent page faults from reinstantiating pages we have released from
5325	 * page cache.
5326	 */
5327	filemap_invalidate_lock(mapping);
5328
5329	ret = ext4_break_layouts(inode);
5330	if (ret)
5331		goto out_mmap;
5332
5333	/*
5334	 * Need to round down offset to be aligned with page size boundary
5335	 * for page size > block size.
5336	 */
5337	ioffset = round_down(offset, PAGE_SIZE);
5338	/*
5339	 * Write tail of the last page before removed range since it will get
5340	 * removed from the page cache below.
5341	 */
5342	ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5343	if (ret)
5344		goto out_mmap;
5345	/*
5346	 * Write data that will be shifted to preserve them when discarding
5347	 * page cache below. We are also protected from pages becoming dirty
5348	 * by i_rwsem and invalidate_lock.
5349	 */
5350	ret = filemap_write_and_wait_range(mapping, offset + len,
5351					   LLONG_MAX);
5352	if (ret)
5353		goto out_mmap;
5354	truncate_pagecache(inode, ioffset);
5355
5356	credits = ext4_writepage_trans_blocks(inode);
5357	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5358	if (IS_ERR(handle)) {
5359		ret = PTR_ERR(handle);
5360		goto out_mmap;
5361	}
5362	ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5363
5364	down_write(&EXT4_I(inode)->i_data_sem);
5365	ext4_discard_preallocations(inode, 0);
5366
5367	ret = ext4_es_remove_extent(inode, punch_start,
5368				    EXT_MAX_BLOCKS - punch_start);
5369	if (ret) {
5370		up_write(&EXT4_I(inode)->i_data_sem);
5371		goto out_stop;
5372	}
5373
5374	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5375	if (ret) {
5376		up_write(&EXT4_I(inode)->i_data_sem);
5377		goto out_stop;
5378	}
5379	ext4_discard_preallocations(inode, 0);
5380
5381	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5382				     punch_stop - punch_start, SHIFT_LEFT);
5383	if (ret) {
5384		up_write(&EXT4_I(inode)->i_data_sem);
5385		goto out_stop;
5386	}
5387
5388	new_size = inode->i_size - len;
5389	i_size_write(inode, new_size);
5390	EXT4_I(inode)->i_disksize = new_size;
5391
5392	up_write(&EXT4_I(inode)->i_data_sem);
5393	if (IS_SYNC(inode))
5394		ext4_handle_sync(handle);
5395	inode->i_mtime = inode->i_ctime = current_time(inode);
5396	ret = ext4_mark_inode_dirty(handle, inode);
5397	ext4_update_inode_fsync_trans(handle, inode, 1);
5398
5399out_stop:
5400	ext4_journal_stop(handle);
5401out_mmap:
5402	filemap_invalidate_unlock(mapping);
5403out_mutex:
5404	inode_unlock(inode);
5405	return ret;
5406}
5407
5408/*
5409 * ext4_insert_range:
5410 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5411 * The data blocks starting from @offset to the EOF are shifted by @len
5412 * towards right to create a hole in the @inode. Inode size is increased
5413 * by len bytes.
5414 * Returns 0 on success, error otherwise.
5415 */
5416static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5417{
5418	struct inode *inode = file_inode(file);
5419	struct super_block *sb = inode->i_sb;
5420	struct address_space *mapping = inode->i_mapping;
5421	handle_t *handle;
5422	struct ext4_ext_path *path;
5423	struct ext4_extent *extent;
5424	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5425	unsigned int credits, ee_len;
5426	int ret = 0, depth, split_flag = 0;
5427	loff_t ioffset;
5428
5429	/*
5430	 * We need to test this early because xfstests assumes that an
5431	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5432	 * system does not support insert range.
5433	 */
5434	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5435		return -EOPNOTSUPP;
5436
5437	/* Insert range works only on fs cluster size aligned regions. */
5438	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5439		return -EINVAL;
5440
5441	trace_ext4_insert_range(inode, offset, len);
5442
5443	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5444	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5445
5446	/* Call ext4_force_commit to flush all data in case of data=journal */
5447	if (ext4_should_journal_data(inode)) {
5448		ret = ext4_force_commit(inode->i_sb);
5449		if (ret)
5450			return ret;
5451	}
5452
5453	inode_lock(inode);
5454	/* Currently just for extent based files */
5455	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5456		ret = -EOPNOTSUPP;
5457		goto out_mutex;
5458	}
5459
5460	/* Check whether the maximum file size would be exceeded */
5461	if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5462		ret = -EFBIG;
5463		goto out_mutex;
5464	}
5465
5466	/* Offset must be less than i_size */
5467	if (offset >= inode->i_size) {
5468		ret = -EINVAL;
5469		goto out_mutex;
5470	}
5471
5472	/* Wait for existing dio to complete */
5473	inode_dio_wait(inode);
5474
5475	ret = file_modified(file);
5476	if (ret)
5477		goto out_mutex;
5478
5479	/*
5480	 * Prevent page faults from reinstantiating pages we have released from
5481	 * page cache.
5482	 */
5483	filemap_invalidate_lock(mapping);
5484
5485	ret = ext4_break_layouts(inode);
5486	if (ret)
5487		goto out_mmap;
5488
5489	/*
5490	 * Need to round down to align start offset to page size boundary
5491	 * for page size > block size.
 
 
5492	 */
5493	ioffset = round_down(offset, PAGE_SIZE);
5494	/* Write out all dirty pages */
5495	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5496			LLONG_MAX);
5497	if (ret)
5498		goto out_mmap;
5499	truncate_pagecache(inode, ioffset);
5500
5501	credits = ext4_writepage_trans_blocks(inode);
5502	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5503	if (IS_ERR(handle)) {
5504		ret = PTR_ERR(handle);
5505		goto out_mmap;
5506	}
5507	ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE, handle);
5508
5509	/* Expand file to avoid data loss if there is error while shifting */
5510	inode->i_size += len;
5511	EXT4_I(inode)->i_disksize += len;
5512	inode->i_mtime = inode->i_ctime = current_time(inode);
5513	ret = ext4_mark_inode_dirty(handle, inode);
5514	if (ret)
5515		goto out_stop;
5516
5517	down_write(&EXT4_I(inode)->i_data_sem);
5518	ext4_discard_preallocations(inode, 0);
5519
5520	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5521	if (IS_ERR(path)) {
5522		up_write(&EXT4_I(inode)->i_data_sem);
5523		goto out_stop;
5524	}
5525
5526	depth = ext_depth(inode);
5527	extent = path[depth].p_ext;
5528	if (extent) {
5529		ee_start_lblk = le32_to_cpu(extent->ee_block);
5530		ee_len = ext4_ext_get_actual_len(extent);
5531
5532		/*
5533		 * If offset_lblk is not the starting block of extent, split
5534		 * the extent @offset_lblk
5535		 */
5536		if ((offset_lblk > ee_start_lblk) &&
5537				(offset_lblk < (ee_start_lblk + ee_len))) {
5538			if (ext4_ext_is_unwritten(extent))
5539				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5540					EXT4_EXT_MARK_UNWRIT2;
5541			ret = ext4_split_extent_at(handle, inode, &path,
5542					offset_lblk, split_flag,
5543					EXT4_EX_NOCACHE |
5544					EXT4_GET_BLOCKS_PRE_IO |
5545					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5546		}
5547
5548		ext4_free_ext_path(path);
5549		if (ret < 0) {
5550			up_write(&EXT4_I(inode)->i_data_sem);
5551			goto out_stop;
5552		}
5553	} else {
5554		ext4_free_ext_path(path);
5555	}
5556
5557	ret = ext4_es_remove_extent(inode, offset_lblk,
5558			EXT_MAX_BLOCKS - offset_lblk);
5559	if (ret) {
5560		up_write(&EXT4_I(inode)->i_data_sem);
5561		goto out_stop;
5562	}
5563
5564	/*
5565	 * if offset_lblk lies in a hole which is at start of file, use
5566	 * ee_start_lblk to shift extents
5567	 */
5568	ret = ext4_ext_shift_extents(inode, handle,
5569		max(ee_start_lblk, offset_lblk), len_lblk, SHIFT_RIGHT);
5570
5571	up_write(&EXT4_I(inode)->i_data_sem);
5572	if (IS_SYNC(inode))
5573		ext4_handle_sync(handle);
5574	if (ret >= 0)
5575		ext4_update_inode_fsync_trans(handle, inode, 1);
5576
5577out_stop:
5578	ext4_journal_stop(handle);
5579out_mmap:
5580	filemap_invalidate_unlock(mapping);
5581out_mutex:
5582	inode_unlock(inode);
5583	return ret;
5584}
5585
5586/**
5587 * ext4_swap_extents() - Swap extents between two inodes
5588 * @handle: handle for this transaction
5589 * @inode1:	First inode
5590 * @inode2:	Second inode
5591 * @lblk1:	Start block for first inode
5592 * @lblk2:	Start block for second inode
5593 * @count:	Number of blocks to swap
5594 * @unwritten: Mark second inode's extents as unwritten after swap
5595 * @erp:	Pointer to save error value
5596 *
5597 * This helper routine does exactly what is promise "swap extents". All other
5598 * stuff such as page-cache locking consistency, bh mapping consistency or
5599 * extent's data copying must be performed by caller.
5600 * Locking:
5601 *		i_rwsem is held for both inodes
5602 * 		i_data_sem is locked for write for both inodes
5603 * Assumptions:
5604 *		All pages from requested range are locked for both inodes
5605 */
5606int
5607ext4_swap_extents(handle_t *handle, struct inode *inode1,
5608		  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5609		  ext4_lblk_t count, int unwritten, int *erp)
5610{
5611	struct ext4_ext_path *path1 = NULL;
5612	struct ext4_ext_path *path2 = NULL;
5613	int replaced_count = 0;
5614
5615	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5616	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5617	BUG_ON(!inode_is_locked(inode1));
5618	BUG_ON(!inode_is_locked(inode2));
5619
5620	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5621	if (unlikely(*erp))
5622		return 0;
5623	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5624	if (unlikely(*erp))
5625		return 0;
5626
5627	while (count) {
5628		struct ext4_extent *ex1, *ex2, tmp_ex;
5629		ext4_lblk_t e1_blk, e2_blk;
5630		int e1_len, e2_len, len;
5631		int split = 0;
5632
5633		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5634		if (IS_ERR(path1)) {
5635			*erp = PTR_ERR(path1);
5636			path1 = NULL;
5637		finish:
5638			count = 0;
5639			goto repeat;
5640		}
5641		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5642		if (IS_ERR(path2)) {
5643			*erp = PTR_ERR(path2);
5644			path2 = NULL;
5645			goto finish;
5646		}
5647		ex1 = path1[path1->p_depth].p_ext;
5648		ex2 = path2[path2->p_depth].p_ext;
5649		/* Do we have something to swap ? */
5650		if (unlikely(!ex2 || !ex1))
5651			goto finish;
5652
5653		e1_blk = le32_to_cpu(ex1->ee_block);
5654		e2_blk = le32_to_cpu(ex2->ee_block);
5655		e1_len = ext4_ext_get_actual_len(ex1);
5656		e2_len = ext4_ext_get_actual_len(ex2);
5657
5658		/* Hole handling */
5659		if (!in_range(lblk1, e1_blk, e1_len) ||
5660		    !in_range(lblk2, e2_blk, e2_len)) {
5661			ext4_lblk_t next1, next2;
5662
5663			/* if hole after extent, then go to next extent */
5664			next1 = ext4_ext_next_allocated_block(path1);
5665			next2 = ext4_ext_next_allocated_block(path2);
5666			/* If hole before extent, then shift to that extent */
5667			if (e1_blk > lblk1)
5668				next1 = e1_blk;
5669			if (e2_blk > lblk2)
5670				next2 = e2_blk;
5671			/* Do we have something to swap */
5672			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5673				goto finish;
5674			/* Move to the rightest boundary */
5675			len = next1 - lblk1;
5676			if (len < next2 - lblk2)
5677				len = next2 - lblk2;
5678			if (len > count)
5679				len = count;
5680			lblk1 += len;
5681			lblk2 += len;
5682			count -= len;
5683			goto repeat;
5684		}
5685
5686		/* Prepare left boundary */
5687		if (e1_blk < lblk1) {
5688			split = 1;
5689			*erp = ext4_force_split_extent_at(handle, inode1,
5690						&path1, lblk1, 0);
5691			if (unlikely(*erp))
5692				goto finish;
5693		}
5694		if (e2_blk < lblk2) {
5695			split = 1;
5696			*erp = ext4_force_split_extent_at(handle, inode2,
5697						&path2,  lblk2, 0);
5698			if (unlikely(*erp))
5699				goto finish;
5700		}
5701		/* ext4_split_extent_at() may result in leaf extent split,
5702		 * path must to be revalidated. */
5703		if (split)
5704			goto repeat;
5705
5706		/* Prepare right boundary */
5707		len = count;
5708		if (len > e1_blk + e1_len - lblk1)
5709			len = e1_blk + e1_len - lblk1;
5710		if (len > e2_blk + e2_len - lblk2)
5711			len = e2_blk + e2_len - lblk2;
5712
5713		if (len != e1_len) {
5714			split = 1;
5715			*erp = ext4_force_split_extent_at(handle, inode1,
5716						&path1, lblk1 + len, 0);
5717			if (unlikely(*erp))
5718				goto finish;
5719		}
5720		if (len != e2_len) {
5721			split = 1;
5722			*erp = ext4_force_split_extent_at(handle, inode2,
5723						&path2, lblk2 + len, 0);
5724			if (*erp)
5725				goto finish;
5726		}
5727		/* ext4_split_extent_at() may result in leaf extent split,
5728		 * path must to be revalidated. */
5729		if (split)
5730			goto repeat;
5731
5732		BUG_ON(e2_len != e1_len);
5733		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5734		if (unlikely(*erp))
5735			goto finish;
5736		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5737		if (unlikely(*erp))
5738			goto finish;
5739
5740		/* Both extents are fully inside boundaries. Swap it now */
5741		tmp_ex = *ex1;
5742		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5743		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5744		ex1->ee_len = cpu_to_le16(e2_len);
5745		ex2->ee_len = cpu_to_le16(e1_len);
5746		if (unwritten)
5747			ext4_ext_mark_unwritten(ex2);
5748		if (ext4_ext_is_unwritten(&tmp_ex))
5749			ext4_ext_mark_unwritten(ex1);
5750
5751		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5752		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5753		*erp = ext4_ext_dirty(handle, inode2, path2 +
5754				      path2->p_depth);
5755		if (unlikely(*erp))
5756			goto finish;
5757		*erp = ext4_ext_dirty(handle, inode1, path1 +
5758				      path1->p_depth);
5759		/*
5760		 * Looks scarry ah..? second inode already points to new blocks,
5761		 * and it was successfully dirtied. But luckily error may happen
5762		 * only due to journal error, so full transaction will be
5763		 * aborted anyway.
5764		 */
5765		if (unlikely(*erp))
5766			goto finish;
5767		lblk1 += len;
5768		lblk2 += len;
5769		replaced_count += len;
5770		count -= len;
5771
5772	repeat:
5773		ext4_free_ext_path(path1);
5774		ext4_free_ext_path(path2);
5775		path1 = path2 = NULL;
5776	}
5777	return replaced_count;
5778}
5779
5780/*
5781 * ext4_clu_mapped - determine whether any block in a logical cluster has
5782 *                   been mapped to a physical cluster
5783 *
5784 * @inode - file containing the logical cluster
5785 * @lclu - logical cluster of interest
5786 *
5787 * Returns 1 if any block in the logical cluster is mapped, signifying
5788 * that a physical cluster has been allocated for it.  Otherwise,
5789 * returns 0.  Can also return negative error codes.  Derived from
5790 * ext4_ext_map_blocks().
5791 */
5792int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5793{
5794	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5795	struct ext4_ext_path *path;
5796	int depth, mapped = 0, err = 0;
5797	struct ext4_extent *extent;
5798	ext4_lblk_t first_lblk, first_lclu, last_lclu;
5799
5800	/*
5801	 * if data can be stored inline, the logical cluster isn't
5802	 * mapped - no physical clusters have been allocated, and the
5803	 * file has no extents
5804	 */
5805	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
5806		return 0;
5807
5808	/* search for the extent closest to the first block in the cluster */
5809	path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5810	if (IS_ERR(path)) {
5811		err = PTR_ERR(path);
5812		path = NULL;
5813		goto out;
5814	}
5815
5816	depth = ext_depth(inode);
5817
5818	/*
5819	 * A consistent leaf must not be empty.  This situation is possible,
5820	 * though, _during_ tree modification, and it's why an assert can't
5821	 * be put in ext4_find_extent().
5822	 */
5823	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5824		EXT4_ERROR_INODE(inode,
5825		    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5826				 (unsigned long) EXT4_C2B(sbi, lclu),
5827				 depth, path[depth].p_block);
5828		err = -EFSCORRUPTED;
5829		goto out;
5830	}
5831
5832	extent = path[depth].p_ext;
5833
5834	/* can't be mapped if the extent tree is empty */
5835	if (extent == NULL)
5836		goto out;
5837
5838	first_lblk = le32_to_cpu(extent->ee_block);
5839	first_lclu = EXT4_B2C(sbi, first_lblk);
 
5840
5841	/*
5842	 * Three possible outcomes at this point - found extent spanning
5843	 * the target cluster, to the left of the target cluster, or to the
5844	 * right of the target cluster.  The first two cases are handled here.
5845	 * The last case indicates the target cluster is not mapped.
5846	 */
5847	if (lclu >= first_lclu) {
5848		last_lclu = EXT4_B2C(sbi, first_lblk +
5849				     ext4_ext_get_actual_len(extent) - 1);
5850		if (lclu <= last_lclu) {
5851			mapped = 1;
5852		} else {
5853			first_lblk = ext4_ext_next_allocated_block(path);
5854			first_lclu = EXT4_B2C(sbi, first_lblk);
5855			if (lclu == first_lclu)
5856				mapped = 1;
5857		}
5858	}
5859
5860out:
5861	ext4_free_ext_path(path);
5862
5863	return err ? err : mapped;
5864}
5865
5866/*
5867 * Updates physical block address and unwritten status of extent
5868 * starting at lblk start and of len. If such an extent doesn't exist,
5869 * this function splits the extent tree appropriately to create an
5870 * extent like this.  This function is called in the fast commit
5871 * replay path.  Returns 0 on success and error on failure.
5872 */
5873int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5874			      int len, int unwritten, ext4_fsblk_t pblk)
5875{
5876	struct ext4_ext_path *path = NULL, *ppath;
5877	struct ext4_extent *ex;
5878	int ret;
5879
5880	path = ext4_find_extent(inode, start, NULL, 0);
5881	if (IS_ERR(path))
5882		return PTR_ERR(path);
5883	ex = path[path->p_depth].p_ext;
5884	if (!ex) {
5885		ret = -EFSCORRUPTED;
5886		goto out;
5887	}
5888
5889	if (le32_to_cpu(ex->ee_block) != start ||
5890		ext4_ext_get_actual_len(ex) != len) {
5891		/* We need to split this extent to match our extent first */
5892		ppath = path;
5893		down_write(&EXT4_I(inode)->i_data_sem);
5894		ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5895		up_write(&EXT4_I(inode)->i_data_sem);
5896		if (ret)
5897			goto out;
5898		kfree(path);
5899		path = ext4_find_extent(inode, start, NULL, 0);
5900		if (IS_ERR(path))
5901			return -1;
5902		ppath = path;
5903		ex = path[path->p_depth].p_ext;
5904		WARN_ON(le32_to_cpu(ex->ee_block) != start);
5905		if (ext4_ext_get_actual_len(ex) != len) {
5906			down_write(&EXT4_I(inode)->i_data_sem);
5907			ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5908							 start + len, 1);
5909			up_write(&EXT4_I(inode)->i_data_sem);
5910			if (ret)
5911				goto out;
5912			kfree(path);
5913			path = ext4_find_extent(inode, start, NULL, 0);
5914			if (IS_ERR(path))
5915				return -EINVAL;
5916			ex = path[path->p_depth].p_ext;
5917		}
5918	}
5919	if (unwritten)
5920		ext4_ext_mark_unwritten(ex);
5921	else
5922		ext4_ext_mark_initialized(ex);
5923	ext4_ext_store_pblock(ex, pblk);
5924	down_write(&EXT4_I(inode)->i_data_sem);
5925	ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5926	up_write(&EXT4_I(inode)->i_data_sem);
5927out:
5928	ext4_free_ext_path(path);
5929	ext4_mark_inode_dirty(NULL, inode);
5930	return ret;
5931}
5932
5933/* Try to shrink the extent tree */
5934void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5935{
5936	struct ext4_ext_path *path = NULL;
5937	struct ext4_extent *ex;
5938	ext4_lblk_t old_cur, cur = 0;
5939
5940	while (cur < end) {
5941		path = ext4_find_extent(inode, cur, NULL, 0);
5942		if (IS_ERR(path))
5943			return;
5944		ex = path[path->p_depth].p_ext;
5945		if (!ex) {
5946			ext4_free_ext_path(path);
5947			ext4_mark_inode_dirty(NULL, inode);
5948			return;
5949		}
5950		old_cur = cur;
5951		cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5952		if (cur <= old_cur)
5953			cur = old_cur + 1;
5954		ext4_ext_try_to_merge(NULL, inode, path, ex);
5955		down_write(&EXT4_I(inode)->i_data_sem);
5956		ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5957		up_write(&EXT4_I(inode)->i_data_sem);
5958		ext4_mark_inode_dirty(NULL, inode);
5959		ext4_free_ext_path(path);
5960	}
5961}
5962
5963/* Check if *cur is a hole and if it is, skip it */
5964static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5965{
5966	int ret;
5967	struct ext4_map_blocks map;
5968
5969	map.m_lblk = *cur;
5970	map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5971
5972	ret = ext4_map_blocks(NULL, inode, &map, 0);
5973	if (ret < 0)
5974		return ret;
5975	if (ret != 0)
5976		return 0;
5977	*cur = *cur + map.m_len;
5978	return 0;
5979}
5980
5981/* Count number of blocks used by this inode and update i_blocks */
5982int ext4_ext_replay_set_iblocks(struct inode *inode)
5983{
5984	struct ext4_ext_path *path = NULL, *path2 = NULL;
5985	struct ext4_extent *ex;
5986	ext4_lblk_t cur = 0, end;
5987	int numblks = 0, i, ret = 0;
5988	ext4_fsblk_t cmp1, cmp2;
5989	struct ext4_map_blocks map;
5990
5991	/* Determin the size of the file first */
5992	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5993					EXT4_EX_NOCACHE);
5994	if (IS_ERR(path))
5995		return PTR_ERR(path);
5996	ex = path[path->p_depth].p_ext;
5997	if (!ex) {
5998		ext4_free_ext_path(path);
5999		goto out;
6000	}
6001	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6002	ext4_free_ext_path(path);
6003
6004	/* Count the number of data blocks */
6005	cur = 0;
6006	while (cur < end) {
6007		map.m_lblk = cur;
6008		map.m_len = end - cur;
6009		ret = ext4_map_blocks(NULL, inode, &map, 0);
6010		if (ret < 0)
6011			break;
6012		if (ret > 0)
6013			numblks += ret;
6014		cur = cur + map.m_len;
6015	}
6016
6017	/*
6018	 * Count the number of extent tree blocks. We do it by looking up
6019	 * two successive extents and determining the difference between
6020	 * their paths. When path is different for 2 successive extents
6021	 * we compare the blocks in the path at each level and increment
6022	 * iblocks by total number of differences found.
6023	 */
6024	cur = 0;
6025	ret = skip_hole(inode, &cur);
6026	if (ret < 0)
6027		goto out;
6028	path = ext4_find_extent(inode, cur, NULL, 0);
6029	if (IS_ERR(path))
6030		goto out;
6031	numblks += path->p_depth;
6032	ext4_free_ext_path(path);
6033	while (cur < end) {
6034		path = ext4_find_extent(inode, cur, NULL, 0);
6035		if (IS_ERR(path))
6036			break;
6037		ex = path[path->p_depth].p_ext;
6038		if (!ex) {
6039			ext4_free_ext_path(path);
6040			return 0;
6041		}
6042		cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6043					ext4_ext_get_actual_len(ex));
6044		ret = skip_hole(inode, &cur);
6045		if (ret < 0) {
6046			ext4_free_ext_path(path);
6047			break;
6048		}
6049		path2 = ext4_find_extent(inode, cur, NULL, 0);
6050		if (IS_ERR(path2)) {
6051			ext4_free_ext_path(path);
6052			break;
6053		}
6054		for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6055			cmp1 = cmp2 = 0;
6056			if (i <= path->p_depth)
6057				cmp1 = path[i].p_bh ?
6058					path[i].p_bh->b_blocknr : 0;
6059			if (i <= path2->p_depth)
6060				cmp2 = path2[i].p_bh ?
6061					path2[i].p_bh->b_blocknr : 0;
6062			if (cmp1 != cmp2 && cmp2 != 0)
6063				numblks++;
6064		}
6065		ext4_free_ext_path(path);
6066		ext4_free_ext_path(path2);
6067	}
6068
6069out:
6070	inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6071	ext4_mark_inode_dirty(NULL, inode);
6072	return 0;
6073}
6074
6075int ext4_ext_clear_bb(struct inode *inode)
6076{
6077	struct ext4_ext_path *path = NULL;
6078	struct ext4_extent *ex;
6079	ext4_lblk_t cur = 0, end;
6080	int j, ret = 0;
6081	struct ext4_map_blocks map;
6082
6083	if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6084		return 0;
 
 
 
6085
6086	/* Determin the size of the file first */
6087	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6088					EXT4_EX_NOCACHE);
6089	if (IS_ERR(path))
6090		return PTR_ERR(path);
6091	ex = path[path->p_depth].p_ext;
6092	if (!ex) {
6093		ext4_free_ext_path(path);
6094		return 0;
6095	}
6096	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6097	ext4_free_ext_path(path);
6098
6099	cur = 0;
6100	while (cur < end) {
6101		map.m_lblk = cur;
6102		map.m_len = end - cur;
6103		ret = ext4_map_blocks(NULL, inode, &map, 0);
6104		if (ret < 0)
6105			break;
6106		if (ret > 0) {
6107			path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6108			if (!IS_ERR_OR_NULL(path)) {
6109				for (j = 0; j < path->p_depth; j++) {
6110
6111					ext4_mb_mark_bb(inode->i_sb,
6112							path[j].p_block, 1, 0);
6113					ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6114							0, path[j].p_block, 1, 1);
6115				}
6116				ext4_free_ext_path(path);
6117			}
6118			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6119			ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6120					map.m_lblk, map.m_pblk, map.m_len, 1);
6121		}
6122		cur = cur + map.m_len;
6123	}
6124
6125	return 0;
6126}
v3.5.6
 
   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 Licens
  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/falloc.h>
  41#include <asm/uaccess.h>
  42#include <linux/fiemap.h>
 
 
  43#include "ext4_jbd2.h"
 
 
  44
  45#include <trace/events/ext4.h>
  46
  47/*
  48 * used by extent splitting.
  49 */
  50#define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
  51					due to ENOSPC */
  52#define EXT4_EXT_MARK_UNINIT1	0x2  /* mark first half uninitialized */
  53#define EXT4_EXT_MARK_UNINIT2	0x4  /* mark second half uninitialized */
 
 
 
  54
  55static __le32 ext4_extent_block_csum(struct inode *inode,
  56				     struct ext4_extent_header *eh)
  57{
  58	struct ext4_inode_info *ei = EXT4_I(inode);
  59	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  60	__u32 csum;
  61
  62	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  63			   EXT4_EXTENT_TAIL_OFFSET(eh));
  64	return cpu_to_le32(csum);
  65}
  66
  67static int ext4_extent_block_csum_verify(struct inode *inode,
  68					 struct ext4_extent_header *eh)
  69{
  70	struct ext4_extent_tail *et;
  71
  72	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  73		EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  74		return 1;
  75
  76	et = find_ext4_extent_tail(eh);
  77	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  78		return 0;
  79	return 1;
  80}
  81
  82static void ext4_extent_block_csum_set(struct inode *inode,
  83				       struct ext4_extent_header *eh)
  84{
  85	struct ext4_extent_tail *et;
  86
  87	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  88		EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  89		return;
  90
  91	et = find_ext4_extent_tail(eh);
  92	et->et_checksum = ext4_extent_block_csum(inode, eh);
  93}
  94
  95static int ext4_split_extent(handle_t *handle,
  96				struct inode *inode,
  97				struct ext4_ext_path *path,
  98				struct ext4_map_blocks *map,
  99				int split_flag,
 100				int flags);
 101
 102static int ext4_split_extent_at(handle_t *handle,
 103			     struct inode *inode,
 104			     struct ext4_ext_path *path,
 105			     ext4_lblk_t split,
 106			     int split_flag,
 107			     int flags);
 108
 109static int ext4_ext_truncate_extend_restart(handle_t *handle,
 110					    struct inode *inode,
 111					    int needed)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 112{
 113	int err;
 
 
 114
 115	if (!ext4_handle_valid(handle))
 116		return 0;
 117	if (handle->h_buffer_credits > needed)
 118		return 0;
 119	err = ext4_journal_extend(handle, needed);
 120	if (err <= 0)
 121		return err;
 122	err = ext4_truncate_restart_trans(handle, inode, needed);
 123	if (err == 0)
 124		err = -EAGAIN;
 
 
 
 
 125
 126	return err;
 
 
 
 
 127}
 128
 129/*
 130 * could return:
 131 *  - EROFS
 132 *  - ENOMEM
 133 */
 134static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 135				struct ext4_ext_path *path)
 136{
 
 
 137	if (path->p_bh) {
 138		/* path points to block */
 139		return ext4_journal_get_write_access(handle, path->p_bh);
 
 
 
 
 
 
 
 
 
 
 140	}
 141	/* path points to leaf/index in inode body */
 142	/* we use in-core data, no need to protect them */
 143	return 0;
 144}
 145
 146/*
 147 * could return:
 148 *  - EROFS
 149 *  - ENOMEM
 150 *  - EIO
 151 */
 152#define ext4_ext_dirty(handle, inode, path) \
 153		__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 154static int __ext4_ext_dirty(const char *where, unsigned int line,
 155			    handle_t *handle, struct inode *inode,
 156			    struct ext4_ext_path *path)
 157{
 158	int err;
 
 
 159	if (path->p_bh) {
 160		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 161		/* path points to block */
 162		err = __ext4_handle_dirty_metadata(where, line, handle,
 163						   inode, path->p_bh);
 
 
 
 164	} else {
 165		/* path points to leaf/index in inode body */
 166		err = ext4_mark_inode_dirty(handle, inode);
 167	}
 168	return err;
 169}
 170
 
 
 
 171static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 172			      struct ext4_ext_path *path,
 173			      ext4_lblk_t block)
 174{
 175	if (path) {
 176		int depth = path->p_depth;
 177		struct ext4_extent *ex;
 178
 179		/*
 180		 * Try to predict block placement assuming that we are
 181		 * filling in a file which will eventually be
 182		 * non-sparse --- i.e., in the case of libbfd writing
 183		 * an ELF object sections out-of-order but in a way
 184		 * the eventually results in a contiguous object or
 185		 * executable file, or some database extending a table
 186		 * space file.  However, this is actually somewhat
 187		 * non-ideal if we are writing a sparse file such as
 188		 * qemu or KVM writing a raw image file that is going
 189		 * to stay fairly sparse, since it will end up
 190		 * fragmenting the file system's free space.  Maybe we
 191		 * should have some hueristics or some way to allow
 192		 * userspace to pass a hint to file system,
 193		 * especially if the latter case turns out to be
 194		 * common.
 195		 */
 196		ex = path[depth].p_ext;
 197		if (ex) {
 198			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 199			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 200
 201			if (block > ext_block)
 202				return ext_pblk + (block - ext_block);
 203			else
 204				return ext_pblk - (ext_block - block);
 205		}
 206
 207		/* it looks like index is empty;
 208		 * try to find starting block from index itself */
 209		if (path[depth].p_bh)
 210			return path[depth].p_bh->b_blocknr;
 211	}
 212
 213	/* OK. use inode's group */
 214	return ext4_inode_to_goal_block(inode);
 215}
 216
 217/*
 218 * Allocation for a meta data block
 219 */
 220static ext4_fsblk_t
 221ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 222			struct ext4_ext_path *path,
 223			struct ext4_extent *ex, int *err, unsigned int flags)
 224{
 225	ext4_fsblk_t goal, newblock;
 226
 227	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 228	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 229					NULL, err);
 230	return newblock;
 231}
 232
 233static inline int ext4_ext_space_block(struct inode *inode, int check)
 234{
 235	int size;
 236
 237	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 238			/ sizeof(struct ext4_extent);
 239#ifdef AGGRESSIVE_TEST
 240	if (!check && size > 6)
 241		size = 6;
 242#endif
 243	return size;
 244}
 245
 246static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 247{
 248	int size;
 249
 250	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 251			/ sizeof(struct ext4_extent_idx);
 252#ifdef AGGRESSIVE_TEST
 253	if (!check && size > 5)
 254		size = 5;
 255#endif
 256	return size;
 257}
 258
 259static inline int ext4_ext_space_root(struct inode *inode, int check)
 260{
 261	int size;
 262
 263	size = sizeof(EXT4_I(inode)->i_data);
 264	size -= sizeof(struct ext4_extent_header);
 265	size /= sizeof(struct ext4_extent);
 266#ifdef AGGRESSIVE_TEST
 267	if (!check && size > 3)
 268		size = 3;
 269#endif
 270	return size;
 271}
 272
 273static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 274{
 275	int size;
 276
 277	size = sizeof(EXT4_I(inode)->i_data);
 278	size -= sizeof(struct ext4_extent_header);
 279	size /= sizeof(struct ext4_extent_idx);
 280#ifdef AGGRESSIVE_TEST
 281	if (!check && size > 4)
 282		size = 4;
 283#endif
 284	return size;
 285}
 286
 287/*
 288 * Calculate the number of metadata blocks needed
 289 * to allocate @blocks
 290 * Worse case is one block per extent
 291 */
 292int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 293{
 294	struct ext4_inode_info *ei = EXT4_I(inode);
 295	int idxs;
 296
 297	idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 298		/ sizeof(struct ext4_extent_idx));
 299
 300	/*
 301	 * If the new delayed allocation block is contiguous with the
 302	 * previous da block, it can share index blocks with the
 303	 * previous block, so we only need to allocate a new index
 304	 * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 305	 * an additional index block, and at ldxs**3 blocks, yet
 306	 * another index blocks.
 307	 */
 308	if (ei->i_da_metadata_calc_len &&
 309	    ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 310		int num = 0;
 311
 312		if ((ei->i_da_metadata_calc_len % idxs) == 0)
 313			num++;
 314		if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 315			num++;
 316		if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 317			num++;
 318			ei->i_da_metadata_calc_len = 0;
 319		} else
 320			ei->i_da_metadata_calc_len++;
 321		ei->i_da_metadata_calc_last_lblock++;
 322		return num;
 323	}
 324
 325	/*
 326	 * In the worst case we need a new set of index blocks at
 327	 * every level of the inode's extent tree.
 328	 */
 329	ei->i_da_metadata_calc_len = 1;
 330	ei->i_da_metadata_calc_last_lblock = lblock;
 331	return ext_depth(inode) + 1;
 332}
 333
 334static int
 335ext4_ext_max_entries(struct inode *inode, int depth)
 336{
 337	int max;
 338
 339	if (depth == ext_depth(inode)) {
 340		if (depth == 0)
 341			max = ext4_ext_space_root(inode, 1);
 342		else
 343			max = ext4_ext_space_root_idx(inode, 1);
 344	} else {
 345		if (depth == 0)
 346			max = ext4_ext_space_block(inode, 1);
 347		else
 348			max = ext4_ext_space_block_idx(inode, 1);
 349	}
 350
 351	return max;
 352}
 353
 354static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 355{
 356	ext4_fsblk_t block = ext4_ext_pblock(ext);
 357	int len = ext4_ext_get_actual_len(ext);
 
 358
 359	if (len == 0)
 
 
 
 
 
 360		return 0;
 361	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 362}
 363
 364static int ext4_valid_extent_idx(struct inode *inode,
 365				struct ext4_extent_idx *ext_idx)
 366{
 367	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 368
 369	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 370}
 371
 372static int ext4_valid_extent_entries(struct inode *inode,
 373				struct ext4_extent_header *eh,
 374				int depth)
 
 375{
 376	unsigned short entries;
 
 
 
 377	if (eh->eh_entries == 0)
 378		return 1;
 379
 380	entries = le16_to_cpu(eh->eh_entries);
 381
 382	if (depth == 0) {
 383		/* leaf entries */
 384		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 
 
 
 
 
 
 
 
 385		while (entries) {
 386			if (!ext4_valid_extent(inode, ext))
 387				return 0;
 
 
 
 
 
 
 
 
 388			ext++;
 389			entries--;
 390		}
 391	} else {
 392		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 
 
 
 
 
 
 
 
 393		while (entries) {
 394			if (!ext4_valid_extent_idx(inode, ext_idx))
 395				return 0;
 
 
 
 
 
 
 
 396			ext_idx++;
 397			entries--;
 
 398		}
 399	}
 400	return 1;
 401}
 402
 403static int __ext4_ext_check(const char *function, unsigned int line,
 404			    struct inode *inode, struct ext4_extent_header *eh,
 405			    int depth)
 406{
 407	const char *error_msg;
 408	int max = 0;
 409
 410	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 411		error_msg = "invalid magic";
 412		goto corrupted;
 413	}
 414	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 415		error_msg = "unexpected eh_depth";
 416		goto corrupted;
 417	}
 418	if (unlikely(eh->eh_max == 0)) {
 419		error_msg = "invalid eh_max";
 420		goto corrupted;
 421	}
 422	max = ext4_ext_max_entries(inode, depth);
 423	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 424		error_msg = "too large eh_max";
 425		goto corrupted;
 426	}
 427	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 428		error_msg = "invalid eh_entries";
 429		goto corrupted;
 430	}
 431	if (!ext4_valid_extent_entries(inode, eh, depth)) {
 
 
 
 
 432		error_msg = "invalid extent entries";
 433		goto corrupted;
 434	}
 
 
 
 
 435	/* Verify checksum on non-root extent tree nodes */
 436	if (ext_depth(inode) != depth &&
 437	    !ext4_extent_block_csum_verify(inode, eh)) {
 438		error_msg = "extent tree corrupted";
 
 439		goto corrupted;
 440	}
 441	return 0;
 442
 443corrupted:
 444	ext4_error_inode(inode, function, line, 0,
 445			"bad header/extent: %s - magic %x, "
 446			"entries %u, max %u(%u), depth %u(%u)",
 447			error_msg, le16_to_cpu(eh->eh_magic),
 448			le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 449			max, le16_to_cpu(eh->eh_depth), depth);
 450
 451	return -EIO;
 
 452}
 453
 454#define ext4_ext_check(inode, eh, depth)	\
 455	__ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 456
 457int ext4_ext_check_inode(struct inode *inode)
 458{
 459	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 460}
 461
 462static int __ext4_ext_check_block(const char *function, unsigned int line,
 463				  struct inode *inode,
 464				  struct ext4_extent_header *eh,
 465				  int depth,
 466				  struct buffer_head *bh)
 
 
 
 
 467{
 468	int ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 469
 470	if (buffer_verified(bh))
 471		return 0;
 472	ret = ext4_ext_check(inode, eh, depth);
 473	if (ret)
 474		return ret;
 475	set_buffer_verified(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 476	return ret;
 477}
 478
 479#define ext4_ext_check_block(inode, eh, depth, bh)	\
 480	__ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
 481
 482#ifdef EXT_DEBUG
 483static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 484{
 485	int k, l = path->p_depth;
 486
 487	ext_debug("path:");
 488	for (k = 0; k <= l; k++, path++) {
 489		if (path->p_idx) {
 490		  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 491			    ext4_idx_pblock(path->p_idx));
 
 492		} else if (path->p_ext) {
 493			ext_debug("  %d:[%d]%d:%llu ",
 494				  le32_to_cpu(path->p_ext->ee_block),
 495				  ext4_ext_is_uninitialized(path->p_ext),
 496				  ext4_ext_get_actual_len(path->p_ext),
 497				  ext4_ext_pblock(path->p_ext));
 498		} else
 499			ext_debug("  []");
 500	}
 501	ext_debug("\n");
 502}
 503
 504static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 505{
 506	int depth = ext_depth(inode);
 507	struct ext4_extent_header *eh;
 508	struct ext4_extent *ex;
 509	int i;
 510
 511	if (!path)
 512		return;
 513
 514	eh = path[depth].p_hdr;
 515	ex = EXT_FIRST_EXTENT(eh);
 516
 517	ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 518
 519	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 520		ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 521			  ext4_ext_is_uninitialized(ex),
 522			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 523	}
 524	ext_debug("\n");
 525}
 526
 527static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 528			ext4_fsblk_t newblock, int level)
 529{
 530	int depth = ext_depth(inode);
 531	struct ext4_extent *ex;
 532
 533	if (depth != level) {
 534		struct ext4_extent_idx *idx;
 535		idx = path[level].p_idx;
 536		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 537			ext_debug("%d: move %d:%llu in new index %llu\n", level,
 538					le32_to_cpu(idx->ei_block),
 539					ext4_idx_pblock(idx),
 540					newblock);
 541			idx++;
 542		}
 543
 544		return;
 545	}
 546
 547	ex = path[depth].p_ext;
 548	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 549		ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 550				le32_to_cpu(ex->ee_block),
 551				ext4_ext_pblock(ex),
 552				ext4_ext_is_uninitialized(ex),
 553				ext4_ext_get_actual_len(ex),
 554				newblock);
 555		ex++;
 556	}
 557}
 558
 559#else
 560#define ext4_ext_show_path(inode, path)
 561#define ext4_ext_show_leaf(inode, path)
 562#define ext4_ext_show_move(inode, path, newblock, level)
 563#endif
 564
 565void ext4_ext_drop_refs(struct ext4_ext_path *path)
 566{
 567	int depth = path->p_depth;
 568	int i;
 569
 570	for (i = 0; i <= depth; i++, path++)
 571		if (path->p_bh) {
 572			brelse(path->p_bh);
 573			path->p_bh = NULL;
 574		}
 575}
 576
 577/*
 578 * ext4_ext_binsearch_idx:
 579 * binary search for the closest index of the given block
 580 * the header must be checked before calling this
 581 */
 582static void
 583ext4_ext_binsearch_idx(struct inode *inode,
 584			struct ext4_ext_path *path, ext4_lblk_t block)
 585{
 586	struct ext4_extent_header *eh = path->p_hdr;
 587	struct ext4_extent_idx *r, *l, *m;
 588
 589
 590	ext_debug("binsearch for %u(idx):  ", block);
 591
 592	l = EXT_FIRST_INDEX(eh) + 1;
 593	r = EXT_LAST_INDEX(eh);
 594	while (l <= r) {
 595		m = l + (r - l) / 2;
 
 
 
 
 596		if (block < le32_to_cpu(m->ei_block))
 597			r = m - 1;
 598		else
 599			l = m + 1;
 600		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 601				m, le32_to_cpu(m->ei_block),
 602				r, le32_to_cpu(r->ei_block));
 603	}
 604
 605	path->p_idx = l - 1;
 606	ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 607		  ext4_idx_pblock(path->p_idx));
 608
 609#ifdef CHECK_BINSEARCH
 610	{
 611		struct ext4_extent_idx *chix, *ix;
 612		int k;
 613
 614		chix = ix = EXT_FIRST_INDEX(eh);
 615		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 616		  if (k != 0 &&
 617		      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 618				printk(KERN_DEBUG "k=%d, ix=0x%p, "
 619				       "first=0x%p\n", k,
 620				       ix, EXT_FIRST_INDEX(eh));
 621				printk(KERN_DEBUG "%u <= %u\n",
 622				       le32_to_cpu(ix->ei_block),
 623				       le32_to_cpu(ix[-1].ei_block));
 624			}
 625			BUG_ON(k && le32_to_cpu(ix->ei_block)
 626					   <= le32_to_cpu(ix[-1].ei_block));
 627			if (block < le32_to_cpu(ix->ei_block))
 628				break;
 629			chix = ix;
 630		}
 631		BUG_ON(chix != path->p_idx);
 632	}
 633#endif
 634
 635}
 636
 637/*
 638 * ext4_ext_binsearch:
 639 * binary search for closest extent of the given block
 640 * the header must be checked before calling this
 641 */
 642static void
 643ext4_ext_binsearch(struct inode *inode,
 644		struct ext4_ext_path *path, ext4_lblk_t block)
 645{
 646	struct ext4_extent_header *eh = path->p_hdr;
 647	struct ext4_extent *r, *l, *m;
 648
 649	if (eh->eh_entries == 0) {
 650		/*
 651		 * this leaf is empty:
 652		 * we get such a leaf in split/add case
 653		 */
 654		return;
 655	}
 656
 657	ext_debug("binsearch for %u:  ", block);
 658
 659	l = EXT_FIRST_EXTENT(eh) + 1;
 660	r = EXT_LAST_EXTENT(eh);
 661
 662	while (l <= r) {
 663		m = l + (r - l) / 2;
 
 
 
 
 664		if (block < le32_to_cpu(m->ee_block))
 665			r = m - 1;
 666		else
 667			l = m + 1;
 668		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 669				m, le32_to_cpu(m->ee_block),
 670				r, le32_to_cpu(r->ee_block));
 671	}
 672
 673	path->p_ext = l - 1;
 674	ext_debug("  -> %d:%llu:[%d]%d ",
 675			le32_to_cpu(path->p_ext->ee_block),
 676			ext4_ext_pblock(path->p_ext),
 677			ext4_ext_is_uninitialized(path->p_ext),
 678			ext4_ext_get_actual_len(path->p_ext));
 679
 680#ifdef CHECK_BINSEARCH
 681	{
 682		struct ext4_extent *chex, *ex;
 683		int k;
 684
 685		chex = ex = EXT_FIRST_EXTENT(eh);
 686		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 687			BUG_ON(k && le32_to_cpu(ex->ee_block)
 688					  <= le32_to_cpu(ex[-1].ee_block));
 689			if (block < le32_to_cpu(ex->ee_block))
 690				break;
 691			chex = ex;
 692		}
 693		BUG_ON(chex != path->p_ext);
 694	}
 695#endif
 696
 697}
 698
 699int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 700{
 701	struct ext4_extent_header *eh;
 702
 703	eh = ext_inode_hdr(inode);
 704	eh->eh_depth = 0;
 705	eh->eh_entries = 0;
 706	eh->eh_magic = EXT4_EXT_MAGIC;
 707	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 
 708	ext4_mark_inode_dirty(handle, inode);
 709	ext4_ext_invalidate_cache(inode);
 710	return 0;
 711}
 712
 713struct ext4_ext_path *
 714ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 715					struct ext4_ext_path *path)
 716{
 717	struct ext4_extent_header *eh;
 718	struct buffer_head *bh;
 719	short int depth, i, ppos = 0, alloc = 0;
 
 
 
 
 
 
 720
 721	eh = ext_inode_hdr(inode);
 722	depth = ext_depth(inode);
 
 
 
 
 
 
 723
 724	/* account possible depth increase */
 
 
 
 
 
 
 725	if (!path) {
 726		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 727				GFP_NOFS);
 728		if (!path)
 
 729			return ERR_PTR(-ENOMEM);
 730		alloc = 1;
 731	}
 732	path[0].p_hdr = eh;
 733	path[0].p_bh = NULL;
 734
 735	i = depth;
 
 
 736	/* walk through the tree */
 737	while (i) {
 738		ext_debug("depth %d: num %d, max %d\n",
 739			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 740
 741		ext4_ext_binsearch_idx(inode, path + ppos, block);
 742		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 743		path[ppos].p_depth = i;
 744		path[ppos].p_ext = NULL;
 745
 746		bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 747		if (unlikely(!bh))
 
 748			goto err;
 749		if (!bh_uptodate_or_lock(bh)) {
 750			trace_ext4_ext_load_extent(inode, block,
 751						path[ppos].p_block);
 752			if (bh_submit_read(bh) < 0) {
 753				put_bh(bh);
 754				goto err;
 755			}
 756		}
 
 757		eh = ext_block_hdr(bh);
 758		ppos++;
 759		if (unlikely(ppos > depth)) {
 760			put_bh(bh);
 761			EXT4_ERROR_INODE(inode,
 762					 "ppos %d > depth %d", ppos, depth);
 763			goto err;
 764		}
 765		path[ppos].p_bh = bh;
 766		path[ppos].p_hdr = eh;
 767		i--;
 768
 769		if (ext4_ext_check_block(inode, eh, i, bh))
 770			goto err;
 771	}
 772
 773	path[ppos].p_depth = i;
 774	path[ppos].p_ext = NULL;
 775	path[ppos].p_idx = NULL;
 776
 777	/* find extent */
 778	ext4_ext_binsearch(inode, path + ppos, block);
 779	/* if not an empty leaf */
 780	if (path[ppos].p_ext)
 781		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 782
 783	ext4_ext_show_path(inode, path);
 784
 785	return path;
 786
 787err:
 788	ext4_ext_drop_refs(path);
 789	if (alloc)
 790		kfree(path);
 791	return ERR_PTR(-EIO);
 792}
 793
 794/*
 795 * ext4_ext_insert_index:
 796 * insert new index [@logical;@ptr] into the block at @curp;
 797 * check where to insert: before @curp or after @curp
 798 */
 799static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 800				 struct ext4_ext_path *curp,
 801				 int logical, ext4_fsblk_t ptr)
 802{
 803	struct ext4_extent_idx *ix;
 804	int len, err;
 805
 806	err = ext4_ext_get_access(handle, inode, curp);
 807	if (err)
 808		return err;
 809
 810	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 811		EXT4_ERROR_INODE(inode,
 812				 "logical %d == ei_block %d!",
 813				 logical, le32_to_cpu(curp->p_idx->ei_block));
 814		return -EIO;
 815	}
 816
 817	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 818			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
 819		EXT4_ERROR_INODE(inode,
 820				 "eh_entries %d >= eh_max %d!",
 821				 le16_to_cpu(curp->p_hdr->eh_entries),
 822				 le16_to_cpu(curp->p_hdr->eh_max));
 823		return -EIO;
 824	}
 825
 826	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 827		/* insert after */
 828		ext_debug("insert new index %d after: %llu\n", logical, ptr);
 
 829		ix = curp->p_idx + 1;
 830	} else {
 831		/* insert before */
 832		ext_debug("insert new index %d before: %llu\n", logical, ptr);
 
 833		ix = curp->p_idx;
 834	}
 835
 836	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 837	BUG_ON(len < 0);
 838	if (len > 0) {
 839		ext_debug("insert new index %d: "
 840				"move %d indices from 0x%p to 0x%p\n",
 841				logical, len, ix, ix + 1);
 842		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 843	}
 844
 845	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 846		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 847		return -EIO;
 848	}
 849
 850	ix->ei_block = cpu_to_le32(logical);
 851	ext4_idx_store_pblock(ix, ptr);
 852	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 853
 854	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 855		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 856		return -EIO;
 857	}
 858
 859	err = ext4_ext_dirty(handle, inode, curp);
 860	ext4_std_error(inode->i_sb, err);
 861
 862	return err;
 863}
 864
 865/*
 866 * ext4_ext_split:
 867 * inserts new subtree into the path, using free index entry
 868 * at depth @at:
 869 * - allocates all needed blocks (new leaf and all intermediate index blocks)
 870 * - makes decision where to split
 871 * - moves remaining extents and index entries (right to the split point)
 872 *   into the newly allocated blocks
 873 * - initializes subtree
 874 */
 875static int ext4_ext_split(handle_t *handle, struct inode *inode,
 876			  unsigned int flags,
 877			  struct ext4_ext_path *path,
 878			  struct ext4_extent *newext, int at)
 879{
 880	struct buffer_head *bh = NULL;
 881	int depth = ext_depth(inode);
 882	struct ext4_extent_header *neh;
 883	struct ext4_extent_idx *fidx;
 884	int i = at, k, m, a;
 885	ext4_fsblk_t newblock, oldblock;
 886	__le32 border;
 887	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 
 888	int err = 0;
 
 
 
 
 889
 890	/* make decision: where to split? */
 891	/* FIXME: now decision is simplest: at current extent */
 892
 893	/* if current leaf will be split, then we should use
 894	 * border from split point */
 895	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 896		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 897		return -EIO;
 898	}
 899	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 900		border = path[depth].p_ext[1].ee_block;
 901		ext_debug("leaf will be split."
 902				" next leaf starts at %d\n",
 903				  le32_to_cpu(border));
 904	} else {
 905		border = newext->ee_block;
 906		ext_debug("leaf will be added."
 907				" next leaf starts at %d\n",
 908				le32_to_cpu(border));
 909	}
 910
 911	/*
 912	 * If error occurs, then we break processing
 913	 * and mark filesystem read-only. index won't
 914	 * be inserted and tree will be in consistent
 915	 * state. Next mount will repair buffers too.
 916	 */
 917
 918	/*
 919	 * Get array to track all allocated blocks.
 920	 * We need this to handle errors and free blocks
 921	 * upon them.
 922	 */
 923	ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 924	if (!ablocks)
 925		return -ENOMEM;
 926
 927	/* allocate all needed blocks */
 928	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 929	for (a = 0; a < depth - at; a++) {
 930		newblock = ext4_ext_new_meta_block(handle, inode, path,
 931						   newext, &err, flags);
 932		if (newblock == 0)
 933			goto cleanup;
 934		ablocks[a] = newblock;
 935	}
 936
 937	/* initialize new leaf */
 938	newblock = ablocks[--a];
 939	if (unlikely(newblock == 0)) {
 940		EXT4_ERROR_INODE(inode, "newblock == 0!");
 941		err = -EIO;
 942		goto cleanup;
 943	}
 944	bh = sb_getblk(inode->i_sb, newblock);
 945	if (!bh) {
 946		err = -EIO;
 947		goto cleanup;
 948	}
 949	lock_buffer(bh);
 950
 951	err = ext4_journal_get_create_access(handle, bh);
 
 952	if (err)
 953		goto cleanup;
 954
 955	neh = ext_block_hdr(bh);
 956	neh->eh_entries = 0;
 957	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 958	neh->eh_magic = EXT4_EXT_MAGIC;
 959	neh->eh_depth = 0;
 
 960
 961	/* move remainder of path[depth] to the new leaf */
 962	if (unlikely(path[depth].p_hdr->eh_entries !=
 963		     path[depth].p_hdr->eh_max)) {
 964		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 965				 path[depth].p_hdr->eh_entries,
 966				 path[depth].p_hdr->eh_max);
 967		err = -EIO;
 968		goto cleanup;
 969	}
 970	/* start copy from next extent */
 971	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 972	ext4_ext_show_move(inode, path, newblock, depth);
 973	if (m) {
 974		struct ext4_extent *ex;
 975		ex = EXT_FIRST_EXTENT(neh);
 976		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 977		le16_add_cpu(&neh->eh_entries, m);
 978	}
 979
 
 
 
 
 980	ext4_extent_block_csum_set(inode, neh);
 981	set_buffer_uptodate(bh);
 982	unlock_buffer(bh);
 983
 984	err = ext4_handle_dirty_metadata(handle, inode, bh);
 985	if (err)
 986		goto cleanup;
 987	brelse(bh);
 988	bh = NULL;
 989
 990	/* correct old leaf */
 991	if (m) {
 992		err = ext4_ext_get_access(handle, inode, path + depth);
 993		if (err)
 994			goto cleanup;
 995		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 996		err = ext4_ext_dirty(handle, inode, path + depth);
 997		if (err)
 998			goto cleanup;
 999
1000	}
1001
1002	/* create intermediate indexes */
1003	k = depth - at - 1;
1004	if (unlikely(k < 0)) {
1005		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1006		err = -EIO;
1007		goto cleanup;
1008	}
1009	if (k)
1010		ext_debug("create %d intermediate indices\n", k);
1011	/* insert new index into current index block */
1012	/* current depth stored in i var */
1013	i = depth - 1;
1014	while (k--) {
1015		oldblock = newblock;
1016		newblock = ablocks[--a];
1017		bh = sb_getblk(inode->i_sb, newblock);
1018		if (!bh) {
1019			err = -EIO;
1020			goto cleanup;
1021		}
1022		lock_buffer(bh);
1023
1024		err = ext4_journal_get_create_access(handle, bh);
 
1025		if (err)
1026			goto cleanup;
1027
1028		neh = ext_block_hdr(bh);
1029		neh->eh_entries = cpu_to_le16(1);
1030		neh->eh_magic = EXT4_EXT_MAGIC;
1031		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1032		neh->eh_depth = cpu_to_le16(depth - i);
 
1033		fidx = EXT_FIRST_INDEX(neh);
1034		fidx->ei_block = border;
1035		ext4_idx_store_pblock(fidx, oldblock);
1036
1037		ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1038				i, newblock, le32_to_cpu(border), oldblock);
1039
1040		/* move remainder of path[i] to the new index block */
1041		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1042					EXT_LAST_INDEX(path[i].p_hdr))) {
1043			EXT4_ERROR_INODE(inode,
1044					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1045					 le32_to_cpu(path[i].p_ext->ee_block));
1046			err = -EIO;
1047			goto cleanup;
1048		}
1049		/* start copy indexes */
1050		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1051		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1052				EXT_MAX_INDEX(path[i].p_hdr));
1053		ext4_ext_show_move(inode, path, newblock, i);
1054		if (m) {
1055			memmove(++fidx, path[i].p_idx,
1056				sizeof(struct ext4_extent_idx) * m);
1057			le16_add_cpu(&neh->eh_entries, m);
1058		}
 
 
 
 
 
1059		ext4_extent_block_csum_set(inode, neh);
1060		set_buffer_uptodate(bh);
1061		unlock_buffer(bh);
1062
1063		err = ext4_handle_dirty_metadata(handle, inode, bh);
1064		if (err)
1065			goto cleanup;
1066		brelse(bh);
1067		bh = NULL;
1068
1069		/* correct old index */
1070		if (m) {
1071			err = ext4_ext_get_access(handle, inode, path + i);
1072			if (err)
1073				goto cleanup;
1074			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1075			err = ext4_ext_dirty(handle, inode, path + i);
1076			if (err)
1077				goto cleanup;
1078		}
1079
1080		i--;
1081	}
1082
1083	/* insert new index */
1084	err = ext4_ext_insert_index(handle, inode, path + at,
1085				    le32_to_cpu(border), newblock);
1086
1087cleanup:
1088	if (bh) {
1089		if (buffer_locked(bh))
1090			unlock_buffer(bh);
1091		brelse(bh);
1092	}
1093
1094	if (err) {
1095		/* free all allocated blocks in error case */
1096		for (i = 0; i < depth; i++) {
1097			if (!ablocks[i])
1098				continue;
1099			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1100					 EXT4_FREE_BLOCKS_METADATA);
1101		}
1102	}
1103	kfree(ablocks);
1104
1105	return err;
1106}
1107
1108/*
1109 * ext4_ext_grow_indepth:
1110 * implements tree growing procedure:
1111 * - allocates new block
1112 * - moves top-level data (index block or leaf) into the new block
1113 * - initializes new top-level, creating index that points to the
1114 *   just created block
1115 */
1116static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1117				 unsigned int flags,
1118				 struct ext4_extent *newext)
1119{
1120	struct ext4_extent_header *neh;
1121	struct buffer_head *bh;
1122	ext4_fsblk_t newblock;
 
1123	int err = 0;
 
1124
1125	newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1126		newext, &err, flags);
 
 
 
 
 
 
 
 
1127	if (newblock == 0)
1128		return err;
1129
1130	bh = sb_getblk(inode->i_sb, newblock);
1131	if (!bh) {
1132		err = -EIO;
1133		ext4_std_error(inode->i_sb, err);
1134		return err;
1135	}
1136	lock_buffer(bh);
1137
1138	err = ext4_journal_get_create_access(handle, bh);
 
1139	if (err) {
1140		unlock_buffer(bh);
1141		goto out;
1142	}
1143
 
1144	/* move top-level index/leaf into new block */
1145	memmove(bh->b_data, EXT4_I(inode)->i_data,
1146		sizeof(EXT4_I(inode)->i_data));
 
1147
1148	/* set size of new block */
1149	neh = ext_block_hdr(bh);
1150	/* old root could have indexes or leaves
1151	 * so calculate e_max right way */
1152	if (ext_depth(inode))
1153		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1154	else
1155		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1156	neh->eh_magic = EXT4_EXT_MAGIC;
1157	ext4_extent_block_csum_set(inode, neh);
1158	set_buffer_uptodate(bh);
 
1159	unlock_buffer(bh);
1160
1161	err = ext4_handle_dirty_metadata(handle, inode, bh);
1162	if (err)
1163		goto out;
1164
1165	/* Update top-level index: num,max,pointer */
1166	neh = ext_inode_hdr(inode);
1167	neh->eh_entries = cpu_to_le16(1);
1168	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1169	if (neh->eh_depth == 0) {
1170		/* Root extent block becomes index block */
1171		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1172		EXT_FIRST_INDEX(neh)->ei_block =
1173			EXT_FIRST_EXTENT(neh)->ee_block;
1174	}
1175	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1176		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1177		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1178		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1179
1180	neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1181	ext4_mark_inode_dirty(handle, inode);
1182out:
1183	brelse(bh);
1184
1185	return err;
1186}
1187
1188/*
1189 * ext4_ext_create_new_leaf:
1190 * finds empty index and adds new leaf.
1191 * if no free index is found, then it requests in-depth growing.
1192 */
1193static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1194				    unsigned int flags,
1195				    struct ext4_ext_path *path,
 
1196				    struct ext4_extent *newext)
1197{
 
1198	struct ext4_ext_path *curp;
1199	int depth, i, err = 0;
1200
1201repeat:
1202	i = depth = ext_depth(inode);
1203
1204	/* walk up to the tree and look for free index entry */
1205	curp = path + depth;
1206	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1207		i--;
1208		curp--;
1209	}
1210
1211	/* we use already allocated block for index block,
1212	 * so subsequent data blocks should be contiguous */
1213	if (EXT_HAS_FREE_INDEX(curp)) {
1214		/* if we found index with free entry, then use that
1215		 * entry: create all needed subtree and add new leaf */
1216		err = ext4_ext_split(handle, inode, flags, path, newext, i);
1217		if (err)
1218			goto out;
1219
1220		/* refill path */
1221		ext4_ext_drop_refs(path);
1222		path = ext4_ext_find_extent(inode,
1223				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1224				    path);
1225		if (IS_ERR(path))
1226			err = PTR_ERR(path);
1227	} else {
1228		/* tree is full, time to grow in depth */
1229		err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1230		if (err)
1231			goto out;
1232
1233		/* refill path */
1234		ext4_ext_drop_refs(path);
1235		path = ext4_ext_find_extent(inode,
1236				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1237				    path);
1238		if (IS_ERR(path)) {
1239			err = PTR_ERR(path);
1240			goto out;
1241		}
1242
1243		/*
1244		 * only first (depth 0 -> 1) produces free space;
1245		 * in all other cases we have to split the grown tree
1246		 */
1247		depth = ext_depth(inode);
1248		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1249			/* now we need to split */
1250			goto repeat;
1251		}
1252	}
1253
1254out:
1255	return err;
1256}
1257
1258/*
1259 * search the closest allocated block to the left for *logical
1260 * and returns it at @logical + it's physical address at @phys
1261 * if *logical is the smallest allocated block, the function
1262 * returns 0 at @phys
1263 * return value contains 0 (success) or error code
1264 */
1265static int ext4_ext_search_left(struct inode *inode,
1266				struct ext4_ext_path *path,
1267				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1268{
1269	struct ext4_extent_idx *ix;
1270	struct ext4_extent *ex;
1271	int depth, ee_len;
1272
1273	if (unlikely(path == NULL)) {
1274		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1275		return -EIO;
1276	}
1277	depth = path->p_depth;
1278	*phys = 0;
1279
1280	if (depth == 0 && path->p_ext == NULL)
1281		return 0;
1282
1283	/* usually extent in the path covers blocks smaller
1284	 * then *logical, but it can be that extent is the
1285	 * first one in the file */
1286
1287	ex = path[depth].p_ext;
1288	ee_len = ext4_ext_get_actual_len(ex);
1289	if (*logical < le32_to_cpu(ex->ee_block)) {
1290		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1291			EXT4_ERROR_INODE(inode,
1292					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1293					 *logical, le32_to_cpu(ex->ee_block));
1294			return -EIO;
1295		}
1296		while (--depth >= 0) {
1297			ix = path[depth].p_idx;
1298			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1299				EXT4_ERROR_INODE(inode,
1300				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1301				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1302				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1303		le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1304				  depth);
1305				return -EIO;
1306			}
1307		}
1308		return 0;
1309	}
1310
1311	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1312		EXT4_ERROR_INODE(inode,
1313				 "logical %d < ee_block %d + ee_len %d!",
1314				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1315		return -EIO;
1316	}
1317
1318	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1319	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1320	return 0;
1321}
1322
1323/*
1324 * search the closest allocated block to the right for *logical
1325 * and returns it at @logical + it's physical address at @phys
1326 * if *logical is the largest allocated block, the function
1327 * returns 0 at @phys
1328 * return value contains 0 (success) or error code
1329 */
1330static int ext4_ext_search_right(struct inode *inode,
1331				 struct ext4_ext_path *path,
1332				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1333				 struct ext4_extent **ret_ex)
1334{
1335	struct buffer_head *bh = NULL;
1336	struct ext4_extent_header *eh;
1337	struct ext4_extent_idx *ix;
1338	struct ext4_extent *ex;
1339	ext4_fsblk_t block;
1340	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1341	int ee_len;
1342
1343	if (unlikely(path == NULL)) {
1344		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1345		return -EIO;
1346	}
1347	depth = path->p_depth;
1348	*phys = 0;
1349
1350	if (depth == 0 && path->p_ext == NULL)
1351		return 0;
1352
1353	/* usually extent in the path covers blocks smaller
1354	 * then *logical, but it can be that extent is the
1355	 * first one in the file */
1356
1357	ex = path[depth].p_ext;
1358	ee_len = ext4_ext_get_actual_len(ex);
1359	if (*logical < le32_to_cpu(ex->ee_block)) {
1360		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1361			EXT4_ERROR_INODE(inode,
1362					 "first_extent(path[%d].p_hdr) != ex",
1363					 depth);
1364			return -EIO;
1365		}
1366		while (--depth >= 0) {
1367			ix = path[depth].p_idx;
1368			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1369				EXT4_ERROR_INODE(inode,
1370						 "ix != EXT_FIRST_INDEX *logical %d!",
1371						 *logical);
1372				return -EIO;
1373			}
1374		}
1375		goto found_extent;
1376	}
1377
1378	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1379		EXT4_ERROR_INODE(inode,
1380				 "logical %d < ee_block %d + ee_len %d!",
1381				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1382		return -EIO;
1383	}
1384
1385	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1386		/* next allocated block in this leaf */
1387		ex++;
1388		goto found_extent;
1389	}
1390
1391	/* go up and search for index to the right */
1392	while (--depth >= 0) {
1393		ix = path[depth].p_idx;
1394		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1395			goto got_index;
1396	}
1397
1398	/* we've gone up to the root and found no index to the right */
1399	return 0;
1400
1401got_index:
1402	/* we've found index to the right, let's
1403	 * follow it and find the closest allocated
1404	 * block to the right */
1405	ix++;
1406	block = ext4_idx_pblock(ix);
1407	while (++depth < path->p_depth) {
1408		bh = sb_bread(inode->i_sb, block);
1409		if (bh == NULL)
1410			return -EIO;
 
1411		eh = ext_block_hdr(bh);
1412		/* subtract from p_depth to get proper eh_depth */
1413		if (ext4_ext_check_block(inode, eh,
1414					 path->p_depth - depth, bh)) {
1415			put_bh(bh);
1416			return -EIO;
1417		}
1418		ix = EXT_FIRST_INDEX(eh);
1419		block = ext4_idx_pblock(ix);
1420		put_bh(bh);
1421	}
1422
1423	bh = sb_bread(inode->i_sb, block);
1424	if (bh == NULL)
1425		return -EIO;
1426	eh = ext_block_hdr(bh);
1427	if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
1428		put_bh(bh);
1429		return -EIO;
1430	}
1431	ex = EXT_FIRST_EXTENT(eh);
1432found_extent:
1433	*logical = le32_to_cpu(ex->ee_block);
1434	*phys = ext4_ext_pblock(ex);
1435	*ret_ex = ex;
 
1436	if (bh)
1437		put_bh(bh);
1438	return 0;
1439}
1440
1441/*
1442 * ext4_ext_next_allocated_block:
1443 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1444 * NOTE: it considers block number from index entry as
1445 * allocated block. Thus, index entries have to be consistent
1446 * with leaves.
1447 */
1448static ext4_lblk_t
1449ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1450{
1451	int depth;
1452
1453	BUG_ON(path == NULL);
1454	depth = path->p_depth;
1455
1456	if (depth == 0 && path->p_ext == NULL)
1457		return EXT_MAX_BLOCKS;
1458
1459	while (depth >= 0) {
 
 
1460		if (depth == path->p_depth) {
1461			/* leaf */
1462			if (path[depth].p_ext &&
1463				path[depth].p_ext !=
1464					EXT_LAST_EXTENT(path[depth].p_hdr))
1465			  return le32_to_cpu(path[depth].p_ext[1].ee_block);
1466		} else {
1467			/* index */
1468			if (path[depth].p_idx !=
1469					EXT_LAST_INDEX(path[depth].p_hdr))
1470			  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1471		}
1472		depth--;
1473	}
1474
1475	return EXT_MAX_BLOCKS;
1476}
1477
1478/*
1479 * ext4_ext_next_leaf_block:
1480 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1481 */
1482static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1483{
1484	int depth;
1485
1486	BUG_ON(path == NULL);
1487	depth = path->p_depth;
1488
1489	/* zero-tree has no leaf blocks at all */
1490	if (depth == 0)
1491		return EXT_MAX_BLOCKS;
1492
1493	/* go to index block */
1494	depth--;
1495
1496	while (depth >= 0) {
1497		if (path[depth].p_idx !=
1498				EXT_LAST_INDEX(path[depth].p_hdr))
1499			return (ext4_lblk_t)
1500				le32_to_cpu(path[depth].p_idx[1].ei_block);
1501		depth--;
1502	}
1503
1504	return EXT_MAX_BLOCKS;
1505}
1506
1507/*
1508 * ext4_ext_correct_indexes:
1509 * if leaf gets modified and modified extent is first in the leaf,
1510 * then we have to correct all indexes above.
1511 * TODO: do we need to correct tree in all cases?
1512 */
1513static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1514				struct ext4_ext_path *path)
1515{
1516	struct ext4_extent_header *eh;
1517	int depth = ext_depth(inode);
1518	struct ext4_extent *ex;
1519	__le32 border;
1520	int k, err = 0;
1521
1522	eh = path[depth].p_hdr;
1523	ex = path[depth].p_ext;
1524
1525	if (unlikely(ex == NULL || eh == NULL)) {
1526		EXT4_ERROR_INODE(inode,
1527				 "ex %p == NULL or eh %p == NULL", ex, eh);
1528		return -EIO;
1529	}
1530
1531	if (depth == 0) {
1532		/* there is no tree at all */
1533		return 0;
1534	}
1535
1536	if (ex != EXT_FIRST_EXTENT(eh)) {
1537		/* we correct tree if first leaf got modified only */
1538		return 0;
1539	}
1540
1541	/*
1542	 * TODO: we need correction if border is smaller than current one
1543	 */
1544	k = depth - 1;
1545	border = path[depth].p_ext->ee_block;
1546	err = ext4_ext_get_access(handle, inode, path + k);
1547	if (err)
1548		return err;
1549	path[k].p_idx->ei_block = border;
1550	err = ext4_ext_dirty(handle, inode, path + k);
1551	if (err)
1552		return err;
1553
1554	while (k--) {
1555		/* change all left-side indexes */
1556		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1557			break;
1558		err = ext4_ext_get_access(handle, inode, path + k);
1559		if (err)
1560			break;
1561		path[k].p_idx->ei_block = border;
1562		err = ext4_ext_dirty(handle, inode, path + k);
1563		if (err)
1564			break;
1565	}
1566
1567	return err;
1568}
1569
1570int
1571ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1572				struct ext4_extent *ex2)
1573{
1574	unsigned short ext1_ee_len, ext2_ee_len, max_len;
1575
1576	/*
1577	 * Make sure that either both extents are uninitialized, or
1578	 * both are _not_.
1579	 */
1580	if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1581		return 0;
1582
1583	if (ext4_ext_is_uninitialized(ex1))
1584		max_len = EXT_UNINIT_MAX_LEN;
1585	else
1586		max_len = EXT_INIT_MAX_LEN;
1587
1588	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1589	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1590
1591	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1592			le32_to_cpu(ex2->ee_block))
1593		return 0;
1594
1595	/*
1596	 * To allow future support for preallocated extents to be added
1597	 * as an RO_COMPAT feature, refuse to merge to extents if
1598	 * this can result in the top bit of ee_len being set.
1599	 */
1600	if (ext1_ee_len + ext2_ee_len > max_len)
1601		return 0;
1602#ifdef AGGRESSIVE_TEST
1603	if (ext1_ee_len >= 4)
1604		return 0;
1605#endif
1606
1607	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1608		return 1;
1609	return 0;
1610}
1611
1612/*
1613 * This function tries to merge the "ex" extent to the next extent in the tree.
1614 * It always tries to merge towards right. If you want to merge towards
1615 * left, pass "ex - 1" as argument instead of "ex".
1616 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1617 * 1 if they got merged.
1618 */
1619static int ext4_ext_try_to_merge_right(struct inode *inode,
1620				 struct ext4_ext_path *path,
1621				 struct ext4_extent *ex)
1622{
1623	struct ext4_extent_header *eh;
1624	unsigned int depth, len;
1625	int merge_done = 0;
1626	int uninitialized = 0;
1627
1628	depth = ext_depth(inode);
1629	BUG_ON(path[depth].p_hdr == NULL);
1630	eh = path[depth].p_hdr;
1631
1632	while (ex < EXT_LAST_EXTENT(eh)) {
1633		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1634			break;
1635		/* merge with next extent! */
1636		if (ext4_ext_is_uninitialized(ex))
1637			uninitialized = 1;
1638		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1639				+ ext4_ext_get_actual_len(ex + 1));
1640		if (uninitialized)
1641			ext4_ext_mark_uninitialized(ex);
1642
1643		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1644			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1645				* sizeof(struct ext4_extent);
1646			memmove(ex + 1, ex + 2, len);
1647		}
1648		le16_add_cpu(&eh->eh_entries, -1);
1649		merge_done = 1;
1650		WARN_ON(eh->eh_entries == 0);
1651		if (!eh->eh_entries)
1652			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1653	}
1654
1655	return merge_done;
1656}
1657
1658/*
1659 * This function tries to merge the @ex extent to neighbours in the tree.
1660 * return 1 if merge left else 0.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1661 */
1662static int ext4_ext_try_to_merge(struct inode *inode,
 
1663				  struct ext4_ext_path *path,
1664				  struct ext4_extent *ex) {
 
1665	struct ext4_extent_header *eh;
1666	unsigned int depth;
1667	int merge_done = 0;
1668	int ret = 0;
1669
1670	depth = ext_depth(inode);
1671	BUG_ON(path[depth].p_hdr == NULL);
1672	eh = path[depth].p_hdr;
1673
1674	if (ex > EXT_FIRST_EXTENT(eh))
1675		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1676
1677	if (!merge_done)
1678		ret = ext4_ext_try_to_merge_right(inode, path, ex);
1679
1680	return ret;
1681}
1682
1683/*
1684 * check if a portion of the "newext" extent overlaps with an
1685 * existing extent.
1686 *
1687 * If there is an overlap discovered, it updates the length of the newext
1688 * such that there will be no overlap, and then returns 1.
1689 * If there is no overlap found, it returns 0.
1690 */
1691static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1692					   struct inode *inode,
1693					   struct ext4_extent *newext,
1694					   struct ext4_ext_path *path)
1695{
1696	ext4_lblk_t b1, b2;
1697	unsigned int depth, len1;
1698	unsigned int ret = 0;
1699
1700	b1 = le32_to_cpu(newext->ee_block);
1701	len1 = ext4_ext_get_actual_len(newext);
1702	depth = ext_depth(inode);
1703	if (!path[depth].p_ext)
1704		goto out;
1705	b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1706	b2 &= ~(sbi->s_cluster_ratio - 1);
1707
1708	/*
1709	 * get the next allocated block if the extent in the path
1710	 * is before the requested block(s)
1711	 */
1712	if (b2 < b1) {
1713		b2 = ext4_ext_next_allocated_block(path);
1714		if (b2 == EXT_MAX_BLOCKS)
1715			goto out;
1716		b2 &= ~(sbi->s_cluster_ratio - 1);
1717	}
1718
1719	/* check for wrap through zero on extent logical start block*/
1720	if (b1 + len1 < b1) {
1721		len1 = EXT_MAX_BLOCKS - b1;
1722		newext->ee_len = cpu_to_le16(len1);
1723		ret = 1;
1724	}
1725
1726	/* check for overlap */
1727	if (b1 + len1 > b2) {
1728		newext->ee_len = cpu_to_le16(b2 - b1);
1729		ret = 1;
1730	}
1731out:
1732	return ret;
1733}
1734
1735/*
1736 * ext4_ext_insert_extent:
1737 * tries to merge requsted extent into the existing extent or
1738 * inserts requested extent as new one into the tree,
1739 * creating new leaf in the no-space case.
1740 */
1741int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1742				struct ext4_ext_path *path,
1743				struct ext4_extent *newext, int flag)
1744{
 
1745	struct ext4_extent_header *eh;
1746	struct ext4_extent *ex, *fex;
1747	struct ext4_extent *nearex; /* nearest extent */
1748	struct ext4_ext_path *npath = NULL;
1749	int depth, len, err;
1750	ext4_lblk_t next;
1751	unsigned uninitialized = 0;
1752	int flags = 0;
1753
 
 
1754	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1755		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1756		return -EIO;
1757	}
1758	depth = ext_depth(inode);
1759	ex = path[depth].p_ext;
 
1760	if (unlikely(path[depth].p_hdr == NULL)) {
1761		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1762		return -EIO;
1763	}
1764
1765	/* try to insert block into found extent and return */
1766	if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1767		&& ext4_can_extents_be_merged(inode, ex, newext)) {
1768		ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1769			  ext4_ext_is_uninitialized(newext),
1770			  ext4_ext_get_actual_len(newext),
1771			  le32_to_cpu(ex->ee_block),
1772			  ext4_ext_is_uninitialized(ex),
1773			  ext4_ext_get_actual_len(ex),
1774			  ext4_ext_pblock(ex));
1775		err = ext4_ext_get_access(handle, inode, path + depth);
1776		if (err)
1777			return err;
1778
1779		/*
1780		 * ext4_can_extents_be_merged should have checked that either
1781		 * both extents are uninitialized, or both aren't. Thus we
1782		 * need to check only one of them here.
 
 
1783		 */
1784		if (ext4_ext_is_uninitialized(ex))
1785			uninitialized = 1;
1786		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1787					+ ext4_ext_get_actual_len(newext));
1788		if (uninitialized)
1789			ext4_ext_mark_uninitialized(ex);
1790		eh = path[depth].p_hdr;
1791		nearex = ex;
1792		goto merge;
1793	}
1794
1795	depth = ext_depth(inode);
1796	eh = path[depth].p_hdr;
1797	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1798		goto has_space;
1799
1800	/* probably next leaf has space for us? */
1801	fex = EXT_LAST_EXTENT(eh);
1802	next = EXT_MAX_BLOCKS;
1803	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1804		next = ext4_ext_next_leaf_block(path);
1805	if (next != EXT_MAX_BLOCKS) {
1806		ext_debug("next leaf block - %u\n", next);
1807		BUG_ON(npath != NULL);
1808		npath = ext4_ext_find_extent(inode, next, NULL);
1809		if (IS_ERR(npath))
1810			return PTR_ERR(npath);
1811		BUG_ON(npath->p_depth != path->p_depth);
1812		eh = npath[depth].p_hdr;
1813		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1814			ext_debug("next leaf isn't full(%d)\n",
1815				  le16_to_cpu(eh->eh_entries));
1816			path = npath;
1817			goto has_space;
1818		}
1819		ext_debug("next leaf has no free space(%d,%d)\n",
1820			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1821	}
1822
1823	/*
1824	 * There is no free space in the found leaf.
1825	 * We're gonna add a new leaf in the tree.
1826	 */
1827	if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1828		flags = EXT4_MB_USE_ROOT_BLOCKS;
1829	err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
 
1830	if (err)
1831		goto cleanup;
1832	depth = ext_depth(inode);
1833	eh = path[depth].p_hdr;
1834
1835has_space:
1836	nearex = path[depth].p_ext;
1837
1838	err = ext4_ext_get_access(handle, inode, path + depth);
1839	if (err)
1840		goto cleanup;
1841
1842	if (!nearex) {
1843		/* there is no extent in this leaf, create first one */
1844		ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1845				le32_to_cpu(newext->ee_block),
1846				ext4_ext_pblock(newext),
1847				ext4_ext_is_uninitialized(newext),
1848				ext4_ext_get_actual_len(newext));
1849		nearex = EXT_FIRST_EXTENT(eh);
1850	} else {
1851		if (le32_to_cpu(newext->ee_block)
1852			   > le32_to_cpu(nearex->ee_block)) {
1853			/* Insert after */
1854			ext_debug("insert %u:%llu:[%d]%d before: "
1855					"nearest %p\n",
1856					le32_to_cpu(newext->ee_block),
1857					ext4_ext_pblock(newext),
1858					ext4_ext_is_uninitialized(newext),
1859					ext4_ext_get_actual_len(newext),
1860					nearex);
1861			nearex++;
1862		} else {
1863			/* Insert before */
1864			BUG_ON(newext->ee_block == nearex->ee_block);
1865			ext_debug("insert %u:%llu:[%d]%d after: "
1866					"nearest %p\n",
1867					le32_to_cpu(newext->ee_block),
1868					ext4_ext_pblock(newext),
1869					ext4_ext_is_uninitialized(newext),
1870					ext4_ext_get_actual_len(newext),
1871					nearex);
1872		}
1873		len = EXT_LAST_EXTENT(eh) - nearex + 1;
1874		if (len > 0) {
1875			ext_debug("insert %u:%llu:[%d]%d: "
1876					"move %d extents from 0x%p to 0x%p\n",
1877					le32_to_cpu(newext->ee_block),
1878					ext4_ext_pblock(newext),
1879					ext4_ext_is_uninitialized(newext),
1880					ext4_ext_get_actual_len(newext),
1881					len, nearex, nearex + 1);
1882			memmove(nearex + 1, nearex,
1883				len * sizeof(struct ext4_extent));
1884		}
1885	}
1886
1887	le16_add_cpu(&eh->eh_entries, 1);
1888	path[depth].p_ext = nearex;
1889	nearex->ee_block = newext->ee_block;
1890	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1891	nearex->ee_len = newext->ee_len;
1892
1893merge:
1894	/* try to merge extents to the right */
1895	if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1896		ext4_ext_try_to_merge(inode, path, nearex);
1897
1898	/* try to merge extents to the left */
1899
1900	/* time to correct all indexes above */
1901	err = ext4_ext_correct_indexes(handle, inode, path);
1902	if (err)
1903		goto cleanup;
1904
1905	err = ext4_ext_dirty(handle, inode, path + depth);
1906
1907cleanup:
1908	if (npath) {
1909		ext4_ext_drop_refs(npath);
1910		kfree(npath);
1911	}
1912	ext4_ext_invalidate_cache(inode);
1913	return err;
1914}
1915
1916static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1917			       ext4_lblk_t num, ext_prepare_callback func,
1918			       void *cbdata)
1919{
1920	struct ext4_ext_path *path = NULL;
1921	struct ext4_ext_cache cbex;
1922	struct ext4_extent *ex;
1923	ext4_lblk_t next, start = 0, end = 0;
1924	ext4_lblk_t last = block + num;
1925	int depth, exists, err = 0;
1926
1927	BUG_ON(func == NULL);
1928	BUG_ON(inode == NULL);
1929
1930	while (block < last && block != EXT_MAX_BLOCKS) {
1931		num = last - block;
1932		/* find extent for this block */
1933		down_read(&EXT4_I(inode)->i_data_sem);
1934		path = ext4_ext_find_extent(inode, block, path);
1935		up_read(&EXT4_I(inode)->i_data_sem);
1936		if (IS_ERR(path)) {
1937			err = PTR_ERR(path);
1938			path = NULL;
1939			break;
1940		}
1941
1942		depth = ext_depth(inode);
1943		if (unlikely(path[depth].p_hdr == NULL)) {
1944			EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1945			err = -EIO;
1946			break;
1947		}
1948		ex = path[depth].p_ext;
1949		next = ext4_ext_next_allocated_block(path);
1950
1951		exists = 0;
1952		if (!ex) {
1953			/* there is no extent yet, so try to allocate
1954			 * all requested space */
1955			start = block;
1956			end = block + num;
1957		} else if (le32_to_cpu(ex->ee_block) > block) {
1958			/* need to allocate space before found extent */
1959			start = block;
1960			end = le32_to_cpu(ex->ee_block);
1961			if (block + num < end)
1962				end = block + num;
1963		} else if (block >= le32_to_cpu(ex->ee_block)
1964					+ ext4_ext_get_actual_len(ex)) {
1965			/* need to allocate space after found extent */
1966			start = block;
1967			end = block + num;
1968			if (end >= next)
1969				end = next;
1970		} else if (block >= le32_to_cpu(ex->ee_block)) {
1971			/*
1972			 * some part of requested space is covered
1973			 * by found extent
1974			 */
1975			start = block;
1976			end = le32_to_cpu(ex->ee_block)
1977				+ ext4_ext_get_actual_len(ex);
1978			if (block + num < end)
1979				end = block + num;
1980			exists = 1;
1981		} else {
1982			BUG();
1983		}
1984		BUG_ON(end <= start);
1985
1986		if (!exists) {
1987			cbex.ec_block = start;
1988			cbex.ec_len = end - start;
1989			cbex.ec_start = 0;
1990		} else {
1991			cbex.ec_block = le32_to_cpu(ex->ee_block);
1992			cbex.ec_len = ext4_ext_get_actual_len(ex);
1993			cbex.ec_start = ext4_ext_pblock(ex);
1994		}
1995
1996		if (unlikely(cbex.ec_len == 0)) {
1997			EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1998			err = -EIO;
1999			break;
2000		}
2001		err = func(inode, next, &cbex, ex, cbdata);
2002		ext4_ext_drop_refs(path);
2003
2004		if (err < 0)
2005			break;
2006
2007		if (err == EXT_REPEAT)
2008			continue;
2009		else if (err == EXT_BREAK) {
2010			err = 0;
2011			break;
2012		}
2013
2014		if (ext_depth(inode) != depth) {
2015			/* depth was changed. we have to realloc path */
2016			kfree(path);
2017			path = NULL;
2018		}
2019
2020		block = cbex.ec_block + cbex.ec_len;
2021	}
2022
2023	if (path) {
2024		ext4_ext_drop_refs(path);
2025		kfree(path);
2026	}
2027
2028	return err;
2029}
2030
2031static void
2032ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
2033			__u32 len, ext4_fsblk_t start)
2034{
2035	struct ext4_ext_cache *cex;
2036	BUG_ON(len == 0);
2037	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2038	trace_ext4_ext_put_in_cache(inode, block, len, start);
2039	cex = &EXT4_I(inode)->i_cached_extent;
2040	cex->ec_block = block;
2041	cex->ec_len = len;
2042	cex->ec_start = start;
2043	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2044}
2045
2046/*
2047 * ext4_ext_put_gap_in_cache:
2048 * calculate boundaries of the gap that the requested block fits into
2049 * and cache this gap
2050 */
2051static void
2052ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2053				ext4_lblk_t block)
 
 
 
 
 
 
 
 
2054{
2055	int depth = ext_depth(inode);
2056	unsigned long len;
2057	ext4_lblk_t lblock;
2058	struct ext4_extent *ex;
 
2059
2060	ex = path[depth].p_ext;
2061	if (ex == NULL) {
2062		/* there is no extent yet, so gap is [0;-] */
2063		lblock = 0;
2064		len = EXT_MAX_BLOCKS;
2065		ext_debug("cache gap(whole file):");
2066	} else if (block < le32_to_cpu(ex->ee_block)) {
2067		lblock = block;
2068		len = le32_to_cpu(ex->ee_block) - block;
2069		ext_debug("cache gap(before): %u [%u:%u]",
2070				block,
2071				le32_to_cpu(ex->ee_block),
2072				 ext4_ext_get_actual_len(ex));
2073	} else if (block >= le32_to_cpu(ex->ee_block)
2074			+ ext4_ext_get_actual_len(ex)) {
2075		ext4_lblk_t next;
2076		lblock = le32_to_cpu(ex->ee_block)
2077			+ ext4_ext_get_actual_len(ex);
2078
 
2079		next = ext4_ext_next_allocated_block(path);
2080		ext_debug("cache gap(after): [%u:%u] %u",
2081				le32_to_cpu(ex->ee_block),
2082				ext4_ext_get_actual_len(ex),
2083				block);
2084		BUG_ON(next == lblock);
2085		len = next - lblock;
2086	} else {
2087		lblock = len = 0;
2088		BUG();
2089	}
2090
2091	ext_debug(" -> %u:%lu\n", lblock, len);
2092	ext4_ext_put_in_cache(inode, lblock, len, 0);
2093}
2094
2095/*
2096 * ext4_ext_check_cache()
2097 * Checks to see if the given block is in the cache.
2098 * If it is, the cached extent is stored in the given
2099 * cache extent pointer.  If the cached extent is a hole,
2100 * this routine should be used instead of
2101 * ext4_ext_in_cache if the calling function needs to
2102 * know the size of the hole.
2103 *
2104 * @inode: The files inode
2105 * @block: The block to look for in the cache
2106 * @ex:    Pointer where the cached extent will be stored
2107 *         if it contains block
2108 *
2109 * Return 0 if cache is invalid; 1 if the cache is valid
2110 */
2111static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2112	struct ext4_ext_cache *ex){
2113	struct ext4_ext_cache *cex;
2114	struct ext4_sb_info *sbi;
2115	int ret = 0;
2116
2117	/*
2118	 * We borrow i_block_reservation_lock to protect i_cached_extent
2119	 */
2120	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2121	cex = &EXT4_I(inode)->i_cached_extent;
2122	sbi = EXT4_SB(inode->i_sb);
2123
2124	/* has cache valid data? */
2125	if (cex->ec_len == 0)
2126		goto errout;
2127
2128	if (in_range(block, cex->ec_block, cex->ec_len)) {
2129		memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2130		ext_debug("%u cached by %u:%u:%llu\n",
2131				block,
2132				cex->ec_block, cex->ec_len, cex->ec_start);
2133		ret = 1;
2134	}
2135errout:
2136	trace_ext4_ext_in_cache(inode, block, ret);
2137	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2138	return ret;
2139}
2140
2141/*
2142 * ext4_ext_in_cache()
2143 * Checks to see if the given block is in the cache.
2144 * If it is, the cached extent is stored in the given
2145 * extent pointer.
2146 *
2147 * @inode: The files inode
2148 * @block: The block to look for in the cache
2149 * @ex:    Pointer where the cached extent will be stored
2150 *         if it contains block
2151 *
2152 * Return 0 if cache is invalid; 1 if the cache is valid
2153 */
2154static int
2155ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2156			struct ext4_extent *ex)
2157{
2158	struct ext4_ext_cache cex;
2159	int ret = 0;
2160
2161	if (ext4_ext_check_cache(inode, block, &cex)) {
2162		ex->ee_block = cpu_to_le32(cex.ec_block);
2163		ext4_ext_store_pblock(ex, cex.ec_start);
2164		ex->ee_len = cpu_to_le16(cex.ec_len);
2165		ret = 1;
2166	}
2167
2168	return ret;
 
 
 
 
 
 
 
 
 
 
2169}
2170
2171
2172/*
2173 * ext4_ext_rm_idx:
2174 * removes index from the index block.
2175 */
2176static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2177			struct ext4_ext_path *path)
2178{
2179	int err;
2180	ext4_fsblk_t leaf;
2181
2182	/* free index block */
2183	path--;
 
2184	leaf = ext4_idx_pblock(path->p_idx);
2185	if (unlikely(path->p_hdr->eh_entries == 0)) {
2186		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2187		return -EIO;
2188	}
2189	err = ext4_ext_get_access(handle, inode, path);
2190	if (err)
2191		return err;
2192
2193	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2194		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2195		len *= sizeof(struct ext4_extent_idx);
2196		memmove(path->p_idx, path->p_idx + 1, len);
2197	}
2198
2199	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2200	err = ext4_ext_dirty(handle, inode, path);
2201	if (err)
2202		return err;
2203	ext_debug("index is empty, remove it, free block %llu\n", leaf);
2204	trace_ext4_ext_rm_idx(inode, leaf);
2205
2206	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2207			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
 
 
 
 
 
 
 
 
 
 
 
 
 
2208	return err;
2209}
2210
2211/*
2212 * ext4_ext_calc_credits_for_single_extent:
2213 * This routine returns max. credits that needed to insert an extent
2214 * to the extent tree.
2215 * When pass the actual path, the caller should calculate credits
2216 * under i_data_sem.
2217 */
2218int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2219						struct ext4_ext_path *path)
2220{
2221	if (path) {
2222		int depth = ext_depth(inode);
2223		int ret = 0;
2224
2225		/* probably there is space in leaf? */
2226		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2227				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2228
2229			/*
2230			 *  There are some space in the leaf tree, no
2231			 *  need to account for leaf block credit
2232			 *
2233			 *  bitmaps and block group descriptor blocks
2234			 *  and other metadata blocks still need to be
2235			 *  accounted.
2236			 */
2237			/* 1 bitmap, 1 block group descriptor */
2238			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2239			return ret;
2240		}
2241	}
2242
2243	return ext4_chunk_trans_blocks(inode, nrblocks);
2244}
2245
2246/*
2247 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2248 *
2249 * if nrblocks are fit in a single extent (chunk flag is 1), then
2250 * in the worse case, each tree level index/leaf need to be changed
2251 * if the tree split due to insert a new extent, then the old tree
2252 * index/leaf need to be updated too
2253 *
2254 * If the nrblocks are discontiguous, they could cause
2255 * the whole tree split more than once, but this is really rare.
2256 */
2257int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2258{
2259	int index;
2260	int depth = ext_depth(inode);
 
 
 
 
 
 
2261
2262	if (chunk)
2263		index = depth * 2;
2264	else
2265		index = depth * 3;
2266
2267	return index;
2268}
2269
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2270static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2271			      struct ext4_extent *ex,
2272			      ext4_fsblk_t *partial_cluster,
2273			      ext4_lblk_t from, ext4_lblk_t to)
2274{
2275	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2276	unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2277	ext4_fsblk_t pblk;
2278	int flags = EXT4_FREE_BLOCKS_FORGET;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2279
2280	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2281		flags |= EXT4_FREE_BLOCKS_METADATA;
2282	/*
2283	 * For bigalloc file systems, we never free a partial cluster
2284	 * at the beginning of the extent.  Instead, we make a note
2285	 * that we tried freeing the cluster, and check to see if we
2286	 * need to free it on a subsequent call to ext4_remove_blocks,
2287	 * or at the end of the ext4_truncate() operation.
2288	 */
2289	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2290
2291	trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2292	/*
2293	 * If we have a partial cluster, and it's different from the
2294	 * cluster of the last block, we need to explicitly free the
2295	 * partial cluster here.
 
2296	 */
2297	pblk = ext4_ext_pblock(ex) + ee_len - 1;
2298	if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
 
 
 
 
 
 
2299		ext4_free_blocks(handle, inode, NULL,
2300				 EXT4_C2B(sbi, *partial_cluster),
2301				 sbi->s_cluster_ratio, flags);
2302		*partial_cluster = 0;
 
 
 
2303	}
2304
2305#ifdef EXTENTS_STATS
2306	{
2307		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2308		spin_lock(&sbi->s_ext_stats_lock);
2309		sbi->s_ext_blocks += ee_len;
2310		sbi->s_ext_extents++;
2311		if (ee_len < sbi->s_ext_min)
2312			sbi->s_ext_min = ee_len;
2313		if (ee_len > sbi->s_ext_max)
2314			sbi->s_ext_max = ee_len;
2315		if (ext_depth(inode) > sbi->s_depth_max)
2316			sbi->s_depth_max = ext_depth(inode);
2317		spin_unlock(&sbi->s_ext_stats_lock);
2318	}
2319#endif
2320	if (from >= le32_to_cpu(ex->ee_block)
2321	    && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2322		/* tail removal */
2323		ext4_lblk_t num;
2324
2325		num = le32_to_cpu(ex->ee_block) + ee_len - from;
2326		pblk = ext4_ext_pblock(ex) + ee_len - num;
2327		ext_debug("free last %u blocks starting %llu\n", num, pblk);
2328		ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2329		/*
2330		 * If the block range to be freed didn't start at the
2331		 * beginning of a cluster, and we removed the entire
2332		 * extent, save the partial cluster here, since we
2333		 * might need to delete if we determine that the
2334		 * truncate operation has removed all of the blocks in
2335		 * the cluster.
2336		 */
2337		if (pblk & (sbi->s_cluster_ratio - 1) &&
2338		    (ee_len == num))
2339			*partial_cluster = EXT4_B2C(sbi, pblk);
2340		else
2341			*partial_cluster = 0;
2342	} else if (from == le32_to_cpu(ex->ee_block)
2343		   && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2344		/* head removal */
2345		ext4_lblk_t num;
2346		ext4_fsblk_t start;
2347
2348		num = to - from;
2349		start = ext4_ext_pblock(ex);
2350
2351		ext_debug("free first %u blocks starting %llu\n", num, start);
2352		ext4_free_blocks(handle, inode, NULL, start, num, flags);
 
 
 
2353
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2354	} else {
2355		printk(KERN_INFO "strange request: removal(2) "
2356				"%u-%u from %u:%u\n",
2357				from, to, le32_to_cpu(ex->ee_block), ee_len);
2358	}
 
2359	return 0;
2360}
2361
2362
2363/*
2364 * ext4_ext_rm_leaf() Removes the extents associated with the
2365 * blocks appearing between "start" and "end", and splits the extents
2366 * if "start" and "end" appear in the same extent
2367 *
2368 * @handle: The journal handle
2369 * @inode:  The files inode
2370 * @path:   The path to the leaf
 
 
 
 
2371 * @start:  The first block to remove
2372 * @end:   The last block to remove
2373 */
2374static int
2375ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2376		 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
 
2377		 ext4_lblk_t start, ext4_lblk_t end)
2378{
2379	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2380	int err = 0, correct_index = 0;
2381	int depth = ext_depth(inode), credits;
2382	struct ext4_extent_header *eh;
2383	ext4_lblk_t a, b;
2384	unsigned num;
2385	ext4_lblk_t ex_ee_block;
2386	unsigned short ex_ee_len;
2387	unsigned uninitialized = 0;
2388	struct ext4_extent *ex;
 
2389
2390	/* the header must be checked already in ext4_ext_remove_space() */
2391	ext_debug("truncate since %u in leaf to %u\n", start, end);
2392	if (!path[depth].p_hdr)
2393		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2394	eh = path[depth].p_hdr;
2395	if (unlikely(path[depth].p_hdr == NULL)) {
2396		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2397		return -EIO;
2398	}
2399	/* find where to start removing */
2400	ex = EXT_LAST_EXTENT(eh);
 
 
2401
2402	ex_ee_block = le32_to_cpu(ex->ee_block);
2403	ex_ee_len = ext4_ext_get_actual_len(ex);
2404
2405	trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2406
2407	while (ex >= EXT_FIRST_EXTENT(eh) &&
2408			ex_ee_block + ex_ee_len > start) {
2409
2410		if (ext4_ext_is_uninitialized(ex))
2411			uninitialized = 1;
2412		else
2413			uninitialized = 0;
2414
2415		ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2416			 uninitialized, ex_ee_len);
2417		path[depth].p_ext = ex;
2418
2419		a = ex_ee_block > start ? ex_ee_block : start;
2420		b = ex_ee_block+ex_ee_len - 1 < end ?
2421			ex_ee_block+ex_ee_len - 1 : end;
2422
2423		ext_debug("  border %u:%u\n", a, b);
2424
2425		/* If this extent is beyond the end of the hole, skip it */
2426		if (end < ex_ee_block) {
 
 
 
 
 
 
 
 
 
 
 
 
2427			ex--;
2428			ex_ee_block = le32_to_cpu(ex->ee_block);
2429			ex_ee_len = ext4_ext_get_actual_len(ex);
2430			continue;
2431		} else if (b != ex_ee_block + ex_ee_len - 1) {
2432			EXT4_ERROR_INODE(inode,
2433					 "can not handle truncate %u:%u "
2434					 "on extent %u:%u",
2435					 start, end, ex_ee_block,
2436					 ex_ee_block + ex_ee_len - 1);
2437			err = -EIO;
2438			goto out;
2439		} else if (a != ex_ee_block) {
2440			/* remove tail of the extent */
2441			num = a - ex_ee_block;
2442		} else {
2443			/* remove whole extent: excellent! */
2444			num = 0;
2445		}
2446		/*
2447		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2448		 * descriptor) for each block group; assume two block
2449		 * groups plus ex_ee_len/blocks_per_block_group for
2450		 * the worst case
2451		 */
2452		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2453		if (ex == EXT_FIRST_EXTENT(eh)) {
2454			correct_index = 1;
2455			credits += (ext_depth(inode)) + 1;
2456		}
2457		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2458
2459		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2460		if (err)
 
 
 
 
 
 
 
 
 
 
 
 
2461			goto out;
 
2462
2463		err = ext4_ext_get_access(handle, inode, path + depth);
2464		if (err)
2465			goto out;
2466
2467		err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2468					 a, b);
2469		if (err)
2470			goto out;
2471
2472		if (num == 0)
2473			/* this extent is removed; mark slot entirely unused */
2474			ext4_ext_store_pblock(ex, 0);
2475
2476		ex->ee_len = cpu_to_le16(num);
2477		/*
2478		 * Do not mark uninitialized if all the blocks in the
2479		 * extent have been removed.
2480		 */
2481		if (uninitialized && num)
2482			ext4_ext_mark_uninitialized(ex);
2483		/*
2484		 * If the extent was completely released,
2485		 * we need to remove it from the leaf
2486		 */
2487		if (num == 0) {
2488			if (end != EXT_MAX_BLOCKS - 1) {
2489				/*
2490				 * For hole punching, we need to scoot all the
2491				 * extents up when an extent is removed so that
2492				 * we dont have blank extents in the middle
2493				 */
2494				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2495					sizeof(struct ext4_extent));
2496
2497				/* Now get rid of the one at the end */
2498				memset(EXT_LAST_EXTENT(eh), 0,
2499					sizeof(struct ext4_extent));
2500			}
2501			le16_add_cpu(&eh->eh_entries, -1);
2502		} else
2503			*partial_cluster = 0;
2504
2505		err = ext4_ext_dirty(handle, inode, path + depth);
2506		if (err)
2507			goto out;
2508
2509		ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2510				ext4_ext_pblock(ex));
2511		ex--;
2512		ex_ee_block = le32_to_cpu(ex->ee_block);
2513		ex_ee_len = ext4_ext_get_actual_len(ex);
2514	}
2515
2516	if (correct_index && eh->eh_entries)
2517		err = ext4_ext_correct_indexes(handle, inode, path);
2518
2519	/*
2520	 * If there is still a entry in the leaf node, check to see if
2521	 * it references the partial cluster.  This is the only place
2522	 * where it could; if it doesn't, we can free the cluster.
2523	 */
2524	if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2525	    (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2526	     *partial_cluster)) {
2527		int flags = EXT4_FREE_BLOCKS_FORGET;
2528
2529		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2530			flags |= EXT4_FREE_BLOCKS_METADATA;
2531
2532		ext4_free_blocks(handle, inode, NULL,
2533				 EXT4_C2B(sbi, *partial_cluster),
2534				 sbi->s_cluster_ratio, flags);
2535		*partial_cluster = 0;
 
 
 
 
2536	}
2537
2538	/* if this leaf is free, then we should
2539	 * remove it from index block above */
2540	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2541		err = ext4_ext_rm_idx(handle, inode, path + depth);
2542
2543out:
2544	return err;
2545}
2546
2547/*
2548 * ext4_ext_more_to_rm:
2549 * returns 1 if current index has to be freed (even partial)
2550 */
2551static int
2552ext4_ext_more_to_rm(struct ext4_ext_path *path)
2553{
2554	BUG_ON(path->p_idx == NULL);
2555
2556	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2557		return 0;
2558
2559	/*
2560	 * if truncate on deeper level happened, it wasn't partial,
2561	 * so we have to consider current index for truncation
2562	 */
2563	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2564		return 0;
2565	return 1;
2566}
2567
2568static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2569				 ext4_lblk_t end)
2570{
2571	struct super_block *sb = inode->i_sb;
2572	int depth = ext_depth(inode);
2573	struct ext4_ext_path *path = NULL;
2574	ext4_fsblk_t partial_cluster = 0;
2575	handle_t *handle;
2576	int i = 0, err;
2577
2578	ext_debug("truncate since %u to %u\n", start, end);
 
 
 
 
2579
2580	/* probably first extent we're gonna free will be last in block */
2581	handle = ext4_journal_start(inode, depth + 1);
 
 
2582	if (IS_ERR(handle))
2583		return PTR_ERR(handle);
2584
2585again:
2586	ext4_ext_invalidate_cache(inode);
2587
2588	trace_ext4_ext_remove_space(inode, start, depth);
2589
2590	/*
2591	 * Check if we are removing extents inside the extent tree. If that
2592	 * is the case, we are going to punch a hole inside the extent tree
2593	 * so we have to check whether we need to split the extent covering
2594	 * the last block to remove so we can easily remove the part of it
2595	 * in ext4_ext_rm_leaf().
2596	 */
2597	if (end < EXT_MAX_BLOCKS - 1) {
2598		struct ext4_extent *ex;
2599		ext4_lblk_t ee_block;
 
2600
2601		/* find extent for this block */
2602		path = ext4_ext_find_extent(inode, end, NULL);
 
2603		if (IS_ERR(path)) {
2604			ext4_journal_stop(handle);
2605			return PTR_ERR(path);
2606		}
2607		depth = ext_depth(inode);
 
2608		ex = path[depth].p_ext;
2609		if (!ex) {
2610			ext4_ext_drop_refs(path);
2611			kfree(path);
2612			path = NULL;
2613			goto cont;
 
 
 
2614		}
2615
2616		ee_block = le32_to_cpu(ex->ee_block);
 
2617
2618		/*
2619		 * See if the last block is inside the extent, if so split
2620		 * the extent at 'end' block so we can easily remove the
2621		 * tail of the first part of the split extent in
2622		 * ext4_ext_rm_leaf().
2623		 */
2624		if (end >= ee_block &&
2625		    end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2626			int split_flag = 0;
2627
2628			if (ext4_ext_is_uninitialized(ex))
2629				split_flag = EXT4_EXT_MARK_UNINIT1 |
2630					     EXT4_EXT_MARK_UNINIT2;
 
 
 
 
 
2631
2632			/*
2633			 * Split the extent in two so that 'end' is the last
2634			 * block in the first new extent
 
 
2635			 */
2636			err = ext4_split_extent_at(handle, inode, path,
2637						end + 1, split_flag,
2638						EXT4_GET_BLOCKS_PRE_IO |
2639						EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2640
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2641			if (err < 0)
2642				goto out;
 
 
 
 
2643		}
2644	}
2645cont:
2646
2647	/*
2648	 * We start scanning from right side, freeing all the blocks
2649	 * after i_size and walking into the tree depth-wise.
2650	 */
2651	depth = ext_depth(inode);
2652	if (path) {
2653		int k = i = depth;
2654		while (--k > 0)
2655			path[k].p_block =
2656				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2657	} else {
2658		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2659			       GFP_NOFS);
2660		if (path == NULL) {
2661			ext4_journal_stop(handle);
2662			return -ENOMEM;
2663		}
2664		path[0].p_depth = depth;
2665		path[0].p_hdr = ext_inode_hdr(inode);
2666		i = 0;
2667
2668		if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2669			err = -EIO;
2670			goto out;
2671		}
2672	}
2673	err = 0;
2674
2675	while (i >= 0 && err == 0) {
2676		if (i == depth) {
2677			/* this is leaf block */
2678			err = ext4_ext_rm_leaf(handle, inode, path,
2679					       &partial_cluster, start,
2680					       end);
2681			/* root level has p_bh == NULL, brelse() eats this */
2682			brelse(path[i].p_bh);
2683			path[i].p_bh = NULL;
2684			i--;
2685			continue;
2686		}
2687
2688		/* this is index block */
2689		if (!path[i].p_hdr) {
2690			ext_debug("initialize header\n");
2691			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2692		}
2693
2694		if (!path[i].p_idx) {
2695			/* this level hasn't been touched yet */
2696			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2697			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2698			ext_debug("init index ptr: hdr 0x%p, num %d\n",
2699				  path[i].p_hdr,
2700				  le16_to_cpu(path[i].p_hdr->eh_entries));
2701		} else {
2702			/* we were already here, see at next index */
2703			path[i].p_idx--;
2704		}
2705
2706		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2707				i, EXT_FIRST_INDEX(path[i].p_hdr),
2708				path[i].p_idx);
2709		if (ext4_ext_more_to_rm(path + i)) {
2710			struct buffer_head *bh;
2711			/* go to the next level */
2712			ext_debug("move to level %d (block %llu)\n",
2713				  i + 1, ext4_idx_pblock(path[i].p_idx));
2714			memset(path + i + 1, 0, sizeof(*path));
2715			bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2716			if (!bh) {
 
 
2717				/* should we reset i_size? */
2718				err = -EIO;
2719				break;
2720			}
 
 
 
2721			if (WARN_ON(i + 1 > depth)) {
2722				err = -EIO;
2723				break;
2724			}
2725			if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2726							depth - i - 1, bh)) {
2727				err = -EIO;
2728				break;
2729			}
2730			path[i + 1].p_bh = bh;
2731
2732			/* save actual number of indexes since this
2733			 * number is changed at the next iteration */
2734			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2735			i++;
2736		} else {
2737			/* we finished processing this index, go up */
2738			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2739				/* index is empty, remove it;
2740				 * handle must be already prepared by the
2741				 * truncatei_leaf() */
2742				err = ext4_ext_rm_idx(handle, inode, path + i);
2743			}
2744			/* root level has p_bh == NULL, brelse() eats this */
2745			brelse(path[i].p_bh);
2746			path[i].p_bh = NULL;
2747			i--;
2748			ext_debug("return to level %d\n", i);
2749		}
2750	}
2751
2752	trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2753			path->p_hdr->eh_entries);
2754
2755	/* If we still have something in the partial cluster and we have removed
2756	 * even the first extent, then we should free the blocks in the partial
2757	 * cluster as well. */
2758	if (partial_cluster && path->p_hdr->eh_entries == 0) {
2759		int flags = EXT4_FREE_BLOCKS_FORGET;
2760
2761		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2762			flags |= EXT4_FREE_BLOCKS_METADATA;
2763
 
 
2764		ext4_free_blocks(handle, inode, NULL,
2765				 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2766				 EXT4_SB(sb)->s_cluster_ratio, flags);
2767		partial_cluster = 0;
 
 
2768	}
2769
2770	/* TODO: flexible tree reduction should be here */
2771	if (path->p_hdr->eh_entries == 0) {
2772		/*
2773		 * truncate to zero freed all the tree,
2774		 * so we need to correct eh_depth
2775		 */
2776		err = ext4_ext_get_access(handle, inode, path);
2777		if (err == 0) {
2778			ext_inode_hdr(inode)->eh_depth = 0;
2779			ext_inode_hdr(inode)->eh_max =
2780				cpu_to_le16(ext4_ext_space_root(inode, 0));
2781			err = ext4_ext_dirty(handle, inode, path);
2782		}
2783	}
2784out:
2785	ext4_ext_drop_refs(path);
2786	kfree(path);
2787	if (err == -EAGAIN) {
2788		path = NULL;
2789		goto again;
2790	}
2791	ext4_journal_stop(handle);
2792
2793	return err;
2794}
2795
2796/*
2797 * called at mount time
2798 */
2799void ext4_ext_init(struct super_block *sb)
2800{
2801	/*
2802	 * possible initialization would be here
2803	 */
2804
2805	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2806#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2807		printk(KERN_INFO "EXT4-fs: file extents enabled"
2808#ifdef AGGRESSIVE_TEST
2809		       ", aggressive tests"
2810#endif
2811#ifdef CHECK_BINSEARCH
2812		       ", check binsearch"
2813#endif
2814#ifdef EXTENTS_STATS
2815		       ", stats"
2816#endif
2817		       "\n");
2818#endif
2819#ifdef EXTENTS_STATS
2820		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2821		EXT4_SB(sb)->s_ext_min = 1 << 30;
2822		EXT4_SB(sb)->s_ext_max = 0;
2823#endif
2824	}
2825}
2826
2827/*
2828 * called at umount time
2829 */
2830void ext4_ext_release(struct super_block *sb)
2831{
2832	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2833		return;
2834
2835#ifdef EXTENTS_STATS
2836	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2837		struct ext4_sb_info *sbi = EXT4_SB(sb);
2838		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2839			sbi->s_ext_blocks, sbi->s_ext_extents,
2840			sbi->s_ext_blocks / sbi->s_ext_extents);
2841		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2842			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2843	}
2844#endif
2845}
2846
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2847/* FIXME!! we need to try to merge to left or right after zero-out  */
2848static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2849{
2850	ext4_fsblk_t ee_pblock;
2851	unsigned int ee_len;
2852	int ret;
2853
2854	ee_len    = ext4_ext_get_actual_len(ex);
2855	ee_pblock = ext4_ext_pblock(ex);
2856
2857	ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2858	if (ret > 0)
2859		ret = 0;
2860
2861	return ret;
2862}
2863
2864/*
2865 * ext4_split_extent_at() splits an extent at given block.
2866 *
2867 * @handle: the journal handle
2868 * @inode: the file inode
2869 * @path: the path to the extent
2870 * @split: the logical block where the extent is splitted.
2871 * @split_flags: indicates if the extent could be zeroout if split fails, and
2872 *		 the states(init or uninit) of new extents.
2873 * @flags: flags used to insert new extent to extent tree.
2874 *
2875 *
2876 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2877 * of which are deterimined by split_flag.
2878 *
2879 * There are two cases:
2880 *  a> the extent are splitted into two extent.
2881 *  b> split is not needed, and just mark the extent.
2882 *
2883 * return 0 on success.
2884 */
2885static int ext4_split_extent_at(handle_t *handle,
2886			     struct inode *inode,
2887			     struct ext4_ext_path *path,
2888			     ext4_lblk_t split,
2889			     int split_flag,
2890			     int flags)
2891{
 
2892	ext4_fsblk_t newblock;
2893	ext4_lblk_t ee_block;
2894	struct ext4_extent *ex, newex, orig_ex;
2895	struct ext4_extent *ex2 = NULL;
2896	unsigned int ee_len, depth;
2897	int err = 0;
2898
2899	ext_debug("ext4_split_extents_at: inode %lu, logical"
2900		"block %llu\n", inode->i_ino, (unsigned long long)split);
 
 
2901
2902	ext4_ext_show_leaf(inode, path);
2903
2904	depth = ext_depth(inode);
2905	ex = path[depth].p_ext;
2906	ee_block = le32_to_cpu(ex->ee_block);
2907	ee_len = ext4_ext_get_actual_len(ex);
2908	newblock = split - ee_block + ext4_ext_pblock(ex);
2909
2910	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
 
 
 
 
2911
2912	err = ext4_ext_get_access(handle, inode, path + depth);
2913	if (err)
2914		goto out;
2915
2916	if (split == ee_block) {
2917		/*
2918		 * case b: block @split is the block that the extent begins with
2919		 * then we just change the state of the extent, and splitting
2920		 * is not needed.
2921		 */
2922		if (split_flag & EXT4_EXT_MARK_UNINIT2)
2923			ext4_ext_mark_uninitialized(ex);
2924		else
2925			ext4_ext_mark_initialized(ex);
2926
2927		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2928			ext4_ext_try_to_merge(inode, path, ex);
2929
2930		err = ext4_ext_dirty(handle, inode, path + depth);
2931		goto out;
2932	}
2933
2934	/* case a */
2935	memcpy(&orig_ex, ex, sizeof(orig_ex));
2936	ex->ee_len = cpu_to_le16(split - ee_block);
2937	if (split_flag & EXT4_EXT_MARK_UNINIT1)
2938		ext4_ext_mark_uninitialized(ex);
2939
2940	/*
2941	 * path may lead to new leaf, not to original leaf any more
2942	 * after ext4_ext_insert_extent() returns,
2943	 */
2944	err = ext4_ext_dirty(handle, inode, path + depth);
2945	if (err)
2946		goto fix_extent_len;
2947
2948	ex2 = &newex;
2949	ex2->ee_block = cpu_to_le32(split);
2950	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2951	ext4_ext_store_pblock(ex2, newblock);
2952	if (split_flag & EXT4_EXT_MARK_UNINIT2)
2953		ext4_ext_mark_uninitialized(ex2);
2954
2955	err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2956	if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2957		err = ext4_ext_zeroout(inode, &orig_ex);
2958		if (err)
2959			goto fix_extent_len;
2960		/* update the extent length and mark as initialized */
2961		ex->ee_len = cpu_to_le16(ee_len);
2962		ext4_ext_try_to_merge(inode, path, ex);
2963		err = ext4_ext_dirty(handle, inode, path + depth);
2964		goto out;
2965	} else if (err)
2966		goto fix_extent_len;
2967
2968out:
2969	ext4_ext_show_leaf(inode, path);
2970	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2971
2972fix_extent_len:
2973	ex->ee_len = orig_ex.ee_len;
2974	ext4_ext_dirty(handle, inode, path + depth);
 
 
 
 
 
 
 
2975	return err;
2976}
2977
2978/*
2979 * ext4_split_extents() splits an extent and mark extent which is covered
2980 * by @map as split_flags indicates
2981 *
2982 * It may result in splitting the extent into multiple extents (upto three)
2983 * There are three possibilities:
2984 *   a> There is no split required
2985 *   b> Splits in two extents: Split is happening at either end of the extent
2986 *   c> Splits in three extents: Somone is splitting in middle of the extent
2987 *
2988 */
2989static int ext4_split_extent(handle_t *handle,
2990			      struct inode *inode,
2991			      struct ext4_ext_path *path,
2992			      struct ext4_map_blocks *map,
2993			      int split_flag,
2994			      int flags)
2995{
 
2996	ext4_lblk_t ee_block;
2997	struct ext4_extent *ex;
2998	unsigned int ee_len, depth;
2999	int err = 0;
3000	int uninitialized;
3001	int split_flag1, flags1;
 
3002
3003	depth = ext_depth(inode);
3004	ex = path[depth].p_ext;
3005	ee_block = le32_to_cpu(ex->ee_block);
3006	ee_len = ext4_ext_get_actual_len(ex);
3007	uninitialized = ext4_ext_is_uninitialized(ex);
3008
3009	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3010		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3011			      EXT4_EXT_MAY_ZEROOUT : 0;
3012		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3013		if (uninitialized)
3014			split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3015				       EXT4_EXT_MARK_UNINIT2;
3016		err = ext4_split_extent_at(handle, inode, path,
 
 
3017				map->m_lblk + map->m_len, split_flag1, flags1);
3018		if (err)
3019			goto out;
 
 
3020	}
3021
3022	ext4_ext_drop_refs(path);
3023	path = ext4_ext_find_extent(inode, map->m_lblk, path);
 
 
3024	if (IS_ERR(path))
3025		return PTR_ERR(path);
 
 
 
 
 
 
 
 
3026
3027	if (map->m_lblk >= ee_block) {
3028		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
3029			      EXT4_EXT_MAY_ZEROOUT : 0;
3030		if (uninitialized)
3031			split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3032		if (split_flag & EXT4_EXT_MARK_UNINIT2)
3033			split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3034		err = ext4_split_extent_at(handle, inode, path,
3035				map->m_lblk, split_flag1, flags);
3036		if (err)
3037			goto out;
3038	}
3039
3040	ext4_ext_show_leaf(inode, path);
3041out:
3042	return err ? err : map->m_len;
3043}
3044
3045#define EXT4_EXT_ZERO_LEN 7
3046/*
3047 * This function is called by ext4_ext_map_blocks() if someone tries to write
3048 * to an uninitialized extent. It may result in splitting the uninitialized
3049 * extent into multiple extents (up to three - one initialized and two
3050 * uninitialized).
3051 * There are three possibilities:
3052 *   a> There is no split required: Entire extent should be initialized
3053 *   b> Splits in two extents: Write is happening at either end of the extent
3054 *   c> Splits in three extents: Somone is writing in middle of the extent
3055 *
3056 * Pre-conditions:
3057 *  - The extent pointed to by 'path' is uninitialized.
3058 *  - The extent pointed to by 'path' contains a superset
3059 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3060 *
3061 * Post-conditions on success:
3062 *  - the returned value is the number of blocks beyond map->l_lblk
3063 *    that are allocated and initialized.
3064 *    It is guaranteed to be >= map->m_len.
3065 */
3066static int ext4_ext_convert_to_initialized(handle_t *handle,
3067					   struct inode *inode,
3068					   struct ext4_map_blocks *map,
3069					   struct ext4_ext_path *path)
 
3070{
 
 
3071	struct ext4_extent_header *eh;
3072	struct ext4_map_blocks split_map;
3073	struct ext4_extent zero_ex;
3074	struct ext4_extent *ex;
3075	ext4_lblk_t ee_block, eof_block;
3076	unsigned int ee_len, depth;
3077	int allocated;
3078	int err = 0;
3079	int split_flag = 0;
3080
3081	ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3082		"block %llu, max_blocks %u\n", inode->i_ino,
3083		(unsigned long long)map->m_lblk, map->m_len);
3084
3085	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3086		inode->i_sb->s_blocksize_bits;
3087	if (eof_block < map->m_lblk + map->m_len)
3088		eof_block = map->m_lblk + map->m_len;
 
3089
3090	depth = ext_depth(inode);
3091	eh = path[depth].p_hdr;
3092	ex = path[depth].p_ext;
3093	ee_block = le32_to_cpu(ex->ee_block);
3094	ee_len = ext4_ext_get_actual_len(ex);
3095	allocated = ee_len - (map->m_lblk - ee_block);
 
3096
3097	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3098
3099	/* Pre-conditions */
3100	BUG_ON(!ext4_ext_is_uninitialized(ex));
3101	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3102
3103	/*
3104	 * Attempt to transfer newly initialized blocks from the currently
3105	 * uninitialized extent to its left neighbor. This is much cheaper
3106	 * than an insertion followed by a merge as those involve costly
3107	 * memmove() calls. This is the common case in steady state for
3108	 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3109	 * writes.
3110	 *
3111	 * Limitations of the current logic:
3112	 *  - L1: we only deal with writes at the start of the extent.
3113	 *    The approach could be extended to writes at the end
3114	 *    of the extent but this scenario was deemed less common.
3115	 *  - L2: we do not deal with writes covering the whole extent.
3116	 *    This would require removing the extent if the transfer
3117	 *    is possible.
3118	 *  - L3: we only attempt to merge with an extent stored in the
3119	 *    same extent tree node.
3120	 */
3121	if ((map->m_lblk == ee_block) &&	/*L1*/
3122		(map->m_len < ee_len) &&	/*L2*/
3123		(ex > EXT_FIRST_EXTENT(eh))) {	/*L3*/
3124		struct ext4_extent *prev_ex;
3125		ext4_lblk_t prev_lblk;
3126		ext4_fsblk_t prev_pblk, ee_pblk;
3127		unsigned int prev_len, write_len;
3128
3129		prev_ex = ex - 1;
3130		prev_lblk = le32_to_cpu(prev_ex->ee_block);
3131		prev_len = ext4_ext_get_actual_len(prev_ex);
3132		prev_pblk = ext4_ext_pblock(prev_ex);
3133		ee_pblk = ext4_ext_pblock(ex);
3134		write_len = map->m_len;
3135
3136		/*
3137		 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3138		 * upon those conditions:
3139		 * - C1: prev_ex is initialized,
3140		 * - C2: prev_ex is logically abutting ex,
3141		 * - C3: prev_ex is physically abutting ex,
3142		 * - C4: prev_ex can receive the additional blocks without
3143		 *   overflowing the (initialized) length limit.
3144		 */
3145		if ((!ext4_ext_is_uninitialized(prev_ex)) &&		/*C1*/
3146			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3147			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3148			(prev_len < (EXT_INIT_MAX_LEN - write_len))) {	/*C4*/
3149			err = ext4_ext_get_access(handle, inode, path + depth);
3150			if (err)
3151				goto out;
3152
3153			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3154				map, ex, prev_ex);
 
 
 
 
 
 
3155
3156			/* Shift the start of ex by 'write_len' blocks */
3157			ex->ee_block = cpu_to_le32(ee_block + write_len);
3158			ext4_ext_store_pblock(ex, ee_pblk + write_len);
3159			ex->ee_len = cpu_to_le16(ee_len - write_len);
3160			ext4_ext_mark_uninitialized(ex); /* Restore the flag */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3161
3162			/* Extend prev_ex by 'write_len' blocks */
3163			prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3164
3165			/* Mark the block containing both extents as dirty */
3166			ext4_ext_dirty(handle, inode, path + depth);
 
 
 
3167
3168			/* Update path to point to the right extent */
3169			path[depth].p_ext = prev_ex;
3170
3171			/* Result: number of initialized blocks past m_lblk */
3172			allocated = write_len;
3173			goto out;
3174		}
3175	}
 
 
 
 
 
 
 
 
 
3176
3177	WARN_ON(map->m_lblk < ee_block);
3178	/*
3179	 * It is safe to convert extent to initialized via explicit
3180	 * zeroout only if extent is fully insde i_size or new_size.
3181	 */
3182	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3183
3184	/* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3185	if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3186	    (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3187		err = ext4_ext_zeroout(inode, ex);
3188		if (err)
3189			goto out;
3190
3191		err = ext4_ext_get_access(handle, inode, path + depth);
3192		if (err)
3193			goto out;
3194		ext4_ext_mark_initialized(ex);
3195		ext4_ext_try_to_merge(inode, path, ex);
3196		err = ext4_ext_dirty(handle, inode, path + depth);
3197		goto out;
3198	}
3199
3200	/*
3201	 * four cases:
3202	 * 1. split the extent into three extents.
3203	 * 2. split the extent into two extents, zeroout the first half.
3204	 * 3. split the extent into two extents, zeroout the second half.
 
 
3205	 * 4. split the extent into two extents with out zeroout.
 
 
3206	 */
3207	split_map.m_lblk = map->m_lblk;
3208	split_map.m_len = map->m_len;
3209
3210	if (allocated > map->m_len) {
3211		if (allocated <= EXT4_EXT_ZERO_LEN &&
3212		    (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3213			/* case 3 */
3214			zero_ex.ee_block =
3215					 cpu_to_le32(map->m_lblk);
3216			zero_ex.ee_len = cpu_to_le16(allocated);
3217			ext4_ext_store_pblock(&zero_ex,
3218				ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3219			err = ext4_ext_zeroout(inode, &zero_ex);
 
 
3220			if (err)
3221				goto out;
3222			split_map.m_lblk = map->m_lblk;
3223			split_map.m_len = allocated;
3224		} else if ((map->m_lblk - ee_block + map->m_len <
3225			   EXT4_EXT_ZERO_LEN) &&
3226			   (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3227			/* case 2 */
3228			if (map->m_lblk != ee_block) {
3229				zero_ex.ee_block = ex->ee_block;
3230				zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3231							ee_block);
3232				ext4_ext_store_pblock(&zero_ex,
3233						      ext4_ext_pblock(ex));
3234				err = ext4_ext_zeroout(inode, &zero_ex);
3235				if (err)
3236					goto out;
3237			}
3238
 
3239			split_map.m_lblk = ee_block;
3240			split_map.m_len = map->m_lblk - ee_block + map->m_len;
3241			allocated = map->m_len;
3242		}
3243	}
3244
3245	allocated = ext4_split_extent(handle, inode, path,
3246				       &split_map, split_flag, 0);
3247	if (allocated < 0)
3248		err = allocated;
3249
3250out:
 
 
 
 
 
 
3251	return err ? err : allocated;
3252}
3253
3254/*
3255 * This function is called by ext4_ext_map_blocks() from
3256 * ext4_get_blocks_dio_write() when DIO to write
3257 * to an uninitialized extent.
3258 *
3259 * Writing to an uninitialized extent may result in splitting the uninitialized
3260 * extent into multiple /initialized uninitialized extents (up to three)
3261 * There are three possibilities:
3262 *   a> There is no split required: Entire extent should be uninitialized
3263 *   b> Splits in two extents: Write is happening at either end of the extent
3264 *   c> Splits in three extents: Somone is writing in middle of the extent
3265 *
 
 
3266 * One of more index blocks maybe needed if the extent tree grow after
3267 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3268 * complete, we need to split the uninitialized extent before DIO submit
3269 * the IO. The uninitialized extent called at this time will be split
3270 * into three uninitialized extent(at most). After IO complete, the part
3271 * being filled will be convert to initialized by the end_io callback function
3272 * via ext4_convert_unwritten_extents().
3273 *
3274 * Returns the size of uninitialized extent to be written on success.
3275 */
3276static int ext4_split_unwritten_extents(handle_t *handle,
3277					struct inode *inode,
3278					struct ext4_map_blocks *map,
3279					struct ext4_ext_path *path,
3280					int flags)
3281{
 
3282	ext4_lblk_t eof_block;
3283	ext4_lblk_t ee_block;
3284	struct ext4_extent *ex;
3285	unsigned int ee_len;
3286	int split_flag = 0, depth;
3287
3288	ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3289		"block %llu, max_blocks %u\n", inode->i_ino,
3290		(unsigned long long)map->m_lblk, map->m_len);
3291
3292	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3293		inode->i_sb->s_blocksize_bits;
3294	if (eof_block < map->m_lblk + map->m_len)
3295		eof_block = map->m_lblk + map->m_len;
3296	/*
3297	 * It is safe to convert extent to initialized via explicit
3298	 * zeroout only if extent is fully insde i_size or new_size.
3299	 */
3300	depth = ext_depth(inode);
3301	ex = path[depth].p_ext;
3302	ee_block = le32_to_cpu(ex->ee_block);
3303	ee_len = ext4_ext_get_actual_len(ex);
3304
3305	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3306	split_flag |= EXT4_EXT_MARK_UNINIT2;
3307
 
 
 
 
 
 
3308	flags |= EXT4_GET_BLOCKS_PRE_IO;
3309	return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3310}
3311
3312static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3313					      struct inode *inode,
3314					      struct ext4_ext_path *path)
 
3315{
 
3316	struct ext4_extent *ex;
 
 
3317	int depth;
3318	int err = 0;
3319
3320	depth = ext_depth(inode);
3321	ex = path[depth].p_ext;
 
 
3322
3323	ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3324		"block %llu, max_blocks %u\n", inode->i_ino,
3325		(unsigned long long)le32_to_cpu(ex->ee_block),
3326		ext4_ext_get_actual_len(ex));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3327
3328	err = ext4_ext_get_access(handle, inode, path + depth);
3329	if (err)
3330		goto out;
3331	/* first mark the extent as initialized */
3332	ext4_ext_mark_initialized(ex);
3333
3334	/* note: ext4_ext_correct_indexes() isn't needed here because
3335	 * borders are not changed
3336	 */
3337	ext4_ext_try_to_merge(inode, path, ex);
3338
3339	/* Mark modified extent as dirty */
3340	err = ext4_ext_dirty(handle, inode, path + depth);
3341out:
3342	ext4_ext_show_leaf(inode, path);
3343	return err;
3344}
3345
3346static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3347			sector_t block, int count)
 
 
 
3348{
3349	int i;
3350	for (i = 0; i < count; i++)
3351                unmap_underlying_metadata(bdev, block + i);
3352}
3353
3354/*
3355 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3356 */
3357static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3358			      ext4_lblk_t lblk,
3359			      struct ext4_ext_path *path,
3360			      unsigned int len)
3361{
3362	int i, depth;
3363	struct ext4_extent_header *eh;
3364	struct ext4_extent *last_ex;
3365
3366	if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3367		return 0;
3368
3369	depth = ext_depth(inode);
3370	eh = path[depth].p_hdr;
3371
3372	/*
3373	 * We're going to remove EOFBLOCKS_FL entirely in future so we
3374	 * do not care for this case anymore. Simply remove the flag
3375	 * if there are no extents.
3376	 */
3377	if (unlikely(!eh->eh_entries))
3378		goto out;
3379	last_ex = EXT_LAST_EXTENT(eh);
3380	/*
3381	 * We should clear the EOFBLOCKS_FL flag if we are writing the
3382	 * last block in the last extent in the file.  We test this by
3383	 * first checking to see if the caller to
3384	 * ext4_ext_get_blocks() was interested in the last block (or
3385	 * a block beyond the last block) in the current extent.  If
3386	 * this turns out to be false, we can bail out from this
3387	 * function immediately.
3388	 */
3389	if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3390	    ext4_ext_get_actual_len(last_ex))
3391		return 0;
3392	/*
3393	 * If the caller does appear to be planning to write at or
3394	 * beyond the end of the current extent, we then test to see
3395	 * if the current extent is the last extent in the file, by
3396	 * checking to make sure it was reached via the rightmost node
3397	 * at each level of the tree.
3398	 */
3399	for (i = depth-1; i >= 0; i--)
3400		if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3401			return 0;
3402out:
3403	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3404	return ext4_mark_inode_dirty(handle, inode);
3405}
3406
3407/**
3408 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3409 *
3410 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3411 * whether there are any buffers marked for delayed allocation. It returns '1'
3412 * on the first delalloc'ed buffer head found. If no buffer head in the given
3413 * range is marked for delalloc, it returns 0.
3414 * lblk_start should always be <= lblk_end.
3415 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3416 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3417 * block sooner). This is useful when blocks are truncated sequentially from
3418 * lblk_start towards lblk_end.
3419 */
3420static int ext4_find_delalloc_range(struct inode *inode,
3421				    ext4_lblk_t lblk_start,
3422				    ext4_lblk_t lblk_end,
3423				    int search_hint_reverse)
3424{
3425	struct address_space *mapping = inode->i_mapping;
3426	struct buffer_head *head, *bh = NULL;
3427	struct page *page;
3428	ext4_lblk_t i, pg_lblk;
3429	pgoff_t index;
3430
3431	if (!test_opt(inode->i_sb, DELALLOC))
3432		return 0;
3433
3434	/* reverse search wont work if fs block size is less than page size */
3435	if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3436		search_hint_reverse = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
3437
3438	if (search_hint_reverse)
3439		i = lblk_end;
3440	else
3441		i = lblk_start;
 
3442
3443	index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
 
 
 
3444
3445	while ((i >= lblk_start) && (i <= lblk_end)) {
3446		page = find_get_page(mapping, index);
3447		if (!page)
3448			goto nextpage;
3449
3450		if (!page_has_buffers(page))
3451			goto nextpage;
3452
3453		head = page_buffers(page);
3454		if (!head)
3455			goto nextpage;
3456
3457		bh = head;
3458		pg_lblk = index << (PAGE_CACHE_SHIFT -
3459						inode->i_blkbits);
3460		do {
3461			if (unlikely(pg_lblk < lblk_start)) {
3462				/*
3463				 * This is possible when fs block size is less
3464				 * than page size and our cluster starts/ends in
3465				 * middle of the page. So we need to skip the
3466				 * initial few blocks till we reach the 'lblk'
3467				 */
3468				pg_lblk++;
3469				continue;
3470			}
3471
3472			/* Check if the buffer is delayed allocated and that it
3473			 * is not yet mapped. (when da-buffers are mapped during
3474			 * their writeout, their da_mapped bit is set.)
3475			 */
3476			if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3477				page_cache_release(page);
3478				trace_ext4_find_delalloc_range(inode,
3479						lblk_start, lblk_end,
3480						search_hint_reverse,
3481						1, i);
3482				return 1;
3483			}
3484			if (search_hint_reverse)
3485				i--;
3486			else
3487				i++;
3488		} while ((i >= lblk_start) && (i <= lblk_end) &&
3489				((bh = bh->b_this_page) != head));
3490nextpage:
3491		if (page)
3492			page_cache_release(page);
3493		/*
3494		 * Move to next page. 'i' will be the first lblk in the next
3495		 * page.
3496		 */
3497		if (search_hint_reverse)
3498			index--;
3499		else
3500			index++;
3501		i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3502	}
3503
3504	trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3505					search_hint_reverse, 0, 0);
 
 
3506	return 0;
3507}
3508
3509int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3510			       int search_hint_reverse)
3511{
3512	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3513	ext4_lblk_t lblk_start, lblk_end;
3514	lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3515	lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3516
3517	return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3518					search_hint_reverse);
3519}
3520
3521/**
3522 * Determines how many complete clusters (out of those specified by the 'map')
3523 * are under delalloc and were reserved quota for.
3524 * This function is called when we are writing out the blocks that were
3525 * originally written with their allocation delayed, but then the space was
3526 * allocated using fallocate() before the delayed allocation could be resolved.
3527 * The cases to look for are:
3528 * ('=' indicated delayed allocated blocks
3529 *  '-' indicates non-delayed allocated blocks)
3530 * (a) partial clusters towards beginning and/or end outside of allocated range
3531 *     are not delalloc'ed.
3532 *	Ex:
3533 *	|----c---=|====c====|====c====|===-c----|
3534 *	         |++++++ allocated ++++++|
3535 *	==> 4 complete clusters in above example
3536 *
3537 * (b) partial cluster (outside of allocated range) towards either end is
3538 *     marked for delayed allocation. In this case, we will exclude that
3539 *     cluster.
3540 *	Ex:
3541 *	|----====c========|========c========|
3542 *	     |++++++ allocated ++++++|
3543 *	==> 1 complete clusters in above example
3544 *
3545 *	Ex:
3546 *	|================c================|
3547 *            |++++++ allocated ++++++|
3548 *	==> 0 complete clusters in above example
3549 *
3550 * The ext4_da_update_reserve_space will be called only if we
3551 * determine here that there were some "entire" clusters that span
3552 * this 'allocated' range.
3553 * In the non-bigalloc case, this function will just end up returning num_blks
3554 * without ever calling ext4_find_delalloc_range.
3555 */
3556static unsigned int
3557get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3558			   unsigned int num_blks)
3559{
3560	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3561	ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3562	ext4_lblk_t lblk_from, lblk_to, c_offset;
3563	unsigned int allocated_clusters = 0;
3564
3565	alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3566	alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3567
3568	/* max possible clusters for this allocation */
3569	allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3570
3571	trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3572
3573	/* Check towards left side */
3574	c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3575	if (c_offset) {
3576		lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3577		lblk_to = lblk_from + c_offset - 1;
3578
3579		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3580			allocated_clusters--;
3581	}
3582
3583	/* Now check towards right. */
3584	c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3585	if (allocated_clusters && c_offset) {
3586		lblk_from = lblk_start + num_blks;
3587		lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3588
3589		if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3590			allocated_clusters--;
3591	}
3592
3593	return allocated_clusters;
3594}
3595
3596static int
3597ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3598			struct ext4_map_blocks *map,
3599			struct ext4_ext_path *path, int flags,
3600			unsigned int allocated, ext4_fsblk_t newblock)
3601{
 
3602	int ret = 0;
3603	int err = 0;
3604	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3605
3606	ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3607		  "block %llu, max_blocks %u, flags %x, allocated %u\n",
3608		  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3609		  flags, allocated);
3610	ext4_ext_show_leaf(inode, path);
3611
3612	trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3613						    newblock);
 
 
 
 
 
 
3614
3615	/* get_block() before submit the IO, split the extent */
3616	if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3617		ret = ext4_split_unwritten_extents(handle, inode, map,
3618						   path, flags);
 
 
 
 
3619		/*
3620		 * Flag the inode(non aio case) or end_io struct (aio case)
3621		 * that this IO needs to conversion to written when IO is
3622		 * completed
3623		 */
3624		if (io)
3625			ext4_set_io_unwritten_flag(inode, io);
3626		else
3627			ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3628		if (ext4_should_dioread_nolock(inode))
3629			map->m_flags |= EXT4_MAP_UNINIT;
 
 
3630		goto out;
3631	}
3632	/* IO end_io complete, convert the filled extent to written */
3633	if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3634		ret = ext4_convert_unwritten_extents_endio(handle, inode,
3635							path);
3636		if (ret >= 0) {
3637			ext4_update_inode_fsync_trans(handle, inode, 1);
3638			err = check_eofblocks_fl(handle, inode, map->m_lblk,
3639						 path, map->m_len);
3640		} else
3641			err = ret;
3642		goto out2;
3643	}
3644	/* buffered IO case */
3645	/*
3646	 * repeat fallocate creation request
3647	 * we already have an unwritten extent
3648	 */
3649	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
 
3650		goto map_out;
 
3651
3652	/* buffered READ or buffered write_begin() lookup */
3653	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3654		/*
3655		 * We have blocks reserved already.  We
3656		 * return allocated blocks so that delalloc
3657		 * won't do block reservation for us.  But
3658		 * the buffer head will be unmapped so that
3659		 * a read from the block returns 0s.
3660		 */
3661		map->m_flags |= EXT4_MAP_UNWRITTEN;
3662		goto out1;
3663	}
3664
3665	/* buffered write, writepage time, convert*/
3666	ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3667	if (ret >= 0)
3668		ext4_update_inode_fsync_trans(handle, inode, 1);
3669out:
3670	if (ret <= 0) {
 
3671		err = ret;
3672		goto out2;
3673	} else
3674		allocated = ret;
3675	map->m_flags |= EXT4_MAP_NEW;
3676	/*
3677	 * if we allocated more blocks than requested
3678	 * we need to make sure we unmap the extra block
3679	 * allocated. The actual needed block will get
3680	 * unmapped later when we find the buffer_head marked
3681	 * new.
3682	 */
3683	if (allocated > map->m_len) {
3684		unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3685					newblock + map->m_len,
3686					allocated - map->m_len);
3687		allocated = map->m_len;
3688	}
3689
3690	/*
3691	 * If we have done fallocate with the offset that is already
3692	 * delayed allocated, we would have block reservation
3693	 * and quota reservation done in the delayed write path.
3694	 * But fallocate would have already updated quota and block
3695	 * count for this offset. So cancel these reservation
3696	 */
3697	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3698		unsigned int reserved_clusters;
3699		reserved_clusters = get_reserved_cluster_alloc(inode,
3700				map->m_lblk, map->m_len);
3701		if (reserved_clusters)
3702			ext4_da_update_reserve_space(inode,
3703						     reserved_clusters,
3704						     0);
3705	}
3706
 
 
 
3707map_out:
3708	map->m_flags |= EXT4_MAP_MAPPED;
3709	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3710		err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3711					 map->m_len);
3712		if (err < 0)
3713			goto out2;
3714	}
3715out1:
 
3716	if (allocated > map->m_len)
3717		allocated = map->m_len;
 
3718	ext4_ext_show_leaf(inode, path);
3719	map->m_pblk = newblock;
3720	map->m_len = allocated;
3721out2:
3722	if (path) {
3723		ext4_ext_drop_refs(path);
3724		kfree(path);
3725	}
3726	return err ? err : allocated;
3727}
3728
3729/*
3730 * get_implied_cluster_alloc - check to see if the requested
3731 * allocation (in the map structure) overlaps with a cluster already
3732 * allocated in an extent.
3733 *	@sb	The filesystem superblock structure
3734 *	@map	The requested lblk->pblk mapping
3735 *	@ex	The extent structure which might contain an implied
3736 *			cluster allocation
3737 *
3738 * This function is called by ext4_ext_map_blocks() after we failed to
3739 * find blocks that were already in the inode's extent tree.  Hence,
3740 * we know that the beginning of the requested region cannot overlap
3741 * the extent from the inode's extent tree.  There are three cases we
3742 * want to catch.  The first is this case:
3743 *
3744 *		 |--- cluster # N--|
3745 *    |--- extent ---|	|---- requested region ---|
3746 *			|==========|
3747 *
3748 * The second case that we need to test for is this one:
3749 *
3750 *   |--------- cluster # N ----------------|
3751 *	   |--- requested region --|   |------- extent ----|
3752 *	   |=======================|
3753 *
3754 * The third case is when the requested region lies between two extents
3755 * within the same cluster:
3756 *          |------------- cluster # N-------------|
3757 * |----- ex -----|                  |---- ex_right ----|
3758 *                  |------ requested region ------|
3759 *                  |================|
3760 *
3761 * In each of the above cases, we need to set the map->m_pblk and
3762 * map->m_len so it corresponds to the return the extent labelled as
3763 * "|====|" from cluster #N, since it is already in use for data in
3764 * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
3765 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3766 * as a new "allocated" block region.  Otherwise, we will return 0 and
3767 * ext4_ext_map_blocks() will then allocate one or more new clusters
3768 * by calling ext4_mb_new_blocks().
3769 */
3770static int get_implied_cluster_alloc(struct super_block *sb,
3771				     struct ext4_map_blocks *map,
3772				     struct ext4_extent *ex,
3773				     struct ext4_ext_path *path)
3774{
3775	struct ext4_sb_info *sbi = EXT4_SB(sb);
3776	ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3777	ext4_lblk_t ex_cluster_start, ex_cluster_end;
3778	ext4_lblk_t rr_cluster_start;
3779	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3780	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3781	unsigned short ee_len = ext4_ext_get_actual_len(ex);
3782
3783	/* The extent passed in that we are trying to match */
3784	ex_cluster_start = EXT4_B2C(sbi, ee_block);
3785	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3786
3787	/* The requested region passed into ext4_map_blocks() */
3788	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3789
3790	if ((rr_cluster_start == ex_cluster_end) ||
3791	    (rr_cluster_start == ex_cluster_start)) {
3792		if (rr_cluster_start == ex_cluster_end)
3793			ee_start += ee_len - 1;
3794		map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3795			c_offset;
3796		map->m_len = min(map->m_len,
3797				 (unsigned) sbi->s_cluster_ratio - c_offset);
3798		/*
3799		 * Check for and handle this case:
3800		 *
3801		 *   |--------- cluster # N-------------|
3802		 *		       |------- extent ----|
3803		 *	   |--- requested region ---|
3804		 *	   |===========|
3805		 */
3806
3807		if (map->m_lblk < ee_block)
3808			map->m_len = min(map->m_len, ee_block - map->m_lblk);
3809
3810		/*
3811		 * Check for the case where there is already another allocated
3812		 * block to the right of 'ex' but before the end of the cluster.
3813		 *
3814		 *          |------------- cluster # N-------------|
3815		 * |----- ex -----|                  |---- ex_right ----|
3816		 *                  |------ requested region ------|
3817		 *                  |================|
3818		 */
3819		if (map->m_lblk > ee_block) {
3820			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3821			map->m_len = min(map->m_len, next - map->m_lblk);
3822		}
3823
3824		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3825		return 1;
3826	}
3827
3828	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3829	return 0;
3830}
3831
3832
3833/*
3834 * Block allocation/map/preallocation routine for extents based files
3835 *
3836 *
3837 * Need to be called with
3838 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3839 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3840 *
3841 * return > 0, number of of blocks already mapped/allocated
3842 *          if create == 0 and these are pre-allocated blocks
3843 *          	buffer head is unmapped
3844 *          otherwise blocks are mapped
3845 *
3846 * return = 0, if plain look up failed (blocks have not been allocated)
3847 *          buffer head is unmapped
3848 *
3849 * return < 0, error case.
3850 */
3851int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3852			struct ext4_map_blocks *map, int flags)
3853{
3854	struct ext4_ext_path *path = NULL;
3855	struct ext4_extent newex, *ex, *ex2;
3856	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3857	ext4_fsblk_t newblock = 0;
3858	int free_on_err = 0, err = 0, depth, ret;
3859	unsigned int allocated = 0, offset = 0;
3860	unsigned int allocated_clusters = 0;
3861	struct ext4_allocation_request ar;
3862	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3863	ext4_lblk_t cluster_offset;
3864
3865	ext_debug("blocks %u/%u requested for inode %lu\n",
3866		  map->m_lblk, map->m_len, inode->i_ino);
3867	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3868
3869	/* check in cache */
3870	if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3871		if (!newex.ee_start_lo && !newex.ee_start_hi) {
3872			if ((sbi->s_cluster_ratio > 1) &&
3873			    ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3874				map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3875
3876			if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3877				/*
3878				 * block isn't allocated yet and
3879				 * user doesn't want to allocate it
3880				 */
3881				goto out2;
3882			}
3883			/* we should allocate requested block */
3884		} else {
3885			/* block is already allocated */
3886			if (sbi->s_cluster_ratio > 1)
3887				map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3888			newblock = map->m_lblk
3889				   - le32_to_cpu(newex.ee_block)
3890				   + ext4_ext_pblock(&newex);
3891			/* number of remaining blocks in the extent */
3892			allocated = ext4_ext_get_actual_len(&newex) -
3893				(map->m_lblk - le32_to_cpu(newex.ee_block));
3894			goto out;
3895		}
3896	}
3897
3898	/* find extent for this block */
3899	path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3900	if (IS_ERR(path)) {
3901		err = PTR_ERR(path);
3902		path = NULL;
3903		goto out2;
3904	}
3905
3906	depth = ext_depth(inode);
3907
3908	/*
3909	 * consistent leaf must not be empty;
3910	 * this situation is possible, though, _during_ tree modification;
3911	 * this is why assert can't be put in ext4_ext_find_extent()
3912	 */
3913	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3914		EXT4_ERROR_INODE(inode, "bad extent address "
3915				 "lblock: %lu, depth: %d pblock %lld",
3916				 (unsigned long) map->m_lblk, depth,
3917				 path[depth].p_block);
3918		err = -EIO;
3919		goto out2;
3920	}
3921
3922	ex = path[depth].p_ext;
3923	if (ex) {
3924		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3925		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3926		unsigned short ee_len;
3927
 
3928		/*
3929		 * Uninitialized extents are treated as holes, except that
3930		 * we split out initialized portions during a write.
3931		 */
3932		ee_len = ext4_ext_get_actual_len(ex);
3933
3934		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3935
3936		/* if found extent covers block, simply return it */
3937		if (in_range(map->m_lblk, ee_block, ee_len)) {
3938			newblock = map->m_lblk - ee_block + ee_start;
3939			/* number of remaining blocks in the extent */
3940			allocated = ee_len - (map->m_lblk - ee_block);
3941			ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3942				  ee_block, ee_len, newblock);
3943
3944			/*
3945			 * Do not put uninitialized extent
3946			 * in the cache
3947			 */
3948			if (!ext4_ext_is_uninitialized(ex)) {
3949				ext4_ext_put_in_cache(inode, ee_block,
3950					ee_len, ee_start);
 
 
 
 
 
 
 
 
 
3951				goto out;
3952			}
3953			ret = ext4_ext_handle_uninitialized_extents(
3954				handle, inode, map, path, flags,
 
3955				allocated, newblock);
3956			return ret;
 
 
 
 
3957		}
3958	}
3959
3960	if ((sbi->s_cluster_ratio > 1) &&
3961	    ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3962		map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3963
3964	/*
3965	 * requested block isn't allocated yet;
3966	 * we couldn't try to create block if create flag is zero
3967	 */
3968	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
 
 
 
 
3969		/*
3970		 * put just found gap into cache to speed up
3971		 * subsequent requests
3972		 */
3973		ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3974		goto out2;
 
 
 
 
 
 
 
3975	}
3976
3977	/*
3978	 * Okay, we need to do block allocation.
3979	 */
3980	map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3981	newex.ee_block = cpu_to_le32(map->m_lblk);
3982	cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3983
3984	/*
3985	 * If we are doing bigalloc, check to see if the extent returned
3986	 * by ext4_ext_find_extent() implies a cluster we can use.
3987	 */
3988	if (cluster_offset && ex &&
3989	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3990		ar.len = allocated = map->m_len;
3991		newblock = map->m_pblk;
3992		map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3993		goto got_allocated_blocks;
3994	}
3995
3996	/* find neighbour allocated blocks */
3997	ar.lleft = map->m_lblk;
3998	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3999	if (err)
4000		goto out2;
4001	ar.lright = map->m_lblk;
4002	ex2 = NULL;
4003	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4004	if (err)
4005		goto out2;
4006
4007	/* Check if the extent after searching to the right implies a
4008	 * cluster we can use. */
4009	if ((sbi->s_cluster_ratio > 1) && ex2 &&
4010	    get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4011		ar.len = allocated = map->m_len;
4012		newblock = map->m_pblk;
4013		map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4014		goto got_allocated_blocks;
4015	}
4016
4017	/*
4018	 * See if request is beyond maximum number of blocks we can have in
4019	 * a single extent. For an initialized extent this limit is
4020	 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4021	 * EXT_UNINIT_MAX_LEN.
4022	 */
4023	if (map->m_len > EXT_INIT_MAX_LEN &&
4024	    !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4025		map->m_len = EXT_INIT_MAX_LEN;
4026	else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4027		 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4028		map->m_len = EXT_UNINIT_MAX_LEN;
4029
4030	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4031	newex.ee_len = cpu_to_le16(map->m_len);
4032	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4033	if (err)
4034		allocated = ext4_ext_get_actual_len(&newex);
4035	else
4036		allocated = map->m_len;
4037
4038	/* allocate new block */
4039	ar.inode = inode;
4040	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4041	ar.logical = map->m_lblk;
4042	/*
4043	 * We calculate the offset from the beginning of the cluster
4044	 * for the logical block number, since when we allocate a
4045	 * physical cluster, the physical block should start at the
4046	 * same offset from the beginning of the cluster.  This is
4047	 * needed so that future calls to get_implied_cluster_alloc()
4048	 * work correctly.
4049	 */
4050	offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4051	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4052	ar.goal -= offset;
4053	ar.logical -= offset;
4054	if (S_ISREG(inode->i_mode))
4055		ar.flags = EXT4_MB_HINT_DATA;
4056	else
4057		/* disable in-core preallocation for non-regular files */
4058		ar.flags = 0;
4059	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4060		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
 
 
 
 
4061	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4062	if (!newblock)
4063		goto out2;
4064	ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4065		  ar.goal, newblock, allocated);
4066	free_on_err = 1;
4067	allocated_clusters = ar.len;
4068	ar.len = EXT4_C2B(sbi, ar.len) - offset;
 
 
4069	if (ar.len > allocated)
4070		ar.len = allocated;
4071
4072got_allocated_blocks:
4073	/* try to insert new extent into found leaf and return */
4074	ext4_ext_store_pblock(&newex, newblock + offset);
 
4075	newex.ee_len = cpu_to_le16(ar.len);
4076	/* Mark uninitialized */
4077	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4078		ext4_ext_mark_uninitialized(&newex);
4079		/*
4080		 * io_end structure was created for every IO write to an
4081		 * uninitialized extent. To avoid unnecessary conversion,
4082		 * here we flag the IO that really needs the conversion.
4083		 * For non asycn direct IO case, flag the inode state
4084		 * that we need to perform conversion when IO is done.
4085		 */
4086		if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4087			if (io)
4088				ext4_set_io_unwritten_flag(inode, io);
4089			else
4090				ext4_set_inode_state(inode,
4091						     EXT4_STATE_DIO_UNWRITTEN);
4092		}
4093		if (ext4_should_dioread_nolock(inode))
4094			map->m_flags |= EXT4_MAP_UNINIT;
4095	}
4096
4097	err = 0;
4098	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4099		err = check_eofblocks_fl(handle, inode, map->m_lblk,
4100					 path, ar.len);
4101	if (!err)
4102		err = ext4_ext_insert_extent(handle, inode, path,
4103					     &newex, flags);
4104	if (err && free_on_err) {
4105		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4106			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4107		/* free data blocks we just allocated */
4108		/* not a good idea to call discard here directly,
4109		 * but otherwise we'd need to call it every free() */
4110		ext4_discard_preallocations(inode);
4111		ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4112				 ext4_ext_get_actual_len(&newex), fb_flags);
4113		goto out2;
 
4114	}
4115
4116	/* previous routine could use block we allocated */
4117	newblock = ext4_ext_pblock(&newex);
4118	allocated = ext4_ext_get_actual_len(&newex);
4119	if (allocated > map->m_len)
4120		allocated = map->m_len;
4121	map->m_flags |= EXT4_MAP_NEW;
4122
4123	/*
4124	 * Update reserved blocks/metadata blocks after successful
4125	 * block allocation which had been deferred till now.
 
 
4126	 */
4127	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4128		unsigned int reserved_clusters;
4129		/*
4130		 * Check how many clusters we had reserved this allocated range
4131		 */
4132		reserved_clusters = get_reserved_cluster_alloc(inode,
4133						map->m_lblk, allocated);
4134		if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4135			if (reserved_clusters) {
4136				/*
4137				 * We have clusters reserved for this range.
4138				 * But since we are not doing actual allocation
4139				 * and are simply using blocks from previously
4140				 * allocated cluster, we should release the
4141				 * reservation and not claim quota.
4142				 */
4143				ext4_da_update_reserve_space(inode,
4144						reserved_clusters, 0);
4145			}
4146		} else {
4147			BUG_ON(allocated_clusters < reserved_clusters);
4148			/* We will claim quota for all newly allocated blocks.*/
4149			ext4_da_update_reserve_space(inode, allocated_clusters,
4150							1);
4151			if (reserved_clusters < allocated_clusters) {
4152				struct ext4_inode_info *ei = EXT4_I(inode);
4153				int reservation = allocated_clusters -
4154						  reserved_clusters;
4155				/*
4156				 * It seems we claimed few clusters outside of
4157				 * the range of this allocation. We should give
4158				 * it back to the reservation pool. This can
4159				 * happen in the following case:
4160				 *
4161				 * * Suppose s_cluster_ratio is 4 (i.e., each
4162				 *   cluster has 4 blocks. Thus, the clusters
4163				 *   are [0-3],[4-7],[8-11]...
4164				 * * First comes delayed allocation write for
4165				 *   logical blocks 10 & 11. Since there were no
4166				 *   previous delayed allocated blocks in the
4167				 *   range [8-11], we would reserve 1 cluster
4168				 *   for this write.
4169				 * * Next comes write for logical blocks 3 to 8.
4170				 *   In this case, we will reserve 2 clusters
4171				 *   (for [0-3] and [4-7]; and not for [8-11] as
4172				 *   that range has a delayed allocated blocks.
4173				 *   Thus total reserved clusters now becomes 3.
4174				 * * Now, during the delayed allocation writeout
4175				 *   time, we will first write blocks [3-8] and
4176				 *   allocate 3 clusters for writing these
4177				 *   blocks. Also, we would claim all these
4178				 *   three clusters above.
4179				 * * Now when we come here to writeout the
4180				 *   blocks [10-11], we would expect to claim
4181				 *   the reservation of 1 cluster we had made
4182				 *   (and we would claim it since there are no
4183				 *   more delayed allocated blocks in the range
4184				 *   [8-11]. But our reserved cluster count had
4185				 *   already gone to 0.
4186				 *
4187				 *   Thus, at the step 4 above when we determine
4188				 *   that there are still some unwritten delayed
4189				 *   allocated blocks outside of our current
4190				 *   block range, we should increment the
4191				 *   reserved clusters count so that when the
4192				 *   remaining blocks finally gets written, we
4193				 *   could claim them.
4194				 */
4195				dquot_reserve_block(inode,
4196						EXT4_C2B(sbi, reservation));
4197				spin_lock(&ei->i_block_reservation_lock);
4198				ei->i_reserved_data_blocks += reservation;
4199				spin_unlock(&ei->i_block_reservation_lock);
4200			}
4201		}
4202	}
4203
4204	/*
4205	 * Cache the extent and update transaction to commit on fdatasync only
4206	 * when it is _not_ an uninitialized extent.
4207	 */
4208	if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4209		ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4210		ext4_update_inode_fsync_trans(handle, inode, 1);
4211	} else
4212		ext4_update_inode_fsync_trans(handle, inode, 0);
 
 
 
 
 
 
4213out:
4214	if (allocated > map->m_len)
4215		allocated = map->m_len;
4216	ext4_ext_show_leaf(inode, path);
4217	map->m_flags |= EXT4_MAP_MAPPED;
4218	map->m_pblk = newblock;
4219	map->m_len = allocated;
4220out2:
4221	if (path) {
4222		ext4_ext_drop_refs(path);
4223		kfree(path);
4224	}
4225
4226	trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4227		newblock, map->m_len, err ? err : allocated);
4228
 
 
4229	return err ? err : allocated;
4230}
4231
4232void ext4_ext_truncate(struct inode *inode)
4233{
4234	struct address_space *mapping = inode->i_mapping;
4235	struct super_block *sb = inode->i_sb;
4236	ext4_lblk_t last_block;
4237	handle_t *handle;
4238	loff_t page_len;
4239	int err = 0;
4240
4241	/*
4242	 * finish any pending end_io work so we won't run the risk of
4243	 * converting any truncated blocks to initialized later
4244	 */
4245	ext4_flush_completed_IO(inode);
4246
4247	/*
4248	 * probably first extent we're gonna free will be last in block
4249	 */
4250	err = ext4_writepage_trans_blocks(inode);
4251	handle = ext4_journal_start(inode, err);
4252	if (IS_ERR(handle))
4253		return;
4254
4255	if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4256		page_len = PAGE_CACHE_SIZE -
4257			(inode->i_size & (PAGE_CACHE_SIZE - 1));
4258
4259		err = ext4_discard_partial_page_buffers(handle,
4260			mapping, inode->i_size, page_len, 0);
4261
4262		if (err)
4263			goto out_stop;
4264	}
4265
4266	if (ext4_orphan_add(handle, inode))
4267		goto out_stop;
4268
4269	down_write(&EXT4_I(inode)->i_data_sem);
4270	ext4_ext_invalidate_cache(inode);
4271
4272	ext4_discard_preallocations(inode);
4273
4274	/*
4275	 * TODO: optimization is possible here.
4276	 * Probably we need not scan at all,
4277	 * because page truncation is enough.
4278	 */
4279
4280	/* we have to know where to truncate from in crash case */
4281	EXT4_I(inode)->i_disksize = inode->i_size;
4282	ext4_mark_inode_dirty(handle, inode);
 
 
4283
4284	last_block = (inode->i_size + sb->s_blocksize - 1)
4285			>> EXT4_BLOCK_SIZE_BITS(sb);
 
 
 
 
 
 
 
 
 
 
4286	err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
 
 
 
 
 
 
4287
4288	/* In a multi-transaction truncate, we only make the final
4289	 * transaction synchronous.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4290	 */
4291	if (IS_SYNC(inode))
4292		ext4_handle_sync(handle);
4293
4294	up_write(&EXT4_I(inode)->i_data_sem);
4295
4296out_stop:
4297	/*
4298	 * If this was a simple ftruncate() and the file will remain alive,
4299	 * then we need to clear up the orphan record which we created above.
4300	 * However, if this was a real unlink then we were called by
4301	 * ext4_delete_inode(), and we allow that function to clean up the
4302	 * orphan info for us.
4303	 */
4304	if (inode->i_nlink)
4305		ext4_orphan_del(handle, inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4306
4307	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4308	ext4_mark_inode_dirty(handle, inode);
4309	ext4_journal_stop(handle);
4310}
4311
4312static void ext4_falloc_update_inode(struct inode *inode,
4313				int mode, loff_t new_size, int update_ctime)
 
 
 
 
4314{
4315	struct timespec now;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4316
4317	if (update_ctime) {
4318		now = current_fs_time(inode->i_sb);
4319		if (!timespec_equal(&inode->i_ctime, &now))
4320			inode->i_ctime = now;
4321	}
4322	/*
4323	 * Update only when preallocation was requested beyond
4324	 * the file size.
4325	 */
4326	if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4327		if (new_size > i_size_read(inode))
4328			i_size_write(inode, new_size);
4329		if (new_size > EXT4_I(inode)->i_disksize)
4330			ext4_update_i_disksize(inode, new_size);
4331	} else {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4332		/*
4333		 * Mark that we allocate beyond EOF so the subsequent truncate
4334		 * can proceed even if the new size is the same as i_size.
4335		 */
4336		if (new_size > i_size_read(inode))
4337			ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4338	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4339
 
 
 
 
 
4340}
4341
4342/*
4343 * preallocate space for a file. This implements ext4's fallocate file
4344 * operation, which gets called from sys_fallocate system call.
4345 * For block-mapped files, posix_fallocate should fall back to the method
4346 * of writing zeroes to the required new blocks (the same behavior which is
4347 * expected for file systems which do not support fallocate() system call).
4348 */
4349long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4350{
4351	struct inode *inode = file->f_path.dentry->d_inode;
4352	handle_t *handle;
4353	loff_t new_size;
4354	unsigned int max_blocks;
4355	int ret = 0;
4356	int ret2 = 0;
4357	int retries = 0;
4358	int flags;
4359	struct ext4_map_blocks map;
4360	unsigned int credits, blkbits = inode->i_blkbits;
4361
4362	/*
4363	 * currently supporting (pre)allocate mode for extent-based
4364	 * files _only_
 
 
4365	 */
4366	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 
4367		return -EOPNOTSUPP;
4368
4369	/* Return error if mode is not supported */
4370	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 
 
4371		return -EOPNOTSUPP;
4372
4373	if (mode & FALLOC_FL_PUNCH_HOLE)
4374		return ext4_punch_hole(file, offset, len);
 
 
 
4375
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4376	trace_ext4_fallocate_enter(inode, offset, len, mode);
4377	map.m_lblk = offset >> blkbits;
 
 
 
 
 
 
4378	/*
4379	 * We can't just convert len to max_blocks because
4380	 * If blocksize = 4096 offset = 3072 and len = 2048
4381	 */
4382	max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4383		- map.m_lblk;
4384	/*
4385	 * credits to insert 1 extent into extent tree
4386	 */
4387	credits = ext4_chunk_trans_blocks(inode, max_blocks);
4388	mutex_lock(&inode->i_mutex);
4389	ret = inode_newsize_ok(inode, (len + offset));
4390	if (ret) {
4391		mutex_unlock(&inode->i_mutex);
4392		trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4393		return ret;
4394	}
4395	flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4396	if (mode & FALLOC_FL_KEEP_SIZE)
4397		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4398	/*
4399	 * Don't normalize the request if it can fit in one extent so
4400	 * that it doesn't get unnecessarily split into multiple
4401	 * extents.
4402	 */
4403	if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4404		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4405retry:
4406	while (ret >= 0 && ret < max_blocks) {
4407		map.m_lblk = map.m_lblk + ret;
4408		map.m_len = max_blocks = max_blocks - ret;
4409		handle = ext4_journal_start(inode, credits);
4410		if (IS_ERR(handle)) {
4411			ret = PTR_ERR(handle);
4412			break;
4413		}
4414		ret = ext4_map_blocks(handle, inode, &map, flags);
4415		if (ret <= 0) {
4416#ifdef EXT4FS_DEBUG
4417			WARN_ON(ret <= 0);
4418			printk(KERN_ERR "%s: ext4_ext_map_blocks "
4419				    "returned error inode#%lu, block=%u, "
4420				    "max_blocks=%u", __func__,
4421				    inode->i_ino, map.m_lblk, max_blocks);
4422#endif
4423			ext4_mark_inode_dirty(handle, inode);
4424			ret2 = ext4_journal_stop(handle);
4425			break;
4426		}
4427		if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4428						blkbits) >> blkbits))
4429			new_size = offset + len;
4430		else
4431			new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4432
4433		ext4_falloc_update_inode(inode, mode, new_size,
4434					 (map.m_flags & EXT4_MAP_NEW));
4435		ext4_mark_inode_dirty(handle, inode);
4436		ret2 = ext4_journal_stop(handle);
4437		if (ret2)
4438			break;
 
4439	}
4440	if (ret == -ENOSPC &&
4441			ext4_should_retry_alloc(inode->i_sb, &retries)) {
4442		ret = 0;
4443		goto retry;
 
 
 
 
 
 
 
 
 
 
 
4444	}
4445	mutex_unlock(&inode->i_mutex);
4446	trace_ext4_fallocate_exit(inode, offset, max_blocks,
4447				ret > 0 ? ret2 : ret);
4448	return ret > 0 ? ret2 : ret;
 
4449}
4450
4451/*
4452 * This function convert a range of blocks to written extents
4453 * The caller of this function will pass the start offset and the size.
4454 * all unwritten extents within this range will be converted to
4455 * written extents.
4456 *
4457 * This function is called from the direct IO end io call back
4458 * function, to convert the fallocated extents after IO is completed.
4459 * Returns 0 on success.
4460 */
4461int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4462				    ssize_t len)
4463{
4464	handle_t *handle;
4465	unsigned int max_blocks;
4466	int ret = 0;
4467	int ret2 = 0;
4468	struct ext4_map_blocks map;
4469	unsigned int credits, blkbits = inode->i_blkbits;
 
4470
4471	map.m_lblk = offset >> blkbits;
4472	/*
4473	 * We can't just convert len to max_blocks because
4474	 * If blocksize = 4096 offset = 3072 and len = 2048
4475	 */
4476	max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4477		      map.m_lblk);
4478	/*
4479	 * credits to insert 1 extent into extent tree
4480	 */
4481	credits = ext4_chunk_trans_blocks(inode, max_blocks);
4482	while (ret >= 0 && ret < max_blocks) {
4483		map.m_lblk += ret;
4484		map.m_len = (max_blocks -= ret);
4485		handle = ext4_journal_start(inode, credits);
4486		if (IS_ERR(handle)) {
4487			ret = PTR_ERR(handle);
4488			break;
 
 
 
4489		}
4490		ret = ext4_map_blocks(handle, inode, &map,
4491				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4492		if (ret <= 0) {
4493			WARN_ON(ret <= 0);
4494			ext4_msg(inode->i_sb, KERN_ERR,
4495				 "%s:%d: inode #%lu: block %u: len %u: "
4496				 "ext4_ext_map_blocks returned %d",
4497				 __func__, __LINE__, inode->i_ino, map.m_lblk,
4498				 map.m_len, ret);
4499		}
4500		ext4_mark_inode_dirty(handle, inode);
4501		ret2 = ext4_journal_stop(handle);
4502		if (ret <= 0 || ret2 )
 
 
 
4503			break;
4504	}
4505	return ret > 0 ? ret2 : ret;
4506}
4507
4508/*
4509 * Callback function called for each extent to gather FIEMAP information.
4510 */
4511static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4512		       struct ext4_ext_cache *newex, struct ext4_extent *ex,
4513		       void *data)
4514{
4515	__u64	logical;
4516	__u64	physical;
4517	__u64	length;
4518	__u32	flags = 0;
4519	int		ret = 0;
4520	struct fiemap_extent_info *fieinfo = data;
4521	unsigned char blksize_bits;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4522
4523	blksize_bits = inode->i_sb->s_blocksize_bits;
4524	logical = (__u64)newex->ec_block << blksize_bits;
 
 
 
 
 
4525
4526	if (newex->ec_start == 0) {
4527		/*
4528		 * No extent in extent-tree contains block @newex->ec_start,
4529		 * then the block may stay in 1)a hole or 2)delayed-extent.
4530		 *
4531		 * Holes or delayed-extents are processed as follows.
4532		 * 1. lookup dirty pages with specified range in pagecache.
4533		 *    If no page is got, then there is no delayed-extent and
4534		 *    return with EXT_CONTINUE.
4535		 * 2. find the 1st mapped buffer,
4536		 * 3. check if the mapped buffer is both in the request range
4537		 *    and a delayed buffer. If not, there is no delayed-extent,
4538		 *    then return.
4539		 * 4. a delayed-extent is found, the extent will be collected.
4540		 */
4541		ext4_lblk_t	end = 0;
4542		pgoff_t		last_offset;
4543		pgoff_t		offset;
4544		pgoff_t		index;
4545		pgoff_t		start_index = 0;
4546		struct page	**pages = NULL;
4547		struct buffer_head *bh = NULL;
4548		struct buffer_head *head = NULL;
4549		unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4550
4551		pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4552		if (pages == NULL)
4553			return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4554
4555		offset = logical >> PAGE_SHIFT;
4556repeat:
4557		last_offset = offset;
4558		head = NULL;
4559		ret = find_get_pages_tag(inode->i_mapping, &offset,
4560					PAGECACHE_TAG_DIRTY, nr_pages, pages);
4561
4562		if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4563			/* First time, try to find a mapped buffer. */
4564			if (ret == 0) {
4565out:
4566				for (index = 0; index < ret; index++)
4567					page_cache_release(pages[index]);
4568				/* just a hole. */
4569				kfree(pages);
4570				return EXT_CONTINUE;
4571			}
4572			index = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4573
4574next_page:
4575			/* Try to find the 1st mapped buffer. */
4576			end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4577				  blksize_bits;
4578			if (!page_has_buffers(pages[index]))
4579				goto out;
4580			head = page_buffers(pages[index]);
4581			if (!head)
4582				goto out;
4583
4584			index++;
4585			bh = head;
4586			do {
4587				if (end >= newex->ec_block +
4588					newex->ec_len)
4589					/* The buffer is out of
4590					 * the request range.
4591					 */
4592					goto out;
4593
4594				if (buffer_mapped(bh) &&
4595				    end >= newex->ec_block) {
4596					start_index = index - 1;
4597					/* get the 1st mapped buffer. */
4598					goto found_mapped_buffer;
4599				}
4600
4601				bh = bh->b_this_page;
4602				end++;
4603			} while (bh != head);
 
 
 
 
 
 
 
 
 
 
 
4604
4605			/* No mapped buffer in the range found in this page,
4606			 * We need to look up next page.
4607			 */
4608			if (index >= ret) {
4609				/* There is no page left, but we need to limit
4610				 * newex->ec_len.
4611				 */
4612				newex->ec_len = end - newex->ec_block;
4613				goto out;
4614			}
4615			goto next_page;
4616		} else {
4617			/*Find contiguous delayed buffers. */
4618			if (ret > 0 && pages[0]->index == last_offset)
4619				head = page_buffers(pages[0]);
4620			bh = head;
4621			index = 1;
4622			start_index = 0;
4623		}
4624
4625found_mapped_buffer:
4626		if (bh != NULL && buffer_delay(bh)) {
4627			/* 1st or contiguous delayed buffer found. */
4628			if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4629				/*
4630				 * 1st delayed buffer found, record
4631				 * the start of extent.
4632				 */
4633				flags |= FIEMAP_EXTENT_DELALLOC;
4634				newex->ec_block = end;
4635				logical = (__u64)end << blksize_bits;
4636			}
4637			/* Find contiguous delayed buffers. */
4638			do {
4639				if (!buffer_delay(bh))
4640					goto found_delayed_extent;
4641				bh = bh->b_this_page;
4642				end++;
4643			} while (bh != head);
4644
4645			for (; index < ret; index++) {
4646				if (!page_has_buffers(pages[index])) {
4647					bh = NULL;
4648					break;
4649				}
4650				head = page_buffers(pages[index]);
4651				if (!head) {
4652					bh = NULL;
4653					break;
4654				}
4655
4656				if (pages[index]->index !=
4657				    pages[start_index]->index + index
4658				    - start_index) {
4659					/* Blocks are not contiguous. */
4660					bh = NULL;
4661					break;
 
 
 
 
 
 
 
 
 
 
 
 
4662				}
4663				bh = head;
4664				do {
4665					if (!buffer_delay(bh))
4666						/* Delayed-extent ends. */
4667						goto found_delayed_extent;
4668					bh = bh->b_this_page;
4669					end++;
4670				} while (bh != head);
4671			}
4672		} else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4673			/* a hole found. */
4674			goto out;
4675
4676found_delayed_extent:
4677		newex->ec_len = min(end - newex->ec_block,
4678						(ext4_lblk_t)EXT_INIT_MAX_LEN);
4679		if (ret == nr_pages && bh != NULL &&
4680			newex->ec_len < EXT_INIT_MAX_LEN &&
4681			buffer_delay(bh)) {
4682			/* Have not collected an extent and continue. */
4683			for (index = 0; index < ret; index++)
4684				page_cache_release(pages[index]);
4685			goto repeat;
4686		}
4687
4688		for (index = 0; index < ret; index++)
4689			page_cache_release(pages[index]);
4690		kfree(pages);
4691	}
4692
4693	physical = (__u64)newex->ec_start << blksize_bits;
4694	length =   (__u64)newex->ec_len << blksize_bits;
 
 
 
 
 
4695
4696	if (ex && ext4_ext_is_uninitialized(ex))
4697		flags |= FIEMAP_EXTENT_UNWRITTEN;
 
4698
4699	if (next == EXT_MAX_BLOCKS)
4700		flags |= FIEMAP_EXTENT_LAST;
4701
4702	ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4703					length, flags);
4704	if (ret < 0)
4705		return ret;
4706	if (ret == 1)
4707		return EXT_BREAK;
4708	return EXT_CONTINUE;
4709}
4710/* fiemap flags we can handle specified here */
4711#define EXT4_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4712
4713static int ext4_xattr_fiemap(struct inode *inode,
4714				struct fiemap_extent_info *fieinfo)
4715{
4716	__u64 physical = 0;
4717	__u64 length;
4718	__u32 flags = FIEMAP_EXTENT_LAST;
4719	int blockbits = inode->i_sb->s_blocksize_bits;
4720	int error = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4721
4722	/* in-inode? */
4723	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4724		struct ext4_iloc iloc;
4725		int offset;	/* offset of xattr in inode */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4726
4727		error = ext4_get_inode_loc(inode, &iloc);
4728		if (error)
4729			return error;
4730		physical = iloc.bh->b_blocknr << blockbits;
4731		offset = EXT4_GOOD_OLD_INODE_SIZE +
4732				EXT4_I(inode)->i_extra_isize;
4733		physical += offset;
4734		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4735		flags |= FIEMAP_EXTENT_DATA_INLINE;
4736		brelse(iloc.bh);
4737	} else { /* external block */
4738		physical = EXT4_I(inode)->i_file_acl << blockbits;
4739		length = inode->i_sb->s_blocksize;
4740	}
4741
4742	if (physical)
4743		error = fiemap_fill_next_extent(fieinfo, 0, physical,
4744						length, flags);
4745	return (error < 0 ? error : 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4746}
4747
4748/*
4749 * ext4_ext_punch_hole
4750 *
4751 * Punches a hole of "length" bytes in a file starting
4752 * at byte "offset"
4753 *
4754 * @inode:  The inode of the file to punch a hole in
4755 * @offset: The starting byte offset of the hole
4756 * @length: The length of the hole
4757 *
4758 * Returns the number of blocks removed or negative on err
4759 */
4760int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4761{
4762	struct inode *inode = file->f_path.dentry->d_inode;
4763	struct super_block *sb = inode->i_sb;
4764	ext4_lblk_t first_block, stop_block;
4765	struct address_space *mapping = inode->i_mapping;
 
4766	handle_t *handle;
4767	loff_t first_page, last_page, page_len;
4768	loff_t first_page_offset, last_page_offset;
4769	int credits, err = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4770
4771	/* No need to punch hole beyond i_size */
4772	if (offset >= inode->i_size)
4773		return 0;
 
 
 
 
 
 
 
 
 
 
4774
4775	/*
4776	 * If the hole extends beyond i_size, set the hole
4777	 * to end after the page that contains i_size
 
 
 
 
 
 
 
 
 
 
 
 
 
4778	 */
4779	if (offset + length > inode->i_size) {
4780		length = inode->i_size +
4781		   PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4782		   offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4783	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4784
4785	first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4786	last_page = (offset + length) >> PAGE_CACHE_SHIFT;
 
 
 
 
 
 
4787
4788	first_page_offset = first_page << PAGE_CACHE_SHIFT;
4789	last_page_offset = last_page << PAGE_CACHE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4790
4791	/*
4792	 * Write out all dirty pages to avoid race conditions
4793	 * Then release them.
 
4794	 */
4795	if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4796		err = filemap_write_and_wait_range(mapping,
4797			offset, offset + length - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4798
4799		if (err)
4800			return err;
 
 
4801	}
4802
4803	/* Now release the pages */
4804	if (last_page_offset > first_page_offset) {
4805		truncate_pagecache_range(inode, first_page_offset,
4806					 last_page_offset - 1);
4807	}
4808
4809	/* finish any pending end_io work */
4810	ext4_flush_completed_IO(inode);
4811
4812	credits = ext4_writepage_trans_blocks(inode);
4813	handle = ext4_journal_start(inode, credits);
4814	if (IS_ERR(handle))
4815		return PTR_ERR(handle);
 
 
 
 
 
4816
4817	err = ext4_orphan_add(handle, inode);
4818	if (err)
4819		goto out;
4820
4821	/*
4822	 * Now we need to zero out the non-page-aligned data in the
4823	 * pages at the start and tail of the hole, and unmap the buffer
4824	 * heads for the block aligned regions of the page that were
4825	 * completely zeroed.
4826	 */
4827	if (first_page > last_page) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4828		/*
4829		 * If the file space being truncated is contained within a page
4830		 * just zero out and unmap the middle of that page
4831		 */
4832		err = ext4_discard_partial_page_buffers(handle,
4833			mapping, offset, length, 0);
 
 
 
 
 
 
 
 
 
4834
4835		if (err)
4836			goto out;
 
 
 
4837	} else {
4838		/*
4839		 * zero out and unmap the partial page that contains
4840		 * the start of the hole
4841		 */
4842		page_len  = first_page_offset - offset;
4843		if (page_len > 0) {
4844			err = ext4_discard_partial_page_buffers(handle, mapping,
4845						   offset, page_len, 0);
4846			if (err)
4847				goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4848		}
4849
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4850		/*
4851		 * zero out and unmap the partial page that contains
4852		 * the end of the hole
 
 
4853		 */
4854		page_len = offset + length - last_page_offset;
4855		if (page_len > 0) {
4856			err = ext4_discard_partial_page_buffers(handle, mapping,
4857					last_page_offset, page_len, 0);
4858			if (err)
4859				goto out;
4860		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4861	}
4862
 
 
4863	/*
4864	 * If i_size is contained in the last page, we need to
4865	 * unmap and zero the partial page after i_size
 
4866	 */
4867	if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4868	   inode->i_size % PAGE_CACHE_SIZE != 0) {
 
 
 
 
 
 
 
 
4869
4870		page_len = PAGE_CACHE_SIZE -
4871			(inode->i_size & (PAGE_CACHE_SIZE - 1));
 
4872
4873		if (page_len > 0) {
4874			err = ext4_discard_partial_page_buffers(handle,
4875			  mapping, inode->i_size, page_len, 0);
4876
4877			if (err)
4878				goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4879		}
4880	}
4881
4882	first_block = (offset + sb->s_blocksize - 1) >>
4883		EXT4_BLOCK_SIZE_BITS(sb);
4884	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
 
 
4885
4886	/* If there are no blocks to remove, return now */
4887	if (first_block >= stop_block)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4888		goto out;
 
4889
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4890	down_write(&EXT4_I(inode)->i_data_sem);
4891	ext4_ext_invalidate_cache(inode);
4892	ext4_discard_preallocations(inode);
 
 
 
 
 
4893
4894	err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
 
 
 
 
 
4895
4896	ext4_ext_invalidate_cache(inode);
4897	ext4_discard_preallocations(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4898
4899	if (IS_SYNC(inode))
4900		ext4_handle_sync(handle);
 
 
 
4901
4902	up_write(&EXT4_I(inode)->i_data_sem);
 
4903
4904out:
4905	ext4_orphan_del(handle, inode);
4906	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4907	ext4_mark_inode_dirty(handle, inode);
4908	ext4_journal_stop(handle);
4909	return err;
 
4910}
4911int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4912		__u64 start, __u64 len)
 
4913{
4914	ext4_lblk_t start_blk;
4915	int error = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4916
4917	/* fallback to generic here if not in extents fmt */
4918	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4919		return generic_block_fiemap(inode, fieinfo, start, len,
4920			ext4_get_block);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4921
4922	if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4923		return -EBADR;
 
 
 
4924
4925	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4926		error = ext4_xattr_fiemap(inode, fieinfo);
4927	} else {
4928		ext4_lblk_t len_blks;
4929		__u64 last_blk;
 
 
4930
4931		start_blk = start >> inode->i_sb->s_blocksize_bits;
4932		last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4933		if (last_blk >= EXT_MAX_BLOCKS)
4934			last_blk = EXT_MAX_BLOCKS-1;
4935		len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4936
4937		/*
4938		 * Walk the extent tree gathering extent information.
4939		 * ext4_ext_fiemap_cb will push extents back to user.
4940		 */
4941		error = ext4_ext_walk_space(inode, start_blk, len_blks,
4942					  ext4_ext_fiemap_cb, fieinfo);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4943	}
4944
4945	return error;
4946}