Linux Audio

Check our new training course

Loading...
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}
v3.1
   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/module.h>
  33#include <linux/fs.h>
  34#include <linux/time.h>
  35#include <linux/jbd2.h>
  36#include <linux/highuid.h>
  37#include <linux/pagemap.h>
  38#include <linux/quotaops.h>
  39#include <linux/string.h>
  40#include <linux/slab.h>
  41#include <linux/falloc.h>
  42#include <asm/uaccess.h>
  43#include <linux/fiemap.h>
  44#include "ext4_jbd2.h"
  45#include "ext4_extents.h"
  46
  47#include <trace/events/ext4.h>
  48
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  49static int ext4_split_extent(handle_t *handle,
  50				struct inode *inode,
  51				struct ext4_ext_path *path,
  52				struct ext4_map_blocks *map,
  53				int split_flag,
  54				int flags);
  55
 
 
 
 
 
 
 
  56static int ext4_ext_truncate_extend_restart(handle_t *handle,
  57					    struct inode *inode,
  58					    int needed)
  59{
  60	int err;
  61
  62	if (!ext4_handle_valid(handle))
  63		return 0;
  64	if (handle->h_buffer_credits > needed)
  65		return 0;
  66	err = ext4_journal_extend(handle, needed);
  67	if (err <= 0)
  68		return err;
  69	err = ext4_truncate_restart_trans(handle, inode, needed);
  70	if (err == 0)
  71		err = -EAGAIN;
  72
  73	return err;
  74}
  75
  76/*
  77 * could return:
  78 *  - EROFS
  79 *  - ENOMEM
  80 */
  81static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  82				struct ext4_ext_path *path)
  83{
  84	if (path->p_bh) {
  85		/* path points to block */
  86		return ext4_journal_get_write_access(handle, path->p_bh);
  87	}
  88	/* path points to leaf/index in inode body */
  89	/* we use in-core data, no need to protect them */
  90	return 0;
  91}
  92
  93/*
  94 * could return:
  95 *  - EROFS
  96 *  - ENOMEM
  97 *  - EIO
  98 */
  99static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
 100				struct ext4_ext_path *path)
 
 
 
 101{
 102	int err;
 103	if (path->p_bh) {
 
 104		/* path points to block */
 105		err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
 
 106	} else {
 107		/* path points to leaf/index in inode body */
 108		err = ext4_mark_inode_dirty(handle, inode);
 109	}
 110	return err;
 111}
 112
 113static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 114			      struct ext4_ext_path *path,
 115			      ext4_lblk_t block)
 116{
 117	int depth;
 118
 119	if (path) {
 
 120		struct ext4_extent *ex;
 121		depth = path->p_depth;
 122
 123		/*
 124		 * Try to predict block placement assuming that we are
 125		 * filling in a file which will eventually be
 126		 * non-sparse --- i.e., in the case of libbfd writing
 127		 * an ELF object sections out-of-order but in a way
 128		 * the eventually results in a contiguous object or
 129		 * executable file, or some database extending a table
 130		 * space file.  However, this is actually somewhat
 131		 * non-ideal if we are writing a sparse file such as
 132		 * qemu or KVM writing a raw image file that is going
 133		 * to stay fairly sparse, since it will end up
 134		 * fragmenting the file system's free space.  Maybe we
 135		 * should have some hueristics or some way to allow
 136		 * userspace to pass a hint to file system,
 137		 * especially if the latter case turns out to be
 138		 * common.
 139		 */
 140		ex = path[depth].p_ext;
 141		if (ex) {
 142			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 143			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 144
 145			if (block > ext_block)
 146				return ext_pblk + (block - ext_block);
 147			else
 148				return ext_pblk - (ext_block - block);
 149		}
 150
 151		/* it looks like index is empty;
 152		 * try to find starting block from index itself */
 153		if (path[depth].p_bh)
 154			return path[depth].p_bh->b_blocknr;
 155	}
 156
 157	/* OK. use inode's group */
 158	return ext4_inode_to_goal_block(inode);
 159}
 160
 161/*
 162 * Allocation for a meta data block
 163 */
 164static ext4_fsblk_t
 165ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 166			struct ext4_ext_path *path,
 167			struct ext4_extent *ex, int *err, unsigned int flags)
 168{
 169	ext4_fsblk_t goal, newblock;
 170
 171	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 172	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 173					NULL, err);
 174	return newblock;
 175}
 176
 177static inline int ext4_ext_space_block(struct inode *inode, int check)
 178{
 179	int size;
 180
 181	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 182			/ sizeof(struct ext4_extent);
 183	if (!check) {
 184#ifdef AGGRESSIVE_TEST
 185		if (size > 6)
 186			size = 6;
 187#endif
 188	}
 189	return size;
 190}
 191
 192static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 193{
 194	int size;
 195
 196	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 197			/ sizeof(struct ext4_extent_idx);
 198	if (!check) {
 199#ifdef AGGRESSIVE_TEST
 200		if (size > 5)
 201			size = 5;
 202#endif
 203	}
 204	return size;
 205}
 206
 207static inline int ext4_ext_space_root(struct inode *inode, int check)
 208{
 209	int size;
 210
 211	size = sizeof(EXT4_I(inode)->i_data);
 212	size -= sizeof(struct ext4_extent_header);
 213	size /= sizeof(struct ext4_extent);
 214	if (!check) {
 215#ifdef AGGRESSIVE_TEST
 216		if (size > 3)
 217			size = 3;
 218#endif
 219	}
 220	return size;
 221}
 222
 223static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 224{
 225	int size;
 226
 227	size = sizeof(EXT4_I(inode)->i_data);
 228	size -= sizeof(struct ext4_extent_header);
 229	size /= sizeof(struct ext4_extent_idx);
 230	if (!check) {
 231#ifdef AGGRESSIVE_TEST
 232		if (size > 4)
 233			size = 4;
 234#endif
 235	}
 236	return size;
 237}
 238
 239/*
 240 * Calculate the number of metadata blocks needed
 241 * to allocate @blocks
 242 * Worse case is one block per extent
 243 */
 244int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 245{
 246	struct ext4_inode_info *ei = EXT4_I(inode);
 247	int idxs, num = 0;
 248
 249	idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 250		/ sizeof(struct ext4_extent_idx));
 251
 252	/*
 253	 * If the new delayed allocation block is contiguous with the
 254	 * previous da block, it can share index blocks with the
 255	 * previous block, so we only need to allocate a new index
 256	 * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 257	 * an additional index block, and at ldxs**3 blocks, yet
 258	 * another index blocks.
 259	 */
 260	if (ei->i_da_metadata_calc_len &&
 261	    ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 
 
 262		if ((ei->i_da_metadata_calc_len % idxs) == 0)
 263			num++;
 264		if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 265			num++;
 266		if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 267			num++;
 268			ei->i_da_metadata_calc_len = 0;
 269		} else
 270			ei->i_da_metadata_calc_len++;
 271		ei->i_da_metadata_calc_last_lblock++;
 272		return num;
 273	}
 274
 275	/*
 276	 * In the worst case we need a new set of index blocks at
 277	 * every level of the inode's extent tree.
 278	 */
 279	ei->i_da_metadata_calc_len = 1;
 280	ei->i_da_metadata_calc_last_lblock = lblock;
 281	return ext_depth(inode) + 1;
 282}
 283
 284static int
 285ext4_ext_max_entries(struct inode *inode, int depth)
 286{
 287	int max;
 288
 289	if (depth == ext_depth(inode)) {
 290		if (depth == 0)
 291			max = ext4_ext_space_root(inode, 1);
 292		else
 293			max = ext4_ext_space_root_idx(inode, 1);
 294	} else {
 295		if (depth == 0)
 296			max = ext4_ext_space_block(inode, 1);
 297		else
 298			max = ext4_ext_space_block_idx(inode, 1);
 299	}
 300
 301	return max;
 302}
 303
 304static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 305{
 306	ext4_fsblk_t block = ext4_ext_pblock(ext);
 307	int len = ext4_ext_get_actual_len(ext);
 308
 
 
 309	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 310}
 311
 312static int ext4_valid_extent_idx(struct inode *inode,
 313				struct ext4_extent_idx *ext_idx)
 314{
 315	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 316
 317	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 318}
 319
 320static int ext4_valid_extent_entries(struct inode *inode,
 321				struct ext4_extent_header *eh,
 322				int depth)
 323{
 324	struct ext4_extent *ext;
 325	struct ext4_extent_idx *ext_idx;
 326	unsigned short entries;
 327	if (eh->eh_entries == 0)
 328		return 1;
 329
 330	entries = le16_to_cpu(eh->eh_entries);
 331
 332	if (depth == 0) {
 333		/* leaf entries */
 334		ext = EXT_FIRST_EXTENT(eh);
 335		while (entries) {
 336			if (!ext4_valid_extent(inode, ext))
 337				return 0;
 338			ext++;
 339			entries--;
 340		}
 341	} else {
 342		ext_idx = EXT_FIRST_INDEX(eh);
 343		while (entries) {
 344			if (!ext4_valid_extent_idx(inode, ext_idx))
 345				return 0;
 346			ext_idx++;
 347			entries--;
 348		}
 349	}
 350	return 1;
 351}
 352
 353static int __ext4_ext_check(const char *function, unsigned int line,
 354			    struct inode *inode, struct ext4_extent_header *eh,
 355			    int depth)
 356{
 357	const char *error_msg;
 358	int max = 0;
 359
 360	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 361		error_msg = "invalid magic";
 362		goto corrupted;
 363	}
 364	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 365		error_msg = "unexpected eh_depth";
 366		goto corrupted;
 367	}
 368	if (unlikely(eh->eh_max == 0)) {
 369		error_msg = "invalid eh_max";
 370		goto corrupted;
 371	}
 372	max = ext4_ext_max_entries(inode, depth);
 373	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 374		error_msg = "too large eh_max";
 375		goto corrupted;
 376	}
 377	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 378		error_msg = "invalid eh_entries";
 379		goto corrupted;
 380	}
 381	if (!ext4_valid_extent_entries(inode, eh, depth)) {
 382		error_msg = "invalid extent entries";
 383		goto corrupted;
 384	}
 
 
 
 
 
 
 385	return 0;
 386
 387corrupted:
 388	ext4_error_inode(inode, function, line, 0,
 389			"bad header/extent: %s - magic %x, "
 390			"entries %u, max %u(%u), depth %u(%u)",
 391			error_msg, le16_to_cpu(eh->eh_magic),
 392			le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 393			max, le16_to_cpu(eh->eh_depth), depth);
 394
 395	return -EIO;
 396}
 397
 398#define ext4_ext_check(inode, eh, depth)	\
 399	__ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 400
 401int ext4_ext_check_inode(struct inode *inode)
 402{
 403	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 404}
 405
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 406#ifdef EXT_DEBUG
 407static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 408{
 409	int k, l = path->p_depth;
 410
 411	ext_debug("path:");
 412	for (k = 0; k <= l; k++, path++) {
 413		if (path->p_idx) {
 414		  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 415			    ext4_idx_pblock(path->p_idx));
 416		} else if (path->p_ext) {
 417			ext_debug("  %d:[%d]%d:%llu ",
 418				  le32_to_cpu(path->p_ext->ee_block),
 419				  ext4_ext_is_uninitialized(path->p_ext),
 420				  ext4_ext_get_actual_len(path->p_ext),
 421				  ext4_ext_pblock(path->p_ext));
 422		} else
 423			ext_debug("  []");
 424	}
 425	ext_debug("\n");
 426}
 427
 428static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 429{
 430	int depth = ext_depth(inode);
 431	struct ext4_extent_header *eh;
 432	struct ext4_extent *ex;
 433	int i;
 434
 435	if (!path)
 436		return;
 437
 438	eh = path[depth].p_hdr;
 439	ex = EXT_FIRST_EXTENT(eh);
 440
 441	ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 442
 443	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 444		ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 445			  ext4_ext_is_uninitialized(ex),
 446			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 447	}
 448	ext_debug("\n");
 449}
 450
 451static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 452			ext4_fsblk_t newblock, int level)
 453{
 454	int depth = ext_depth(inode);
 455	struct ext4_extent *ex;
 456
 457	if (depth != level) {
 458		struct ext4_extent_idx *idx;
 459		idx = path[level].p_idx;
 460		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 461			ext_debug("%d: move %d:%llu in new index %llu\n", level,
 462					le32_to_cpu(idx->ei_block),
 463					ext4_idx_pblock(idx),
 464					newblock);
 465			idx++;
 466		}
 467
 468		return;
 469	}
 470
 471	ex = path[depth].p_ext;
 472	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 473		ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 474				le32_to_cpu(ex->ee_block),
 475				ext4_ext_pblock(ex),
 476				ext4_ext_is_uninitialized(ex),
 477				ext4_ext_get_actual_len(ex),
 478				newblock);
 479		ex++;
 480	}
 481}
 482
 483#else
 484#define ext4_ext_show_path(inode, path)
 485#define ext4_ext_show_leaf(inode, path)
 486#define ext4_ext_show_move(inode, path, newblock, level)
 487#endif
 488
 489void ext4_ext_drop_refs(struct ext4_ext_path *path)
 490{
 491	int depth = path->p_depth;
 492	int i;
 493
 494	for (i = 0; i <= depth; i++, path++)
 495		if (path->p_bh) {
 496			brelse(path->p_bh);
 497			path->p_bh = NULL;
 498		}
 499}
 500
 501/*
 502 * ext4_ext_binsearch_idx:
 503 * binary search for the closest index of the given block
 504 * the header must be checked before calling this
 505 */
 506static void
 507ext4_ext_binsearch_idx(struct inode *inode,
 508			struct ext4_ext_path *path, ext4_lblk_t block)
 509{
 510	struct ext4_extent_header *eh = path->p_hdr;
 511	struct ext4_extent_idx *r, *l, *m;
 512
 513
 514	ext_debug("binsearch for %u(idx):  ", block);
 515
 516	l = EXT_FIRST_INDEX(eh) + 1;
 517	r = EXT_LAST_INDEX(eh);
 518	while (l <= r) {
 519		m = l + (r - l) / 2;
 520		if (block < le32_to_cpu(m->ei_block))
 521			r = m - 1;
 522		else
 523			l = m + 1;
 524		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 525				m, le32_to_cpu(m->ei_block),
 526				r, le32_to_cpu(r->ei_block));
 527	}
 528
 529	path->p_idx = l - 1;
 530	ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 531		  ext4_idx_pblock(path->p_idx));
 532
 533#ifdef CHECK_BINSEARCH
 534	{
 535		struct ext4_extent_idx *chix, *ix;
 536		int k;
 537
 538		chix = ix = EXT_FIRST_INDEX(eh);
 539		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 540		  if (k != 0 &&
 541		      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 542				printk(KERN_DEBUG "k=%d, ix=0x%p, "
 543				       "first=0x%p\n", k,
 544				       ix, EXT_FIRST_INDEX(eh));
 545				printk(KERN_DEBUG "%u <= %u\n",
 546				       le32_to_cpu(ix->ei_block),
 547				       le32_to_cpu(ix[-1].ei_block));
 548			}
 549			BUG_ON(k && le32_to_cpu(ix->ei_block)
 550					   <= le32_to_cpu(ix[-1].ei_block));
 551			if (block < le32_to_cpu(ix->ei_block))
 552				break;
 553			chix = ix;
 554		}
 555		BUG_ON(chix != path->p_idx);
 556	}
 557#endif
 558
 559}
 560
 561/*
 562 * ext4_ext_binsearch:
 563 * binary search for closest extent of the given block
 564 * the header must be checked before calling this
 565 */
 566static void
 567ext4_ext_binsearch(struct inode *inode,
 568		struct ext4_ext_path *path, ext4_lblk_t block)
 569{
 570	struct ext4_extent_header *eh = path->p_hdr;
 571	struct ext4_extent *r, *l, *m;
 572
 573	if (eh->eh_entries == 0) {
 574		/*
 575		 * this leaf is empty:
 576		 * we get such a leaf in split/add case
 577		 */
 578		return;
 579	}
 580
 581	ext_debug("binsearch for %u:  ", block);
 582
 583	l = EXT_FIRST_EXTENT(eh) + 1;
 584	r = EXT_LAST_EXTENT(eh);
 585
 586	while (l <= r) {
 587		m = l + (r - l) / 2;
 588		if (block < le32_to_cpu(m->ee_block))
 589			r = m - 1;
 590		else
 591			l = m + 1;
 592		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 593				m, le32_to_cpu(m->ee_block),
 594				r, le32_to_cpu(r->ee_block));
 595	}
 596
 597	path->p_ext = l - 1;
 598	ext_debug("  -> %d:%llu:[%d]%d ",
 599			le32_to_cpu(path->p_ext->ee_block),
 600			ext4_ext_pblock(path->p_ext),
 601			ext4_ext_is_uninitialized(path->p_ext),
 602			ext4_ext_get_actual_len(path->p_ext));
 603
 604#ifdef CHECK_BINSEARCH
 605	{
 606		struct ext4_extent *chex, *ex;
 607		int k;
 608
 609		chex = ex = EXT_FIRST_EXTENT(eh);
 610		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 611			BUG_ON(k && le32_to_cpu(ex->ee_block)
 612					  <= le32_to_cpu(ex[-1].ee_block));
 613			if (block < le32_to_cpu(ex->ee_block))
 614				break;
 615			chex = ex;
 616		}
 617		BUG_ON(chex != path->p_ext);
 618	}
 619#endif
 620
 621}
 622
 623int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 624{
 625	struct ext4_extent_header *eh;
 626
 627	eh = ext_inode_hdr(inode);
 628	eh->eh_depth = 0;
 629	eh->eh_entries = 0;
 630	eh->eh_magic = EXT4_EXT_MAGIC;
 631	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 632	ext4_mark_inode_dirty(handle, inode);
 633	ext4_ext_invalidate_cache(inode);
 634	return 0;
 635}
 636
 637struct ext4_ext_path *
 638ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 639					struct ext4_ext_path *path)
 640{
 641	struct ext4_extent_header *eh;
 642	struct buffer_head *bh;
 643	short int depth, i, ppos = 0, alloc = 0;
 644
 645	eh = ext_inode_hdr(inode);
 646	depth = ext_depth(inode);
 647
 648	/* account possible depth increase */
 649	if (!path) {
 650		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 651				GFP_NOFS);
 652		if (!path)
 653			return ERR_PTR(-ENOMEM);
 654		alloc = 1;
 655	}
 656	path[0].p_hdr = eh;
 657	path[0].p_bh = NULL;
 658
 659	i = depth;
 660	/* walk through the tree */
 661	while (i) {
 662		int need_to_validate = 0;
 663
 664		ext_debug("depth %d: num %d, max %d\n",
 665			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 666
 667		ext4_ext_binsearch_idx(inode, path + ppos, block);
 668		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 669		path[ppos].p_depth = i;
 670		path[ppos].p_ext = NULL;
 671
 672		bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 673		if (unlikely(!bh))
 674			goto err;
 675		if (!bh_uptodate_or_lock(bh)) {
 676			trace_ext4_ext_load_extent(inode, block,
 677						path[ppos].p_block);
 678			if (bh_submit_read(bh) < 0) {
 679				put_bh(bh);
 680				goto err;
 681			}
 682			/* validate the extent entries */
 683			need_to_validate = 1;
 684		}
 685		eh = ext_block_hdr(bh);
 686		ppos++;
 687		if (unlikely(ppos > depth)) {
 688			put_bh(bh);
 689			EXT4_ERROR_INODE(inode,
 690					 "ppos %d > depth %d", ppos, depth);
 691			goto err;
 692		}
 693		path[ppos].p_bh = bh;
 694		path[ppos].p_hdr = eh;
 695		i--;
 696
 697		if (need_to_validate && ext4_ext_check(inode, eh, i))
 698			goto err;
 699	}
 700
 701	path[ppos].p_depth = i;
 702	path[ppos].p_ext = NULL;
 703	path[ppos].p_idx = NULL;
 704
 705	/* find extent */
 706	ext4_ext_binsearch(inode, path + ppos, block);
 707	/* if not an empty leaf */
 708	if (path[ppos].p_ext)
 709		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 710
 711	ext4_ext_show_path(inode, path);
 712
 713	return path;
 714
 715err:
 716	ext4_ext_drop_refs(path);
 717	if (alloc)
 718		kfree(path);
 719	return ERR_PTR(-EIO);
 720}
 721
 722/*
 723 * ext4_ext_insert_index:
 724 * insert new index [@logical;@ptr] into the block at @curp;
 725 * check where to insert: before @curp or after @curp
 726 */
 727static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 728				 struct ext4_ext_path *curp,
 729				 int logical, ext4_fsblk_t ptr)
 730{
 731	struct ext4_extent_idx *ix;
 732	int len, err;
 733
 734	err = ext4_ext_get_access(handle, inode, curp);
 735	if (err)
 736		return err;
 737
 738	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 739		EXT4_ERROR_INODE(inode,
 740				 "logical %d == ei_block %d!",
 741				 logical, le32_to_cpu(curp->p_idx->ei_block));
 742		return -EIO;
 743	}
 744
 745	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 746			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
 747		EXT4_ERROR_INODE(inode,
 748				 "eh_entries %d >= eh_max %d!",
 749				 le16_to_cpu(curp->p_hdr->eh_entries),
 750				 le16_to_cpu(curp->p_hdr->eh_max));
 751		return -EIO;
 752	}
 753
 754	len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 755	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 756		/* insert after */
 757		if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 758			len = (len - 1) * sizeof(struct ext4_extent_idx);
 759			len = len < 0 ? 0 : len;
 760			ext_debug("insert new index %d after: %llu. "
 761					"move %d from 0x%p to 0x%p\n",
 762					logical, ptr, len,
 763					(curp->p_idx + 1), (curp->p_idx + 2));
 764			memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 765		}
 766		ix = curp->p_idx + 1;
 767	} else {
 768		/* insert before */
 769		len = len * sizeof(struct ext4_extent_idx);
 770		len = len < 0 ? 0 : len;
 771		ext_debug("insert new index %d before: %llu. "
 772				"move %d from 0x%p to 0x%p\n",
 773				logical, ptr, len,
 774				curp->p_idx, (curp->p_idx + 1));
 775		memmove(curp->p_idx + 1, curp->p_idx, len);
 776		ix = curp->p_idx;
 777	}
 778
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 779	ix->ei_block = cpu_to_le32(logical);
 780	ext4_idx_store_pblock(ix, ptr);
 781	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 782
 783	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 784		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 785		return -EIO;
 786	}
 787
 788	err = ext4_ext_dirty(handle, inode, curp);
 789	ext4_std_error(inode->i_sb, err);
 790
 791	return err;
 792}
 793
 794/*
 795 * ext4_ext_split:
 796 * inserts new subtree into the path, using free index entry
 797 * at depth @at:
 798 * - allocates all needed blocks (new leaf and all intermediate index blocks)
 799 * - makes decision where to split
 800 * - moves remaining extents and index entries (right to the split point)
 801 *   into the newly allocated blocks
 802 * - initializes subtree
 803 */
 804static int ext4_ext_split(handle_t *handle, struct inode *inode,
 805			  unsigned int flags,
 806			  struct ext4_ext_path *path,
 807			  struct ext4_extent *newext, int at)
 808{
 809	struct buffer_head *bh = NULL;
 810	int depth = ext_depth(inode);
 811	struct ext4_extent_header *neh;
 812	struct ext4_extent_idx *fidx;
 813	int i = at, k, m, a;
 814	ext4_fsblk_t newblock, oldblock;
 815	__le32 border;
 816	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 817	int err = 0;
 818
 819	/* make decision: where to split? */
 820	/* FIXME: now decision is simplest: at current extent */
 821
 822	/* if current leaf will be split, then we should use
 823	 * border from split point */
 824	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 825		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 826		return -EIO;
 827	}
 828	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 829		border = path[depth].p_ext[1].ee_block;
 830		ext_debug("leaf will be split."
 831				" next leaf starts at %d\n",
 832				  le32_to_cpu(border));
 833	} else {
 834		border = newext->ee_block;
 835		ext_debug("leaf will be added."
 836				" next leaf starts at %d\n",
 837				le32_to_cpu(border));
 838	}
 839
 840	/*
 841	 * If error occurs, then we break processing
 842	 * and mark filesystem read-only. index won't
 843	 * be inserted and tree will be in consistent
 844	 * state. Next mount will repair buffers too.
 845	 */
 846
 847	/*
 848	 * Get array to track all allocated blocks.
 849	 * We need this to handle errors and free blocks
 850	 * upon them.
 851	 */
 852	ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 853	if (!ablocks)
 854		return -ENOMEM;
 855
 856	/* allocate all needed blocks */
 857	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 858	for (a = 0; a < depth - at; a++) {
 859		newblock = ext4_ext_new_meta_block(handle, inode, path,
 860						   newext, &err, flags);
 861		if (newblock == 0)
 862			goto cleanup;
 863		ablocks[a] = newblock;
 864	}
 865
 866	/* initialize new leaf */
 867	newblock = ablocks[--a];
 868	if (unlikely(newblock == 0)) {
 869		EXT4_ERROR_INODE(inode, "newblock == 0!");
 870		err = -EIO;
 871		goto cleanup;
 872	}
 873	bh = sb_getblk(inode->i_sb, newblock);
 874	if (!bh) {
 875		err = -EIO;
 876		goto cleanup;
 877	}
 878	lock_buffer(bh);
 879
 880	err = ext4_journal_get_create_access(handle, bh);
 881	if (err)
 882		goto cleanup;
 883
 884	neh = ext_block_hdr(bh);
 885	neh->eh_entries = 0;
 886	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 887	neh->eh_magic = EXT4_EXT_MAGIC;
 888	neh->eh_depth = 0;
 889
 890	/* move remainder of path[depth] to the new leaf */
 891	if (unlikely(path[depth].p_hdr->eh_entries !=
 892		     path[depth].p_hdr->eh_max)) {
 893		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 894				 path[depth].p_hdr->eh_entries,
 895				 path[depth].p_hdr->eh_max);
 896		err = -EIO;
 897		goto cleanup;
 898	}
 899	/* start copy from next extent */
 900	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 901	ext4_ext_show_move(inode, path, newblock, depth);
 902	if (m) {
 903		struct ext4_extent *ex;
 904		ex = EXT_FIRST_EXTENT(neh);
 905		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 906		le16_add_cpu(&neh->eh_entries, m);
 907	}
 908
 
 909	set_buffer_uptodate(bh);
 910	unlock_buffer(bh);
 911
 912	err = ext4_handle_dirty_metadata(handle, inode, bh);
 913	if (err)
 914		goto cleanup;
 915	brelse(bh);
 916	bh = NULL;
 917
 918	/* correct old leaf */
 919	if (m) {
 920		err = ext4_ext_get_access(handle, inode, path + depth);
 921		if (err)
 922			goto cleanup;
 923		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 924		err = ext4_ext_dirty(handle, inode, path + depth);
 925		if (err)
 926			goto cleanup;
 927
 928	}
 929
 930	/* create intermediate indexes */
 931	k = depth - at - 1;
 932	if (unlikely(k < 0)) {
 933		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 934		err = -EIO;
 935		goto cleanup;
 936	}
 937	if (k)
 938		ext_debug("create %d intermediate indices\n", k);
 939	/* insert new index into current index block */
 940	/* current depth stored in i var */
 941	i = depth - 1;
 942	while (k--) {
 943		oldblock = newblock;
 944		newblock = ablocks[--a];
 945		bh = sb_getblk(inode->i_sb, newblock);
 946		if (!bh) {
 947			err = -EIO;
 948			goto cleanup;
 949		}
 950		lock_buffer(bh);
 951
 952		err = ext4_journal_get_create_access(handle, bh);
 953		if (err)
 954			goto cleanup;
 955
 956		neh = ext_block_hdr(bh);
 957		neh->eh_entries = cpu_to_le16(1);
 958		neh->eh_magic = EXT4_EXT_MAGIC;
 959		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 960		neh->eh_depth = cpu_to_le16(depth - i);
 961		fidx = EXT_FIRST_INDEX(neh);
 962		fidx->ei_block = border;
 963		ext4_idx_store_pblock(fidx, oldblock);
 964
 965		ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 966				i, newblock, le32_to_cpu(border), oldblock);
 967
 968		/* move remainder of path[i] to the new index block */
 969		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 970					EXT_LAST_INDEX(path[i].p_hdr))) {
 971			EXT4_ERROR_INODE(inode,
 972					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 973					 le32_to_cpu(path[i].p_ext->ee_block));
 974			err = -EIO;
 975			goto cleanup;
 976		}
 977		/* start copy indexes */
 978		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
 979		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 980				EXT_MAX_INDEX(path[i].p_hdr));
 981		ext4_ext_show_move(inode, path, newblock, i);
 982		if (m) {
 983			memmove(++fidx, path[i].p_idx,
 984				sizeof(struct ext4_extent_idx) * m);
 985			le16_add_cpu(&neh->eh_entries, m);
 986		}
 
 987		set_buffer_uptodate(bh);
 988		unlock_buffer(bh);
 989
 990		err = ext4_handle_dirty_metadata(handle, inode, bh);
 991		if (err)
 992			goto cleanup;
 993		brelse(bh);
 994		bh = NULL;
 995
 996		/* correct old index */
 997		if (m) {
 998			err = ext4_ext_get_access(handle, inode, path + i);
 999			if (err)
1000				goto cleanup;
1001			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1002			err = ext4_ext_dirty(handle, inode, path + i);
1003			if (err)
1004				goto cleanup;
1005		}
1006
1007		i--;
1008	}
1009
1010	/* insert new index */
1011	err = ext4_ext_insert_index(handle, inode, path + at,
1012				    le32_to_cpu(border), newblock);
1013
1014cleanup:
1015	if (bh) {
1016		if (buffer_locked(bh))
1017			unlock_buffer(bh);
1018		brelse(bh);
1019	}
1020
1021	if (err) {
1022		/* free all allocated blocks in error case */
1023		for (i = 0; i < depth; i++) {
1024			if (!ablocks[i])
1025				continue;
1026			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1027					 EXT4_FREE_BLOCKS_METADATA);
1028		}
1029	}
1030	kfree(ablocks);
1031
1032	return err;
1033}
1034
1035/*
1036 * ext4_ext_grow_indepth:
1037 * implements tree growing procedure:
1038 * - allocates new block
1039 * - moves top-level data (index block or leaf) into the new block
1040 * - initializes new top-level, creating index that points to the
1041 *   just created block
1042 */
1043static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1044				 unsigned int flags,
1045				 struct ext4_ext_path *path,
1046				 struct ext4_extent *newext)
1047{
1048	struct ext4_ext_path *curp = path;
1049	struct ext4_extent_header *neh;
1050	struct buffer_head *bh;
1051	ext4_fsblk_t newblock;
1052	int err = 0;
1053
1054	newblock = ext4_ext_new_meta_block(handle, inode, path,
1055		newext, &err, flags);
1056	if (newblock == 0)
1057		return err;
1058
1059	bh = sb_getblk(inode->i_sb, newblock);
1060	if (!bh) {
1061		err = -EIO;
1062		ext4_std_error(inode->i_sb, err);
1063		return err;
1064	}
1065	lock_buffer(bh);
1066
1067	err = ext4_journal_get_create_access(handle, bh);
1068	if (err) {
1069		unlock_buffer(bh);
1070		goto out;
1071	}
1072
1073	/* move top-level index/leaf into new block */
1074	memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
 
1075
1076	/* set size of new block */
1077	neh = ext_block_hdr(bh);
1078	/* old root could have indexes or leaves
1079	 * so calculate e_max right way */
1080	if (ext_depth(inode))
1081		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1082	else
1083		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1084	neh->eh_magic = EXT4_EXT_MAGIC;
 
1085	set_buffer_uptodate(bh);
1086	unlock_buffer(bh);
1087
1088	err = ext4_handle_dirty_metadata(handle, inode, bh);
1089	if (err)
1090		goto out;
1091
1092	/* create index in new top-level index: num,max,pointer */
1093	err = ext4_ext_get_access(handle, inode, curp);
1094	if (err)
1095		goto out;
1096
1097	curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1098	curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1099	curp->p_hdr->eh_entries = cpu_to_le16(1);
1100	curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1101
1102	if (path[0].p_hdr->eh_depth)
1103		curp->p_idx->ei_block =
1104			EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1105	else
1106		curp->p_idx->ei_block =
1107			EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1108	ext4_idx_store_pblock(curp->p_idx, newblock);
1109
1110	neh = ext_inode_hdr(inode);
 
 
 
 
 
 
 
 
1111	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1112		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1113		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1114		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1115
1116	neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1117	err = ext4_ext_dirty(handle, inode, curp);
1118out:
1119	brelse(bh);
1120
1121	return err;
1122}
1123
1124/*
1125 * ext4_ext_create_new_leaf:
1126 * finds empty index and adds new leaf.
1127 * if no free index is found, then it requests in-depth growing.
1128 */
1129static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1130				    unsigned int flags,
1131				    struct ext4_ext_path *path,
1132				    struct ext4_extent *newext)
1133{
1134	struct ext4_ext_path *curp;
1135	int depth, i, err = 0;
1136
1137repeat:
1138	i = depth = ext_depth(inode);
1139
1140	/* walk up to the tree and look for free index entry */
1141	curp = path + depth;
1142	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1143		i--;
1144		curp--;
1145	}
1146
1147	/* we use already allocated block for index block,
1148	 * so subsequent data blocks should be contiguous */
1149	if (EXT_HAS_FREE_INDEX(curp)) {
1150		/* if we found index with free entry, then use that
1151		 * entry: create all needed subtree and add new leaf */
1152		err = ext4_ext_split(handle, inode, flags, path, newext, i);
1153		if (err)
1154			goto out;
1155
1156		/* refill path */
1157		ext4_ext_drop_refs(path);
1158		path = ext4_ext_find_extent(inode,
1159				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1160				    path);
1161		if (IS_ERR(path))
1162			err = PTR_ERR(path);
1163	} else {
1164		/* tree is full, time to grow in depth */
1165		err = ext4_ext_grow_indepth(handle, inode, flags,
1166					    path, newext);
1167		if (err)
1168			goto out;
1169
1170		/* refill path */
1171		ext4_ext_drop_refs(path);
1172		path = ext4_ext_find_extent(inode,
1173				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1174				    path);
1175		if (IS_ERR(path)) {
1176			err = PTR_ERR(path);
1177			goto out;
1178		}
1179
1180		/*
1181		 * only first (depth 0 -> 1) produces free space;
1182		 * in all other cases we have to split the grown tree
1183		 */
1184		depth = ext_depth(inode);
1185		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1186			/* now we need to split */
1187			goto repeat;
1188		}
1189	}
1190
1191out:
1192	return err;
1193}
1194
1195/*
1196 * search the closest allocated block to the left for *logical
1197 * and returns it at @logical + it's physical address at @phys
1198 * if *logical is the smallest allocated block, the function
1199 * returns 0 at @phys
1200 * return value contains 0 (success) or error code
1201 */
1202static int ext4_ext_search_left(struct inode *inode,
1203				struct ext4_ext_path *path,
1204				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1205{
1206	struct ext4_extent_idx *ix;
1207	struct ext4_extent *ex;
1208	int depth, ee_len;
1209
1210	if (unlikely(path == NULL)) {
1211		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1212		return -EIO;
1213	}
1214	depth = path->p_depth;
1215	*phys = 0;
1216
1217	if (depth == 0 && path->p_ext == NULL)
1218		return 0;
1219
1220	/* usually extent in the path covers blocks smaller
1221	 * then *logical, but it can be that extent is the
1222	 * first one in the file */
1223
1224	ex = path[depth].p_ext;
1225	ee_len = ext4_ext_get_actual_len(ex);
1226	if (*logical < le32_to_cpu(ex->ee_block)) {
1227		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1228			EXT4_ERROR_INODE(inode,
1229					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1230					 *logical, le32_to_cpu(ex->ee_block));
1231			return -EIO;
1232		}
1233		while (--depth >= 0) {
1234			ix = path[depth].p_idx;
1235			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1236				EXT4_ERROR_INODE(inode,
1237				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1238				  ix != NULL ? ix->ei_block : 0,
1239				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1240				    EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1241				  depth);
1242				return -EIO;
1243			}
1244		}
1245		return 0;
1246	}
1247
1248	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1249		EXT4_ERROR_INODE(inode,
1250				 "logical %d < ee_block %d + ee_len %d!",
1251				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1252		return -EIO;
1253	}
1254
1255	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1256	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1257	return 0;
1258}
1259
1260/*
1261 * search the closest allocated block to the right for *logical
1262 * and returns it at @logical + it's physical address at @phys
1263 * if *logical is the smallest allocated block, the function
1264 * returns 0 at @phys
1265 * return value contains 0 (success) or error code
1266 */
1267static int ext4_ext_search_right(struct inode *inode,
1268				 struct ext4_ext_path *path,
1269				 ext4_lblk_t *logical, ext4_fsblk_t *phys)
 
1270{
1271	struct buffer_head *bh = NULL;
1272	struct ext4_extent_header *eh;
1273	struct ext4_extent_idx *ix;
1274	struct ext4_extent *ex;
1275	ext4_fsblk_t block;
1276	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1277	int ee_len;
1278
1279	if (unlikely(path == NULL)) {
1280		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1281		return -EIO;
1282	}
1283	depth = path->p_depth;
1284	*phys = 0;
1285
1286	if (depth == 0 && path->p_ext == NULL)
1287		return 0;
1288
1289	/* usually extent in the path covers blocks smaller
1290	 * then *logical, but it can be that extent is the
1291	 * first one in the file */
1292
1293	ex = path[depth].p_ext;
1294	ee_len = ext4_ext_get_actual_len(ex);
1295	if (*logical < le32_to_cpu(ex->ee_block)) {
1296		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1297			EXT4_ERROR_INODE(inode,
1298					 "first_extent(path[%d].p_hdr) != ex",
1299					 depth);
1300			return -EIO;
1301		}
1302		while (--depth >= 0) {
1303			ix = path[depth].p_idx;
1304			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1305				EXT4_ERROR_INODE(inode,
1306						 "ix != EXT_FIRST_INDEX *logical %d!",
1307						 *logical);
1308				return -EIO;
1309			}
1310		}
1311		*logical = le32_to_cpu(ex->ee_block);
1312		*phys = ext4_ext_pblock(ex);
1313		return 0;
1314	}
1315
1316	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1317		EXT4_ERROR_INODE(inode,
1318				 "logical %d < ee_block %d + ee_len %d!",
1319				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1320		return -EIO;
1321	}
1322
1323	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1324		/* next allocated block in this leaf */
1325		ex++;
1326		*logical = le32_to_cpu(ex->ee_block);
1327		*phys = ext4_ext_pblock(ex);
1328		return 0;
1329	}
1330
1331	/* go up and search for index to the right */
1332	while (--depth >= 0) {
1333		ix = path[depth].p_idx;
1334		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1335			goto got_index;
1336	}
1337
1338	/* we've gone up to the root and found no index to the right */
1339	return 0;
1340
1341got_index:
1342	/* we've found index to the right, let's
1343	 * follow it and find the closest allocated
1344	 * block to the right */
1345	ix++;
1346	block = ext4_idx_pblock(ix);
1347	while (++depth < path->p_depth) {
1348		bh = sb_bread(inode->i_sb, block);
1349		if (bh == NULL)
1350			return -EIO;
1351		eh = ext_block_hdr(bh);
1352		/* subtract from p_depth to get proper eh_depth */
1353		if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
 
1354			put_bh(bh);
1355			return -EIO;
1356		}
1357		ix = EXT_FIRST_INDEX(eh);
1358		block = ext4_idx_pblock(ix);
1359		put_bh(bh);
1360	}
1361
1362	bh = sb_bread(inode->i_sb, block);
1363	if (bh == NULL)
1364		return -EIO;
1365	eh = ext_block_hdr(bh);
1366	if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1367		put_bh(bh);
1368		return -EIO;
1369	}
1370	ex = EXT_FIRST_EXTENT(eh);
 
1371	*logical = le32_to_cpu(ex->ee_block);
1372	*phys = ext4_ext_pblock(ex);
1373	put_bh(bh);
 
 
1374	return 0;
1375}
1376
1377/*
1378 * ext4_ext_next_allocated_block:
1379 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1380 * NOTE: it considers block number from index entry as
1381 * allocated block. Thus, index entries have to be consistent
1382 * with leaves.
1383 */
1384static ext4_lblk_t
1385ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1386{
1387	int depth;
1388
1389	BUG_ON(path == NULL);
1390	depth = path->p_depth;
1391
1392	if (depth == 0 && path->p_ext == NULL)
1393		return EXT_MAX_BLOCKS;
1394
1395	while (depth >= 0) {
1396		if (depth == path->p_depth) {
1397			/* leaf */
1398			if (path[depth].p_ext !=
 
1399					EXT_LAST_EXTENT(path[depth].p_hdr))
1400			  return le32_to_cpu(path[depth].p_ext[1].ee_block);
1401		} else {
1402			/* index */
1403			if (path[depth].p_idx !=
1404					EXT_LAST_INDEX(path[depth].p_hdr))
1405			  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1406		}
1407		depth--;
1408	}
1409
1410	return EXT_MAX_BLOCKS;
1411}
1412
1413/*
1414 * ext4_ext_next_leaf_block:
1415 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1416 */
1417static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1418{
1419	int depth;
1420
1421	BUG_ON(path == NULL);
1422	depth = path->p_depth;
1423
1424	/* zero-tree has no leaf blocks at all */
1425	if (depth == 0)
1426		return EXT_MAX_BLOCKS;
1427
1428	/* go to index block */
1429	depth--;
1430
1431	while (depth >= 0) {
1432		if (path[depth].p_idx !=
1433				EXT_LAST_INDEX(path[depth].p_hdr))
1434			return (ext4_lblk_t)
1435				le32_to_cpu(path[depth].p_idx[1].ei_block);
1436		depth--;
1437	}
1438
1439	return EXT_MAX_BLOCKS;
1440}
1441
1442/*
1443 * ext4_ext_correct_indexes:
1444 * if leaf gets modified and modified extent is first in the leaf,
1445 * then we have to correct all indexes above.
1446 * TODO: do we need to correct tree in all cases?
1447 */
1448static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1449				struct ext4_ext_path *path)
1450{
1451	struct ext4_extent_header *eh;
1452	int depth = ext_depth(inode);
1453	struct ext4_extent *ex;
1454	__le32 border;
1455	int k, err = 0;
1456
1457	eh = path[depth].p_hdr;
1458	ex = path[depth].p_ext;
1459
1460	if (unlikely(ex == NULL || eh == NULL)) {
1461		EXT4_ERROR_INODE(inode,
1462				 "ex %p == NULL or eh %p == NULL", ex, eh);
1463		return -EIO;
1464	}
1465
1466	if (depth == 0) {
1467		/* there is no tree at all */
1468		return 0;
1469	}
1470
1471	if (ex != EXT_FIRST_EXTENT(eh)) {
1472		/* we correct tree if first leaf got modified only */
1473		return 0;
1474	}
1475
1476	/*
1477	 * TODO: we need correction if border is smaller than current one
1478	 */
1479	k = depth - 1;
1480	border = path[depth].p_ext->ee_block;
1481	err = ext4_ext_get_access(handle, inode, path + k);
1482	if (err)
1483		return err;
1484	path[k].p_idx->ei_block = border;
1485	err = ext4_ext_dirty(handle, inode, path + k);
1486	if (err)
1487		return err;
1488
1489	while (k--) {
1490		/* change all left-side indexes */
1491		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1492			break;
1493		err = ext4_ext_get_access(handle, inode, path + k);
1494		if (err)
1495			break;
1496		path[k].p_idx->ei_block = border;
1497		err = ext4_ext_dirty(handle, inode, path + k);
1498		if (err)
1499			break;
1500	}
1501
1502	return err;
1503}
1504
1505int
1506ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1507				struct ext4_extent *ex2)
1508{
1509	unsigned short ext1_ee_len, ext2_ee_len, max_len;
1510
1511	/*
1512	 * Make sure that either both extents are uninitialized, or
1513	 * both are _not_.
1514	 */
1515	if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1516		return 0;
1517
1518	if (ext4_ext_is_uninitialized(ex1))
1519		max_len = EXT_UNINIT_MAX_LEN;
1520	else
1521		max_len = EXT_INIT_MAX_LEN;
1522
1523	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1524	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1525
1526	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1527			le32_to_cpu(ex2->ee_block))
1528		return 0;
1529
1530	/*
1531	 * To allow future support for preallocated extents to be added
1532	 * as an RO_COMPAT feature, refuse to merge to extents if
1533	 * this can result in the top bit of ee_len being set.
1534	 */
1535	if (ext1_ee_len + ext2_ee_len > max_len)
1536		return 0;
1537#ifdef AGGRESSIVE_TEST
1538	if (ext1_ee_len >= 4)
1539		return 0;
1540#endif
1541
1542	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1543		return 1;
1544	return 0;
1545}
1546
1547/*
1548 * This function tries to merge the "ex" extent to the next extent in the tree.
1549 * It always tries to merge towards right. If you want to merge towards
1550 * left, pass "ex - 1" as argument instead of "ex".
1551 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1552 * 1 if they got merged.
1553 */
1554static int ext4_ext_try_to_merge_right(struct inode *inode,
1555				 struct ext4_ext_path *path,
1556				 struct ext4_extent *ex)
1557{
1558	struct ext4_extent_header *eh;
1559	unsigned int depth, len;
1560	int merge_done = 0;
1561	int uninitialized = 0;
1562
1563	depth = ext_depth(inode);
1564	BUG_ON(path[depth].p_hdr == NULL);
1565	eh = path[depth].p_hdr;
1566
1567	while (ex < EXT_LAST_EXTENT(eh)) {
1568		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1569			break;
1570		/* merge with next extent! */
1571		if (ext4_ext_is_uninitialized(ex))
1572			uninitialized = 1;
1573		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1574				+ ext4_ext_get_actual_len(ex + 1));
1575		if (uninitialized)
1576			ext4_ext_mark_uninitialized(ex);
1577
1578		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1579			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1580				* sizeof(struct ext4_extent);
1581			memmove(ex + 1, ex + 2, len);
1582		}
1583		le16_add_cpu(&eh->eh_entries, -1);
1584		merge_done = 1;
1585		WARN_ON(eh->eh_entries == 0);
1586		if (!eh->eh_entries)
1587			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1588	}
1589
1590	return merge_done;
1591}
1592
1593/*
1594 * This function tries to merge the @ex extent to neighbours in the tree.
1595 * return 1 if merge left else 0.
1596 */
1597static int ext4_ext_try_to_merge(struct inode *inode,
1598				  struct ext4_ext_path *path,
1599				  struct ext4_extent *ex) {
1600	struct ext4_extent_header *eh;
1601	unsigned int depth;
1602	int merge_done = 0;
1603	int ret = 0;
1604
1605	depth = ext_depth(inode);
1606	BUG_ON(path[depth].p_hdr == NULL);
1607	eh = path[depth].p_hdr;
1608
1609	if (ex > EXT_FIRST_EXTENT(eh))
1610		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1611
1612	if (!merge_done)
1613		ret = ext4_ext_try_to_merge_right(inode, path, ex);
1614
1615	return ret;
1616}
1617
1618/*
1619 * check if a portion of the "newext" extent overlaps with an
1620 * existing extent.
1621 *
1622 * If there is an overlap discovered, it updates the length of the newext
1623 * such that there will be no overlap, and then returns 1.
1624 * If there is no overlap found, it returns 0.
1625 */
1626static unsigned int ext4_ext_check_overlap(struct inode *inode,
 
1627					   struct ext4_extent *newext,
1628					   struct ext4_ext_path *path)
1629{
1630	ext4_lblk_t b1, b2;
1631	unsigned int depth, len1;
1632	unsigned int ret = 0;
1633
1634	b1 = le32_to_cpu(newext->ee_block);
1635	len1 = ext4_ext_get_actual_len(newext);
1636	depth = ext_depth(inode);
1637	if (!path[depth].p_ext)
1638		goto out;
1639	b2 = le32_to_cpu(path[depth].p_ext->ee_block);
 
1640
1641	/*
1642	 * get the next allocated block if the extent in the path
1643	 * is before the requested block(s)
1644	 */
1645	if (b2 < b1) {
1646		b2 = ext4_ext_next_allocated_block(path);
1647		if (b2 == EXT_MAX_BLOCKS)
1648			goto out;
 
1649	}
1650
1651	/* check for wrap through zero on extent logical start block*/
1652	if (b1 + len1 < b1) {
1653		len1 = EXT_MAX_BLOCKS - b1;
1654		newext->ee_len = cpu_to_le16(len1);
1655		ret = 1;
1656	}
1657
1658	/* check for overlap */
1659	if (b1 + len1 > b2) {
1660		newext->ee_len = cpu_to_le16(b2 - b1);
1661		ret = 1;
1662	}
1663out:
1664	return ret;
1665}
1666
1667/*
1668 * ext4_ext_insert_extent:
1669 * tries to merge requsted extent into the existing extent or
1670 * inserts requested extent as new one into the tree,
1671 * creating new leaf in the no-space case.
1672 */
1673int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1674				struct ext4_ext_path *path,
1675				struct ext4_extent *newext, int flag)
1676{
1677	struct ext4_extent_header *eh;
1678	struct ext4_extent *ex, *fex;
1679	struct ext4_extent *nearex; /* nearest extent */
1680	struct ext4_ext_path *npath = NULL;
1681	int depth, len, err;
1682	ext4_lblk_t next;
1683	unsigned uninitialized = 0;
1684	int flags = 0;
1685
1686	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1687		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1688		return -EIO;
1689	}
1690	depth = ext_depth(inode);
1691	ex = path[depth].p_ext;
1692	if (unlikely(path[depth].p_hdr == NULL)) {
1693		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1694		return -EIO;
1695	}
1696
1697	/* try to insert block into found extent and return */
1698	if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1699		&& ext4_can_extents_be_merged(inode, ex, newext)) {
1700		ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1701			  ext4_ext_is_uninitialized(newext),
1702			  ext4_ext_get_actual_len(newext),
1703			  le32_to_cpu(ex->ee_block),
1704			  ext4_ext_is_uninitialized(ex),
1705			  ext4_ext_get_actual_len(ex),
1706			  ext4_ext_pblock(ex));
1707		err = ext4_ext_get_access(handle, inode, path + depth);
1708		if (err)
1709			return err;
1710
1711		/*
1712		 * ext4_can_extents_be_merged should have checked that either
1713		 * both extents are uninitialized, or both aren't. Thus we
1714		 * need to check only one of them here.
1715		 */
1716		if (ext4_ext_is_uninitialized(ex))
1717			uninitialized = 1;
1718		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1719					+ ext4_ext_get_actual_len(newext));
1720		if (uninitialized)
1721			ext4_ext_mark_uninitialized(ex);
1722		eh = path[depth].p_hdr;
1723		nearex = ex;
1724		goto merge;
1725	}
1726
1727	depth = ext_depth(inode);
1728	eh = path[depth].p_hdr;
1729	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1730		goto has_space;
1731
1732	/* probably next leaf has space for us? */
1733	fex = EXT_LAST_EXTENT(eh);
1734	next = EXT_MAX_BLOCKS;
1735	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1736		next = ext4_ext_next_leaf_block(path);
1737	if (next != EXT_MAX_BLOCKS) {
1738		ext_debug("next leaf block - %d\n", next);
1739		BUG_ON(npath != NULL);
1740		npath = ext4_ext_find_extent(inode, next, NULL);
1741		if (IS_ERR(npath))
1742			return PTR_ERR(npath);
1743		BUG_ON(npath->p_depth != path->p_depth);
1744		eh = npath[depth].p_hdr;
1745		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1746			ext_debug("next leaf isn't full(%d)\n",
1747				  le16_to_cpu(eh->eh_entries));
1748			path = npath;
1749			goto has_space;
1750		}
1751		ext_debug("next leaf has no free space(%d,%d)\n",
1752			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1753	}
1754
1755	/*
1756	 * There is no free space in the found leaf.
1757	 * We're gonna add a new leaf in the tree.
1758	 */
1759	if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1760		flags = EXT4_MB_USE_ROOT_BLOCKS;
1761	err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1762	if (err)
1763		goto cleanup;
1764	depth = ext_depth(inode);
1765	eh = path[depth].p_hdr;
1766
1767has_space:
1768	nearex = path[depth].p_ext;
1769
1770	err = ext4_ext_get_access(handle, inode, path + depth);
1771	if (err)
1772		goto cleanup;
1773
1774	if (!nearex) {
1775		/* there is no extent in this leaf, create first one */
1776		ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1777				le32_to_cpu(newext->ee_block),
1778				ext4_ext_pblock(newext),
1779				ext4_ext_is_uninitialized(newext),
1780				ext4_ext_get_actual_len(newext));
1781		path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1782	} else if (le32_to_cpu(newext->ee_block)
 
1783			   > le32_to_cpu(nearex->ee_block)) {
1784/*		BUG_ON(newext->ee_block == nearex->ee_block); */
1785		if (nearex != EXT_LAST_EXTENT(eh)) {
1786			len = EXT_MAX_EXTENT(eh) - nearex;
1787			len = (len - 1) * sizeof(struct ext4_extent);
1788			len = len < 0 ? 0 : len;
1789			ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1790					"move %d from 0x%p to 0x%p\n",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1791					le32_to_cpu(newext->ee_block),
1792					ext4_ext_pblock(newext),
1793					ext4_ext_is_uninitialized(newext),
1794					ext4_ext_get_actual_len(newext),
1795					nearex, len, nearex + 1, nearex + 2);
1796			memmove(nearex + 2, nearex + 1, len);
 
1797		}
1798		path[depth].p_ext = nearex + 1;
1799	} else {
1800		BUG_ON(newext->ee_block == nearex->ee_block);
1801		len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1802		len = len < 0 ? 0 : len;
1803		ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1804				"move %d from 0x%p to 0x%p\n",
1805				le32_to_cpu(newext->ee_block),
1806				ext4_ext_pblock(newext),
1807				ext4_ext_is_uninitialized(newext),
1808				ext4_ext_get_actual_len(newext),
1809				nearex, len, nearex, nearex + 1);
1810		memmove(nearex + 1, nearex, len);
1811		path[depth].p_ext = nearex;
1812	}
1813
1814	le16_add_cpu(&eh->eh_entries, 1);
1815	nearex = path[depth].p_ext;
1816	nearex->ee_block = newext->ee_block;
1817	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1818	nearex->ee_len = newext->ee_len;
1819
1820merge:
1821	/* try to merge extents to the right */
1822	if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1823		ext4_ext_try_to_merge(inode, path, nearex);
1824
1825	/* try to merge extents to the left */
1826
1827	/* time to correct all indexes above */
1828	err = ext4_ext_correct_indexes(handle, inode, path);
1829	if (err)
1830		goto cleanup;
1831
1832	err = ext4_ext_dirty(handle, inode, path + depth);
1833
1834cleanup:
1835	if (npath) {
1836		ext4_ext_drop_refs(npath);
1837		kfree(npath);
1838	}
1839	ext4_ext_invalidate_cache(inode);
1840	return err;
1841}
1842
1843static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1844			       ext4_lblk_t num, ext_prepare_callback func,
1845			       void *cbdata)
1846{
1847	struct ext4_ext_path *path = NULL;
1848	struct ext4_ext_cache cbex;
1849	struct ext4_extent *ex;
1850	ext4_lblk_t next, start = 0, end = 0;
1851	ext4_lblk_t last = block + num;
1852	int depth, exists, err = 0;
1853
1854	BUG_ON(func == NULL);
1855	BUG_ON(inode == NULL);
1856
1857	while (block < last && block != EXT_MAX_BLOCKS) {
1858		num = last - block;
1859		/* find extent for this block */
1860		down_read(&EXT4_I(inode)->i_data_sem);
1861		path = ext4_ext_find_extent(inode, block, path);
1862		up_read(&EXT4_I(inode)->i_data_sem);
1863		if (IS_ERR(path)) {
1864			err = PTR_ERR(path);
1865			path = NULL;
1866			break;
1867		}
1868
1869		depth = ext_depth(inode);
1870		if (unlikely(path[depth].p_hdr == NULL)) {
1871			EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1872			err = -EIO;
1873			break;
1874		}
1875		ex = path[depth].p_ext;
1876		next = ext4_ext_next_allocated_block(path);
1877
1878		exists = 0;
1879		if (!ex) {
1880			/* there is no extent yet, so try to allocate
1881			 * all requested space */
1882			start = block;
1883			end = block + num;
1884		} else if (le32_to_cpu(ex->ee_block) > block) {
1885			/* need to allocate space before found extent */
1886			start = block;
1887			end = le32_to_cpu(ex->ee_block);
1888			if (block + num < end)
1889				end = block + num;
1890		} else if (block >= le32_to_cpu(ex->ee_block)
1891					+ ext4_ext_get_actual_len(ex)) {
1892			/* need to allocate space after found extent */
1893			start = block;
1894			end = block + num;
1895			if (end >= next)
1896				end = next;
1897		} else if (block >= le32_to_cpu(ex->ee_block)) {
1898			/*
1899			 * some part of requested space is covered
1900			 * by found extent
1901			 */
1902			start = block;
1903			end = le32_to_cpu(ex->ee_block)
1904				+ ext4_ext_get_actual_len(ex);
1905			if (block + num < end)
1906				end = block + num;
1907			exists = 1;
1908		} else {
1909			BUG();
1910		}
1911		BUG_ON(end <= start);
1912
1913		if (!exists) {
1914			cbex.ec_block = start;
1915			cbex.ec_len = end - start;
1916			cbex.ec_start = 0;
1917		} else {
1918			cbex.ec_block = le32_to_cpu(ex->ee_block);
1919			cbex.ec_len = ext4_ext_get_actual_len(ex);
1920			cbex.ec_start = ext4_ext_pblock(ex);
1921		}
1922
1923		if (unlikely(cbex.ec_len == 0)) {
1924			EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1925			err = -EIO;
1926			break;
1927		}
1928		err = func(inode, next, &cbex, ex, cbdata);
1929		ext4_ext_drop_refs(path);
1930
1931		if (err < 0)
1932			break;
1933
1934		if (err == EXT_REPEAT)
1935			continue;
1936		else if (err == EXT_BREAK) {
1937			err = 0;
1938			break;
1939		}
1940
1941		if (ext_depth(inode) != depth) {
1942			/* depth was changed. we have to realloc path */
1943			kfree(path);
1944			path = NULL;
1945		}
1946
1947		block = cbex.ec_block + cbex.ec_len;
1948	}
1949
1950	if (path) {
1951		ext4_ext_drop_refs(path);
1952		kfree(path);
1953	}
1954
1955	return err;
1956}
1957
1958static void
1959ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1960			__u32 len, ext4_fsblk_t start)
1961{
1962	struct ext4_ext_cache *cex;
1963	BUG_ON(len == 0);
1964	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
 
1965	cex = &EXT4_I(inode)->i_cached_extent;
1966	cex->ec_block = block;
1967	cex->ec_len = len;
1968	cex->ec_start = start;
1969	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1970}
1971
1972/*
1973 * ext4_ext_put_gap_in_cache:
1974 * calculate boundaries of the gap that the requested block fits into
1975 * and cache this gap
1976 */
1977static void
1978ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1979				ext4_lblk_t block)
1980{
1981	int depth = ext_depth(inode);
1982	unsigned long len;
1983	ext4_lblk_t lblock;
1984	struct ext4_extent *ex;
1985
1986	ex = path[depth].p_ext;
1987	if (ex == NULL) {
1988		/* there is no extent yet, so gap is [0;-] */
1989		lblock = 0;
1990		len = EXT_MAX_BLOCKS;
1991		ext_debug("cache gap(whole file):");
1992	} else if (block < le32_to_cpu(ex->ee_block)) {
1993		lblock = block;
1994		len = le32_to_cpu(ex->ee_block) - block;
1995		ext_debug("cache gap(before): %u [%u:%u]",
1996				block,
1997				le32_to_cpu(ex->ee_block),
1998				 ext4_ext_get_actual_len(ex));
1999	} else if (block >= le32_to_cpu(ex->ee_block)
2000			+ ext4_ext_get_actual_len(ex)) {
2001		ext4_lblk_t next;
2002		lblock = le32_to_cpu(ex->ee_block)
2003			+ ext4_ext_get_actual_len(ex);
2004
2005		next = ext4_ext_next_allocated_block(path);
2006		ext_debug("cache gap(after): [%u:%u] %u",
2007				le32_to_cpu(ex->ee_block),
2008				ext4_ext_get_actual_len(ex),
2009				block);
2010		BUG_ON(next == lblock);
2011		len = next - lblock;
2012	} else {
2013		lblock = len = 0;
2014		BUG();
2015	}
2016
2017	ext_debug(" -> %u:%lu\n", lblock, len);
2018	ext4_ext_put_in_cache(inode, lblock, len, 0);
2019}
2020
2021/*
2022 * ext4_ext_check_cache()
2023 * Checks to see if the given block is in the cache.
2024 * If it is, the cached extent is stored in the given
2025 * cache extent pointer.  If the cached extent is a hole,
2026 * this routine should be used instead of
2027 * ext4_ext_in_cache if the calling function needs to
2028 * know the size of the hole.
2029 *
2030 * @inode: The files inode
2031 * @block: The block to look for in the cache
2032 * @ex:    Pointer where the cached extent will be stored
2033 *         if it contains block
2034 *
2035 * Return 0 if cache is invalid; 1 if the cache is valid
2036 */
2037static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2038	struct ext4_ext_cache *ex){
2039	struct ext4_ext_cache *cex;
2040	struct ext4_sb_info *sbi;
2041	int ret = 0;
2042
2043	/*
2044	 * We borrow i_block_reservation_lock to protect i_cached_extent
2045	 */
2046	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2047	cex = &EXT4_I(inode)->i_cached_extent;
2048	sbi = EXT4_SB(inode->i_sb);
2049
2050	/* has cache valid data? */
2051	if (cex->ec_len == 0)
2052		goto errout;
2053
2054	if (in_range(block, cex->ec_block, cex->ec_len)) {
2055		memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2056		ext_debug("%u cached by %u:%u:%llu\n",
2057				block,
2058				cex->ec_block, cex->ec_len, cex->ec_start);
2059		ret = 1;
2060	}
2061errout:
2062	if (!ret)
2063		sbi->extent_cache_misses++;
2064	else
2065		sbi->extent_cache_hits++;
2066	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2067	return ret;
2068}
2069
2070/*
2071 * ext4_ext_in_cache()
2072 * Checks to see if the given block is in the cache.
2073 * If it is, the cached extent is stored in the given
2074 * extent pointer.
2075 *
2076 * @inode: The files inode
2077 * @block: The block to look for in the cache
2078 * @ex:    Pointer where the cached extent will be stored
2079 *         if it contains block
2080 *
2081 * Return 0 if cache is invalid; 1 if the cache is valid
2082 */
2083static int
2084ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2085			struct ext4_extent *ex)
2086{
2087	struct ext4_ext_cache cex;
2088	int ret = 0;
2089
2090	if (ext4_ext_check_cache(inode, block, &cex)) {
2091		ex->ee_block = cpu_to_le32(cex.ec_block);
2092		ext4_ext_store_pblock(ex, cex.ec_start);
2093		ex->ee_len = cpu_to_le16(cex.ec_len);
2094		ret = 1;
2095	}
2096
2097	return ret;
2098}
2099
2100
2101/*
2102 * ext4_ext_rm_idx:
2103 * removes index from the index block.
2104 */
2105static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2106			struct ext4_ext_path *path)
2107{
2108	int err;
2109	ext4_fsblk_t leaf;
2110
2111	/* free index block */
2112	path--;
2113	leaf = ext4_idx_pblock(path->p_idx);
2114	if (unlikely(path->p_hdr->eh_entries == 0)) {
2115		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2116		return -EIO;
2117	}
2118	err = ext4_ext_get_access(handle, inode, path);
2119	if (err)
2120		return err;
2121
2122	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2123		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2124		len *= sizeof(struct ext4_extent_idx);
2125		memmove(path->p_idx, path->p_idx + 1, len);
2126	}
2127
2128	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2129	err = ext4_ext_dirty(handle, inode, path);
2130	if (err)
2131		return err;
2132	ext_debug("index is empty, remove it, free block %llu\n", leaf);
 
 
2133	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2134			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2135	return err;
2136}
2137
2138/*
2139 * ext4_ext_calc_credits_for_single_extent:
2140 * This routine returns max. credits that needed to insert an extent
2141 * to the extent tree.
2142 * When pass the actual path, the caller should calculate credits
2143 * under i_data_sem.
2144 */
2145int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2146						struct ext4_ext_path *path)
2147{
2148	if (path) {
2149		int depth = ext_depth(inode);
2150		int ret = 0;
2151
2152		/* probably there is space in leaf? */
2153		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2154				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2155
2156			/*
2157			 *  There are some space in the leaf tree, no
2158			 *  need to account for leaf block credit
2159			 *
2160			 *  bitmaps and block group descriptor blocks
2161			 *  and other metadat blocks still need to be
2162			 *  accounted.
2163			 */
2164			/* 1 bitmap, 1 block group descriptor */
2165			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2166			return ret;
2167		}
2168	}
2169
2170	return ext4_chunk_trans_blocks(inode, nrblocks);
2171}
2172
2173/*
2174 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2175 *
2176 * if nrblocks are fit in a single extent (chunk flag is 1), then
2177 * in the worse case, each tree level index/leaf need to be changed
2178 * if the tree split due to insert a new extent, then the old tree
2179 * index/leaf need to be updated too
2180 *
2181 * If the nrblocks are discontiguous, they could cause
2182 * the whole tree split more than once, but this is really rare.
2183 */
2184int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2185{
2186	int index;
2187	int depth = ext_depth(inode);
2188
2189	if (chunk)
2190		index = depth * 2;
2191	else
2192		index = depth * 3;
2193
2194	return index;
2195}
2196
2197static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2198				struct ext4_extent *ex,
2199				ext4_lblk_t from, ext4_lblk_t to)
 
2200{
 
2201	unsigned short ee_len =  ext4_ext_get_actual_len(ex);
 
2202	int flags = EXT4_FREE_BLOCKS_FORGET;
2203
2204	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2205		flags |= EXT4_FREE_BLOCKS_METADATA;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2206#ifdef EXTENTS_STATS
2207	{
2208		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2209		spin_lock(&sbi->s_ext_stats_lock);
2210		sbi->s_ext_blocks += ee_len;
2211		sbi->s_ext_extents++;
2212		if (ee_len < sbi->s_ext_min)
2213			sbi->s_ext_min = ee_len;
2214		if (ee_len > sbi->s_ext_max)
2215			sbi->s_ext_max = ee_len;
2216		if (ext_depth(inode) > sbi->s_depth_max)
2217			sbi->s_depth_max = ext_depth(inode);
2218		spin_unlock(&sbi->s_ext_stats_lock);
2219	}
2220#endif
2221	if (from >= le32_to_cpu(ex->ee_block)
2222	    && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2223		/* tail removal */
2224		ext4_lblk_t num;
2225		ext4_fsblk_t start;
2226
2227		num = le32_to_cpu(ex->ee_block) + ee_len - from;
2228		start = ext4_ext_pblock(ex) + ee_len - num;
2229		ext_debug("free last %u blocks starting %llu\n", num, start);
2230		ext4_free_blocks(handle, inode, NULL, start, num, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
2231	} else if (from == le32_to_cpu(ex->ee_block)
2232		   && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2233		/* head removal */
2234		ext4_lblk_t num;
2235		ext4_fsblk_t start;
2236
2237		num = to - from;
2238		start = ext4_ext_pblock(ex);
2239
2240		ext_debug("free first %u blocks starting %llu\n", num, start);
2241		ext4_free_blocks(handle, inode, 0, start, num, flags);
2242
2243	} else {
2244		printk(KERN_INFO "strange request: removal(2) "
2245				"%u-%u from %u:%u\n",
2246				from, to, le32_to_cpu(ex->ee_block), ee_len);
2247	}
2248	return 0;
2249}
2250
2251
2252/*
2253 * ext4_ext_rm_leaf() Removes the extents associated with the
2254 * blocks appearing between "start" and "end", and splits the extents
2255 * if "start" and "end" appear in the same extent
2256 *
2257 * @handle: The journal handle
2258 * @inode:  The files inode
2259 * @path:   The path to the leaf
2260 * @start:  The first block to remove
2261 * @end:   The last block to remove
2262 */
2263static int
2264ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2265		struct ext4_ext_path *path, ext4_lblk_t start,
2266		ext4_lblk_t end)
2267{
 
2268	int err = 0, correct_index = 0;
2269	int depth = ext_depth(inode), credits;
2270	struct ext4_extent_header *eh;
2271	ext4_lblk_t a, b, block;
2272	unsigned num;
2273	ext4_lblk_t ex_ee_block;
2274	unsigned short ex_ee_len;
2275	unsigned uninitialized = 0;
2276	struct ext4_extent *ex;
2277	struct ext4_map_blocks map;
2278
2279	/* the header must be checked already in ext4_ext_remove_space() */
2280	ext_debug("truncate since %u in leaf\n", start);
2281	if (!path[depth].p_hdr)
2282		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2283	eh = path[depth].p_hdr;
2284	if (unlikely(path[depth].p_hdr == NULL)) {
2285		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2286		return -EIO;
2287	}
2288	/* find where to start removing */
2289	ex = EXT_LAST_EXTENT(eh);
2290
2291	ex_ee_block = le32_to_cpu(ex->ee_block);
2292	ex_ee_len = ext4_ext_get_actual_len(ex);
2293
 
 
2294	while (ex >= EXT_FIRST_EXTENT(eh) &&
2295			ex_ee_block + ex_ee_len > start) {
2296
2297		if (ext4_ext_is_uninitialized(ex))
2298			uninitialized = 1;
2299		else
2300			uninitialized = 0;
2301
2302		ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2303			 uninitialized, ex_ee_len);
2304		path[depth].p_ext = ex;
2305
2306		a = ex_ee_block > start ? ex_ee_block : start;
2307		b = ex_ee_block+ex_ee_len - 1 < end ?
2308			ex_ee_block+ex_ee_len - 1 : end;
2309
2310		ext_debug("  border %u:%u\n", a, b);
2311
2312		/* If this extent is beyond the end of the hole, skip it */
2313		if (end <= ex_ee_block) {
2314			ex--;
2315			ex_ee_block = le32_to_cpu(ex->ee_block);
2316			ex_ee_len = ext4_ext_get_actual_len(ex);
2317			continue;
2318		} else if (a != ex_ee_block &&
2319			b != ex_ee_block + ex_ee_len - 1) {
2320			/*
2321			 * If this is a truncate, then this condition should
2322			 * never happen because at least one of the end points
2323			 * needs to be on the edge of the extent.
2324			 */
2325			if (end == EXT_MAX_BLOCKS - 1) {
2326				ext_debug("  bad truncate %u:%u\n",
2327						start, end);
2328				block = 0;
2329				num = 0;
2330				err = -EIO;
2331				goto out;
2332			}
2333			/*
2334			 * else this is a hole punch, so the extent needs to
2335			 * be split since neither edge of the hole is on the
2336			 * extent edge
2337			 */
2338			else{
2339				map.m_pblk = ext4_ext_pblock(ex);
2340				map.m_lblk = ex_ee_block;
2341				map.m_len = b - ex_ee_block;
2342
2343				err = ext4_split_extent(handle,
2344					inode, path, &map, 0,
2345					EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2346					EXT4_GET_BLOCKS_PRE_IO);
2347
2348				if (err < 0)
2349					goto out;
2350
2351				ex_ee_len = ext4_ext_get_actual_len(ex);
2352
2353				b = ex_ee_block+ex_ee_len - 1 < end ?
2354					ex_ee_block+ex_ee_len - 1 : end;
2355
2356				/* Then remove tail of this extent */
2357				block = ex_ee_block;
2358				num = a - block;
2359			}
2360		} else if (a != ex_ee_block) {
2361			/* remove tail of the extent */
2362			block = ex_ee_block;
2363			num = a - block;
2364		} else if (b != ex_ee_block + ex_ee_len - 1) {
2365			/* remove head of the extent */
2366			block = b;
2367			num =  ex_ee_block + ex_ee_len - b;
2368
2369			/*
2370			 * If this is a truncate, this condition
2371			 * should never happen
2372			 */
2373			if (end == EXT_MAX_BLOCKS - 1) {
2374				ext_debug("  bad truncate %u:%u\n",
2375					start, end);
2376				err = -EIO;
2377				goto out;
2378			}
2379		} else {
2380			/* remove whole extent: excellent! */
2381			block = ex_ee_block;
2382			num = 0;
2383			if (a != ex_ee_block) {
2384				ext_debug("  bad truncate %u:%u\n",
2385					start, end);
2386				err = -EIO;
2387				goto out;
2388			}
2389
2390			if (b != ex_ee_block + ex_ee_len - 1) {
2391				ext_debug("  bad truncate %u:%u\n",
2392					start, end);
2393				err = -EIO;
2394				goto out;
2395			}
2396		}
2397
2398		/*
2399		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2400		 * descriptor) for each block group; assume two block
2401		 * groups plus ex_ee_len/blocks_per_block_group for
2402		 * the worst case
2403		 */
2404		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2405		if (ex == EXT_FIRST_EXTENT(eh)) {
2406			correct_index = 1;
2407			credits += (ext_depth(inode)) + 1;
2408		}
2409		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2410
2411		err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2412		if (err)
2413			goto out;
2414
2415		err = ext4_ext_get_access(handle, inode, path + depth);
2416		if (err)
2417			goto out;
2418
2419		err = ext4_remove_blocks(handle, inode, ex, a, b);
 
2420		if (err)
2421			goto out;
2422
2423		if (num == 0) {
2424			/* this extent is removed; mark slot entirely unused */
2425			ext4_ext_store_pblock(ex, 0);
2426		} else if (block != ex_ee_block) {
2427			/*
2428			 * If this was a head removal, then we need to update
2429			 * the physical block since it is now at a different
2430			 * location
2431			 */
2432			ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2433		}
2434
2435		ex->ee_block = cpu_to_le32(block);
2436		ex->ee_len = cpu_to_le16(num);
2437		/*
2438		 * Do not mark uninitialized if all the blocks in the
2439		 * extent have been removed.
2440		 */
2441		if (uninitialized && num)
2442			ext4_ext_mark_uninitialized(ex);
2443
2444		err = ext4_ext_dirty(handle, inode, path + depth);
2445		if (err)
2446			goto out;
2447
2448		/*
2449		 * If the extent was completely released,
2450		 * we need to remove it from the leaf
2451		 */
2452		if (num == 0) {
2453			if (end != EXT_MAX_BLOCKS - 1) {
2454				/*
2455				 * For hole punching, we need to scoot all the
2456				 * extents up when an extent is removed so that
2457				 * we dont have blank extents in the middle
2458				 */
2459				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2460					sizeof(struct ext4_extent));
2461
2462				/* Now get rid of the one at the end */
2463				memset(EXT_LAST_EXTENT(eh), 0,
2464					sizeof(struct ext4_extent));
2465			}
2466			le16_add_cpu(&eh->eh_entries, -1);
2467		}
 
 
 
 
 
2468
2469		ext_debug("new extent: %u:%u:%llu\n", block, num,
2470				ext4_ext_pblock(ex));
2471		ex--;
2472		ex_ee_block = le32_to_cpu(ex->ee_block);
2473		ex_ee_len = ext4_ext_get_actual_len(ex);
2474	}
2475
2476	if (correct_index && eh->eh_entries)
2477		err = ext4_ext_correct_indexes(handle, inode, path);
2478
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2479	/* if this leaf is free, then we should
2480	 * remove it from index block above */
2481	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2482		err = ext4_ext_rm_idx(handle, inode, path + depth);
2483
2484out:
2485	return err;
2486}
2487
2488/*
2489 * ext4_ext_more_to_rm:
2490 * returns 1 if current index has to be freed (even partial)
2491 */
2492static int
2493ext4_ext_more_to_rm(struct ext4_ext_path *path)
2494{
2495	BUG_ON(path->p_idx == NULL);
2496
2497	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2498		return 0;
2499
2500	/*
2501	 * if truncate on deeper level happened, it wasn't partial,
2502	 * so we have to consider current index for truncation
2503	 */
2504	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2505		return 0;
2506	return 1;
2507}
2508
2509static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
 
2510{
2511	struct super_block *sb = inode->i_sb;
2512	int depth = ext_depth(inode);
2513	struct ext4_ext_path *path;
 
2514	handle_t *handle;
2515	int i, err;
2516
2517	ext_debug("truncate since %u\n", start);
2518
2519	/* probably first extent we're gonna free will be last in block */
2520	handle = ext4_journal_start(inode, depth + 1);
2521	if (IS_ERR(handle))
2522		return PTR_ERR(handle);
2523
2524again:
2525	ext4_ext_invalidate_cache(inode);
2526
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2527	/*
2528	 * We start scanning from right side, freeing all the blocks
2529	 * after i_size and walking into the tree depth-wise.
2530	 */
2531	depth = ext_depth(inode);
2532	path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2533	if (path == NULL) {
2534		ext4_journal_stop(handle);
2535		return -ENOMEM;
2536	}
2537	path[0].p_depth = depth;
2538	path[0].p_hdr = ext_inode_hdr(inode);
2539	if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2540		err = -EIO;
2541		goto out;
 
 
 
 
 
 
 
 
 
 
2542	}
2543	i = err = 0;
2544
2545	while (i >= 0 && err == 0) {
2546		if (i == depth) {
2547			/* this is leaf block */
2548			err = ext4_ext_rm_leaf(handle, inode, path,
2549					start, EXT_MAX_BLOCKS - 1);
 
2550			/* root level has p_bh == NULL, brelse() eats this */
2551			brelse(path[i].p_bh);
2552			path[i].p_bh = NULL;
2553			i--;
2554			continue;
2555		}
2556
2557		/* this is index block */
2558		if (!path[i].p_hdr) {
2559			ext_debug("initialize header\n");
2560			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2561		}
2562
2563		if (!path[i].p_idx) {
2564			/* this level hasn't been touched yet */
2565			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2566			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2567			ext_debug("init index ptr: hdr 0x%p, num %d\n",
2568				  path[i].p_hdr,
2569				  le16_to_cpu(path[i].p_hdr->eh_entries));
2570		} else {
2571			/* we were already here, see at next index */
2572			path[i].p_idx--;
2573		}
2574
2575		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2576				i, EXT_FIRST_INDEX(path[i].p_hdr),
2577				path[i].p_idx);
2578		if (ext4_ext_more_to_rm(path + i)) {
2579			struct buffer_head *bh;
2580			/* go to the next level */
2581			ext_debug("move to level %d (block %llu)\n",
2582				  i + 1, ext4_idx_pblock(path[i].p_idx));
2583			memset(path + i + 1, 0, sizeof(*path));
2584			bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2585			if (!bh) {
2586				/* should we reset i_size? */
2587				err = -EIO;
2588				break;
2589			}
2590			if (WARN_ON(i + 1 > depth)) {
2591				err = -EIO;
2592				break;
2593			}
2594			if (ext4_ext_check(inode, ext_block_hdr(bh),
2595							depth - i - 1)) {
2596				err = -EIO;
2597				break;
2598			}
2599			path[i + 1].p_bh = bh;
2600
2601			/* save actual number of indexes since this
2602			 * number is changed at the next iteration */
2603			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2604			i++;
2605		} else {
2606			/* we finished processing this index, go up */
2607			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2608				/* index is empty, remove it;
2609				 * handle must be already prepared by the
2610				 * truncatei_leaf() */
2611				err = ext4_ext_rm_idx(handle, inode, path + i);
2612			}
2613			/* root level has p_bh == NULL, brelse() eats this */
2614			brelse(path[i].p_bh);
2615			path[i].p_bh = NULL;
2616			i--;
2617			ext_debug("return to level %d\n", i);
2618		}
2619	}
2620
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2621	/* TODO: flexible tree reduction should be here */
2622	if (path->p_hdr->eh_entries == 0) {
2623		/*
2624		 * truncate to zero freed all the tree,
2625		 * so we need to correct eh_depth
2626		 */
2627		err = ext4_ext_get_access(handle, inode, path);
2628		if (err == 0) {
2629			ext_inode_hdr(inode)->eh_depth = 0;
2630			ext_inode_hdr(inode)->eh_max =
2631				cpu_to_le16(ext4_ext_space_root(inode, 0));
2632			err = ext4_ext_dirty(handle, inode, path);
2633		}
2634	}
2635out:
2636	ext4_ext_drop_refs(path);
2637	kfree(path);
2638	if (err == -EAGAIN)
 
2639		goto again;
 
2640	ext4_journal_stop(handle);
2641
2642	return err;
2643}
2644
2645/*
2646 * called at mount time
2647 */
2648void ext4_ext_init(struct super_block *sb)
2649{
2650	/*
2651	 * possible initialization would be here
2652	 */
2653
2654	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2655#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2656		printk(KERN_INFO "EXT4-fs: file extents enabled");
2657#ifdef AGGRESSIVE_TEST
2658		printk(", aggressive tests");
2659#endif
2660#ifdef CHECK_BINSEARCH
2661		printk(", check binsearch");
2662#endif
2663#ifdef EXTENTS_STATS
2664		printk(", stats");
2665#endif
2666		printk("\n");
2667#endif
2668#ifdef EXTENTS_STATS
2669		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2670		EXT4_SB(sb)->s_ext_min = 1 << 30;
2671		EXT4_SB(sb)->s_ext_max = 0;
2672#endif
2673	}
2674}
2675
2676/*
2677 * called at umount time
2678 */
2679void ext4_ext_release(struct super_block *sb)
2680{
2681	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2682		return;
2683
2684#ifdef EXTENTS_STATS
2685	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2686		struct ext4_sb_info *sbi = EXT4_SB(sb);
2687		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2688			sbi->s_ext_blocks, sbi->s_ext_extents,
2689			sbi->s_ext_blocks / sbi->s_ext_extents);
2690		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2691			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2692	}
2693#endif
2694}
2695
2696/* FIXME!! we need to try to merge to left or right after zero-out  */
2697static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2698{
2699	ext4_fsblk_t ee_pblock;
2700	unsigned int ee_len;
2701	int ret;
2702
2703	ee_len    = ext4_ext_get_actual_len(ex);
2704	ee_pblock = ext4_ext_pblock(ex);
2705
2706	ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2707	if (ret > 0)
2708		ret = 0;
2709
2710	return ret;
2711}
2712
2713/*
2714 * used by extent splitting.
2715 */
2716#define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
2717					due to ENOSPC */
2718#define EXT4_EXT_MARK_UNINIT1	0x2  /* mark first half uninitialized */
2719#define EXT4_EXT_MARK_UNINIT2	0x4  /* mark second half uninitialized */
2720
2721/*
2722 * ext4_split_extent_at() splits an extent at given block.
2723 *
2724 * @handle: the journal handle
2725 * @inode: the file inode
2726 * @path: the path to the extent
2727 * @split: the logical block where the extent is splitted.
2728 * @split_flags: indicates if the extent could be zeroout if split fails, and
2729 *		 the states(init or uninit) of new extents.
2730 * @flags: flags used to insert new extent to extent tree.
2731 *
2732 *
2733 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2734 * of which are deterimined by split_flag.
2735 *
2736 * There are two cases:
2737 *  a> the extent are splitted into two extent.
2738 *  b> split is not needed, and just mark the extent.
2739 *
2740 * return 0 on success.
2741 */
2742static int ext4_split_extent_at(handle_t *handle,
2743			     struct inode *inode,
2744			     struct ext4_ext_path *path,
2745			     ext4_lblk_t split,
2746			     int split_flag,
2747			     int flags)
2748{
2749	ext4_fsblk_t newblock;
2750	ext4_lblk_t ee_block;
2751	struct ext4_extent *ex, newex, orig_ex;
2752	struct ext4_extent *ex2 = NULL;
2753	unsigned int ee_len, depth;
2754	int err = 0;
2755
2756	ext_debug("ext4_split_extents_at: inode %lu, logical"
2757		"block %llu\n", inode->i_ino, (unsigned long long)split);
2758
2759	ext4_ext_show_leaf(inode, path);
2760
2761	depth = ext_depth(inode);
2762	ex = path[depth].p_ext;
2763	ee_block = le32_to_cpu(ex->ee_block);
2764	ee_len = ext4_ext_get_actual_len(ex);
2765	newblock = split - ee_block + ext4_ext_pblock(ex);
2766
2767	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2768
2769	err = ext4_ext_get_access(handle, inode, path + depth);
2770	if (err)
2771		goto out;
2772
2773	if (split == ee_block) {
2774		/*
2775		 * case b: block @split is the block that the extent begins with
2776		 * then we just change the state of the extent, and splitting
2777		 * is not needed.
2778		 */
2779		if (split_flag & EXT4_EXT_MARK_UNINIT2)
2780			ext4_ext_mark_uninitialized(ex);
2781		else
2782			ext4_ext_mark_initialized(ex);
2783
2784		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2785			ext4_ext_try_to_merge(inode, path, ex);
2786
2787		err = ext4_ext_dirty(handle, inode, path + depth);
2788		goto out;
2789	}
2790
2791	/* case a */
2792	memcpy(&orig_ex, ex, sizeof(orig_ex));
2793	ex->ee_len = cpu_to_le16(split - ee_block);
2794	if (split_flag & EXT4_EXT_MARK_UNINIT1)
2795		ext4_ext_mark_uninitialized(ex);
2796
2797	/*
2798	 * path may lead to new leaf, not to original leaf any more
2799	 * after ext4_ext_insert_extent() returns,
2800	 */
2801	err = ext4_ext_dirty(handle, inode, path + depth);
2802	if (err)
2803		goto fix_extent_len;
2804
2805	ex2 = &newex;
2806	ex2->ee_block = cpu_to_le32(split);
2807	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2808	ext4_ext_store_pblock(ex2, newblock);
2809	if (split_flag & EXT4_EXT_MARK_UNINIT2)
2810		ext4_ext_mark_uninitialized(ex2);
2811
2812	err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2813	if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2814		err = ext4_ext_zeroout(inode, &orig_ex);
2815		if (err)
2816			goto fix_extent_len;
2817		/* update the extent length and mark as initialized */
2818		ex->ee_len = cpu_to_le32(ee_len);
2819		ext4_ext_try_to_merge(inode, path, ex);
2820		err = ext4_ext_dirty(handle, inode, path + depth);
2821		goto out;
2822	} else if (err)
2823		goto fix_extent_len;
2824
2825out:
2826	ext4_ext_show_leaf(inode, path);
2827	return err;
2828
2829fix_extent_len:
2830	ex->ee_len = orig_ex.ee_len;
2831	ext4_ext_dirty(handle, inode, path + depth);
2832	return err;
2833}
2834
2835/*
2836 * ext4_split_extents() splits an extent and mark extent which is covered
2837 * by @map as split_flags indicates
2838 *
2839 * It may result in splitting the extent into multiple extents (upto three)
2840 * There are three possibilities:
2841 *   a> There is no split required
2842 *   b> Splits in two extents: Split is happening at either end of the extent
2843 *   c> Splits in three extents: Somone is splitting in middle of the extent
2844 *
2845 */
2846static int ext4_split_extent(handle_t *handle,
2847			      struct inode *inode,
2848			      struct ext4_ext_path *path,
2849			      struct ext4_map_blocks *map,
2850			      int split_flag,
2851			      int flags)
2852{
2853	ext4_lblk_t ee_block;
2854	struct ext4_extent *ex;
2855	unsigned int ee_len, depth;
2856	int err = 0;
2857	int uninitialized;
2858	int split_flag1, flags1;
2859
2860	depth = ext_depth(inode);
2861	ex = path[depth].p_ext;
2862	ee_block = le32_to_cpu(ex->ee_block);
2863	ee_len = ext4_ext_get_actual_len(ex);
2864	uninitialized = ext4_ext_is_uninitialized(ex);
2865
2866	if (map->m_lblk + map->m_len < ee_block + ee_len) {
2867		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2868			      EXT4_EXT_MAY_ZEROOUT : 0;
2869		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2870		if (uninitialized)
2871			split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2872				       EXT4_EXT_MARK_UNINIT2;
2873		err = ext4_split_extent_at(handle, inode, path,
2874				map->m_lblk + map->m_len, split_flag1, flags1);
2875		if (err)
2876			goto out;
2877	}
2878
2879	ext4_ext_drop_refs(path);
2880	path = ext4_ext_find_extent(inode, map->m_lblk, path);
2881	if (IS_ERR(path))
2882		return PTR_ERR(path);
2883
2884	if (map->m_lblk >= ee_block) {
2885		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2886			      EXT4_EXT_MAY_ZEROOUT : 0;
2887		if (uninitialized)
2888			split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2889		if (split_flag & EXT4_EXT_MARK_UNINIT2)
2890			split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2891		err = ext4_split_extent_at(handle, inode, path,
2892				map->m_lblk, split_flag1, flags);
2893		if (err)
2894			goto out;
2895	}
2896
2897	ext4_ext_show_leaf(inode, path);
2898out:
2899	return err ? err : map->m_len;
2900}
2901
2902#define EXT4_EXT_ZERO_LEN 7
2903/*
2904 * This function is called by ext4_ext_map_blocks() if someone tries to write
2905 * to an uninitialized extent. It may result in splitting the uninitialized
2906 * extent into multiple extents (up to three - one initialized and two
2907 * uninitialized).
2908 * There are three possibilities:
2909 *   a> There is no split required: Entire extent should be initialized
2910 *   b> Splits in two extents: Write is happening at either end of the extent
2911 *   c> Splits in three extents: Somone is writing in middle of the extent
 
 
 
 
 
 
 
 
 
 
2912 */
2913static int ext4_ext_convert_to_initialized(handle_t *handle,
2914					   struct inode *inode,
2915					   struct ext4_map_blocks *map,
2916					   struct ext4_ext_path *path)
2917{
 
2918	struct ext4_map_blocks split_map;
2919	struct ext4_extent zero_ex;
2920	struct ext4_extent *ex;
2921	ext4_lblk_t ee_block, eof_block;
2922	unsigned int allocated, ee_len, depth;
 
2923	int err = 0;
2924	int split_flag = 0;
2925
2926	ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2927		"block %llu, max_blocks %u\n", inode->i_ino,
2928		(unsigned long long)map->m_lblk, map->m_len);
2929
2930	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2931		inode->i_sb->s_blocksize_bits;
2932	if (eof_block < map->m_lblk + map->m_len)
2933		eof_block = map->m_lblk + map->m_len;
2934
2935	depth = ext_depth(inode);
 
2936	ex = path[depth].p_ext;
2937	ee_block = le32_to_cpu(ex->ee_block);
2938	ee_len = ext4_ext_get_actual_len(ex);
2939	allocated = ee_len - (map->m_lblk - ee_block);
2940
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2941	WARN_ON(map->m_lblk < ee_block);
2942	/*
2943	 * It is safe to convert extent to initialized via explicit
2944	 * zeroout only if extent is fully insde i_size or new_size.
2945	 */
2946	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
2947
2948	/* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2949	if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2950	    (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2951		err = ext4_ext_zeroout(inode, ex);
2952		if (err)
2953			goto out;
2954
2955		err = ext4_ext_get_access(handle, inode, path + depth);
2956		if (err)
2957			goto out;
2958		ext4_ext_mark_initialized(ex);
2959		ext4_ext_try_to_merge(inode, path, ex);
2960		err = ext4_ext_dirty(handle, inode, path + depth);
2961		goto out;
2962	}
2963
2964	/*
2965	 * four cases:
2966	 * 1. split the extent into three extents.
2967	 * 2. split the extent into two extents, zeroout the first half.
2968	 * 3. split the extent into two extents, zeroout the second half.
2969	 * 4. split the extent into two extents with out zeroout.
2970	 */
2971	split_map.m_lblk = map->m_lblk;
2972	split_map.m_len = map->m_len;
2973
2974	if (allocated > map->m_len) {
2975		if (allocated <= EXT4_EXT_ZERO_LEN &&
2976		    (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2977			/* case 3 */
2978			zero_ex.ee_block =
2979					 cpu_to_le32(map->m_lblk);
2980			zero_ex.ee_len = cpu_to_le16(allocated);
2981			ext4_ext_store_pblock(&zero_ex,
2982				ext4_ext_pblock(ex) + map->m_lblk - ee_block);
2983			err = ext4_ext_zeroout(inode, &zero_ex);
2984			if (err)
2985				goto out;
2986			split_map.m_lblk = map->m_lblk;
2987			split_map.m_len = allocated;
2988		} else if ((map->m_lblk - ee_block + map->m_len <
2989			   EXT4_EXT_ZERO_LEN) &&
2990			   (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2991			/* case 2 */
2992			if (map->m_lblk != ee_block) {
2993				zero_ex.ee_block = ex->ee_block;
2994				zero_ex.ee_len = cpu_to_le16(map->m_lblk -
2995							ee_block);
2996				ext4_ext_store_pblock(&zero_ex,
2997						      ext4_ext_pblock(ex));
2998				err = ext4_ext_zeroout(inode, &zero_ex);
2999				if (err)
3000					goto out;
3001			}
3002
3003			split_map.m_lblk = ee_block;
3004			split_map.m_len = map->m_lblk - ee_block + map->m_len;
3005			allocated = map->m_len;
3006		}
3007	}
3008
3009	allocated = ext4_split_extent(handle, inode, path,
3010				       &split_map, split_flag, 0);
3011	if (allocated < 0)
3012		err = allocated;
3013
3014out:
3015	return err ? err : allocated;
3016}
3017
3018/*
3019 * This function is called by ext4_ext_map_blocks() from
3020 * ext4_get_blocks_dio_write() when DIO to write
3021 * to an uninitialized extent.
3022 *
3023 * Writing to an uninitialized extent may result in splitting the uninitialized
3024 * extent into multiple /initialized uninitialized extents (up to three)
3025 * There are three possibilities:
3026 *   a> There is no split required: Entire extent should be uninitialized
3027 *   b> Splits in two extents: Write is happening at either end of the extent
3028 *   c> Splits in three extents: Somone is writing in middle of the extent
3029 *
3030 * One of more index blocks maybe needed if the extent tree grow after
3031 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3032 * complete, we need to split the uninitialized extent before DIO submit
3033 * the IO. The uninitialized extent called at this time will be split
3034 * into three uninitialized extent(at most). After IO complete, the part
3035 * being filled will be convert to initialized by the end_io callback function
3036 * via ext4_convert_unwritten_extents().
3037 *
3038 * Returns the size of uninitialized extent to be written on success.
3039 */
3040static int ext4_split_unwritten_extents(handle_t *handle,
3041					struct inode *inode,
3042					struct ext4_map_blocks *map,
3043					struct ext4_ext_path *path,
3044					int flags)
3045{
3046	ext4_lblk_t eof_block;
3047	ext4_lblk_t ee_block;
3048	struct ext4_extent *ex;
3049	unsigned int ee_len;
3050	int split_flag = 0, depth;
3051
3052	ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3053		"block %llu, max_blocks %u\n", inode->i_ino,
3054		(unsigned long long)map->m_lblk, map->m_len);
3055
3056	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3057		inode->i_sb->s_blocksize_bits;
3058	if (eof_block < map->m_lblk + map->m_len)
3059		eof_block = map->m_lblk + map->m_len;
3060	/*
3061	 * It is safe to convert extent to initialized via explicit
3062	 * zeroout only if extent is fully insde i_size or new_size.
3063	 */
3064	depth = ext_depth(inode);
3065	ex = path[depth].p_ext;
3066	ee_block = le32_to_cpu(ex->ee_block);
3067	ee_len = ext4_ext_get_actual_len(ex);
3068
3069	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3070	split_flag |= EXT4_EXT_MARK_UNINIT2;
3071
3072	flags |= EXT4_GET_BLOCKS_PRE_IO;
3073	return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3074}
3075
3076static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3077					      struct inode *inode,
3078					      struct ext4_ext_path *path)
3079{
3080	struct ext4_extent *ex;
3081	int depth;
3082	int err = 0;
3083
3084	depth = ext_depth(inode);
3085	ex = path[depth].p_ext;
3086
3087	ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3088		"block %llu, max_blocks %u\n", inode->i_ino,
3089		(unsigned long long)le32_to_cpu(ex->ee_block),
3090		ext4_ext_get_actual_len(ex));
3091
3092	err = ext4_ext_get_access(handle, inode, path + depth);
3093	if (err)
3094		goto out;
3095	/* first mark the extent as initialized */
3096	ext4_ext_mark_initialized(ex);
3097
3098	/* note: ext4_ext_correct_indexes() isn't needed here because
3099	 * borders are not changed
3100	 */
3101	ext4_ext_try_to_merge(inode, path, ex);
3102
3103	/* Mark modified extent as dirty */
3104	err = ext4_ext_dirty(handle, inode, path + depth);
3105out:
3106	ext4_ext_show_leaf(inode, path);
3107	return err;
3108}
3109
3110static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3111			sector_t block, int count)
3112{
3113	int i;
3114	for (i = 0; i < count; i++)
3115                unmap_underlying_metadata(bdev, block + i);
3116}
3117
3118/*
3119 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3120 */
3121static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3122			      ext4_lblk_t lblk,
3123			      struct ext4_ext_path *path,
3124			      unsigned int len)
3125{
3126	int i, depth;
3127	struct ext4_extent_header *eh;
3128	struct ext4_extent *last_ex;
3129
3130	if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3131		return 0;
3132
3133	depth = ext_depth(inode);
3134	eh = path[depth].p_hdr;
3135
3136	if (unlikely(!eh->eh_entries)) {
3137		EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3138				 "EOFBLOCKS_FL set");
3139		return -EIO;
3140	}
 
 
3141	last_ex = EXT_LAST_EXTENT(eh);
3142	/*
3143	 * We should clear the EOFBLOCKS_FL flag if we are writing the
3144	 * last block in the last extent in the file.  We test this by
3145	 * first checking to see if the caller to
3146	 * ext4_ext_get_blocks() was interested in the last block (or
3147	 * a block beyond the last block) in the current extent.  If
3148	 * this turns out to be false, we can bail out from this
3149	 * function immediately.
3150	 */
3151	if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3152	    ext4_ext_get_actual_len(last_ex))
3153		return 0;
3154	/*
3155	 * If the caller does appear to be planning to write at or
3156	 * beyond the end of the current extent, we then test to see
3157	 * if the current extent is the last extent in the file, by
3158	 * checking to make sure it was reached via the rightmost node
3159	 * at each level of the tree.
3160	 */
3161	for (i = depth-1; i >= 0; i--)
3162		if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3163			return 0;
 
3164	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3165	return ext4_mark_inode_dirty(handle, inode);
3166}
3167
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3168static int
3169ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3170			struct ext4_map_blocks *map,
3171			struct ext4_ext_path *path, int flags,
3172			unsigned int allocated, ext4_fsblk_t newblock)
3173{
3174	int ret = 0;
3175	int err = 0;
3176	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3177
3178	ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3179		  "block %llu, max_blocks %u, flags %d, allocated %u",
3180		  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3181		  flags, allocated);
3182	ext4_ext_show_leaf(inode, path);
3183
 
 
 
3184	/* get_block() before submit the IO, split the extent */
3185	if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3186		ret = ext4_split_unwritten_extents(handle, inode, map,
3187						   path, flags);
3188		/*
3189		 * Flag the inode(non aio case) or end_io struct (aio case)
3190		 * that this IO needs to conversion to written when IO is
3191		 * completed
3192		 */
3193		if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3194			io->flag = EXT4_IO_END_UNWRITTEN;
3195			atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3196		} else
3197			ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3198		if (ext4_should_dioread_nolock(inode))
3199			map->m_flags |= EXT4_MAP_UNINIT;
3200		goto out;
3201	}
3202	/* IO end_io complete, convert the filled extent to written */
3203	if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3204		ret = ext4_convert_unwritten_extents_endio(handle, inode,
3205							path);
3206		if (ret >= 0) {
3207			ext4_update_inode_fsync_trans(handle, inode, 1);
3208			err = check_eofblocks_fl(handle, inode, map->m_lblk,
3209						 path, map->m_len);
3210		} else
3211			err = ret;
3212		goto out2;
3213	}
3214	/* buffered IO case */
3215	/*
3216	 * repeat fallocate creation request
3217	 * we already have an unwritten extent
3218	 */
3219	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3220		goto map_out;
3221
3222	/* buffered READ or buffered write_begin() lookup */
3223	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3224		/*
3225		 * We have blocks reserved already.  We
3226		 * return allocated blocks so that delalloc
3227		 * won't do block reservation for us.  But
3228		 * the buffer head will be unmapped so that
3229		 * a read from the block returns 0s.
3230		 */
3231		map->m_flags |= EXT4_MAP_UNWRITTEN;
3232		goto out1;
3233	}
3234
3235	/* buffered write, writepage time, convert*/
3236	ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3237	if (ret >= 0) {
3238		ext4_update_inode_fsync_trans(handle, inode, 1);
3239		err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3240					 map->m_len);
3241		if (err < 0)
3242			goto out2;
3243	}
3244
3245out:
3246	if (ret <= 0) {
3247		err = ret;
3248		goto out2;
3249	} else
3250		allocated = ret;
3251	map->m_flags |= EXT4_MAP_NEW;
3252	/*
3253	 * if we allocated more blocks than requested
3254	 * we need to make sure we unmap the extra block
3255	 * allocated. The actual needed block will get
3256	 * unmapped later when we find the buffer_head marked
3257	 * new.
3258	 */
3259	if (allocated > map->m_len) {
3260		unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3261					newblock + map->m_len,
3262					allocated - map->m_len);
3263		allocated = map->m_len;
3264	}
3265
3266	/*
3267	 * If we have done fallocate with the offset that is already
3268	 * delayed allocated, we would have block reservation
3269	 * and quota reservation done in the delayed write path.
3270	 * But fallocate would have already updated quota and block
3271	 * count for this offset. So cancel these reservation
3272	 */
3273	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3274		ext4_da_update_reserve_space(inode, allocated, 0);
 
 
 
 
 
 
 
3275
3276map_out:
3277	map->m_flags |= EXT4_MAP_MAPPED;
 
 
 
 
 
 
3278out1:
3279	if (allocated > map->m_len)
3280		allocated = map->m_len;
3281	ext4_ext_show_leaf(inode, path);
3282	map->m_pblk = newblock;
3283	map->m_len = allocated;
3284out2:
3285	if (path) {
3286		ext4_ext_drop_refs(path);
3287		kfree(path);
3288	}
3289	return err ? err : allocated;
3290}
3291
3292/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3293 * Block allocation/map/preallocation routine for extents based files
3294 *
3295 *
3296 * Need to be called with
3297 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3298 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3299 *
3300 * return > 0, number of of blocks already mapped/allocated
3301 *          if create == 0 and these are pre-allocated blocks
3302 *          	buffer head is unmapped
3303 *          otherwise blocks are mapped
3304 *
3305 * return = 0, if plain look up failed (blocks have not been allocated)
3306 *          buffer head is unmapped
3307 *
3308 * return < 0, error case.
3309 */
3310int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3311			struct ext4_map_blocks *map, int flags)
3312{
3313	struct ext4_ext_path *path = NULL;
3314	struct ext4_extent newex, *ex;
 
3315	ext4_fsblk_t newblock = 0;
3316	int err = 0, depth, ret;
3317	unsigned int allocated = 0;
3318	unsigned int punched_out = 0;
3319	unsigned int result = 0;
3320	struct ext4_allocation_request ar;
3321	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3322	struct ext4_map_blocks punch_map;
3323
3324	ext_debug("blocks %u/%u requested for inode %lu\n",
3325		  map->m_lblk, map->m_len, inode->i_ino);
3326	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3327
3328	/* check in cache */
3329	if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3330		ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3331		if (!newex.ee_start_lo && !newex.ee_start_hi) {
 
 
 
 
3332			if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3333				/*
3334				 * block isn't allocated yet and
3335				 * user doesn't want to allocate it
3336				 */
3337				goto out2;
3338			}
3339			/* we should allocate requested block */
3340		} else {
3341			/* block is already allocated */
 
 
3342			newblock = map->m_lblk
3343				   - le32_to_cpu(newex.ee_block)
3344				   + ext4_ext_pblock(&newex);
3345			/* number of remaining blocks in the extent */
3346			allocated = ext4_ext_get_actual_len(&newex) -
3347				(map->m_lblk - le32_to_cpu(newex.ee_block));
3348			goto out;
3349		}
3350	}
3351
3352	/* find extent for this block */
3353	path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3354	if (IS_ERR(path)) {
3355		err = PTR_ERR(path);
3356		path = NULL;
3357		goto out2;
3358	}
3359
3360	depth = ext_depth(inode);
3361
3362	/*
3363	 * consistent leaf must not be empty;
3364	 * this situation is possible, though, _during_ tree modification;
3365	 * this is why assert can't be put in ext4_ext_find_extent()
3366	 */
3367	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3368		EXT4_ERROR_INODE(inode, "bad extent address "
3369				 "lblock: %lu, depth: %d pblock %lld",
3370				 (unsigned long) map->m_lblk, depth,
3371				 path[depth].p_block);
3372		err = -EIO;
3373		goto out2;
3374	}
3375
3376	ex = path[depth].p_ext;
3377	if (ex) {
3378		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3379		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3380		unsigned short ee_len;
3381
3382		/*
3383		 * Uninitialized extents are treated as holes, except that
3384		 * we split out initialized portions during a write.
3385		 */
3386		ee_len = ext4_ext_get_actual_len(ex);
 
 
 
3387		/* if found extent covers block, simply return it */
3388		if (in_range(map->m_lblk, ee_block, ee_len)) {
3389			newblock = map->m_lblk - ee_block + ee_start;
3390			/* number of remaining blocks in the extent */
3391			allocated = ee_len - (map->m_lblk - ee_block);
3392			ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3393				  ee_block, ee_len, newblock);
3394
3395			if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3396				/*
3397				 * Do not put uninitialized extent
3398				 * in the cache
3399				 */
3400				if (!ext4_ext_is_uninitialized(ex)) {
3401					ext4_ext_put_in_cache(inode, ee_block,
3402						ee_len, ee_start);
3403					goto out;
3404				}
3405				ret = ext4_ext_handle_uninitialized_extents(
3406					handle, inode, map, path, flags,
3407					allocated, newblock);
3408				return ret;
3409			}
3410
3411			/*
3412			 * Punch out the map length, but only to the
3413			 * end of the extent
3414			 */
3415			punched_out = allocated < map->m_len ?
3416				allocated : map->m_len;
3417
3418			/*
3419			 * Sense extents need to be converted to
3420			 * uninitialized, they must fit in an
3421			 * uninitialized extent
3422			 */
3423			if (punched_out > EXT_UNINIT_MAX_LEN)
3424				punched_out = EXT_UNINIT_MAX_LEN;
3425
3426			punch_map.m_lblk = map->m_lblk;
3427			punch_map.m_pblk = newblock;
3428			punch_map.m_len = punched_out;
3429			punch_map.m_flags = 0;
3430
3431			/* Check to see if the extent needs to be split */
3432			if (punch_map.m_len != ee_len ||
3433				punch_map.m_lblk != ee_block) {
3434
3435				ret = ext4_split_extent(handle, inode,
3436				path, &punch_map, 0,
3437				EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3438				EXT4_GET_BLOCKS_PRE_IO);
3439
3440				if (ret < 0) {
3441					err = ret;
3442					goto out2;
3443				}
3444				/*
3445				 * find extent for the block at
3446				 * the start of the hole
3447				 */
3448				ext4_ext_drop_refs(path);
3449				kfree(path);
3450
3451				path = ext4_ext_find_extent(inode,
3452				map->m_lblk, NULL);
3453				if (IS_ERR(path)) {
3454					err = PTR_ERR(path);
3455					path = NULL;
3456					goto out2;
3457				}
3458
3459				depth = ext_depth(inode);
3460				ex = path[depth].p_ext;
3461				ee_len = ext4_ext_get_actual_len(ex);
3462				ee_block = le32_to_cpu(ex->ee_block);
3463				ee_start = ext4_ext_pblock(ex);
3464
3465			}
3466
3467			ext4_ext_mark_uninitialized(ex);
3468
3469			ext4_ext_invalidate_cache(inode);
3470
3471			err = ext4_ext_rm_leaf(handle, inode, path,
3472				map->m_lblk, map->m_lblk + punched_out);
3473
3474			if (!err && path->p_hdr->eh_entries == 0) {
3475				/*
3476				 * Punch hole freed all of this sub tree,
3477				 * so we need to correct eh_depth
3478				 */
3479				err = ext4_ext_get_access(handle, inode, path);
3480				if (err == 0) {
3481					ext_inode_hdr(inode)->eh_depth = 0;
3482					ext_inode_hdr(inode)->eh_max =
3483					cpu_to_le16(ext4_ext_space_root(
3484						inode, 0));
3485
3486					err = ext4_ext_dirty(
3487						handle, inode, path);
3488				}
3489			}
3490
3491			goto out2;
3492		}
3493	}
3494
 
 
 
 
3495	/*
3496	 * requested block isn't allocated yet;
3497	 * we couldn't try to create block if create flag is zero
3498	 */
3499	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3500		/*
3501		 * put just found gap into cache to speed up
3502		 * subsequent requests
3503		 */
3504		ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3505		goto out2;
3506	}
 
3507	/*
3508	 * Okay, we need to do block allocation.
3509	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3510
3511	/* find neighbour allocated blocks */
3512	ar.lleft = map->m_lblk;
3513	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3514	if (err)
3515		goto out2;
3516	ar.lright = map->m_lblk;
3517	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
 
3518	if (err)
3519		goto out2;
3520
 
 
 
 
 
 
 
 
 
 
3521	/*
3522	 * See if request is beyond maximum number of blocks we can have in
3523	 * a single extent. For an initialized extent this limit is
3524	 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3525	 * EXT_UNINIT_MAX_LEN.
3526	 */
3527	if (map->m_len > EXT_INIT_MAX_LEN &&
3528	    !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3529		map->m_len = EXT_INIT_MAX_LEN;
3530	else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3531		 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3532		map->m_len = EXT_UNINIT_MAX_LEN;
3533
3534	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3535	newex.ee_block = cpu_to_le32(map->m_lblk);
3536	newex.ee_len = cpu_to_le16(map->m_len);
3537	err = ext4_ext_check_overlap(inode, &newex, path);
3538	if (err)
3539		allocated = ext4_ext_get_actual_len(&newex);
3540	else
3541		allocated = map->m_len;
3542
3543	/* allocate new block */
3544	ar.inode = inode;
3545	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3546	ar.logical = map->m_lblk;
3547	ar.len = allocated;
 
 
 
 
 
 
 
 
 
 
 
3548	if (S_ISREG(inode->i_mode))
3549		ar.flags = EXT4_MB_HINT_DATA;
3550	else
3551		/* disable in-core preallocation for non-regular files */
3552		ar.flags = 0;
3553	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3554		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
3555	newblock = ext4_mb_new_blocks(handle, &ar, &err);
3556	if (!newblock)
3557		goto out2;
3558	ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3559		  ar.goal, newblock, allocated);
 
 
 
 
 
3560
 
3561	/* try to insert new extent into found leaf and return */
3562	ext4_ext_store_pblock(&newex, newblock);
3563	newex.ee_len = cpu_to_le16(ar.len);
3564	/* Mark uninitialized */
3565	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3566		ext4_ext_mark_uninitialized(&newex);
3567		/*
3568		 * io_end structure was created for every IO write to an
3569		 * uninitialized extent. To avoid unnecessary conversion,
3570		 * here we flag the IO that really needs the conversion.
3571		 * For non asycn direct IO case, flag the inode state
3572		 * that we need to perform conversion when IO is done.
3573		 */
3574		if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3575			if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3576				io->flag = EXT4_IO_END_UNWRITTEN;
3577				atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3578			} else
3579				ext4_set_inode_state(inode,
3580						     EXT4_STATE_DIO_UNWRITTEN);
3581		}
3582		if (ext4_should_dioread_nolock(inode))
3583			map->m_flags |= EXT4_MAP_UNINIT;
3584	}
3585
3586	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
 
 
 
3587	if (!err)
3588		err = ext4_ext_insert_extent(handle, inode, path,
3589					     &newex, flags);
3590	if (err) {
3591		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3592			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3593		/* free data blocks we just allocated */
3594		/* not a good idea to call discard here directly,
3595		 * but otherwise we'd need to call it every free() */
3596		ext4_discard_preallocations(inode);
3597		ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3598				 ext4_ext_get_actual_len(&newex), fb_flags);
3599		goto out2;
3600	}
3601
3602	/* previous routine could use block we allocated */
3603	newblock = ext4_ext_pblock(&newex);
3604	allocated = ext4_ext_get_actual_len(&newex);
3605	if (allocated > map->m_len)
3606		allocated = map->m_len;
3607	map->m_flags |= EXT4_MAP_NEW;
3608
3609	/*
3610	 * Update reserved blocks/metadata blocks after successful
3611	 * block allocation which had been deferred till now.
3612	 */
3613	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3614		ext4_da_update_reserve_space(inode, allocated, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3615
3616	/*
3617	 * Cache the extent and update transaction to commit on fdatasync only
3618	 * when it is _not_ an uninitialized extent.
3619	 */
3620	if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3621		ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3622		ext4_update_inode_fsync_trans(handle, inode, 1);
3623	} else
3624		ext4_update_inode_fsync_trans(handle, inode, 0);
3625out:
3626	if (allocated > map->m_len)
3627		allocated = map->m_len;
3628	ext4_ext_show_leaf(inode, path);
3629	map->m_flags |= EXT4_MAP_MAPPED;
3630	map->m_pblk = newblock;
3631	map->m_len = allocated;
3632out2:
3633	if (path) {
3634		ext4_ext_drop_refs(path);
3635		kfree(path);
3636	}
 
3637	trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3638		newblock, map->m_len, err ? err : allocated);
3639
3640	result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3641			punched_out : allocated;
3642
3643	return err ? err : result;
3644}
3645
3646void ext4_ext_truncate(struct inode *inode)
3647{
3648	struct address_space *mapping = inode->i_mapping;
3649	struct super_block *sb = inode->i_sb;
3650	ext4_lblk_t last_block;
3651	handle_t *handle;
 
3652	int err = 0;
3653
3654	/*
3655	 * finish any pending end_io work so we won't run the risk of
3656	 * converting any truncated blocks to initialized later
3657	 */
3658	ext4_flush_completed_IO(inode);
3659
3660	/*
3661	 * probably first extent we're gonna free will be last in block
3662	 */
3663	err = ext4_writepage_trans_blocks(inode);
3664	handle = ext4_journal_start(inode, err);
3665	if (IS_ERR(handle))
3666		return;
3667
3668	if (inode->i_size & (sb->s_blocksize - 1))
3669		ext4_block_truncate_page(handle, mapping, inode->i_size);
 
 
 
 
 
 
 
 
3670
3671	if (ext4_orphan_add(handle, inode))
3672		goto out_stop;
3673
3674	down_write(&EXT4_I(inode)->i_data_sem);
3675	ext4_ext_invalidate_cache(inode);
3676
3677	ext4_discard_preallocations(inode);
3678
3679	/*
3680	 * TODO: optimization is possible here.
3681	 * Probably we need not scan at all,
3682	 * because page truncation is enough.
3683	 */
3684
3685	/* we have to know where to truncate from in crash case */
3686	EXT4_I(inode)->i_disksize = inode->i_size;
3687	ext4_mark_inode_dirty(handle, inode);
3688
3689	last_block = (inode->i_size + sb->s_blocksize - 1)
3690			>> EXT4_BLOCK_SIZE_BITS(sb);
3691	err = ext4_ext_remove_space(inode, last_block);
3692
3693	/* In a multi-transaction truncate, we only make the final
3694	 * transaction synchronous.
3695	 */
3696	if (IS_SYNC(inode))
3697		ext4_handle_sync(handle);
3698
3699	up_write(&EXT4_I(inode)->i_data_sem);
3700
3701out_stop:
3702	/*
3703	 * If this was a simple ftruncate() and the file will remain alive,
3704	 * then we need to clear up the orphan record which we created above.
3705	 * However, if this was a real unlink then we were called by
3706	 * ext4_delete_inode(), and we allow that function to clean up the
3707	 * orphan info for us.
3708	 */
3709	if (inode->i_nlink)
3710		ext4_orphan_del(handle, inode);
3711
3712	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3713	ext4_mark_inode_dirty(handle, inode);
3714	ext4_journal_stop(handle);
3715}
3716
3717static void ext4_falloc_update_inode(struct inode *inode,
3718				int mode, loff_t new_size, int update_ctime)
3719{
3720	struct timespec now;
3721
3722	if (update_ctime) {
3723		now = current_fs_time(inode->i_sb);
3724		if (!timespec_equal(&inode->i_ctime, &now))
3725			inode->i_ctime = now;
3726	}
3727	/*
3728	 * Update only when preallocation was requested beyond
3729	 * the file size.
3730	 */
3731	if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3732		if (new_size > i_size_read(inode))
3733			i_size_write(inode, new_size);
3734		if (new_size > EXT4_I(inode)->i_disksize)
3735			ext4_update_i_disksize(inode, new_size);
3736	} else {
3737		/*
3738		 * Mark that we allocate beyond EOF so the subsequent truncate
3739		 * can proceed even if the new size is the same as i_size.
3740		 */
3741		if (new_size > i_size_read(inode))
3742			ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3743	}
3744
3745}
3746
3747/*
3748 * preallocate space for a file. This implements ext4's fallocate file
3749 * operation, which gets called from sys_fallocate system call.
3750 * For block-mapped files, posix_fallocate should fall back to the method
3751 * of writing zeroes to the required new blocks (the same behavior which is
3752 * expected for file systems which do not support fallocate() system call).
3753 */
3754long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3755{
3756	struct inode *inode = file->f_path.dentry->d_inode;
3757	handle_t *handle;
3758	loff_t new_size;
3759	unsigned int max_blocks;
3760	int ret = 0;
3761	int ret2 = 0;
3762	int retries = 0;
 
3763	struct ext4_map_blocks map;
3764	unsigned int credits, blkbits = inode->i_blkbits;
3765
3766	/*
3767	 * currently supporting (pre)allocate mode for extent-based
3768	 * files _only_
3769	 */
3770	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3771		return -EOPNOTSUPP;
3772
3773	/* Return error if mode is not supported */
3774	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3775		return -EOPNOTSUPP;
3776
3777	if (mode & FALLOC_FL_PUNCH_HOLE)
3778		return ext4_punch_hole(file, offset, len);
3779
3780	trace_ext4_fallocate_enter(inode, offset, len, mode);
3781	map.m_lblk = offset >> blkbits;
3782	/*
3783	 * We can't just convert len to max_blocks because
3784	 * If blocksize = 4096 offset = 3072 and len = 2048
3785	 */
3786	max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3787		- map.m_lblk;
3788	/*
3789	 * credits to insert 1 extent into extent tree
3790	 */
3791	credits = ext4_chunk_trans_blocks(inode, max_blocks);
3792	mutex_lock(&inode->i_mutex);
3793	ret = inode_newsize_ok(inode, (len + offset));
3794	if (ret) {
3795		mutex_unlock(&inode->i_mutex);
3796		trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3797		return ret;
3798	}
 
 
 
 
 
 
 
 
 
 
3799retry:
3800	while (ret >= 0 && ret < max_blocks) {
3801		map.m_lblk = map.m_lblk + ret;
3802		map.m_len = max_blocks = max_blocks - ret;
3803		handle = ext4_journal_start(inode, credits);
3804		if (IS_ERR(handle)) {
3805			ret = PTR_ERR(handle);
3806			break;
3807		}
3808		ret = ext4_map_blocks(handle, inode, &map,
3809				      EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
3810				      EXT4_GET_BLOCKS_NO_NORMALIZE);
3811		if (ret <= 0) {
3812#ifdef EXT4FS_DEBUG
3813			WARN_ON(ret <= 0);
3814			printk(KERN_ERR "%s: ext4_ext_map_blocks "
3815				    "returned error inode#%lu, block=%u, "
3816				    "max_blocks=%u", __func__,
3817				    inode->i_ino, map.m_lblk, max_blocks);
3818#endif
3819			ext4_mark_inode_dirty(handle, inode);
3820			ret2 = ext4_journal_stop(handle);
3821			break;
3822		}
3823		if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3824						blkbits) >> blkbits))
3825			new_size = offset + len;
3826		else
3827			new_size = ((loff_t) map.m_lblk + ret) << blkbits;
3828
3829		ext4_falloc_update_inode(inode, mode, new_size,
3830					 (map.m_flags & EXT4_MAP_NEW));
3831		ext4_mark_inode_dirty(handle, inode);
3832		ret2 = ext4_journal_stop(handle);
3833		if (ret2)
3834			break;
3835	}
3836	if (ret == -ENOSPC &&
3837			ext4_should_retry_alloc(inode->i_sb, &retries)) {
3838		ret = 0;
3839		goto retry;
3840	}
3841	mutex_unlock(&inode->i_mutex);
3842	trace_ext4_fallocate_exit(inode, offset, max_blocks,
3843				ret > 0 ? ret2 : ret);
3844	return ret > 0 ? ret2 : ret;
3845}
3846
3847/*
3848 * This function convert a range of blocks to written extents
3849 * The caller of this function will pass the start offset and the size.
3850 * all unwritten extents within this range will be converted to
3851 * written extents.
3852 *
3853 * This function is called from the direct IO end io call back
3854 * function, to convert the fallocated extents after IO is completed.
3855 * Returns 0 on success.
3856 */
3857int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3858				    ssize_t len)
3859{
3860	handle_t *handle;
3861	unsigned int max_blocks;
3862	int ret = 0;
3863	int ret2 = 0;
3864	struct ext4_map_blocks map;
3865	unsigned int credits, blkbits = inode->i_blkbits;
3866
3867	map.m_lblk = offset >> blkbits;
3868	/*
3869	 * We can't just convert len to max_blocks because
3870	 * If blocksize = 4096 offset = 3072 and len = 2048
3871	 */
3872	max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3873		      map.m_lblk);
3874	/*
3875	 * credits to insert 1 extent into extent tree
3876	 */
3877	credits = ext4_chunk_trans_blocks(inode, max_blocks);
3878	while (ret >= 0 && ret < max_blocks) {
3879		map.m_lblk += ret;
3880		map.m_len = (max_blocks -= ret);
3881		handle = ext4_journal_start(inode, credits);
3882		if (IS_ERR(handle)) {
3883			ret = PTR_ERR(handle);
3884			break;
3885		}
3886		ret = ext4_map_blocks(handle, inode, &map,
3887				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3888		if (ret <= 0) {
3889			WARN_ON(ret <= 0);
3890			printk(KERN_ERR "%s: ext4_ext_map_blocks "
3891				    "returned error inode#%lu, block=%u, "
3892				    "max_blocks=%u", __func__,
3893				    inode->i_ino, map.m_lblk, map.m_len);
 
3894		}
3895		ext4_mark_inode_dirty(handle, inode);
3896		ret2 = ext4_journal_stop(handle);
3897		if (ret <= 0 || ret2 )
3898			break;
3899	}
3900	return ret > 0 ? ret2 : ret;
3901}
3902
3903/*
3904 * Callback function called for each extent to gather FIEMAP information.
3905 */
3906static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
3907		       struct ext4_ext_cache *newex, struct ext4_extent *ex,
3908		       void *data)
3909{
3910	__u64	logical;
3911	__u64	physical;
3912	__u64	length;
3913	__u32	flags = 0;
3914	int		ret = 0;
3915	struct fiemap_extent_info *fieinfo = data;
3916	unsigned char blksize_bits;
3917
3918	blksize_bits = inode->i_sb->s_blocksize_bits;
3919	logical = (__u64)newex->ec_block << blksize_bits;
3920
3921	if (newex->ec_start == 0) {
3922		/*
3923		 * No extent in extent-tree contains block @newex->ec_start,
3924		 * then the block may stay in 1)a hole or 2)delayed-extent.
3925		 *
3926		 * Holes or delayed-extents are processed as follows.
3927		 * 1. lookup dirty pages with specified range in pagecache.
3928		 *    If no page is got, then there is no delayed-extent and
3929		 *    return with EXT_CONTINUE.
3930		 * 2. find the 1st mapped buffer,
3931		 * 3. check if the mapped buffer is both in the request range
3932		 *    and a delayed buffer. If not, there is no delayed-extent,
3933		 *    then return.
3934		 * 4. a delayed-extent is found, the extent will be collected.
3935		 */
3936		ext4_lblk_t	end = 0;
3937		pgoff_t		last_offset;
3938		pgoff_t		offset;
3939		pgoff_t		index;
3940		pgoff_t		start_index = 0;
3941		struct page	**pages = NULL;
3942		struct buffer_head *bh = NULL;
3943		struct buffer_head *head = NULL;
3944		unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
3945
3946		pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
3947		if (pages == NULL)
3948			return -ENOMEM;
3949
3950		offset = logical >> PAGE_SHIFT;
3951repeat:
3952		last_offset = offset;
3953		head = NULL;
3954		ret = find_get_pages_tag(inode->i_mapping, &offset,
3955					PAGECACHE_TAG_DIRTY, nr_pages, pages);
3956
3957		if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3958			/* First time, try to find a mapped buffer. */
3959			if (ret == 0) {
3960out:
3961				for (index = 0; index < ret; index++)
3962					page_cache_release(pages[index]);
3963				/* just a hole. */
3964				kfree(pages);
3965				return EXT_CONTINUE;
3966			}
3967			index = 0;
3968
3969next_page:
3970			/* Try to find the 1st mapped buffer. */
3971			end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
3972				  blksize_bits;
3973			if (!page_has_buffers(pages[index]))
3974				goto out;
3975			head = page_buffers(pages[index]);
3976			if (!head)
3977				goto out;
3978
3979			index++;
3980			bh = head;
3981			do {
3982				if (end >= newex->ec_block +
3983					newex->ec_len)
3984					/* The buffer is out of
3985					 * the request range.
3986					 */
3987					goto out;
3988
3989				if (buffer_mapped(bh) &&
3990				    end >= newex->ec_block) {
3991					start_index = index - 1;
3992					/* get the 1st mapped buffer. */
3993					goto found_mapped_buffer;
3994				}
3995
3996				bh = bh->b_this_page;
3997				end++;
3998			} while (bh != head);
3999
4000			/* No mapped buffer in the range found in this page,
4001			 * We need to look up next page.
4002			 */
4003			if (index >= ret) {
4004				/* There is no page left, but we need to limit
4005				 * newex->ec_len.
4006				 */
4007				newex->ec_len = end - newex->ec_block;
4008				goto out;
4009			}
4010			goto next_page;
4011		} else {
4012			/*Find contiguous delayed buffers. */
4013			if (ret > 0 && pages[0]->index == last_offset)
4014				head = page_buffers(pages[0]);
4015			bh = head;
4016			index = 1;
4017			start_index = 0;
4018		}
4019
4020found_mapped_buffer:
4021		if (bh != NULL && buffer_delay(bh)) {
4022			/* 1st or contiguous delayed buffer found. */
4023			if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4024				/*
4025				 * 1st delayed buffer found, record
4026				 * the start of extent.
4027				 */
4028				flags |= FIEMAP_EXTENT_DELALLOC;
4029				newex->ec_block = end;
4030				logical = (__u64)end << blksize_bits;
4031			}
4032			/* Find contiguous delayed buffers. */
4033			do {
4034				if (!buffer_delay(bh))
4035					goto found_delayed_extent;
4036				bh = bh->b_this_page;
4037				end++;
4038			} while (bh != head);
4039
4040			for (; index < ret; index++) {
4041				if (!page_has_buffers(pages[index])) {
4042					bh = NULL;
4043					break;
4044				}
4045				head = page_buffers(pages[index]);
4046				if (!head) {
4047					bh = NULL;
4048					break;
4049				}
4050
4051				if (pages[index]->index !=
4052				    pages[start_index]->index + index
4053				    - start_index) {
4054					/* Blocks are not contiguous. */
4055					bh = NULL;
4056					break;
4057				}
4058				bh = head;
4059				do {
4060					if (!buffer_delay(bh))
4061						/* Delayed-extent ends. */
4062						goto found_delayed_extent;
4063					bh = bh->b_this_page;
4064					end++;
4065				} while (bh != head);
4066			}
4067		} else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4068			/* a hole found. */
4069			goto out;
4070
4071found_delayed_extent:
4072		newex->ec_len = min(end - newex->ec_block,
4073						(ext4_lblk_t)EXT_INIT_MAX_LEN);
4074		if (ret == nr_pages && bh != NULL &&
4075			newex->ec_len < EXT_INIT_MAX_LEN &&
4076			buffer_delay(bh)) {
4077			/* Have not collected an extent and continue. */
4078			for (index = 0; index < ret; index++)
4079				page_cache_release(pages[index]);
4080			goto repeat;
4081		}
4082
4083		for (index = 0; index < ret; index++)
4084			page_cache_release(pages[index]);
4085		kfree(pages);
4086	}
4087
4088	physical = (__u64)newex->ec_start << blksize_bits;
4089	length =   (__u64)newex->ec_len << blksize_bits;
4090
4091	if (ex && ext4_ext_is_uninitialized(ex))
4092		flags |= FIEMAP_EXTENT_UNWRITTEN;
4093
4094	if (next == EXT_MAX_BLOCKS)
4095		flags |= FIEMAP_EXTENT_LAST;
4096
4097	ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4098					length, flags);
4099	if (ret < 0)
4100		return ret;
4101	if (ret == 1)
4102		return EXT_BREAK;
4103	return EXT_CONTINUE;
4104}
4105
4106/* fiemap flags we can handle specified here */
4107#define EXT4_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4108
4109static int ext4_xattr_fiemap(struct inode *inode,
4110				struct fiemap_extent_info *fieinfo)
4111{
4112	__u64 physical = 0;
4113	__u64 length;
4114	__u32 flags = FIEMAP_EXTENT_LAST;
4115	int blockbits = inode->i_sb->s_blocksize_bits;
4116	int error = 0;
4117
4118	/* in-inode? */
4119	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4120		struct ext4_iloc iloc;
4121		int offset;	/* offset of xattr in inode */
4122
4123		error = ext4_get_inode_loc(inode, &iloc);
4124		if (error)
4125			return error;
4126		physical = iloc.bh->b_blocknr << blockbits;
4127		offset = EXT4_GOOD_OLD_INODE_SIZE +
4128				EXT4_I(inode)->i_extra_isize;
4129		physical += offset;
4130		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4131		flags |= FIEMAP_EXTENT_DATA_INLINE;
4132		brelse(iloc.bh);
4133	} else { /* external block */
4134		physical = EXT4_I(inode)->i_file_acl << blockbits;
4135		length = inode->i_sb->s_blocksize;
4136	}
4137
4138	if (physical)
4139		error = fiemap_fill_next_extent(fieinfo, 0, physical,
4140						length, flags);
4141	return (error < 0 ? error : 0);
4142}
4143
4144/*
4145 * ext4_ext_punch_hole
4146 *
4147 * Punches a hole of "length" bytes in a file starting
4148 * at byte "offset"
4149 *
4150 * @inode:  The inode of the file to punch a hole in
4151 * @offset: The starting byte offset of the hole
4152 * @length: The length of the hole
4153 *
4154 * Returns the number of blocks removed or negative on err
4155 */
4156int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4157{
4158	struct inode *inode = file->f_path.dentry->d_inode;
4159	struct super_block *sb = inode->i_sb;
4160	struct ext4_ext_cache cache_ex;
4161	ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4162	struct address_space *mapping = inode->i_mapping;
4163	struct ext4_map_blocks map;
4164	handle_t *handle;
4165	loff_t first_block_offset, last_block_offset, block_len;
4166	loff_t first_page, last_page, first_page_offset, last_page_offset;
4167	int ret, credits, blocks_released, err = 0;
4168
4169	first_block = (offset + sb->s_blocksize - 1) >>
4170		EXT4_BLOCK_SIZE_BITS(sb);
4171	last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4172
4173	first_block_offset = first_block << EXT4_BLOCK_SIZE_BITS(sb);
4174	last_block_offset = last_block << EXT4_BLOCK_SIZE_BITS(sb);
 
 
 
 
 
 
 
4175
4176	first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4177	last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4178
4179	first_page_offset = first_page << PAGE_CACHE_SHIFT;
4180	last_page_offset = last_page << PAGE_CACHE_SHIFT;
4181
4182	/*
4183	 * Write out all dirty pages to avoid race conditions
4184	 * Then release them.
4185	 */
4186	if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4187		err = filemap_write_and_wait_range(mapping,
4188			first_page_offset == 0 ? 0 : first_page_offset-1,
4189			last_page_offset);
4190
4191			if (err)
4192				return err;
4193	}
4194
4195	/* Now release the pages */
4196	if (last_page_offset > first_page_offset) {
4197		truncate_inode_pages_range(mapping, first_page_offset,
4198					   last_page_offset-1);
4199	}
4200
4201	/* finish any pending end_io work */
4202	ext4_flush_completed_IO(inode);
4203
4204	credits = ext4_writepage_trans_blocks(inode);
4205	handle = ext4_journal_start(inode, credits);
4206	if (IS_ERR(handle))
4207		return PTR_ERR(handle);
4208
4209	err = ext4_orphan_add(handle, inode);
4210	if (err)
4211		goto out;
4212
4213	/*
4214	 * Now we need to zero out the un block aligned data.
4215	 * If the file is smaller than a block, just
4216	 * zero out the middle
4217	 */
4218	if (first_block > last_block)
4219		ext4_block_zero_page_range(handle, mapping, offset, length);
4220	else {
4221		/* zero out the head of the hole before the first block */
4222		block_len  = first_block_offset - offset;
4223		if (block_len > 0)
4224			ext4_block_zero_page_range(handle, mapping,
4225						   offset, block_len);
4226
4227		/* zero out the tail of the hole after the last block */
4228		block_len = offset + length - last_block_offset;
4229		if (block_len > 0) {
4230			ext4_block_zero_page_range(handle, mapping,
4231					last_block_offset, block_len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4232		}
4233	}
4234
 
 
 
 
4235	/* If there are no blocks to remove, return now */
4236	if (first_block >= last_block)
4237		goto out;
4238
4239	down_write(&EXT4_I(inode)->i_data_sem);
4240	ext4_ext_invalidate_cache(inode);
4241	ext4_discard_preallocations(inode);
4242
4243	/*
4244	 * Loop over all the blocks and identify blocks
4245	 * that need to be punched out
4246	 */
4247	iblock = first_block;
4248	blocks_released = 0;
4249	while (iblock < last_block) {
4250		max_blocks = last_block - iblock;
4251		num_blocks = 1;
4252		memset(&map, 0, sizeof(map));
4253		map.m_lblk = iblock;
4254		map.m_len = max_blocks;
4255		ret = ext4_ext_map_blocks(handle, inode, &map,
4256			EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4257
4258		if (ret > 0) {
4259			blocks_released += ret;
4260			num_blocks = ret;
4261		} else if (ret == 0) {
4262			/*
4263			 * If map blocks could not find the block,
4264			 * then it is in a hole.  If the hole was
4265			 * not already cached, then map blocks should
4266			 * put it in the cache.  So we can get the hole
4267			 * out of the cache
4268			 */
4269			memset(&cache_ex, 0, sizeof(cache_ex));
4270			if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4271				!cache_ex.ec_start) {
4272
4273				/* The hole is cached */
4274				num_blocks = cache_ex.ec_block +
4275				cache_ex.ec_len - iblock;
4276
4277			} else {
4278				/* The block could not be identified */
4279				err = -EIO;
4280				break;
4281			}
4282		} else {
4283			/* Map blocks error */
4284			err = ret;
4285			break;
4286		}
4287
4288		if (num_blocks == 0) {
4289			/* This condition should never happen */
4290			ext_debug("Block lookup failed");
4291			err = -EIO;
4292			break;
4293		}
4294
4295		iblock += num_blocks;
4296	}
4297
4298	if (blocks_released > 0) {
4299		ext4_ext_invalidate_cache(inode);
4300		ext4_discard_preallocations(inode);
4301	}
4302
4303	if (IS_SYNC(inode))
4304		ext4_handle_sync(handle);
4305
4306	up_write(&EXT4_I(inode)->i_data_sem);
4307
4308out:
4309	ext4_orphan_del(handle, inode);
4310	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4311	ext4_mark_inode_dirty(handle, inode);
4312	ext4_journal_stop(handle);
4313	return err;
4314}
4315int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4316		__u64 start, __u64 len)
4317{
4318	ext4_lblk_t start_blk;
4319	int error = 0;
4320
4321	/* fallback to generic here if not in extents fmt */
4322	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4323		return generic_block_fiemap(inode, fieinfo, start, len,
4324			ext4_get_block);
4325
4326	if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4327		return -EBADR;
4328
4329	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4330		error = ext4_xattr_fiemap(inode, fieinfo);
4331	} else {
4332		ext4_lblk_t len_blks;
4333		__u64 last_blk;
4334
4335		start_blk = start >> inode->i_sb->s_blocksize_bits;
4336		last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4337		if (last_blk >= EXT_MAX_BLOCKS)
4338			last_blk = EXT_MAX_BLOCKS-1;
4339		len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4340
4341		/*
4342		 * Walk the extent tree gathering extent information.
4343		 * ext4_ext_fiemap_cb will push extents back to user.
4344		 */
4345		error = ext4_ext_walk_space(inode, start_blk, len_blks,
4346					  ext4_ext_fiemap_cb, fieinfo);
4347	}
4348
4349	return error;
4350}