1/*
   2 *  linux/fs/ext4/indirect.c
   3 *
   4 *  from
   5 *
   6 *  linux/fs/ext4/inode.c
   7 *
   8 * Copyright (C) 1992, 1993, 1994, 1995
   9 * Remy Card (card@masi.ibp.fr)
  10 * Laboratoire MASI - Institut Blaise Pascal
  11 * Universite Pierre et Marie Curie (Paris VI)
  12 *
  13 *  from
  14 *
  15 *  linux/fs/minix/inode.c
  16 *
  17 *  Copyright (C) 1991, 1992  Linus Torvalds
  18 *
  19 *  Goal-directed block allocation by Stephen Tweedie
  20 *	(sct@redhat.com), 1993, 1998
  21 */
  22
 
  23#include "ext4_jbd2.h"
  24#include "truncate.h"
  25
  26#include <trace/events/ext4.h>
  27
  28typedef struct {
  29	__le32	*p;
  30	__le32	key;
  31	struct buffer_head *bh;
  32} Indirect;
  33
  34static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
  35{
  36	p->key = *(p->p = v);
  37	p->bh = bh;
  38}
  39
  40/**
  41 *	ext4_block_to_path - parse the block number into array of offsets
  42 *	@inode: inode in question (we are only interested in its superblock)
  43 *	@i_block: block number to be parsed
  44 *	@offsets: array to store the offsets in
  45 *	@boundary: set this non-zero if the referred-to block is likely to be
  46 *	       followed (on disk) by an indirect block.
  47 *
  48 *	To store the locations of file's data ext4 uses a data structure common
  49 *	for UNIX filesystems - tree of pointers anchored in the inode, with
  50 *	data blocks at leaves and indirect blocks in intermediate nodes.
  51 *	This function translates the block number into path in that tree -
  52 *	return value is the path length and @offsets[n] is the offset of
  53 *	pointer to (n+1)th node in the nth one. If @block is out of range
  54 *	(negative or too large) warning is printed and zero returned.
  55 *
  56 *	Note: function doesn't find node addresses, so no IO is needed. All
  57 *	we need to know is the capacity of indirect blocks (taken from the
  58 *	inode->i_sb).
  59 */
  60
  61/*
  62 * Portability note: the last comparison (check that we fit into triple
  63 * indirect block) is spelled differently, because otherwise on an
  64 * architecture with 32-bit longs and 8Kb pages we might get into trouble
  65 * if our filesystem had 8Kb blocks. We might use long long, but that would
  66 * kill us on x86. Oh, well, at least the sign propagation does not matter -
  67 * i_block would have to be negative in the very beginning, so we would not
  68 * get there at all.
  69 */
  70
  71static int ext4_block_to_path(struct inode *inode,
  72			      ext4_lblk_t i_block,
  73			      ext4_lblk_t offsets[4], int *boundary)
  74{
  75	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  76	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
  77	const long direct_blocks = EXT4_NDIR_BLOCKS,
  78		indirect_blocks = ptrs,
  79		double_blocks = (1 << (ptrs_bits * 2));
  80	int n = 0;
  81	int final = 0;
  82
  83	if (i_block < direct_blocks) {
  84		offsets[n++] = i_block;
  85		final = direct_blocks;
  86	} else if ((i_block -= direct_blocks) < indirect_blocks) {
  87		offsets[n++] = EXT4_IND_BLOCK;
  88		offsets[n++] = i_block;
  89		final = ptrs;
  90	} else if ((i_block -= indirect_blocks) < double_blocks) {
  91		offsets[n++] = EXT4_DIND_BLOCK;
  92		offsets[n++] = i_block >> ptrs_bits;
  93		offsets[n++] = i_block & (ptrs - 1);
  94		final = ptrs;
  95	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
  96		offsets[n++] = EXT4_TIND_BLOCK;
  97		offsets[n++] = i_block >> (ptrs_bits * 2);
  98		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
  99		offsets[n++] = i_block & (ptrs - 1);
 100		final = ptrs;
 101	} else {
 102		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
 103			     i_block + direct_blocks +
 104			     indirect_blocks + double_blocks, inode->i_ino);
 105	}
 106	if (boundary)
 107		*boundary = final - 1 - (i_block & (ptrs - 1));
 108	return n;
 109}
 110
 111/**
 112 *	ext4_get_branch - read the chain of indirect blocks leading to data
 113 *	@inode: inode in question
 114 *	@depth: depth of the chain (1 - direct pointer, etc.)
 115 *	@offsets: offsets of pointers in inode/indirect blocks
 116 *	@chain: place to store the result
 117 *	@err: here we store the error value
 118 *
 119 *	Function fills the array of triples <key, p, bh> and returns %NULL
 120 *	if everything went OK or the pointer to the last filled triple
 121 *	(incomplete one) otherwise. Upon the return chain[i].key contains
 122 *	the number of (i+1)-th block in the chain (as it is stored in memory,
 123 *	i.e. little-endian 32-bit), chain[i].p contains the address of that
 124 *	number (it points into struct inode for i==0 and into the bh->b_data
 125 *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 126 *	block for i>0 and NULL for i==0. In other words, it holds the block
 127 *	numbers of the chain, addresses they were taken from (and where we can
 128 *	verify that chain did not change) and buffer_heads hosting these
 129 *	numbers.
 130 *
 131 *	Function stops when it stumbles upon zero pointer (absent block)
 132 *		(pointer to last triple returned, *@err == 0)
 133 *	or when it gets an IO error reading an indirect block
 134 *		(ditto, *@err == -EIO)
 135 *	or when it reads all @depth-1 indirect blocks successfully and finds
 136 *	the whole chain, all way to the data (returns %NULL, *err == 0).
 137 *
 138 *      Need to be called with
 139 *      down_read(&EXT4_I(inode)->i_data_sem)
 140 */
 141static Indirect *ext4_get_branch(struct inode *inode, int depth,
 142				 ext4_lblk_t  *offsets,
 143				 Indirect chain[4], int *err)
 144{
 145	struct super_block *sb = inode->i_sb;
 146	Indirect *p = chain;
 147	struct buffer_head *bh;
 148
 149	*err = 0;
 150	/* i_data is not going away, no lock needed */
 151	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
 152	if (!p->key)
 153		goto no_block;
 154	while (--depth) {
 155		bh = sb_getblk(sb, le32_to_cpu(p->key));
 156		if (unlikely(!bh))
 157			goto failure;
 158
 159		if (!bh_uptodate_or_lock(bh)) {
 160			if (bh_submit_read(bh) < 0) {
 161				put_bh(bh);
 162				goto failure;
 163			}
 164			/* validate block references */
 165			if (ext4_check_indirect_blockref(inode, bh)) {
 166				put_bh(bh);
 167				goto failure;
 168			}
 169		}
 170
 171		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
 172		/* Reader: end */
 173		if (!p->key)
 174			goto no_block;
 175	}
 176	return NULL;
 177
 178failure:
 179	*err = -EIO;
 180no_block:
 181	return p;
 182}
 183
 184/**
 185 *	ext4_find_near - find a place for allocation with sufficient locality
 186 *	@inode: owner
 187 *	@ind: descriptor of indirect block.
 188 *
 189 *	This function returns the preferred place for block allocation.
 190 *	It is used when heuristic for sequential allocation fails.
 191 *	Rules are:
 192 *	  + if there is a block to the left of our position - allocate near it.
 193 *	  + if pointer will live in indirect block - allocate near that block.
 194 *	  + if pointer will live in inode - allocate in the same
 195 *	    cylinder group.
 196 *
 197 * In the latter case we colour the starting block by the callers PID to
 198 * prevent it from clashing with concurrent allocations for a different inode
 199 * in the same block group.   The PID is used here so that functionally related
 200 * files will be close-by on-disk.
 201 *
 202 *	Caller must make sure that @ind is valid and will stay that way.
 203 */
 204static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
 205{
 206	struct ext4_inode_info *ei = EXT4_I(inode);
 207	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
 208	__le32 *p;
 209
 210	/* Try to find previous block */
 211	for (p = ind->p - 1; p >= start; p--) {
 212		if (*p)
 213			return le32_to_cpu(*p);
 214	}
 215
 216	/* No such thing, so let's try location of indirect block */
 217	if (ind->bh)
 218		return ind->bh->b_blocknr;
 219
 220	/*
 221	 * It is going to be referred to from the inode itself? OK, just put it
 222	 * into the same cylinder group then.
 223	 */
 224	return ext4_inode_to_goal_block(inode);
 225}
 226
 227/**
 228 *	ext4_find_goal - find a preferred place for allocation.
 229 *	@inode: owner
 230 *	@block:  block we want
 231 *	@partial: pointer to the last triple within a chain
 232 *
 233 *	Normally this function find the preferred place for block allocation,
 234 *	returns it.
 235 *	Because this is only used for non-extent files, we limit the block nr
 236 *	to 32 bits.
 237 */
 238static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
 239				   Indirect *partial)
 240{
 241	ext4_fsblk_t goal;
 242
 243	/*
 244	 * XXX need to get goal block from mballoc's data structures
 245	 */
 246
 247	goal = ext4_find_near(inode, partial);
 248	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 249	return goal;
 250}
 251
 252/**
 253 *	ext4_blks_to_allocate - Look up the block map and count the number
 254 *	of direct blocks need to be allocated for the given branch.
 255 *
 256 *	@branch: chain of indirect blocks
 257 *	@k: number of blocks need for indirect blocks
 258 *	@blks: number of data blocks to be mapped.
 259 *	@blocks_to_boundary:  the offset in the indirect block
 260 *
 261 *	return the total number of blocks to be allocate, including the
 262 *	direct and indirect blocks.
 263 */
 264static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
 265				 int blocks_to_boundary)
 266{
 267	unsigned int count = 0;
 268
 269	/*
 270	 * Simple case, [t,d]Indirect block(s) has not allocated yet
 271	 * then it's clear blocks on that path have not allocated
 272	 */
 273	if (k > 0) {
 274		/* right now we don't handle cross boundary allocation */
 275		if (blks < blocks_to_boundary + 1)
 276			count += blks;
 277		else
 278			count += blocks_to_boundary + 1;
 279		return count;
 280	}
 281
 282	count++;
 283	while (count < blks && count <= blocks_to_boundary &&
 284		le32_to_cpu(*(branch[0].p + count)) == 0) {
 285		count++;
 286	}
 287	return count;
 288}
 289
 290/**
 291 *	ext4_alloc_blocks: multiple allocate blocks needed for a branch
 292 *	@handle: handle for this transaction
 293 *	@inode: inode which needs allocated blocks
 294 *	@iblock: the logical block to start allocated at
 295 *	@goal: preferred physical block of allocation
 296 *	@indirect_blks: the number of blocks need to allocate for indirect
 297 *			blocks
 298 *	@blks: number of desired blocks
 299 *	@new_blocks: on return it will store the new block numbers for
 300 *	the indirect blocks(if needed) and the first direct block,
 301 *	@err: on return it will store the error code
 302 *
 303 *	This function will return the number of blocks allocated as
 304 *	requested by the passed-in parameters.
 305 */
 306static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
 307			     ext4_lblk_t iblock, ext4_fsblk_t goal,
 308			     int indirect_blks, int blks,
 309			     ext4_fsblk_t new_blocks[4], int *err)
 310{
 311	struct ext4_allocation_request ar;
 312	int target, i;
 313	unsigned long count = 0, blk_allocated = 0;
 314	int index = 0;
 315	ext4_fsblk_t current_block = 0;
 316	int ret = 0;
 317
 318	/*
 319	 * Here we try to allocate the requested multiple blocks at once,
 320	 * on a best-effort basis.
 321	 * To build a branch, we should allocate blocks for
 322	 * the indirect blocks(if not allocated yet), and at least
 323	 * the first direct block of this branch.  That's the
 324	 * minimum number of blocks need to allocate(required)
 325	 */
 326	/* first we try to allocate the indirect blocks */
 327	target = indirect_blks;
 328	while (target > 0) {
 329		count = target;
 330		/* allocating blocks for indirect blocks and direct blocks */
 331		current_block = ext4_new_meta_blocks(handle, inode, goal,
 332						     0, &count, err);
 333		if (*err)
 334			goto failed_out;
 335
 336		if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
 337			EXT4_ERROR_INODE(inode,
 338					 "current_block %llu + count %lu > %d!",
 339					 current_block, count,
 340					 EXT4_MAX_BLOCK_FILE_PHYS);
 341			*err = -EIO;
 342			goto failed_out;
 343		}
 344
 345		target -= count;
 346		/* allocate blocks for indirect blocks */
 347		while (index < indirect_blks && count) {
 348			new_blocks[index++] = current_block++;
 349			count--;
 350		}
 351		if (count > 0) {
 352			/*
 353			 * save the new block number
 354			 * for the first direct block
 355			 */
 356			new_blocks[index] = current_block;
 357			printk(KERN_INFO "%s returned more blocks than "
 358						"requested\n", __func__);
 359			WARN_ON(1);
 360			break;
 361		}
 362	}
 363
 364	target = blks - count ;
 365	blk_allocated = count;
 366	if (!target)
 367		goto allocated;
 368	/* Now allocate data blocks */
 369	memset(&ar, 0, sizeof(ar));
 370	ar.inode = inode;
 371	ar.goal = goal;
 372	ar.len = target;
 373	ar.logical = iblock;
 374	if (S_ISREG(inode->i_mode))
 375		/* enable in-core preallocation only for regular files */
 376		ar.flags = EXT4_MB_HINT_DATA;
 377
 378	current_block = ext4_mb_new_blocks(handle, &ar, err);
 379	if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
 380		EXT4_ERROR_INODE(inode,
 381				 "current_block %llu + ar.len %d > %d!",
 382				 current_block, ar.len,
 383				 EXT4_MAX_BLOCK_FILE_PHYS);
 384		*err = -EIO;
 385		goto failed_out;
 386	}
 387
 388	if (*err && (target == blks)) {
 389		/*
 390		 * if the allocation failed and we didn't allocate
 391		 * any blocks before
 392		 */
 393		goto failed_out;
 394	}
 395	if (!*err) {
 396		if (target == blks) {
 397			/*
 398			 * save the new block number
 399			 * for the first direct block
 400			 */
 401			new_blocks[index] = current_block;
 402		}
 403		blk_allocated += ar.len;
 404	}
 405allocated:
 406	/* total number of blocks allocated for direct blocks */
 407	ret = blk_allocated;
 408	*err = 0;
 409	return ret;
 410failed_out:
 411	for (i = 0; i < index; i++)
 412		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
 413	return ret;
 414}
 415
 416/**
 417 *	ext4_alloc_branch - allocate and set up a chain of blocks.
 418 *	@handle: handle for this transaction
 419 *	@inode: owner
 420 *	@indirect_blks: number of allocated indirect blocks
 421 *	@blks: number of allocated direct blocks
 422 *	@goal: preferred place for allocation
 423 *	@offsets: offsets (in the blocks) to store the pointers to next.
 424 *	@branch: place to store the chain in.
 425 *
 426 *	This function allocates blocks, zeroes out all but the last one,
 427 *	links them into chain and (if we are synchronous) writes them to disk.
 428 *	In other words, it prepares a branch that can be spliced onto the
 429 *	inode. It stores the information about that chain in the branch[], in
 430 *	the same format as ext4_get_branch() would do. We are calling it after
 431 *	we had read the existing part of chain and partial points to the last
 432 *	triple of that (one with zero ->key). Upon the exit we have the same
 433 *	picture as after the successful ext4_get_block(), except that in one
 434 *	place chain is disconnected - *branch->p is still zero (we did not
 435 *	set the last link), but branch->key contains the number that should
 436 *	be placed into *branch->p to fill that gap.
 437 *
 438 *	If allocation fails we free all blocks we've allocated (and forget
 439 *	their buffer_heads) and return the error value the from failed
 440 *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 441 *	as described above and return 0.
 442 */
 443static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
 444			     ext4_lblk_t iblock, int indirect_blks,
 445			     int *blks, ext4_fsblk_t goal,
 446			     ext4_lblk_t *offsets, Indirect *branch)
 447{
 448	int blocksize = inode->i_sb->s_blocksize;
 449	int i, n = 0;
 450	int err = 0;
 451	struct buffer_head *bh;
 452	int num;
 453	ext4_fsblk_t new_blocks[4];
 454	ext4_fsblk_t current_block;
 455
 456	num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
 457				*blks, new_blocks, &err);
 458	if (err)
 459		return err;
 460
 461	branch[0].key = cpu_to_le32(new_blocks[0]);
 462	/*
 463	 * metadata blocks and data blocks are allocated.
 464	 */
 465	for (n = 1; n <= indirect_blks;  n++) {
 466		/*
 467		 * Get buffer_head for parent block, zero it out
 468		 * and set the pointer to new one, then send
 469		 * parent to disk.
 470		 */
 471		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
 472		if (unlikely(!bh)) {
 473			err = -EIO;
 474			goto failed;
 475		}
 476
 477		branch[n].bh = bh;
 478		lock_buffer(bh);
 479		BUFFER_TRACE(bh, "call get_create_access");
 480		err = ext4_journal_get_create_access(handle, bh);
 481		if (err) {
 482			/* Don't brelse(bh) here; it's done in
 483			 * ext4_journal_forget() below */
 484			unlock_buffer(bh);
 485			goto failed;
 486		}
 487
 488		memset(bh->b_data, 0, blocksize);
 489		branch[n].p = (__le32 *) bh->b_data + offsets[n];
 490		branch[n].key = cpu_to_le32(new_blocks[n]);
 491		*branch[n].p = branch[n].key;
 492		if (n == indirect_blks) {
 493			current_block = new_blocks[n];
 494			/*
 495			 * End of chain, update the last new metablock of
 496			 * the chain to point to the new allocated
 497			 * data blocks numbers
 498			 */
 499			for (i = 1; i < num; i++)
 500				*(branch[n].p + i) = cpu_to_le32(++current_block);
 501		}
 502		BUFFER_TRACE(bh, "marking uptodate");
 503		set_buffer_uptodate(bh);
 504		unlock_buffer(bh);
 505
 506		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 507		err = ext4_handle_dirty_metadata(handle, inode, bh);
 508		if (err)
 509			goto failed;
 510	}
 511	*blks = num;
 512	return err;
 513failed:
 514	/* Allocation failed, free what we already allocated */
 515	ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
 516	for (i = 1; i <= n ; i++) {
 517		/*
 518		 * branch[i].bh is newly allocated, so there is no
 519		 * need to revoke the block, which is why we don't
 520		 * need to set EXT4_FREE_BLOCKS_METADATA.
 521		 */
 522		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
 523				 EXT4_FREE_BLOCKS_FORGET);
 524	}
 525	for (i = n+1; i < indirect_blks; i++)
 526		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
 527
 528	ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
 529
 530	return err;
 531}
 532
 533/**
 534 * ext4_splice_branch - splice the allocated branch onto inode.
 535 * @handle: handle for this transaction
 536 * @inode: owner
 537 * @block: (logical) number of block we are adding
 538 * @chain: chain of indirect blocks (with a missing link - see
 539 *	ext4_alloc_branch)
 540 * @where: location of missing link
 541 * @num:   number of indirect blocks we are adding
 542 * @blks:  number of direct blocks we are adding
 543 *
 544 * This function fills the missing link and does all housekeeping needed in
 545 * inode (->i_blocks, etc.). In case of success we end up with the full
 546 * chain to new block and return 0.
 547 */
 548static int ext4_splice_branch(handle_t *handle, struct inode *inode,
 549			      ext4_lblk_t block, Indirect *where, int num,
 550			      int blks)
 551{
 552	int i;
 553	int err = 0;
 554	ext4_fsblk_t current_block;
 555
 556	/*
 557	 * If we're splicing into a [td]indirect block (as opposed to the
 558	 * inode) then we need to get write access to the [td]indirect block
 559	 * before the splice.
 560	 */
 561	if (where->bh) {
 562		BUFFER_TRACE(where->bh, "get_write_access");
 563		err = ext4_journal_get_write_access(handle, where->bh);
 564		if (err)
 565			goto err_out;
 566	}
 567	/* That's it */
 568
 569	*where->p = where->key;
 570
 571	/*
 572	 * Update the host buffer_head or inode to point to more just allocated
 573	 * direct blocks blocks
 574	 */
 575	if (num == 0 && blks > 1) {
 576		current_block = le32_to_cpu(where->key) + 1;
 577		for (i = 1; i < blks; i++)
 578			*(where->p + i) = cpu_to_le32(current_block++);
 579	}
 580
 581	/* We are done with atomic stuff, now do the rest of housekeeping */
 582	/* had we spliced it onto indirect block? */
 583	if (where->bh) {
 584		/*
 585		 * If we spliced it onto an indirect block, we haven't
 586		 * altered the inode.  Note however that if it is being spliced
 587		 * onto an indirect block at the very end of the file (the
 588		 * file is growing) then we *will* alter the inode to reflect
 589		 * the new i_size.  But that is not done here - it is done in
 590		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
 591		 */
 592		jbd_debug(5, "splicing indirect only\n");
 593		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
 594		err = ext4_handle_dirty_metadata(handle, inode, where->bh);
 595		if (err)
 596			goto err_out;
 597	} else {
 598		/*
 599		 * OK, we spliced it into the inode itself on a direct block.
 600		 */
 601		ext4_mark_inode_dirty(handle, inode);
 602		jbd_debug(5, "splicing direct\n");
 603	}
 604	return err;
 605
 606err_out:
 607	for (i = 1; i <= num; i++) {
 608		/*
 609		 * branch[i].bh is newly allocated, so there is no
 610		 * need to revoke the block, which is why we don't
 611		 * need to set EXT4_FREE_BLOCKS_METADATA.
 612		 */
 613		ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
 614				 EXT4_FREE_BLOCKS_FORGET);
 615	}
 616	ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
 617			 blks, 0);
 618
 619	return err;
 620}
 621
 622/*
 623 * The ext4_ind_map_blocks() function handles non-extents inodes
 624 * (i.e., using the traditional indirect/double-indirect i_blocks
 625 * scheme) for ext4_map_blocks().
 626 *
 627 * Allocation strategy is simple: if we have to allocate something, we will
 628 * have to go the whole way to leaf. So let's do it before attaching anything
 629 * to tree, set linkage between the newborn blocks, write them if sync is
 630 * required, recheck the path, free and repeat if check fails, otherwise
 631 * set the last missing link (that will protect us from any truncate-generated
 632 * removals - all blocks on the path are immune now) and possibly force the
 633 * write on the parent block.
 634 * That has a nice additional property: no special recovery from the failed
 635 * allocations is needed - we simply release blocks and do not touch anything
 636 * reachable from inode.
 637 *
 638 * `handle' can be NULL if create == 0.
 639 *
 640 * return > 0, # of blocks mapped or allocated.
 641 * return = 0, if plain lookup failed.
 642 * return < 0, error case.
 643 *
 644 * The ext4_ind_get_blocks() function should be called with
 645 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
 646 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
 647 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
 648 * blocks.
 649 */
 650int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
 651			struct ext4_map_blocks *map,
 652			int flags)
 653{
 654	int err = -EIO;
 655	ext4_lblk_t offsets[4];
 656	Indirect chain[4];
 657	Indirect *partial;
 658	ext4_fsblk_t goal;
 659	int indirect_blks;
 660	int blocks_to_boundary = 0;
 661	int depth;
 662	int count = 0;
 663	ext4_fsblk_t first_block = 0;
 664
 665	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
 666	J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
 667	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
 668	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
 669				   &blocks_to_boundary);
 670
 671	if (depth == 0)
 672		goto out;
 673
 674	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
 675
 676	/* Simplest case - block found, no allocation needed */
 677	if (!partial) {
 678		first_block = le32_to_cpu(chain[depth - 1].key);
 679		count++;
 680		/*map more blocks*/
 681		while (count < map->m_len && count <= blocks_to_boundary) {
 682			ext4_fsblk_t blk;
 683
 684			blk = le32_to_cpu(*(chain[depth-1].p + count));
 685
 686			if (blk == first_block + count)
 687				count++;
 688			else
 689				break;
 690		}
 691		goto got_it;
 692	}
 693
 694	/* Next simple case - plain lookup or failed read of indirect block */
 695	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
 696		goto cleanup;
 697
 698	/*
 699	 * Okay, we need to do block allocation.
 700	*/
 701	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 702				       EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
 703		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
 704				 "non-extent mapped inodes with bigalloc");
 705		return -ENOSPC;
 706	}
 707
 708	goal = ext4_find_goal(inode, map->m_lblk, partial);
 709
 710	/* the number of blocks need to allocate for [d,t]indirect blocks */
 711	indirect_blks = (chain + depth) - partial - 1;
 712
 713	/*
 714	 * Next look up the indirect map to count the totoal number of
 715	 * direct blocks to allocate for this branch.
 716	 */
 717	count = ext4_blks_to_allocate(partial, indirect_blks,
 718				      map->m_len, blocks_to_boundary);
 719	/*
 720	 * Block out ext4_truncate while we alter the tree
 721	 */
 722	err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
 723				&count, goal,
 724				offsets + (partial - chain), partial);
 725
 726	/*
 727	 * The ext4_splice_branch call will free and forget any buffers
 728	 * on the new chain if there is a failure, but that risks using
 729	 * up transaction credits, especially for bitmaps where the
 730	 * credits cannot be returned.  Can we handle this somehow?  We
 731	 * may need to return -EAGAIN upwards in the worst case.  --sct
 732	 */
 733	if (!err)
 734		err = ext4_splice_branch(handle, inode, map->m_lblk,
 735					 partial, indirect_blks, count);
 736	if (err)
 737		goto cleanup;
 738
 739	map->m_flags |= EXT4_MAP_NEW;
 740
 741	ext4_update_inode_fsync_trans(handle, inode, 1);
 742got_it:
 743	map->m_flags |= EXT4_MAP_MAPPED;
 744	map->m_pblk = le32_to_cpu(chain[depth-1].key);
 745	map->m_len = count;
 746	if (count > blocks_to_boundary)
 747		map->m_flags |= EXT4_MAP_BOUNDARY;
 748	err = count;
 749	/* Clean up and exit */
 750	partial = chain + depth - 1;	/* the whole chain */
 751cleanup:
 752	while (partial > chain) {
 753		BUFFER_TRACE(partial->bh, "call brelse");
 754		brelse(partial->bh);
 755		partial--;
 756	}
 757out:
 758	trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
 759				map->m_pblk, map->m_len, err);
 760	return err;
 761}
 762
 763/*
 764 * O_DIRECT for ext3 (or indirect map) based files
 765 *
 766 * If the O_DIRECT write will extend the file then add this inode to the
 767 * orphan list.  So recovery will truncate it back to the original size
 768 * if the machine crashes during the write.
 769 *
 770 * If the O_DIRECT write is intantiating holes inside i_size and the machine
 771 * crashes then stale disk data _may_ be exposed inside the file. But current
 772 * VFS code falls back into buffered path in that case so we are safe.
 773 */
 774ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
 775			   const struct iovec *iov, loff_t offset,
 776			   unsigned long nr_segs)
 777{
 778	struct file *file = iocb->ki_filp;
 779	struct inode *inode = file->f_mapping->host;
 780	struct ext4_inode_info *ei = EXT4_I(inode);
 781	handle_t *handle;
 782	ssize_t ret;
 783	int orphan = 0;
 784	size_t count = iov_length(iov, nr_segs);
 785	int retries = 0;
 786
 787	if (rw == WRITE) {
 788		loff_t final_size = offset + count;
 789
 790		if (final_size > inode->i_size) {
 791			/* Credits for sb + inode write */
 792			handle = ext4_journal_start(inode, 2);
 793			if (IS_ERR(handle)) {
 794				ret = PTR_ERR(handle);
 795				goto out;
 796			}
 797			ret = ext4_orphan_add(handle, inode);
 798			if (ret) {
 799				ext4_journal_stop(handle);
 800				goto out;
 801			}
 802			orphan = 1;
 803			ei->i_disksize = inode->i_size;
 804			ext4_journal_stop(handle);
 805		}
 806	}
 807
 808retry:
 809	if (rw == READ && ext4_should_dioread_nolock(inode)) {
 810		if (unlikely(!list_empty(&ei->i_completed_io_list))) {
 811			mutex_lock(&inode->i_mutex);
 812			ext4_flush_completed_IO(inode);
 813			mutex_unlock(&inode->i_mutex);
 814		}
 815		ret = __blockdev_direct_IO(rw, iocb, inode,
 816				 inode->i_sb->s_bdev, iov,
 817				 offset, nr_segs,
 818				 ext4_get_block, NULL, NULL, 0);
 819	} else {
 820		ret = blockdev_direct_IO(rw, iocb, inode, iov,
 821				 offset, nr_segs, ext4_get_block);
 822
 823		if (unlikely((rw & WRITE) && ret < 0)) {
 824			loff_t isize = i_size_read(inode);
 825			loff_t end = offset + iov_length(iov, nr_segs);
 826
 827			if (end > isize)
 828				ext4_truncate_failed_write(inode);
 829		}
 830	}
 831	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 832		goto retry;
 833
 834	if (orphan) {
 835		int err;
 836
 837		/* Credits for sb + inode write */
 838		handle = ext4_journal_start(inode, 2);
 839		if (IS_ERR(handle)) {
 840			/* This is really bad luck. We've written the data
 841			 * but cannot extend i_size. Bail out and pretend
 842			 * the write failed... */
 843			ret = PTR_ERR(handle);
 844			if (inode->i_nlink)
 845				ext4_orphan_del(NULL, inode);
 846
 847			goto out;
 848		}
 849		if (inode->i_nlink)
 850			ext4_orphan_del(handle, inode);
 851		if (ret > 0) {
 852			loff_t end = offset + ret;
 853			if (end > inode->i_size) {
 854				ei->i_disksize = end;
 855				i_size_write(inode, end);
 856				/*
 857				 * We're going to return a positive `ret'
 858				 * here due to non-zero-length I/O, so there's
 859				 * no way of reporting error returns from
 860				 * ext4_mark_inode_dirty() to userspace.  So
 861				 * ignore it.
 862				 */
 863				ext4_mark_inode_dirty(handle, inode);
 864			}
 865		}
 866		err = ext4_journal_stop(handle);
 867		if (ret == 0)
 868			ret = err;
 869	}
 870out:
 871	return ret;
 872}
 873
 874/*
 875 * Calculate the number of metadata blocks need to reserve
 876 * to allocate a new block at @lblocks for non extent file based file
 877 */
 878int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
 879{
 880	struct ext4_inode_info *ei = EXT4_I(inode);
 881	sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
 882	int blk_bits;
 883
 884	if (lblock < EXT4_NDIR_BLOCKS)
 885		return 0;
 886
 887	lblock -= EXT4_NDIR_BLOCKS;
 888
 889	if (ei->i_da_metadata_calc_len &&
 890	    (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
 891		ei->i_da_metadata_calc_len++;
 892		return 0;
 893	}
 894	ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
 895	ei->i_da_metadata_calc_len = 1;
 896	blk_bits = order_base_2(lblock);
 897	return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
 898}
 899
 900int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 901{
 902	int indirects;
 903
 904	/* if nrblocks are contiguous */
 905	if (chunk) {
 906		/*
 907		 * With N contiguous data blocks, we need at most
 908		 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
 909		 * 2 dindirect blocks, and 1 tindirect block
 910		 */
 911		return DIV_ROUND_UP(nrblocks,
 912				    EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
 913	}
 914	/*
 915	 * if nrblocks are not contiguous, worse case, each block touch
 916	 * a indirect block, and each indirect block touch a double indirect
 917	 * block, plus a triple indirect block
 918	 */
 919	indirects = nrblocks * 2 + 1;
 920	return indirects;
 921}
 922
 923/*
 924 * Truncate transactions can be complex and absolutely huge.  So we need to
 925 * be able to restart the transaction at a conventient checkpoint to make
 926 * sure we don't overflow the journal.
 927 *
 928 * start_transaction gets us a new handle for a truncate transaction,
 929 * and extend_transaction tries to extend the existing one a bit.  If
 930 * extend fails, we need to propagate the failure up and restart the
 931 * transaction in the top-level truncate loop. --sct
 932 */
 933static handle_t *start_transaction(struct inode *inode)
 934{
 935	handle_t *result;
 936
 937	result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
 938	if (!IS_ERR(result))
 939		return result;
 940
 941	ext4_std_error(inode->i_sb, PTR_ERR(result));
 942	return result;
 943}
 944
 945/*
 946 * Try to extend this transaction for the purposes of truncation.
 947 *
 948 * Returns 0 if we managed to create more room.  If we can't create more
 949 * room, and the transaction must be restarted we return 1.
 950 */
 951static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
 952{
 953	if (!ext4_handle_valid(handle))
 954		return 0;
 955	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
 956		return 0;
 957	if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
 958		return 0;
 959	return 1;
 960}
 961
 962/*
 963 * Probably it should be a library function... search for first non-zero word
 964 * or memcmp with zero_page, whatever is better for particular architecture.
 965 * Linus?
 966 */
 967static inline int all_zeroes(__le32 *p, __le32 *q)
 968{
 969	while (p < q)
 970		if (*p++)
 971			return 0;
 972	return 1;
 973}
 974
 975/**
 976 *	ext4_find_shared - find the indirect blocks for partial truncation.
 977 *	@inode:	  inode in question
 978 *	@depth:	  depth of the affected branch
 979 *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
 980 *	@chain:	  place to store the pointers to partial indirect blocks
 981 *	@top:	  place to the (detached) top of branch
 982 *
 983 *	This is a helper function used by ext4_truncate().
 984 *
 985 *	When we do truncate() we may have to clean the ends of several
 986 *	indirect blocks but leave the blocks themselves alive. Block is
 987 *	partially truncated if some data below the new i_size is referred
 988 *	from it (and it is on the path to the first completely truncated
 989 *	data block, indeed).  We have to free the top of that path along
 990 *	with everything to the right of the path. Since no allocation
 991 *	past the truncation point is possible until ext4_truncate()
 992 *	finishes, we may safely do the latter, but top of branch may
 993 *	require special attention - pageout below the truncation point
 994 *	might try to populate it.
 995 *
 996 *	We atomically detach the top of branch from the tree, store the
 997 *	block number of its root in *@top, pointers to buffer_heads of
 998 *	partially truncated blocks - in @chain[].bh and pointers to
 999 *	their last elements that should not be removed - in
1000 *	@chain[].p. Return value is the pointer to last filled element
1001 *	of @chain.
1002 *
1003 *	The work left to caller to do the actual freeing of subtrees:
1004 *		a) free the subtree starting from *@top
1005 *		b) free the subtrees whose roots are stored in
1006 *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1007 *		c) free the subtrees growing from the inode past the @chain[0].
1008 *			(no partially truncated stuff there).  */
1009
1010static Indirect *ext4_find_shared(struct inode *inode, int depth,
1011				  ext4_lblk_t offsets[4], Indirect chain[4],
1012				  __le32 *top)
1013{
1014	Indirect *partial, *p;
1015	int k, err;
1016
1017	*top = 0;
1018	/* Make k index the deepest non-null offset + 1 */
1019	for (k = depth; k > 1 && !offsets[k-1]; k--)
1020		;
1021	partial = ext4_get_branch(inode, k, offsets, chain, &err);
1022	/* Writer: pointers */
1023	if (!partial)
1024		partial = chain + k-1;
1025	/*
1026	 * If the branch acquired continuation since we've looked at it -
1027	 * fine, it should all survive and (new) top doesn't belong to us.
1028	 */
1029	if (!partial->key && *partial->p)
1030		/* Writer: end */
1031		goto no_top;
1032	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1033		;
1034	/*
1035	 * OK, we've found the last block that must survive. The rest of our
1036	 * branch should be detached before unlocking. However, if that rest
1037	 * of branch is all ours and does not grow immediately from the inode
1038	 * it's easier to cheat and just decrement partial->p.
1039	 */
1040	if (p == chain + k - 1 && p > chain) {
1041		p->p--;
1042	} else {
1043		*top = *p->p;
1044		/* Nope, don't do this in ext4.  Must leave the tree intact */
1045#if 0
1046		*p->p = 0;
1047#endif
1048	}
1049	/* Writer: end */
1050
1051	while (partial > p) {
1052		brelse(partial->bh);
1053		partial--;
1054	}
1055no_top:
1056	return partial;
1057}
1058
1059/*
1060 * Zero a number of block pointers in either an inode or an indirect block.
1061 * If we restart the transaction we must again get write access to the
1062 * indirect block for further modification.
1063 *
1064 * We release `count' blocks on disk, but (last - first) may be greater
1065 * than `count' because there can be holes in there.
1066 *
1067 * Return 0 on success, 1 on invalid block range
1068 * and < 0 on fatal error.
1069 */
1070static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1071			     struct buffer_head *bh,
1072			     ext4_fsblk_t block_to_free,
1073			     unsigned long count, __le32 *first,
1074			     __le32 *last)
1075{
1076	__le32 *p;
1077	int	flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1078	int	err;
1079
1080	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1081		flags |= EXT4_FREE_BLOCKS_METADATA;
1082
1083	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1084				   count)) {
1085		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1086				 "blocks %llu len %lu",
1087				 (unsigned long long) block_to_free, count);
1088		return 1;
1089	}
1090
1091	if (try_to_extend_transaction(handle, inode)) {
1092		if (bh) {
1093			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1094			err = ext4_handle_dirty_metadata(handle, inode, bh);
1095			if (unlikely(err))
1096				goto out_err;
1097		}
1098		err = ext4_mark_inode_dirty(handle, inode);
1099		if (unlikely(err))
1100			goto out_err;
1101		err = ext4_truncate_restart_trans(handle, inode,
1102					ext4_blocks_for_truncate(inode));
1103		if (unlikely(err))
1104			goto out_err;
1105		if (bh) {
1106			BUFFER_TRACE(bh, "retaking write access");
1107			err = ext4_journal_get_write_access(handle, bh);
1108			if (unlikely(err))
1109				goto out_err;
1110		}
1111	}
1112
1113	for (p = first; p < last; p++)
1114		*p = 0;
1115
1116	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1117	return 0;
1118out_err:
1119	ext4_std_error(inode->i_sb, err);
1120	return err;
1121}
1122
1123/**
1124 * ext4_free_data - free a list of data blocks
1125 * @handle:	handle for this transaction
1126 * @inode:	inode we are dealing with
1127 * @this_bh:	indirect buffer_head which contains *@first and *@last
1128 * @first:	array of block numbers
1129 * @last:	points immediately past the end of array
1130 *
1131 * We are freeing all blocks referred from that array (numbers are stored as
1132 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1133 *
1134 * We accumulate contiguous runs of blocks to free.  Conveniently, if these
1135 * blocks are contiguous then releasing them at one time will only affect one
1136 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1137 * actually use a lot of journal space.
1138 *
1139 * @this_bh will be %NULL if @first and @last point into the inode's direct
1140 * block pointers.
1141 */
1142static void ext4_free_data(handle_t *handle, struct inode *inode,
1143			   struct buffer_head *this_bh,
1144			   __le32 *first, __le32 *last)
1145{
1146	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
1147	unsigned long count = 0;	    /* Number of blocks in the run */
1148	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
1149					       corresponding to
1150					       block_to_free */
1151	ext4_fsblk_t nr;		    /* Current block # */
1152	__le32 *p;			    /* Pointer into inode/ind
1153					       for current block */
1154	int err = 0;
1155
1156	if (this_bh) {				/* For indirect block */
1157		BUFFER_TRACE(this_bh, "get_write_access");
1158		err = ext4_journal_get_write_access(handle, this_bh);
1159		/* Important: if we can't update the indirect pointers
1160		 * to the blocks, we can't free them. */
1161		if (err)
1162			return;
1163	}
1164
1165	for (p = first; p < last; p++) {
1166		nr = le32_to_cpu(*p);
1167		if (nr) {
1168			/* accumulate blocks to free if they're contiguous */
1169			if (count == 0) {
1170				block_to_free = nr;
1171				block_to_free_p = p;
1172				count = 1;
1173			} else if (nr == block_to_free + count) {
1174				count++;
1175			} else {
1176				err = ext4_clear_blocks(handle, inode, this_bh,
1177						        block_to_free, count,
1178						        block_to_free_p, p);
1179				if (err)
1180					break;
1181				block_to_free = nr;
1182				block_to_free_p = p;
1183				count = 1;
1184			}
1185		}
1186	}
1187
1188	if (!err && count > 0)
1189		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1190					count, block_to_free_p, p);
1191	if (err < 0)
1192		/* fatal error */
1193		return;
1194
1195	if (this_bh) {
1196		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1197
1198		/*
1199		 * The buffer head should have an attached journal head at this
1200		 * point. However, if the data is corrupted and an indirect
1201		 * block pointed to itself, it would have been detached when
1202		 * the block was cleared. Check for this instead of OOPSing.
1203		 */
1204		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1205			ext4_handle_dirty_metadata(handle, inode, this_bh);
1206		else
1207			EXT4_ERROR_INODE(inode,
1208					 "circular indirect block detected at "
1209					 "block %llu",
1210				(unsigned long long) this_bh->b_blocknr);
1211	}
1212}
1213
1214/**
1215 *	ext4_free_branches - free an array of branches
1216 *	@handle: JBD handle for this transaction
1217 *	@inode:	inode we are dealing with
1218 *	@parent_bh: the buffer_head which contains *@first and *@last
1219 *	@first:	array of block numbers
1220 *	@last:	pointer immediately past the end of array
1221 *	@depth:	depth of the branches to free
1222 *
1223 *	We are freeing all blocks referred from these branches (numbers are
1224 *	stored as little-endian 32-bit) and updating @inode->i_blocks
1225 *	appropriately.
1226 */
1227static void ext4_free_branches(handle_t *handle, struct inode *inode,
1228			       struct buffer_head *parent_bh,
1229			       __le32 *first, __le32 *last, int depth)
1230{
1231	ext4_fsblk_t nr;
1232	__le32 *p;
1233
1234	if (ext4_handle_is_aborted(handle))
1235		return;
1236
1237	if (depth--) {
1238		struct buffer_head *bh;
1239		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1240		p = last;
1241		while (--p >= first) {
1242			nr = le32_to_cpu(*p);
1243			if (!nr)
1244				continue;		/* A hole */
1245
1246			if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1247						   nr, 1)) {
1248				EXT4_ERROR_INODE(inode,
1249						 "invalid indirect mapped "
1250						 "block %lu (level %d)",
1251						 (unsigned long) nr, depth);
1252				break;
1253			}
1254
1255			/* Go read the buffer for the next level down */
1256			bh = sb_bread(inode->i_sb, nr);
1257
1258			/*
1259			 * A read failure? Report error and clear slot
1260			 * (should be rare).
1261			 */
1262			if (!bh) {
1263				EXT4_ERROR_INODE_BLOCK(inode, nr,
1264						       "Read failure");
1265				continue;
1266			}
1267
1268			/* This zaps the entire block.  Bottom up. */
1269			BUFFER_TRACE(bh, "free child branches");
1270			ext4_free_branches(handle, inode, bh,
1271					(__le32 *) bh->b_data,
1272					(__le32 *) bh->b_data + addr_per_block,
1273					depth);
1274			brelse(bh);
1275
1276			/*
1277			 * Everything below this this pointer has been
1278			 * released.  Now let this top-of-subtree go.
1279			 *
1280			 * We want the freeing of this indirect block to be
1281			 * atomic in the journal with the updating of the
1282			 * bitmap block which owns it.  So make some room in
1283			 * the journal.
1284			 *
1285			 * We zero the parent pointer *after* freeing its
1286			 * pointee in the bitmaps, so if extend_transaction()
1287			 * for some reason fails to put the bitmap changes and
1288			 * the release into the same transaction, recovery
1289			 * will merely complain about releasing a free block,
1290			 * rather than leaking blocks.
1291			 */
1292			if (ext4_handle_is_aborted(handle))
1293				return;
1294			if (try_to_extend_transaction(handle, inode)) {
1295				ext4_mark_inode_dirty(handle, inode);
1296				ext4_truncate_restart_trans(handle, inode,
1297					    ext4_blocks_for_truncate(inode));
1298			}
1299
1300			/*
1301			 * The forget flag here is critical because if
1302			 * we are journaling (and not doing data
1303			 * journaling), we have to make sure a revoke
1304			 * record is written to prevent the journal
1305			 * replay from overwriting the (former)
1306			 * indirect block if it gets reallocated as a
1307			 * data block.  This must happen in the same
1308			 * transaction where the data blocks are
1309			 * actually freed.
1310			 */
1311			ext4_free_blocks(handle, inode, NULL, nr, 1,
1312					 EXT4_FREE_BLOCKS_METADATA|
1313					 EXT4_FREE_BLOCKS_FORGET);
1314
1315			if (parent_bh) {
1316				/*
1317				 * The block which we have just freed is
1318				 * pointed to by an indirect block: journal it
1319				 */
1320				BUFFER_TRACE(parent_bh, "get_write_access");
1321				if (!ext4_journal_get_write_access(handle,
1322								   parent_bh)){
1323					*p = 0;
1324					BUFFER_TRACE(parent_bh,
1325					"call ext4_handle_dirty_metadata");
1326					ext4_handle_dirty_metadata(handle,
1327								   inode,
1328								   parent_bh);
1329				}
1330			}
1331		}
1332	} else {
1333		/* We have reached the bottom of the tree. */
1334		BUFFER_TRACE(parent_bh, "free data blocks");
1335		ext4_free_data(handle, inode, parent_bh, first, last);
1336	}
1337}
1338
1339void ext4_ind_truncate(struct inode *inode)
1340{
1341	handle_t *handle;
1342	struct ext4_inode_info *ei = EXT4_I(inode);
1343	__le32 *i_data = ei->i_data;
1344	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1345	struct address_space *mapping = inode->i_mapping;
1346	ext4_lblk_t offsets[4];
1347	Indirect chain[4];
1348	Indirect *partial;
1349	__le32 nr = 0;
1350	int n = 0;
1351	ext4_lblk_t last_block, max_block;
1352	loff_t page_len;
1353	unsigned blocksize = inode->i_sb->s_blocksize;
1354	int err;
1355
1356	handle = start_transaction(inode);
1357	if (IS_ERR(handle))
1358		return;		/* AKPM: return what? */
1359
1360	last_block = (inode->i_size + blocksize-1)
1361					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1362	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1363					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1364
1365	if (inode->i_size % PAGE_CACHE_SIZE != 0) {
1366		page_len = PAGE_CACHE_SIZE -
1367			(inode->i_size & (PAGE_CACHE_SIZE - 1));
1368
1369		err = ext4_discard_partial_page_buffers(handle,
1370			mapping, inode->i_size, page_len, 0);
1371
1372		if (err)
1373			goto out_stop;
1374	}
1375
1376	if (last_block != max_block) {
1377		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1378		if (n == 0)
1379			goto out_stop;	/* error */
1380	}
1381
1382	/*
1383	 * OK.  This truncate is going to happen.  We add the inode to the
1384	 * orphan list, so that if this truncate spans multiple transactions,
1385	 * and we crash, we will resume the truncate when the filesystem
1386	 * recovers.  It also marks the inode dirty, to catch the new size.
1387	 *
1388	 * Implication: the file must always be in a sane, consistent
1389	 * truncatable state while each transaction commits.
1390	 */
1391	if (ext4_orphan_add(handle, inode))
1392		goto out_stop;
1393
1394	/*
1395	 * From here we block out all ext4_get_block() callers who want to
1396	 * modify the block allocation tree.
1397	 */
1398	down_write(&ei->i_data_sem);
1399
1400	ext4_discard_preallocations(inode);
1401
1402	/*
1403	 * The orphan list entry will now protect us from any crash which
1404	 * occurs before the truncate completes, so it is now safe to propagate
1405	 * the new, shorter inode size (held for now in i_size) into the
1406	 * on-disk inode. We do this via i_disksize, which is the value which
1407	 * ext4 *really* writes onto the disk inode.
1408	 */
1409	ei->i_disksize = inode->i_size;
1410
1411	if (last_block == max_block) {
1412		/*
1413		 * It is unnecessary to free any data blocks if last_block is
1414		 * equal to the indirect block limit.
1415		 */
1416		goto out_unlock;
1417	} else if (n == 1) {		/* direct blocks */
1418		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1419			       i_data + EXT4_NDIR_BLOCKS);
1420		goto do_indirects;
1421	}
1422
1423	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1424	/* Kill the top of shared branch (not detached) */
1425	if (nr) {
1426		if (partial == chain) {
1427			/* Shared branch grows from the inode */
1428			ext4_free_branches(handle, inode, NULL,
1429					   &nr, &nr+1, (chain+n-1) - partial);
1430			*partial->p = 0;
1431			/*
1432			 * We mark the inode dirty prior to restart,
1433			 * and prior to stop.  No need for it here.
1434			 */
1435		} else {
1436			/* Shared branch grows from an indirect block */
1437			BUFFER_TRACE(partial->bh, "get_write_access");
1438			ext4_free_branches(handle, inode, partial->bh,
1439					partial->p,
1440					partial->p+1, (chain+n-1) - partial);
1441		}
1442	}
1443	/* Clear the ends of indirect blocks on the shared branch */
1444	while (partial > chain) {
1445		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1446				   (__le32*)partial->bh->b_data+addr_per_block,
1447				   (chain+n-1) - partial);
1448		BUFFER_TRACE(partial->bh, "call brelse");
1449		brelse(partial->bh);
1450		partial--;
1451	}
1452do_indirects:
1453	/* Kill the remaining (whole) subtrees */
1454	switch (offsets[0]) {
1455	default:
1456		nr = i_data[EXT4_IND_BLOCK];
1457		if (nr) {
1458			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1459			i_data[EXT4_IND_BLOCK] = 0;
1460		}
1461	case EXT4_IND_BLOCK:
1462		nr = i_data[EXT4_DIND_BLOCK];
1463		if (nr) {
1464			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1465			i_data[EXT4_DIND_BLOCK] = 0;
1466		}
1467	case EXT4_DIND_BLOCK:
1468		nr = i_data[EXT4_TIND_BLOCK];
1469		if (nr) {
1470			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1471			i_data[EXT4_TIND_BLOCK] = 0;
1472		}
1473	case EXT4_TIND_BLOCK:
1474		;
1475	}
1476
1477out_unlock:
1478	up_write(&ei->i_data_sem);
1479	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1480	ext4_mark_inode_dirty(handle, inode);
1481
1482	/*
1483	 * In a multi-transaction truncate, we only make the final transaction
1484	 * synchronous
1485	 */
1486	if (IS_SYNC(inode))
1487		ext4_handle_sync(handle);
1488out_stop:
1489	/*
1490	 * If this was a simple ftruncate(), and the file will remain alive
1491	 * then we need to clear up the orphan record which we created above.
1492	 * However, if this was a real unlink then we were called by
1493	 * ext4_delete_inode(), and we allow that function to clean up the
1494	 * orphan info for us.
1495	 */
1496	if (inode->i_nlink)
1497		ext4_orphan_del(handle, inode);
1498
1499	ext4_journal_stop(handle);
1500	trace_ext4_truncate_exit(inode);
1501}
1502