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 <linux/module.h>
  24#include "ext4_jbd2.h"
  25#include "truncate.h"
 
 
  26
  27#include <trace/events/ext4.h>
  28
  29typedef struct {
  30	__le32	*p;
  31	__le32	key;
  32	struct buffer_head *bh;
  33} Indirect;
  34
  35static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
  36{
  37	p->key = *(p->p = v);
  38	p->bh = bh;
  39}
  40
  41/**
  42 *	ext4_block_to_path - parse the block number into array of offsets
  43 *	@inode: inode in question (we are only interested in its superblock)
  44 *	@i_block: block number to be parsed
  45 *	@offsets: array to store the offsets in
  46 *	@boundary: set this non-zero if the referred-to block is likely to be
  47 *	       followed (on disk) by an indirect block.
  48 *
  49 *	To store the locations of file's data ext4 uses a data structure common
  50 *	for UNIX filesystems - tree of pointers anchored in the inode, with
  51 *	data blocks at leaves and indirect blocks in intermediate nodes.
  52 *	This function translates the block number into path in that tree -
  53 *	return value is the path length and @offsets[n] is the offset of
  54 *	pointer to (n+1)th node in the nth one. If @block is out of range
  55 *	(negative or too large) warning is printed and zero returned.
  56 *
  57 *	Note: function doesn't find node addresses, so no IO is needed. All
  58 *	we need to know is the capacity of indirect blocks (taken from the
  59 *	inode->i_sb).
  60 */
  61
  62/*
  63 * Portability note: the last comparison (check that we fit into triple
  64 * indirect block) is spelled differently, because otherwise on an
  65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
  66 * if our filesystem had 8Kb blocks. We might use long long, but that would
  67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
  68 * i_block would have to be negative in the very beginning, so we would not
  69 * get there at all.
  70 */
  71
  72static int ext4_block_to_path(struct inode *inode,
  73			      ext4_lblk_t i_block,
  74			      ext4_lblk_t offsets[4], int *boundary)
  75{
  76	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  77	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
  78	const long direct_blocks = EXT4_NDIR_BLOCKS,
  79		indirect_blocks = ptrs,
  80		double_blocks = (1 << (ptrs_bits * 2));
  81	int n = 0;
  82	int final = 0;
  83
  84	if (i_block < direct_blocks) {
  85		offsets[n++] = i_block;
  86		final = direct_blocks;
  87	} else if ((i_block -= direct_blocks) < indirect_blocks) {
  88		offsets[n++] = EXT4_IND_BLOCK;
  89		offsets[n++] = i_block;
  90		final = ptrs;
  91	} else if ((i_block -= indirect_blocks) < double_blocks) {
  92		offsets[n++] = EXT4_DIND_BLOCK;
  93		offsets[n++] = i_block >> ptrs_bits;
  94		offsets[n++] = i_block & (ptrs - 1);
  95		final = ptrs;
  96	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
  97		offsets[n++] = EXT4_TIND_BLOCK;
  98		offsets[n++] = i_block >> (ptrs_bits * 2);
  99		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
 100		offsets[n++] = i_block & (ptrs - 1);
 101		final = ptrs;
 102	} else {
 103		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
 104			     i_block + direct_blocks +
 105			     indirect_blocks + double_blocks, inode->i_ino);
 106	}
 107	if (boundary)
 108		*boundary = final - 1 - (i_block & (ptrs - 1));
 109	return n;
 110}
 111
 112/**
 113 *	ext4_get_branch - read the chain of indirect blocks leading to data
 114 *	@inode: inode in question
 115 *	@depth: depth of the chain (1 - direct pointer, etc.)
 116 *	@offsets: offsets of pointers in inode/indirect blocks
 117 *	@chain: place to store the result
 118 *	@err: here we store the error value
 119 *
 120 *	Function fills the array of triples <key, p, bh> and returns %NULL
 121 *	if everything went OK or the pointer to the last filled triple
 122 *	(incomplete one) otherwise. Upon the return chain[i].key contains
 123 *	the number of (i+1)-th block in the chain (as it is stored in memory,
 124 *	i.e. little-endian 32-bit), chain[i].p contains the address of that
 125 *	number (it points into struct inode for i==0 and into the bh->b_data
 126 *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 127 *	block for i>0 and NULL for i==0. In other words, it holds the block
 128 *	numbers of the chain, addresses they were taken from (and where we can
 129 *	verify that chain did not change) and buffer_heads hosting these
 130 *	numbers.
 131 *
 132 *	Function stops when it stumbles upon zero pointer (absent block)
 133 *		(pointer to last triple returned, *@err == 0)
 134 *	or when it gets an IO error reading an indirect block
 135 *		(ditto, *@err == -EIO)
 136 *	or when it reads all @depth-1 indirect blocks successfully and finds
 137 *	the whole chain, all way to the data (returns %NULL, *err == 0).
 138 *
 139 *      Need to be called with
 140 *      down_read(&EXT4_I(inode)->i_data_sem)
 141 */
 142static Indirect *ext4_get_branch(struct inode *inode, int depth,
 143				 ext4_lblk_t  *offsets,
 144				 Indirect chain[4], int *err)
 145{
 146	struct super_block *sb = inode->i_sb;
 147	Indirect *p = chain;
 148	struct buffer_head *bh;
 
 149
 150	*err = 0;
 151	/* i_data is not going away, no lock needed */
 152	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
 153	if (!p->key)
 154		goto no_block;
 155	while (--depth) {
 156		bh = sb_getblk(sb, le32_to_cpu(p->key));
 157		if (unlikely(!bh))
 
 158			goto failure;
 
 159
 160		if (!bh_uptodate_or_lock(bh)) {
 161			if (bh_submit_read(bh) < 0) {
 162				put_bh(bh);
 163				goto failure;
 164			}
 165			/* validate block references */
 166			if (ext4_check_indirect_blockref(inode, bh)) {
 167				put_bh(bh);
 168				goto failure;
 169			}
 170		}
 171
 172		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
 173		/* Reader: end */
 174		if (!p->key)
 175			goto no_block;
 176	}
 177	return NULL;
 178
 179failure:
 180	*err = -EIO;
 181no_block:
 182	return p;
 183}
 184
 185/**
 186 *	ext4_find_near - find a place for allocation with sufficient locality
 187 *	@inode: owner
 188 *	@ind: descriptor of indirect block.
 189 *
 190 *	This function returns the preferred place for block allocation.
 191 *	It is used when heuristic for sequential allocation fails.
 192 *	Rules are:
 193 *	  + if there is a block to the left of our position - allocate near it.
 194 *	  + if pointer will live in indirect block - allocate near that block.
 195 *	  + if pointer will live in inode - allocate in the same
 196 *	    cylinder group.
 197 *
 198 * In the latter case we colour the starting block by the callers PID to
 199 * prevent it from clashing with concurrent allocations for a different inode
 200 * in the same block group.   The PID is used here so that functionally related
 201 * files will be close-by on-disk.
 202 *
 203 *	Caller must make sure that @ind is valid and will stay that way.
 204 */
 205static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
 206{
 207	struct ext4_inode_info *ei = EXT4_I(inode);
 208	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
 209	__le32 *p;
 210
 211	/* Try to find previous block */
 212	for (p = ind->p - 1; p >= start; p--) {
 213		if (*p)
 214			return le32_to_cpu(*p);
 215	}
 216
 217	/* No such thing, so let's try location of indirect block */
 218	if (ind->bh)
 219		return ind->bh->b_blocknr;
 220
 221	/*
 222	 * It is going to be referred to from the inode itself? OK, just put it
 223	 * into the same cylinder group then.
 224	 */
 225	return ext4_inode_to_goal_block(inode);
 226}
 227
 228/**
 229 *	ext4_find_goal - find a preferred place for allocation.
 230 *	@inode: owner
 231 *	@block:  block we want
 232 *	@partial: pointer to the last triple within a chain
 233 *
 234 *	Normally this function find the preferred place for block allocation,
 235 *	returns it.
 236 *	Because this is only used for non-extent files, we limit the block nr
 237 *	to 32 bits.
 238 */
 239static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
 240				   Indirect *partial)
 241{
 242	ext4_fsblk_t goal;
 243
 244	/*
 245	 * XXX need to get goal block from mballoc's data structures
 246	 */
 247
 248	goal = ext4_find_near(inode, partial);
 249	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 250	return goal;
 251}
 252
 253/**
 254 *	ext4_blks_to_allocate - Look up the block map and count the number
 255 *	of direct blocks need to be allocated for the given branch.
 256 *
 257 *	@branch: chain of indirect blocks
 258 *	@k: number of blocks need for indirect blocks
 259 *	@blks: number of data blocks to be mapped.
 260 *	@blocks_to_boundary:  the offset in the indirect block
 261 *
 262 *	return the total number of blocks to be allocate, including the
 263 *	direct and indirect blocks.
 264 */
 265static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
 266				 int blocks_to_boundary)
 267{
 268	unsigned int count = 0;
 269
 270	/*
 271	 * Simple case, [t,d]Indirect block(s) has not allocated yet
 272	 * then it's clear blocks on that path have not allocated
 273	 */
 274	if (k > 0) {
 275		/* right now we don't handle cross boundary allocation */
 276		if (blks < blocks_to_boundary + 1)
 277			count += blks;
 278		else
 279			count += blocks_to_boundary + 1;
 280		return count;
 281	}
 282
 283	count++;
 284	while (count < blks && count <= blocks_to_boundary &&
 285		le32_to_cpu(*(branch[0].p + count)) == 0) {
 286		count++;
 287	}
 288	return count;
 289}
 290
 291/**
 292 *	ext4_alloc_blocks: multiple allocate blocks needed for a branch
 293 *	@handle: handle for this transaction
 294 *	@inode: inode which needs allocated blocks
 295 *	@iblock: the logical block to start allocated at
 296 *	@goal: preferred physical block of allocation
 297 *	@indirect_blks: the number of blocks need to allocate for indirect
 298 *			blocks
 299 *	@blks: number of desired blocks
 300 *	@new_blocks: on return it will store the new block numbers for
 301 *	the indirect blocks(if needed) and the first direct block,
 302 *	@err: on return it will store the error code
 303 *
 304 *	This function will return the number of blocks allocated as
 305 *	requested by the passed-in parameters.
 306 */
 307static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
 308			     ext4_lblk_t iblock, ext4_fsblk_t goal,
 309			     int indirect_blks, int blks,
 310			     ext4_fsblk_t new_blocks[4], int *err)
 311{
 312	struct ext4_allocation_request ar;
 313	int target, i;
 314	unsigned long count = 0, blk_allocated = 0;
 315	int index = 0;
 316	ext4_fsblk_t current_block = 0;
 317	int ret = 0;
 318
 319	/*
 320	 * Here we try to allocate the requested multiple blocks at once,
 321	 * on a best-effort basis.
 322	 * To build a branch, we should allocate blocks for
 323	 * the indirect blocks(if not allocated yet), and at least
 324	 * the first direct block of this branch.  That's the
 325	 * minimum number of blocks need to allocate(required)
 326	 */
 327	/* first we try to allocate the indirect blocks */
 328	target = indirect_blks;
 329	while (target > 0) {
 330		count = target;
 331		/* allocating blocks for indirect blocks and direct blocks */
 332		current_block = ext4_new_meta_blocks(handle, inode, goal,
 333						     0, &count, err);
 334		if (*err)
 335			goto failed_out;
 336
 337		if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
 338			EXT4_ERROR_INODE(inode,
 339					 "current_block %llu + count %lu > %d!",
 340					 current_block, count,
 341					 EXT4_MAX_BLOCK_FILE_PHYS);
 342			*err = -EIO;
 343			goto failed_out;
 344		}
 345
 346		target -= count;
 347		/* allocate blocks for indirect blocks */
 348		while (index < indirect_blks && count) {
 349			new_blocks[index++] = current_block++;
 350			count--;
 351		}
 352		if (count > 0) {
 353			/*
 354			 * save the new block number
 355			 * for the first direct block
 356			 */
 357			new_blocks[index] = current_block;
 358			printk(KERN_INFO "%s returned more blocks than "
 359						"requested\n", __func__);
 360			WARN_ON(1);
 361			break;
 362		}
 363	}
 364
 365	target = blks - count ;
 366	blk_allocated = count;
 367	if (!target)
 368		goto allocated;
 369	/* Now allocate data blocks */
 370	memset(&ar, 0, sizeof(ar));
 371	ar.inode = inode;
 372	ar.goal = goal;
 373	ar.len = target;
 374	ar.logical = iblock;
 375	if (S_ISREG(inode->i_mode))
 376		/* enable in-core preallocation only for regular files */
 377		ar.flags = EXT4_MB_HINT_DATA;
 378
 379	current_block = ext4_mb_new_blocks(handle, &ar, err);
 380	if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
 381		EXT4_ERROR_INODE(inode,
 382				 "current_block %llu + ar.len %d > %d!",
 383				 current_block, ar.len,
 384				 EXT4_MAX_BLOCK_FILE_PHYS);
 385		*err = -EIO;
 386		goto failed_out;
 387	}
 388
 389	if (*err && (target == blks)) {
 390		/*
 391		 * if the allocation failed and we didn't allocate
 392		 * any blocks before
 393		 */
 394		goto failed_out;
 395	}
 396	if (!*err) {
 397		if (target == blks) {
 398			/*
 399			 * save the new block number
 400			 * for the first direct block
 401			 */
 402			new_blocks[index] = current_block;
 403		}
 404		blk_allocated += ar.len;
 405	}
 406allocated:
 407	/* total number of blocks allocated for direct blocks */
 408	ret = blk_allocated;
 409	*err = 0;
 410	return ret;
 411failed_out:
 412	for (i = 0; i < index; i++)
 413		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
 414	return ret;
 415}
 416
 417/**
 418 *	ext4_alloc_branch - allocate and set up a chain of blocks.
 419 *	@handle: handle for this transaction
 420 *	@inode: owner
 421 *	@indirect_blks: number of allocated indirect blocks
 422 *	@blks: number of allocated direct blocks
 423 *	@goal: preferred place for allocation
 424 *	@offsets: offsets (in the blocks) to store the pointers to next.
 425 *	@branch: place to store the chain in.
 426 *
 427 *	This function allocates blocks, zeroes out all but the last one,
 428 *	links them into chain and (if we are synchronous) writes them to disk.
 429 *	In other words, it prepares a branch that can be spliced onto the
 430 *	inode. It stores the information about that chain in the branch[], in
 431 *	the same format as ext4_get_branch() would do. We are calling it after
 432 *	we had read the existing part of chain and partial points to the last
 433 *	triple of that (one with zero ->key). Upon the exit we have the same
 434 *	picture as after the successful ext4_get_block(), except that in one
 435 *	place chain is disconnected - *branch->p is still zero (we did not
 436 *	set the last link), but branch->key contains the number that should
 437 *	be placed into *branch->p to fill that gap.
 438 *
 439 *	If allocation fails we free all blocks we've allocated (and forget
 440 *	their buffer_heads) and return the error value the from failed
 441 *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 442 *	as described above and return 0.
 443 */
 444static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
 445			     ext4_lblk_t iblock, int indirect_blks,
 446			     int *blks, ext4_fsblk_t goal,
 447			     ext4_lblk_t *offsets, Indirect *branch)
 448{
 449	int blocksize = inode->i_sb->s_blocksize;
 450	int i, n = 0;
 451	int err = 0;
 452	struct buffer_head *bh;
 453	int num;
 454	ext4_fsblk_t new_blocks[4];
 455	ext4_fsblk_t current_block;
 456
 457	num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
 458				*blks, new_blocks, &err);
 459	if (err)
 460		return err;
 
 
 
 
 
 
 
 
 461
 462	branch[0].key = cpu_to_le32(new_blocks[0]);
 463	/*
 464	 * metadata blocks and data blocks are allocated.
 465	 */
 466	for (n = 1; n <= indirect_blks;  n++) {
 467		/*
 468		 * Get buffer_head for parent block, zero it out
 469		 * and set the pointer to new one, then send
 470		 * parent to disk.
 471		 */
 472		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
 473		if (unlikely(!bh)) {
 474			err = -EIO;
 475			goto failed;
 476		}
 477
 478		branch[n].bh = bh;
 479		lock_buffer(bh);
 480		BUFFER_TRACE(bh, "call get_create_access");
 481		err = ext4_journal_get_create_access(handle, bh);
 482		if (err) {
 483			/* Don't brelse(bh) here; it's done in
 484			 * ext4_journal_forget() below */
 485			unlock_buffer(bh);
 486			goto failed;
 487		}
 488
 489		memset(bh->b_data, 0, blocksize);
 490		branch[n].p = (__le32 *) bh->b_data + offsets[n];
 491		branch[n].key = cpu_to_le32(new_blocks[n]);
 492		*branch[n].p = branch[n].key;
 493		if (n == indirect_blks) {
 494			current_block = new_blocks[n];
 495			/*
 496			 * End of chain, update the last new metablock of
 497			 * the chain to point to the new allocated
 498			 * data blocks numbers
 499			 */
 500			for (i = 1; i < num; i++)
 501				*(branch[n].p + i) = cpu_to_le32(++current_block);
 502		}
 503		BUFFER_TRACE(bh, "marking uptodate");
 504		set_buffer_uptodate(bh);
 505		unlock_buffer(bh);
 506
 507		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 508		err = ext4_handle_dirty_metadata(handle, inode, bh);
 509		if (err)
 510			goto failed;
 511	}
 512	*blks = num;
 513	return err;
 514failed:
 515	/* Allocation failed, free what we already allocated */
 516	ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
 517	for (i = 1; i <= n ; i++) {
 518		/*
 519		 * branch[i].bh is newly allocated, so there is no
 520		 * need to revoke the block, which is why we don't
 521		 * need to set EXT4_FREE_BLOCKS_METADATA.
 
 522		 */
 523		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
 524				 EXT4_FREE_BLOCKS_FORGET);
 
 
 
 525	}
 526	for (i = n+1; i < indirect_blks; i++)
 527		ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
 528
 529	ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
 530
 531	return err;
 532}
 533
 534/**
 535 * ext4_splice_branch - splice the allocated branch onto inode.
 536 * @handle: handle for this transaction
 537 * @inode: owner
 538 * @block: (logical) number of block we are adding
 539 * @chain: chain of indirect blocks (with a missing link - see
 540 *	ext4_alloc_branch)
 541 * @where: location of missing link
 542 * @num:   number of indirect blocks we are adding
 543 * @blks:  number of direct blocks we are adding
 544 *
 545 * This function fills the missing link and does all housekeeping needed in
 546 * inode (->i_blocks, etc.). In case of success we end up with the full
 547 * chain to new block and return 0.
 548 */
 549static int ext4_splice_branch(handle_t *handle, struct inode *inode,
 550			      ext4_lblk_t block, Indirect *where, int num,
 551			      int blks)
 552{
 553	int i;
 554	int err = 0;
 555	ext4_fsblk_t current_block;
 556
 557	/*
 558	 * If we're splicing into a [td]indirect block (as opposed to the
 559	 * inode) then we need to get write access to the [td]indirect block
 560	 * before the splice.
 561	 */
 562	if (where->bh) {
 563		BUFFER_TRACE(where->bh, "get_write_access");
 564		err = ext4_journal_get_write_access(handle, where->bh);
 565		if (err)
 566			goto err_out;
 567	}
 568	/* That's it */
 569
 570	*where->p = where->key;
 571
 572	/*
 573	 * Update the host buffer_head or inode to point to more just allocated
 574	 * direct blocks blocks
 575	 */
 576	if (num == 0 && blks > 1) {
 577		current_block = le32_to_cpu(where->key) + 1;
 578		for (i = 1; i < blks; i++)
 579			*(where->p + i) = cpu_to_le32(current_block++);
 580	}
 581
 582	/* We are done with atomic stuff, now do the rest of housekeeping */
 583	/* had we spliced it onto indirect block? */
 584	if (where->bh) {
 585		/*
 586		 * If we spliced it onto an indirect block, we haven't
 587		 * altered the inode.  Note however that if it is being spliced
 588		 * onto an indirect block at the very end of the file (the
 589		 * file is growing) then we *will* alter the inode to reflect
 590		 * the new i_size.  But that is not done here - it is done in
 591		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
 592		 */
 593		jbd_debug(5, "splicing indirect only\n");
 594		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
 595		err = ext4_handle_dirty_metadata(handle, inode, where->bh);
 596		if (err)
 597			goto err_out;
 598	} else {
 599		/*
 600		 * OK, we spliced it into the inode itself on a direct block.
 601		 */
 602		ext4_mark_inode_dirty(handle, inode);
 603		jbd_debug(5, "splicing direct\n");
 604	}
 605	return err;
 606
 607err_out:
 608	for (i = 1; i <= num; i++) {
 609		/*
 610		 * branch[i].bh is newly allocated, so there is no
 611		 * need to revoke the block, which is why we don't
 612		 * need to set EXT4_FREE_BLOCKS_METADATA.
 613		 */
 614		ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
 615				 EXT4_FREE_BLOCKS_FORGET);
 616	}
 617	ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
 618			 blks, 0);
 619
 620	return err;
 621}
 622
 623/*
 624 * The ext4_ind_map_blocks() function handles non-extents inodes
 625 * (i.e., using the traditional indirect/double-indirect i_blocks
 626 * scheme) for ext4_map_blocks().
 627 *
 628 * Allocation strategy is simple: if we have to allocate something, we will
 629 * have to go the whole way to leaf. So let's do it before attaching anything
 630 * to tree, set linkage between the newborn blocks, write them if sync is
 631 * required, recheck the path, free and repeat if check fails, otherwise
 632 * set the last missing link (that will protect us from any truncate-generated
 633 * removals - all blocks on the path are immune now) and possibly force the
 634 * write on the parent block.
 635 * That has a nice additional property: no special recovery from the failed
 636 * allocations is needed - we simply release blocks and do not touch anything
 637 * reachable from inode.
 638 *
 639 * `handle' can be NULL if create == 0.
 640 *
 641 * return > 0, # of blocks mapped or allocated.
 642 * return = 0, if plain lookup failed.
 643 * return < 0, error case.
 644 *
 645 * The ext4_ind_get_blocks() function should be called with
 646 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
 647 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
 648 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
 649 * blocks.
 650 */
 651int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
 652			struct ext4_map_blocks *map,
 653			int flags)
 654{
 
 655	int err = -EIO;
 656	ext4_lblk_t offsets[4];
 657	Indirect chain[4];
 658	Indirect *partial;
 659	ext4_fsblk_t goal;
 660	int indirect_blks;
 661	int blocks_to_boundary = 0;
 662	int depth;
 663	int count = 0;
 664	ext4_fsblk_t first_block = 0;
 665
 666	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
 667	J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
 668	J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
 669	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
 670				   &blocks_to_boundary);
 671
 672	if (depth == 0)
 673		goto out;
 674
 675	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
 676
 677	/* Simplest case - block found, no allocation needed */
 678	if (!partial) {
 679		first_block = le32_to_cpu(chain[depth - 1].key);
 680		count++;
 681		/*map more blocks*/
 682		while (count < map->m_len && count <= blocks_to_boundary) {
 683			ext4_fsblk_t blk;
 684
 685			blk = le32_to_cpu(*(chain[depth-1].p + count));
 686
 687			if (blk == first_block + count)
 688				count++;
 689			else
 690				break;
 691		}
 692		goto got_it;
 693	}
 694
 695	/* Next simple case - plain lookup or failed read of indirect block */
 696	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 697		goto cleanup;
 698
 699	/*
 700	 * Okay, we need to do block allocation.
 701	*/
 702	goal = ext4_find_goal(inode, map->m_lblk, partial);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 703
 704	/* the number of blocks need to allocate for [d,t]indirect blocks */
 705	indirect_blks = (chain + depth) - partial - 1;
 706
 707	/*
 708	 * Next look up the indirect map to count the totoal number of
 709	 * direct blocks to allocate for this branch.
 710	 */
 711	count = ext4_blks_to_allocate(partial, indirect_blks,
 712				      map->m_len, blocks_to_boundary);
 
 713	/*
 714	 * Block out ext4_truncate while we alter the tree
 715	 */
 716	err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
 717				&count, goal,
 718				offsets + (partial - chain), partial);
 719
 720	/*
 721	 * The ext4_splice_branch call will free and forget any buffers
 722	 * on the new chain if there is a failure, but that risks using
 723	 * up transaction credits, especially for bitmaps where the
 724	 * credits cannot be returned.  Can we handle this somehow?  We
 725	 * may need to return -EAGAIN upwards in the worst case.  --sct
 726	 */
 727	if (!err)
 728		err = ext4_splice_branch(handle, inode, map->m_lblk,
 729					 partial, indirect_blks, count);
 730	if (err)
 731		goto cleanup;
 732
 733	map->m_flags |= EXT4_MAP_NEW;
 734
 735	ext4_update_inode_fsync_trans(handle, inode, 1);
 
 736got_it:
 737	map->m_flags |= EXT4_MAP_MAPPED;
 738	map->m_pblk = le32_to_cpu(chain[depth-1].key);
 739	map->m_len = count;
 740	if (count > blocks_to_boundary)
 741		map->m_flags |= EXT4_MAP_BOUNDARY;
 742	err = count;
 743	/* Clean up and exit */
 744	partial = chain + depth - 1;	/* the whole chain */
 745cleanup:
 746	while (partial > chain) {
 747		BUFFER_TRACE(partial->bh, "call brelse");
 748		brelse(partial->bh);
 749		partial--;
 750	}
 751out:
 752	trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
 753				map->m_pblk, map->m_len, err);
 754	return err;
 755}
 756
 757/*
 758 * O_DIRECT for ext3 (or indirect map) based files
 759 *
 760 * If the O_DIRECT write will extend the file then add this inode to the
 761 * orphan list.  So recovery will truncate it back to the original size
 762 * if the machine crashes during the write.
 763 *
 764 * If the O_DIRECT write is intantiating holes inside i_size and the machine
 765 * crashes then stale disk data _may_ be exposed inside the file. But current
 766 * VFS code falls back into buffered path in that case so we are safe.
 767 */
 768ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
 769			   const struct iovec *iov, loff_t offset,
 770			   unsigned long nr_segs)
 771{
 772	struct file *file = iocb->ki_filp;
 773	struct inode *inode = file->f_mapping->host;
 774	struct ext4_inode_info *ei = EXT4_I(inode);
 775	handle_t *handle;
 776	ssize_t ret;
 777	int orphan = 0;
 778	size_t count = iov_length(iov, nr_segs);
 779	int retries = 0;
 780
 781	if (rw == WRITE) {
 782		loff_t final_size = offset + count;
 783
 784		if (final_size > inode->i_size) {
 785			/* Credits for sb + inode write */
 786			handle = ext4_journal_start(inode, 2);
 787			if (IS_ERR(handle)) {
 788				ret = PTR_ERR(handle);
 789				goto out;
 790			}
 791			ret = ext4_orphan_add(handle, inode);
 792			if (ret) {
 793				ext4_journal_stop(handle);
 794				goto out;
 795			}
 796			orphan = 1;
 797			ei->i_disksize = inode->i_size;
 798			ext4_journal_stop(handle);
 799		}
 800	}
 801
 802retry:
 803	if (rw == READ && ext4_should_dioread_nolock(inode)) {
 804		if (unlikely(!list_empty(&ei->i_completed_io_list))) {
 805			mutex_lock(&inode->i_mutex);
 806			ext4_flush_completed_IO(inode);
 807			mutex_unlock(&inode->i_mutex);
 808		}
 809		ret = __blockdev_direct_IO(rw, iocb, inode,
 810				 inode->i_sb->s_bdev, iov,
 811				 offset, nr_segs,
 812				 ext4_get_block, NULL, NULL, 0);
 813	} else {
 814		ret = blockdev_direct_IO(rw, iocb, inode, iov,
 815				 offset, nr_segs, ext4_get_block);
 816
 817		if (unlikely((rw & WRITE) && ret < 0)) {
 818			loff_t isize = i_size_read(inode);
 819			loff_t end = offset + iov_length(iov, nr_segs);
 820
 821			if (end > isize)
 822				ext4_truncate_failed_write(inode);
 823		}
 824	}
 825	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 826		goto retry;
 827
 828	if (orphan) {
 829		int err;
 830
 831		/* Credits for sb + inode write */
 832		handle = ext4_journal_start(inode, 2);
 833		if (IS_ERR(handle)) {
 834			/* This is really bad luck. We've written the data
 835			 * but cannot extend i_size. Bail out and pretend
 836			 * the write failed... */
 837			ret = PTR_ERR(handle);
 838			if (inode->i_nlink)
 839				ext4_orphan_del(NULL, inode);
 840
 841			goto out;
 842		}
 843		if (inode->i_nlink)
 844			ext4_orphan_del(handle, inode);
 845		if (ret > 0) {
 846			loff_t end = offset + ret;
 847			if (end > inode->i_size) {
 848				ei->i_disksize = end;
 849				i_size_write(inode, end);
 850				/*
 851				 * We're going to return a positive `ret'
 852				 * here due to non-zero-length I/O, so there's
 853				 * no way of reporting error returns from
 854				 * ext4_mark_inode_dirty() to userspace.  So
 855				 * ignore it.
 856				 */
 857				ext4_mark_inode_dirty(handle, inode);
 858			}
 859		}
 860		err = ext4_journal_stop(handle);
 861		if (ret == 0)
 862			ret = err;
 863	}
 864out:
 865	return ret;
 866}
 867
 868/*
 869 * Calculate the number of metadata blocks need to reserve
 870 * to allocate a new block at @lblocks for non extent file based file
 871 */
 872int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
 873{
 874	struct ext4_inode_info *ei = EXT4_I(inode);
 875	sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
 876	int blk_bits;
 877
 878	if (lblock < EXT4_NDIR_BLOCKS)
 879		return 0;
 880
 881	lblock -= EXT4_NDIR_BLOCKS;
 882
 883	if (ei->i_da_metadata_calc_len &&
 884	    (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
 885		ei->i_da_metadata_calc_len++;
 886		return 0;
 887	}
 888	ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
 889	ei->i_da_metadata_calc_len = 1;
 890	blk_bits = order_base_2(lblock);
 891	return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
 892}
 893
 894int ext4_ind_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
 
 
 
 895{
 896	int indirects;
 897
 898	/* if nrblocks are contiguous */
 899	if (chunk) {
 900		/*
 901		 * With N contiguous data blocks, we need at most
 902		 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
 903		 * 2 dindirect blocks, and 1 tindirect block
 904		 */
 905		return DIV_ROUND_UP(nrblocks,
 906				    EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
 907	}
 908	/*
 909	 * if nrblocks are not contiguous, worse case, each block touch
 910	 * a indirect block, and each indirect block touch a double indirect
 911	 * block, plus a triple indirect block
 912	 */
 913	indirects = nrblocks * 2 + 1;
 914	return indirects;
 915}
 916
 917/*
 918 * Truncate transactions can be complex and absolutely huge.  So we need to
 919 * be able to restart the transaction at a conventient checkpoint to make
 920 * sure we don't overflow the journal.
 921 *
 922 * start_transaction gets us a new handle for a truncate transaction,
 923 * and extend_transaction tries to extend the existing one a bit.  If
 924 * extend fails, we need to propagate the failure up and restart the
 925 * transaction in the top-level truncate loop. --sct
 926 */
 927static handle_t *start_transaction(struct inode *inode)
 928{
 929	handle_t *result;
 930
 931	result = ext4_journal_start(inode, ext4_blocks_for_truncate(inode));
 932	if (!IS_ERR(result))
 933		return result;
 934
 935	ext4_std_error(inode->i_sb, PTR_ERR(result));
 936	return result;
 937}
 938
 939/*
 940 * Try to extend this transaction for the purposes of truncation.
 941 *
 942 * Returns 0 if we managed to create more room.  If we can't create more
 943 * room, and the transaction must be restarted we return 1.
 944 */
 945static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
 946{
 947	if (!ext4_handle_valid(handle))
 948		return 0;
 949	if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
 950		return 0;
 951	if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
 952		return 0;
 953	return 1;
 954}
 955
 956/*
 957 * Probably it should be a library function... search for first non-zero word
 958 * or memcmp with zero_page, whatever is better for particular architecture.
 959 * Linus?
 960 */
 961static inline int all_zeroes(__le32 *p, __le32 *q)
 962{
 963	while (p < q)
 964		if (*p++)
 965			return 0;
 966	return 1;
 967}
 968
 969/**
 970 *	ext4_find_shared - find the indirect blocks for partial truncation.
 971 *	@inode:	  inode in question
 972 *	@depth:	  depth of the affected branch
 973 *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
 974 *	@chain:	  place to store the pointers to partial indirect blocks
 975 *	@top:	  place to the (detached) top of branch
 976 *
 977 *	This is a helper function used by ext4_truncate().
 978 *
 979 *	When we do truncate() we may have to clean the ends of several
 980 *	indirect blocks but leave the blocks themselves alive. Block is
 981 *	partially truncated if some data below the new i_size is referred
 982 *	from it (and it is on the path to the first completely truncated
 983 *	data block, indeed).  We have to free the top of that path along
 984 *	with everything to the right of the path. Since no allocation
 985 *	past the truncation point is possible until ext4_truncate()
 986 *	finishes, we may safely do the latter, but top of branch may
 987 *	require special attention - pageout below the truncation point
 988 *	might try to populate it.
 989 *
 990 *	We atomically detach the top of branch from the tree, store the
 991 *	block number of its root in *@top, pointers to buffer_heads of
 992 *	partially truncated blocks - in @chain[].bh and pointers to
 993 *	their last elements that should not be removed - in
 994 *	@chain[].p. Return value is the pointer to last filled element
 995 *	of @chain.
 996 *
 997 *	The work left to caller to do the actual freeing of subtrees:
 998 *		a) free the subtree starting from *@top
 999 *		b) free the subtrees whose roots are stored in
1000 *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1001 *		c) free the subtrees growing from the inode past the @chain[0].
1002 *			(no partially truncated stuff there).  */
1003
1004static Indirect *ext4_find_shared(struct inode *inode, int depth,
1005				  ext4_lblk_t offsets[4], Indirect chain[4],
1006				  __le32 *top)
1007{
1008	Indirect *partial, *p;
1009	int k, err;
1010
1011	*top = 0;
1012	/* Make k index the deepest non-null offset + 1 */
1013	for (k = depth; k > 1 && !offsets[k-1]; k--)
1014		;
1015	partial = ext4_get_branch(inode, k, offsets, chain, &err);
1016	/* Writer: pointers */
1017	if (!partial)
1018		partial = chain + k-1;
1019	/*
1020	 * If the branch acquired continuation since we've looked at it -
1021	 * fine, it should all survive and (new) top doesn't belong to us.
1022	 */
1023	if (!partial->key && *partial->p)
1024		/* Writer: end */
1025		goto no_top;
1026	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
1027		;
1028	/*
1029	 * OK, we've found the last block that must survive. The rest of our
1030	 * branch should be detached before unlocking. However, if that rest
1031	 * of branch is all ours and does not grow immediately from the inode
1032	 * it's easier to cheat and just decrement partial->p.
1033	 */
1034	if (p == chain + k - 1 && p > chain) {
1035		p->p--;
1036	} else {
1037		*top = *p->p;
1038		/* Nope, don't do this in ext4.  Must leave the tree intact */
1039#if 0
1040		*p->p = 0;
1041#endif
1042	}
1043	/* Writer: end */
1044
1045	while (partial > p) {
1046		brelse(partial->bh);
1047		partial--;
1048	}
1049no_top:
1050	return partial;
1051}
1052
1053/*
1054 * Zero a number of block pointers in either an inode or an indirect block.
1055 * If we restart the transaction we must again get write access to the
1056 * indirect block for further modification.
1057 *
1058 * We release `count' blocks on disk, but (last - first) may be greater
1059 * than `count' because there can be holes in there.
1060 *
1061 * Return 0 on success, 1 on invalid block range
1062 * and < 0 on fatal error.
1063 */
1064static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
1065			     struct buffer_head *bh,
1066			     ext4_fsblk_t block_to_free,
1067			     unsigned long count, __le32 *first,
1068			     __le32 *last)
1069{
1070	__le32 *p;
1071	int	flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
1072	int	err;
1073
1074	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1075		flags |= EXT4_FREE_BLOCKS_METADATA;
 
 
1076
1077	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
1078				   count)) {
1079		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
1080				 "blocks %llu len %lu",
1081				 (unsigned long long) block_to_free, count);
1082		return 1;
1083	}
1084
1085	if (try_to_extend_transaction(handle, inode)) {
1086		if (bh) {
1087			BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1088			err = ext4_handle_dirty_metadata(handle, inode, bh);
1089			if (unlikely(err))
1090				goto out_err;
1091		}
1092		err = ext4_mark_inode_dirty(handle, inode);
1093		if (unlikely(err))
1094			goto out_err;
1095		err = ext4_truncate_restart_trans(handle, inode,
1096					ext4_blocks_for_truncate(inode));
1097		if (unlikely(err))
1098			goto out_err;
1099		if (bh) {
1100			BUFFER_TRACE(bh, "retaking write access");
1101			err = ext4_journal_get_write_access(handle, bh);
1102			if (unlikely(err))
1103				goto out_err;
1104		}
1105	}
1106
1107	for (p = first; p < last; p++)
1108		*p = 0;
1109
1110	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
1111	return 0;
1112out_err:
1113	ext4_std_error(inode->i_sb, err);
1114	return err;
1115}
1116
1117/**
1118 * ext4_free_data - free a list of data blocks
1119 * @handle:	handle for this transaction
1120 * @inode:	inode we are dealing with
1121 * @this_bh:	indirect buffer_head which contains *@first and *@last
1122 * @first:	array of block numbers
1123 * @last:	points immediately past the end of array
1124 *
1125 * We are freeing all blocks referred from that array (numbers are stored as
1126 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1127 *
1128 * We accumulate contiguous runs of blocks to free.  Conveniently, if these
1129 * blocks are contiguous then releasing them at one time will only affect one
1130 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1131 * actually use a lot of journal space.
1132 *
1133 * @this_bh will be %NULL if @first and @last point into the inode's direct
1134 * block pointers.
1135 */
1136static void ext4_free_data(handle_t *handle, struct inode *inode,
1137			   struct buffer_head *this_bh,
1138			   __le32 *first, __le32 *last)
1139{
1140	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
1141	unsigned long count = 0;	    /* Number of blocks in the run */
1142	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
1143					       corresponding to
1144					       block_to_free */
1145	ext4_fsblk_t nr;		    /* Current block # */
1146	__le32 *p;			    /* Pointer into inode/ind
1147					       for current block */
1148	int err = 0;
1149
1150	if (this_bh) {				/* For indirect block */
1151		BUFFER_TRACE(this_bh, "get_write_access");
1152		err = ext4_journal_get_write_access(handle, this_bh);
1153		/* Important: if we can't update the indirect pointers
1154		 * to the blocks, we can't free them. */
1155		if (err)
1156			return;
1157	}
1158
1159	for (p = first; p < last; p++) {
1160		nr = le32_to_cpu(*p);
1161		if (nr) {
1162			/* accumulate blocks to free if they're contiguous */
1163			if (count == 0) {
1164				block_to_free = nr;
1165				block_to_free_p = p;
1166				count = 1;
1167			} else if (nr == block_to_free + count) {
1168				count++;
1169			} else {
1170				err = ext4_clear_blocks(handle, inode, this_bh,
1171						        block_to_free, count,
1172						        block_to_free_p, p);
1173				if (err)
1174					break;
1175				block_to_free = nr;
1176				block_to_free_p = p;
1177				count = 1;
1178			}
1179		}
1180	}
1181
1182	if (!err && count > 0)
1183		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1184					count, block_to_free_p, p);
1185	if (err < 0)
1186		/* fatal error */
1187		return;
1188
1189	if (this_bh) {
1190		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1191
1192		/*
1193		 * The buffer head should have an attached journal head at this
1194		 * point. However, if the data is corrupted and an indirect
1195		 * block pointed to itself, it would have been detached when
1196		 * the block was cleared. Check for this instead of OOPSing.
1197		 */
1198		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1199			ext4_handle_dirty_metadata(handle, inode, this_bh);
1200		else
1201			EXT4_ERROR_INODE(inode,
1202					 "circular indirect block detected at "
1203					 "block %llu",
1204				(unsigned long long) this_bh->b_blocknr);
1205	}
1206}
1207
1208/**
1209 *	ext4_free_branches - free an array of branches
1210 *	@handle: JBD handle for this transaction
1211 *	@inode:	inode we are dealing with
1212 *	@parent_bh: the buffer_head which contains *@first and *@last
1213 *	@first:	array of block numbers
1214 *	@last:	pointer immediately past the end of array
1215 *	@depth:	depth of the branches to free
1216 *
1217 *	We are freeing all blocks referred from these branches (numbers are
1218 *	stored as little-endian 32-bit) and updating @inode->i_blocks
1219 *	appropriately.
1220 */
1221static void ext4_free_branches(handle_t *handle, struct inode *inode,
1222			       struct buffer_head *parent_bh,
1223			       __le32 *first, __le32 *last, int depth)
1224{
1225	ext4_fsblk_t nr;
1226	__le32 *p;
1227
1228	if (ext4_handle_is_aborted(handle))
1229		return;
1230
1231	if (depth--) {
1232		struct buffer_head *bh;
1233		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1234		p = last;
1235		while (--p >= first) {
1236			nr = le32_to_cpu(*p);
1237			if (!nr)
1238				continue;		/* A hole */
1239
1240			if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1241						   nr, 1)) {
1242				EXT4_ERROR_INODE(inode,
1243						 "invalid indirect mapped "
1244						 "block %lu (level %d)",
1245						 (unsigned long) nr, depth);
1246				break;
1247			}
1248
1249			/* Go read the buffer for the next level down */
1250			bh = sb_bread(inode->i_sb, nr);
1251
1252			/*
1253			 * A read failure? Report error and clear slot
1254			 * (should be rare).
1255			 */
1256			if (!bh) {
1257				EXT4_ERROR_INODE_BLOCK(inode, nr,
1258						       "Read failure");
1259				continue;
1260			}
1261
1262			/* This zaps the entire block.  Bottom up. */
1263			BUFFER_TRACE(bh, "free child branches");
1264			ext4_free_branches(handle, inode, bh,
1265					(__le32 *) bh->b_data,
1266					(__le32 *) bh->b_data + addr_per_block,
1267					depth);
1268			brelse(bh);
1269
1270			/*
1271			 * Everything below this this pointer has been
1272			 * released.  Now let this top-of-subtree go.
1273			 *
1274			 * We want the freeing of this indirect block to be
1275			 * atomic in the journal with the updating of the
1276			 * bitmap block which owns it.  So make some room in
1277			 * the journal.
1278			 *
1279			 * We zero the parent pointer *after* freeing its
1280			 * pointee in the bitmaps, so if extend_transaction()
1281			 * for some reason fails to put the bitmap changes and
1282			 * the release into the same transaction, recovery
1283			 * will merely complain about releasing a free block,
1284			 * rather than leaking blocks.
1285			 */
1286			if (ext4_handle_is_aborted(handle))
1287				return;
1288			if (try_to_extend_transaction(handle, inode)) {
1289				ext4_mark_inode_dirty(handle, inode);
1290				ext4_truncate_restart_trans(handle, inode,
1291					    ext4_blocks_for_truncate(inode));
1292			}
1293
1294			/*
1295			 * The forget flag here is critical because if
1296			 * we are journaling (and not doing data
1297			 * journaling), we have to make sure a revoke
1298			 * record is written to prevent the journal
1299			 * replay from overwriting the (former)
1300			 * indirect block if it gets reallocated as a
1301			 * data block.  This must happen in the same
1302			 * transaction where the data blocks are
1303			 * actually freed.
1304			 */
1305			ext4_free_blocks(handle, inode, NULL, nr, 1,
1306					 EXT4_FREE_BLOCKS_METADATA|
1307					 EXT4_FREE_BLOCKS_FORGET);
1308
1309			if (parent_bh) {
1310				/*
1311				 * The block which we have just freed is
1312				 * pointed to by an indirect block: journal it
1313				 */
1314				BUFFER_TRACE(parent_bh, "get_write_access");
1315				if (!ext4_journal_get_write_access(handle,
1316								   parent_bh)){
1317					*p = 0;
1318					BUFFER_TRACE(parent_bh,
1319					"call ext4_handle_dirty_metadata");
1320					ext4_handle_dirty_metadata(handle,
1321								   inode,
1322								   parent_bh);
1323				}
1324			}
1325		}
1326	} else {
1327		/* We have reached the bottom of the tree. */
1328		BUFFER_TRACE(parent_bh, "free data blocks");
1329		ext4_free_data(handle, inode, parent_bh, first, last);
1330	}
1331}
1332
1333void ext4_ind_truncate(struct inode *inode)
1334{
1335	handle_t *handle;
1336	struct ext4_inode_info *ei = EXT4_I(inode);
1337	__le32 *i_data = ei->i_data;
1338	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1339	struct address_space *mapping = inode->i_mapping;
1340	ext4_lblk_t offsets[4];
1341	Indirect chain[4];
1342	Indirect *partial;
1343	__le32 nr = 0;
1344	int n = 0;
1345	ext4_lblk_t last_block, max_block;
1346	unsigned blocksize = inode->i_sb->s_blocksize;
1347
1348	handle = start_transaction(inode);
1349	if (IS_ERR(handle))
1350		return;		/* AKPM: return what? */
1351
1352	last_block = (inode->i_size + blocksize-1)
1353					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1354	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1355					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1356
1357	if (inode->i_size & (blocksize - 1))
1358		if (ext4_block_truncate_page(handle, mapping, inode->i_size))
1359			goto out_stop;
1360
1361	if (last_block != max_block) {
1362		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1363		if (n == 0)
1364			goto out_stop;	/* error */
1365	}
1366
1367	/*
1368	 * OK.  This truncate is going to happen.  We add the inode to the
1369	 * orphan list, so that if this truncate spans multiple transactions,
1370	 * and we crash, we will resume the truncate when the filesystem
1371	 * recovers.  It also marks the inode dirty, to catch the new size.
1372	 *
1373	 * Implication: the file must always be in a sane, consistent
1374	 * truncatable state while each transaction commits.
1375	 */
1376	if (ext4_orphan_add(handle, inode))
1377		goto out_stop;
1378
1379	/*
1380	 * From here we block out all ext4_get_block() callers who want to
1381	 * modify the block allocation tree.
1382	 */
1383	down_write(&ei->i_data_sem);
1384
1385	ext4_discard_preallocations(inode);
1386
1387	/*
1388	 * The orphan list entry will now protect us from any crash which
1389	 * occurs before the truncate completes, so it is now safe to propagate
1390	 * the new, shorter inode size (held for now in i_size) into the
1391	 * on-disk inode. We do this via i_disksize, which is the value which
1392	 * ext4 *really* writes onto the disk inode.
1393	 */
1394	ei->i_disksize = inode->i_size;
1395
1396	if (last_block == max_block) {
1397		/*
1398		 * It is unnecessary to free any data blocks if last_block is
1399		 * equal to the indirect block limit.
1400		 */
1401		goto out_unlock;
1402	} else if (n == 1) {		/* direct blocks */
1403		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1404			       i_data + EXT4_NDIR_BLOCKS);
1405		goto do_indirects;
1406	}
1407
1408	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1409	/* Kill the top of shared branch (not detached) */
1410	if (nr) {
1411		if (partial == chain) {
1412			/* Shared branch grows from the inode */
1413			ext4_free_branches(handle, inode, NULL,
1414					   &nr, &nr+1, (chain+n-1) - partial);
1415			*partial->p = 0;
1416			/*
1417			 * We mark the inode dirty prior to restart,
1418			 * and prior to stop.  No need for it here.
1419			 */
1420		} else {
1421			/* Shared branch grows from an indirect block */
1422			BUFFER_TRACE(partial->bh, "get_write_access");
1423			ext4_free_branches(handle, inode, partial->bh,
1424					partial->p,
1425					partial->p+1, (chain+n-1) - partial);
1426		}
1427	}
1428	/* Clear the ends of indirect blocks on the shared branch */
1429	while (partial > chain) {
1430		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1431				   (__le32*)partial->bh->b_data+addr_per_block,
1432				   (chain+n-1) - partial);
1433		BUFFER_TRACE(partial->bh, "call brelse");
1434		brelse(partial->bh);
1435		partial--;
1436	}
1437do_indirects:
1438	/* Kill the remaining (whole) subtrees */
1439	switch (offsets[0]) {
1440	default:
1441		nr = i_data[EXT4_IND_BLOCK];
1442		if (nr) {
1443			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1444			i_data[EXT4_IND_BLOCK] = 0;
1445		}
1446	case EXT4_IND_BLOCK:
1447		nr = i_data[EXT4_DIND_BLOCK];
1448		if (nr) {
1449			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1450			i_data[EXT4_DIND_BLOCK] = 0;
1451		}
1452	case EXT4_DIND_BLOCK:
1453		nr = i_data[EXT4_TIND_BLOCK];
1454		if (nr) {
1455			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1456			i_data[EXT4_TIND_BLOCK] = 0;
1457		}
1458	case EXT4_TIND_BLOCK:
1459		;
1460	}
 
1461
1462out_unlock:
1463	up_write(&ei->i_data_sem);
1464	inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
1465	ext4_mark_inode_dirty(handle, inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1466
1467	/*
1468	 * In a multi-transaction truncate, we only make the final transaction
1469	 * synchronous
1470	 */
1471	if (IS_SYNC(inode))
1472		ext4_handle_sync(handle);
1473out_stop:
1474	/*
1475	 * If this was a simple ftruncate(), and the file will remain alive
1476	 * then we need to clear up the orphan record which we created above.
1477	 * However, if this was a real unlink then we were called by
1478	 * ext4_delete_inode(), and we allow that function to clean up the
1479	 * orphan info for us.
1480	 */
1481	if (inode->i_nlink)
1482		ext4_orphan_del(handle, inode);
1483
1484	ext4_journal_stop(handle);
1485	trace_ext4_truncate_exit(inode);
1486}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1487