1/*
   2 *  linux/fs/ext3/balloc.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  10 *  Big-endian to little-endian byte-swapping/bitmaps by
  11 *        David S. Miller (davem@caip.rutgers.edu), 1995
  12 */
  13
 
 
 
 
 
 
 
  14#include <linux/quotaops.h>
 
  15#include <linux/blkdev.h>
  16#include "ext3.h"
  17
  18/*
  19 * balloc.c contains the blocks allocation and deallocation routines
  20 */
  21
  22/*
  23 * The free blocks are managed by bitmaps.  A file system contains several
  24 * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  25 * block for inodes, N blocks for the inode table and data blocks.
  26 *
  27 * The file system contains group descriptors which are located after the
  28 * super block.  Each descriptor contains the number of the bitmap block and
  29 * the free blocks count in the block.  The descriptors are loaded in memory
  30 * when a file system is mounted (see ext3_fill_super).
  31 */
  32
  33
  34#define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
  35
  36/*
  37 * Calculate the block group number and offset, given a block number
  38 */
  39static void ext3_get_group_no_and_offset(struct super_block *sb,
  40	ext3_fsblk_t blocknr, unsigned long *blockgrpp, ext3_grpblk_t *offsetp)
  41{
  42	struct ext3_super_block *es = EXT3_SB(sb)->s_es;
  43
  44	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  45	if (offsetp)
  46		*offsetp = blocknr % EXT3_BLOCKS_PER_GROUP(sb);
  47	if (blockgrpp)
  48		*blockgrpp = blocknr / EXT3_BLOCKS_PER_GROUP(sb);
  49}
  50
  51/**
  52 * ext3_get_group_desc() -- load group descriptor from disk
  53 * @sb:			super block
  54 * @block_group:	given block group
  55 * @bh:			pointer to the buffer head to store the block
  56 *			group descriptor
  57 */
  58struct ext3_group_desc * ext3_get_group_desc(struct super_block * sb,
  59					     unsigned int block_group,
  60					     struct buffer_head ** bh)
  61{
  62	unsigned long group_desc;
  63	unsigned long offset;
  64	struct ext3_group_desc * desc;
  65	struct ext3_sb_info *sbi = EXT3_SB(sb);
  66
  67	if (block_group >= sbi->s_groups_count) {
  68		ext3_error (sb, "ext3_get_group_desc",
  69			    "block_group >= groups_count - "
  70			    "block_group = %d, groups_count = %lu",
  71			    block_group, sbi->s_groups_count);
  72
  73		return NULL;
  74	}
  75	smp_rmb();
  76
  77	group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
  78	offset = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
  79	if (!sbi->s_group_desc[group_desc]) {
  80		ext3_error (sb, "ext3_get_group_desc",
  81			    "Group descriptor not loaded - "
  82			    "block_group = %d, group_desc = %lu, desc = %lu",
  83			     block_group, group_desc, offset);
  84		return NULL;
  85	}
  86
  87	desc = (struct ext3_group_desc *) sbi->s_group_desc[group_desc]->b_data;
  88	if (bh)
  89		*bh = sbi->s_group_desc[group_desc];
  90	return desc + offset;
  91}
  92
  93static int ext3_valid_block_bitmap(struct super_block *sb,
  94					struct ext3_group_desc *desc,
  95					unsigned int block_group,
  96					struct buffer_head *bh)
  97{
  98	ext3_grpblk_t offset;
  99	ext3_grpblk_t next_zero_bit;
 100	ext3_fsblk_t bitmap_blk;
 101	ext3_fsblk_t group_first_block;
 102
 103	group_first_block = ext3_group_first_block_no(sb, block_group);
 104
 105	/* check whether block bitmap block number is set */
 106	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
 107	offset = bitmap_blk - group_first_block;
 108	if (!ext3_test_bit(offset, bh->b_data))
 109		/* bad block bitmap */
 110		goto err_out;
 111
 112	/* check whether the inode bitmap block number is set */
 113	bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
 114	offset = bitmap_blk - group_first_block;
 115	if (!ext3_test_bit(offset, bh->b_data))
 116		/* bad block bitmap */
 117		goto err_out;
 118
 119	/* check whether the inode table block number is set */
 120	bitmap_blk = le32_to_cpu(desc->bg_inode_table);
 121	offset = bitmap_blk - group_first_block;
 122	next_zero_bit = ext3_find_next_zero_bit(bh->b_data,
 123				offset + EXT3_SB(sb)->s_itb_per_group,
 124				offset);
 125	if (next_zero_bit >= offset + EXT3_SB(sb)->s_itb_per_group)
 126		/* good bitmap for inode tables */
 127		return 1;
 128
 129err_out:
 130	ext3_error(sb, __func__,
 131			"Invalid block bitmap - "
 132			"block_group = %d, block = %lu",
 133			block_group, bitmap_blk);
 134	return 0;
 135}
 136
 137/**
 138 * read_block_bitmap()
 139 * @sb:			super block
 140 * @block_group:	given block group
 141 *
 142 * Read the bitmap for a given block_group,and validate the
 143 * bits for block/inode/inode tables are set in the bitmaps
 144 *
 145 * Return buffer_head on success or NULL in case of failure.
 146 */
 147static struct buffer_head *
 148read_block_bitmap(struct super_block *sb, unsigned int block_group)
 149{
 150	struct ext3_group_desc * desc;
 151	struct buffer_head * bh = NULL;
 152	ext3_fsblk_t bitmap_blk;
 153
 154	desc = ext3_get_group_desc(sb, block_group, NULL);
 155	if (!desc)
 156		return NULL;
 157	trace_ext3_read_block_bitmap(sb, block_group);
 158	bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
 159	bh = sb_getblk(sb, bitmap_blk);
 160	if (unlikely(!bh)) {
 161		ext3_error(sb, __func__,
 162			    "Cannot read block bitmap - "
 163			    "block_group = %d, block_bitmap = %u",
 164			    block_group, le32_to_cpu(desc->bg_block_bitmap));
 165		return NULL;
 166	}
 167	if (likely(bh_uptodate_or_lock(bh)))
 168		return bh;
 169
 170	if (bh_submit_read(bh) < 0) {
 171		brelse(bh);
 172		ext3_error(sb, __func__,
 173			    "Cannot read block bitmap - "
 174			    "block_group = %d, block_bitmap = %u",
 175			    block_group, le32_to_cpu(desc->bg_block_bitmap));
 176		return NULL;
 177	}
 178	ext3_valid_block_bitmap(sb, desc, block_group, bh);
 179	/*
 180	 * file system mounted not to panic on error, continue with corrupt
 181	 * bitmap
 182	 */
 183	return bh;
 184}
 185/*
 186 * The reservation window structure operations
 187 * --------------------------------------------
 188 * Operations include:
 189 * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
 190 *
 191 * We use a red-black tree to represent per-filesystem reservation
 192 * windows.
 193 *
 194 */
 195
 196/**
 197 * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
 198 * @rb_root:		root of per-filesystem reservation rb tree
 199 * @verbose:		verbose mode
 200 * @fn:			function which wishes to dump the reservation map
 201 *
 202 * If verbose is turned on, it will print the whole block reservation
 203 * windows(start, end).	Otherwise, it will only print out the "bad" windows,
 204 * those windows that overlap with their immediate neighbors.
 205 */
 206#if 1
 207static void __rsv_window_dump(struct rb_root *root, int verbose,
 208			      const char *fn)
 209{
 210	struct rb_node *n;
 211	struct ext3_reserve_window_node *rsv, *prev;
 212	int bad;
 213
 214restart:
 215	n = rb_first(root);
 216	bad = 0;
 217	prev = NULL;
 218
 219	printk("Block Allocation Reservation Windows Map (%s):\n", fn);
 220	while (n) {
 221		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 222		if (verbose)
 223			printk("reservation window 0x%p "
 224			       "start:  %lu, end:  %lu\n",
 225			       rsv, rsv->rsv_start, rsv->rsv_end);
 226		if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
 227			printk("Bad reservation %p (start >= end)\n",
 228			       rsv);
 229			bad = 1;
 230		}
 231		if (prev && prev->rsv_end >= rsv->rsv_start) {
 232			printk("Bad reservation %p (prev->end >= start)\n",
 233			       rsv);
 234			bad = 1;
 235		}
 236		if (bad) {
 237			if (!verbose) {
 238				printk("Restarting reservation walk in verbose mode\n");
 239				verbose = 1;
 240				goto restart;
 241			}
 242		}
 243		n = rb_next(n);
 244		prev = rsv;
 245	}
 246	printk("Window map complete.\n");
 247	BUG_ON(bad);
 248}
 249#define rsv_window_dump(root, verbose) \
 250	__rsv_window_dump((root), (verbose), __func__)
 251#else
 252#define rsv_window_dump(root, verbose) do {} while (0)
 253#endif
 254
 255/**
 256 * goal_in_my_reservation()
 257 * @rsv:		inode's reservation window
 258 * @grp_goal:		given goal block relative to the allocation block group
 259 * @group:		the current allocation block group
 260 * @sb:			filesystem super block
 261 *
 262 * Test if the given goal block (group relative) is within the file's
 263 * own block reservation window range.
 264 *
 265 * If the reservation window is outside the goal allocation group, return 0;
 266 * grp_goal (given goal block) could be -1, which means no specific
 267 * goal block. In this case, always return 1.
 268 * If the goal block is within the reservation window, return 1;
 269 * otherwise, return 0;
 270 */
 271static int
 272goal_in_my_reservation(struct ext3_reserve_window *rsv, ext3_grpblk_t grp_goal,
 273			unsigned int group, struct super_block * sb)
 274{
 275	ext3_fsblk_t group_first_block, group_last_block;
 276
 277	group_first_block = ext3_group_first_block_no(sb, group);
 278	group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
 279
 280	if ((rsv->_rsv_start > group_last_block) ||
 281	    (rsv->_rsv_end < group_first_block))
 282		return 0;
 283	if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
 284		|| (grp_goal + group_first_block > rsv->_rsv_end)))
 285		return 0;
 286	return 1;
 287}
 288
 289/**
 290 * search_reserve_window()
 291 * @rb_root:		root of reservation tree
 292 * @goal:		target allocation block
 293 *
 294 * Find the reserved window which includes the goal, or the previous one
 295 * if the goal is not in any window.
 296 * Returns NULL if there are no windows or if all windows start after the goal.
 297 */
 298static struct ext3_reserve_window_node *
 299search_reserve_window(struct rb_root *root, ext3_fsblk_t goal)
 300{
 301	struct rb_node *n = root->rb_node;
 302	struct ext3_reserve_window_node *rsv;
 303
 304	if (!n)
 305		return NULL;
 306
 307	do {
 308		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 309
 310		if (goal < rsv->rsv_start)
 311			n = n->rb_left;
 312		else if (goal > rsv->rsv_end)
 313			n = n->rb_right;
 314		else
 315			return rsv;
 316	} while (n);
 317	/*
 318	 * We've fallen off the end of the tree: the goal wasn't inside
 319	 * any particular node.  OK, the previous node must be to one
 320	 * side of the interval containing the goal.  If it's the RHS,
 321	 * we need to back up one.
 322	 */
 323	if (rsv->rsv_start > goal) {
 324		n = rb_prev(&rsv->rsv_node);
 325		rsv = rb_entry(n, struct ext3_reserve_window_node, rsv_node);
 326	}
 327	return rsv;
 328}
 329
 330/**
 331 * ext3_rsv_window_add() -- Insert a window to the block reservation rb tree.
 332 * @sb:			super block
 333 * @rsv:		reservation window to add
 334 *
 335 * Must be called with rsv_lock hold.
 336 */
 337void ext3_rsv_window_add(struct super_block *sb,
 338		    struct ext3_reserve_window_node *rsv)
 339{
 340	struct rb_root *root = &EXT3_SB(sb)->s_rsv_window_root;
 341	struct rb_node *node = &rsv->rsv_node;
 342	ext3_fsblk_t start = rsv->rsv_start;
 343
 344	struct rb_node ** p = &root->rb_node;
 345	struct rb_node * parent = NULL;
 346	struct ext3_reserve_window_node *this;
 347
 348	trace_ext3_rsv_window_add(sb, rsv);
 349	while (*p)
 350	{
 351		parent = *p;
 352		this = rb_entry(parent, struct ext3_reserve_window_node, rsv_node);
 353
 354		if (start < this->rsv_start)
 355			p = &(*p)->rb_left;
 356		else if (start > this->rsv_end)
 357			p = &(*p)->rb_right;
 358		else {
 359			rsv_window_dump(root, 1);
 360			BUG();
 361		}
 362	}
 363
 364	rb_link_node(node, parent, p);
 365	rb_insert_color(node, root);
 366}
 367
 368/**
 369 * ext3_rsv_window_remove() -- unlink a window from the reservation rb tree
 370 * @sb:			super block
 371 * @rsv:		reservation window to remove
 372 *
 373 * Mark the block reservation window as not allocated, and unlink it
 374 * from the filesystem reservation window rb tree. Must be called with
 375 * rsv_lock hold.
 376 */
 377static void rsv_window_remove(struct super_block *sb,
 378			      struct ext3_reserve_window_node *rsv)
 379{
 380	rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 381	rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 382	rsv->rsv_alloc_hit = 0;
 383	rb_erase(&rsv->rsv_node, &EXT3_SB(sb)->s_rsv_window_root);
 384}
 385
 386/*
 387 * rsv_is_empty() -- Check if the reservation window is allocated.
 388 * @rsv:		given reservation window to check
 389 *
 390 * returns 1 if the end block is EXT3_RESERVE_WINDOW_NOT_ALLOCATED.
 391 */
 392static inline int rsv_is_empty(struct ext3_reserve_window *rsv)
 393{
 394	/* a valid reservation end block could not be 0 */
 395	return rsv->_rsv_end == EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 396}
 397
 398/**
 399 * ext3_init_block_alloc_info()
 400 * @inode:		file inode structure
 401 *
 402 * Allocate and initialize the	reservation window structure, and
 403 * link the window to the ext3 inode structure at last
 404 *
 405 * The reservation window structure is only dynamically allocated
 406 * and linked to ext3 inode the first time the open file
 407 * needs a new block. So, before every ext3_new_block(s) call, for
 408 * regular files, we should check whether the reservation window
 409 * structure exists or not. In the latter case, this function is called.
 410 * Fail to do so will result in block reservation being turned off for that
 411 * open file.
 412 *
 413 * This function is called from ext3_get_blocks_handle(), also called
 414 * when setting the reservation window size through ioctl before the file
 415 * is open for write (needs block allocation).
 416 *
 417 * Needs truncate_mutex protection prior to call this function.
 418 */
 419void ext3_init_block_alloc_info(struct inode *inode)
 420{
 421	struct ext3_inode_info *ei = EXT3_I(inode);
 422	struct ext3_block_alloc_info *block_i;
 423	struct super_block *sb = inode->i_sb;
 424
 425	block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
 426	if (block_i) {
 427		struct ext3_reserve_window_node *rsv = &block_i->rsv_window_node;
 428
 429		rsv->rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 430		rsv->rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 431
 432		/*
 433		 * if filesystem is mounted with NORESERVATION, the goal
 434		 * reservation window size is set to zero to indicate
 435		 * block reservation is off
 436		 */
 437		if (!test_opt(sb, RESERVATION))
 438			rsv->rsv_goal_size = 0;
 439		else
 440			rsv->rsv_goal_size = EXT3_DEFAULT_RESERVE_BLOCKS;
 441		rsv->rsv_alloc_hit = 0;
 442		block_i->last_alloc_logical_block = 0;
 443		block_i->last_alloc_physical_block = 0;
 444	}
 445	ei->i_block_alloc_info = block_i;
 446}
 447
 448/**
 449 * ext3_discard_reservation()
 450 * @inode:		inode
 451 *
 452 * Discard(free) block reservation window on last file close, or truncate
 453 * or at last iput().
 454 *
 455 * It is being called in three cases:
 456 *	ext3_release_file(): last writer close the file
 457 *	ext3_clear_inode(): last iput(), when nobody link to this file.
 458 *	ext3_truncate(): when the block indirect map is about to change.
 459 *
 460 */
 461void ext3_discard_reservation(struct inode *inode)
 462{
 463	struct ext3_inode_info *ei = EXT3_I(inode);
 464	struct ext3_block_alloc_info *block_i = ei->i_block_alloc_info;
 465	struct ext3_reserve_window_node *rsv;
 466	spinlock_t *rsv_lock = &EXT3_SB(inode->i_sb)->s_rsv_window_lock;
 467
 468	if (!block_i)
 469		return;
 470
 471	rsv = &block_i->rsv_window_node;
 472	if (!rsv_is_empty(&rsv->rsv_window)) {
 473		spin_lock(rsv_lock);
 474		if (!rsv_is_empty(&rsv->rsv_window)) {
 475			trace_ext3_discard_reservation(inode, rsv);
 476			rsv_window_remove(inode->i_sb, rsv);
 477		}
 478		spin_unlock(rsv_lock);
 479	}
 480}
 481
 482/**
 483 * ext3_free_blocks_sb() -- Free given blocks and update quota
 484 * @handle:			handle to this transaction
 485 * @sb:				super block
 486 * @block:			start physcial block to free
 487 * @count:			number of blocks to free
 488 * @pdquot_freed_blocks:	pointer to quota
 489 */
 490void ext3_free_blocks_sb(handle_t *handle, struct super_block *sb,
 491			 ext3_fsblk_t block, unsigned long count,
 492			 unsigned long *pdquot_freed_blocks)
 493{
 494	struct buffer_head *bitmap_bh = NULL;
 495	struct buffer_head *gd_bh;
 496	unsigned long block_group;
 497	ext3_grpblk_t bit;
 498	unsigned long i;
 499	unsigned long overflow;
 500	struct ext3_group_desc * desc;
 501	struct ext3_super_block * es;
 502	struct ext3_sb_info *sbi;
 503	int err = 0, ret;
 504	ext3_grpblk_t group_freed;
 505
 506	*pdquot_freed_blocks = 0;
 507	sbi = EXT3_SB(sb);
 508	es = sbi->s_es;
 509	if (block < le32_to_cpu(es->s_first_data_block) ||
 510	    block + count < block ||
 511	    block + count > le32_to_cpu(es->s_blocks_count)) {
 512		ext3_error (sb, "ext3_free_blocks",
 513			    "Freeing blocks not in datazone - "
 514			    "block = "E3FSBLK", count = %lu", block, count);
 515		goto error_return;
 516	}
 517
 518	ext3_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
 519
 520do_more:
 521	overflow = 0;
 522	block_group = (block - le32_to_cpu(es->s_first_data_block)) /
 523		      EXT3_BLOCKS_PER_GROUP(sb);
 524	bit = (block - le32_to_cpu(es->s_first_data_block)) %
 525		      EXT3_BLOCKS_PER_GROUP(sb);
 526	/*
 527	 * Check to see if we are freeing blocks across a group
 528	 * boundary.
 529	 */
 530	if (bit + count > EXT3_BLOCKS_PER_GROUP(sb)) {
 531		overflow = bit + count - EXT3_BLOCKS_PER_GROUP(sb);
 532		count -= overflow;
 533	}
 534	brelse(bitmap_bh);
 535	bitmap_bh = read_block_bitmap(sb, block_group);
 536	if (!bitmap_bh)
 537		goto error_return;
 538	desc = ext3_get_group_desc (sb, block_group, &gd_bh);
 539	if (!desc)
 540		goto error_return;
 541
 542	if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
 543	    in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
 544	    in_range (block, le32_to_cpu(desc->bg_inode_table),
 545		      sbi->s_itb_per_group) ||
 546	    in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
 547		      sbi->s_itb_per_group)) {
 548		ext3_error (sb, "ext3_free_blocks",
 549			    "Freeing blocks in system zones - "
 550			    "Block = "E3FSBLK", count = %lu",
 551			    block, count);
 552		goto error_return;
 553	}
 554
 555	/*
 556	 * We are about to start releasing blocks in the bitmap,
 557	 * so we need undo access.
 558	 */
 559	/* @@@ check errors */
 560	BUFFER_TRACE(bitmap_bh, "getting undo access");
 561	err = ext3_journal_get_undo_access(handle, bitmap_bh);
 562	if (err)
 563		goto error_return;
 564
 565	/*
 566	 * We are about to modify some metadata.  Call the journal APIs
 567	 * to unshare ->b_data if a currently-committing transaction is
 568	 * using it
 569	 */
 570	BUFFER_TRACE(gd_bh, "get_write_access");
 571	err = ext3_journal_get_write_access(handle, gd_bh);
 572	if (err)
 573		goto error_return;
 574
 575	jbd_lock_bh_state(bitmap_bh);
 576
 577	for (i = 0, group_freed = 0; i < count; i++) {
 578		/*
 579		 * An HJ special.  This is expensive...
 580		 */
 581#ifdef CONFIG_JBD_DEBUG
 582		jbd_unlock_bh_state(bitmap_bh);
 583		{
 584			struct buffer_head *debug_bh;
 585			debug_bh = sb_find_get_block(sb, block + i);
 586			if (debug_bh) {
 587				BUFFER_TRACE(debug_bh, "Deleted!");
 588				if (!bh2jh(bitmap_bh)->b_committed_data)
 589					BUFFER_TRACE(debug_bh,
 590						"No committed data in bitmap");
 591				BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
 592				__brelse(debug_bh);
 593			}
 594		}
 595		jbd_lock_bh_state(bitmap_bh);
 596#endif
 597		if (need_resched()) {
 598			jbd_unlock_bh_state(bitmap_bh);
 599			cond_resched();
 600			jbd_lock_bh_state(bitmap_bh);
 601		}
 602		/* @@@ This prevents newly-allocated data from being
 603		 * freed and then reallocated within the same
 604		 * transaction.
 605		 *
 606		 * Ideally we would want to allow that to happen, but to
 607		 * do so requires making journal_forget() capable of
 608		 * revoking the queued write of a data block, which
 609		 * implies blocking on the journal lock.  *forget()
 610		 * cannot block due to truncate races.
 611		 *
 612		 * Eventually we can fix this by making journal_forget()
 613		 * return a status indicating whether or not it was able
 614		 * to revoke the buffer.  On successful revoke, it is
 615		 * safe not to set the allocation bit in the committed
 616		 * bitmap, because we know that there is no outstanding
 617		 * activity on the buffer any more and so it is safe to
 618		 * reallocate it.
 619		 */
 620		BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
 621		J_ASSERT_BH(bitmap_bh,
 622				bh2jh(bitmap_bh)->b_committed_data != NULL);
 623		ext3_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
 624				bh2jh(bitmap_bh)->b_committed_data);
 625
 626		/*
 627		 * We clear the bit in the bitmap after setting the committed
 628		 * data bit, because this is the reverse order to that which
 629		 * the allocator uses.
 630		 */
 631		BUFFER_TRACE(bitmap_bh, "clear bit");
 632		if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
 633						bit + i, bitmap_bh->b_data)) {
 634			jbd_unlock_bh_state(bitmap_bh);
 635			ext3_error(sb, __func__,
 636				"bit already cleared for block "E3FSBLK,
 637				 block + i);
 638			jbd_lock_bh_state(bitmap_bh);
 639			BUFFER_TRACE(bitmap_bh, "bit already cleared");
 640		} else {
 641			group_freed++;
 642		}
 643	}
 644	jbd_unlock_bh_state(bitmap_bh);
 645
 646	spin_lock(sb_bgl_lock(sbi, block_group));
 647	le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
 648	spin_unlock(sb_bgl_lock(sbi, block_group));
 649	percpu_counter_add(&sbi->s_freeblocks_counter, count);
 650
 651	/* We dirtied the bitmap block */
 652	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 653	err = ext3_journal_dirty_metadata(handle, bitmap_bh);
 654
 655	/* And the group descriptor block */
 656	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
 657	ret = ext3_journal_dirty_metadata(handle, gd_bh);
 658	if (!err) err = ret;
 659	*pdquot_freed_blocks += group_freed;
 660
 661	if (overflow && !err) {
 662		block += count;
 663		count = overflow;
 664		goto do_more;
 665	}
 666
 667error_return:
 668	brelse(bitmap_bh);
 669	ext3_std_error(sb, err);
 670	return;
 671}
 672
 673/**
 674 * ext3_free_blocks() -- Free given blocks and update quota
 675 * @handle:		handle for this transaction
 676 * @inode:		inode
 677 * @block:		start physical block to free
 678 * @count:		number of blocks to count
 679 */
 680void ext3_free_blocks(handle_t *handle, struct inode *inode,
 681			ext3_fsblk_t block, unsigned long count)
 682{
 683	struct super_block *sb = inode->i_sb;
 684	unsigned long dquot_freed_blocks;
 685
 686	trace_ext3_free_blocks(inode, block, count);
 687	ext3_free_blocks_sb(handle, sb, block, count, &dquot_freed_blocks);
 688	if (dquot_freed_blocks)
 689		dquot_free_block(inode, dquot_freed_blocks);
 690	return;
 691}
 692
 693/**
 694 * ext3_test_allocatable()
 695 * @nr:			given allocation block group
 696 * @bh:			bufferhead contains the bitmap of the given block group
 697 *
 698 * For ext3 allocations, we must not reuse any blocks which are
 699 * allocated in the bitmap buffer's "last committed data" copy.  This
 700 * prevents deletes from freeing up the page for reuse until we have
 701 * committed the delete transaction.
 702 *
 703 * If we didn't do this, then deleting something and reallocating it as
 704 * data would allow the old block to be overwritten before the
 705 * transaction committed (because we force data to disk before commit).
 706 * This would lead to corruption if we crashed between overwriting the
 707 * data and committing the delete.
 708 *
 709 * @@@ We may want to make this allocation behaviour conditional on
 710 * data-writes at some point, and disable it for metadata allocations or
 711 * sync-data inodes.
 712 */
 713static int ext3_test_allocatable(ext3_grpblk_t nr, struct buffer_head *bh)
 714{
 715	int ret;
 716	struct journal_head *jh = bh2jh(bh);
 717
 718	if (ext3_test_bit(nr, bh->b_data))
 719		return 0;
 720
 721	jbd_lock_bh_state(bh);
 722	if (!jh->b_committed_data)
 723		ret = 1;
 724	else
 725		ret = !ext3_test_bit(nr, jh->b_committed_data);
 726	jbd_unlock_bh_state(bh);
 727	return ret;
 728}
 729
 730/**
 731 * bitmap_search_next_usable_block()
 732 * @start:		the starting block (group relative) of the search
 733 * @bh:			bufferhead contains the block group bitmap
 734 * @maxblocks:		the ending block (group relative) of the reservation
 735 *
 736 * The bitmap search --- search forward alternately through the actual
 737 * bitmap on disk and the last-committed copy in journal, until we find a
 738 * bit free in both bitmaps.
 739 */
 740static ext3_grpblk_t
 741bitmap_search_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
 742					ext3_grpblk_t maxblocks)
 743{
 744	ext3_grpblk_t next;
 745	struct journal_head *jh = bh2jh(bh);
 746
 747	while (start < maxblocks) {
 748		next = ext3_find_next_zero_bit(bh->b_data, maxblocks, start);
 749		if (next >= maxblocks)
 750			return -1;
 751		if (ext3_test_allocatable(next, bh))
 752			return next;
 753		jbd_lock_bh_state(bh);
 754		if (jh->b_committed_data)
 755			start = ext3_find_next_zero_bit(jh->b_committed_data,
 756							maxblocks, next);
 757		jbd_unlock_bh_state(bh);
 758	}
 759	return -1;
 760}
 761
 762/**
 763 * find_next_usable_block()
 764 * @start:		the starting block (group relative) to find next
 765 *			allocatable block in bitmap.
 766 * @bh:			bufferhead contains the block group bitmap
 767 * @maxblocks:		the ending block (group relative) for the search
 768 *
 769 * Find an allocatable block in a bitmap.  We honor both the bitmap and
 770 * its last-committed copy (if that exists), and perform the "most
 771 * appropriate allocation" algorithm of looking for a free block near
 772 * the initial goal; then for a free byte somewhere in the bitmap; then
 773 * for any free bit in the bitmap.
 774 */
 775static ext3_grpblk_t
 776find_next_usable_block(ext3_grpblk_t start, struct buffer_head *bh,
 777			ext3_grpblk_t maxblocks)
 778{
 779	ext3_grpblk_t here, next;
 780	char *p, *r;
 781
 782	if (start > 0) {
 783		/*
 784		 * The goal was occupied; search forward for a free
 785		 * block within the next XX blocks.
 786		 *
 787		 * end_goal is more or less random, but it has to be
 788		 * less than EXT3_BLOCKS_PER_GROUP. Aligning up to the
 789		 * next 64-bit boundary is simple..
 790		 */
 791		ext3_grpblk_t end_goal = (start + 63) & ~63;
 792		if (end_goal > maxblocks)
 793			end_goal = maxblocks;
 794		here = ext3_find_next_zero_bit(bh->b_data, end_goal, start);
 795		if (here < end_goal && ext3_test_allocatable(here, bh))
 796			return here;
 797		ext3_debug("Bit not found near goal\n");
 798	}
 799
 800	here = start;
 801	if (here < 0)
 802		here = 0;
 803
 804	p = bh->b_data + (here >> 3);
 805	r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
 806	next = (r - bh->b_data) << 3;
 807
 808	if (next < maxblocks && next >= start && ext3_test_allocatable(next, bh))
 809		return next;
 810
 811	/*
 812	 * The bitmap search --- search forward alternately through the actual
 813	 * bitmap and the last-committed copy until we find a bit free in
 814	 * both
 815	 */
 816	here = bitmap_search_next_usable_block(here, bh, maxblocks);
 817	return here;
 818}
 819
 820/**
 821 * claim_block()
 822 * @lock:		the spin lock for this block group
 823 * @block:		the free block (group relative) to allocate
 824 * @bh:			the buffer_head contains the block group bitmap
 825 *
 826 * We think we can allocate this block in this bitmap.  Try to set the bit.
 827 * If that succeeds then check that nobody has allocated and then freed the
 828 * block since we saw that is was not marked in b_committed_data.  If it _was_
 829 * allocated and freed then clear the bit in the bitmap again and return
 830 * zero (failure).
 831 */
 832static inline int
 833claim_block(spinlock_t *lock, ext3_grpblk_t block, struct buffer_head *bh)
 834{
 835	struct journal_head *jh = bh2jh(bh);
 836	int ret;
 837
 838	if (ext3_set_bit_atomic(lock, block, bh->b_data))
 839		return 0;
 840	jbd_lock_bh_state(bh);
 841	if (jh->b_committed_data && ext3_test_bit(block,jh->b_committed_data)) {
 842		ext3_clear_bit_atomic(lock, block, bh->b_data);
 843		ret = 0;
 844	} else {
 845		ret = 1;
 846	}
 847	jbd_unlock_bh_state(bh);
 848	return ret;
 849}
 850
 851/**
 852 * ext3_try_to_allocate()
 853 * @sb:			superblock
 854 * @handle:		handle to this transaction
 855 * @group:		given allocation block group
 856 * @bitmap_bh:		bufferhead holds the block bitmap
 857 * @grp_goal:		given target block within the group
 858 * @count:		target number of blocks to allocate
 859 * @my_rsv:		reservation window
 860 *
 861 * Attempt to allocate blocks within a give range. Set the range of allocation
 862 * first, then find the first free bit(s) from the bitmap (within the range),
 863 * and at last, allocate the blocks by claiming the found free bit as allocated.
 864 *
 865 * To set the range of this allocation:
 866 *	if there is a reservation window, only try to allocate block(s) from the
 867 *	file's own reservation window;
 868 *	Otherwise, the allocation range starts from the give goal block, ends at
 869 *	the block group's last block.
 870 *
 871 * If we failed to allocate the desired block then we may end up crossing to a
 872 * new bitmap.  In that case we must release write access to the old one via
 873 * ext3_journal_release_buffer(), else we'll run out of credits.
 874 */
 875static ext3_grpblk_t
 876ext3_try_to_allocate(struct super_block *sb, handle_t *handle, int group,
 877			struct buffer_head *bitmap_bh, ext3_grpblk_t grp_goal,
 878			unsigned long *count, struct ext3_reserve_window *my_rsv)
 879{
 880	ext3_fsblk_t group_first_block;
 881	ext3_grpblk_t start, end;
 882	unsigned long num = 0;
 883
 884	/* we do allocation within the reservation window if we have a window */
 885	if (my_rsv) {
 886		group_first_block = ext3_group_first_block_no(sb, group);
 887		if (my_rsv->_rsv_start >= group_first_block)
 888			start = my_rsv->_rsv_start - group_first_block;
 889		else
 890			/* reservation window cross group boundary */
 891			start = 0;
 892		end = my_rsv->_rsv_end - group_first_block + 1;
 893		if (end > EXT3_BLOCKS_PER_GROUP(sb))
 894			/* reservation window crosses group boundary */
 895			end = EXT3_BLOCKS_PER_GROUP(sb);
 896		if ((start <= grp_goal) && (grp_goal < end))
 897			start = grp_goal;
 898		else
 899			grp_goal = -1;
 900	} else {
 901		if (grp_goal > 0)
 902			start = grp_goal;
 903		else
 904			start = 0;
 905		end = EXT3_BLOCKS_PER_GROUP(sb);
 906	}
 907
 908	BUG_ON(start > EXT3_BLOCKS_PER_GROUP(sb));
 909
 910repeat:
 911	if (grp_goal < 0 || !ext3_test_allocatable(grp_goal, bitmap_bh)) {
 912		grp_goal = find_next_usable_block(start, bitmap_bh, end);
 913		if (grp_goal < 0)
 914			goto fail_access;
 915		if (!my_rsv) {
 916			int i;
 917
 918			for (i = 0; i < 7 && grp_goal > start &&
 919					ext3_test_allocatable(grp_goal - 1,
 920								bitmap_bh);
 921					i++, grp_goal--)
 922				;
 923		}
 924	}
 925	start = grp_goal;
 926
 927	if (!claim_block(sb_bgl_lock(EXT3_SB(sb), group),
 928		grp_goal, bitmap_bh)) {
 929		/*
 930		 * The block was allocated by another thread, or it was
 931		 * allocated and then freed by another thread
 932		 */
 933		start++;
 934		grp_goal++;
 935		if (start >= end)
 936			goto fail_access;
 937		goto repeat;
 938	}
 939	num++;
 940	grp_goal++;
 941	while (num < *count && grp_goal < end
 942		&& ext3_test_allocatable(grp_goal, bitmap_bh)
 943		&& claim_block(sb_bgl_lock(EXT3_SB(sb), group),
 944				grp_goal, bitmap_bh)) {
 945		num++;
 946		grp_goal++;
 947	}
 948	*count = num;
 949	return grp_goal - num;
 950fail_access:
 951	*count = num;
 952	return -1;
 953}
 954
 955/**
 956 *	find_next_reservable_window():
 957 *		find a reservable space within the given range.
 958 *		It does not allocate the reservation window for now:
 959 *		alloc_new_reservation() will do the work later.
 960 *
 961 *	@search_head: the head of the searching list;
 962 *		This is not necessarily the list head of the whole filesystem
 963 *
 964 *		We have both head and start_block to assist the search
 965 *		for the reservable space. The list starts from head,
 966 *		but we will shift to the place where start_block is,
 967 *		then start from there, when looking for a reservable space.
 968 *
 969 *	@my_rsv: the reservation window
 970 *
 971 *	@sb: the super block
 972 *
 973 *	@start_block: the first block we consider to start
 974 *			the real search from
 975 *
 976 *	@last_block:
 977 *		the maximum block number that our goal reservable space
 978 *		could start from. This is normally the last block in this
 979 *		group. The search will end when we found the start of next
 980 *		possible reservable space is out of this boundary.
 981 *		This could handle the cross boundary reservation window
 982 *		request.
 983 *
 984 *	basically we search from the given range, rather than the whole
 985 *	reservation double linked list, (start_block, last_block)
 986 *	to find a free region that is of my size and has not
 987 *	been reserved.
 988 *
 989 */
 990static int find_next_reservable_window(
 991				struct ext3_reserve_window_node *search_head,
 992				struct ext3_reserve_window_node *my_rsv,
 993				struct super_block * sb,
 994				ext3_fsblk_t start_block,
 995				ext3_fsblk_t last_block)
 996{
 997	struct rb_node *next;
 998	struct ext3_reserve_window_node *rsv, *prev;
 999	ext3_fsblk_t cur;
1000	int size = my_rsv->rsv_goal_size;
1001
1002	/* TODO: make the start of the reservation window byte-aligned */
1003	/* cur = *start_block & ~7;*/
1004	cur = start_block;
1005	rsv = search_head;
1006	if (!rsv)
1007		return -1;
1008
1009	while (1) {
1010		if (cur <= rsv->rsv_end)
1011			cur = rsv->rsv_end + 1;
1012
1013		/* TODO?
1014		 * in the case we could not find a reservable space
1015		 * that is what is expected, during the re-search, we could
1016		 * remember what's the largest reservable space we could have
1017		 * and return that one.
1018		 *
1019		 * For now it will fail if we could not find the reservable
1020		 * space with expected-size (or more)...
1021		 */
1022		if (cur > last_block)
1023			return -1;		/* fail */
1024
1025		prev = rsv;
1026		next = rb_next(&rsv->rsv_node);
1027		rsv = rb_entry(next,struct ext3_reserve_window_node,rsv_node);
1028
1029		/*
1030		 * Reached the last reservation, we can just append to the
1031		 * previous one.
1032		 */
1033		if (!next)
1034			break;
1035
1036		if (cur + size <= rsv->rsv_start) {
1037			/*
1038			 * Found a reserveable space big enough.  We could
1039			 * have a reservation across the group boundary here
1040			 */
1041			break;
1042		}
1043	}
1044	/*
1045	 * we come here either :
1046	 * when we reach the end of the whole list,
1047	 * and there is empty reservable space after last entry in the list.
1048	 * append it to the end of the list.
1049	 *
1050	 * or we found one reservable space in the middle of the list,
1051	 * return the reservation window that we could append to.
1052	 * succeed.
1053	 */
1054
1055	if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
1056		rsv_window_remove(sb, my_rsv);
1057
1058	/*
1059	 * Let's book the whole available window for now.  We will check the
1060	 * disk bitmap later and then, if there are free blocks then we adjust
1061	 * the window size if it's larger than requested.
1062	 * Otherwise, we will remove this node from the tree next time
1063	 * call find_next_reservable_window.
1064	 */
1065	my_rsv->rsv_start = cur;
1066	my_rsv->rsv_end = cur + size - 1;
1067	my_rsv->rsv_alloc_hit = 0;
1068
1069	if (prev != my_rsv)
1070		ext3_rsv_window_add(sb, my_rsv);
1071
1072	return 0;
1073}
1074
1075/**
1076 *	alloc_new_reservation()--allocate a new reservation window
1077 *
1078 *		To make a new reservation, we search part of the filesystem
1079 *		reservation list (the list that inside the group). We try to
1080 *		allocate a new reservation window near the allocation goal,
1081 *		or the beginning of the group, if there is no goal.
1082 *
1083 *		We first find a reservable space after the goal, then from
1084 *		there, we check the bitmap for the first free block after
1085 *		it. If there is no free block until the end of group, then the
1086 *		whole group is full, we failed. Otherwise, check if the free
1087 *		block is inside the expected reservable space, if so, we
1088 *		succeed.
1089 *		If the first free block is outside the reservable space, then
1090 *		start from the first free block, we search for next available
1091 *		space, and go on.
1092 *
1093 *	on succeed, a new reservation will be found and inserted into the list
1094 *	It contains at least one free block, and it does not overlap with other
1095 *	reservation windows.
1096 *
1097 *	failed: we failed to find a reservation window in this group
1098 *
1099 *	@my_rsv: the reservation window
1100 *
1101 *	@grp_goal: The goal (group-relative).  It is where the search for a
1102 *		free reservable space should start from.
1103 *		if we have a grp_goal(grp_goal >0 ), then start from there,
1104 *		no grp_goal(grp_goal = -1), we start from the first block
1105 *		of the group.
1106 *
1107 *	@sb: the super block
1108 *	@group: the group we are trying to allocate in
1109 *	@bitmap_bh: the block group block bitmap
1110 *
1111 */
1112static int alloc_new_reservation(struct ext3_reserve_window_node *my_rsv,
1113		ext3_grpblk_t grp_goal, struct super_block *sb,
1114		unsigned int group, struct buffer_head *bitmap_bh)
1115{
1116	struct ext3_reserve_window_node *search_head;
1117	ext3_fsblk_t group_first_block, group_end_block, start_block;
1118	ext3_grpblk_t first_free_block;
1119	struct rb_root *fs_rsv_root = &EXT3_SB(sb)->s_rsv_window_root;
1120	unsigned long size;
1121	int ret;
1122	spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1123
1124	group_first_block = ext3_group_first_block_no(sb, group);
1125	group_end_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1126
1127	if (grp_goal < 0)
1128		start_block = group_first_block;
1129	else
1130		start_block = grp_goal + group_first_block;
1131
1132	trace_ext3_alloc_new_reservation(sb, start_block);
1133	size = my_rsv->rsv_goal_size;
1134
1135	if (!rsv_is_empty(&my_rsv->rsv_window)) {
1136		/*
1137		 * if the old reservation is cross group boundary
1138		 * and if the goal is inside the old reservation window,
1139		 * we will come here when we just failed to allocate from
1140		 * the first part of the window. We still have another part
1141		 * that belongs to the next group. In this case, there is no
1142		 * point to discard our window and try to allocate a new one
1143		 * in this group(which will fail). we should
1144		 * keep the reservation window, just simply move on.
1145		 *
1146		 * Maybe we could shift the start block of the reservation
1147		 * window to the first block of next group.
1148		 */
1149
1150		if ((my_rsv->rsv_start <= group_end_block) &&
1151				(my_rsv->rsv_end > group_end_block) &&
1152				(start_block >= my_rsv->rsv_start))
1153			return -1;
1154
1155		if ((my_rsv->rsv_alloc_hit >
1156		     (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
1157			/*
1158			 * if the previously allocation hit ratio is
1159			 * greater than 1/2, then we double the size of
1160			 * the reservation window the next time,
1161			 * otherwise we keep the same size window
1162			 */
1163			size = size * 2;
1164			if (size > EXT3_MAX_RESERVE_BLOCKS)
1165				size = EXT3_MAX_RESERVE_BLOCKS;
1166			my_rsv->rsv_goal_size= size;
1167		}
1168	}
1169
1170	spin_lock(rsv_lock);
1171	/*
1172	 * shift the search start to the window near the goal block
1173	 */
1174	search_head = search_reserve_window(fs_rsv_root, start_block);
1175
1176	/*
1177	 * find_next_reservable_window() simply finds a reservable window
1178	 * inside the given range(start_block, group_end_block).
1179	 *
1180	 * To make sure the reservation window has a free bit inside it, we
1181	 * need to check the bitmap after we found a reservable window.
1182	 */
1183retry:
1184	ret = find_next_reservable_window(search_head, my_rsv, sb,
1185						start_block, group_end_block);
1186
1187	if (ret == -1) {
1188		if (!rsv_is_empty(&my_rsv->rsv_window))
1189			rsv_window_remove(sb, my_rsv);
1190		spin_unlock(rsv_lock);
1191		return -1;
1192	}
1193
1194	/*
1195	 * On success, find_next_reservable_window() returns the
1196	 * reservation window where there is a reservable space after it.
1197	 * Before we reserve this reservable space, we need
1198	 * to make sure there is at least a free block inside this region.
1199	 *
1200	 * searching the first free bit on the block bitmap and copy of
1201	 * last committed bitmap alternatively, until we found a allocatable
1202	 * block. Search start from the start block of the reservable space
1203	 * we just found.
1204	 */
1205	spin_unlock(rsv_lock);
1206	first_free_block = bitmap_search_next_usable_block(
1207			my_rsv->rsv_start - group_first_block,
1208			bitmap_bh, group_end_block - group_first_block + 1);
1209
1210	if (first_free_block < 0) {
1211		/*
1212		 * no free block left on the bitmap, no point
1213		 * to reserve the space. return failed.
1214		 */
1215		spin_lock(rsv_lock);
1216		if (!rsv_is_empty(&my_rsv->rsv_window))
1217			rsv_window_remove(sb, my_rsv);
1218		spin_unlock(rsv_lock);
1219		return -1;		/* failed */
1220	}
1221
1222	start_block = first_free_block + group_first_block;
1223	/*
1224	 * check if the first free block is within the
1225	 * free space we just reserved
1226	 */
1227	if (start_block >= my_rsv->rsv_start &&
1228	    start_block <= my_rsv->rsv_end) {
1229		trace_ext3_reserved(sb, start_block, my_rsv);
1230		return 0;		/* success */
1231	}
1232	/*
1233	 * if the first free bit we found is out of the reservable space
1234	 * continue search for next reservable space,
1235	 * start from where the free block is,
1236	 * we also shift the list head to where we stopped last time
1237	 */
1238	search_head = my_rsv;
1239	spin_lock(rsv_lock);
1240	goto retry;
1241}
1242
1243/**
1244 * try_to_extend_reservation()
1245 * @my_rsv:		given reservation window
1246 * @sb:			super block
1247 * @size:		the delta to extend
1248 *
1249 * Attempt to expand the reservation window large enough to have
1250 * required number of free blocks
1251 *
1252 * Since ext3_try_to_allocate() will always allocate blocks within
1253 * the reservation window range, if the window size is too small,
1254 * multiple blocks allocation has to stop at the end of the reservation
1255 * window. To make this more efficient, given the total number of
1256 * blocks needed and the current size of the window, we try to
1257 * expand the reservation window size if necessary on a best-effort
1258 * basis before ext3_new_blocks() tries to allocate blocks,
1259 */
1260static void try_to_extend_reservation(struct ext3_reserve_window_node *my_rsv,
1261			struct super_block *sb, int size)
1262{
1263	struct ext3_reserve_window_node *next_rsv;
1264	struct rb_node *next;
1265	spinlock_t *rsv_lock = &EXT3_SB(sb)->s_rsv_window_lock;
1266
1267	if (!spin_trylock(rsv_lock))
1268		return;
1269
1270	next = rb_next(&my_rsv->rsv_node);
1271
1272	if (!next)
1273		my_rsv->rsv_end += size;
1274	else {
1275		next_rsv = rb_entry(next, struct ext3_reserve_window_node, rsv_node);
1276
1277		if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
1278			my_rsv->rsv_end += size;
1279		else
1280			my_rsv->rsv_end = next_rsv->rsv_start - 1;
1281	}
1282	spin_unlock(rsv_lock);
1283}
1284
1285/**
1286 * ext3_try_to_allocate_with_rsv()
1287 * @sb:			superblock
1288 * @handle:		handle to this transaction
1289 * @group:		given allocation block group
1290 * @bitmap_bh:		bufferhead holds the block bitmap
1291 * @grp_goal:		given target block within the group
1292 * @my_rsv:		reservation window
1293 * @count:		target number of blocks to allocate
1294 * @errp:		pointer to store the error code
1295 *
1296 * This is the main function used to allocate a new block and its reservation
1297 * window.
1298 *
1299 * Each time when a new block allocation is need, first try to allocate from
1300 * its own reservation.  If it does not have a reservation window, instead of
1301 * looking for a free bit on bitmap first, then look up the reservation list to
1302 * see if it is inside somebody else's reservation window, we try to allocate a
1303 * reservation window for it starting from the goal first. Then do the block
1304 * allocation within the reservation window.
1305 *
1306 * This will avoid keeping on searching the reservation list again and
1307 * again when somebody is looking for a free block (without
1308 * reservation), and there are lots of free blocks, but they are all
1309 * being reserved.
1310 *
1311 * We use a red-black tree for the per-filesystem reservation list.
1312 *
1313 */
1314static ext3_grpblk_t
1315ext3_try_to_allocate_with_rsv(struct super_block *sb, handle_t *handle,
1316			unsigned int group, struct buffer_head *bitmap_bh,
1317			ext3_grpblk_t grp_goal,
1318			struct ext3_reserve_window_node * my_rsv,
1319			unsigned long *count, int *errp)
1320{
1321	ext3_fsblk_t group_first_block, group_last_block;
1322	ext3_grpblk_t ret = 0;
1323	int fatal;
1324	unsigned long num = *count;
1325
1326	*errp = 0;
1327
1328	/*
1329	 * Make sure we use undo access for the bitmap, because it is critical
1330	 * that we do the frozen_data COW on bitmap buffers in all cases even
1331	 * if the buffer is in BJ_Forget state in the committing transaction.
1332	 */
1333	BUFFER_TRACE(bitmap_bh, "get undo access for new block");
1334	fatal = ext3_journal_get_undo_access(handle, bitmap_bh);
1335	if (fatal) {
1336		*errp = fatal;
1337		return -1;
1338	}
1339
1340	/*
1341	 * we don't deal with reservation when
1342	 * filesystem is mounted without reservation
1343	 * or the file is not a regular file
1344	 * or last attempt to allocate a block with reservation turned on failed
1345	 */
1346	if (my_rsv == NULL ) {
1347		ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1348						grp_goal, count, NULL);
1349		goto out;
1350	}
1351	/*
1352	 * grp_goal is a group relative block number (if there is a goal)
1353	 * 0 <= grp_goal < EXT3_BLOCKS_PER_GROUP(sb)
1354	 * first block is a filesystem wide block number
1355	 * first block is the block number of the first block in this group
1356	 */
1357	group_first_block = ext3_group_first_block_no(sb, group);
1358	group_last_block = group_first_block + (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1359
1360	/*
1361	 * Basically we will allocate a new block from inode's reservation
1362	 * window.
1363	 *
1364	 * We need to allocate a new reservation window, if:
1365	 * a) inode does not have a reservation window; or
1366	 * b) last attempt to allocate a block from existing reservation
1367	 *    failed; or
1368	 * c) we come here with a goal and with a reservation window
1369	 *
1370	 * We do not need to allocate a new reservation window if we come here
1371	 * at the beginning with a goal and the goal is inside the window, or
1372	 * we don't have a goal but already have a reservation window.
1373	 * then we could go to allocate from the reservation window directly.
1374	 */
1375	while (1) {
1376		if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
1377			!goal_in_my_reservation(&my_rsv->rsv_window,
1378						grp_goal, group, sb)) {
1379			if (my_rsv->rsv_goal_size < *count)
1380				my_rsv->rsv_goal_size = *count;
1381			ret = alloc_new_reservation(my_rsv, grp_goal, sb,
1382							group, bitmap_bh);
1383			if (ret < 0)
1384				break;			/* failed */
1385
1386			if (!goal_in_my_reservation(&my_rsv->rsv_window,
1387							grp_goal, group, sb))
1388				grp_goal = -1;
1389		} else if (grp_goal >= 0) {
1390			int curr = my_rsv->rsv_end -
1391					(grp_goal + group_first_block) + 1;
1392
1393			if (curr < *count)
1394				try_to_extend_reservation(my_rsv, sb,
1395							*count - curr);
1396		}
1397
1398		if ((my_rsv->rsv_start > group_last_block) ||
1399				(my_rsv->rsv_end < group_first_block)) {
1400			rsv_window_dump(&EXT3_SB(sb)->s_rsv_window_root, 1);
1401			BUG();
1402		}
1403		ret = ext3_try_to_allocate(sb, handle, group, bitmap_bh,
1404					   grp_goal, &num, &my_rsv->rsv_window);
1405		if (ret >= 0) {
1406			my_rsv->rsv_alloc_hit += num;
1407			*count = num;
1408			break;				/* succeed */
1409		}
1410		num = *count;
1411	}
1412out:
1413	if (ret >= 0) {
1414		BUFFER_TRACE(bitmap_bh, "journal_dirty_metadata for "
1415					"bitmap block");
1416		fatal = ext3_journal_dirty_metadata(handle, bitmap_bh);
1417		if (fatal) {
1418			*errp = fatal;
1419			return -1;
1420		}
1421		return ret;
1422	}
1423
1424	BUFFER_TRACE(bitmap_bh, "journal_release_buffer");
1425	ext3_journal_release_buffer(handle, bitmap_bh);
1426	return ret;
1427}
1428
1429/**
1430 * ext3_has_free_blocks()
1431 * @sbi:		in-core super block structure.
1432 *
1433 * Check if filesystem has at least 1 free block available for allocation.
1434 */
1435static int ext3_has_free_blocks(struct ext3_sb_info *sbi, int use_reservation)
1436{
1437	ext3_fsblk_t free_blocks, root_blocks;
1438
1439	free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1440	root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
1441	if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
1442		!use_reservation && !uid_eq(sbi->s_resuid, current_fsuid()) &&
1443		(gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) ||
1444		 !in_group_p (sbi->s_resgid))) {
1445		return 0;
1446	}
1447	return 1;
1448}
1449
1450/**
1451 * ext3_should_retry_alloc()
1452 * @sb:			super block
1453 * @retries		number of attemps has been made
1454 *
1455 * ext3_should_retry_alloc() is called when ENOSPC is returned, and if
1456 * it is profitable to retry the operation, this function will wait
1457 * for the current or committing transaction to complete, and then
1458 * return TRUE.
1459 *
1460 * if the total number of retries exceed three times, return FALSE.
1461 */
1462int ext3_should_retry_alloc(struct super_block *sb, int *retries)
1463{
1464	if (!ext3_has_free_blocks(EXT3_SB(sb), 0) || (*retries)++ > 3)
1465		return 0;
1466
1467	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
1468
1469	return journal_force_commit_nested(EXT3_SB(sb)->s_journal);
1470}
1471
1472/**
1473 * ext3_new_blocks() -- core block(s) allocation function
1474 * @handle:		handle to this transaction
1475 * @inode:		file inode
1476 * @goal:		given target block(filesystem wide)
1477 * @count:		target number of blocks to allocate
1478 * @errp:		error code
1479 *
1480 * ext3_new_blocks uses a goal block to assist allocation.  It tries to
1481 * allocate block(s) from the block group contains the goal block first. If that
1482 * fails, it will try to allocate block(s) from other block groups without
1483 * any specific goal block.
1484 *
1485 */
1486ext3_fsblk_t ext3_new_blocks(handle_t *handle, struct inode *inode,
1487			ext3_fsblk_t goal, unsigned long *count, int *errp)
1488{
1489	struct buffer_head *bitmap_bh = NULL;
1490	struct buffer_head *gdp_bh;
1491	int group_no;
1492	int goal_group;
1493	ext3_grpblk_t grp_target_blk;	/* blockgroup relative goal block */
1494	ext3_grpblk_t grp_alloc_blk;	/* blockgroup-relative allocated block*/
1495	ext3_fsblk_t ret_block;		/* filesyetem-wide allocated block */
1496	int bgi;			/* blockgroup iteration index */
1497	int fatal = 0, err;
1498	int performed_allocation = 0;
1499	ext3_grpblk_t free_blocks;	/* number of free blocks in a group */
1500	struct super_block *sb;
1501	struct ext3_group_desc *gdp;
1502	struct ext3_super_block *es;
1503	struct ext3_sb_info *sbi;
1504	struct ext3_reserve_window_node *my_rsv = NULL;
1505	struct ext3_block_alloc_info *block_i;
1506	unsigned short windowsz = 0;
1507#ifdef EXT3FS_DEBUG
1508	static int goal_hits, goal_attempts;
1509#endif
1510	unsigned long ngroups;
1511	unsigned long num = *count;
1512
1513	*errp = -ENOSPC;
1514	sb = inode->i_sb;
1515
1516	/*
1517	 * Check quota for allocation of this block.
1518	 */
1519	err = dquot_alloc_block(inode, num);
1520	if (err) {
1521		*errp = err;
1522		return 0;
1523	}
1524
1525	trace_ext3_request_blocks(inode, goal, num);
1526
1527	sbi = EXT3_SB(sb);
1528	es = sbi->s_es;
1529	ext3_debug("goal=%lu.\n", goal);
1530	/*
1531	 * Allocate a block from reservation only when
1532	 * filesystem is mounted with reservation(default,-o reservation), and
1533	 * it's a regular file, and
1534	 * the desired window size is greater than 0 (One could use ioctl
1535	 * command EXT3_IOC_SETRSVSZ to set the window size to 0 to turn off
1536	 * reservation on that particular file)
1537	 */
1538	block_i = EXT3_I(inode)->i_block_alloc_info;
1539	if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1540		my_rsv = &block_i->rsv_window_node;
1541
1542	if (!ext3_has_free_blocks(sbi, IS_NOQUOTA(inode))) {
1543		*errp = -ENOSPC;
1544		goto out;
1545	}
1546
1547	/*
1548	 * First, test whether the goal block is free.
1549	 */
1550	if (goal < le32_to_cpu(es->s_first_data_block) ||
1551	    goal >= le32_to_cpu(es->s_blocks_count))
1552		goal = le32_to_cpu(es->s_first_data_block);
1553	group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
1554			EXT3_BLOCKS_PER_GROUP(sb);
1555	goal_group = group_no;
1556retry_alloc:
1557	gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1558	if (!gdp)
1559		goto io_error;
1560
1561	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1562	/*
1563	 * if there is not enough free blocks to make a new resevation
1564	 * turn off reservation for this allocation
1565	 */
1566	if (my_rsv && (free_blocks < windowsz)
1567		&& (free_blocks > 0)
1568		&& (rsv_is_empty(&my_rsv->rsv_window)))
1569		my_rsv = NULL;
1570
1571	if (free_blocks > 0) {
1572		grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
1573				EXT3_BLOCKS_PER_GROUP(sb));
1574		bitmap_bh = read_block_bitmap(sb, group_no);
1575		if (!bitmap_bh)
1576			goto io_error;
1577		grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1578					group_no, bitmap_bh, grp_target_blk,
1579					my_rsv,	&num, &fatal);
1580		if (fatal)
1581			goto out;
1582		if (grp_alloc_blk >= 0)
1583			goto allocated;
1584	}
1585
1586	ngroups = EXT3_SB(sb)->s_groups_count;
1587	smp_rmb();
1588
1589	/*
1590	 * Now search the rest of the groups.  We assume that
1591	 * group_no and gdp correctly point to the last group visited.
1592	 */
1593	for (bgi = 0; bgi < ngroups; bgi++) {
1594		group_no++;
1595		if (group_no >= ngroups)
1596			group_no = 0;
1597		gdp = ext3_get_group_desc(sb, group_no, &gdp_bh);
1598		if (!gdp)
1599			goto io_error;
1600		free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1601		/*
1602		 * skip this group (and avoid loading bitmap) if there
1603		 * are no free blocks
1604		 */
1605		if (!free_blocks)
1606			continue;
1607		/*
1608		 * skip this group if the number of
1609		 * free blocks is less than half of the reservation
1610		 * window size.
1611		 */
1612		if (my_rsv && (free_blocks <= (windowsz/2)))
1613			continue;
1614
1615		brelse(bitmap_bh);
1616		bitmap_bh = read_block_bitmap(sb, group_no);
1617		if (!bitmap_bh)
1618			goto io_error;
1619		/*
1620		 * try to allocate block(s) from this group, without a goal(-1).
1621		 */
1622		grp_alloc_blk = ext3_try_to_allocate_with_rsv(sb, handle,
1623					group_no, bitmap_bh, -1, my_rsv,
1624					&num, &fatal);
1625		if (fatal)
1626			goto out;
1627		if (grp_alloc_blk >= 0)
1628			goto allocated;
1629	}
1630	/*
1631	 * We may end up a bogus earlier ENOSPC error due to
1632	 * filesystem is "full" of reservations, but
1633	 * there maybe indeed free blocks available on disk
1634	 * In this case, we just forget about the reservations
1635	 * just do block allocation as without reservations.
1636	 */
1637	if (my_rsv) {
1638		my_rsv = NULL;
1639		windowsz = 0;
1640		group_no = goal_group;
1641		goto retry_alloc;
1642	}
1643	/* No space left on the device */
1644	*errp = -ENOSPC;
1645	goto out;
1646
1647allocated:
1648
1649	ext3_debug("using block group %d(%d)\n",
1650			group_no, gdp->bg_free_blocks_count);
1651
1652	BUFFER_TRACE(gdp_bh, "get_write_access");
1653	fatal = ext3_journal_get_write_access(handle, gdp_bh);
1654	if (fatal)
1655		goto out;
1656
1657	ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no);
1658
1659	if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
1660	    in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
1661	    in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
1662		      EXT3_SB(sb)->s_itb_per_group) ||
1663	    in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
1664		      EXT3_SB(sb)->s_itb_per_group)) {
1665		ext3_error(sb, "ext3_new_block",
1666			    "Allocating block in system zone - "
1667			    "blocks from "E3FSBLK", length %lu",
1668			     ret_block, num);
1669		/*
1670		 * claim_block() marked the blocks we allocated as in use. So we
1671		 * may want to selectively mark some of the blocks as free.
1672		 */
1673		goto retry_alloc;
1674	}
1675
1676	performed_allocation = 1;
1677
1678#ifdef CONFIG_JBD_DEBUG
1679	{
1680		struct buffer_head *debug_bh;
1681
1682		/* Record bitmap buffer state in the newly allocated block */
1683		debug_bh = sb_find_get_block(sb, ret_block);
1684		if (debug_bh) {
1685			BUFFER_TRACE(debug_bh, "state when allocated");
1686			BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap state");
1687			brelse(debug_bh);
1688		}
1689	}
1690	jbd_lock_bh_state(bitmap_bh);
1691	spin_lock(sb_bgl_lock(sbi, group_no));
1692	if (buffer_jbd(bitmap_bh) && bh2jh(bitmap_bh)->b_committed_data) {
1693		int i;
1694
1695		for (i = 0; i < num; i++) {
1696			if (ext3_test_bit(grp_alloc_blk+i,
1697					bh2jh(bitmap_bh)->b_committed_data)) {
1698				printk("%s: block was unexpectedly set in "
1699					"b_committed_data\n", __func__);
1700			}
1701		}
1702	}
1703	ext3_debug("found bit %d\n", grp_alloc_blk);
1704	spin_unlock(sb_bgl_lock(sbi, group_no));
1705	jbd_unlock_bh_state(bitmap_bh);
1706#endif
1707
1708	if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
1709		ext3_error(sb, "ext3_new_block",
1710			    "block("E3FSBLK") >= blocks count(%d) - "
1711			    "block_group = %d, es == %p ", ret_block,
1712			le32_to_cpu(es->s_blocks_count), group_no, es);
1713		goto out;
1714	}
1715
1716	/*
1717	 * It is up to the caller to add the new buffer to a journal
1718	 * list of some description.  We don't know in advance whether
1719	 * the caller wants to use it as metadata or data.
1720	 */
1721	ext3_debug("allocating block %lu. Goal hits %d of %d.\n",
1722			ret_block, goal_hits, goal_attempts);
1723
1724	spin_lock(sb_bgl_lock(sbi, group_no));
1725	le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1726	spin_unlock(sb_bgl_lock(sbi, group_no));
1727	percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1728
1729	BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1730	err = ext3_journal_dirty_metadata(handle, gdp_bh);
1731	if (!fatal)
1732		fatal = err;
1733
1734	if (fatal)
1735		goto out;
1736
1737	*errp = 0;
1738	brelse(bitmap_bh);
1739
1740	if (num < *count) {
1741		dquot_free_block(inode, *count-num);
1742		*count = num;
1743	}
1744
1745	trace_ext3_allocate_blocks(inode, goal, num,
1746				   (unsigned long long)ret_block);
1747
1748	return ret_block;
1749
1750io_error:
1751	*errp = -EIO;
1752out:
1753	if (fatal) {
1754		*errp = fatal;
1755		ext3_std_error(sb, fatal);
1756	}
1757	/*
1758	 * Undo the block allocation
1759	 */
1760	if (!performed_allocation)
1761		dquot_free_block(inode, *count);
1762	brelse(bitmap_bh);
1763	return 0;
1764}
1765
1766ext3_fsblk_t ext3_new_block(handle_t *handle, struct inode *inode,
1767			ext3_fsblk_t goal, int *errp)
1768{
1769	unsigned long count = 1;
1770
1771	return ext3_new_blocks(handle, inode, goal, &count, errp);
1772}
1773
1774/**
1775 * ext3_count_free_blocks() -- count filesystem free blocks
1776 * @sb:		superblock
1777 *
1778 * Adds up the number of free blocks from each block group.
1779 */
1780ext3_fsblk_t ext3_count_free_blocks(struct super_block *sb)
1781{
1782	ext3_fsblk_t desc_count;
1783	struct ext3_group_desc *gdp;
1784	int i;
1785	unsigned long ngroups = EXT3_SB(sb)->s_groups_count;
1786#ifdef EXT3FS_DEBUG
1787	struct ext3_super_block *es;
1788	ext3_fsblk_t bitmap_count;
1789	unsigned long x;
1790	struct buffer_head *bitmap_bh = NULL;
1791
1792	es = EXT3_SB(sb)->s_es;
1793	desc_count = 0;
1794	bitmap_count = 0;
1795	gdp = NULL;
1796
1797	smp_rmb();
1798	for (i = 0; i < ngroups; i++) {
1799		gdp = ext3_get_group_desc(sb, i, NULL);
1800		if (!gdp)
1801			continue;
1802		desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1803		brelse(bitmap_bh);
1804		bitmap_bh = read_block_bitmap(sb, i);
1805		if (bitmap_bh == NULL)
1806			continue;
1807
1808		x = ext3_count_free(bitmap_bh, sb->s_blocksize);
1809		printk("group %d: stored = %d, counted = %lu\n",
1810			i, le16_to_cpu(gdp->bg_free_blocks_count), x);
1811		bitmap_count += x;
1812	}
1813	brelse(bitmap_bh);
1814	printk("ext3_count_free_blocks: stored = "E3FSBLK
1815		", computed = "E3FSBLK", "E3FSBLK"\n",
1816	       le32_to_cpu(es->s_free_blocks_count),
1817		desc_count, bitmap_count);
1818	return bitmap_count;
1819#else
1820	desc_count = 0;
1821	smp_rmb();
1822	for (i = 0; i < ngroups; i++) {
1823		gdp = ext3_get_group_desc(sb, i, NULL);
1824		if (!gdp)
1825			continue;
1826		desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1827	}
1828
1829	return desc_count;
1830#endif
1831}
1832
1833static inline int test_root(int a, int b)
1834{
1835	int num = b;
1836
1837	while (a > num)
1838		num *= b;
1839	return num == a;
1840}
1841
1842static int ext3_group_sparse(int group)
1843{
1844	if (group <= 1)
1845		return 1;
1846	if (!(group & 1))
1847		return 0;
1848	return (test_root(group, 7) || test_root(group, 5) ||
1849		test_root(group, 3));
1850}
1851
1852/**
1853 *	ext3_bg_has_super - number of blocks used by the superblock in group
1854 *	@sb: superblock for filesystem
1855 *	@group: group number to check
1856 *
1857 *	Return the number of blocks used by the superblock (primary or backup)
1858 *	in this group.  Currently this will be only 0 or 1.
1859 */
1860int ext3_bg_has_super(struct super_block *sb, int group)
1861{
1862	if (EXT3_HAS_RO_COMPAT_FEATURE(sb,
1863				EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
1864			!ext3_group_sparse(group))
1865		return 0;
1866	return 1;
1867}
1868
1869static unsigned long ext3_bg_num_gdb_meta(struct super_block *sb, int group)
1870{
1871	unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1872	unsigned long first = metagroup * EXT3_DESC_PER_BLOCK(sb);
1873	unsigned long last = first + EXT3_DESC_PER_BLOCK(sb) - 1;
1874
1875	if (group == first || group == first + 1 || group == last)
1876		return 1;
1877	return 0;
1878}
1879
1880static unsigned long ext3_bg_num_gdb_nometa(struct super_block *sb, int group)
1881{
1882	return ext3_bg_has_super(sb, group) ? EXT3_SB(sb)->s_gdb_count : 0;
1883}
1884
1885/**
1886 *	ext3_bg_num_gdb - number of blocks used by the group table in group
1887 *	@sb: superblock for filesystem
1888 *	@group: group number to check
1889 *
1890 *	Return the number of blocks used by the group descriptor table
1891 *	(primary or backup) in this group.  In the future there may be a
1892 *	different number of descriptor blocks in each group.
1893 */
1894unsigned long ext3_bg_num_gdb(struct super_block *sb, int group)
1895{
1896	unsigned long first_meta_bg =
1897			le32_to_cpu(EXT3_SB(sb)->s_es->s_first_meta_bg);
1898	unsigned long metagroup = group / EXT3_DESC_PER_BLOCK(sb);
1899
1900	if (!EXT3_HAS_INCOMPAT_FEATURE(sb,EXT3_FEATURE_INCOMPAT_META_BG) ||
1901			metagroup < first_meta_bg)
1902		return ext3_bg_num_gdb_nometa(sb,group);
1903
1904	return ext3_bg_num_gdb_meta(sb,group);
1905
1906}
1907
1908/**
1909 * ext3_trim_all_free -- function to trim all free space in alloc. group
1910 * @sb:			super block for file system
1911 * @group:		allocation group to trim
1912 * @start:		first group block to examine
1913 * @max:		last group block to examine
1914 * @gdp:		allocation group description structure
1915 * @minblocks:		minimum extent block count
1916 *
1917 * ext3_trim_all_free walks through group's block bitmap searching for free
1918 * blocks. When the free block is found, it tries to allocate this block and
1919 * consequent free block to get the biggest free extent possible, until it
1920 * reaches any used block. Then issue a TRIM command on this extent and free
1921 * the extent in the block bitmap. This is done until whole group is scanned.
1922 */
1923static ext3_grpblk_t ext3_trim_all_free(struct super_block *sb,
1924					unsigned int group,
1925					ext3_grpblk_t start, ext3_grpblk_t max,
1926					ext3_grpblk_t minblocks)
1927{
1928	handle_t *handle;
1929	ext3_grpblk_t next, free_blocks, bit, freed, count = 0;
1930	ext3_fsblk_t discard_block;
1931	struct ext3_sb_info *sbi;
1932	struct buffer_head *gdp_bh, *bitmap_bh = NULL;
1933	struct ext3_group_desc *gdp;
1934	int err = 0, ret = 0;
1935
1936	/*
1937	 * We will update one block bitmap, and one group descriptor
1938	 */
1939	handle = ext3_journal_start_sb(sb, 2);
1940	if (IS_ERR(handle))
1941		return PTR_ERR(handle);
1942
1943	bitmap_bh = read_block_bitmap(sb, group);
1944	if (!bitmap_bh) {
1945		err = -EIO;
1946		goto err_out;
1947	}
1948
1949	BUFFER_TRACE(bitmap_bh, "getting undo access");
1950	err = ext3_journal_get_undo_access(handle, bitmap_bh);
1951	if (err)
1952		goto err_out;
1953
1954	gdp = ext3_get_group_desc(sb, group, &gdp_bh);
1955	if (!gdp) {
1956		err = -EIO;
1957		goto err_out;
1958	}
1959
1960	BUFFER_TRACE(gdp_bh, "get_write_access");
1961	err = ext3_journal_get_write_access(handle, gdp_bh);
1962	if (err)
1963		goto err_out;
1964
1965	free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
1966	sbi = EXT3_SB(sb);
1967
1968	 /* Walk through the whole group */
1969	while (start <= max) {
1970		start = bitmap_search_next_usable_block(start, bitmap_bh, max);
1971		if (start < 0)
1972			break;
1973		next = start;
1974
1975		/*
1976		 * Allocate contiguous free extents by setting bits in the
1977		 * block bitmap
1978		 */
1979		while (next <= max
1980			&& claim_block(sb_bgl_lock(sbi, group),
1981					next, bitmap_bh)) {
1982			next++;
1983		}
1984
1985		 /* We did not claim any blocks */
1986		if (next == start)
1987			continue;
1988
1989		discard_block = (ext3_fsblk_t)start +
1990				ext3_group_first_block_no(sb, group);
1991
1992		/* Update counters */
1993		spin_lock(sb_bgl_lock(sbi, group));
1994		le16_add_cpu(&gdp->bg_free_blocks_count, start - next);
1995		spin_unlock(sb_bgl_lock(sbi, group));
1996		percpu_counter_sub(&sbi->s_freeblocks_counter, next - start);
1997
1998		free_blocks -= next - start;
1999		/* Do not issue a TRIM on extents smaller than minblocks */
2000		if ((next - start) < minblocks)
2001			goto free_extent;
2002
2003		trace_ext3_discard_blocks(sb, discard_block, next - start);
2004		 /* Send the TRIM command down to the device */
2005		err = sb_issue_discard(sb, discard_block, next - start,
2006				       GFP_NOFS, 0);
2007		count += (next - start);
2008free_extent:
2009		freed = 0;
2010
2011		/*
2012		 * Clear bits in the bitmap
2013		 */
2014		for (bit = start; bit < next; bit++) {
2015			BUFFER_TRACE(bitmap_bh, "clear bit");
2016			if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, group),
2017						bit, bitmap_bh->b_data)) {
2018				ext3_error(sb, __func__,
2019					"bit already cleared for block "E3FSBLK,
2020					 (unsigned long)bit);
2021				BUFFER_TRACE(bitmap_bh, "bit already cleared");
2022			} else {
2023				freed++;
2024			}
2025		}
2026
2027		/* Update couters */
2028		spin_lock(sb_bgl_lock(sbi, group));
2029		le16_add_cpu(&gdp->bg_free_blocks_count, freed);
2030		spin_unlock(sb_bgl_lock(sbi, group));
2031		percpu_counter_add(&sbi->s_freeblocks_counter, freed);
2032
2033		start = next;
2034		if (err < 0) {
2035			if (err != -EOPNOTSUPP)
2036				ext3_warning(sb, __func__, "Discard command "
2037					     "returned error %d\n", err);
2038			break;
2039		}
2040
2041		if (fatal_signal_pending(current)) {
2042			err = -ERESTARTSYS;
2043			break;
2044		}
2045
2046		cond_resched();
2047
2048		/* No more suitable extents */
2049		if (free_blocks < minblocks)
2050			break;
2051	}
2052
2053	/* We dirtied the bitmap block */
2054	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
2055	ret = ext3_journal_dirty_metadata(handle, bitmap_bh);
2056	if (!err)
2057		err = ret;
2058
2059	/* And the group descriptor block */
2060	BUFFER_TRACE(gdp_bh, "dirtied group descriptor block");
2061	ret = ext3_journal_dirty_metadata(handle, gdp_bh);
2062	if (!err)
2063		err = ret;
2064
2065	ext3_debug("trimmed %d blocks in the group %d\n",
2066		count, group);
2067
2068err_out:
2069	if (err)
2070		count = err;
2071	ext3_journal_stop(handle);
2072	brelse(bitmap_bh);
2073
2074	return count;
2075}
2076
2077/**
2078 * ext3_trim_fs() -- trim ioctl handle function
2079 * @sb:			superblock for filesystem
2080 * @start:		First Byte to trim
2081 * @len:		number of Bytes to trim from start
2082 * @minlen:		minimum extent length in Bytes
2083 *
2084 * ext3_trim_fs goes through all allocation groups containing Bytes from
2085 * start to start+len. For each such a group ext3_trim_all_free function
2086 * is invoked to trim all free space.
2087 */
2088int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range)
2089{
2090	ext3_grpblk_t last_block, first_block;
2091	unsigned long group, first_group, last_group;
 
2092	struct ext3_group_desc *gdp;
2093	struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2094	uint64_t start, minlen, end, trimmed = 0;
2095	ext3_fsblk_t first_data_blk =
2096			le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block);
2097	ext3_fsblk_t max_blks = le32_to_cpu(es->s_blocks_count);
2098	int ret = 0;
2099
2100	start = range->start >> sb->s_blocksize_bits;
2101	end = start + (range->len >> sb->s_blocksize_bits) - 1;
 
2102	minlen = range->minlen >> sb->s_blocksize_bits;
 
2103
2104	if (unlikely(minlen > EXT3_BLOCKS_PER_GROUP(sb)) ||
2105	    unlikely(start >= max_blks))
2106		return -EINVAL;
2107	if (end >= max_blks)
2108		end = max_blks - 1;
2109	if (end <= first_data_blk)
2110		goto out;
2111	if (start < first_data_blk)
2112		start = first_data_blk;
2113
 
2114	smp_rmb();
2115
2116	/* Determine first and last group to examine based on start and len */
2117	ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) start,
2118				     &first_group, &first_block);
2119	ext3_get_group_no_and_offset(sb, (ext3_fsblk_t) end,
2120				     &last_group, &last_block);
 
 
2121
2122	/* end now represents the last block to discard in this group */
2123	end = EXT3_BLOCKS_PER_GROUP(sb) - 1;
2124
2125	for (group = first_group; group <= last_group; group++) {
2126		gdp = ext3_get_group_desc(sb, group, NULL);
2127		if (!gdp)
2128			break;
2129
 
 
 
 
2130		/*
2131		 * For all the groups except the last one, last block will
2132		 * always be EXT3_BLOCKS_PER_GROUP(sb)-1, so we only need to
2133		 * change it for the last group, note that last_block is
2134		 * already computed earlier by ext3_get_group_no_and_offset()
2135		 */
2136		if (group == last_group)
2137			end = last_block;
2138
2139		if (le16_to_cpu(gdp->bg_free_blocks_count) >= minlen) {
2140			ret = ext3_trim_all_free(sb, group, first_block,
2141						 end, minlen);
2142			if (ret < 0)
2143				break;
2144			trimmed += ret;
2145		}
2146
2147		/*
2148		 * For every group except the first one, we are sure
2149		 * that the first block to discard will be block #0.
2150		 */
2151		first_block = 0;
2152	}
2153
2154	if (ret > 0)
2155		ret = 0;
2156
2157out:
2158	range->len = trimmed * sb->s_blocksize;
 
2159	return ret;
2160}