mballoc.c 131.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/*
 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
 * Written by Alex Tomas <alex@clusterfs.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public Licens
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
 */


/*
 * mballoc.c contains the multiblocks allocation routines
 */

24
#include "mballoc.h"
25
#include <linux/debugfs.h>
26 27
#include <trace/events/ext4.h>

28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
/*
 * MUSTDO:
 *   - test ext4_ext_search_left() and ext4_ext_search_right()
 *   - search for metadata in few groups
 *
 * TODO v4:
 *   - normalization should take into account whether file is still open
 *   - discard preallocations if no free space left (policy?)
 *   - don't normalize tails
 *   - quota
 *   - reservation for superuser
 *
 * TODO v3:
 *   - bitmap read-ahead (proposed by Oleg Drokin aka green)
 *   - track min/max extents in each group for better group selection
 *   - mb_mark_used() may allocate chunk right after splitting buddy
 *   - tree of groups sorted by number of free blocks
 *   - error handling
 */

/*
 * The allocation request involve request for multiple number of blocks
 * near to the goal(block) value specified.
 *
T
Theodore Ts'o 已提交
52 53 54 55 56 57 58 59 60
 * During initialization phase of the allocator we decide to use the
 * group preallocation or inode preallocation depending on the size of
 * the file. The size of the file could be the resulting file size we
 * would have after allocation, or the current file size, which ever
 * is larger. If the size is less than sbi->s_mb_stream_request we
 * select to use the group preallocation. The default value of
 * s_mb_stream_request is 16 blocks. This can also be tuned via
 * /sys/fs/ext4/<partition>/mb_stream_req. The value is represented in
 * terms of number of blocks.
61 62
 *
 * The main motivation for having small file use group preallocation is to
T
Theodore Ts'o 已提交
63
 * ensure that we have small files closer together on the disk.
64
 *
T
Theodore Ts'o 已提交
65 66 67 68
 * First stage the allocator looks at the inode prealloc list,
 * ext4_inode_info->i_prealloc_list, which contains list of prealloc
 * spaces for this particular inode. The inode prealloc space is
 * represented as:
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106
 *
 * pa_lstart -> the logical start block for this prealloc space
 * pa_pstart -> the physical start block for this prealloc space
 * pa_len    -> lenght for this prealloc space
 * pa_free   ->  free space available in this prealloc space
 *
 * The inode preallocation space is used looking at the _logical_ start
 * block. If only the logical file block falls within the range of prealloc
 * space we will consume the particular prealloc space. This make sure that
 * that the we have contiguous physical blocks representing the file blocks
 *
 * The important thing to be noted in case of inode prealloc space is that
 * we don't modify the values associated to inode prealloc space except
 * pa_free.
 *
 * If we are not able to find blocks in the inode prealloc space and if we
 * have the group allocation flag set then we look at the locality group
 * prealloc space. These are per CPU prealloc list repreasented as
 *
 * ext4_sb_info.s_locality_groups[smp_processor_id()]
 *
 * The reason for having a per cpu locality group is to reduce the contention
 * between CPUs. It is possible to get scheduled at this point.
 *
 * The locality group prealloc space is used looking at whether we have
 * enough free space (pa_free) withing the prealloc space.
 *
 * If we can't allocate blocks via inode prealloc or/and locality group
 * prealloc then we look at the buddy cache. The buddy cache is represented
 * by ext4_sb_info.s_buddy_cache (struct inode) whose file offset gets
 * mapped to the buddy and bitmap information regarding different
 * groups. The buddy information is attached to buddy cache inode so that
 * we can access them through the page cache. The information regarding
 * each group is loaded via ext4_mb_load_buddy.  The information involve
 * block bitmap and buddy information. The information are stored in the
 * inode as:
 *
 *  {                        page                        }
107
 *  [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
 *
 *
 * one block each for bitmap and buddy information.  So for each group we
 * take up 2 blocks. A page can contain blocks_per_page (PAGE_CACHE_SIZE /
 * blocksize) blocks.  So it can have information regarding groups_per_page
 * which is blocks_per_page/2
 *
 * The buddy cache inode is not stored on disk. The inode is thrown
 * away when the filesystem is unmounted.
 *
 * We look for count number of blocks in the buddy cache. If we were able
 * to locate that many free blocks we return with additional information
 * regarding rest of the contiguous physical block available
 *
 * Before allocating blocks via buddy cache we normalize the request
 * blocks. This ensure we ask for more blocks that we needed. The extra
 * blocks that we get after allocation is added to the respective prealloc
 * list. In case of inode preallocation we follow a list of heuristics
 * based on file size. This can be found in ext4_mb_normalize_request. If
 * we are doing a group prealloc we try to normalize the request to
T
Theodore Ts'o 已提交
128
 * sbi->s_mb_group_prealloc. Default value of s_mb_group_prealloc is
129
 * 512 blocks. This can be tuned via
T
Theodore Ts'o 已提交
130
 * /sys/fs/ext4/<partition/mb_group_prealloc. The value is represented in
131 132 133 134
 * terms of number of blocks. If we have mounted the file system with -O
 * stripe=<value> option the group prealloc request is normalized to the
 * stripe value (sbi->s_stripe)
 *
T
Theodore Ts'o 已提交
135
 * The regular allocator(using the buddy cache) supports few tunables.
136
 *
T
Theodore Ts'o 已提交
137 138 139
 * /sys/fs/ext4/<partition>/mb_min_to_scan
 * /sys/fs/ext4/<partition>/mb_max_to_scan
 * /sys/fs/ext4/<partition>/mb_order2_req
140
 *
T
Theodore Ts'o 已提交
141
 * The regular allocator uses buddy scan only if the request len is power of
142 143
 * 2 blocks and the order of allocation is >= sbi->s_mb_order2_reqs. The
 * value of s_mb_order2_reqs can be tuned via
T
Theodore Ts'o 已提交
144
 * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
145
 * stripe size (sbi->s_stripe), we try to search for contigous block in
T
Theodore Ts'o 已提交
146 147 148
 * stripe size. This should result in better allocation on RAID setups. If
 * not, we search in the specific group using bitmap for best extents. The
 * tunable min_to_scan and max_to_scan control the behaviour here.
149
 * min_to_scan indicate how long the mballoc __must__ look for a best
T
Theodore Ts'o 已提交
150
 * extent and max_to_scan indicates how long the mballoc __can__ look for a
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
 * best extent in the found extents. Searching for the blocks starts with
 * the group specified as the goal value in allocation context via
 * ac_g_ex. Each group is first checked based on the criteria whether it
 * can used for allocation. ext4_mb_good_group explains how the groups are
 * checked.
 *
 * Both the prealloc space are getting populated as above. So for the first
 * request we will hit the buddy cache which will result in this prealloc
 * space getting filled. The prealloc space is then later used for the
 * subsequent request.
 */

/*
 * mballoc operates on the following data:
 *  - on-disk bitmap
 *  - in-core buddy (actually includes buddy and bitmap)
 *  - preallocation descriptors (PAs)
 *
 * there are two types of preallocations:
 *  - inode
 *    assiged to specific inode and can be used for this inode only.
 *    it describes part of inode's space preallocated to specific
 *    physical blocks. any block from that preallocated can be used
 *    independent. the descriptor just tracks number of blocks left
 *    unused. so, before taking some block from descriptor, one must
 *    make sure corresponded logical block isn't allocated yet. this
 *    also means that freeing any block within descriptor's range
 *    must discard all preallocated blocks.
 *  - locality group
 *    assigned to specific locality group which does not translate to
 *    permanent set of inodes: inode can join and leave group. space
 *    from this type of preallocation can be used for any inode. thus
 *    it's consumed from the beginning to the end.
 *
 * relation between them can be expressed as:
 *    in-core buddy = on-disk bitmap + preallocation descriptors
 *
 * this mean blocks mballoc considers used are:
 *  - allocated blocks (persistent)
 *  - preallocated blocks (non-persistent)
 *
 * consistency in mballoc world means that at any time a block is either
 * free or used in ALL structures. notice: "any time" should not be read
 * literally -- time is discrete and delimited by locks.
 *
 *  to keep it simple, we don't use block numbers, instead we count number of
 *  blocks: how many blocks marked used/free in on-disk bitmap, buddy and PA.
 *
 * all operations can be expressed as:
 *  - init buddy:			buddy = on-disk + PAs
 *  - new PA:				buddy += N; PA = N
 *  - use inode PA:			on-disk += N; PA -= N
 *  - discard inode PA			buddy -= on-disk - PA; PA = 0
 *  - use locality group PA		on-disk += N; PA -= N
 *  - discard locality group PA		buddy -= PA; PA = 0
 *  note: 'buddy -= on-disk - PA' is used to show that on-disk bitmap
 *        is used in real operation because we can't know actual used
 *        bits from PA, only from on-disk bitmap
 *
 * if we follow this strict logic, then all operations above should be atomic.
 * given some of them can block, we'd have to use something like semaphores
 * killing performance on high-end SMP hardware. let's try to relax it using
 * the following knowledge:
 *  1) if buddy is referenced, it's already initialized
 *  2) while block is used in buddy and the buddy is referenced,
 *     nobody can re-allocate that block
 *  3) we work on bitmaps and '+' actually means 'set bits'. if on-disk has
 *     bit set and PA claims same block, it's OK. IOW, one can set bit in
 *     on-disk bitmap if buddy has same bit set or/and PA covers corresponded
 *     block
 *
 * so, now we're building a concurrency table:
 *  - init buddy vs.
 *    - new PA
 *      blocks for PA are allocated in the buddy, buddy must be referenced
 *      until PA is linked to allocation group to avoid concurrent buddy init
 *    - use inode PA
 *      we need to make sure that either on-disk bitmap or PA has uptodate data
 *      given (3) we care that PA-=N operation doesn't interfere with init
 *    - discard inode PA
 *      the simplest way would be to have buddy initialized by the discard
 *    - use locality group PA
 *      again PA-=N must be serialized with init
 *    - discard locality group PA
 *      the simplest way would be to have buddy initialized by the discard
 *  - new PA vs.
 *    - use inode PA
 *      i_data_sem serializes them
 *    - discard inode PA
 *      discard process must wait until PA isn't used by another process
 *    - use locality group PA
 *      some mutex should serialize them
 *    - discard locality group PA
 *      discard process must wait until PA isn't used by another process
 *  - use inode PA
 *    - use inode PA
 *      i_data_sem or another mutex should serializes them
 *    - discard inode PA
 *      discard process must wait until PA isn't used by another process
 *    - use locality group PA
 *      nothing wrong here -- they're different PAs covering different blocks
 *    - discard locality group PA
 *      discard process must wait until PA isn't used by another process
 *
 * now we're ready to make few consequences:
 *  - PA is referenced and while it is no discard is possible
 *  - PA is referenced until block isn't marked in on-disk bitmap
 *  - PA changes only after on-disk bitmap
 *  - discard must not compete with init. either init is done before
 *    any discard or they're serialized somehow
 *  - buddy init as sum of on-disk bitmap and PAs is done atomically
 *
 * a special case when we've used PA to emptiness. no need to modify buddy
 * in this case, but we should care about concurrent init
 *
 */

 /*
 * Logic in few words:
 *
 *  - allocation:
 *    load group
 *    find blocks
 *    mark bits in on-disk bitmap
 *    release group
 *
 *  - use preallocation:
 *    find proper PA (per-inode or group)
 *    load group
 *    mark bits in on-disk bitmap
 *    release group
 *    release PA
 *
 *  - free:
 *    load group
 *    mark bits in on-disk bitmap
 *    release group
 *
 *  - discard preallocations in group:
 *    mark PAs deleted
 *    move them onto local list
 *    load on-disk bitmap
 *    load group
 *    remove PA from object (inode or locality group)
 *    mark free blocks in-core
 *
 *  - discard inode's preallocations:
 */

/*
 * Locking rules
 *
 * Locks:
 *  - bitlock on a group	(group)
 *  - object (inode/locality)	(object)
 *  - per-pa lock		(pa)
 *
 * Paths:
 *  - new pa
 *    object
 *    group
 *
 *  - find and use pa:
 *    pa
 *
 *  - release consumed pa:
 *    pa
 *    group
 *    object
 *
 *  - generate in-core bitmap:
 *    group
 *        pa
 *
 *  - discard all for given object (inode, locality group):
 *    object
 *        pa
 *    group
 *
 *  - discard all for given group:
 *    group
 *        pa
 *    group
 *        object
 *
 */
337 338 339 340 341
static struct kmem_cache *ext4_pspace_cachep;
static struct kmem_cache *ext4_ac_cachep;
static struct kmem_cache *ext4_free_ext_cachep;
static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
					ext4_group_t group);
342 343
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group);
344 345
static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);

346 347
static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
{
348
#if BITS_PER_LONG == 64
349 350
	*bit += ((unsigned long) addr & 7UL) << 3;
	addr = (void *) ((unsigned long) addr & ~7UL);
351
#elif BITS_PER_LONG == 32
352 353
	*bit += ((unsigned long) addr & 3UL) << 3;
	addr = (void *) ((unsigned long) addr & ~3UL);
354 355 356
#else
#error "how many bits you are?!"
#endif
357 358
	return addr;
}
359 360 361 362 363 364 365

static inline int mb_test_bit(int bit, void *addr)
{
	/*
	 * ext4_test_bit on architecture like powerpc
	 * needs unsigned long aligned address
	 */
366
	addr = mb_correct_addr_and_bit(&bit, addr);
367 368 369 370 371
	return ext4_test_bit(bit, addr);
}

static inline void mb_set_bit(int bit, void *addr)
{
372
	addr = mb_correct_addr_and_bit(&bit, addr);
373 374 375 376 377
	ext4_set_bit(bit, addr);
}

static inline void mb_clear_bit(int bit, void *addr)
{
378
	addr = mb_correct_addr_and_bit(&bit, addr);
379 380 381
	ext4_clear_bit(bit, addr);
}

382 383
static inline int mb_find_next_zero_bit(void *addr, int max, int start)
{
384
	int fix = 0, ret, tmpmax;
385
	addr = mb_correct_addr_and_bit(&fix, addr);
386
	tmpmax = max + fix;
387 388
	start += fix;

389 390 391 392
	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
393 394 395 396
}

static inline int mb_find_next_bit(void *addr, int max, int start)
{
397
	int fix = 0, ret, tmpmax;
398
	addr = mb_correct_addr_and_bit(&fix, addr);
399
	tmpmax = max + fix;
400 401
	start += fix;

402 403 404 405
	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
406 407
}

408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
{
	char *bb;

	BUG_ON(EXT4_MB_BITMAP(e4b) == EXT4_MB_BUDDY(e4b));
	BUG_ON(max == NULL);

	if (order > e4b->bd_blkbits + 1) {
		*max = 0;
		return NULL;
	}

	/* at order 0 we see each particular block */
	*max = 1 << (e4b->bd_blkbits + 3);
	if (order == 0)
		return EXT4_MB_BITMAP(e4b);

	bb = EXT4_MB_BUDDY(e4b) + EXT4_SB(e4b->bd_sb)->s_mb_offsets[order];
	*max = EXT4_SB(e4b->bd_sb)->s_mb_maxs[order];

	return bb;
}

#ifdef DOUBLE_CHECK
static void mb_free_blocks_double(struct inode *inode, struct ext4_buddy *e4b,
			   int first, int count)
{
	int i;
	struct super_block *sb = e4b->bd_sb;

	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
		return;
440
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
441 442 443 444 445 446 447
	for (i = 0; i < count; i++) {
		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
			ext4_fsblk_t blocknr;
			blocknr = e4b->bd_group * EXT4_BLOCKS_PER_GROUP(sb);
			blocknr += first + i;
			blocknr +=
			    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
448 449
			ext4_grp_locked_error(sb, e4b->bd_group,
				   __func__, "double-free of inode"
450
				   " %lu's block %llu(bit %u in group %u)",
451 452 453 454 455 456 457 458 459 460 461 462 463
				   inode ? inode->i_ino : 0, blocknr,
				   first + i, e4b->bd_group);
		}
		mb_clear_bit(first + i, e4b->bd_info->bb_bitmap);
	}
}

static void mb_mark_used_double(struct ext4_buddy *e4b, int first, int count)
{
	int i;

	if (unlikely(e4b->bd_info->bb_bitmap == NULL))
		return;
464
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
	for (i = 0; i < count; i++) {
		BUG_ON(mb_test_bit(first + i, e4b->bd_info->bb_bitmap));
		mb_set_bit(first + i, e4b->bd_info->bb_bitmap);
	}
}

static void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
{
	if (memcmp(e4b->bd_info->bb_bitmap, bitmap, e4b->bd_sb->s_blocksize)) {
		unsigned char *b1, *b2;
		int i;
		b1 = (unsigned char *) e4b->bd_info->bb_bitmap;
		b2 = (unsigned char *) bitmap;
		for (i = 0; i < e4b->bd_sb->s_blocksize; i++) {
			if (b1[i] != b2[i]) {
480
				printk(KERN_ERR "corruption in group %u "
481 482 483
				       "at byte %u(%u): %x in copy != %x "
				       "on disk/prealloc\n",
				       e4b->bd_group, i, i * 8, b1[i], b2[i]);
484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
				BUG();
			}
		}
	}
}

#else
static inline void mb_free_blocks_double(struct inode *inode,
				struct ext4_buddy *e4b, int first, int count)
{
	return;
}
static inline void mb_mark_used_double(struct ext4_buddy *e4b,
						int first, int count)
{
	return;
}
static inline void mb_cmp_bitmaps(struct ext4_buddy *e4b, void *bitmap)
{
	return;
}
#endif

#ifdef AGGRESSIVE_CHECK

#define MB_CHECK_ASSERT(assert)						\
do {									\
	if (!(assert)) {						\
		printk(KERN_EMERG					\
			"Assertion failure in %s() at %s:%d: \"%s\"\n",	\
			function, file, line, # assert);		\
		BUG();							\
	}								\
} while (0)

static int __mb_check_buddy(struct ext4_buddy *e4b, char *file,
				const char *function, int line)
{
	struct super_block *sb = e4b->bd_sb;
	int order = e4b->bd_blkbits + 1;
	int max;
	int max2;
	int i;
	int j;
	int k;
	int count;
	struct ext4_group_info *grp;
	int fragments = 0;
	int fstart;
	struct list_head *cur;
	void *buddy;
	void *buddy2;

	{
		static int mb_check_counter;
		if (mb_check_counter++ % 100 != 0)
			return 0;
	}

	while (order > 1) {
		buddy = mb_find_buddy(e4b, order, &max);
		MB_CHECK_ASSERT(buddy);
		buddy2 = mb_find_buddy(e4b, order - 1, &max2);
		MB_CHECK_ASSERT(buddy2);
		MB_CHECK_ASSERT(buddy != buddy2);
		MB_CHECK_ASSERT(max * 2 == max2);

		count = 0;
		for (i = 0; i < max; i++) {

			if (mb_test_bit(i, buddy)) {
				/* only single bit in buddy2 may be 1 */
				if (!mb_test_bit(i << 1, buddy2)) {
					MB_CHECK_ASSERT(
						mb_test_bit((i<<1)+1, buddy2));
				} else if (!mb_test_bit((i << 1) + 1, buddy2)) {
					MB_CHECK_ASSERT(
						mb_test_bit(i << 1, buddy2));
				}
				continue;
			}

			/* both bits in buddy2 must be 0 */
			MB_CHECK_ASSERT(mb_test_bit(i << 1, buddy2));
			MB_CHECK_ASSERT(mb_test_bit((i << 1) + 1, buddy2));

			for (j = 0; j < (1 << order); j++) {
				k = (i * (1 << order)) + j;
				MB_CHECK_ASSERT(
					!mb_test_bit(k, EXT4_MB_BITMAP(e4b)));
			}
			count++;
		}
		MB_CHECK_ASSERT(e4b->bd_info->bb_counters[order] == count);
		order--;
	}

	fstart = -1;
	buddy = mb_find_buddy(e4b, 0, &max);
	for (i = 0; i < max; i++) {
		if (!mb_test_bit(i, buddy)) {
			MB_CHECK_ASSERT(i >= e4b->bd_info->bb_first_free);
			if (fstart == -1) {
				fragments++;
				fstart = i;
			}
			continue;
		}
		fstart = -1;
		/* check used bits only */
		for (j = 0; j < e4b->bd_blkbits + 1; j++) {
			buddy2 = mb_find_buddy(e4b, j, &max2);
			k = i >> j;
			MB_CHECK_ASSERT(k < max2);
			MB_CHECK_ASSERT(mb_test_bit(k, buddy2));
		}
	}
	MB_CHECK_ASSERT(!EXT4_MB_GRP_NEED_INIT(e4b->bd_info));
	MB_CHECK_ASSERT(e4b->bd_info->bb_fragments == fragments);

	grp = ext4_get_group_info(sb, e4b->bd_group);
	buddy = mb_find_buddy(e4b, 0, &max);
	list_for_each(cur, &grp->bb_prealloc_list) {
		ext4_group_t groupnr;
		struct ext4_prealloc_space *pa;
609 610
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
611
		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
612
		for (i = 0; i < pa->pa_len; i++)
613 614 615 616 617 618
			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
	}
	return 0;
}
#undef MB_CHECK_ASSERT
#define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
619
					__FILE__, __func__, __LINE__)
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
#else
#define mb_check_buddy(e4b)
#endif

/* FIXME!! need more doc */
static void ext4_mb_mark_free_simple(struct super_block *sb,
				void *buddy, unsigned first, int len,
					struct ext4_group_info *grp)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	unsigned short min;
	unsigned short max;
	unsigned short chunk;
	unsigned short border;

635
	BUG_ON(len > EXT4_BLOCKS_PER_GROUP(sb));
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660

	border = 2 << sb->s_blocksize_bits;

	while (len > 0) {
		/* find how many blocks can be covered since this position */
		max = ffs(first | border) - 1;

		/* find how many blocks of power 2 we need to mark */
		min = fls(len) - 1;

		if (max < min)
			min = max;
		chunk = 1 << min;

		/* mark multiblock chunks only */
		grp->bb_counters[min]++;
		if (min > 0)
			mb_clear_bit(first >> min,
				     buddy + sbi->s_mb_offsets[min]);

		len -= chunk;
		first += chunk;
	}
}

661 662
static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
663 664 665 666 667 668 669 670 671 672 673 674 675
				void *buddy, void *bitmap, ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
	unsigned short max = EXT4_BLOCKS_PER_GROUP(sb);
	unsigned short i = 0;
	unsigned short first;
	unsigned short len;
	unsigned free = 0;
	unsigned fragments = 0;
	unsigned long long period = get_cycles();

	/* initialize buddy from bitmap which is aggregation
	 * of on-disk bitmap and preallocations */
676
	i = mb_find_next_zero_bit(bitmap, max, 0);
677 678 679 680
	grp->bb_first_free = i;
	while (i < max) {
		fragments++;
		first = i;
681
		i = mb_find_next_bit(bitmap, max, i);
682 683 684 685 686 687 688
		len = i - first;
		free += len;
		if (len > 1)
			ext4_mb_mark_free_simple(sb, buddy, first, len, grp);
		else
			grp->bb_counters[0]++;
		if (i < max)
689
			i = mb_find_next_zero_bit(bitmap, max, i);
690 691 692 693
	}
	grp->bb_fragments = fragments;

	if (free != grp->bb_free) {
694
		ext4_grp_locked_error(sb, group,  __func__,
695
			"EXT4-fs: group %u: %u blocks in bitmap, %u in gd",
696
			group, free, grp->bb_free);
697 698 699 700
		/*
		 * If we intent to continue, we consider group descritor
		 * corrupt and update bb_free using bitmap value
		 */
701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719
		grp->bb_free = free;
	}

	clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));

	period = get_cycles() - period;
	spin_lock(&EXT4_SB(sb)->s_bal_lock);
	EXT4_SB(sb)->s_mb_buddies_generated++;
	EXT4_SB(sb)->s_mb_generation_time += period;
	spin_unlock(&EXT4_SB(sb)->s_bal_lock);
}

/* The buddy information is attached the buddy cache inode
 * for convenience. The information regarding each group
 * is loaded via ext4_mb_load_buddy. The information involve
 * block bitmap and buddy information. The information are
 * stored in the inode as
 *
 * {                        page                        }
720
 * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
721 722 723 724 725 726 727 728 729 730 731
 *
 *
 * one block each for bitmap and buddy information.
 * So for each group we take up 2 blocks. A page can
 * contain blocks_per_page (PAGE_CACHE_SIZE / blocksize)  blocks.
 * So it can have information regarding groups_per_page which
 * is blocks_per_page/2
 */

static int ext4_mb_init_cache(struct page *page, char *incore)
{
732
	ext4_group_t ngroups;
733 734 735 736 737 738 739 740 741 742 743 744 745 746
	int blocksize;
	int blocks_per_page;
	int groups_per_page;
	int err = 0;
	int i;
	ext4_group_t first_group;
	int first_block;
	struct super_block *sb;
	struct buffer_head *bhs;
	struct buffer_head **bh;
	struct inode *inode;
	char *data;
	char *bitmap;

747
	mb_debug(1, "init page %lu\n", page->index);
748 749 750

	inode = page->mapping->host;
	sb = inode->i_sb;
751
	ngroups = ext4_get_groups_count(sb);
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774
	blocksize = 1 << inode->i_blkbits;
	blocks_per_page = PAGE_CACHE_SIZE / blocksize;

	groups_per_page = blocks_per_page >> 1;
	if (groups_per_page == 0)
		groups_per_page = 1;

	/* allocate buffer_heads to read bitmaps */
	if (groups_per_page > 1) {
		err = -ENOMEM;
		i = sizeof(struct buffer_head *) * groups_per_page;
		bh = kzalloc(i, GFP_NOFS);
		if (bh == NULL)
			goto out;
	} else
		bh = &bhs;

	first_group = page->index * blocks_per_page / 2;

	/* read all groups the page covers into the cache */
	for (i = 0; i < groups_per_page; i++) {
		struct ext4_group_desc *desc;

775
		if (first_group + i >= ngroups)
776 777 778 779 780 781 782 783 784 785 786 787
			break;

		err = -EIO;
		desc = ext4_get_group_desc(sb, first_group + i, NULL);
		if (desc == NULL)
			goto out;

		err = -ENOMEM;
		bh[i] = sb_getblk(sb, ext4_block_bitmap(sb, desc));
		if (bh[i] == NULL)
			goto out;

788
		if (bitmap_uptodate(bh[i]))
789 790
			continue;

791
		lock_buffer(bh[i]);
792 793 794 795
		if (bitmap_uptodate(bh[i])) {
			unlock_buffer(bh[i]);
			continue;
		}
796
		ext4_lock_group(sb, first_group + i);
797 798 799
		if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
			ext4_init_block_bitmap(sb, bh[i],
						first_group + i, desc);
800
			set_bitmap_uptodate(bh[i]);
801
			set_buffer_uptodate(bh[i]);
802
			ext4_unlock_group(sb, first_group + i);
A
Aneesh Kumar K.V 已提交
803
			unlock_buffer(bh[i]);
804 805
			continue;
		}
806
		ext4_unlock_group(sb, first_group + i);
807 808 809 810 811 812 813 814 815
		if (buffer_uptodate(bh[i])) {
			/*
			 * if not uninit if bh is uptodate,
			 * bitmap is also uptodate
			 */
			set_bitmap_uptodate(bh[i]);
			unlock_buffer(bh[i]);
			continue;
		}
816
		get_bh(bh[i]);
817 818 819 820 821 822 823
		/*
		 * submit the buffer_head for read. We can
		 * safely mark the bitmap as uptodate now.
		 * We do it here so the bitmap uptodate bit
		 * get set with buffer lock held.
		 */
		set_bitmap_uptodate(bh[i]);
824 825
		bh[i]->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh[i]);
826
		mb_debug(1, "read bitmap for group %u\n", first_group + i);
827 828 829 830 831 832 833 834 835 836 837
	}

	/* wait for I/O completion */
	for (i = 0; i < groups_per_page && bh[i]; i++)
		wait_on_buffer(bh[i]);

	err = -EIO;
	for (i = 0; i < groups_per_page && bh[i]; i++)
		if (!buffer_uptodate(bh[i]))
			goto out;

838
	err = 0;
839
	first_block = page->index * blocks_per_page;
840 841
	/* init the page  */
	memset(page_address(page), 0xff, PAGE_CACHE_SIZE);
842 843 844 845 846
	for (i = 0; i < blocks_per_page; i++) {
		int group;
		struct ext4_group_info *grinfo;

		group = (first_block + i) >> 1;
847
		if (group >= ngroups)
848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
			break;

		/*
		 * data carry information regarding this
		 * particular group in the format specified
		 * above
		 *
		 */
		data = page_address(page) + (i * blocksize);
		bitmap = bh[group - first_group]->b_data;

		/*
		 * We place the buddy block and bitmap block
		 * close together
		 */
		if ((first_block + i) & 1) {
			/* this is block of buddy */
			BUG_ON(incore == NULL);
866
			mb_debug(1, "put buddy for group %u in page %lu/%x\n",
867 868 869 870
				group, page->index, i * blocksize);
			grinfo = ext4_get_group_info(sb, group);
			grinfo->bb_fragments = 0;
			memset(grinfo->bb_counters, 0,
871 872
			       sizeof(*grinfo->bb_counters) *
				(sb->s_blocksize_bits+2));
873 874 875
			/*
			 * incore got set to the group block bitmap below
			 */
876
			ext4_lock_group(sb, group);
877
			ext4_mb_generate_buddy(sb, data, incore, group);
878
			ext4_unlock_group(sb, group);
879 880 881 882
			incore = NULL;
		} else {
			/* this is block of bitmap */
			BUG_ON(incore != NULL);
883
			mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
884 885 886 887 888 889 890 891
				group, page->index, i * blocksize);

			/* see comments in ext4_mb_put_pa() */
			ext4_lock_group(sb, group);
			memcpy(data, bitmap, blocksize);

			/* mark all preallocated blks used in in-core bitmap */
			ext4_mb_generate_from_pa(sb, data, group);
892
			ext4_mb_generate_from_freelist(sb, data, group);
893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
			ext4_unlock_group(sb, group);

			/* set incore so that the buddy information can be
			 * generated using this
			 */
			incore = data;
		}
	}
	SetPageUptodate(page);

out:
	if (bh) {
		for (i = 0; i < groups_per_page && bh[i]; i++)
			brelse(bh[i]);
		if (bh != &bhs)
			kfree(bh);
	}
	return err;
}

913 914 915
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
					struct ext4_buddy *e4b)
916 917 918 919 920 921
{
	int blocks_per_page;
	int block;
	int pnum;
	int poff;
	struct page *page;
922
	int ret;
923 924 925
	struct ext4_group_info *grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;
926

927
	mb_debug(1, "load group %u\n", group);
928 929

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
930
	grp = ext4_get_group_info(sb, group);
931 932 933 934 935 936 937

	e4b->bd_blkbits = sb->s_blocksize_bits;
	e4b->bd_info = ext4_get_group_info(sb, group);
	e4b->bd_sb = sb;
	e4b->bd_group = group;
	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;
938 939 940 941 942 943 944 945 946
	e4b->alloc_semp = &grp->alloc_sem;

	/* Take the read lock on the group alloc
	 * sem. This would make sure a parallel
	 * ext4_mb_init_group happening on other
	 * groups mapped by the page is blocked
	 * till we are done with allocation
	 */
	down_read(e4b->alloc_semp);
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961

	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;

	/* we could use find_or_create_page(), but it locks page
	 * what we'd like to avoid in fast path ... */
	page = find_get_page(inode->i_mapping, pnum);
	if (page == NULL || !PageUptodate(page)) {
		if (page)
962 963 964 965 966 967 968 969
			/*
			 * drop the page reference and try
			 * to get the page with lock. If we
			 * are not uptodate that implies
			 * somebody just created the page but
			 * is yet to initialize the same. So
			 * wait for it to initialize.
			 */
970 971 972 973 974
			page_cache_release(page);
		page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
		if (page) {
			BUG_ON(page->mapping != inode->i_mapping);
			if (!PageUptodate(page)) {
975 976 977 978 979
				ret = ext4_mb_init_cache(page, NULL);
				if (ret) {
					unlock_page(page);
					goto err;
				}
980 981 982 983 984 985
				mb_cmp_bitmaps(e4b, page_address(page) +
					       (poff * sb->s_blocksize));
			}
			unlock_page(page);
		}
	}
986 987
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
988
		goto err;
989
	}
990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004
	e4b->bd_bitmap_page = page;
	e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);
	mark_page_accessed(page);

	block++;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;

	page = find_get_page(inode->i_mapping, pnum);
	if (page == NULL || !PageUptodate(page)) {
		if (page)
			page_cache_release(page);
		page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
		if (page) {
			BUG_ON(page->mapping != inode->i_mapping);
1005 1006 1007 1008 1009 1010 1011
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
				if (ret) {
					unlock_page(page);
					goto err;
				}
			}
1012 1013 1014
			unlock_page(page);
		}
	}
1015 1016
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1017
		goto err;
1018
	}
1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
	e4b->bd_buddy_page = page;
	e4b->bd_buddy = page_address(page) + (poff * sb->s_blocksize);
	mark_page_accessed(page);

	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

	return 0;

err:
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
	e4b->bd_buddy = NULL;
	e4b->bd_bitmap = NULL;
1035 1036 1037

	/* Done with the buddy cache */
	up_read(e4b->alloc_semp);
1038
	return ret;
1039 1040 1041 1042 1043 1044 1045 1046
}

static void ext4_mb_release_desc(struct ext4_buddy *e4b)
{
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
1047
	/* Done with the buddy cache */
1048 1049
	if (e4b->alloc_semp)
		up_read(e4b->alloc_semp);
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073
}


static int mb_find_order_for_block(struct ext4_buddy *e4b, int block)
{
	int order = 1;
	void *bb;

	BUG_ON(EXT4_MB_BITMAP(e4b) == EXT4_MB_BUDDY(e4b));
	BUG_ON(block >= (1 << (e4b->bd_blkbits + 3)));

	bb = EXT4_MB_BUDDY(e4b);
	while (order <= e4b->bd_blkbits + 1) {
		block = block >> 1;
		if (!mb_test_bit(block, bb)) {
			/* this block is part of buddy of order 'order' */
			return order;
		}
		bb += 1 << (e4b->bd_blkbits - order);
		order++;
	}
	return 0;
}

1074
static void mb_clear_bits(void *bm, int cur, int len)
1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
{
	__u32 *addr;

	len = cur + len;
	while (cur < len) {
		if ((cur & 31) == 0 && (len - cur) >= 32) {
			/* fast path: clear whole word at once */
			addr = bm + (cur >> 3);
			*addr = 0;
			cur += 32;
			continue;
		}
1087
		mb_clear_bit(cur, bm);
1088 1089 1090 1091
		cur++;
	}
}

1092
static void mb_set_bits(void *bm, int cur, int len)
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
{
	__u32 *addr;

	len = cur + len;
	while (cur < len) {
		if ((cur & 31) == 0 && (len - cur) >= 32) {
			/* fast path: set whole word at once */
			addr = bm + (cur >> 3);
			*addr = 0xffffffff;
			cur += 32;
			continue;
		}
1105
		mb_set_bit(cur, bm);
1106 1107 1108 1109
		cur++;
	}
}

1110
static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
			  int first, int count)
{
	int block = 0;
	int max = 0;
	int order;
	void *buddy;
	void *buddy2;
	struct super_block *sb = e4b->bd_sb;

	BUG_ON(first + count > (sb->s_blocksize << 3));
1121
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149
	mb_check_buddy(e4b);
	mb_free_blocks_double(inode, e4b, first, count);

	e4b->bd_info->bb_free += count;
	if (first < e4b->bd_info->bb_first_free)
		e4b->bd_info->bb_first_free = first;

	/* let's maintain fragments counter */
	if (first != 0)
		block = !mb_test_bit(first - 1, EXT4_MB_BITMAP(e4b));
	if (first + count < EXT4_SB(sb)->s_mb_maxs[0])
		max = !mb_test_bit(first + count, EXT4_MB_BITMAP(e4b));
	if (block && max)
		e4b->bd_info->bb_fragments--;
	else if (!block && !max)
		e4b->bd_info->bb_fragments++;

	/* let's maintain buddy itself */
	while (count-- > 0) {
		block = first++;
		order = 0;

		if (!mb_test_bit(block, EXT4_MB_BITMAP(e4b))) {
			ext4_fsblk_t blocknr;
			blocknr = e4b->bd_group * EXT4_BLOCKS_PER_GROUP(sb);
			blocknr += block;
			blocknr +=
			    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
1150 1151
			ext4_grp_locked_error(sb, e4b->bd_group,
				   __func__, "double-free of inode"
1152
				   " %lu's block %llu(bit %u in group %u)",
1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
				   inode ? inode->i_ino : 0, blocknr, block,
				   e4b->bd_group);
		}
		mb_clear_bit(block, EXT4_MB_BITMAP(e4b));
		e4b->bd_info->bb_counters[order]++;

		/* start of the buddy */
		buddy = mb_find_buddy(e4b, order, &max);

		do {
			block &= ~1UL;
			if (mb_test_bit(block, buddy) ||
					mb_test_bit(block + 1, buddy))
				break;

			/* both the buddies are free, try to coalesce them */
			buddy2 = mb_find_buddy(e4b, order + 1, &max);

			if (!buddy2)
				break;

			if (order > 0) {
				/* for special purposes, we don't set
				 * free bits in bitmap */
				mb_set_bit(block, buddy);
				mb_set_bit(block + 1, buddy);
			}
			e4b->bd_info->bb_counters[order]--;
			e4b->bd_info->bb_counters[order]--;

			block = block >> 1;
			order++;
			e4b->bd_info->bb_counters[order]++;

			mb_clear_bit(block, buddy2);
			buddy = buddy2;
		} while (1);
	}
	mb_check_buddy(e4b);
}

static int mb_find_extent(struct ext4_buddy *e4b, int order, int block,
				int needed, struct ext4_free_extent *ex)
{
	int next = block;
	int max;
	int ord;
	void *buddy;

1202
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265
	BUG_ON(ex == NULL);

	buddy = mb_find_buddy(e4b, order, &max);
	BUG_ON(buddy == NULL);
	BUG_ON(block >= max);
	if (mb_test_bit(block, buddy)) {
		ex->fe_len = 0;
		ex->fe_start = 0;
		ex->fe_group = 0;
		return 0;
	}

	/* FIXME dorp order completely ? */
	if (likely(order == 0)) {
		/* find actual order */
		order = mb_find_order_for_block(e4b, block);
		block = block >> order;
	}

	ex->fe_len = 1 << order;
	ex->fe_start = block << order;
	ex->fe_group = e4b->bd_group;

	/* calc difference from given start */
	next = next - ex->fe_start;
	ex->fe_len -= next;
	ex->fe_start += next;

	while (needed > ex->fe_len &&
	       (buddy = mb_find_buddy(e4b, order, &max))) {

		if (block + 1 >= max)
			break;

		next = (block + 1) * (1 << order);
		if (mb_test_bit(next, EXT4_MB_BITMAP(e4b)))
			break;

		ord = mb_find_order_for_block(e4b, next);

		order = ord;
		block = next >> order;
		ex->fe_len += 1 << order;
	}

	BUG_ON(ex->fe_start + ex->fe_len > (1 << (e4b->bd_blkbits + 3)));
	return ex->fe_len;
}

static int mb_mark_used(struct ext4_buddy *e4b, struct ext4_free_extent *ex)
{
	int ord;
	int mlen = 0;
	int max = 0;
	int cur;
	int start = ex->fe_start;
	int len = ex->fe_len;
	unsigned ret = 0;
	int len0 = len;
	void *buddy;

	BUG_ON(start + len > (e4b->bd_sb->s_blocksize << 3));
	BUG_ON(e4b->bd_group != ex->fe_group);
1266
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319
	mb_check_buddy(e4b);
	mb_mark_used_double(e4b, start, len);

	e4b->bd_info->bb_free -= len;
	if (e4b->bd_info->bb_first_free == start)
		e4b->bd_info->bb_first_free += len;

	/* let's maintain fragments counter */
	if (start != 0)
		mlen = !mb_test_bit(start - 1, EXT4_MB_BITMAP(e4b));
	if (start + len < EXT4_SB(e4b->bd_sb)->s_mb_maxs[0])
		max = !mb_test_bit(start + len, EXT4_MB_BITMAP(e4b));
	if (mlen && max)
		e4b->bd_info->bb_fragments++;
	else if (!mlen && !max)
		e4b->bd_info->bb_fragments--;

	/* let's maintain buddy itself */
	while (len) {
		ord = mb_find_order_for_block(e4b, start);

		if (((start >> ord) << ord) == start && len >= (1 << ord)) {
			/* the whole chunk may be allocated at once! */
			mlen = 1 << ord;
			buddy = mb_find_buddy(e4b, ord, &max);
			BUG_ON((start >> ord) >= max);
			mb_set_bit(start >> ord, buddy);
			e4b->bd_info->bb_counters[ord]--;
			start += mlen;
			len -= mlen;
			BUG_ON(len < 0);
			continue;
		}

		/* store for history */
		if (ret == 0)
			ret = len | (ord << 16);

		/* we have to split large buddy */
		BUG_ON(ord <= 0);
		buddy = mb_find_buddy(e4b, ord, &max);
		mb_set_bit(start >> ord, buddy);
		e4b->bd_info->bb_counters[ord]--;

		ord--;
		cur = (start >> ord) & ~1U;
		buddy = mb_find_buddy(e4b, ord, &max);
		mb_clear_bit(cur, buddy);
		mb_clear_bit(cur + 1, buddy);
		e4b->bd_info->bb_counters[ord]++;
		e4b->bd_info->bb_counters[ord]++;
	}

1320
	mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
	mb_check_buddy(e4b);

	return ret;
}

/*
 * Must be called under group lock!
 */
static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	int ret;

	BUG_ON(ac->ac_b_ex.fe_group != e4b->bd_group);
	BUG_ON(ac->ac_status == AC_STATUS_FOUND);

	ac->ac_b_ex.fe_len = min(ac->ac_b_ex.fe_len, ac->ac_g_ex.fe_len);
	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
	ret = mb_mark_used(e4b, &ac->ac_b_ex);

	/* preallocation can change ac_b_ex, thus we store actually
	 * allocated blocks for history */
	ac->ac_f_ex = ac->ac_b_ex;

	ac->ac_status = AC_STATUS_FOUND;
	ac->ac_tail = ret & 0xffff;
	ac->ac_buddy = ret >> 16;

1350 1351 1352 1353 1354 1355 1356
	/*
	 * take the page reference. We want the page to be pinned
	 * so that we don't get a ext4_mb_init_cache_call for this
	 * group until we update the bitmap. That would mean we
	 * double allocate blocks. The reference is dropped
	 * in ext4_mb_release_context
	 */
1357 1358 1359 1360
	ac->ac_bitmap_page = e4b->bd_bitmap_page;
	get_page(ac->ac_bitmap_page);
	ac->ac_buddy_page = e4b->bd_buddy_page;
	get_page(ac->ac_buddy_page);
1361 1362 1363
	/* on allocation we use ac to track the held semaphore */
	ac->alloc_semp =  e4b->alloc_semp;
	e4b->alloc_semp = NULL;
1364
	/* store last allocated for subsequent stream allocation */
1365
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
		spin_lock(&sbi->s_md_lock);
		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
		spin_unlock(&sbi->s_md_lock);
	}
}

/*
 * regular allocator, for general purposes allocation
 */

static void ext4_mb_check_limits(struct ext4_allocation_context *ac,
					struct ext4_buddy *e4b,
					int finish_group)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	struct ext4_free_extent *bex = &ac->ac_b_ex;
	struct ext4_free_extent *gex = &ac->ac_g_ex;
	struct ext4_free_extent ex;
	int max;

1387 1388
	if (ac->ac_status == AC_STATUS_FOUND)
		return;
1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434
	/*
	 * We don't want to scan for a whole year
	 */
	if (ac->ac_found > sbi->s_mb_max_to_scan &&
			!(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
		ac->ac_status = AC_STATUS_BREAK;
		return;
	}

	/*
	 * Haven't found good chunk so far, let's continue
	 */
	if (bex->fe_len < gex->fe_len)
		return;

	if ((finish_group || ac->ac_found > sbi->s_mb_min_to_scan)
			&& bex->fe_group == e4b->bd_group) {
		/* recheck chunk's availability - we don't know
		 * when it was found (within this lock-unlock
		 * period or not) */
		max = mb_find_extent(e4b, 0, bex->fe_start, gex->fe_len, &ex);
		if (max >= gex->fe_len) {
			ext4_mb_use_best_found(ac, e4b);
			return;
		}
	}
}

/*
 * The routine checks whether found extent is good enough. If it is,
 * then the extent gets marked used and flag is set to the context
 * to stop scanning. Otherwise, the extent is compared with the
 * previous found extent and if new one is better, then it's stored
 * in the context. Later, the best found extent will be used, if
 * mballoc can't find good enough extent.
 *
 * FIXME: real allocation policy is to be designed yet!
 */
static void ext4_mb_measure_extent(struct ext4_allocation_context *ac,
					struct ext4_free_extent *ex,
					struct ext4_buddy *e4b)
{
	struct ext4_free_extent *bex = &ac->ac_b_ex;
	struct ext4_free_extent *gex = &ac->ac_g_ex;

	BUG_ON(ex->fe_len <= 0);
1435
	BUG_ON(ex->fe_len > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
	BUG_ON(ex->fe_start >= EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
	BUG_ON(ac->ac_status != AC_STATUS_CONTINUE);

	ac->ac_found++;

	/*
	 * The special case - take what you catch first
	 */
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
		*bex = *ex;
		ext4_mb_use_best_found(ac, e4b);
		return;
	}

	/*
	 * Let's check whether the chuck is good enough
	 */
	if (ex->fe_len == gex->fe_len) {
		*bex = *ex;
		ext4_mb_use_best_found(ac, e4b);
		return;
	}

	/*
	 * If this is first found extent, just store it in the context
	 */
	if (bex->fe_len == 0) {
		*bex = *ex;
		return;
	}

	/*
	 * If new found extent is better, store it in the context
	 */
	if (bex->fe_len < gex->fe_len) {
		/* if the request isn't satisfied, any found extent
		 * larger than previous best one is better */
		if (ex->fe_len > bex->fe_len)
			*bex = *ex;
	} else if (ex->fe_len > gex->fe_len) {
		/* if the request is satisfied, then we try to find
		 * an extent that still satisfy the request, but is
		 * smaller than previous one */
		if (ex->fe_len < bex->fe_len)
			*bex = *ex;
	}

	ext4_mb_check_limits(ac, e4b, 0);
}

1486 1487
static noinline_for_stack
int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513
					struct ext4_buddy *e4b)
{
	struct ext4_free_extent ex = ac->ac_b_ex;
	ext4_group_t group = ex.fe_group;
	int max;
	int err;

	BUG_ON(ex.fe_len <= 0);
	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
	if (err)
		return err;

	ext4_lock_group(ac->ac_sb, group);
	max = mb_find_extent(e4b, 0, ex.fe_start, ex.fe_len, &ex);

	if (max > 0) {
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	}

	ext4_unlock_group(ac->ac_sb, group);
	ext4_mb_release_desc(e4b);

	return 0;
}

1514 1515
static noinline_for_stack
int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573
				struct ext4_buddy *e4b)
{
	ext4_group_t group = ac->ac_g_ex.fe_group;
	int max;
	int err;
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	struct ext4_super_block *es = sbi->s_es;
	struct ext4_free_extent ex;

	if (!(ac->ac_flags & EXT4_MB_HINT_TRY_GOAL))
		return 0;

	err = ext4_mb_load_buddy(ac->ac_sb, group, e4b);
	if (err)
		return err;

	ext4_lock_group(ac->ac_sb, group);
	max = mb_find_extent(e4b, 0, ac->ac_g_ex.fe_start,
			     ac->ac_g_ex.fe_len, &ex);

	if (max >= ac->ac_g_ex.fe_len && ac->ac_g_ex.fe_len == sbi->s_stripe) {
		ext4_fsblk_t start;

		start = (e4b->bd_group * EXT4_BLOCKS_PER_GROUP(ac->ac_sb)) +
			ex.fe_start + le32_to_cpu(es->s_first_data_block);
		/* use do_div to get remainder (would be 64-bit modulo) */
		if (do_div(start, sbi->s_stripe) == 0) {
			ac->ac_found++;
			ac->ac_b_ex = ex;
			ext4_mb_use_best_found(ac, e4b);
		}
	} else if (max >= ac->ac_g_ex.fe_len) {
		BUG_ON(ex.fe_len <= 0);
		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
		ac->ac_found++;
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	} else if (max > 0 && (ac->ac_flags & EXT4_MB_HINT_MERGE)) {
		/* Sometimes, caller may want to merge even small
		 * number of blocks to an existing extent */
		BUG_ON(ex.fe_len <= 0);
		BUG_ON(ex.fe_group != ac->ac_g_ex.fe_group);
		BUG_ON(ex.fe_start != ac->ac_g_ex.fe_start);
		ac->ac_found++;
		ac->ac_b_ex = ex;
		ext4_mb_use_best_found(ac, e4b);
	}
	ext4_unlock_group(ac->ac_sb, group);
	ext4_mb_release_desc(e4b);

	return 0;
}

/*
 * The routine scans buddy structures (not bitmap!) from given order
 * to max order and tries to find big enough chunk to satisfy the req
 */
1574 1575
static noinline_for_stack
void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592
					struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_group_info *grp = e4b->bd_info;
	void *buddy;
	int i;
	int k;
	int max;

	BUG_ON(ac->ac_2order <= 0);
	for (i = ac->ac_2order; i <= sb->s_blocksize_bits + 1; i++) {
		if (grp->bb_counters[i] == 0)
			continue;

		buddy = mb_find_buddy(e4b, i, &max);
		BUG_ON(buddy == NULL);

1593
		k = mb_find_next_zero_bit(buddy, max, 0);
1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617
		BUG_ON(k >= max);

		ac->ac_found++;

		ac->ac_b_ex.fe_len = 1 << i;
		ac->ac_b_ex.fe_start = k << i;
		ac->ac_b_ex.fe_group = e4b->bd_group;

		ext4_mb_use_best_found(ac, e4b);

		BUG_ON(ac->ac_b_ex.fe_len != ac->ac_g_ex.fe_len);

		if (EXT4_SB(sb)->s_mb_stats)
			atomic_inc(&EXT4_SB(sb)->s_bal_2orders);

		break;
	}
}

/*
 * The routine scans the group and measures all found extents.
 * In order to optimize scanning, caller must pass number of
 * free blocks in the group, so the routine can know upper limit.
 */
1618 1619
static noinline_for_stack
void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633
					struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
	void *bitmap = EXT4_MB_BITMAP(e4b);
	struct ext4_free_extent ex;
	int i;
	int free;

	free = e4b->bd_info->bb_free;
	BUG_ON(free <= 0);

	i = e4b->bd_info->bb_first_free;

	while (free && ac->ac_status == AC_STATUS_CONTINUE) {
1634
		i = mb_find_next_zero_bit(bitmap,
1635 1636
						EXT4_BLOCKS_PER_GROUP(sb), i);
		if (i >= EXT4_BLOCKS_PER_GROUP(sb)) {
1637
			/*
1638
			 * IF we have corrupt bitmap, we won't find any
1639 1640 1641
			 * free blocks even though group info says we
			 * we have free blocks
			 */
1642 1643
			ext4_grp_locked_error(sb, e4b->bd_group,
					__func__, "%d free blocks as per "
1644
					"group info. But bitmap says 0",
1645
					free);
1646 1647 1648 1649 1650
			break;
		}

		mb_find_extent(e4b, 0, i, ac->ac_g_ex.fe_len, &ex);
		BUG_ON(ex.fe_len <= 0);
1651
		if (free < ex.fe_len) {
1652 1653
			ext4_grp_locked_error(sb, e4b->bd_group,
					__func__, "%d free blocks as per "
1654
					"group info. But got %d blocks",
1655
					free, ex.fe_len);
1656 1657 1658 1659 1660 1661
			/*
			 * The number of free blocks differs. This mostly
			 * indicate that the bitmap is corrupt. So exit
			 * without claiming the space.
			 */
			break;
1662
		}
1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677

		ext4_mb_measure_extent(ac, &ex, e4b);

		i += ex.fe_len;
		free -= ex.fe_len;
	}

	ext4_mb_check_limits(ac, e4b, 1);
}

/*
 * This is a special case for storages like raid5
 * we try to find stripe-aligned chunks for stripe-size requests
 * XXX should do so at least for multiples of stripe size as well
 */
1678 1679
static noinline_for_stack
void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718
				 struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	void *bitmap = EXT4_MB_BITMAP(e4b);
	struct ext4_free_extent ex;
	ext4_fsblk_t first_group_block;
	ext4_fsblk_t a;
	ext4_grpblk_t i;
	int max;

	BUG_ON(sbi->s_stripe == 0);

	/* find first stripe-aligned block in group */
	first_group_block = e4b->bd_group * EXT4_BLOCKS_PER_GROUP(sb)
		+ le32_to_cpu(sbi->s_es->s_first_data_block);
	a = first_group_block + sbi->s_stripe - 1;
	do_div(a, sbi->s_stripe);
	i = (a * sbi->s_stripe) - first_group_block;

	while (i < EXT4_BLOCKS_PER_GROUP(sb)) {
		if (!mb_test_bit(i, bitmap)) {
			max = mb_find_extent(e4b, 0, i, sbi->s_stripe, &ex);
			if (max >= sbi->s_stripe) {
				ac->ac_found++;
				ac->ac_b_ex = ex;
				ext4_mb_use_best_found(ac, e4b);
				break;
			}
		}
		i += sbi->s_stripe;
	}
}

static int ext4_mb_good_group(struct ext4_allocation_context *ac,
				ext4_group_t group, int cr)
{
	unsigned free, fragments;
	unsigned i, bits;
1719
	int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);

	BUG_ON(cr < 0 || cr >= 4);
	BUG_ON(EXT4_MB_GRP_NEED_INIT(grp));

	free = grp->bb_free;
	fragments = grp->bb_fragments;
	if (free == 0)
		return 0;
	if (fragments == 0)
		return 0;

	switch (cr) {
	case 0:
		BUG_ON(ac->ac_2order == 0);

1736 1737 1738 1739 1740 1741
		/* Avoid using the first bg of a flexgroup for data files */
		if ((ac->ac_flags & EXT4_MB_HINT_DATA) &&
		    (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) &&
		    ((group % flex_size) == 0))
			return 0;

1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763
		bits = ac->ac_sb->s_blocksize_bits + 1;
		for (i = ac->ac_2order; i <= bits; i++)
			if (grp->bb_counters[i] > 0)
				return 1;
		break;
	case 1:
		if ((free / fragments) >= ac->ac_g_ex.fe_len)
			return 1;
		break;
	case 2:
		if (free >= ac->ac_g_ex.fe_len)
			return 1;
		break;
	case 3:
		return 1;
	default:
		BUG();
	}

	return 0;
}

1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776
/*
 * lock the group_info alloc_sem of all the groups
 * belonging to the same buddy cache page. This
 * make sure other parallel operation on the buddy
 * cache doesn't happen  whild holding the buddy cache
 * lock
 */
int ext4_mb_get_buddy_cache_lock(struct super_block *sb, ext4_group_t group)
{
	int i;
	int block, pnum;
	int blocks_per_page;
	int groups_per_page;
1777
	ext4_group_t ngroups = ext4_get_groups_count(sb);
1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796
	ext4_group_t first_group;
	struct ext4_group_info *grp;

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	first_group = pnum * blocks_per_page / 2;

	groups_per_page = blocks_per_page >> 1;
	if (groups_per_page == 0)
		groups_per_page = 1;
	/* read all groups the page covers into the cache */
	for (i = 0; i < groups_per_page; i++) {

1797
		if ((first_group + i) >= ngroups)
1798 1799 1800 1801 1802 1803 1804
			break;
		grp = ext4_get_group_info(sb, first_group + i);
		/* take all groups write allocation
		 * semaphore. This make sure there is
		 * no block allocation going on in any
		 * of that groups
		 */
1805
		down_write_nested(&grp->alloc_sem, i);
1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
	}
	return i;
}

void ext4_mb_put_buddy_cache_lock(struct super_block *sb,
					ext4_group_t group, int locked_group)
{
	int i;
	int block, pnum;
	int blocks_per_page;
	ext4_group_t first_group;
	struct ext4_group_info *grp;

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	first_group = pnum * blocks_per_page / 2;
	/* release locks on all the groups */
	for (i = 0; i < locked_group; i++) {

		grp = ext4_get_group_info(sb, first_group + i);
		/* take all groups write allocation
		 * semaphore. This make sure there is
		 * no block allocation going on in any
		 * of that groups
		 */
		up_write(&grp->alloc_sem);
	}

}

1842 1843
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855
{

	int ret;
	void *bitmap;
	int blocks_per_page;
	int block, pnum, poff;
	int num_grp_locked = 0;
	struct ext4_group_info *this_grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;
	struct page *page = NULL, *bitmap_page = NULL;

1856
	mb_debug(1, "init group %u\n", group);
1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932
	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	this_grp = ext4_get_group_info(sb, group);
	/*
	 * This ensures we don't add group
	 * to this buddy cache via resize
	 */
	num_grp_locked =  ext4_mb_get_buddy_cache_lock(sb, group);
	if (!EXT4_MB_GRP_NEED_INIT(this_grp)) {
		/*
		 * somebody initialized the group
		 * return without doing anything
		 */
		ret = 0;
		goto err;
	}
	/*
	 * the buddy cache inode stores the block bitmap
	 * and buddy information in consecutive blocks.
	 * So for each group we need two blocks.
	 */
	block = group * 2;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (page) {
		BUG_ON(page->mapping != inode->i_mapping);
		ret = ext4_mb_init_cache(page, NULL);
		if (ret) {
			unlock_page(page);
			goto err;
		}
		unlock_page(page);
	}
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);
	bitmap_page = page;
	bitmap = page_address(page) + (poff * sb->s_blocksize);

	/* init buddy cache */
	block++;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (page == bitmap_page) {
		/*
		 * If both the bitmap and buddy are in
		 * the same page we don't need to force
		 * init the buddy
		 */
		unlock_page(page);
	} else if (page) {
		BUG_ON(page->mapping != inode->i_mapping);
		ret = ext4_mb_init_cache(page, bitmap);
		if (ret) {
			unlock_page(page);
			goto err;
		}
		unlock_page(page);
	}
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);
err:
	ext4_mb_put_buddy_cache_lock(sb, group, num_grp_locked);
	if (bitmap_page)
		page_cache_release(bitmap_page);
	if (page)
		page_cache_release(page);
	return ret;
}

1933 1934
static noinline_for_stack int
ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1935
{
1936
	ext4_group_t ngroups, group, i;
1937 1938 1939 1940 1941 1942 1943 1944 1945
	int cr;
	int err = 0;
	int bsbits;
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	struct ext4_buddy e4b;

	sb = ac->ac_sb;
	sbi = EXT4_SB(sb);
1946
	ngroups = ext4_get_groups_count(sb);
1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966
	BUG_ON(ac->ac_status == AC_STATUS_FOUND);

	/* first, try the goal */
	err = ext4_mb_find_by_goal(ac, &e4b);
	if (err || ac->ac_status == AC_STATUS_FOUND)
		goto out;

	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		goto out;

	/*
	 * ac->ac2_order is set only if the fe_len is a power of 2
	 * if ac2_order is set we also set criteria to 0 so that we
	 * try exact allocation using buddy.
	 */
	i = fls(ac->ac_g_ex.fe_len);
	ac->ac_2order = 0;
	/*
	 * We search using buddy data only if the order of the request
	 * is greater than equal to the sbi_s_mb_order2_reqs
T
Theodore Ts'o 已提交
1967
	 * You can tune it via /sys/fs/ext4/<partition>/mb_order2_req
1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
	 */
	if (i >= sbi->s_mb_order2_reqs) {
		/*
		 * This should tell if fe_len is exactly power of 2
		 */
		if ((ac->ac_g_ex.fe_len & (~(1 << (i - 1)))) == 0)
			ac->ac_2order = i - 1;
	}

	bsbits = ac->ac_sb->s_blocksize_bits;

1979 1980
	/* if stream allocation is enabled, use global goal */
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1981 1982 1983 1984 1985 1986
		/* TBD: may be hot point */
		spin_lock(&sbi->s_md_lock);
		ac->ac_g_ex.fe_group = sbi->s_mb_last_group;
		ac->ac_g_ex.fe_start = sbi->s_mb_last_start;
		spin_unlock(&sbi->s_md_lock);
	}
1987

1988 1989 1990 1991 1992 1993 1994 1995 1996
	/* Let's just scan groups to find more-less suitable blocks */
	cr = ac->ac_2order ? 0 : 1;
	/*
	 * cr == 0 try to get exact allocation,
	 * cr == 3  try to get anything
	 */
repeat:
	for (; cr < 4 && ac->ac_status == AC_STATUS_CONTINUE; cr++) {
		ac->ac_criteria = cr;
1997 1998 1999 2000 2001 2002
		/*
		 * searching for the right group start
		 * from the goal value specified
		 */
		group = ac->ac_g_ex.fe_group;

2003
		for (i = 0; i < ngroups; group++, i++) {
2004 2005 2006
			struct ext4_group_info *grp;
			struct ext4_group_desc *desc;

2007
			if (group == ngroups)
2008 2009 2010
				group = 0;

			/* quick check to skip empty groups */
2011
			grp = ext4_get_group_info(sb, group);
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
			if (grp->bb_free == 0)
				continue;

			/*
			 * if the group is already init we check whether it is
			 * a good group and if not we don't load the buddy
			 */
			if (EXT4_MB_GRP_NEED_INIT(grp)) {
				/*
				 * we need full data about the group
				 * to make a good selection
				 */
2024
				err = ext4_mb_init_group(sb, group);
2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049
				if (err)
					goto out;
			}

			/*
			 * If the particular group doesn't satisfy our
			 * criteria we continue with the next group
			 */
			if (!ext4_mb_good_group(ac, group, cr))
				continue;

			err = ext4_mb_load_buddy(sb, group, &e4b);
			if (err)
				goto out;

			ext4_lock_group(sb, group);
			if (!ext4_mb_good_group(ac, group, cr)) {
				/* someone did allocation from this group */
				ext4_unlock_group(sb, group);
				ext4_mb_release_desc(&e4b);
				continue;
			}

			ac->ac_groups_scanned++;
			desc = ext4_get_group_desc(sb, group, NULL);
2050
			if (cr == 0)
2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155
				ext4_mb_simple_scan_group(ac, &e4b);
			else if (cr == 1 &&
					ac->ac_g_ex.fe_len == sbi->s_stripe)
				ext4_mb_scan_aligned(ac, &e4b);
			else
				ext4_mb_complex_scan_group(ac, &e4b);

			ext4_unlock_group(sb, group);
			ext4_mb_release_desc(&e4b);

			if (ac->ac_status != AC_STATUS_CONTINUE)
				break;
		}
	}

	if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
	    !(ac->ac_flags & EXT4_MB_HINT_FIRST)) {
		/*
		 * We've been searching too long. Let's try to allocate
		 * the best chunk we've found so far
		 */

		ext4_mb_try_best_found(ac, &e4b);
		if (ac->ac_status != AC_STATUS_FOUND) {
			/*
			 * Someone more lucky has already allocated it.
			 * The only thing we can do is just take first
			 * found block(s)
			printk(KERN_DEBUG "EXT4-fs: someone won our chunk\n");
			 */
			ac->ac_b_ex.fe_group = 0;
			ac->ac_b_ex.fe_start = 0;
			ac->ac_b_ex.fe_len = 0;
			ac->ac_status = AC_STATUS_CONTINUE;
			ac->ac_flags |= EXT4_MB_HINT_FIRST;
			cr = 3;
			atomic_inc(&sbi->s_mb_lost_chunks);
			goto repeat;
		}
	}
out:
	return err;
}

#ifdef EXT4_MB_HISTORY
struct ext4_mb_proc_session {
	struct ext4_mb_history *history;
	struct super_block *sb;
	int start;
	int max;
};

static void *ext4_mb_history_skip_empty(struct ext4_mb_proc_session *s,
					struct ext4_mb_history *hs,
					int first)
{
	if (hs == s->history + s->max)
		hs = s->history;
	if (!first && hs == s->history + s->start)
		return NULL;
	while (hs->orig.fe_len == 0) {
		hs++;
		if (hs == s->history + s->max)
			hs = s->history;
		if (hs == s->history + s->start)
			return NULL;
	}
	return hs;
}

static void *ext4_mb_seq_history_start(struct seq_file *seq, loff_t *pos)
{
	struct ext4_mb_proc_session *s = seq->private;
	struct ext4_mb_history *hs;
	int l = *pos;

	if (l == 0)
		return SEQ_START_TOKEN;
	hs = ext4_mb_history_skip_empty(s, s->history + s->start, 1);
	if (!hs)
		return NULL;
	while (--l && (hs = ext4_mb_history_skip_empty(s, ++hs, 0)) != NULL);
	return hs;
}

static void *ext4_mb_seq_history_next(struct seq_file *seq, void *v,
				      loff_t *pos)
{
	struct ext4_mb_proc_session *s = seq->private;
	struct ext4_mb_history *hs = v;

	++*pos;
	if (v == SEQ_START_TOKEN)
		return ext4_mb_history_skip_empty(s, s->history + s->start, 1);
	else
		return ext4_mb_history_skip_empty(s, ++hs, 0);
}

static int ext4_mb_seq_history_show(struct seq_file *seq, void *v)
{
	char buf[25], buf2[25], buf3[25], *fmt;
	struct ext4_mb_history *hs = v;

	if (v == SEQ_START_TOKEN) {
		seq_printf(seq, "%-5s %-8s %-23s %-23s %-23s %-5s "
2156
				"%-5s %-2s %-6s %-5s %-5s %-6s\n",
2157 2158 2159 2160 2161 2162 2163
			  "pid", "inode", "original", "goal", "result", "found",
			   "grps", "cr", "flags", "merge", "tail", "broken");
		return 0;
	}

	if (hs->op == EXT4_MB_HISTORY_ALLOC) {
		fmt = "%-5u %-8u %-23s %-23s %-23s %-5u %-5u %-2u "
2164
			"0x%04x %-5s %-5u %-6u\n";
2165
		sprintf(buf2, "%u/%d/%u@%u", hs->result.fe_group,
2166 2167
			hs->result.fe_start, hs->result.fe_len,
			hs->result.fe_logical);
2168
		sprintf(buf, "%u/%d/%u@%u", hs->orig.fe_group,
2169 2170
			hs->orig.fe_start, hs->orig.fe_len,
			hs->orig.fe_logical);
2171
		sprintf(buf3, "%u/%d/%u@%u", hs->goal.fe_group,
2172 2173 2174 2175 2176 2177 2178 2179
			hs->goal.fe_start, hs->goal.fe_len,
			hs->goal.fe_logical);
		seq_printf(seq, fmt, hs->pid, hs->ino, buf, buf3, buf2,
				hs->found, hs->groups, hs->cr, hs->flags,
				hs->merged ? "M" : "", hs->tail,
				hs->buddy ? 1 << hs->buddy : 0);
	} else if (hs->op == EXT4_MB_HISTORY_PREALLOC) {
		fmt = "%-5u %-8u %-23s %-23s %-23s\n";
2180
		sprintf(buf2, "%u/%d/%u@%u", hs->result.fe_group,
2181 2182
			hs->result.fe_start, hs->result.fe_len,
			hs->result.fe_logical);
2183
		sprintf(buf, "%u/%d/%u@%u", hs->orig.fe_group,
2184 2185 2186 2187
			hs->orig.fe_start, hs->orig.fe_len,
			hs->orig.fe_logical);
		seq_printf(seq, fmt, hs->pid, hs->ino, buf, "", buf2);
	} else if (hs->op == EXT4_MB_HISTORY_DISCARD) {
2188
		sprintf(buf2, "%u/%d/%u", hs->result.fe_group,
2189 2190 2191 2192
			hs->result.fe_start, hs->result.fe_len);
		seq_printf(seq, "%-5u %-8u %-23s discard\n",
				hs->pid, hs->ino, buf2);
	} else if (hs->op == EXT4_MB_HISTORY_FREE) {
2193
		sprintf(buf2, "%u/%d/%u", hs->result.fe_group,
2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219
			hs->result.fe_start, hs->result.fe_len);
		seq_printf(seq, "%-5u %-8u %-23s free\n",
				hs->pid, hs->ino, buf2);
	}
	return 0;
}

static void ext4_mb_seq_history_stop(struct seq_file *seq, void *v)
{
}

static struct seq_operations ext4_mb_seq_history_ops = {
	.start  = ext4_mb_seq_history_start,
	.next   = ext4_mb_seq_history_next,
	.stop   = ext4_mb_seq_history_stop,
	.show   = ext4_mb_seq_history_show,
};

static int ext4_mb_seq_history_open(struct inode *inode, struct file *file)
{
	struct super_block *sb = PDE(inode)->data;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_mb_proc_session *s;
	int rc;
	int size;

2220 2221
	if (unlikely(sbi->s_mb_history == NULL))
		return -ENOMEM;
2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300
	s = kmalloc(sizeof(*s), GFP_KERNEL);
	if (s == NULL)
		return -ENOMEM;
	s->sb = sb;
	size = sizeof(struct ext4_mb_history) * sbi->s_mb_history_max;
	s->history = kmalloc(size, GFP_KERNEL);
	if (s->history == NULL) {
		kfree(s);
		return -ENOMEM;
	}

	spin_lock(&sbi->s_mb_history_lock);
	memcpy(s->history, sbi->s_mb_history, size);
	s->max = sbi->s_mb_history_max;
	s->start = sbi->s_mb_history_cur % s->max;
	spin_unlock(&sbi->s_mb_history_lock);

	rc = seq_open(file, &ext4_mb_seq_history_ops);
	if (rc == 0) {
		struct seq_file *m = (struct seq_file *)file->private_data;
		m->private = s;
	} else {
		kfree(s->history);
		kfree(s);
	}
	return rc;

}

static int ext4_mb_seq_history_release(struct inode *inode, struct file *file)
{
	struct seq_file *seq = (struct seq_file *)file->private_data;
	struct ext4_mb_proc_session *s = seq->private;
	kfree(s->history);
	kfree(s);
	return seq_release(inode, file);
}

static ssize_t ext4_mb_seq_history_write(struct file *file,
				const char __user *buffer,
				size_t count, loff_t *ppos)
{
	struct seq_file *seq = (struct seq_file *)file->private_data;
	struct ext4_mb_proc_session *s = seq->private;
	struct super_block *sb = s->sb;
	char str[32];
	int value;

	if (count >= sizeof(str)) {
		printk(KERN_ERR "EXT4-fs: %s string too long, max %u bytes\n",
				"mb_history", (int)sizeof(str));
		return -EOVERFLOW;
	}

	if (copy_from_user(str, buffer, count))
		return -EFAULT;

	value = simple_strtol(str, NULL, 0);
	if (value < 0)
		return -ERANGE;
	EXT4_SB(sb)->s_mb_history_filter = value;

	return count;
}

static struct file_operations ext4_mb_seq_history_fops = {
	.owner		= THIS_MODULE,
	.open		= ext4_mb_seq_history_open,
	.read		= seq_read,
	.write		= ext4_mb_seq_history_write,
	.llseek		= seq_lseek,
	.release	= ext4_mb_seq_history_release,
};

static void *ext4_mb_seq_groups_start(struct seq_file *seq, loff_t *pos)
{
	struct super_block *sb = seq->private;
	ext4_group_t group;

2301
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2302 2303
		return NULL;
	group = *pos + 1;
2304
	return (void *) ((unsigned long) group);
2305 2306 2307 2308 2309 2310 2311 2312
}

static void *ext4_mb_seq_groups_next(struct seq_file *seq, void *v, loff_t *pos)
{
	struct super_block *sb = seq->private;
	ext4_group_t group;

	++*pos;
2313
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2314 2315
		return NULL;
	group = *pos + 1;
2316
	return (void *) ((unsigned long) group);
2317 2318 2319 2320 2321
}

static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
{
	struct super_block *sb = seq->private;
2322
	ext4_group_t group = (ext4_group_t) ((unsigned long) v);
2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343
	int i;
	int err;
	struct ext4_buddy e4b;
	struct sg {
		struct ext4_group_info info;
		unsigned short counters[16];
	} sg;

	group--;
	if (group == 0)
		seq_printf(seq, "#%-5s: %-5s %-5s %-5s "
				"[ %-5s %-5s %-5s %-5s %-5s %-5s %-5s "
				  "%-5s %-5s %-5s %-5s %-5s %-5s %-5s ]\n",
			   "group", "free", "frags", "first",
			   "2^0", "2^1", "2^2", "2^3", "2^4", "2^5", "2^6",
			   "2^7", "2^8", "2^9", "2^10", "2^11", "2^12", "2^13");

	i = (sb->s_blocksize_bits + 2) * sizeof(sg.info.bb_counters[0]) +
		sizeof(struct ext4_group_info);
	err = ext4_mb_load_buddy(sb, group, &e4b);
	if (err) {
2344
		seq_printf(seq, "#%-5u: I/O error\n", group);
2345 2346 2347 2348 2349 2350 2351
		return 0;
	}
	ext4_lock_group(sb, group);
	memcpy(&sg, ext4_get_group_info(sb, group), i);
	ext4_unlock_group(sb, group);
	ext4_mb_release_desc(&e4b);

2352
	seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398
			sg.info.bb_fragments, sg.info.bb_first_free);
	for (i = 0; i <= 13; i++)
		seq_printf(seq, " %-5u", i <= sb->s_blocksize_bits + 1 ?
				sg.info.bb_counters[i] : 0);
	seq_printf(seq, " ]\n");

	return 0;
}

static void ext4_mb_seq_groups_stop(struct seq_file *seq, void *v)
{
}

static struct seq_operations ext4_mb_seq_groups_ops = {
	.start  = ext4_mb_seq_groups_start,
	.next   = ext4_mb_seq_groups_next,
	.stop   = ext4_mb_seq_groups_stop,
	.show   = ext4_mb_seq_groups_show,
};

static int ext4_mb_seq_groups_open(struct inode *inode, struct file *file)
{
	struct super_block *sb = PDE(inode)->data;
	int rc;

	rc = seq_open(file, &ext4_mb_seq_groups_ops);
	if (rc == 0) {
		struct seq_file *m = (struct seq_file *)file->private_data;
		m->private = sb;
	}
	return rc;

}

static struct file_operations ext4_mb_seq_groups_fops = {
	.owner		= THIS_MODULE,
	.open		= ext4_mb_seq_groups_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= seq_release,
};

static void ext4_mb_history_release(struct super_block *sb)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);

2399 2400
	if (sbi->s_proc != NULL) {
		remove_proc_entry("mb_groups", sbi->s_proc);
2401 2402
		if (sbi->s_mb_history_max)
			remove_proc_entry("mb_history", sbi->s_proc);
2403
	}
2404 2405 2406 2407 2408 2409 2410 2411
	kfree(sbi->s_mb_history);
}

static void ext4_mb_history_init(struct super_block *sb)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	int i;

2412
	if (sbi->s_proc != NULL) {
2413 2414 2415
		if (sbi->s_mb_history_max)
			proc_create_data("mb_history", S_IRUGO, sbi->s_proc,
					 &ext4_mb_seq_history_fops, sb);
2416
		proc_create_data("mb_groups", S_IRUGO, sbi->s_proc,
2417
				 &ext4_mb_seq_groups_fops, sb);
2418 2419 2420 2421 2422
	}

	sbi->s_mb_history_cur = 0;
	spin_lock_init(&sbi->s_mb_history_lock);
	i = sbi->s_mb_history_max * sizeof(struct ext4_mb_history);
2423
	sbi->s_mb_history = i ? kzalloc(i, GFP_KERNEL) : NULL;
2424 2425 2426
	/* if we can't allocate history, then we simple won't use it */
}

2427 2428
static noinline_for_stack void
ext4_mb_store_history(struct ext4_allocation_context *ac)
2429 2430 2431 2432
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	struct ext4_mb_history h;

2433
	if (sbi->s_mb_history == NULL)
2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
		return;

	if (!(ac->ac_op & sbi->s_mb_history_filter))
		return;

	h.op = ac->ac_op;
	h.pid = current->pid;
	h.ino = ac->ac_inode ? ac->ac_inode->i_ino : 0;
	h.orig = ac->ac_o_ex;
	h.result = ac->ac_b_ex;
	h.flags = ac->ac_flags;
	h.found = ac->ac_found;
	h.groups = ac->ac_groups_scanned;
	h.cr = ac->ac_criteria;
	h.tail = ac->ac_tail;
	h.buddy = ac->ac_buddy;
	h.merged = 0;
	if (ac->ac_op == EXT4_MB_HISTORY_ALLOC) {
		if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
				ac->ac_g_ex.fe_group == ac->ac_b_ex.fe_group)
			h.merged = 1;
		h.goal = ac->ac_g_ex;
		h.result = ac->ac_f_ex;
	}

	spin_lock(&sbi->s_mb_history_lock);
	memcpy(sbi->s_mb_history + sbi->s_mb_history_cur, &h, sizeof(h));
	if (++sbi->s_mb_history_cur >= sbi->s_mb_history_max)
		sbi->s_mb_history_cur = 0;
	spin_unlock(&sbi->s_mb_history_lock);
}

#else
#define ext4_mb_history_release(sb)
#define ext4_mb_history_init(sb)
#endif

2471 2472

/* Create and initialize ext4_group_info data for the given group. */
2473
int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526
			  struct ext4_group_desc *desc)
{
	int i, len;
	int metalen = 0;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_group_info **meta_group_info;

	/*
	 * First check if this group is the first of a reserved block.
	 * If it's true, we have to allocate a new table of pointers
	 * to ext4_group_info structures
	 */
	if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
		metalen = sizeof(*meta_group_info) <<
			EXT4_DESC_PER_BLOCK_BITS(sb);
		meta_group_info = kmalloc(metalen, GFP_KERNEL);
		if (meta_group_info == NULL) {
			printk(KERN_ERR "EXT4-fs: can't allocate mem for a "
			       "buddy group\n");
			goto exit_meta_group_info;
		}
		sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] =
			meta_group_info;
	}

	/*
	 * calculate needed size. if change bb_counters size,
	 * don't forget about ext4_mb_generate_buddy()
	 */
	len = offsetof(typeof(**meta_group_info),
		       bb_counters[sb->s_blocksize_bits + 2]);

	meta_group_info =
		sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)];
	i = group & (EXT4_DESC_PER_BLOCK(sb) - 1);

	meta_group_info[i] = kzalloc(len, GFP_KERNEL);
	if (meta_group_info[i] == NULL) {
		printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
		goto exit_group_info;
	}
	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT,
		&(meta_group_info[i]->bb_state));

	/*
	 * initialize bb_free to be able to skip
	 * empty groups without initialization
	 */
	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
		meta_group_info[i]->bb_free =
			ext4_free_blocks_after_init(sb, group, desc);
	} else {
		meta_group_info[i]->bb_free =
2527
			ext4_free_blks_count(sb, desc);
2528 2529 2530
	}

	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2531
	init_rwsem(&meta_group_info[i]->alloc_sem);
2532
	meta_group_info[i]->bb_free_root.rb_node = NULL;
2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557

#ifdef DOUBLE_CHECK
	{
		struct buffer_head *bh;
		meta_group_info[i]->bb_bitmap =
			kmalloc(sb->s_blocksize, GFP_KERNEL);
		BUG_ON(meta_group_info[i]->bb_bitmap == NULL);
		bh = ext4_read_block_bitmap(sb, group);
		BUG_ON(bh == NULL);
		memcpy(meta_group_info[i]->bb_bitmap, bh->b_data,
			sb->s_blocksize);
		put_bh(bh);
	}
#endif

	return 0;

exit_group_info:
	/* If a meta_group_info table has been allocated, release it now */
	if (group % EXT4_DESC_PER_BLOCK(sb) == 0)
		kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]);
exit_meta_group_info:
	return -ENOMEM;
} /* ext4_mb_add_groupinfo */

2558 2559
static int ext4_mb_init_backend(struct super_block *sb)
{
2560
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2561 2562
	ext4_group_t i;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2563 2564 2565 2566 2567 2568 2569
	struct ext4_super_block *es = sbi->s_es;
	int num_meta_group_infos;
	int num_meta_group_infos_max;
	int array_size;
	struct ext4_group_desc *desc;

	/* This is the number of blocks used by GDT */
2570
	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585
				1) >> EXT4_DESC_PER_BLOCK_BITS(sb);

	/*
	 * This is the total number of blocks used by GDT including
	 * the number of reserved blocks for GDT.
	 * The s_group_info array is allocated with this value
	 * to allow a clean online resize without a complex
	 * manipulation of pointer.
	 * The drawback is the unused memory when no resize
	 * occurs but it's very low in terms of pages
	 * (see comments below)
	 * Need to handle this properly when META_BG resizing is allowed
	 */
	num_meta_group_infos_max = num_meta_group_infos +
				le16_to_cpu(es->s_reserved_gdt_blocks);
2586

2587 2588 2589 2590 2591 2592 2593 2594 2595 2596
	/*
	 * array_size is the size of s_group_info array. We round it
	 * to the next power of two because this approximation is done
	 * internally by kmalloc so we can have some more memory
	 * for free here (e.g. may be used for META_BG resize).
	 */
	array_size = 1;
	while (array_size < sizeof(*sbi->s_group_info) *
	       num_meta_group_infos_max)
		array_size = array_size << 1;
2597 2598 2599
	/* An 8TB filesystem with 64-bit pointers requires a 4096 byte
	 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
	 * So a two level scheme suffices for now. */
2600
	sbi->s_group_info = kmalloc(array_size, GFP_KERNEL);
2601 2602 2603 2604 2605 2606 2607 2608 2609 2610
	if (sbi->s_group_info == NULL) {
		printk(KERN_ERR "EXT4-fs: can't allocate buddy meta group\n");
		return -ENOMEM;
	}
	sbi->s_buddy_cache = new_inode(sb);
	if (sbi->s_buddy_cache == NULL) {
		printk(KERN_ERR "EXT4-fs: can't get new inode\n");
		goto err_freesgi;
	}
	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
2611
	for (i = 0; i < ngroups; i++) {
2612 2613 2614
		desc = ext4_get_group_desc(sb, i, NULL);
		if (desc == NULL) {
			printk(KERN_ERR
2615
				"EXT4-fs: can't read descriptor %u\n", i);
2616 2617
			goto err_freebuddy;
		}
2618 2619
		if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
			goto err_freebuddy;
2620 2621 2622 2623 2624
	}

	return 0;

err_freebuddy:
2625
	while (i-- > 0)
2626 2627
		kfree(ext4_get_group_info(sb, i));
	i = num_meta_group_infos;
2628
	while (i-- > 0)
2629 2630 2631 2632 2633 2634 2635 2636 2637 2638
		kfree(sbi->s_group_info[i]);
	iput(sbi->s_buddy_cache);
err_freesgi:
	kfree(sbi->s_group_info);
	return -ENOMEM;
}

int ext4_mb_init(struct super_block *sb, int needs_recovery)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2639
	unsigned i, j;
2640 2641
	unsigned offset;
	unsigned max;
2642
	int ret;
2643

2644
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
2645 2646 2647 2648 2649

	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_offsets == NULL) {
		return -ENOMEM;
	}
2650

2651
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
2652 2653
	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_maxs == NULL) {
2654
		kfree(sbi->s_mb_offsets);
2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673
		return -ENOMEM;
	}

	/* order 0 is regular bitmap */
	sbi->s_mb_maxs[0] = sb->s_blocksize << 3;
	sbi->s_mb_offsets[0] = 0;

	i = 1;
	offset = 0;
	max = sb->s_blocksize << 2;
	do {
		sbi->s_mb_offsets[i] = offset;
		sbi->s_mb_maxs[i] = max;
		offset += 1 << (sb->s_blocksize_bits - i);
		max = max >> 1;
		i++;
	} while (i <= sb->s_blocksize_bits + 1);

	/* init file for buddy data */
2674 2675
	ret = ext4_mb_init_backend(sb);
	if (ret != 0) {
2676 2677
		kfree(sbi->s_mb_offsets);
		kfree(sbi->s_mb_maxs);
2678
		return ret;
2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
	}

	spin_lock_init(&sbi->s_md_lock);
	spin_lock_init(&sbi->s_bal_lock);

	sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
	sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
	sbi->s_mb_stats = MB_DEFAULT_STATS;
	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
	sbi->s_mb_history_filter = EXT4_MB_HISTORY_DEFAULT;
	sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;

2692
	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2693 2694 2695 2696 2697
	if (sbi->s_locality_groups == NULL) {
		kfree(sbi->s_mb_offsets);
		kfree(sbi->s_mb_maxs);
		return -ENOMEM;
	}
2698
	for_each_possible_cpu(i) {
2699
		struct ext4_locality_group *lg;
2700
		lg = per_cpu_ptr(sbi->s_locality_groups, i);
2701
		mutex_init(&lg->lg_mutex);
2702 2703
		for (j = 0; j < PREALLOC_TB_SIZE; j++)
			INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
2704 2705 2706 2707 2708
		spin_lock_init(&lg->lg_prealloc_lock);
	}

	ext4_mb_history_init(sb);

2709 2710
	if (sbi->s_journal)
		sbi->s_journal->j_commit_callback = release_blocks_on_commit;
2711

2712
	printk(KERN_INFO "EXT4-fs: mballoc enabled\n");
2713 2714 2715
	return 0;
}

2716
/* need to called with the ext4 group lock held */
2717 2718 2719 2720 2721 2722 2723 2724 2725 2726
static void ext4_mb_cleanup_pa(struct ext4_group_info *grp)
{
	struct ext4_prealloc_space *pa;
	struct list_head *cur, *tmp;
	int count = 0;

	list_for_each_safe(cur, tmp, &grp->bb_prealloc_list) {
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		list_del(&pa->pa_group_list);
		count++;
2727
		kmem_cache_free(ext4_pspace_cachep, pa);
2728 2729
	}
	if (count)
2730
		mb_debug(1, "mballoc: %u PAs left\n", count);
2731 2732 2733 2734 2735

}

int ext4_mb_release(struct super_block *sb)
{
2736
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2737 2738 2739 2740 2741 2742
	ext4_group_t i;
	int num_meta_group_infos;
	struct ext4_group_info *grinfo;
	struct ext4_sb_info *sbi = EXT4_SB(sb);

	if (sbi->s_group_info) {
2743
		for (i = 0; i < ngroups; i++) {
2744 2745 2746 2747 2748 2749 2750 2751 2752
			grinfo = ext4_get_group_info(sb, i);
#ifdef DOUBLE_CHECK
			kfree(grinfo->bb_bitmap);
#endif
			ext4_lock_group(sb, i);
			ext4_mb_cleanup_pa(grinfo);
			ext4_unlock_group(sb, i);
			kfree(grinfo);
		}
2753
		num_meta_group_infos = (ngroups +
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787
				EXT4_DESC_PER_BLOCK(sb) - 1) >>
			EXT4_DESC_PER_BLOCK_BITS(sb);
		for (i = 0; i < num_meta_group_infos; i++)
			kfree(sbi->s_group_info[i]);
		kfree(sbi->s_group_info);
	}
	kfree(sbi->s_mb_offsets);
	kfree(sbi->s_mb_maxs);
	if (sbi->s_buddy_cache)
		iput(sbi->s_buddy_cache);
	if (sbi->s_mb_stats) {
		printk(KERN_INFO
		       "EXT4-fs: mballoc: %u blocks %u reqs (%u success)\n",
				atomic_read(&sbi->s_bal_allocated),
				atomic_read(&sbi->s_bal_reqs),
				atomic_read(&sbi->s_bal_success));
		printk(KERN_INFO
		      "EXT4-fs: mballoc: %u extents scanned, %u goal hits, "
				"%u 2^N hits, %u breaks, %u lost\n",
				atomic_read(&sbi->s_bal_ex_scanned),
				atomic_read(&sbi->s_bal_goals),
				atomic_read(&sbi->s_bal_2orders),
				atomic_read(&sbi->s_bal_breaks),
				atomic_read(&sbi->s_mb_lost_chunks));
		printk(KERN_INFO
		       "EXT4-fs: mballoc: %lu generated and it took %Lu\n",
				sbi->s_mb_buddies_generated++,
				sbi->s_mb_generation_time);
		printk(KERN_INFO
		       "EXT4-fs: mballoc: %u preallocated, %u discarded\n",
				atomic_read(&sbi->s_mb_preallocated),
				atomic_read(&sbi->s_mb_discarded));
	}

2788
	free_percpu(sbi->s_locality_groups);
2789 2790 2791 2792 2793
	ext4_mb_history_release(sb);

	return 0;
}

2794 2795 2796 2797 2798
/*
 * This function is called by the jbd2 layer once the commit has finished,
 * so we know we can free the blocks that were released with that commit.
 */
static void release_blocks_on_commit(journal_t *journal, transaction_t *txn)
2799
{
2800
	struct super_block *sb = journal->j_private;
2801
	struct ext4_buddy e4b;
2802 2803 2804
	struct ext4_group_info *db;
	int err, count = 0, count2 = 0;
	struct ext4_free_data *entry;
2805
	ext4_fsblk_t discard_block;
2806
	struct list_head *l, *ltmp;
2807

2808 2809
	list_for_each_safe(l, ltmp, &txn->t_private_list) {
		entry = list_entry(l, struct ext4_free_data, list);
2810

2811
		mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
2812
			 entry->count, entry->group, entry);
2813

2814
		err = ext4_mb_load_buddy(sb, entry->group, &e4b);
2815 2816 2817
		/* we expect to find existing buddy because it's pinned */
		BUG_ON(err != 0);

2818
		db = e4b.bd_info;
2819
		/* there are blocks to put in buddy to make them really free */
2820
		count += entry->count;
2821
		count2++;
2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832
		ext4_lock_group(sb, entry->group);
		/* Take it out of per group rb tree */
		rb_erase(&entry->node, &(db->bb_free_root));
		mb_free_blocks(NULL, &e4b, entry->start_blk, entry->count);

		if (!db->bb_free_root.rb_node) {
			/* No more items in the per group rb tree
			 * balance refcounts from ext4_mb_free_metadata()
			 */
			page_cache_release(e4b.bd_buddy_page);
			page_cache_release(e4b.bd_bitmap_page);
2833
		}
2834
		ext4_unlock_group(sb, entry->group);
2835 2836 2837
		discard_block = (ext4_fsblk_t) entry->group * EXT4_BLOCKS_PER_GROUP(sb)
			+ entry->start_blk
			+ le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
2838 2839
		trace_ext4_discard_blocks(sb, (unsigned long long)discard_block,
					  entry->count);
2840
		sb_issue_discard(sb, discard_block, entry->count);
2841

2842
		kmem_cache_free(ext4_free_ext_cachep, entry);
2843
		ext4_mb_release_desc(&e4b);
2844
	}
2845

2846
	mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
2847 2848
}

2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882
#ifdef CONFIG_EXT4_DEBUG
u8 mb_enable_debug __read_mostly;

static struct dentry *debugfs_dir;
static struct dentry *debugfs_debug;

static void __init ext4_create_debugfs_entry(void)
{
	debugfs_dir = debugfs_create_dir("ext4", NULL);
	if (debugfs_dir)
		debugfs_debug = debugfs_create_u8("mballoc-debug",
						  S_IRUGO | S_IWUSR,
						  debugfs_dir,
						  &mb_enable_debug);
}

static void ext4_remove_debugfs_entry(void)
{
	debugfs_remove(debugfs_debug);
	debugfs_remove(debugfs_dir);
}

#else

static void __init ext4_create_debugfs_entry(void)
{
}

static void ext4_remove_debugfs_entry(void)
{
}

#endif

2883 2884 2885 2886 2887 2888 2889 2890 2891
int __init init_ext4_mballoc(void)
{
	ext4_pspace_cachep =
		kmem_cache_create("ext4_prealloc_space",
				     sizeof(struct ext4_prealloc_space),
				     0, SLAB_RECLAIM_ACCOUNT, NULL);
	if (ext4_pspace_cachep == NULL)
		return -ENOMEM;

2892 2893 2894 2895 2896 2897 2898 2899
	ext4_ac_cachep =
		kmem_cache_create("ext4_alloc_context",
				     sizeof(struct ext4_allocation_context),
				     0, SLAB_RECLAIM_ACCOUNT, NULL);
	if (ext4_ac_cachep == NULL) {
		kmem_cache_destroy(ext4_pspace_cachep);
		return -ENOMEM;
	}
2900 2901 2902 2903 2904 2905 2906 2907 2908 2909

	ext4_free_ext_cachep =
		kmem_cache_create("ext4_free_block_extents",
				     sizeof(struct ext4_free_data),
				     0, SLAB_RECLAIM_ACCOUNT, NULL);
	if (ext4_free_ext_cachep == NULL) {
		kmem_cache_destroy(ext4_pspace_cachep);
		kmem_cache_destroy(ext4_ac_cachep);
		return -ENOMEM;
	}
2910
	ext4_create_debugfs_entry();
2911 2912 2913 2914 2915
	return 0;
}

void exit_ext4_mballoc(void)
{
2916 2917 2918 2919 2920
	/* 
	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
	 * before destroying the slab cache.
	 */
	rcu_barrier();
2921
	kmem_cache_destroy(ext4_pspace_cachep);
2922
	kmem_cache_destroy(ext4_ac_cachep);
2923
	kmem_cache_destroy(ext4_free_ext_cachep);
2924
	ext4_remove_debugfs_entry();
2925 2926 2927 2928 2929 2930 2931
}


/*
 * Check quota and mark choosed space (ac->ac_b_ex) non-free in bitmaps
 * Returns 0 if success or error code
 */
2932 2933
static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2934
				handle_t *handle, unsigned int reserv_blks)
2935 2936 2937 2938 2939 2940 2941 2942
{
	struct buffer_head *bitmap_bh = NULL;
	struct ext4_super_block *es;
	struct ext4_group_desc *gdp;
	struct buffer_head *gdp_bh;
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block;
2943
	int err, len;
2944 2945 2946 2947 2948 2949 2950 2951 2952 2953

	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
	BUG_ON(ac->ac_b_ex.fe_len <= 0);

	sb = ac->ac_sb;
	sbi = EXT4_SB(sb);
	es = sbi->s_es;


	err = -EIO;
2954
	bitmap_bh = ext4_read_block_bitmap(sb, ac->ac_b_ex.fe_group);
2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966
	if (!bitmap_bh)
		goto out_err;

	err = ext4_journal_get_write_access(handle, bitmap_bh);
	if (err)
		goto out_err;

	err = -EIO;
	gdp = ext4_get_group_desc(sb, ac->ac_b_ex.fe_group, &gdp_bh);
	if (!gdp)
		goto out_err;

2967
	ext4_debug("using block group %u(%d)\n", ac->ac_b_ex.fe_group,
2968
			ext4_free_blks_count(sb, gdp));
2969

2970 2971 2972 2973 2974 2975 2976 2977
	err = ext4_journal_get_write_access(handle, gdp_bh);
	if (err)
		goto out_err;

	block = ac->ac_b_ex.fe_group * EXT4_BLOCKS_PER_GROUP(sb)
		+ ac->ac_b_ex.fe_start
		+ le32_to_cpu(es->s_first_data_block);

2978
	len = ac->ac_b_ex.fe_len;
2979
	if (!ext4_data_block_valid(sbi, block, len)) {
2980
		ext4_error(sb, __func__,
2981 2982
			   "Allocating blocks %llu-%llu which overlap "
			   "fs metadata\n", block, block+len);
2983 2984 2985 2986
		/* File system mounted not to panic on error
		 * Fix the bitmap and repeat the block allocation
		 * We leak some of the blocks here.
		 */
2987 2988 2989 2990
		ext4_lock_group(sb, ac->ac_b_ex.fe_group);
		mb_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
			    ac->ac_b_ex.fe_len);
		ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2991
		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2992 2993 2994
		if (!err)
			err = -EAGAIN;
		goto out_err;
2995
	}
2996 2997

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2998 2999 3000 3001 3002 3003 3004 3005 3006
#ifdef AGGRESSIVE_CHECK
	{
		int i;
		for (i = 0; i < ac->ac_b_ex.fe_len; i++) {
			BUG_ON(mb_test_bit(ac->ac_b_ex.fe_start + i,
						bitmap_bh->b_data));
		}
	}
#endif
3007
	mb_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,ac->ac_b_ex.fe_len);
3008 3009
	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
		gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
3010 3011 3012
		ext4_free_blks_set(sb, gdp,
					ext4_free_blocks_after_init(sb,
					ac->ac_b_ex.fe_group, gdp));
3013
	}
3014 3015
	len = ext4_free_blks_count(sb, gdp) - ac->ac_b_ex.fe_len;
	ext4_free_blks_set(sb, gdp, len);
3016
	gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
3017 3018

	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
3019
	percpu_counter_sub(&sbi->s_freeblocks_counter, ac->ac_b_ex.fe_len);
3020
	/*
3021
	 * Now reduce the dirty block count also. Should not go negative
3022
	 */
3023 3024 3025
	if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
		/* release all the reserved blocks if non delalloc */
		percpu_counter_sub(&sbi->s_dirtyblocks_counter, reserv_blks);
3026
	else {
3027 3028
		percpu_counter_sub(&sbi->s_dirtyblocks_counter,
						ac->ac_b_ex.fe_len);
3029 3030 3031
		/* convert reserved quota blocks to real quota blocks */
		vfs_dq_claim_block(ac->ac_inode, ac->ac_b_ex.fe_len);
	}
3032

3033 3034 3035
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi,
							  ac->ac_b_ex.fe_group);
3036 3037
		atomic_sub(ac->ac_b_ex.fe_len,
			   &sbi->s_flex_groups[flex_group].free_blocks);
3038 3039
	}

3040
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
3041 3042
	if (err)
		goto out_err;
3043
	err = ext4_handle_dirty_metadata(handle, NULL, gdp_bh);
3044 3045 3046

out_err:
	sb->s_dirt = 1;
3047
	brelse(bitmap_bh);
3048 3049 3050 3051 3052 3053 3054
	return err;
}

/*
 * here we normalize request for locality group
 * Group request are normalized to s_strip size if we set the same via mount
 * option. If not we set it to s_mb_group_prealloc which can be configured via
T
Theodore Ts'o 已提交
3055
 * /sys/fs/ext4/<partition>/mb_group_prealloc
3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068
 *
 * XXX: should we try to preallocate more than the group has now?
 */
static void ext4_mb_normalize_group_request(struct ext4_allocation_context *ac)
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_locality_group *lg = ac->ac_lg;

	BUG_ON(lg == NULL);
	if (EXT4_SB(sb)->s_stripe)
		ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_stripe;
	else
		ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
3069
	mb_debug(1, "#%u: goal %u blocks for locality group\n",
3070 3071 3072 3073 3074 3075 3076
		current->pid, ac->ac_g_ex.fe_len);
}

/*
 * Normalization means making request better in terms of
 * size and alignment
 */
3077 3078
static noinline_for_stack void
ext4_mb_normalize_request(struct ext4_allocation_context *ac,
3079 3080 3081 3082 3083 3084 3085
				struct ext4_allocation_request *ar)
{
	int bsbits, max;
	ext4_lblk_t end;
	loff_t size, orig_size, start_off;
	ext4_lblk_t start, orig_start;
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
3086
	struct ext4_prealloc_space *pa;
3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115

	/* do normalize only data requests, metadata requests
	   do not need preallocation */
	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
		return;

	/* sometime caller may want exact blocks */
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		return;

	/* caller may indicate that preallocation isn't
	 * required (it's a tail, for example) */
	if (ac->ac_flags & EXT4_MB_HINT_NOPREALLOC)
		return;

	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC) {
		ext4_mb_normalize_group_request(ac);
		return ;
	}

	bsbits = ac->ac_sb->s_blocksize_bits;

	/* first, let's learn actual file size
	 * given current request is allocated */
	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
	size = size << bsbits;
	if (size < i_size_read(ac->ac_inode))
		size = i_size_read(ac->ac_inode);

3116 3117
	/* max size of free chunks */
	max = 2 << bsbits;
3118

3119 3120
#define NRL_CHECK_SIZE(req, size, max, chunk_size)	\
		(req <= (size) || max <= (chunk_size))
3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138

	/* first, try to predict filesize */
	/* XXX: should this table be tunable? */
	start_off = 0;
	if (size <= 16 * 1024) {
		size = 16 * 1024;
	} else if (size <= 32 * 1024) {
		size = 32 * 1024;
	} else if (size <= 64 * 1024) {
		size = 64 * 1024;
	} else if (size <= 128 * 1024) {
		size = 128 * 1024;
	} else if (size <= 256 * 1024) {
		size = 256 * 1024;
	} else if (size <= 512 * 1024) {
		size = 512 * 1024;
	} else if (size <= 1024 * 1024) {
		size = 1024 * 1024;
3139
	} else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
3140
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
3141 3142 3143
						(21 - bsbits)) << 21;
		size = 2 * 1024 * 1024;
	} else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
3144 3145 3146 3147
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
							(22 - bsbits)) << 22;
		size = 4 * 1024 * 1024;
	} else if (NRL_CHECK_SIZE(ac->ac_o_ex.fe_len,
3148
					(8<<20)>>bsbits, max, 8 * 1024)) {
3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
							(23 - bsbits)) << 23;
		size = 8 * 1024 * 1024;
	} else {
		start_off = (loff_t)ac->ac_o_ex.fe_logical << bsbits;
		size	  = ac->ac_o_ex.fe_len << bsbits;
	}
	orig_size = size = size >> bsbits;
	orig_start = start = start_off >> bsbits;

	/* don't cover already allocated blocks in selected range */
	if (ar->pleft && start <= ar->lleft) {
		size -= ar->lleft + 1 - start;
		start = ar->lleft + 1;
	}
	if (ar->pright && start + size - 1 >= ar->lright)
		size -= start + size - ar->lright;

	end = start + size;

	/* check we don't cross already preallocated blocks */
	rcu_read_lock();
3171
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3172
		ext4_lblk_t pa_end;
3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187

		if (pa->pa_deleted)
			continue;
		spin_lock(&pa->pa_lock);
		if (pa->pa_deleted) {
			spin_unlock(&pa->pa_lock);
			continue;
		}

		pa_end = pa->pa_lstart + pa->pa_len;

		/* PA must not overlap original request */
		BUG_ON(!(ac->ac_o_ex.fe_logical >= pa_end ||
			ac->ac_o_ex.fe_logical < pa->pa_lstart));

3188 3189
		/* skip PAs this normalized request doesn't overlap with */
		if (pa->pa_lstart >= end || pa_end <= start) {
3190 3191 3192 3193 3194
			spin_unlock(&pa->pa_lock);
			continue;
		}
		BUG_ON(pa->pa_lstart <= start && pa_end >= end);

3195
		/* adjust start or end to be adjacent to this pa */
3196 3197 3198
		if (pa_end <= ac->ac_o_ex.fe_logical) {
			BUG_ON(pa_end < start);
			start = pa_end;
3199
		} else if (pa->pa_lstart > ac->ac_o_ex.fe_logical) {
3200 3201 3202 3203 3204 3205 3206 3207 3208 3209
			BUG_ON(pa->pa_lstart > end);
			end = pa->pa_lstart;
		}
		spin_unlock(&pa->pa_lock);
	}
	rcu_read_unlock();
	size = end - start;

	/* XXX: extra loop to check we really don't overlap preallocations */
	rcu_read_lock();
3210
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3211
		ext4_lblk_t pa_end;
3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228
		spin_lock(&pa->pa_lock);
		if (pa->pa_deleted == 0) {
			pa_end = pa->pa_lstart + pa->pa_len;
			BUG_ON(!(start >= pa_end || end <= pa->pa_lstart));
		}
		spin_unlock(&pa->pa_lock);
	}
	rcu_read_unlock();

	if (start + size <= ac->ac_o_ex.fe_logical &&
			start > ac->ac_o_ex.fe_logical) {
		printk(KERN_ERR "start %lu, size %lu, fe_logical %lu\n",
			(unsigned long) start, (unsigned long) size,
			(unsigned long) ac->ac_o_ex.fe_logical);
	}
	BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
			start > ac->ac_o_ex.fe_logical);
3229
	BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253

	/* now prepare goal request */

	/* XXX: is it better to align blocks WRT to logical
	 * placement or satisfy big request as is */
	ac->ac_g_ex.fe_logical = start;
	ac->ac_g_ex.fe_len = size;

	/* define goal start in order to merge */
	if (ar->pright && (ar->lright == (start + size))) {
		/* merge to the right */
		ext4_get_group_no_and_offset(ac->ac_sb, ar->pright - size,
						&ac->ac_f_ex.fe_group,
						&ac->ac_f_ex.fe_start);
		ac->ac_flags |= EXT4_MB_HINT_TRY_GOAL;
	}
	if (ar->pleft && (ar->lleft + 1 == start)) {
		/* merge to the left */
		ext4_get_group_no_and_offset(ac->ac_sb, ar->pleft + 1,
						&ac->ac_f_ex.fe_group,
						&ac->ac_f_ex.fe_start);
		ac->ac_flags |= EXT4_MB_HINT_TRY_GOAL;
	}

3254
	mb_debug(1, "goal: %u(was %u) blocks at %u\n", (unsigned) size,
3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302
		(unsigned) orig_size, (unsigned) start);
}

static void ext4_mb_collect_stats(struct ext4_allocation_context *ac)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);

	if (sbi->s_mb_stats && ac->ac_g_ex.fe_len > 1) {
		atomic_inc(&sbi->s_bal_reqs);
		atomic_add(ac->ac_b_ex.fe_len, &sbi->s_bal_allocated);
		if (ac->ac_o_ex.fe_len >= ac->ac_g_ex.fe_len)
			atomic_inc(&sbi->s_bal_success);
		atomic_add(ac->ac_found, &sbi->s_bal_ex_scanned);
		if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
				ac->ac_g_ex.fe_group == ac->ac_b_ex.fe_group)
			atomic_inc(&sbi->s_bal_goals);
		if (ac->ac_found > sbi->s_mb_max_to_scan)
			atomic_inc(&sbi->s_bal_breaks);
	}

	ext4_mb_store_history(ac);
}

/*
 * use blocks preallocated to inode
 */
static void ext4_mb_use_inode_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
	ext4_fsblk_t start;
	ext4_fsblk_t end;
	int len;

	/* found preallocated blocks, use them */
	start = pa->pa_pstart + (ac->ac_o_ex.fe_logical - pa->pa_lstart);
	end = min(pa->pa_pstart + pa->pa_len, start + ac->ac_o_ex.fe_len);
	len = end - start;
	ext4_get_group_no_and_offset(ac->ac_sb, start, &ac->ac_b_ex.fe_group,
					&ac->ac_b_ex.fe_start);
	ac->ac_b_ex.fe_len = len;
	ac->ac_status = AC_STATUS_FOUND;
	ac->ac_pa = pa;

	BUG_ON(start < pa->pa_pstart);
	BUG_ON(start + len > pa->pa_pstart + pa->pa_len);
	BUG_ON(pa->pa_free < len);
	pa->pa_free -= len;

3303
	mb_debug(1, "use %llu/%u from inode pa %p\n", start, len, pa);
3304 3305 3306 3307 3308 3309 3310 3311
}

/*
 * use blocks preallocated to locality group
 */
static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
3312
	unsigned int len = ac->ac_o_ex.fe_len;
3313

3314 3315 3316 3317 3318 3319 3320 3321
	ext4_get_group_no_and_offset(ac->ac_sb, pa->pa_pstart,
					&ac->ac_b_ex.fe_group,
					&ac->ac_b_ex.fe_start);
	ac->ac_b_ex.fe_len = len;
	ac->ac_status = AC_STATUS_FOUND;
	ac->ac_pa = pa;

	/* we don't correct pa_pstart or pa_plen here to avoid
3322
	 * possible race when the group is being loaded concurrently
3323
	 * instead we correct pa later, after blocks are marked
3324 3325
	 * in on-disk bitmap -- see ext4_mb_release_context()
	 * Other CPUs are prevented from allocating from this pa by lg_mutex
3326
	 */
3327
	mb_debug(1, "use %u/%u from group pa %p\n", pa->pa_lstart-len, len, pa);
3328 3329
}

3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358
/*
 * Return the prealloc space that have minimal distance
 * from the goal block. @cpa is the prealloc
 * space that is having currently known minimal distance
 * from the goal block.
 */
static struct ext4_prealloc_space *
ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
			struct ext4_prealloc_space *pa,
			struct ext4_prealloc_space *cpa)
{
	ext4_fsblk_t cur_distance, new_distance;

	if (cpa == NULL) {
		atomic_inc(&pa->pa_count);
		return pa;
	}
	cur_distance = abs(goal_block - cpa->pa_pstart);
	new_distance = abs(goal_block - pa->pa_pstart);

	if (cur_distance < new_distance)
		return cpa;

	/* drop the previous reference */
	atomic_dec(&cpa->pa_count);
	atomic_inc(&pa->pa_count);
	return pa;
}

3359 3360 3361
/*
 * search goal blocks in preallocated space
 */
3362 3363
static noinline_for_stack int
ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
3364
{
3365
	int order, i;
3366 3367
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
	struct ext4_locality_group *lg;
3368 3369
	struct ext4_prealloc_space *pa, *cpa = NULL;
	ext4_fsblk_t goal_block;
3370 3371 3372 3373 3374 3375 3376

	/* only data can be preallocated */
	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
		return 0;

	/* first, try per-file preallocation */
	rcu_read_lock();
3377
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406

		/* all fields in this condition don't change,
		 * so we can skip locking for them */
		if (ac->ac_o_ex.fe_logical < pa->pa_lstart ||
			ac->ac_o_ex.fe_logical >= pa->pa_lstart + pa->pa_len)
			continue;

		/* found preallocated blocks, use them */
		spin_lock(&pa->pa_lock);
		if (pa->pa_deleted == 0 && pa->pa_free) {
			atomic_inc(&pa->pa_count);
			ext4_mb_use_inode_pa(ac, pa);
			spin_unlock(&pa->pa_lock);
			ac->ac_criteria = 10;
			rcu_read_unlock();
			return 1;
		}
		spin_unlock(&pa->pa_lock);
	}
	rcu_read_unlock();

	/* can we use group allocation? */
	if (!(ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC))
		return 0;

	/* inode may have no locality group for some reason */
	lg = ac->ac_lg;
	if (lg == NULL)
		return 0;
3407 3408 3409 3410 3411
	order  = fls(ac->ac_o_ex.fe_len) - 1;
	if (order > PREALLOC_TB_SIZE - 1)
		/* The max size of hash table is PREALLOC_TB_SIZE */
		order = PREALLOC_TB_SIZE - 1;

3412 3413 3414 3415 3416 3417 3418
	goal_block = ac->ac_g_ex.fe_group * EXT4_BLOCKS_PER_GROUP(ac->ac_sb) +
		     ac->ac_g_ex.fe_start +
		     le32_to_cpu(EXT4_SB(ac->ac_sb)->s_es->s_first_data_block);
	/*
	 * search for the prealloc space that is having
	 * minimal distance from the goal block.
	 */
3419 3420 3421 3422 3423 3424 3425
	for (i = order; i < PREALLOC_TB_SIZE; i++) {
		rcu_read_lock();
		list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[i],
					pa_inode_list) {
			spin_lock(&pa->pa_lock);
			if (pa->pa_deleted == 0 &&
					pa->pa_free >= ac->ac_o_ex.fe_len) {
3426 3427 3428

				cpa = ext4_mb_check_group_pa(goal_block,
								pa, cpa);
3429
			}
3430 3431
			spin_unlock(&pa->pa_lock);
		}
3432
		rcu_read_unlock();
3433
	}
3434 3435 3436 3437 3438
	if (cpa) {
		ext4_mb_use_group_pa(ac, cpa);
		ac->ac_criteria = 20;
		return 1;
	}
3439 3440 3441
	return 0;
}

3442 3443 3444 3445
/*
 * the function goes through all block freed in the group
 * but not yet committed and marks them used in in-core bitmap.
 * buddy must be generated from this bitmap
3446
 * Need to be called with the ext4 group lock held
3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459
 */
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group)
{
	struct rb_node *n;
	struct ext4_group_info *grp;
	struct ext4_free_data *entry;

	grp = ext4_get_group_info(sb, group);
	n = rb_first(&(grp->bb_free_root));

	while (n) {
		entry = rb_entry(n, struct ext4_free_data, node);
3460
		mb_set_bits(bitmap, entry->start_blk, entry->count);
3461 3462 3463 3464 3465
		n = rb_next(n);
	}
	return;
}

3466 3467 3468
/*
 * the function goes through all preallocation in this group and marks them
 * used in in-core bitmap. buddy must be generated from this bitmap
3469
 * Need to be called with ext4 group lock held
3470
 */
3471 3472
static noinline_for_stack
void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501
					ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
	struct ext4_prealloc_space *pa;
	struct list_head *cur;
	ext4_group_t groupnr;
	ext4_grpblk_t start;
	int preallocated = 0;
	int count = 0;
	int len;

	/* all form of preallocation discards first load group,
	 * so the only competing code is preallocation use.
	 * we don't need any locking here
	 * notice we do NOT ignore preallocations with pa_deleted
	 * otherwise we could leave used blocks available for
	 * allocation in buddy when concurrent ext4_mb_put_pa()
	 * is dropping preallocation
	 */
	list_for_each(cur, &grp->bb_prealloc_list) {
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		spin_lock(&pa->pa_lock);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart,
					     &groupnr, &start);
		len = pa->pa_len;
		spin_unlock(&pa->pa_lock);
		if (unlikely(len == 0))
			continue;
		BUG_ON(groupnr != group);
3502
		mb_set_bits(bitmap, start, len);
3503 3504 3505
		preallocated += len;
		count++;
	}
3506
	mb_debug(1, "prellocated %u for group %u\n", preallocated, group);
3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522
}

static void ext4_mb_pa_callback(struct rcu_head *head)
{
	struct ext4_prealloc_space *pa;
	pa = container_of(head, struct ext4_prealloc_space, u.pa_rcu);
	kmem_cache_free(ext4_pspace_cachep, pa);
}

/*
 * drops a reference to preallocated space descriptor
 * if this was the last reference and the space is consumed
 */
static void ext4_mb_put_pa(struct ext4_allocation_context *ac,
			struct super_block *sb, struct ext4_prealloc_space *pa)
{
3523
	ext4_group_t grp;
3524
	ext4_fsblk_t grp_blk;
3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538

	if (!atomic_dec_and_test(&pa->pa_count) || pa->pa_free != 0)
		return;

	/* in this short window concurrent discard can set pa_deleted */
	spin_lock(&pa->pa_lock);
	if (pa->pa_deleted == 1) {
		spin_unlock(&pa->pa_lock);
		return;
	}

	pa->pa_deleted = 1;
	spin_unlock(&pa->pa_lock);

3539
	grp_blk = pa->pa_pstart;
3540 3541 3542 3543 3544
	/* 
	 * If doing group-based preallocation, pa_pstart may be in the
	 * next group when pa is used up
	 */
	if (pa->pa_type == MB_GROUP_PA)
3545 3546 3547
		grp_blk--;

	ext4_get_group_no_and_offset(sb, grp_blk, &grp, NULL);
3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576

	/*
	 * possible race:
	 *
	 *  P1 (buddy init)			P2 (regular allocation)
	 *					find block B in PA
	 *  copy on-disk bitmap to buddy
	 *  					mark B in on-disk bitmap
	 *					drop PA from group
	 *  mark all PAs in buddy
	 *
	 * thus, P1 initializes buddy with B available. to prevent this
	 * we make "copy" and "mark all PAs" atomic and serialize "drop PA"
	 * against that pair
	 */
	ext4_lock_group(sb, grp);
	list_del(&pa->pa_group_list);
	ext4_unlock_group(sb, grp);

	spin_lock(pa->pa_obj_lock);
	list_del_rcu(&pa->pa_inode_list);
	spin_unlock(pa->pa_obj_lock);

	call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
}

/*
 * creates new preallocated space for given inode
 */
3577 3578
static noinline_for_stack int
ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_prealloc_space *pa;
	struct ext4_group_info *grp;
	struct ext4_inode_info *ei;

	/* preallocate only when found space is larger then requested */
	BUG_ON(ac->ac_o_ex.fe_len >= ac->ac_b_ex.fe_len);
	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
	BUG_ON(!S_ISREG(ac->ac_inode->i_mode));

	pa = kmem_cache_alloc(ext4_pspace_cachep, GFP_NOFS);
	if (pa == NULL)
		return -ENOMEM;

	if (ac->ac_b_ex.fe_len < ac->ac_g_ex.fe_len) {
		int winl;
		int wins;
		int win;
		int offs;

		/* we can't allocate as much as normalizer wants.
		 * so, found space must get proper lstart
		 * to cover original request */
		BUG_ON(ac->ac_g_ex.fe_logical > ac->ac_o_ex.fe_logical);
		BUG_ON(ac->ac_g_ex.fe_len < ac->ac_o_ex.fe_len);

		/* we're limited by original request in that
		 * logical block must be covered any way
		 * winl is window we can move our chunk within */
		winl = ac->ac_o_ex.fe_logical - ac->ac_g_ex.fe_logical;

		/* also, we should cover whole original request */
		wins = ac->ac_b_ex.fe_len - ac->ac_o_ex.fe_len;

		/* the smallest one defines real window */
		win = min(winl, wins);

		offs = ac->ac_o_ex.fe_logical % ac->ac_b_ex.fe_len;
		if (offs && offs < win)
			win = offs;

		ac->ac_b_ex.fe_logical = ac->ac_o_ex.fe_logical - win;
		BUG_ON(ac->ac_o_ex.fe_logical < ac->ac_b_ex.fe_logical);
		BUG_ON(ac->ac_o_ex.fe_len > ac->ac_b_ex.fe_len);
	}

	/* preallocation can change ac_b_ex, thus we store actually
	 * allocated blocks for history */
	ac->ac_f_ex = ac->ac_b_ex;

	pa->pa_lstart = ac->ac_b_ex.fe_logical;
	pa->pa_pstart = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
	pa->pa_len = ac->ac_b_ex.fe_len;
	pa->pa_free = pa->pa_len;
	atomic_set(&pa->pa_count, 1);
	spin_lock_init(&pa->pa_lock);
3636 3637
	INIT_LIST_HEAD(&pa->pa_inode_list);
	INIT_LIST_HEAD(&pa->pa_group_list);
3638
	pa->pa_deleted = 0;
3639
	pa->pa_type = MB_INODE_PA;
3640

3641
	mb_debug(1, "new inode pa %p: %llu/%u for %u\n", pa,
3642
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
3643
	trace_ext4_mb_new_inode_pa(ac, pa);
3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667

	ext4_mb_use_inode_pa(ac, pa);
	atomic_add(pa->pa_free, &EXT4_SB(sb)->s_mb_preallocated);

	ei = EXT4_I(ac->ac_inode);
	grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);

	pa->pa_obj_lock = &ei->i_prealloc_lock;
	pa->pa_inode = ac->ac_inode;

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
	list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);

	spin_lock(pa->pa_obj_lock);
	list_add_rcu(&pa->pa_inode_list, &ei->i_prealloc_list);
	spin_unlock(pa->pa_obj_lock);

	return 0;
}

/*
 * creates new preallocated space for locality group inodes belongs to
 */
3668 3669
static noinline_for_stack int
ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_locality_group *lg;
	struct ext4_prealloc_space *pa;
	struct ext4_group_info *grp;

	/* preallocate only when found space is larger then requested */
	BUG_ON(ac->ac_o_ex.fe_len >= ac->ac_b_ex.fe_len);
	BUG_ON(ac->ac_status != AC_STATUS_FOUND);
	BUG_ON(!S_ISREG(ac->ac_inode->i_mode));

	BUG_ON(ext4_pspace_cachep == NULL);
	pa = kmem_cache_alloc(ext4_pspace_cachep, GFP_NOFS);
	if (pa == NULL)
		return -ENOMEM;

	/* preallocation can change ac_b_ex, thus we store actually
	 * allocated blocks for history */
	ac->ac_f_ex = ac->ac_b_ex;

	pa->pa_pstart = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
	pa->pa_lstart = pa->pa_pstart;
	pa->pa_len = ac->ac_b_ex.fe_len;
	pa->pa_free = pa->pa_len;
	atomic_set(&pa->pa_count, 1);
	spin_lock_init(&pa->pa_lock);
3696
	INIT_LIST_HEAD(&pa->pa_inode_list);
3697
	INIT_LIST_HEAD(&pa->pa_group_list);
3698
	pa->pa_deleted = 0;
3699
	pa->pa_type = MB_GROUP_PA;
3700

3701
	mb_debug(1, "new group pa %p: %llu/%u for %u\n", pa,
3702 3703
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
	trace_ext4_mb_new_group_pa(ac, pa);
3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718

	ext4_mb_use_group_pa(ac, pa);
	atomic_add(pa->pa_free, &EXT4_SB(sb)->s_mb_preallocated);

	grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);
	lg = ac->ac_lg;
	BUG_ON(lg == NULL);

	pa->pa_obj_lock = &lg->lg_prealloc_lock;
	pa->pa_inode = NULL;

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
	list_add(&pa->pa_group_list, &grp->bb_prealloc_list);
	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);

3719 3720 3721 3722
	/*
	 * We will later add the new pa to the right bucket
	 * after updating the pa_free in ext4_mb_release_context
	 */
3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744
	return 0;
}

static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac)
{
	int err;

	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
		err = ext4_mb_new_group_pa(ac);
	else
		err = ext4_mb_new_inode_pa(ac);
	return err;
}

/*
 * finds all unused blocks in on-disk bitmap, frees them in
 * in-core bitmap and buddy.
 * @pa must be unlinked from inode and group lists, so that
 * nobody else can find/use it.
 * the caller MUST hold group/inode locks.
 * TODO: optimize the case when there are no in-core structures yet
 */
3745 3746
static noinline_for_stack int
ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3747 3748
			struct ext4_prealloc_space *pa,
			struct ext4_allocation_context *ac)
3749 3750 3751
{
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
3752 3753
	unsigned int end;
	unsigned int next;
3754 3755
	ext4_group_t group;
	ext4_grpblk_t bit;
3756
	unsigned long long grp_blk_start;
3757 3758 3759 3760 3761 3762
	sector_t start;
	int err = 0;
	int free = 0;

	BUG_ON(pa->pa_deleted == 0);
	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
3763
	grp_blk_start = pa->pa_pstart - bit;
3764 3765 3766
	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
	end = bit + pa->pa_len;

3767 3768 3769 3770 3771
	if (ac) {
		ac->ac_sb = sb;
		ac->ac_inode = pa->pa_inode;
		ac->ac_op = EXT4_MB_HISTORY_DISCARD;
	}
3772 3773

	while (bit < end) {
3774
		bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
3775 3776
		if (bit >= end)
			break;
3777
		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
3778 3779
		start = group * EXT4_BLOCKS_PER_GROUP(sb) + bit +
				le32_to_cpu(sbi->s_es->s_first_data_block);
3780
		mb_debug(1, "    free preallocated %u/%u in group %u\n",
3781 3782 3783 3784
				(unsigned) start, (unsigned) next - bit,
				(unsigned) group);
		free += next - bit;

3785 3786 3787 3788 3789 3790 3791
		if (ac) {
			ac->ac_b_ex.fe_group = group;
			ac->ac_b_ex.fe_start = bit;
			ac->ac_b_ex.fe_len = next - bit;
			ac->ac_b_ex.fe_logical = 0;
			ext4_mb_store_history(ac);
		}
3792

3793 3794
		trace_ext4_mb_release_inode_pa(ac, pa, grp_blk_start + bit,
					       next - bit);
3795 3796 3797 3798
		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
		bit = next + 1;
	}
	if (free != pa->pa_free) {
3799
		printk(KERN_CRIT "pa %p: logic %lu, phys. %lu, len %lu\n",
3800 3801 3802
			pa, (unsigned long) pa->pa_lstart,
			(unsigned long) pa->pa_pstart,
			(unsigned long) pa->pa_len);
3803 3804 3805
		ext4_grp_locked_error(sb, group,
					__func__, "free %u, pa_free %u",
					free, pa->pa_free);
3806 3807 3808 3809
		/*
		 * pa is already deleted so we use the value obtained
		 * from the bitmap and continue.
		 */
3810 3811 3812 3813 3814 3815
	}
	atomic_add(free, &sbi->s_mb_discarded);

	return err;
}

3816 3817
static noinline_for_stack int
ext4_mb_release_group_pa(struct ext4_buddy *e4b,
3818 3819
				struct ext4_prealloc_space *pa,
				struct ext4_allocation_context *ac)
3820 3821 3822 3823 3824
{
	struct super_block *sb = e4b->bd_sb;
	ext4_group_t group;
	ext4_grpblk_t bit;

3825 3826
	if (ac)
		ac->ac_op = EXT4_MB_HISTORY_DISCARD;
3827

3828
	trace_ext4_mb_release_group_pa(ac, pa);
3829 3830 3831 3832 3833 3834
	BUG_ON(pa->pa_deleted == 0);
	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
	mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len);
	atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded);

3835 3836 3837 3838 3839 3840 3841 3842 3843
	if (ac) {
		ac->ac_sb = sb;
		ac->ac_inode = NULL;
		ac->ac_b_ex.fe_group = group;
		ac->ac_b_ex.fe_start = bit;
		ac->ac_b_ex.fe_len = pa->pa_len;
		ac->ac_b_ex.fe_logical = 0;
		ext4_mb_store_history(ac);
	}
3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856

	return 0;
}

/*
 * releases all preallocations in given group
 *
 * first, we need to decide discard policy:
 * - when do we discard
 *   1) ENOSPC
 * - how many do we discard
 *   1) how many requested
 */
3857 3858
static noinline_for_stack int
ext4_mb_discard_group_preallocations(struct super_block *sb,
3859 3860 3861 3862 3863
					ext4_group_t group, int needed)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
	struct buffer_head *bitmap_bh = NULL;
	struct ext4_prealloc_space *pa, *tmp;
3864
	struct ext4_allocation_context *ac;
3865 3866 3867 3868 3869 3870
	struct list_head list;
	struct ext4_buddy e4b;
	int err;
	int busy = 0;
	int free = 0;

3871
	mb_debug(1, "discard preallocation for group %u\n", group);
3872 3873 3874 3875

	if (list_empty(&grp->bb_prealloc_list))
		return 0;

3876
	bitmap_bh = ext4_read_block_bitmap(sb, group);
3877
	if (bitmap_bh == NULL) {
3878
		ext4_error(sb, __func__, "Error in reading block "
3879
				"bitmap for %u", group);
3880
		return 0;
3881 3882 3883
	}

	err = ext4_mb_load_buddy(sb, group, &e4b);
3884 3885
	if (err) {
		ext4_error(sb, __func__, "Error in loading buddy "
3886
				"information for %u", group);
3887 3888 3889
		put_bh(bitmap_bh);
		return 0;
	}
3890 3891 3892 3893 3894

	if (needed == 0)
		needed = EXT4_BLOCKS_PER_GROUP(sb) + 1;

	INIT_LIST_HEAD(&list);
3895
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
3896 3897
	if (ac)
		ac->ac_sb = sb;
3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950
repeat:
	ext4_lock_group(sb, group);
	list_for_each_entry_safe(pa, tmp,
				&grp->bb_prealloc_list, pa_group_list) {
		spin_lock(&pa->pa_lock);
		if (atomic_read(&pa->pa_count)) {
			spin_unlock(&pa->pa_lock);
			busy = 1;
			continue;
		}
		if (pa->pa_deleted) {
			spin_unlock(&pa->pa_lock);
			continue;
		}

		/* seems this one can be freed ... */
		pa->pa_deleted = 1;

		/* we can trust pa_free ... */
		free += pa->pa_free;

		spin_unlock(&pa->pa_lock);

		list_del(&pa->pa_group_list);
		list_add(&pa->u.pa_tmp_list, &list);
	}

	/* if we still need more blocks and some PAs were used, try again */
	if (free < needed && busy) {
		busy = 0;
		ext4_unlock_group(sb, group);
		/*
		 * Yield the CPU here so that we don't get soft lockup
		 * in non preempt case.
		 */
		yield();
		goto repeat;
	}

	/* found anything to free? */
	if (list_empty(&list)) {
		BUG_ON(free != 0);
		goto out;
	}

	/* now free all selected PAs */
	list_for_each_entry_safe(pa, tmp, &list, u.pa_tmp_list) {

		/* remove from object (inode or locality group) */
		spin_lock(pa->pa_obj_lock);
		list_del_rcu(&pa->pa_inode_list);
		spin_unlock(pa->pa_obj_lock);

3951
		if (pa->pa_type == MB_GROUP_PA)
3952
			ext4_mb_release_group_pa(&e4b, pa, ac);
3953
		else
3954
			ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac);
3955 3956 3957 3958 3959 3960 3961

		list_del(&pa->u.pa_tmp_list);
		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
	}

out:
	ext4_unlock_group(sb, group);
3962 3963
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977
	ext4_mb_release_desc(&e4b);
	put_bh(bitmap_bh);
	return free;
}

/*
 * releases all non-used preallocated blocks for given inode
 *
 * It's important to discard preallocations under i_data_sem
 * We don't want another block to be served from the prealloc
 * space when we are discarding the inode prealloc space.
 *
 * FIXME!! Make sure it is valid at all the call sites
 */
3978
void ext4_discard_preallocations(struct inode *inode)
3979 3980 3981 3982 3983
{
	struct ext4_inode_info *ei = EXT4_I(inode);
	struct super_block *sb = inode->i_sb;
	struct buffer_head *bitmap_bh = NULL;
	struct ext4_prealloc_space *pa, *tmp;
3984
	struct ext4_allocation_context *ac;
3985 3986 3987 3988 3989
	ext4_group_t group = 0;
	struct list_head list;
	struct ext4_buddy e4b;
	int err;

3990
	if (!S_ISREG(inode->i_mode)) {
3991 3992 3993 3994
		/*BUG_ON(!list_empty(&ei->i_prealloc_list));*/
		return;
	}

3995
	mb_debug(1, "discard preallocation for inode %lu\n", inode->i_ino);
3996
	trace_ext4_discard_preallocations(inode);
3997 3998 3999

	INIT_LIST_HEAD(&list);

4000
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4001 4002 4003 4004
	if (ac) {
		ac->ac_sb = sb;
		ac->ac_inode = inode;
	}
4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053
repeat:
	/* first, collect all pa's in the inode */
	spin_lock(&ei->i_prealloc_lock);
	while (!list_empty(&ei->i_prealloc_list)) {
		pa = list_entry(ei->i_prealloc_list.next,
				struct ext4_prealloc_space, pa_inode_list);
		BUG_ON(pa->pa_obj_lock != &ei->i_prealloc_lock);
		spin_lock(&pa->pa_lock);
		if (atomic_read(&pa->pa_count)) {
			/* this shouldn't happen often - nobody should
			 * use preallocation while we're discarding it */
			spin_unlock(&pa->pa_lock);
			spin_unlock(&ei->i_prealloc_lock);
			printk(KERN_ERR "uh-oh! used pa while discarding\n");
			WARN_ON(1);
			schedule_timeout_uninterruptible(HZ);
			goto repeat;

		}
		if (pa->pa_deleted == 0) {
			pa->pa_deleted = 1;
			spin_unlock(&pa->pa_lock);
			list_del_rcu(&pa->pa_inode_list);
			list_add(&pa->u.pa_tmp_list, &list);
			continue;
		}

		/* someone is deleting pa right now */
		spin_unlock(&pa->pa_lock);
		spin_unlock(&ei->i_prealloc_lock);

		/* we have to wait here because pa_deleted
		 * doesn't mean pa is already unlinked from
		 * the list. as we might be called from
		 * ->clear_inode() the inode will get freed
		 * and concurrent thread which is unlinking
		 * pa from inode's list may access already
		 * freed memory, bad-bad-bad */

		/* XXX: if this happens too often, we can
		 * add a flag to force wait only in case
		 * of ->clear_inode(), but not in case of
		 * regular truncate */
		schedule_timeout_uninterruptible(HZ);
		goto repeat;
	}
	spin_unlock(&ei->i_prealloc_lock);

	list_for_each_entry_safe(pa, tmp, &list, u.pa_tmp_list) {
4054
		BUG_ON(pa->pa_type != MB_INODE_PA);
4055 4056 4057
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, NULL);

		err = ext4_mb_load_buddy(sb, group, &e4b);
4058 4059
		if (err) {
			ext4_error(sb, __func__, "Error in loading buddy "
4060
					"information for %u", group);
4061 4062
			continue;
		}
4063

4064
		bitmap_bh = ext4_read_block_bitmap(sb, group);
4065
		if (bitmap_bh == NULL) {
4066
			ext4_error(sb, __func__, "Error in reading block "
4067
					"bitmap for %u", group);
4068
			ext4_mb_release_desc(&e4b);
4069
			continue;
4070 4071 4072 4073
		}

		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
4074
		ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac);
4075 4076 4077 4078 4079 4080 4081 4082
		ext4_unlock_group(sb, group);

		ext4_mb_release_desc(&e4b);
		put_bh(bitmap_bh);

		list_del(&pa->u.pa_tmp_list);
		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
	}
4083 4084
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099
}

/*
 * finds all preallocated spaces and return blocks being freed to them
 * if preallocated space becomes full (no block is used from the space)
 * then the function frees space in buddy
 * XXX: at the moment, truncate (which is the only way to free blocks)
 * discards all preallocations
 */
static void ext4_mb_return_to_preallocation(struct inode *inode,
					struct ext4_buddy *e4b,
					sector_t block, int count)
{
	BUG_ON(!list_empty(&EXT4_I(inode)->i_prealloc_list));
}
4100
#ifdef CONFIG_EXT4_DEBUG
4101 4102 4103
static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
{
	struct super_block *sb = ac->ac_sb;
4104
	ext4_group_t ngroups, i;
4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127

	printk(KERN_ERR "EXT4-fs: Can't allocate:"
			" Allocation context details:\n");
	printk(KERN_ERR "EXT4-fs: status %d flags %d\n",
			ac->ac_status, ac->ac_flags);
	printk(KERN_ERR "EXT4-fs: orig %lu/%lu/%lu@%lu, goal %lu/%lu/%lu@%lu, "
			"best %lu/%lu/%lu@%lu cr %d\n",
			(unsigned long)ac->ac_o_ex.fe_group,
			(unsigned long)ac->ac_o_ex.fe_start,
			(unsigned long)ac->ac_o_ex.fe_len,
			(unsigned long)ac->ac_o_ex.fe_logical,
			(unsigned long)ac->ac_g_ex.fe_group,
			(unsigned long)ac->ac_g_ex.fe_start,
			(unsigned long)ac->ac_g_ex.fe_len,
			(unsigned long)ac->ac_g_ex.fe_logical,
			(unsigned long)ac->ac_b_ex.fe_group,
			(unsigned long)ac->ac_b_ex.fe_start,
			(unsigned long)ac->ac_b_ex.fe_len,
			(unsigned long)ac->ac_b_ex.fe_logical,
			(int)ac->ac_criteria);
	printk(KERN_ERR "EXT4-fs: %lu scanned, %d found\n", ac->ac_ex_scanned,
		ac->ac_found);
	printk(KERN_ERR "EXT4-fs: groups: \n");
4128 4129
	ngroups = ext4_get_groups_count(sb);
	for (i = 0; i < ngroups; i++) {
4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141
		struct ext4_group_info *grp = ext4_get_group_info(sb, i);
		struct ext4_prealloc_space *pa;
		ext4_grpblk_t start;
		struct list_head *cur;
		ext4_lock_group(sb, i);
		list_for_each(cur, &grp->bb_prealloc_list) {
			pa = list_entry(cur, struct ext4_prealloc_space,
					pa_group_list);
			spin_lock(&pa->pa_lock);
			ext4_get_group_no_and_offset(sb, pa->pa_pstart,
						     NULL, &start);
			spin_unlock(&pa->pa_lock);
4142 4143
			printk(KERN_ERR "PA:%u:%d:%u \n", i,
			       start, pa->pa_len);
4144
		}
4145
		ext4_unlock_group(sb, i);
4146 4147 4148

		if (grp->bb_free == 0)
			continue;
4149
		printk(KERN_ERR "%u: %d/%d \n",
4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165
		       i, grp->bb_free, grp->bb_fragments);
	}
	printk(KERN_ERR "\n");
}
#else
static inline void ext4_mb_show_ac(struct ext4_allocation_context *ac)
{
	return;
}
#endif

/*
 * We use locality group preallocation for small size file. The size of the
 * file is determined by the current size or the resulting size after
 * allocation which ever is larger
 *
T
Theodore Ts'o 已提交
4166
 * One can tune this size via /sys/fs/ext4/<partition>/mb_stream_req
4167 4168 4169 4170 4171 4172 4173 4174 4175 4176
 */
static void ext4_mb_group_or_file(struct ext4_allocation_context *ac)
{
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
	int bsbits = ac->ac_sb->s_blocksize_bits;
	loff_t size, isize;

	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
		return;

4177 4178 4179
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		return;

4180
	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
4181 4182
	isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1)
		>> bsbits;
4183 4184
	size = max(size, isize);

4185 4186 4187 4188 4189 4190 4191
	if ((size == isize) &&
	    !ext4_fs_is_busy(sbi) &&
	    (atomic_read(&ac->ac_inode->i_writecount) == 0)) {
		ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC;
		return;
	}

4192
	/* don't use group allocation for large files */
4193 4194
	if (size >= sbi->s_mb_stream_request) {
		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
4195
		return;
4196
	}
4197 4198 4199 4200 4201 4202 4203

	BUG_ON(ac->ac_lg != NULL);
	/*
	 * locality group prealloc space are per cpu. The reason for having
	 * per cpu locality group is to reduce the contention between block
	 * request from multiple CPUs.
	 */
4204
	ac->ac_lg = per_cpu_ptr(sbi->s_locality_groups, raw_smp_processor_id());
4205 4206 4207 4208 4209 4210 4211 4212

	/* we're going to use group allocation */
	ac->ac_flags |= EXT4_MB_HINT_GROUP_ALLOC;

	/* serialize all allocations in the group */
	mutex_lock(&ac->ac_lg->lg_mutex);
}

4213 4214
static noinline_for_stack int
ext4_mb_initialize_context(struct ext4_allocation_context *ac,
4215 4216 4217 4218 4219 4220
				struct ext4_allocation_request *ar)
{
	struct super_block *sb = ar->inode->i_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_super_block *es = sbi->s_es;
	ext4_group_t group;
4221 4222
	unsigned int len;
	ext4_fsblk_t goal;
4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239
	ext4_grpblk_t block;

	/* we can't allocate > group size */
	len = ar->len;

	/* just a dirty hack to filter too big requests  */
	if (len >= EXT4_BLOCKS_PER_GROUP(sb) - 10)
		len = EXT4_BLOCKS_PER_GROUP(sb) - 10;

	/* start searching from the goal */
	goal = ar->goal;
	if (goal < le32_to_cpu(es->s_first_data_block) ||
			goal >= ext4_blocks_count(es))
		goal = le32_to_cpu(es->s_first_data_block);
	ext4_get_group_no_and_offset(sb, goal, &group, &block);

	/* set up allocation goals */
4240
	memset(ac, 0, sizeof(struct ext4_allocation_context));
4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258
	ac->ac_b_ex.fe_logical = ar->logical;
	ac->ac_status = AC_STATUS_CONTINUE;
	ac->ac_sb = sb;
	ac->ac_inode = ar->inode;
	ac->ac_o_ex.fe_logical = ar->logical;
	ac->ac_o_ex.fe_group = group;
	ac->ac_o_ex.fe_start = block;
	ac->ac_o_ex.fe_len = len;
	ac->ac_g_ex.fe_logical = ar->logical;
	ac->ac_g_ex.fe_group = group;
	ac->ac_g_ex.fe_start = block;
	ac->ac_g_ex.fe_len = len;
	ac->ac_flags = ar->flags;

	/* we have to define context: we'll we work with a file or
	 * locality group. this is a policy, actually */
	ext4_mb_group_or_file(ac);

4259
	mb_debug(1, "init ac: %u blocks @ %u, goal %u, flags %x, 2^%d, "
4260 4261 4262 4263 4264 4265 4266 4267 4268 4269
			"left: %u/%u, right %u/%u to %swritable\n",
			(unsigned) ar->len, (unsigned) ar->logical,
			(unsigned) ar->goal, ac->ac_flags, ac->ac_2order,
			(unsigned) ar->lleft, (unsigned) ar->pleft,
			(unsigned) ar->lright, (unsigned) ar->pright,
			atomic_read(&ar->inode->i_writecount) ? "" : "non-");
	return 0;

}

4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280
static noinline_for_stack void
ext4_mb_discard_lg_preallocations(struct super_block *sb,
					struct ext4_locality_group *lg,
					int order, int total_entries)
{
	ext4_group_t group = 0;
	struct ext4_buddy e4b;
	struct list_head discard_list;
	struct ext4_prealloc_space *pa, *tmp;
	struct ext4_allocation_context *ac;

4281
	mb_debug(1, "discard locality group preallocation\n");
4282 4283 4284

	INIT_LIST_HEAD(&discard_list);
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4285 4286
	if (ac)
		ac->ac_sb = sb;
4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305

	spin_lock(&lg->lg_prealloc_lock);
	list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[order],
						pa_inode_list) {
		spin_lock(&pa->pa_lock);
		if (atomic_read(&pa->pa_count)) {
			/*
			 * This is the pa that we just used
			 * for block allocation. So don't
			 * free that
			 */
			spin_unlock(&pa->pa_lock);
			continue;
		}
		if (pa->pa_deleted) {
			spin_unlock(&pa->pa_lock);
			continue;
		}
		/* only lg prealloc space */
4306
		BUG_ON(pa->pa_type != MB_GROUP_PA);
4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332

		/* seems this one can be freed ... */
		pa->pa_deleted = 1;
		spin_unlock(&pa->pa_lock);

		list_del_rcu(&pa->pa_inode_list);
		list_add(&pa->u.pa_tmp_list, &discard_list);

		total_entries--;
		if (total_entries <= 5) {
			/*
			 * we want to keep only 5 entries
			 * allowing it to grow to 8. This
			 * mak sure we don't call discard
			 * soon for this list.
			 */
			break;
		}
	}
	spin_unlock(&lg->lg_prealloc_lock);

	list_for_each_entry_safe(pa, tmp, &discard_list, u.pa_tmp_list) {

		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, NULL);
		if (ext4_mb_load_buddy(sb, group, &e4b)) {
			ext4_error(sb, __func__, "Error in loading buddy "
4333
					"information for %u", group);
4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374
			continue;
		}
		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
		ext4_mb_release_group_pa(&e4b, pa, ac);
		ext4_unlock_group(sb, group);

		ext4_mb_release_desc(&e4b);
		list_del(&pa->u.pa_tmp_list);
		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
	}
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
}

/*
 * We have incremented pa_count. So it cannot be freed at this
 * point. Also we hold lg_mutex. So no parallel allocation is
 * possible from this lg. That means pa_free cannot be updated.
 *
 * A parallel ext4_mb_discard_group_preallocations is possible.
 * which can cause the lg_prealloc_list to be updated.
 */

static void ext4_mb_add_n_trim(struct ext4_allocation_context *ac)
{
	int order, added = 0, lg_prealloc_count = 1;
	struct super_block *sb = ac->ac_sb;
	struct ext4_locality_group *lg = ac->ac_lg;
	struct ext4_prealloc_space *tmp_pa, *pa = ac->ac_pa;

	order = fls(pa->pa_free) - 1;
	if (order > PREALLOC_TB_SIZE - 1)
		/* The max size of hash table is PREALLOC_TB_SIZE */
		order = PREALLOC_TB_SIZE - 1;
	/* Add the prealloc space to lg */
	rcu_read_lock();
	list_for_each_entry_rcu(tmp_pa, &lg->lg_prealloc_list[order],
						pa_inode_list) {
		spin_lock(&tmp_pa->pa_lock);
		if (tmp_pa->pa_deleted) {
4375
			spin_unlock(&tmp_pa->pa_lock);
4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404
			continue;
		}
		if (!added && pa->pa_free < tmp_pa->pa_free) {
			/* Add to the tail of the previous entry */
			list_add_tail_rcu(&pa->pa_inode_list,
						&tmp_pa->pa_inode_list);
			added = 1;
			/*
			 * we want to count the total
			 * number of entries in the list
			 */
		}
		spin_unlock(&tmp_pa->pa_lock);
		lg_prealloc_count++;
	}
	if (!added)
		list_add_tail_rcu(&pa->pa_inode_list,
					&lg->lg_prealloc_list[order]);
	rcu_read_unlock();

	/* Now trim the list to be not more than 8 elements */
	if (lg_prealloc_count > 8) {
		ext4_mb_discard_lg_preallocations(sb, lg,
						order, lg_prealloc_count);
		return;
	}
	return ;
}

4405 4406 4407 4408 4409
/*
 * release all resource we used in allocation
 */
static int ext4_mb_release_context(struct ext4_allocation_context *ac)
{
4410 4411
	struct ext4_prealloc_space *pa = ac->ac_pa;
	if (pa) {
4412
		if (pa->pa_type == MB_GROUP_PA) {
4413
			/* see comment in ext4_mb_use_group_pa() */
4414 4415 4416 4417 4418 4419
			spin_lock(&pa->pa_lock);
			pa->pa_pstart += ac->ac_b_ex.fe_len;
			pa->pa_lstart += ac->ac_b_ex.fe_len;
			pa->pa_free -= ac->ac_b_ex.fe_len;
			pa->pa_len -= ac->ac_b_ex.fe_len;
			spin_unlock(&pa->pa_lock);
4420 4421
		}
	}
4422 4423
	if (ac->alloc_semp)
		up_read(ac->alloc_semp);
A
Aneesh Kumar K.V 已提交
4424 4425 4426 4427 4428 4429 4430 4431
	if (pa) {
		/*
		 * We want to add the pa to the right bucket.
		 * Remove it from the list and while adding
		 * make sure the list to which we are adding
		 * doesn't grow big.  We need to release
		 * alloc_semp before calling ext4_mb_add_n_trim()
		 */
4432
		if ((pa->pa_type == MB_GROUP_PA) && likely(pa->pa_free)) {
A
Aneesh Kumar K.V 已提交
4433 4434 4435 4436 4437 4438 4439
			spin_lock(pa->pa_obj_lock);
			list_del_rcu(&pa->pa_inode_list);
			spin_unlock(pa->pa_obj_lock);
			ext4_mb_add_n_trim(ac);
		}
		ext4_mb_put_pa(ac, ac->ac_sb, pa);
	}
4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451
	if (ac->ac_bitmap_page)
		page_cache_release(ac->ac_bitmap_page);
	if (ac->ac_buddy_page)
		page_cache_release(ac->ac_buddy_page);
	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
		mutex_unlock(&ac->ac_lg->lg_mutex);
	ext4_mb_collect_stats(ac);
	return 0;
}

static int ext4_mb_discard_preallocations(struct super_block *sb, int needed)
{
4452
	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4453 4454 4455
	int ret;
	int freed = 0;

4456
	trace_ext4_mb_discard_preallocations(sb, needed);
4457
	for (i = 0; i < ngroups && needed > 0; i++) {
4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473
		ret = ext4_mb_discard_group_preallocations(sb, i, needed);
		freed += ret;
		needed -= ret;
	}

	return freed;
}

/*
 * Main entry point into mballoc to allocate blocks
 * it tries to use preallocation first, then falls back
 * to usual allocation
 */
ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
				 struct ext4_allocation_request *ar, int *errp)
{
4474
	int freed;
4475
	struct ext4_allocation_context *ac = NULL;
4476 4477 4478
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block = 0;
4479
	unsigned int inquota = 0;
4480
	unsigned int reserv_blks = 0;
4481 4482 4483 4484

	sb = ar->inode->i_sb;
	sbi = EXT4_SB(sb);

4485
	trace_ext4_request_blocks(ar);
4486

4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497
	/*
	 * For delayed allocation, we could skip the ENOSPC and
	 * EDQUOT check, as blocks and quotas have been already
	 * reserved when data being copied into pagecache.
	 */
	if (EXT4_I(ar->inode)->i_delalloc_reserved_flag)
		ar->flags |= EXT4_MB_DELALLOC_RESERVED;
	else {
		/* Without delayed allocation we need to verify
		 * there is enough free blocks to do block allocation
		 * and verify allocation doesn't exceed the quota limits.
4498
		 */
A
Aneesh Kumar K.V 已提交
4499 4500 4501 4502 4503 4504
		while (ar->len && ext4_claim_free_blocks(sbi, ar->len)) {
			/* let others to free the space */
			yield();
			ar->len = ar->len >> 1;
		}
		if (!ar->len) {
4505 4506 4507
			*errp = -ENOSPC;
			return 0;
		}
4508
		reserv_blks = ar->len;
4509
		while (ar->len && vfs_dq_alloc_block(ar->inode, ar->len)) {
4510 4511 4512 4513 4514 4515 4516 4517
			ar->flags |= EXT4_MB_HINT_NOPREALLOC;
			ar->len--;
		}
		inquota = ar->len;
		if (ar->len == 0) {
			*errp = -EDQUOT;
			goto out3;
		}
4518
	}
4519

4520
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4521
	if (!ac) {
4522
		ar->len = 0;
4523
		*errp = -ENOMEM;
4524
		goto out1;
4525 4526 4527
	}

	*errp = ext4_mb_initialize_context(ac, ar);
4528 4529
	if (*errp) {
		ar->len = 0;
4530
		goto out2;
4531 4532
	}

4533 4534 4535 4536
	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
	if (!ext4_mb_use_preallocated(ac)) {
		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
		ext4_mb_normalize_request(ac, ar);
4537 4538
repeat:
		/* allocate space in core */
4539
		ext4_mb_regular_allocator(ac);
4540 4541 4542 4543

		/* as we've just preallocated more space than
		 * user requested orinally, we store allocated
		 * space in a special descriptor */
4544 4545 4546
		if (ac->ac_status == AC_STATUS_FOUND &&
				ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
			ext4_mb_new_preallocation(ac);
4547
	}
4548
	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
4549
		*errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_blks);
4550
		if (*errp ==  -EAGAIN) {
4551 4552 4553 4554 4555
			/*
			 * drop the reference that we took
			 * in ext4_mb_use_best_found
			 */
			ext4_mb_release_context(ac);
4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568
			ac->ac_b_ex.fe_group = 0;
			ac->ac_b_ex.fe_start = 0;
			ac->ac_b_ex.fe_len = 0;
			ac->ac_status = AC_STATUS_CONTINUE;
			goto repeat;
		} else if (*errp) {
			ac->ac_b_ex.fe_len = 0;
			ar->len = 0;
			ext4_mb_show_ac(ac);
		} else {
			block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
			ar->len = ac->ac_b_ex.fe_len;
		}
4569
	} else {
4570
		freed  = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
4571 4572 4573
		if (freed)
			goto repeat;
		*errp = -ENOSPC;
4574
		ac->ac_b_ex.fe_len = 0;
4575
		ar->len = 0;
4576
		ext4_mb_show_ac(ac);
4577 4578
	}

4579
	ext4_mb_release_context(ac);
4580

4581 4582 4583
out2:
	kmem_cache_free(ext4_ac_cachep, ac);
out1:
4584
	if (inquota && ar->len < inquota)
4585
		vfs_dq_free_block(ar->inode, inquota - ar->len);
4586 4587 4588 4589 4590 4591 4592
out3:
	if (!ar->len) {
		if (!EXT4_I(ar->inode)->i_delalloc_reserved_flag)
			/* release all the reserved blocks if non delalloc */
			percpu_counter_sub(&sbi->s_dirtyblocks_counter,
						reserv_blks);
	}
4593

4594
	trace_ext4_allocate_blocks(ar, (unsigned long long)block);
4595

4596 4597 4598
	return block;
}

4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613
/*
 * We can merge two free data extents only if the physical blocks
 * are contiguous, AND the extents were freed by the same transaction,
 * AND the blocks are associated with the same group.
 */
static int can_merge(struct ext4_free_data *entry1,
			struct ext4_free_data *entry2)
{
	if ((entry1->t_tid == entry2->t_tid) &&
	    (entry1->group == entry2->group) &&
	    ((entry1->start_blk + entry1->count) == entry2->start_blk))
		return 1;
	return 0;
}

4614 4615
static noinline_for_stack int
ext4_mb_free_metadata(handle_t *handle, struct ext4_buddy *e4b,
4616
		      struct ext4_free_data *new_entry)
4617
{
4618 4619
	ext4_grpblk_t block;
	struct ext4_free_data *entry;
4620 4621 4622
	struct ext4_group_info *db = e4b->bd_info;
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
4623 4624 4625
	struct rb_node **n = &db->bb_free_root.rb_node, *node;
	struct rb_node *parent = NULL, *new_node;

4626
	BUG_ON(!ext4_handle_valid(handle));
4627 4628 4629
	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

4630
	new_node = &new_entry->node;
4631
	block = new_entry->start_blk;
4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649

	if (!*n) {
		/* first free block exent. We need to
		   protect buddy cache from being freed,
		 * otherwise we'll refresh it from
		 * on-disk bitmap and lose not-yet-available
		 * blocks */
		page_cache_get(e4b->bd_buddy_page);
		page_cache_get(e4b->bd_bitmap_page);
	}
	while (*n) {
		parent = *n;
		entry = rb_entry(parent, struct ext4_free_data, node);
		if (block < entry->start_blk)
			n = &(*n)->rb_left;
		else if (block >= (entry->start_blk + entry->count))
			n = &(*n)->rb_right;
		else {
4650 4651 4652
			ext4_grp_locked_error(sb, e4b->bd_group, __func__,
					"Double free of blocks %d (%d %d)",
					block, entry->start_blk, entry->count);
4653
			return 0;
4654
		}
4655
	}
4656

4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671
	rb_link_node(new_node, parent, n);
	rb_insert_color(new_node, &db->bb_free_root);

	/* Now try to see the extent can be merged to left and right */
	node = rb_prev(new_node);
	if (node) {
		entry = rb_entry(node, struct ext4_free_data, node);
		if (can_merge(entry, new_entry)) {
			new_entry->start_blk = entry->start_blk;
			new_entry->count += entry->count;
			rb_erase(node, &(db->bb_free_root));
			spin_lock(&sbi->s_md_lock);
			list_del(&entry->list);
			spin_unlock(&sbi->s_md_lock);
			kmem_cache_free(ext4_free_ext_cachep, entry);
4672
		}
4673
	}
4674

4675 4676 4677 4678 4679 4680 4681 4682 4683 4684
	node = rb_next(new_node);
	if (node) {
		entry = rb_entry(node, struct ext4_free_data, node);
		if (can_merge(new_entry, entry)) {
			new_entry->count += entry->count;
			rb_erase(node, &(db->bb_free_root));
			spin_lock(&sbi->s_md_lock);
			list_del(&entry->list);
			spin_unlock(&sbi->s_md_lock);
			kmem_cache_free(ext4_free_ext_cachep, entry);
4685 4686
		}
	}
4687
	/* Add the extent to transaction's private list */
4688
	spin_lock(&sbi->s_md_lock);
4689
	list_add(&new_entry->list, &handle->h_transaction->t_private_list);
4690
	spin_unlock(&sbi->s_md_lock);
4691 4692 4693 4694 4695 4696 4697
	return 0;
}

/*
 * Main entry point into mballoc to free blocks
 */
void ext4_mb_free_blocks(handle_t *handle, struct inode *inode,
4698
			ext4_fsblk_t block, unsigned long count,
4699 4700
			int metadata, unsigned long *freed)
{
4701
	struct buffer_head *bitmap_bh = NULL;
4702
	struct super_block *sb = inode->i_sb;
4703
	struct ext4_allocation_context *ac = NULL;
4704 4705
	struct ext4_group_desc *gdp;
	struct ext4_super_block *es;
4706
	unsigned int overflow;
4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721
	ext4_grpblk_t bit;
	struct buffer_head *gd_bh;
	ext4_group_t block_group;
	struct ext4_sb_info *sbi;
	struct ext4_buddy e4b;
	int err = 0;
	int ret;

	*freed = 0;

	sbi = EXT4_SB(sb);
	es = EXT4_SB(sb)->s_es;
	if (block < le32_to_cpu(es->s_first_data_block) ||
	    block + count < block ||
	    block + count > ext4_blocks_count(es)) {
4722
		ext4_error(sb, __func__,
4723
			    "Freeing blocks not in datazone - "
4724
			    "block = %llu, count = %lu", block, count);
4725 4726 4727
		goto error_return;
	}

4728
	ext4_debug("freeing block %llu\n", block);
4729
	trace_ext4_free_blocks(inode, block, count, metadata);
4730

4731 4732 4733 4734 4735 4736
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
	if (ac) {
		ac->ac_op = EXT4_MB_HISTORY_FREE;
		ac->ac_inode = inode;
		ac->ac_sb = sb;
	}
4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749

do_more:
	overflow = 0;
	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);

	/*
	 * Check to see if we are freeing blocks across a group
	 * boundary.
	 */
	if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
		overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
		count -= overflow;
	}
4750
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4751 4752
	if (!bitmap_bh) {
		err = -EIO;
4753
		goto error_return;
4754
	}
4755
	gdp = ext4_get_group_desc(sb, block_group, &gd_bh);
4756 4757
	if (!gdp) {
		err = -EIO;
4758
		goto error_return;
4759
	}
4760 4761 4762 4763 4764 4765 4766 4767

	if (in_range(ext4_block_bitmap(sb, gdp), block, count) ||
	    in_range(ext4_inode_bitmap(sb, gdp), block, count) ||
	    in_range(block, ext4_inode_table(sb, gdp),
		      EXT4_SB(sb)->s_itb_per_group) ||
	    in_range(block + count - 1, ext4_inode_table(sb, gdp),
		      EXT4_SB(sb)->s_itb_per_group)) {

4768
		ext4_error(sb, __func__,
4769
			   "Freeing blocks in system zone - "
4770
			   "Block = %llu, count = %lu", block, count);
4771 4772
		/* err = 0. ext4_std_error should be a no op */
		goto error_return;
4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795
	}

	BUFFER_TRACE(bitmap_bh, "getting write access");
	err = ext4_journal_get_write_access(handle, bitmap_bh);
	if (err)
		goto error_return;

	/*
	 * We are about to modify some metadata.  Call the journal APIs
	 * to unshare ->b_data if a currently-committing transaction is
	 * using it
	 */
	BUFFER_TRACE(gd_bh, "get_write_access");
	err = ext4_journal_get_write_access(handle, gd_bh);
	if (err)
		goto error_return;
#ifdef AGGRESSIVE_CHECK
	{
		int i;
		for (i = 0; i < count; i++)
			BUG_ON(!mb_test_bit(bit + i, bitmap_bh->b_data));
	}
#endif
4796 4797 4798 4799 4800 4801
	if (ac) {
		ac->ac_b_ex.fe_group = block_group;
		ac->ac_b_ex.fe_start = bit;
		ac->ac_b_ex.fe_len = count;
		ext4_mb_store_history(ac);
	}
4802

4803 4804 4805
	err = ext4_mb_load_buddy(sb, block_group, &e4b);
	if (err)
		goto error_return;
4806
	if (metadata && ext4_handle_valid(handle)) {
4807 4808 4809 4810 4811 4812 4813 4814 4815 4816
		struct ext4_free_data *new_entry;
		/*
		 * blocks being freed are metadata. these blocks shouldn't
		 * be used until this transaction is committed
		 */
		new_entry  = kmem_cache_alloc(ext4_free_ext_cachep, GFP_NOFS);
		new_entry->start_blk = bit;
		new_entry->group  = block_group;
		new_entry->count = count;
		new_entry->t_tid = handle->h_transaction->t_tid;
4817

4818
		ext4_lock_group(sb, block_group);
4819
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4820
		ext4_mb_free_metadata(handle, &e4b, new_entry);
4821
	} else {
4822 4823 4824 4825
		/* need to update group_info->bb_free and bitmap
		 * with group lock held. generate_buddy look at
		 * them with group lock_held
		 */
4826 4827
		ext4_lock_group(sb, block_group);
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4828
		mb_free_blocks(inode, &e4b, bit, count);
4829 4830 4831
		ext4_mb_return_to_preallocation(inode, &e4b, block, count);
	}

4832 4833
	ret = ext4_free_blks_count(sb, gdp) + count;
	ext4_free_blks_set(sb, gdp, ret);
4834
	gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
4835
	ext4_unlock_group(sb, block_group);
4836 4837
	percpu_counter_add(&sbi->s_freeblocks_counter, count);

4838 4839
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4840
		atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks);
4841 4842
	}

4843 4844 4845 4846
	ext4_mb_release_desc(&e4b);

	*freed += count;

4847 4848 4849 4850
	/* We dirtied the bitmap block */
	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);

4851 4852
	/* And the group descriptor block */
	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
4853
	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866
	if (!err)
		err = ret;

	if (overflow && !err) {
		block += count;
		count = overflow;
		put_bh(bitmap_bh);
		goto do_more;
	}
	sb->s_dirt = 1;
error_return:
	brelse(bitmap_bh);
	ext4_std_error(sb, err);
4867 4868
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
4869 4870
	return;
}