mballoc.c 127.1 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
#include <linux/slab.h>
27 28
#include <trace/events/ext4.h>

29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
/*
 * 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
53 54 55 56 57 58 59 60 61
 * 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.
62 63
 *
 * The main motivation for having small file use group preallocation is to
T
Theodore Ts'o 已提交
64
 * ensure that we have small files closer together on the disk.
65
 *
T
Theodore Ts'o 已提交
66 67 68 69
 * 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:
70 71 72
 *
 * pa_lstart -> the logical start block for this prealloc space
 * pa_pstart -> the physical start block for this prealloc space
73
 * pa_len    -> length for this prealloc space
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 99 100 101 102 103 104 105 106 107
 * 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                        }
108
 *  [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
 *
 *
 * 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
129
 * sbi->s_mb_group_prealloc. Default value of s_mb_group_prealloc is
130
 * 512 blocks. This can be tuned via
T
Theodore Ts'o 已提交
131
 * /sys/fs/ext4/<partition/mb_group_prealloc. The value is represented in
132 133 134 135
 * 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
136
 * The regular allocator(using the buddy cache) supports few tunables.
137
 *
T
Theodore Ts'o 已提交
138 139 140
 * /sys/fs/ext4/<partition>/mb_min_to_scan
 * /sys/fs/ext4/<partition>/mb_max_to_scan
 * /sys/fs/ext4/<partition>/mb_order2_req
141
 *
T
Theodore Ts'o 已提交
142
 * The regular allocator uses buddy scan only if the request len is power of
143 144
 * 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
145
 * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
146
 * stripe size (sbi->s_stripe), we try to search for contiguous block in
T
Theodore Ts'o 已提交
147 148 149
 * 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.
150
 * min_to_scan indicate how long the mballoc __must__ look for a best
T
Theodore Ts'o 已提交
151
 * extent and max_to_scan indicates how long the mballoc __can__ look for a
152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
 * 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
 *
 */
338 339 340
static struct kmem_cache *ext4_pspace_cachep;
static struct kmem_cache *ext4_ac_cachep;
static struct kmem_cache *ext4_free_ext_cachep;
341 342 343 344 345 346 347 348

/* We create slab caches for groupinfo data structures based on the
 * superblock block size.  There will be one per mounted filesystem for
 * each unique s_blocksize_bits */
#define NR_GRPINFO_CACHES	\
	(EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE + 1)
static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];

349 350
static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
					ext4_group_t group);
351 352
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group);
353 354
static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);

355 356
static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
{
357
#if BITS_PER_LONG == 64
358 359
	*bit += ((unsigned long) addr & 7UL) << 3;
	addr = (void *) ((unsigned long) addr & ~7UL);
360
#elif BITS_PER_LONG == 32
361 362
	*bit += ((unsigned long) addr & 3UL) << 3;
	addr = (void *) ((unsigned long) addr & ~3UL);
363 364 365
#else
#error "how many bits you are?!"
#endif
366 367
	return addr;
}
368 369 370 371 372 373 374

static inline int mb_test_bit(int bit, void *addr)
{
	/*
	 * ext4_test_bit on architecture like powerpc
	 * needs unsigned long aligned address
	 */
375
	addr = mb_correct_addr_and_bit(&bit, addr);
376 377 378 379 380
	return ext4_test_bit(bit, addr);
}

static inline void mb_set_bit(int bit, void *addr)
{
381
	addr = mb_correct_addr_and_bit(&bit, addr);
382 383 384 385 386
	ext4_set_bit(bit, addr);
}

static inline void mb_clear_bit(int bit, void *addr)
{
387
	addr = mb_correct_addr_and_bit(&bit, addr);
388 389 390
	ext4_clear_bit(bit, addr);
}

391 392
static inline int mb_find_next_zero_bit(void *addr, int max, int start)
{
393
	int fix = 0, ret, tmpmax;
394
	addr = mb_correct_addr_and_bit(&fix, addr);
395
	tmpmax = max + fix;
396 397
	start += fix;

398 399 400 401
	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
402 403 404 405
}

static inline int mb_find_next_bit(void *addr, int max, int start)
{
406
	int fix = 0, ret, tmpmax;
407
	addr = mb_correct_addr_and_bit(&fix, addr);
408
	tmpmax = max + fix;
409 410
	start += fix;

411 412 413 414
	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
415 416
}

417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448
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;
449
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
450 451 452
	for (i = 0; i < count; i++) {
		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
			ext4_fsblk_t blocknr;
453 454

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
455
			blocknr += first + i;
456
			ext4_grp_locked_error(sb, e4b->bd_group,
457 458 459 460 461
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing block already freed "
					      "(bit %u)",
					      first + i);
462 463 464 465 466 467 468 469 470 471 472
		}
		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;
473
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
	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]) {
489
				printk(KERN_ERR "corruption in group %u "
490 491 492
				       "at byte %u(%u): %x in copy != %x "
				       "on disk/prealloc\n",
				       e4b->bd_group, i, i * 8, b1[i], b2[i]);
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617
				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;
618 619
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
620
		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
621
		for (i = 0; i < pa->pa_len; i++)
622 623 624 625 626 627
			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
	}
	return 0;
}
#undef MB_CHECK_ASSERT
#define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
628
					__FILE__, __func__, __LINE__)
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,
635
				void *buddy, ext4_grpblk_t first, ext4_grpblk_t len,
636 637 638
					struct ext4_group_info *grp)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
639 640 641
	ext4_grpblk_t min;
	ext4_grpblk_t max;
	ext4_grpblk_t chunk;
642 643
	unsigned short border;

644
	BUG_ON(len > EXT4_BLOCKS_PER_GROUP(sb));
645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669

	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;
	}
}

670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690
/*
 * Cache the order of the largest free extent we have available in this block
 * group.
 */
static void
mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
{
	int i;
	int bits;

	grp->bb_largest_free_order = -1; /* uninit */

	bits = sb->s_blocksize_bits + 1;
	for (i = bits; i >= 0; i--) {
		if (grp->bb_counters[i] > 0) {
			grp->bb_largest_free_order = i;
			break;
		}
	}
}

691 692
static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
693 694 695
				void *buddy, void *bitmap, ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
696 697 698 699
	ext4_grpblk_t max = EXT4_BLOCKS_PER_GROUP(sb);
	ext4_grpblk_t i = 0;
	ext4_grpblk_t first;
	ext4_grpblk_t len;
700 701 702 703 704 705
	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 */
706
	i = mb_find_next_zero_bit(bitmap, max, 0);
707 708 709 710
	grp->bb_first_free = i;
	while (i < max) {
		fragments++;
		first = i;
711
		i = mb_find_next_bit(bitmap, max, i);
712 713 714 715 716 717 718
		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)
719
			i = mb_find_next_zero_bit(bitmap, max, i);
720 721 722 723
	}
	grp->bb_fragments = fragments;

	if (free != grp->bb_free) {
724 725 726
		ext4_grp_locked_error(sb, group, 0, 0,
				      "%u blocks in bitmap, %u in gd",
				      free, grp->bb_free);
727 728 729 730
		/*
		 * If we intent to continue, we consider group descritor
		 * corrupt and update bb_free using bitmap value
		 */
731 732
		grp->bb_free = free;
	}
733
	mb_set_largest_free_order(sb, grp);
734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750

	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                        }
751
 * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
752 753 754 755 756 757 758
 *
 *
 * 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
759 760 761
 *
 * Locking note:  This routine takes the block group lock of all groups
 * for this page; do not hold this lock when calling this routine!
762 763 764 765
 */

static int ext4_mb_init_cache(struct page *page, char *incore)
{
766
	ext4_group_t ngroups;
767 768 769 770 771 772 773 774 775 776 777 778 779 780
	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;

781
	mb_debug(1, "init page %lu\n", page->index);
782 783 784

	inode = page->mapping->host;
	sb = inode->i_sb;
785
	ngroups = ext4_get_groups_count(sb);
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808
	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;

809
		if (first_group + i >= ngroups)
810 811 812 813 814 815 816 817 818 819 820 821
			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;

822
		if (bitmap_uptodate(bh[i]))
823 824
			continue;

825
		lock_buffer(bh[i]);
826 827 828 829
		if (bitmap_uptodate(bh[i])) {
			unlock_buffer(bh[i]);
			continue;
		}
830
		ext4_lock_group(sb, first_group + i);
831 832 833
		if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
			ext4_init_block_bitmap(sb, bh[i],
						first_group + i, desc);
834
			set_bitmap_uptodate(bh[i]);
835
			set_buffer_uptodate(bh[i]);
836
			ext4_unlock_group(sb, first_group + i);
A
Aneesh Kumar K.V 已提交
837
			unlock_buffer(bh[i]);
838 839
			continue;
		}
840
		ext4_unlock_group(sb, first_group + i);
841 842 843 844 845 846 847 848 849
		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;
		}
850
		get_bh(bh[i]);
851 852 853 854 855 856 857
		/*
		 * 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]);
858 859
		bh[i]->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh[i]);
860
		mb_debug(1, "read bitmap for group %u\n", first_group + i);
861 862 863 864 865 866 867 868 869 870 871
	}

	/* 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;

872
	err = 0;
873
	first_block = page->index * blocks_per_page;
874 875
	/* init the page  */
	memset(page_address(page), 0xff, PAGE_CACHE_SIZE);
876 877 878 879 880
	for (i = 0; i < blocks_per_page; i++) {
		int group;
		struct ext4_group_info *grinfo;

		group = (first_block + i) >> 1;
881
		if (group >= ngroups)
882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899
			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);
900
			mb_debug(1, "put buddy for group %u in page %lu/%x\n",
901
				group, page->index, i * blocksize);
902
			trace_ext4_mb_buddy_bitmap_load(sb, group);
903 904 905
			grinfo = ext4_get_group_info(sb, group);
			grinfo->bb_fragments = 0;
			memset(grinfo->bb_counters, 0,
906 907
			       sizeof(*grinfo->bb_counters) *
				(sb->s_blocksize_bits+2));
908 909 910
			/*
			 * incore got set to the group block bitmap below
			 */
911
			ext4_lock_group(sb, group);
912
			ext4_mb_generate_buddy(sb, data, incore, group);
913
			ext4_unlock_group(sb, group);
914 915 916 917
			incore = NULL;
		} else {
			/* this is block of bitmap */
			BUG_ON(incore != NULL);
918
			mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
919
				group, page->index, i * blocksize);
920
			trace_ext4_mb_bitmap_load(sb, group);
921 922 923 924 925 926 927

			/* 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);
928
			ext4_mb_generate_from_freelist(sb, data, group);
929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948
			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;
}

949 950 951 952 953
/*
 * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
 * block group lock of all groups for this page; do not hold the BG lock when
 * calling this routine!
 */
954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
{

	int ret = 0;
	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;

	mb_debug(1, "init group %u\n", group);
	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
	this_grp = ext4_get_group_info(sb, group);
	/*
972 973 974 975 976
	 * This ensures that we don't reinit the buddy cache
	 * page which map to the group from which we are already
	 * allocating. If we are looking at the buddy cache we would
	 * have taken a reference using ext4_mb_load_buddy and that
	 * would have taken the alloc_sem lock.
977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047
	 */
	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;
}

1048 1049 1050 1051 1052
/*
 * Locking note:  This routine calls ext4_mb_init_cache(), which takes the
 * block group lock of all groups for this page; do not hold the BG lock when
 * calling this routine!
 */
1053 1054 1055
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
					struct ext4_buddy *e4b)
1056 1057 1058 1059 1060 1061
{
	int blocks_per_page;
	int block;
	int pnum;
	int poff;
	struct page *page;
1062
	int ret;
1063 1064 1065
	struct ext4_group_info *grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;
1066

1067
	mb_debug(1, "load group %u\n", group);
1068 1069

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1070
	grp = ext4_get_group_info(sb, group);
1071 1072 1073 1074 1075 1076 1077

	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;
1078 1079 1080 1081 1082 1083 1084 1085
	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
	 */
1086
repeat_load_buddy:
1087
	down_read(e4b->alloc_semp);
1088

1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
		/* we need to check for group need init flag
		 * with alloc_semp held so that we can be sure
		 * that new blocks didn't get added to the group
		 * when we are loading the buddy cache
		 */
		up_read(e4b->alloc_semp);
		/*
		 * we need full data about the group
		 * to make a good selection
		 */
		ret = ext4_mb_init_group(sb, group);
		if (ret)
			return ret;
		goto repeat_load_buddy;
	}

1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
	/*
	 * 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)
1120 1121 1122 1123 1124 1125 1126 1127
			/*
			 * 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.
			 */
1128 1129 1130 1131 1132
			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)) {
1133 1134 1135 1136 1137
				ret = ext4_mb_init_cache(page, NULL);
				if (ret) {
					unlock_page(page);
					goto err;
				}
1138 1139 1140 1141 1142 1143
				mb_cmp_bitmaps(e4b, page_address(page) +
					       (poff * sb->s_blocksize));
			}
			unlock_page(page);
		}
	}
1144 1145
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1146
		goto err;
1147
	}
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162
	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);
1163 1164 1165 1166 1167 1168 1169
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
				if (ret) {
					unlock_page(page);
					goto err;
				}
			}
1170 1171 1172
			unlock_page(page);
		}
	}
1173 1174
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1175
		goto err;
1176
	}
1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
	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;
1193 1194 1195

	/* Done with the buddy cache */
	up_read(e4b->alloc_semp);
1196
	return ret;
1197 1198
}

1199
static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
1200 1201 1202 1203 1204
{
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
1205
	/* Done with the buddy cache */
1206 1207
	if (e4b->alloc_semp)
		up_read(e4b->alloc_semp);
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231
}


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;
}

1232
static void mb_clear_bits(void *bm, int cur, int len)
1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
{
	__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;
		}
1245
		mb_clear_bit(cur, bm);
1246 1247 1248 1249
		cur++;
	}
}

1250
static void mb_set_bits(void *bm, int cur, int len)
1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
{
	__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;
		}
1263
		mb_set_bit(cur, bm);
1264 1265 1266 1267
		cur++;
	}
}

1268
static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
			  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));
1279
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
	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;
1304 1305

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
1306
			blocknr += block;
1307
			ext4_grp_locked_error(sb, e4b->bd_group,
1308 1309 1310 1311
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing already freed block "
					      "(bit %u)", block);
1312 1313 1314 1315 1316 1317 1318 1319 1320 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
		}
		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);
	}
1348
	mb_set_largest_free_order(sb, e4b->bd_info);
1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359
	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;

1360
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 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
	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);
1424
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 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
	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]++;
	}
1477
	mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
1478

1479
	mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508
	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;

1509 1510 1511 1512 1513 1514 1515
	/*
	 * 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
	 */
1516 1517 1518 1519
	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);
1520 1521 1522
	/* on allocation we use ac to track the held semaphore */
	ac->alloc_semp =  e4b->alloc_semp;
	e4b->alloc_semp = NULL;
1523
	/* store last allocated for subsequent stream allocation */
1524
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
		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;

1546 1547
	if (ac->ac_status == AC_STATUS_FOUND)
		return;
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 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593
	/*
	 * 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);
1594
	BUG_ON(ex->fe_len > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644
	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);
}

1645 1646
static noinline_for_stack
int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667
					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);
1668
	ext4_mb_unload_buddy(e4b);
1669 1670 1671 1672

	return 0;
}

1673 1674
static noinline_for_stack
int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696
				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_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;

1697 1698
		start = ext4_group_first_block_no(ac->ac_sb, e4b->bd_group) +
			ex.fe_start;
1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
		/* 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);
1723
	ext4_mb_unload_buddy(e4b);
1724 1725 1726 1727 1728 1729 1730 1731

	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
 */
1732 1733
static noinline_for_stack
void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750
					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);

1751
		k = mb_find_next_zero_bit(buddy, max, 0);
1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775
		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.
 */
1776 1777
static noinline_for_stack
void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791
					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) {
1792
		i = mb_find_next_zero_bit(bitmap,
1793 1794
						EXT4_BLOCKS_PER_GROUP(sb), i);
		if (i >= EXT4_BLOCKS_PER_GROUP(sb)) {
1795
			/*
1796
			 * IF we have corrupt bitmap, we won't find any
1797 1798 1799
			 * free blocks even though group info says we
			 * we have free blocks
			 */
1800 1801
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
					"%d free blocks as per "
1802
					"group info. But bitmap says 0",
1803
					free);
1804 1805 1806 1807 1808
			break;
		}

		mb_find_extent(e4b, 0, i, ac->ac_g_ex.fe_len, &ex);
		BUG_ON(ex.fe_len <= 0);
1809
		if (free < ex.fe_len) {
1810 1811
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
					"%d free blocks as per "
1812
					"group info. But got %d blocks",
1813
					free, ex.fe_len);
1814 1815 1816 1817 1818 1819
			/*
			 * The number of free blocks differs. This mostly
			 * indicate that the bitmap is corrupt. So exit
			 * without claiming the space.
			 */
			break;
1820
		}
1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832

		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
1833
 * we try to find stripe-aligned chunks for stripe-size-multiple requests
1834
 */
1835 1836
static noinline_for_stack
void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850
				 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 */
1851 1852
	first_group_block = ext4_group_first_block_no(sb, e4b->bd_group);

1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
	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;
	}
}

1871
/* This is now called BEFORE we load the buddy bitmap. */
1872 1873 1874 1875
static int ext4_mb_good_group(struct ext4_allocation_context *ac,
				ext4_group_t group, int cr)
{
	unsigned free, fragments;
1876
	int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
1877 1878 1879
	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);

	BUG_ON(cr < 0 || cr >= 4);
1880 1881 1882 1883 1884 1885 1886

	/* We only do this if the grp has never been initialized */
	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
		int ret = ext4_mb_init_group(ac->ac_sb, group);
		if (ret)
			return 0;
	}
1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898

	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);

1899 1900 1901
		if (grp->bb_largest_free_order < ac->ac_2order)
			return 0;

1902 1903 1904 1905 1906 1907
		/* 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;

1908
		return 1;
1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925
	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;
}

1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
/*
 * 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;
1939
	ext4_group_t ngroups = ext4_get_groups_count(sb);
1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958
	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++) {

1959
		if ((first_group + i) >= ngroups)
1960 1961 1962 1963 1964 1965 1966
			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
		 */
1967
		down_write_nested(&grp->alloc_sem, i);
1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
	}
	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);
	}

}

2004 2005
static noinline_for_stack int
ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
2006
{
2007
	ext4_group_t ngroups, group, i;
2008 2009 2010 2011 2012 2013 2014 2015
	int cr;
	int err = 0;
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	struct ext4_buddy e4b;

	sb = ac->ac_sb;
	sbi = EXT4_SB(sb);
2016
	ngroups = ext4_get_groups_count(sb);
2017
	/* non-extent files are limited to low blocks/groups */
2018
	if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)))
2019 2020
		ngroups = sbi->s_blockfile_groups;

2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040
	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
2041
	 * You can tune it via /sys/fs/ext4/<partition>/mb_order2_req
2042 2043 2044 2045 2046 2047 2048 2049 2050
	 */
	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;
	}

2051 2052
	/* if stream allocation is enabled, use global goal */
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
2053 2054 2055 2056 2057 2058
		/* 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);
	}
2059

2060 2061 2062 2063 2064 2065 2066 2067 2068
	/* 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;
2069 2070 2071 2072 2073 2074
		/*
		 * searching for the right group start
		 * from the goal value specified
		 */
		group = ac->ac_g_ex.fe_group;

2075 2076
		for (i = 0; i < ngroups; group++, i++) {
			if (group == ngroups)
2077 2078
				group = 0;

2079 2080
			/* This now checks without needing the buddy page */
			if (!ext4_mb_good_group(ac, group, cr))
2081 2082 2083 2084 2085 2086 2087
				continue;

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

			ext4_lock_group(sb, group);
2088 2089 2090 2091 2092

			/*
			 * We need to check again after locking the
			 * block group
			 */
2093 2094
			if (!ext4_mb_good_group(ac, group, cr)) {
				ext4_unlock_group(sb, group);
2095
				ext4_mb_unload_buddy(&e4b);
2096 2097 2098 2099
				continue;
			}

			ac->ac_groups_scanned++;
2100
			if (cr == 0)
2101
				ext4_mb_simple_scan_group(ac, &e4b);
2102 2103
			else if (cr == 1 && sbi->s_stripe &&
					!(ac->ac_g_ex.fe_len % sbi->s_stripe))
2104 2105 2106 2107 2108
				ext4_mb_scan_aligned(ac, &e4b);
			else
				ext4_mb_complex_scan_group(ac, &e4b);

			ext4_unlock_group(sb, group);
2109
			ext4_mb_unload_buddy(&e4b);
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

			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;
}

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

2150
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2151 2152
		return NULL;
	group = *pos + 1;
2153
	return (void *) ((unsigned long) group);
2154 2155 2156 2157 2158 2159 2160 2161
}

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;
2162
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2163 2164
		return NULL;
	group = *pos + 1;
2165
	return (void *) ((unsigned long) group);
2166 2167 2168 2169 2170
}

static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
{
	struct super_block *sb = seq->private;
2171
	ext4_group_t group = (ext4_group_t) ((unsigned long) v);
2172 2173 2174 2175 2176
	int i;
	int err;
	struct ext4_buddy e4b;
	struct sg {
		struct ext4_group_info info;
2177
		ext4_grpblk_t counters[16];
2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192
	} 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) {
2193
		seq_printf(seq, "#%-5u: I/O error\n", group);
2194 2195 2196 2197 2198
		return 0;
	}
	ext4_lock_group(sb, group);
	memcpy(&sg, ext4_get_group_info(sb, group), i);
	ext4_unlock_group(sb, group);
2199
	ext4_mb_unload_buddy(&e4b);
2200

2201
	seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214
			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)
{
}

2215
static const struct seq_operations ext4_mb_seq_groups_ops = {
2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228
	.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) {
2229
		struct seq_file *m = file->private_data;
2230 2231 2232 2233 2234 2235
		m->private = sb;
	}
	return rc;

}

2236
static const struct file_operations ext4_mb_seq_groups_fops = {
2237 2238 2239 2240 2241 2242 2243
	.owner		= THIS_MODULE,
	.open		= ext4_mb_seq_groups_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= seq_release,
};

2244 2245 2246 2247 2248 2249 2250 2251
static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
{
	int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
	struct kmem_cache *cachep = ext4_groupinfo_caches[cache_index];

	BUG_ON(!cachep);
	return cachep;
}
2252 2253

/* Create and initialize ext4_group_info data for the given group. */
2254
int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2255 2256
			  struct ext4_group_desc *desc)
{
2257
	int i;
2258 2259 2260
	int metalen = 0;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_group_info **meta_group_info;
2261
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284

	/*
	 * 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;
	}

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

2285
	meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
2286 2287 2288 2289
	if (meta_group_info[i] == NULL) {
		printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
		goto exit_group_info;
	}
2290
	memset(meta_group_info[i], 0, kmem_cache_size(cachep));
2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302
	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 =
2303
			ext4_free_blks_count(sb, desc);
2304 2305 2306
	}

	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2307
	init_rwsem(&meta_group_info[i]->alloc_sem);
2308
	meta_group_info[i]->bb_free_root = RB_ROOT;
2309
	meta_group_info[i]->bb_largest_free_order = -1;  /* uninit */
2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334

#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 */

2335 2336
static int ext4_mb_init_backend(struct super_block *sb)
{
2337
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2338 2339
	ext4_group_t i;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2340 2341 2342 2343 2344
	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;
2345
	struct kmem_cache *cachep;
2346 2347

	/* This is the number of blocks used by GDT */
2348
	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363
				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);
2364

2365 2366 2367 2368 2369 2370 2371 2372 2373 2374
	/*
	 * 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;
2375 2376 2377
	/* 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. */
2378
	sbi->s_group_info = kmalloc(array_size, GFP_KERNEL);
2379 2380 2381 2382 2383 2384 2385 2386 2387 2388
	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;
2389
	for (i = 0; i < ngroups; i++) {
2390 2391 2392
		desc = ext4_get_group_desc(sb, i, NULL);
		if (desc == NULL) {
			printk(KERN_ERR
2393
				"EXT4-fs: can't read descriptor %u\n", i);
2394 2395
			goto err_freebuddy;
		}
2396 2397
		if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
			goto err_freebuddy;
2398 2399 2400 2401 2402
	}

	return 0;

err_freebuddy:
2403
	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2404
	while (i-- > 0)
2405
		kmem_cache_free(cachep, ext4_get_group_info(sb, i));
2406
	i = num_meta_group_infos;
2407
	while (i-- > 0)
2408 2409 2410 2411 2412 2413 2414 2415 2416 2417
		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);
2418
	unsigned i, j;
2419 2420
	unsigned offset;
	unsigned max;
2421
	int ret;
2422 2423 2424
	int cache_index;
	struct kmem_cache *cachep;
	char *namep = NULL;
2425

2426
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
2427 2428 2429

	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_offsets == NULL) {
2430 2431
		ret = -ENOMEM;
		goto out;
2432
	}
2433

2434
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
2435 2436
	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_maxs == NULL) {
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
		ret = -ENOMEM;
		goto out;
	}

	cache_index = sb->s_blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
	cachep = ext4_groupinfo_caches[cache_index];
	if (!cachep) {
		char name[32];
		int len = offsetof(struct ext4_group_info,
					bb_counters[sb->s_blocksize_bits + 2]);

		sprintf(name, "ext4_groupinfo_%d", sb->s_blocksize_bits);
		namep = kstrdup(name, GFP_KERNEL);
		if (!namep) {
			ret = -ENOMEM;
			goto out;
		}

		/* Need to free the kmem_cache_name() when we
		 * destroy the slab */
		cachep = kmem_cache_create(namep, len, 0,
					     SLAB_RECLAIM_ACCOUNT, NULL);
		if (!cachep) {
			ret = -ENOMEM;
			goto out;
		}
		ext4_groupinfo_caches[cache_index] = cachep;
2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481
	}

	/* 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 */
2482 2483
	ret = ext4_mb_init_backend(sb);
	if (ret != 0) {
2484
		goto out;
2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496
	}

	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_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;

2497
	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2498
	if (sbi->s_locality_groups == NULL) {
2499 2500
		ret = -ENOMEM;
		goto out;
2501
	}
2502
	for_each_possible_cpu(i) {
2503
		struct ext4_locality_group *lg;
2504
		lg = per_cpu_ptr(sbi->s_locality_groups, i);
2505
		mutex_init(&lg->lg_mutex);
2506 2507
		for (j = 0; j < PREALLOC_TB_SIZE; j++)
			INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
2508 2509 2510
		spin_lock_init(&lg->lg_prealloc_lock);
	}

2511 2512 2513
	if (sbi->s_proc)
		proc_create_data("mb_groups", S_IRUGO, sbi->s_proc,
				 &ext4_mb_seq_groups_fops, sb);
2514

2515 2516
	if (sbi->s_journal)
		sbi->s_journal->j_commit_callback = release_blocks_on_commit;
2517 2518 2519 2520 2521 2522 2523
out:
	if (ret) {
		kfree(sbi->s_mb_offsets);
		kfree(sbi->s_mb_maxs);
		kfree(namep);
	}
	return ret;
2524 2525
}

2526
/* need to called with the ext4 group lock held */
2527 2528 2529 2530 2531 2532 2533 2534 2535 2536
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++;
2537
		kmem_cache_free(ext4_pspace_cachep, pa);
2538 2539
	}
	if (count)
2540
		mb_debug(1, "mballoc: %u PAs left\n", count);
2541 2542 2543 2544 2545

}

int ext4_mb_release(struct super_block *sb)
{
2546
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2547 2548 2549 2550
	ext4_group_t i;
	int num_meta_group_infos;
	struct ext4_group_info *grinfo;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2551
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2552 2553

	if (sbi->s_group_info) {
2554
		for (i = 0; i < ngroups; i++) {
2555 2556 2557 2558 2559 2560 2561
			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);
2562
			kmem_cache_free(cachep, grinfo);
2563
		}
2564
		num_meta_group_infos = (ngroups +
2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598
				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));
	}

2599
	free_percpu(sbi->s_locality_groups);
2600 2601
	if (sbi->s_proc)
		remove_proc_entry("mb_groups", sbi->s_proc);
2602 2603 2604 2605

	return 0;
}

2606 2607 2608 2609 2610 2611 2612 2613 2614 2615
static inline void ext4_issue_discard(struct super_block *sb,
		ext4_group_t block_group, ext4_grpblk_t block, int count)
{
	int ret;
	ext4_fsblk_t discard_block;

	discard_block = block + ext4_group_first_block_no(sb, block_group);
	trace_ext4_discard_blocks(sb,
			(unsigned long long) discard_block, count);
	ret = sb_issue_discard(sb, discard_block, count);
2616
	if (ret == -EOPNOTSUPP) {
2617 2618 2619 2620 2621
		ext4_warning(sb, "discard not supported, disabling");
		clear_opt(EXT4_SB(sb)->s_mount_opt, DISCARD);
	}
}

2622 2623 2624 2625 2626
/*
 * 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)
2627
{
2628
	struct super_block *sb = journal->j_private;
2629
	struct ext4_buddy e4b;
2630 2631 2632
	struct ext4_group_info *db;
	int err, count = 0, count2 = 0;
	struct ext4_free_data *entry;
2633
	struct list_head *l, *ltmp;
2634

2635 2636
	list_for_each_safe(l, ltmp, &txn->t_private_list) {
		entry = list_entry(l, struct ext4_free_data, list);
2637

2638
		mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
2639
			 entry->count, entry->group, entry);
2640

2641 2642 2643
		if (test_opt(sb, DISCARD))
			ext4_issue_discard(sb, entry->group,
					entry->start_blk, entry->count);
2644

2645
		err = ext4_mb_load_buddy(sb, entry->group, &e4b);
2646 2647 2648
		/* we expect to find existing buddy because it's pinned */
		BUG_ON(err != 0);

2649
		db = e4b.bd_info;
2650
		/* there are blocks to put in buddy to make them really free */
2651
		count += entry->count;
2652
		count2++;
2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663
		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);
2664
		}
2665 2666
		ext4_unlock_group(sb, entry->group);
		kmem_cache_free(ext4_free_ext_cachep, entry);
2667
		ext4_mb_unload_buddy(&e4b);
2668
	}
2669

2670
	mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
2671 2672
}

2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706
#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

2707 2708 2709 2710 2711 2712 2713 2714 2715
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;

2716 2717 2718 2719 2720 2721 2722 2723
	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;
	}
2724 2725 2726 2727 2728 2729 2730 2731 2732 2733

	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;
	}
2734
	ext4_create_debugfs_entry();
2735 2736 2737 2738 2739
	return 0;
}

void exit_ext4_mballoc(void)
{
2740
	int i;
2741
	/*
2742 2743 2744 2745
	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
	 * before destroying the slab cache.
	 */
	rcu_barrier();
2746
	kmem_cache_destroy(ext4_pspace_cachep);
2747
	kmem_cache_destroy(ext4_ac_cachep);
2748
	kmem_cache_destroy(ext4_free_ext_cachep);
2749 2750 2751 2752 2753 2754 2755 2756 2757

	for (i = 0; i < NR_GRPINFO_CACHES; i++) {
		struct kmem_cache *cachep = ext4_groupinfo_caches[i];
		if (cachep) {
			char *name = (char *)kmem_cache_name(cachep);
			kmem_cache_destroy(cachep);
			kfree(name);
		}
	}
2758
	ext4_remove_debugfs_entry();
2759 2760 2761 2762
}


/*
2763
 * Check quota and mark chosen space (ac->ac_b_ex) non-free in bitmaps
2764 2765
 * Returns 0 if success or error code
 */
2766 2767
static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2768
				handle_t *handle, unsigned int reserv_blks)
2769 2770 2771 2772 2773 2774 2775
{
	struct buffer_head *bitmap_bh = NULL;
	struct ext4_group_desc *gdp;
	struct buffer_head *gdp_bh;
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block;
2776
	int err, len;
2777 2778 2779 2780 2781 2782 2783 2784

	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);

	err = -EIO;
2785
	bitmap_bh = ext4_read_block_bitmap(sb, ac->ac_b_ex.fe_group);
2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797
	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;

2798
	ext4_debug("using block group %u(%d)\n", ac->ac_b_ex.fe_group,
2799
			ext4_free_blks_count(sb, gdp));
2800

2801 2802 2803 2804
	err = ext4_journal_get_write_access(handle, gdp_bh);
	if (err)
		goto out_err;

2805
	block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
2806

2807
	len = ac->ac_b_ex.fe_len;
2808
	if (!ext4_data_block_valid(sbi, block, len)) {
2809
		ext4_error(sb, "Allocating blocks %llu-%llu which overlap "
2810
			   "fs metadata\n", block, block+len);
2811 2812 2813 2814
		/* File system mounted not to panic on error
		 * Fix the bitmap and repeat the block allocation
		 * We leak some of the blocks here.
		 */
2815 2816 2817 2818
		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);
2819
		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2820 2821 2822
		if (!err)
			err = -EAGAIN;
		goto out_err;
2823
	}
2824 2825

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2826 2827 2828 2829 2830 2831 2832 2833 2834
#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
2835
	mb_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,ac->ac_b_ex.fe_len);
2836 2837
	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
		gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
2838 2839 2840
		ext4_free_blks_set(sb, gdp,
					ext4_free_blocks_after_init(sb,
					ac->ac_b_ex.fe_group, gdp));
2841
	}
2842 2843
	len = ext4_free_blks_count(sb, gdp) - ac->ac_b_ex.fe_len;
	ext4_free_blks_set(sb, gdp, len);
2844
	gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
2845 2846

	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2847
	percpu_counter_sub(&sbi->s_freeblocks_counter, ac->ac_b_ex.fe_len);
2848
	/*
2849
	 * Now reduce the dirty block count also. Should not go negative
2850
	 */
2851 2852 2853
	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);
2854

2855 2856 2857
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi,
							  ac->ac_b_ex.fe_group);
2858 2859
		atomic_sub(ac->ac_b_ex.fe_len,
			   &sbi->s_flex_groups[flex_group].free_blocks);
2860 2861
	}

2862
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2863 2864
	if (err)
		goto out_err;
2865
	err = ext4_handle_dirty_metadata(handle, NULL, gdp_bh);
2866 2867

out_err:
T
Theodore Ts'o 已提交
2868
	ext4_mark_super_dirty(sb);
2869
	brelse(bitmap_bh);
2870 2871 2872 2873 2874 2875 2876
	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
2877
 * /sys/fs/ext4/<partition>/mb_group_prealloc
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890
 *
 * 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;
2891
	mb_debug(1, "#%u: goal %u blocks for locality group\n",
2892 2893 2894 2895 2896 2897 2898
		current->pid, ac->ac_g_ex.fe_len);
}

/*
 * Normalization means making request better in terms of
 * size and alignment
 */
2899 2900
static noinline_for_stack void
ext4_mb_normalize_request(struct ext4_allocation_context *ac,
2901 2902 2903 2904 2905
				struct ext4_allocation_request *ar)
{
	int bsbits, max;
	ext4_lblk_t end;
	loff_t size, orig_size, start_off;
2906
	ext4_lblk_t start;
2907
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
2908
	struct ext4_prealloc_space *pa;
2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936

	/* 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);
2937
	orig_size = size;
2938

2939 2940
	/* max size of free chunks */
	max = 2 << bsbits;
2941

2942 2943
#define NRL_CHECK_SIZE(req, size, max, chunk_size)	\
		(req <= (size) || max <= (chunk_size))
2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961

	/* 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;
2962
	} else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
2963
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
2964 2965 2966
						(21 - bsbits)) << 21;
		size = 2 * 1024 * 1024;
	} else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
2967 2968 2969 2970
		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,
2971
					(8<<20)>>bsbits, max, 8 * 1024)) {
2972 2973 2974 2975 2976 2977 2978
		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;
	}
2979 2980
	size = size >> bsbits;
	start = start_off >> bsbits;
2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993

	/* 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();
2994
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
2995
		ext4_lblk_t pa_end;
2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010

		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));

3011 3012
		/* skip PAs this normalized request doesn't overlap with */
		if (pa->pa_lstart >= end || pa_end <= start) {
3013 3014 3015 3016 3017
			spin_unlock(&pa->pa_lock);
			continue;
		}
		BUG_ON(pa->pa_lstart <= start && pa_end >= end);

3018
		/* adjust start or end to be adjacent to this pa */
3019 3020 3021
		if (pa_end <= ac->ac_o_ex.fe_logical) {
			BUG_ON(pa_end < start);
			start = pa_end;
3022
		} else if (pa->pa_lstart > ac->ac_o_ex.fe_logical) {
3023 3024 3025 3026 3027 3028 3029 3030 3031 3032
			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();
3033
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3034
		ext4_lblk_t pa_end;
3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051
		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);
3052
	BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076

	/* 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;
	}

3077
	mb_debug(1, "goal: %u(was %u) blocks at %u\n", (unsigned) size,
3078 3079 3080 3081 3082 3083 3084 3085 3086 3087
		(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);
3088
		if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
3089 3090 3091 3092 3093 3094 3095 3096 3097
			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);
	}

3098 3099 3100 3101
	if (ac->ac_op == EXT4_MB_HISTORY_ALLOC)
		trace_ext4_mballoc_alloc(ac);
	else
		trace_ext4_mballoc_prealloc(ac);
3102 3103
}

3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121
/*
 * Called on failure; free up any blocks from the inode PA for this
 * context.  We don't need this for MB_GROUP_PA because we only change
 * pa_free in ext4_mb_release_context(), but on failure, we've already
 * zeroed out ac->ac_b_ex.fe_len, so group_pa->pa_free is not changed.
 */
static void ext4_discard_allocated_blocks(struct ext4_allocation_context *ac)
{
	struct ext4_prealloc_space *pa = ac->ac_pa;
	int len;

	if (pa && pa->pa_type == MB_INODE_PA) {
		len = ac->ac_b_ex.fe_len;
		pa->pa_free += len;
	}

}

3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146
/*
 * 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;

3147
	mb_debug(1, "use %llu/%u from inode pa %p\n", start, len, pa);
3148 3149 3150 3151 3152 3153 3154 3155
}

/*
 * use blocks preallocated to locality group
 */
static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
3156
	unsigned int len = ac->ac_o_ex.fe_len;
3157

3158 3159 3160 3161 3162 3163 3164 3165
	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
3166
	 * possible race when the group is being loaded concurrently
3167
	 * instead we correct pa later, after blocks are marked
3168 3169
	 * in on-disk bitmap -- see ext4_mb_release_context()
	 * Other CPUs are prevented from allocating from this pa by lg_mutex
3170
	 */
3171
	mb_debug(1, "use %u/%u from group pa %p\n", pa->pa_lstart-len, len, pa);
3172 3173
}

3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202
/*
 * 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;
}

3203 3204 3205
/*
 * search goal blocks in preallocated space
 */
3206 3207
static noinline_for_stack int
ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
3208
{
3209
	int order, i;
3210 3211
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
	struct ext4_locality_group *lg;
3212 3213
	struct ext4_prealloc_space *pa, *cpa = NULL;
	ext4_fsblk_t goal_block;
3214 3215 3216 3217 3218 3219 3220

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

	/* first, try per-file preallocation */
	rcu_read_lock();
3221
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3222 3223 3224 3225 3226 3227 3228

		/* 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;

3229
		/* non-extent files can't have physical blocks past 2^32 */
3230
		if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)) &&
3231 3232 3233
			pa->pa_pstart + pa->pa_len > EXT4_MAX_BLOCK_FILE_PHYS)
			continue;

3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255
		/* 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;
3256 3257 3258 3259 3260
	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;

3261
	goal_block = ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex);
3262 3263 3264 3265
	/*
	 * search for the prealloc space that is having
	 * minimal distance from the goal block.
	 */
3266 3267 3268 3269 3270 3271 3272
	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) {
3273 3274 3275

				cpa = ext4_mb_check_group_pa(goal_block,
								pa, cpa);
3276
			}
3277 3278
			spin_unlock(&pa->pa_lock);
		}
3279
		rcu_read_unlock();
3280
	}
3281 3282 3283 3284 3285
	if (cpa) {
		ext4_mb_use_group_pa(ac, cpa);
		ac->ac_criteria = 20;
		return 1;
	}
3286 3287 3288
	return 0;
}

3289 3290 3291 3292
/*
 * 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
3293
 * Need to be called with the ext4 group lock held
3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306
 */
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);
3307
		mb_set_bits(bitmap, entry->start_blk, entry->count);
3308 3309 3310 3311 3312
		n = rb_next(n);
	}
	return;
}

3313 3314 3315
/*
 * the function goes through all preallocation in this group and marks them
 * used in in-core bitmap. buddy must be generated from this bitmap
3316
 * Need to be called with ext4 group lock held
3317
 */
3318 3319
static noinline_for_stack
void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348
					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);
3349
		mb_set_bits(bitmap, start, len);
3350 3351 3352
		preallocated += len;
		count++;
	}
3353
	mb_debug(1, "prellocated %u for group %u\n", preallocated, group);
3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369
}

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)
{
3370
	ext4_group_t grp;
3371
	ext4_fsblk_t grp_blk;
3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385

	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);

3386
	grp_blk = pa->pa_pstart;
3387
	/*
3388 3389 3390 3391
	 * 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)
3392 3393 3394
		grp_blk--;

	ext4_get_group_no_and_offset(sb, grp_blk, &grp, NULL);
3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423

	/*
	 * 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
 */
3424 3425
static noinline_for_stack int
ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482
{
	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);
3483 3484
	INIT_LIST_HEAD(&pa->pa_inode_list);
	INIT_LIST_HEAD(&pa->pa_group_list);
3485
	pa->pa_deleted = 0;
3486
	pa->pa_type = MB_INODE_PA;
3487

3488
	mb_debug(1, "new inode pa %p: %llu/%u for %u\n", pa,
3489
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
3490
	trace_ext4_mb_new_inode_pa(ac, pa);
3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514

	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
 */
3515 3516
static noinline_for_stack int
ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542
{
	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);
3543
	INIT_LIST_HEAD(&pa->pa_inode_list);
3544
	INIT_LIST_HEAD(&pa->pa_group_list);
3545
	pa->pa_deleted = 0;
3546
	pa->pa_type = MB_GROUP_PA;
3547

3548
	mb_debug(1, "new group pa %p: %llu/%u for %u\n", pa,
3549 3550
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
	trace_ext4_mb_new_group_pa(ac, pa);
3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565

	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);

3566 3567 3568 3569
	/*
	 * We will later add the new pa to the right bucket
	 * after updating the pa_free in ext4_mb_release_context
	 */
3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591
	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
 */
3592 3593
static noinline_for_stack int
ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3594
			struct ext4_prealloc_space *pa)
3595 3596 3597
{
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
3598 3599
	unsigned int end;
	unsigned int next;
3600 3601
	ext4_group_t group;
	ext4_grpblk_t bit;
3602
	unsigned long long grp_blk_start;
3603 3604 3605 3606 3607
	int err = 0;
	int free = 0;

	BUG_ON(pa->pa_deleted == 0);
	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
3608
	grp_blk_start = pa->pa_pstart - bit;
3609 3610 3611 3612
	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
	end = bit + pa->pa_len;

	while (bit < end) {
3613
		bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
3614 3615
		if (bit >= end)
			break;
3616
		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
3617
		mb_debug(1, "    free preallocated %u/%u in group %u\n",
3618 3619
			 (unsigned) ext4_group_first_block_no(sb, group) + bit,
			 (unsigned) next - bit, (unsigned) group);
3620 3621
		free += next - bit;

3622 3623 3624
		trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
		trace_ext4_mb_release_inode_pa(sb, pa->pa_inode, pa,
					       grp_blk_start + bit, next - bit);
3625 3626 3627 3628
		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
		bit = next + 1;
	}
	if (free != pa->pa_free) {
3629
		printk(KERN_CRIT "pa %p: logic %lu, phys. %lu, len %lu\n",
3630 3631 3632
			pa, (unsigned long) pa->pa_lstart,
			(unsigned long) pa->pa_pstart,
			(unsigned long) pa->pa_len);
3633
		ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
3634
					free, pa->pa_free);
3635 3636 3637 3638
		/*
		 * pa is already deleted so we use the value obtained
		 * from the bitmap and continue.
		 */
3639 3640 3641 3642 3643 3644
	}
	atomic_add(free, &sbi->s_mb_discarded);

	return err;
}

3645 3646
static noinline_for_stack int
ext4_mb_release_group_pa(struct ext4_buddy *e4b,
3647
				struct ext4_prealloc_space *pa)
3648 3649 3650 3651 3652
{
	struct super_block *sb = e4b->bd_sb;
	ext4_group_t group;
	ext4_grpblk_t bit;

3653
	trace_ext4_mb_release_group_pa(sb, pa);
3654 3655 3656 3657 3658
	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);
3659
	trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672

	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
 */
3673 3674
static noinline_for_stack int
ext4_mb_discard_group_preallocations(struct super_block *sb,
3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685
					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;
	struct list_head list;
	struct ext4_buddy e4b;
	int err;
	int busy = 0;
	int free = 0;

3686
	mb_debug(1, "discard preallocation for group %u\n", group);
3687 3688 3689 3690

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

3691
	bitmap_bh = ext4_read_block_bitmap(sb, group);
3692
	if (bitmap_bh == NULL) {
3693
		ext4_error(sb, "Error reading block bitmap for %u", group);
3694
		return 0;
3695 3696 3697
	}

	err = ext4_mb_load_buddy(sb, group, &e4b);
3698
	if (err) {
3699
		ext4_error(sb, "Error loading buddy information for %u", group);
3700 3701 3702
		put_bh(bitmap_bh);
		return 0;
	}
3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760

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

	INIT_LIST_HEAD(&list);
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);

3761
		if (pa->pa_type == MB_GROUP_PA)
3762
			ext4_mb_release_group_pa(&e4b, pa);
3763
		else
3764
			ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3765 3766 3767 3768 3769 3770 3771

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

out:
	ext4_unlock_group(sb, group);
3772
	ext4_mb_unload_buddy(&e4b);
3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785
	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
 */
3786
void ext4_discard_preallocations(struct inode *inode)
3787 3788 3789 3790 3791 3792 3793 3794 3795 3796
{
	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;
	ext4_group_t group = 0;
	struct list_head list;
	struct ext4_buddy e4b;
	int err;

3797
	if (!S_ISREG(inode->i_mode)) {
3798 3799 3800 3801
		/*BUG_ON(!list_empty(&ei->i_prealloc_list));*/
		return;
	}

3802
	mb_debug(1, "discard preallocation for inode %lu\n", inode->i_ino);
3803
	trace_ext4_discard_preallocations(inode);
3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855

	INIT_LIST_HEAD(&list);

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) {
3856
		BUG_ON(pa->pa_type != MB_INODE_PA);
3857 3858 3859
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, NULL);

		err = ext4_mb_load_buddy(sb, group, &e4b);
3860
		if (err) {
3861 3862
			ext4_error(sb, "Error loading buddy information for %u",
					group);
3863 3864
			continue;
		}
3865

3866
		bitmap_bh = ext4_read_block_bitmap(sb, group);
3867
		if (bitmap_bh == NULL) {
3868 3869
			ext4_error(sb, "Error reading block bitmap for %u",
					group);
3870
			ext4_mb_unload_buddy(&e4b);
3871
			continue;
3872 3873 3874 3875
		}

		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
3876
		ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3877 3878
		ext4_unlock_group(sb, group);

3879
		ext4_mb_unload_buddy(&e4b);
3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899
		put_bh(bitmap_bh);

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

/*
 * 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));
}
3900
#ifdef CONFIG_EXT4_DEBUG
3901 3902 3903
static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
{
	struct super_block *sb = ac->ac_sb;
3904
	ext4_group_t ngroups, i;
3905

3906 3907 3908
	if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
		return;

3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930
	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");
3931 3932
	ngroups = ext4_get_groups_count(sb);
	for (i = 0; i < ngroups; i++) {
3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944
		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);
3945 3946
			printk(KERN_ERR "PA:%u:%d:%u \n", i,
			       start, pa->pa_len);
3947
		}
3948
		ext4_unlock_group(sb, i);
3949 3950 3951

		if (grp->bb_free == 0)
			continue;
3952
		printk(KERN_ERR "%u: %d/%d \n",
3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968
		       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
3969
 * One can tune this size via /sys/fs/ext4/<partition>/mb_stream_req
3970 3971 3972 3973 3974 3975 3976 3977 3978 3979
 */
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;

3980 3981 3982
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		return;

3983
	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
3984 3985
	isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1)
		>> bsbits;
3986

3987 3988 3989 3990 3991 3992 3993
	if ((size == isize) &&
	    !ext4_fs_is_busy(sbi) &&
	    (atomic_read(&ac->ac_inode->i_writecount) == 0)) {
		ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC;
		return;
	}

3994
	/* don't use group allocation for large files */
3995
	size = max(size, isize);
3996
	if (size > sbi->s_mb_stream_request) {
3997
		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
3998
		return;
3999
	}
4000 4001 4002 4003 4004 4005 4006

	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.
	 */
4007
	ac->ac_lg = __this_cpu_ptr(sbi->s_locality_groups);
4008 4009 4010 4011 4012 4013 4014 4015

	/* 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);
}

4016 4017
static noinline_for_stack int
ext4_mb_initialize_context(struct ext4_allocation_context *ac,
4018 4019 4020 4021 4022 4023
				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;
4024 4025
	unsigned int len;
	ext4_fsblk_t goal;
4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042
	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 */
4043
	memset(ac, 0, sizeof(struct ext4_allocation_context));
4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061
	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);

4062
	mb_debug(1, "init ac: %u blocks @ %u, goal %u, flags %x, 2^%d, "
4063 4064 4065 4066 4067 4068 4069 4070 4071 4072
			"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;

}

4073 4074 4075 4076 4077 4078 4079 4080 4081 4082
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;

4083
	mb_debug(1, "discard locality group preallocation\n");
4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104

	INIT_LIST_HEAD(&discard_list);

	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 */
4105
		BUG_ON(pa->pa_type != MB_GROUP_PA);
4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130

		/* 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)) {
4131 4132
			ext4_error(sb, "Error loading buddy information for %u",
					group);
4133 4134 4135 4136
			continue;
		}
		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
4137
		ext4_mb_release_group_pa(&e4b, pa);
4138 4139
		ext4_unlock_group(sb, group);

4140
		ext4_mb_unload_buddy(&e4b);
4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171
		list_del(&pa->u.pa_tmp_list);
		call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
	}
}

/*
 * 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) {
4172
			spin_unlock(&tmp_pa->pa_lock);
4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201
			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 ;
}

4202 4203 4204 4205 4206
/*
 * release all resource we used in allocation
 */
static int ext4_mb_release_context(struct ext4_allocation_context *ac)
{
4207 4208
	struct ext4_prealloc_space *pa = ac->ac_pa;
	if (pa) {
4209
		if (pa->pa_type == MB_GROUP_PA) {
4210
			/* see comment in ext4_mb_use_group_pa() */
4211 4212 4213 4214 4215 4216
			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);
4217 4218
		}
	}
4219 4220
	if (ac->alloc_semp)
		up_read(ac->alloc_semp);
A
Aneesh Kumar K.V 已提交
4221 4222 4223 4224 4225 4226 4227 4228
	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()
		 */
4229
		if ((pa->pa_type == MB_GROUP_PA) && likely(pa->pa_free)) {
A
Aneesh Kumar K.V 已提交
4230 4231 4232 4233 4234 4235 4236
			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);
	}
4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248
	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)
{
4249
	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4250 4251 4252
	int ret;
	int freed = 0;

4253
	trace_ext4_mb_discard_preallocations(sb, needed);
4254
	for (i = 0; i < ngroups && needed > 0; i++) {
4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268
		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,
4269
				struct ext4_allocation_request *ar, int *errp)
4270
{
4271
	int freed;
4272
	struct ext4_allocation_context *ac = NULL;
4273 4274 4275
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block = 0;
4276
	unsigned int inquota = 0;
4277
	unsigned int reserv_blks = 0;
4278 4279 4280 4281

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

4282
	trace_ext4_request_blocks(ar);
4283

4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294
	/*
	 * 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.
4295
		 */
A
Aneesh Kumar K.V 已提交
4296 4297 4298 4299 4300 4301
		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) {
4302 4303 4304
			*errp = -ENOSPC;
			return 0;
		}
4305
		reserv_blks = ar->len;
4306
		while (ar->len && dquot_alloc_block(ar->inode, ar->len)) {
4307 4308 4309 4310 4311 4312
			ar->flags |= EXT4_MB_HINT_NOPREALLOC;
			ar->len--;
		}
		inquota = ar->len;
		if (ar->len == 0) {
			*errp = -EDQUOT;
4313
			goto out;
4314
		}
4315
	}
4316

4317
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4318
	if (!ac) {
4319
		ar->len = 0;
4320
		*errp = -ENOMEM;
4321
		goto out;
4322 4323 4324
	}

	*errp = ext4_mb_initialize_context(ac, ar);
4325 4326
	if (*errp) {
		ar->len = 0;
4327
		goto out;
4328 4329
	}

4330 4331 4332 4333
	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);
4334 4335
repeat:
		/* allocate space in core */
4336 4337 4338
		*errp = ext4_mb_regular_allocator(ac);
		if (*errp)
			goto errout;
4339 4340 4341 4342

		/* as we've just preallocated more space than
		 * user requested orinally, we store allocated
		 * space in a special descriptor */
4343 4344 4345
		if (ac->ac_status == AC_STATUS_FOUND &&
				ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
			ext4_mb_new_preallocation(ac);
4346
	}
4347
	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
4348
		*errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_blks);
4349
		if (*errp == -EAGAIN) {
4350 4351 4352 4353 4354
			/*
			 * drop the reference that we took
			 * in ext4_mb_use_best_found
			 */
			ext4_mb_release_context(ac);
4355 4356 4357 4358 4359
			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;
4360 4361
		} else if (*errp)
		errout:
4362
			ext4_discard_allocated_blocks(ac);
4363
		else {
4364 4365 4366
			block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
			ar->len = ac->ac_b_ex.fe_len;
		}
4367
	} else {
4368
		freed  = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
4369 4370 4371
		if (freed)
			goto repeat;
		*errp = -ENOSPC;
4372 4373 4374
	}

	if (*errp) {
4375
		ac->ac_b_ex.fe_len = 0;
4376
		ar->len = 0;
4377
		ext4_mb_show_ac(ac);
4378
	}
4379
	ext4_mb_release_context(ac);
4380 4381 4382
out:
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
4383
	if (inquota && ar->len < inquota)
4384
		dquot_free_block(ar->inode, inquota - ar->len);
4385 4386 4387 4388 4389 4390
	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);
	}
4391

4392
	trace_ext4_allocate_blocks(ar, (unsigned long long)block);
4393

4394 4395 4396
	return block;
}

4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411
/*
 * 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;
}

4412 4413
static noinline_for_stack int
ext4_mb_free_metadata(handle_t *handle, struct ext4_buddy *e4b,
4414
		      struct ext4_free_data *new_entry)
4415
{
4416
	ext4_group_t group = e4b->bd_group;
4417 4418
	ext4_grpblk_t block;
	struct ext4_free_data *entry;
4419 4420 4421
	struct ext4_group_info *db = e4b->bd_info;
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
4422 4423 4424
	struct rb_node **n = &db->bb_free_root.rb_node, *node;
	struct rb_node *parent = NULL, *new_node;

4425
	BUG_ON(!ext4_handle_valid(handle));
4426 4427 4428
	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

4429
	new_node = &new_entry->node;
4430
	block = new_entry->start_blk;
4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448

	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 {
4449 4450 4451
			ext4_grp_locked_error(sb, group, 0,
				ext4_group_first_block_no(sb, group) + block,
				"Block already on to-be-freed list");
4452
			return 0;
4453
		}
4454
	}
4455

4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470
	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);
4471
		}
4472
	}
4473

4474 4475 4476 4477 4478 4479 4480 4481 4482 4483
	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);
4484 4485
		}
	}
4486
	/* Add the extent to transaction's private list */
4487
	spin_lock(&sbi->s_md_lock);
4488
	list_add(&new_entry->list, &handle->h_transaction->t_private_list);
4489
	spin_unlock(&sbi->s_md_lock);
4490 4491 4492
	return 0;
}

4493 4494 4495 4496 4497 4498 4499
/**
 * ext4_free_blocks() -- Free given blocks and update quota
 * @handle:		handle for this transaction
 * @inode:		inode
 * @block:		start physical block to free
 * @count:		number of blocks to count
 * @metadata: 		Are these metadata blocks
4500
 */
4501
void ext4_free_blocks(handle_t *handle, struct inode *inode,
4502 4503
		      struct buffer_head *bh, ext4_fsblk_t block,
		      unsigned long count, int flags)
4504
{
4505
	struct buffer_head *bitmap_bh = NULL;
4506 4507
	struct super_block *sb = inode->i_sb;
	struct ext4_group_desc *gdp;
4508
	unsigned long freed = 0;
4509
	unsigned int overflow;
4510 4511 4512 4513 4514 4515 4516 4517
	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;

4518 4519 4520 4521 4522 4523
	if (bh) {
		if (block)
			BUG_ON(block != bh->b_blocknr);
		else
			block = bh->b_blocknr;
	}
4524 4525

	sbi = EXT4_SB(sb);
4526 4527
	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
	    !ext4_data_block_valid(sbi, block, count)) {
4528
		ext4_error(sb, "Freeing blocks not in datazone - "
4529
			   "block = %llu, count = %lu", block, count);
4530 4531 4532
		goto error_return;
	}

4533
	ext4_debug("freeing block %llu\n", block);
4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545
	trace_ext4_free_blocks(inode, block, count, flags);

	if (flags & EXT4_FREE_BLOCKS_FORGET) {
		struct buffer_head *tbh = bh;
		int i;

		BUG_ON(bh && (count > 1));

		for (i = 0; i < count; i++) {
			if (!bh)
				tbh = sb_find_get_block(inode->i_sb,
							block + i);
4546
			ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
4547 4548 4549 4550
				    inode, tbh, block + i);
		}
	}

4551
	/*
4552 4553 4554 4555 4556 4557 4558 4559
	 * We need to make sure we don't reuse the freed block until
	 * after the transaction is committed, which we can do by
	 * treating the block as metadata, below.  We make an
	 * exception if the inode is to be written in writeback mode
	 * since writeback mode has weak data consistency guarantees.
	 */
	if (!ext4_should_writeback_data(inode))
		flags |= EXT4_FREE_BLOCKS_METADATA;
4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572

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;
	}
4573
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4574 4575
	if (!bitmap_bh) {
		err = -EIO;
4576
		goto error_return;
4577
	}
4578
	gdp = ext4_get_group_desc(sb, block_group, &gd_bh);
4579 4580
	if (!gdp) {
		err = -EIO;
4581
		goto error_return;
4582
	}
4583 4584 4585 4586 4587 4588 4589 4590

	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)) {

4591
		ext4_error(sb, "Freeing blocks in system zone - "
4592
			   "Block = %llu, count = %lu", block, count);
4593 4594
		/* err = 0. ext4_std_error should be a no op */
		goto error_return;
4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617
	}

	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
4618
	trace_ext4_mballoc_free(sb, inode, block_group, bit, count);
4619

4620 4621 4622
	err = ext4_mb_load_buddy(sb, block_group, &e4b);
	if (err)
		goto error_return;
4623 4624

	if ((flags & EXT4_FREE_BLOCKS_METADATA) && ext4_handle_valid(handle)) {
4625 4626 4627 4628 4629 4630 4631 4632 4633 4634
		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;
4635

4636
		ext4_lock_group(sb, block_group);
4637
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4638
		ext4_mb_free_metadata(handle, &e4b, new_entry);
4639
	} else {
4640 4641 4642 4643
		/* need to update group_info->bb_free and bitmap
		 * with group lock held. generate_buddy look at
		 * them with group lock_held
		 */
4644 4645
		if (test_opt(sb, DISCARD))
			ext4_issue_discard(sb, block_group, bit, count);
4646 4647
		ext4_lock_group(sb, block_group);
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4648
		mb_free_blocks(inode, &e4b, bit, count);
4649 4650 4651
		ext4_mb_return_to_preallocation(inode, &e4b, block, count);
	}

4652 4653
	ret = ext4_free_blks_count(sb, gdp) + count;
	ext4_free_blks_set(sb, gdp, ret);
4654
	gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
4655
	ext4_unlock_group(sb, block_group);
4656 4657
	percpu_counter_add(&sbi->s_freeblocks_counter, count);

4658 4659
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4660
		atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks);
4661 4662
	}

4663
	ext4_mb_unload_buddy(&e4b);
4664

4665
	freed += count;
4666

4667 4668 4669 4670
	/* We dirtied the bitmap block */
	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);

4671 4672
	/* And the group descriptor block */
	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
4673
	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
4674 4675 4676 4677 4678 4679 4680 4681 4682
	if (!err)
		err = ret;

	if (overflow && !err) {
		block += count;
		count = overflow;
		put_bh(bitmap_bh);
		goto do_more;
	}
T
Theodore Ts'o 已提交
4683
	ext4_mark_super_dirty(sb);
4684
error_return:
4685
	if (freed)
4686
		dquot_free_block(inode, freed);
4687 4688 4689 4690
	brelse(bitmap_bh);
	ext4_std_error(sb, err);
	return;
}