mballoc.c 136.1 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
 * 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
78 79
 * space we will consume the particular prealloc space. This makes sure that
 * we have contiguous physical blocks representing the file blocks
80 81 82 83 84 85 86
 *
 * 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
87
 * prealloc space. These are per CPU prealloc list represented as
88 89 90 91 92 93 94
 *
 * 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
L
Lucas De Marchi 已提交
95
 * enough free space (pa_free) within the prealloc space.
96 97 98 99 100 101 102 103 104 105 106 107
 *
 * 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
131
 * /sys/fs/ext4/<partition>/mb_group_prealloc. The value is represented in
132 133
 * 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
134 135
 * the smallest multiple of the stripe value (sbi->s_stripe) which is
 * greater than the default mb_group_prealloc.
136
 *
137
 * The regular allocator (using the buddy cache) supports a few tunables.
138
 *
T
Theodore Ts'o 已提交
139 140 141
 * /sys/fs/ext4/<partition>/mb_min_to_scan
 * /sys/fs/ext4/<partition>/mb_max_to_scan
 * /sys/fs/ext4/<partition>/mb_order2_req
142
 *
T
Theodore Ts'o 已提交
143
 * The regular allocator uses buddy scan only if the request len is power of
144 145
 * 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
146
 * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
147
 * stripe size (sbi->s_stripe), we try to search for contiguous block in
T
Theodore Ts'o 已提交
148 149 150
 * 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.
151
 * min_to_scan indicate how long the mballoc __must__ look for a best
T
Theodore Ts'o 已提交
152
 * extent and max_to_scan indicates how long the mballoc __can__ look for a
153 154 155
 * 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
156
 * can be used for allocation. ext4_mb_good_group explains how the groups are
157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338
 * 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
 *
 */
339 340 341
static struct kmem_cache *ext4_pspace_cachep;
static struct kmem_cache *ext4_ac_cachep;
static struct kmem_cache *ext4_free_ext_cachep;
342 343 344 345

/* 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 */
346
#define NR_GRPINFO_CACHES 8
347 348
static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];

349 350 351 352 353 354
static const char *ext4_groupinfo_slab_names[NR_GRPINFO_CACHES] = {
	"ext4_groupinfo_1k", "ext4_groupinfo_2k", "ext4_groupinfo_4k",
	"ext4_groupinfo_8k", "ext4_groupinfo_16k", "ext4_groupinfo_32k",
	"ext4_groupinfo_64k", "ext4_groupinfo_128k"
};

355 356
static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
					ext4_group_t group);
357 358
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group);
359 360
static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);

361 362
static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
{
363
#if BITS_PER_LONG == 64
364 365
	*bit += ((unsigned long) addr & 7UL) << 3;
	addr = (void *) ((unsigned long) addr & ~7UL);
366
#elif BITS_PER_LONG == 32
367 368
	*bit += ((unsigned long) addr & 3UL) << 3;
	addr = (void *) ((unsigned long) addr & ~3UL);
369 370 371
#else
#error "how many bits you are?!"
#endif
372 373
	return addr;
}
374 375 376 377 378 379 380

static inline int mb_test_bit(int bit, void *addr)
{
	/*
	 * ext4_test_bit on architecture like powerpc
	 * needs unsigned long aligned address
	 */
381
	addr = mb_correct_addr_and_bit(&bit, addr);
382 383 384 385 386
	return ext4_test_bit(bit, addr);
}

static inline void mb_set_bit(int bit, void *addr)
{
387
	addr = mb_correct_addr_and_bit(&bit, addr);
388 389 390 391 392
	ext4_set_bit(bit, addr);
}

static inline void mb_clear_bit(int bit, void *addr)
{
393
	addr = mb_correct_addr_and_bit(&bit, addr);
394 395 396
	ext4_clear_bit(bit, addr);
}

397 398
static inline int mb_find_next_zero_bit(void *addr, int max, int start)
{
399
	int fix = 0, ret, tmpmax;
400
	addr = mb_correct_addr_and_bit(&fix, addr);
401
	tmpmax = max + fix;
402 403
	start += fix;

404 405 406 407
	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
408 409 410 411
}

static inline int mb_find_next_bit(void *addr, int max, int start)
{
412
	int fix = 0, ret, tmpmax;
413
	addr = mb_correct_addr_and_bit(&fix, addr);
414
	tmpmax = max + fix;
415 416
	start += fix;

417 418 419 420
	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
421 422
}

423 424 425 426 427 428 429 430 431 432 433 434 435
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 */
C
Coly Li 已提交
436 437
	if (order == 0) {
		*max = 1 << (e4b->bd_blkbits + 3);
438
		return EXT4_MB_BITMAP(e4b);
C
Coly Li 已提交
439
	}
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455

	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;
456
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
457 458 459
	for (i = 0; i < count; i++) {
		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
			ext4_fsblk_t blocknr;
460 461

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
462
			blocknr += first + i;
463
			ext4_grp_locked_error(sb, e4b->bd_group,
464 465 466 467 468
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing block already freed "
					      "(bit %u)",
					      first + i);
469 470 471 472 473 474 475 476 477 478 479
		}
		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;
480
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
	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]) {
496 497 498 499 500
				ext4_msg(e4b->bd_sb, KERN_ERR,
					 "corruption in group %u "
					 "at byte %u(%u): %x in copy != %x "
					 "on disk/prealloc",
					 e4b->bd_group, i, i * 8, b1[i], b2[i]);
501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624
				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);
	list_for_each(cur, &grp->bb_prealloc_list) {
		ext4_group_t groupnr;
		struct ext4_prealloc_space *pa;
625 626
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
627
		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
628
		for (i = 0; i < pa->pa_len; i++)
629 630 631 632 633 634
			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
	}
	return 0;
}
#undef MB_CHECK_ASSERT
#define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
635
					__FILE__, __func__, __LINE__)
636 637 638 639
#else
#define mb_check_buddy(e4b)
#endif

640 641 642 643 644 645
/*
 * Divide blocks started from @first with length @len into
 * smaller chunks with power of 2 blocks.
 * Clear the bits in bitmap which the blocks of the chunk(s) covered,
 * then increase bb_counters[] for corresponded chunk size.
 */
646
static void ext4_mb_mark_free_simple(struct super_block *sb,
647
				void *buddy, ext4_grpblk_t first, ext4_grpblk_t len,
648 649 650
					struct ext4_group_info *grp)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
651 652 653
	ext4_grpblk_t min;
	ext4_grpblk_t max;
	ext4_grpblk_t chunk;
654 655
	unsigned short border;

656
	BUG_ON(len > EXT4_BLOCKS_PER_GROUP(sb));
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681

	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;
	}
}

682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702
/*
 * 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;
		}
	}
}

703 704
static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
705 706 707
				void *buddy, void *bitmap, ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
708 709 710 711
	ext4_grpblk_t max = EXT4_BLOCKS_PER_GROUP(sb);
	ext4_grpblk_t i = 0;
	ext4_grpblk_t first;
	ext4_grpblk_t len;
712 713 714 715 716 717
	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 */
718
	i = mb_find_next_zero_bit(bitmap, max, 0);
719 720 721 722
	grp->bb_first_free = i;
	while (i < max) {
		fragments++;
		first = i;
723
		i = mb_find_next_bit(bitmap, max, i);
724 725 726 727 728 729 730
		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)
731
			i = mb_find_next_zero_bit(bitmap, max, i);
732 733 734 735
	}
	grp->bb_fragments = fragments;

	if (free != grp->bb_free) {
736 737 738
		ext4_grp_locked_error(sb, group, 0, 0,
				      "%u blocks in bitmap, %u in gd",
				      free, grp->bb_free);
739 740 741 742
		/*
		 * If we intent to continue, we consider group descritor
		 * corrupt and update bb_free using bitmap value
		 */
743 744
		grp->bb_free = free;
	}
745
	mb_set_largest_free_order(sb, grp);
746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762

	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                        }
763
 * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
764 765 766 767 768 769 770
 *
 *
 * 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
771 772 773
 *
 * Locking note:  This routine takes the block group lock of all groups
 * for this page; do not hold this lock when calling this routine!
774 775 776 777
 */

static int ext4_mb_init_cache(struct page *page, char *incore)
{
778
	ext4_group_t ngroups;
779 780 781 782 783 784 785 786 787 788 789 790 791
	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;
792
	struct ext4_group_info *grinfo;
793

794
	mb_debug(1, "init page %lu\n", page->index);
795 796 797

	inode = page->mapping->host;
	sb = inode->i_sb;
798
	ngroups = ext4_get_groups_count(sb);
799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821
	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;

822
		if (first_group + i >= ngroups)
823 824
			break;

825 826 827 828 829 830 831 832 833 834 835 836
		grinfo = ext4_get_group_info(sb, first_group + i);
		/*
		 * If page is uptodate then we came here after online resize
		 * which added some new uninitialized group info structs, so
		 * we must skip all initialized uptodate buddies on the page,
		 * which may be currently in use by an allocating task.
		 */
		if (PageUptodate(page) && !EXT4_MB_GRP_NEED_INIT(grinfo)) {
			bh[i] = NULL;
			continue;
		}

837 838 839 840 841 842 843 844 845 846
		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;

847
		if (bitmap_uptodate(bh[i]))
848 849
			continue;

850
		lock_buffer(bh[i]);
851 852 853 854
		if (bitmap_uptodate(bh[i])) {
			unlock_buffer(bh[i]);
			continue;
		}
855
		ext4_lock_group(sb, first_group + i);
856 857 858
		if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
			ext4_init_block_bitmap(sb, bh[i],
						first_group + i, desc);
859
			set_bitmap_uptodate(bh[i]);
860
			set_buffer_uptodate(bh[i]);
861
			ext4_unlock_group(sb, first_group + i);
A
Aneesh Kumar K.V 已提交
862
			unlock_buffer(bh[i]);
863 864
			continue;
		}
865
		ext4_unlock_group(sb, first_group + i);
866 867 868 869 870 871 872 873 874
		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;
		}
875
		get_bh(bh[i]);
876 877 878 879 880 881 882
		/*
		 * 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]);
883 884
		bh[i]->b_end_io = end_buffer_read_sync;
		submit_bh(READ, bh[i]);
885
		mb_debug(1, "read bitmap for group %u\n", first_group + i);
886 887 888
	}

	/* wait for I/O completion */
889 890 891
	for (i = 0; i < groups_per_page; i++)
		if (bh[i])
			wait_on_buffer(bh[i]);
892 893

	err = -EIO;
894 895
	for (i = 0; i < groups_per_page; i++)
		if (bh[i] && !buffer_uptodate(bh[i]))
896 897
			goto out;

898
	err = 0;
899 900 901 902 903
	first_block = page->index * blocks_per_page;
	for (i = 0; i < blocks_per_page; i++) {
		int group;

		group = (first_block + i) >> 1;
904
		if (group >= ngroups)
905 906
			break;

907 908 909 910
		if (!bh[group - first_group])
			/* skip initialized uptodate buddy */
			continue;

911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926
		/*
		 * 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);
927
			mb_debug(1, "put buddy for group %u in page %lu/%x\n",
928
				group, page->index, i * blocksize);
929
			trace_ext4_mb_buddy_bitmap_load(sb, group);
930 931 932
			grinfo = ext4_get_group_info(sb, group);
			grinfo->bb_fragments = 0;
			memset(grinfo->bb_counters, 0,
933 934
			       sizeof(*grinfo->bb_counters) *
				(sb->s_blocksize_bits+2));
935 936 937
			/*
			 * incore got set to the group block bitmap below
			 */
938
			ext4_lock_group(sb, group);
939 940
			/* init the buddy */
			memset(data, 0xff, blocksize);
941
			ext4_mb_generate_buddy(sb, data, incore, group);
942
			ext4_unlock_group(sb, group);
943 944 945 946
			incore = NULL;
		} else {
			/* this is block of bitmap */
			BUG_ON(incore != NULL);
947
			mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
948
				group, page->index, i * blocksize);
949
			trace_ext4_mb_bitmap_load(sb, group);
950 951 952 953 954 955 956

			/* 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);
957
			ext4_mb_generate_from_freelist(sb, data, group);
958 959 960 961 962 963 964 965 966 967 968 969
			ext4_unlock_group(sb, group);

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

out:
	if (bh) {
970
		for (i = 0; i < groups_per_page; i++)
971 972 973 974 975 976 977
			brelse(bh[i]);
		if (bh != &bhs)
			kfree(bh);
	}
	return err;
}

978
/*
979 980 981 982
 * Lock the buddy and bitmap pages. This make sure other parallel init_group
 * on the same buddy page doesn't happen whild holding the buddy page lock.
 * Return locked buddy and bitmap pages on e4b struct. If buddy and bitmap
 * are on the same page e4b->bd_buddy_page is NULL and return value is 0.
983
 */
984 985
static int ext4_mb_get_buddy_page_lock(struct super_block *sb,
		ext4_group_t group, struct ext4_buddy *e4b)
986
{
987 988
	struct inode *inode = EXT4_SB(sb)->s_buddy_cache;
	int block, pnum, poff;
989
	int blocks_per_page;
990 991 992 993
	struct page *page;

	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;
994 995 996 997 998 999 1000 1001 1002

	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;
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (!page)
		return -EIO;
	BUG_ON(page->mapping != inode->i_mapping);
	e4b->bd_bitmap_page = page;
	e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);

	if (blocks_per_page >= 2) {
		/* buddy and bitmap are on the same page */
		return 0;
1014
	}
1015 1016 1017 1018 1019 1020 1021 1022 1023 1024

	block++;
	pnum = block / blocks_per_page;
	poff = block % blocks_per_page;
	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
	if (!page)
		return -EIO;
	BUG_ON(page->mapping != inode->i_mapping);
	e4b->bd_buddy_page = page;
	return 0;
1025 1026
}

1027
static void ext4_mb_put_buddy_page_lock(struct ext4_buddy *e4b)
1028
{
1029 1030 1031 1032 1033 1034 1035
	if (e4b->bd_bitmap_page) {
		unlock_page(e4b->bd_bitmap_page);
		page_cache_release(e4b->bd_bitmap_page);
	}
	if (e4b->bd_buddy_page) {
		unlock_page(e4b->bd_buddy_page);
		page_cache_release(e4b->bd_buddy_page);
1036 1037 1038
	}
}

1039 1040 1041 1042 1043
/*
 * 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!
 */
1044 1045 1046 1047 1048
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
{

	struct ext4_group_info *this_grp;
1049 1050 1051
	struct ext4_buddy e4b;
	struct page *page;
	int ret = 0;
1052 1053 1054 1055

	mb_debug(1, "init group %u\n", group);
	this_grp = ext4_get_group_info(sb, group);
	/*
1056 1057 1058 1059
	 * 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
1060
	 * would have pinned buddy page to page cache.
1061
	 */
1062 1063
	ret = ext4_mb_get_buddy_page_lock(sb, group, &e4b);
	if (ret || !EXT4_MB_GRP_NEED_INIT(this_grp)) {
1064 1065 1066 1067 1068 1069
		/*
		 * somebody initialized the group
		 * return without doing anything
		 */
		goto err;
	}
1070 1071 1072 1073 1074 1075

	page = e4b.bd_bitmap_page;
	ret = ext4_mb_init_cache(page, NULL);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
1076 1077 1078 1079 1080
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);

1081
	if (e4b.bd_buddy_page == NULL) {
1082 1083 1084 1085 1086
		/*
		 * If both the bitmap and buddy are in
		 * the same page we don't need to force
		 * init the buddy
		 */
1087 1088
		ret = 0;
		goto err;
1089
	}
1090 1091 1092 1093 1094 1095
	/* init buddy cache */
	page = e4b.bd_buddy_page;
	ret = ext4_mb_init_cache(page, e4b.bd_bitmap);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
1096 1097 1098 1099 1100
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);
err:
1101
	ext4_mb_put_buddy_page_lock(&e4b);
1102 1103 1104
	return ret;
}

1105 1106 1107 1108 1109
/*
 * 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!
 */
1110 1111 1112
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
					struct ext4_buddy *e4b)
1113 1114 1115 1116 1117 1118
{
	int blocks_per_page;
	int block;
	int pnum;
	int poff;
	struct page *page;
1119
	int ret;
1120 1121 1122
	struct ext4_group_info *grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;
1123

1124
	mb_debug(1, "load group %u\n", group);
1125 1126

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1127
	grp = ext4_get_group_info(sb, group);
1128 1129

	e4b->bd_blkbits = sb->s_blocksize_bits;
1130
	e4b->bd_info = grp;
1131 1132 1133 1134 1135
	e4b->bd_sb = sb;
	e4b->bd_group = group;
	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;

1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
	if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
		/*
		 * we need full data about the group
		 * to make a good selection
		 */
		ret = ext4_mb_init_group(sb, group);
		if (ret)
			return ret;
	}

1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
	/*
	 * 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)
1160 1161 1162 1163 1164 1165 1166 1167
			/*
			 * 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.
			 */
1168 1169 1170 1171 1172
			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)) {
1173 1174 1175 1176 1177
				ret = ext4_mb_init_cache(page, NULL);
				if (ret) {
					unlock_page(page);
					goto err;
				}
1178 1179 1180 1181 1182 1183
				mb_cmp_bitmaps(e4b, page_address(page) +
					       (poff * sb->s_blocksize));
			}
			unlock_page(page);
		}
	}
1184 1185
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1186
		goto err;
1187
	}
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
	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);
1203 1204 1205 1206 1207 1208 1209
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
				if (ret) {
					unlock_page(page);
					goto err;
				}
			}
1210 1211 1212
			unlock_page(page);
		}
	}
1213 1214
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1215
		goto err;
1216
	}
1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
	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:
1227 1228
	if (page)
		page_cache_release(page);
1229 1230 1231 1232 1233 1234
	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;
1235
	return ret;
1236 1237
}

1238
static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
{
	if (e4b->bd_bitmap_page)
		page_cache_release(e4b->bd_bitmap_page);
	if (e4b->bd_buddy_page)
		page_cache_release(e4b->bd_buddy_page);
}


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;
}

1268
static void mb_clear_bits(void *bm, int cur, int len)
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
{
	__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;
		}
1281
		mb_clear_bit(cur, bm);
1282 1283 1284 1285
		cur++;
	}
}

1286
void ext4_set_bits(void *bm, int cur, int len)
1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298
{
	__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;
		}
1299
		mb_set_bit(cur, bm);
1300 1301 1302 1303
		cur++;
	}
}

1304
static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
			  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));
1315
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
	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;
1340 1341

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
1342
			blocknr += block;
1343
			ext4_grp_locked_error(sb, e4b->bd_group,
1344 1345 1346 1347
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing already freed block "
					      "(bit %u)", block);
1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
		}
		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);
	}
1384
	mb_set_largest_free_order(sb, e4b->bd_info);
1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395
	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;

1396
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 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
	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);
1460
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
	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]++;
	}
1513
	mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
1514

1515
	ext4_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544
	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;

1545 1546 1547 1548 1549 1550 1551
	/*
	 * 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
	 */
1552 1553 1554 1555 1556
	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);
	/* store last allocated for subsequent stream allocation */
1557
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
		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;

1579 1580
	if (ac->ac_status == AC_STATUS_FOUND)
		return;
1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 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
	/*
	 * 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);
1627
	BUG_ON(ex->fe_len > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677
	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);
}

1678 1679
static noinline_for_stack
int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
					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);
1701
	ext4_mb_unload_buddy(e4b);
1702 1703 1704 1705

	return 0;
}

1706 1707
static noinline_for_stack
int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729
				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;

1730 1731
		start = ext4_group_first_block_no(ac->ac_sb, e4b->bd_group) +
			ex.fe_start;
1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755
		/* 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);
1756
	ext4_mb_unload_buddy(e4b);
1757 1758 1759 1760 1761 1762 1763 1764

	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
 */
1765 1766
static noinline_for_stack
void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783
					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);

1784
		k = mb_find_next_zero_bit(buddy, max, 0);
1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808
		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.
 */
1809 1810
static noinline_for_stack
void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824
					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) {
1825
		i = mb_find_next_zero_bit(bitmap,
1826 1827
						EXT4_BLOCKS_PER_GROUP(sb), i);
		if (i >= EXT4_BLOCKS_PER_GROUP(sb)) {
1828
			/*
1829
			 * IF we have corrupt bitmap, we won't find any
1830 1831 1832
			 * free blocks even though group info says we
			 * we have free blocks
			 */
1833 1834
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
					"%d free blocks as per "
1835
					"group info. But bitmap says 0",
1836
					free);
1837 1838 1839 1840 1841
			break;
		}

		mb_find_extent(e4b, 0, i, ac->ac_g_ex.fe_len, &ex);
		BUG_ON(ex.fe_len <= 0);
1842
		if (free < ex.fe_len) {
1843 1844
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
					"%d free blocks as per "
1845
					"group info. But got %d blocks",
1846
					free, ex.fe_len);
1847 1848 1849 1850 1851 1852
			/*
			 * The number of free blocks differs. This mostly
			 * indicate that the bitmap is corrupt. So exit
			 * without claiming the space.
			 */
			break;
1853
		}
1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865

		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
1866
 * we try to find stripe-aligned chunks for stripe-size-multiple requests
1867
 */
1868 1869
static noinline_for_stack
void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883
				 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 */
1884 1885
	first_group_block = ext4_group_first_block_no(sb, e4b->bd_group);

1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903
	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;
	}
}

1904
/* This is now called BEFORE we load the buddy bitmap. */
1905 1906 1907 1908
static int ext4_mb_good_group(struct ext4_allocation_context *ac,
				ext4_group_t group, int cr)
{
	unsigned free, fragments;
1909
	int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
1910 1911 1912
	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);

	BUG_ON(cr < 0 || cr >= 4);
1913 1914 1915 1916 1917 1918 1919

	/* 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;
	}
1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931

	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);

1932 1933 1934
		if (grp->bb_largest_free_order < ac->ac_2order)
			return 0;

1935 1936 1937 1938 1939 1940
		/* 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;

1941
		return 1;
1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958
	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;
}

1959 1960
static noinline_for_stack int
ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1961
{
1962
	ext4_group_t ngroups, group, i;
1963 1964 1965 1966 1967 1968 1969 1970
	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);
1971
	ngroups = ext4_get_groups_count(sb);
1972
	/* non-extent files are limited to low blocks/groups */
1973
	if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)))
1974 1975
		ngroups = sbi->s_blockfile_groups;

1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995
	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
1996
	 * You can tune it via /sys/fs/ext4/<partition>/mb_order2_req
1997 1998 1999 2000 2001 2002 2003 2004 2005
	 */
	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;
	}

2006 2007
	/* if stream allocation is enabled, use global goal */
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
2008 2009 2010 2011 2012 2013
		/* 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);
	}
2014

2015 2016 2017 2018 2019 2020 2021 2022 2023
	/* 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;
2024 2025 2026 2027 2028 2029
		/*
		 * searching for the right group start
		 * from the goal value specified
		 */
		group = ac->ac_g_ex.fe_group;

2030 2031
		for (i = 0; i < ngroups; group++, i++) {
			if (group == ngroups)
2032 2033
				group = 0;

2034 2035
			/* This now checks without needing the buddy page */
			if (!ext4_mb_good_group(ac, group, cr))
2036 2037 2038 2039 2040 2041 2042
				continue;

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

			ext4_lock_group(sb, group);
2043 2044 2045 2046 2047

			/*
			 * We need to check again after locking the
			 * block group
			 */
2048 2049
			if (!ext4_mb_good_group(ac, group, cr)) {
				ext4_unlock_group(sb, group);
2050
				ext4_mb_unload_buddy(&e4b);
2051 2052 2053 2054
				continue;
			}

			ac->ac_groups_scanned++;
2055
			if (cr == 0)
2056
				ext4_mb_simple_scan_group(ac, &e4b);
2057 2058
			else if (cr == 1 && sbi->s_stripe &&
					!(ac->ac_g_ex.fe_len % sbi->s_stripe))
2059 2060 2061 2062 2063
				ext4_mb_scan_aligned(ac, &e4b);
			else
				ext4_mb_complex_scan_group(ac, &e4b);

			ext4_unlock_group(sb, group);
2064
			ext4_mb_unload_buddy(&e4b);
2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104

			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;

2105
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2106 2107
		return NULL;
	group = *pos + 1;
2108
	return (void *) ((unsigned long) group);
2109 2110 2111 2112 2113 2114 2115 2116
}

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;
2117
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2118 2119
		return NULL;
	group = *pos + 1;
2120
	return (void *) ((unsigned long) group);
2121 2122 2123 2124 2125
}

static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
{
	struct super_block *sb = seq->private;
2126
	ext4_group_t group = (ext4_group_t) ((unsigned long) v);
2127 2128 2129 2130 2131
	int i;
	int err;
	struct ext4_buddy e4b;
	struct sg {
		struct ext4_group_info info;
2132
		ext4_grpblk_t counters[16];
2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147
	} 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) {
2148
		seq_printf(seq, "#%-5u: I/O error\n", group);
2149 2150 2151 2152 2153
		return 0;
	}
	ext4_lock_group(sb, group);
	memcpy(&sg, ext4_get_group_info(sb, group), i);
	ext4_unlock_group(sb, group);
2154
	ext4_mb_unload_buddy(&e4b);
2155

2156
	seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169
			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)
{
}

2170
static const struct seq_operations ext4_mb_seq_groups_ops = {
2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183
	.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) {
2184
		struct seq_file *m = file->private_data;
2185 2186 2187 2188 2189 2190
		m->private = sb;
	}
	return rc;

}

2191
static const struct file_operations ext4_mb_seq_groups_fops = {
2192 2193 2194 2195 2196 2197 2198
	.owner		= THIS_MODULE,
	.open		= ext4_mb_seq_groups_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= seq_release,
};

2199 2200 2201 2202 2203 2204 2205 2206
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;
}
2207 2208

/* Create and initialize ext4_group_info data for the given group. */
2209
int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2210 2211
			  struct ext4_group_desc *desc)
{
2212
	int i;
2213 2214 2215
	int metalen = 0;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_group_info **meta_group_info;
2216
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227

	/*
	 * 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) {
2228 2229
			ext4_msg(sb, KERN_ERR, "EXT4-fs: can't allocate mem "
				 "for a buddy group");
2230 2231 2232 2233 2234 2235 2236 2237 2238 2239
			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);

2240
	meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
2241
	if (meta_group_info[i] == NULL) {
2242
		ext4_msg(sb, KERN_ERR, "EXT4-fs: can't allocate buddy mem");
2243 2244
		goto exit_group_info;
	}
2245
	memset(meta_group_info[i], 0, kmem_cache_size(cachep));
2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257
	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 =
2258
			ext4_free_blks_count(sb, desc);
2259 2260 2261
	}

	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2262
	init_rwsem(&meta_group_info[i]->alloc_sem);
2263
	meta_group_info[i]->bb_free_root = RB_ROOT;
2264
	meta_group_info[i]->bb_largest_free_order = -1;  /* uninit */
2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283

#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 */
2284
	if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
2285
		kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]);
2286 2287
		sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] = NULL;
	}
2288 2289 2290 2291
exit_meta_group_info:
	return -ENOMEM;
} /* ext4_mb_add_groupinfo */

2292 2293
static int ext4_mb_init_backend(struct super_block *sb)
{
2294
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2295 2296
	ext4_group_t i;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2297 2298 2299 2300 2301
	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;
2302
	struct kmem_cache *cachep;
2303 2304

	/* This is the number of blocks used by GDT */
2305
	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320
				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);
2321

2322 2323 2324 2325 2326 2327 2328 2329 2330 2331
	/*
	 * 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;
2332 2333 2334
	/* 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. */
2335
	sbi->s_group_info = ext4_kvzalloc(array_size, GFP_KERNEL);
2336
	if (sbi->s_group_info == NULL) {
2337
		ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
2338 2339 2340 2341
		return -ENOMEM;
	}
	sbi->s_buddy_cache = new_inode(sb);
	if (sbi->s_buddy_cache == NULL) {
2342
		ext4_msg(sb, KERN_ERR, "can't get new inode");
2343 2344
		goto err_freesgi;
	}
2345 2346 2347 2348 2349
	/* To avoid potentially colliding with an valid on-disk inode number,
	 * use EXT4_BAD_INO for the buddy cache inode number.  This inode is
	 * not in the inode hash, so it should never be found by iget(), but
	 * this will avoid confusion if it ever shows up during debugging. */
	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
2350
	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
2351
	for (i = 0; i < ngroups; i++) {
2352 2353
		desc = ext4_get_group_desc(sb, i, NULL);
		if (desc == NULL) {
2354
			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
2355 2356
			goto err_freebuddy;
		}
2357 2358
		if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
			goto err_freebuddy;
2359 2360 2361 2362 2363
	}

	return 0;

err_freebuddy:
2364
	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2365
	while (i-- > 0)
2366
		kmem_cache_free(cachep, ext4_get_group_info(sb, i));
2367
	i = num_meta_group_infos;
2368
	while (i-- > 0)
2369 2370 2371
		kfree(sbi->s_group_info[i]);
	iput(sbi->s_buddy_cache);
err_freesgi:
2372
	ext4_kvfree(sbi->s_group_info);
2373 2374 2375
	return -ENOMEM;
}

2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413
static void ext4_groupinfo_destroy_slabs(void)
{
	int i;

	for (i = 0; i < NR_GRPINFO_CACHES; i++) {
		if (ext4_groupinfo_caches[i])
			kmem_cache_destroy(ext4_groupinfo_caches[i]);
		ext4_groupinfo_caches[i] = NULL;
	}
}

static int ext4_groupinfo_create_slab(size_t size)
{
	static DEFINE_MUTEX(ext4_grpinfo_slab_create_mutex);
	int slab_size;
	int blocksize_bits = order_base_2(size);
	int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
	struct kmem_cache *cachep;

	if (cache_index >= NR_GRPINFO_CACHES)
		return -EINVAL;

	if (unlikely(cache_index < 0))
		cache_index = 0;

	mutex_lock(&ext4_grpinfo_slab_create_mutex);
	if (ext4_groupinfo_caches[cache_index]) {
		mutex_unlock(&ext4_grpinfo_slab_create_mutex);
		return 0;	/* Already created */
	}

	slab_size = offsetof(struct ext4_group_info,
				bb_counters[blocksize_bits + 2]);

	cachep = kmem_cache_create(ext4_groupinfo_slab_names[cache_index],
					slab_size, 0, SLAB_RECLAIM_ACCOUNT,
					NULL);

2414 2415
	ext4_groupinfo_caches[cache_index] = cachep;

2416 2417
	mutex_unlock(&ext4_grpinfo_slab_create_mutex);
	if (!cachep) {
2418 2419
		printk(KERN_EMERG
		       "EXT4-fs: no memory for groupinfo slab cache\n");
2420 2421 2422 2423 2424 2425
		return -ENOMEM;
	}

	return 0;
}

2426 2427 2428
int ext4_mb_init(struct super_block *sb, int needs_recovery)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2429
	unsigned i, j;
2430 2431
	unsigned offset;
	unsigned max;
2432
	int ret;
2433

2434
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
2435 2436 2437

	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_offsets == NULL) {
2438 2439
		ret = -ENOMEM;
		goto out;
2440
	}
2441

2442
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
2443 2444
	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_maxs == NULL) {
2445 2446 2447 2448
		ret = -ENOMEM;
		goto out;
	}

2449 2450 2451
	ret = ext4_groupinfo_create_slab(sb->s_blocksize);
	if (ret < 0)
		goto out;
2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468

	/* 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 */
2469 2470
	ret = ext4_mb_init_backend(sb);
	if (ret != 0) {
2471
		goto out;
2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482
	}

	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;
2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494
	/*
	 * If there is a s_stripe > 1, then we set the s_mb_group_prealloc
	 * to the lowest multiple of s_stripe which is bigger than
	 * the s_mb_group_prealloc as determined above. We want
	 * the preallocation size to be an exact multiple of the
	 * RAID stripe size so that preallocations don't fragment
	 * the stripes.
	 */
	if (sbi->s_stripe > 1) {
		sbi->s_mb_group_prealloc = roundup(
			sbi->s_mb_group_prealloc, sbi->s_stripe);
	}
2495

2496
	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2497
	if (sbi->s_locality_groups == NULL) {
2498 2499
		ret = -ENOMEM;
		goto out;
2500
	}
2501
	for_each_possible_cpu(i) {
2502
		struct ext4_locality_group *lg;
2503
		lg = per_cpu_ptr(sbi->s_locality_groups, i);
2504
		mutex_init(&lg->lg_mutex);
2505 2506
		for (j = 0; j < PREALLOC_TB_SIZE; j++)
			INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
2507 2508 2509
		spin_lock_init(&lg->lg_prealloc_lock);
	}

2510 2511 2512
	if (sbi->s_proc)
		proc_create_data("mb_groups", S_IRUGO, sbi->s_proc,
				 &ext4_mb_seq_groups_fops, sb);
2513

2514 2515
	if (sbi->s_journal)
		sbi->s_journal->j_commit_callback = release_blocks_on_commit;
2516 2517 2518 2519 2520 2521
out:
	if (ret) {
		kfree(sbi->s_mb_offsets);
		kfree(sbi->s_mb_maxs);
	}
	return ret;
2522 2523
}

2524
/* need to called with the ext4 group lock held */
2525 2526 2527 2528 2529 2530 2531 2532 2533 2534
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++;
2535
		kmem_cache_free(ext4_pspace_cachep, pa);
2536 2537
	}
	if (count)
2538
		mb_debug(1, "mballoc: %u PAs left\n", count);
2539 2540 2541 2542 2543

}

int ext4_mb_release(struct super_block *sb)
{
2544
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2545 2546 2547 2548
	ext4_group_t i;
	int num_meta_group_infos;
	struct ext4_group_info *grinfo;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2549
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2550 2551

	if (sbi->s_group_info) {
2552
		for (i = 0; i < ngroups; i++) {
2553 2554 2555 2556 2557 2558 2559
			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);
2560
			kmem_cache_free(cachep, grinfo);
2561
		}
2562
		num_meta_group_infos = (ngroups +
2563 2564 2565 2566
				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]);
2567
		ext4_kvfree(sbi->s_group_info);
2568 2569 2570 2571 2572 2573
	}
	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) {
2574 2575
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %u blocks %u reqs (%u success)",
2576 2577 2578
				atomic_read(&sbi->s_bal_allocated),
				atomic_read(&sbi->s_bal_reqs),
				atomic_read(&sbi->s_bal_success));
2579 2580 2581
		ext4_msg(sb, KERN_INFO,
		      "mballoc: %u extents scanned, %u goal hits, "
				"%u 2^N hits, %u breaks, %u lost",
2582 2583 2584 2585 2586
				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));
2587 2588
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %lu generated and it took %Lu",
2589
				sbi->s_mb_buddies_generated,
2590
				sbi->s_mb_generation_time);
2591 2592
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %u preallocated, %u discarded",
2593 2594 2595 2596
				atomic_read(&sbi->s_mb_preallocated),
				atomic_read(&sbi->s_mb_discarded));
	}

2597
	free_percpu(sbi->s_locality_groups);
2598 2599
	if (sbi->s_proc)
		remove_proc_entry("mb_groups", sbi->s_proc);
2600 2601 2602 2603

	return 0;
}

2604
static inline int ext4_issue_discard(struct super_block *sb,
2605 2606 2607 2608 2609 2610 2611
		ext4_group_t block_group, ext4_grpblk_t block, int count)
{
	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);
2612
	return sb_issue_discard(sb, discard_block, count, GFP_NOFS, 0);
2613 2614
}

2615 2616 2617 2618 2619
/*
 * 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)
2620
{
2621
	struct super_block *sb = journal->j_private;
2622
	struct ext4_buddy e4b;
2623
	struct ext4_group_info *db;
2624
	int err, count = 0, count2 = 0;
2625
	struct ext4_free_data *entry;
2626
	struct list_head *l, *ltmp;
2627

2628 2629
	list_for_each_safe(l, ltmp, &txn->t_private_list) {
		entry = list_entry(l, struct ext4_free_data, list);
2630

2631
		mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
2632
			 entry->count, entry->group, entry);
2633

2634 2635 2636
		if (test_opt(sb, DISCARD))
			ext4_issue_discard(sb, entry->group,
					   entry->start_blk, entry->count);
2637

2638
		err = ext4_mb_load_buddy(sb, entry->group, &e4b);
2639 2640 2641
		/* we expect to find existing buddy because it's pinned */
		BUG_ON(err != 0);

2642
		db = e4b.bd_info;
2643
		/* there are blocks to put in buddy to make them really free */
2644
		count += entry->count;
2645
		count2++;
2646 2647 2648 2649 2650
		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);

2651 2652 2653 2654 2655 2656 2657 2658 2659
		/*
		 * Clear the trimmed flag for the group so that the next
		 * ext4_trim_fs can trim it.
		 * If the volume is mounted with -o discard, online discard
		 * is supported and the free blocks will be trimmed online.
		 */
		if (!test_opt(sb, DISCARD))
			EXT4_MB_GRP_CLEAR_TRIMMED(db);

2660 2661 2662 2663 2664 2665
		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);
2666
		}
2667 2668
		ext4_unlock_group(sb, entry->group);
		kmem_cache_free(ext4_free_ext_cachep, entry);
2669
		ext4_mb_unload_buddy(&e4b);
2670
	}
2671

2672
	mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
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 2707 2708
#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

2709
int __init ext4_init_mballoc(void)
2710
{
2711 2712
	ext4_pspace_cachep = KMEM_CACHE(ext4_prealloc_space,
					SLAB_RECLAIM_ACCOUNT);
2713 2714 2715
	if (ext4_pspace_cachep == NULL)
		return -ENOMEM;

2716 2717
	ext4_ac_cachep = KMEM_CACHE(ext4_allocation_context,
				    SLAB_RECLAIM_ACCOUNT);
2718 2719 2720 2721
	if (ext4_ac_cachep == NULL) {
		kmem_cache_destroy(ext4_pspace_cachep);
		return -ENOMEM;
	}
2722

2723 2724
	ext4_free_ext_cachep = KMEM_CACHE(ext4_free_data,
					  SLAB_RECLAIM_ACCOUNT);
2725 2726 2727 2728 2729
	if (ext4_free_ext_cachep == NULL) {
		kmem_cache_destroy(ext4_pspace_cachep);
		kmem_cache_destroy(ext4_ac_cachep);
		return -ENOMEM;
	}
2730
	ext4_create_debugfs_entry();
2731 2732 2733
	return 0;
}

2734
void ext4_exit_mballoc(void)
2735
{
2736
	/*
2737 2738 2739 2740
	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
	 * before destroying the slab cache.
	 */
	rcu_barrier();
2741
	kmem_cache_destroy(ext4_pspace_cachep);
2742
	kmem_cache_destroy(ext4_ac_cachep);
2743
	kmem_cache_destroy(ext4_free_ext_cachep);
2744
	ext4_groupinfo_destroy_slabs();
2745
	ext4_remove_debugfs_entry();
2746 2747 2748 2749
}


/*
2750
 * Check quota and mark chosen space (ac->ac_b_ex) non-free in bitmaps
2751 2752
 * Returns 0 if success or error code
 */
2753 2754
static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2755
				handle_t *handle, unsigned int reserv_blks)
2756 2757 2758 2759 2760 2761 2762
{
	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;
2763
	int err, len;
2764 2765 2766 2767 2768 2769 2770 2771

	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;
2772
	bitmap_bh = ext4_read_block_bitmap(sb, ac->ac_b_ex.fe_group);
2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784
	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;

2785
	ext4_debug("using block group %u(%d)\n", ac->ac_b_ex.fe_group,
2786
			ext4_free_blks_count(sb, gdp));
2787

2788 2789 2790 2791
	err = ext4_journal_get_write_access(handle, gdp_bh);
	if (err)
		goto out_err;

2792
	block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
2793

2794
	len = ac->ac_b_ex.fe_len;
2795
	if (!ext4_data_block_valid(sbi, block, len)) {
2796
		ext4_error(sb, "Allocating blocks %llu-%llu which overlap "
2797
			   "fs metadata\n", block, block+len);
2798 2799 2800 2801
		/* File system mounted not to panic on error
		 * Fix the bitmap and repeat the block allocation
		 * We leak some of the blocks here.
		 */
2802
		ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2803 2804
		ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
			      ac->ac_b_ex.fe_len);
2805
		ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2806
		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2807 2808 2809
		if (!err)
			err = -EAGAIN;
		goto out_err;
2810
	}
2811 2812

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2813 2814 2815 2816 2817 2818 2819 2820 2821
#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
2822 2823
	ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
		      ac->ac_b_ex.fe_len);
2824 2825
	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
		gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
2826 2827 2828
		ext4_free_blks_set(sb, gdp,
					ext4_free_blocks_after_init(sb,
					ac->ac_b_ex.fe_group, gdp));
2829
	}
2830 2831
	len = ext4_free_blks_count(sb, gdp) - ac->ac_b_ex.fe_len;
	ext4_free_blks_set(sb, gdp, len);
2832
	gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
2833 2834

	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2835
	percpu_counter_sub(&sbi->s_freeblocks_counter, ac->ac_b_ex.fe_len);
2836
	/*
2837
	 * Now reduce the dirty block count also. Should not go negative
2838
	 */
2839 2840 2841
	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);
2842

2843 2844 2845
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi,
							  ac->ac_b_ex.fe_group);
2846 2847
		atomic_sub(ac->ac_b_ex.fe_len,
			   &sbi->s_flex_groups[flex_group].free_blocks);
2848 2849
	}

2850
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2851 2852
	if (err)
		goto out_err;
2853
	err = ext4_handle_dirty_metadata(handle, NULL, gdp_bh);
2854 2855

out_err:
T
Theodore Ts'o 已提交
2856
	ext4_mark_super_dirty(sb);
2857
	brelse(bitmap_bh);
2858 2859 2860 2861 2862
	return err;
}

/*
 * here we normalize request for locality group
2863 2864 2865
 * Group request are normalized to s_mb_group_prealloc, which goes to
 * s_strip if we set the same via mount option.
 * s_mb_group_prealloc can be configured via
T
Theodore Ts'o 已提交
2866
 * /sys/fs/ext4/<partition>/mb_group_prealloc
2867 2868 2869 2870 2871 2872 2873 2874 2875
 *
 * 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);
2876
	ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
2877
	mb_debug(1, "#%u: goal %u blocks for locality group\n",
2878 2879 2880 2881 2882 2883 2884
		current->pid, ac->ac_g_ex.fe_len);
}

/*
 * Normalization means making request better in terms of
 * size and alignment
 */
2885 2886
static noinline_for_stack void
ext4_mb_normalize_request(struct ext4_allocation_context *ac,
2887 2888 2889 2890 2891
				struct ext4_allocation_request *ar)
{
	int bsbits, max;
	ext4_lblk_t end;
	loff_t size, orig_size, start_off;
2892
	ext4_lblk_t start;
2893
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
2894
	struct ext4_prealloc_space *pa;
2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922

	/* 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);
2923
	orig_size = size;
2924

2925 2926
	/* max size of free chunks */
	max = 2 << bsbits;
2927

2928 2929
#define NRL_CHECK_SIZE(req, size, max, chunk_size)	\
		(req <= (size) || max <= (chunk_size))
2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947

	/* 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;
2948
	} else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
2949
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
2950 2951 2952
						(21 - bsbits)) << 21;
		size = 2 * 1024 * 1024;
	} else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
2953 2954 2955 2956
		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,
2957
					(8<<20)>>bsbits, max, 8 * 1024)) {
2958 2959 2960 2961 2962 2963 2964
		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;
	}
2965 2966
	size = size >> bsbits;
	start = start_off >> bsbits;
2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979

	/* 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();
2980
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
2981
		ext4_lblk_t pa_end;
2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996

		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));

2997 2998
		/* skip PAs this normalized request doesn't overlap with */
		if (pa->pa_lstart >= end || pa_end <= start) {
2999 3000 3001 3002 3003
			spin_unlock(&pa->pa_lock);
			continue;
		}
		BUG_ON(pa->pa_lstart <= start && pa_end >= end);

3004
		/* adjust start or end to be adjacent to this pa */
3005 3006 3007
		if (pa_end <= ac->ac_o_ex.fe_logical) {
			BUG_ON(pa_end < start);
			start = pa_end;
3008
		} else if (pa->pa_lstart > ac->ac_o_ex.fe_logical) {
3009 3010 3011 3012 3013 3014 3015 3016 3017 3018
			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();
3019
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3020
		ext4_lblk_t pa_end;
3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031
		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) {
3032 3033 3034 3035
		ext4_msg(ac->ac_sb, KERN_ERR,
			 "start %lu, size %lu, fe_logical %lu",
			 (unsigned long) start, (unsigned long) size,
			 (unsigned long) ac->ac_o_ex.fe_logical);
3036 3037 3038
	}
	BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
			start > ac->ac_o_ex.fe_logical);
3039
	BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063

	/* 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;
	}

3064
	mb_debug(1, "goal: %u(was %u) blocks at %u\n", (unsigned) size,
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074
		(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);
3075
		if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
3076 3077 3078 3079 3080 3081 3082 3083 3084
			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);
	}

3085 3086 3087 3088
	if (ac->ac_op == EXT4_MB_HISTORY_ALLOC)
		trace_ext4_mballoc_alloc(ac);
	else
		trace_ext4_mballoc_prealloc(ac);
3089 3090
}

3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108
/*
 * 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;
	}

}

3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133
/*
 * 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;

3134
	mb_debug(1, "use %llu/%u from inode pa %p\n", start, len, pa);
3135 3136 3137 3138 3139 3140 3141 3142
}

/*
 * use blocks preallocated to locality group
 */
static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
3143
	unsigned int len = ac->ac_o_ex.fe_len;
3144

3145 3146 3147 3148 3149 3150 3151 3152
	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
3153
	 * possible race when the group is being loaded concurrently
3154
	 * instead we correct pa later, after blocks are marked
3155 3156
	 * in on-disk bitmap -- see ext4_mb_release_context()
	 * Other CPUs are prevented from allocating from this pa by lg_mutex
3157
	 */
3158
	mb_debug(1, "use %u/%u from group pa %p\n", pa->pa_lstart-len, len, pa);
3159 3160
}

3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180
/*
 * 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);

3181
	if (cur_distance <= new_distance)
3182 3183 3184 3185 3186 3187 3188 3189
		return cpa;

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

3190 3191 3192
/*
 * search goal blocks in preallocated space
 */
3193 3194
static noinline_for_stack int
ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
3195
{
3196
	int order, i;
3197 3198
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
	struct ext4_locality_group *lg;
3199 3200
	struct ext4_prealloc_space *pa, *cpa = NULL;
	ext4_fsblk_t goal_block;
3201 3202 3203 3204 3205 3206 3207

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

	/* first, try per-file preallocation */
	rcu_read_lock();
3208
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3209 3210 3211 3212 3213 3214 3215

		/* 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;

3216
		/* non-extent files can't have physical blocks past 2^32 */
3217
		if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)) &&
3218 3219 3220
			pa->pa_pstart + pa->pa_len > EXT4_MAX_BLOCK_FILE_PHYS)
			continue;

3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242
		/* 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;
3243 3244 3245 3246 3247
	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;

3248
	goal_block = ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex);
3249 3250 3251 3252
	/*
	 * search for the prealloc space that is having
	 * minimal distance from the goal block.
	 */
3253 3254 3255 3256 3257 3258 3259
	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) {
3260 3261 3262

				cpa = ext4_mb_check_group_pa(goal_block,
								pa, cpa);
3263
			}
3264 3265
			spin_unlock(&pa->pa_lock);
		}
3266
		rcu_read_unlock();
3267
	}
3268 3269 3270 3271 3272
	if (cpa) {
		ext4_mb_use_group_pa(ac, cpa);
		ac->ac_criteria = 20;
		return 1;
	}
3273 3274 3275
	return 0;
}

3276 3277 3278 3279
/*
 * 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
3280
 * Need to be called with the ext4 group lock held
3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293
 */
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);
3294
		ext4_set_bits(bitmap, entry->start_blk, entry->count);
3295 3296 3297 3298 3299
		n = rb_next(n);
	}
	return;
}

3300 3301 3302
/*
 * the function goes through all preallocation in this group and marks them
 * used in in-core bitmap. buddy must be generated from this bitmap
3303
 * Need to be called with ext4 group lock held
3304
 */
3305 3306
static noinline_for_stack
void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335
					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);
3336
		ext4_set_bits(bitmap, start, len);
3337 3338 3339
		preallocated += len;
		count++;
	}
3340
	mb_debug(1, "prellocated %u for group %u\n", preallocated, group);
3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356
}

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)
{
3357
	ext4_group_t grp;
3358
	ext4_fsblk_t grp_blk;
3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372

	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);

3373
	grp_blk = pa->pa_pstart;
3374
	/*
3375 3376 3377 3378
	 * 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)
3379 3380 3381
		grp_blk--;

	ext4_get_group_no_and_offset(sb, grp_blk, &grp, NULL);
3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410

	/*
	 * 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
 */
3411 3412
static noinline_for_stack int
ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 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
{
	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);
3470 3471
	INIT_LIST_HEAD(&pa->pa_inode_list);
	INIT_LIST_HEAD(&pa->pa_group_list);
3472
	pa->pa_deleted = 0;
3473
	pa->pa_type = MB_INODE_PA;
3474

3475
	mb_debug(1, "new inode pa %p: %llu/%u for %u\n", pa,
3476
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
3477
	trace_ext4_mb_new_inode_pa(ac, pa);
3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501

	ext4_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
 */
3502 3503
static noinline_for_stack int
ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529
{
	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);
3530
	INIT_LIST_HEAD(&pa->pa_inode_list);
3531
	INIT_LIST_HEAD(&pa->pa_group_list);
3532
	pa->pa_deleted = 0;
3533
	pa->pa_type = MB_GROUP_PA;
3534

3535
	mb_debug(1, "new group pa %p: %llu/%u for %u\n", pa,
3536 3537
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
	trace_ext4_mb_new_group_pa(ac, pa);
3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552

	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);

3553 3554 3555 3556
	/*
	 * We will later add the new pa to the right bucket
	 * after updating the pa_free in ext4_mb_release_context
	 */
3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578
	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
 */
3579 3580
static noinline_for_stack int
ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3581
			struct ext4_prealloc_space *pa)
3582 3583 3584
{
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
3585 3586
	unsigned int end;
	unsigned int next;
3587 3588
	ext4_group_t group;
	ext4_grpblk_t bit;
3589
	unsigned long long grp_blk_start;
3590 3591 3592 3593 3594
	int err = 0;
	int free = 0;

	BUG_ON(pa->pa_deleted == 0);
	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
3595
	grp_blk_start = pa->pa_pstart - bit;
3596 3597 3598 3599
	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
	end = bit + pa->pa_len;

	while (bit < end) {
3600
		bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
3601 3602
		if (bit >= end)
			break;
3603
		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
3604
		mb_debug(1, "    free preallocated %u/%u in group %u\n",
3605 3606
			 (unsigned) ext4_group_first_block_no(sb, group) + bit,
			 (unsigned) next - bit, (unsigned) group);
3607 3608
		free += next - bit;

3609
		trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
L
Lukas Czerner 已提交
3610 3611
		trace_ext4_mb_release_inode_pa(pa, grp_blk_start + bit,
					       next - bit);
3612 3613 3614 3615
		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
		bit = next + 1;
	}
	if (free != pa->pa_free) {
3616 3617 3618 3619 3620
		ext4_msg(e4b->bd_sb, KERN_CRIT,
			 "pa %p: logic %lu, phys. %lu, len %lu",
			 pa, (unsigned long) pa->pa_lstart,
			 (unsigned long) pa->pa_pstart,
			 (unsigned long) pa->pa_len);
3621
		ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
3622
					free, pa->pa_free);
3623 3624 3625 3626
		/*
		 * pa is already deleted so we use the value obtained
		 * from the bitmap and continue.
		 */
3627 3628 3629 3630 3631 3632
	}
	atomic_add(free, &sbi->s_mb_discarded);

	return err;
}

3633 3634
static noinline_for_stack int
ext4_mb_release_group_pa(struct ext4_buddy *e4b,
3635
				struct ext4_prealloc_space *pa)
3636 3637 3638 3639 3640
{
	struct super_block *sb = e4b->bd_sb;
	ext4_group_t group;
	ext4_grpblk_t bit;

L
Lukas Czerner 已提交
3641
	trace_ext4_mb_release_group_pa(pa);
3642 3643 3644 3645 3646
	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);
3647
	trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660

	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
 */
3661 3662
static noinline_for_stack int
ext4_mb_discard_group_preallocations(struct super_block *sb,
3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
					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;

3674
	mb_debug(1, "discard preallocation for group %u\n", group);
3675 3676 3677 3678

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

3679
	bitmap_bh = ext4_read_block_bitmap(sb, group);
3680
	if (bitmap_bh == NULL) {
3681
		ext4_error(sb, "Error reading block bitmap for %u", group);
3682
		return 0;
3683 3684 3685
	}

	err = ext4_mb_load_buddy(sb, group, &e4b);
3686
	if (err) {
3687
		ext4_error(sb, "Error loading buddy information for %u", group);
3688 3689 3690
		put_bh(bitmap_bh);
		return 0;
	}
3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 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

	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);

3749
		if (pa->pa_type == MB_GROUP_PA)
3750
			ext4_mb_release_group_pa(&e4b, pa);
3751
		else
3752
			ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3753 3754 3755 3756 3757 3758 3759

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

out:
	ext4_unlock_group(sb, group);
3760
	ext4_mb_unload_buddy(&e4b);
3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773
	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
 */
3774
void ext4_discard_preallocations(struct inode *inode)
3775 3776 3777 3778 3779 3780 3781 3782 3783 3784
{
	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;

3785
	if (!S_ISREG(inode->i_mode)) {
3786 3787 3788 3789
		/*BUG_ON(!list_empty(&ei->i_prealloc_list));*/
		return;
	}

3790
	mb_debug(1, "discard preallocation for inode %lu\n", inode->i_ino);
3791
	trace_ext4_discard_preallocations(inode);
3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807

	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);
3808 3809
			ext4_msg(sb, KERN_ERR,
				 "uh-oh! used pa while discarding");
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
			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) {
3845
		BUG_ON(pa->pa_type != MB_INODE_PA);
3846 3847 3848
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, NULL);

		err = ext4_mb_load_buddy(sb, group, &e4b);
3849
		if (err) {
3850 3851
			ext4_error(sb, "Error loading buddy information for %u",
					group);
3852 3853
			continue;
		}
3854

3855
		bitmap_bh = ext4_read_block_bitmap(sb, group);
3856
		if (bitmap_bh == NULL) {
3857 3858
			ext4_error(sb, "Error reading block bitmap for %u",
					group);
3859
			ext4_mb_unload_buddy(&e4b);
3860
			continue;
3861 3862 3863 3864
		}

		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
3865
		ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3866 3867
		ext4_unlock_group(sb, group);

3868
		ext4_mb_unload_buddy(&e4b);
3869 3870 3871 3872 3873 3874 3875
		put_bh(bitmap_bh);

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

3876
#ifdef CONFIG_EXT4_DEBUG
3877 3878 3879
static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
{
	struct super_block *sb = ac->ac_sb;
3880
	ext4_group_t ngroups, i;
3881

3882 3883
	if (!mb_enable_debug ||
	    (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED))
3884 3885
		return;

3886 3887 3888
	ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: Can't allocate:"
			" Allocation context details:");
	ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: status %d flags %d",
3889
			ac->ac_status, ac->ac_flags);
3890 3891 3892
	ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: orig %lu/%lu/%lu@%lu, "
		 	"goal %lu/%lu/%lu@%lu, "
			"best %lu/%lu/%lu@%lu cr %d",
3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905
			(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);
3906 3907 3908
	ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: %lu scanned, %d found",
		 ac->ac_ex_scanned, ac->ac_found);
	ext4_msg(ac->ac_sb, KERN_ERR, "EXT4-fs: groups: ");
3909 3910
	ngroups = ext4_get_groups_count(sb);
	for (i = 0; i < ngroups; i++) {
3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922
		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);
3923 3924
			printk(KERN_ERR "PA:%u:%d:%u \n", i,
			       start, pa->pa_len);
3925
		}
3926
		ext4_unlock_group(sb, i);
3927 3928 3929

		if (grp->bb_free == 0)
			continue;
3930
		printk(KERN_ERR "%u: %d/%d \n",
3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946
		       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
3947
 * One can tune this size via /sys/fs/ext4/<partition>/mb_stream_req
3948 3949 3950 3951 3952 3953 3954 3955 3956 3957
 */
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;

3958 3959 3960
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		return;

3961
	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
3962 3963
	isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1)
		>> bsbits;
3964

3965 3966 3967 3968 3969 3970 3971
	if ((size == isize) &&
	    !ext4_fs_is_busy(sbi) &&
	    (atomic_read(&ac->ac_inode->i_writecount) == 0)) {
		ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC;
		return;
	}

3972
	/* don't use group allocation for large files */
3973
	size = max(size, isize);
3974
	if (size > sbi->s_mb_stream_request) {
3975
		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
3976
		return;
3977
	}
3978 3979 3980 3981 3982 3983 3984

	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.
	 */
3985
	ac->ac_lg = __this_cpu_ptr(sbi->s_locality_groups);
3986 3987 3988 3989 3990 3991 3992 3993

	/* 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);
}

3994 3995
static noinline_for_stack int
ext4_mb_initialize_context(struct ext4_allocation_context *ac,
3996 3997 3998 3999 4000 4001
				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;
4002 4003
	unsigned int len;
	ext4_fsblk_t goal;
4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020
	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 */
4021
	memset(ac, 0, sizeof(struct ext4_allocation_context));
4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039
	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);

4040
	mb_debug(1, "init ac: %u blocks @ %u, goal %u, flags %x, 2^%d, "
4041 4042 4043 4044 4045 4046 4047 4048 4049 4050
			"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;

}

4051 4052 4053 4054 4055 4056 4057 4058 4059 4060
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;

4061
	mb_debug(1, "discard locality group preallocation\n");
4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082

	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 */
4083
		BUG_ON(pa->pa_type != MB_GROUP_PA);
4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108

		/* 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)) {
4109 4110
			ext4_error(sb, "Error loading buddy information for %u",
					group);
4111 4112 4113 4114
			continue;
		}
		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
4115
		ext4_mb_release_group_pa(&e4b, pa);
4116 4117
		ext4_unlock_group(sb, group);

4118
		ext4_mb_unload_buddy(&e4b);
4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149
		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) {
4150
			spin_unlock(&tmp_pa->pa_lock);
4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179
			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 ;
}

4180 4181 4182 4183 4184
/*
 * release all resource we used in allocation
 */
static int ext4_mb_release_context(struct ext4_allocation_context *ac)
{
4185 4186
	struct ext4_prealloc_space *pa = ac->ac_pa;
	if (pa) {
4187
		if (pa->pa_type == MB_GROUP_PA) {
4188
			/* see comment in ext4_mb_use_group_pa() */
4189 4190 4191 4192 4193 4194
			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);
4195 4196
		}
	}
A
Aneesh Kumar K.V 已提交
4197 4198 4199 4200 4201
	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
A
Amir Goldstein 已提交
4202
		 * doesn't grow big.
A
Aneesh Kumar K.V 已提交
4203
		 */
4204
		if ((pa->pa_type == MB_GROUP_PA) && likely(pa->pa_free)) {
A
Aneesh Kumar K.V 已提交
4205 4206 4207 4208 4209 4210 4211
			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);
	}
4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223
	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)
{
4224
	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4225 4226 4227
	int ret;
	int freed = 0;

4228
	trace_ext4_mb_discard_preallocations(sb, needed);
4229
	for (i = 0; i < ngroups && needed > 0; i++) {
4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243
		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,
4244
				struct ext4_allocation_request *ar, int *errp)
4245
{
4246
	int freed;
4247
	struct ext4_allocation_context *ac = NULL;
4248 4249 4250
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block = 0;
4251
	unsigned int inquota = 0;
4252
	unsigned int reserv_blks = 0;
4253 4254 4255 4256

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

4257
	trace_ext4_request_blocks(ar);
4258

4259 4260 4261 4262 4263
	/*
	 * 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.
	 */
4264
	if (ext4_test_inode_state(ar->inode, EXT4_STATE_DELALLOC_RESERVED))
4265 4266 4267 4268 4269
		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.
4270
		 */
4271 4272 4273
		while (ar->len &&
			ext4_claim_free_blocks(sbi, ar->len, ar->flags)) {

A
Aneesh Kumar K.V 已提交
4274 4275 4276 4277 4278
			/* let others to free the space */
			yield();
			ar->len = ar->len >> 1;
		}
		if (!ar->len) {
4279 4280 4281
			*errp = -ENOSPC;
			return 0;
		}
4282
		reserv_blks = ar->len;
4283 4284 4285 4286 4287 4288 4289 4290 4291
		if (ar->flags & EXT4_MB_USE_ROOT_BLOCKS) {
			dquot_alloc_block_nofail(ar->inode, ar->len);
		} else {
			while (ar->len &&
				dquot_alloc_block(ar->inode, ar->len)) {

				ar->flags |= EXT4_MB_HINT_NOPREALLOC;
				ar->len--;
			}
4292 4293 4294 4295
		}
		inquota = ar->len;
		if (ar->len == 0) {
			*errp = -EDQUOT;
4296
			goto out;
4297
		}
4298
	}
4299

4300
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4301
	if (!ac) {
4302
		ar->len = 0;
4303
		*errp = -ENOMEM;
4304
		goto out;
4305 4306 4307
	}

	*errp = ext4_mb_initialize_context(ac, ar);
4308 4309
	if (*errp) {
		ar->len = 0;
4310
		goto out;
4311 4312
	}

4313 4314 4315 4316
	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);
4317 4318
repeat:
		/* allocate space in core */
4319 4320 4321
		*errp = ext4_mb_regular_allocator(ac);
		if (*errp)
			goto errout;
4322 4323 4324 4325

		/* as we've just preallocated more space than
		 * user requested orinally, we store allocated
		 * space in a special descriptor */
4326 4327 4328
		if (ac->ac_status == AC_STATUS_FOUND &&
				ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
			ext4_mb_new_preallocation(ac);
4329
	}
4330
	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
4331
		*errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_blks);
4332
		if (*errp == -EAGAIN) {
4333 4334 4335 4336 4337
			/*
			 * drop the reference that we took
			 * in ext4_mb_use_best_found
			 */
			ext4_mb_release_context(ac);
4338 4339 4340 4341 4342
			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;
4343 4344
		} else if (*errp)
		errout:
4345
			ext4_discard_allocated_blocks(ac);
4346
		else {
4347 4348 4349
			block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
			ar->len = ac->ac_b_ex.fe_len;
		}
4350
	} else {
4351
		freed  = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
4352 4353 4354
		if (freed)
			goto repeat;
		*errp = -ENOSPC;
4355 4356 4357
	}

	if (*errp) {
4358
		ac->ac_b_ex.fe_len = 0;
4359
		ar->len = 0;
4360
		ext4_mb_show_ac(ac);
4361
	}
4362
	ext4_mb_release_context(ac);
4363 4364 4365
out:
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
4366
	if (inquota && ar->len < inquota)
4367
		dquot_free_block(ar->inode, inquota - ar->len);
4368
	if (!ar->len) {
4369 4370
		if (!ext4_test_inode_state(ar->inode,
					   EXT4_STATE_DELALLOC_RESERVED))
4371 4372 4373 4374
			/* release all the reserved blocks if non delalloc */
			percpu_counter_sub(&sbi->s_dirtyblocks_counter,
						reserv_blks);
	}
4375

4376
	trace_ext4_allocate_blocks(ar, (unsigned long long)block);
4377

4378 4379 4380
	return block;
}

4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395
/*
 * 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;
}

4396 4397
static noinline_for_stack int
ext4_mb_free_metadata(handle_t *handle, struct ext4_buddy *e4b,
4398
		      struct ext4_free_data *new_entry)
4399
{
4400
	ext4_group_t group = e4b->bd_group;
4401 4402
	ext4_grpblk_t block;
	struct ext4_free_data *entry;
4403 4404 4405
	struct ext4_group_info *db = e4b->bd_info;
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
4406 4407 4408
	struct rb_node **n = &db->bb_free_root.rb_node, *node;
	struct rb_node *parent = NULL, *new_node;

4409
	BUG_ON(!ext4_handle_valid(handle));
4410 4411 4412
	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

4413
	new_node = &new_entry->node;
4414
	block = new_entry->start_blk;
4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432

	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 {
4433 4434 4435
			ext4_grp_locked_error(sb, group, 0,
				ext4_group_first_block_no(sb, group) + block,
				"Block already on to-be-freed list");
4436
			return 0;
4437
		}
4438
	}
4439

4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454
	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);
4455
		}
4456
	}
4457

4458 4459 4460 4461 4462 4463 4464 4465 4466 4467
	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);
4468 4469
		}
	}
4470
	/* Add the extent to transaction's private list */
4471
	spin_lock(&sbi->s_md_lock);
4472
	list_add(&new_entry->list, &handle->h_transaction->t_private_list);
4473
	spin_unlock(&sbi->s_md_lock);
4474 4475 4476
	return 0;
}

4477 4478 4479 4480 4481 4482
/**
 * 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
4483
 * @flags:		flags used by ext4_free_blocks
4484
 */
4485
void ext4_free_blocks(handle_t *handle, struct inode *inode,
4486 4487
		      struct buffer_head *bh, ext4_fsblk_t block,
		      unsigned long count, int flags)
4488
{
4489
	struct buffer_head *bitmap_bh = NULL;
4490 4491
	struct super_block *sb = inode->i_sb;
	struct ext4_group_desc *gdp;
4492
	unsigned long freed = 0;
4493
	unsigned int overflow;
4494 4495 4496 4497 4498 4499 4500 4501
	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;

4502 4503 4504 4505 4506 4507
	if (bh) {
		if (block)
			BUG_ON(block != bh->b_blocknr);
		else
			block = bh->b_blocknr;
	}
4508 4509

	sbi = EXT4_SB(sb);
4510 4511
	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
	    !ext4_data_block_valid(sbi, block, count)) {
4512
		ext4_error(sb, "Freeing blocks not in datazone - "
4513
			   "block = %llu, count = %lu", block, count);
4514 4515 4516
		goto error_return;
	}

4517
	ext4_debug("freeing block %llu\n", block);
4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529
	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);
4530 4531
			if (unlikely(!tbh))
				continue;
4532
			ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
4533 4534 4535 4536
				    inode, tbh, block + i);
		}
	}

4537
	/*
4538 4539 4540 4541 4542 4543 4544 4545
	 * 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;
4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558

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;
	}
4559
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4560 4561
	if (!bitmap_bh) {
		err = -EIO;
4562
		goto error_return;
4563
	}
4564
	gdp = ext4_get_group_desc(sb, block_group, &gd_bh);
4565 4566
	if (!gdp) {
		err = -EIO;
4567
		goto error_return;
4568
	}
4569 4570 4571 4572 4573 4574 4575 4576

	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)) {

4577
		ext4_error(sb, "Freeing blocks in system zone - "
4578
			   "Block = %llu, count = %lu", block, count);
4579 4580
		/* err = 0. ext4_std_error should be a no op */
		goto error_return;
4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603
	}

	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
4604
	trace_ext4_mballoc_free(sb, inode, block_group, bit, count);
4605

4606 4607 4608
	err = ext4_mb_load_buddy(sb, block_group, &e4b);
	if (err)
		goto error_return;
4609 4610

	if ((flags & EXT4_FREE_BLOCKS_METADATA) && ext4_handle_valid(handle)) {
4611 4612 4613 4614 4615
		struct ext4_free_data *new_entry;
		/*
		 * blocks being freed are metadata. these blocks shouldn't
		 * be used until this transaction is committed
		 */
4616 4617 4618 4619 4620
		new_entry = kmem_cache_alloc(ext4_free_ext_cachep, GFP_NOFS);
		if (!new_entry) {
			err = -ENOMEM;
			goto error_return;
		}
4621 4622 4623 4624
		new_entry->start_blk = bit;
		new_entry->group  = block_group;
		new_entry->count = count;
		new_entry->t_tid = handle->h_transaction->t_tid;
4625

4626
		ext4_lock_group(sb, block_group);
4627
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4628
		ext4_mb_free_metadata(handle, &e4b, new_entry);
4629
	} else {
4630 4631 4632 4633
		/* need to update group_info->bb_free and bitmap
		 * with group lock held. generate_buddy look at
		 * them with group lock_held
		 */
4634 4635
		ext4_lock_group(sb, block_group);
		mb_clear_bits(bitmap_bh->b_data, bit, count);
4636
		mb_free_blocks(inode, &e4b, bit, count);
4637 4638
	}

4639 4640
	ret = ext4_free_blks_count(sb, gdp) + count;
	ext4_free_blks_set(sb, gdp, ret);
4641
	gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
4642
	ext4_unlock_group(sb, block_group);
4643 4644
	percpu_counter_add(&sbi->s_freeblocks_counter, count);

4645 4646
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4647
		atomic_add(count, &sbi->s_flex_groups[flex_group].free_blocks);
4648 4649
	}

4650
	ext4_mb_unload_buddy(&e4b);
4651

4652
	freed += count;
4653

4654 4655 4656 4657
	/* We dirtied the bitmap block */
	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);

4658 4659
	/* And the group descriptor block */
	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
4660
	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
4661 4662 4663 4664 4665 4666 4667 4668 4669
	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
4670
	ext4_mark_super_dirty(sb);
4671
error_return:
4672
	if (freed && !(flags & EXT4_FREE_BLOCKS_NO_QUOT_UPDATE))
4673
		dquot_free_block(inode, freed);
4674 4675 4676 4677
	brelse(bitmap_bh);
	ext4_std_error(sb, err);
	return;
}
4678

4679
/**
4680
 * ext4_group_add_blocks() -- Add given blocks to an existing group
4681 4682 4683 4684 4685
 * @handle:			handle to this transaction
 * @sb:				super block
 * @block:			start physcial block to add to the block group
 * @count:			number of blocks to free
 *
4686
 * This marks the blocks as free in the bitmap and buddy.
4687
 */
4688
int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
4689 4690 4691 4692 4693 4694 4695 4696 4697
			 ext4_fsblk_t block, unsigned long count)
{
	struct buffer_head *bitmap_bh = NULL;
	struct buffer_head *gd_bh;
	ext4_group_t block_group;
	ext4_grpblk_t bit;
	unsigned int i;
	struct ext4_group_desc *desc;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
4698
	struct ext4_buddy e4b;
4699 4700 4701 4702 4703
	int err = 0, ret, blk_free_count;
	ext4_grpblk_t blocks_freed;

	ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);

4704 4705 4706
	if (count == 0)
		return 0;

4707 4708 4709 4710 4711
	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
	/*
	 * Check to see if we are freeing blocks across a group
	 * boundary.
	 */
4712 4713 4714 4715
	if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
		ext4_warning(sb, "too much blocks added to group %u\n",
			     block_group);
		err = -EINVAL;
4716
		goto error_return;
4717
	}
4718

4719
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4720 4721
	if (!bitmap_bh) {
		err = -EIO;
4722
		goto error_return;
4723 4724
	}

4725
	desc = ext4_get_group_desc(sb, block_group, &gd_bh);
4726 4727
	if (!desc) {
		err = -EIO;
4728
		goto error_return;
4729
	}
4730 4731 4732 4733 4734 4735 4736 4737 4738

	if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
	    in_range(ext4_inode_bitmap(sb, desc), block, count) ||
	    in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
	    in_range(block + count - 1, ext4_inode_table(sb, desc),
		     sbi->s_itb_per_group)) {
		ext4_error(sb, "Adding blocks in system zones - "
			   "Block = %llu, count = %lu",
			   block, count);
4739
		err = -EINVAL;
4740 4741 4742
		goto error_return;
	}

4743 4744
	BUFFER_TRACE(bitmap_bh, "getting write access");
	err = ext4_journal_get_write_access(handle, bitmap_bh);
4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756
	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;
4757

4758 4759
	for (i = 0, blocks_freed = 0; i < count; i++) {
		BUFFER_TRACE(bitmap_bh, "clear bit");
4760
		if (!mb_test_bit(bit + i, bitmap_bh->b_data)) {
4761 4762 4763 4764 4765 4766 4767
			ext4_error(sb, "bit already cleared for block %llu",
				   (ext4_fsblk_t)(block + i));
			BUFFER_TRACE(bitmap_bh, "bit already cleared");
		} else {
			blocks_freed++;
		}
	}
4768 4769 4770 4771 4772 4773 4774 4775 4776 4777

	err = ext4_mb_load_buddy(sb, block_group, &e4b);
	if (err)
		goto error_return;

	/*
	 * need to update group_info->bb_free and bitmap
	 * with group lock held. generate_buddy look at
	 * them with group lock_held
	 */
4778
	ext4_lock_group(sb, block_group);
4779 4780
	mb_clear_bits(bitmap_bh->b_data, bit, count);
	mb_free_blocks(NULL, &e4b, bit, count);
4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791
	blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
	ext4_free_blks_set(sb, desc, blk_free_count);
	desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
	ext4_unlock_group(sb, block_group);
	percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);

	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
		atomic_add(blocks_freed,
			   &sbi->s_flex_groups[flex_group].free_blocks);
	}
4792 4793

	ext4_mb_unload_buddy(&e4b);
4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807

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

	/* And the group descriptor block */
	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
	if (!err)
		err = ret;

error_return:
	brelse(bitmap_bh);
	ext4_std_error(sb, err);
4808
	return err;
4809 4810
}

4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822
/**
 * ext4_trim_extent -- function to TRIM one single free extent in the group
 * @sb:		super block for the file system
 * @start:	starting block of the free extent in the alloc. group
 * @count:	number of blocks to TRIM
 * @group:	alloc. group we are working with
 * @e4b:	ext4 buddy for the group
 *
 * Trim "count" blocks starting at "start" in the "group". To assure that no
 * one will allocate those blocks, mark it as used in buddy bitmap. This must
 * be called with under the group lock.
 */
4823 4824
static void ext4_trim_extent(struct super_block *sb, int start, int count,
			     ext4_group_t group, struct ext4_buddy *e4b)
4825 4826 4827
{
	struct ext4_free_extent ex;

T
Tao Ma 已提交
4828 4829
	trace_ext4_trim_extent(sb, group, start, count);

4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841
	assert_spin_locked(ext4_group_lock_ptr(sb, group));

	ex.fe_start = start;
	ex.fe_group = group;
	ex.fe_len = count;

	/*
	 * Mark blocks used, so no one can reuse them while
	 * being trimmed.
	 */
	mb_mark_used(e4b, &ex);
	ext4_unlock_group(sb, group);
4842
	ext4_issue_discard(sb, group, start, count);
4843 4844 4845 4846 4847 4848 4849
	ext4_lock_group(sb, group);
	mb_free_blocks(NULL, e4b, start, ex.fe_len);
}

/**
 * ext4_trim_all_free -- function to trim all free space in alloc. group
 * @sb:			super block for file system
4850
 * @group:		group to be trimmed
4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864
 * @start:		first group block to examine
 * @max:		last group block to examine
 * @minblocks:		minimum extent block count
 *
 * ext4_trim_all_free walks through group's buddy bitmap searching for free
 * extents. When the free block is found, ext4_trim_extent is called to TRIM
 * the extent.
 *
 *
 * ext4_trim_all_free walks through group's block bitmap searching for free
 * extents. When the free extent is found, mark it as used in group buddy
 * bitmap. Then issue a TRIM command on this extent and free the extent in
 * the group buddy bitmap. This is done until whole group is scanned.
 */
4865
static ext4_grpblk_t
4866 4867 4868
ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
		   ext4_grpblk_t start, ext4_grpblk_t max,
		   ext4_grpblk_t minblocks)
4869 4870
{
	void *bitmap;
4871
	ext4_grpblk_t next, count = 0, free_count = 0;
4872 4873
	struct ext4_buddy e4b;
	int ret;
4874

T
Tao Ma 已提交
4875 4876
	trace_ext4_trim_all_free(sb, group, start, max);

4877 4878 4879 4880 4881 4882 4883
	ret = ext4_mb_load_buddy(sb, group, &e4b);
	if (ret) {
		ext4_error(sb, "Error in loading buddy "
				"information for %u", group);
		return ret;
	}
	bitmap = e4b.bd_bitmap;
4884 4885

	ext4_lock_group(sb, group);
4886 4887 4888 4889
	if (EXT4_MB_GRP_WAS_TRIMMED(e4b.bd_info) &&
	    minblocks >= atomic_read(&EXT4_SB(sb)->s_last_trim_minblks))
		goto out;

4890 4891
	start = (e4b.bd_info->bb_first_free > start) ?
		e4b.bd_info->bb_first_free : start;
4892 4893 4894 4895 4896 4897 4898 4899

	while (start < max) {
		start = mb_find_next_zero_bit(bitmap, max, start);
		if (start >= max)
			break;
		next = mb_find_next_bit(bitmap, max, start);

		if ((next - start) >= minblocks) {
4900
			ext4_trim_extent(sb, start,
4901
					 next - start, group, &e4b);
4902 4903
			count += next - start;
		}
4904
		free_count += next - start;
4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917
		start = next + 1;

		if (fatal_signal_pending(current)) {
			count = -ERESTARTSYS;
			break;
		}

		if (need_resched()) {
			ext4_unlock_group(sb, group);
			cond_resched();
			ext4_lock_group(sb, group);
		}

4918
		if ((e4b.bd_info->bb_free - free_count) < minblocks)
4919 4920
			break;
	}
4921 4922 4923 4924

	if (!ret)
		EXT4_MB_GRP_SET_TRIMMED(e4b.bd_info);
out:
4925
	ext4_unlock_group(sb, group);
4926
	ext4_mb_unload_buddy(&e4b);
4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947

	ext4_debug("trimmed %d blocks in the group %d\n",
		count, group);

	return count;
}

/**
 * ext4_trim_fs() -- trim ioctl handle function
 * @sb:			superblock for filesystem
 * @range:		fstrim_range structure
 *
 * start:	First Byte to trim
 * len:		number of Bytes to trim from start
 * minlen:	minimum extent length in Bytes
 * ext4_trim_fs goes through all allocation groups containing Bytes from
 * start to start+len. For each such a group ext4_trim_all_free function
 * is invoked to trim all free space.
 */
int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
{
4948
	struct ext4_group_info *grp;
4949 4950 4951
	ext4_group_t first_group, last_group;
	ext4_group_t group, ngroups = ext4_get_groups_count(sb);
	ext4_grpblk_t cnt = 0, first_block, last_block;
4952
	uint64_t start, len, minlen, trimmed = 0;
4953 4954
	ext4_fsblk_t first_data_blk =
			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
4955 4956 4957 4958 4959 4960 4961 4962
	int ret = 0;

	start = range->start >> sb->s_blocksize_bits;
	len = range->len >> sb->s_blocksize_bits;
	minlen = range->minlen >> sb->s_blocksize_bits;

	if (unlikely(minlen > EXT4_BLOCKS_PER_GROUP(sb)))
		return -EINVAL;
4963 4964
	if (start + len <= first_data_blk)
		goto out;
4965 4966 4967 4968
	if (start < first_data_blk) {
		len -= first_data_blk - start;
		start = first_data_blk;
	}
4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981

	/* Determine first and last group to examine based on start and len */
	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) start,
				     &first_group, &first_block);
	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) (start + len),
				     &last_group, &last_block);
	last_group = (last_group > ngroups - 1) ? ngroups - 1 : last_group;
	last_block = EXT4_BLOCKS_PER_GROUP(sb);

	if (first_group > last_group)
		return -EINVAL;

	for (group = first_group; group <= last_group; group++) {
4982 4983 4984 4985 4986 4987
		grp = ext4_get_group_info(sb, group);
		/* We only do this if the grp has never been initialized */
		if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
			ret = ext4_mb_init_group(sb, group);
			if (ret)
				break;
4988 4989
		}

4990 4991 4992 4993 4994 4995 4996
		/*
		 * For all the groups except the last one, last block will
		 * always be EXT4_BLOCKS_PER_GROUP(sb), so we only need to
		 * change it for the last group in which case start +
		 * len < EXT4_BLOCKS_PER_GROUP(sb).
		 */
		if (first_block + len < EXT4_BLOCKS_PER_GROUP(sb))
J
Jan Kara 已提交
4997
			last_block = first_block + len;
4998
		len -= last_block - first_block;
4999

5000 5001
		if (grp->bb_free >= minlen) {
			cnt = ext4_trim_all_free(sb, group, first_block,
5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012
						last_block, minlen);
			if (cnt < 0) {
				ret = cnt;
				break;
			}
		}
		trimmed += cnt;
		first_block = 0;
	}
	range->len = trimmed * sb->s_blocksize;

5013 5014 5015
	if (!ret)
		atomic_set(&EXT4_SB(sb)->s_last_trim_minblks, minlen);

5016
out:
5017 5018
	return ret;
}