mballoc.c 137.9 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
 */

B
Bobi Jam 已提交
24
#include "ext4_jbd2.h"
25
#include "mballoc.h"
26
#include <linux/debugfs.h>
27
#include <linux/slab.h>
28 29
#include <trace/events/ext4.h>

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
/*
 * 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
54 55 56 57 58 59 60 61 62
 * 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.
63 64
 *
 * The main motivation for having small file use group preallocation is to
T
Theodore Ts'o 已提交
65
 * ensure that we have small files closer together on the disk.
66
 *
T
Theodore Ts'o 已提交
67 68 69 70
 * 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:
71 72 73
 *
 * pa_lstart -> the logical start block for this prealloc space
 * pa_pstart -> the physical start block for this prealloc space
74 75
 * pa_len    -> length for this prealloc space (in clusters)
 * pa_free   ->  free space available in this prealloc space (in clusters)
76 77 78
 *
 * The inode preallocation space is used looking at the _logical_ start
 * block. If only the logical file block falls within the range of prealloc
79 80
 * space we will consume the particular prealloc space. This makes sure that
 * we have contiguous physical blocks representing the file blocks
81 82 83 84 85 86 87
 *
 * 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
88
 * prealloc space. These are per CPU prealloc list represented as
89 90 91 92 93 94 95
 *
 * 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
96
 * enough free space (pa_free) within the prealloc space.
97 98 99 100 101 102 103 104 105 106 107 108
 *
 * 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                        }
109
 *  [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
 *
 *
 * 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
130 131
 * sbi->s_mb_group_prealloc.  The default value of s_mb_group_prealloc is
 * dependent on the cluster size; for non-bigalloc file systems, it is
132
 * 512 blocks. This can be tuned via
133
 * /sys/fs/ext4/<partition>/mb_group_prealloc. The value is represented in
134 135
 * terms of number of blocks. If we have mounted the file system with -O
 * stripe=<value> option the group prealloc request is normalized to the
136 137
 * the smallest multiple of the stripe value (sbi->s_stripe) which is
 * greater than the default mb_group_prealloc.
138
 *
139
 * The regular allocator (using the buddy cache) supports a few tunables.
140
 *
T
Theodore Ts'o 已提交
141 142 143
 * /sys/fs/ext4/<partition>/mb_min_to_scan
 * /sys/fs/ext4/<partition>/mb_max_to_scan
 * /sys/fs/ext4/<partition>/mb_order2_req
144
 *
T
Theodore Ts'o 已提交
145
 * The regular allocator uses buddy scan only if the request len is power of
146 147
 * 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
148
 * /sys/fs/ext4/<partition>/mb_order2_req.  If the request len is equal to
149
 * stripe size (sbi->s_stripe), we try to search for contiguous block in
T
Theodore Ts'o 已提交
150 151 152
 * 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.
153
 * min_to_scan indicate how long the mballoc __must__ look for a best
T
Theodore Ts'o 已提交
154
 * extent and max_to_scan indicates how long the mballoc __can__ look for a
155 156 157
 * 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
158
 * can be used for allocation. ext4_mb_good_group explains how the groups are
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
 * 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
 *
 */
341 342
static struct kmem_cache *ext4_pspace_cachep;
static struct kmem_cache *ext4_ac_cachep;
B
Bobi Jam 已提交
343
static struct kmem_cache *ext4_free_data_cachep;
344 345 346 347

/* 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 */
348
#define NR_GRPINFO_CACHES 8
349 350
static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];

351 352 353 354 355 356
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"
};

357 358
static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
					ext4_group_t group);
359 360
static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
						ext4_group_t group);
B
Bobi Jam 已提交
361 362
static void ext4_free_data_callback(struct super_block *sb,
				struct ext4_journal_cb_entry *jce, int rc);
363

364 365
static inline void *mb_correct_addr_and_bit(int *bit, void *addr)
{
366
#if BITS_PER_LONG == 64
367 368
	*bit += ((unsigned long) addr & 7UL) << 3;
	addr = (void *) ((unsigned long) addr & ~7UL);
369
#elif BITS_PER_LONG == 32
370 371
	*bit += ((unsigned long) addr & 3UL) << 3;
	addr = (void *) ((unsigned long) addr & ~3UL);
372 373 374
#else
#error "how many bits you are?!"
#endif
375 376
	return addr;
}
377 378 379 380 381 382 383

static inline int mb_test_bit(int bit, void *addr)
{
	/*
	 * ext4_test_bit on architecture like powerpc
	 * needs unsigned long aligned address
	 */
384
	addr = mb_correct_addr_and_bit(&bit, addr);
385 386 387 388 389
	return ext4_test_bit(bit, addr);
}

static inline void mb_set_bit(int bit, void *addr)
{
390
	addr = mb_correct_addr_and_bit(&bit, addr);
391 392 393 394 395
	ext4_set_bit(bit, addr);
}

static inline void mb_clear_bit(int bit, void *addr)
{
396
	addr = mb_correct_addr_and_bit(&bit, addr);
397 398 399
	ext4_clear_bit(bit, addr);
}

400 401
static inline int mb_find_next_zero_bit(void *addr, int max, int start)
{
402
	int fix = 0, ret, tmpmax;
403
	addr = mb_correct_addr_and_bit(&fix, addr);
404
	tmpmax = max + fix;
405 406
	start += fix;

407 408 409 410
	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
411 412 413 414
}

static inline int mb_find_next_bit(void *addr, int max, int start)
{
415
	int fix = 0, ret, tmpmax;
416
	addr = mb_correct_addr_and_bit(&fix, addr);
417
	tmpmax = max + fix;
418 419
	start += fix;

420 421 422 423
	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
	if (ret > max)
		return max;
	return ret;
424 425
}

426 427 428 429
static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
{
	char *bb;

430
	BUG_ON(e4b->bd_bitmap == e4b->bd_buddy);
431 432 433 434 435 436 437 438
	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
439 440
	if (order == 0) {
		*max = 1 << (e4b->bd_blkbits + 3);
441
		return e4b->bd_bitmap;
C
Coly Li 已提交
442
	}
443

444
	bb = e4b->bd_buddy + EXT4_SB(e4b->bd_sb)->s_mb_offsets[order];
445 446 447 448 449 450 451 452 453 454 455 456 457 458
	*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;
459
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
460 461 462
	for (i = 0; i < count; i++) {
		if (!mb_test_bit(first + i, e4b->bd_info->bb_bitmap)) {
			ext4_fsblk_t blocknr;
463 464

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
465
			blocknr += EXT4_C2B(EXT4_SB(sb), first + i);
466
			ext4_grp_locked_error(sb, e4b->bd_group,
467 468 469 470 471
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing block already freed "
					      "(bit %u)",
					      first + i);
472 473 474 475 476 477 478 479 480 481 482
		}
		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;
483
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
	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]) {
499 500 501 502 503
				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]);
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
				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;
			}

586
			/* both bits in buddy2 must be 1 */
587 588 589 590 591 592
			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(
593
					!mb_test_bit(k, e4b->bd_bitmap));
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627
			}
			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;
628 629
		pa = list_entry(cur, struct ext4_prealloc_space, pa_group_list);
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &groupnr, &k);
630
		MB_CHECK_ASSERT(groupnr == e4b->bd_group);
631
		for (i = 0; i < pa->pa_len; i++)
632 633 634 635 636 637
			MB_CHECK_ASSERT(mb_test_bit(k + i, buddy));
	}
	return 0;
}
#undef MB_CHECK_ASSERT
#define mb_check_buddy(e4b) __mb_check_buddy(e4b,	\
638
					__FILE__, __func__, __LINE__)
639 640 641 642
#else
#define mb_check_buddy(e4b)
#endif

643 644 645 646 647 648
/*
 * 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.
 */
649
static void ext4_mb_mark_free_simple(struct super_block *sb,
650
				void *buddy, ext4_grpblk_t first, ext4_grpblk_t len,
651 652 653
					struct ext4_group_info *grp)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
654 655 656
	ext4_grpblk_t min;
	ext4_grpblk_t max;
	ext4_grpblk_t chunk;
657 658
	unsigned short border;

659
	BUG_ON(len > EXT4_CLUSTERS_PER_GROUP(sb));
660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684

	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;
	}
}

685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
/*
 * 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;
		}
	}
}

706 707
static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
708 709 710
				void *buddy, void *bitmap, ext4_group_t group)
{
	struct ext4_group_info *grp = ext4_get_group_info(sb, group);
711
	ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb);
712 713 714
	ext4_grpblk_t i = 0;
	ext4_grpblk_t first;
	ext4_grpblk_t len;
715 716 717 718 719 720
	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 */
721
	i = mb_find_next_zero_bit(bitmap, max, 0);
722 723 724 725
	grp->bb_first_free = i;
	while (i < max) {
		fragments++;
		first = i;
726
		i = mb_find_next_bit(bitmap, max, i);
727 728 729 730 731 732 733
		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)
734
			i = mb_find_next_zero_bit(bitmap, max, i);
735 736 737 738
	}
	grp->bb_fragments = fragments;

	if (free != grp->bb_free) {
739
		ext4_grp_locked_error(sb, group, 0, 0,
740
				      "%u clusters in bitmap, %u in gd",
741
				      free, grp->bb_free);
742 743 744 745
		/*
		 * If we intent to continue, we consider group descritor
		 * corrupt and update bb_free using bitmap value
		 */
746 747
		grp->bb_free = free;
	}
748
	mb_set_largest_free_order(sb, grp);
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765

	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                        }
766
 * [ group 0 bitmap][ group 0 buddy] [group 1][ group 1]...
767 768 769 770 771 772 773
 *
 *
 * 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
774 775 776
 *
 * Locking note:  This routine takes the block group lock of all groups
 * for this page; do not hold this lock when calling this routine!
777 778 779 780
 */

static int ext4_mb_init_cache(struct page *page, char *incore)
{
781
	ext4_group_t ngroups;
782 783 784 785 786
	int blocksize;
	int blocks_per_page;
	int groups_per_page;
	int err = 0;
	int i;
787
	ext4_group_t first_group, group;
788 789 790 791 792 793 794
	int first_block;
	struct super_block *sb;
	struct buffer_head *bhs;
	struct buffer_head **bh;
	struct inode *inode;
	char *data;
	char *bitmap;
795
	struct ext4_group_info *grinfo;
796

797
	mb_debug(1, "init page %lu\n", page->index);
798 799 800

	inode = page->mapping->host;
	sb = inode->i_sb;
801
	ngroups = ext4_get_groups_count(sb);
802 803 804 805 806 807 808 809 810 811 812
	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) {
		i = sizeof(struct buffer_head *) * groups_per_page;
		bh = kzalloc(i, GFP_NOFS);
813 814
		if (bh == NULL) {
			err = -ENOMEM;
815
			goto out;
816
		}
817 818 819 820 821 822
	} else
		bh = &bhs;

	first_group = page->index * blocks_per_page / 2;

	/* read all groups the page covers into the cache */
823 824
	for (i = 0, group = first_group; i < groups_per_page; i++, group++) {
		if (group >= ngroups)
825 826
			break;

827
		grinfo = ext4_get_group_info(sb, group);
828 829 830 831 832 833 834 835 836 837
		/*
		 * 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;
		}
838 839
		if (!(bh[i] = ext4_read_block_bitmap_nowait(sb, group))) {
			err = -ENOMEM;
840
			goto out;
841
		}
842
		mb_debug(1, "read bitmap for group %u\n", group);
843 844 845
	}

	/* wait for I/O completion */
846 847 848
	for (i = 0, group = first_group; i < groups_per_page; i++, group++) {
		if (bh[i] && ext4_wait_block_bitmap(sb, group, bh[i])) {
			err = -EIO;
849
			goto out;
850 851
		}
	}
852 853 854 855 856 857

	first_block = page->index * blocks_per_page;
	for (i = 0; i < blocks_per_page; i++) {
		int group;

		group = (first_block + i) >> 1;
858
		if (group >= ngroups)
859 860
			break;

861 862 863 864
		if (!bh[group - first_group])
			/* skip initialized uptodate buddy */
			continue;

865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
		/*
		 * 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);
881
			mb_debug(1, "put buddy for group %u in page %lu/%x\n",
882
				group, page->index, i * blocksize);
883
			trace_ext4_mb_buddy_bitmap_load(sb, group);
884 885 886
			grinfo = ext4_get_group_info(sb, group);
			grinfo->bb_fragments = 0;
			memset(grinfo->bb_counters, 0,
887 888
			       sizeof(*grinfo->bb_counters) *
				(sb->s_blocksize_bits+2));
889 890 891
			/*
			 * incore got set to the group block bitmap below
			 */
892
			ext4_lock_group(sb, group);
893 894
			/* init the buddy */
			memset(data, 0xff, blocksize);
895
			ext4_mb_generate_buddy(sb, data, incore, group);
896
			ext4_unlock_group(sb, group);
897 898 899 900
			incore = NULL;
		} else {
			/* this is block of bitmap */
			BUG_ON(incore != NULL);
901
			mb_debug(1, "put bitmap for group %u in page %lu/%x\n",
902
				group, page->index, i * blocksize);
903
			trace_ext4_mb_bitmap_load(sb, group);
904 905 906 907 908 909 910

			/* 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);
911
			ext4_mb_generate_from_freelist(sb, data, group);
912 913 914 915 916 917 918 919 920 921 922 923
			ext4_unlock_group(sb, group);

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

out:
	if (bh) {
924
		for (i = 0; i < groups_per_page; i++)
925 926 927 928 929 930 931
			brelse(bh[i]);
		if (bh != &bhs)
			kfree(bh);
	}
	return err;
}

932
/*
933 934 935 936
 * 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.
937
 */
938 939
static int ext4_mb_get_buddy_page_lock(struct super_block *sb,
		ext4_group_t group, struct ext4_buddy *e4b)
940
{
941 942
	struct inode *inode = EXT4_SB(sb)->s_buddy_cache;
	int block, pnum, poff;
943
	int blocks_per_page;
944 945 946 947
	struct page *page;

	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;
948 949 950 951 952 953 954 955 956

	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;
957 958 959 960 961 962 963 964 965 966 967
	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;
968
	}
969 970 971 972 973 974 975 976 977 978

	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;
979 980
}

981
static void ext4_mb_put_buddy_page_lock(struct ext4_buddy *e4b)
982
{
983 984 985 986 987 988 989
	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);
990 991 992
	}
}

993 994 995 996 997
/*
 * 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!
 */
998 999 1000 1001 1002
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
{

	struct ext4_group_info *this_grp;
1003 1004 1005
	struct ext4_buddy e4b;
	struct page *page;
	int ret = 0;
1006 1007 1008 1009

	mb_debug(1, "init group %u\n", group);
	this_grp = ext4_get_group_info(sb, group);
	/*
1010 1011 1012 1013
	 * 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
1014
	 * would have pinned buddy page to page cache.
1015
	 */
1016 1017
	ret = ext4_mb_get_buddy_page_lock(sb, group, &e4b);
	if (ret || !EXT4_MB_GRP_NEED_INIT(this_grp)) {
1018 1019 1020 1021 1022 1023
		/*
		 * somebody initialized the group
		 * return without doing anything
		 */
		goto err;
	}
1024 1025 1026 1027 1028 1029

	page = e4b.bd_bitmap_page;
	ret = ext4_mb_init_cache(page, NULL);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
1030 1031 1032 1033 1034
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);

1035
	if (e4b.bd_buddy_page == NULL) {
1036 1037 1038 1039 1040
		/*
		 * If both the bitmap and buddy are in
		 * the same page we don't need to force
		 * init the buddy
		 */
1041 1042
		ret = 0;
		goto err;
1043
	}
1044 1045 1046 1047 1048 1049
	/* init buddy cache */
	page = e4b.bd_buddy_page;
	ret = ext4_mb_init_cache(page, e4b.bd_bitmap);
	if (ret)
		goto err;
	if (!PageUptodate(page)) {
1050 1051 1052 1053 1054
		ret = -EIO;
		goto err;
	}
	mark_page_accessed(page);
err:
1055
	ext4_mb_put_buddy_page_lock(&e4b);
1056 1057 1058
	return ret;
}

1059 1060 1061 1062 1063
/*
 * 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!
 */
1064 1065 1066
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
					struct ext4_buddy *e4b)
1067 1068 1069 1070 1071 1072
{
	int blocks_per_page;
	int block;
	int pnum;
	int poff;
	struct page *page;
1073
	int ret;
1074 1075 1076
	struct ext4_group_info *grp;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct inode *inode = sbi->s_buddy_cache;
1077

1078
	mb_debug(1, "load group %u\n", group);
1079 1080

	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1081
	grp = ext4_get_group_info(sb, group);
1082 1083

	e4b->bd_blkbits = sb->s_blocksize_bits;
1084
	e4b->bd_info = grp;
1085 1086 1087 1088 1089
	e4b->bd_sb = sb;
	e4b->bd_group = group;
	e4b->bd_buddy_page = NULL;
	e4b->bd_bitmap_page = NULL;

1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
	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;
	}

1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113
	/*
	 * 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)
1114 1115 1116 1117 1118 1119 1120 1121
			/*
			 * 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.
			 */
1122 1123 1124 1125 1126
			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)) {
1127 1128 1129 1130 1131
				ret = ext4_mb_init_cache(page, NULL);
				if (ret) {
					unlock_page(page);
					goto err;
				}
1132 1133 1134 1135 1136 1137
				mb_cmp_bitmaps(e4b, page_address(page) +
					       (poff * sb->s_blocksize));
			}
			unlock_page(page);
		}
	}
1138 1139
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1140
		goto err;
1141
	}
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156
	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);
1157 1158 1159 1160 1161 1162 1163
			if (!PageUptodate(page)) {
				ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
				if (ret) {
					unlock_page(page);
					goto err;
				}
			}
1164 1165 1166
			unlock_page(page);
		}
	}
1167 1168
	if (page == NULL || !PageUptodate(page)) {
		ret = -EIO;
1169
		goto err;
1170
	}
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180
	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:
1181 1182
	if (page)
		page_cache_release(page);
1183 1184 1185 1186 1187 1188
	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;
1189
	return ret;
1190 1191
}

1192
static void ext4_mb_unload_buddy(struct ext4_buddy *e4b)
1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
{
	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;

1206
	BUG_ON(e4b->bd_bitmap == e4b->bd_buddy);
1207 1208
	BUG_ON(block >= (1 << (e4b->bd_blkbits + 3)));

1209
	bb = e4b->bd_buddy;
1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
	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;
}

1222
static void mb_clear_bits(void *bm, int cur, int len)
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
{
	__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;
		}
1235
		mb_clear_bit(cur, bm);
1236 1237 1238 1239
		cur++;
	}
}

1240
void ext4_set_bits(void *bm, int cur, int len)
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
{
	__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;
		}
1253
		mb_set_bit(cur, bm);
1254 1255 1256 1257
		cur++;
	}
}

1258
static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
			  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));
1269
	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
1270 1271 1272 1273 1274 1275 1276 1277 1278
	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)
1279
		block = !mb_test_bit(first - 1, e4b->bd_bitmap);
1280
	if (first + count < EXT4_SB(sb)->s_mb_maxs[0])
1281
		max = !mb_test_bit(first + count, e4b->bd_bitmap);
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
	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;

1292
		if (!mb_test_bit(block, e4b->bd_bitmap)) {
1293
			ext4_fsblk_t blocknr;
1294 1295

			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
1296
			blocknr += EXT4_C2B(EXT4_SB(sb), block);
1297
			ext4_grp_locked_error(sb, e4b->bd_group,
1298 1299 1300 1301
					      inode ? inode->i_ino : 0,
					      blocknr,
					      "freeing already freed block "
					      "(bit %u)", block);
1302
		}
1303
		mb_clear_bit(block, e4b->bd_bitmap);
1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337
		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);
	}
1338
	mb_set_largest_free_order(sb, e4b->bd_info);
1339 1340 1341 1342 1343 1344 1345 1346 1347 1348
	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;
	void *buddy;

1349
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
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 1384
	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);
1385
		if (mb_test_bit(next, e4b->bd_bitmap))
1386 1387
			break;

1388
		order = mb_find_order_for_block(e4b, next);
1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411

		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);
1412
	assert_spin_locked(ext4_group_lock_ptr(e4b->bd_sb, e4b->bd_group));
1413 1414 1415 1416 1417 1418 1419 1420 1421
	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)
1422
		mlen = !mb_test_bit(start - 1, e4b->bd_bitmap);
1423
	if (start + len < EXT4_SB(e4b->bd_sb)->s_mb_maxs[0])
1424
		max = !mb_test_bit(start + len, e4b->bd_bitmap);
1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464
	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]++;
	}
1465
	mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
1466

1467
	ext4_set_bits(e4b->bd_bitmap, ex->fe_start, len0);
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
	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;

1497 1498 1499 1500 1501 1502 1503
	/*
	 * 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
	 */
1504 1505 1506 1507 1508
	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 */
1509
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530
		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;

1531 1532
	if (ac->ac_status == AC_STATUS_FOUND)
		return;
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
	/*
	 * 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);
1579 1580
	BUG_ON(ex->fe_len > EXT4_CLUSTERS_PER_GROUP(ac->ac_sb));
	BUG_ON(ex->fe_start >= EXT4_CLUSTERS_PER_GROUP(ac->ac_sb));
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 1627 1628 1629
	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);
}

1630 1631
static noinline_for_stack
int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652
					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);
1653
	ext4_mb_unload_buddy(e4b);
1654 1655 1656 1657

	return 0;
}

1658 1659
static noinline_for_stack
int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681
				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;

1682 1683
		start = ext4_group_first_block_no(ac->ac_sb, e4b->bd_group) +
			ex.fe_start;
1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707
		/* 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);
1708
	ext4_mb_unload_buddy(e4b);
1709 1710 1711 1712 1713 1714 1715 1716

	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
 */
1717 1718
static noinline_for_stack
void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
					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);

1736
		k = mb_find_next_zero_bit(buddy, max, 0);
1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760
		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.
 */
1761 1762
static noinline_for_stack
void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1763 1764 1765
					struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
1766
	void *bitmap = e4b->bd_bitmap;
1767 1768 1769 1770 1771 1772 1773 1774 1775 1776
	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) {
1777
		i = mb_find_next_zero_bit(bitmap,
1778 1779
						EXT4_CLUSTERS_PER_GROUP(sb), i);
		if (i >= EXT4_CLUSTERS_PER_GROUP(sb)) {
1780
			/*
1781
			 * IF we have corrupt bitmap, we won't find any
1782 1783 1784
			 * free blocks even though group info says we
			 * we have free blocks
			 */
1785
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
1786
					"%d free clusters as per "
1787
					"group info. But bitmap says 0",
1788
					free);
1789 1790 1791 1792 1793
			break;
		}

		mb_find_extent(e4b, 0, i, ac->ac_g_ex.fe_len, &ex);
		BUG_ON(ex.fe_len <= 0);
1794
		if (free < ex.fe_len) {
1795
			ext4_grp_locked_error(sb, e4b->bd_group, 0, 0,
1796
					"%d free clusters as per "
1797
					"group info. But got %d blocks",
1798
					free, ex.fe_len);
1799 1800 1801 1802 1803 1804
			/*
			 * The number of free blocks differs. This mostly
			 * indicate that the bitmap is corrupt. So exit
			 * without claiming the space.
			 */
			break;
1805
		}
1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817

		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
1818
 * we try to find stripe-aligned chunks for stripe-size-multiple requests
1819
 */
1820 1821
static noinline_for_stack
void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1822 1823 1824 1825
				 struct ext4_buddy *e4b)
{
	struct super_block *sb = ac->ac_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
1826
	void *bitmap = e4b->bd_bitmap;
1827 1828 1829 1830 1831 1832 1833 1834 1835
	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 */
1836 1837
	first_group_block = ext4_group_first_block_no(sb, e4b->bd_group);

1838 1839 1840 1841
	a = first_group_block + sbi->s_stripe - 1;
	do_div(a, sbi->s_stripe);
	i = (a * sbi->s_stripe) - first_group_block;

1842
	while (i < EXT4_CLUSTERS_PER_GROUP(sb)) {
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855
		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;
	}
}

1856
/* This is now called BEFORE we load the buddy bitmap. */
1857 1858 1859 1860
static int ext4_mb_good_group(struct ext4_allocation_context *ac,
				ext4_group_t group, int cr)
{
	unsigned free, fragments;
1861
	int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
1862 1863 1864
	struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);

	BUG_ON(cr < 0 || cr >= 4);
1865 1866 1867 1868 1869 1870 1871

	/* 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;
	}
1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883

	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);

1884 1885 1886
		if (grp->bb_largest_free_order < ac->ac_2order)
			return 0;

1887 1888 1889 1890 1891 1892
		/* 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;

1893
		return 1;
1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910
	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;
}

1911 1912
static noinline_for_stack int
ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1913
{
1914
	ext4_group_t ngroups, group, i;
1915 1916 1917 1918 1919 1920 1921 1922
	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);
1923
	ngroups = ext4_get_groups_count(sb);
1924
	/* non-extent files are limited to low blocks/groups */
1925
	if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)))
1926 1927
		ngroups = sbi->s_blockfile_groups;

1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947
	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
1948
	 * You can tune it via /sys/fs/ext4/<partition>/mb_order2_req
1949 1950 1951 1952 1953 1954 1955 1956 1957
	 */
	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;
	}

1958 1959
	/* if stream allocation is enabled, use global goal */
	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
1960 1961 1962 1963 1964 1965
		/* 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);
	}
1966

1967 1968 1969 1970 1971 1972 1973 1974 1975
	/* 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;
1976 1977 1978 1979 1980 1981
		/*
		 * searching for the right group start
		 * from the goal value specified
		 */
		group = ac->ac_g_ex.fe_group;

1982 1983
		for (i = 0; i < ngroups; group++, i++) {
			if (group == ngroups)
1984 1985
				group = 0;

1986 1987
			/* This now checks without needing the buddy page */
			if (!ext4_mb_good_group(ac, group, cr))
1988 1989 1990 1991 1992 1993 1994
				continue;

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

			ext4_lock_group(sb, group);
1995 1996 1997 1998 1999

			/*
			 * We need to check again after locking the
			 * block group
			 */
2000 2001
			if (!ext4_mb_good_group(ac, group, cr)) {
				ext4_unlock_group(sb, group);
2002
				ext4_mb_unload_buddy(&e4b);
2003 2004 2005 2006
				continue;
			}

			ac->ac_groups_scanned++;
2007
			if (cr == 0)
2008
				ext4_mb_simple_scan_group(ac, &e4b);
2009 2010
			else if (cr == 1 && sbi->s_stripe &&
					!(ac->ac_g_ex.fe_len % sbi->s_stripe))
2011 2012 2013 2014 2015
				ext4_mb_scan_aligned(ac, &e4b);
			else
				ext4_mb_complex_scan_group(ac, &e4b);

			ext4_unlock_group(sb, group);
2016
			ext4_mb_unload_buddy(&e4b);
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056

			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;

2057
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2058 2059
		return NULL;
	group = *pos + 1;
2060
	return (void *) ((unsigned long) group);
2061 2062 2063 2064 2065 2066 2067 2068
}

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;
2069
	if (*pos < 0 || *pos >= ext4_get_groups_count(sb))
2070 2071
		return NULL;
	group = *pos + 1;
2072
	return (void *) ((unsigned long) group);
2073 2074 2075 2076 2077
}

static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
{
	struct super_block *sb = seq->private;
2078
	ext4_group_t group = (ext4_group_t) ((unsigned long) v);
2079 2080 2081 2082 2083
	int i;
	int err;
	struct ext4_buddy e4b;
	struct sg {
		struct ext4_group_info info;
2084
		ext4_grpblk_t counters[16];
2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
	} 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) {
2100
		seq_printf(seq, "#%-5u: I/O error\n", group);
2101 2102 2103 2104 2105
		return 0;
	}
	ext4_lock_group(sb, group);
	memcpy(&sg, ext4_get_group_info(sb, group), i);
	ext4_unlock_group(sb, group);
2106
	ext4_mb_unload_buddy(&e4b);
2107

2108
	seq_printf(seq, "#%-5u: %-5u %-5u %-5u [", group, sg.info.bb_free,
2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121
			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)
{
}

2122
static const struct seq_operations ext4_mb_seq_groups_ops = {
2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135
	.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) {
2136
		struct seq_file *m = file->private_data;
2137 2138 2139 2140 2141 2142
		m->private = sb;
	}
	return rc;

}

2143
static const struct file_operations ext4_mb_seq_groups_fops = {
2144 2145 2146 2147 2148 2149 2150
	.owner		= THIS_MODULE,
	.open		= ext4_mb_seq_groups_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= seq_release,
};

2151 2152 2153 2154 2155 2156 2157 2158
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;
}
2159 2160

/* Create and initialize ext4_group_info data for the given group. */
2161
int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2162 2163
			  struct ext4_group_desc *desc)
{
2164
	int i;
2165 2166 2167
	int metalen = 0;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
	struct ext4_group_info **meta_group_info;
2168
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179

	/*
	 * 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) {
2180
			ext4_msg(sb, KERN_ERR, "can't allocate mem "
2181
				 "for a buddy group");
2182 2183 2184 2185 2186 2187 2188 2189 2190 2191
			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);

2192
	meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
2193
	if (meta_group_info[i] == NULL) {
2194
		ext4_msg(sb, KERN_ERR, "can't allocate buddy mem");
2195 2196
		goto exit_group_info;
	}
2197
	memset(meta_group_info[i], 0, kmem_cache_size(cachep));
2198 2199 2200 2201 2202 2203 2204 2205 2206
	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 =
2207
			ext4_free_clusters_after_init(sb, group, desc);
2208 2209
	} else {
		meta_group_info[i]->bb_free =
2210
			ext4_free_group_clusters(sb, desc);
2211 2212 2213
	}

	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2214
	init_rwsem(&meta_group_info[i]->alloc_sem);
2215
	meta_group_info[i]->bb_free_root = RB_ROOT;
2216
	meta_group_info[i]->bb_largest_free_order = -1;  /* uninit */
2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235

#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 */
2236
	if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
2237
		kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]);
2238 2239
		sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] = NULL;
	}
2240 2241 2242 2243
exit_meta_group_info:
	return -ENOMEM;
} /* ext4_mb_add_groupinfo */

2244 2245
static int ext4_mb_init_backend(struct super_block *sb)
{
2246
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2247 2248
	ext4_group_t i;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2249 2250 2251 2252 2253
	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;
2254
	struct kmem_cache *cachep;
2255 2256

	/* This is the number of blocks used by GDT */
2257
	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272
				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);
2273

2274 2275 2276 2277 2278 2279 2280 2281 2282 2283
	/*
	 * 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;
2284 2285 2286
	/* 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. */
2287
	sbi->s_group_info = ext4_kvzalloc(array_size, GFP_KERNEL);
2288
	if (sbi->s_group_info == NULL) {
2289
		ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
2290 2291 2292 2293
		return -ENOMEM;
	}
	sbi->s_buddy_cache = new_inode(sb);
	if (sbi->s_buddy_cache == NULL) {
2294
		ext4_msg(sb, KERN_ERR, "can't get new inode");
2295 2296
		goto err_freesgi;
	}
2297 2298 2299 2300 2301
	/* 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;
2302
	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
2303
	for (i = 0; i < ngroups; i++) {
2304 2305
		desc = ext4_get_group_desc(sb, i, NULL);
		if (desc == NULL) {
2306
			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
2307 2308
			goto err_freebuddy;
		}
2309 2310
		if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
			goto err_freebuddy;
2311 2312 2313 2314 2315
	}

	return 0;

err_freebuddy:
2316
	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2317
	while (i-- > 0)
2318
		kmem_cache_free(cachep, ext4_get_group_info(sb, i));
2319
	i = num_meta_group_infos;
2320
	while (i-- > 0)
2321 2322 2323
		kfree(sbi->s_group_info[i]);
	iput(sbi->s_buddy_cache);
err_freesgi:
2324
	ext4_kvfree(sbi->s_group_info);
2325 2326 2327
	return -ENOMEM;
}

2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365
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);

2366 2367
	ext4_groupinfo_caches[cache_index] = cachep;

2368 2369
	mutex_unlock(&ext4_grpinfo_slab_create_mutex);
	if (!cachep) {
2370 2371
		printk(KERN_EMERG
		       "EXT4-fs: no memory for groupinfo slab cache\n");
2372 2373 2374 2375 2376 2377
		return -ENOMEM;
	}

	return 0;
}

2378 2379 2380
int ext4_mb_init(struct super_block *sb, int needs_recovery)
{
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2381
	unsigned i, j;
2382 2383
	unsigned offset;
	unsigned max;
2384
	int ret;
2385

2386
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
2387 2388 2389

	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_offsets == NULL) {
2390 2391
		ret = -ENOMEM;
		goto out;
2392
	}
2393

2394
	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
2395 2396
	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
	if (sbi->s_mb_maxs == NULL) {
2397 2398 2399 2400
		ret = -ENOMEM;
		goto out;
	}

2401 2402 2403
	ret = ext4_groupinfo_create_slab(sb->s_blocksize);
	if (ret < 0)
		goto out;
2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427

	/* 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);

	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;
2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441
	/*
	 * The default group preallocation is 512, which for 4k block
	 * sizes translates to 2 megabytes.  However for bigalloc file
	 * systems, this is probably too big (i.e, if the cluster size
	 * is 1 megabyte, then group preallocation size becomes half a
	 * gigabyte!).  As a default, we will keep a two megabyte
	 * group pralloc size for cluster sizes up to 64k, and after
	 * that, we will force a minimum group preallocation size of
	 * 32 clusters.  This translates to 8 megs when the cluster
	 * size is 256k, and 32 megs when the cluster size is 1 meg,
	 * which seems reasonable as a default.
	 */
	sbi->s_mb_group_prealloc = max(MB_DEFAULT_GROUP_PREALLOC >>
				       sbi->s_cluster_bits, 32);
2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453
	/*
	 * 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);
	}
2454

2455
	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
2456
	if (sbi->s_locality_groups == NULL) {
2457
		ret = -ENOMEM;
2458
		goto out_free_groupinfo_slab;
2459
	}
2460
	for_each_possible_cpu(i) {
2461
		struct ext4_locality_group *lg;
2462
		lg = per_cpu_ptr(sbi->s_locality_groups, i);
2463
		mutex_init(&lg->lg_mutex);
2464 2465
		for (j = 0; j < PREALLOC_TB_SIZE; j++)
			INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
2466 2467 2468
		spin_lock_init(&lg->lg_prealloc_lock);
	}

2469 2470
	/* init file for buddy data */
	ret = ext4_mb_init_backend(sb);
2471 2472
	if (ret != 0)
		goto out_free_locality_groups;
2473

2474 2475 2476
	if (sbi->s_proc)
		proc_create_data("mb_groups", S_IRUGO, sbi->s_proc,
				 &ext4_mb_seq_groups_fops, sb);
2477

2478 2479 2480 2481 2482 2483 2484
	return 0;

out_free_locality_groups:
	free_percpu(sbi->s_locality_groups);
	sbi->s_locality_groups = NULL;
out_free_groupinfo_slab:
	ext4_groupinfo_destroy_slabs();
2485
out:
2486 2487 2488 2489
	kfree(sbi->s_mb_offsets);
	sbi->s_mb_offsets = NULL;
	kfree(sbi->s_mb_maxs);
	sbi->s_mb_maxs = NULL;
2490
	return ret;
2491 2492
}

2493
/* need to called with the ext4 group lock held */
2494 2495 2496 2497 2498 2499 2500 2501 2502 2503
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++;
2504
		kmem_cache_free(ext4_pspace_cachep, pa);
2505 2506
	}
	if (count)
2507
		mb_debug(1, "mballoc: %u PAs left\n", count);
2508 2509 2510 2511 2512

}

int ext4_mb_release(struct super_block *sb)
{
2513
	ext4_group_t ngroups = ext4_get_groups_count(sb);
2514 2515 2516 2517
	ext4_group_t i;
	int num_meta_group_infos;
	struct ext4_group_info *grinfo;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
2518
	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
2519 2520

	if (sbi->s_group_info) {
2521
		for (i = 0; i < ngroups; i++) {
2522 2523 2524 2525 2526 2527 2528
			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);
2529
			kmem_cache_free(cachep, grinfo);
2530
		}
2531
		num_meta_group_infos = (ngroups +
2532 2533 2534 2535
				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]);
2536
		ext4_kvfree(sbi->s_group_info);
2537 2538 2539 2540 2541 2542
	}
	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) {
2543 2544
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %u blocks %u reqs (%u success)",
2545 2546 2547
				atomic_read(&sbi->s_bal_allocated),
				atomic_read(&sbi->s_bal_reqs),
				atomic_read(&sbi->s_bal_success));
2548 2549 2550
		ext4_msg(sb, KERN_INFO,
		      "mballoc: %u extents scanned, %u goal hits, "
				"%u 2^N hits, %u breaks, %u lost",
2551 2552 2553 2554 2555
				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));
2556 2557
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %lu generated and it took %Lu",
2558
				sbi->s_mb_buddies_generated,
2559
				sbi->s_mb_generation_time);
2560 2561
		ext4_msg(sb, KERN_INFO,
		       "mballoc: %u preallocated, %u discarded",
2562 2563 2564 2565
				atomic_read(&sbi->s_mb_preallocated),
				atomic_read(&sbi->s_mb_discarded));
	}

2566
	free_percpu(sbi->s_locality_groups);
2567 2568
	if (sbi->s_proc)
		remove_proc_entry("mb_groups", sbi->s_proc);
2569 2570 2571 2572

	return 0;
}

2573
static inline int ext4_issue_discard(struct super_block *sb,
2574
		ext4_group_t block_group, ext4_grpblk_t cluster, int count)
2575 2576 2577
{
	ext4_fsblk_t discard_block;

2578 2579 2580
	discard_block = (EXT4_C2B(EXT4_SB(sb), cluster) +
			 ext4_group_first_block_no(sb, block_group));
	count = EXT4_C2B(EXT4_SB(sb), count);
2581 2582
	trace_ext4_discard_blocks(sb,
			(unsigned long long) discard_block, count);
2583
	return sb_issue_discard(sb, discard_block, count, GFP_NOFS, 0);
2584 2585
}

2586 2587 2588 2589
/*
 * 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.
 */
B
Bobi Jam 已提交
2590 2591 2592
static void ext4_free_data_callback(struct super_block *sb,
				    struct ext4_journal_cb_entry *jce,
				    int rc)
2593
{
B
Bobi Jam 已提交
2594
	struct ext4_free_data *entry = (struct ext4_free_data *)jce;
2595
	struct ext4_buddy e4b;
2596
	struct ext4_group_info *db;
2597
	int err, count = 0, count2 = 0;
2598

B
Bobi Jam 已提交
2599 2600
	mb_debug(1, "gonna free %u blocks in group %u (0x%p):",
		 entry->efd_count, entry->efd_group, entry);
2601

B
Bobi Jam 已提交
2602 2603 2604
	if (test_opt(sb, DISCARD))
		ext4_issue_discard(sb, entry->efd_group,
				   entry->efd_start_cluster, entry->efd_count);
2605

B
Bobi Jam 已提交
2606 2607 2608
	err = ext4_mb_load_buddy(sb, entry->efd_group, &e4b);
	/* we expect to find existing buddy because it's pinned */
	BUG_ON(err != 0);
2609

2610

B
Bobi Jam 已提交
2611 2612 2613 2614 2615 2616 2617 2618
	db = e4b.bd_info;
	/* there are blocks to put in buddy to make them really free */
	count += entry->efd_count;
	count2++;
	ext4_lock_group(sb, entry->efd_group);
	/* Take it out of per group rb tree */
	rb_erase(&entry->efd_node, &(db->bb_free_root));
	mb_free_blocks(NULL, &e4b, entry->efd_start_cluster, entry->efd_count);
2619

B
Bobi Jam 已提交
2620 2621 2622 2623 2624 2625 2626 2627
	/*
	 * 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);
2628

B
Bobi Jam 已提交
2629 2630 2631 2632 2633 2634
	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);
2635
	}
B
Bobi Jam 已提交
2636 2637 2638
	ext4_unlock_group(sb, entry->efd_group);
	kmem_cache_free(ext4_free_data_cachep, entry);
	ext4_mb_unload_buddy(&e4b);
2639

2640
	mb_debug(1, "freed %u blocks in %u structures\n", count, count2);
2641 2642
}

2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676
#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

2677
int __init ext4_init_mballoc(void)
2678
{
2679 2680
	ext4_pspace_cachep = KMEM_CACHE(ext4_prealloc_space,
					SLAB_RECLAIM_ACCOUNT);
2681 2682 2683
	if (ext4_pspace_cachep == NULL)
		return -ENOMEM;

2684 2685
	ext4_ac_cachep = KMEM_CACHE(ext4_allocation_context,
				    SLAB_RECLAIM_ACCOUNT);
2686 2687 2688 2689
	if (ext4_ac_cachep == NULL) {
		kmem_cache_destroy(ext4_pspace_cachep);
		return -ENOMEM;
	}
2690

B
Bobi Jam 已提交
2691 2692 2693
	ext4_free_data_cachep = KMEM_CACHE(ext4_free_data,
					   SLAB_RECLAIM_ACCOUNT);
	if (ext4_free_data_cachep == NULL) {
2694 2695 2696 2697
		kmem_cache_destroy(ext4_pspace_cachep);
		kmem_cache_destroy(ext4_ac_cachep);
		return -ENOMEM;
	}
2698
	ext4_create_debugfs_entry();
2699 2700 2701
	return 0;
}

2702
void ext4_exit_mballoc(void)
2703
{
2704
	/*
2705 2706 2707 2708
	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
	 * before destroying the slab cache.
	 */
	rcu_barrier();
2709
	kmem_cache_destroy(ext4_pspace_cachep);
2710
	kmem_cache_destroy(ext4_ac_cachep);
B
Bobi Jam 已提交
2711
	kmem_cache_destroy(ext4_free_data_cachep);
2712
	ext4_groupinfo_destroy_slabs();
2713
	ext4_remove_debugfs_entry();
2714 2715 2716 2717
}


/*
2718
 * Check quota and mark chosen space (ac->ac_b_ex) non-free in bitmaps
2719 2720
 * Returns 0 if success or error code
 */
2721 2722
static noinline_for_stack int
ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2723
				handle_t *handle, unsigned int reserv_clstrs)
2724 2725 2726 2727 2728 2729 2730
{
	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;
2731
	int err, len;
2732 2733 2734 2735 2736 2737 2738 2739

	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;
2740
	bitmap_bh = ext4_read_block_bitmap(sb, ac->ac_b_ex.fe_group);
2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
	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;

2753
	ext4_debug("using block group %u(%d)\n", ac->ac_b_ex.fe_group,
2754
			ext4_free_group_clusters(sb, gdp));
2755

2756 2757 2758 2759
	err = ext4_journal_get_write_access(handle, gdp_bh);
	if (err)
		goto out_err;

2760
	block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
2761

2762
	len = EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
2763
	if (!ext4_data_block_valid(sbi, block, len)) {
2764
		ext4_error(sb, "Allocating blocks %llu-%llu which overlap "
2765
			   "fs metadata", block, block+len);
2766 2767 2768 2769
		/* File system mounted not to panic on error
		 * Fix the bitmap and repeat the block allocation
		 * We leak some of the blocks here.
		 */
2770
		ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2771 2772
		ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
			      ac->ac_b_ex.fe_len);
2773
		ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2774
		err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2775 2776 2777
		if (!err)
			err = -EAGAIN;
		goto out_err;
2778
	}
2779 2780

	ext4_lock_group(sb, ac->ac_b_ex.fe_group);
2781 2782 2783 2784 2785 2786 2787 2788 2789
#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
2790 2791
	ext4_set_bits(bitmap_bh->b_data, ac->ac_b_ex.fe_start,
		      ac->ac_b_ex.fe_len);
2792 2793
	if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
		gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
2794
		ext4_free_group_clusters_set(sb, gdp,
2795
					     ext4_free_clusters_after_init(sb,
2796
						ac->ac_b_ex.fe_group, gdp));
2797
	}
2798 2799
	len = ext4_free_group_clusters(sb, gdp) - ac->ac_b_ex.fe_len;
	ext4_free_group_clusters_set(sb, gdp, len);
2800
	gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
2801 2802

	ext4_unlock_group(sb, ac->ac_b_ex.fe_group);
2803
	percpu_counter_sub(&sbi->s_freeclusters_counter, ac->ac_b_ex.fe_len);
2804
	/*
2805
	 * Now reduce the dirty block count also. Should not go negative
2806
	 */
2807 2808
	if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
		/* release all the reserved blocks if non delalloc */
2809 2810
		percpu_counter_sub(&sbi->s_dirtyclusters_counter,
				   reserv_clstrs);
2811

2812 2813 2814
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi,
							  ac->ac_b_ex.fe_group);
2815
		atomic_sub(ac->ac_b_ex.fe_len,
2816
			   &sbi->s_flex_groups[flex_group].free_clusters);
2817 2818
	}

2819
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
2820 2821
	if (err)
		goto out_err;
2822
	err = ext4_handle_dirty_metadata(handle, NULL, gdp_bh);
2823 2824

out_err:
T
Theodore Ts'o 已提交
2825
	ext4_mark_super_dirty(sb);
2826
	brelse(bitmap_bh);
2827 2828 2829 2830 2831
	return err;
}

/*
 * here we normalize request for locality group
2832 2833 2834
 * 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
2835
 * /sys/fs/ext4/<partition>/mb_group_prealloc
2836 2837 2838 2839 2840 2841 2842 2843 2844
 *
 * 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);
2845
	ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
2846
	mb_debug(1, "#%u: goal %u blocks for locality group\n",
2847 2848 2849 2850 2851 2852 2853
		current->pid, ac->ac_g_ex.fe_len);
}

/*
 * Normalization means making request better in terms of
 * size and alignment
 */
2854 2855
static noinline_for_stack void
ext4_mb_normalize_request(struct ext4_allocation_context *ac,
2856 2857
				struct ext4_allocation_request *ar)
{
2858
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
2859 2860
	int bsbits, max;
	ext4_lblk_t end;
2861 2862
	loff_t size, start_off;
	loff_t orig_size __maybe_unused;
2863
	ext4_lblk_t start;
2864
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
2865
	struct ext4_prealloc_space *pa;
2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889

	/* 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 */
2890
	size = ac->ac_o_ex.fe_logical + EXT4_C2B(sbi, ac->ac_o_ex.fe_len);
2891 2892 2893
	size = size << bsbits;
	if (size < i_size_read(ac->ac_inode))
		size = i_size_read(ac->ac_inode);
2894
	orig_size = size;
2895

2896 2897
	/* max size of free chunks */
	max = 2 << bsbits;
2898

2899 2900
#define NRL_CHECK_SIZE(req, size, max, chunk_size)	\
		(req <= (size) || max <= (chunk_size))
2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918

	/* 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;
2919
	} else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {
2920
		start_off = ((loff_t)ac->ac_o_ex.fe_logical >>
2921 2922 2923
						(21 - bsbits)) << 21;
		size = 2 * 1024 * 1024;
	} else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {
2924 2925 2926 2927
		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,
2928
					(8<<20)>>bsbits, max, 8 * 1024)) {
2929 2930 2931 2932 2933 2934 2935
		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;
	}
2936 2937
	size = size >> bsbits;
	start = start_off >> bsbits;
2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950

	/* 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();
2951
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
2952
		ext4_lblk_t pa_end;
2953 2954 2955 2956 2957 2958 2959 2960 2961

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

2962 2963
		pa_end = pa->pa_lstart + EXT4_C2B(EXT4_SB(ac->ac_sb),
						  pa->pa_len);
2964 2965 2966 2967 2968

		/* 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));

2969 2970
		/* skip PAs this normalized request doesn't overlap with */
		if (pa->pa_lstart >= end || pa_end <= start) {
2971 2972 2973 2974 2975
			spin_unlock(&pa->pa_lock);
			continue;
		}
		BUG_ON(pa->pa_lstart <= start && pa_end >= end);

2976
		/* adjust start or end to be adjacent to this pa */
2977 2978 2979
		if (pa_end <= ac->ac_o_ex.fe_logical) {
			BUG_ON(pa_end < start);
			start = pa_end;
2980
		} else if (pa->pa_lstart > ac->ac_o_ex.fe_logical) {
2981 2982 2983 2984 2985 2986 2987 2988 2989 2990
			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();
2991
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
2992
		ext4_lblk_t pa_end;
2993

2994 2995
		spin_lock(&pa->pa_lock);
		if (pa->pa_deleted == 0) {
2996 2997
			pa_end = pa->pa_lstart + EXT4_C2B(EXT4_SB(ac->ac_sb),
							  pa->pa_len);
2998 2999 3000 3001 3002 3003 3004 3005
			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) {
3006 3007 3008 3009
		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);
3010 3011 3012
	}
	BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
			start > ac->ac_o_ex.fe_logical);
3013
	BUG_ON(size <= 0 || size > EXT4_CLUSTERS_PER_GROUP(ac->ac_sb));
3014 3015 3016 3017 3018 3019

	/* 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;
3020
	ac->ac_g_ex.fe_len = EXT4_NUM_B2C(sbi, size);
3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037

	/* 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;
	}

3038
	mb_debug(1, "goal: %u(was %u) blocks at %u\n", (unsigned) size,
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048
		(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);
3049
		if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
3050 3051 3052 3053 3054 3055 3056 3057 3058
			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);
	}

3059 3060 3061 3062
	if (ac->ac_op == EXT4_MB_HISTORY_ALLOC)
		trace_ext4_mballoc_alloc(ac);
	else
		trace_ext4_mballoc_prealloc(ac);
3063 3064
}

3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082
/*
 * 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;
	}

}

3083 3084 3085 3086 3087 3088
/*
 * use blocks preallocated to inode
 */
static void ext4_mb_use_inode_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
3089
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
3090 3091 3092 3093 3094 3095
	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);
3096 3097 3098
	end = min(pa->pa_pstart + EXT4_C2B(sbi, pa->pa_len),
		  start + EXT4_C2B(sbi, ac->ac_o_ex.fe_len));
	len = EXT4_NUM_B2C(sbi, end - start);
3099 3100 3101 3102 3103 3104 3105
	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);
3106
	BUG_ON(end > pa->pa_pstart + EXT4_C2B(sbi, pa->pa_len));
3107 3108 3109
	BUG_ON(pa->pa_free < len);
	pa->pa_free -= len;

3110
	mb_debug(1, "use %llu/%u from inode pa %p\n", start, len, pa);
3111 3112 3113 3114 3115 3116 3117 3118
}

/*
 * use blocks preallocated to locality group
 */
static void ext4_mb_use_group_pa(struct ext4_allocation_context *ac,
				struct ext4_prealloc_space *pa)
{
3119
	unsigned int len = ac->ac_o_ex.fe_len;
3120

3121 3122 3123 3124 3125 3126 3127 3128
	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
3129
	 * possible race when the group is being loaded concurrently
3130
	 * instead we correct pa later, after blocks are marked
3131 3132
	 * in on-disk bitmap -- see ext4_mb_release_context()
	 * Other CPUs are prevented from allocating from this pa by lg_mutex
3133
	 */
3134
	mb_debug(1, "use %u/%u from group pa %p\n", pa->pa_lstart-len, len, pa);
3135 3136
}

3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156
/*
 * 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);

3157
	if (cur_distance <= new_distance)
3158 3159 3160 3161 3162 3163 3164 3165
		return cpa;

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

3166 3167 3168
/*
 * search goal blocks in preallocated space
 */
3169 3170
static noinline_for_stack int
ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
3171
{
3172
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
3173
	int order, i;
3174 3175
	struct ext4_inode_info *ei = EXT4_I(ac->ac_inode);
	struct ext4_locality_group *lg;
3176 3177
	struct ext4_prealloc_space *pa, *cpa = NULL;
	ext4_fsblk_t goal_block;
3178 3179 3180 3181 3182 3183 3184

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

	/* first, try per-file preallocation */
	rcu_read_lock();
3185
	list_for_each_entry_rcu(pa, &ei->i_prealloc_list, pa_inode_list) {
3186 3187 3188 3189

		/* 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 ||
3190 3191
		    ac->ac_o_ex.fe_logical >= (pa->pa_lstart +
					       EXT4_C2B(sbi, pa->pa_len)))
3192 3193
			continue;

3194
		/* non-extent files can't have physical blocks past 2^32 */
3195
		if (!(ext4_test_inode_flag(ac->ac_inode, EXT4_INODE_EXTENTS)) &&
3196 3197
		    (pa->pa_pstart + EXT4_C2B(sbi, pa->pa_len) >
		     EXT4_MAX_BLOCK_FILE_PHYS))
3198 3199
			continue;

3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221
		/* 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;
3222 3223 3224 3225 3226
	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;

3227
	goal_block = ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex);
3228 3229 3230 3231
	/*
	 * search for the prealloc space that is having
	 * minimal distance from the goal block.
	 */
3232 3233 3234 3235 3236 3237 3238
	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) {
3239 3240 3241

				cpa = ext4_mb_check_group_pa(goal_block,
								pa, cpa);
3242
			}
3243 3244
			spin_unlock(&pa->pa_lock);
		}
3245
		rcu_read_unlock();
3246
	}
3247 3248 3249 3250 3251
	if (cpa) {
		ext4_mb_use_group_pa(ac, cpa);
		ac->ac_criteria = 20;
		return 1;
	}
3252 3253 3254
	return 0;
}

3255 3256 3257 3258
/*
 * 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
3259
 * Need to be called with the ext4 group lock held
3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271
 */
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) {
B
Bobi Jam 已提交
3272 3273
		entry = rb_entry(n, struct ext4_free_data, efd_node);
		ext4_set_bits(bitmap, entry->efd_start_cluster, entry->efd_count);
3274 3275 3276 3277 3278
		n = rb_next(n);
	}
	return;
}

3279 3280 3281
/*
 * the function goes through all preallocation in this group and marks them
 * used in in-core bitmap. buddy must be generated from this bitmap
3282
 * Need to be called with ext4 group lock held
3283
 */
3284 3285
static noinline_for_stack
void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313
					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 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);
3314
		ext4_set_bits(bitmap, start, len);
3315 3316
		preallocated += len;
	}
3317
	mb_debug(1, "prellocated %u for group %u\n", preallocated, group);
3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333
}

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)
{
3334
	ext4_group_t grp;
3335
	ext4_fsblk_t grp_blk;
3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349

	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);

3350
	grp_blk = pa->pa_pstart;
3351
	/*
3352 3353 3354 3355
	 * 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)
3356 3357 3358
		grp_blk--;

	ext4_get_group_no_and_offset(sb, grp_blk, &grp, NULL);
3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387

	/*
	 * 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
 */
3388 3389
static noinline_for_stack int
ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
3390 3391
{
	struct super_block *sb = ac->ac_sb;
3392
	struct ext4_sb_info *sbi = EXT4_SB(sb);
3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423
	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 */
3424
		wins = EXT4_C2B(sbi, ac->ac_b_ex.fe_len - ac->ac_o_ex.fe_len);
3425 3426 3427 3428

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

3429 3430
		offs = ac->ac_o_ex.fe_logical %
			EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
3431 3432 3433
		if (offs && offs < win)
			win = offs;

3434 3435
		ac->ac_b_ex.fe_logical = ac->ac_o_ex.fe_logical -
			EXT4_B2C(sbi, win);
3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449
		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);
3450 3451
	INIT_LIST_HEAD(&pa->pa_inode_list);
	INIT_LIST_HEAD(&pa->pa_group_list);
3452
	pa->pa_deleted = 0;
3453
	pa->pa_type = MB_INODE_PA;
3454

3455
	mb_debug(1, "new inode pa %p: %llu/%u for %u\n", pa,
3456
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
3457
	trace_ext4_mb_new_inode_pa(ac, pa);
3458 3459

	ext4_mb_use_inode_pa(ac, pa);
3460
	atomic_add(pa->pa_free, &sbi->s_mb_preallocated);
3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481

	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
 */
3482 3483
static noinline_for_stack int
ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509
{
	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);
3510
	INIT_LIST_HEAD(&pa->pa_inode_list);
3511
	INIT_LIST_HEAD(&pa->pa_group_list);
3512
	pa->pa_deleted = 0;
3513
	pa->pa_type = MB_GROUP_PA;
3514

3515
	mb_debug(1, "new group pa %p: %llu/%u for %u\n", pa,
3516 3517
			pa->pa_pstart, pa->pa_len, pa->pa_lstart);
	trace_ext4_mb_new_group_pa(ac, pa);
3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532

	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);

3533 3534 3535 3536
	/*
	 * We will later add the new pa to the right bucket
	 * after updating the pa_free in ext4_mb_release_context
	 */
3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558
	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
 */
3559 3560
static noinline_for_stack int
ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3561
			struct ext4_prealloc_space *pa)
3562 3563 3564
{
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
3565 3566
	unsigned int end;
	unsigned int next;
3567 3568
	ext4_group_t group;
	ext4_grpblk_t bit;
3569
	unsigned long long grp_blk_start;
3570 3571 3572 3573 3574
	int err = 0;
	int free = 0;

	BUG_ON(pa->pa_deleted == 0);
	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
3575
	grp_blk_start = pa->pa_pstart - EXT4_C2B(sbi, bit);
3576 3577 3578 3579
	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
	end = bit + pa->pa_len;

	while (bit < end) {
3580
		bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
3581 3582
		if (bit >= end)
			break;
3583
		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
3584
		mb_debug(1, "    free preallocated %u/%u in group %u\n",
3585 3586
			 (unsigned) ext4_group_first_block_no(sb, group) + bit,
			 (unsigned) next - bit, (unsigned) group);
3587 3588
		free += next - bit;

3589
		trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
3590 3591
		trace_ext4_mb_release_inode_pa(pa, (grp_blk_start +
						    EXT4_C2B(sbi, bit)),
L
Lukas Czerner 已提交
3592
					       next - bit);
3593 3594 3595 3596
		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
		bit = next + 1;
	}
	if (free != pa->pa_free) {
3597 3598 3599 3600 3601
		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);
3602
		ext4_grp_locked_error(sb, group, 0, 0, "free %u, pa_free %u",
3603
					free, pa->pa_free);
3604 3605 3606 3607
		/*
		 * pa is already deleted so we use the value obtained
		 * from the bitmap and continue.
		 */
3608 3609 3610 3611 3612 3613
	}
	atomic_add(free, &sbi->s_mb_discarded);

	return err;
}

3614 3615
static noinline_for_stack int
ext4_mb_release_group_pa(struct ext4_buddy *e4b,
3616
				struct ext4_prealloc_space *pa)
3617 3618 3619 3620 3621
{
	struct super_block *sb = e4b->bd_sb;
	ext4_group_t group;
	ext4_grpblk_t bit;

3622
	trace_ext4_mb_release_group_pa(sb, pa);
3623 3624 3625 3626 3627
	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);
3628
	trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641

	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
 */
3642 3643
static noinline_for_stack int
ext4_mb_discard_group_preallocations(struct super_block *sb,
3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654
					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;

3655
	mb_debug(1, "discard preallocation for group %u\n", group);
3656 3657 3658 3659

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

3660
	bitmap_bh = ext4_read_block_bitmap(sb, group);
3661
	if (bitmap_bh == NULL) {
3662
		ext4_error(sb, "Error reading block bitmap for %u", group);
3663
		return 0;
3664 3665 3666
	}

	err = ext4_mb_load_buddy(sb, group, &e4b);
3667
	if (err) {
3668
		ext4_error(sb, "Error loading buddy information for %u", group);
3669 3670 3671
		put_bh(bitmap_bh);
		return 0;
	}
3672 3673

	if (needed == 0)
3674
		needed = EXT4_CLUSTERS_PER_GROUP(sb) + 1;
3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 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

	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);

3730
		if (pa->pa_type == MB_GROUP_PA)
3731
			ext4_mb_release_group_pa(&e4b, pa);
3732
		else
3733
			ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3734 3735 3736 3737 3738 3739 3740

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

out:
	ext4_unlock_group(sb, group);
3741
	ext4_mb_unload_buddy(&e4b);
3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754
	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
 */
3755
void ext4_discard_preallocations(struct inode *inode)
3756 3757 3758 3759 3760 3761 3762 3763 3764 3765
{
	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;

3766
	if (!S_ISREG(inode->i_mode)) {
3767 3768 3769 3770
		/*BUG_ON(!list_empty(&ei->i_prealloc_list));*/
		return;
	}

3771
	mb_debug(1, "discard preallocation for inode %lu\n", inode->i_ino);
3772
	trace_ext4_discard_preallocations(inode);
3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788

	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);
3789 3790
			ext4_msg(sb, KERN_ERR,
				 "uh-oh! used pa while discarding");
3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825
			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) {
3826
		BUG_ON(pa->pa_type != MB_INODE_PA);
3827 3828 3829
		ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, NULL);

		err = ext4_mb_load_buddy(sb, group, &e4b);
3830
		if (err) {
3831 3832
			ext4_error(sb, "Error loading buddy information for %u",
					group);
3833 3834
			continue;
		}
3835

3836
		bitmap_bh = ext4_read_block_bitmap(sb, group);
3837
		if (bitmap_bh == NULL) {
3838 3839
			ext4_error(sb, "Error reading block bitmap for %u",
					group);
3840
			ext4_mb_unload_buddy(&e4b);
3841
			continue;
3842 3843 3844 3845
		}

		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
3846
		ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
3847 3848
		ext4_unlock_group(sb, group);

3849
		ext4_mb_unload_buddy(&e4b);
3850 3851 3852 3853 3854 3855 3856
		put_bh(bitmap_bh);

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

3857
#ifdef CONFIG_EXT4_DEBUG
3858 3859 3860
static void ext4_mb_show_ac(struct ext4_allocation_context *ac)
{
	struct super_block *sb = ac->ac_sb;
3861
	ext4_group_t ngroups, i;
3862

3863 3864
	if (!mb_enable_debug ||
	    (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED))
3865 3866
		return;

3867
	ext4_msg(ac->ac_sb, KERN_ERR, "Can't allocate:"
3868
			" Allocation context details:");
3869
	ext4_msg(ac->ac_sb, KERN_ERR, "status %d flags %d",
3870
			ac->ac_status, ac->ac_flags);
3871
	ext4_msg(ac->ac_sb, KERN_ERR, "orig %lu/%lu/%lu@%lu, "
3872 3873
		 	"goal %lu/%lu/%lu@%lu, "
			"best %lu/%lu/%lu@%lu cr %d",
3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886
			(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);
3887
	ext4_msg(ac->ac_sb, KERN_ERR, "%lu scanned, %d found",
3888
		 ac->ac_ex_scanned, ac->ac_found);
3889
	ext4_msg(ac->ac_sb, KERN_ERR, "groups: ");
3890 3891
	ngroups = ext4_get_groups_count(sb);
	for (i = 0; i < ngroups; i++) {
3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903
		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);
3904 3905
			printk(KERN_ERR "PA:%u:%d:%u \n", i,
			       start, pa->pa_len);
3906
		}
3907
		ext4_unlock_group(sb, i);
3908 3909 3910

		if (grp->bb_free == 0)
			continue;
3911
		printk(KERN_ERR "%u: %d/%d \n",
3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927
		       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
3928
 * One can tune this size via /sys/fs/ext4/<partition>/mb_stream_req
3929 3930 3931 3932 3933 3934 3935 3936 3937 3938
 */
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;

3939 3940 3941
	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
		return;

3942
	size = ac->ac_o_ex.fe_logical + EXT4_C2B(sbi, ac->ac_o_ex.fe_len);
3943 3944
	isize = (i_size_read(ac->ac_inode) + ac->ac_sb->s_blocksize - 1)
		>> bsbits;
3945

3946 3947 3948 3949 3950 3951 3952
	if ((size == isize) &&
	    !ext4_fs_is_busy(sbi) &&
	    (atomic_read(&ac->ac_inode->i_writecount) == 0)) {
		ac->ac_flags |= EXT4_MB_HINT_NOPREALLOC;
		return;
	}

3953 3954 3955 3956 3957
	if (sbi->s_mb_group_prealloc <= 0) {
		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
		return;
	}

3958
	/* don't use group allocation for large files */
3959
	size = max(size, isize);
3960
	if (size > sbi->s_mb_stream_request) {
3961
		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
3962
		return;
3963
	}
3964 3965 3966 3967 3968 3969 3970

	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.
	 */
3971
	ac->ac_lg = __this_cpu_ptr(sbi->s_locality_groups);
3972 3973 3974 3975 3976 3977 3978 3979

	/* 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);
}

3980 3981
static noinline_for_stack int
ext4_mb_initialize_context(struct ext4_allocation_context *ac,
3982 3983 3984 3985 3986 3987
				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;
3988 3989
	unsigned int len;
	ext4_fsblk_t goal;
3990 3991 3992 3993 3994 3995
	ext4_grpblk_t block;

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

	/* just a dirty hack to filter too big requests  */
3996 3997
	if (len >= EXT4_CLUSTERS_PER_GROUP(sb) - 10)
		len = EXT4_CLUSTERS_PER_GROUP(sb) - 10;
3998 3999 4000 4001 4002 4003 4004 4005 4006

	/* 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 */
4007
	memset(ac, 0, sizeof(struct ext4_allocation_context));
4008
	ac->ac_b_ex.fe_logical = ar->logical & ~(sbi->s_cluster_ratio - 1);
4009 4010 4011
	ac->ac_status = AC_STATUS_CONTINUE;
	ac->ac_sb = sb;
	ac->ac_inode = ar->inode;
4012
	ac->ac_o_ex.fe_logical = ac->ac_b_ex.fe_logical;
4013 4014 4015
	ac->ac_o_ex.fe_group = group;
	ac->ac_o_ex.fe_start = block;
	ac->ac_o_ex.fe_len = len;
4016
	ac->ac_g_ex = ac->ac_o_ex;
4017 4018 4019 4020 4021 4022
	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);

4023
	mb_debug(1, "init ac: %u blocks @ %u, goal %u, flags %x, 2^%d, "
4024 4025 4026 4027 4028 4029 4030 4031 4032 4033
			"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;

}

4034 4035 4036 4037 4038 4039 4040 4041 4042 4043
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;

4044
	mb_debug(1, "discard locality group preallocation\n");
4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065

	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 */
4066
		BUG_ON(pa->pa_type != MB_GROUP_PA);
4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091

		/* 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)) {
4092 4093
			ext4_error(sb, "Error loading buddy information for %u",
					group);
4094 4095 4096 4097
			continue;
		}
		ext4_lock_group(sb, group);
		list_del(&pa->pa_group_list);
4098
		ext4_mb_release_group_pa(&e4b, pa);
4099 4100
		ext4_unlock_group(sb, group);

4101
		ext4_mb_unload_buddy(&e4b);
4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132
		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) {
4133
			spin_unlock(&tmp_pa->pa_lock);
4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162
			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 ;
}

4163 4164 4165 4166 4167
/*
 * release all resource we used in allocation
 */
static int ext4_mb_release_context(struct ext4_allocation_context *ac)
{
4168
	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
4169 4170
	struct ext4_prealloc_space *pa = ac->ac_pa;
	if (pa) {
4171
		if (pa->pa_type == MB_GROUP_PA) {
4172
			/* see comment in ext4_mb_use_group_pa() */
4173
			spin_lock(&pa->pa_lock);
4174 4175
			pa->pa_pstart += EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
			pa->pa_lstart += EXT4_C2B(sbi, ac->ac_b_ex.fe_len);
4176 4177 4178
			pa->pa_free -= ac->ac_b_ex.fe_len;
			pa->pa_len -= ac->ac_b_ex.fe_len;
			spin_unlock(&pa->pa_lock);
4179 4180
		}
	}
A
Aneesh Kumar K.V 已提交
4181 4182 4183 4184 4185
	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
4186
		 * doesn't grow big.
A
Aneesh Kumar K.V 已提交
4187
		 */
4188
		if ((pa->pa_type == MB_GROUP_PA) && likely(pa->pa_free)) {
A
Aneesh Kumar K.V 已提交
4189 4190 4191 4192 4193 4194 4195
			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);
	}
4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207
	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)
{
4208
	ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4209 4210 4211
	int ret;
	int freed = 0;

4212
	trace_ext4_mb_discard_preallocations(sb, needed);
4213
	for (i = 0; i < ngroups && needed > 0; i++) {
4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227
		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,
4228
				struct ext4_allocation_request *ar, int *errp)
4229
{
4230
	int freed;
4231
	struct ext4_allocation_context *ac = NULL;
4232 4233 4234
	struct ext4_sb_info *sbi;
	struct super_block *sb;
	ext4_fsblk_t block = 0;
4235
	unsigned int inquota = 0;
4236
	unsigned int reserv_clstrs = 0;
4237 4238 4239 4240

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

4241
	trace_ext4_request_blocks(ar);
4242

4243 4244 4245 4246
	/* Allow to use superuser reservation for quota file */
	if (IS_NOQUOTA(ar->inode))
		ar->flags |= EXT4_MB_USE_ROOT_BLOCKS;

4247 4248 4249 4250 4251
	/*
	 * 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.
	 */
4252
	if (ext4_test_inode_state(ar->inode, EXT4_STATE_DELALLOC_RESERVED))
4253 4254 4255 4256 4257
		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.
4258
		 */
4259
		while (ar->len &&
4260
			ext4_claim_free_clusters(sbi, ar->len, ar->flags)) {
4261

A
Aneesh Kumar K.V 已提交
4262 4263 4264 4265 4266
			/* let others to free the space */
			yield();
			ar->len = ar->len >> 1;
		}
		if (!ar->len) {
4267 4268 4269
			*errp = -ENOSPC;
			return 0;
		}
4270
		reserv_clstrs = ar->len;
4271
		if (ar->flags & EXT4_MB_USE_ROOT_BLOCKS) {
4272 4273
			dquot_alloc_block_nofail(ar->inode,
						 EXT4_C2B(sbi, ar->len));
4274 4275
		} else {
			while (ar->len &&
4276 4277
				dquot_alloc_block(ar->inode,
						  EXT4_C2B(sbi, ar->len))) {
4278 4279 4280 4281

				ar->flags |= EXT4_MB_HINT_NOPREALLOC;
				ar->len--;
			}
4282 4283 4284 4285
		}
		inquota = ar->len;
		if (ar->len == 0) {
			*errp = -EDQUOT;
4286
			goto out;
4287
		}
4288
	}
4289

4290
	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4291
	if (!ac) {
4292
		ar->len = 0;
4293
		*errp = -ENOMEM;
4294
		goto out;
4295 4296 4297
	}

	*errp = ext4_mb_initialize_context(ac, ar);
4298 4299
	if (*errp) {
		ar->len = 0;
4300
		goto out;
4301 4302
	}

4303 4304 4305 4306
	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);
4307 4308
repeat:
		/* allocate space in core */
4309 4310 4311
		*errp = ext4_mb_regular_allocator(ac);
		if (*errp)
			goto errout;
4312 4313 4314 4315

		/* as we've just preallocated more space than
		 * user requested orinally, we store allocated
		 * space in a special descriptor */
4316 4317 4318
		if (ac->ac_status == AC_STATUS_FOUND &&
				ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
			ext4_mb_new_preallocation(ac);
4319
	}
4320
	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
4321
		*errp = ext4_mb_mark_diskspace_used(ac, handle, reserv_clstrs);
4322
		if (*errp == -EAGAIN) {
4323 4324 4325 4326 4327
			/*
			 * drop the reference that we took
			 * in ext4_mb_use_best_found
			 */
			ext4_mb_release_context(ac);
4328 4329 4330 4331 4332
			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;
4333 4334
		} else if (*errp)
		errout:
4335
			ext4_discard_allocated_blocks(ac);
4336
		else {
4337 4338 4339
			block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
			ar->len = ac->ac_b_ex.fe_len;
		}
4340
	} else {
4341
		freed  = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
4342 4343 4344
		if (freed)
			goto repeat;
		*errp = -ENOSPC;
4345 4346 4347
	}

	if (*errp) {
4348
		ac->ac_b_ex.fe_len = 0;
4349
		ar->len = 0;
4350
		ext4_mb_show_ac(ac);
4351
	}
4352
	ext4_mb_release_context(ac);
4353 4354 4355
out:
	if (ac)
		kmem_cache_free(ext4_ac_cachep, ac);
4356
	if (inquota && ar->len < inquota)
4357
		dquot_free_block(ar->inode, EXT4_C2B(sbi, inquota - ar->len));
4358
	if (!ar->len) {
4359 4360
		if (!ext4_test_inode_state(ar->inode,
					   EXT4_STATE_DELALLOC_RESERVED))
4361
			/* release all the reserved blocks if non delalloc */
4362
			percpu_counter_sub(&sbi->s_dirtyclusters_counter,
4363
						reserv_clstrs);
4364
	}
4365

4366
	trace_ext4_allocate_blocks(ar, (unsigned long long)block);
4367

4368 4369 4370
	return block;
}

4371 4372 4373 4374 4375 4376 4377 4378
/*
 * 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)
{
B
Bobi Jam 已提交
4379 4380 4381
	if ((entry1->efd_tid == entry2->efd_tid) &&
	    (entry1->efd_group == entry2->efd_group) &&
	    ((entry1->efd_start_cluster + entry1->efd_count) == entry2->efd_start_cluster))
4382 4383 4384 4385
		return 1;
	return 0;
}

4386 4387
static noinline_for_stack int
ext4_mb_free_metadata(handle_t *handle, struct ext4_buddy *e4b,
4388
		      struct ext4_free_data *new_entry)
4389
{
4390
	ext4_group_t group = e4b->bd_group;
4391
	ext4_grpblk_t cluster;
4392
	struct ext4_free_data *entry;
4393 4394 4395
	struct ext4_group_info *db = e4b->bd_info;
	struct super_block *sb = e4b->bd_sb;
	struct ext4_sb_info *sbi = EXT4_SB(sb);
4396 4397 4398
	struct rb_node **n = &db->bb_free_root.rb_node, *node;
	struct rb_node *parent = NULL, *new_node;

4399
	BUG_ON(!ext4_handle_valid(handle));
4400 4401 4402
	BUG_ON(e4b->bd_bitmap_page == NULL);
	BUG_ON(e4b->bd_buddy_page == NULL);

B
Bobi Jam 已提交
4403 4404
	new_node = &new_entry->efd_node;
	cluster = new_entry->efd_start_cluster;
4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416

	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;
B
Bobi Jam 已提交
4417 4418
		entry = rb_entry(parent, struct ext4_free_data, efd_node);
		if (cluster < entry->efd_start_cluster)
4419
			n = &(*n)->rb_left;
B
Bobi Jam 已提交
4420
		else if (cluster >= (entry->efd_start_cluster + entry->efd_count))
4421 4422
			n = &(*n)->rb_right;
		else {
4423
			ext4_grp_locked_error(sb, group, 0,
4424 4425
				ext4_group_first_block_no(sb, group) +
				EXT4_C2B(sbi, cluster),
4426
				"Block already on to-be-freed list");
4427
			return 0;
4428
		}
4429
	}
4430

4431 4432 4433 4434 4435 4436
	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) {
B
Bobi Jam 已提交
4437
		entry = rb_entry(node, struct ext4_free_data, efd_node);
4438
		if (can_merge(entry, new_entry)) {
B
Bobi Jam 已提交
4439 4440
			new_entry->efd_start_cluster = entry->efd_start_cluster;
			new_entry->efd_count += entry->efd_count;
4441
			rb_erase(node, &(db->bb_free_root));
B
Bobi Jam 已提交
4442 4443
			ext4_journal_callback_del(handle, &entry->efd_jce);
			kmem_cache_free(ext4_free_data_cachep, entry);
4444
		}
4445
	}
4446

4447 4448
	node = rb_next(new_node);
	if (node) {
B
Bobi Jam 已提交
4449
		entry = rb_entry(node, struct ext4_free_data, efd_node);
4450
		if (can_merge(new_entry, entry)) {
B
Bobi Jam 已提交
4451
			new_entry->efd_count += entry->efd_count;
4452
			rb_erase(node, &(db->bb_free_root));
B
Bobi Jam 已提交
4453 4454
			ext4_journal_callback_del(handle, &entry->efd_jce);
			kmem_cache_free(ext4_free_data_cachep, entry);
4455 4456
		}
	}
4457
	/* Add the extent to transaction's private list */
B
Bobi Jam 已提交
4458 4459
	ext4_journal_callback_add(handle, ext4_free_data_callback,
				  &new_entry->efd_jce);
4460 4461 4462
	return 0;
}

4463 4464 4465 4466 4467 4468
/**
 * 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
4469
 * @flags:		flags used by ext4_free_blocks
4470
 */
4471
void ext4_free_blocks(handle_t *handle, struct inode *inode,
4472 4473
		      struct buffer_head *bh, ext4_fsblk_t block,
		      unsigned long count, int flags)
4474
{
4475
	struct buffer_head *bitmap_bh = NULL;
4476 4477
	struct super_block *sb = inode->i_sb;
	struct ext4_group_desc *gdp;
4478
	unsigned long freed = 0;
4479
	unsigned int overflow;
4480 4481 4482 4483 4484
	ext4_grpblk_t bit;
	struct buffer_head *gd_bh;
	ext4_group_t block_group;
	struct ext4_sb_info *sbi;
	struct ext4_buddy e4b;
4485
	unsigned int count_clusters;
4486 4487 4488
	int err = 0;
	int ret;

4489 4490 4491 4492 4493 4494
	if (bh) {
		if (block)
			BUG_ON(block != bh->b_blocknr);
		else
			block = bh->b_blocknr;
	}
4495 4496

	sbi = EXT4_SB(sb);
4497 4498
	if (!(flags & EXT4_FREE_BLOCKS_VALIDATED) &&
	    !ext4_data_block_valid(sbi, block, count)) {
4499
		ext4_error(sb, "Freeing blocks not in datazone - "
4500
			   "block = %llu, count = %lu", block, count);
4501 4502 4503
		goto error_return;
	}

4504
	ext4_debug("freeing block %llu\n", block);
4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516
	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);
4517 4518
			if (unlikely(!tbh))
				continue;
4519
			ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
4520 4521 4522 4523
				    inode, tbh, block + i);
		}
	}

4524
	/*
4525 4526 4527 4528 4529 4530 4531 4532
	 * 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;
4533

4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565
	/*
	 * If the extent to be freed does not begin on a cluster
	 * boundary, we need to deal with partial clusters at the
	 * beginning and end of the extent.  Normally we will free
	 * blocks at the beginning or the end unless we are explicitly
	 * requested to avoid doing so.
	 */
	overflow = block & (sbi->s_cluster_ratio - 1);
	if (overflow) {
		if (flags & EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER) {
			overflow = sbi->s_cluster_ratio - overflow;
			block += overflow;
			if (count > overflow)
				count -= overflow;
			else
				return;
		} else {
			block -= overflow;
			count += overflow;
		}
	}
	overflow = count & (sbi->s_cluster_ratio - 1);
	if (overflow) {
		if (flags & EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER) {
			if (count > overflow)
				count -= overflow;
			else
				return;
		} else
			count += sbi->s_cluster_ratio - overflow;
	}

4566 4567 4568 4569 4570 4571 4572 4573
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.
	 */
4574 4575 4576
	if (EXT4_C2B(sbi, bit) + count > EXT4_BLOCKS_PER_GROUP(sb)) {
		overflow = EXT4_C2B(sbi, bit) + count -
			EXT4_BLOCKS_PER_GROUP(sb);
4577 4578
		count -= overflow;
	}
4579
	count_clusters = EXT4_B2C(sbi, count);
4580
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4581 4582
	if (!bitmap_bh) {
		err = -EIO;
4583
		goto error_return;
4584
	}
4585
	gdp = ext4_get_group_desc(sb, block_group, &gd_bh);
4586 4587
	if (!gdp) {
		err = -EIO;
4588
		goto error_return;
4589
	}
4590 4591 4592 4593

	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),
4594
		     EXT4_SB(sb)->s_itb_per_group) ||
4595
	    in_range(block + count - 1, ext4_inode_table(sb, gdp),
4596
		     EXT4_SB(sb)->s_itb_per_group)) {
4597

4598
		ext4_error(sb, "Freeing blocks in system zone - "
4599
			   "Block = %llu, count = %lu", block, count);
4600 4601
		/* err = 0. ext4_std_error should be a no op */
		goto error_return;
4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620
	}

	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;
4621
		for (i = 0; i < count_clusters; i++)
4622 4623 4624
			BUG_ON(!mb_test_bit(bit + i, bitmap_bh->b_data));
	}
#endif
4625
	trace_ext4_mballoc_free(sb, inode, block_group, bit, count_clusters);
4626

4627 4628 4629
	err = ext4_mb_load_buddy(sb, block_group, &e4b);
	if (err)
		goto error_return;
4630 4631

	if ((flags & EXT4_FREE_BLOCKS_METADATA) && ext4_handle_valid(handle)) {
4632 4633 4634 4635 4636
		struct ext4_free_data *new_entry;
		/*
		 * blocks being freed are metadata. these blocks shouldn't
		 * be used until this transaction is committed
		 */
B
Bobi Jam 已提交
4637
		new_entry = kmem_cache_alloc(ext4_free_data_cachep, GFP_NOFS);
4638 4639 4640 4641
		if (!new_entry) {
			err = -ENOMEM;
			goto error_return;
		}
B
Bobi Jam 已提交
4642 4643 4644 4645
		new_entry->efd_start_cluster = bit;
		new_entry->efd_group = block_group;
		new_entry->efd_count = count_clusters;
		new_entry->efd_tid = handle->h_transaction->t_tid;
4646

4647
		ext4_lock_group(sb, block_group);
4648
		mb_clear_bits(bitmap_bh->b_data, bit, count_clusters);
4649
		ext4_mb_free_metadata(handle, &e4b, new_entry);
4650
	} else {
4651 4652 4653 4654
		/* need to update group_info->bb_free and bitmap
		 * with group lock held. generate_buddy look at
		 * them with group lock_held
		 */
4655
		ext4_lock_group(sb, block_group);
4656 4657
		mb_clear_bits(bitmap_bh->b_data, bit, count_clusters);
		mb_free_blocks(inode, &e4b, bit, count_clusters);
4658 4659
	}

4660 4661
	ret = ext4_free_group_clusters(sb, gdp) + count_clusters;
	ext4_free_group_clusters_set(sb, gdp, ret);
4662
	gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
4663
	ext4_unlock_group(sb, block_group);
4664
	percpu_counter_add(&sbi->s_freeclusters_counter, count_clusters);
4665

4666 4667
	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4668 4669
		atomic_add(count_clusters,
			   &sbi->s_flex_groups[flex_group].free_clusters);
4670 4671
	}

4672
	ext4_mb_unload_buddy(&e4b);
4673

4674
	freed += count;
4675

4676 4677 4678
	if (!(flags & EXT4_FREE_BLOCKS_NO_QUOT_UPDATE))
		dquot_free_block(inode, EXT4_C2B(sbi, count_clusters));

4679 4680 4681 4682
	/* We dirtied the bitmap block */
	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);

4683 4684
	/* And the group descriptor block */
	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
4685
	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
4686 4687 4688 4689 4690 4691 4692 4693 4694
	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
4695
	ext4_mark_super_dirty(sb);
4696 4697 4698 4699 4700
error_return:
	brelse(bitmap_bh);
	ext4_std_error(sb, err);
	return;
}
4701

4702
/**
4703
 * ext4_group_add_blocks() -- Add given blocks to an existing group
4704 4705 4706 4707 4708
 * @handle:			handle to this transaction
 * @sb:				super block
 * @block:			start physcial block to add to the block group
 * @count:			number of blocks to free
 *
4709
 * This marks the blocks as free in the bitmap and buddy.
4710
 */
4711
int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
4712 4713 4714 4715 4716 4717 4718 4719 4720
			 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);
4721
	struct ext4_buddy e4b;
4722 4723 4724 4725 4726
	int err = 0, ret, blk_free_count;
	ext4_grpblk_t blocks_freed;

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

4727 4728 4729
	if (count == 0)
		return 0;

4730 4731 4732 4733 4734
	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
	/*
	 * Check to see if we are freeing blocks across a group
	 * boundary.
	 */
4735 4736 4737 4738
	if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
		ext4_warning(sb, "too much blocks added to group %u\n",
			     block_group);
		err = -EINVAL;
4739
		goto error_return;
4740
	}
4741

4742
	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4743 4744
	if (!bitmap_bh) {
		err = -EIO;
4745
		goto error_return;
4746 4747
	}

4748
	desc = ext4_get_group_desc(sb, block_group, &gd_bh);
4749 4750
	if (!desc) {
		err = -EIO;
4751
		goto error_return;
4752
	}
4753 4754 4755 4756 4757 4758 4759 4760 4761

	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);
4762
		err = -EINVAL;
4763 4764 4765
		goto error_return;
	}

4766 4767
	BUFFER_TRACE(bitmap_bh, "getting write access");
	err = ext4_journal_get_write_access(handle, bitmap_bh);
4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779
	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;
4780

4781 4782
	for (i = 0, blocks_freed = 0; i < count; i++) {
		BUFFER_TRACE(bitmap_bh, "clear bit");
4783
		if (!mb_test_bit(bit + i, bitmap_bh->b_data)) {
4784 4785 4786 4787 4788 4789 4790
			ext4_error(sb, "bit already cleared for block %llu",
				   (ext4_fsblk_t)(block + i));
			BUFFER_TRACE(bitmap_bh, "bit already cleared");
		} else {
			blocks_freed++;
		}
	}
4791 4792 4793 4794 4795 4796 4797 4798 4799 4800

	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
	 */
4801
	ext4_lock_group(sb, block_group);
4802 4803
	mb_clear_bits(bitmap_bh->b_data, bit, count);
	mb_free_blocks(NULL, &e4b, bit, count);
4804 4805
	blk_free_count = blocks_freed + ext4_free_group_clusters(sb, desc);
	ext4_free_group_clusters_set(sb, desc, blk_free_count);
4806 4807
	desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
	ext4_unlock_group(sb, block_group);
4808 4809
	percpu_counter_add(&sbi->s_freeclusters_counter,
			   EXT4_B2C(sbi, blocks_freed));
4810 4811 4812

	if (sbi->s_log_groups_per_flex) {
		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4813 4814
		atomic_add(EXT4_B2C(sbi, blocks_freed),
			   &sbi->s_flex_groups[flex_group].free_clusters);
4815
	}
4816 4817

	ext4_mb_unload_buddy(&e4b);
4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831

	/* 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);
4832
	return err;
4833 4834
}

4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846
/**
 * 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.
 */
4847 4848
static void ext4_trim_extent(struct super_block *sb, int start, int count,
			     ext4_group_t group, struct ext4_buddy *e4b)
4849 4850 4851
{
	struct ext4_free_extent ex;

T
Tao Ma 已提交
4852 4853
	trace_ext4_trim_extent(sb, group, start, count);

4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865
	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);
4866
	ext4_issue_discard(sb, group, start, count);
4867 4868 4869 4870 4871 4872 4873
	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
4874
 * @group:		group to be trimmed
4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888
 * @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.
 */
4889
static ext4_grpblk_t
4890 4891 4892
ext4_trim_all_free(struct super_block *sb, ext4_group_t group,
		   ext4_grpblk_t start, ext4_grpblk_t max,
		   ext4_grpblk_t minblocks)
4893 4894
{
	void *bitmap;
4895
	ext4_grpblk_t next, count = 0, free_count = 0;
4896 4897
	struct ext4_buddy e4b;
	int ret;
4898

T
Tao Ma 已提交
4899 4900
	trace_ext4_trim_all_free(sb, group, start, max);

4901 4902 4903 4904 4905 4906 4907
	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;
4908 4909

	ext4_lock_group(sb, group);
4910 4911 4912 4913
	if (EXT4_MB_GRP_WAS_TRIMMED(e4b.bd_info) &&
	    minblocks >= atomic_read(&EXT4_SB(sb)->s_last_trim_minblks))
		goto out;

4914 4915
	start = (e4b.bd_info->bb_first_free > start) ?
		e4b.bd_info->bb_first_free : start;
4916

4917 4918 4919
	while (start <= max) {
		start = mb_find_next_zero_bit(bitmap, max + 1, start);
		if (start > max)
4920
			break;
4921
		next = mb_find_next_bit(bitmap, max + 1, start);
4922 4923

		if ((next - start) >= minblocks) {
4924
			ext4_trim_extent(sb, start,
4925
					 next - start, group, &e4b);
4926 4927
			count += next - start;
		}
4928
		free_count += next - start;
4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941
		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);
		}

4942
		if ((e4b.bd_info->bb_free - free_count) < minblocks)
4943 4944
			break;
	}
4945 4946 4947 4948

	if (!ret)
		EXT4_MB_GRP_SET_TRIMMED(e4b.bd_info);
out:
4949
	ext4_unlock_group(sb, group);
4950
	ext4_mb_unload_buddy(&e4b);
4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971

	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)
{
4972
	struct ext4_group_info *grp;
4973
	ext4_group_t group, first_group, last_group;
4974
	ext4_grpblk_t cnt = 0, first_cluster, last_cluster;
4975
	uint64_t start, end, minlen, trimmed = 0;
4976 4977
	ext4_fsblk_t first_data_blk =
			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
4978
	ext4_fsblk_t max_blks = ext4_blocks_count(EXT4_SB(sb)->s_es);
4979 4980 4981
	int ret = 0;

	start = range->start >> sb->s_blocksize_bits;
4982
	end = start + (range->len >> sb->s_blocksize_bits) - 1;
4983 4984
	minlen = range->minlen >> sb->s_blocksize_bits;

4985 4986
	if (unlikely(minlen > EXT4_CLUSTERS_PER_GROUP(sb)) ||
	    unlikely(start >= max_blks))
4987
		return -EINVAL;
4988 4989 4990
	if (end >= max_blks)
		end = max_blks - 1;
	if (end <= first_data_blk)
4991
		goto out;
4992
	if (start < first_data_blk)
4993
		start = first_data_blk;
4994

4995
	/* Determine first and last group to examine based on start and end */
4996
	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) start,
4997
				     &first_group, &first_cluster);
4998
	ext4_get_group_no_and_offset(sb, (ext4_fsblk_t) end,
4999
				     &last_group, &last_cluster);
5000

5001 5002
	/* end now represents the last cluster to discard in this group */
	end = EXT4_CLUSTERS_PER_GROUP(sb) - 1;
5003 5004

	for (group = first_group; group <= last_group; group++) {
5005 5006 5007 5008 5009 5010
		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;
5011 5012
		}

5013
		/*
5014 5015 5016 5017
		 * For all the groups except the last one, last cluster will
		 * always be EXT4_CLUSTERS_PER_GROUP(sb)-1, so we only need to
		 * change it for the last group, note that last_cluster is
		 * already computed earlier by ext4_get_group_no_and_offset()
5018
		 */
5019 5020
		if (group == last_group)
			end = last_cluster;
5021

5022
		if (grp->bb_free >= minlen) {
5023
			cnt = ext4_trim_all_free(sb, group, first_cluster,
5024
						end, minlen);
5025 5026 5027 5028 5029 5030
			if (cnt < 0) {
				ret = cnt;
				break;
			}
		}
		trimmed += cnt;
5031 5032 5033 5034 5035

		/*
		 * For every group except the first one, we are sure
		 * that the first cluster to discard will be cluster #0.
		 */
5036
		first_cluster = 0;
5037 5038 5039
	}
	range->len = trimmed * sb->s_blocksize;

5040 5041 5042
	if (!ret)
		atomic_set(&EXT4_SB(sb)->s_last_trim_minblks, minlen);

5043
out:
5044 5045
	return ret;
}