ext4: fix block bitmap validation when bigalloc, ^flex_bg
[deliverable/linux.git] / fs / ext4 / balloc.c
1 /*
2 * linux/fs/ext4/balloc.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10 * Big-endian to little-endian byte-swapping/bitmaps by
11 * David S. Miller (davem@caip.rutgers.edu), 1995
12 */
13
14 #include <linux/time.h>
15 #include <linux/capability.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/quotaops.h>
19 #include <linux/buffer_head.h>
20 #include "ext4.h"
21 #include "ext4_jbd2.h"
22 #include "mballoc.h"
23
24 #include <trace/events/ext4.h>
25
26 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
27 ext4_group_t block_group);
28 /*
29 * balloc.c contains the blocks allocation and deallocation routines
30 */
31
32 /*
33 * Calculate block group number for a given block number
34 */
35 ext4_group_t ext4_get_group_number(struct super_block *sb,
36 ext4_fsblk_t block)
37 {
38 ext4_group_t group;
39
40 if (test_opt2(sb, STD_GROUP_SIZE))
41 group = (block -
42 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
43 (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
44 else
45 ext4_get_group_no_and_offset(sb, block, &group, NULL);
46 return group;
47 }
48
49 /*
50 * Calculate the block group number and offset into the block/cluster
51 * allocation bitmap, given a block number
52 */
53 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
54 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
55 {
56 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
57 ext4_grpblk_t offset;
58
59 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
60 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
61 EXT4_SB(sb)->s_cluster_bits;
62 if (offsetp)
63 *offsetp = offset;
64 if (blockgrpp)
65 *blockgrpp = blocknr;
66
67 }
68
69 /*
70 * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
71 * and 0 otherwise.
72 */
73 static inline int ext4_block_in_group(struct super_block *sb,
74 ext4_fsblk_t block,
75 ext4_group_t block_group)
76 {
77 ext4_group_t actual_group;
78
79 actual_group = ext4_get_group_number(sb, block);
80 return (actual_group == block_group) ? 1 : 0;
81 }
82
83 /* Return the number of clusters used for file system metadata; this
84 * represents the overhead needed by the file system.
85 */
86 unsigned ext4_num_overhead_clusters(struct super_block *sb,
87 ext4_group_t block_group,
88 struct ext4_group_desc *gdp)
89 {
90 unsigned num_clusters;
91 int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
92 ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
93 ext4_fsblk_t itbl_blk;
94 struct ext4_sb_info *sbi = EXT4_SB(sb);
95
96 /* This is the number of clusters used by the superblock,
97 * block group descriptors, and reserved block group
98 * descriptor blocks */
99 num_clusters = ext4_num_base_meta_clusters(sb, block_group);
100
101 /*
102 * For the allocation bitmaps and inode table, we first need
103 * to check to see if the block is in the block group. If it
104 * is, then check to see if the cluster is already accounted
105 * for in the clusters used for the base metadata cluster, or
106 * if we can increment the base metadata cluster to include
107 * that block. Otherwise, we will have to track the cluster
108 * used for the allocation bitmap or inode table explicitly.
109 * Normally all of these blocks are contiguous, so the special
110 * case handling shouldn't be necessary except for *very*
111 * unusual file system layouts.
112 */
113 if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
114 block_cluster = EXT4_B2C(sbi,
115 ext4_block_bitmap(sb, gdp) - start);
116 if (block_cluster < num_clusters)
117 block_cluster = -1;
118 else if (block_cluster == num_clusters) {
119 num_clusters++;
120 block_cluster = -1;
121 }
122 }
123
124 if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
125 inode_cluster = EXT4_B2C(sbi,
126 ext4_inode_bitmap(sb, gdp) - start);
127 if (inode_cluster < num_clusters)
128 inode_cluster = -1;
129 else if (inode_cluster == num_clusters) {
130 num_clusters++;
131 inode_cluster = -1;
132 }
133 }
134
135 itbl_blk = ext4_inode_table(sb, gdp);
136 for (i = 0; i < sbi->s_itb_per_group; i++) {
137 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
138 c = EXT4_B2C(sbi, itbl_blk + i - start);
139 if ((c < num_clusters) || (c == inode_cluster) ||
140 (c == block_cluster) || (c == itbl_cluster))
141 continue;
142 if (c == num_clusters) {
143 num_clusters++;
144 continue;
145 }
146 num_clusters++;
147 itbl_cluster = c;
148 }
149 }
150
151 if (block_cluster != -1)
152 num_clusters++;
153 if (inode_cluster != -1)
154 num_clusters++;
155
156 return num_clusters;
157 }
158
159 static unsigned int num_clusters_in_group(struct super_block *sb,
160 ext4_group_t block_group)
161 {
162 unsigned int blocks;
163
164 if (block_group == ext4_get_groups_count(sb) - 1) {
165 /*
166 * Even though mke2fs always initializes the first and
167 * last group, just in case some other tool was used,
168 * we need to make sure we calculate the right free
169 * blocks.
170 */
171 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
172 ext4_group_first_block_no(sb, block_group);
173 } else
174 blocks = EXT4_BLOCKS_PER_GROUP(sb);
175 return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
176 }
177
178 /* Initializes an uninitialized block bitmap */
179 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
180 ext4_group_t block_group,
181 struct ext4_group_desc *gdp)
182 {
183 unsigned int bit, bit_max;
184 struct ext4_sb_info *sbi = EXT4_SB(sb);
185 ext4_fsblk_t start, tmp;
186 int flex_bg = 0;
187 struct ext4_group_info *grp;
188
189 J_ASSERT_BH(bh, buffer_locked(bh));
190
191 /* If checksum is bad mark all blocks used to prevent allocation
192 * essentially implementing a per-group read-only flag. */
193 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
194 ext4_error(sb, "Checksum bad for group %u", block_group);
195 grp = ext4_get_group_info(sb, block_group);
196 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
197 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
198 return;
199 }
200 memset(bh->b_data, 0, sb->s_blocksize);
201
202 bit_max = ext4_num_base_meta_clusters(sb, block_group);
203 for (bit = 0; bit < bit_max; bit++)
204 ext4_set_bit(bit, bh->b_data);
205
206 start = ext4_group_first_block_no(sb, block_group);
207
208 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
209 flex_bg = 1;
210
211 /* Set bits for block and inode bitmaps, and inode table */
212 tmp = ext4_block_bitmap(sb, gdp);
213 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
214 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
215
216 tmp = ext4_inode_bitmap(sb, gdp);
217 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
218 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
219
220 tmp = ext4_inode_table(sb, gdp);
221 for (; tmp < ext4_inode_table(sb, gdp) +
222 sbi->s_itb_per_group; tmp++) {
223 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
224 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
225 }
226
227 /*
228 * Also if the number of blocks within the group is less than
229 * the blocksize * 8 ( which is the size of bitmap ), set rest
230 * of the block bitmap to 1
231 */
232 ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
233 sb->s_blocksize * 8, bh->b_data);
234 ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
235 ext4_group_desc_csum_set(sb, block_group, gdp);
236 }
237
238 /* Return the number of free blocks in a block group. It is used when
239 * the block bitmap is uninitialized, so we can't just count the bits
240 * in the bitmap. */
241 unsigned ext4_free_clusters_after_init(struct super_block *sb,
242 ext4_group_t block_group,
243 struct ext4_group_desc *gdp)
244 {
245 return num_clusters_in_group(sb, block_group) -
246 ext4_num_overhead_clusters(sb, block_group, gdp);
247 }
248
249 /*
250 * The free blocks are managed by bitmaps. A file system contains several
251 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
252 * block for inodes, N blocks for the inode table and data blocks.
253 *
254 * The file system contains group descriptors which are located after the
255 * super block. Each descriptor contains the number of the bitmap block and
256 * the free blocks count in the block. The descriptors are loaded in memory
257 * when a file system is mounted (see ext4_fill_super).
258 */
259
260 /**
261 * ext4_get_group_desc() -- load group descriptor from disk
262 * @sb: super block
263 * @block_group: given block group
264 * @bh: pointer to the buffer head to store the block
265 * group descriptor
266 */
267 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
268 ext4_group_t block_group,
269 struct buffer_head **bh)
270 {
271 unsigned int group_desc;
272 unsigned int offset;
273 ext4_group_t ngroups = ext4_get_groups_count(sb);
274 struct ext4_group_desc *desc;
275 struct ext4_sb_info *sbi = EXT4_SB(sb);
276
277 if (block_group >= ngroups) {
278 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
279 " groups_count = %u", block_group, ngroups);
280
281 return NULL;
282 }
283
284 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
285 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
286 if (!sbi->s_group_desc[group_desc]) {
287 ext4_error(sb, "Group descriptor not loaded - "
288 "block_group = %u, group_desc = %u, desc = %u",
289 block_group, group_desc, offset);
290 return NULL;
291 }
292
293 desc = (struct ext4_group_desc *)(
294 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
295 offset * EXT4_DESC_SIZE(sb));
296 if (bh)
297 *bh = sbi->s_group_desc[group_desc];
298 return desc;
299 }
300
301 /*
302 * Return the block number which was discovered to be invalid, or 0 if
303 * the block bitmap is valid.
304 */
305 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
306 struct ext4_group_desc *desc,
307 ext4_group_t block_group,
308 struct buffer_head *bh)
309 {
310 struct ext4_sb_info *sbi = EXT4_SB(sb);
311 ext4_grpblk_t offset;
312 ext4_grpblk_t next_zero_bit;
313 ext4_fsblk_t blk;
314 ext4_fsblk_t group_first_block;
315
316 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
317 /* with FLEX_BG, the inode/block bitmaps and itable
318 * blocks may not be in the group at all
319 * so the bitmap validation will be skipped for those groups
320 * or it has to also read the block group where the bitmaps
321 * are located to verify they are set.
322 */
323 return 0;
324 }
325 group_first_block = ext4_group_first_block_no(sb, block_group);
326
327 /* check whether block bitmap block number is set */
328 blk = ext4_block_bitmap(sb, desc);
329 offset = blk - group_first_block;
330 if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
331 /* bad block bitmap */
332 return blk;
333
334 /* check whether the inode bitmap block number is set */
335 blk = ext4_inode_bitmap(sb, desc);
336 offset = blk - group_first_block;
337 if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
338 /* bad block bitmap */
339 return blk;
340
341 /* check whether the inode table block number is set */
342 blk = ext4_inode_table(sb, desc);
343 offset = blk - group_first_block;
344 next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
345 EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
346 EXT4_B2C(sbi, offset));
347 if (next_zero_bit <
348 EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
349 /* bad bitmap for inode tables */
350 return blk;
351 return 0;
352 }
353
354 void ext4_validate_block_bitmap(struct super_block *sb,
355 struct ext4_group_desc *desc,
356 ext4_group_t block_group,
357 struct buffer_head *bh)
358 {
359 ext4_fsblk_t blk;
360 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
361
362 if (buffer_verified(bh))
363 return;
364
365 ext4_lock_group(sb, block_group);
366 blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
367 if (unlikely(blk != 0)) {
368 ext4_unlock_group(sb, block_group);
369 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
370 block_group, blk);
371 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
372 return;
373 }
374 if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
375 desc, bh))) {
376 ext4_unlock_group(sb, block_group);
377 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
378 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
379 return;
380 }
381 set_buffer_verified(bh);
382 ext4_unlock_group(sb, block_group);
383 }
384
385 /**
386 * ext4_read_block_bitmap_nowait()
387 * @sb: super block
388 * @block_group: given block group
389 *
390 * Read the bitmap for a given block_group,and validate the
391 * bits for block/inode/inode tables are set in the bitmaps
392 *
393 * Return buffer_head on success or NULL in case of failure.
394 */
395 struct buffer_head *
396 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
397 {
398 struct ext4_group_desc *desc;
399 struct buffer_head *bh;
400 ext4_fsblk_t bitmap_blk;
401
402 desc = ext4_get_group_desc(sb, block_group, NULL);
403 if (!desc)
404 return NULL;
405 bitmap_blk = ext4_block_bitmap(sb, desc);
406 bh = sb_getblk(sb, bitmap_blk);
407 if (unlikely(!bh)) {
408 ext4_error(sb, "Cannot get buffer for block bitmap - "
409 "block_group = %u, block_bitmap = %llu",
410 block_group, bitmap_blk);
411 return NULL;
412 }
413
414 if (bitmap_uptodate(bh))
415 goto verify;
416
417 lock_buffer(bh);
418 if (bitmap_uptodate(bh)) {
419 unlock_buffer(bh);
420 goto verify;
421 }
422 ext4_lock_group(sb, block_group);
423 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
424 ext4_init_block_bitmap(sb, bh, block_group, desc);
425 set_bitmap_uptodate(bh);
426 set_buffer_uptodate(bh);
427 ext4_unlock_group(sb, block_group);
428 unlock_buffer(bh);
429 return bh;
430 }
431 ext4_unlock_group(sb, block_group);
432 if (buffer_uptodate(bh)) {
433 /*
434 * if not uninit if bh is uptodate,
435 * bitmap is also uptodate
436 */
437 set_bitmap_uptodate(bh);
438 unlock_buffer(bh);
439 goto verify;
440 }
441 /*
442 * submit the buffer_head for reading
443 */
444 set_buffer_new(bh);
445 trace_ext4_read_block_bitmap_load(sb, block_group);
446 bh->b_end_io = ext4_end_bitmap_read;
447 get_bh(bh);
448 submit_bh(READ | REQ_META | REQ_PRIO, bh);
449 return bh;
450 verify:
451 ext4_validate_block_bitmap(sb, desc, block_group, bh);
452 if (buffer_verified(bh))
453 return bh;
454 put_bh(bh);
455 return NULL;
456 }
457
458 /* Returns 0 on success, 1 on error */
459 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
460 struct buffer_head *bh)
461 {
462 struct ext4_group_desc *desc;
463
464 if (!buffer_new(bh))
465 return 0;
466 desc = ext4_get_group_desc(sb, block_group, NULL);
467 if (!desc)
468 return 1;
469 wait_on_buffer(bh);
470 if (!buffer_uptodate(bh)) {
471 ext4_error(sb, "Cannot read block bitmap - "
472 "block_group = %u, block_bitmap = %llu",
473 block_group, (unsigned long long) bh->b_blocknr);
474 return 1;
475 }
476 clear_buffer_new(bh);
477 /* Panic or remount fs read-only if block bitmap is invalid */
478 ext4_validate_block_bitmap(sb, desc, block_group, bh);
479 /* ...but check for error just in case errors=continue. */
480 return !buffer_verified(bh);
481 }
482
483 struct buffer_head *
484 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
485 {
486 struct buffer_head *bh;
487
488 bh = ext4_read_block_bitmap_nowait(sb, block_group);
489 if (!bh)
490 return NULL;
491 if (ext4_wait_block_bitmap(sb, block_group, bh)) {
492 put_bh(bh);
493 return NULL;
494 }
495 return bh;
496 }
497
498 /**
499 * ext4_has_free_clusters()
500 * @sbi: in-core super block structure.
501 * @nclusters: number of needed blocks
502 * @flags: flags from ext4_mb_new_blocks()
503 *
504 * Check if filesystem has nclusters free & available for allocation.
505 * On success return 1, return 0 on failure.
506 */
507 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
508 s64 nclusters, unsigned int flags)
509 {
510 s64 free_clusters, dirty_clusters, rsv, resv_clusters;
511 struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
512 struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
513
514 free_clusters = percpu_counter_read_positive(fcc);
515 dirty_clusters = percpu_counter_read_positive(dcc);
516 resv_clusters = atomic64_read(&sbi->s_resv_clusters);
517
518 /*
519 * r_blocks_count should always be multiple of the cluster ratio so
520 * we are safe to do a plane bit shift only.
521 */
522 rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
523 resv_clusters;
524
525 if (free_clusters - (nclusters + rsv + dirty_clusters) <
526 EXT4_FREECLUSTERS_WATERMARK) {
527 free_clusters = percpu_counter_sum_positive(fcc);
528 dirty_clusters = percpu_counter_sum_positive(dcc);
529 }
530 /* Check whether we have space after accounting for current
531 * dirty clusters & root reserved clusters.
532 */
533 if (free_clusters >= (rsv + nclusters + dirty_clusters))
534 return 1;
535
536 /* Hm, nope. Are (enough) root reserved clusters available? */
537 if (uid_eq(sbi->s_resuid, current_fsuid()) ||
538 (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
539 capable(CAP_SYS_RESOURCE) ||
540 (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
541
542 if (free_clusters >= (nclusters + dirty_clusters +
543 resv_clusters))
544 return 1;
545 }
546 /* No free blocks. Let's see if we can dip into reserved pool */
547 if (flags & EXT4_MB_USE_RESERVED) {
548 if (free_clusters >= (nclusters + dirty_clusters))
549 return 1;
550 }
551
552 return 0;
553 }
554
555 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
556 s64 nclusters, unsigned int flags)
557 {
558 if (ext4_has_free_clusters(sbi, nclusters, flags)) {
559 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
560 return 0;
561 } else
562 return -ENOSPC;
563 }
564
565 /**
566 * ext4_should_retry_alloc()
567 * @sb: super block
568 * @retries number of attemps has been made
569 *
570 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
571 * it is profitable to retry the operation, this function will wait
572 * for the current or committing transaction to complete, and then
573 * return TRUE.
574 *
575 * if the total number of retries exceed three times, return FALSE.
576 */
577 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
578 {
579 if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
580 (*retries)++ > 3 ||
581 !EXT4_SB(sb)->s_journal)
582 return 0;
583
584 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
585
586 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
587 }
588
589 /*
590 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
591 *
592 * @handle: handle to this transaction
593 * @inode: file inode
594 * @goal: given target block(filesystem wide)
595 * @count: pointer to total number of clusters needed
596 * @errp: error code
597 *
598 * Return 1st allocated block number on success, *count stores total account
599 * error stores in errp pointer
600 */
601 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
602 ext4_fsblk_t goal, unsigned int flags,
603 unsigned long *count, int *errp)
604 {
605 struct ext4_allocation_request ar;
606 ext4_fsblk_t ret;
607
608 memset(&ar, 0, sizeof(ar));
609 /* Fill with neighbour allocated blocks */
610 ar.inode = inode;
611 ar.goal = goal;
612 ar.len = count ? *count : 1;
613 ar.flags = flags;
614
615 ret = ext4_mb_new_blocks(handle, &ar, errp);
616 if (count)
617 *count = ar.len;
618 /*
619 * Account for the allocated meta blocks. We will never
620 * fail EDQUOT for metdata, but we do account for it.
621 */
622 if (!(*errp) &&
623 ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
624 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
625 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
626 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
627 dquot_alloc_block_nofail(inode,
628 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
629 }
630 return ret;
631 }
632
633 /**
634 * ext4_count_free_clusters() -- count filesystem free clusters
635 * @sb: superblock
636 *
637 * Adds up the number of free clusters from each block group.
638 */
639 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
640 {
641 ext4_fsblk_t desc_count;
642 struct ext4_group_desc *gdp;
643 ext4_group_t i;
644 ext4_group_t ngroups = ext4_get_groups_count(sb);
645 struct ext4_group_info *grp;
646 #ifdef EXT4FS_DEBUG
647 struct ext4_super_block *es;
648 ext4_fsblk_t bitmap_count;
649 unsigned int x;
650 struct buffer_head *bitmap_bh = NULL;
651
652 es = EXT4_SB(sb)->s_es;
653 desc_count = 0;
654 bitmap_count = 0;
655 gdp = NULL;
656
657 for (i = 0; i < ngroups; i++) {
658 gdp = ext4_get_group_desc(sb, i, NULL);
659 if (!gdp)
660 continue;
661 grp = NULL;
662 if (EXT4_SB(sb)->s_group_info)
663 grp = ext4_get_group_info(sb, i);
664 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
665 desc_count += ext4_free_group_clusters(sb, gdp);
666 brelse(bitmap_bh);
667 bitmap_bh = ext4_read_block_bitmap(sb, i);
668 if (bitmap_bh == NULL)
669 continue;
670
671 x = ext4_count_free(bitmap_bh->b_data,
672 EXT4_CLUSTERS_PER_GROUP(sb) / 8);
673 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
674 i, ext4_free_group_clusters(sb, gdp), x);
675 bitmap_count += x;
676 }
677 brelse(bitmap_bh);
678 printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
679 ", computed = %llu, %llu\n",
680 EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
681 desc_count, bitmap_count);
682 return bitmap_count;
683 #else
684 desc_count = 0;
685 for (i = 0; i < ngroups; i++) {
686 gdp = ext4_get_group_desc(sb, i, NULL);
687 if (!gdp)
688 continue;
689 grp = NULL;
690 if (EXT4_SB(sb)->s_group_info)
691 grp = ext4_get_group_info(sb, i);
692 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
693 desc_count += ext4_free_group_clusters(sb, gdp);
694 }
695
696 return desc_count;
697 #endif
698 }
699
700 static inline int test_root(ext4_group_t a, int b)
701 {
702 while (1) {
703 if (a < b)
704 return 0;
705 if (a == b)
706 return 1;
707 if ((a % b) != 0)
708 return 0;
709 a = a / b;
710 }
711 }
712
713 /**
714 * ext4_bg_has_super - number of blocks used by the superblock in group
715 * @sb: superblock for filesystem
716 * @group: group number to check
717 *
718 * Return the number of blocks used by the superblock (primary or backup)
719 * in this group. Currently this will be only 0 or 1.
720 */
721 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
722 {
723 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
724
725 if (group == 0)
726 return 1;
727 if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_SPARSE_SUPER2)) {
728 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
729 group == le32_to_cpu(es->s_backup_bgs[1]))
730 return 1;
731 return 0;
732 }
733 if ((group <= 1) || !EXT4_HAS_RO_COMPAT_FEATURE(sb,
734 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER))
735 return 1;
736 if (!(group & 1))
737 return 0;
738 if (test_root(group, 3) || (test_root(group, 5)) ||
739 test_root(group, 7))
740 return 1;
741
742 return 0;
743 }
744
745 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
746 ext4_group_t group)
747 {
748 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
749 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
750 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
751
752 if (group == first || group == first + 1 || group == last)
753 return 1;
754 return 0;
755 }
756
757 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
758 ext4_group_t group)
759 {
760 if (!ext4_bg_has_super(sb, group))
761 return 0;
762
763 if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
764 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
765 else
766 return EXT4_SB(sb)->s_gdb_count;
767 }
768
769 /**
770 * ext4_bg_num_gdb - number of blocks used by the group table in group
771 * @sb: superblock for filesystem
772 * @group: group number to check
773 *
774 * Return the number of blocks used by the group descriptor table
775 * (primary or backup) in this group. In the future there may be a
776 * different number of descriptor blocks in each group.
777 */
778 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
779 {
780 unsigned long first_meta_bg =
781 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
782 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
783
784 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
785 metagroup < first_meta_bg)
786 return ext4_bg_num_gdb_nometa(sb, group);
787
788 return ext4_bg_num_gdb_meta(sb,group);
789
790 }
791
792 /*
793 * This function returns the number of file system metadata clusters at
794 * the beginning of a block group, including the reserved gdt blocks.
795 */
796 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
797 ext4_group_t block_group)
798 {
799 struct ext4_sb_info *sbi = EXT4_SB(sb);
800 unsigned num;
801
802 /* Check for superblock and gdt backups in this group */
803 num = ext4_bg_has_super(sb, block_group);
804
805 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
806 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
807 sbi->s_desc_per_block) {
808 if (num) {
809 num += ext4_bg_num_gdb(sb, block_group);
810 num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
811 }
812 } else { /* For META_BG_BLOCK_GROUPS */
813 num += ext4_bg_num_gdb(sb, block_group);
814 }
815 return EXT4_NUM_B2C(sbi, num);
816 }
817 /**
818 * ext4_inode_to_goal_block - return a hint for block allocation
819 * @inode: inode for block allocation
820 *
821 * Return the ideal location to start allocating blocks for a
822 * newly created inode.
823 */
824 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
825 {
826 struct ext4_inode_info *ei = EXT4_I(inode);
827 ext4_group_t block_group;
828 ext4_grpblk_t colour;
829 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
830 ext4_fsblk_t bg_start;
831 ext4_fsblk_t last_block;
832
833 block_group = ei->i_block_group;
834 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
835 /*
836 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
837 * block groups per flexgroup, reserve the first block
838 * group for directories and special files. Regular
839 * files will start at the second block group. This
840 * tends to speed up directory access and improves
841 * fsck times.
842 */
843 block_group &= ~(flex_size-1);
844 if (S_ISREG(inode->i_mode))
845 block_group++;
846 }
847 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
848 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
849
850 /*
851 * If we are doing delayed allocation, we don't need take
852 * colour into account.
853 */
854 if (test_opt(inode->i_sb, DELALLOC))
855 return bg_start;
856
857 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
858 colour = (current->pid % 16) *
859 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
860 else
861 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
862 return bg_start + colour;
863 }
864
This page took 0.04832 seconds and 5 git commands to generate.