Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/ext4/balloc.c
4 : *
5 : * Copyright (C) 1992, 1993, 1994, 1995
6 : * Remy Card (card@masi.ibp.fr)
7 : * Laboratoire MASI - Institut Blaise Pascal
8 : * Universite Pierre et Marie Curie (Paris VI)
9 : *
10 : * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11 : * Big-endian to little-endian byte-swapping/bitmaps by
12 : * David S. Miller (davem@caip.rutgers.edu), 1995
13 : */
14 :
15 : #include <linux/time.h>
16 : #include <linux/capability.h>
17 : #include <linux/fs.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 2 : ext4_group_t ext4_get_group_number(struct super_block *sb,
36 : ext4_fsblk_t block)
37 : {
38 2 : ext4_group_t group;
39 :
40 2 : if (test_opt2(sb, STD_GROUP_SIZE))
41 2 : group = (block -
42 2 : le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
43 2 : (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
44 : else
45 2 : ext4_get_group_no_and_offset(sb, block, &group, NULL);
46 2 : 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 : */
53 515 : 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 : {
56 515 : struct ext4_super_block *es = EXT4_SB(sb)->s_es;
57 515 : ext4_grpblk_t offset;
58 :
59 515 : blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
60 515 : offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
61 515 : EXT4_SB(sb)->s_cluster_bits;
62 515 : if (offsetp)
63 515 : *offsetp = offset;
64 515 : if (blockgrpp)
65 515 : *blockgrpp = blocknr;
66 :
67 515 : }
68 :
69 : /*
70 : * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
71 : * and 0 otherwise.
72 : */
73 0 : static inline int ext4_block_in_group(struct super_block *sb,
74 : ext4_fsblk_t block,
75 : ext4_group_t block_group)
76 : {
77 0 : ext4_group_t actual_group;
78 :
79 0 : actual_group = ext4_get_group_number(sb, block);
80 0 : 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 0 : static unsigned ext4_num_overhead_clusters(struct super_block *sb,
87 : ext4_group_t block_group,
88 : struct ext4_group_desc *gdp)
89 : {
90 0 : unsigned num_clusters;
91 0 : int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
92 0 : ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
93 0 : ext4_fsblk_t itbl_blk;
94 0 : 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 0 : 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 0 : if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
114 0 : block_cluster = EXT4_B2C(sbi,
115 : ext4_block_bitmap(sb, gdp) - start);
116 0 : if (block_cluster < num_clusters)
117 : block_cluster = -1;
118 0 : else if (block_cluster == num_clusters) {
119 0 : num_clusters++;
120 0 : block_cluster = -1;
121 : }
122 : }
123 :
124 0 : if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
125 0 : inode_cluster = EXT4_B2C(sbi,
126 : ext4_inode_bitmap(sb, gdp) - start);
127 0 : if (inode_cluster < num_clusters)
128 : inode_cluster = -1;
129 0 : else if (inode_cluster == num_clusters) {
130 0 : num_clusters++;
131 0 : inode_cluster = -1;
132 : }
133 : }
134 :
135 0 : itbl_blk = ext4_inode_table(sb, gdp);
136 0 : for (i = 0; i < sbi->s_itb_per_group; i++) {
137 0 : if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
138 0 : c = EXT4_B2C(sbi, itbl_blk + i - start);
139 0 : if ((c < num_clusters) || (c == inode_cluster) ||
140 0 : (c == block_cluster) || (c == itbl_cluster))
141 0 : continue;
142 0 : if (c == num_clusters) {
143 0 : num_clusters++;
144 0 : continue;
145 : }
146 0 : num_clusters++;
147 0 : itbl_cluster = c;
148 : }
149 : }
150 :
151 0 : if (block_cluster != -1)
152 0 : num_clusters++;
153 0 : if (inode_cluster != -1)
154 0 : num_clusters++;
155 :
156 0 : return num_clusters;
157 : }
158 :
159 0 : static unsigned int num_clusters_in_group(struct super_block *sb,
160 : ext4_group_t block_group)
161 : {
162 0 : unsigned int blocks;
163 :
164 0 : 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 0 : blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
172 0 : ext4_group_first_block_no(sb, block_group);
173 : } else
174 0 : blocks = EXT4_BLOCKS_PER_GROUP(sb);
175 0 : return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
176 : }
177 :
178 : /* Initializes an uninitialized block bitmap */
179 0 : static int ext4_init_block_bitmap(struct super_block *sb,
180 : struct buffer_head *bh,
181 : ext4_group_t block_group,
182 : struct ext4_group_desc *gdp)
183 : {
184 0 : unsigned int bit, bit_max;
185 0 : struct ext4_sb_info *sbi = EXT4_SB(sb);
186 0 : ext4_fsblk_t start, tmp;
187 :
188 0 : ASSERT(buffer_locked(bh));
189 :
190 : /* If checksum is bad mark all blocks used to prevent allocation
191 : * essentially implementing a per-group read-only flag. */
192 0 : if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
193 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
194 : EXT4_GROUP_INFO_BBITMAP_CORRUPT |
195 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
196 0 : return -EFSBADCRC;
197 : }
198 0 : memset(bh->b_data, 0, sb->s_blocksize);
199 :
200 0 : bit_max = ext4_num_base_meta_clusters(sb, block_group);
201 0 : if ((bit_max >> 3) >= bh->b_size)
202 : return -EFSCORRUPTED;
203 :
204 0 : for (bit = 0; bit < bit_max; bit++)
205 0 : ext4_set_bit(bit, bh->b_data);
206 :
207 0 : start = ext4_group_first_block_no(sb, block_group);
208 :
209 : /* Set bits for block and inode bitmaps, and inode table */
210 0 : tmp = ext4_block_bitmap(sb, gdp);
211 0 : if (ext4_block_in_group(sb, tmp, block_group))
212 0 : ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
213 :
214 0 : tmp = ext4_inode_bitmap(sb, gdp);
215 0 : if (ext4_block_in_group(sb, tmp, block_group))
216 0 : ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
217 :
218 0 : tmp = ext4_inode_table(sb, gdp);
219 0 : for (; tmp < ext4_inode_table(sb, gdp) +
220 0 : sbi->s_itb_per_group; tmp++) {
221 0 : if (ext4_block_in_group(sb, tmp, block_group))
222 0 : ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
223 : }
224 :
225 : /*
226 : * Also if the number of blocks within the group is less than
227 : * the blocksize * 8 ( which is the size of bitmap ), set rest
228 : * of the block bitmap to 1
229 : */
230 0 : ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
231 0 : sb->s_blocksize * 8, bh->b_data);
232 0 : return 0;
233 : }
234 :
235 : /* Return the number of free blocks in a block group. It is used when
236 : * the block bitmap is uninitialized, so we can't just count the bits
237 : * in the bitmap. */
238 0 : unsigned ext4_free_clusters_after_init(struct super_block *sb,
239 : ext4_group_t block_group,
240 : struct ext4_group_desc *gdp)
241 : {
242 0 : return num_clusters_in_group(sb, block_group) -
243 0 : ext4_num_overhead_clusters(sb, block_group, gdp);
244 : }
245 :
246 : /*
247 : * The free blocks are managed by bitmaps. A file system contains several
248 : * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
249 : * block for inodes, N blocks for the inode table and data blocks.
250 : *
251 : * The file system contains group descriptors which are located after the
252 : * super block. Each descriptor contains the number of the bitmap block and
253 : * the free blocks count in the block. The descriptors are loaded in memory
254 : * when a file system is mounted (see ext4_fill_super).
255 : */
256 :
257 : /**
258 : * ext4_get_group_desc() -- load group descriptor from disk
259 : * @sb: super block
260 : * @block_group: given block group
261 : * @bh: pointer to the buffer head to store the block
262 : * group descriptor
263 : */
264 23140 : struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
265 : ext4_group_t block_group,
266 : struct buffer_head **bh)
267 : {
268 23140 : unsigned int group_desc;
269 23140 : unsigned int offset;
270 23140 : ext4_group_t ngroups = ext4_get_groups_count(sb);
271 23140 : struct ext4_group_desc *desc;
272 23140 : struct ext4_sb_info *sbi = EXT4_SB(sb);
273 23140 : struct buffer_head *bh_p;
274 :
275 23140 : if (block_group >= ngroups) {
276 0 : ext4_error(sb, "block_group >= groups_count - block_group = %u,"
277 : " groups_count = %u", block_group, ngroups);
278 :
279 0 : return NULL;
280 : }
281 :
282 23140 : group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
283 23140 : offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
284 46281 : bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
285 : /*
286 : * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
287 : * the pointer being dereferenced won't be dereferenced again. By
288 : * looking at the usage in add_new_gdb() the value isn't modified,
289 : * just the pointer, and so it remains valid.
290 : */
291 23140 : if (!bh_p) {
292 0 : ext4_error(sb, "Group descriptor not loaded - "
293 : "block_group = %u, group_desc = %u, desc = %u",
294 : block_group, group_desc, offset);
295 0 : return NULL;
296 : }
297 :
298 23140 : desc = (struct ext4_group_desc *)(
299 23140 : (__u8 *)bh_p->b_data +
300 23140 : offset * EXT4_DESC_SIZE(sb));
301 23140 : if (bh)
302 1261 : *bh = bh_p;
303 : return desc;
304 : }
305 :
306 : /*
307 : * Return the block number which was discovered to be invalid, or 0 if
308 : * the block bitmap is valid.
309 : */
310 16 : static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
311 : struct ext4_group_desc *desc,
312 : ext4_group_t block_group,
313 : struct buffer_head *bh)
314 : {
315 16 : struct ext4_sb_info *sbi = EXT4_SB(sb);
316 16 : ext4_grpblk_t offset;
317 16 : ext4_grpblk_t next_zero_bit;
318 16 : ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
319 16 : ext4_fsblk_t blk;
320 16 : ext4_fsblk_t group_first_block;
321 :
322 16 : if (ext4_has_feature_flex_bg(sb)) {
323 : /* with FLEX_BG, the inode/block bitmaps and itable
324 : * blocks may not be in the group at all
325 : * so the bitmap validation will be skipped for those groups
326 : * or it has to also read the block group where the bitmaps
327 : * are located to verify they are set.
328 : */
329 : return 0;
330 : }
331 0 : group_first_block = ext4_group_first_block_no(sb, block_group);
332 :
333 : /* check whether block bitmap block number is set */
334 0 : blk = ext4_block_bitmap(sb, desc);
335 0 : offset = blk - group_first_block;
336 0 : if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
337 0 : !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
338 : /* bad block bitmap */
339 0 : return blk;
340 :
341 : /* check whether the inode bitmap block number is set */
342 0 : blk = ext4_inode_bitmap(sb, desc);
343 0 : offset = blk - group_first_block;
344 0 : if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
345 0 : !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
346 : /* bad block bitmap */
347 0 : return blk;
348 :
349 : /* check whether the inode table block number is set */
350 0 : blk = ext4_inode_table(sb, desc);
351 0 : offset = blk - group_first_block;
352 0 : if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
353 0 : EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
354 : return blk;
355 0 : next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
356 : EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
357 : EXT4_B2C(sbi, offset));
358 0 : if (next_zero_bit <
359 0 : EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
360 : /* bad bitmap for inode tables */
361 0 : return blk;
362 : return 0;
363 : }
364 :
365 500 : static int ext4_validate_block_bitmap(struct super_block *sb,
366 : struct ext4_group_desc *desc,
367 : ext4_group_t block_group,
368 : struct buffer_head *bh)
369 : {
370 500 : ext4_fsblk_t blk;
371 500 : struct ext4_group_info *grp;
372 :
373 500 : if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
374 : return 0;
375 :
376 500 : grp = ext4_get_group_info(sb, block_group);
377 :
378 500 : if (buffer_verified(bh))
379 : return 0;
380 16 : if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
381 : return -EFSCORRUPTED;
382 :
383 16 : ext4_lock_group(sb, block_group);
384 16 : if (buffer_verified(bh))
385 0 : goto verified;
386 16 : if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
387 : desc, bh) ||
388 : ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
389 0 : ext4_unlock_group(sb, block_group);
390 0 : ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
391 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
392 : EXT4_GROUP_INFO_BBITMAP_CORRUPT);
393 0 : return -EFSBADCRC;
394 : }
395 16 : blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
396 16 : if (unlikely(blk != 0)) {
397 0 : ext4_unlock_group(sb, block_group);
398 0 : ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
399 : block_group, blk);
400 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
401 : EXT4_GROUP_INFO_BBITMAP_CORRUPT);
402 0 : return -EFSCORRUPTED;
403 : }
404 16 : set_buffer_verified(bh);
405 16 : verified:
406 16 : ext4_unlock_group(sb, block_group);
407 16 : return 0;
408 : }
409 :
410 : /**
411 : * ext4_read_block_bitmap_nowait()
412 : * @sb: super block
413 : * @block_group: given block group
414 : *
415 : * Read the bitmap for a given block_group,and validate the
416 : * bits for block/inode/inode tables are set in the bitmaps
417 : *
418 : * Return buffer_head on success or an ERR_PTR in case of failure.
419 : */
420 : struct buffer_head *
421 500 : ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
422 : bool ignore_locked)
423 : {
424 500 : struct ext4_group_desc *desc;
425 500 : struct ext4_sb_info *sbi = EXT4_SB(sb);
426 500 : struct buffer_head *bh;
427 500 : ext4_fsblk_t bitmap_blk;
428 500 : int err;
429 :
430 500 : desc = ext4_get_group_desc(sb, block_group, NULL);
431 500 : if (!desc)
432 500 : return ERR_PTR(-EFSCORRUPTED);
433 500 : bitmap_blk = ext4_block_bitmap(sb, desc);
434 500 : if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
435 500 : (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
436 0 : ext4_error(sb, "Invalid block bitmap block %llu in "
437 : "block_group %u", bitmap_blk, block_group);
438 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
439 : EXT4_GROUP_INFO_BBITMAP_CORRUPT);
440 0 : return ERR_PTR(-EFSCORRUPTED);
441 : }
442 500 : bh = sb_getblk(sb, bitmap_blk);
443 500 : if (unlikely(!bh)) {
444 0 : ext4_warning(sb, "Cannot get buffer for block bitmap - "
445 : "block_group = %u, block_bitmap = %llu",
446 : block_group, bitmap_blk);
447 0 : return ERR_PTR(-ENOMEM);
448 : }
449 :
450 515 : if (ignore_locked && buffer_locked(bh)) {
451 : /* buffer under IO already, return if called for prefetching */
452 0 : put_bh(bh);
453 0 : return NULL;
454 : }
455 :
456 500 : if (bitmap_uptodate(bh))
457 484 : goto verify;
458 :
459 16 : lock_buffer(bh);
460 16 : if (bitmap_uptodate(bh)) {
461 0 : unlock_buffer(bh);
462 0 : goto verify;
463 : }
464 16 : ext4_lock_group(sb, block_group);
465 16 : if (ext4_has_group_desc_csum(sb) &&
466 16 : (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
467 0 : if (block_group == 0) {
468 0 : ext4_unlock_group(sb, block_group);
469 0 : unlock_buffer(bh);
470 0 : ext4_error(sb, "Block bitmap for bg 0 marked "
471 : "uninitialized");
472 0 : err = -EFSCORRUPTED;
473 0 : goto out;
474 : }
475 0 : err = ext4_init_block_bitmap(sb, bh, block_group, desc);
476 0 : set_bitmap_uptodate(bh);
477 0 : set_buffer_uptodate(bh);
478 0 : set_buffer_verified(bh);
479 0 : ext4_unlock_group(sb, block_group);
480 0 : unlock_buffer(bh);
481 0 : if (err) {
482 0 : ext4_error(sb, "Failed to init block bitmap for group "
483 : "%u: %d", block_group, err);
484 0 : goto out;
485 : }
486 0 : goto verify;
487 : }
488 16 : ext4_unlock_group(sb, block_group);
489 16 : if (buffer_uptodate(bh)) {
490 : /*
491 : * if not uninit if bh is uptodate,
492 : * bitmap is also uptodate
493 : */
494 0 : set_bitmap_uptodate(bh);
495 0 : unlock_buffer(bh);
496 0 : goto verify;
497 : }
498 : /*
499 : * submit the buffer_head for reading
500 : */
501 16 : set_buffer_new(bh);
502 16 : trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked);
503 16 : ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO |
504 : (ignore_locked ? REQ_RAHEAD : 0),
505 : ext4_end_bitmap_read);
506 16 : return bh;
507 484 : verify:
508 484 : err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
509 484 : if (err)
510 0 : goto out;
511 : return bh;
512 0 : out:
513 0 : put_bh(bh);
514 0 : return ERR_PTR(err);
515 : }
516 :
517 : /* Returns 0 on success, -errno on error */
518 485 : int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
519 : struct buffer_head *bh)
520 : {
521 485 : struct ext4_group_desc *desc;
522 :
523 485 : if (!buffer_new(bh))
524 : return 0;
525 16 : desc = ext4_get_group_desc(sb, block_group, NULL);
526 16 : if (!desc)
527 : return -EFSCORRUPTED;
528 16 : wait_on_buffer(bh);
529 16 : ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
530 16 : if (!buffer_uptodate(bh)) {
531 0 : ext4_error_err(sb, EIO, "Cannot read block bitmap - "
532 : "block_group = %u, block_bitmap = %llu",
533 : block_group, (unsigned long long) bh->b_blocknr);
534 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
535 : EXT4_GROUP_INFO_BBITMAP_CORRUPT);
536 0 : return -EIO;
537 : }
538 16 : clear_buffer_new(bh);
539 : /* Panic or remount fs read-only if block bitmap is invalid */
540 16 : return ext4_validate_block_bitmap(sb, desc, block_group, bh);
541 : }
542 :
543 : struct buffer_head *
544 453 : ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
545 : {
546 453 : struct buffer_head *bh;
547 453 : int err;
548 :
549 453 : bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
550 453 : if (IS_ERR(bh))
551 : return bh;
552 453 : err = ext4_wait_block_bitmap(sb, block_group, bh);
553 453 : if (err) {
554 0 : put_bh(bh);
555 0 : return ERR_PTR(err);
556 : }
557 : return bh;
558 : }
559 :
560 : /**
561 : * ext4_has_free_clusters()
562 : * @sbi: in-core super block structure.
563 : * @nclusters: number of needed blocks
564 : * @flags: flags from ext4_mb_new_blocks()
565 : *
566 : * Check if filesystem has nclusters free & available for allocation.
567 : * On success return 1, return 0 on failure.
568 : */
569 1865 : static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
570 : s64 nclusters, unsigned int flags)
571 : {
572 1865 : s64 free_clusters, dirty_clusters, rsv, resv_clusters;
573 1865 : struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
574 1865 : struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
575 :
576 1865 : free_clusters = percpu_counter_read_positive(fcc);
577 1865 : dirty_clusters = percpu_counter_read_positive(dcc);
578 1865 : resv_clusters = atomic64_read(&sbi->s_resv_clusters);
579 :
580 : /*
581 : * r_blocks_count should always be multiple of the cluster ratio so
582 : * we are safe to do a plane bit shift only.
583 : */
584 1865 : rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
585 : resv_clusters;
586 :
587 1865 : if (free_clusters - (nclusters + rsv + dirty_clusters) <
588 1865 : EXT4_FREECLUSTERS_WATERMARK) {
589 0 : free_clusters = percpu_counter_sum_positive(fcc);
590 0 : dirty_clusters = percpu_counter_sum_positive(dcc);
591 : }
592 : /* Check whether we have space after accounting for current
593 : * dirty clusters & root reserved clusters.
594 : */
595 1865 : if (free_clusters >= (rsv + nclusters + dirty_clusters))
596 : return 1;
597 :
598 : /* Hm, nope. Are (enough) root reserved clusters available? */
599 0 : if (uid_eq(sbi->s_resuid, current_fsuid()) ||
600 0 : (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
601 0 : capable(CAP_SYS_RESOURCE) ||
602 0 : (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
603 :
604 0 : if (free_clusters >= (nclusters + dirty_clusters +
605 : resv_clusters))
606 : return 1;
607 : }
608 : /* No free blocks. Let's see if we can dip into reserved pool */
609 0 : if (flags & EXT4_MB_USE_RESERVED) {
610 0 : if (free_clusters >= (nclusters + dirty_clusters))
611 0 : return 1;
612 : }
613 :
614 : return 0;
615 : }
616 :
617 1865 : int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
618 : s64 nclusters, unsigned int flags)
619 : {
620 1865 : if (ext4_has_free_clusters(sbi, nclusters, flags)) {
621 1865 : percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
622 1865 : return 0;
623 : } else
624 : return -ENOSPC;
625 : }
626 :
627 : /**
628 : * ext4_should_retry_alloc() - check if a block allocation should be retried
629 : * @sb: super block
630 : * @retries: number of attemps has been made
631 : *
632 : * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
633 : * it is profitable to retry the operation, this function will wait
634 : * for the current or committing transaction to complete, and then
635 : * return TRUE. We will only retry once.
636 : */
637 0 : int ext4_should_retry_alloc(struct super_block *sb, int *retries)
638 : {
639 0 : if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
640 0 : (*retries)++ > 1 ||
641 0 : !EXT4_SB(sb)->s_journal)
642 0 : return 0;
643 :
644 0 : smp_mb();
645 0 : if (EXT4_SB(sb)->s_mb_free_pending == 0)
646 : return 0;
647 :
648 0 : jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
649 0 : jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
650 0 : return 1;
651 : }
652 :
653 : /*
654 : * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
655 : *
656 : * @handle: handle to this transaction
657 : * @inode: file inode
658 : * @goal: given target block(filesystem wide)
659 : * @count: pointer to total number of clusters needed
660 : * @errp: error code
661 : *
662 : * Return 1st allocated block number on success, *count stores total account
663 : * error stores in errp pointer
664 : */
665 2 : ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
666 : ext4_fsblk_t goal, unsigned int flags,
667 : unsigned long *count, int *errp)
668 : {
669 2 : struct ext4_allocation_request ar;
670 2 : ext4_fsblk_t ret;
671 :
672 2 : memset(&ar, 0, sizeof(ar));
673 : /* Fill with neighbour allocated blocks */
674 2 : ar.inode = inode;
675 2 : ar.goal = goal;
676 2 : ar.len = count ? *count : 1;
677 2 : ar.flags = flags;
678 :
679 2 : ret = ext4_mb_new_blocks(handle, &ar, errp);
680 2 : if (count)
681 0 : *count = ar.len;
682 : /*
683 : * Account for the allocated meta blocks. We will never
684 : * fail EDQUOT for metdata, but we do account for it.
685 : */
686 2 : if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
687 4 : dquot_alloc_block_nofail(inode,
688 2 : EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
689 : }
690 2 : return ret;
691 : }
692 :
693 : /**
694 : * ext4_count_free_clusters() -- count filesystem free clusters
695 : * @sb: superblock
696 : *
697 : * Adds up the number of free clusters from each block group.
698 : */
699 1 : ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
700 : {
701 1 : ext4_fsblk_t desc_count;
702 1 : struct ext4_group_desc *gdp;
703 1 : ext4_group_t i;
704 1 : ext4_group_t ngroups = ext4_get_groups_count(sb);
705 1 : struct ext4_group_info *grp;
706 : #ifdef EXT4FS_DEBUG
707 : struct ext4_super_block *es;
708 : ext4_fsblk_t bitmap_count;
709 : unsigned int x;
710 : struct buffer_head *bitmap_bh = NULL;
711 :
712 : es = EXT4_SB(sb)->s_es;
713 : desc_count = 0;
714 : bitmap_count = 0;
715 : gdp = NULL;
716 :
717 : for (i = 0; i < ngroups; i++) {
718 : gdp = ext4_get_group_desc(sb, i, NULL);
719 : if (!gdp)
720 : continue;
721 : grp = NULL;
722 : if (EXT4_SB(sb)->s_group_info)
723 : grp = ext4_get_group_info(sb, i);
724 : if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
725 : desc_count += ext4_free_group_clusters(sb, gdp);
726 : brelse(bitmap_bh);
727 : bitmap_bh = ext4_read_block_bitmap(sb, i);
728 : if (IS_ERR(bitmap_bh)) {
729 : bitmap_bh = NULL;
730 : continue;
731 : }
732 :
733 : x = ext4_count_free(bitmap_bh->b_data,
734 : EXT4_CLUSTERS_PER_GROUP(sb) / 8);
735 : printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
736 : i, ext4_free_group_clusters(sb, gdp), x);
737 : bitmap_count += x;
738 : }
739 : brelse(bitmap_bh);
740 : printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
741 : ", computed = %llu, %llu\n",
742 : EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
743 : desc_count, bitmap_count);
744 : return bitmap_count;
745 : #else
746 1 : desc_count = 0;
747 18 : for (i = 0; i < ngroups; i++) {
748 16 : gdp = ext4_get_group_desc(sb, i, NULL);
749 16 : if (!gdp)
750 0 : continue;
751 16 : grp = NULL;
752 16 : if (EXT4_SB(sb)->s_group_info)
753 16 : grp = ext4_get_group_info(sb, i);
754 16 : if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
755 16 : desc_count += ext4_free_group_clusters(sb, gdp);
756 : }
757 :
758 1 : return desc_count;
759 : #endif
760 : }
761 :
762 : static inline int test_root(ext4_group_t a, int b)
763 : {
764 75 : while (1) {
765 33 : if (a < b)
766 : return 0;
767 62 : if (a == b)
768 : return 1;
769 46 : if ((a % b) != 0)
770 : return 0;
771 10 : a = a / b;
772 : }
773 : }
774 :
775 : /**
776 : * ext4_bg_has_super - number of blocks used by the superblock in group
777 : * @sb: superblock for filesystem
778 : * @group: group number to check
779 : *
780 : * Return the number of blocks used by the superblock (primary or backup)
781 : * in this group. Currently this will be only 0 or 1.
782 : */
783 55 : int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
784 : {
785 55 : struct ext4_super_block *es = EXT4_SB(sb)->s_es;
786 :
787 55 : if (group == 0)
788 : return 1;
789 50 : if (ext4_has_feature_sparse_super2(sb)) {
790 0 : if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
791 0 : group == le32_to_cpu(es->s_backup_bgs[1]))
792 : return 1;
793 0 : return 0;
794 : }
795 50 : if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
796 : return 1;
797 46 : if (!(group & 1))
798 : return 0;
799 74 : if (test_root(group, 3) || (test_root(group, 5)) ||
800 13 : test_root(group, 7))
801 16 : return 1;
802 :
803 : return 0;
804 : }
805 :
806 0 : static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
807 : ext4_group_t group)
808 : {
809 0 : unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
810 0 : ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
811 0 : ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
812 :
813 0 : if (group == first || group == first + 1 || group == last)
814 0 : return 1;
815 : return 0;
816 : }
817 :
818 23 : static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
819 : ext4_group_t group)
820 : {
821 23 : if (!ext4_bg_has_super(sb, group))
822 : return 0;
823 :
824 13 : if (ext4_has_feature_meta_bg(sb))
825 0 : return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
826 : else
827 13 : return EXT4_SB(sb)->s_gdb_count;
828 : }
829 :
830 : /**
831 : * ext4_bg_num_gdb - number of blocks used by the group table in group
832 : * @sb: superblock for filesystem
833 : * @group: group number to check
834 : *
835 : * Return the number of blocks used by the group descriptor table
836 : * (primary or backup) in this group. In the future there may be a
837 : * different number of descriptor blocks in each group.
838 : */
839 23 : unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
840 : {
841 23 : unsigned long first_meta_bg =
842 23 : le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
843 23 : unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
844 :
845 23 : if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
846 23 : return ext4_bg_num_gdb_nometa(sb, group);
847 :
848 0 : return ext4_bg_num_gdb_meta(sb,group);
849 :
850 : }
851 :
852 : /*
853 : * This function returns the number of file system metadata clusters at
854 : * the beginning of a block group, including the reserved gdt blocks.
855 : */
856 0 : static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
857 : ext4_group_t block_group)
858 : {
859 0 : struct ext4_sb_info *sbi = EXT4_SB(sb);
860 0 : unsigned num;
861 :
862 : /* Check for superblock and gdt backups in this group */
863 0 : num = ext4_bg_has_super(sb, block_group);
864 :
865 0 : if (!ext4_has_feature_meta_bg(sb) ||
866 0 : block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
867 0 : sbi->s_desc_per_block) {
868 0 : if (num) {
869 0 : num += ext4_bg_num_gdb(sb, block_group);
870 0 : num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
871 : }
872 : } else { /* For META_BG_BLOCK_GROUPS */
873 0 : num += ext4_bg_num_gdb(sb, block_group);
874 : }
875 0 : return EXT4_NUM_B2C(sbi, num);
876 : }
877 : /**
878 : * ext4_inode_to_goal_block - return a hint for block allocation
879 : * @inode: inode for block allocation
880 : *
881 : * Return the ideal location to start allocating blocks for a
882 : * newly created inode.
883 : */
884 223 : ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
885 : {
886 223 : struct ext4_inode_info *ei = EXT4_I(inode);
887 223 : ext4_group_t block_group;
888 223 : ext4_grpblk_t colour;
889 223 : int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
890 223 : ext4_fsblk_t bg_start;
891 223 : ext4_fsblk_t last_block;
892 :
893 223 : block_group = ei->i_block_group;
894 223 : if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
895 : /*
896 : * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
897 : * block groups per flexgroup, reserve the first block
898 : * group for directories and special files. Regular
899 : * files will start at the second block group. This
900 : * tends to speed up directory access and improves
901 : * fsck times.
902 : */
903 223 : block_group &= ~(flex_size-1);
904 223 : if (S_ISREG(inode->i_mode))
905 50 : block_group++;
906 : }
907 223 : bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
908 223 : last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
909 :
910 : /*
911 : * If we are doing delayed allocation, we don't need take
912 : * colour into account.
913 : */
914 223 : if (test_opt(inode->i_sb, DELALLOC))
915 : return bg_start;
916 :
917 0 : if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
918 0 : colour = (task_pid_nr(current) % 16) *
919 0 : (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
920 : else
921 0 : colour = (task_pid_nr(current) % 16) *
922 0 : ((last_block - bg_start) / 16);
923 0 : return bg_start + colour;
924 : }
925 :
|
