1 /*
2 * linux/fs/ext4/ialloc.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 * BSD ufs-inspired inode and directory allocation by
10 * 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/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/stat.h>
19 #include <linux/string.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
22 #include <linux/random.h>
23 #include <linux/bitops.h>
24 #include <linux/blkdev.h>
25 #include <asm/byteorder.h>
26
27 #include "ext4.h"
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
31
32 #include <trace/events/ext4.h>
33
34 /*
35 * ialloc.c contains the inodes allocation and deallocation routines
36 */
37
38 /*
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
42 *
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
46 */
47
48 /*
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
52 */
ext4_mark_bitmap_end(int start_bit,int end_bit,char * bitmap)53 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54 {
55 int i;
56
57 if (start_bit >= end_bit)
58 return;
59
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
63 if (i < end_bit)
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65 }
66
67 /* Initializes an uninitialized inode bitmap */
ext4_init_inode_bitmap(struct super_block * sb,struct buffer_head * bh,ext4_group_t block_group,struct ext4_group_desc * gdp)68 static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
72 {
73 struct ext4_group_info *grp;
74 struct ext4_sb_info *sbi = EXT4_SB(sb);
75 J_ASSERT_BH(bh, buffer_locked(bh));
76
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
80 ext4_error(sb, "Checksum bad for group %u", block_group);
81 grp = ext4_get_group_info(sb, block_group);
82 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
83 percpu_counter_sub(&sbi->s_freeclusters_counter,
84 grp->bb_free);
85 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
86 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
87 int count;
88 count = ext4_free_inodes_count(sb, gdp);
89 percpu_counter_sub(&sbi->s_freeinodes_counter,
90 count);
91 }
92 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
93 return 0;
94 }
95
96 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
97 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
98 bh->b_data);
99 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
100 EXT4_INODES_PER_GROUP(sb) / 8);
101 ext4_group_desc_csum_set(sb, block_group, gdp);
102
103 return EXT4_INODES_PER_GROUP(sb);
104 }
105
ext4_end_bitmap_read(struct buffer_head * bh,int uptodate)106 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
107 {
108 if (uptodate) {
109 set_buffer_uptodate(bh);
110 set_bitmap_uptodate(bh);
111 }
112 unlock_buffer(bh);
113 put_bh(bh);
114 }
115
116 /*
117 * Read the inode allocation bitmap for a given block_group, reading
118 * into the specified slot in the superblock's bitmap cache.
119 *
120 * Return buffer_head of bitmap on success or NULL.
121 */
122 static struct buffer_head *
ext4_read_inode_bitmap(struct super_block * sb,ext4_group_t block_group)123 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
124 {
125 struct ext4_group_desc *desc;
126 struct buffer_head *bh = NULL;
127 ext4_fsblk_t bitmap_blk;
128 struct ext4_group_info *grp;
129 struct ext4_sb_info *sbi = EXT4_SB(sb);
130
131 desc = ext4_get_group_desc(sb, block_group, NULL);
132 if (!desc)
133 return NULL;
134
135 bitmap_blk = ext4_inode_bitmap(sb, desc);
136 bh = sb_getblk(sb, bitmap_blk);
137 if (unlikely(!bh)) {
138 ext4_error(sb, "Cannot read inode bitmap - "
139 "block_group = %u, inode_bitmap = %llu",
140 block_group, bitmap_blk);
141 return NULL;
142 }
143 if (bitmap_uptodate(bh))
144 goto verify;
145
146 lock_buffer(bh);
147 if (bitmap_uptodate(bh)) {
148 unlock_buffer(bh);
149 goto verify;
150 }
151
152 ext4_lock_group(sb, block_group);
153 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
154 ext4_init_inode_bitmap(sb, bh, block_group, desc);
155 set_bitmap_uptodate(bh);
156 set_buffer_uptodate(bh);
157 set_buffer_verified(bh);
158 ext4_unlock_group(sb, block_group);
159 unlock_buffer(bh);
160 return bh;
161 }
162 ext4_unlock_group(sb, block_group);
163
164 if (buffer_uptodate(bh)) {
165 /*
166 * if not uninit if bh is uptodate,
167 * bitmap is also uptodate
168 */
169 set_bitmap_uptodate(bh);
170 unlock_buffer(bh);
171 goto verify;
172 }
173 /*
174 * submit the buffer_head for reading
175 */
176 trace_ext4_load_inode_bitmap(sb, block_group);
177 bh->b_end_io = ext4_end_bitmap_read;
178 get_bh(bh);
179 submit_bh(READ | REQ_META | REQ_PRIO, bh);
180 wait_on_buffer(bh);
181 if (!buffer_uptodate(bh)) {
182 put_bh(bh);
183 ext4_error(sb, "Cannot read inode bitmap - "
184 "block_group = %u, inode_bitmap = %llu",
185 block_group, bitmap_blk);
186 return NULL;
187 }
188
189 verify:
190 ext4_lock_group(sb, block_group);
191 if (!buffer_verified(bh) &&
192 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
193 EXT4_INODES_PER_GROUP(sb) / 8)) {
194 ext4_unlock_group(sb, block_group);
195 put_bh(bh);
196 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
197 "inode_bitmap = %llu", block_group, bitmap_blk);
198 grp = ext4_get_group_info(sb, block_group);
199 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
200 int count;
201 count = ext4_free_inodes_count(sb, desc);
202 percpu_counter_sub(&sbi->s_freeinodes_counter,
203 count);
204 }
205 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
206 return NULL;
207 }
208 ext4_unlock_group(sb, block_group);
209 set_buffer_verified(bh);
210 return bh;
211 }
212
213 /*
214 * NOTE! When we get the inode, we're the only people
215 * that have access to it, and as such there are no
216 * race conditions we have to worry about. The inode
217 * is not on the hash-lists, and it cannot be reached
218 * through the filesystem because the directory entry
219 * has been deleted earlier.
220 *
221 * HOWEVER: we must make sure that we get no aliases,
222 * which means that we have to call "clear_inode()"
223 * _before_ we mark the inode not in use in the inode
224 * bitmaps. Otherwise a newly created file might use
225 * the same inode number (not actually the same pointer
226 * though), and then we'd have two inodes sharing the
227 * same inode number and space on the harddisk.
228 */
ext4_free_inode(handle_t * handle,struct inode * inode)229 void ext4_free_inode(handle_t *handle, struct inode *inode)
230 {
231 struct super_block *sb = inode->i_sb;
232 int is_directory;
233 unsigned long ino;
234 struct buffer_head *bitmap_bh = NULL;
235 struct buffer_head *bh2;
236 ext4_group_t block_group;
237 unsigned long bit;
238 struct ext4_group_desc *gdp;
239 struct ext4_super_block *es;
240 struct ext4_sb_info *sbi;
241 int fatal = 0, err, count, cleared;
242 struct ext4_group_info *grp;
243
244 if (!sb) {
245 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
246 "nonexistent device\n", __func__, __LINE__);
247 return;
248 }
249 if (atomic_read(&inode->i_count) > 1) {
250 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
251 __func__, __LINE__, inode->i_ino,
252 atomic_read(&inode->i_count));
253 return;
254 }
255 if (inode->i_nlink) {
256 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
257 __func__, __LINE__, inode->i_ino, inode->i_nlink);
258 return;
259 }
260 sbi = EXT4_SB(sb);
261
262 ino = inode->i_ino;
263 ext4_debug("freeing inode %lu\n", ino);
264 trace_ext4_free_inode(inode);
265
266 /*
267 * Note: we must free any quota before locking the superblock,
268 * as writing the quota to disk may need the lock as well.
269 */
270 dquot_initialize(inode);
271 ext4_xattr_delete_inode(handle, inode);
272 dquot_free_inode(inode);
273 dquot_drop(inode);
274
275 is_directory = S_ISDIR(inode->i_mode);
276
277 /* Do this BEFORE marking the inode not in use or returning an error */
278 ext4_clear_inode(inode);
279
280 es = EXT4_SB(sb)->s_es;
281 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 goto error_return;
284 }
285 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 /* Don't bother if the inode bitmap is corrupt. */
289 grp = ext4_get_group_info(sb, block_group);
290 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
291 goto error_return;
292
293 BUFFER_TRACE(bitmap_bh, "get_write_access");
294 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
295 if (fatal)
296 goto error_return;
297
298 fatal = -ESRCH;
299 gdp = ext4_get_group_desc(sb, block_group, &bh2);
300 if (gdp) {
301 BUFFER_TRACE(bh2, "get_write_access");
302 fatal = ext4_journal_get_write_access(handle, bh2);
303 }
304 ext4_lock_group(sb, block_group);
305 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
306 if (fatal || !cleared) {
307 ext4_unlock_group(sb, block_group);
308 goto out;
309 }
310
311 count = ext4_free_inodes_count(sb, gdp) + 1;
312 ext4_free_inodes_set(sb, gdp, count);
313 if (is_directory) {
314 count = ext4_used_dirs_count(sb, gdp) - 1;
315 ext4_used_dirs_set(sb, gdp, count);
316 percpu_counter_dec(&sbi->s_dirs_counter);
317 }
318 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
319 EXT4_INODES_PER_GROUP(sb) / 8);
320 ext4_group_desc_csum_set(sb, block_group, gdp);
321 ext4_unlock_group(sb, block_group);
322
323 percpu_counter_inc(&sbi->s_freeinodes_counter);
324 if (sbi->s_log_groups_per_flex) {
325 ext4_group_t f = ext4_flex_group(sbi, block_group);
326
327 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
328 if (is_directory)
329 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
330 }
331 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
332 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
333 out:
334 if (cleared) {
335 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
336 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
337 if (!fatal)
338 fatal = err;
339 } else {
340 ext4_error(sb, "bit already cleared for inode %lu", ino);
341 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
342 int count;
343 count = ext4_free_inodes_count(sb, gdp);
344 percpu_counter_sub(&sbi->s_freeinodes_counter,
345 count);
346 }
347 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
348 }
349
350 error_return:
351 brelse(bitmap_bh);
352 ext4_std_error(sb, fatal);
353 }
354
355 struct orlov_stats {
356 __u64 free_clusters;
357 __u32 free_inodes;
358 __u32 used_dirs;
359 };
360
361 /*
362 * Helper function for Orlov's allocator; returns critical information
363 * for a particular block group or flex_bg. If flex_size is 1, then g
364 * is a block group number; otherwise it is flex_bg number.
365 */
get_orlov_stats(struct super_block * sb,ext4_group_t g,int flex_size,struct orlov_stats * stats)366 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
367 int flex_size, struct orlov_stats *stats)
368 {
369 struct ext4_group_desc *desc;
370 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
371
372 if (flex_size > 1) {
373 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
374 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
375 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
376 return;
377 }
378
379 desc = ext4_get_group_desc(sb, g, NULL);
380 if (desc) {
381 stats->free_inodes = ext4_free_inodes_count(sb, desc);
382 stats->free_clusters = ext4_free_group_clusters(sb, desc);
383 stats->used_dirs = ext4_used_dirs_count(sb, desc);
384 } else {
385 stats->free_inodes = 0;
386 stats->free_clusters = 0;
387 stats->used_dirs = 0;
388 }
389 }
390
391 /*
392 * Orlov's allocator for directories.
393 *
394 * We always try to spread first-level directories.
395 *
396 * If there are blockgroups with both free inodes and free blocks counts
397 * not worse than average we return one with smallest directory count.
398 * Otherwise we simply return a random group.
399 *
400 * For the rest rules look so:
401 *
402 * It's OK to put directory into a group unless
403 * it has too many directories already (max_dirs) or
404 * it has too few free inodes left (min_inodes) or
405 * it has too few free blocks left (min_blocks) or
406 * Parent's group is preferred, if it doesn't satisfy these
407 * conditions we search cyclically through the rest. If none
408 * of the groups look good we just look for a group with more
409 * free inodes than average (starting at parent's group).
410 */
411
find_group_orlov(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode,const struct qstr * qstr)412 static int find_group_orlov(struct super_block *sb, struct inode *parent,
413 ext4_group_t *group, umode_t mode,
414 const struct qstr *qstr)
415 {
416 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
417 struct ext4_sb_info *sbi = EXT4_SB(sb);
418 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
419 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
420 unsigned int freei, avefreei, grp_free;
421 ext4_fsblk_t freeb, avefreec;
422 unsigned int ndirs;
423 int max_dirs, min_inodes;
424 ext4_grpblk_t min_clusters;
425 ext4_group_t i, grp, g, ngroups;
426 struct ext4_group_desc *desc;
427 struct orlov_stats stats;
428 int flex_size = ext4_flex_bg_size(sbi);
429 struct dx_hash_info hinfo;
430
431 ngroups = real_ngroups;
432 if (flex_size > 1) {
433 ngroups = (real_ngroups + flex_size - 1) >>
434 sbi->s_log_groups_per_flex;
435 parent_group >>= sbi->s_log_groups_per_flex;
436 }
437
438 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
439 avefreei = freei / ngroups;
440 freeb = EXT4_C2B(sbi,
441 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
442 avefreec = freeb;
443 do_div(avefreec, ngroups);
444 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
445
446 if (S_ISDIR(mode) &&
447 ((parent == sb->s_root->d_inode) ||
448 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
449 int best_ndir = inodes_per_group;
450 int ret = -1;
451
452 if (qstr) {
453 hinfo.hash_version = DX_HASH_HALF_MD4;
454 hinfo.seed = sbi->s_hash_seed;
455 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
456 grp = hinfo.hash;
457 } else
458 grp = prandom_u32();
459 parent_group = (unsigned)grp % ngroups;
460 for (i = 0; i < ngroups; i++) {
461 g = (parent_group + i) % ngroups;
462 get_orlov_stats(sb, g, flex_size, &stats);
463 if (!stats.free_inodes)
464 continue;
465 if (stats.used_dirs >= best_ndir)
466 continue;
467 if (stats.free_inodes < avefreei)
468 continue;
469 if (stats.free_clusters < avefreec)
470 continue;
471 grp = g;
472 ret = 0;
473 best_ndir = stats.used_dirs;
474 }
475 if (ret)
476 goto fallback;
477 found_flex_bg:
478 if (flex_size == 1) {
479 *group = grp;
480 return 0;
481 }
482
483 /*
484 * We pack inodes at the beginning of the flexgroup's
485 * inode tables. Block allocation decisions will do
486 * something similar, although regular files will
487 * start at 2nd block group of the flexgroup. See
488 * ext4_ext_find_goal() and ext4_find_near().
489 */
490 grp *= flex_size;
491 for (i = 0; i < flex_size; i++) {
492 if (grp+i >= real_ngroups)
493 break;
494 desc = ext4_get_group_desc(sb, grp+i, NULL);
495 if (desc && ext4_free_inodes_count(sb, desc)) {
496 *group = grp+i;
497 return 0;
498 }
499 }
500 goto fallback;
501 }
502
503 max_dirs = ndirs / ngroups + inodes_per_group / 16;
504 min_inodes = avefreei - inodes_per_group*flex_size / 4;
505 if (min_inodes < 1)
506 min_inodes = 1;
507 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
508
509 /*
510 * Start looking in the flex group where we last allocated an
511 * inode for this parent directory
512 */
513 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
514 parent_group = EXT4_I(parent)->i_last_alloc_group;
515 if (flex_size > 1)
516 parent_group >>= sbi->s_log_groups_per_flex;
517 }
518
519 for (i = 0; i < ngroups; i++) {
520 grp = (parent_group + i) % ngroups;
521 get_orlov_stats(sb, grp, flex_size, &stats);
522 if (stats.used_dirs >= max_dirs)
523 continue;
524 if (stats.free_inodes < min_inodes)
525 continue;
526 if (stats.free_clusters < min_clusters)
527 continue;
528 goto found_flex_bg;
529 }
530
531 fallback:
532 ngroups = real_ngroups;
533 avefreei = freei / ngroups;
534 fallback_retry:
535 parent_group = EXT4_I(parent)->i_block_group;
536 for (i = 0; i < ngroups; i++) {
537 grp = (parent_group + i) % ngroups;
538 desc = ext4_get_group_desc(sb, grp, NULL);
539 if (desc) {
540 grp_free = ext4_free_inodes_count(sb, desc);
541 if (grp_free && grp_free >= avefreei) {
542 *group = grp;
543 return 0;
544 }
545 }
546 }
547
548 if (avefreei) {
549 /*
550 * The free-inodes counter is approximate, and for really small
551 * filesystems the above test can fail to find any blockgroups
552 */
553 avefreei = 0;
554 goto fallback_retry;
555 }
556
557 return -1;
558 }
559
find_group_other(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode)560 static int find_group_other(struct super_block *sb, struct inode *parent,
561 ext4_group_t *group, umode_t mode)
562 {
563 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
564 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
565 struct ext4_group_desc *desc;
566 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
567
568 /*
569 * Try to place the inode is the same flex group as its
570 * parent. If we can't find space, use the Orlov algorithm to
571 * find another flex group, and store that information in the
572 * parent directory's inode information so that use that flex
573 * group for future allocations.
574 */
575 if (flex_size > 1) {
576 int retry = 0;
577
578 try_again:
579 parent_group &= ~(flex_size-1);
580 last = parent_group + flex_size;
581 if (last > ngroups)
582 last = ngroups;
583 for (i = parent_group; i < last; i++) {
584 desc = ext4_get_group_desc(sb, i, NULL);
585 if (desc && ext4_free_inodes_count(sb, desc)) {
586 *group = i;
587 return 0;
588 }
589 }
590 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
591 retry = 1;
592 parent_group = EXT4_I(parent)->i_last_alloc_group;
593 goto try_again;
594 }
595 /*
596 * If this didn't work, use the Orlov search algorithm
597 * to find a new flex group; we pass in the mode to
598 * avoid the topdir algorithms.
599 */
600 *group = parent_group + flex_size;
601 if (*group > ngroups)
602 *group = 0;
603 return find_group_orlov(sb, parent, group, mode, NULL);
604 }
605
606 /*
607 * Try to place the inode in its parent directory
608 */
609 *group = parent_group;
610 desc = ext4_get_group_desc(sb, *group, NULL);
611 if (desc && ext4_free_inodes_count(sb, desc) &&
612 ext4_free_group_clusters(sb, desc))
613 return 0;
614
615 /*
616 * We're going to place this inode in a different blockgroup from its
617 * parent. We want to cause files in a common directory to all land in
618 * the same blockgroup. But we want files which are in a different
619 * directory which shares a blockgroup with our parent to land in a
620 * different blockgroup.
621 *
622 * So add our directory's i_ino into the starting point for the hash.
623 */
624 *group = (*group + parent->i_ino) % ngroups;
625
626 /*
627 * Use a quadratic hash to find a group with a free inode and some free
628 * blocks.
629 */
630 for (i = 1; i < ngroups; i <<= 1) {
631 *group += i;
632 if (*group >= ngroups)
633 *group -= ngroups;
634 desc = ext4_get_group_desc(sb, *group, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc) &&
636 ext4_free_group_clusters(sb, desc))
637 return 0;
638 }
639
640 /*
641 * That failed: try linear search for a free inode, even if that group
642 * has no free blocks.
643 */
644 *group = parent_group;
645 for (i = 0; i < ngroups; i++) {
646 if (++*group >= ngroups)
647 *group = 0;
648 desc = ext4_get_group_desc(sb, *group, NULL);
649 if (desc && ext4_free_inodes_count(sb, desc))
650 return 0;
651 }
652
653 return -1;
654 }
655
656 /*
657 * In no journal mode, if an inode has recently been deleted, we want
658 * to avoid reusing it until we're reasonably sure the inode table
659 * block has been written back to disk. (Yes, these values are
660 * somewhat arbitrary...)
661 */
662 #define RECENTCY_MIN 5
663 #define RECENTCY_DIRTY 30
664
recently_deleted(struct super_block * sb,ext4_group_t group,int ino)665 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
666 {
667 struct ext4_group_desc *gdp;
668 struct ext4_inode *raw_inode;
669 struct buffer_head *bh;
670 unsigned long dtime, now;
671 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
672 int offset, ret = 0, recentcy = RECENTCY_MIN;
673
674 gdp = ext4_get_group_desc(sb, group, NULL);
675 if (unlikely(!gdp))
676 return 0;
677
678 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
679 (ino / inodes_per_block));
680 if (unlikely(!bh) || !buffer_uptodate(bh))
681 /*
682 * If the block is not in the buffer cache, then it
683 * must have been written out.
684 */
685 goto out;
686
687 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
688 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
689 dtime = le32_to_cpu(raw_inode->i_dtime);
690 now = get_seconds();
691 if (buffer_dirty(bh))
692 recentcy += RECENTCY_DIRTY;
693
694 if (dtime && (dtime < now) && (now < dtime + recentcy))
695 ret = 1;
696 out:
697 brelse(bh);
698 return ret;
699 }
700
701 /*
702 * There are two policies for allocating an inode. If the new inode is
703 * a directory, then a forward search is made for a block group with both
704 * free space and a low directory-to-inode ratio; if that fails, then of
705 * the groups with above-average free space, that group with the fewest
706 * directories already is chosen.
707 *
708 * For other inodes, search forward from the parent directory's block
709 * group to find a free inode.
710 */
__ext4_new_inode(handle_t * handle,struct inode * dir,umode_t mode,const struct qstr * qstr,__u32 goal,uid_t * owner,int handle_type,unsigned int line_no,int nblocks)711 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
712 umode_t mode, const struct qstr *qstr,
713 __u32 goal, uid_t *owner, int handle_type,
714 unsigned int line_no, int nblocks)
715 {
716 struct super_block *sb;
717 struct buffer_head *inode_bitmap_bh = NULL;
718 struct buffer_head *group_desc_bh;
719 ext4_group_t ngroups, group = 0;
720 unsigned long ino = 0;
721 struct inode *inode;
722 struct ext4_group_desc *gdp = NULL;
723 struct ext4_inode_info *ei;
724 struct ext4_sb_info *sbi;
725 int ret2, err = 0;
726 struct inode *ret;
727 ext4_group_t i;
728 ext4_group_t flex_group;
729 struct ext4_group_info *grp;
730 int encrypt = 0;
731
732 /* Cannot create files in a deleted directory */
733 if (!dir || !dir->i_nlink)
734 return ERR_PTR(-EPERM);
735
736 if ((ext4_encrypted_inode(dir) ||
737 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
738 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
739 err = ext4_get_encryption_info(dir);
740 if (err)
741 return ERR_PTR(err);
742 if (ext4_encryption_info(dir) == NULL)
743 return ERR_PTR(-EPERM);
744 if (!handle)
745 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
746 encrypt = 1;
747 }
748
749 sb = dir->i_sb;
750 ngroups = ext4_get_groups_count(sb);
751 trace_ext4_request_inode(dir, mode);
752 inode = new_inode(sb);
753 if (!inode)
754 return ERR_PTR(-ENOMEM);
755 ei = EXT4_I(inode);
756 sbi = EXT4_SB(sb);
757
758 /*
759 * Initalize owners and quota early so that we don't have to account
760 * for quota initialization worst case in standard inode creating
761 * transaction
762 */
763 if (owner) {
764 inode->i_mode = mode;
765 i_uid_write(inode, owner[0]);
766 i_gid_write(inode, owner[1]);
767 } else if (test_opt(sb, GRPID)) {
768 inode->i_mode = mode;
769 inode->i_uid = current_fsuid();
770 inode->i_gid = dir->i_gid;
771 } else
772 inode_init_owner(inode, dir, mode);
773 dquot_initialize(inode);
774
775 if (!goal)
776 goal = sbi->s_inode_goal;
777
778 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
779 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
780 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
781 ret2 = 0;
782 goto got_group;
783 }
784
785 if (S_ISDIR(mode))
786 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
787 else
788 ret2 = find_group_other(sb, dir, &group, mode);
789
790 got_group:
791 EXT4_I(dir)->i_last_alloc_group = group;
792 err = -ENOSPC;
793 if (ret2 == -1)
794 goto out;
795
796 /*
797 * Normally we will only go through one pass of this loop,
798 * unless we get unlucky and it turns out the group we selected
799 * had its last inode grabbed by someone else.
800 */
801 for (i = 0; i < ngroups; i++, ino = 0) {
802 err = -EIO;
803
804 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
805 if (!gdp)
806 goto out;
807
808 /*
809 * Check free inodes count before loading bitmap.
810 */
811 if (ext4_free_inodes_count(sb, gdp) == 0) {
812 if (++group == ngroups)
813 group = 0;
814 continue;
815 }
816
817 grp = ext4_get_group_info(sb, group);
818 /* Skip groups with already-known suspicious inode tables */
819 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
820 if (++group == ngroups)
821 group = 0;
822 continue;
823 }
824
825 brelse(inode_bitmap_bh);
826 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
827 /* Skip groups with suspicious inode tables */
828 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
829 if (++group == ngroups)
830 group = 0;
831 continue;
832 }
833
834 repeat_in_this_group:
835 ino = ext4_find_next_zero_bit((unsigned long *)
836 inode_bitmap_bh->b_data,
837 EXT4_INODES_PER_GROUP(sb), ino);
838 if (ino >= EXT4_INODES_PER_GROUP(sb))
839 goto next_group;
840 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
841 ext4_error(sb, "reserved inode found cleared - "
842 "inode=%lu", ino + 1);
843 continue;
844 }
845 if ((EXT4_SB(sb)->s_journal == NULL) &&
846 recently_deleted(sb, group, ino)) {
847 ino++;
848 goto next_inode;
849 }
850 if (!handle) {
851 BUG_ON(nblocks <= 0);
852 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
853 handle_type, nblocks,
854 0);
855 if (IS_ERR(handle)) {
856 err = PTR_ERR(handle);
857 ext4_std_error(sb, err);
858 goto out;
859 }
860 }
861 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
862 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
863 if (err) {
864 ext4_std_error(sb, err);
865 goto out;
866 }
867 ext4_lock_group(sb, group);
868 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
869 ext4_unlock_group(sb, group);
870 ino++; /* the inode bitmap is zero-based */
871 if (!ret2)
872 goto got; /* we grabbed the inode! */
873 next_inode:
874 if (ino < EXT4_INODES_PER_GROUP(sb))
875 goto repeat_in_this_group;
876 next_group:
877 if (++group == ngroups)
878 group = 0;
879 }
880 err = -ENOSPC;
881 goto out;
882
883 got:
884 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
885 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
886 if (err) {
887 ext4_std_error(sb, err);
888 goto out;
889 }
890
891 BUFFER_TRACE(group_desc_bh, "get_write_access");
892 err = ext4_journal_get_write_access(handle, group_desc_bh);
893 if (err) {
894 ext4_std_error(sb, err);
895 goto out;
896 }
897
898 /* We may have to initialize the block bitmap if it isn't already */
899 if (ext4_has_group_desc_csum(sb) &&
900 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
901 struct buffer_head *block_bitmap_bh;
902
903 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
904 if (!block_bitmap_bh) {
905 err = -EIO;
906 goto out;
907 }
908 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
909 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
910 if (err) {
911 brelse(block_bitmap_bh);
912 ext4_std_error(sb, err);
913 goto out;
914 }
915
916 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
917 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
918
919 /* recheck and clear flag under lock if we still need to */
920 ext4_lock_group(sb, group);
921 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
922 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
923 ext4_free_group_clusters_set(sb, gdp,
924 ext4_free_clusters_after_init(sb, group, gdp));
925 ext4_block_bitmap_csum_set(sb, group, gdp,
926 block_bitmap_bh);
927 ext4_group_desc_csum_set(sb, group, gdp);
928 }
929 ext4_unlock_group(sb, group);
930 brelse(block_bitmap_bh);
931
932 if (err) {
933 ext4_std_error(sb, err);
934 goto out;
935 }
936 }
937
938 /* Update the relevant bg descriptor fields */
939 if (ext4_has_group_desc_csum(sb)) {
940 int free;
941 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
942
943 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
944 ext4_lock_group(sb, group); /* while we modify the bg desc */
945 free = EXT4_INODES_PER_GROUP(sb) -
946 ext4_itable_unused_count(sb, gdp);
947 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
948 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
949 free = 0;
950 }
951 /*
952 * Check the relative inode number against the last used
953 * relative inode number in this group. if it is greater
954 * we need to update the bg_itable_unused count
955 */
956 if (ino > free)
957 ext4_itable_unused_set(sb, gdp,
958 (EXT4_INODES_PER_GROUP(sb) - ino));
959 up_read(&grp->alloc_sem);
960 } else {
961 ext4_lock_group(sb, group);
962 }
963
964 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
965 if (S_ISDIR(mode)) {
966 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
967 if (sbi->s_log_groups_per_flex) {
968 ext4_group_t f = ext4_flex_group(sbi, group);
969
970 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
971 }
972 }
973 if (ext4_has_group_desc_csum(sb)) {
974 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
975 EXT4_INODES_PER_GROUP(sb) / 8);
976 ext4_group_desc_csum_set(sb, group, gdp);
977 }
978 ext4_unlock_group(sb, group);
979
980 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
981 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
982 if (err) {
983 ext4_std_error(sb, err);
984 goto out;
985 }
986
987 percpu_counter_dec(&sbi->s_freeinodes_counter);
988 if (S_ISDIR(mode))
989 percpu_counter_inc(&sbi->s_dirs_counter);
990
991 if (sbi->s_log_groups_per_flex) {
992 flex_group = ext4_flex_group(sbi, group);
993 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
994 }
995
996 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
997 /* This is the optimal IO size (for stat), not the fs block size */
998 inode->i_blocks = 0;
999 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1000 ext4_current_time(inode);
1001
1002 memset(ei->i_data, 0, sizeof(ei->i_data));
1003 ei->i_dir_start_lookup = 0;
1004 ei->i_disksize = 0;
1005
1006 /* Don't inherit extent flag from directory, amongst others. */
1007 ei->i_flags =
1008 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1009 ei->i_file_acl = 0;
1010 ei->i_dtime = 0;
1011 ei->i_block_group = group;
1012 ei->i_last_alloc_group = ~0;
1013
1014 ext4_set_inode_flags(inode);
1015 if (IS_DIRSYNC(inode))
1016 ext4_handle_sync(handle);
1017 if (insert_inode_locked(inode) < 0) {
1018 /*
1019 * Likely a bitmap corruption causing inode to be allocated
1020 * twice.
1021 */
1022 err = -EIO;
1023 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1024 inode->i_ino);
1025 goto out;
1026 }
1027 spin_lock(&sbi->s_next_gen_lock);
1028 inode->i_generation = sbi->s_next_generation++;
1029 spin_unlock(&sbi->s_next_gen_lock);
1030
1031 /* Precompute checksum seed for inode metadata */
1032 if (ext4_has_metadata_csum(sb)) {
1033 __u32 csum;
1034 __le32 inum = cpu_to_le32(inode->i_ino);
1035 __le32 gen = cpu_to_le32(inode->i_generation);
1036 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1037 sizeof(inum));
1038 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1039 sizeof(gen));
1040 }
1041
1042 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1043 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1044
1045 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1046 ei->i_inline_off = 0;
1047 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1048 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1049 ret = inode;
1050 err = dquot_alloc_inode(inode);
1051 if (err)
1052 goto fail_drop;
1053
1054 err = ext4_init_acl(handle, inode, dir);
1055 if (err)
1056 goto fail_free_drop;
1057
1058 err = ext4_init_security(handle, inode, dir, qstr);
1059 if (err)
1060 goto fail_free_drop;
1061
1062 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1063 /* set extent flag only for directory, file and normal symlink*/
1064 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1065 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1066 ext4_ext_tree_init(handle, inode);
1067 }
1068 }
1069
1070 if (ext4_handle_valid(handle)) {
1071 ei->i_sync_tid = handle->h_transaction->t_tid;
1072 ei->i_datasync_tid = handle->h_transaction->t_tid;
1073 }
1074
1075 if (encrypt) {
1076 err = ext4_inherit_context(dir, inode);
1077 if (err)
1078 goto fail_free_drop;
1079 }
1080
1081 err = ext4_mark_inode_dirty(handle, inode);
1082 if (err) {
1083 ext4_std_error(sb, err);
1084 goto fail_free_drop;
1085 }
1086
1087 ext4_debug("allocating inode %lu\n", inode->i_ino);
1088 trace_ext4_allocate_inode(inode, dir, mode);
1089 brelse(inode_bitmap_bh);
1090 return ret;
1091
1092 fail_free_drop:
1093 dquot_free_inode(inode);
1094 fail_drop:
1095 clear_nlink(inode);
1096 unlock_new_inode(inode);
1097 out:
1098 dquot_drop(inode);
1099 inode->i_flags |= S_NOQUOTA;
1100 iput(inode);
1101 brelse(inode_bitmap_bh);
1102 return ERR_PTR(err);
1103 }
1104
1105 /* Verify that we are loading a valid orphan from disk */
ext4_orphan_get(struct super_block * sb,unsigned long ino)1106 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1107 {
1108 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1109 ext4_group_t block_group;
1110 int bit;
1111 struct buffer_head *bitmap_bh = NULL;
1112 struct inode *inode = NULL;
1113 int err = -EIO;
1114
1115 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1116 goto bad_orphan;
1117
1118 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1119 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1120 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1121 if (!bitmap_bh) {
1122 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1123 ino, PTR_ERR(bitmap_bh));
1124 return (struct inode *) bitmap_bh;
1125 }
1126
1127 /* Having the inode bit set should be a 100% indicator that this
1128 * is a valid orphan (no e2fsck run on fs). Orphans also include
1129 * inodes that were being truncated, so we can't check i_nlink==0.
1130 */
1131 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1132 goto bad_orphan;
1133
1134 inode = ext4_iget(sb, ino);
1135 if (IS_ERR(inode)) {
1136 err = PTR_ERR(inode);
1137 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1138 ino, err);
1139 return inode;
1140 }
1141
1142 /*
1143 * If the orphans has i_nlinks > 0 then it should be able to
1144 * be truncated, otherwise it won't be removed from the orphan
1145 * list during processing and an infinite loop will result.
1146 * Similarly, it must not be a bad inode.
1147 */
1148 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1149 is_bad_inode(inode))
1150 goto bad_orphan;
1151
1152 if (NEXT_ORPHAN(inode) > max_ino)
1153 goto bad_orphan;
1154 brelse(bitmap_bh);
1155 return inode;
1156
1157 bad_orphan:
1158 ext4_error(sb, "bad orphan inode %lu", ino);
1159 if (bitmap_bh)
1160 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1161 bit, (unsigned long long)bitmap_bh->b_blocknr,
1162 ext4_test_bit(bit, bitmap_bh->b_data));
1163 if (inode) {
1164 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1165 is_bad_inode(inode));
1166 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1167 NEXT_ORPHAN(inode));
1168 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1169 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1170 /* Avoid freeing blocks if we got a bad deleted inode */
1171 if (inode->i_nlink == 0)
1172 inode->i_blocks = 0;
1173 iput(inode);
1174 }
1175 brelse(bitmap_bh);
1176 return ERR_PTR(err);
1177 }
1178
ext4_count_free_inodes(struct super_block * sb)1179 unsigned long ext4_count_free_inodes(struct super_block *sb)
1180 {
1181 unsigned long desc_count;
1182 struct ext4_group_desc *gdp;
1183 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1184 #ifdef EXT4FS_DEBUG
1185 struct ext4_super_block *es;
1186 unsigned long bitmap_count, x;
1187 struct buffer_head *bitmap_bh = NULL;
1188
1189 es = EXT4_SB(sb)->s_es;
1190 desc_count = 0;
1191 bitmap_count = 0;
1192 gdp = NULL;
1193 for (i = 0; i < ngroups; i++) {
1194 gdp = ext4_get_group_desc(sb, i, NULL);
1195 if (!gdp)
1196 continue;
1197 desc_count += ext4_free_inodes_count(sb, gdp);
1198 brelse(bitmap_bh);
1199 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1200 if (!bitmap_bh)
1201 continue;
1202
1203 x = ext4_count_free(bitmap_bh->b_data,
1204 EXT4_INODES_PER_GROUP(sb) / 8);
1205 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1206 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1207 bitmap_count += x;
1208 }
1209 brelse(bitmap_bh);
1210 printk(KERN_DEBUG "ext4_count_free_inodes: "
1211 "stored = %u, computed = %lu, %lu\n",
1212 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1213 return desc_count;
1214 #else
1215 desc_count = 0;
1216 for (i = 0; i < ngroups; i++) {
1217 gdp = ext4_get_group_desc(sb, i, NULL);
1218 if (!gdp)
1219 continue;
1220 desc_count += ext4_free_inodes_count(sb, gdp);
1221 cond_resched();
1222 }
1223 return desc_count;
1224 #endif
1225 }
1226
1227 /* Called at mount-time, super-block is locked */
ext4_count_dirs(struct super_block * sb)1228 unsigned long ext4_count_dirs(struct super_block * sb)
1229 {
1230 unsigned long count = 0;
1231 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1232
1233 for (i = 0; i < ngroups; i++) {
1234 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1235 if (!gdp)
1236 continue;
1237 count += ext4_used_dirs_count(sb, gdp);
1238 }
1239 return count;
1240 }
1241
1242 /*
1243 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1244 * inode table. Must be called without any spinlock held. The only place
1245 * where it is called from on active part of filesystem is ext4lazyinit
1246 * thread, so we do not need any special locks, however we have to prevent
1247 * inode allocation from the current group, so we take alloc_sem lock, to
1248 * block ext4_new_inode() until we are finished.
1249 */
ext4_init_inode_table(struct super_block * sb,ext4_group_t group,int barrier)1250 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1251 int barrier)
1252 {
1253 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1254 struct ext4_sb_info *sbi = EXT4_SB(sb);
1255 struct ext4_group_desc *gdp = NULL;
1256 struct buffer_head *group_desc_bh;
1257 handle_t *handle;
1258 ext4_fsblk_t blk;
1259 int num, ret = 0, used_blks = 0;
1260
1261 /* This should not happen, but just to be sure check this */
1262 if (sb->s_flags & MS_RDONLY) {
1263 ret = 1;
1264 goto out;
1265 }
1266
1267 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1268 if (!gdp)
1269 goto out;
1270
1271 /*
1272 * We do not need to lock this, because we are the only one
1273 * handling this flag.
1274 */
1275 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1276 goto out;
1277
1278 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1279 if (IS_ERR(handle)) {
1280 ret = PTR_ERR(handle);
1281 goto out;
1282 }
1283
1284 down_write(&grp->alloc_sem);
1285 /*
1286 * If inode bitmap was already initialized there may be some
1287 * used inodes so we need to skip blocks with used inodes in
1288 * inode table.
1289 */
1290 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1291 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1292 ext4_itable_unused_count(sb, gdp)),
1293 sbi->s_inodes_per_block);
1294
1295 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1296 ext4_error(sb, "Something is wrong with group %u: "
1297 "used itable blocks: %d; "
1298 "itable unused count: %u",
1299 group, used_blks,
1300 ext4_itable_unused_count(sb, gdp));
1301 ret = 1;
1302 goto err_out;
1303 }
1304
1305 blk = ext4_inode_table(sb, gdp) + used_blks;
1306 num = sbi->s_itb_per_group - used_blks;
1307
1308 BUFFER_TRACE(group_desc_bh, "get_write_access");
1309 ret = ext4_journal_get_write_access(handle,
1310 group_desc_bh);
1311 if (ret)
1312 goto err_out;
1313
1314 /*
1315 * Skip zeroout if the inode table is full. But we set the ZEROED
1316 * flag anyway, because obviously, when it is full it does not need
1317 * further zeroing.
1318 */
1319 if (unlikely(num == 0))
1320 goto skip_zeroout;
1321
1322 ext4_debug("going to zero out inode table in group %d\n",
1323 group);
1324 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1325 if (ret < 0)
1326 goto err_out;
1327 if (barrier)
1328 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1329
1330 skip_zeroout:
1331 ext4_lock_group(sb, group);
1332 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1333 ext4_group_desc_csum_set(sb, group, gdp);
1334 ext4_unlock_group(sb, group);
1335
1336 BUFFER_TRACE(group_desc_bh,
1337 "call ext4_handle_dirty_metadata");
1338 ret = ext4_handle_dirty_metadata(handle, NULL,
1339 group_desc_bh);
1340
1341 err_out:
1342 up_write(&grp->alloc_sem);
1343 ext4_journal_stop(handle);
1344 out:
1345 return ret;
1346 }
1347