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/stat.h>
18 #include <linux/string.h>
19 #include <linux/quotaops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/random.h>
22 #include <linux/bitops.h>
23 #include <linux/blkdev.h>
24 #include <asm/byteorder.h>
25
26 #include "ext4.h"
27 #include "ext4_jbd2.h"
28 #include "xattr.h"
29 #include "acl.h"
30
31 #include <trace/events/ext4.h>
32
33 /*
34 * ialloc.c contains the inodes allocation and deallocation routines
35 */
36
37 /*
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
41 *
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
45 */
46
47 /*
48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
49 * need to use it within a single byte (to ensure we get endianness right).
50 * We can use memset for the rest of the bitmap as there are no other users.
51 */
ext4_mark_bitmap_end(int start_bit,int end_bit,char * bitmap)52 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
53 {
54 int i;
55
56 if (start_bit >= end_bit)
57 return;
58
59 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 ext4_set_bit(i, bitmap);
62 if (i < end_bit)
63 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64 }
65
ext4_end_bitmap_read(struct buffer_head * bh,int uptodate)66 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
67 {
68 if (uptodate) {
69 set_buffer_uptodate(bh);
70 set_bitmap_uptodate(bh);
71 }
72 unlock_buffer(bh);
73 put_bh(bh);
74 }
75
ext4_validate_inode_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)76 static int ext4_validate_inode_bitmap(struct super_block *sb,
77 struct ext4_group_desc *desc,
78 ext4_group_t block_group,
79 struct buffer_head *bh)
80 {
81 ext4_fsblk_t blk;
82 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
83 struct ext4_sb_info *sbi = EXT4_SB(sb);
84
85 if (buffer_verified(bh))
86 return 0;
87 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
88 return -EFSCORRUPTED;
89
90 ext4_lock_group(sb, block_group);
91 blk = ext4_inode_bitmap(sb, desc);
92 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
93 EXT4_INODES_PER_GROUP(sb) / 8)) {
94 ext4_unlock_group(sb, block_group);
95 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
96 "inode_bitmap = %llu", block_group, blk);
97 grp = ext4_get_group_info(sb, block_group);
98 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
99 int count;
100 count = ext4_free_inodes_count(sb, desc);
101 percpu_counter_sub(&sbi->s_freeinodes_counter,
102 count);
103 }
104 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
105 return -EFSBADCRC;
106 }
107 set_buffer_verified(bh);
108 ext4_unlock_group(sb, block_group);
109 return 0;
110 }
111
112 /*
113 * Read the inode allocation bitmap for a given block_group, reading
114 * into the specified slot in the superblock's bitmap cache.
115 *
116 * Return buffer_head of bitmap on success or NULL.
117 */
118 static struct buffer_head *
ext4_read_inode_bitmap(struct super_block * sb,ext4_group_t block_group)119 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
120 {
121 struct ext4_group_desc *desc;
122 struct buffer_head *bh = NULL;
123 ext4_fsblk_t bitmap_blk;
124 int err;
125
126 desc = ext4_get_group_desc(sb, block_group, NULL);
127 if (!desc)
128 return ERR_PTR(-EFSCORRUPTED);
129
130 bitmap_blk = ext4_inode_bitmap(sb, desc);
131 bh = sb_getblk(sb, bitmap_blk);
132 if (unlikely(!bh)) {
133 ext4_error(sb, "Cannot read inode bitmap - "
134 "block_group = %u, inode_bitmap = %llu",
135 block_group, bitmap_blk);
136 return ERR_PTR(-EIO);
137 }
138 if (bitmap_uptodate(bh))
139 goto verify;
140
141 lock_buffer(bh);
142 if (bitmap_uptodate(bh)) {
143 unlock_buffer(bh);
144 goto verify;
145 }
146
147 ext4_lock_group(sb, block_group);
148 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
149 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
150 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
151 sb->s_blocksize * 8, bh->b_data);
152 set_bitmap_uptodate(bh);
153 set_buffer_uptodate(bh);
154 set_buffer_verified(bh);
155 ext4_unlock_group(sb, block_group);
156 unlock_buffer(bh);
157 return bh;
158 }
159 ext4_unlock_group(sb, block_group);
160
161 if (buffer_uptodate(bh)) {
162 /*
163 * if not uninit if bh is uptodate,
164 * bitmap is also uptodate
165 */
166 set_bitmap_uptodate(bh);
167 unlock_buffer(bh);
168 goto verify;
169 }
170 /*
171 * submit the buffer_head for reading
172 */
173 trace_ext4_load_inode_bitmap(sb, block_group);
174 bh->b_end_io = ext4_end_bitmap_read;
175 get_bh(bh);
176 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
177 wait_on_buffer(bh);
178 if (!buffer_uptodate(bh)) {
179 put_bh(bh);
180 ext4_error(sb, "Cannot read inode bitmap - "
181 "block_group = %u, inode_bitmap = %llu",
182 block_group, bitmap_blk);
183 return ERR_PTR(-EIO);
184 }
185
186 verify:
187 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
188 if (err)
189 goto out;
190 return bh;
191 out:
192 put_bh(bh);
193 return ERR_PTR(err);
194 }
195
196 /*
197 * NOTE! When we get the inode, we're the only people
198 * that have access to it, and as such there are no
199 * race conditions we have to worry about. The inode
200 * is not on the hash-lists, and it cannot be reached
201 * through the filesystem because the directory entry
202 * has been deleted earlier.
203 *
204 * HOWEVER: we must make sure that we get no aliases,
205 * which means that we have to call "clear_inode()"
206 * _before_ we mark the inode not in use in the inode
207 * bitmaps. Otherwise a newly created file might use
208 * the same inode number (not actually the same pointer
209 * though), and then we'd have two inodes sharing the
210 * same inode number and space on the harddisk.
211 */
ext4_free_inode(handle_t * handle,struct inode * inode)212 void ext4_free_inode(handle_t *handle, struct inode *inode)
213 {
214 struct super_block *sb = inode->i_sb;
215 int is_directory;
216 unsigned long ino;
217 struct buffer_head *bitmap_bh = NULL;
218 struct buffer_head *bh2;
219 ext4_group_t block_group;
220 unsigned long bit;
221 struct ext4_group_desc *gdp;
222 struct ext4_super_block *es;
223 struct ext4_sb_info *sbi;
224 int fatal = 0, err, count, cleared;
225 struct ext4_group_info *grp;
226
227 if (!sb) {
228 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
229 "nonexistent device\n", __func__, __LINE__);
230 return;
231 }
232 if (atomic_read(&inode->i_count) > 1) {
233 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
234 __func__, __LINE__, inode->i_ino,
235 atomic_read(&inode->i_count));
236 return;
237 }
238 if (inode->i_nlink) {
239 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
240 __func__, __LINE__, inode->i_ino, inode->i_nlink);
241 return;
242 }
243 sbi = EXT4_SB(sb);
244
245 ino = inode->i_ino;
246 ext4_debug("freeing inode %lu\n", ino);
247 trace_ext4_free_inode(inode);
248
249 /*
250 * Note: we must free any quota before locking the superblock,
251 * as writing the quota to disk may need the lock as well.
252 */
253 dquot_initialize(inode);
254 ext4_xattr_delete_inode(handle, inode);
255 dquot_free_inode(inode);
256 dquot_drop(inode);
257
258 is_directory = S_ISDIR(inode->i_mode);
259
260 /* Do this BEFORE marking the inode not in use or returning an error */
261 ext4_clear_inode(inode);
262
263 es = EXT4_SB(sb)->s_es;
264 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
265 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
266 goto error_return;
267 }
268 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
269 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
270 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
271 /* Don't bother if the inode bitmap is corrupt. */
272 grp = ext4_get_group_info(sb, block_group);
273 if (IS_ERR(bitmap_bh)) {
274 fatal = PTR_ERR(bitmap_bh);
275 bitmap_bh = NULL;
276 goto error_return;
277 }
278 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
279 fatal = -EFSCORRUPTED;
280 goto error_return;
281 }
282
283 BUFFER_TRACE(bitmap_bh, "get_write_access");
284 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
285 if (fatal)
286 goto error_return;
287
288 fatal = -ESRCH;
289 gdp = ext4_get_group_desc(sb, block_group, &bh2);
290 if (gdp) {
291 BUFFER_TRACE(bh2, "get_write_access");
292 fatal = ext4_journal_get_write_access(handle, bh2);
293 }
294 ext4_lock_group(sb, block_group);
295 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
296 if (fatal || !cleared) {
297 ext4_unlock_group(sb, block_group);
298 goto out;
299 }
300
301 count = ext4_free_inodes_count(sb, gdp) + 1;
302 ext4_free_inodes_set(sb, gdp, count);
303 if (is_directory) {
304 count = ext4_used_dirs_count(sb, gdp) - 1;
305 ext4_used_dirs_set(sb, gdp, count);
306 percpu_counter_dec(&sbi->s_dirs_counter);
307 }
308 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
309 EXT4_INODES_PER_GROUP(sb) / 8);
310 ext4_group_desc_csum_set(sb, block_group, gdp);
311 ext4_unlock_group(sb, block_group);
312
313 percpu_counter_inc(&sbi->s_freeinodes_counter);
314 if (sbi->s_log_groups_per_flex) {
315 ext4_group_t f = ext4_flex_group(sbi, block_group);
316
317 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
318 if (is_directory)
319 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
320 }
321 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
322 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
323 out:
324 if (cleared) {
325 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
326 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
327 if (!fatal)
328 fatal = err;
329 } else {
330 ext4_error(sb, "bit already cleared for inode %lu", ino);
331 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
332 int count;
333 count = ext4_free_inodes_count(sb, gdp);
334 percpu_counter_sub(&sbi->s_freeinodes_counter,
335 count);
336 }
337 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
338 }
339
340 error_return:
341 brelse(bitmap_bh);
342 ext4_std_error(sb, fatal);
343 }
344
345 struct orlov_stats {
346 __u64 free_clusters;
347 __u32 free_inodes;
348 __u32 used_dirs;
349 };
350
351 /*
352 * Helper function for Orlov's allocator; returns critical information
353 * for a particular block group or flex_bg. If flex_size is 1, then g
354 * is a block group number; otherwise it is flex_bg number.
355 */
get_orlov_stats(struct super_block * sb,ext4_group_t g,int flex_size,struct orlov_stats * stats)356 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
357 int flex_size, struct orlov_stats *stats)
358 {
359 struct ext4_group_desc *desc;
360 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
361
362 if (flex_size > 1) {
363 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
364 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
365 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
366 return;
367 }
368
369 desc = ext4_get_group_desc(sb, g, NULL);
370 if (desc) {
371 stats->free_inodes = ext4_free_inodes_count(sb, desc);
372 stats->free_clusters = ext4_free_group_clusters(sb, desc);
373 stats->used_dirs = ext4_used_dirs_count(sb, desc);
374 } else {
375 stats->free_inodes = 0;
376 stats->free_clusters = 0;
377 stats->used_dirs = 0;
378 }
379 }
380
381 /*
382 * Orlov's allocator for directories.
383 *
384 * We always try to spread first-level directories.
385 *
386 * If there are blockgroups with both free inodes and free blocks counts
387 * not worse than average we return one with smallest directory count.
388 * Otherwise we simply return a random group.
389 *
390 * For the rest rules look so:
391 *
392 * It's OK to put directory into a group unless
393 * it has too many directories already (max_dirs) or
394 * it has too few free inodes left (min_inodes) or
395 * it has too few free blocks left (min_blocks) or
396 * Parent's group is preferred, if it doesn't satisfy these
397 * conditions we search cyclically through the rest. If none
398 * of the groups look good we just look for a group with more
399 * free inodes than average (starting at parent's group).
400 */
401
find_group_orlov(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode,const struct qstr * qstr)402 static int find_group_orlov(struct super_block *sb, struct inode *parent,
403 ext4_group_t *group, umode_t mode,
404 const struct qstr *qstr)
405 {
406 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
407 struct ext4_sb_info *sbi = EXT4_SB(sb);
408 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
409 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
410 unsigned int freei, avefreei, grp_free;
411 ext4_fsblk_t freeb, avefreec;
412 unsigned int ndirs;
413 int max_dirs, min_inodes;
414 ext4_grpblk_t min_clusters;
415 ext4_group_t i, grp, g, ngroups;
416 struct ext4_group_desc *desc;
417 struct orlov_stats stats;
418 int flex_size = ext4_flex_bg_size(sbi);
419 struct dx_hash_info hinfo;
420
421 ngroups = real_ngroups;
422 if (flex_size > 1) {
423 ngroups = (real_ngroups + flex_size - 1) >>
424 sbi->s_log_groups_per_flex;
425 parent_group >>= sbi->s_log_groups_per_flex;
426 }
427
428 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
429 avefreei = freei / ngroups;
430 freeb = EXT4_C2B(sbi,
431 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
432 avefreec = freeb;
433 do_div(avefreec, ngroups);
434 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
435
436 if (S_ISDIR(mode) &&
437 ((parent == d_inode(sb->s_root)) ||
438 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
439 int best_ndir = inodes_per_group;
440 int ret = -1;
441
442 if (qstr) {
443 hinfo.hash_version = DX_HASH_HALF_MD4;
444 hinfo.seed = sbi->s_hash_seed;
445 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
446 grp = hinfo.hash;
447 } else
448 grp = prandom_u32();
449 parent_group = (unsigned)grp % ngroups;
450 for (i = 0; i < ngroups; i++) {
451 g = (parent_group + i) % ngroups;
452 get_orlov_stats(sb, g, flex_size, &stats);
453 if (!stats.free_inodes)
454 continue;
455 if (stats.used_dirs >= best_ndir)
456 continue;
457 if (stats.free_inodes < avefreei)
458 continue;
459 if (stats.free_clusters < avefreec)
460 continue;
461 grp = g;
462 ret = 0;
463 best_ndir = stats.used_dirs;
464 }
465 if (ret)
466 goto fallback;
467 found_flex_bg:
468 if (flex_size == 1) {
469 *group = grp;
470 return 0;
471 }
472
473 /*
474 * We pack inodes at the beginning of the flexgroup's
475 * inode tables. Block allocation decisions will do
476 * something similar, although regular files will
477 * start at 2nd block group of the flexgroup. See
478 * ext4_ext_find_goal() and ext4_find_near().
479 */
480 grp *= flex_size;
481 for (i = 0; i < flex_size; i++) {
482 if (grp+i >= real_ngroups)
483 break;
484 desc = ext4_get_group_desc(sb, grp+i, NULL);
485 if (desc && ext4_free_inodes_count(sb, desc)) {
486 *group = grp+i;
487 return 0;
488 }
489 }
490 goto fallback;
491 }
492
493 max_dirs = ndirs / ngroups + inodes_per_group / 16;
494 min_inodes = avefreei - inodes_per_group*flex_size / 4;
495 if (min_inodes < 1)
496 min_inodes = 1;
497 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
498
499 /*
500 * Start looking in the flex group where we last allocated an
501 * inode for this parent directory
502 */
503 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
504 parent_group = EXT4_I(parent)->i_last_alloc_group;
505 if (flex_size > 1)
506 parent_group >>= sbi->s_log_groups_per_flex;
507 }
508
509 for (i = 0; i < ngroups; i++) {
510 grp = (parent_group + i) % ngroups;
511 get_orlov_stats(sb, grp, flex_size, &stats);
512 if (stats.used_dirs >= max_dirs)
513 continue;
514 if (stats.free_inodes < min_inodes)
515 continue;
516 if (stats.free_clusters < min_clusters)
517 continue;
518 goto found_flex_bg;
519 }
520
521 fallback:
522 ngroups = real_ngroups;
523 avefreei = freei / ngroups;
524 fallback_retry:
525 parent_group = EXT4_I(parent)->i_block_group;
526 for (i = 0; i < ngroups; i++) {
527 grp = (parent_group + i) % ngroups;
528 desc = ext4_get_group_desc(sb, grp, NULL);
529 if (desc) {
530 grp_free = ext4_free_inodes_count(sb, desc);
531 if (grp_free && grp_free >= avefreei) {
532 *group = grp;
533 return 0;
534 }
535 }
536 }
537
538 if (avefreei) {
539 /*
540 * The free-inodes counter is approximate, and for really small
541 * filesystems the above test can fail to find any blockgroups
542 */
543 avefreei = 0;
544 goto fallback_retry;
545 }
546
547 return -1;
548 }
549
find_group_other(struct super_block * sb,struct inode * parent,ext4_group_t * group,umode_t mode)550 static int find_group_other(struct super_block *sb, struct inode *parent,
551 ext4_group_t *group, umode_t mode)
552 {
553 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
554 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
555 struct ext4_group_desc *desc;
556 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
557
558 /*
559 * Try to place the inode is the same flex group as its
560 * parent. If we can't find space, use the Orlov algorithm to
561 * find another flex group, and store that information in the
562 * parent directory's inode information so that use that flex
563 * group for future allocations.
564 */
565 if (flex_size > 1) {
566 int retry = 0;
567
568 try_again:
569 parent_group &= ~(flex_size-1);
570 last = parent_group + flex_size;
571 if (last > ngroups)
572 last = ngroups;
573 for (i = parent_group; i < last; i++) {
574 desc = ext4_get_group_desc(sb, i, NULL);
575 if (desc && ext4_free_inodes_count(sb, desc)) {
576 *group = i;
577 return 0;
578 }
579 }
580 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
581 retry = 1;
582 parent_group = EXT4_I(parent)->i_last_alloc_group;
583 goto try_again;
584 }
585 /*
586 * If this didn't work, use the Orlov search algorithm
587 * to find a new flex group; we pass in the mode to
588 * avoid the topdir algorithms.
589 */
590 *group = parent_group + flex_size;
591 if (*group > ngroups)
592 *group = 0;
593 return find_group_orlov(sb, parent, group, mode, NULL);
594 }
595
596 /*
597 * Try to place the inode in its parent directory
598 */
599 *group = parent_group;
600 desc = ext4_get_group_desc(sb, *group, NULL);
601 if (desc && ext4_free_inodes_count(sb, desc) &&
602 ext4_free_group_clusters(sb, desc))
603 return 0;
604
605 /*
606 * We're going to place this inode in a different blockgroup from its
607 * parent. We want to cause files in a common directory to all land in
608 * the same blockgroup. But we want files which are in a different
609 * directory which shares a blockgroup with our parent to land in a
610 * different blockgroup.
611 *
612 * So add our directory's i_ino into the starting point for the hash.
613 */
614 *group = (*group + parent->i_ino) % ngroups;
615
616 /*
617 * Use a quadratic hash to find a group with a free inode and some free
618 * blocks.
619 */
620 for (i = 1; i < ngroups; i <<= 1) {
621 *group += i;
622 if (*group >= ngroups)
623 *group -= ngroups;
624 desc = ext4_get_group_desc(sb, *group, NULL);
625 if (desc && ext4_free_inodes_count(sb, desc) &&
626 ext4_free_group_clusters(sb, desc))
627 return 0;
628 }
629
630 /*
631 * That failed: try linear search for a free inode, even if that group
632 * has no free blocks.
633 */
634 *group = parent_group;
635 for (i = 0; i < ngroups; i++) {
636 if (++*group >= ngroups)
637 *group = 0;
638 desc = ext4_get_group_desc(sb, *group, NULL);
639 if (desc && ext4_free_inodes_count(sb, desc))
640 return 0;
641 }
642
643 return -1;
644 }
645
646 /*
647 * In no journal mode, if an inode has recently been deleted, we want
648 * to avoid reusing it until we're reasonably sure the inode table
649 * block has been written back to disk. (Yes, these values are
650 * somewhat arbitrary...)
651 */
652 #define RECENTCY_MIN 5
653 #define RECENTCY_DIRTY 30
654
recently_deleted(struct super_block * sb,ext4_group_t group,int ino)655 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
656 {
657 struct ext4_group_desc *gdp;
658 struct ext4_inode *raw_inode;
659 struct buffer_head *bh;
660 unsigned long dtime, now;
661 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
662 int offset, ret = 0, recentcy = RECENTCY_MIN;
663
664 gdp = ext4_get_group_desc(sb, group, NULL);
665 if (unlikely(!gdp))
666 return 0;
667
668 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
669 (ino / inodes_per_block));
670 if (unlikely(!bh) || !buffer_uptodate(bh))
671 /*
672 * If the block is not in the buffer cache, then it
673 * must have been written out.
674 */
675 goto out;
676
677 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
678 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
679 dtime = le32_to_cpu(raw_inode->i_dtime);
680 now = get_seconds();
681 if (buffer_dirty(bh))
682 recentcy += RECENTCY_DIRTY;
683
684 if (dtime && (dtime < now) && (now < dtime + recentcy))
685 ret = 1;
686 out:
687 brelse(bh);
688 return ret;
689 }
690
691 /*
692 * There are two policies for allocating an inode. If the new inode is
693 * a directory, then a forward search is made for a block group with both
694 * free space and a low directory-to-inode ratio; if that fails, then of
695 * the groups with above-average free space, that group with the fewest
696 * directories already is chosen.
697 *
698 * For other inodes, search forward from the parent directory's block
699 * group to find a free inode.
700 */
__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)701 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
702 umode_t mode, const struct qstr *qstr,
703 __u32 goal, uid_t *owner, int handle_type,
704 unsigned int line_no, int nblocks)
705 {
706 struct super_block *sb;
707 struct buffer_head *inode_bitmap_bh = NULL;
708 struct buffer_head *group_desc_bh;
709 ext4_group_t ngroups, group = 0;
710 unsigned long ino = 0;
711 struct inode *inode;
712 struct ext4_group_desc *gdp = NULL;
713 struct ext4_inode_info *ei;
714 struct ext4_sb_info *sbi;
715 int ret2, err;
716 struct inode *ret;
717 ext4_group_t i;
718 ext4_group_t flex_group;
719 struct ext4_group_info *grp;
720 int encrypt = 0;
721
722 /* Cannot create files in a deleted directory */
723 if (!dir || !dir->i_nlink)
724 return ERR_PTR(-EPERM);
725
726 if ((ext4_encrypted_inode(dir) ||
727 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
728 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
729 err = fscrypt_get_encryption_info(dir);
730 if (err)
731 return ERR_PTR(err);
732 if (!fscrypt_has_encryption_key(dir))
733 return ERR_PTR(-ENOKEY);
734 if (!handle)
735 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
736 encrypt = 1;
737 }
738
739 sb = dir->i_sb;
740 ngroups = ext4_get_groups_count(sb);
741 trace_ext4_request_inode(dir, mode);
742 inode = new_inode(sb);
743 if (!inode)
744 return ERR_PTR(-ENOMEM);
745 ei = EXT4_I(inode);
746 sbi = EXT4_SB(sb);
747
748 /*
749 * Initialize owners and quota early so that we don't have to account
750 * for quota initialization worst case in standard inode creating
751 * transaction
752 */
753 if (owner) {
754 inode->i_mode = mode;
755 i_uid_write(inode, owner[0]);
756 i_gid_write(inode, owner[1]);
757 } else if (test_opt(sb, GRPID)) {
758 inode->i_mode = mode;
759 inode->i_uid = current_fsuid();
760 inode->i_gid = dir->i_gid;
761 } else
762 inode_init_owner(inode, dir, mode);
763
764 if (ext4_has_feature_project(sb) &&
765 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
766 ei->i_projid = EXT4_I(dir)->i_projid;
767 else
768 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
769
770 err = dquot_initialize(inode);
771 if (err)
772 goto out;
773
774 if (!goal)
775 goal = sbi->s_inode_goal;
776
777 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
778 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
779 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
780 ret2 = 0;
781 goto got_group;
782 }
783
784 if (S_ISDIR(mode))
785 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
786 else
787 ret2 = find_group_other(sb, dir, &group, mode);
788
789 got_group:
790 EXT4_I(dir)->i_last_alloc_group = group;
791 err = -ENOSPC;
792 if (ret2 == -1)
793 goto out;
794
795 /*
796 * Normally we will only go through one pass of this loop,
797 * unless we get unlucky and it turns out the group we selected
798 * had its last inode grabbed by someone else.
799 */
800 for (i = 0; i < ngroups; i++, ino = 0) {
801 err = -EIO;
802
803 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
804 if (!gdp)
805 goto out;
806
807 /*
808 * Check free inodes count before loading bitmap.
809 */
810 if (ext4_free_inodes_count(sb, gdp) == 0) {
811 if (++group == ngroups)
812 group = 0;
813 continue;
814 }
815
816 grp = ext4_get_group_info(sb, group);
817 /* Skip groups with already-known suspicious inode tables */
818 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
819 if (++group == ngroups)
820 group = 0;
821 continue;
822 }
823
824 brelse(inode_bitmap_bh);
825 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
826 /* Skip groups with suspicious inode tables */
827 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
828 IS_ERR(inode_bitmap_bh)) {
829 inode_bitmap_bh = NULL;
830 if (++group == ngroups)
831 group = 0;
832 continue;
833 }
834
835 repeat_in_this_group:
836 ino = ext4_find_next_zero_bit((unsigned long *)
837 inode_bitmap_bh->b_data,
838 EXT4_INODES_PER_GROUP(sb), ino);
839 if (ino >= EXT4_INODES_PER_GROUP(sb))
840 goto next_group;
841 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
842 ext4_error(sb, "reserved inode found cleared - "
843 "inode=%lu", ino + 1);
844 continue;
845 }
846 if ((EXT4_SB(sb)->s_journal == NULL) &&
847 recently_deleted(sb, group, ino)) {
848 ino++;
849 goto next_inode;
850 }
851 if (!handle) {
852 BUG_ON(nblocks <= 0);
853 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
854 handle_type, nblocks,
855 0);
856 if (IS_ERR(handle)) {
857 err = PTR_ERR(handle);
858 ext4_std_error(sb, err);
859 goto out;
860 }
861 }
862 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
863 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
864 if (err) {
865 ext4_std_error(sb, err);
866 goto out;
867 }
868 ext4_lock_group(sb, group);
869 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
870 ext4_unlock_group(sb, group);
871 ino++; /* the inode bitmap is zero-based */
872 if (!ret2)
873 goto got; /* we grabbed the inode! */
874 next_inode:
875 if (ino < EXT4_INODES_PER_GROUP(sb))
876 goto repeat_in_this_group;
877 next_group:
878 if (++group == ngroups)
879 group = 0;
880 }
881 err = -ENOSPC;
882 goto out;
883
884 got:
885 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
886 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
887 if (err) {
888 ext4_std_error(sb, err);
889 goto out;
890 }
891
892 BUFFER_TRACE(group_desc_bh, "get_write_access");
893 err = ext4_journal_get_write_access(handle, group_desc_bh);
894 if (err) {
895 ext4_std_error(sb, err);
896 goto out;
897 }
898
899 /* We may have to initialize the block bitmap if it isn't already */
900 if (ext4_has_group_desc_csum(sb) &&
901 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
902 struct buffer_head *block_bitmap_bh;
903
904 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
905 if (IS_ERR(block_bitmap_bh)) {
906 err = PTR_ERR(block_bitmap_bh);
907 goto out;
908 }
909 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
910 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
911 if (err) {
912 brelse(block_bitmap_bh);
913 ext4_std_error(sb, err);
914 goto out;
915 }
916
917 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
918 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
919
920 /* recheck and clear flag under lock if we still need to */
921 ext4_lock_group(sb, group);
922 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
923 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
924 ext4_free_group_clusters_set(sb, gdp,
925 ext4_free_clusters_after_init(sb, group, gdp));
926 ext4_block_bitmap_csum_set(sb, group, gdp,
927 block_bitmap_bh);
928 ext4_group_desc_csum_set(sb, group, gdp);
929 }
930 ext4_unlock_group(sb, group);
931 brelse(block_bitmap_bh);
932
933 if (err) {
934 ext4_std_error(sb, err);
935 goto out;
936 }
937 }
938
939 /* Update the relevant bg descriptor fields */
940 if (ext4_has_group_desc_csum(sb)) {
941 int free;
942 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
943
944 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
945 ext4_lock_group(sb, group); /* while we modify the bg desc */
946 free = EXT4_INODES_PER_GROUP(sb) -
947 ext4_itable_unused_count(sb, gdp);
948 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
949 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
950 free = 0;
951 }
952 /*
953 * Check the relative inode number against the last used
954 * relative inode number in this group. if it is greater
955 * we need to update the bg_itable_unused count
956 */
957 if (ino > free)
958 ext4_itable_unused_set(sb, gdp,
959 (EXT4_INODES_PER_GROUP(sb) - ino));
960 up_read(&grp->alloc_sem);
961 } else {
962 ext4_lock_group(sb, group);
963 }
964
965 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
966 if (S_ISDIR(mode)) {
967 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
968 if (sbi->s_log_groups_per_flex) {
969 ext4_group_t f = ext4_flex_group(sbi, group);
970
971 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
972 }
973 }
974 if (ext4_has_group_desc_csum(sb)) {
975 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
976 EXT4_INODES_PER_GROUP(sb) / 8);
977 ext4_group_desc_csum_set(sb, group, gdp);
978 }
979 ext4_unlock_group(sb, group);
980
981 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
982 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
983 if (err) {
984 ext4_std_error(sb, err);
985 goto out;
986 }
987
988 percpu_counter_dec(&sbi->s_freeinodes_counter);
989 if (S_ISDIR(mode))
990 percpu_counter_inc(&sbi->s_dirs_counter);
991
992 if (sbi->s_log_groups_per_flex) {
993 flex_group = ext4_flex_group(sbi, group);
994 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
995 }
996
997 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
998 /* This is the optimal IO size (for stat), not the fs block size */
999 inode->i_blocks = 0;
1000 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1001 ext4_current_time(inode);
1002
1003 memset(ei->i_data, 0, sizeof(ei->i_data));
1004 ei->i_dir_start_lookup = 0;
1005 ei->i_disksize = 0;
1006
1007 /* Don't inherit extent flag from directory, amongst others. */
1008 ei->i_flags =
1009 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1010 ei->i_file_acl = 0;
1011 ei->i_dtime = 0;
1012 ei->i_block_group = group;
1013 ei->i_last_alloc_group = ~0;
1014
1015 ext4_set_inode_flags(inode);
1016 if (IS_DIRSYNC(inode))
1017 ext4_handle_sync(handle);
1018 if (insert_inode_locked(inode) < 0) {
1019 /*
1020 * Likely a bitmap corruption causing inode to be allocated
1021 * twice.
1022 */
1023 err = -EIO;
1024 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1025 inode->i_ino);
1026 goto out;
1027 }
1028 spin_lock(&sbi->s_next_gen_lock);
1029 inode->i_generation = sbi->s_next_generation++;
1030 spin_unlock(&sbi->s_next_gen_lock);
1031
1032 /* Precompute checksum seed for inode metadata */
1033 if (ext4_has_metadata_csum(sb)) {
1034 __u32 csum;
1035 __le32 inum = cpu_to_le32(inode->i_ino);
1036 __le32 gen = cpu_to_le32(inode->i_generation);
1037 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1038 sizeof(inum));
1039 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1040 sizeof(gen));
1041 }
1042
1043 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1044 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1045
1046 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1047 ei->i_inline_off = 0;
1048 if (ext4_has_feature_inline_data(sb))
1049 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1050 ret = inode;
1051 err = dquot_alloc_inode(inode);
1052 if (err)
1053 goto fail_drop;
1054
1055 /*
1056 * Since the encryption xattr will always be unique, create it first so
1057 * that it's less likely to end up in an external xattr block and
1058 * prevent its deduplication.
1059 */
1060 if (encrypt) {
1061 err = fscrypt_inherit_context(dir, inode, handle, true);
1062 if (err)
1063 goto fail_free_drop;
1064 }
1065
1066 err = ext4_init_acl(handle, inode, dir);
1067 if (err)
1068 goto fail_free_drop;
1069
1070 err = ext4_init_security(handle, inode, dir, qstr);
1071 if (err)
1072 goto fail_free_drop;
1073
1074 if (ext4_has_feature_extents(sb)) {
1075 /* set extent flag only for directory, file and normal symlink*/
1076 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1077 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1078 ext4_ext_tree_init(handle, inode);
1079 }
1080 }
1081
1082 if (ext4_handle_valid(handle)) {
1083 ei->i_sync_tid = handle->h_transaction->t_tid;
1084 ei->i_datasync_tid = handle->h_transaction->t_tid;
1085 }
1086
1087 err = ext4_mark_inode_dirty(handle, inode);
1088 if (err) {
1089 ext4_std_error(sb, err);
1090 goto fail_free_drop;
1091 }
1092
1093 ext4_debug("allocating inode %lu\n", inode->i_ino);
1094 trace_ext4_allocate_inode(inode, dir, mode);
1095 brelse(inode_bitmap_bh);
1096 return ret;
1097
1098 fail_free_drop:
1099 dquot_free_inode(inode);
1100 fail_drop:
1101 clear_nlink(inode);
1102 unlock_new_inode(inode);
1103 out:
1104 dquot_drop(inode);
1105 inode->i_flags |= S_NOQUOTA;
1106 iput(inode);
1107 brelse(inode_bitmap_bh);
1108 return ERR_PTR(err);
1109 }
1110
1111 /* Verify that we are loading a valid orphan from disk */
ext4_orphan_get(struct super_block * sb,unsigned long ino)1112 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1113 {
1114 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1115 ext4_group_t block_group;
1116 int bit;
1117 struct buffer_head *bitmap_bh = NULL;
1118 struct inode *inode = NULL;
1119 int err = -EFSCORRUPTED;
1120
1121 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1122 goto bad_orphan;
1123
1124 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1125 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1126 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1127 if (IS_ERR(bitmap_bh)) {
1128 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1129 ino, PTR_ERR(bitmap_bh));
1130 return (struct inode *) bitmap_bh;
1131 }
1132
1133 /* Having the inode bit set should be a 100% indicator that this
1134 * is a valid orphan (no e2fsck run on fs). Orphans also include
1135 * inodes that were being truncated, so we can't check i_nlink==0.
1136 */
1137 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1138 goto bad_orphan;
1139
1140 inode = ext4_iget(sb, ino);
1141 if (IS_ERR(inode)) {
1142 err = PTR_ERR(inode);
1143 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1144 ino, err);
1145 return inode;
1146 }
1147
1148 /*
1149 * If the orphans has i_nlinks > 0 then it should be able to
1150 * be truncated, otherwise it won't be removed from the orphan
1151 * list during processing and an infinite loop will result.
1152 * Similarly, it must not be a bad inode.
1153 */
1154 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1155 is_bad_inode(inode))
1156 goto bad_orphan;
1157
1158 if (NEXT_ORPHAN(inode) > max_ino)
1159 goto bad_orphan;
1160 brelse(bitmap_bh);
1161 return inode;
1162
1163 bad_orphan:
1164 ext4_error(sb, "bad orphan inode %lu", ino);
1165 if (bitmap_bh)
1166 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1167 bit, (unsigned long long)bitmap_bh->b_blocknr,
1168 ext4_test_bit(bit, bitmap_bh->b_data));
1169 if (inode) {
1170 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1171 is_bad_inode(inode));
1172 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1173 NEXT_ORPHAN(inode));
1174 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1175 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1176 /* Avoid freeing blocks if we got a bad deleted inode */
1177 if (inode->i_nlink == 0)
1178 inode->i_blocks = 0;
1179 iput(inode);
1180 }
1181 brelse(bitmap_bh);
1182 return ERR_PTR(err);
1183 }
1184
ext4_count_free_inodes(struct super_block * sb)1185 unsigned long ext4_count_free_inodes(struct super_block *sb)
1186 {
1187 unsigned long desc_count;
1188 struct ext4_group_desc *gdp;
1189 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1190 #ifdef EXT4FS_DEBUG
1191 struct ext4_super_block *es;
1192 unsigned long bitmap_count, x;
1193 struct buffer_head *bitmap_bh = NULL;
1194
1195 es = EXT4_SB(sb)->s_es;
1196 desc_count = 0;
1197 bitmap_count = 0;
1198 gdp = NULL;
1199 for (i = 0; i < ngroups; i++) {
1200 gdp = ext4_get_group_desc(sb, i, NULL);
1201 if (!gdp)
1202 continue;
1203 desc_count += ext4_free_inodes_count(sb, gdp);
1204 brelse(bitmap_bh);
1205 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1206 if (IS_ERR(bitmap_bh)) {
1207 bitmap_bh = NULL;
1208 continue;
1209 }
1210
1211 x = ext4_count_free(bitmap_bh->b_data,
1212 EXT4_INODES_PER_GROUP(sb) / 8);
1213 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1214 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1215 bitmap_count += x;
1216 }
1217 brelse(bitmap_bh);
1218 printk(KERN_DEBUG "ext4_count_free_inodes: "
1219 "stored = %u, computed = %lu, %lu\n",
1220 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1221 return desc_count;
1222 #else
1223 desc_count = 0;
1224 for (i = 0; i < ngroups; i++) {
1225 gdp = ext4_get_group_desc(sb, i, NULL);
1226 if (!gdp)
1227 continue;
1228 desc_count += ext4_free_inodes_count(sb, gdp);
1229 cond_resched();
1230 }
1231 return desc_count;
1232 #endif
1233 }
1234
1235 /* Called at mount-time, super-block is locked */
ext4_count_dirs(struct super_block * sb)1236 unsigned long ext4_count_dirs(struct super_block * sb)
1237 {
1238 unsigned long count = 0;
1239 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1240
1241 for (i = 0; i < ngroups; i++) {
1242 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1243 if (!gdp)
1244 continue;
1245 count += ext4_used_dirs_count(sb, gdp);
1246 }
1247 return count;
1248 }
1249
1250 /*
1251 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1252 * inode table. Must be called without any spinlock held. The only place
1253 * where it is called from on active part of filesystem is ext4lazyinit
1254 * thread, so we do not need any special locks, however we have to prevent
1255 * inode allocation from the current group, so we take alloc_sem lock, to
1256 * block ext4_new_inode() until we are finished.
1257 */
ext4_init_inode_table(struct super_block * sb,ext4_group_t group,int barrier)1258 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1259 int barrier)
1260 {
1261 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1262 struct ext4_sb_info *sbi = EXT4_SB(sb);
1263 struct ext4_group_desc *gdp = NULL;
1264 struct buffer_head *group_desc_bh;
1265 handle_t *handle;
1266 ext4_fsblk_t blk;
1267 int num, ret = 0, used_blks = 0;
1268
1269 /* This should not happen, but just to be sure check this */
1270 if (sb->s_flags & MS_RDONLY) {
1271 ret = 1;
1272 goto out;
1273 }
1274
1275 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1276 if (!gdp)
1277 goto out;
1278
1279 /*
1280 * We do not need to lock this, because we are the only one
1281 * handling this flag.
1282 */
1283 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1284 goto out;
1285
1286 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1287 if (IS_ERR(handle)) {
1288 ret = PTR_ERR(handle);
1289 goto out;
1290 }
1291
1292 down_write(&grp->alloc_sem);
1293 /*
1294 * If inode bitmap was already initialized there may be some
1295 * used inodes so we need to skip blocks with used inodes in
1296 * inode table.
1297 */
1298 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1299 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1300 ext4_itable_unused_count(sb, gdp)),
1301 sbi->s_inodes_per_block);
1302
1303 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1304 ext4_error(sb, "Something is wrong with group %u: "
1305 "used itable blocks: %d; "
1306 "itable unused count: %u",
1307 group, used_blks,
1308 ext4_itable_unused_count(sb, gdp));
1309 ret = 1;
1310 goto err_out;
1311 }
1312
1313 blk = ext4_inode_table(sb, gdp) + used_blks;
1314 num = sbi->s_itb_per_group - used_blks;
1315
1316 BUFFER_TRACE(group_desc_bh, "get_write_access");
1317 ret = ext4_journal_get_write_access(handle,
1318 group_desc_bh);
1319 if (ret)
1320 goto err_out;
1321
1322 /*
1323 * Skip zeroout if the inode table is full. But we set the ZEROED
1324 * flag anyway, because obviously, when it is full it does not need
1325 * further zeroing.
1326 */
1327 if (unlikely(num == 0))
1328 goto skip_zeroout;
1329
1330 ext4_debug("going to zero out inode table in group %d\n",
1331 group);
1332 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1333 if (ret < 0)
1334 goto err_out;
1335 if (barrier)
1336 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1337
1338 skip_zeroout:
1339 ext4_lock_group(sb, group);
1340 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1341 ext4_group_desc_csum_set(sb, group, gdp);
1342 ext4_unlock_group(sb, group);
1343
1344 BUFFER_TRACE(group_desc_bh,
1345 "call ext4_handle_dirty_metadata");
1346 ret = ext4_handle_dirty_metadata(handle, NULL,
1347 group_desc_bh);
1348
1349 err_out:
1350 up_write(&grp->alloc_sem);
1351 ext4_journal_stop(handle);
1352 out:
1353 return ret;
1354 }
1355