Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/ext4/ialloc.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 : * BSD ufs-inspired inode and directory allocation by
11 : * Stephen Tweedie (sct@redhat.com), 1993
12 : * Big-endian to little-endian byte-swapping/bitmaps by
13 : * David S. Miller (davem@caip.rutgers.edu), 1995
14 : */
15 :
16 : #include <linux/time.h>
17 : #include <linux/fs.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 <linux/cred.h>
26 :
27 : #include <asm/byteorder.h>
28 :
29 : #include "ext4.h"
30 : #include "ext4_jbd2.h"
31 : #include "xattr.h"
32 : #include "acl.h"
33 :
34 : #include <trace/events/ext4.h>
35 :
36 : /*
37 : * ialloc.c contains the inodes allocation and deallocation routines
38 : */
39 :
40 : /*
41 : * The free inodes are managed by bitmaps. A file system contains several
42 : * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 : * block for inodes, N blocks for the inode table and data blocks.
44 : *
45 : * The file system contains group descriptors which are located after the
46 : * super block. Each descriptor contains the number of the bitmap block and
47 : * the free blocks count in the block.
48 : */
49 :
50 : /*
51 : * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 : * need to use it within a single byte (to ensure we get endianness right).
53 : * We can use memset for the rest of the bitmap as there are no other users.
54 : */
55 0 : void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56 : {
57 0 : int i;
58 :
59 0 : if (start_bit >= end_bit)
60 : return;
61 :
62 : ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 0 : for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 0 : ext4_set_bit(i, bitmap);
65 0 : if (i < end_bit)
66 0 : memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67 : }
68 :
69 22 : void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70 : {
71 22 : if (uptodate) {
72 22 : set_buffer_uptodate(bh);
73 22 : set_bitmap_uptodate(bh);
74 : }
75 22 : unlock_buffer(bh);
76 22 : put_bh(bh);
77 22 : }
78 :
79 821 : static int ext4_validate_inode_bitmap(struct super_block *sb,
80 : struct ext4_group_desc *desc,
81 : ext4_group_t block_group,
82 : struct buffer_head *bh)
83 : {
84 821 : ext4_fsblk_t blk;
85 821 : struct ext4_group_info *grp;
86 :
87 821 : if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
88 : return 0;
89 :
90 821 : grp = ext4_get_group_info(sb, block_group);
91 :
92 821 : if (buffer_verified(bh))
93 : return 0;
94 6 : if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
95 : return -EFSCORRUPTED;
96 :
97 6 : ext4_lock_group(sb, block_group);
98 6 : if (buffer_verified(bh))
99 0 : goto verified;
100 6 : blk = ext4_inode_bitmap(sb, desc);
101 6 : if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
102 6 : EXT4_INODES_PER_GROUP(sb) / 8) ||
103 6 : ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
104 0 : ext4_unlock_group(sb, block_group);
105 0 : ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
106 : "inode_bitmap = %llu", block_group, blk);
107 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
108 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
109 0 : return -EFSBADCRC;
110 : }
111 6 : set_buffer_verified(bh);
112 6 : verified:
113 6 : ext4_unlock_group(sb, block_group);
114 6 : return 0;
115 : }
116 :
117 : /*
118 : * Read the inode allocation bitmap for a given block_group, reading
119 : * into the specified slot in the superblock's bitmap cache.
120 : *
121 : * Return buffer_head of bitmap on success, or an ERR_PTR on error.
122 : */
123 : static struct buffer_head *
124 821 : ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
125 : {
126 821 : struct ext4_group_desc *desc;
127 821 : struct ext4_sb_info *sbi = EXT4_SB(sb);
128 821 : struct buffer_head *bh = NULL;
129 821 : ext4_fsblk_t bitmap_blk;
130 821 : int err;
131 :
132 821 : desc = ext4_get_group_desc(sb, block_group, NULL);
133 821 : if (!desc)
134 821 : return ERR_PTR(-EFSCORRUPTED);
135 :
136 821 : bitmap_blk = ext4_inode_bitmap(sb, desc);
137 821 : if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
138 821 : (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
139 0 : ext4_error(sb, "Invalid inode bitmap blk %llu in "
140 : "block_group %u", bitmap_blk, block_group);
141 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
142 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
143 0 : return ERR_PTR(-EFSCORRUPTED);
144 : }
145 821 : bh = sb_getblk(sb, bitmap_blk);
146 821 : if (unlikely(!bh)) {
147 0 : ext4_warning(sb, "Cannot read inode bitmap - "
148 : "block_group = %u, inode_bitmap = %llu",
149 : block_group, bitmap_blk);
150 0 : return ERR_PTR(-ENOMEM);
151 : }
152 821 : if (bitmap_uptodate(bh))
153 815 : goto verify;
154 :
155 6 : lock_buffer(bh);
156 6 : if (bitmap_uptodate(bh)) {
157 0 : unlock_buffer(bh);
158 0 : goto verify;
159 : }
160 :
161 6 : ext4_lock_group(sb, block_group);
162 6 : if (ext4_has_group_desc_csum(sb) &&
163 6 : (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
164 0 : if (block_group == 0) {
165 0 : ext4_unlock_group(sb, block_group);
166 0 : unlock_buffer(bh);
167 0 : ext4_error(sb, "Inode bitmap for bg 0 marked "
168 : "uninitialized");
169 0 : err = -EFSCORRUPTED;
170 0 : goto out;
171 : }
172 0 : memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
173 0 : ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
174 0 : sb->s_blocksize * 8, bh->b_data);
175 0 : set_bitmap_uptodate(bh);
176 0 : set_buffer_uptodate(bh);
177 0 : set_buffer_verified(bh);
178 0 : ext4_unlock_group(sb, block_group);
179 0 : unlock_buffer(bh);
180 0 : return bh;
181 : }
182 6 : ext4_unlock_group(sb, block_group);
183 :
184 6 : if (buffer_uptodate(bh)) {
185 : /*
186 : * if not uninit if bh is uptodate,
187 : * bitmap is also uptodate
188 : */
189 0 : set_bitmap_uptodate(bh);
190 0 : unlock_buffer(bh);
191 0 : goto verify;
192 : }
193 : /*
194 : * submit the buffer_head for reading
195 : */
196 6 : trace_ext4_load_inode_bitmap(sb, block_group);
197 6 : ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
198 6 : ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
199 6 : if (!buffer_uptodate(bh)) {
200 0 : put_bh(bh);
201 0 : ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
202 : "block_group = %u, inode_bitmap = %llu",
203 : block_group, bitmap_blk);
204 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
205 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
206 0 : return ERR_PTR(-EIO);
207 : }
208 :
209 6 : verify:
210 821 : err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
211 821 : if (err)
212 0 : goto out;
213 : return bh;
214 0 : out:
215 0 : put_bh(bh);
216 0 : return ERR_PTR(err);
217 : }
218 :
219 : /*
220 : * NOTE! When we get the inode, we're the only people
221 : * that have access to it, and as such there are no
222 : * race conditions we have to worry about. The inode
223 : * is not on the hash-lists, and it cannot be reached
224 : * through the filesystem because the directory entry
225 : * has been deleted earlier.
226 : *
227 : * HOWEVER: we must make sure that we get no aliases,
228 : * which means that we have to call "clear_inode()"
229 : * _before_ we mark the inode not in use in the inode
230 : * bitmaps. Otherwise a newly created file might use
231 : * the same inode number (not actually the same pointer
232 : * though), and then we'd have two inodes sharing the
233 : * same inode number and space on the harddisk.
234 : */
235 193 : void ext4_free_inode(handle_t *handle, struct inode *inode)
236 : {
237 193 : struct super_block *sb = inode->i_sb;
238 193 : int is_directory;
239 193 : unsigned long ino;
240 193 : struct buffer_head *bitmap_bh = NULL;
241 193 : struct buffer_head *bh2;
242 193 : ext4_group_t block_group;
243 193 : unsigned long bit;
244 193 : struct ext4_group_desc *gdp;
245 193 : struct ext4_super_block *es;
246 193 : struct ext4_sb_info *sbi;
247 193 : int fatal = 0, err, count, cleared;
248 193 : struct ext4_group_info *grp;
249 :
250 193 : if (!sb) {
251 0 : printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
252 : "nonexistent device\n", __func__, __LINE__);
253 0 : return;
254 : }
255 193 : if (atomic_read(&inode->i_count) > 1) {
256 0 : ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
257 : __func__, __LINE__, inode->i_ino,
258 : atomic_read(&inode->i_count));
259 0 : return;
260 : }
261 193 : if (inode->i_nlink) {
262 0 : ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
263 : __func__, __LINE__, inode->i_ino, inode->i_nlink);
264 0 : return;
265 : }
266 193 : sbi = EXT4_SB(sb);
267 :
268 193 : ino = inode->i_ino;
269 193 : ext4_debug("freeing inode %lu\n", ino);
270 193 : trace_ext4_free_inode(inode);
271 :
272 193 : dquot_initialize(inode);
273 193 : dquot_free_inode(inode);
274 :
275 193 : is_directory = S_ISDIR(inode->i_mode);
276 :
277 : /* Do this BEFORE marking the inode not in use or returning an error */
278 193 : ext4_clear_inode(inode);
279 :
280 193 : es = sbi->s_es;
281 193 : if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
282 0 : ext4_error(sb, "reserved or nonexistent inode %lu", ino);
283 0 : goto error_return;
284 : }
285 193 : block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
286 193 : bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
287 193 : bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
288 : /* Don't bother if the inode bitmap is corrupt. */
289 193 : if (IS_ERR(bitmap_bh)) {
290 0 : fatal = PTR_ERR(bitmap_bh);
291 0 : bitmap_bh = NULL;
292 0 : goto error_return;
293 : }
294 193 : if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
295 193 : grp = ext4_get_group_info(sb, block_group);
296 193 : if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 0 : fatal = -EFSCORRUPTED;
298 0 : goto error_return;
299 : }
300 : }
301 :
302 193 : BUFFER_TRACE(bitmap_bh, "get_write_access");
303 193 : fatal = ext4_journal_get_write_access(handle, bitmap_bh);
304 193 : if (fatal)
305 0 : goto error_return;
306 :
307 193 : fatal = -ESRCH;
308 193 : gdp = ext4_get_group_desc(sb, block_group, &bh2);
309 193 : if (gdp) {
310 193 : BUFFER_TRACE(bh2, "get_write_access");
311 193 : fatal = ext4_journal_get_write_access(handle, bh2);
312 : }
313 193 : ext4_lock_group(sb, block_group);
314 193 : cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
315 193 : if (fatal || !cleared) {
316 0 : ext4_unlock_group(sb, block_group);
317 0 : goto out;
318 : }
319 :
320 193 : count = ext4_free_inodes_count(sb, gdp) + 1;
321 193 : ext4_free_inodes_set(sb, gdp, count);
322 193 : if (is_directory) {
323 71 : count = ext4_used_dirs_count(sb, gdp) - 1;
324 71 : ext4_used_dirs_set(sb, gdp, count);
325 71 : percpu_counter_dec(&sbi->s_dirs_counter);
326 : }
327 193 : ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
328 193 : EXT4_INODES_PER_GROUP(sb) / 8);
329 193 : ext4_group_desc_csum_set(sb, block_group, gdp);
330 193 : ext4_unlock_group(sb, block_group);
331 :
332 193 : percpu_counter_inc(&sbi->s_freeinodes_counter);
333 193 : if (sbi->s_log_groups_per_flex) {
334 193 : struct flex_groups *fg;
335 :
336 386 : fg = sbi_array_rcu_deref(sbi, s_flex_groups,
337 : ext4_flex_group(sbi, block_group));
338 193 : atomic_inc(&fg->free_inodes);
339 193 : if (is_directory)
340 71 : atomic_dec(&fg->used_dirs);
341 : }
342 193 : BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
343 193 : fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
344 193 : out:
345 193 : if (cleared) {
346 193 : BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
347 193 : err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
348 193 : if (!fatal)
349 193 : fatal = err;
350 : } else {
351 0 : ext4_error(sb, "bit already cleared for inode %lu", ino);
352 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
353 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
354 : }
355 :
356 193 : error_return:
357 193 : brelse(bitmap_bh);
358 193 : ext4_std_error(sb, fatal);
359 : }
360 :
361 : struct orlov_stats {
362 : __u64 free_clusters;
363 : __u32 free_inodes;
364 : __u32 used_dirs;
365 : };
366 :
367 : /*
368 : * Helper function for Orlov's allocator; returns critical information
369 : * for a particular block group or flex_bg. If flex_size is 1, then g
370 : * is a block group number; otherwise it is flex_bg number.
371 : */
372 173 : static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
373 : int flex_size, struct orlov_stats *stats)
374 : {
375 173 : struct ext4_group_desc *desc;
376 :
377 173 : if (flex_size > 1) {
378 346 : struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
379 : s_flex_groups, g);
380 173 : stats->free_inodes = atomic_read(&fg->free_inodes);
381 173 : stats->free_clusters = atomic64_read(&fg->free_clusters);
382 173 : stats->used_dirs = atomic_read(&fg->used_dirs);
383 173 : return;
384 : }
385 :
386 0 : desc = ext4_get_group_desc(sb, g, NULL);
387 0 : if (desc) {
388 0 : stats->free_inodes = ext4_free_inodes_count(sb, desc);
389 0 : stats->free_clusters = ext4_free_group_clusters(sb, desc);
390 0 : stats->used_dirs = ext4_used_dirs_count(sb, desc);
391 : } else {
392 0 : stats->free_inodes = 0;
393 0 : stats->free_clusters = 0;
394 0 : stats->used_dirs = 0;
395 : }
396 : }
397 :
398 : /*
399 : * Orlov's allocator for directories.
400 : *
401 : * We always try to spread first-level directories.
402 : *
403 : * If there are blockgroups with both free inodes and free blocks counts
404 : * not worse than average we return one with smallest directory count.
405 : * Otherwise we simply return a random group.
406 : *
407 : * For the rest rules look so:
408 : *
409 : * It's OK to put directory into a group unless
410 : * it has too many directories already (max_dirs) or
411 : * it has too few free inodes left (min_inodes) or
412 : * it has too few free blocks left (min_blocks) or
413 : * Parent's group is preferred, if it doesn't satisfy these
414 : * conditions we search cyclically through the rest. If none
415 : * of the groups look good we just look for a group with more
416 : * free inodes than average (starting at parent's group).
417 : */
418 :
419 173 : static int find_group_orlov(struct super_block *sb, struct inode *parent,
420 : ext4_group_t *group, umode_t mode,
421 : const struct qstr *qstr)
422 : {
423 173 : ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
424 173 : struct ext4_sb_info *sbi = EXT4_SB(sb);
425 173 : ext4_group_t real_ngroups = ext4_get_groups_count(sb);
426 173 : int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
427 173 : unsigned int freei, avefreei, grp_free;
428 173 : ext4_fsblk_t freeb, avefreec;
429 173 : unsigned int ndirs;
430 173 : int max_dirs, min_inodes;
431 173 : ext4_grpblk_t min_clusters;
432 173 : ext4_group_t i, grp, g, ngroups;
433 173 : struct ext4_group_desc *desc;
434 173 : struct orlov_stats stats;
435 173 : int flex_size = ext4_flex_bg_size(sbi);
436 173 : struct dx_hash_info hinfo;
437 :
438 173 : ngroups = real_ngroups;
439 173 : if (flex_size > 1) {
440 173 : ngroups = (real_ngroups + flex_size - 1) >>
441 : sbi->s_log_groups_per_flex;
442 173 : parent_group >>= sbi->s_log_groups_per_flex;
443 : }
444 :
445 173 : freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
446 173 : avefreei = freei / ngroups;
447 173 : freeb = EXT4_C2B(sbi,
448 : percpu_counter_read_positive(&sbi->s_freeclusters_counter));
449 173 : avefreec = freeb;
450 173 : do_div(avefreec, ngroups);
451 173 : ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
452 :
453 173 : if (S_ISDIR(mode) &&
454 173 : ((parent == d_inode(sb->s_root)) ||
455 173 : (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
456 0 : int best_ndir = inodes_per_group;
457 0 : int ret = -1;
458 :
459 0 : if (qstr) {
460 0 : hinfo.hash_version = DX_HASH_HALF_MD4;
461 0 : hinfo.seed = sbi->s_hash_seed;
462 0 : ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
463 0 : grp = hinfo.hash;
464 : } else
465 0 : grp = prandom_u32();
466 0 : parent_group = (unsigned)grp % ngroups;
467 0 : for (i = 0; i < ngroups; i++) {
468 0 : g = (parent_group + i) % ngroups;
469 0 : get_orlov_stats(sb, g, flex_size, &stats);
470 0 : if (!stats.free_inodes)
471 0 : continue;
472 0 : if (stats.used_dirs >= best_ndir)
473 0 : continue;
474 0 : if (stats.free_inodes < avefreei)
475 0 : continue;
476 0 : if (stats.free_clusters < avefreec)
477 0 : continue;
478 0 : grp = g;
479 0 : ret = 0;
480 0 : best_ndir = stats.used_dirs;
481 : }
482 0 : if (ret)
483 0 : goto fallback;
484 0 : found_flex_bg:
485 0 : if (flex_size == 1) {
486 0 : *group = grp;
487 0 : return 0;
488 : }
489 :
490 : /*
491 : * We pack inodes at the beginning of the flexgroup's
492 : * inode tables. Block allocation decisions will do
493 : * something similar, although regular files will
494 : * start at 2nd block group of the flexgroup. See
495 : * ext4_ext_find_goal() and ext4_find_near().
496 : */
497 0 : grp *= flex_size;
498 0 : for (i = 0; i < flex_size; i++) {
499 0 : if (grp+i >= real_ngroups)
500 : break;
501 0 : desc = ext4_get_group_desc(sb, grp+i, NULL);
502 0 : if (desc && ext4_free_inodes_count(sb, desc)) {
503 0 : *group = grp+i;
504 0 : return 0;
505 : }
506 : }
507 0 : goto fallback;
508 : }
509 :
510 173 : max_dirs = ndirs / ngroups + inodes_per_group / 16;
511 173 : min_inodes = avefreei - inodes_per_group*flex_size / 4;
512 173 : if (min_inodes < 1)
513 : min_inodes = 1;
514 173 : min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
515 :
516 : /*
517 : * Start looking in the flex group where we last allocated an
518 : * inode for this parent directory
519 : */
520 173 : if (EXT4_I(parent)->i_last_alloc_group != ~0) {
521 135 : parent_group = EXT4_I(parent)->i_last_alloc_group;
522 135 : if (flex_size > 1)
523 135 : parent_group >>= sbi->s_log_groups_per_flex;
524 : }
525 :
526 346 : for (i = 0; i < ngroups; i++) {
527 173 : grp = (parent_group + i) % ngroups;
528 173 : get_orlov_stats(sb, grp, flex_size, &stats);
529 173 : if (stats.used_dirs >= max_dirs)
530 0 : continue;
531 173 : if (stats.free_inodes < min_inodes)
532 0 : continue;
533 173 : if (stats.free_clusters < min_clusters)
534 173 : continue;
535 0 : goto found_flex_bg;
536 : }
537 :
538 173 : fallback:
539 173 : ngroups = real_ngroups;
540 173 : avefreei = freei / ngroups;
541 173 : fallback_retry:
542 173 : parent_group = EXT4_I(parent)->i_block_group;
543 241 : for (i = 0; i < ngroups; i++) {
544 241 : grp = (parent_group + i) % ngroups;
545 241 : desc = ext4_get_group_desc(sb, grp, NULL);
546 241 : if (desc) {
547 241 : grp_free = ext4_free_inodes_count(sb, desc);
548 241 : if (grp_free && grp_free >= avefreei) {
549 173 : *group = grp;
550 173 : return 0;
551 : }
552 : }
553 : }
554 :
555 0 : if (avefreei) {
556 : /*
557 : * The free-inodes counter is approximate, and for really small
558 : * filesystems the above test can fail to find any blockgroups
559 : */
560 0 : avefreei = 0;
561 0 : goto fallback_retry;
562 : }
563 :
564 : return -1;
565 : }
566 :
567 455 : static int find_group_other(struct super_block *sb, struct inode *parent,
568 : ext4_group_t *group, umode_t mode)
569 : {
570 455 : ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
571 455 : ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
572 455 : struct ext4_group_desc *desc;
573 455 : int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
574 :
575 : /*
576 : * Try to place the inode is the same flex group as its
577 : * parent. If we can't find space, use the Orlov algorithm to
578 : * find another flex group, and store that information in the
579 : * parent directory's inode information so that use that flex
580 : * group for future allocations.
581 : */
582 455 : if (flex_size > 1) {
583 : int retry = 0;
584 :
585 455 : try_again:
586 455 : parent_group &= ~(flex_size-1);
587 455 : last = parent_group + flex_size;
588 455 : if (last > ngroups)
589 : last = ngroups;
590 1650 : for (i = parent_group; i < last; i++) {
591 1650 : desc = ext4_get_group_desc(sb, i, NULL);
592 1650 : if (desc && ext4_free_inodes_count(sb, desc)) {
593 455 : *group = i;
594 455 : return 0;
595 : }
596 : }
597 0 : if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
598 0 : retry = 1;
599 0 : parent_group = EXT4_I(parent)->i_last_alloc_group;
600 0 : goto try_again;
601 : }
602 : /*
603 : * If this didn't work, use the Orlov search algorithm
604 : * to find a new flex group; we pass in the mode to
605 : * avoid the topdir algorithms.
606 : */
607 0 : *group = parent_group + flex_size;
608 0 : if (*group > ngroups)
609 0 : *group = 0;
610 0 : return find_group_orlov(sb, parent, group, mode, NULL);
611 : }
612 :
613 : /*
614 : * Try to place the inode in its parent directory
615 : */
616 0 : *group = parent_group;
617 0 : desc = ext4_get_group_desc(sb, *group, NULL);
618 0 : if (desc && ext4_free_inodes_count(sb, desc) &&
619 0 : ext4_free_group_clusters(sb, desc))
620 : return 0;
621 :
622 : /*
623 : * We're going to place this inode in a different blockgroup from its
624 : * parent. We want to cause files in a common directory to all land in
625 : * the same blockgroup. But we want files which are in a different
626 : * directory which shares a blockgroup with our parent to land in a
627 : * different blockgroup.
628 : *
629 : * So add our directory's i_ino into the starting point for the hash.
630 : */
631 0 : *group = (*group + parent->i_ino) % ngroups;
632 :
633 : /*
634 : * Use a quadratic hash to find a group with a free inode and some free
635 : * blocks.
636 : */
637 0 : for (i = 1; i < ngroups; i <<= 1) {
638 0 : *group += i;
639 0 : if (*group >= ngroups)
640 0 : *group -= ngroups;
641 0 : desc = ext4_get_group_desc(sb, *group, NULL);
642 0 : if (desc && ext4_free_inodes_count(sb, desc) &&
643 0 : ext4_free_group_clusters(sb, desc))
644 : return 0;
645 : }
646 :
647 : /*
648 : * That failed: try linear search for a free inode, even if that group
649 : * has no free blocks.
650 : */
651 0 : *group = parent_group;
652 0 : for (i = 0; i < ngroups; i++) {
653 0 : if (++*group >= ngroups)
654 0 : *group = 0;
655 0 : desc = ext4_get_group_desc(sb, *group, NULL);
656 0 : if (desc && ext4_free_inodes_count(sb, desc))
657 : return 0;
658 : }
659 :
660 : return -1;
661 : }
662 :
663 : /*
664 : * In no journal mode, if an inode has recently been deleted, we want
665 : * to avoid reusing it until we're reasonably sure the inode table
666 : * block has been written back to disk. (Yes, these values are
667 : * somewhat arbitrary...)
668 : */
669 : #define RECENTCY_MIN 60
670 : #define RECENTCY_DIRTY 300
671 :
672 0 : static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
673 : {
674 0 : struct ext4_group_desc *gdp;
675 0 : struct ext4_inode *raw_inode;
676 0 : struct buffer_head *bh;
677 0 : int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
678 0 : int offset, ret = 0;
679 0 : int recentcy = RECENTCY_MIN;
680 0 : u32 dtime, now;
681 :
682 0 : gdp = ext4_get_group_desc(sb, group, NULL);
683 0 : if (unlikely(!gdp))
684 : return 0;
685 :
686 0 : bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
687 0 : (ino / inodes_per_block));
688 0 : if (!bh || !buffer_uptodate(bh))
689 : /*
690 : * If the block is not in the buffer cache, then it
691 : * must have been written out.
692 : */
693 0 : goto out;
694 :
695 0 : offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
696 0 : raw_inode = (struct ext4_inode *) (bh->b_data + offset);
697 :
698 : /* i_dtime is only 32 bits on disk, but we only care about relative
699 : * times in the range of a few minutes (i.e. long enough to sync a
700 : * recently-deleted inode to disk), so using the low 32 bits of the
701 : * clock (a 68 year range) is enough, see time_before32() */
702 0 : dtime = le32_to_cpu(raw_inode->i_dtime);
703 0 : now = ktime_get_real_seconds();
704 0 : if (buffer_dirty(bh))
705 0 : recentcy += RECENTCY_DIRTY;
706 :
707 0 : if (dtime && time_before32(dtime, now) &&
708 0 : time_before32(now, dtime + recentcy))
709 0 : ret = 1;
710 0 : out:
711 0 : brelse(bh);
712 : return ret;
713 : }
714 :
715 628 : static int find_inode_bit(struct super_block *sb, ext4_group_t group,
716 : struct buffer_head *bitmap, unsigned long *ino)
717 : {
718 628 : bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
719 628 : unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
720 :
721 628 : next:
722 628 : *ino = ext4_find_next_zero_bit((unsigned long *)
723 628 : bitmap->b_data,
724 628 : EXT4_INODES_PER_GROUP(sb), *ino);
725 628 : if (*ino >= EXT4_INODES_PER_GROUP(sb))
726 0 : goto not_found;
727 :
728 628 : if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
729 0 : recently_deleted_ino = *ino;
730 0 : *ino = *ino + 1;
731 0 : if (*ino < EXT4_INODES_PER_GROUP(sb))
732 0 : goto next;
733 0 : goto not_found;
734 : }
735 : return 1;
736 0 : not_found:
737 0 : if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
738 : return 0;
739 : /*
740 : * Not reusing recently deleted inodes is mostly a preference. We don't
741 : * want to report ENOSPC or skew allocation patterns because of that.
742 : * So return even recently deleted inode if we could find better in the
743 : * given range.
744 : */
745 0 : *ino = recently_deleted_ino;
746 0 : return 1;
747 : }
748 :
749 0 : int ext4_mark_inode_used(struct super_block *sb, int ino)
750 : {
751 0 : unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
752 0 : struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
753 0 : struct ext4_group_desc *gdp;
754 0 : ext4_group_t group;
755 0 : int bit;
756 0 : int err = -EFSCORRUPTED;
757 :
758 0 : if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
759 0 : goto out;
760 :
761 0 : group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
762 0 : bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
763 0 : inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
764 0 : if (IS_ERR(inode_bitmap_bh))
765 0 : return PTR_ERR(inode_bitmap_bh);
766 :
767 0 : if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
768 0 : err = 0;
769 0 : goto out;
770 : }
771 :
772 0 : gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
773 0 : if (!gdp || !group_desc_bh) {
774 0 : err = -EINVAL;
775 0 : goto out;
776 : }
777 :
778 0 : ext4_set_bit(bit, inode_bitmap_bh->b_data);
779 :
780 0 : BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
781 0 : err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
782 0 : if (err) {
783 0 : ext4_std_error(sb, err);
784 0 : goto out;
785 : }
786 0 : err = sync_dirty_buffer(inode_bitmap_bh);
787 0 : if (err) {
788 0 : ext4_std_error(sb, err);
789 0 : goto out;
790 : }
791 :
792 : /* We may have to initialize the block bitmap if it isn't already */
793 0 : if (ext4_has_group_desc_csum(sb) &&
794 0 : gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
795 0 : struct buffer_head *block_bitmap_bh;
796 :
797 0 : block_bitmap_bh = ext4_read_block_bitmap(sb, group);
798 0 : if (IS_ERR(block_bitmap_bh)) {
799 0 : err = PTR_ERR(block_bitmap_bh);
800 0 : goto out;
801 : }
802 :
803 0 : BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
804 0 : err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
805 0 : sync_dirty_buffer(block_bitmap_bh);
806 :
807 : /* recheck and clear flag under lock if we still need to */
808 0 : ext4_lock_group(sb, group);
809 0 : if (ext4_has_group_desc_csum(sb) &&
810 0 : (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
811 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
812 0 : ext4_free_group_clusters_set(sb, gdp,
813 : ext4_free_clusters_after_init(sb, group, gdp));
814 0 : ext4_block_bitmap_csum_set(sb, group, gdp,
815 : block_bitmap_bh);
816 0 : ext4_group_desc_csum_set(sb, group, gdp);
817 : }
818 0 : ext4_unlock_group(sb, group);
819 0 : brelse(block_bitmap_bh);
820 :
821 0 : if (err) {
822 0 : ext4_std_error(sb, err);
823 0 : goto out;
824 : }
825 : }
826 :
827 : /* Update the relevant bg descriptor fields */
828 0 : if (ext4_has_group_desc_csum(sb)) {
829 0 : int free;
830 :
831 0 : ext4_lock_group(sb, group); /* while we modify the bg desc */
832 0 : free = EXT4_INODES_PER_GROUP(sb) -
833 0 : ext4_itable_unused_count(sb, gdp);
834 0 : if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
835 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
836 0 : free = 0;
837 : }
838 :
839 : /*
840 : * Check the relative inode number against the last used
841 : * relative inode number in this group. if it is greater
842 : * we need to update the bg_itable_unused count
843 : */
844 0 : if (bit >= free)
845 0 : ext4_itable_unused_set(sb, gdp,
846 0 : (EXT4_INODES_PER_GROUP(sb) - bit - 1));
847 : } else {
848 0 : ext4_lock_group(sb, group);
849 : }
850 :
851 0 : ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
852 0 : if (ext4_has_group_desc_csum(sb)) {
853 0 : ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
854 0 : EXT4_INODES_PER_GROUP(sb) / 8);
855 0 : ext4_group_desc_csum_set(sb, group, gdp);
856 : }
857 :
858 0 : ext4_unlock_group(sb, group);
859 0 : err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
860 0 : sync_dirty_buffer(group_desc_bh);
861 : out:
862 : return err;
863 : }
864 :
865 628 : static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
866 : bool encrypt)
867 : {
868 628 : struct super_block *sb = dir->i_sb;
869 628 : int nblocks = 0;
870 : #ifdef CONFIG_EXT4_FS_POSIX_ACL
871 : struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
872 :
873 : if (IS_ERR(p))
874 : return PTR_ERR(p);
875 : if (p) {
876 : int acl_size = p->a_count * sizeof(ext4_acl_entry);
877 :
878 : nblocks += (S_ISDIR(mode) ? 2 : 1) *
879 : __ext4_xattr_set_credits(sb, NULL /* inode */,
880 : NULL /* block_bh */, acl_size,
881 : true /* is_create */);
882 : posix_acl_release(p);
883 : }
884 : #endif
885 :
886 : #ifdef CONFIG_SECURITY
887 : {
888 628 : int num_security_xattrs = 1;
889 :
890 : #ifdef CONFIG_INTEGRITY
891 : num_security_xattrs++;
892 : #endif
893 : /*
894 : * We assume that security xattrs are never more than 1k.
895 : * In practice they are under 128 bytes.
896 : */
897 1256 : nblocks += num_security_xattrs *
898 628 : __ext4_xattr_set_credits(sb, NULL /* inode */,
899 : NULL /* block_bh */, 1024,
900 : true /* is_create */);
901 : }
902 : #endif
903 628 : if (encrypt)
904 0 : nblocks += __ext4_xattr_set_credits(sb,
905 : NULL /* inode */,
906 : NULL /* block_bh */,
907 : FSCRYPT_SET_CONTEXT_MAX_SIZE,
908 : true /* is_create */);
909 628 : return nblocks;
910 : }
911 :
912 : /*
913 : * There are two policies for allocating an inode. If the new inode is
914 : * a directory, then a forward search is made for a block group with both
915 : * free space and a low directory-to-inode ratio; if that fails, then of
916 : * the groups with above-average free space, that group with the fewest
917 : * directories already is chosen.
918 : *
919 : * For other inodes, search forward from the parent directory's block
920 : * group to find a free inode.
921 : */
922 628 : struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
923 : handle_t *handle, struct inode *dir,
924 : umode_t mode, const struct qstr *qstr,
925 : __u32 goal, uid_t *owner, __u32 i_flags,
926 : int handle_type, unsigned int line_no,
927 : int nblocks)
928 : {
929 628 : struct super_block *sb;
930 628 : struct buffer_head *inode_bitmap_bh = NULL;
931 628 : struct buffer_head *group_desc_bh;
932 628 : ext4_group_t ngroups, group = 0;
933 628 : unsigned long ino = 0;
934 628 : struct inode *inode;
935 628 : struct ext4_group_desc *gdp = NULL;
936 628 : struct ext4_inode_info *ei;
937 628 : struct ext4_sb_info *sbi;
938 628 : int ret2, err;
939 628 : struct inode *ret;
940 628 : ext4_group_t i;
941 628 : ext4_group_t flex_group;
942 628 : struct ext4_group_info *grp = NULL;
943 628 : bool encrypt = false;
944 :
945 : /* Cannot create files in a deleted directory */
946 628 : if (!dir || !dir->i_nlink)
947 628 : return ERR_PTR(-EPERM);
948 :
949 628 : sb = dir->i_sb;
950 628 : sbi = EXT4_SB(sb);
951 :
952 628 : if (unlikely(ext4_forced_shutdown(sbi)))
953 628 : return ERR_PTR(-EIO);
954 :
955 628 : ngroups = ext4_get_groups_count(sb);
956 628 : trace_ext4_request_inode(dir, mode);
957 628 : inode = new_inode(sb);
958 628 : if (!inode)
959 628 : return ERR_PTR(-ENOMEM);
960 628 : ei = EXT4_I(inode);
961 :
962 : /*
963 : * Initialize owners and quota early so that we don't have to account
964 : * for quota initialization worst case in standard inode creating
965 : * transaction
966 : */
967 628 : if (owner) {
968 0 : inode->i_mode = mode;
969 0 : i_uid_write(inode, owner[0]);
970 0 : i_gid_write(inode, owner[1]);
971 628 : } else if (test_opt(sb, GRPID)) {
972 0 : inode->i_mode = mode;
973 0 : inode->i_uid = fsuid_into_mnt(mnt_userns);
974 0 : inode->i_gid = dir->i_gid;
975 : } else
976 628 : inode_init_owner(mnt_userns, inode, dir, mode);
977 :
978 628 : if (ext4_has_feature_project(sb) &&
979 0 : ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
980 0 : ei->i_projid = EXT4_I(dir)->i_projid;
981 : else
982 628 : ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
983 :
984 628 : if (!(i_flags & EXT4_EA_INODE_FL)) {
985 628 : err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
986 0 : if (err)
987 0 : goto out;
988 : }
989 :
990 628 : err = dquot_initialize(inode);
991 628 : if (err)
992 : goto out;
993 :
994 628 : if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
995 628 : ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
996 628 : if (ret2 < 0) {
997 0 : err = ret2;
998 0 : goto out;
999 : }
1000 628 : nblocks += ret2;
1001 : }
1002 :
1003 628 : if (!goal)
1004 628 : goal = sbi->s_inode_goal;
1005 :
1006 628 : if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1007 0 : group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1008 0 : ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1009 0 : ret2 = 0;
1010 0 : goto got_group;
1011 : }
1012 :
1013 628 : if (S_ISDIR(mode))
1014 173 : ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1015 : else
1016 455 : ret2 = find_group_other(sb, dir, &group, mode);
1017 :
1018 628 : got_group:
1019 628 : EXT4_I(dir)->i_last_alloc_group = group;
1020 628 : err = -ENOSPC;
1021 628 : if (ret2 == -1)
1022 0 : goto out;
1023 :
1024 : /*
1025 : * Normally we will only go through one pass of this loop,
1026 : * unless we get unlucky and it turns out the group we selected
1027 : * had its last inode grabbed by someone else.
1028 : */
1029 628 : for (i = 0; i < ngroups; i++, ino = 0) {
1030 628 : err = -EIO;
1031 :
1032 628 : gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1033 628 : if (!gdp)
1034 0 : goto out;
1035 :
1036 : /*
1037 : * Check free inodes count before loading bitmap.
1038 : */
1039 628 : if (ext4_free_inodes_count(sb, gdp) == 0)
1040 0 : goto next_group;
1041 :
1042 628 : if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1043 628 : grp = ext4_get_group_info(sb, group);
1044 : /*
1045 : * Skip groups with already-known suspicious inode
1046 : * tables
1047 : */
1048 628 : if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1049 0 : goto next_group;
1050 : }
1051 :
1052 628 : brelse(inode_bitmap_bh);
1053 628 : inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1054 : /* Skip groups with suspicious inode tables */
1055 628 : if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1056 628 : && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1057 628 : IS_ERR(inode_bitmap_bh)) {
1058 0 : inode_bitmap_bh = NULL;
1059 0 : goto next_group;
1060 : }
1061 :
1062 628 : repeat_in_this_group:
1063 628 : ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1064 628 : if (!ret2)
1065 0 : goto next_group;
1066 :
1067 628 : if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1068 0 : ext4_error(sb, "reserved inode found cleared - "
1069 : "inode=%lu", ino + 1);
1070 0 : ext4_mark_group_bitmap_corrupted(sb, group,
1071 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1072 0 : goto next_group;
1073 : }
1074 :
1075 628 : if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1076 628 : BUG_ON(nblocks <= 0);
1077 628 : handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1078 : handle_type, nblocks, 0,
1079 : ext4_trans_default_revoke_credits(sb));
1080 628 : if (IS_ERR(handle)) {
1081 0 : err = PTR_ERR(handle);
1082 0 : ext4_std_error(sb, err);
1083 0 : goto out;
1084 : }
1085 : }
1086 628 : BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1087 628 : err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1088 628 : if (err) {
1089 0 : ext4_std_error(sb, err);
1090 0 : goto out;
1091 : }
1092 628 : ext4_lock_group(sb, group);
1093 628 : ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1094 628 : if (ret2) {
1095 : /* Someone already took the bit. Repeat the search
1096 : * with lock held.
1097 : */
1098 0 : ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1099 0 : if (ret2) {
1100 0 : ext4_set_bit(ino, inode_bitmap_bh->b_data);
1101 0 : ret2 = 0;
1102 : } else {
1103 : ret2 = 1; /* we didn't grab the inode */
1104 : }
1105 : }
1106 628 : ext4_unlock_group(sb, group);
1107 628 : ino++; /* the inode bitmap is zero-based */
1108 628 : if (!ret2)
1109 628 : goto got; /* we grabbed the inode! */
1110 :
1111 0 : if (ino < EXT4_INODES_PER_GROUP(sb))
1112 0 : goto repeat_in_this_group;
1113 0 : next_group:
1114 0 : if (++group == ngroups)
1115 0 : group = 0;
1116 : }
1117 0 : err = -ENOSPC;
1118 0 : goto out;
1119 :
1120 628 : got:
1121 628 : BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1122 628 : err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1123 628 : if (err) {
1124 0 : ext4_std_error(sb, err);
1125 0 : goto out;
1126 : }
1127 :
1128 628 : BUFFER_TRACE(group_desc_bh, "get_write_access");
1129 628 : err = ext4_journal_get_write_access(handle, group_desc_bh);
1130 628 : if (err) {
1131 0 : ext4_std_error(sb, err);
1132 0 : goto out;
1133 : }
1134 :
1135 : /* We may have to initialize the block bitmap if it isn't already */
1136 628 : if (ext4_has_group_desc_csum(sb) &&
1137 628 : gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1138 0 : struct buffer_head *block_bitmap_bh;
1139 :
1140 0 : block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1141 0 : if (IS_ERR(block_bitmap_bh)) {
1142 0 : err = PTR_ERR(block_bitmap_bh);
1143 0 : goto out;
1144 : }
1145 0 : BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1146 0 : err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1147 0 : if (err) {
1148 0 : brelse(block_bitmap_bh);
1149 0 : ext4_std_error(sb, err);
1150 0 : goto out;
1151 : }
1152 :
1153 0 : BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1154 0 : err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1155 :
1156 : /* recheck and clear flag under lock if we still need to */
1157 0 : ext4_lock_group(sb, group);
1158 0 : if (ext4_has_group_desc_csum(sb) &&
1159 0 : (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1160 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1161 0 : ext4_free_group_clusters_set(sb, gdp,
1162 : ext4_free_clusters_after_init(sb, group, gdp));
1163 0 : ext4_block_bitmap_csum_set(sb, group, gdp,
1164 : block_bitmap_bh);
1165 0 : ext4_group_desc_csum_set(sb, group, gdp);
1166 : }
1167 0 : ext4_unlock_group(sb, group);
1168 0 : brelse(block_bitmap_bh);
1169 :
1170 0 : if (err) {
1171 0 : ext4_std_error(sb, err);
1172 0 : goto out;
1173 : }
1174 : }
1175 :
1176 : /* Update the relevant bg descriptor fields */
1177 628 : if (ext4_has_group_desc_csum(sb)) {
1178 628 : int free;
1179 628 : struct ext4_group_info *grp = NULL;
1180 :
1181 628 : if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1182 628 : grp = ext4_get_group_info(sb, group);
1183 628 : down_read(&grp->alloc_sem); /*
1184 : * protect vs itable
1185 : * lazyinit
1186 : */
1187 : }
1188 628 : ext4_lock_group(sb, group); /* while we modify the bg desc */
1189 628 : free = EXT4_INODES_PER_GROUP(sb) -
1190 628 : ext4_itable_unused_count(sb, gdp);
1191 628 : if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1192 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1193 0 : free = 0;
1194 : }
1195 : /*
1196 : * Check the relative inode number against the last used
1197 : * relative inode number in this group. if it is greater
1198 : * we need to update the bg_itable_unused count
1199 : */
1200 628 : if (ino > free)
1201 0 : ext4_itable_unused_set(sb, gdp,
1202 0 : (EXT4_INODES_PER_GROUP(sb) - ino));
1203 628 : if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1204 628 : up_read(&grp->alloc_sem);
1205 : } else {
1206 0 : ext4_lock_group(sb, group);
1207 : }
1208 :
1209 628 : ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1210 628 : if (S_ISDIR(mode)) {
1211 173 : ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1212 173 : if (sbi->s_log_groups_per_flex) {
1213 173 : ext4_group_t f = ext4_flex_group(sbi, group);
1214 :
1215 346 : atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1216 : f)->used_dirs);
1217 : }
1218 : }
1219 628 : if (ext4_has_group_desc_csum(sb)) {
1220 628 : ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1221 628 : EXT4_INODES_PER_GROUP(sb) / 8);
1222 628 : ext4_group_desc_csum_set(sb, group, gdp);
1223 : }
1224 628 : ext4_unlock_group(sb, group);
1225 :
1226 628 : BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1227 628 : err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1228 628 : if (err) {
1229 0 : ext4_std_error(sb, err);
1230 0 : goto out;
1231 : }
1232 :
1233 628 : percpu_counter_dec(&sbi->s_freeinodes_counter);
1234 628 : if (S_ISDIR(mode))
1235 173 : percpu_counter_inc(&sbi->s_dirs_counter);
1236 :
1237 628 : if (sbi->s_log_groups_per_flex) {
1238 628 : flex_group = ext4_flex_group(sbi, group);
1239 1256 : atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1240 : flex_group)->free_inodes);
1241 : }
1242 :
1243 628 : inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1244 : /* This is the optimal IO size (for stat), not the fs block size */
1245 628 : inode->i_blocks = 0;
1246 628 : inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1247 628 : ei->i_crtime = inode->i_mtime;
1248 :
1249 628 : memset(ei->i_data, 0, sizeof(ei->i_data));
1250 628 : ei->i_dir_start_lookup = 0;
1251 628 : ei->i_disksize = 0;
1252 :
1253 : /* Don't inherit extent flag from directory, amongst others. */
1254 628 : ei->i_flags =
1255 628 : ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1256 628 : ei->i_flags |= i_flags;
1257 628 : ei->i_file_acl = 0;
1258 628 : ei->i_dtime = 0;
1259 628 : ei->i_block_group = group;
1260 628 : ei->i_last_alloc_group = ~0;
1261 :
1262 628 : ext4_set_inode_flags(inode, true);
1263 628 : if (IS_DIRSYNC(inode))
1264 0 : ext4_handle_sync(handle);
1265 628 : if (insert_inode_locked(inode) < 0) {
1266 : /*
1267 : * Likely a bitmap corruption causing inode to be allocated
1268 : * twice.
1269 : */
1270 0 : err = -EIO;
1271 0 : ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1272 : inode->i_ino);
1273 0 : ext4_mark_group_bitmap_corrupted(sb, group,
1274 : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1275 0 : goto out;
1276 : }
1277 628 : inode->i_generation = prandom_u32();
1278 :
1279 : /* Precompute checksum seed for inode metadata */
1280 628 : if (ext4_has_metadata_csum(sb)) {
1281 0 : __u32 csum;
1282 0 : __le32 inum = cpu_to_le32(inode->i_ino);
1283 0 : __le32 gen = cpu_to_le32(inode->i_generation);
1284 0 : csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1285 : sizeof(inum));
1286 0 : ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1287 : sizeof(gen));
1288 : }
1289 :
1290 628 : ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1291 628 : ext4_set_inode_state(inode, EXT4_STATE_NEW);
1292 :
1293 628 : ei->i_extra_isize = sbi->s_want_extra_isize;
1294 628 : ei->i_inline_off = 0;
1295 628 : if (ext4_has_feature_inline_data(sb))
1296 0 : ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1297 628 : ret = inode;
1298 628 : err = dquot_alloc_inode(inode);
1299 628 : if (err)
1300 : goto fail_drop;
1301 :
1302 : /*
1303 : * Since the encryption xattr will always be unique, create it first so
1304 : * that it's less likely to end up in an external xattr block and
1305 : * prevent its deduplication.
1306 : */
1307 628 : if (encrypt) {
1308 : err = fscrypt_set_context(inode, handle);
1309 : if (err)
1310 : goto fail_free_drop;
1311 : }
1312 :
1313 628 : if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1314 628 : err = ext4_init_acl(handle, inode, dir);
1315 : if (err)
1316 : goto fail_free_drop;
1317 :
1318 628 : err = ext4_init_security(handle, inode, dir, qstr);
1319 : if (err)
1320 : goto fail_free_drop;
1321 : }
1322 :
1323 628 : if (ext4_has_feature_extents(sb)) {
1324 : /* set extent flag only for directory, file and normal symlink*/
1325 628 : if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1326 628 : ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1327 628 : ext4_ext_tree_init(handle, inode);
1328 : }
1329 : }
1330 :
1331 628 : if (ext4_handle_valid(handle)) {
1332 628 : ei->i_sync_tid = handle->h_transaction->t_tid;
1333 628 : ei->i_datasync_tid = handle->h_transaction->t_tid;
1334 : }
1335 :
1336 628 : err = ext4_mark_inode_dirty(handle, inode);
1337 628 : if (err) {
1338 0 : ext4_std_error(sb, err);
1339 0 : goto fail_free_drop;
1340 : }
1341 :
1342 628 : ext4_debug("allocating inode %lu\n", inode->i_ino);
1343 628 : trace_ext4_allocate_inode(inode, dir, mode);
1344 628 : brelse(inode_bitmap_bh);
1345 : return ret;
1346 :
1347 0 : fail_free_drop:
1348 0 : dquot_free_inode(inode);
1349 0 : fail_drop:
1350 0 : clear_nlink(inode);
1351 0 : unlock_new_inode(inode);
1352 0 : out:
1353 0 : dquot_drop(inode);
1354 0 : inode->i_flags |= S_NOQUOTA;
1355 0 : iput(inode);
1356 0 : brelse(inode_bitmap_bh);
1357 0 : return ERR_PTR(err);
1358 : }
1359 :
1360 : /* Verify that we are loading a valid orphan from disk */
1361 0 : struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1362 : {
1363 0 : unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1364 0 : ext4_group_t block_group;
1365 0 : int bit;
1366 0 : struct buffer_head *bitmap_bh = NULL;
1367 0 : struct inode *inode = NULL;
1368 0 : int err = -EFSCORRUPTED;
1369 :
1370 0 : if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1371 0 : goto bad_orphan;
1372 :
1373 0 : block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1374 0 : bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1375 0 : bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1376 0 : if (IS_ERR(bitmap_bh))
1377 0 : return ERR_CAST(bitmap_bh);
1378 :
1379 : /* Having the inode bit set should be a 100% indicator that this
1380 : * is a valid orphan (no e2fsck run on fs). Orphans also include
1381 : * inodes that were being truncated, so we can't check i_nlink==0.
1382 : */
1383 0 : if (!ext4_test_bit(bit, bitmap_bh->b_data))
1384 0 : goto bad_orphan;
1385 :
1386 0 : inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1387 0 : if (IS_ERR(inode)) {
1388 0 : err = PTR_ERR(inode);
1389 0 : ext4_error_err(sb, -err,
1390 : "couldn't read orphan inode %lu (err %d)",
1391 : ino, err);
1392 0 : brelse(bitmap_bh);
1393 0 : return inode;
1394 : }
1395 :
1396 : /*
1397 : * If the orphans has i_nlinks > 0 then it should be able to
1398 : * be truncated, otherwise it won't be removed from the orphan
1399 : * list during processing and an infinite loop will result.
1400 : * Similarly, it must not be a bad inode.
1401 : */
1402 0 : if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1403 0 : is_bad_inode(inode))
1404 0 : goto bad_orphan;
1405 :
1406 0 : if (NEXT_ORPHAN(inode) > max_ino)
1407 0 : goto bad_orphan;
1408 0 : brelse(bitmap_bh);
1409 : return inode;
1410 :
1411 0 : bad_orphan:
1412 0 : ext4_error(sb, "bad orphan inode %lu", ino);
1413 0 : if (bitmap_bh)
1414 0 : printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1415 0 : bit, (unsigned long long)bitmap_bh->b_blocknr,
1416 0 : ext4_test_bit(bit, bitmap_bh->b_data));
1417 0 : if (inode) {
1418 0 : printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1419 0 : is_bad_inode(inode));
1420 0 : printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1421 0 : NEXT_ORPHAN(inode));
1422 0 : printk(KERN_ERR "max_ino=%lu\n", max_ino);
1423 0 : printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1424 : /* Avoid freeing blocks if we got a bad deleted inode */
1425 0 : if (inode->i_nlink == 0)
1426 0 : inode->i_blocks = 0;
1427 0 : iput(inode);
1428 : }
1429 0 : brelse(bitmap_bh);
1430 0 : return ERR_PTR(err);
1431 : }
1432 :
1433 1 : unsigned long ext4_count_free_inodes(struct super_block *sb)
1434 : {
1435 1 : unsigned long desc_count;
1436 1 : struct ext4_group_desc *gdp;
1437 1 : ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1438 : #ifdef EXT4FS_DEBUG
1439 : struct ext4_super_block *es;
1440 : unsigned long bitmap_count, x;
1441 : struct buffer_head *bitmap_bh = NULL;
1442 :
1443 : es = EXT4_SB(sb)->s_es;
1444 : desc_count = 0;
1445 : bitmap_count = 0;
1446 : gdp = NULL;
1447 : for (i = 0; i < ngroups; i++) {
1448 : gdp = ext4_get_group_desc(sb, i, NULL);
1449 : if (!gdp)
1450 : continue;
1451 : desc_count += ext4_free_inodes_count(sb, gdp);
1452 : brelse(bitmap_bh);
1453 : bitmap_bh = ext4_read_inode_bitmap(sb, i);
1454 : if (IS_ERR(bitmap_bh)) {
1455 : bitmap_bh = NULL;
1456 : continue;
1457 : }
1458 :
1459 : x = ext4_count_free(bitmap_bh->b_data,
1460 : EXT4_INODES_PER_GROUP(sb) / 8);
1461 : printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1462 : (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1463 : bitmap_count += x;
1464 : }
1465 : brelse(bitmap_bh);
1466 : printk(KERN_DEBUG "ext4_count_free_inodes: "
1467 : "stored = %u, computed = %lu, %lu\n",
1468 : le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1469 : return desc_count;
1470 : #else
1471 1 : desc_count = 0;
1472 18 : for (i = 0; i < ngroups; i++) {
1473 16 : gdp = ext4_get_group_desc(sb, i, NULL);
1474 16 : if (!gdp)
1475 0 : continue;
1476 16 : desc_count += ext4_free_inodes_count(sb, gdp);
1477 16 : cond_resched();
1478 : }
1479 1 : return desc_count;
1480 : #endif
1481 : }
1482 :
1483 : /* Called at mount-time, super-block is locked */
1484 1 : unsigned long ext4_count_dirs(struct super_block * sb)
1485 : {
1486 1 : unsigned long count = 0;
1487 1 : ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1488 :
1489 18 : for (i = 0; i < ngroups; i++) {
1490 16 : struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1491 16 : if (!gdp)
1492 0 : continue;
1493 16 : count += ext4_used_dirs_count(sb, gdp);
1494 : }
1495 1 : return count;
1496 : }
1497 :
1498 : /*
1499 : * Zeroes not yet zeroed inode table - just write zeroes through the whole
1500 : * inode table. Must be called without any spinlock held. The only place
1501 : * where it is called from on active part of filesystem is ext4lazyinit
1502 : * thread, so we do not need any special locks, however we have to prevent
1503 : * inode allocation from the current group, so we take alloc_sem lock, to
1504 : * block ext4_new_inode() until we are finished.
1505 : */
1506 0 : int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1507 : int barrier)
1508 : {
1509 0 : struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1510 0 : struct ext4_sb_info *sbi = EXT4_SB(sb);
1511 0 : struct ext4_group_desc *gdp = NULL;
1512 0 : struct buffer_head *group_desc_bh;
1513 0 : handle_t *handle;
1514 0 : ext4_fsblk_t blk;
1515 0 : int num, ret = 0, used_blks = 0;
1516 :
1517 : /* This should not happen, but just to be sure check this */
1518 0 : if (sb_rdonly(sb)) {
1519 0 : ret = 1;
1520 0 : goto out;
1521 : }
1522 :
1523 0 : gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1524 0 : if (!gdp)
1525 0 : goto out;
1526 :
1527 : /*
1528 : * We do not need to lock this, because we are the only one
1529 : * handling this flag.
1530 : */
1531 0 : if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1532 0 : goto out;
1533 :
1534 0 : handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1535 0 : if (IS_ERR(handle)) {
1536 0 : ret = PTR_ERR(handle);
1537 0 : goto out;
1538 : }
1539 :
1540 0 : down_write(&grp->alloc_sem);
1541 : /*
1542 : * If inode bitmap was already initialized there may be some
1543 : * used inodes so we need to skip blocks with used inodes in
1544 : * inode table.
1545 : */
1546 0 : if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1547 0 : used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1548 : ext4_itable_unused_count(sb, gdp)),
1549 : sbi->s_inodes_per_block);
1550 :
1551 0 : if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1552 0 : ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1553 0 : ext4_itable_unused_count(sb, gdp)) <
1554 0 : EXT4_FIRST_INO(sb)))) {
1555 0 : ext4_error(sb, "Something is wrong with group %u: "
1556 : "used itable blocks: %d; "
1557 : "itable unused count: %u",
1558 : group, used_blks,
1559 : ext4_itable_unused_count(sb, gdp));
1560 0 : ret = 1;
1561 0 : goto err_out;
1562 : }
1563 :
1564 0 : blk = ext4_inode_table(sb, gdp) + used_blks;
1565 0 : num = sbi->s_itb_per_group - used_blks;
1566 :
1567 0 : BUFFER_TRACE(group_desc_bh, "get_write_access");
1568 0 : ret = ext4_journal_get_write_access(handle,
1569 : group_desc_bh);
1570 0 : if (ret)
1571 0 : goto err_out;
1572 :
1573 : /*
1574 : * Skip zeroout if the inode table is full. But we set the ZEROED
1575 : * flag anyway, because obviously, when it is full it does not need
1576 : * further zeroing.
1577 : */
1578 0 : if (unlikely(num == 0))
1579 0 : goto skip_zeroout;
1580 :
1581 0 : ext4_debug("going to zero out inode table in group %d\n",
1582 : group);
1583 0 : ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1584 0 : if (ret < 0)
1585 0 : goto err_out;
1586 0 : if (barrier)
1587 0 : blkdev_issue_flush(sb->s_bdev);
1588 :
1589 0 : skip_zeroout:
1590 0 : ext4_lock_group(sb, group);
1591 0 : gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1592 0 : ext4_group_desc_csum_set(sb, group, gdp);
1593 0 : ext4_unlock_group(sb, group);
1594 :
1595 0 : BUFFER_TRACE(group_desc_bh,
1596 : "call ext4_handle_dirty_metadata");
1597 0 : ret = ext4_handle_dirty_metadata(handle, NULL,
1598 : group_desc_bh);
1599 :
1600 0 : err_out:
1601 0 : up_write(&grp->alloc_sem);
1602 0 : ext4_journal_stop(handle);
1603 0 : out:
1604 0 : return ret;
1605 : }
|
