Branch data 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 : : int i;
58 : :
59 [ # # ]: 0 : if (start_bit >= end_bit)
60 : 0 : 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 : : 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 : 3982 : void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
70 : : {
71 [ + - ]: 3982 : if (uptodate) {
72 : : set_buffer_uptodate(bh);
73 : : set_bitmap_uptodate(bh);
74 : : }
75 : 3982 : unlock_buffer(bh);
76 : 3982 : put_bh(bh);
77 : 3982 : }
78 : :
79 : 147557 : 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 : : ext4_fsblk_t blk;
85 : 147557 : struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
86 : :
87 [ + + ]: 147556 : if (buffer_verified(bh))
88 : : return 0;
89 [ + - ]: 1404 : if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
90 : : return -EFSCORRUPTED;
91 : :
92 : 1404 : ext4_lock_group(sb, block_group);
93 [ + - ]: 1404 : if (buffer_verified(bh))
94 : : goto verified;
95 : 1404 : blk = ext4_inode_bitmap(sb, desc);
96 [ - + ]: 1404 : if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
97 : 1404 : EXT4_INODES_PER_GROUP(sb) / 8)) {
98 : : ext4_unlock_group(sb, block_group);
99 : 0 : ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
100 : : "inode_bitmap = %llu", block_group, blk);
101 : 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
102 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
103 : 0 : return -EFSBADCRC;
104 : : }
105 : : set_buffer_verified(bh);
106 : : verified:
107 : : ext4_unlock_group(sb, block_group);
108 : 1404 : return 0;
109 : : }
110 : :
111 : : /*
112 : : * Read the inode allocation bitmap for a given block_group, reading
113 : : * into the specified slot in the superblock's bitmap cache.
114 : : *
115 : : * Return buffer_head of bitmap on success or NULL.
116 : : */
117 : : static struct buffer_head *
118 : 147555 : ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
119 : : {
120 : : struct ext4_group_desc *desc;
121 : : struct ext4_sb_info *sbi = EXT4_SB(sb);
122 : : struct buffer_head *bh = NULL;
123 : : ext4_fsblk_t bitmap_blk;
124 : : int err;
125 : :
126 : 147555 : desc = ext4_get_group_desc(sb, block_group, NULL);
127 [ + - ]: 147557 : if (!desc)
128 : : return ERR_PTR(-EFSCORRUPTED);
129 : :
130 : 147557 : bitmap_blk = ext4_inode_bitmap(sb, desc);
131 [ + - + + ]: 295113 : if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
132 : : (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
133 : 1 : ext4_error(sb, "Invalid inode bitmap blk %llu in "
134 : : "block_group %u", bitmap_blk, block_group);
135 : 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
136 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
137 : 0 : return ERR_PTR(-EFSCORRUPTED);
138 : : }
139 : : bh = sb_getblk(sb, bitmap_blk);
140 [ - + ]: 147557 : if (unlikely(!bh)) {
141 : 0 : ext4_warning(sb, "Cannot read inode bitmap - "
142 : : "block_group = %u, inode_bitmap = %llu",
143 : : block_group, bitmap_blk);
144 : 0 : return ERR_PTR(-ENOMEM);
145 : : }
146 [ + + ]: 147557 : if (bitmap_uptodate(bh))
147 : : goto verify;
148 : :
149 : 1404 : lock_buffer(bh);
150 [ - + ]: 1404 : if (bitmap_uptodate(bh)) {
151 : 0 : unlock_buffer(bh);
152 : 0 : goto verify;
153 : : }
154 : :
155 : 1404 : ext4_lock_group(sb, block_group);
156 [ - + # # ]: 1404 : if (ext4_has_group_desc_csum(sb) &&
157 : 0 : (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
158 [ # # ]: 0 : if (block_group == 0) {
159 : : ext4_unlock_group(sb, block_group);
160 : 0 : unlock_buffer(bh);
161 : 0 : ext4_error(sb, "Inode bitmap for bg 0 marked "
162 : : "uninitialized");
163 : : err = -EFSCORRUPTED;
164 : 0 : goto out;
165 : : }
166 : 0 : memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
167 : 0 : ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
168 : 0 : sb->s_blocksize * 8, bh->b_data);
169 : : set_bitmap_uptodate(bh);
170 : : set_buffer_uptodate(bh);
171 : : set_buffer_verified(bh);
172 : : ext4_unlock_group(sb, block_group);
173 : 0 : unlock_buffer(bh);
174 : 0 : return bh;
175 : : }
176 : : ext4_unlock_group(sb, block_group);
177 : :
178 [ - + ]: 1404 : if (buffer_uptodate(bh)) {
179 : : /*
180 : : * if not uninit if bh is uptodate,
181 : : * bitmap is also uptodate
182 : : */
183 : : set_bitmap_uptodate(bh);
184 : 0 : unlock_buffer(bh);
185 : 0 : goto verify;
186 : : }
187 : : /*
188 : : * submit the buffer_head for reading
189 : : */
190 : 1404 : trace_ext4_load_inode_bitmap(sb, block_group);
191 : 1404 : bh->b_end_io = ext4_end_bitmap_read;
192 : : get_bh(bh);
193 : 1404 : submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
194 : 1404 : wait_on_buffer(bh);
195 [ - + ]: 1404 : if (!buffer_uptodate(bh)) {
196 : 0 : put_bh(bh);
197 : 0 : ext4_error(sb, "Cannot read inode bitmap - "
198 : : "block_group = %u, inode_bitmap = %llu",
199 : : block_group, bitmap_blk);
200 : 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
201 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
202 : 0 : return ERR_PTR(-EIO);
203 : : }
204 : :
205 : : verify:
206 : 147557 : err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
207 [ - + ]: 147556 : if (err)
208 : : goto out;
209 : : return bh;
210 : : out:
211 : 0 : put_bh(bh);
212 : 0 : return ERR_PTR(err);
213 : : }
214 : :
215 : : /*
216 : : * NOTE! When we get the inode, we're the only people
217 : : * that have access to it, and as such there are no
218 : : * race conditions we have to worry about. The inode
219 : : * is not on the hash-lists, and it cannot be reached
220 : : * through the filesystem because the directory entry
221 : : * has been deleted earlier.
222 : : *
223 : : * HOWEVER: we must make sure that we get no aliases,
224 : : * which means that we have to call "clear_inode()"
225 : : * _before_ we mark the inode not in use in the inode
226 : : * bitmaps. Otherwise a newly created file might use
227 : : * the same inode number (not actually the same pointer
228 : : * though), and then we'd have two inodes sharing the
229 : : * same inode number and space on the harddisk.
230 : : */
231 : 3292 : void ext4_free_inode(handle_t *handle, struct inode *inode)
232 : : {
233 : 3292 : struct super_block *sb = inode->i_sb;
234 : : int is_directory;
235 : : unsigned long ino;
236 : : struct buffer_head *bitmap_bh = NULL;
237 : : struct buffer_head *bh2;
238 : : ext4_group_t block_group;
239 : : unsigned long bit;
240 : : struct ext4_group_desc *gdp;
241 : : struct ext4_super_block *es;
242 : : struct ext4_sb_info *sbi;
243 : : int fatal = 0, err, count, cleared;
244 : : struct ext4_group_info *grp;
245 : :
246 [ - + ]: 3292 : if (!sb) {
247 : 0 : printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
248 : : "nonexistent device\n", __func__, __LINE__);
249 : 0 : return;
250 : : }
251 [ - + ]: 3292 : if (atomic_read(&inode->i_count) > 1) {
252 : 0 : ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
253 : : __func__, __LINE__, inode->i_ino,
254 : : atomic_read(&inode->i_count));
255 : 0 : return;
256 : : }
257 [ - + ]: 3292 : if (inode->i_nlink) {
258 : 0 : ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
259 : : __func__, __LINE__, inode->i_ino, inode->i_nlink);
260 : 0 : return;
261 : : }
262 : : sbi = EXT4_SB(sb);
263 : :
264 : 3292 : ino = inode->i_ino;
265 : : ext4_debug("freeing inode %lu\n", ino);
266 : 3292 : trace_ext4_free_inode(inode);
267 : :
268 : 3292 : dquot_initialize(inode);
269 : 3292 : dquot_free_inode(inode);
270 : :
271 : 3291 : is_directory = S_ISDIR(inode->i_mode);
272 : :
273 : : /* Do this BEFORE marking the inode not in use or returning an error */
274 : 3291 : ext4_clear_inode(inode);
275 : :
276 : 3290 : es = sbi->s_es;
277 [ + - - + ]: 3290 : if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
278 : 0 : ext4_error(sb, "reserved or nonexistent inode %lu", ino);
279 : 0 : goto error_return;
280 : : }
281 : 3292 : block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
282 : 3292 : bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
283 : 3292 : bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
284 : : /* Don't bother if the inode bitmap is corrupt. */
285 : 3292 : grp = ext4_get_group_info(sb, block_group);
286 [ - + ]: 3290 : if (IS_ERR(bitmap_bh)) {
287 : : fatal = PTR_ERR(bitmap_bh);
288 : : bitmap_bh = NULL;
289 : 0 : goto error_return;
290 : : }
291 [ + - ]: 3290 : if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
292 : : fatal = -EFSCORRUPTED;
293 : : goto error_return;
294 : : }
295 : :
296 : : BUFFER_TRACE(bitmap_bh, "get_write_access");
297 : 3291 : fatal = ext4_journal_get_write_access(handle, bitmap_bh);
298 [ + - ]: 3292 : if (fatal)
299 : : goto error_return;
300 : :
301 : : fatal = -ESRCH;
302 : 3292 : gdp = ext4_get_group_desc(sb, block_group, &bh2);
303 [ + - ]: 3292 : if (gdp) {
304 : : BUFFER_TRACE(bh2, "get_write_access");
305 : 3292 : fatal = ext4_journal_get_write_access(handle, bh2);
306 : : }
307 : 3292 : ext4_lock_group(sb, block_group);
308 : 3291 : cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
309 [ - + ]: 3291 : if (fatal || !cleared) {
310 : : ext4_unlock_group(sb, block_group);
311 : : goto out;
312 : : }
313 : :
314 : 3291 : count = ext4_free_inodes_count(sb, gdp) + 1;
315 : 3292 : ext4_free_inodes_set(sb, gdp, count);
316 [ + + ]: 3292 : if (is_directory) {
317 : 1462 : count = ext4_used_dirs_count(sb, gdp) - 1;
318 : 1462 : ext4_used_dirs_set(sb, gdp, count);
319 : 1462 : percpu_counter_dec(&sbi->s_dirs_counter);
320 : : }
321 : 3292 : ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
322 : 3292 : EXT4_INODES_PER_GROUP(sb) / 8);
323 : 3292 : ext4_group_desc_csum_set(sb, block_group, gdp);
324 : : ext4_unlock_group(sb, block_group);
325 : :
326 : 3292 : percpu_counter_inc(&sbi->s_freeinodes_counter);
327 [ + - ]: 3291 : if (sbi->s_log_groups_per_flex) {
328 : : struct flex_groups *fg;
329 : :
330 : 6580 : fg = sbi_array_rcu_deref(sbi, s_flex_groups,
331 : : ext4_flex_group(sbi, block_group));
332 : 3291 : atomic_inc(&fg->free_inodes);
333 [ + + ]: 3292 : if (is_directory)
334 : 1460 : atomic_dec(&fg->used_dirs);
335 : : }
336 : : BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
337 : 3292 : fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
338 : : out:
339 [ + - ]: 3292 : if (cleared) {
340 : : BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
341 : 3292 : err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
342 [ + - ]: 3292 : if (!fatal)
343 : : fatal = err;
344 : : } else {
345 : 0 : ext4_error(sb, "bit already cleared for inode %lu", ino);
346 : 0 : ext4_mark_group_bitmap_corrupted(sb, block_group,
347 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
348 : : }
349 : :
350 : : error_return:
351 : : brelse(bitmap_bh);
352 [ - + ]: 3292 : ext4_std_error(sb, fatal);
353 : : }
354 : :
355 : : struct orlov_stats {
356 : : __u64 free_clusters;
357 : : __u32 free_inodes;
358 : : __u32 used_dirs;
359 : : };
360 : :
361 : : /*
362 : : * Helper function for Orlov's allocator; returns critical information
363 : : * for a particular block group or flex_bg. If flex_size is 1, then g
364 : : * is a block group number; otherwise it is flex_bg number.
365 : : */
366 : 183036 : static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
367 : : int flex_size, struct orlov_stats *stats)
368 : : {
369 : : struct ext4_group_desc *desc;
370 : :
371 [ + - ]: 183036 : if (flex_size > 1) {
372 : 183036 : struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
373 : : s_flex_groups, g);
374 : 183036 : stats->free_inodes = atomic_read(&fg->free_inodes);
375 : 366072 : stats->free_clusters = atomic64_read(&fg->free_clusters);
376 : 183036 : stats->used_dirs = atomic_read(&fg->used_dirs);
377 : 366072 : return;
378 : : }
379 : :
380 : 0 : desc = ext4_get_group_desc(sb, g, NULL);
381 [ # # ]: 0 : if (desc) {
382 : 0 : stats->free_inodes = ext4_free_inodes_count(sb, desc);
383 : 0 : stats->free_clusters = ext4_free_group_clusters(sb, desc);
384 : 0 : stats->used_dirs = ext4_used_dirs_count(sb, desc);
385 : : } else {
386 : 0 : stats->free_inodes = 0;
387 : 0 : stats->free_clusters = 0;
388 : 0 : stats->used_dirs = 0;
389 : : }
390 : : }
391 : :
392 : : /*
393 : : * Orlov's allocator for directories.
394 : : *
395 : : * We always try to spread first-level directories.
396 : : *
397 : : * If there are blockgroups with both free inodes and free blocks counts
398 : : * not worse than average we return one with smallest directory count.
399 : : * Otherwise we simply return a random group.
400 : : *
401 : : * For the rest rules look so:
402 : : *
403 : : * It's OK to put directory into a group unless
404 : : * it has too many directories already (max_dirs) or
405 : : * it has too few free inodes left (min_inodes) or
406 : : * it has too few free blocks left (min_blocks) or
407 : : * Parent's group is preferred, if it doesn't satisfy these
408 : : * conditions we search cyclically through the rest. If none
409 : : * of the groups look good we just look for a group with more
410 : : * free inodes than average (starting at parent's group).
411 : : */
412 : :
413 : 45759 : static int find_group_orlov(struct super_block *sb, struct inode *parent,
414 : : ext4_group_t *group, umode_t mode,
415 : : const struct qstr *qstr)
416 : : {
417 : 45759 : ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
418 : : struct ext4_sb_info *sbi = EXT4_SB(sb);
419 : : ext4_group_t real_ngroups = ext4_get_groups_count(sb);
420 : 45759 : int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
421 : : unsigned int freei, avefreei, grp_free;
422 : : ext4_fsblk_t freeb, avefreec;
423 : : unsigned int ndirs;
424 : : int max_dirs, min_inodes;
425 : : ext4_grpblk_t min_clusters;
426 : : ext4_group_t i, grp, g, ngroups;
427 : : struct ext4_group_desc *desc;
428 : : struct orlov_stats stats;
429 : : int flex_size = ext4_flex_bg_size(sbi);
430 : : struct dx_hash_info hinfo;
431 : :
432 : : ngroups = real_ngroups;
433 [ + - ]: 45759 : if (flex_size > 1) {
434 : 45759 : ngroups = (real_ngroups + flex_size - 1) >>
435 : : sbi->s_log_groups_per_flex;
436 : 45759 : parent_group >>= sbi->s_log_groups_per_flex;
437 : : }
438 : :
439 : 45759 : freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
440 : 45759 : avefreei = freei / ngroups;
441 : 45759 : freeb = EXT4_C2B(sbi,
442 : : percpu_counter_read_positive(&sbi->s_freeclusters_counter));
443 : 45759 : avefreec = freeb;
444 [ - + # # : 45759 : do_div(avefreec, ngroups);
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # - +
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
+ - ]
445 : 45759 : ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
446 : :
447 [ + - + - ]: 91518 : if (S_ISDIR(mode) &&
448 [ - + ]: 91518 : ((parent == d_inode(sb->s_root)) ||
449 : : (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
450 : : int best_ndir = inodes_per_group;
451 : : int ret = -1;
452 : :
453 [ # # ]: 0 : if (qstr) {
454 : 0 : hinfo.hash_version = DX_HASH_HALF_MD4;
455 : 0 : hinfo.seed = sbi->s_hash_seed;
456 : 0 : ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
457 : 0 : grp = hinfo.hash;
458 : : } else
459 : 0 : grp = prandom_u32();
460 : 0 : parent_group = (unsigned)grp % ngroups;
461 [ # # ]: 0 : for (i = 0; i < ngroups; i++) {
462 : 0 : g = (parent_group + i) % ngroups;
463 : 0 : get_orlov_stats(sb, g, flex_size, &stats);
464 [ # # ]: 0 : if (!stats.free_inodes)
465 : 0 : continue;
466 [ # # ]: 0 : if (stats.used_dirs >= best_ndir)
467 : 0 : continue;
468 [ # # ]: 0 : if (stats.free_inodes < avefreei)
469 : 0 : continue;
470 [ # # ]: 0 : if (stats.free_clusters < avefreec)
471 : 0 : continue;
472 : : grp = g;
473 : : ret = 0;
474 : 0 : best_ndir = stats.used_dirs;
475 : : }
476 [ # # ]: 0 : if (ret)
477 : : goto fallback;
478 : : found_flex_bg:
479 [ # # ]: 0 : if (flex_size == 1) {
480 : 0 : *group = grp;
481 : 0 : return 0;
482 : : }
483 : :
484 : : /*
485 : : * We pack inodes at the beginning of the flexgroup's
486 : : * inode tables. Block allocation decisions will do
487 : : * something similar, although regular files will
488 : : * start at 2nd block group of the flexgroup. See
489 : : * ext4_ext_find_goal() and ext4_find_near().
490 : : */
491 : 0 : grp *= flex_size;
492 [ # # ]: 0 : for (i = 0; i < flex_size; i++) {
493 [ # # ]: 0 : if (grp+i >= real_ngroups)
494 : : break;
495 : 0 : desc = ext4_get_group_desc(sb, grp+i, NULL);
496 [ # # # # ]: 0 : if (desc && ext4_free_inodes_count(sb, desc)) {
497 : 0 : *group = grp+i;
498 : 0 : return 0;
499 : : }
500 : : }
501 : : goto fallback;
502 : : }
503 : :
504 : 45759 : max_dirs = ndirs / ngroups + inodes_per_group / 16;
505 : 45759 : min_inodes = avefreei - inodes_per_group*flex_size / 4;
506 [ - + ]: 45759 : if (min_inodes < 1)
507 : : min_inodes = 1;
508 : 91518 : min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
509 : :
510 : : /*
511 : : * Start looking in the flex group where we last allocated an
512 : : * inode for this parent directory
513 : : */
514 [ + + ]: 45759 : if (EXT4_I(parent)->i_last_alloc_group != ~0) {
515 : : parent_group = EXT4_I(parent)->i_last_alloc_group;
516 [ + - ]: 33333 : if (flex_size > 1)
517 : 33333 : parent_group >>= sbi->s_log_groups_per_flex;
518 : : }
519 : :
520 [ + + ]: 228795 : for (i = 0; i < ngroups; i++) {
521 : 183036 : grp = (parent_group + i) % ngroups;
522 : 183036 : get_orlov_stats(sb, grp, flex_size, &stats);
523 [ + + ]: 183036 : if (stats.used_dirs >= max_dirs)
524 : 45759 : continue;
525 [ - + ]: 137277 : if (stats.free_inodes < min_inodes)
526 : 0 : continue;
527 [ - + ]: 137277 : if (stats.free_clusters < min_clusters)
528 : 137277 : continue;
529 : : goto found_flex_bg;
530 : : }
531 : :
532 : : fallback:
533 : : ngroups = real_ngroups;
534 : 45759 : avefreei = freei / ngroups;
535 : : fallback_retry:
536 : 45759 : parent_group = EXT4_I(parent)->i_block_group;
537 [ + - ]: 54264 : for (i = 0; i < ngroups; i++) {
538 : 54264 : grp = (parent_group + i) % ngroups;
539 : 54264 : desc = ext4_get_group_desc(sb, grp, NULL);
540 [ + - ]: 54264 : if (desc) {
541 : 54264 : grp_free = ext4_free_inodes_count(sb, desc);
542 [ + + ]: 54264 : if (grp_free && grp_free >= avefreei) {
543 : 45759 : *group = grp;
544 : 45759 : return 0;
545 : : }
546 : : }
547 : : }
548 : :
549 [ # # ]: 0 : if (avefreei) {
550 : : /*
551 : : * The free-inodes counter is approximate, and for really small
552 : : * filesystems the above test can fail to find any blockgroups
553 : : */
554 : : avefreei = 0;
555 : : goto fallback_retry;
556 : : }
557 : :
558 : : return -1;
559 : : }
560 : :
561 : 98506 : static int find_group_other(struct super_block *sb, struct inode *parent,
562 : : ext4_group_t *group, umode_t mode)
563 : : {
564 : 98506 : ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
565 : : ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
566 : : struct ext4_group_desc *desc;
567 : : int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
568 : :
569 : : /*
570 : : * Try to place the inode is the same flex group as its
571 : : * parent. If we can't find space, use the Orlov algorithm to
572 : : * find another flex group, and store that information in the
573 : : * parent directory's inode information so that use that flex
574 : : * group for future allocations.
575 : : */
576 [ + - ]: 98506 : if (flex_size > 1) {
577 : : int retry = 0;
578 : :
579 : : try_again:
580 : 98506 : parent_group &= ~(flex_size-1);
581 : 98506 : last = parent_group + flex_size;
582 [ - + ]: 98506 : if (last > ngroups)
583 : : last = ngroups;
584 [ + - ]: 360224 : for (i = parent_group; i < last; i++) {
585 : 360224 : desc = ext4_get_group_desc(sb, i, NULL);
586 [ + - + + ]: 360224 : if (desc && ext4_free_inodes_count(sb, desc)) {
587 : 98506 : *group = i;
588 : 98506 : return 0;
589 : : }
590 : : }
591 [ # # # # ]: 0 : if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
592 : : retry = 1;
593 : : parent_group = EXT4_I(parent)->i_last_alloc_group;
594 : : goto try_again;
595 : : }
596 : : /*
597 : : * If this didn't work, use the Orlov search algorithm
598 : : * to find a new flex group; we pass in the mode to
599 : : * avoid the topdir algorithms.
600 : : */
601 : 0 : *group = parent_group + flex_size;
602 [ # # ]: 0 : if (*group > ngroups)
603 : 0 : *group = 0;
604 : 0 : return find_group_orlov(sb, parent, group, mode, NULL);
605 : : }
606 : :
607 : : /*
608 : : * Try to place the inode in its parent directory
609 : : */
610 : 0 : *group = parent_group;
611 : 0 : desc = ext4_get_group_desc(sb, *group, NULL);
612 [ # # # # : 0 : if (desc && ext4_free_inodes_count(sb, desc) &&
# # ]
613 : 0 : ext4_free_group_clusters(sb, desc))
614 : : return 0;
615 : :
616 : : /*
617 : : * We're going to place this inode in a different blockgroup from its
618 : : * parent. We want to cause files in a common directory to all land in
619 : : * the same blockgroup. But we want files which are in a different
620 : : * directory which shares a blockgroup with our parent to land in a
621 : : * different blockgroup.
622 : : *
623 : : * So add our directory's i_ino into the starting point for the hash.
624 : : */
625 : 0 : *group = (*group + parent->i_ino) % ngroups;
626 : :
627 : : /*
628 : : * Use a quadratic hash to find a group with a free inode and some free
629 : : * blocks.
630 : : */
631 [ # # ]: 0 : for (i = 1; i < ngroups; i <<= 1) {
632 : 0 : *group += i;
633 [ # # ]: 0 : if (*group >= ngroups)
634 : 0 : *group -= ngroups;
635 : 0 : desc = ext4_get_group_desc(sb, *group, NULL);
636 [ # # # # : 0 : if (desc && ext4_free_inodes_count(sb, desc) &&
# # ]
637 : 0 : ext4_free_group_clusters(sb, desc))
638 : : return 0;
639 : : }
640 : :
641 : : /*
642 : : * That failed: try linear search for a free inode, even if that group
643 : : * has no free blocks.
644 : : */
645 : 0 : *group = parent_group;
646 [ # # ]: 0 : for (i = 0; i < ngroups; i++) {
647 [ # # ]: 0 : if (++*group >= ngroups)
648 : 0 : *group = 0;
649 : 0 : desc = ext4_get_group_desc(sb, *group, NULL);
650 [ # # # # ]: 0 : if (desc && ext4_free_inodes_count(sb, desc))
651 : : return 0;
652 : : }
653 : :
654 : : return -1;
655 : : }
656 : :
657 : : /*
658 : : * In no journal mode, if an inode has recently been deleted, we want
659 : : * to avoid reusing it until we're reasonably sure the inode table
660 : : * block has been written back to disk. (Yes, these values are
661 : : * somewhat arbitrary...)
662 : : */
663 : : #define RECENTCY_MIN 60
664 : : #define RECENTCY_DIRTY 300
665 : :
666 : 0 : static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
667 : : {
668 : : struct ext4_group_desc *gdp;
669 : : struct ext4_inode *raw_inode;
670 : : struct buffer_head *bh;
671 : 0 : int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
672 : : int offset, ret = 0;
673 : : int recentcy = RECENTCY_MIN;
674 : : u32 dtime, now;
675 : :
676 : 0 : gdp = ext4_get_group_desc(sb, group, NULL);
677 [ # # ]: 0 : if (unlikely(!gdp))
678 : : return 0;
679 : :
680 : 0 : bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
681 : 0 : (ino / inodes_per_block));
682 [ # # # # ]: 0 : if (!bh || !buffer_uptodate(bh))
683 : : /*
684 : : * If the block is not in the buffer cache, then it
685 : : * must have been written out.
686 : : */
687 : : goto out;
688 : :
689 : 0 : offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
690 : 0 : raw_inode = (struct ext4_inode *) (bh->b_data + offset);
691 : :
692 : : /* i_dtime is only 32 bits on disk, but we only care about relative
693 : : * times in the range of a few minutes (i.e. long enough to sync a
694 : : * recently-deleted inode to disk), so using the low 32 bits of the
695 : : * clock (a 68 year range) is enough, see time_before32() */
696 : 0 : dtime = le32_to_cpu(raw_inode->i_dtime);
697 : 0 : now = ktime_get_real_seconds();
698 [ # # ]: 0 : if (buffer_dirty(bh))
699 : : recentcy += RECENTCY_DIRTY;
700 : :
701 [ # # # # : 0 : if (dtime && time_before32(dtime, now) &&
# # ]
702 : 0 : time_before32(now, dtime + recentcy))
703 : : ret = 1;
704 : : out:
705 : : brelse(bh);
706 : 0 : return ret;
707 : : }
708 : :
709 : 144265 : static int find_inode_bit(struct super_block *sb, ext4_group_t group,
710 : : struct buffer_head *bitmap, unsigned long *ino)
711 : : {
712 : : next:
713 : 288530 : *ino = ext4_find_next_zero_bit((unsigned long *)
714 : 144265 : bitmap->b_data,
715 : : EXT4_INODES_PER_GROUP(sb), *ino);
716 [ + - ]: 144265 : if (*ino >= EXT4_INODES_PER_GROUP(sb))
717 : : return 0;
718 : :
719 [ - + # # ]: 144265 : if ((EXT4_SB(sb)->s_journal == NULL) &&
720 : 0 : recently_deleted(sb, group, *ino)) {
721 : 0 : *ino = *ino + 1;
722 [ # # ]: 0 : if (*ino < EXT4_INODES_PER_GROUP(sb))
723 : : goto next;
724 : : return 0;
725 : : }
726 : :
727 : : return 1;
728 : : }
729 : :
730 : : /*
731 : : * There are two policies for allocating an inode. If the new inode is
732 : : * a directory, then a forward search is made for a block group with both
733 : : * free space and a low directory-to-inode ratio; if that fails, then of
734 : : * the groups with above-average free space, that group with the fewest
735 : : * directories already is chosen.
736 : : *
737 : : * For other inodes, search forward from the parent directory's block
738 : : * group to find a free inode.
739 : : */
740 : 144265 : struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
741 : : umode_t mode, const struct qstr *qstr,
742 : : __u32 goal, uid_t *owner, __u32 i_flags,
743 : : int handle_type, unsigned int line_no,
744 : : int nblocks)
745 : : {
746 : : struct super_block *sb;
747 : : struct buffer_head *inode_bitmap_bh = NULL;
748 : : struct buffer_head *group_desc_bh;
749 : 144265 : ext4_group_t ngroups, group = 0;
750 : 144265 : unsigned long ino = 0;
751 : : struct inode *inode;
752 : : struct ext4_group_desc *gdp = NULL;
753 : : struct ext4_inode_info *ei;
754 : : struct ext4_sb_info *sbi;
755 : : int ret2, err;
756 : : struct inode *ret;
757 : : ext4_group_t i;
758 : : ext4_group_t flex_group;
759 : : struct ext4_group_info *grp;
760 : : int encrypt = 0;
761 : :
762 : : /* Cannot create files in a deleted directory */
763 [ + - + - ]: 144265 : if (!dir || !dir->i_nlink)
764 : : return ERR_PTR(-EPERM);
765 : :
766 : 144265 : sb = dir->i_sb;
767 : : sbi = EXT4_SB(sb);
768 : :
769 [ + - ]: 144265 : if (unlikely(ext4_forced_shutdown(sbi)))
770 : : return ERR_PTR(-EIO);
771 : :
772 [ + - - + : 144265 : if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
# # ]
773 [ # # # # ]: 0 : (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
774 : 0 : !(i_flags & EXT4_EA_INODE_FL)) {
775 : 0 : err = fscrypt_get_encryption_info(dir);
776 [ # # ]: 0 : if (err)
777 : 0 : return ERR_PTR(err);
778 [ # # ]: 0 : if (!fscrypt_has_encryption_key(dir))
779 : : return ERR_PTR(-ENOKEY);
780 : : encrypt = 1;
781 : : }
782 : :
783 [ + - + - : 144265 : if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
+ - ]
784 : : #ifdef CONFIG_EXT4_FS_POSIX_ACL
785 : 144265 : struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
786 : :
787 [ + - ]: 144265 : if (IS_ERR(p))
788 : : return ERR_CAST(p);
789 [ - + ]: 144265 : if (p) {
790 : 0 : int acl_size = p->a_count * sizeof(ext4_acl_entry);
791 : :
792 [ # # ]: 0 : nblocks += (S_ISDIR(mode) ? 2 : 1) *
793 : 0 : __ext4_xattr_set_credits(sb, NULL /* inode */,
794 : : NULL /* block_bh */, acl_size,
795 : : true /* is_create */);
796 : 0 : posix_acl_release(p);
797 : : }
798 : : #endif
799 : :
800 : : #ifdef CONFIG_SECURITY
801 : : {
802 : : int num_security_xattrs = 1;
803 : :
804 : : #ifdef CONFIG_INTEGRITY
805 : : num_security_xattrs++;
806 : : #endif
807 : : /*
808 : : * We assume that security xattrs are never
809 : : * more than 1k. In practice they are under
810 : : * 128 bytes.
811 : : */
812 : 144265 : nblocks += num_security_xattrs *
813 : 144265 : __ext4_xattr_set_credits(sb, NULL /* inode */,
814 : : NULL /* block_bh */, 1024,
815 : : true /* is_create */);
816 : : }
817 : : #endif
818 [ - + ]: 144265 : if (encrypt)
819 : 0 : nblocks += __ext4_xattr_set_credits(sb,
820 : : NULL /* inode */, NULL /* block_bh */,
821 : : FSCRYPT_SET_CONTEXT_MAX_SIZE,
822 : : true /* is_create */);
823 : : }
824 : :
825 : : ngroups = ext4_get_groups_count(sb);
826 : 144265 : trace_ext4_request_inode(dir, mode);
827 : 144265 : inode = new_inode(sb);
828 [ + - ]: 144265 : if (!inode)
829 : : return ERR_PTR(-ENOMEM);
830 : 144265 : ei = EXT4_I(inode);
831 : :
832 : : /*
833 : : * Initialize owners and quota early so that we don't have to account
834 : : * for quota initialization worst case in standard inode creating
835 : : * transaction
836 : : */
837 [ - + ]: 144265 : if (owner) {
838 : 0 : inode->i_mode = mode;
839 : 0 : i_uid_write(inode, owner[0]);
840 : 0 : i_gid_write(inode, owner[1]);
841 [ - + ]: 144265 : } else if (test_opt(sb, GRPID)) {
842 : 0 : inode->i_mode = mode;
843 : 0 : inode->i_uid = current_fsuid();
844 : 0 : inode->i_gid = dir->i_gid;
845 : : } else
846 : 144265 : inode_init_owner(inode, dir, mode);
847 : :
848 [ - + # # ]: 144265 : if (ext4_has_feature_project(sb) &&
849 : : ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
850 : 0 : ei->i_projid = EXT4_I(dir)->i_projid;
851 : : else
852 : 144265 : ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
853 : :
854 : 144265 : err = dquot_initialize(inode);
855 [ + - ]: 144265 : if (err)
856 : : goto out;
857 : :
858 [ + - ]: 144265 : if (!goal)
859 : 144265 : goal = sbi->s_inode_goal;
860 : :
861 [ - + # # ]: 144265 : if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
862 : 0 : group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
863 : 0 : ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
864 : : ret2 = 0;
865 : 0 : goto got_group;
866 : : }
867 : :
868 [ + + ]: 144265 : if (S_ISDIR(mode))
869 : 45759 : ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
870 : : else
871 : 98506 : ret2 = find_group_other(sb, dir, &group, mode);
872 : :
873 : : got_group:
874 : 144265 : EXT4_I(dir)->i_last_alloc_group = group;
875 : : err = -ENOSPC;
876 [ + - ]: 144265 : if (ret2 == -1)
877 : : goto out;
878 : :
879 : : /*
880 : : * Normally we will only go through one pass of this loop,
881 : : * unless we get unlucky and it turns out the group we selected
882 : : * had its last inode grabbed by someone else.
883 : : */
884 [ + - ]: 0 : for (i = 0; i < ngroups; i++, ino = 0) {
885 : : err = -EIO;
886 : :
887 : 144265 : gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
888 [ + - ]: 144265 : if (!gdp)
889 : : goto out;
890 : :
891 : : /*
892 : : * Check free inodes count before loading bitmap.
893 : : */
894 [ + - ]: 144265 : if (ext4_free_inodes_count(sb, gdp) == 0)
895 : : goto next_group;
896 : :
897 : 144265 : grp = ext4_get_group_info(sb, group);
898 : : /* Skip groups with already-known suspicious inode tables */
899 [ + - ]: 144265 : if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
900 : : goto next_group;
901 : :
902 : : brelse(inode_bitmap_bh);
903 : 144265 : inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
904 : : /* Skip groups with suspicious inode tables */
905 [ + - + - ]: 288530 : if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
906 : : IS_ERR(inode_bitmap_bh)) {
907 : : inode_bitmap_bh = NULL;
908 : : goto next_group;
909 : : }
910 : :
911 : : repeat_in_this_group:
912 : 144265 : ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
913 [ + - ]: 144265 : if (!ret2)
914 : : goto next_group;
915 : :
916 [ - + # # ]: 144265 : if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
917 : 0 : ext4_error(sb, "reserved inode found cleared - "
918 : : "inode=%lu", ino + 1);
919 : 0 : ext4_mark_group_bitmap_corrupted(sb, group,
920 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
921 : 0 : goto next_group;
922 : : }
923 : :
924 [ + - ]: 144265 : if (!handle) {
925 [ - + ]: 144265 : BUG_ON(nblocks <= 0);
926 : 144265 : handle = __ext4_journal_start_sb(dir->i_sb, line_no,
927 : : handle_type, nblocks,
928 : : 0);
929 [ - + ]: 144265 : if (IS_ERR(handle)) {
930 : : err = PTR_ERR(handle);
931 [ # # ]: 0 : ext4_std_error(sb, err);
932 : : goto out;
933 : : }
934 : : }
935 : : BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
936 : 144265 : err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
937 [ - + ]: 144265 : if (err) {
938 [ # # ]: 0 : ext4_std_error(sb, err);
939 : : goto out;
940 : : }
941 : 144265 : ext4_lock_group(sb, group);
942 : 144265 : ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
943 [ - + ]: 144265 : if (ret2) {
944 : : /* Someone already took the bit. Repeat the search
945 : : * with lock held.
946 : : */
947 : 0 : ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
948 [ # # ]: 0 : if (ret2) {
949 : 0 : ext4_set_bit(ino, inode_bitmap_bh->b_data);
950 : : ret2 = 0;
951 : : } else {
952 : : ret2 = 1; /* we didn't grab the inode */
953 : : }
954 : : }
955 : 144265 : ext4_unlock_group(sb, group);
956 : 144265 : ino++; /* the inode bitmap is zero-based */
957 [ - + ]: 144265 : if (!ret2)
958 : : goto got; /* we grabbed the inode! */
959 : :
960 [ # # ]: 0 : if (ino < EXT4_INODES_PER_GROUP(sb))
961 : : goto repeat_in_this_group;
962 : : next_group:
963 [ # # ]: 0 : if (++group == ngroups)
964 : 0 : group = 0;
965 : : }
966 : : err = -ENOSPC;
967 : : goto out;
968 : :
969 : : got:
970 : : BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
971 : 144265 : err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
972 [ - + ]: 144265 : if (err) {
973 [ # # ]: 0 : ext4_std_error(sb, err);
974 : : goto out;
975 : : }
976 : :
977 : : BUFFER_TRACE(group_desc_bh, "get_write_access");
978 : 144265 : err = ext4_journal_get_write_access(handle, group_desc_bh);
979 [ - + ]: 144265 : if (err) {
980 [ # # ]: 0 : ext4_std_error(sb, err);
981 : : goto out;
982 : : }
983 : :
984 : : /* We may have to initialize the block bitmap if it isn't already */
985 [ - + # # ]: 144265 : if (ext4_has_group_desc_csum(sb) &&
986 : 0 : gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
987 : : struct buffer_head *block_bitmap_bh;
988 : :
989 : 0 : block_bitmap_bh = ext4_read_block_bitmap(sb, group);
990 [ # # ]: 0 : if (IS_ERR(block_bitmap_bh)) {
991 : : err = PTR_ERR(block_bitmap_bh);
992 : 0 : goto out;
993 : : }
994 : : BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
995 : 0 : err = ext4_journal_get_write_access(handle, block_bitmap_bh);
996 [ # # ]: 0 : if (err) {
997 : : brelse(block_bitmap_bh);
998 [ # # ]: 0 : ext4_std_error(sb, err);
999 : : goto out;
1000 : : }
1001 : :
1002 : : BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1003 : 0 : err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1004 : :
1005 : : /* recheck and clear flag under lock if we still need to */
1006 : 0 : ext4_lock_group(sb, group);
1007 [ # # # # ]: 0 : if (ext4_has_group_desc_csum(sb) &&
1008 : 0 : (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1009 : 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1010 : 0 : ext4_free_group_clusters_set(sb, gdp,
1011 : : ext4_free_clusters_after_init(sb, group, gdp));
1012 : 0 : ext4_block_bitmap_csum_set(sb, group, gdp,
1013 : : block_bitmap_bh);
1014 : 0 : ext4_group_desc_csum_set(sb, group, gdp);
1015 : : }
1016 : 0 : ext4_unlock_group(sb, group);
1017 : : brelse(block_bitmap_bh);
1018 : :
1019 [ # # ]: 0 : if (err) {
1020 [ # # ]: 0 : ext4_std_error(sb, err);
1021 : : goto out;
1022 : : }
1023 : : }
1024 : :
1025 : : /* Update the relevant bg descriptor fields */
1026 [ - + ]: 144265 : if (ext4_has_group_desc_csum(sb)) {
1027 : : int free;
1028 : 0 : struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1029 : :
1030 : 0 : down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1031 : 0 : ext4_lock_group(sb, group); /* while we modify the bg desc */
1032 : 0 : free = EXT4_INODES_PER_GROUP(sb) -
1033 : 0 : ext4_itable_unused_count(sb, gdp);
1034 [ # # ]: 0 : if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1035 : 0 : gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1036 : : free = 0;
1037 : : }
1038 : : /*
1039 : : * Check the relative inode number against the last used
1040 : : * relative inode number in this group. if it is greater
1041 : : * we need to update the bg_itable_unused count
1042 : : */
1043 [ # # ]: 0 : if (ino > free)
1044 : 0 : ext4_itable_unused_set(sb, gdp,
1045 : 0 : (EXT4_INODES_PER_GROUP(sb) - ino));
1046 : 0 : up_read(&grp->alloc_sem);
1047 : : } else {
1048 : 144265 : ext4_lock_group(sb, group);
1049 : : }
1050 : :
1051 : 144265 : ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1052 [ + + ]: 144265 : if (S_ISDIR(mode)) {
1053 : 45759 : ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1054 [ + - ]: 45759 : if (sbi->s_log_groups_per_flex) {
1055 : 45759 : ext4_group_t f = ext4_flex_group(sbi, group);
1056 : :
1057 : 91518 : atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1058 : : f)->used_dirs);
1059 : : }
1060 : : }
1061 [ - + ]: 144265 : if (ext4_has_group_desc_csum(sb)) {
1062 : 0 : ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1063 : 0 : EXT4_INODES_PER_GROUP(sb) / 8);
1064 : 0 : ext4_group_desc_csum_set(sb, group, gdp);
1065 : : }
1066 : 144265 : ext4_unlock_group(sb, group);
1067 : :
1068 : : BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1069 : 144265 : err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1070 [ - + ]: 144265 : if (err) {
1071 [ # # ]: 0 : ext4_std_error(sb, err);
1072 : : goto out;
1073 : : }
1074 : :
1075 : 144265 : percpu_counter_dec(&sbi->s_freeinodes_counter);
1076 [ + + ]: 144265 : if (S_ISDIR(mode))
1077 : 45759 : percpu_counter_inc(&sbi->s_dirs_counter);
1078 : :
1079 [ + - ]: 144265 : if (sbi->s_log_groups_per_flex) {
1080 : 144265 : flex_group = ext4_flex_group(sbi, group);
1081 : 288530 : atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1082 : : flex_group)->free_inodes);
1083 : : }
1084 : :
1085 : 144265 : inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1086 : : /* This is the optimal IO size (for stat), not the fs block size */
1087 : 144265 : inode->i_blocks = 0;
1088 : 144265 : inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1089 : 144265 : ei->i_crtime = inode->i_mtime;
1090 : :
1091 : 144265 : memset(ei->i_data, 0, sizeof(ei->i_data));
1092 : 144265 : ei->i_dir_start_lookup = 0;
1093 : 144265 : ei->i_disksize = 0;
1094 : :
1095 : : /* Don't inherit extent flag from directory, amongst others. */
1096 : 144265 : ei->i_flags =
1097 : 144265 : ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1098 : 144265 : ei->i_flags |= i_flags;
1099 : 144265 : ei->i_file_acl = 0;
1100 : 144265 : ei->i_dtime = 0;
1101 : 144265 : ei->i_block_group = group;
1102 : 144265 : ei->i_last_alloc_group = ~0;
1103 : :
1104 : 144265 : ext4_set_inode_flags(inode);
1105 [ + - - + ]: 144265 : if (IS_DIRSYNC(inode))
1106 : : ext4_handle_sync(handle);
1107 [ - + ]: 144265 : if (insert_inode_locked(inode) < 0) {
1108 : : /*
1109 : : * Likely a bitmap corruption causing inode to be allocated
1110 : : * twice.
1111 : : */
1112 : : err = -EIO;
1113 : 0 : ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1114 : : inode->i_ino);
1115 : 0 : ext4_mark_group_bitmap_corrupted(sb, group,
1116 : : EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1117 : 0 : goto out;
1118 : : }
1119 : 144265 : inode->i_generation = prandom_u32();
1120 : :
1121 : : /* Precompute checksum seed for inode metadata */
1122 [ - + ]: 144265 : if (ext4_has_metadata_csum(sb)) {
1123 : : __u32 csum;
1124 : 0 : __le32 inum = cpu_to_le32(inode->i_ino);
1125 : 0 : __le32 gen = cpu_to_le32(inode->i_generation);
1126 : 0 : csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1127 : : sizeof(inum));
1128 : 0 : ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1129 : : sizeof(gen));
1130 : : }
1131 : :
1132 : : ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1133 : : ext4_set_inode_state(inode, EXT4_STATE_NEW);
1134 : :
1135 : 144265 : ei->i_extra_isize = sbi->s_want_extra_isize;
1136 : 144265 : ei->i_inline_off = 0;
1137 [ - + ]: 144265 : if (ext4_has_feature_inline_data(sb))
1138 : : ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1139 : : ret = inode;
1140 : 144265 : err = dquot_alloc_inode(inode);
1141 [ + - ]: 144265 : if (err)
1142 : : goto fail_drop;
1143 : :
1144 : : /*
1145 : : * Since the encryption xattr will always be unique, create it first so
1146 : : * that it's less likely to end up in an external xattr block and
1147 : : * prevent its deduplication.
1148 : : */
1149 [ - + ]: 144265 : if (encrypt) {
1150 : 0 : err = fscrypt_inherit_context(dir, inode, handle, true);
1151 [ # # ]: 0 : if (err)
1152 : : goto fail_free_drop;
1153 : : }
1154 : :
1155 [ + - ]: 144265 : if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1156 : 144265 : err = ext4_init_acl(handle, inode, dir);
1157 [ + - ]: 144265 : if (err)
1158 : : goto fail_free_drop;
1159 : :
1160 : 144265 : err = ext4_init_security(handle, inode, dir, qstr);
1161 [ + - ]: 144265 : if (err)
1162 : : goto fail_free_drop;
1163 : : }
1164 : :
1165 [ + - ]: 144265 : if (ext4_has_feature_extents(sb)) {
1166 : : /* set extent flag only for directory, file and normal symlink*/
1167 [ + + - + ]: 144265 : if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1168 : : ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1169 : 143644 : ext4_ext_tree_init(handle, inode);
1170 : : }
1171 : : }
1172 : :
1173 [ + - ]: 144265 : if (ext4_handle_valid(handle)) {
1174 : 144265 : ei->i_sync_tid = handle->h_transaction->t_tid;
1175 : 144265 : ei->i_datasync_tid = handle->h_transaction->t_tid;
1176 : : }
1177 : :
1178 : 144265 : err = ext4_mark_inode_dirty(handle, inode);
1179 [ - + ]: 144265 : if (err) {
1180 [ # # ]: 0 : ext4_std_error(sb, err);
1181 : : goto fail_free_drop;
1182 : : }
1183 : :
1184 : : ext4_debug("allocating inode %lu\n", inode->i_ino);
1185 : 144265 : trace_ext4_allocate_inode(inode, dir, mode);
1186 : : brelse(inode_bitmap_bh);
1187 : 144265 : return ret;
1188 : :
1189 : : fail_free_drop:
1190 : 0 : dquot_free_inode(inode);
1191 : : fail_drop:
1192 : 0 : clear_nlink(inode);
1193 : 0 : unlock_new_inode(inode);
1194 : : out:
1195 : 0 : dquot_drop(inode);
1196 : 0 : inode->i_flags |= S_NOQUOTA;
1197 : 0 : iput(inode);
1198 : : brelse(inode_bitmap_bh);
1199 : 0 : return ERR_PTR(err);
1200 : : }
1201 : :
1202 : : /* Verify that we are loading a valid orphan from disk */
1203 : 0 : struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1204 : : {
1205 : 0 : unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1206 : : ext4_group_t block_group;
1207 : : int bit;
1208 : : struct buffer_head *bitmap_bh = NULL;
1209 : : struct inode *inode = NULL;
1210 : : int err = -EFSCORRUPTED;
1211 : :
1212 [ # # # # ]: 0 : if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1213 : : goto bad_orphan;
1214 : :
1215 : 0 : block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1216 : 0 : bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1217 : 0 : bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1218 [ # # ]: 0 : if (IS_ERR(bitmap_bh))
1219 : : return ERR_CAST(bitmap_bh);
1220 : :
1221 : : /* Having the inode bit set should be a 100% indicator that this
1222 : : * is a valid orphan (no e2fsck run on fs). Orphans also include
1223 : : * inodes that were being truncated, so we can't check i_nlink==0.
1224 : : */
1225 [ # # ]: 0 : if (!ext4_test_bit(bit, bitmap_bh->b_data))
1226 : : goto bad_orphan;
1227 : :
1228 : 0 : inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1229 [ # # ]: 0 : if (IS_ERR(inode)) {
1230 : : err = PTR_ERR(inode);
1231 : 0 : ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1232 : : ino, err);
1233 : 0 : return inode;
1234 : : }
1235 : :
1236 : : /*
1237 : : * If the orphans has i_nlinks > 0 then it should be able to
1238 : : * be truncated, otherwise it won't be removed from the orphan
1239 : : * list during processing and an infinite loop will result.
1240 : : * Similarly, it must not be a bad inode.
1241 : : */
1242 [ # # # # : 0 : if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
# # ]
1243 : 0 : is_bad_inode(inode))
1244 : : goto bad_orphan;
1245 : :
1246 [ # # ]: 0 : if (NEXT_ORPHAN(inode) > max_ino)
1247 : : goto bad_orphan;
1248 : : brelse(bitmap_bh);
1249 : 0 : return inode;
1250 : :
1251 : : bad_orphan:
1252 : 0 : ext4_error(sb, "bad orphan inode %lu", ino);
1253 [ # # ]: 0 : if (bitmap_bh)
1254 : 0 : printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1255 : : bit, (unsigned long long)bitmap_bh->b_blocknr,
1256 : 0 : ext4_test_bit(bit, bitmap_bh->b_data));
1257 [ # # ]: 0 : if (inode) {
1258 : 0 : printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1259 : 0 : is_bad_inode(inode));
1260 : 0 : printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1261 : : NEXT_ORPHAN(inode));
1262 : 0 : printk(KERN_ERR "max_ino=%lu\n", max_ino);
1263 : 0 : printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1264 : : /* Avoid freeing blocks if we got a bad deleted inode */
1265 [ # # ]: 0 : if (inode->i_nlink == 0)
1266 : 0 : inode->i_blocks = 0;
1267 : 0 : iput(inode);
1268 : : }
1269 : : brelse(bitmap_bh);
1270 : : return ERR_PTR(err);
1271 : : }
1272 : :
1273 : 207 : unsigned long ext4_count_free_inodes(struct super_block *sb)
1274 : : {
1275 : : unsigned long desc_count;
1276 : : struct ext4_group_desc *gdp;
1277 : : ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1278 : : #ifdef EXT4FS_DEBUG
1279 : : struct ext4_super_block *es;
1280 : : unsigned long bitmap_count, x;
1281 : : struct buffer_head *bitmap_bh = NULL;
1282 : :
1283 : : es = EXT4_SB(sb)->s_es;
1284 : : desc_count = 0;
1285 : : bitmap_count = 0;
1286 : : gdp = NULL;
1287 : : for (i = 0; i < ngroups; i++) {
1288 : : gdp = ext4_get_group_desc(sb, i, NULL);
1289 : : if (!gdp)
1290 : : continue;
1291 : : desc_count += ext4_free_inodes_count(sb, gdp);
1292 : : brelse(bitmap_bh);
1293 : : bitmap_bh = ext4_read_inode_bitmap(sb, i);
1294 : : if (IS_ERR(bitmap_bh)) {
1295 : : bitmap_bh = NULL;
1296 : : continue;
1297 : : }
1298 : :
1299 : : x = ext4_count_free(bitmap_bh->b_data,
1300 : : EXT4_INODES_PER_GROUP(sb) / 8);
1301 : : printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1302 : : (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1303 : : bitmap_count += x;
1304 : : }
1305 : : brelse(bitmap_bh);
1306 : : printk(KERN_DEBUG "ext4_count_free_inodes: "
1307 : : "stored = %u, computed = %lu, %lu\n",
1308 : : le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1309 : : return desc_count;
1310 : : #else
1311 : : desc_count = 0;
1312 [ + + ]: 11178 : for (i = 0; i < ngroups; i++) {
1313 : 10971 : gdp = ext4_get_group_desc(sb, i, NULL);
1314 [ - + ]: 10971 : if (!gdp)
1315 : 0 : continue;
1316 : 10971 : desc_count += ext4_free_inodes_count(sb, gdp);
1317 : 10971 : cond_resched();
1318 : : }
1319 : 207 : return desc_count;
1320 : : #endif
1321 : : }
1322 : :
1323 : : /* Called at mount-time, super-block is locked */
1324 : 207 : unsigned long ext4_count_dirs(struct super_block * sb)
1325 : : {
1326 : : unsigned long count = 0;
1327 : : ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1328 : :
1329 [ + + ]: 11178 : for (i = 0; i < ngroups; i++) {
1330 : 10971 : struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1331 [ - + ]: 10971 : if (!gdp)
1332 : 0 : continue;
1333 : 10971 : count += ext4_used_dirs_count(sb, gdp);
1334 : : }
1335 : 207 : return count;
1336 : : }
1337 : :
1338 : : /*
1339 : : * Zeroes not yet zeroed inode table - just write zeroes through the whole
1340 : : * inode table. Must be called without any spinlock held. The only place
1341 : : * where it is called from on active part of filesystem is ext4lazyinit
1342 : : * thread, so we do not need any special locks, however we have to prevent
1343 : : * inode allocation from the current group, so we take alloc_sem lock, to
1344 : : * block ext4_new_inode() until we are finished.
1345 : : */
1346 : 0 : int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1347 : : int barrier)
1348 : : {
1349 : 0 : struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1350 : : struct ext4_sb_info *sbi = EXT4_SB(sb);
1351 : : struct ext4_group_desc *gdp = NULL;
1352 : : struct buffer_head *group_desc_bh;
1353 : : handle_t *handle;
1354 : : ext4_fsblk_t blk;
1355 : : int num, ret = 0, used_blks = 0;
1356 : :
1357 : : /* This should not happen, but just to be sure check this */
1358 [ # # ]: 0 : if (sb_rdonly(sb)) {
1359 : : ret = 1;
1360 : : goto out;
1361 : : }
1362 : :
1363 : 0 : gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1364 [ # # ]: 0 : if (!gdp)
1365 : : goto out;
1366 : :
1367 : : /*
1368 : : * We do not need to lock this, because we are the only one
1369 : : * handling this flag.
1370 : : */
1371 [ # # ]: 0 : if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1372 : : goto out;
1373 : :
1374 : 0 : handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1375 [ # # ]: 0 : if (IS_ERR(handle)) {
1376 : : ret = PTR_ERR(handle);
1377 : 0 : goto out;
1378 : : }
1379 : :
1380 : 0 : down_write(&grp->alloc_sem);
1381 : : /*
1382 : : * If inode bitmap was already initialized there may be some
1383 : : * used inodes so we need to skip blocks with used inodes in
1384 : : * inode table.
1385 : : */
1386 [ # # ]: 0 : if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1387 : 0 : used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1388 : : ext4_itable_unused_count(sb, gdp)),
1389 : : sbi->s_inodes_per_block);
1390 : :
1391 [ # # # # : 0 : if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
# # ]
1392 [ # # ]: 0 : ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1393 : 0 : ext4_itable_unused_count(sb, gdp)) <
1394 : 0 : EXT4_FIRST_INO(sb)))) {
1395 : 0 : ext4_error(sb, "Something is wrong with group %u: "
1396 : : "used itable blocks: %d; "
1397 : : "itable unused count: %u",
1398 : : group, used_blks,
1399 : : ext4_itable_unused_count(sb, gdp));
1400 : : ret = 1;
1401 : 0 : goto err_out;
1402 : : }
1403 : :
1404 : 0 : blk = ext4_inode_table(sb, gdp) + used_blks;
1405 : 0 : num = sbi->s_itb_per_group - used_blks;
1406 : :
1407 : : BUFFER_TRACE(group_desc_bh, "get_write_access");
1408 : 0 : ret = ext4_journal_get_write_access(handle,
1409 : : group_desc_bh);
1410 [ # # ]: 0 : if (ret)
1411 : : goto err_out;
1412 : :
1413 : : /*
1414 : : * Skip zeroout if the inode table is full. But we set the ZEROED
1415 : : * flag anyway, because obviously, when it is full it does not need
1416 : : * further zeroing.
1417 : : */
1418 [ # # ]: 0 : if (unlikely(num == 0))
1419 : : goto skip_zeroout;
1420 : :
1421 : : ext4_debug("going to zero out inode table in group %d\n",
1422 : : group);
1423 : 0 : ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1424 [ # # ]: 0 : if (ret < 0)
1425 : : goto err_out;
1426 [ # # ]: 0 : if (barrier)
1427 : 0 : blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1428 : :
1429 : : skip_zeroout:
1430 : 0 : ext4_lock_group(sb, group);
1431 : 0 : gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1432 : 0 : ext4_group_desc_csum_set(sb, group, gdp);
1433 : : ext4_unlock_group(sb, group);
1434 : :
1435 : : BUFFER_TRACE(group_desc_bh,
1436 : : "call ext4_handle_dirty_metadata");
1437 : 0 : ret = ext4_handle_dirty_metadata(handle, NULL,
1438 : : group_desc_bh);
1439 : :
1440 : : err_out:
1441 : 0 : up_write(&grp->alloc_sem);
1442 : 0 : ext4_journal_stop(handle);
1443 : : out:
1444 : 0 : return ret;
1445 : : }
|
