?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 6.13-rc4
[linux.git] / fs / ext4 / super.c
blob785809f33ff4a9bda0dcf3f03092fb7816c5b49d
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/super.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 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 */
19
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/vmalloc.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/blkdev.h>
28 #include <linux/backing-dev.h>
29 #include <linux/parser.h>
30 #include <linux/buffer_head.h>
31 #include <linux/exportfs.h>
32 #include <linux/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38 #include <linux/ctype.h>
39 #include <linux/log2.h>
40 #include <linux/crc16.h>
41 #include <linux/dax.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44 #include <linux/unicode.h>
45 #include <linux/part_stat.h>
46 #include <linux/kthread.h>
47 #include <linux/freezer.h>
48 #include <linux/fsnotify.h>
49 #include <linux/fs_context.h>
50 #include <linux/fs_parser.h>
51
52 #include "ext4.h"
53 #include "ext4_extents.h" /* Needed for trace points definition */
54 #include "ext4_jbd2.h"
55 #include "xattr.h"
56 #include "acl.h"
57 #include "mballoc.h"
58 #include "fsmap.h"
59
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/ext4.h>
62
63 static struct ext4_lazy_init *ext4_li_info;
64 static DEFINE_MUTEX(ext4_li_mtx);
65 static struct ratelimit_state ext4_mount_msg_ratelimit;
66
67 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
68 unsigned long journal_devnum);
69 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
70 static void ext4_update_super(struct super_block *sb);
71 static int ext4_commit_super(struct super_block *sb);
72 static int ext4_mark_recovery_complete(struct super_block *sb,
73 struct ext4_super_block *es);
74 static int ext4_clear_journal_err(struct super_block *sb,
75 struct ext4_super_block *es);
76 static int ext4_sync_fs(struct super_block *sb, int wait);
77 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
78 static int ext4_unfreeze(struct super_block *sb);
79 static int ext4_freeze(struct super_block *sb);
80 static inline int ext2_feature_set_ok(struct super_block *sb);
81 static inline int ext3_feature_set_ok(struct super_block *sb);
82 static void ext4_destroy_lazyinit_thread(void);
83 static void ext4_unregister_li_request(struct super_block *sb);
84 static void ext4_clear_request_list(void);
85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
86 unsigned int journal_inum);
87 static int ext4_validate_options(struct fs_context *fc);
88 static int ext4_check_opt_consistency(struct fs_context *fc,
89 struct super_block *sb);
90 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb);
91 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param);
92 static int ext4_get_tree(struct fs_context *fc);
93 static int ext4_reconfigure(struct fs_context *fc);
94 static void ext4_fc_free(struct fs_context *fc);
95 static int ext4_init_fs_context(struct fs_context *fc);
96 static void ext4_kill_sb(struct super_block *sb);
97 static const struct fs_parameter_spec ext4_param_specs[];
98
99 /*
100 * Lock ordering
101 *
102 * page fault path:
103 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
104 * -> page lock -> i_data_sem (rw)
105 *
106 * buffered write path:
107 * sb_start_write -> i_mutex -> mmap_lock
108 * sb_start_write -> i_mutex -> transaction start -> page lock ->
109 * i_data_sem (rw)
110 *
111 * truncate:
112 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
113 * page lock
114 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
115 * i_data_sem (rw)
116 *
117 * direct IO:
118 * sb_start_write -> i_mutex -> mmap_lock
119 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
120 *
121 * writepages:
122 * transaction start -> page lock(s) -> i_data_sem (rw)
123 */
124
125 static const struct fs_context_operations ext4_context_ops = {
126 .parse_param = ext4_parse_param,
127 .get_tree = ext4_get_tree,
128 .reconfigure = ext4_reconfigure,
129 .free = ext4_fc_free,
130 };
131
132
133 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
134 static struct file_system_type ext2_fs_type = {
135 .owner = THIS_MODULE,
136 .name = "ext2",
137 .init_fs_context = ext4_init_fs_context,
138 .parameters = ext4_param_specs,
139 .kill_sb = ext4_kill_sb,
140 .fs_flags = FS_REQUIRES_DEV,
141 };
142 MODULE_ALIAS_FS("ext2");
143 MODULE_ALIAS("ext2");
144 #define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type)
145 #else
146 #define IS_EXT2_SB(sb) (0)
147 #endif
148
149
150 static struct file_system_type ext3_fs_type = {
151 .owner = THIS_MODULE,
152 .name = "ext3",
153 .init_fs_context = ext4_init_fs_context,
154 .parameters = ext4_param_specs,
155 .kill_sb = ext4_kill_sb,
156 .fs_flags = FS_REQUIRES_DEV,
157 };
158 MODULE_ALIAS_FS("ext3");
159 MODULE_ALIAS("ext3");
160 #define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type)
161
162
163 static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
164 bh_end_io_t *end_io, bool simu_fail)
165 {
166 if (simu_fail) {
167 clear_buffer_uptodate(bh);
168 unlock_buffer(bh);
169 return;
170 }
171
172 /*
173 * buffer's verified bit is no longer valid after reading from
174 * disk again due to write out error, clear it to make sure we
175 * recheck the buffer contents.
176 */
177 clear_buffer_verified(bh);
178
179 bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
180 get_bh(bh);
181 submit_bh(REQ_OP_READ | op_flags, bh);
182 }
183
184 void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags,
185 bh_end_io_t *end_io, bool simu_fail)
186 {
187 BUG_ON(!buffer_locked(bh));
188
189 if (ext4_buffer_uptodate(bh)) {
190 unlock_buffer(bh);
191 return;
192 }
193 __ext4_read_bh(bh, op_flags, end_io, simu_fail);
194 }
195
196 int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
197 bh_end_io_t *end_io, bool simu_fail)
198 {
199 BUG_ON(!buffer_locked(bh));
200
201 if (ext4_buffer_uptodate(bh)) {
202 unlock_buffer(bh);
203 return 0;
204 }
205
206 __ext4_read_bh(bh, op_flags, end_io, simu_fail);
207
208 wait_on_buffer(bh);
209 if (buffer_uptodate(bh))
210 return 0;
211 return -EIO;
212 }
213
214 int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
215 {
216 lock_buffer(bh);
217 if (!wait) {
218 ext4_read_bh_nowait(bh, op_flags, NULL, false);
219 return 0;
220 }
221 return ext4_read_bh(bh, op_flags, NULL, false);
222 }
223
224 /*
225 * This works like __bread_gfp() except it uses ERR_PTR for error
226 * returns. Currently with sb_bread it's impossible to distinguish
227 * between ENOMEM and EIO situations (since both result in a NULL
228 * return.
229 */
230 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
231 sector_t block,
232 blk_opf_t op_flags, gfp_t gfp)
233 {
234 struct buffer_head *bh;
235 int ret;
236
237 bh = sb_getblk_gfp(sb, block, gfp);
238 if (bh == NULL)
239 return ERR_PTR(-ENOMEM);
240 if (ext4_buffer_uptodate(bh))
241 return bh;
242
243 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
244 if (ret) {
245 put_bh(bh);
246 return ERR_PTR(ret);
247 }
248 return bh;
249 }
250
251 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
252 blk_opf_t op_flags)
253 {
254 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping,
255 ~__GFP_FS) | __GFP_MOVABLE;
256
257 return __ext4_sb_bread_gfp(sb, block, op_flags, gfp);
258 }
259
260 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
261 sector_t block)
262 {
263 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_mapping,
264 ~__GFP_FS);
265
266 return __ext4_sb_bread_gfp(sb, block, 0, gfp);
267 }
268
269 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
270 {
271 struct buffer_head *bh = bdev_getblk(sb->s_bdev, block,
272 sb->s_blocksize, GFP_NOWAIT | __GFP_NOWARN);
273
274 if (likely(bh)) {
275 if (trylock_buffer(bh))
276 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL, false);
277 brelse(bh);
278 }
279 }
280
281 static int ext4_verify_csum_type(struct super_block *sb,
282 struct ext4_super_block *es)
283 {
284 if (!ext4_has_feature_metadata_csum(sb))
285 return 1;
286
287 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
288 }
289
290 __le32 ext4_superblock_csum(struct super_block *sb,
291 struct ext4_super_block *es)
292 {
293 struct ext4_sb_info *sbi = EXT4_SB(sb);
294 int offset = offsetof(struct ext4_super_block, s_checksum);
295 __u32 csum;
296
297 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
298
299 return cpu_to_le32(csum);
300 }
301
302 static int ext4_superblock_csum_verify(struct super_block *sb,
303 struct ext4_super_block *es)
304 {
305 if (!ext4_has_metadata_csum(sb))
306 return 1;
307
308 return es->s_checksum == ext4_superblock_csum(sb, es);
309 }
310
311 void ext4_superblock_csum_set(struct super_block *sb)
312 {
313 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
314
315 if (!ext4_has_metadata_csum(sb))
316 return;
317
318 es->s_checksum = ext4_superblock_csum(sb, es);
319 }
320
321 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
322 struct ext4_group_desc *bg)
323 {
324 return le32_to_cpu(bg->bg_block_bitmap_lo) |
325 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
326 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
327 }
328
329 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
330 struct ext4_group_desc *bg)
331 {
332 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
333 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
334 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
335 }
336
337 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
338 struct ext4_group_desc *bg)
339 {
340 return le32_to_cpu(bg->bg_inode_table_lo) |
341 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
342 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
343 }
344
345 __u32 ext4_free_group_clusters(struct super_block *sb,
346 struct ext4_group_desc *bg)
347 {
348 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
349 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
350 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
351 }
352
353 __u32 ext4_free_inodes_count(struct super_block *sb,
354 struct ext4_group_desc *bg)
355 {
356 return le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_lo)) |
357 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
358 (__u32)le16_to_cpu(READ_ONCE(bg->bg_free_inodes_count_hi)) << 16 : 0);
359 }
360
361 __u32 ext4_used_dirs_count(struct super_block *sb,
362 struct ext4_group_desc *bg)
363 {
364 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
365 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
366 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
367 }
368
369 __u32 ext4_itable_unused_count(struct super_block *sb,
370 struct ext4_group_desc *bg)
371 {
372 return le16_to_cpu(bg->bg_itable_unused_lo) |
373 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
374 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
375 }
376
377 void ext4_block_bitmap_set(struct super_block *sb,
378 struct ext4_group_desc *bg, ext4_fsblk_t blk)
379 {
380 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
381 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
382 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
383 }
384
385 void ext4_inode_bitmap_set(struct super_block *sb,
386 struct ext4_group_desc *bg, ext4_fsblk_t blk)
387 {
388 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
389 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
390 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
391 }
392
393 void ext4_inode_table_set(struct super_block *sb,
394 struct ext4_group_desc *bg, ext4_fsblk_t blk)
395 {
396 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
397 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
398 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
399 }
400
401 void ext4_free_group_clusters_set(struct super_block *sb,
402 struct ext4_group_desc *bg, __u32 count)
403 {
404 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
405 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
406 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
407 }
408
409 void ext4_free_inodes_set(struct super_block *sb,
410 struct ext4_group_desc *bg, __u32 count)
411 {
412 WRITE_ONCE(bg->bg_free_inodes_count_lo, cpu_to_le16((__u16)count));
413 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
414 WRITE_ONCE(bg->bg_free_inodes_count_hi, cpu_to_le16(count >> 16));
415 }
416
417 void ext4_used_dirs_set(struct super_block *sb,
418 struct ext4_group_desc *bg, __u32 count)
419 {
420 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
421 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
422 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
423 }
424
425 void ext4_itable_unused_set(struct super_block *sb,
426 struct ext4_group_desc *bg, __u32 count)
427 {
428 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
429 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
430 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
431 }
432
433 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
434 {
435 now = clamp_val(now, 0, (1ull << 40) - 1);
436
437 *lo = cpu_to_le32(lower_32_bits(now));
438 *hi = upper_32_bits(now);
439 }
440
441 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
442 {
443 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
444 }
445 #define ext4_update_tstamp(es, tstamp) \
446 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
447 ktime_get_real_seconds())
448 #define ext4_get_tstamp(es, tstamp) \
449 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
450
451 #define EXT4_SB_REFRESH_INTERVAL_SEC (3600) /* seconds (1 hour) */
452 #define EXT4_SB_REFRESH_INTERVAL_KB (16384) /* kilobytes (16MB) */
453
454 /*
455 * The ext4_maybe_update_superblock() function checks and updates the
456 * superblock if needed.
457 *
458 * This function is designed to update the on-disk superblock only under
459 * certain conditions to prevent excessive disk writes and unnecessary
460 * waking of the disk from sleep. The superblock will be updated if:
461 * 1. More than an hour has passed since the last superblock update, and
462 * 2. More than 16MB have been written since the last superblock update.
463 *
464 * @sb: The superblock
465 */
466 static void ext4_maybe_update_superblock(struct super_block *sb)
467 {
468 struct ext4_sb_info *sbi = EXT4_SB(sb);
469 struct ext4_super_block *es = sbi->s_es;
470 journal_t *journal = sbi->s_journal;
471 time64_t now;
472 __u64 last_update;
473 __u64 lifetime_write_kbytes;
474 __u64 diff_size;
475
476 if (sb_rdonly(sb) || !(sb->s_flags & SB_ACTIVE) ||
477 !journal || (journal->j_flags & JBD2_UNMOUNT))
478 return;
479
480 now = ktime_get_real_seconds();
481 last_update = ext4_get_tstamp(es, s_wtime);
482
483 if (likely(now - last_update < EXT4_SB_REFRESH_INTERVAL_SEC))
484 return;
485
486 lifetime_write_kbytes = sbi->s_kbytes_written +
487 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
488 sbi->s_sectors_written_start) >> 1);
489
490 /* Get the number of kilobytes not written to disk to account
491 * for statistics and compare with a multiple of 16 MB. This
492 * is used to determine when the next superblock commit should
493 * occur (i.e. not more often than once per 16MB if there was
494 * less written in an hour).
495 */
496 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written);
497
498 if (diff_size > EXT4_SB_REFRESH_INTERVAL_KB)
499 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
500 }
501
502 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
503 {
504 struct super_block *sb = journal->j_private;
505 struct ext4_sb_info *sbi = EXT4_SB(sb);
506 int error = is_journal_aborted(journal);
507 struct ext4_journal_cb_entry *jce;
508
509 BUG_ON(txn->t_state == T_FINISHED);
510
511 ext4_process_freed_data(sb, txn->t_tid);
512 ext4_maybe_update_superblock(sb);
513
514 spin_lock(&sbi->s_md_lock);
515 while (!list_empty(&txn->t_private_list)) {
516 jce = list_entry(txn->t_private_list.next,
517 struct ext4_journal_cb_entry, jce_list);
518 list_del_init(&jce->jce_list);
519 spin_unlock(&sbi->s_md_lock);
520 jce->jce_func(sb, jce, error);
521 spin_lock(&sbi->s_md_lock);
522 }
523 spin_unlock(&sbi->s_md_lock);
524 }
525
526 /*
527 * This writepage callback for write_cache_pages()
528 * takes care of a few cases after page cleaning.
529 *
530 * write_cache_pages() already checks for dirty pages
531 * and calls clear_page_dirty_for_io(), which we want,
532 * to write protect the pages.
533 *
534 * However, we may have to redirty a page (see below.)
535 */
536 static int ext4_journalled_writepage_callback(struct folio *folio,
537 struct writeback_control *wbc,
538 void *data)
539 {
540 transaction_t *transaction = (transaction_t *) data;
541 struct buffer_head *bh, *head;
542 struct journal_head *jh;
543
544 bh = head = folio_buffers(folio);
545 do {
546 /*
547 * We have to redirty a page in these cases:
548 * 1) If buffer is dirty, it means the page was dirty because it
549 * contains a buffer that needs checkpointing. So the dirty bit
550 * needs to be preserved so that checkpointing writes the buffer
551 * properly.
552 * 2) If buffer is not part of the committing transaction
553 * (we may have just accidentally come across this buffer because
554 * inode range tracking is not exact) or if the currently running
555 * transaction already contains this buffer as well, dirty bit
556 * needs to be preserved so that the buffer gets writeprotected
557 * properly on running transaction's commit.
558 */
559 jh = bh2jh(bh);
560 if (buffer_dirty(bh) ||
561 (jh && (jh->b_transaction != transaction ||
562 jh->b_next_transaction))) {
563 folio_redirty_for_writepage(wbc, folio);
564 goto out;
565 }
566 } while ((bh = bh->b_this_page) != head);
567
568 out:
569 return AOP_WRITEPAGE_ACTIVATE;
570 }
571
572 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
573 {
574 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
575 struct writeback_control wbc = {
576 .sync_mode = WB_SYNC_ALL,
577 .nr_to_write = LONG_MAX,
578 .range_start = jinode->i_dirty_start,
579 .range_end = jinode->i_dirty_end,
580 };
581
582 return write_cache_pages(mapping, &wbc,
583 ext4_journalled_writepage_callback,
584 jinode->i_transaction);
585 }
586
587 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
588 {
589 int ret;
590
591 if (ext4_should_journal_data(jinode->i_vfs_inode))
592 ret = ext4_journalled_submit_inode_data_buffers(jinode);
593 else
594 ret = ext4_normal_submit_inode_data_buffers(jinode);
595 return ret;
596 }
597
598 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
599 {
600 int ret = 0;
601
602 if (!ext4_should_journal_data(jinode->i_vfs_inode))
603 ret = jbd2_journal_finish_inode_data_buffers(jinode);
604
605 return ret;
606 }
607
608 static bool system_going_down(void)
609 {
610 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
611 || system_state == SYSTEM_RESTART;
612 }
613
614 struct ext4_err_translation {
615 int code;
616 int errno;
617 };
618
619 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
620
621 static struct ext4_err_translation err_translation[] = {
622 EXT4_ERR_TRANSLATE(EIO),
623 EXT4_ERR_TRANSLATE(ENOMEM),
624 EXT4_ERR_TRANSLATE(EFSBADCRC),
625 EXT4_ERR_TRANSLATE(EFSCORRUPTED),
626 EXT4_ERR_TRANSLATE(ENOSPC),
627 EXT4_ERR_TRANSLATE(ENOKEY),
628 EXT4_ERR_TRANSLATE(EROFS),
629 EXT4_ERR_TRANSLATE(EFBIG),
630 EXT4_ERR_TRANSLATE(EEXIST),
631 EXT4_ERR_TRANSLATE(ERANGE),
632 EXT4_ERR_TRANSLATE(EOVERFLOW),
633 EXT4_ERR_TRANSLATE(EBUSY),
634 EXT4_ERR_TRANSLATE(ENOTDIR),
635 EXT4_ERR_TRANSLATE(ENOTEMPTY),
636 EXT4_ERR_TRANSLATE(ESHUTDOWN),
637 EXT4_ERR_TRANSLATE(EFAULT),
638 };
639
640 static int ext4_errno_to_code(int errno)
641 {
642 int i;
643
644 for (i = 0; i < ARRAY_SIZE(err_translation); i++)
645 if (err_translation[i].errno == errno)
646 return err_translation[i].code;
647 return EXT4_ERR_UNKNOWN;
648 }
649
650 static void save_error_info(struct super_block *sb, int error,
651 __u32 ino, __u64 block,
652 const char *func, unsigned int line)
653 {
654 struct ext4_sb_info *sbi = EXT4_SB(sb);
655
656 /* We default to EFSCORRUPTED error... */
657 if (error == 0)
658 error = EFSCORRUPTED;
659
660 spin_lock(&sbi->s_error_lock);
661 sbi->s_add_error_count++;
662 sbi->s_last_error_code = error;
663 sbi->s_last_error_line = line;
664 sbi->s_last_error_ino = ino;
665 sbi->s_last_error_block = block;
666 sbi->s_last_error_func = func;
667 sbi->s_last_error_time = ktime_get_real_seconds();
668 if (!sbi->s_first_error_time) {
669 sbi->s_first_error_code = error;
670 sbi->s_first_error_line = line;
671 sbi->s_first_error_ino = ino;
672 sbi->s_first_error_block = block;
673 sbi->s_first_error_func = func;
674 sbi->s_first_error_time = sbi->s_last_error_time;
675 }
676 spin_unlock(&sbi->s_error_lock);
677 }
678
679 /* Deal with the reporting of failure conditions on a filesystem such as
680 * inconsistencies detected or read IO failures.
681 *
682 * On ext2, we can store the error state of the filesystem in the
683 * superblock. That is not possible on ext4, because we may have other
684 * write ordering constraints on the superblock which prevent us from
685 * writing it out straight away; and given that the journal is about to
686 * be aborted, we can't rely on the current, or future, transactions to
687 * write out the superblock safely.
688 *
689 * We'll just use the jbd2_journal_abort() error code to record an error in
690 * the journal instead. On recovery, the journal will complain about
691 * that error until we've noted it down and cleared it.
692 *
693 * If force_ro is set, we unconditionally force the filesystem into an
694 * ABORT|READONLY state, unless the error response on the fs has been set to
695 * panic in which case we take the easy way out and panic immediately. This is
696 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
697 * at a critical moment in log management.
698 */
699 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
700 __u32 ino, __u64 block,
701 const char *func, unsigned int line)
702 {
703 journal_t *journal = EXT4_SB(sb)->s_journal;
704 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
705
706 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
707 if (test_opt(sb, WARN_ON_ERROR))
708 WARN_ON_ONCE(1);
709
710 if (!continue_fs && !sb_rdonly(sb)) {
711 set_bit(EXT4_FLAGS_SHUTDOWN, &EXT4_SB(sb)->s_ext4_flags);
712 if (journal)
713 jbd2_journal_abort(journal, -EIO);
714 }
715
716 if (!bdev_read_only(sb->s_bdev)) {
717 save_error_info(sb, error, ino, block, func, line);
718 /*
719 * In case the fs should keep running, we need to writeout
720 * superblock through the journal. Due to lock ordering
721 * constraints, it may not be safe to do it right here so we
722 * defer superblock flushing to a workqueue.
723 */
724 if (continue_fs && journal)
725 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
726 else
727 ext4_commit_super(sb);
728 }
729
730 /*
731 * We force ERRORS_RO behavior when system is rebooting. Otherwise we
732 * could panic during 'reboot -f' as the underlying device got already
733 * disabled.
734 */
735 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
736 panic("EXT4-fs (device %s): panic forced after error\n",
737 sb->s_id);
738 }
739
740 if (sb_rdonly(sb) || continue_fs)
741 return;
742
743 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
744 /*
745 * EXT4_FLAGS_SHUTDOWN was set which stops all filesystem
746 * modifications. We don't set SB_RDONLY because that requires
747 * sb->s_umount semaphore and setting it without proper remount
748 * procedure is confusing code such as freeze_super() leading to
749 * deadlocks and other problems.
750 */
751 }
752
753 static void update_super_work(struct work_struct *work)
754 {
755 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
756 s_sb_upd_work);
757 journal_t *journal = sbi->s_journal;
758 handle_t *handle;
759
760 /*
761 * If the journal is still running, we have to write out superblock
762 * through the journal to avoid collisions of other journalled sb
763 * updates.
764 *
765 * We use directly jbd2 functions here to avoid recursing back into
766 * ext4 error handling code during handling of previous errors.
767 */
768 if (!sb_rdonly(sbi->s_sb) && journal) {
769 struct buffer_head *sbh = sbi->s_sbh;
770 bool call_notify_err = false;
771
772 handle = jbd2_journal_start(journal, 1);
773 if (IS_ERR(handle))
774 goto write_directly;
775 if (jbd2_journal_get_write_access(handle, sbh)) {
776 jbd2_journal_stop(handle);
777 goto write_directly;
778 }
779
780 if (sbi->s_add_error_count > 0)
781 call_notify_err = true;
782
783 ext4_update_super(sbi->s_sb);
784 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
785 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
786 "superblock detected");
787 clear_buffer_write_io_error(sbh);
788 set_buffer_uptodate(sbh);
789 }
790
791 if (jbd2_journal_dirty_metadata(handle, sbh)) {
792 jbd2_journal_stop(handle);
793 goto write_directly;
794 }
795 jbd2_journal_stop(handle);
796
797 if (call_notify_err)
798 ext4_notify_error_sysfs(sbi);
799
800 return;
801 }
802 write_directly:
803 /*
804 * Write through journal failed. Write sb directly to get error info
805 * out and hope for the best.
806 */
807 ext4_commit_super(sbi->s_sb);
808 ext4_notify_error_sysfs(sbi);
809 }
810
811 #define ext4_error_ratelimit(sb) \
812 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \
813 "EXT4-fs error")
814
815 void __ext4_error(struct super_block *sb, const char *function,
816 unsigned int line, bool force_ro, int error, __u64 block,
817 const char *fmt, ...)
818 {
819 struct va_format vaf;
820 va_list args;
821
822 if (unlikely(ext4_forced_shutdown(sb)))
823 return;
824
825 trace_ext4_error(sb, function, line);
826 if (ext4_error_ratelimit(sb)) {
827 va_start(args, fmt);
828 vaf.fmt = fmt;
829 vaf.va = &args;
830 printk(KERN_CRIT
831 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
832 sb->s_id, function, line, current->comm, &vaf);
833 va_end(args);
834 }
835 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
836
837 ext4_handle_error(sb, force_ro, error, 0, block, function, line);
838 }
839
840 void __ext4_error_inode(struct inode *inode, const char *function,
841 unsigned int line, ext4_fsblk_t block, int error,
842 const char *fmt, ...)
843 {
844 va_list args;
845 struct va_format vaf;
846
847 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
848 return;
849
850 trace_ext4_error(inode->i_sb, function, line);
851 if (ext4_error_ratelimit(inode->i_sb)) {
852 va_start(args, fmt);
853 vaf.fmt = fmt;
854 vaf.va = &args;
855 if (block)
856 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
857 "inode #%lu: block %llu: comm %s: %pV\n",
858 inode->i_sb->s_id, function, line, inode->i_ino,
859 block, current->comm, &vaf);
860 else
861 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
862 "inode #%lu: comm %s: %pV\n",
863 inode->i_sb->s_id, function, line, inode->i_ino,
864 current->comm, &vaf);
865 va_end(args);
866 }
867 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
868
869 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
870 function, line);
871 }
872
873 void __ext4_error_file(struct file *file, const char *function,
874 unsigned int line, ext4_fsblk_t block,
875 const char *fmt, ...)
876 {
877 va_list args;
878 struct va_format vaf;
879 struct inode *inode = file_inode(file);
880 char pathname[80], *path;
881
882 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
883 return;
884
885 trace_ext4_error(inode->i_sb, function, line);
886 if (ext4_error_ratelimit(inode->i_sb)) {
887 path = file_path(file, pathname, sizeof(pathname));
888 if (IS_ERR(path))
889 path = "(unknown)";
890 va_start(args, fmt);
891 vaf.fmt = fmt;
892 vaf.va = &args;
893 if (block)
894 printk(KERN_CRIT
895 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
896 "block %llu: comm %s: path %s: %pV\n",
897 inode->i_sb->s_id, function, line, inode->i_ino,
898 block, current->comm, path, &vaf);
899 else
900 printk(KERN_CRIT
901 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
902 "comm %s: path %s: %pV\n",
903 inode->i_sb->s_id, function, line, inode->i_ino,
904 current->comm, path, &vaf);
905 va_end(args);
906 }
907 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
908
909 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
910 function, line);
911 }
912
913 const char *ext4_decode_error(struct super_block *sb, int errno,
914 char nbuf[16])
915 {
916 char *errstr = NULL;
917
918 switch (errno) {
919 case -EFSCORRUPTED:
920 errstr = "Corrupt filesystem";
921 break;
922 case -EFSBADCRC:
923 errstr = "Filesystem failed CRC";
924 break;
925 case -EIO:
926 errstr = "IO failure";
927 break;
928 case -ENOMEM:
929 errstr = "Out of memory";
930 break;
931 case -EROFS:
932 if (!sb || (EXT4_SB(sb)->s_journal &&
933 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
934 errstr = "Journal has aborted";
935 else
936 errstr = "Readonly filesystem";
937 break;
938 default:
939 /* If the caller passed in an extra buffer for unknown
940 * errors, textualise them now. Else we just return
941 * NULL. */
942 if (nbuf) {
943 /* Check for truncated error codes... */
944 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
945 errstr = nbuf;
946 }
947 break;
948 }
949
950 return errstr;
951 }
952
953 /* __ext4_std_error decodes expected errors from journaling functions
954 * automatically and invokes the appropriate error response. */
955
956 void __ext4_std_error(struct super_block *sb, const char *function,
957 unsigned int line, int errno)
958 {
959 char nbuf[16];
960 const char *errstr;
961
962 if (unlikely(ext4_forced_shutdown(sb)))
963 return;
964
965 /* Special case: if the error is EROFS, and we're not already
966 * inside a transaction, then there's really no point in logging
967 * an error. */
968 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
969 return;
970
971 if (ext4_error_ratelimit(sb)) {
972 errstr = ext4_decode_error(sb, errno, nbuf);
973 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
974 sb->s_id, function, line, errstr);
975 }
976 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
977
978 ext4_handle_error(sb, false, -errno, 0, 0, function, line);
979 }
980
981 void __ext4_msg(struct super_block *sb,
982 const char *prefix, const char *fmt, ...)
983 {
984 struct va_format vaf;
985 va_list args;
986
987 if (sb) {
988 atomic_inc(&EXT4_SB(sb)->s_msg_count);
989 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
990 "EXT4-fs"))
991 return;
992 }
993
994 va_start(args, fmt);
995 vaf.fmt = fmt;
996 vaf.va = &args;
997 if (sb)
998 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
999 else
1000 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
1001 va_end(args);
1002 }
1003
1004 static int ext4_warning_ratelimit(struct super_block *sb)
1005 {
1006 atomic_inc(&EXT4_SB(sb)->s_warning_count);
1007 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
1008 "EXT4-fs warning");
1009 }
1010
1011 void __ext4_warning(struct super_block *sb, const char *function,
1012 unsigned int line, const char *fmt, ...)
1013 {
1014 struct va_format vaf;
1015 va_list args;
1016
1017 if (!ext4_warning_ratelimit(sb))
1018 return;
1019
1020 va_start(args, fmt);
1021 vaf.fmt = fmt;
1022 vaf.va = &args;
1023 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
1024 sb->s_id, function, line, &vaf);
1025 va_end(args);
1026 }
1027
1028 void __ext4_warning_inode(const struct inode *inode, const char *function,
1029 unsigned int line, const char *fmt, ...)
1030 {
1031 struct va_format vaf;
1032 va_list args;
1033
1034 if (!ext4_warning_ratelimit(inode->i_sb))
1035 return;
1036
1037 va_start(args, fmt);
1038 vaf.fmt = fmt;
1039 vaf.va = &args;
1040 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
1041 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
1042 function, line, inode->i_ino, current->comm, &vaf);
1043 va_end(args);
1044 }
1045
1046 void __ext4_grp_locked_error(const char *function, unsigned int line,
1047 struct super_block *sb, ext4_group_t grp,
1048 unsigned long ino, ext4_fsblk_t block,
1049 const char *fmt, ...)
1050 __releases(bitlock)
1051 __acquires(bitlock)
1052 {
1053 struct va_format vaf;
1054 va_list args;
1055
1056 if (unlikely(ext4_forced_shutdown(sb)))
1057 return;
1058
1059 trace_ext4_error(sb, function, line);
1060 if (ext4_error_ratelimit(sb)) {
1061 va_start(args, fmt);
1062 vaf.fmt = fmt;
1063 vaf.va = &args;
1064 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1065 sb->s_id, function, line, grp);
1066 if (ino)
1067 printk(KERN_CONT "inode %lu: ", ino);
1068 if (block)
1069 printk(KERN_CONT "block %llu:",
1070 (unsigned long long) block);
1071 printk(KERN_CONT "%pV\n", &vaf);
1072 va_end(args);
1073 }
1074
1075 if (test_opt(sb, ERRORS_CONT)) {
1076 if (test_opt(sb, WARN_ON_ERROR))
1077 WARN_ON_ONCE(1);
1078 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1079 if (!bdev_read_only(sb->s_bdev)) {
1080 save_error_info(sb, EFSCORRUPTED, ino, block, function,
1081 line);
1082 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
1083 }
1084 return;
1085 }
1086 ext4_unlock_group(sb, grp);
1087 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1088 /*
1089 * We only get here in the ERRORS_RO case; relocking the group
1090 * may be dangerous, but nothing bad will happen since the
1091 * filesystem will have already been marked read/only and the
1092 * journal has been aborted. We return 1 as a hint to callers
1093 * who might what to use the return value from
1094 * ext4_grp_locked_error() to distinguish between the
1095 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1096 * aggressively from the ext4 function in question, with a
1097 * more appropriate error code.
1098 */
1099 ext4_lock_group(sb, grp);
1100 return;
1101 }
1102
1103 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1104 ext4_group_t group,
1105 unsigned int flags)
1106 {
1107 struct ext4_sb_info *sbi = EXT4_SB(sb);
1108 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1109 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1110 int ret;
1111
1112 if (!grp || !gdp)
1113 return;
1114 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1115 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1116 &grp->bb_state);
1117 if (!ret)
1118 percpu_counter_sub(&sbi->s_freeclusters_counter,
1119 grp->bb_free);
1120 }
1121
1122 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1123 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1124 &grp->bb_state);
1125 if (!ret && gdp) {
1126 int count;
1127
1128 count = ext4_free_inodes_count(sb, gdp);
1129 percpu_counter_sub(&sbi->s_freeinodes_counter,
1130 count);
1131 }
1132 }
1133 }
1134
1135 void ext4_update_dynamic_rev(struct super_block *sb)
1136 {
1137 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1138
1139 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1140 return;
1141
1142 ext4_warning(sb,
1143 "updating to rev %d because of new feature flag, "
1144 "running e2fsck is recommended",
1145 EXT4_DYNAMIC_REV);
1146
1147 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1148 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1149 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1150 /* leave es->s_feature_*compat flags alone */
1151 /* es->s_uuid will be set by e2fsck if empty */
1152
1153 /*
1154 * The rest of the superblock fields should be zero, and if not it
1155 * means they are likely already in use, so leave them alone. We
1156 * can leave it up to e2fsck to clean up any inconsistencies there.
1157 */
1158 }
1159
1160 static inline struct inode *orphan_list_entry(struct list_head *l)
1161 {
1162 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1163 }
1164
1165 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1166 {
1167 struct list_head *l;
1168
1169 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1170 le32_to_cpu(sbi->s_es->s_last_orphan));
1171
1172 printk(KERN_ERR "sb_info orphan list:\n");
1173 list_for_each(l, &sbi->s_orphan) {
1174 struct inode *inode = orphan_list_entry(l);
1175 printk(KERN_ERR " "
1176 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1177 inode->i_sb->s_id, inode->i_ino, inode,
1178 inode->i_mode, inode->i_nlink,
1179 NEXT_ORPHAN(inode));
1180 }
1181 }
1182
1183 #ifdef CONFIG_QUOTA
1184 static int ext4_quota_off(struct super_block *sb, int type);
1185
1186 static inline void ext4_quotas_off(struct super_block *sb, int type)
1187 {
1188 BUG_ON(type > EXT4_MAXQUOTAS);
1189
1190 /* Use our quota_off function to clear inode flags etc. */
1191 for (type--; type >= 0; type--)
1192 ext4_quota_off(sb, type);
1193 }
1194
1195 /*
1196 * This is a helper function which is used in the mount/remount
1197 * codepaths (which holds s_umount) to fetch the quota file name.
1198 */
1199 static inline char *get_qf_name(struct super_block *sb,
1200 struct ext4_sb_info *sbi,
1201 int type)
1202 {
1203 return rcu_dereference_protected(sbi->s_qf_names[type],
1204 lockdep_is_held(&sb->s_umount));
1205 }
1206 #else
1207 static inline void ext4_quotas_off(struct super_block *sb, int type)
1208 {
1209 }
1210 #endif
1211
1212 static int ext4_percpu_param_init(struct ext4_sb_info *sbi)
1213 {
1214 ext4_fsblk_t block;
1215 int err;
1216
1217 block = ext4_count_free_clusters(sbi->s_sb);
1218 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block));
1219 err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
1220 GFP_KERNEL);
1221 if (!err) {
1222 unsigned long freei = ext4_count_free_inodes(sbi->s_sb);
1223 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
1224 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
1225 GFP_KERNEL);
1226 }
1227 if (!err)
1228 err = percpu_counter_init(&sbi->s_dirs_counter,
1229 ext4_count_dirs(sbi->s_sb), GFP_KERNEL);
1230 if (!err)
1231 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
1232 GFP_KERNEL);
1233 if (!err)
1234 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
1235 GFP_KERNEL);
1236 if (!err)
1237 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
1238
1239 if (err)
1240 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory");
1241
1242 return err;
1243 }
1244
1245 static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi)
1246 {
1247 percpu_counter_destroy(&sbi->s_freeclusters_counter);
1248 percpu_counter_destroy(&sbi->s_freeinodes_counter);
1249 percpu_counter_destroy(&sbi->s_dirs_counter);
1250 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1251 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1252 percpu_free_rwsem(&sbi->s_writepages_rwsem);
1253 }
1254
1255 static void ext4_group_desc_free(struct ext4_sb_info *sbi)
1256 {
1257 struct buffer_head **group_desc;
1258 int i;
1259
1260 rcu_read_lock();
1261 group_desc = rcu_dereference(sbi->s_group_desc);
1262 for (i = 0; i < sbi->s_gdb_count; i++)
1263 brelse(group_desc[i]);
1264 kvfree(group_desc);
1265 rcu_read_unlock();
1266 }
1267
1268 static void ext4_flex_groups_free(struct ext4_sb_info *sbi)
1269 {
1270 struct flex_groups **flex_groups;
1271 int i;
1272
1273 rcu_read_lock();
1274 flex_groups = rcu_dereference(sbi->s_flex_groups);
1275 if (flex_groups) {
1276 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1277 kvfree(flex_groups[i]);
1278 kvfree(flex_groups);
1279 }
1280 rcu_read_unlock();
1281 }
1282
1283 static void ext4_put_super(struct super_block *sb)
1284 {
1285 struct ext4_sb_info *sbi = EXT4_SB(sb);
1286 struct ext4_super_block *es = sbi->s_es;
1287 int aborted = 0;
1288 int err;
1289
1290 /*
1291 * Unregister sysfs before destroying jbd2 journal.
1292 * Since we could still access attr_journal_task attribute via sysfs
1293 * path which could have sbi->s_journal->j_task as NULL
1294 * Unregister sysfs before flush sbi->s_sb_upd_work.
1295 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1296 * read metadata verify failed then will queue error work.
1297 * update_super_work will call start_this_handle may trigger
1298 * BUG_ON.
1299 */
1300 ext4_unregister_sysfs(sb);
1301
1302 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1303 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.",
1304 &sb->s_uuid);
1305
1306 ext4_unregister_li_request(sb);
1307 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
1308
1309 flush_work(&sbi->s_sb_upd_work);
1310 destroy_workqueue(sbi->rsv_conversion_wq);
1311 ext4_release_orphan_info(sb);
1312
1313 if (sbi->s_journal) {
1314 aborted = is_journal_aborted(sbi->s_journal);
1315 err = jbd2_journal_destroy(sbi->s_journal);
1316 sbi->s_journal = NULL;
1317 if ((err < 0) && !aborted) {
1318 ext4_abort(sb, -err, "Couldn't clean up the journal");
1319 }
1320 }
1321
1322 ext4_es_unregister_shrinker(sbi);
1323 timer_shutdown_sync(&sbi->s_err_report);
1324 ext4_release_system_zone(sb);
1325 ext4_mb_release(sb);
1326 ext4_ext_release(sb);
1327
1328 if (!sb_rdonly(sb) && !aborted) {
1329 ext4_clear_feature_journal_needs_recovery(sb);
1330 ext4_clear_feature_orphan_present(sb);
1331 es->s_state = cpu_to_le16(sbi->s_mount_state);
1332 }
1333 if (!sb_rdonly(sb))
1334 ext4_commit_super(sb);
1335
1336 ext4_group_desc_free(sbi);
1337 ext4_flex_groups_free(sbi);
1338
1339 WARN_ON_ONCE(!(sbi->s_mount_state & EXT4_ERROR_FS) &&
1340 percpu_counter_sum(&sbi->s_dirtyclusters_counter));
1341 ext4_percpu_param_destroy(sbi);
1342 #ifdef CONFIG_QUOTA
1343 for (int i = 0; i < EXT4_MAXQUOTAS; i++)
1344 kfree(get_qf_name(sb, sbi, i));
1345 #endif
1346
1347 /* Debugging code just in case the in-memory inode orphan list
1348 * isn't empty. The on-disk one can be non-empty if we've
1349 * detected an error and taken the fs readonly, but the
1350 * in-memory list had better be clean by this point. */
1351 if (!list_empty(&sbi->s_orphan))
1352 dump_orphan_list(sb, sbi);
1353 ASSERT(list_empty(&sbi->s_orphan));
1354
1355 sync_blockdev(sb->s_bdev);
1356 invalidate_bdev(sb->s_bdev);
1357 if (sbi->s_journal_bdev_file) {
1358 /*
1359 * Invalidate the journal device's buffers. We don't want them
1360 * floating about in memory - the physical journal device may
1361 * hotswapped, and it breaks the `ro-after' testing code.
1362 */
1363 sync_blockdev(file_bdev(sbi->s_journal_bdev_file));
1364 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
1365 }
1366
1367 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1368 sbi->s_ea_inode_cache = NULL;
1369
1370 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1371 sbi->s_ea_block_cache = NULL;
1372
1373 ext4_stop_mmpd(sbi);
1374
1375 brelse(sbi->s_sbh);
1376 sb->s_fs_info = NULL;
1377 /*
1378 * Now that we are completely done shutting down the
1379 * superblock, we need to actually destroy the kobject.
1380 */
1381 kobject_put(&sbi->s_kobj);
1382 wait_for_completion(&sbi->s_kobj_unregister);
1383 if (sbi->s_chksum_driver)
1384 crypto_free_shash(sbi->s_chksum_driver);
1385 kfree(sbi->s_blockgroup_lock);
1386 fs_put_dax(sbi->s_daxdev, NULL);
1387 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1388 #if IS_ENABLED(CONFIG_UNICODE)
1389 utf8_unload(sb->s_encoding);
1390 #endif
1391 kfree(sbi);
1392 }
1393
1394 static struct kmem_cache *ext4_inode_cachep;
1395
1396 /*
1397 * Called inside transaction, so use GFP_NOFS
1398 */
1399 static struct inode *ext4_alloc_inode(struct super_block *sb)
1400 {
1401 struct ext4_inode_info *ei;
1402
1403 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1404 if (!ei)
1405 return NULL;
1406
1407 inode_set_iversion(&ei->vfs_inode, 1);
1408 ei->i_flags = 0;
1409 spin_lock_init(&ei->i_raw_lock);
1410 ei->i_prealloc_node = RB_ROOT;
1411 atomic_set(&ei->i_prealloc_active, 0);
1412 rwlock_init(&ei->i_prealloc_lock);
1413 ext4_es_init_tree(&ei->i_es_tree);
1414 rwlock_init(&ei->i_es_lock);
1415 INIT_LIST_HEAD(&ei->i_es_list);
1416 ei->i_es_all_nr = 0;
1417 ei->i_es_shk_nr = 0;
1418 ei->i_es_shrink_lblk = 0;
1419 ei->i_reserved_data_blocks = 0;
1420 spin_lock_init(&(ei->i_block_reservation_lock));
1421 ext4_init_pending_tree(&ei->i_pending_tree);
1422 #ifdef CONFIG_QUOTA
1423 ei->i_reserved_quota = 0;
1424 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1425 #endif
1426 ei->jinode = NULL;
1427 INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1428 spin_lock_init(&ei->i_completed_io_lock);
1429 ei->i_sync_tid = 0;
1430 ei->i_datasync_tid = 0;
1431 atomic_set(&ei->i_unwritten, 0);
1432 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1433 ext4_fc_init_inode(&ei->vfs_inode);
1434 mutex_init(&ei->i_fc_lock);
1435 return &ei->vfs_inode;
1436 }
1437
1438 static int ext4_drop_inode(struct inode *inode)
1439 {
1440 int drop = generic_drop_inode(inode);
1441
1442 if (!drop)
1443 drop = fscrypt_drop_inode(inode);
1444
1445 trace_ext4_drop_inode(inode, drop);
1446 return drop;
1447 }
1448
1449 static void ext4_free_in_core_inode(struct inode *inode)
1450 {
1451 fscrypt_free_inode(inode);
1452 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1453 pr_warn("%s: inode %ld still in fc list",
1454 __func__, inode->i_ino);
1455 }
1456 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1457 }
1458
1459 static void ext4_destroy_inode(struct inode *inode)
1460 {
1461 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1462 ext4_msg(inode->i_sb, KERN_ERR,
1463 "Inode %lu (%p): orphan list check failed!",
1464 inode->i_ino, EXT4_I(inode));
1465 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1466 EXT4_I(inode), sizeof(struct ext4_inode_info),
1467 true);
1468 dump_stack();
1469 }
1470
1471 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ERROR_FS) &&
1472 WARN_ON_ONCE(EXT4_I(inode)->i_reserved_data_blocks))
1473 ext4_msg(inode->i_sb, KERN_ERR,
1474 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1475 inode->i_ino, EXT4_I(inode),
1476 EXT4_I(inode)->i_reserved_data_blocks);
1477 }
1478
1479 static void ext4_shutdown(struct super_block *sb)
1480 {
1481 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH);
1482 }
1483
1484 static void init_once(void *foo)
1485 {
1486 struct ext4_inode_info *ei = foo;
1487
1488 INIT_LIST_HEAD(&ei->i_orphan);
1489 init_rwsem(&ei->xattr_sem);
1490 init_rwsem(&ei->i_data_sem);
1491 inode_init_once(&ei->vfs_inode);
1492 ext4_fc_init_inode(&ei->vfs_inode);
1493 }
1494
1495 static int __init init_inodecache(void)
1496 {
1497 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1498 sizeof(struct ext4_inode_info), 0,
1499 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
1500 offsetof(struct ext4_inode_info, i_data),
1501 sizeof_field(struct ext4_inode_info, i_data),
1502 init_once);
1503 if (ext4_inode_cachep == NULL)
1504 return -ENOMEM;
1505 return 0;
1506 }
1507
1508 static void destroy_inodecache(void)
1509 {
1510 /*
1511 * Make sure all delayed rcu free inodes are flushed before we
1512 * destroy cache.
1513 */
1514 rcu_barrier();
1515 kmem_cache_destroy(ext4_inode_cachep);
1516 }
1517
1518 void ext4_clear_inode(struct inode *inode)
1519 {
1520 ext4_fc_del(inode);
1521 invalidate_inode_buffers(inode);
1522 clear_inode(inode);
1523 ext4_discard_preallocations(inode);
1524 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1525 dquot_drop(inode);
1526 if (EXT4_I(inode)->jinode) {
1527 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1528 EXT4_I(inode)->jinode);
1529 jbd2_free_inode(EXT4_I(inode)->jinode);
1530 EXT4_I(inode)->jinode = NULL;
1531 }
1532 fscrypt_put_encryption_info(inode);
1533 fsverity_cleanup_inode(inode);
1534 }
1535
1536 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1537 u64 ino, u32 generation)
1538 {
1539 struct inode *inode;
1540
1541 /*
1542 * Currently we don't know the generation for parent directory, so
1543 * a generation of 0 means "accept any"
1544 */
1545 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1546 if (IS_ERR(inode))
1547 return ERR_CAST(inode);
1548 if (generation && inode->i_generation != generation) {
1549 iput(inode);
1550 return ERR_PTR(-ESTALE);
1551 }
1552
1553 return inode;
1554 }
1555
1556 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1557 int fh_len, int fh_type)
1558 {
1559 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1560 ext4_nfs_get_inode);
1561 }
1562
1563 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1564 int fh_len, int fh_type)
1565 {
1566 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1567 ext4_nfs_get_inode);
1568 }
1569
1570 static int ext4_nfs_commit_metadata(struct inode *inode)
1571 {
1572 struct writeback_control wbc = {
1573 .sync_mode = WB_SYNC_ALL
1574 };
1575
1576 trace_ext4_nfs_commit_metadata(inode);
1577 return ext4_write_inode(inode, &wbc);
1578 }
1579
1580 #ifdef CONFIG_QUOTA
1581 static const char * const quotatypes[] = INITQFNAMES;
1582 #define QTYPE2NAME(t) (quotatypes[t])
1583
1584 static int ext4_write_dquot(struct dquot *dquot);
1585 static int ext4_acquire_dquot(struct dquot *dquot);
1586 static int ext4_release_dquot(struct dquot *dquot);
1587 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1588 static int ext4_write_info(struct super_block *sb, int type);
1589 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1590 const struct path *path);
1591 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1592 size_t len, loff_t off);
1593 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1594 const char *data, size_t len, loff_t off);
1595 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1596 unsigned int flags);
1597
1598 static struct dquot __rcu **ext4_get_dquots(struct inode *inode)
1599 {
1600 return EXT4_I(inode)->i_dquot;
1601 }
1602
1603 static const struct dquot_operations ext4_quota_operations = {
1604 .get_reserved_space = ext4_get_reserved_space,
1605 .write_dquot = ext4_write_dquot,
1606 .acquire_dquot = ext4_acquire_dquot,
1607 .release_dquot = ext4_release_dquot,
1608 .mark_dirty = ext4_mark_dquot_dirty,
1609 .write_info = ext4_write_info,
1610 .alloc_dquot = dquot_alloc,
1611 .destroy_dquot = dquot_destroy,
1612 .get_projid = ext4_get_projid,
1613 .get_inode_usage = ext4_get_inode_usage,
1614 .get_next_id = dquot_get_next_id,
1615 };
1616
1617 static const struct quotactl_ops ext4_qctl_operations = {
1618 .quota_on = ext4_quota_on,
1619 .quota_off = ext4_quota_off,
1620 .quota_sync = dquot_quota_sync,
1621 .get_state = dquot_get_state,
1622 .set_info = dquot_set_dqinfo,
1623 .get_dqblk = dquot_get_dqblk,
1624 .set_dqblk = dquot_set_dqblk,
1625 .get_nextdqblk = dquot_get_next_dqblk,
1626 };
1627 #endif
1628
1629 static const struct super_operations ext4_sops = {
1630 .alloc_inode = ext4_alloc_inode,
1631 .free_inode = ext4_free_in_core_inode,
1632 .destroy_inode = ext4_destroy_inode,
1633 .write_inode = ext4_write_inode,
1634 .dirty_inode = ext4_dirty_inode,
1635 .drop_inode = ext4_drop_inode,
1636 .evict_inode = ext4_evict_inode,
1637 .put_super = ext4_put_super,
1638 .sync_fs = ext4_sync_fs,
1639 .freeze_fs = ext4_freeze,
1640 .unfreeze_fs = ext4_unfreeze,
1641 .statfs = ext4_statfs,
1642 .show_options = ext4_show_options,
1643 .shutdown = ext4_shutdown,
1644 #ifdef CONFIG_QUOTA
1645 .quota_read = ext4_quota_read,
1646 .quota_write = ext4_quota_write,
1647 .get_dquots = ext4_get_dquots,
1648 #endif
1649 };
1650
1651 static const struct export_operations ext4_export_ops = {
1652 .encode_fh = generic_encode_ino32_fh,
1653 .fh_to_dentry = ext4_fh_to_dentry,
1654 .fh_to_parent = ext4_fh_to_parent,
1655 .get_parent = ext4_get_parent,
1656 .commit_metadata = ext4_nfs_commit_metadata,
1657 };
1658
1659 enum {
1660 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1661 Opt_resgid, Opt_resuid, Opt_sb,
1662 Opt_nouid32, Opt_debug, Opt_removed,
1663 Opt_user_xattr, Opt_acl,
1664 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1665 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1666 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1667 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1668 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1669 Opt_inlinecrypt,
1670 Opt_usrjquota, Opt_grpjquota, Opt_quota,
1671 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1672 Opt_usrquota, Opt_grpquota, Opt_prjquota,
1673 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1674 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1675 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1676 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1677 Opt_inode_readahead_blks, Opt_journal_ioprio,
1678 Opt_dioread_nolock, Opt_dioread_lock,
1679 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1680 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1681 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1682 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1683 #ifdef CONFIG_EXT4_DEBUG
1684 Opt_fc_debug_max_replay, Opt_fc_debug_force
1685 #endif
1686 };
1687
1688 static const struct constant_table ext4_param_errors[] = {
1689 {"continue", EXT4_MOUNT_ERRORS_CONT},
1690 {"panic", EXT4_MOUNT_ERRORS_PANIC},
1691 {"remount-ro", EXT4_MOUNT_ERRORS_RO},
1692 {}
1693 };
1694
1695 static const struct constant_table ext4_param_data[] = {
1696 {"journal", EXT4_MOUNT_JOURNAL_DATA},
1697 {"ordered", EXT4_MOUNT_ORDERED_DATA},
1698 {"writeback", EXT4_MOUNT_WRITEBACK_DATA},
1699 {}
1700 };
1701
1702 static const struct constant_table ext4_param_data_err[] = {
1703 {"abort", Opt_data_err_abort},
1704 {"ignore", Opt_data_err_ignore},
1705 {}
1706 };
1707
1708 static const struct constant_table ext4_param_jqfmt[] = {
1709 {"vfsold", QFMT_VFS_OLD},
1710 {"vfsv0", QFMT_VFS_V0},
1711 {"vfsv1", QFMT_VFS_V1},
1712 {}
1713 };
1714
1715 static const struct constant_table ext4_param_dax[] = {
1716 {"always", Opt_dax_always},
1717 {"inode", Opt_dax_inode},
1718 {"never", Opt_dax_never},
1719 {}
1720 };
1721
1722 /*
1723 * Mount option specification
1724 * We don't use fsparam_flag_no because of the way we set the
1725 * options and the way we show them in _ext4_show_options(). To
1726 * keep the changes to a minimum, let's keep the negative options
1727 * separate for now.
1728 */
1729 static const struct fs_parameter_spec ext4_param_specs[] = {
1730 fsparam_flag ("bsddf", Opt_bsd_df),
1731 fsparam_flag ("minixdf", Opt_minix_df),
1732 fsparam_flag ("grpid", Opt_grpid),
1733 fsparam_flag ("bsdgroups", Opt_grpid),
1734 fsparam_flag ("nogrpid", Opt_nogrpid),
1735 fsparam_flag ("sysvgroups", Opt_nogrpid),
1736 fsparam_gid ("resgid", Opt_resgid),
1737 fsparam_uid ("resuid", Opt_resuid),
1738 fsparam_u32 ("sb", Opt_sb),
1739 fsparam_enum ("errors", Opt_errors, ext4_param_errors),
1740 fsparam_flag ("nouid32", Opt_nouid32),
1741 fsparam_flag ("debug", Opt_debug),
1742 fsparam_flag ("oldalloc", Opt_removed),
1743 fsparam_flag ("orlov", Opt_removed),
1744 fsparam_flag ("user_xattr", Opt_user_xattr),
1745 fsparam_flag ("acl", Opt_acl),
1746 fsparam_flag ("norecovery", Opt_noload),
1747 fsparam_flag ("noload", Opt_noload),
1748 fsparam_flag ("bh", Opt_removed),
1749 fsparam_flag ("nobh", Opt_removed),
1750 fsparam_u32 ("commit", Opt_commit),
1751 fsparam_u32 ("min_batch_time", Opt_min_batch_time),
1752 fsparam_u32 ("max_batch_time", Opt_max_batch_time),
1753 fsparam_u32 ("journal_dev", Opt_journal_dev),
1754 fsparam_bdev ("journal_path", Opt_journal_path),
1755 fsparam_flag ("journal_checksum", Opt_journal_checksum),
1756 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum),
1757 fsparam_flag ("journal_async_commit",Opt_journal_async_commit),
1758 fsparam_flag ("abort", Opt_abort),
1759 fsparam_enum ("data", Opt_data, ext4_param_data),
1760 fsparam_enum ("data_err", Opt_data_err,
1761 ext4_param_data_err),
1762 fsparam_string_empty
1763 ("usrjquota", Opt_usrjquota),
1764 fsparam_string_empty
1765 ("grpjquota", Opt_grpjquota),
1766 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt),
1767 fsparam_flag ("grpquota", Opt_grpquota),
1768 fsparam_flag ("quota", Opt_quota),
1769 fsparam_flag ("noquota", Opt_noquota),
1770 fsparam_flag ("usrquota", Opt_usrquota),
1771 fsparam_flag ("prjquota", Opt_prjquota),
1772 fsparam_flag ("barrier", Opt_barrier),
1773 fsparam_u32 ("barrier", Opt_barrier),
1774 fsparam_flag ("nobarrier", Opt_nobarrier),
1775 fsparam_flag ("i_version", Opt_removed),
1776 fsparam_flag ("dax", Opt_dax),
1777 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax),
1778 fsparam_u32 ("stripe", Opt_stripe),
1779 fsparam_flag ("delalloc", Opt_delalloc),
1780 fsparam_flag ("nodelalloc", Opt_nodelalloc),
1781 fsparam_flag ("warn_on_error", Opt_warn_on_error),
1782 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error),
1783 fsparam_u32 ("debug_want_extra_isize",
1784 Opt_debug_want_extra_isize),
1785 fsparam_flag ("mblk_io_submit", Opt_removed),
1786 fsparam_flag ("nomblk_io_submit", Opt_removed),
1787 fsparam_flag ("block_validity", Opt_block_validity),
1788 fsparam_flag ("noblock_validity", Opt_noblock_validity),
1789 fsparam_u32 ("inode_readahead_blks",
1790 Opt_inode_readahead_blks),
1791 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio),
1792 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc),
1793 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc),
1794 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc),
1795 fsparam_flag ("dioread_nolock", Opt_dioread_nolock),
1796 fsparam_flag ("nodioread_nolock", Opt_dioread_lock),
1797 fsparam_flag ("dioread_lock", Opt_dioread_lock),
1798 fsparam_flag ("discard", Opt_discard),
1799 fsparam_flag ("nodiscard", Opt_nodiscard),
1800 fsparam_u32 ("init_itable", Opt_init_itable),
1801 fsparam_flag ("init_itable", Opt_init_itable),
1802 fsparam_flag ("noinit_itable", Opt_noinit_itable),
1803 #ifdef CONFIG_EXT4_DEBUG
1804 fsparam_flag ("fc_debug_force", Opt_fc_debug_force),
1805 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1806 #endif
1807 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb),
1808 fsparam_flag ("test_dummy_encryption",
1809 Opt_test_dummy_encryption),
1810 fsparam_string ("test_dummy_encryption",
1811 Opt_test_dummy_encryption),
1812 fsparam_flag ("inlinecrypt", Opt_inlinecrypt),
1813 fsparam_flag ("nombcache", Opt_nombcache),
1814 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */
1815 fsparam_flag ("prefetch_block_bitmaps",
1816 Opt_removed),
1817 fsparam_flag ("no_prefetch_block_bitmaps",
1818 Opt_no_prefetch_block_bitmaps),
1819 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan),
1820 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */
1821 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */
1822 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */
1823 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */
1824 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */
1825 {}
1826 };
1827
1828 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1829
1830 #define MOPT_SET 0x0001
1831 #define MOPT_CLEAR 0x0002
1832 #define MOPT_NOSUPPORT 0x0004
1833 #define MOPT_EXPLICIT 0x0008
1834 #ifdef CONFIG_QUOTA
1835 #define MOPT_Q 0
1836 #define MOPT_QFMT 0x0010
1837 #else
1838 #define MOPT_Q MOPT_NOSUPPORT
1839 #define MOPT_QFMT MOPT_NOSUPPORT
1840 #endif
1841 #define MOPT_NO_EXT2 0x0020
1842 #define MOPT_NO_EXT3 0x0040
1843 #define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1844 #define MOPT_SKIP 0x0080
1845 #define MOPT_2 0x0100
1846
1847 static const struct mount_opts {
1848 int token;
1849 int mount_opt;
1850 int flags;
1851 } ext4_mount_opts[] = {
1852 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1853 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1854 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1855 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1856 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1857 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1858 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1859 MOPT_EXT4_ONLY | MOPT_SET},
1860 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1861 MOPT_EXT4_ONLY | MOPT_CLEAR},
1862 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1863 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1864 {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1865 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1866 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1867 MOPT_EXT4_ONLY | MOPT_CLEAR},
1868 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1869 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1870 {Opt_commit, 0, MOPT_NO_EXT2},
1871 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1872 MOPT_EXT4_ONLY | MOPT_CLEAR},
1873 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1874 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1875 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1876 EXT4_MOUNT_JOURNAL_CHECKSUM),
1877 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1878 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1879 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1880 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1881 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1882 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1883 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1884 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1885 {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1886 {Opt_journal_dev, 0, MOPT_NO_EXT2},
1887 {Opt_journal_path, 0, MOPT_NO_EXT2},
1888 {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1889 {Opt_data, 0, MOPT_NO_EXT2},
1890 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1891 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1892 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1893 #else
1894 {Opt_acl, 0, MOPT_NOSUPPORT},
1895 #endif
1896 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1897 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1898 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1899 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1900 MOPT_SET | MOPT_Q},
1901 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1902 MOPT_SET | MOPT_Q},
1903 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1904 MOPT_SET | MOPT_Q},
1905 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1906 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1907 MOPT_CLEAR | MOPT_Q},
1908 {Opt_usrjquota, 0, MOPT_Q},
1909 {Opt_grpjquota, 0, MOPT_Q},
1910 {Opt_jqfmt, 0, MOPT_QFMT},
1911 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1912 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1913 MOPT_SET},
1914 #ifdef CONFIG_EXT4_DEBUG
1915 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1916 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1917 #endif
1918 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2},
1919 {Opt_err, 0, 0}
1920 };
1921
1922 #if IS_ENABLED(CONFIG_UNICODE)
1923 static const struct ext4_sb_encodings {
1924 __u16 magic;
1925 char *name;
1926 unsigned int version;
1927 } ext4_sb_encoding_map[] = {
1928 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1929 };
1930
1931 static const struct ext4_sb_encodings *
1932 ext4_sb_read_encoding(const struct ext4_super_block *es)
1933 {
1934 __u16 magic = le16_to_cpu(es->s_encoding);
1935 int i;
1936
1937 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1938 if (magic == ext4_sb_encoding_map[i].magic)
1939 return &ext4_sb_encoding_map[i];
1940
1941 return NULL;
1942 }
1943 #endif
1944
1945 #define EXT4_SPEC_JQUOTA (1 << 0)
1946 #define EXT4_SPEC_JQFMT (1 << 1)
1947 #define EXT4_SPEC_DATAJ (1 << 2)
1948 #define EXT4_SPEC_SB_BLOCK (1 << 3)
1949 #define EXT4_SPEC_JOURNAL_DEV (1 << 4)
1950 #define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5)
1951 #define EXT4_SPEC_s_want_extra_isize (1 << 7)
1952 #define EXT4_SPEC_s_max_batch_time (1 << 8)
1953 #define EXT4_SPEC_s_min_batch_time (1 << 9)
1954 #define EXT4_SPEC_s_inode_readahead_blks (1 << 10)
1955 #define EXT4_SPEC_s_li_wait_mult (1 << 11)
1956 #define EXT4_SPEC_s_max_dir_size_kb (1 << 12)
1957 #define EXT4_SPEC_s_stripe (1 << 13)
1958 #define EXT4_SPEC_s_resuid (1 << 14)
1959 #define EXT4_SPEC_s_resgid (1 << 15)
1960 #define EXT4_SPEC_s_commit_interval (1 << 16)
1961 #define EXT4_SPEC_s_fc_debug_max_replay (1 << 17)
1962 #define EXT4_SPEC_s_sb_block (1 << 18)
1963 #define EXT4_SPEC_mb_optimize_scan (1 << 19)
1964
1965 struct ext4_fs_context {
1966 char *s_qf_names[EXT4_MAXQUOTAS];
1967 struct fscrypt_dummy_policy dummy_enc_policy;
1968 int s_jquota_fmt; /* Format of quota to use */
1969 #ifdef CONFIG_EXT4_DEBUG
1970 int s_fc_debug_max_replay;
1971 #endif
1972 unsigned short qname_spec;
1973 unsigned long vals_s_flags; /* Bits to set in s_flags */
1974 unsigned long mask_s_flags; /* Bits changed in s_flags */
1975 unsigned long journal_devnum;
1976 unsigned long s_commit_interval;
1977 unsigned long s_stripe;
1978 unsigned int s_inode_readahead_blks;
1979 unsigned int s_want_extra_isize;
1980 unsigned int s_li_wait_mult;
1981 unsigned int s_max_dir_size_kb;
1982 unsigned int journal_ioprio;
1983 unsigned int vals_s_mount_opt;
1984 unsigned int mask_s_mount_opt;
1985 unsigned int vals_s_mount_opt2;
1986 unsigned int mask_s_mount_opt2;
1987 unsigned int opt_flags; /* MOPT flags */
1988 unsigned int spec;
1989 u32 s_max_batch_time;
1990 u32 s_min_batch_time;
1991 kuid_t s_resuid;
1992 kgid_t s_resgid;
1993 ext4_fsblk_t s_sb_block;
1994 };
1995
1996 static void ext4_fc_free(struct fs_context *fc)
1997 {
1998 struct ext4_fs_context *ctx = fc->fs_private;
1999 int i;
2000
2001 if (!ctx)
2002 return;
2003
2004 for (i = 0; i < EXT4_MAXQUOTAS; i++)
2005 kfree(ctx->s_qf_names[i]);
2006
2007 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
2008 kfree(ctx);
2009 }
2010
2011 int ext4_init_fs_context(struct fs_context *fc)
2012 {
2013 struct ext4_fs_context *ctx;
2014
2015 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2016 if (!ctx)
2017 return -ENOMEM;
2018
2019 fc->fs_private = ctx;
2020 fc->ops = &ext4_context_ops;
2021
2022 return 0;
2023 }
2024
2025 #ifdef CONFIG_QUOTA
2026 /*
2027 * Note the name of the specified quota file.
2028 */
2029 static int note_qf_name(struct fs_context *fc, int qtype,
2030 struct fs_parameter *param)
2031 {
2032 struct ext4_fs_context *ctx = fc->fs_private;
2033 char *qname;
2034
2035 if (param->size < 1) {
2036 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2037 return -EINVAL;
2038 }
2039 if (strchr(param->string, '/')) {
2040 ext4_msg(NULL, KERN_ERR,
2041 "quotafile must be on filesystem root");
2042 return -EINVAL;
2043 }
2044 if (ctx->s_qf_names[qtype]) {
2045 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2046 ext4_msg(NULL, KERN_ERR,
2047 "%s quota file already specified",
2048 QTYPE2NAME(qtype));
2049 return -EINVAL;
2050 }
2051 return 0;
2052 }
2053
2054 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2055 if (!qname) {
2056 ext4_msg(NULL, KERN_ERR,
2057 "Not enough memory for storing quotafile name");
2058 return -ENOMEM;
2059 }
2060 ctx->s_qf_names[qtype] = qname;
2061 ctx->qname_spec |= 1 << qtype;
2062 ctx->spec |= EXT4_SPEC_JQUOTA;
2063 return 0;
2064 }
2065
2066 /*
2067 * Clear the name of the specified quota file.
2068 */
2069 static int unnote_qf_name(struct fs_context *fc, int qtype)
2070 {
2071 struct ext4_fs_context *ctx = fc->fs_private;
2072
2073 kfree(ctx->s_qf_names[qtype]);
2074
2075 ctx->s_qf_names[qtype] = NULL;
2076 ctx->qname_spec |= 1 << qtype;
2077 ctx->spec |= EXT4_SPEC_JQUOTA;
2078 return 0;
2079 }
2080 #endif
2081
2082 static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2083 struct ext4_fs_context *ctx)
2084 {
2085 int err;
2086
2087 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2088 ext4_msg(NULL, KERN_WARNING,
2089 "test_dummy_encryption option not supported");
2090 return -EINVAL;
2091 }
2092 err = fscrypt_parse_test_dummy_encryption(param,
2093 &ctx->dummy_enc_policy);
2094 if (err == -EINVAL) {
2095 ext4_msg(NULL, KERN_WARNING,
2096 "Value of option \"%s\" is unrecognized", param->key);
2097 } else if (err == -EEXIST) {
2098 ext4_msg(NULL, KERN_WARNING,
2099 "Conflicting test_dummy_encryption options");
2100 return -EINVAL;
2101 }
2102 return err;
2103 }
2104
2105 #define EXT4_SET_CTX(name) \
2106 static inline __maybe_unused \
2107 void ctx_set_##name(struct ext4_fs_context *ctx, unsigned long flag) \
2108 { \
2109 ctx->mask_s_##name |= flag; \
2110 ctx->vals_s_##name |= flag; \
2111 }
2112
2113 #define EXT4_CLEAR_CTX(name) \
2114 static inline __maybe_unused \
2115 void ctx_clear_##name(struct ext4_fs_context *ctx, unsigned long flag) \
2116 { \
2117 ctx->mask_s_##name |= flag; \
2118 ctx->vals_s_##name &= ~flag; \
2119 }
2120
2121 #define EXT4_TEST_CTX(name) \
2122 static inline unsigned long \
2123 ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \
2124 { \
2125 return (ctx->vals_s_##name & flag); \
2126 }
2127
2128 EXT4_SET_CTX(flags); /* set only */
2129 EXT4_SET_CTX(mount_opt);
2130 EXT4_CLEAR_CTX(mount_opt);
2131 EXT4_TEST_CTX(mount_opt);
2132 EXT4_SET_CTX(mount_opt2);
2133 EXT4_CLEAR_CTX(mount_opt2);
2134 EXT4_TEST_CTX(mount_opt2);
2135
2136 static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2137 {
2138 struct ext4_fs_context *ctx = fc->fs_private;
2139 struct fs_parse_result result;
2140 const struct mount_opts *m;
2141 int is_remount;
2142 int token;
2143
2144 token = fs_parse(fc, ext4_param_specs, param, &result);
2145 if (token < 0)
2146 return token;
2147 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2148
2149 for (m = ext4_mount_opts; m->token != Opt_err; m++)
2150 if (token == m->token)
2151 break;
2152
2153 ctx->opt_flags |= m->flags;
2154
2155 if (m->flags & MOPT_EXPLICIT) {
2156 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2157 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2158 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2159 ctx_set_mount_opt2(ctx,
2160 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2161 } else
2162 return -EINVAL;
2163 }
2164
2165 if (m->flags & MOPT_NOSUPPORT) {
2166 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2167 param->key);
2168 return 0;
2169 }
2170
2171 switch (token) {
2172 #ifdef CONFIG_QUOTA
2173 case Opt_usrjquota:
2174 if (!*param->string)
2175 return unnote_qf_name(fc, USRQUOTA);
2176 else
2177 return note_qf_name(fc, USRQUOTA, param);
2178 case Opt_grpjquota:
2179 if (!*param->string)
2180 return unnote_qf_name(fc, GRPQUOTA);
2181 else
2182 return note_qf_name(fc, GRPQUOTA, param);
2183 #endif
2184 case Opt_sb:
2185 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2186 ext4_msg(NULL, KERN_WARNING,
2187 "Ignoring %s option on remount", param->key);
2188 } else {
2189 ctx->s_sb_block = result.uint_32;
2190 ctx->spec |= EXT4_SPEC_s_sb_block;
2191 }
2192 return 0;
2193 case Opt_removed:
2194 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2195 param->key);
2196 return 0;
2197 case Opt_inlinecrypt:
2198 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2199 ctx_set_flags(ctx, SB_INLINECRYPT);
2200 #else
2201 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2202 #endif
2203 return 0;
2204 case Opt_errors:
2205 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2206 ctx_set_mount_opt(ctx, result.uint_32);
2207 return 0;
2208 #ifdef CONFIG_QUOTA
2209 case Opt_jqfmt:
2210 ctx->s_jquota_fmt = result.uint_32;
2211 ctx->spec |= EXT4_SPEC_JQFMT;
2212 return 0;
2213 #endif
2214 case Opt_data:
2215 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2216 ctx_set_mount_opt(ctx, result.uint_32);
2217 ctx->spec |= EXT4_SPEC_DATAJ;
2218 return 0;
2219 case Opt_commit:
2220 if (result.uint_32 == 0)
2221 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE;
2222 else if (result.uint_32 > INT_MAX / HZ) {
2223 ext4_msg(NULL, KERN_ERR,
2224 "Invalid commit interval %d, "
2225 "must be smaller than %d",
2226 result.uint_32, INT_MAX / HZ);
2227 return -EINVAL;
2228 }
2229 ctx->s_commit_interval = HZ * result.uint_32;
2230 ctx->spec |= EXT4_SPEC_s_commit_interval;
2231 return 0;
2232 case Opt_debug_want_extra_isize:
2233 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2234 ext4_msg(NULL, KERN_ERR,
2235 "Invalid want_extra_isize %d", result.uint_32);
2236 return -EINVAL;
2237 }
2238 ctx->s_want_extra_isize = result.uint_32;
2239 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2240 return 0;
2241 case Opt_max_batch_time:
2242 ctx->s_max_batch_time = result.uint_32;
2243 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2244 return 0;
2245 case Opt_min_batch_time:
2246 ctx->s_min_batch_time = result.uint_32;
2247 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2248 return 0;
2249 case Opt_inode_readahead_blks:
2250 if (result.uint_32 &&
2251 (result.uint_32 > (1 << 30) ||
2252 !is_power_of_2(result.uint_32))) {
2253 ext4_msg(NULL, KERN_ERR,
2254 "EXT4-fs: inode_readahead_blks must be "
2255 "0 or a power of 2 smaller than 2^31");
2256 return -EINVAL;
2257 }
2258 ctx->s_inode_readahead_blks = result.uint_32;
2259 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2260 return 0;
2261 case Opt_init_itable:
2262 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2263 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2264 if (param->type == fs_value_is_string)
2265 ctx->s_li_wait_mult = result.uint_32;
2266 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2267 return 0;
2268 case Opt_max_dir_size_kb:
2269 ctx->s_max_dir_size_kb = result.uint_32;
2270 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2271 return 0;
2272 #ifdef CONFIG_EXT4_DEBUG
2273 case Opt_fc_debug_max_replay:
2274 ctx->s_fc_debug_max_replay = result.uint_32;
2275 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2276 return 0;
2277 #endif
2278 case Opt_stripe:
2279 ctx->s_stripe = result.uint_32;
2280 ctx->spec |= EXT4_SPEC_s_stripe;
2281 return 0;
2282 case Opt_resuid:
2283 ctx->s_resuid = result.uid;
2284 ctx->spec |= EXT4_SPEC_s_resuid;
2285 return 0;
2286 case Opt_resgid:
2287 ctx->s_resgid = result.gid;
2288 ctx->spec |= EXT4_SPEC_s_resgid;
2289 return 0;
2290 case Opt_journal_dev:
2291 if (is_remount) {
2292 ext4_msg(NULL, KERN_ERR,
2293 "Cannot specify journal on remount");
2294 return -EINVAL;
2295 }
2296 ctx->journal_devnum = result.uint_32;
2297 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2298 return 0;
2299 case Opt_journal_path:
2300 {
2301 struct inode *journal_inode;
2302 struct path path;
2303 int error;
2304
2305 if (is_remount) {
2306 ext4_msg(NULL, KERN_ERR,
2307 "Cannot specify journal on remount");
2308 return -EINVAL;
2309 }
2310
2311 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path);
2312 if (error) {
2313 ext4_msg(NULL, KERN_ERR, "error: could not find "
2314 "journal device path");
2315 return -EINVAL;
2316 }
2317
2318 journal_inode = d_inode(path.dentry);
2319 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2320 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2321 path_put(&path);
2322 return 0;
2323 }
2324 case Opt_journal_ioprio:
2325 if (result.uint_32 > 7) {
2326 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2327 " (must be 0-7)");
2328 return -EINVAL;
2329 }
2330 ctx->journal_ioprio =
2331 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2332 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2333 return 0;
2334 case Opt_test_dummy_encryption:
2335 return ext4_parse_test_dummy_encryption(param, ctx);
2336 case Opt_dax:
2337 case Opt_dax_type:
2338 #ifdef CONFIG_FS_DAX
2339 {
2340 int type = (token == Opt_dax) ?
2341 Opt_dax : result.uint_32;
2342
2343 switch (type) {
2344 case Opt_dax:
2345 case Opt_dax_always:
2346 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2347 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2348 break;
2349 case Opt_dax_never:
2350 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2351 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2352 break;
2353 case Opt_dax_inode:
2354 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2355 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2356 /* Strictly for printing options */
2357 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2358 break;
2359 }
2360 return 0;
2361 }
2362 #else
2363 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2364 return -EINVAL;
2365 #endif
2366 case Opt_data_err:
2367 if (result.uint_32 == Opt_data_err_abort)
2368 ctx_set_mount_opt(ctx, m->mount_opt);
2369 else if (result.uint_32 == Opt_data_err_ignore)
2370 ctx_clear_mount_opt(ctx, m->mount_opt);
2371 return 0;
2372 case Opt_mb_optimize_scan:
2373 if (result.int_32 == 1) {
2374 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2375 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2376 } else if (result.int_32 == 0) {
2377 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2378 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2379 } else {
2380 ext4_msg(NULL, KERN_WARNING,
2381 "mb_optimize_scan should be set to 0 or 1.");
2382 return -EINVAL;
2383 }
2384 return 0;
2385 }
2386
2387 /*
2388 * At this point we should only be getting options requiring MOPT_SET,
2389 * or MOPT_CLEAR. Anything else is a bug
2390 */
2391 if (m->token == Opt_err) {
2392 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2393 param->key);
2394 WARN_ON(1);
2395 return -EINVAL;
2396 }
2397
2398 else {
2399 unsigned int set = 0;
2400
2401 if ((param->type == fs_value_is_flag) ||
2402 result.uint_32 > 0)
2403 set = 1;
2404
2405 if (m->flags & MOPT_CLEAR)
2406 set = !set;
2407 else if (unlikely(!(m->flags & MOPT_SET))) {
2408 ext4_msg(NULL, KERN_WARNING,
2409 "buggy handling of option %s",
2410 param->key);
2411 WARN_ON(1);
2412 return -EINVAL;
2413 }
2414 if (m->flags & MOPT_2) {
2415 if (set != 0)
2416 ctx_set_mount_opt2(ctx, m->mount_opt);
2417 else
2418 ctx_clear_mount_opt2(ctx, m->mount_opt);
2419 } else {
2420 if (set != 0)
2421 ctx_set_mount_opt(ctx, m->mount_opt);
2422 else
2423 ctx_clear_mount_opt(ctx, m->mount_opt);
2424 }
2425 }
2426
2427 return 0;
2428 }
2429
2430 static int parse_options(struct fs_context *fc, char *options)
2431 {
2432 struct fs_parameter param;
2433 int ret;
2434 char *key;
2435
2436 if (!options)
2437 return 0;
2438
2439 while ((key = strsep(&options, ",")) != NULL) {
2440 if (*key) {
2441 size_t v_len = 0;
2442 char *value = strchr(key, '=');
2443
2444 param.type = fs_value_is_flag;
2445 param.string = NULL;
2446
2447 if (value) {
2448 if (value == key)
2449 continue;
2450
2451 *value++ = 0;
2452 v_len = strlen(value);
2453 param.string = kmemdup_nul(value, v_len,
2454 GFP_KERNEL);
2455 if (!param.string)
2456 return -ENOMEM;
2457 param.type = fs_value_is_string;
2458 }
2459
2460 param.key = key;
2461 param.size = v_len;
2462
2463 ret = ext4_parse_param(fc, &param);
2464 kfree(param.string);
2465 if (ret < 0)
2466 return ret;
2467 }
2468 }
2469
2470 ret = ext4_validate_options(fc);
2471 if (ret < 0)
2472 return ret;
2473
2474 return 0;
2475 }
2476
2477 static int parse_apply_sb_mount_options(struct super_block *sb,
2478 struct ext4_fs_context *m_ctx)
2479 {
2480 struct ext4_sb_info *sbi = EXT4_SB(sb);
2481 char *s_mount_opts = NULL;
2482 struct ext4_fs_context *s_ctx = NULL;
2483 struct fs_context *fc = NULL;
2484 int ret = -ENOMEM;
2485
2486 if (!sbi->s_es->s_mount_opts[0])
2487 return 0;
2488
2489 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2490 sizeof(sbi->s_es->s_mount_opts),
2491 GFP_KERNEL);
2492 if (!s_mount_opts)
2493 return ret;
2494
2495 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2496 if (!fc)
2497 goto out_free;
2498
2499 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2500 if (!s_ctx)
2501 goto out_free;
2502
2503 fc->fs_private = s_ctx;
2504 fc->s_fs_info = sbi;
2505
2506 ret = parse_options(fc, s_mount_opts);
2507 if (ret < 0)
2508 goto parse_failed;
2509
2510 ret = ext4_check_opt_consistency(fc, sb);
2511 if (ret < 0) {
2512 parse_failed:
2513 ext4_msg(sb, KERN_WARNING,
2514 "failed to parse options in superblock: %s",
2515 s_mount_opts);
2516 ret = 0;
2517 goto out_free;
2518 }
2519
2520 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2521 m_ctx->journal_devnum = s_ctx->journal_devnum;
2522 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2523 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2524
2525 ext4_apply_options(fc, sb);
2526 ret = 0;
2527
2528 out_free:
2529 if (fc) {
2530 ext4_fc_free(fc);
2531 kfree(fc);
2532 }
2533 kfree(s_mount_opts);
2534 return ret;
2535 }
2536
2537 static void ext4_apply_quota_options(struct fs_context *fc,
2538 struct super_block *sb)
2539 {
2540 #ifdef CONFIG_QUOTA
2541 bool quota_feature = ext4_has_feature_quota(sb);
2542 struct ext4_fs_context *ctx = fc->fs_private;
2543 struct ext4_sb_info *sbi = EXT4_SB(sb);
2544 char *qname;
2545 int i;
2546
2547 if (quota_feature)
2548 return;
2549
2550 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2551 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2552 if (!(ctx->qname_spec & (1 << i)))
2553 continue;
2554
2555 qname = ctx->s_qf_names[i]; /* May be NULL */
2556 if (qname)
2557 set_opt(sb, QUOTA);
2558 ctx->s_qf_names[i] = NULL;
2559 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2560 lockdep_is_held(&sb->s_umount));
2561 if (qname)
2562 kfree_rcu_mightsleep(qname);
2563 }
2564 }
2565
2566 if (ctx->spec & EXT4_SPEC_JQFMT)
2567 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2568 #endif
2569 }
2570
2571 /*
2572 * Check quota settings consistency.
2573 */
2574 static int ext4_check_quota_consistency(struct fs_context *fc,
2575 struct super_block *sb)
2576 {
2577 #ifdef CONFIG_QUOTA
2578 struct ext4_fs_context *ctx = fc->fs_private;
2579 struct ext4_sb_info *sbi = EXT4_SB(sb);
2580 bool quota_feature = ext4_has_feature_quota(sb);
2581 bool quota_loaded = sb_any_quota_loaded(sb);
2582 bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2583 int quota_flags, i;
2584
2585 /*
2586 * We do the test below only for project quotas. 'usrquota' and
2587 * 'grpquota' mount options are allowed even without quota feature
2588 * to support legacy quotas in quota files.
2589 */
2590 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2591 !ext4_has_feature_project(sb)) {
2592 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2593 "Cannot enable project quota enforcement.");
2594 return -EINVAL;
2595 }
2596
2597 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2598 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2599 if (quota_loaded &&
2600 ctx->mask_s_mount_opt & quota_flags &&
2601 !ctx_test_mount_opt(ctx, quota_flags))
2602 goto err_quota_change;
2603
2604 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2605
2606 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2607 if (!(ctx->qname_spec & (1 << i)))
2608 continue;
2609
2610 if (quota_loaded &&
2611 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2612 goto err_jquota_change;
2613
2614 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2615 strcmp(get_qf_name(sb, sbi, i),
2616 ctx->s_qf_names[i]) != 0)
2617 goto err_jquota_specified;
2618 }
2619
2620 if (quota_feature) {
2621 ext4_msg(NULL, KERN_INFO,
2622 "Journaled quota options ignored when "
2623 "QUOTA feature is enabled");
2624 return 0;
2625 }
2626 }
2627
2628 if (ctx->spec & EXT4_SPEC_JQFMT) {
2629 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2630 goto err_jquota_change;
2631 if (quota_feature) {
2632 ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2633 "ignored when QUOTA feature is enabled");
2634 return 0;
2635 }
2636 }
2637
2638 /* Make sure we don't mix old and new quota format */
2639 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2640 ctx->s_qf_names[USRQUOTA]);
2641 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2642 ctx->s_qf_names[GRPQUOTA]);
2643
2644 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2645 test_opt(sb, USRQUOTA));
2646
2647 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2648 test_opt(sb, GRPQUOTA));
2649
2650 if (usr_qf_name) {
2651 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2652 usrquota = false;
2653 }
2654 if (grp_qf_name) {
2655 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2656 grpquota = false;
2657 }
2658
2659 if (usr_qf_name || grp_qf_name) {
2660 if (usrquota || grpquota) {
2661 ext4_msg(NULL, KERN_ERR, "old and new quota "
2662 "format mixing");
2663 return -EINVAL;
2664 }
2665
2666 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2667 ext4_msg(NULL, KERN_ERR, "journaled quota format "
2668 "not specified");
2669 return -EINVAL;
2670 }
2671 }
2672
2673 return 0;
2674
2675 err_quota_change:
2676 ext4_msg(NULL, KERN_ERR,
2677 "Cannot change quota options when quota turned on");
2678 return -EINVAL;
2679 err_jquota_change:
2680 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2681 "options when quota turned on");
2682 return -EINVAL;
2683 err_jquota_specified:
2684 ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2685 QTYPE2NAME(i));
2686 return -EINVAL;
2687 #else
2688 return 0;
2689 #endif
2690 }
2691
2692 static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2693 struct super_block *sb)
2694 {
2695 const struct ext4_fs_context *ctx = fc->fs_private;
2696 const struct ext4_sb_info *sbi = EXT4_SB(sb);
2697
2698 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2699 return 0;
2700
2701 if (!ext4_has_feature_encrypt(sb)) {
2702 ext4_msg(NULL, KERN_WARNING,
2703 "test_dummy_encryption requires encrypt feature");
2704 return -EINVAL;
2705 }
2706 /*
2707 * This mount option is just for testing, and it's not worthwhile to
2708 * implement the extra complexity (e.g. RCU protection) that would be
2709 * needed to allow it to be set or changed during remount. We do allow
2710 * it to be specified during remount, but only if there is no change.
2711 */
2712 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2713 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2714 &ctx->dummy_enc_policy))
2715 return 0;
2716 ext4_msg(NULL, KERN_WARNING,
2717 "Can't set or change test_dummy_encryption on remount");
2718 return -EINVAL;
2719 }
2720 /* Also make sure s_mount_opts didn't contain a conflicting value. */
2721 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2722 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2723 &ctx->dummy_enc_policy))
2724 return 0;
2725 ext4_msg(NULL, KERN_WARNING,
2726 "Conflicting test_dummy_encryption options");
2727 return -EINVAL;
2728 }
2729 return 0;
2730 }
2731
2732 static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2733 struct super_block *sb)
2734 {
2735 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2736 /* if already set, it was already verified to be the same */
2737 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2738 return;
2739 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2740 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2741 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2742 }
2743
2744 static int ext4_check_opt_consistency(struct fs_context *fc,
2745 struct super_block *sb)
2746 {
2747 struct ext4_fs_context *ctx = fc->fs_private;
2748 struct ext4_sb_info *sbi = fc->s_fs_info;
2749 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2750 int err;
2751
2752 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2753 ext4_msg(NULL, KERN_ERR,
2754 "Mount option(s) incompatible with ext2");
2755 return -EINVAL;
2756 }
2757 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2758 ext4_msg(NULL, KERN_ERR,
2759 "Mount option(s) incompatible with ext3");
2760 return -EINVAL;
2761 }
2762
2763 if (ctx->s_want_extra_isize >
2764 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2765 ext4_msg(NULL, KERN_ERR,
2766 "Invalid want_extra_isize %d",
2767 ctx->s_want_extra_isize);
2768 return -EINVAL;
2769 }
2770
2771 err = ext4_check_test_dummy_encryption(fc, sb);
2772 if (err)
2773 return err;
2774
2775 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2776 if (!sbi->s_journal) {
2777 ext4_msg(NULL, KERN_WARNING,
2778 "Remounting file system with no journal "
2779 "so ignoring journalled data option");
2780 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2781 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2782 test_opt(sb, DATA_FLAGS)) {
2783 ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2784 "on remount");
2785 return -EINVAL;
2786 }
2787 }
2788
2789 if (is_remount) {
2790 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2791 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2792 ext4_msg(NULL, KERN_ERR, "can't mount with "
2793 "both data=journal and dax");
2794 return -EINVAL;
2795 }
2796
2797 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2798 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2799 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2800 fail_dax_change_remount:
2801 ext4_msg(NULL, KERN_ERR, "can't change "
2802 "dax mount option while remounting");
2803 return -EINVAL;
2804 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2805 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2806 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2807 goto fail_dax_change_remount;
2808 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2809 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2810 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2811 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2812 goto fail_dax_change_remount;
2813 }
2814 }
2815
2816 return ext4_check_quota_consistency(fc, sb);
2817 }
2818
2819 static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2820 {
2821 struct ext4_fs_context *ctx = fc->fs_private;
2822 struct ext4_sb_info *sbi = fc->s_fs_info;
2823
2824 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2825 sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2826 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2827 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2828 sb->s_flags &= ~ctx->mask_s_flags;
2829 sb->s_flags |= ctx->vals_s_flags;
2830
2831 #define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2832 APPLY(s_commit_interval);
2833 APPLY(s_stripe);
2834 APPLY(s_max_batch_time);
2835 APPLY(s_min_batch_time);
2836 APPLY(s_want_extra_isize);
2837 APPLY(s_inode_readahead_blks);
2838 APPLY(s_max_dir_size_kb);
2839 APPLY(s_li_wait_mult);
2840 APPLY(s_resgid);
2841 APPLY(s_resuid);
2842
2843 #ifdef CONFIG_EXT4_DEBUG
2844 APPLY(s_fc_debug_max_replay);
2845 #endif
2846
2847 ext4_apply_quota_options(fc, sb);
2848 ext4_apply_test_dummy_encryption(ctx, sb);
2849 }
2850
2851
2852 static int ext4_validate_options(struct fs_context *fc)
2853 {
2854 #ifdef CONFIG_QUOTA
2855 struct ext4_fs_context *ctx = fc->fs_private;
2856 char *usr_qf_name, *grp_qf_name;
2857
2858 usr_qf_name = ctx->s_qf_names[USRQUOTA];
2859 grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2860
2861 if (usr_qf_name || grp_qf_name) {
2862 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2863 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2864
2865 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2866 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2867
2868 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2869 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2870 ext4_msg(NULL, KERN_ERR, "old and new quota "
2871 "format mixing");
2872 return -EINVAL;
2873 }
2874 }
2875 #endif
2876 return 1;
2877 }
2878
2879 static inline void ext4_show_quota_options(struct seq_file *seq,
2880 struct super_block *sb)
2881 {
2882 #if defined(CONFIG_QUOTA)
2883 struct ext4_sb_info *sbi = EXT4_SB(sb);
2884 char *usr_qf_name, *grp_qf_name;
2885
2886 if (sbi->s_jquota_fmt) {
2887 char *fmtname = "";
2888
2889 switch (sbi->s_jquota_fmt) {
2890 case QFMT_VFS_OLD:
2891 fmtname = "vfsold";
2892 break;
2893 case QFMT_VFS_V0:
2894 fmtname = "vfsv0";
2895 break;
2896 case QFMT_VFS_V1:
2897 fmtname = "vfsv1";
2898 break;
2899 }
2900 seq_printf(seq, ",jqfmt=%s", fmtname);
2901 }
2902
2903 rcu_read_lock();
2904 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2905 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2906 if (usr_qf_name)
2907 seq_show_option(seq, "usrjquota", usr_qf_name);
2908 if (grp_qf_name)
2909 seq_show_option(seq, "grpjquota", grp_qf_name);
2910 rcu_read_unlock();
2911 #endif
2912 }
2913
2914 static const char *token2str(int token)
2915 {
2916 const struct fs_parameter_spec *spec;
2917
2918 for (spec = ext4_param_specs; spec->name != NULL; spec++)
2919 if (spec->opt == token && !spec->type)
2920 break;
2921 return spec->name;
2922 }
2923
2924 /*
2925 * Show an option if
2926 * - it's set to a non-default value OR
2927 * - if the per-sb default is different from the global default
2928 */
2929 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2930 int nodefs)
2931 {
2932 struct ext4_sb_info *sbi = EXT4_SB(sb);
2933 struct ext4_super_block *es = sbi->s_es;
2934 int def_errors;
2935 const struct mount_opts *m;
2936 char sep = nodefs ? '\n' : ',';
2937
2938 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2939 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2940
2941 if (sbi->s_sb_block != 1)
2942 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2943
2944 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2945 int want_set = m->flags & MOPT_SET;
2946 int opt_2 = m->flags & MOPT_2;
2947 unsigned int mount_opt, def_mount_opt;
2948
2949 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2950 m->flags & MOPT_SKIP)
2951 continue;
2952
2953 if (opt_2) {
2954 mount_opt = sbi->s_mount_opt2;
2955 def_mount_opt = sbi->s_def_mount_opt2;
2956 } else {
2957 mount_opt = sbi->s_mount_opt;
2958 def_mount_opt = sbi->s_def_mount_opt;
2959 }
2960 /* skip if same as the default */
2961 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt)))
2962 continue;
2963 /* select Opt_noFoo vs Opt_Foo */
2964 if ((want_set &&
2965 (mount_opt & m->mount_opt) != m->mount_opt) ||
2966 (!want_set && (mount_opt & m->mount_opt)))
2967 continue;
2968 SEQ_OPTS_PRINT("%s", token2str(m->token));
2969 }
2970
2971 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2972 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2973 SEQ_OPTS_PRINT("resuid=%u",
2974 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2975 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2976 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2977 SEQ_OPTS_PRINT("resgid=%u",
2978 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2979 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2980 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2981 SEQ_OPTS_PUTS("errors=remount-ro");
2982 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2983 SEQ_OPTS_PUTS("errors=continue");
2984 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2985 SEQ_OPTS_PUTS("errors=panic");
2986 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2987 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2988 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2989 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2990 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2991 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2992 if (nodefs || sbi->s_stripe)
2993 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2994 if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2995 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
2996 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2997 SEQ_OPTS_PUTS("data=journal");
2998 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2999 SEQ_OPTS_PUTS("data=ordered");
3000 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3001 SEQ_OPTS_PUTS("data=writeback");
3002 }
3003 if (nodefs ||
3004 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3005 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3006 sbi->s_inode_readahead_blks);
3007
3008 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3009 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3010 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3011 if (nodefs || sbi->s_max_dir_size_kb)
3012 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3013 if (test_opt(sb, DATA_ERR_ABORT))
3014 SEQ_OPTS_PUTS("data_err=abort");
3015
3016 fscrypt_show_test_dummy_encryption(seq, sep, sb);
3017
3018 if (sb->s_flags & SB_INLINECRYPT)
3019 SEQ_OPTS_PUTS("inlinecrypt");
3020
3021 if (test_opt(sb, DAX_ALWAYS)) {
3022 if (IS_EXT2_SB(sb))
3023 SEQ_OPTS_PUTS("dax");
3024 else
3025 SEQ_OPTS_PUTS("dax=always");
3026 } else if (test_opt2(sb, DAX_NEVER)) {
3027 SEQ_OPTS_PUTS("dax=never");
3028 } else if (test_opt2(sb, DAX_INODE)) {
3029 SEQ_OPTS_PUTS("dax=inode");
3030 }
3031
3032 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3033 !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3034 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3035 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3036 test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3037 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3038 }
3039
3040 if (nodefs && !test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS))
3041 SEQ_OPTS_PUTS("prefetch_block_bitmaps");
3042
3043 ext4_show_quota_options(seq, sb);
3044 return 0;
3045 }
3046
3047 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3048 {
3049 return _ext4_show_options(seq, root->d_sb, 0);
3050 }
3051
3052 int ext4_seq_options_show(struct seq_file *seq, void *offset)
3053 {
3054 struct super_block *sb = seq->private;
3055 int rc;
3056
3057 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3058 rc = _ext4_show_options(seq, sb, 1);
3059 seq_putc(seq, '\n');
3060 return rc;
3061 }
3062
3063 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3064 int read_only)
3065 {
3066 struct ext4_sb_info *sbi = EXT4_SB(sb);
3067 int err = 0;
3068
3069 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3070 ext4_msg(sb, KERN_ERR, "revision level too high, "
3071 "forcing read-only mode");
3072 err = -EROFS;
3073 goto done;
3074 }
3075 if (read_only)
3076 goto done;
3077 if (!(sbi->s_mount_state & EXT4_VALID_FS))
3078 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3079 "running e2fsck is recommended");
3080 else if (sbi->s_mount_state & EXT4_ERROR_FS)
3081 ext4_msg(sb, KERN_WARNING,
3082 "warning: mounting fs with errors, "
3083 "running e2fsck is recommended");
3084 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3085 le16_to_cpu(es->s_mnt_count) >=
3086 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3087 ext4_msg(sb, KERN_WARNING,
3088 "warning: maximal mount count reached, "
3089 "running e2fsck is recommended");
3090 else if (le32_to_cpu(es->s_checkinterval) &&
3091 (ext4_get_tstamp(es, s_lastcheck) +
3092 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3093 ext4_msg(sb, KERN_WARNING,
3094 "warning: checktime reached, "
3095 "running e2fsck is recommended");
3096 if (!sbi->s_journal)
3097 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3098 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3099 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3100 le16_add_cpu(&es->s_mnt_count, 1);
3101 ext4_update_tstamp(es, s_mtime);
3102 if (sbi->s_journal) {
3103 ext4_set_feature_journal_needs_recovery(sb);
3104 if (ext4_has_feature_orphan_file(sb))
3105 ext4_set_feature_orphan_present(sb);
3106 }
3107
3108 err = ext4_commit_super(sb);
3109 done:
3110 if (test_opt(sb, DEBUG))
3111 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3112 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3113 sb->s_blocksize,
3114 sbi->s_groups_count,
3115 EXT4_BLOCKS_PER_GROUP(sb),
3116 EXT4_INODES_PER_GROUP(sb),
3117 sbi->s_mount_opt, sbi->s_mount_opt2);
3118 return err;
3119 }
3120
3121 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3122 {
3123 struct ext4_sb_info *sbi = EXT4_SB(sb);
3124 struct flex_groups **old_groups, **new_groups;
3125 int size, i, j;
3126
3127 if (!sbi->s_log_groups_per_flex)
3128 return 0;
3129
3130 size = ext4_flex_group(sbi, ngroup - 1) + 1;
3131 if (size <= sbi->s_flex_groups_allocated)
3132 return 0;
3133
3134 new_groups = kvzalloc(roundup_pow_of_two(size *
3135 sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3136 if (!new_groups) {
3137 ext4_msg(sb, KERN_ERR,
3138 "not enough memory for %d flex group pointers", size);
3139 return -ENOMEM;
3140 }
3141 for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3142 new_groups[i] = kvzalloc(roundup_pow_of_two(
3143 sizeof(struct flex_groups)),
3144 GFP_KERNEL);
3145 if (!new_groups[i]) {
3146 for (j = sbi->s_flex_groups_allocated; j < i; j++)
3147 kvfree(new_groups[j]);
3148 kvfree(new_groups);
3149 ext4_msg(sb, KERN_ERR,
3150 "not enough memory for %d flex groups", size);
3151 return -ENOMEM;
3152 }
3153 }
3154 rcu_read_lock();
3155 old_groups = rcu_dereference(sbi->s_flex_groups);
3156 if (old_groups)
3157 memcpy(new_groups, old_groups,
3158 (sbi->s_flex_groups_allocated *
3159 sizeof(struct flex_groups *)));
3160 rcu_read_unlock();
3161 rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3162 sbi->s_flex_groups_allocated = size;
3163 if (old_groups)
3164 ext4_kvfree_array_rcu(old_groups);
3165 return 0;
3166 }
3167
3168 static int ext4_fill_flex_info(struct super_block *sb)
3169 {
3170 struct ext4_sb_info *sbi = EXT4_SB(sb);
3171 struct ext4_group_desc *gdp = NULL;
3172 struct flex_groups *fg;
3173 ext4_group_t flex_group;
3174 int i, err;
3175
3176 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3177 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3178 sbi->s_log_groups_per_flex = 0;
3179 return 1;
3180 }
3181
3182 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3183 if (err)
3184 goto failed;
3185
3186 for (i = 0; i < sbi->s_groups_count; i++) {
3187 gdp = ext4_get_group_desc(sb, i, NULL);
3188
3189 flex_group = ext4_flex_group(sbi, i);
3190 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3191 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3192 atomic64_add(ext4_free_group_clusters(sb, gdp),
3193 &fg->free_clusters);
3194 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3195 }
3196
3197 return 1;
3198 failed:
3199 return 0;
3200 }
3201
3202 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3203 struct ext4_group_desc *gdp)
3204 {
3205 int offset = offsetof(struct ext4_group_desc, bg_checksum);
3206 __u16 crc = 0;
3207 __le32 le_group = cpu_to_le32(block_group);
3208 struct ext4_sb_info *sbi = EXT4_SB(sb);
3209
3210 if (ext4_has_metadata_csum(sbi->s_sb)) {
3211 /* Use new metadata_csum algorithm */
3212 __u32 csum32;
3213 __u16 dummy_csum = 0;
3214
3215 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3216 sizeof(le_group));
3217 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3218 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3219 sizeof(dummy_csum));
3220 offset += sizeof(dummy_csum);
3221 if (offset < sbi->s_desc_size)
3222 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3223 sbi->s_desc_size - offset);
3224
3225 crc = csum32 & 0xFFFF;
3226 goto out;
3227 }
3228
3229 /* old crc16 code */
3230 if (!ext4_has_feature_gdt_csum(sb))
3231 return 0;
3232
3233 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3234 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3235 crc = crc16(crc, (__u8 *)gdp, offset);
3236 offset += sizeof(gdp->bg_checksum); /* skip checksum */
3237 /* for checksum of struct ext4_group_desc do the rest...*/
3238 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size)
3239 crc = crc16(crc, (__u8 *)gdp + offset,
3240 sbi->s_desc_size - offset);
3241
3242 out:
3243 return cpu_to_le16(crc);
3244 }
3245
3246 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3247 struct ext4_group_desc *gdp)
3248 {
3249 if (ext4_has_group_desc_csum(sb) &&
3250 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3251 return 0;
3252
3253 return 1;
3254 }
3255
3256 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3257 struct ext4_group_desc *gdp)
3258 {
3259 if (!ext4_has_group_desc_csum(sb))
3260 return;
3261 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3262 }
3263
3264 /* Called at mount-time, super-block is locked */
3265 static int ext4_check_descriptors(struct super_block *sb,
3266 ext4_fsblk_t sb_block,
3267 ext4_group_t *first_not_zeroed)
3268 {
3269 struct ext4_sb_info *sbi = EXT4_SB(sb);
3270 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3271 ext4_fsblk_t last_block;
3272 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3273 ext4_fsblk_t block_bitmap;
3274 ext4_fsblk_t inode_bitmap;
3275 ext4_fsblk_t inode_table;
3276 int flexbg_flag = 0;
3277 ext4_group_t i, grp = sbi->s_groups_count;
3278
3279 if (ext4_has_feature_flex_bg(sb))
3280 flexbg_flag = 1;
3281
3282 ext4_debug("Checking group descriptors");
3283
3284 for (i = 0; i < sbi->s_groups_count; i++) {
3285 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3286
3287 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3288 last_block = ext4_blocks_count(sbi->s_es) - 1;
3289 else
3290 last_block = first_block +
3291 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3292
3293 if ((grp == sbi->s_groups_count) &&
3294 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3295 grp = i;
3296
3297 block_bitmap = ext4_block_bitmap(sb, gdp);
3298 if (block_bitmap == sb_block) {
3299 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3300 "Block bitmap for group %u overlaps "
3301 "superblock", i);
3302 if (!sb_rdonly(sb))
3303 return 0;
3304 }
3305 if (block_bitmap >= sb_block + 1 &&
3306 block_bitmap <= last_bg_block) {
3307 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3308 "Block bitmap for group %u overlaps "
3309 "block group descriptors", i);
3310 if (!sb_rdonly(sb))
3311 return 0;
3312 }
3313 if (block_bitmap < first_block || block_bitmap > last_block) {
3314 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3315 "Block bitmap for group %u not in group "
3316 "(block %llu)!", i, block_bitmap);
3317 return 0;
3318 }
3319 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3320 if (inode_bitmap == sb_block) {
3321 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3322 "Inode bitmap for group %u overlaps "
3323 "superblock", i);
3324 if (!sb_rdonly(sb))
3325 return 0;
3326 }
3327 if (inode_bitmap >= sb_block + 1 &&
3328 inode_bitmap <= last_bg_block) {
3329 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3330 "Inode bitmap for group %u overlaps "
3331 "block group descriptors", i);
3332 if (!sb_rdonly(sb))
3333 return 0;
3334 }
3335 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3336 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3337 "Inode bitmap for group %u not in group "
3338 "(block %llu)!", i, inode_bitmap);
3339 return 0;
3340 }
3341 inode_table = ext4_inode_table(sb, gdp);
3342 if (inode_table == sb_block) {
3343 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3344 "Inode table for group %u overlaps "
3345 "superblock", i);
3346 if (!sb_rdonly(sb))
3347 return 0;
3348 }
3349 if (inode_table >= sb_block + 1 &&
3350 inode_table <= last_bg_block) {
3351 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3352 "Inode table for group %u overlaps "
3353 "block group descriptors", i);
3354 if (!sb_rdonly(sb))
3355 return 0;
3356 }
3357 if (inode_table < first_block ||
3358 inode_table + sbi->s_itb_per_group - 1 > last_block) {
3359 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3360 "Inode table for group %u not in group "
3361 "(block %llu)!", i, inode_table);
3362 return 0;
3363 }
3364 ext4_lock_group(sb, i);
3365 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3366 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3367 "Checksum for group %u failed (%u!=%u)",
3368 i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3369 gdp)), le16_to_cpu(gdp->bg_checksum));
3370 if (!sb_rdonly(sb)) {
3371 ext4_unlock_group(sb, i);
3372 return 0;
3373 }
3374 }
3375 ext4_unlock_group(sb, i);
3376 if (!flexbg_flag)
3377 first_block += EXT4_BLOCKS_PER_GROUP(sb);
3378 }
3379 if (NULL != first_not_zeroed)
3380 *first_not_zeroed = grp;
3381 return 1;
3382 }
3383
3384 /*
3385 * Maximal extent format file size.
3386 * Resulting logical blkno at s_maxbytes must fit in our on-disk
3387 * extent format containers, within a sector_t, and within i_blocks
3388 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
3389 * so that won't be a limiting factor.
3390 *
3391 * However there is other limiting factor. We do store extents in the form
3392 * of starting block and length, hence the resulting length of the extent
3393 * covering maximum file size must fit into on-disk format containers as
3394 * well. Given that length is always by 1 unit bigger than max unit (because
3395 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3396 *
3397 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3398 */
3399 static loff_t ext4_max_size(int blkbits, int has_huge_files)
3400 {
3401 loff_t res;
3402 loff_t upper_limit = MAX_LFS_FILESIZE;
3403
3404 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3405
3406 if (!has_huge_files) {
3407 upper_limit = (1LL << 32) - 1;
3408
3409 /* total blocks in file system block size */
3410 upper_limit >>= (blkbits - 9);
3411 upper_limit <<= blkbits;
3412 }
3413
3414 /*
3415 * 32-bit extent-start container, ee_block. We lower the maxbytes
3416 * by one fs block, so ee_len can cover the extent of maximum file
3417 * size
3418 */
3419 res = (1LL << 32) - 1;
3420 res <<= blkbits;
3421
3422 /* Sanity check against vm- & vfs- imposed limits */
3423 if (res > upper_limit)
3424 res = upper_limit;
3425
3426 return res;
3427 }
3428
3429 /*
3430 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
3431 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3432 * We need to be 1 filesystem block less than the 2^48 sector limit.
3433 */
3434 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3435 {
3436 loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3437 int meta_blocks;
3438 unsigned int ppb = 1 << (bits - 2);
3439
3440 /*
3441 * This is calculated to be the largest file size for a dense, block
3442 * mapped file such that the file's total number of 512-byte sectors,
3443 * including data and all indirect blocks, does not exceed (2^48 - 1).
3444 *
3445 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3446 * number of 512-byte sectors of the file.
3447 */
3448 if (!has_huge_files) {
3449 /*
3450 * !has_huge_files or implies that the inode i_block field
3451 * represents total file blocks in 2^32 512-byte sectors ==
3452 * size of vfs inode i_blocks * 8
3453 */
3454 upper_limit = (1LL << 32) - 1;
3455
3456 /* total blocks in file system block size */
3457 upper_limit >>= (bits - 9);
3458
3459 } else {
3460 /*
3461 * We use 48 bit ext4_inode i_blocks
3462 * With EXT4_HUGE_FILE_FL set the i_blocks
3463 * represent total number of blocks in
3464 * file system block size
3465 */
3466 upper_limit = (1LL << 48) - 1;
3467
3468 }
3469
3470 /* Compute how many blocks we can address by block tree */
3471 res += ppb;
3472 res += ppb * ppb;
3473 res += ((loff_t)ppb) * ppb * ppb;
3474 /* Compute how many metadata blocks are needed */
3475 meta_blocks = 1;
3476 meta_blocks += 1 + ppb;
3477 meta_blocks += 1 + ppb + ppb * ppb;
3478 /* Does block tree limit file size? */
3479 if (res + meta_blocks <= upper_limit)
3480 goto check_lfs;
3481
3482 res = upper_limit;
3483 /* How many metadata blocks are needed for addressing upper_limit? */
3484 upper_limit -= EXT4_NDIR_BLOCKS;
3485 /* indirect blocks */
3486 meta_blocks = 1;
3487 upper_limit -= ppb;
3488 /* double indirect blocks */
3489 if (upper_limit < ppb * ppb) {
3490 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3491 res -= meta_blocks;
3492 goto check_lfs;
3493 }
3494 meta_blocks += 1 + ppb;
3495 upper_limit -= ppb * ppb;
3496 /* tripple indirect blocks for the rest */
3497 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3498 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3499 res -= meta_blocks;
3500 check_lfs:
3501 res <<= bits;
3502 if (res > MAX_LFS_FILESIZE)
3503 res = MAX_LFS_FILESIZE;
3504
3505 return res;
3506 }
3507
3508 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3509 ext4_fsblk_t logical_sb_block, int nr)
3510 {
3511 struct ext4_sb_info *sbi = EXT4_SB(sb);
3512 ext4_group_t bg, first_meta_bg;
3513 int has_super = 0;
3514
3515 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3516
3517 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3518 return logical_sb_block + nr + 1;
3519 bg = sbi->s_desc_per_block * nr;
3520 if (ext4_bg_has_super(sb, bg))
3521 has_super = 1;
3522
3523 /*
3524 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3525 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled
3526 * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3527 * compensate.
3528 */
3529 if (sb->s_blocksize == 1024 && nr == 0 &&
3530 le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3531 has_super++;
3532
3533 return (has_super + ext4_group_first_block_no(sb, bg));
3534 }
3535
3536 /**
3537 * ext4_get_stripe_size: Get the stripe size.
3538 * @sbi: In memory super block info
3539 *
3540 * If we have specified it via mount option, then
3541 * use the mount option value. If the value specified at mount time is
3542 * greater than the blocks per group use the super block value.
3543 * If the super block value is greater than blocks per group return 0.
3544 * Allocator needs it be less than blocks per group.
3545 *
3546 */
3547 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3548 {
3549 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3550 unsigned long stripe_width =
3551 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3552 int ret;
3553
3554 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3555 ret = sbi->s_stripe;
3556 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3557 ret = stripe_width;
3558 else if (stride && stride <= sbi->s_blocks_per_group)
3559 ret = stride;
3560 else
3561 ret = 0;
3562
3563 /*
3564 * If the stripe width is 1, this makes no sense and
3565 * we set it to 0 to turn off stripe handling code.
3566 */
3567 if (ret <= 1)
3568 ret = 0;
3569
3570 return ret;
3571 }
3572
3573 /*
3574 * Check whether this filesystem can be mounted based on
3575 * the features present and the RDONLY/RDWR mount requested.
3576 * Returns 1 if this filesystem can be mounted as requested,
3577 * 0 if it cannot be.
3578 */
3579 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3580 {
3581 if (ext4_has_unknown_ext4_incompat_features(sb)) {
3582 ext4_msg(sb, KERN_ERR,
3583 "Couldn't mount because of "
3584 "unsupported optional features (%x)",
3585 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3586 ~EXT4_FEATURE_INCOMPAT_SUPP));
3587 return 0;
3588 }
3589
3590 if (!IS_ENABLED(CONFIG_UNICODE) && ext4_has_feature_casefold(sb)) {
3591 ext4_msg(sb, KERN_ERR,
3592 "Filesystem with casefold feature cannot be "
3593 "mounted without CONFIG_UNICODE");
3594 return 0;
3595 }
3596
3597 if (readonly)
3598 return 1;
3599
3600 if (ext4_has_feature_readonly(sb)) {
3601 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3602 sb->s_flags |= SB_RDONLY;
3603 return 1;
3604 }
3605
3606 /* Check that feature set is OK for a read-write mount */
3607 if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3608 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3609 "unsupported optional features (%x)",
3610 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3611 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3612 return 0;
3613 }
3614 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3615 ext4_msg(sb, KERN_ERR,
3616 "Can't support bigalloc feature without "
3617 "extents feature\n");
3618 return 0;
3619 }
3620
3621 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3622 if (!readonly && (ext4_has_feature_quota(sb) ||
3623 ext4_has_feature_project(sb))) {
3624 ext4_msg(sb, KERN_ERR,
3625 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3626 return 0;
3627 }
3628 #endif /* CONFIG_QUOTA */
3629 return 1;
3630 }
3631
3632 /*
3633 * This function is called once a day if we have errors logged
3634 * on the file system
3635 */
3636 static void print_daily_error_info(struct timer_list *t)
3637 {
3638 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3639 struct super_block *sb = sbi->s_sb;
3640 struct ext4_super_block *es = sbi->s_es;
3641
3642 if (es->s_error_count)
3643 /* fsck newer than v1.41.13 is needed to clean this condition. */
3644 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3645 le32_to_cpu(es->s_error_count));
3646 if (es->s_first_error_time) {
3647 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3648 sb->s_id,
3649 ext4_get_tstamp(es, s_first_error_time),
3650 (int) sizeof(es->s_first_error_func),
3651 es->s_first_error_func,
3652 le32_to_cpu(es->s_first_error_line));
3653 if (es->s_first_error_ino)
3654 printk(KERN_CONT ": inode %u",
3655 le32_to_cpu(es->s_first_error_ino));
3656 if (es->s_first_error_block)
3657 printk(KERN_CONT ": block %llu", (unsigned long long)
3658 le64_to_cpu(es->s_first_error_block));
3659 printk(KERN_CONT "\n");
3660 }
3661 if (es->s_last_error_time) {
3662 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3663 sb->s_id,
3664 ext4_get_tstamp(es, s_last_error_time),
3665 (int) sizeof(es->s_last_error_func),
3666 es->s_last_error_func,
3667 le32_to_cpu(es->s_last_error_line));
3668 if (es->s_last_error_ino)
3669 printk(KERN_CONT ": inode %u",
3670 le32_to_cpu(es->s_last_error_ino));
3671 if (es->s_last_error_block)
3672 printk(KERN_CONT ": block %llu", (unsigned long long)
3673 le64_to_cpu(es->s_last_error_block));
3674 printk(KERN_CONT "\n");
3675 }
3676 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
3677 }
3678
3679 /* Find next suitable group and run ext4_init_inode_table */
3680 static int ext4_run_li_request(struct ext4_li_request *elr)
3681 {
3682 struct ext4_group_desc *gdp = NULL;
3683 struct super_block *sb = elr->lr_super;
3684 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3685 ext4_group_t group = elr->lr_next_group;
3686 unsigned int prefetch_ios = 0;
3687 int ret = 0;
3688 int nr = EXT4_SB(sb)->s_mb_prefetch;
3689 u64 start_time;
3690
3691 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3692 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios);
3693 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr);
3694 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr);
3695 if (group >= elr->lr_next_group) {
3696 ret = 1;
3697 if (elr->lr_first_not_zeroed != ngroups &&
3698 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3699 elr->lr_next_group = elr->lr_first_not_zeroed;
3700 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3701 ret = 0;
3702 }
3703 }
3704 return ret;
3705 }
3706
3707 for (; group < ngroups; group++) {
3708 gdp = ext4_get_group_desc(sb, group, NULL);
3709 if (!gdp) {
3710 ret = 1;
3711 break;
3712 }
3713
3714 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3715 break;
3716 }
3717
3718 if (group >= ngroups)
3719 ret = 1;
3720
3721 if (!ret) {
3722 start_time = ktime_get_ns();
3723 ret = ext4_init_inode_table(sb, group,
3724 elr->lr_timeout ? 0 : 1);
3725 trace_ext4_lazy_itable_init(sb, group);
3726 if (elr->lr_timeout == 0) {
3727 elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) *
3728 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3729 }
3730 elr->lr_next_sched = jiffies + elr->lr_timeout;
3731 elr->lr_next_group = group + 1;
3732 }
3733 return ret;
3734 }
3735
3736 /*
3737 * Remove lr_request from the list_request and free the
3738 * request structure. Should be called with li_list_mtx held
3739 */
3740 static void ext4_remove_li_request(struct ext4_li_request *elr)
3741 {
3742 if (!elr)
3743 return;
3744
3745 list_del(&elr->lr_request);
3746 EXT4_SB(elr->lr_super)->s_li_request = NULL;
3747 kfree(elr);
3748 }
3749
3750 static void ext4_unregister_li_request(struct super_block *sb)
3751 {
3752 mutex_lock(&ext4_li_mtx);
3753 if (!ext4_li_info) {
3754 mutex_unlock(&ext4_li_mtx);
3755 return;
3756 }
3757
3758 mutex_lock(&ext4_li_info->li_list_mtx);
3759 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3760 mutex_unlock(&ext4_li_info->li_list_mtx);
3761 mutex_unlock(&ext4_li_mtx);
3762 }
3763
3764 static struct task_struct *ext4_lazyinit_task;
3765
3766 /*
3767 * This is the function where ext4lazyinit thread lives. It walks
3768 * through the request list searching for next scheduled filesystem.
3769 * When such a fs is found, run the lazy initialization request
3770 * (ext4_rn_li_request) and keep track of the time spend in this
3771 * function. Based on that time we compute next schedule time of
3772 * the request. When walking through the list is complete, compute
3773 * next waking time and put itself into sleep.
3774 */
3775 static int ext4_lazyinit_thread(void *arg)
3776 {
3777 struct ext4_lazy_init *eli = arg;
3778 struct list_head *pos, *n;
3779 struct ext4_li_request *elr;
3780 unsigned long next_wakeup, cur;
3781
3782 BUG_ON(NULL == eli);
3783 set_freezable();
3784
3785 cont_thread:
3786 while (true) {
3787 bool next_wakeup_initialized = false;
3788
3789 next_wakeup = 0;
3790 mutex_lock(&eli->li_list_mtx);
3791 if (list_empty(&eli->li_request_list)) {
3792 mutex_unlock(&eli->li_list_mtx);
3793 goto exit_thread;
3794 }
3795 list_for_each_safe(pos, n, &eli->li_request_list) {
3796 int err = 0;
3797 int progress = 0;
3798 elr = list_entry(pos, struct ext4_li_request,
3799 lr_request);
3800
3801 if (time_before(jiffies, elr->lr_next_sched)) {
3802 if (!next_wakeup_initialized ||
3803 time_before(elr->lr_next_sched, next_wakeup)) {
3804 next_wakeup = elr->lr_next_sched;
3805 next_wakeup_initialized = true;
3806 }
3807 continue;
3808 }
3809 if (down_read_trylock(&elr->lr_super->s_umount)) {
3810 if (sb_start_write_trylock(elr->lr_super)) {
3811 progress = 1;
3812 /*
3813 * We hold sb->s_umount, sb can not
3814 * be removed from the list, it is
3815 * now safe to drop li_list_mtx
3816 */
3817 mutex_unlock(&eli->li_list_mtx);
3818 err = ext4_run_li_request(elr);
3819 sb_end_write(elr->lr_super);
3820 mutex_lock(&eli->li_list_mtx);
3821 n = pos->next;
3822 }
3823 up_read((&elr->lr_super->s_umount));
3824 }
3825 /* error, remove the lazy_init job */
3826 if (err) {
3827 ext4_remove_li_request(elr);
3828 continue;
3829 }
3830 if (!progress) {
3831 elr->lr_next_sched = jiffies +
3832 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3833 }
3834 if (!next_wakeup_initialized ||
3835 time_before(elr->lr_next_sched, next_wakeup)) {
3836 next_wakeup = elr->lr_next_sched;
3837 next_wakeup_initialized = true;
3838 }
3839 }
3840 mutex_unlock(&eli->li_list_mtx);
3841
3842 try_to_freeze();
3843
3844 cur = jiffies;
3845 if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) {
3846 cond_resched();
3847 continue;
3848 }
3849
3850 schedule_timeout_interruptible(next_wakeup - cur);
3851
3852 if (kthread_should_stop()) {
3853 ext4_clear_request_list();
3854 goto exit_thread;
3855 }
3856 }
3857
3858 exit_thread:
3859 /*
3860 * It looks like the request list is empty, but we need
3861 * to check it under the li_list_mtx lock, to prevent any
3862 * additions into it, and of course we should lock ext4_li_mtx
3863 * to atomically free the list and ext4_li_info, because at
3864 * this point another ext4 filesystem could be registering
3865 * new one.
3866 */
3867 mutex_lock(&ext4_li_mtx);
3868 mutex_lock(&eli->li_list_mtx);
3869 if (!list_empty(&eli->li_request_list)) {
3870 mutex_unlock(&eli->li_list_mtx);
3871 mutex_unlock(&ext4_li_mtx);
3872 goto cont_thread;
3873 }
3874 mutex_unlock(&eli->li_list_mtx);
3875 kfree(ext4_li_info);
3876 ext4_li_info = NULL;
3877 mutex_unlock(&ext4_li_mtx);
3878
3879 return 0;
3880 }
3881
3882 static void ext4_clear_request_list(void)
3883 {
3884 struct list_head *pos, *n;
3885 struct ext4_li_request *elr;
3886
3887 mutex_lock(&ext4_li_info->li_list_mtx);
3888 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3889 elr = list_entry(pos, struct ext4_li_request,
3890 lr_request);
3891 ext4_remove_li_request(elr);
3892 }
3893 mutex_unlock(&ext4_li_info->li_list_mtx);
3894 }
3895
3896 static int ext4_run_lazyinit_thread(void)
3897 {
3898 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3899 ext4_li_info, "ext4lazyinit");
3900 if (IS_ERR(ext4_lazyinit_task)) {
3901 int err = PTR_ERR(ext4_lazyinit_task);
3902 ext4_clear_request_list();
3903 kfree(ext4_li_info);
3904 ext4_li_info = NULL;
3905 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3906 "initialization thread\n",
3907 err);
3908 return err;
3909 }
3910 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3911 return 0;
3912 }
3913
3914 /*
3915 * Check whether it make sense to run itable init. thread or not.
3916 * If there is at least one uninitialized inode table, return
3917 * corresponding group number, else the loop goes through all
3918 * groups and return total number of groups.
3919 */
3920 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3921 {
3922 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3923 struct ext4_group_desc *gdp = NULL;
3924
3925 if (!ext4_has_group_desc_csum(sb))
3926 return ngroups;
3927
3928 for (group = 0; group < ngroups; group++) {
3929 gdp = ext4_get_group_desc(sb, group, NULL);
3930 if (!gdp)
3931 continue;
3932
3933 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3934 break;
3935 }
3936
3937 return group;
3938 }
3939
3940 static int ext4_li_info_new(void)
3941 {
3942 struct ext4_lazy_init *eli = NULL;
3943
3944 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3945 if (!eli)
3946 return -ENOMEM;
3947
3948 INIT_LIST_HEAD(&eli->li_request_list);
3949 mutex_init(&eli->li_list_mtx);
3950
3951 eli->li_state |= EXT4_LAZYINIT_QUIT;
3952
3953 ext4_li_info = eli;
3954
3955 return 0;
3956 }
3957
3958 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3959 ext4_group_t start)
3960 {
3961 struct ext4_li_request *elr;
3962
3963 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3964 if (!elr)
3965 return NULL;
3966
3967 elr->lr_super = sb;
3968 elr->lr_first_not_zeroed = start;
3969 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3970 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3971 elr->lr_next_group = start;
3972 } else {
3973 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3974 }
3975
3976 /*
3977 * Randomize first schedule time of the request to
3978 * spread the inode table initialization requests
3979 * better.
3980 */
3981 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3982 return elr;
3983 }
3984
3985 int ext4_register_li_request(struct super_block *sb,
3986 ext4_group_t first_not_zeroed)
3987 {
3988 struct ext4_sb_info *sbi = EXT4_SB(sb);
3989 struct ext4_li_request *elr = NULL;
3990 ext4_group_t ngroups = sbi->s_groups_count;
3991 int ret = 0;
3992
3993 mutex_lock(&ext4_li_mtx);
3994 if (sbi->s_li_request != NULL) {
3995 /*
3996 * Reset timeout so it can be computed again, because
3997 * s_li_wait_mult might have changed.
3998 */
3999 sbi->s_li_request->lr_timeout = 0;
4000 goto out;
4001 }
4002
4003 if (sb_rdonly(sb) ||
4004 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4005 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE))))
4006 goto out;
4007
4008 elr = ext4_li_request_new(sb, first_not_zeroed);
4009 if (!elr) {
4010 ret = -ENOMEM;
4011 goto out;
4012 }
4013
4014 if (NULL == ext4_li_info) {
4015 ret = ext4_li_info_new();
4016 if (ret)
4017 goto out;
4018 }
4019
4020 mutex_lock(&ext4_li_info->li_list_mtx);
4021 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4022 mutex_unlock(&ext4_li_info->li_list_mtx);
4023
4024 sbi->s_li_request = elr;
4025 /*
4026 * set elr to NULL here since it has been inserted to
4027 * the request_list and the removal and free of it is
4028 * handled by ext4_clear_request_list from now on.
4029 */
4030 elr = NULL;
4031
4032 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4033 ret = ext4_run_lazyinit_thread();
4034 if (ret)
4035 goto out;
4036 }
4037 out:
4038 mutex_unlock(&ext4_li_mtx);
4039 if (ret)
4040 kfree(elr);
4041 return ret;
4042 }
4043
4044 /*
4045 * We do not need to lock anything since this is called on
4046 * module unload.
4047 */
4048 static void ext4_destroy_lazyinit_thread(void)
4049 {
4050 /*
4051 * If thread exited earlier
4052 * there's nothing to be done.
4053 */
4054 if (!ext4_li_info || !ext4_lazyinit_task)
4055 return;
4056
4057 kthread_stop(ext4_lazyinit_task);
4058 }
4059
4060 static int set_journal_csum_feature_set(struct super_block *sb)
4061 {
4062 int ret = 1;
4063 int compat, incompat;
4064 struct ext4_sb_info *sbi = EXT4_SB(sb);
4065
4066 if (ext4_has_metadata_csum(sb)) {
4067 /* journal checksum v3 */
4068 compat = 0;
4069 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4070 } else {
4071 /* journal checksum v1 */
4072 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4073 incompat = 0;
4074 }
4075
4076 jbd2_journal_clear_features(sbi->s_journal,
4077 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4078 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4079 JBD2_FEATURE_INCOMPAT_CSUM_V2);
4080 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4081 ret = jbd2_journal_set_features(sbi->s_journal,
4082 compat, 0,
4083 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4084 incompat);
4085 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4086 ret = jbd2_journal_set_features(sbi->s_journal,
4087 compat, 0,
4088 incompat);
4089 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4090 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4091 } else {
4092 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4093 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4094 }
4095
4096 return ret;
4097 }
4098
4099 /*
4100 * Note: calculating the overhead so we can be compatible with
4101 * historical BSD practice is quite difficult in the face of
4102 * clusters/bigalloc. This is because multiple metadata blocks from
4103 * different block group can end up in the same allocation cluster.
4104 * Calculating the exact overhead in the face of clustered allocation
4105 * requires either O(all block bitmaps) in memory or O(number of block
4106 * groups**2) in time. We will still calculate the superblock for
4107 * older file systems --- and if we come across with a bigalloc file
4108 * system with zero in s_overhead_clusters the estimate will be close to
4109 * correct especially for very large cluster sizes --- but for newer
4110 * file systems, it's better to calculate this figure once at mkfs
4111 * time, and store it in the superblock. If the superblock value is
4112 * present (even for non-bigalloc file systems), we will use it.
4113 */
4114 static int count_overhead(struct super_block *sb, ext4_group_t grp,
4115 char *buf)
4116 {
4117 struct ext4_sb_info *sbi = EXT4_SB(sb);
4118 struct ext4_group_desc *gdp;
4119 ext4_fsblk_t first_block, last_block, b;
4120 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4121 int s, j, count = 0;
4122 int has_super = ext4_bg_has_super(sb, grp);
4123
4124 if (!ext4_has_feature_bigalloc(sb))
4125 return (has_super + ext4_bg_num_gdb(sb, grp) +
4126 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4127 sbi->s_itb_per_group + 2);
4128
4129 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4130 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4131 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4132 for (i = 0; i < ngroups; i++) {
4133 gdp = ext4_get_group_desc(sb, i, NULL);
4134 b = ext4_block_bitmap(sb, gdp);
4135 if (b >= first_block && b <= last_block) {
4136 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4137 count++;
4138 }
4139 b = ext4_inode_bitmap(sb, gdp);
4140 if (b >= first_block && b <= last_block) {
4141 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4142 count++;
4143 }
4144 b = ext4_inode_table(sb, gdp);
4145 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4146 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4147 int c = EXT4_B2C(sbi, b - first_block);
4148 ext4_set_bit(c, buf);
4149 count++;
4150 }
4151 if (i != grp)
4152 continue;
4153 s = 0;
4154 if (ext4_bg_has_super(sb, grp)) {
4155 ext4_set_bit(s++, buf);
4156 count++;
4157 }
4158 j = ext4_bg_num_gdb(sb, grp);
4159 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4160 ext4_error(sb, "Invalid number of block group "
4161 "descriptor blocks: %d", j);
4162 j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4163 }
4164 count += j;
4165 for (; j > 0; j--)
4166 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4167 }
4168 if (!count)
4169 return 0;
4170 return EXT4_CLUSTERS_PER_GROUP(sb) -
4171 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4172 }
4173
4174 /*
4175 * Compute the overhead and stash it in sbi->s_overhead
4176 */
4177 int ext4_calculate_overhead(struct super_block *sb)
4178 {
4179 struct ext4_sb_info *sbi = EXT4_SB(sb);
4180 struct ext4_super_block *es = sbi->s_es;
4181 struct inode *j_inode;
4182 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4183 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4184 ext4_fsblk_t overhead = 0;
4185 char *buf = (char *) get_zeroed_page(GFP_NOFS);
4186
4187 if (!buf)
4188 return -ENOMEM;
4189
4190 /*
4191 * Compute the overhead (FS structures). This is constant
4192 * for a given filesystem unless the number of block groups
4193 * changes so we cache the previous value until it does.
4194 */
4195
4196 /*
4197 * All of the blocks before first_data_block are overhead
4198 */
4199 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4200
4201 /*
4202 * Add the overhead found in each block group
4203 */
4204 for (i = 0; i < ngroups; i++) {
4205 int blks;
4206
4207 blks = count_overhead(sb, i, buf);
4208 overhead += blks;
4209 if (blks)
4210 memset(buf, 0, PAGE_SIZE);
4211 cond_resched();
4212 }
4213
4214 /*
4215 * Add the internal journal blocks whether the journal has been
4216 * loaded or not
4217 */
4218 if (sbi->s_journal && !sbi->s_journal_bdev_file)
4219 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4220 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4221 /* j_inum for internal journal is non-zero */
4222 j_inode = ext4_get_journal_inode(sb, j_inum);
4223 if (!IS_ERR(j_inode)) {
4224 j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4225 overhead += EXT4_NUM_B2C(sbi, j_blocks);
4226 iput(j_inode);
4227 } else {
4228 ext4_msg(sb, KERN_ERR, "can't get journal size");
4229 }
4230 }
4231 sbi->s_overhead = overhead;
4232 smp_wmb();
4233 free_page((unsigned long) buf);
4234 return 0;
4235 }
4236
4237 static void ext4_set_resv_clusters(struct super_block *sb)
4238 {
4239 ext4_fsblk_t resv_clusters;
4240 struct ext4_sb_info *sbi = EXT4_SB(sb);
4241
4242 /*
4243 * There's no need to reserve anything when we aren't using extents.
4244 * The space estimates are exact, there are no unwritten extents,
4245 * hole punching doesn't need new metadata... This is needed especially
4246 * to keep ext2/3 backward compatibility.
4247 */
4248 if (!ext4_has_feature_extents(sb))
4249 return;
4250 /*
4251 * By default we reserve 2% or 4096 clusters, whichever is smaller.
4252 * This should cover the situations where we can not afford to run
4253 * out of space like for example punch hole, or converting
4254 * unwritten extents in delalloc path. In most cases such
4255 * allocation would require 1, or 2 blocks, higher numbers are
4256 * very rare.
4257 */
4258 resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4259 sbi->s_cluster_bits);
4260
4261 do_div(resv_clusters, 50);
4262 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4263
4264 atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4265 }
4266
4267 static const char *ext4_quota_mode(struct super_block *sb)
4268 {
4269 #ifdef CONFIG_QUOTA
4270 if (!ext4_quota_capable(sb))
4271 return "none";
4272
4273 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4274 return "journalled";
4275 else
4276 return "writeback";
4277 #else
4278 return "disabled";
4279 #endif
4280 }
4281
4282 static void ext4_setup_csum_trigger(struct super_block *sb,
4283 enum ext4_journal_trigger_type type,
4284 void (*trigger)(
4285 struct jbd2_buffer_trigger_type *type,
4286 struct buffer_head *bh,
4287 void *mapped_data,
4288 size_t size))
4289 {
4290 struct ext4_sb_info *sbi = EXT4_SB(sb);
4291
4292 sbi->s_journal_triggers[type].sb = sb;
4293 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4294 }
4295
4296 static void ext4_free_sbi(struct ext4_sb_info *sbi)
4297 {
4298 if (!sbi)
4299 return;
4300
4301 kfree(sbi->s_blockgroup_lock);
4302 fs_put_dax(sbi->s_daxdev, NULL);
4303 kfree(sbi);
4304 }
4305
4306 static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4307 {
4308 struct ext4_sb_info *sbi;
4309
4310 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4311 if (!sbi)
4312 return NULL;
4313
4314 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off,
4315 NULL, NULL);
4316
4317 sbi->s_blockgroup_lock =
4318 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4319
4320 if (!sbi->s_blockgroup_lock)
4321 goto err_out;
4322
4323 sb->s_fs_info = sbi;
4324 sbi->s_sb = sb;
4325 return sbi;
4326 err_out:
4327 fs_put_dax(sbi->s_daxdev, NULL);
4328 kfree(sbi);
4329 return NULL;
4330 }
4331
4332 static void ext4_set_def_opts(struct super_block *sb,
4333 struct ext4_super_block *es)
4334 {
4335 unsigned long def_mount_opts;
4336
4337 /* Set defaults before we parse the mount options */
4338 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4339 set_opt(sb, INIT_INODE_TABLE);
4340 if (def_mount_opts & EXT4_DEFM_DEBUG)
4341 set_opt(sb, DEBUG);
4342 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4343 set_opt(sb, GRPID);
4344 if (def_mount_opts & EXT4_DEFM_UID16)
4345 set_opt(sb, NO_UID32);
4346 /* xattr user namespace & acls are now defaulted on */
4347 set_opt(sb, XATTR_USER);
4348 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4349 set_opt(sb, POSIX_ACL);
4350 #endif
4351 if (ext4_has_feature_fast_commit(sb))
4352 set_opt2(sb, JOURNAL_FAST_COMMIT);
4353 /* don't forget to enable journal_csum when metadata_csum is enabled. */
4354 if (ext4_has_metadata_csum(sb))
4355 set_opt(sb, JOURNAL_CHECKSUM);
4356
4357 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4358 set_opt(sb, JOURNAL_DATA);
4359 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4360 set_opt(sb, ORDERED_DATA);
4361 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4362 set_opt(sb, WRITEBACK_DATA);
4363
4364 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC)
4365 set_opt(sb, ERRORS_PANIC);
4366 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE)
4367 set_opt(sb, ERRORS_CONT);
4368 else
4369 set_opt(sb, ERRORS_RO);
4370 /* block_validity enabled by default; disable with noblock_validity */
4371 set_opt(sb, BLOCK_VALIDITY);
4372 if (def_mount_opts & EXT4_DEFM_DISCARD)
4373 set_opt(sb, DISCARD);
4374
4375 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4376 set_opt(sb, BARRIER);
4377
4378 /*
4379 * enable delayed allocation by default
4380 * Use -o nodelalloc to turn it off
4381 */
4382 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4383 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4384 set_opt(sb, DELALLOC);
4385
4386 if (sb->s_blocksize <= PAGE_SIZE)
4387 set_opt(sb, DIOREAD_NOLOCK);
4388 }
4389
4390 static int ext4_handle_clustersize(struct super_block *sb)
4391 {
4392 struct ext4_sb_info *sbi = EXT4_SB(sb);
4393 struct ext4_super_block *es = sbi->s_es;
4394 int clustersize;
4395
4396 /* Handle clustersize */
4397 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4398 if (ext4_has_feature_bigalloc(sb)) {
4399 if (clustersize < sb->s_blocksize) {
4400 ext4_msg(sb, KERN_ERR,
4401 "cluster size (%d) smaller than "
4402 "block size (%lu)", clustersize, sb->s_blocksize);
4403 return -EINVAL;
4404 }
4405 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4406 le32_to_cpu(es->s_log_block_size);
4407 } else {
4408 if (clustersize != sb->s_blocksize) {
4409 ext4_msg(sb, KERN_ERR,
4410 "fragment/cluster size (%d) != "
4411 "block size (%lu)", clustersize, sb->s_blocksize);
4412 return -EINVAL;
4413 }
4414 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
4415 ext4_msg(sb, KERN_ERR,
4416 "#blocks per group too big: %lu",
4417 sbi->s_blocks_per_group);
4418 return -EINVAL;
4419 }
4420 sbi->s_cluster_bits = 0;
4421 }
4422 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group);
4423 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) {
4424 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu",
4425 sbi->s_clusters_per_group);
4426 return -EINVAL;
4427 }
4428 if (sbi->s_blocks_per_group !=
4429 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) {
4430 ext4_msg(sb, KERN_ERR,
4431 "blocks per group (%lu) and clusters per group (%lu) inconsistent",
4432 sbi->s_blocks_per_group, sbi->s_clusters_per_group);
4433 return -EINVAL;
4434 }
4435 sbi->s_cluster_ratio = clustersize / sb->s_blocksize;
4436
4437 /* Do we have standard group size of clustersize * 8 blocks ? */
4438 if (sbi->s_blocks_per_group == clustersize << 3)
4439 set_opt2(sb, STD_GROUP_SIZE);
4440
4441 return 0;
4442 }
4443
4444 /*
4445 * ext4_atomic_write_init: Initializes filesystem min & max atomic write units.
4446 * @sb: super block
4447 * TODO: Later add support for bigalloc
4448 */
4449 static void ext4_atomic_write_init(struct super_block *sb)
4450 {
4451 struct ext4_sb_info *sbi = EXT4_SB(sb);
4452 struct block_device *bdev = sb->s_bdev;
4453
4454 if (!bdev_can_atomic_write(bdev))
4455 return;
4456
4457 if (!ext4_has_feature_extents(sb))
4458 return;
4459
4460 sbi->s_awu_min = max(sb->s_blocksize,
4461 bdev_atomic_write_unit_min_bytes(bdev));
4462 sbi->s_awu_max = min(sb->s_blocksize,
4463 bdev_atomic_write_unit_max_bytes(bdev));
4464 if (sbi->s_awu_min && sbi->s_awu_max &&
4465 sbi->s_awu_min <= sbi->s_awu_max) {
4466 ext4_msg(sb, KERN_NOTICE, "Supports (experimental) DIO atomic writes awu_min: %u, awu_max: %u",
4467 sbi->s_awu_min, sbi->s_awu_max);
4468 } else {
4469 sbi->s_awu_min = 0;
4470 sbi->s_awu_max = 0;
4471 }
4472 }
4473
4474 static void ext4_fast_commit_init(struct super_block *sb)
4475 {
4476 struct ext4_sb_info *sbi = EXT4_SB(sb);
4477
4478 /* Initialize fast commit stuff */
4479 atomic_set(&sbi->s_fc_subtid, 0);
4480 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4481 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4482 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4483 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4484 sbi->s_fc_bytes = 0;
4485 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4486 sbi->s_fc_ineligible_tid = 0;
4487 spin_lock_init(&sbi->s_fc_lock);
4488 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4489 sbi->s_fc_replay_state.fc_regions = NULL;
4490 sbi->s_fc_replay_state.fc_regions_size = 0;
4491 sbi->s_fc_replay_state.fc_regions_used = 0;
4492 sbi->s_fc_replay_state.fc_regions_valid = 0;
4493 sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4494 sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4495 sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4496 }
4497
4498 static int ext4_inode_info_init(struct super_block *sb,
4499 struct ext4_super_block *es)
4500 {
4501 struct ext4_sb_info *sbi = EXT4_SB(sb);
4502
4503 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4504 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4505 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4506 } else {
4507 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4508 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4509 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4510 ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4511 sbi->s_first_ino);
4512 return -EINVAL;
4513 }
4514 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4515 (!is_power_of_2(sbi->s_inode_size)) ||
4516 (sbi->s_inode_size > sb->s_blocksize)) {
4517 ext4_msg(sb, KERN_ERR,
4518 "unsupported inode size: %d",
4519 sbi->s_inode_size);
4520 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize);
4521 return -EINVAL;
4522 }
4523 /*
4524 * i_atime_extra is the last extra field available for
4525 * [acm]times in struct ext4_inode. Checking for that
4526 * field should suffice to ensure we have extra space
4527 * for all three.
4528 */
4529 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4530 sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4531 sb->s_time_gran = 1;
4532 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4533 } else {
4534 sb->s_time_gran = NSEC_PER_SEC;
4535 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4536 }
4537 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4538 }
4539
4540 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4541 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4542 EXT4_GOOD_OLD_INODE_SIZE;
4543 if (ext4_has_feature_extra_isize(sb)) {
4544 unsigned v, max = (sbi->s_inode_size -
4545 EXT4_GOOD_OLD_INODE_SIZE);
4546
4547 v = le16_to_cpu(es->s_want_extra_isize);
4548 if (v > max) {
4549 ext4_msg(sb, KERN_ERR,
4550 "bad s_want_extra_isize: %d", v);
4551 return -EINVAL;
4552 }
4553 if (sbi->s_want_extra_isize < v)
4554 sbi->s_want_extra_isize = v;
4555
4556 v = le16_to_cpu(es->s_min_extra_isize);
4557 if (v > max) {
4558 ext4_msg(sb, KERN_ERR,
4559 "bad s_min_extra_isize: %d", v);
4560 return -EINVAL;
4561 }
4562 if (sbi->s_want_extra_isize < v)
4563 sbi->s_want_extra_isize = v;
4564 }
4565 }
4566
4567 return 0;
4568 }
4569
4570 #if IS_ENABLED(CONFIG_UNICODE)
4571 static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4572 {
4573 const struct ext4_sb_encodings *encoding_info;
4574 struct unicode_map *encoding;
4575 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4576
4577 if (!ext4_has_feature_casefold(sb) || sb->s_encoding)
4578 return 0;
4579
4580 encoding_info = ext4_sb_read_encoding(es);
4581 if (!encoding_info) {
4582 ext4_msg(sb, KERN_ERR,
4583 "Encoding requested by superblock is unknown");
4584 return -EINVAL;
4585 }
4586
4587 encoding = utf8_load(encoding_info->version);
4588 if (IS_ERR(encoding)) {
4589 ext4_msg(sb, KERN_ERR,
4590 "can't mount with superblock charset: %s-%u.%u.%u "
4591 "not supported by the kernel. flags: 0x%x.",
4592 encoding_info->name,
4593 unicode_major(encoding_info->version),
4594 unicode_minor(encoding_info->version),
4595 unicode_rev(encoding_info->version),
4596 encoding_flags);
4597 return -EINVAL;
4598 }
4599 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4600 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4601 unicode_major(encoding_info->version),
4602 unicode_minor(encoding_info->version),
4603 unicode_rev(encoding_info->version),
4604 encoding_flags);
4605
4606 sb->s_encoding = encoding;
4607 sb->s_encoding_flags = encoding_flags;
4608
4609 return 0;
4610 }
4611 #else
4612 static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4613 {
4614 return 0;
4615 }
4616 #endif
4617
4618 static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es)
4619 {
4620 struct ext4_sb_info *sbi = EXT4_SB(sb);
4621
4622 /* Warn if metadata_csum and gdt_csum are both set. */
4623 if (ext4_has_feature_metadata_csum(sb) &&
4624 ext4_has_feature_gdt_csum(sb))
4625 ext4_warning(sb, "metadata_csum and uninit_bg are "
4626 "redundant flags; please run fsck.");
4627
4628 /* Check for a known checksum algorithm */
4629 if (!ext4_verify_csum_type(sb, es)) {
4630 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4631 "unknown checksum algorithm.");
4632 return -EINVAL;
4633 }
4634 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4635 ext4_orphan_file_block_trigger);
4636
4637 /* Load the checksum driver */
4638 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4639 if (IS_ERR(sbi->s_chksum_driver)) {
4640 int ret = PTR_ERR(sbi->s_chksum_driver);
4641 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4642 sbi->s_chksum_driver = NULL;
4643 return ret;
4644 }
4645
4646 /* Check superblock checksum */
4647 if (!ext4_superblock_csum_verify(sb, es)) {
4648 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4649 "invalid superblock checksum. Run e2fsck?");
4650 return -EFSBADCRC;
4651 }
4652
4653 /* Precompute checksum seed for all metadata */
4654 if (ext4_has_feature_csum_seed(sb))
4655 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4656 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4657 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4658 sizeof(es->s_uuid));
4659 return 0;
4660 }
4661
4662 static int ext4_check_feature_compatibility(struct super_block *sb,
4663 struct ext4_super_block *es,
4664 int silent)
4665 {
4666 struct ext4_sb_info *sbi = EXT4_SB(sb);
4667
4668 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4669 (ext4_has_compat_features(sb) ||
4670 ext4_has_ro_compat_features(sb) ||
4671 ext4_has_incompat_features(sb)))
4672 ext4_msg(sb, KERN_WARNING,
4673 "feature flags set on rev 0 fs, "
4674 "running e2fsck is recommended");
4675
4676 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4677 set_opt2(sb, HURD_COMPAT);
4678 if (ext4_has_feature_64bit(sb)) {
4679 ext4_msg(sb, KERN_ERR,
4680 "The Hurd can't support 64-bit file systems");
4681 return -EINVAL;
4682 }
4683
4684 /*
4685 * ea_inode feature uses l_i_version field which is not
4686 * available in HURD_COMPAT mode.
4687 */
4688 if (ext4_has_feature_ea_inode(sb)) {
4689 ext4_msg(sb, KERN_ERR,
4690 "ea_inode feature is not supported for Hurd");
4691 return -EINVAL;
4692 }
4693 }
4694
4695 if (IS_EXT2_SB(sb)) {
4696 if (ext2_feature_set_ok(sb))
4697 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4698 "using the ext4 subsystem");
4699 else {
4700 /*
4701 * If we're probing be silent, if this looks like
4702 * it's actually an ext[34] filesystem.
4703 */
4704 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4705 return -EINVAL;
4706 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4707 "to feature incompatibilities");
4708 return -EINVAL;
4709 }
4710 }
4711
4712 if (IS_EXT3_SB(sb)) {
4713 if (ext3_feature_set_ok(sb))
4714 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4715 "using the ext4 subsystem");
4716 else {
4717 /*
4718 * If we're probing be silent, if this looks like
4719 * it's actually an ext4 filesystem.
4720 */
4721 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4722 return -EINVAL;
4723 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4724 "to feature incompatibilities");
4725 return -EINVAL;
4726 }
4727 }
4728
4729 /*
4730 * Check feature flags regardless of the revision level, since we
4731 * previously didn't change the revision level when setting the flags,
4732 * so there is a chance incompat flags are set on a rev 0 filesystem.
4733 */
4734 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4735 return -EINVAL;
4736
4737 if (sbi->s_daxdev) {
4738 if (sb->s_blocksize == PAGE_SIZE)
4739 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4740 else
4741 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4742 }
4743
4744 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4745 if (ext4_has_feature_inline_data(sb)) {
4746 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4747 " that may contain inline data");
4748 return -EINVAL;
4749 }
4750 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4751 ext4_msg(sb, KERN_ERR,
4752 "DAX unsupported by block device.");
4753 return -EINVAL;
4754 }
4755 }
4756
4757 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4758 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4759 es->s_encryption_level);
4760 return -EINVAL;
4761 }
4762
4763 return 0;
4764 }
4765
4766 static int ext4_check_geometry(struct super_block *sb,
4767 struct ext4_super_block *es)
4768 {
4769 struct ext4_sb_info *sbi = EXT4_SB(sb);
4770 __u64 blocks_count;
4771 int err;
4772
4773 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) {
4774 ext4_msg(sb, KERN_ERR,
4775 "Number of reserved GDT blocks insanely large: %d",
4776 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4777 return -EINVAL;
4778 }
4779 /*
4780 * Test whether we have more sectors than will fit in sector_t,
4781 * and whether the max offset is addressable by the page cache.
4782 */
4783 err = generic_check_addressable(sb->s_blocksize_bits,
4784 ext4_blocks_count(es));
4785 if (err) {
4786 ext4_msg(sb, KERN_ERR, "filesystem"
4787 " too large to mount safely on this system");
4788 return err;
4789 }
4790
4791 /* check blocks count against device size */
4792 blocks_count = sb_bdev_nr_blocks(sb);
4793 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4794 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4795 "exceeds size of device (%llu blocks)",
4796 ext4_blocks_count(es), blocks_count);
4797 return -EINVAL;
4798 }
4799
4800 /*
4801 * It makes no sense for the first data block to be beyond the end
4802 * of the filesystem.
4803 */
4804 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4805 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4806 "block %u is beyond end of filesystem (%llu)",
4807 le32_to_cpu(es->s_first_data_block),
4808 ext4_blocks_count(es));
4809 return -EINVAL;
4810 }
4811 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4812 (sbi->s_cluster_ratio == 1)) {
4813 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4814 "block is 0 with a 1k block and cluster size");
4815 return -EINVAL;
4816 }
4817
4818 blocks_count = (ext4_blocks_count(es) -
4819 le32_to_cpu(es->s_first_data_block) +
4820 EXT4_BLOCKS_PER_GROUP(sb) - 1);
4821 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4822 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4823 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4824 "(block count %llu, first data block %u, "
4825 "blocks per group %lu)", blocks_count,
4826 ext4_blocks_count(es),
4827 le32_to_cpu(es->s_first_data_block),
4828 EXT4_BLOCKS_PER_GROUP(sb));
4829 return -EINVAL;
4830 }
4831 sbi->s_groups_count = blocks_count;
4832 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4833 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4834 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4835 le32_to_cpu(es->s_inodes_count)) {
4836 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4837 le32_to_cpu(es->s_inodes_count),
4838 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4839 return -EINVAL;
4840 }
4841
4842 return 0;
4843 }
4844
4845 static int ext4_group_desc_init(struct super_block *sb,
4846 struct ext4_super_block *es,
4847 ext4_fsblk_t logical_sb_block,
4848 ext4_group_t *first_not_zeroed)
4849 {
4850 struct ext4_sb_info *sbi = EXT4_SB(sb);
4851 unsigned int db_count;
4852 ext4_fsblk_t block;
4853 int i;
4854
4855 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4856 EXT4_DESC_PER_BLOCK(sb);
4857 if (ext4_has_feature_meta_bg(sb)) {
4858 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4859 ext4_msg(sb, KERN_WARNING,
4860 "first meta block group too large: %u "
4861 "(group descriptor block count %u)",
4862 le32_to_cpu(es->s_first_meta_bg), db_count);
4863 return -EINVAL;
4864 }
4865 }
4866 rcu_assign_pointer(sbi->s_group_desc,
4867 kvmalloc_array(db_count,
4868 sizeof(struct buffer_head *),
4869 GFP_KERNEL));
4870 if (sbi->s_group_desc == NULL) {
4871 ext4_msg(sb, KERN_ERR, "not enough memory");
4872 return -ENOMEM;
4873 }
4874
4875 bgl_lock_init(sbi->s_blockgroup_lock);
4876
4877 /* Pre-read the descriptors into the buffer cache */
4878 for (i = 0; i < db_count; i++) {
4879 block = descriptor_loc(sb, logical_sb_block, i);
4880 ext4_sb_breadahead_unmovable(sb, block);
4881 }
4882
4883 for (i = 0; i < db_count; i++) {
4884 struct buffer_head *bh;
4885
4886 block = descriptor_loc(sb, logical_sb_block, i);
4887 bh = ext4_sb_bread_unmovable(sb, block);
4888 if (IS_ERR(bh)) {
4889 ext4_msg(sb, KERN_ERR,
4890 "can't read group descriptor %d", i);
4891 sbi->s_gdb_count = i;
4892 return PTR_ERR(bh);
4893 }
4894 rcu_read_lock();
4895 rcu_dereference(sbi->s_group_desc)[i] = bh;
4896 rcu_read_unlock();
4897 }
4898 sbi->s_gdb_count = db_count;
4899 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) {
4900 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4901 return -EFSCORRUPTED;
4902 }
4903
4904 return 0;
4905 }
4906
4907 static int ext4_load_and_init_journal(struct super_block *sb,
4908 struct ext4_super_block *es,
4909 struct ext4_fs_context *ctx)
4910 {
4911 struct ext4_sb_info *sbi = EXT4_SB(sb);
4912 int err;
4913
4914 err = ext4_load_journal(sb, es, ctx->journal_devnum);
4915 if (err)
4916 return err;
4917
4918 if (ext4_has_feature_64bit(sb) &&
4919 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4920 JBD2_FEATURE_INCOMPAT_64BIT)) {
4921 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4922 goto out;
4923 }
4924
4925 if (!set_journal_csum_feature_set(sb)) {
4926 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4927 "feature set");
4928 goto out;
4929 }
4930
4931 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4932 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4933 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4934 ext4_msg(sb, KERN_ERR,
4935 "Failed to set fast commit journal feature");
4936 goto out;
4937 }
4938
4939 /* We have now updated the journal if required, so we can
4940 * validate the data journaling mode. */
4941 switch (test_opt(sb, DATA_FLAGS)) {
4942 case 0:
4943 /* No mode set, assume a default based on the journal
4944 * capabilities: ORDERED_DATA if the journal can
4945 * cope, else JOURNAL_DATA
4946 */
4947 if (jbd2_journal_check_available_features
4948 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4949 set_opt(sb, ORDERED_DATA);
4950 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4951 } else {
4952 set_opt(sb, JOURNAL_DATA);
4953 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4954 }
4955 break;
4956
4957 case EXT4_MOUNT_ORDERED_DATA:
4958 case EXT4_MOUNT_WRITEBACK_DATA:
4959 if (!jbd2_journal_check_available_features
4960 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4961 ext4_msg(sb, KERN_ERR, "Journal does not support "
4962 "requested data journaling mode");
4963 goto out;
4964 }
4965 break;
4966 default:
4967 break;
4968 }
4969
4970 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4971 test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4972 ext4_msg(sb, KERN_ERR, "can't mount with "
4973 "journal_async_commit in data=ordered mode");
4974 goto out;
4975 }
4976
4977 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
4978
4979 sbi->s_journal->j_submit_inode_data_buffers =
4980 ext4_journal_submit_inode_data_buffers;
4981 sbi->s_journal->j_finish_inode_data_buffers =
4982 ext4_journal_finish_inode_data_buffers;
4983
4984 return 0;
4985
4986 out:
4987 /* flush s_sb_upd_work before destroying the journal. */
4988 flush_work(&sbi->s_sb_upd_work);
4989 jbd2_journal_destroy(sbi->s_journal);
4990 sbi->s_journal = NULL;
4991 return -EINVAL;
4992 }
4993
4994 static int ext4_check_journal_data_mode(struct super_block *sb)
4995 {
4996 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4997 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with "
4998 "data=journal disables delayed allocation, "
4999 "dioread_nolock, O_DIRECT and fast_commit support!\n");
5000 /* can't mount with both data=journal and dioread_nolock. */
5001 clear_opt(sb, DIOREAD_NOLOCK);
5002 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5003 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5004 ext4_msg(sb, KERN_ERR, "can't mount with "
5005 "both data=journal and delalloc");
5006 return -EINVAL;
5007 }
5008 if (test_opt(sb, DAX_ALWAYS)) {
5009 ext4_msg(sb, KERN_ERR, "can't mount with "
5010 "both data=journal and dax");
5011 return -EINVAL;
5012 }
5013 if (ext4_has_feature_encrypt(sb)) {
5014 ext4_msg(sb, KERN_WARNING,
5015 "encrypted files will use data=ordered "
5016 "instead of data journaling mode");
5017 }
5018 if (test_opt(sb, DELALLOC))
5019 clear_opt(sb, DELALLOC);
5020 } else {
5021 sb->s_iflags |= SB_I_CGROUPWB;
5022 }
5023
5024 return 0;
5025 }
5026
5027 static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb,
5028 int silent)
5029 {
5030 struct ext4_sb_info *sbi = EXT4_SB(sb);
5031 struct ext4_super_block *es;
5032 ext4_fsblk_t logical_sb_block;
5033 unsigned long offset = 0;
5034 struct buffer_head *bh;
5035 int ret = -EINVAL;
5036 int blocksize;
5037
5038 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
5039 if (!blocksize) {
5040 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
5041 return -EINVAL;
5042 }
5043
5044 /*
5045 * The ext4 superblock will not be buffer aligned for other than 1kB
5046 * block sizes. We need to calculate the offset from buffer start.
5047 */
5048 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
5049 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5050 offset = do_div(logical_sb_block, blocksize);
5051 } else {
5052 logical_sb_block = sbi->s_sb_block;
5053 }
5054
5055 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5056 if (IS_ERR(bh)) {
5057 ext4_msg(sb, KERN_ERR, "unable to read superblock");
5058 return PTR_ERR(bh);
5059 }
5060 /*
5061 * Note: s_es must be initialized as soon as possible because
5062 * some ext4 macro-instructions depend on its value
5063 */
5064 es = (struct ext4_super_block *) (bh->b_data + offset);
5065 sbi->s_es = es;
5066 sb->s_magic = le16_to_cpu(es->s_magic);
5067 if (sb->s_magic != EXT4_SUPER_MAGIC) {
5068 if (!silent)
5069 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5070 goto out;
5071 }
5072
5073 if (le32_to_cpu(es->s_log_block_size) >
5074 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5075 ext4_msg(sb, KERN_ERR,
5076 "Invalid log block size: %u",
5077 le32_to_cpu(es->s_log_block_size));
5078 goto out;
5079 }
5080 if (le32_to_cpu(es->s_log_cluster_size) >
5081 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5082 ext4_msg(sb, KERN_ERR,
5083 "Invalid log cluster size: %u",
5084 le32_to_cpu(es->s_log_cluster_size));
5085 goto out;
5086 }
5087
5088 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
5089
5090 /*
5091 * If the default block size is not the same as the real block size,
5092 * we need to reload it.
5093 */
5094 if (sb->s_blocksize == blocksize) {
5095 *lsb = logical_sb_block;
5096 sbi->s_sbh = bh;
5097 return 0;
5098 }
5099
5100 /*
5101 * bh must be released before kill_bdev(), otherwise
5102 * it won't be freed and its page also. kill_bdev()
5103 * is called by sb_set_blocksize().
5104 */
5105 brelse(bh);
5106 /* Validate the filesystem blocksize */
5107 if (!sb_set_blocksize(sb, blocksize)) {
5108 ext4_msg(sb, KERN_ERR, "bad block size %d",
5109 blocksize);
5110 bh = NULL;
5111 goto out;
5112 }
5113
5114 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5115 offset = do_div(logical_sb_block, blocksize);
5116 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5117 if (IS_ERR(bh)) {
5118 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try");
5119 ret = PTR_ERR(bh);
5120 bh = NULL;
5121 goto out;
5122 }
5123 es = (struct ext4_super_block *)(bh->b_data + offset);
5124 sbi->s_es = es;
5125 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
5126 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!");
5127 goto out;
5128 }
5129 *lsb = logical_sb_block;
5130 sbi->s_sbh = bh;
5131 return 0;
5132 out:
5133 brelse(bh);
5134 return ret;
5135 }
5136
5137 static int ext4_hash_info_init(struct super_block *sb)
5138 {
5139 struct ext4_sb_info *sbi = EXT4_SB(sb);
5140 struct ext4_super_block *es = sbi->s_es;
5141 unsigned int i;
5142
5143 sbi->s_def_hash_version = es->s_def_hash_version;
5144
5145 if (sbi->s_def_hash_version > DX_HASH_LAST) {
5146 ext4_msg(sb, KERN_ERR,
5147 "Invalid default hash set in the superblock");
5148 return -EINVAL;
5149 } else if (sbi->s_def_hash_version == DX_HASH_SIPHASH) {
5150 ext4_msg(sb, KERN_ERR,
5151 "SIPHASH is not a valid default hash value");
5152 return -EINVAL;
5153 }
5154
5155 for (i = 0; i < 4; i++)
5156 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
5157
5158 if (ext4_has_feature_dir_index(sb)) {
5159 i = le32_to_cpu(es->s_flags);
5160 if (i & EXT2_FLAGS_UNSIGNED_HASH)
5161 sbi->s_hash_unsigned = 3;
5162 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
5163 #ifdef __CHAR_UNSIGNED__
5164 if (!sb_rdonly(sb))
5165 es->s_flags |=
5166 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
5167 sbi->s_hash_unsigned = 3;
5168 #else
5169 if (!sb_rdonly(sb))
5170 es->s_flags |=
5171 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
5172 #endif
5173 }
5174 }
5175 return 0;
5176 }
5177
5178 static int ext4_block_group_meta_init(struct super_block *sb, int silent)
5179 {
5180 struct ext4_sb_info *sbi = EXT4_SB(sb);
5181 struct ext4_super_block *es = sbi->s_es;
5182 int has_huge_files;
5183
5184 has_huge_files = ext4_has_feature_huge_file(sb);
5185 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
5186 has_huge_files);
5187 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
5188
5189 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
5190 if (ext4_has_feature_64bit(sb)) {
5191 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
5192 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
5193 !is_power_of_2(sbi->s_desc_size)) {
5194 ext4_msg(sb, KERN_ERR,
5195 "unsupported descriptor size %lu",
5196 sbi->s_desc_size);
5197 return -EINVAL;
5198 }
5199 } else
5200 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
5201
5202 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
5203 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
5204
5205 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb);
5206 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) {
5207 if (!silent)
5208 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5209 return -EINVAL;
5210 }
5211 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
5212 sbi->s_inodes_per_group > sb->s_blocksize * 8) {
5213 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
5214 sbi->s_inodes_per_group);
5215 return -EINVAL;
5216 }
5217 sbi->s_itb_per_group = sbi->s_inodes_per_group /
5218 sbi->s_inodes_per_block;
5219 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb);
5220 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
5221 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
5222 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
5223
5224 return 0;
5225 }
5226
5227 /*
5228 * It's hard to get stripe aligned blocks if stripe is not aligned with
5229 * cluster, just disable stripe and alert user to simplify code and avoid
5230 * stripe aligned allocation which will rarely succeed.
5231 */
5232 static bool ext4_is_stripe_incompatible(struct super_block *sb, unsigned long stripe)
5233 {
5234 struct ext4_sb_info *sbi = EXT4_SB(sb);
5235 return (stripe > 0 && sbi->s_cluster_ratio > 1 &&
5236 stripe % sbi->s_cluster_ratio != 0);
5237 }
5238
5239 static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
5240 {
5241 struct ext4_super_block *es = NULL;
5242 struct ext4_sb_info *sbi = EXT4_SB(sb);
5243 ext4_fsblk_t logical_sb_block;
5244 struct inode *root;
5245 int needs_recovery;
5246 int err;
5247 ext4_group_t first_not_zeroed;
5248 struct ext4_fs_context *ctx = fc->fs_private;
5249 int silent = fc->sb_flags & SB_SILENT;
5250
5251 /* Set defaults for the variables that will be set during parsing */
5252 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
5253 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5254
5255 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
5256 sbi->s_sectors_written_start =
5257 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
5258
5259 err = ext4_load_super(sb, &logical_sb_block, silent);
5260 if (err)
5261 goto out_fail;
5262
5263 es = sbi->s_es;
5264 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
5265
5266 err = ext4_init_metadata_csum(sb, es);
5267 if (err)
5268 goto failed_mount;
5269
5270 ext4_set_def_opts(sb, es);
5271
5272 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
5273 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
5274 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
5275 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
5276 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
5277
5278 /*
5279 * set default s_li_wait_mult for lazyinit, for the case there is
5280 * no mount option specified.
5281 */
5282 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
5283
5284 err = ext4_inode_info_init(sb, es);
5285 if (err)
5286 goto failed_mount;
5287
5288 err = parse_apply_sb_mount_options(sb, ctx);
5289 if (err < 0)
5290 goto failed_mount;
5291
5292 sbi->s_def_mount_opt = sbi->s_mount_opt;
5293 sbi->s_def_mount_opt2 = sbi->s_mount_opt2;
5294
5295 err = ext4_check_opt_consistency(fc, sb);
5296 if (err < 0)
5297 goto failed_mount;
5298
5299 ext4_apply_options(fc, sb);
5300
5301 err = ext4_encoding_init(sb, es);
5302 if (err)
5303 goto failed_mount;
5304
5305 err = ext4_check_journal_data_mode(sb);
5306 if (err)
5307 goto failed_mount;
5308
5309 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5310 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5311
5312 /* i_version is always enabled now */
5313 sb->s_flags |= SB_I_VERSION;
5314
5315 err = ext4_check_feature_compatibility(sb, es, silent);
5316 if (err)
5317 goto failed_mount;
5318
5319 err = ext4_block_group_meta_init(sb, silent);
5320 if (err)
5321 goto failed_mount;
5322
5323 err = ext4_hash_info_init(sb);
5324 if (err)
5325 goto failed_mount;
5326
5327 err = ext4_handle_clustersize(sb);
5328 if (err)
5329 goto failed_mount;
5330
5331 err = ext4_check_geometry(sb, es);
5332 if (err)
5333 goto failed_mount;
5334
5335 timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5336 spin_lock_init(&sbi->s_error_lock);
5337 INIT_WORK(&sbi->s_sb_upd_work, update_super_work);
5338
5339 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed);
5340 if (err)
5341 goto failed_mount3;
5342
5343 err = ext4_es_register_shrinker(sbi);
5344 if (err)
5345 goto failed_mount3;
5346
5347 sbi->s_stripe = ext4_get_stripe_size(sbi);
5348 if (ext4_is_stripe_incompatible(sb, sbi->s_stripe)) {
5349 ext4_msg(sb, KERN_WARNING,
5350 "stripe (%lu) is not aligned with cluster size (%u), "
5351 "stripe is disabled",
5352 sbi->s_stripe, sbi->s_cluster_ratio);
5353 sbi->s_stripe = 0;
5354 }
5355 sbi->s_extent_max_zeroout_kb = 32;
5356
5357 /*
5358 * set up enough so that it can read an inode
5359 */
5360 sb->s_op = &ext4_sops;
5361 sb->s_export_op = &ext4_export_ops;
5362 sb->s_xattr = ext4_xattr_handlers;
5363 #ifdef CONFIG_FS_ENCRYPTION
5364 sb->s_cop = &ext4_cryptops;
5365 #endif
5366 #ifdef CONFIG_FS_VERITY
5367 sb->s_vop = &ext4_verityops;
5368 #endif
5369 #ifdef CONFIG_QUOTA
5370 sb->dq_op = &ext4_quota_operations;
5371 if (ext4_has_feature_quota(sb))
5372 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5373 else
5374 sb->s_qcop = &ext4_qctl_operations;
5375 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5376 #endif
5377 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid));
5378 super_set_sysfs_name_bdev(sb);
5379
5380 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5381 mutex_init(&sbi->s_orphan_lock);
5382
5383 spin_lock_init(&sbi->s_bdev_wb_lock);
5384
5385 ext4_atomic_write_init(sb);
5386 ext4_fast_commit_init(sb);
5387
5388 sb->s_root = NULL;
5389
5390 needs_recovery = (es->s_last_orphan != 0 ||
5391 ext4_has_feature_orphan_present(sb) ||
5392 ext4_has_feature_journal_needs_recovery(sb));
5393
5394 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) {
5395 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block));
5396 if (err)
5397 goto failed_mount3a;
5398 }
5399
5400 err = -EINVAL;
5401 /*
5402 * The first inode we look at is the journal inode. Don't try
5403 * root first: it may be modified in the journal!
5404 */
5405 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5406 err = ext4_load_and_init_journal(sb, es, ctx);
5407 if (err)
5408 goto failed_mount3a;
5409 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5410 ext4_has_feature_journal_needs_recovery(sb)) {
5411 ext4_msg(sb, KERN_ERR, "required journal recovery "
5412 "suppressed and not mounted read-only");
5413 goto failed_mount3a;
5414 } else {
5415 /* Nojournal mode, all journal mount options are illegal */
5416 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5417 ext4_msg(sb, KERN_ERR, "can't mount with "
5418 "journal_async_commit, fs mounted w/o journal");
5419 goto failed_mount3a;
5420 }
5421
5422 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5423 ext4_msg(sb, KERN_ERR, "can't mount with "
5424 "journal_checksum, fs mounted w/o journal");
5425 goto failed_mount3a;
5426 }
5427 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5428 ext4_msg(sb, KERN_ERR, "can't mount with "
5429 "commit=%lu, fs mounted w/o journal",
5430 sbi->s_commit_interval / HZ);
5431 goto failed_mount3a;
5432 }
5433 if (EXT4_MOUNT_DATA_FLAGS &
5434 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5435 ext4_msg(sb, KERN_ERR, "can't mount with "
5436 "data=, fs mounted w/o journal");
5437 goto failed_mount3a;
5438 }
5439 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5440 clear_opt(sb, JOURNAL_CHECKSUM);
5441 clear_opt(sb, DATA_FLAGS);
5442 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5443 sbi->s_journal = NULL;
5444 needs_recovery = 0;
5445 }
5446
5447 if (!test_opt(sb, NO_MBCACHE)) {
5448 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5449 if (!sbi->s_ea_block_cache) {
5450 ext4_msg(sb, KERN_ERR,
5451 "Failed to create ea_block_cache");
5452 err = -EINVAL;
5453 goto failed_mount_wq;
5454 }
5455
5456 if (ext4_has_feature_ea_inode(sb)) {
5457 sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5458 if (!sbi->s_ea_inode_cache) {
5459 ext4_msg(sb, KERN_ERR,
5460 "Failed to create ea_inode_cache");
5461 err = -EINVAL;
5462 goto failed_mount_wq;
5463 }
5464 }
5465 }
5466
5467 /*
5468 * Get the # of file system overhead blocks from the
5469 * superblock if present.
5470 */
5471 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5472 /* ignore the precalculated value if it is ridiculous */
5473 if (sbi->s_overhead > ext4_blocks_count(es))
5474 sbi->s_overhead = 0;
5475 /*
5476 * If the bigalloc feature is not enabled recalculating the
5477 * overhead doesn't take long, so we might as well just redo
5478 * it to make sure we are using the correct value.
5479 */
5480 if (!ext4_has_feature_bigalloc(sb))
5481 sbi->s_overhead = 0;
5482 if (sbi->s_overhead == 0) {
5483 err = ext4_calculate_overhead(sb);
5484 if (err)
5485 goto failed_mount_wq;
5486 }
5487
5488 /*
5489 * The maximum number of concurrent works can be high and
5490 * concurrency isn't really necessary. Limit it to 1.
5491 */
5492 EXT4_SB(sb)->rsv_conversion_wq =
5493 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5494 if (!EXT4_SB(sb)->rsv_conversion_wq) {
5495 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5496 err = -ENOMEM;
5497 goto failed_mount4;
5498 }
5499
5500 /*
5501 * The jbd2_journal_load will have done any necessary log recovery,
5502 * so we can safely mount the rest of the filesystem now.
5503 */
5504
5505 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5506 if (IS_ERR(root)) {
5507 ext4_msg(sb, KERN_ERR, "get root inode failed");
5508 err = PTR_ERR(root);
5509 root = NULL;
5510 goto failed_mount4;
5511 }
5512 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5513 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5514 iput(root);
5515 err = -EFSCORRUPTED;
5516 goto failed_mount4;
5517 }
5518
5519 generic_set_sb_d_ops(sb);
5520 sb->s_root = d_make_root(root);
5521 if (!sb->s_root) {
5522 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5523 err = -ENOMEM;
5524 goto failed_mount4;
5525 }
5526
5527 err = ext4_setup_super(sb, es, sb_rdonly(sb));
5528 if (err == -EROFS) {
5529 sb->s_flags |= SB_RDONLY;
5530 } else if (err)
5531 goto failed_mount4a;
5532
5533 ext4_set_resv_clusters(sb);
5534
5535 if (test_opt(sb, BLOCK_VALIDITY)) {
5536 err = ext4_setup_system_zone(sb);
5537 if (err) {
5538 ext4_msg(sb, KERN_ERR, "failed to initialize system "
5539 "zone (%d)", err);
5540 goto failed_mount4a;
5541 }
5542 }
5543 ext4_fc_replay_cleanup(sb);
5544
5545 ext4_ext_init(sb);
5546
5547 /*
5548 * Enable optimize_scan if number of groups is > threshold. This can be
5549 * turned off by passing "mb_optimize_scan=0". This can also be
5550 * turned on forcefully by passing "mb_optimize_scan=1".
5551 */
5552 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5553 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5554 set_opt2(sb, MB_OPTIMIZE_SCAN);
5555 else
5556 clear_opt2(sb, MB_OPTIMIZE_SCAN);
5557 }
5558
5559 err = ext4_mb_init(sb);
5560 if (err) {
5561 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5562 err);
5563 goto failed_mount5;
5564 }
5565
5566 /*
5567 * We can only set up the journal commit callback once
5568 * mballoc is initialized
5569 */
5570 if (sbi->s_journal)
5571 sbi->s_journal->j_commit_callback =
5572 ext4_journal_commit_callback;
5573
5574 err = ext4_percpu_param_init(sbi);
5575 if (err)
5576 goto failed_mount6;
5577
5578 if (ext4_has_feature_flex_bg(sb))
5579 if (!ext4_fill_flex_info(sb)) {
5580 ext4_msg(sb, KERN_ERR,
5581 "unable to initialize "
5582 "flex_bg meta info!");
5583 err = -ENOMEM;
5584 goto failed_mount6;
5585 }
5586
5587 err = ext4_register_li_request(sb, first_not_zeroed);
5588 if (err)
5589 goto failed_mount6;
5590
5591 err = ext4_init_orphan_info(sb);
5592 if (err)
5593 goto failed_mount7;
5594 #ifdef CONFIG_QUOTA
5595 /* Enable quota usage during mount. */
5596 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5597 err = ext4_enable_quotas(sb);
5598 if (err)
5599 goto failed_mount8;
5600 }
5601 #endif /* CONFIG_QUOTA */
5602
5603 /*
5604 * Save the original bdev mapping's wb_err value which could be
5605 * used to detect the metadata async write error.
5606 */
5607 errseq_check_and_advance(&sb->s_bdev->bd_mapping->wb_err,
5608 &sbi->s_bdev_wb_err);
5609 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5610 ext4_orphan_cleanup(sb, es);
5611 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5612 /*
5613 * Update the checksum after updating free space/inode counters and
5614 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5615 * checksum in the buffer cache until it is written out and
5616 * e2fsprogs programs trying to open a file system immediately
5617 * after it is mounted can fail.
5618 */
5619 ext4_superblock_csum_set(sb);
5620 if (needs_recovery) {
5621 ext4_msg(sb, KERN_INFO, "recovery complete");
5622 err = ext4_mark_recovery_complete(sb, es);
5623 if (err)
5624 goto failed_mount9;
5625 }
5626
5627 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5628 ext4_msg(sb, KERN_WARNING,
5629 "mounting with \"discard\" option, but the device does not support discard");
5630
5631 if (es->s_error_count)
5632 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5633
5634 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5635 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5636 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5637 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5638 atomic_set(&sbi->s_warning_count, 0);
5639 atomic_set(&sbi->s_msg_count, 0);
5640
5641 /* Register sysfs after all initializations are complete. */
5642 err = ext4_register_sysfs(sb);
5643 if (err)
5644 goto failed_mount9;
5645
5646 return 0;
5647
5648 failed_mount9:
5649 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
5650 failed_mount8: __maybe_unused
5651 ext4_release_orphan_info(sb);
5652 failed_mount7:
5653 ext4_unregister_li_request(sb);
5654 failed_mount6:
5655 ext4_mb_release(sb);
5656 ext4_flex_groups_free(sbi);
5657 ext4_percpu_param_destroy(sbi);
5658 failed_mount5:
5659 ext4_ext_release(sb);
5660 ext4_release_system_zone(sb);
5661 failed_mount4a:
5662 dput(sb->s_root);
5663 sb->s_root = NULL;
5664 failed_mount4:
5665 ext4_msg(sb, KERN_ERR, "mount failed");
5666 if (EXT4_SB(sb)->rsv_conversion_wq)
5667 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5668 failed_mount_wq:
5669 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5670 sbi->s_ea_inode_cache = NULL;
5671
5672 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5673 sbi->s_ea_block_cache = NULL;
5674
5675 if (sbi->s_journal) {
5676 /* flush s_sb_upd_work before journal destroy. */
5677 flush_work(&sbi->s_sb_upd_work);
5678 jbd2_journal_destroy(sbi->s_journal);
5679 sbi->s_journal = NULL;
5680 }
5681 failed_mount3a:
5682 ext4_es_unregister_shrinker(sbi);
5683 failed_mount3:
5684 /* flush s_sb_upd_work before sbi destroy */
5685 flush_work(&sbi->s_sb_upd_work);
5686 ext4_stop_mmpd(sbi);
5687 del_timer_sync(&sbi->s_err_report);
5688 ext4_group_desc_free(sbi);
5689 failed_mount:
5690 if (sbi->s_chksum_driver)
5691 crypto_free_shash(sbi->s_chksum_driver);
5692
5693 #if IS_ENABLED(CONFIG_UNICODE)
5694 utf8_unload(sb->s_encoding);
5695 #endif
5696
5697 #ifdef CONFIG_QUOTA
5698 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++)
5699 kfree(get_qf_name(sb, sbi, i));
5700 #endif
5701 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5702 brelse(sbi->s_sbh);
5703 if (sbi->s_journal_bdev_file) {
5704 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
5705 bdev_fput(sbi->s_journal_bdev_file);
5706 }
5707 out_fail:
5708 invalidate_bdev(sb->s_bdev);
5709 sb->s_fs_info = NULL;
5710 return err;
5711 }
5712
5713 static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5714 {
5715 struct ext4_fs_context *ctx = fc->fs_private;
5716 struct ext4_sb_info *sbi;
5717 const char *descr;
5718 int ret;
5719
5720 sbi = ext4_alloc_sbi(sb);
5721 if (!sbi)
5722 return -ENOMEM;
5723
5724 fc->s_fs_info = sbi;
5725
5726 /* Cleanup superblock name */
5727 strreplace(sb->s_id, '/', '!');
5728
5729 sbi->s_sb_block = 1; /* Default super block location */
5730 if (ctx->spec & EXT4_SPEC_s_sb_block)
5731 sbi->s_sb_block = ctx->s_sb_block;
5732
5733 ret = __ext4_fill_super(fc, sb);
5734 if (ret < 0)
5735 goto free_sbi;
5736
5737 if (sbi->s_journal) {
5738 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5739 descr = " journalled data mode";
5740 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5741 descr = " ordered data mode";
5742 else
5743 descr = " writeback data mode";
5744 } else
5745 descr = "out journal";
5746
5747 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5748 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. "
5749 "Quota mode: %s.", &sb->s_uuid,
5750 sb_rdonly(sb) ? "ro" : "r/w", descr,
5751 ext4_quota_mode(sb));
5752
5753 /* Update the s_overhead_clusters if necessary */
5754 ext4_update_overhead(sb, false);
5755 return 0;
5756
5757 free_sbi:
5758 ext4_free_sbi(sbi);
5759 fc->s_fs_info = NULL;
5760 return ret;
5761 }
5762
5763 static int ext4_get_tree(struct fs_context *fc)
5764 {
5765 return get_tree_bdev(fc, ext4_fill_super);
5766 }
5767
5768 /*
5769 * Setup any per-fs journal parameters now. We'll do this both on
5770 * initial mount, once the journal has been initialised but before we've
5771 * done any recovery; and again on any subsequent remount.
5772 */
5773 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5774 {
5775 struct ext4_sb_info *sbi = EXT4_SB(sb);
5776
5777 journal->j_commit_interval = sbi->s_commit_interval;
5778 journal->j_min_batch_time = sbi->s_min_batch_time;
5779 journal->j_max_batch_time = sbi->s_max_batch_time;
5780 ext4_fc_init(sb, journal);
5781
5782 write_lock(&journal->j_state_lock);
5783 if (test_opt(sb, BARRIER))
5784 journal->j_flags |= JBD2_BARRIER;
5785 else
5786 journal->j_flags &= ~JBD2_BARRIER;
5787 if (test_opt(sb, DATA_ERR_ABORT))
5788 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5789 else
5790 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5791 /*
5792 * Always enable journal cycle record option, letting the journal
5793 * records log transactions continuously between each mount.
5794 */
5795 journal->j_flags |= JBD2_CYCLE_RECORD;
5796 write_unlock(&journal->j_state_lock);
5797 }
5798
5799 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5800 unsigned int journal_inum)
5801 {
5802 struct inode *journal_inode;
5803
5804 /*
5805 * Test for the existence of a valid inode on disk. Bad things
5806 * happen if we iget() an unused inode, as the subsequent iput()
5807 * will try to delete it.
5808 */
5809 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5810 if (IS_ERR(journal_inode)) {
5811 ext4_msg(sb, KERN_ERR, "no journal found");
5812 return ERR_CAST(journal_inode);
5813 }
5814 if (!journal_inode->i_nlink) {
5815 make_bad_inode(journal_inode);
5816 iput(journal_inode);
5817 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5818 return ERR_PTR(-EFSCORRUPTED);
5819 }
5820 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
5821 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5822 iput(journal_inode);
5823 return ERR_PTR(-EFSCORRUPTED);
5824 }
5825
5826 ext4_debug("Journal inode found at %p: %lld bytes\n",
5827 journal_inode, journal_inode->i_size);
5828 return journal_inode;
5829 }
5830
5831 static int ext4_journal_bmap(journal_t *journal, sector_t *block)
5832 {
5833 struct ext4_map_blocks map;
5834 int ret;
5835
5836 if (journal->j_inode == NULL)
5837 return 0;
5838
5839 map.m_lblk = *block;
5840 map.m_len = 1;
5841 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0);
5842 if (ret <= 0) {
5843 ext4_msg(journal->j_inode->i_sb, KERN_CRIT,
5844 "journal bmap failed: block %llu ret %d\n",
5845 *block, ret);
5846 jbd2_journal_abort(journal, ret ? ret : -EIO);
5847 return ret;
5848 }
5849 *block = map.m_pblk;
5850 return 0;
5851 }
5852
5853 static journal_t *ext4_open_inode_journal(struct super_block *sb,
5854 unsigned int journal_inum)
5855 {
5856 struct inode *journal_inode;
5857 journal_t *journal;
5858
5859 journal_inode = ext4_get_journal_inode(sb, journal_inum);
5860 if (IS_ERR(journal_inode))
5861 return ERR_CAST(journal_inode);
5862
5863 journal = jbd2_journal_init_inode(journal_inode);
5864 if (IS_ERR(journal)) {
5865 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5866 iput(journal_inode);
5867 return ERR_CAST(journal);
5868 }
5869 journal->j_private = sb;
5870 journal->j_bmap = ext4_journal_bmap;
5871 ext4_init_journal_params(sb, journal);
5872 return journal;
5873 }
5874
5875 static struct file *ext4_get_journal_blkdev(struct super_block *sb,
5876 dev_t j_dev, ext4_fsblk_t *j_start,
5877 ext4_fsblk_t *j_len)
5878 {
5879 struct buffer_head *bh;
5880 struct block_device *bdev;
5881 struct file *bdev_file;
5882 int hblock, blocksize;
5883 ext4_fsblk_t sb_block;
5884 unsigned long offset;
5885 struct ext4_super_block *es;
5886 int errno;
5887
5888 bdev_file = bdev_file_open_by_dev(j_dev,
5889 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES,
5890 sb, &fs_holder_ops);
5891 if (IS_ERR(bdev_file)) {
5892 ext4_msg(sb, KERN_ERR,
5893 "failed to open journal device unknown-block(%u,%u) %ld",
5894 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file));
5895 return bdev_file;
5896 }
5897
5898 bdev = file_bdev(bdev_file);
5899 blocksize = sb->s_blocksize;
5900 hblock = bdev_logical_block_size(bdev);
5901 if (blocksize < hblock) {
5902 ext4_msg(sb, KERN_ERR,
5903 "blocksize too small for journal device");
5904 errno = -EINVAL;
5905 goto out_bdev;
5906 }
5907
5908 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5909 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5910 set_blocksize(bdev_file, blocksize);
5911 bh = __bread(bdev, sb_block, blocksize);
5912 if (!bh) {
5913 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5914 "external journal");
5915 errno = -EINVAL;
5916 goto out_bdev;
5917 }
5918
5919 es = (struct ext4_super_block *) (bh->b_data + offset);
5920 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5921 !(le32_to_cpu(es->s_feature_incompat) &
5922 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5923 ext4_msg(sb, KERN_ERR, "external journal has bad superblock");
5924 errno = -EFSCORRUPTED;
5925 goto out_bh;
5926 }
5927
5928 if ((le32_to_cpu(es->s_feature_ro_compat) &
5929 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5930 es->s_checksum != ext4_superblock_csum(sb, es)) {
5931 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock");
5932 errno = -EFSCORRUPTED;
5933 goto out_bh;
5934 }
5935
5936 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5937 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5938 errno = -EFSCORRUPTED;
5939 goto out_bh;
5940 }
5941
5942 *j_start = sb_block + 1;
5943 *j_len = ext4_blocks_count(es);
5944 brelse(bh);
5945 return bdev_file;
5946
5947 out_bh:
5948 brelse(bh);
5949 out_bdev:
5950 bdev_fput(bdev_file);
5951 return ERR_PTR(errno);
5952 }
5953
5954 static journal_t *ext4_open_dev_journal(struct super_block *sb,
5955 dev_t j_dev)
5956 {
5957 journal_t *journal;
5958 ext4_fsblk_t j_start;
5959 ext4_fsblk_t j_len;
5960 struct file *bdev_file;
5961 int errno = 0;
5962
5963 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len);
5964 if (IS_ERR(bdev_file))
5965 return ERR_CAST(bdev_file);
5966
5967 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start,
5968 j_len, sb->s_blocksize);
5969 if (IS_ERR(journal)) {
5970 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5971 errno = PTR_ERR(journal);
5972 goto out_bdev;
5973 }
5974 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5975 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5976 "user (unsupported) - %d",
5977 be32_to_cpu(journal->j_superblock->s_nr_users));
5978 errno = -EINVAL;
5979 goto out_journal;
5980 }
5981 journal->j_private = sb;
5982 EXT4_SB(sb)->s_journal_bdev_file = bdev_file;
5983 ext4_init_journal_params(sb, journal);
5984 return journal;
5985
5986 out_journal:
5987 jbd2_journal_destroy(journal);
5988 out_bdev:
5989 bdev_fput(bdev_file);
5990 return ERR_PTR(errno);
5991 }
5992
5993 static int ext4_load_journal(struct super_block *sb,
5994 struct ext4_super_block *es,
5995 unsigned long journal_devnum)
5996 {
5997 journal_t *journal;
5998 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5999 dev_t journal_dev;
6000 int err = 0;
6001 int really_read_only;
6002 int journal_dev_ro;
6003
6004 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
6005 return -EFSCORRUPTED;
6006
6007 if (journal_devnum &&
6008 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
6009 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
6010 "numbers have changed");
6011 journal_dev = new_decode_dev(journal_devnum);
6012 } else
6013 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
6014
6015 if (journal_inum && journal_dev) {
6016 ext4_msg(sb, KERN_ERR,
6017 "filesystem has both journal inode and journal device!");
6018 return -EINVAL;
6019 }
6020
6021 if (journal_inum) {
6022 journal = ext4_open_inode_journal(sb, journal_inum);
6023 if (IS_ERR(journal))
6024 return PTR_ERR(journal);
6025 } else {
6026 journal = ext4_open_dev_journal(sb, journal_dev);
6027 if (IS_ERR(journal))
6028 return PTR_ERR(journal);
6029 }
6030
6031 journal_dev_ro = bdev_read_only(journal->j_dev);
6032 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
6033
6034 if (journal_dev_ro && !sb_rdonly(sb)) {
6035 ext4_msg(sb, KERN_ERR,
6036 "journal device read-only, try mounting with '-o ro'");
6037 err = -EROFS;
6038 goto err_out;
6039 }
6040
6041 /*
6042 * Are we loading a blank journal or performing recovery after a
6043 * crash? For recovery, we need to check in advance whether we
6044 * can get read-write access to the device.
6045 */
6046 if (ext4_has_feature_journal_needs_recovery(sb)) {
6047 if (sb_rdonly(sb)) {
6048 ext4_msg(sb, KERN_INFO, "INFO: recovery "
6049 "required on readonly filesystem");
6050 if (really_read_only) {
6051 ext4_msg(sb, KERN_ERR, "write access "
6052 "unavailable, cannot proceed "
6053 "(try mounting with noload)");
6054 err = -EROFS;
6055 goto err_out;
6056 }
6057 ext4_msg(sb, KERN_INFO, "write access will "
6058 "be enabled during recovery");
6059 }
6060 }
6061
6062 if (!(journal->j_flags & JBD2_BARRIER))
6063 ext4_msg(sb, KERN_INFO, "barriers disabled");
6064
6065 if (!ext4_has_feature_journal_needs_recovery(sb))
6066 err = jbd2_journal_wipe(journal, !really_read_only);
6067 if (!err) {
6068 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
6069 __le16 orig_state;
6070 bool changed = false;
6071
6072 if (save)
6073 memcpy(save, ((char *) es) +
6074 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
6075 err = jbd2_journal_load(journal);
6076 if (save && memcmp(((char *) es) + EXT4_S_ERR_START,
6077 save, EXT4_S_ERR_LEN)) {
6078 memcpy(((char *) es) + EXT4_S_ERR_START,
6079 save, EXT4_S_ERR_LEN);
6080 changed = true;
6081 }
6082 kfree(save);
6083 orig_state = es->s_state;
6084 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state &
6085 EXT4_ERROR_FS);
6086 if (orig_state != es->s_state)
6087 changed = true;
6088 /* Write out restored error information to the superblock */
6089 if (changed && !really_read_only) {
6090 int err2;
6091 err2 = ext4_commit_super(sb);
6092 err = err ? : err2;
6093 }
6094 }
6095
6096 if (err) {
6097 ext4_msg(sb, KERN_ERR, "error loading journal");
6098 goto err_out;
6099 }
6100
6101 EXT4_SB(sb)->s_journal = journal;
6102 err = ext4_clear_journal_err(sb, es);
6103 if (err) {
6104 EXT4_SB(sb)->s_journal = NULL;
6105 jbd2_journal_destroy(journal);
6106 return err;
6107 }
6108
6109 if (!really_read_only && journal_devnum &&
6110 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
6111 es->s_journal_dev = cpu_to_le32(journal_devnum);
6112 ext4_commit_super(sb);
6113 }
6114 if (!really_read_only && journal_inum &&
6115 journal_inum != le32_to_cpu(es->s_journal_inum)) {
6116 es->s_journal_inum = cpu_to_le32(journal_inum);
6117 ext4_commit_super(sb);
6118 }
6119
6120 return 0;
6121
6122 err_out:
6123 jbd2_journal_destroy(journal);
6124 return err;
6125 }
6126
6127 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
6128 static void ext4_update_super(struct super_block *sb)
6129 {
6130 struct ext4_sb_info *sbi = EXT4_SB(sb);
6131 struct ext4_super_block *es = sbi->s_es;
6132 struct buffer_head *sbh = sbi->s_sbh;
6133
6134 lock_buffer(sbh);
6135 /*
6136 * If the file system is mounted read-only, don't update the
6137 * superblock write time. This avoids updating the superblock
6138 * write time when we are mounting the root file system
6139 * read/only but we need to replay the journal; at that point,
6140 * for people who are east of GMT and who make their clock
6141 * tick in localtime for Windows bug-for-bug compatibility,
6142 * the clock is set in the future, and this will cause e2fsck
6143 * to complain and force a full file system check.
6144 */
6145 if (!sb_rdonly(sb))
6146 ext4_update_tstamp(es, s_wtime);
6147 es->s_kbytes_written =
6148 cpu_to_le64(sbi->s_kbytes_written +
6149 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
6150 sbi->s_sectors_written_start) >> 1));
6151 if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
6152 ext4_free_blocks_count_set(es,
6153 EXT4_C2B(sbi, percpu_counter_sum_positive(
6154 &sbi->s_freeclusters_counter)));
6155 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
6156 es->s_free_inodes_count =
6157 cpu_to_le32(percpu_counter_sum_positive(
6158 &sbi->s_freeinodes_counter));
6159 /* Copy error information to the on-disk superblock */
6160 spin_lock(&sbi->s_error_lock);
6161 if (sbi->s_add_error_count > 0) {
6162 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6163 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
6164 __ext4_update_tstamp(&es->s_first_error_time,
6165 &es->s_first_error_time_hi,
6166 sbi->s_first_error_time);
6167 strtomem_pad(es->s_first_error_func,
6168 sbi->s_first_error_func, 0);
6169 es->s_first_error_line =
6170 cpu_to_le32(sbi->s_first_error_line);
6171 es->s_first_error_ino =
6172 cpu_to_le32(sbi->s_first_error_ino);
6173 es->s_first_error_block =
6174 cpu_to_le64(sbi->s_first_error_block);
6175 es->s_first_error_errcode =
6176 ext4_errno_to_code(sbi->s_first_error_code);
6177 }
6178 __ext4_update_tstamp(&es->s_last_error_time,
6179 &es->s_last_error_time_hi,
6180 sbi->s_last_error_time);
6181 strtomem_pad(es->s_last_error_func, sbi->s_last_error_func, 0);
6182 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
6183 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
6184 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
6185 es->s_last_error_errcode =
6186 ext4_errno_to_code(sbi->s_last_error_code);
6187 /*
6188 * Start the daily error reporting function if it hasn't been
6189 * started already
6190 */
6191 if (!es->s_error_count)
6192 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
6193 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
6194 sbi->s_add_error_count = 0;
6195 }
6196 spin_unlock(&sbi->s_error_lock);
6197
6198 ext4_superblock_csum_set(sb);
6199 unlock_buffer(sbh);
6200 }
6201
6202 static int ext4_commit_super(struct super_block *sb)
6203 {
6204 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6205
6206 if (!sbh)
6207 return -EINVAL;
6208
6209 ext4_update_super(sb);
6210
6211 lock_buffer(sbh);
6212 /* Buffer got discarded which means block device got invalidated */
6213 if (!buffer_mapped(sbh)) {
6214 unlock_buffer(sbh);
6215 return -EIO;
6216 }
6217
6218 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6219 /*
6220 * Oh, dear. A previous attempt to write the
6221 * superblock failed. This could happen because the
6222 * USB device was yanked out. Or it could happen to
6223 * be a transient write error and maybe the block will
6224 * be remapped. Nothing we can do but to retry the
6225 * write and hope for the best.
6226 */
6227 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6228 "superblock detected");
6229 clear_buffer_write_io_error(sbh);
6230 set_buffer_uptodate(sbh);
6231 }
6232 get_bh(sbh);
6233 /* Clear potential dirty bit if it was journalled update */
6234 clear_buffer_dirty(sbh);
6235 sbh->b_end_io = end_buffer_write_sync;
6236 submit_bh(REQ_OP_WRITE | REQ_SYNC |
6237 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
6238 wait_on_buffer(sbh);
6239 if (buffer_write_io_error(sbh)) {
6240 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6241 "superblock");
6242 clear_buffer_write_io_error(sbh);
6243 set_buffer_uptodate(sbh);
6244 return -EIO;
6245 }
6246 return 0;
6247 }
6248
6249 /*
6250 * Have we just finished recovery? If so, and if we are mounting (or
6251 * remounting) the filesystem readonly, then we will end up with a
6252 * consistent fs on disk. Record that fact.
6253 */
6254 static int ext4_mark_recovery_complete(struct super_block *sb,
6255 struct ext4_super_block *es)
6256 {
6257 int err;
6258 journal_t *journal = EXT4_SB(sb)->s_journal;
6259
6260 if (!ext4_has_feature_journal(sb)) {
6261 if (journal != NULL) {
6262 ext4_error(sb, "Journal got removed while the fs was "
6263 "mounted!");
6264 return -EFSCORRUPTED;
6265 }
6266 return 0;
6267 }
6268 jbd2_journal_lock_updates(journal);
6269 err = jbd2_journal_flush(journal, 0);
6270 if (err < 0)
6271 goto out;
6272
6273 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6274 ext4_has_feature_orphan_present(sb))) {
6275 if (!ext4_orphan_file_empty(sb)) {
6276 ext4_error(sb, "Orphan file not empty on read-only fs.");
6277 err = -EFSCORRUPTED;
6278 goto out;
6279 }
6280 ext4_clear_feature_journal_needs_recovery(sb);
6281 ext4_clear_feature_orphan_present(sb);
6282 ext4_commit_super(sb);
6283 }
6284 out:
6285 jbd2_journal_unlock_updates(journal);
6286 return err;
6287 }
6288
6289 /*
6290 * If we are mounting (or read-write remounting) a filesystem whose journal
6291 * has recorded an error from a previous lifetime, move that error to the
6292 * main filesystem now.
6293 */
6294 static int ext4_clear_journal_err(struct super_block *sb,
6295 struct ext4_super_block *es)
6296 {
6297 journal_t *journal;
6298 int j_errno;
6299 const char *errstr;
6300
6301 if (!ext4_has_feature_journal(sb)) {
6302 ext4_error(sb, "Journal got removed while the fs was mounted!");
6303 return -EFSCORRUPTED;
6304 }
6305
6306 journal = EXT4_SB(sb)->s_journal;
6307
6308 /*
6309 * Now check for any error status which may have been recorded in the
6310 * journal by a prior ext4_error() or ext4_abort()
6311 */
6312
6313 j_errno = jbd2_journal_errno(journal);
6314 if (j_errno) {
6315 char nbuf[16];
6316
6317 errstr = ext4_decode_error(sb, j_errno, nbuf);
6318 ext4_warning(sb, "Filesystem error recorded "
6319 "from previous mount: %s", errstr);
6320
6321 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6322 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6323 j_errno = ext4_commit_super(sb);
6324 if (j_errno)
6325 return j_errno;
6326 ext4_warning(sb, "Marked fs in need of filesystem check.");
6327
6328 jbd2_journal_clear_err(journal);
6329 jbd2_journal_update_sb_errno(journal);
6330 }
6331 return 0;
6332 }
6333
6334 /*
6335 * Force the running and committing transactions to commit,
6336 * and wait on the commit.
6337 */
6338 int ext4_force_commit(struct super_block *sb)
6339 {
6340 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal);
6341 }
6342
6343 static int ext4_sync_fs(struct super_block *sb, int wait)
6344 {
6345 int ret = 0;
6346 tid_t target;
6347 bool needs_barrier = false;
6348 struct ext4_sb_info *sbi = EXT4_SB(sb);
6349
6350 if (unlikely(ext4_forced_shutdown(sb)))
6351 return -EIO;
6352
6353 trace_ext4_sync_fs(sb, wait);
6354 flush_workqueue(sbi->rsv_conversion_wq);
6355 /*
6356 * Writeback quota in non-journalled quota case - journalled quota has
6357 * no dirty dquots
6358 */
6359 dquot_writeback_dquots(sb, -1);
6360 /*
6361 * Data writeback is possible w/o journal transaction, so barrier must
6362 * being sent at the end of the function. But we can skip it if
6363 * transaction_commit will do it for us.
6364 */
6365 if (sbi->s_journal) {
6366 target = jbd2_get_latest_transaction(sbi->s_journal);
6367 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6368 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6369 needs_barrier = true;
6370
6371 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6372 if (wait)
6373 ret = jbd2_log_wait_commit(sbi->s_journal,
6374 target);
6375 }
6376 } else if (wait && test_opt(sb, BARRIER))
6377 needs_barrier = true;
6378 if (needs_barrier) {
6379 int err;
6380 err = blkdev_issue_flush(sb->s_bdev);
6381 if (!ret)
6382 ret = err;
6383 }
6384
6385 return ret;
6386 }
6387
6388 /*
6389 * LVM calls this function before a (read-only) snapshot is created. This
6390 * gives us a chance to flush the journal completely and mark the fs clean.
6391 *
6392 * Note that only this function cannot bring a filesystem to be in a clean
6393 * state independently. It relies on upper layer to stop all data & metadata
6394 * modifications.
6395 */
6396 static int ext4_freeze(struct super_block *sb)
6397 {
6398 int error = 0;
6399 journal_t *journal = EXT4_SB(sb)->s_journal;
6400
6401 if (journal) {
6402 /* Now we set up the journal barrier. */
6403 jbd2_journal_lock_updates(journal);
6404
6405 /*
6406 * Don't clear the needs_recovery flag if we failed to
6407 * flush the journal.
6408 */
6409 error = jbd2_journal_flush(journal, 0);
6410 if (error < 0)
6411 goto out;
6412
6413 /* Journal blocked and flushed, clear needs_recovery flag. */
6414 ext4_clear_feature_journal_needs_recovery(sb);
6415 if (ext4_orphan_file_empty(sb))
6416 ext4_clear_feature_orphan_present(sb);
6417 }
6418
6419 error = ext4_commit_super(sb);
6420 out:
6421 if (journal)
6422 /* we rely on upper layer to stop further updates */
6423 jbd2_journal_unlock_updates(journal);
6424 return error;
6425 }
6426
6427 /*
6428 * Called by LVM after the snapshot is done. We need to reset the RECOVER
6429 * flag here, even though the filesystem is not technically dirty yet.
6430 */
6431 static int ext4_unfreeze(struct super_block *sb)
6432 {
6433 if (ext4_forced_shutdown(sb))
6434 return 0;
6435
6436 if (EXT4_SB(sb)->s_journal) {
6437 /* Reset the needs_recovery flag before the fs is unlocked. */
6438 ext4_set_feature_journal_needs_recovery(sb);
6439 if (ext4_has_feature_orphan_file(sb))
6440 ext4_set_feature_orphan_present(sb);
6441 }
6442
6443 ext4_commit_super(sb);
6444 return 0;
6445 }
6446
6447 /*
6448 * Structure to save mount options for ext4_remount's benefit
6449 */
6450 struct ext4_mount_options {
6451 unsigned long s_mount_opt;
6452 unsigned long s_mount_opt2;
6453 kuid_t s_resuid;
6454 kgid_t s_resgid;
6455 unsigned long s_commit_interval;
6456 u32 s_min_batch_time, s_max_batch_time;
6457 #ifdef CONFIG_QUOTA
6458 int s_jquota_fmt;
6459 char *s_qf_names[EXT4_MAXQUOTAS];
6460 #endif
6461 };
6462
6463 static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6464 {
6465 struct ext4_fs_context *ctx = fc->fs_private;
6466 struct ext4_super_block *es;
6467 struct ext4_sb_info *sbi = EXT4_SB(sb);
6468 unsigned long old_sb_flags;
6469 struct ext4_mount_options old_opts;
6470 ext4_group_t g;
6471 int err = 0;
6472 int alloc_ctx;
6473 #ifdef CONFIG_QUOTA
6474 int enable_quota = 0;
6475 int i, j;
6476 char *to_free[EXT4_MAXQUOTAS];
6477 #endif
6478
6479
6480 /* Store the original options */
6481 old_sb_flags = sb->s_flags;
6482 old_opts.s_mount_opt = sbi->s_mount_opt;
6483 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6484 old_opts.s_resuid = sbi->s_resuid;
6485 old_opts.s_resgid = sbi->s_resgid;
6486 old_opts.s_commit_interval = sbi->s_commit_interval;
6487 old_opts.s_min_batch_time = sbi->s_min_batch_time;
6488 old_opts.s_max_batch_time = sbi->s_max_batch_time;
6489 #ifdef CONFIG_QUOTA
6490 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6491 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6492 if (sbi->s_qf_names[i]) {
6493 char *qf_name = get_qf_name(sb, sbi, i);
6494
6495 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6496 if (!old_opts.s_qf_names[i]) {
6497 for (j = 0; j < i; j++)
6498 kfree(old_opts.s_qf_names[j]);
6499 return -ENOMEM;
6500 }
6501 } else
6502 old_opts.s_qf_names[i] = NULL;
6503 #endif
6504 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6505 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6506 ctx->journal_ioprio =
6507 sbi->s_journal->j_task->io_context->ioprio;
6508 else
6509 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6510
6511 }
6512
6513 if ((ctx->spec & EXT4_SPEC_s_stripe) &&
6514 ext4_is_stripe_incompatible(sb, ctx->s_stripe)) {
6515 ext4_msg(sb, KERN_WARNING,
6516 "stripe (%lu) is not aligned with cluster size (%u), "
6517 "stripe is disabled",
6518 ctx->s_stripe, sbi->s_cluster_ratio);
6519 ctx->s_stripe = 0;
6520 }
6521
6522 /*
6523 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause
6524 * two calls to ext4_should_dioread_nolock() to return inconsistent
6525 * values, triggering WARN_ON in ext4_add_complete_io(). we grab
6526 * here s_writepages_rwsem to avoid race between writepages ops and
6527 * remount.
6528 */
6529 alloc_ctx = ext4_writepages_down_write(sb);
6530 ext4_apply_options(fc, sb);
6531 ext4_writepages_up_write(sb, alloc_ctx);
6532
6533 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6534 test_opt(sb, JOURNAL_CHECKSUM)) {
6535 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6536 "during remount not supported; ignoring");
6537 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6538 }
6539
6540 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6541 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6542 ext4_msg(sb, KERN_ERR, "can't mount with "
6543 "both data=journal and delalloc");
6544 err = -EINVAL;
6545 goto restore_opts;
6546 }
6547 if (test_opt(sb, DIOREAD_NOLOCK)) {
6548 ext4_msg(sb, KERN_ERR, "can't mount with "
6549 "both data=journal and dioread_nolock");
6550 err = -EINVAL;
6551 goto restore_opts;
6552 }
6553 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6554 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6555 ext4_msg(sb, KERN_ERR, "can't mount with "
6556 "journal_async_commit in data=ordered mode");
6557 err = -EINVAL;
6558 goto restore_opts;
6559 }
6560 }
6561
6562 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6563 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6564 err = -EINVAL;
6565 goto restore_opts;
6566 }
6567
6568 if ((old_opts.s_mount_opt & EXT4_MOUNT_DELALLOC) &&
6569 !test_opt(sb, DELALLOC)) {
6570 ext4_msg(sb, KERN_ERR, "can't disable delalloc during remount");
6571 err = -EINVAL;
6572 goto restore_opts;
6573 }
6574
6575 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6576 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6577
6578 es = sbi->s_es;
6579
6580 if (sbi->s_journal) {
6581 ext4_init_journal_params(sb, sbi->s_journal);
6582 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6583 }
6584
6585 /* Flush outstanding errors before changing fs state */
6586 flush_work(&sbi->s_sb_upd_work);
6587
6588 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6589 if (ext4_forced_shutdown(sb)) {
6590 err = -EROFS;
6591 goto restore_opts;
6592 }
6593
6594 if (fc->sb_flags & SB_RDONLY) {
6595 err = sync_filesystem(sb);
6596 if (err < 0)
6597 goto restore_opts;
6598 err = dquot_suspend(sb, -1);
6599 if (err < 0)
6600 goto restore_opts;
6601
6602 /*
6603 * First of all, the unconditional stuff we have to do
6604 * to disable replay of the journal when we next remount
6605 */
6606 sb->s_flags |= SB_RDONLY;
6607
6608 /*
6609 * OK, test if we are remounting a valid rw partition
6610 * readonly, and if so set the rdonly flag and then
6611 * mark the partition as valid again.
6612 */
6613 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6614 (sbi->s_mount_state & EXT4_VALID_FS))
6615 es->s_state = cpu_to_le16(sbi->s_mount_state);
6616
6617 if (sbi->s_journal) {
6618 /*
6619 * We let remount-ro finish even if marking fs
6620 * as clean failed...
6621 */
6622 ext4_mark_recovery_complete(sb, es);
6623 }
6624 } else {
6625 /* Make sure we can mount this feature set readwrite */
6626 if (ext4_has_feature_readonly(sb) ||
6627 !ext4_feature_set_ok(sb, 0)) {
6628 err = -EROFS;
6629 goto restore_opts;
6630 }
6631 /*
6632 * Make sure the group descriptor checksums
6633 * are sane. If they aren't, refuse to remount r/w.
6634 */
6635 for (g = 0; g < sbi->s_groups_count; g++) {
6636 struct ext4_group_desc *gdp =
6637 ext4_get_group_desc(sb, g, NULL);
6638
6639 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6640 ext4_msg(sb, KERN_ERR,
6641 "ext4_remount: Checksum for group %u failed (%u!=%u)",
6642 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6643 le16_to_cpu(gdp->bg_checksum));
6644 err = -EFSBADCRC;
6645 goto restore_opts;
6646 }
6647 }
6648
6649 /*
6650 * If we have an unprocessed orphan list hanging
6651 * around from a previously readonly bdev mount,
6652 * require a full umount/remount for now.
6653 */
6654 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6655 ext4_msg(sb, KERN_WARNING, "Couldn't "
6656 "remount RDWR because of unprocessed "
6657 "orphan inode list. Please "
6658 "umount/remount instead");
6659 err = -EINVAL;
6660 goto restore_opts;
6661 }
6662
6663 /*
6664 * Mounting a RDONLY partition read-write, so reread
6665 * and store the current valid flag. (It may have
6666 * been changed by e2fsck since we originally mounted
6667 * the partition.)
6668 */
6669 if (sbi->s_journal) {
6670 err = ext4_clear_journal_err(sb, es);
6671 if (err)
6672 goto restore_opts;
6673 }
6674 sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6675 ~EXT4_FC_REPLAY);
6676
6677 err = ext4_setup_super(sb, es, 0);
6678 if (err)
6679 goto restore_opts;
6680
6681 sb->s_flags &= ~SB_RDONLY;
6682 if (ext4_has_feature_mmp(sb)) {
6683 err = ext4_multi_mount_protect(sb,
6684 le64_to_cpu(es->s_mmp_block));
6685 if (err)
6686 goto restore_opts;
6687 }
6688 #ifdef CONFIG_QUOTA
6689 enable_quota = 1;
6690 #endif
6691 }
6692 }
6693
6694 /*
6695 * Handle creation of system zone data early because it can fail.
6696 * Releasing of existing data is done when we are sure remount will
6697 * succeed.
6698 */
6699 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6700 err = ext4_setup_system_zone(sb);
6701 if (err)
6702 goto restore_opts;
6703 }
6704
6705 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6706 err = ext4_commit_super(sb);
6707 if (err)
6708 goto restore_opts;
6709 }
6710
6711 #ifdef CONFIG_QUOTA
6712 if (enable_quota) {
6713 if (sb_any_quota_suspended(sb))
6714 dquot_resume(sb, -1);
6715 else if (ext4_has_feature_quota(sb)) {
6716 err = ext4_enable_quotas(sb);
6717 if (err)
6718 goto restore_opts;
6719 }
6720 }
6721 /* Release old quota file names */
6722 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6723 kfree(old_opts.s_qf_names[i]);
6724 #endif
6725 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6726 ext4_release_system_zone(sb);
6727
6728 /*
6729 * Reinitialize lazy itable initialization thread based on
6730 * current settings
6731 */
6732 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6733 ext4_unregister_li_request(sb);
6734 else {
6735 ext4_group_t first_not_zeroed;
6736 first_not_zeroed = ext4_has_uninit_itable(sb);
6737 ext4_register_li_request(sb, first_not_zeroed);
6738 }
6739
6740 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6741 ext4_stop_mmpd(sbi);
6742
6743 /*
6744 * Handle aborting the filesystem as the last thing during remount to
6745 * avoid obsure errors during remount when some option changes fail to
6746 * apply due to shutdown filesystem.
6747 */
6748 if (test_opt2(sb, ABORT))
6749 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6750
6751 return 0;
6752
6753 restore_opts:
6754 /*
6755 * If there was a failing r/w to ro transition, we may need to
6756 * re-enable quota
6757 */
6758 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) &&
6759 sb_any_quota_suspended(sb))
6760 dquot_resume(sb, -1);
6761
6762 alloc_ctx = ext4_writepages_down_write(sb);
6763 sb->s_flags = old_sb_flags;
6764 sbi->s_mount_opt = old_opts.s_mount_opt;
6765 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6766 sbi->s_resuid = old_opts.s_resuid;
6767 sbi->s_resgid = old_opts.s_resgid;
6768 sbi->s_commit_interval = old_opts.s_commit_interval;
6769 sbi->s_min_batch_time = old_opts.s_min_batch_time;
6770 sbi->s_max_batch_time = old_opts.s_max_batch_time;
6771 ext4_writepages_up_write(sb, alloc_ctx);
6772
6773 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6774 ext4_release_system_zone(sb);
6775 #ifdef CONFIG_QUOTA
6776 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6777 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6778 to_free[i] = get_qf_name(sb, sbi, i);
6779 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6780 }
6781 synchronize_rcu();
6782 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6783 kfree(to_free[i]);
6784 #endif
6785 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6786 ext4_stop_mmpd(sbi);
6787 return err;
6788 }
6789
6790 static int ext4_reconfigure(struct fs_context *fc)
6791 {
6792 struct super_block *sb = fc->root->d_sb;
6793 int ret;
6794
6795 fc->s_fs_info = EXT4_SB(sb);
6796
6797 ret = ext4_check_opt_consistency(fc, sb);
6798 if (ret < 0)
6799 return ret;
6800
6801 ret = __ext4_remount(fc, sb);
6802 if (ret < 0)
6803 return ret;
6804
6805 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.",
6806 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w",
6807 ext4_quota_mode(sb));
6808
6809 return 0;
6810 }
6811
6812 #ifdef CONFIG_QUOTA
6813 static int ext4_statfs_project(struct super_block *sb,
6814 kprojid_t projid, struct kstatfs *buf)
6815 {
6816 struct kqid qid;
6817 struct dquot *dquot;
6818 u64 limit;
6819 u64 curblock;
6820
6821 qid = make_kqid_projid(projid);
6822 dquot = dqget(sb, qid);
6823 if (IS_ERR(dquot))
6824 return PTR_ERR(dquot);
6825 spin_lock(&dquot->dq_dqb_lock);
6826
6827 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6828 dquot->dq_dqb.dqb_bhardlimit);
6829 limit >>= sb->s_blocksize_bits;
6830
6831 if (limit && buf->f_blocks > limit) {
6832 curblock = (dquot->dq_dqb.dqb_curspace +
6833 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6834 buf->f_blocks = limit;
6835 buf->f_bfree = buf->f_bavail =
6836 (buf->f_blocks > curblock) ?
6837 (buf->f_blocks - curblock) : 0;
6838 }
6839
6840 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6841 dquot->dq_dqb.dqb_ihardlimit);
6842 if (limit && buf->f_files > limit) {
6843 buf->f_files = limit;
6844 buf->f_ffree =
6845 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6846 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6847 }
6848
6849 spin_unlock(&dquot->dq_dqb_lock);
6850 dqput(dquot);
6851 return 0;
6852 }
6853 #endif
6854
6855 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6856 {
6857 struct super_block *sb = dentry->d_sb;
6858 struct ext4_sb_info *sbi = EXT4_SB(sb);
6859 struct ext4_super_block *es = sbi->s_es;
6860 ext4_fsblk_t overhead = 0, resv_blocks;
6861 s64 bfree;
6862 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6863
6864 if (!test_opt(sb, MINIX_DF))
6865 overhead = sbi->s_overhead;
6866
6867 buf->f_type = EXT4_SUPER_MAGIC;
6868 buf->f_bsize = sb->s_blocksize;
6869 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6870 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6871 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6872 /* prevent underflow in case that few free space is available */
6873 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6874 buf->f_bavail = buf->f_bfree -
6875 (ext4_r_blocks_count(es) + resv_blocks);
6876 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6877 buf->f_bavail = 0;
6878 buf->f_files = le32_to_cpu(es->s_inodes_count);
6879 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6880 buf->f_namelen = EXT4_NAME_LEN;
6881 buf->f_fsid = uuid_to_fsid(es->s_uuid);
6882
6883 #ifdef CONFIG_QUOTA
6884 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6885 sb_has_quota_limits_enabled(sb, PRJQUOTA))
6886 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6887 #endif
6888 return 0;
6889 }
6890
6891
6892 #ifdef CONFIG_QUOTA
6893
6894 /*
6895 * Helper functions so that transaction is started before we acquire dqio_sem
6896 * to keep correct lock ordering of transaction > dqio_sem
6897 */
6898 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6899 {
6900 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6901 }
6902
6903 static int ext4_write_dquot(struct dquot *dquot)
6904 {
6905 int ret, err;
6906 handle_t *handle;
6907 struct inode *inode;
6908
6909 inode = dquot_to_inode(dquot);
6910 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6911 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6912 if (IS_ERR(handle))
6913 return PTR_ERR(handle);
6914 ret = dquot_commit(dquot);
6915 if (ret < 0)
6916 ext4_error_err(dquot->dq_sb, -ret,
6917 "Failed to commit dquot type %d",
6918 dquot->dq_id.type);
6919 err = ext4_journal_stop(handle);
6920 if (!ret)
6921 ret = err;
6922 return ret;
6923 }
6924
6925 static int ext4_acquire_dquot(struct dquot *dquot)
6926 {
6927 int ret, err;
6928 handle_t *handle;
6929
6930 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6931 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6932 if (IS_ERR(handle))
6933 return PTR_ERR(handle);
6934 ret = dquot_acquire(dquot);
6935 if (ret < 0)
6936 ext4_error_err(dquot->dq_sb, -ret,
6937 "Failed to acquire dquot type %d",
6938 dquot->dq_id.type);
6939 err = ext4_journal_stop(handle);
6940 if (!ret)
6941 ret = err;
6942 return ret;
6943 }
6944
6945 static int ext4_release_dquot(struct dquot *dquot)
6946 {
6947 int ret, err;
6948 handle_t *handle;
6949
6950 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6951 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6952 if (IS_ERR(handle)) {
6953 /* Release dquot anyway to avoid endless cycle in dqput() */
6954 dquot_release(dquot);
6955 return PTR_ERR(handle);
6956 }
6957 ret = dquot_release(dquot);
6958 if (ret < 0)
6959 ext4_error_err(dquot->dq_sb, -ret,
6960 "Failed to release dquot type %d",
6961 dquot->dq_id.type);
6962 err = ext4_journal_stop(handle);
6963 if (!ret)
6964 ret = err;
6965 return ret;
6966 }
6967
6968 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6969 {
6970 struct super_block *sb = dquot->dq_sb;
6971
6972 if (ext4_is_quota_journalled(sb)) {
6973 dquot_mark_dquot_dirty(dquot);
6974 return ext4_write_dquot(dquot);
6975 } else {
6976 return dquot_mark_dquot_dirty(dquot);
6977 }
6978 }
6979
6980 static int ext4_write_info(struct super_block *sb, int type)
6981 {
6982 int ret, err;
6983 handle_t *handle;
6984
6985 /* Data block + inode block */
6986 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
6987 if (IS_ERR(handle))
6988 return PTR_ERR(handle);
6989 ret = dquot_commit_info(sb, type);
6990 err = ext4_journal_stop(handle);
6991 if (!ret)
6992 ret = err;
6993 return ret;
6994 }
6995
6996 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6997 {
6998 struct ext4_inode_info *ei = EXT4_I(inode);
6999
7000 /* The first argument of lockdep_set_subclass has to be
7001 * *exactly* the same as the argument to init_rwsem() --- in
7002 * this case, in init_once() --- or lockdep gets unhappy
7003 * because the name of the lock is set using the
7004 * stringification of the argument to init_rwsem().
7005 */
7006 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
7007 lockdep_set_subclass(&ei->i_data_sem, subclass);
7008 }
7009
7010 /*
7011 * Standard function to be called on quota_on
7012 */
7013 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
7014 const struct path *path)
7015 {
7016 int err;
7017
7018 if (!test_opt(sb, QUOTA))
7019 return -EINVAL;
7020
7021 /* Quotafile not on the same filesystem? */
7022 if (path->dentry->d_sb != sb)
7023 return -EXDEV;
7024
7025 /* Quota already enabled for this file? */
7026 if (IS_NOQUOTA(d_inode(path->dentry)))
7027 return -EBUSY;
7028
7029 /* Journaling quota? */
7030 if (EXT4_SB(sb)->s_qf_names[type]) {
7031 /* Quotafile not in fs root? */
7032 if (path->dentry->d_parent != sb->s_root)
7033 ext4_msg(sb, KERN_WARNING,
7034 "Quota file not on filesystem root. "
7035 "Journaled quota will not work");
7036 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
7037 } else {
7038 /*
7039 * Clear the flag just in case mount options changed since
7040 * last time.
7041 */
7042 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
7043 }
7044
7045 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
7046 err = dquot_quota_on(sb, type, format_id, path);
7047 if (!err) {
7048 struct inode *inode = d_inode(path->dentry);
7049 handle_t *handle;
7050
7051 /*
7052 * Set inode flags to prevent userspace from messing with quota
7053 * files. If this fails, we return success anyway since quotas
7054 * are already enabled and this is not a hard failure.
7055 */
7056 inode_lock(inode);
7057 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7058 if (IS_ERR(handle))
7059 goto unlock_inode;
7060 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
7061 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
7062 S_NOATIME | S_IMMUTABLE);
7063 err = ext4_mark_inode_dirty(handle, inode);
7064 ext4_journal_stop(handle);
7065 unlock_inode:
7066 inode_unlock(inode);
7067 if (err)
7068 dquot_quota_off(sb, type);
7069 }
7070 if (err)
7071 lockdep_set_quota_inode(path->dentry->d_inode,
7072 I_DATA_SEM_NORMAL);
7073 return err;
7074 }
7075
7076 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
7077 {
7078 switch (type) {
7079 case USRQUOTA:
7080 return qf_inum == EXT4_USR_QUOTA_INO;
7081 case GRPQUOTA:
7082 return qf_inum == EXT4_GRP_QUOTA_INO;
7083 case PRJQUOTA:
7084 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
7085 default:
7086 BUG();
7087 }
7088 }
7089
7090 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
7091 unsigned int flags)
7092 {
7093 int err;
7094 struct inode *qf_inode;
7095 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
7096 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
7097 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
7098 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
7099 };
7100
7101 BUG_ON(!ext4_has_feature_quota(sb));
7102
7103 if (!qf_inums[type])
7104 return -EPERM;
7105
7106 if (!ext4_check_quota_inum(type, qf_inums[type])) {
7107 ext4_error(sb, "Bad quota inum: %lu, type: %d",
7108 qf_inums[type], type);
7109 return -EUCLEAN;
7110 }
7111
7112 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
7113 if (IS_ERR(qf_inode)) {
7114 ext4_error(sb, "Bad quota inode: %lu, type: %d",
7115 qf_inums[type], type);
7116 return PTR_ERR(qf_inode);
7117 }
7118
7119 /* Don't account quota for quota files to avoid recursion */
7120 qf_inode->i_flags |= S_NOQUOTA;
7121 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
7122 err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
7123 if (err)
7124 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
7125 iput(qf_inode);
7126
7127 return err;
7128 }
7129
7130 /* Enable usage tracking for all quota types. */
7131 int ext4_enable_quotas(struct super_block *sb)
7132 {
7133 int type, err = 0;
7134 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
7135 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
7136 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
7137 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
7138 };
7139 bool quota_mopt[EXT4_MAXQUOTAS] = {
7140 test_opt(sb, USRQUOTA),
7141 test_opt(sb, GRPQUOTA),
7142 test_opt(sb, PRJQUOTA),
7143 };
7144
7145 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
7146 for (type = 0; type < EXT4_MAXQUOTAS; type++) {
7147 if (qf_inums[type]) {
7148 err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
7149 DQUOT_USAGE_ENABLED |
7150 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
7151 if (err) {
7152 ext4_warning(sb,
7153 "Failed to enable quota tracking "
7154 "(type=%d, err=%d, ino=%lu). "
7155 "Please run e2fsck to fix.", type,
7156 err, qf_inums[type]);
7157
7158 ext4_quotas_off(sb, type);
7159 return err;
7160 }
7161 }
7162 }
7163 return 0;
7164 }
7165
7166 static int ext4_quota_off(struct super_block *sb, int type)
7167 {
7168 struct inode *inode = sb_dqopt(sb)->files[type];
7169 handle_t *handle;
7170 int err;
7171
7172 /* Force all delayed allocation blocks to be allocated.
7173 * Caller already holds s_umount sem */
7174 if (test_opt(sb, DELALLOC))
7175 sync_filesystem(sb);
7176
7177 if (!inode || !igrab(inode))
7178 goto out;
7179
7180 err = dquot_quota_off(sb, type);
7181 if (err || ext4_has_feature_quota(sb))
7182 goto out_put;
7183 /*
7184 * When the filesystem was remounted read-only first, we cannot cleanup
7185 * inode flags here. Bad luck but people should be using QUOTA feature
7186 * these days anyway.
7187 */
7188 if (sb_rdonly(sb))
7189 goto out_put;
7190
7191 inode_lock(inode);
7192 /*
7193 * Update modification times of quota files when userspace can
7194 * start looking at them. If we fail, we return success anyway since
7195 * this is not a hard failure and quotas are already disabled.
7196 */
7197 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7198 if (IS_ERR(handle)) {
7199 err = PTR_ERR(handle);
7200 goto out_unlock;
7201 }
7202 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
7203 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
7204 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
7205 err = ext4_mark_inode_dirty(handle, inode);
7206 ext4_journal_stop(handle);
7207 out_unlock:
7208 inode_unlock(inode);
7209 out_put:
7210 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
7211 iput(inode);
7212 return err;
7213 out:
7214 return dquot_quota_off(sb, type);
7215 }
7216
7217 /* Read data from quotafile - avoid pagecache and such because we cannot afford
7218 * acquiring the locks... As quota files are never truncated and quota code
7219 * itself serializes the operations (and no one else should touch the files)
7220 * we don't have to be afraid of races */
7221 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
7222 size_t len, loff_t off)
7223 {
7224 struct inode *inode = sb_dqopt(sb)->files[type];
7225 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7226 int offset = off & (sb->s_blocksize - 1);
7227 int tocopy;
7228 size_t toread;
7229 struct buffer_head *bh;
7230 loff_t i_size = i_size_read(inode);
7231
7232 if (off > i_size)
7233 return 0;
7234 if (off+len > i_size)
7235 len = i_size-off;
7236 toread = len;
7237 while (toread > 0) {
7238 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread);
7239 bh = ext4_bread(NULL, inode, blk, 0);
7240 if (IS_ERR(bh))
7241 return PTR_ERR(bh);
7242 if (!bh) /* A hole? */
7243 memset(data, 0, tocopy);
7244 else
7245 memcpy(data, bh->b_data+offset, tocopy);
7246 brelse(bh);
7247 offset = 0;
7248 toread -= tocopy;
7249 data += tocopy;
7250 blk++;
7251 }
7252 return len;
7253 }
7254
7255 /* Write to quotafile (we know the transaction is already started and has
7256 * enough credits) */
7257 static ssize_t ext4_quota_write(struct super_block *sb, int type,
7258 const char *data, size_t len, loff_t off)
7259 {
7260 struct inode *inode = sb_dqopt(sb)->files[type];
7261 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7262 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7263 int retries = 0;
7264 struct buffer_head *bh;
7265 handle_t *handle = journal_current_handle();
7266
7267 if (!handle) {
7268 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7269 " cancelled because transaction is not started",
7270 (unsigned long long)off, (unsigned long long)len);
7271 return -EIO;
7272 }
7273 /*
7274 * Since we account only one data block in transaction credits,
7275 * then it is impossible to cross a block boundary.
7276 */
7277 if (sb->s_blocksize - offset < len) {
7278 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7279 " cancelled because not block aligned",
7280 (unsigned long long)off, (unsigned long long)len);
7281 return -EIO;
7282 }
7283
7284 do {
7285 bh = ext4_bread(handle, inode, blk,
7286 EXT4_GET_BLOCKS_CREATE |
7287 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7288 } while (PTR_ERR(bh) == -ENOSPC &&
7289 ext4_should_retry_alloc(inode->i_sb, &retries));
7290 if (IS_ERR(bh))
7291 return PTR_ERR(bh);
7292 if (!bh)
7293 goto out;
7294 BUFFER_TRACE(bh, "get write access");
7295 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7296 if (err) {
7297 brelse(bh);
7298 return err;
7299 }
7300 lock_buffer(bh);
7301 memcpy(bh->b_data+offset, data, len);
7302 flush_dcache_page(bh->b_page);
7303 unlock_buffer(bh);
7304 err = ext4_handle_dirty_metadata(handle, NULL, bh);
7305 brelse(bh);
7306 out:
7307 if (inode->i_size < off + len) {
7308 i_size_write(inode, off + len);
7309 EXT4_I(inode)->i_disksize = inode->i_size;
7310 err2 = ext4_mark_inode_dirty(handle, inode);
7311 if (unlikely(err2 && !err))
7312 err = err2;
7313 }
7314 return err ? err : len;
7315 }
7316 #endif
7317
7318 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7319 static inline void register_as_ext2(void)
7320 {
7321 int err = register_filesystem(&ext2_fs_type);
7322 if (err)
7323 printk(KERN_WARNING
7324 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7325 }
7326
7327 static inline void unregister_as_ext2(void)
7328 {
7329 unregister_filesystem(&ext2_fs_type);
7330 }
7331
7332 static inline int ext2_feature_set_ok(struct super_block *sb)
7333 {
7334 if (ext4_has_unknown_ext2_incompat_features(sb))
7335 return 0;
7336 if (sb_rdonly(sb))
7337 return 1;
7338 if (ext4_has_unknown_ext2_ro_compat_features(sb))
7339 return 0;
7340 return 1;
7341 }
7342 #else
7343 static inline void register_as_ext2(void) { }
7344 static inline void unregister_as_ext2(void) { }
7345 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7346 #endif
7347
7348 static inline void register_as_ext3(void)
7349 {
7350 int err = register_filesystem(&ext3_fs_type);
7351 if (err)
7352 printk(KERN_WARNING
7353 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7354 }
7355
7356 static inline void unregister_as_ext3(void)
7357 {
7358 unregister_filesystem(&ext3_fs_type);
7359 }
7360
7361 static inline int ext3_feature_set_ok(struct super_block *sb)
7362 {
7363 if (ext4_has_unknown_ext3_incompat_features(sb))
7364 return 0;
7365 if (!ext4_has_feature_journal(sb))
7366 return 0;
7367 if (sb_rdonly(sb))
7368 return 1;
7369 if (ext4_has_unknown_ext3_ro_compat_features(sb))
7370 return 0;
7371 return 1;
7372 }
7373
7374 static void ext4_kill_sb(struct super_block *sb)
7375 {
7376 struct ext4_sb_info *sbi = EXT4_SB(sb);
7377 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL;
7378
7379 kill_block_super(sb);
7380
7381 if (bdev_file)
7382 bdev_fput(bdev_file);
7383 }
7384
7385 static struct file_system_type ext4_fs_type = {
7386 .owner = THIS_MODULE,
7387 .name = "ext4",
7388 .init_fs_context = ext4_init_fs_context,
7389 .parameters = ext4_param_specs,
7390 .kill_sb = ext4_kill_sb,
7391 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME,
7392 };
7393 MODULE_ALIAS_FS("ext4");
7394
7395 /* Shared across all ext4 file systems */
7396 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7397
7398 static int __init ext4_init_fs(void)
7399 {
7400 int i, err;
7401
7402 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7403 ext4_li_info = NULL;
7404
7405 /* Build-time check for flags consistency */
7406 ext4_check_flag_values();
7407
7408 for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7409 init_waitqueue_head(&ext4__ioend_wq[i]);
7410
7411 err = ext4_init_es();
7412 if (err)
7413 return err;
7414
7415 err = ext4_init_pending();
7416 if (err)
7417 goto out7;
7418
7419 err = ext4_init_post_read_processing();
7420 if (err)
7421 goto out6;
7422
7423 err = ext4_init_pageio();
7424 if (err)
7425 goto out5;
7426
7427 err = ext4_init_system_zone();
7428 if (err)
7429 goto out4;
7430
7431 err = ext4_init_sysfs();
7432 if (err)
7433 goto out3;
7434
7435 err = ext4_init_mballoc();
7436 if (err)
7437 goto out2;
7438 err = init_inodecache();
7439 if (err)
7440 goto out1;
7441
7442 err = ext4_fc_init_dentry_cache();
7443 if (err)
7444 goto out05;
7445
7446 register_as_ext3();
7447 register_as_ext2();
7448 err = register_filesystem(&ext4_fs_type);
7449 if (err)
7450 goto out;
7451
7452 return 0;
7453 out:
7454 unregister_as_ext2();
7455 unregister_as_ext3();
7456 ext4_fc_destroy_dentry_cache();
7457 out05:
7458 destroy_inodecache();
7459 out1:
7460 ext4_exit_mballoc();
7461 out2:
7462 ext4_exit_sysfs();
7463 out3:
7464 ext4_exit_system_zone();
7465 out4:
7466 ext4_exit_pageio();
7467 out5:
7468 ext4_exit_post_read_processing();
7469 out6:
7470 ext4_exit_pending();
7471 out7:
7472 ext4_exit_es();
7473
7474 return err;
7475 }
7476
7477 static void __exit ext4_exit_fs(void)
7478 {
7479 ext4_destroy_lazyinit_thread();
7480 unregister_as_ext2();
7481 unregister_as_ext3();
7482 unregister_filesystem(&ext4_fs_type);
7483 ext4_fc_destroy_dentry_cache();
7484 destroy_inodecache();
7485 ext4_exit_mballoc();
7486 ext4_exit_sysfs();
7487 ext4_exit_system_zone();
7488 ext4_exit_pageio();
7489 ext4_exit_post_read_processing();
7490 ext4_exit_es();
7491 ext4_exit_pending();
7492 }
7493
7494 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7495 MODULE_DESCRIPTION("Fourth Extended Filesystem");
7496 MODULE_LICENSE("GPL");
7497 MODULE_SOFTDEP("pre: crc32c");
7498 module_init(ext4_init_fs)
7499 module_exit(ext4_exit_fs)
Linux Kernel