projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
ext4: avoid updating the superblock on a r/o mount if not needed
[linux-block.git] / fs / ext4 / super.c
... / ...
CommitLineData
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
63static struct ext4_lazy_init *ext4_li_info;
64static DEFINE_MUTEX(ext4_li_mtx);
65static struct ratelimit_state ext4_mount_msg_ratelimit;
66
67static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
68 unsigned long journal_devnum);
69static int ext4_show_options(struct seq_file *seq, struct dentry *root);
70static void ext4_update_super(struct super_block *sb);
71static int ext4_commit_super(struct super_block *sb);
72static int ext4_mark_recovery_complete(struct super_block *sb,
73 struct ext4_super_block *es);
74static int ext4_clear_journal_err(struct super_block *sb,
75 struct ext4_super_block *es);
76static int ext4_sync_fs(struct super_block *sb, int wait);
77static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
78static int ext4_unfreeze(struct super_block *sb);
79static int ext4_freeze(struct super_block *sb);
80static inline int ext2_feature_set_ok(struct super_block *sb);
81static inline int ext3_feature_set_ok(struct super_block *sb);
82static void ext4_destroy_lazyinit_thread(void);
83static void ext4_unregister_li_request(struct super_block *sb);
84static void ext4_clear_request_list(void);
85static struct inode *ext4_get_journal_inode(struct super_block *sb,
86 unsigned int journal_inum);
87static int ext4_validate_options(struct fs_context *fc);
88static int ext4_check_opt_consistency(struct fs_context *fc,
89 struct super_block *sb);
90static void ext4_apply_options(struct fs_context *fc, struct super_block *sb);
91static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param);
92static int ext4_get_tree(struct fs_context *fc);
93static int ext4_reconfigure(struct fs_context *fc);
94static void ext4_fc_free(struct fs_context *fc);
95static int ext4_init_fs_context(struct fs_context *fc);
96static const struct fs_parameter_spec ext4_param_specs[];
97
98/*
99 * Lock ordering
100 *
101 * page fault path:
102 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
103 * -> page lock -> i_data_sem (rw)
104 *
105 * buffered write path:
106 * sb_start_write -> i_mutex -> mmap_lock
107 * sb_start_write -> i_mutex -> transaction start -> page lock ->
108 * i_data_sem (rw)
109 *
110 * truncate:
111 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
112 * page lock
113 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
114 * i_data_sem (rw)
115 *
116 * direct IO:
117 * sb_start_write -> i_mutex -> mmap_lock
118 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
119 *
120 * writepages:
121 * transaction start -> page lock(s) -> i_data_sem (rw)
122 */
123
124static const struct fs_context_operations ext4_context_ops = {
125 .parse_param = ext4_parse_param,
126 .get_tree = ext4_get_tree,
127 .reconfigure = ext4_reconfigure,
128 .free = ext4_fc_free,
129};
130
131
132#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
133static struct file_system_type ext2_fs_type = {
134 .owner = THIS_MODULE,
135 .name = "ext2",
136 .init_fs_context = ext4_init_fs_context,
137 .parameters = ext4_param_specs,
138 .kill_sb = kill_block_super,
139 .fs_flags = FS_REQUIRES_DEV,
140};
141MODULE_ALIAS_FS("ext2");
142MODULE_ALIAS("ext2");
143#define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
144#else
145#define IS_EXT2_SB(sb) (0)
146#endif
147
148
149static struct file_system_type ext3_fs_type = {
150 .owner = THIS_MODULE,
151 .name = "ext3",
152 .init_fs_context = ext4_init_fs_context,
153 .parameters = ext4_param_specs,
154 .kill_sb = kill_block_super,
155 .fs_flags = FS_REQUIRES_DEV,
156};
157MODULE_ALIAS_FS("ext3");
158MODULE_ALIAS("ext3");
159#define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
160
161
162static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
163 bh_end_io_t *end_io)
164{
165 /*
166 * buffer's verified bit is no longer valid after reading from
167 * disk again due to write out error, clear it to make sure we
168 * recheck the buffer contents.
169 */
170 clear_buffer_verified(bh);
171
172 bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
173 get_bh(bh);
174 submit_bh(REQ_OP_READ | op_flags, bh);
175}
176
177void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags,
178 bh_end_io_t *end_io)
179{
180 BUG_ON(!buffer_locked(bh));
181
182 if (ext4_buffer_uptodate(bh)) {
183 unlock_buffer(bh);
184 return;
185 }
186 __ext4_read_bh(bh, op_flags, end_io);
187}
188
189int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io)
190{
191 BUG_ON(!buffer_locked(bh));
192
193 if (ext4_buffer_uptodate(bh)) {
194 unlock_buffer(bh);
195 return 0;
196 }
197
198 __ext4_read_bh(bh, op_flags, end_io);
199
200 wait_on_buffer(bh);
201 if (buffer_uptodate(bh))
202 return 0;
203 return -EIO;
204}
205
206int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
207{
208 lock_buffer(bh);
209 if (!wait) {
210 ext4_read_bh_nowait(bh, op_flags, NULL);
211 return 0;
212 }
213 return ext4_read_bh(bh, op_flags, NULL);
214}
215
216/*
217 * This works like __bread_gfp() except it uses ERR_PTR for error
218 * returns. Currently with sb_bread it's impossible to distinguish
219 * between ENOMEM and EIO situations (since both result in a NULL
220 * return.
221 */
222static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
223 sector_t block,
224 blk_opf_t op_flags, gfp_t gfp)
225{
226 struct buffer_head *bh;
227 int ret;
228
229 bh = sb_getblk_gfp(sb, block, gfp);
230 if (bh == NULL)
231 return ERR_PTR(-ENOMEM);
232 if (ext4_buffer_uptodate(bh))
233 return bh;
234
235 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
236 if (ret) {
237 put_bh(bh);
238 return ERR_PTR(ret);
239 }
240 return bh;
241}
242
243struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
244 blk_opf_t op_flags)
245{
246 return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
247}
248
249struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
250 sector_t block)
251{
252 return __ext4_sb_bread_gfp(sb, block, 0, 0);
253}
254
255void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
256{
257 struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
258
259 if (likely(bh)) {
260 if (trylock_buffer(bh))
261 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL);
262 brelse(bh);
263 }
264}
265
266static int ext4_verify_csum_type(struct super_block *sb,
267 struct ext4_super_block *es)
268{
269 if (!ext4_has_feature_metadata_csum(sb))
270 return 1;
271
272 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
273}
274
275__le32 ext4_superblock_csum(struct super_block *sb,
276 struct ext4_super_block *es)
277{
278 struct ext4_sb_info *sbi = EXT4_SB(sb);
279 int offset = offsetof(struct ext4_super_block, s_checksum);
280 __u32 csum;
281
282 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
283
284 return cpu_to_le32(csum);
285}
286
287static int ext4_superblock_csum_verify(struct super_block *sb,
288 struct ext4_super_block *es)
289{
290 if (!ext4_has_metadata_csum(sb))
291 return 1;
292
293 return es->s_checksum == ext4_superblock_csum(sb, es);
294}
295
296void ext4_superblock_csum_set(struct super_block *sb)
297{
298 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
299
300 if (!ext4_has_metadata_csum(sb))
301 return;
302
303 es->s_checksum = ext4_superblock_csum(sb, es);
304}
305
306ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
307 struct ext4_group_desc *bg)
308{
309 return le32_to_cpu(bg->bg_block_bitmap_lo) |
310 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
311 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
312}
313
314ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
315 struct ext4_group_desc *bg)
316{
317 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
319 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
320}
321
322ext4_fsblk_t ext4_inode_table(struct super_block *sb,
323 struct ext4_group_desc *bg)
324{
325 return le32_to_cpu(bg->bg_inode_table_lo) |
326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
328}
329
330__u32 ext4_free_group_clusters(struct super_block *sb,
331 struct ext4_group_desc *bg)
332{
333 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
335 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
336}
337
338__u32 ext4_free_inodes_count(struct super_block *sb,
339 struct ext4_group_desc *bg)
340{
341 return le16_to_cpu(bg->bg_free_inodes_count_lo) |
342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
343 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
344}
345
346__u32 ext4_used_dirs_count(struct super_block *sb,
347 struct ext4_group_desc *bg)
348{
349 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
351 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
352}
353
354__u32 ext4_itable_unused_count(struct super_block *sb,
355 struct ext4_group_desc *bg)
356{
357 return le16_to_cpu(bg->bg_itable_unused_lo) |
358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
359 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
360}
361
362void ext4_block_bitmap_set(struct super_block *sb,
363 struct ext4_group_desc *bg, ext4_fsblk_t blk)
364{
365 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
366 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
367 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
368}
369
370void ext4_inode_bitmap_set(struct super_block *sb,
371 struct ext4_group_desc *bg, ext4_fsblk_t blk)
372{
373 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
375 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
376}
377
378void ext4_inode_table_set(struct super_block *sb,
379 struct ext4_group_desc *bg, ext4_fsblk_t blk)
380{
381 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
383 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
384}
385
386void ext4_free_group_clusters_set(struct super_block *sb,
387 struct ext4_group_desc *bg, __u32 count)
388{
389 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
391 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
392}
393
394void ext4_free_inodes_set(struct super_block *sb,
395 struct ext4_group_desc *bg, __u32 count)
396{
397 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
399 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
400}
401
402void ext4_used_dirs_set(struct super_block *sb,
403 struct ext4_group_desc *bg, __u32 count)
404{
405 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
407 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
408}
409
410void ext4_itable_unused_set(struct super_block *sb,
411 struct ext4_group_desc *bg, __u32 count)
412{
413 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
415 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
416}
417
418static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
419{
420 now = clamp_val(now, 0, (1ull << 40) - 1);
421
422 *lo = cpu_to_le32(lower_32_bits(now));
423 *hi = upper_32_bits(now);
424}
425
426static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
427{
428 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
429}
430#define ext4_update_tstamp(es, tstamp) \
431 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
432 ktime_get_real_seconds())
433#define ext4_get_tstamp(es, tstamp) \
434 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
435
436/*
437 * The del_gendisk() function uninitializes the disk-specific data
438 * structures, including the bdi structure, without telling anyone
439 * else. Once this happens, any attempt to call mark_buffer_dirty()
440 * (for example, by ext4_commit_super), will cause a kernel OOPS.
441 * This is a kludge to prevent these oops until we can put in a proper
442 * hook in del_gendisk() to inform the VFS and file system layers.
443 */
444static int block_device_ejected(struct super_block *sb)
445{
446 struct inode *bd_inode = sb->s_bdev->bd_inode;
447 struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
448
449 return bdi->dev == NULL;
450}
451
452static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
453{
454 struct super_block *sb = journal->j_private;
455 struct ext4_sb_info *sbi = EXT4_SB(sb);
456 int error = is_journal_aborted(journal);
457 struct ext4_journal_cb_entry *jce;
458
459 BUG_ON(txn->t_state == T_FINISHED);
460
461 ext4_process_freed_data(sb, txn->t_tid);
462
463 spin_lock(&sbi->s_md_lock);
464 while (!list_empty(&txn->t_private_list)) {
465 jce = list_entry(txn->t_private_list.next,
466 struct ext4_journal_cb_entry, jce_list);
467 list_del_init(&jce->jce_list);
468 spin_unlock(&sbi->s_md_lock);
469 jce->jce_func(sb, jce, error);
470 spin_lock(&sbi->s_md_lock);
471 }
472 spin_unlock(&sbi->s_md_lock);
473}
474
475/*
476 * This writepage callback for write_cache_pages()
477 * takes care of a few cases after page cleaning.
478 *
479 * write_cache_pages() already checks for dirty pages
480 * and calls clear_page_dirty_for_io(), which we want,
481 * to write protect the pages.
482 *
483 * However, we may have to redirty a page (see below.)
484 */
485static int ext4_journalled_writepage_callback(struct folio *folio,
486 struct writeback_control *wbc,
487 void *data)
488{
489 transaction_t *transaction = (transaction_t *) data;
490 struct buffer_head *bh, *head;
491 struct journal_head *jh;
492
493 bh = head = folio_buffers(folio);
494 do {
495 /*
496 * We have to redirty a page in these cases:
497 * 1) If buffer is dirty, it means the page was dirty because it
498 * contains a buffer that needs checkpointing. So the dirty bit
499 * needs to be preserved so that checkpointing writes the buffer
500 * properly.
501 * 2) If buffer is not part of the committing transaction
502 * (we may have just accidentally come across this buffer because
503 * inode range tracking is not exact) or if the currently running
504 * transaction already contains this buffer as well, dirty bit
505 * needs to be preserved so that the buffer gets writeprotected
506 * properly on running transaction's commit.
507 */
508 jh = bh2jh(bh);
509 if (buffer_dirty(bh) ||
510 (jh && (jh->b_transaction != transaction ||
511 jh->b_next_transaction))) {
512 folio_redirty_for_writepage(wbc, folio);
513 goto out;
514 }
515 } while ((bh = bh->b_this_page) != head);
516
517out:
518 return AOP_WRITEPAGE_ACTIVATE;
519}
520
521static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
522{
523 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
524 struct writeback_control wbc = {
525 .sync_mode = WB_SYNC_ALL,
526 .nr_to_write = LONG_MAX,
527 .range_start = jinode->i_dirty_start,
528 .range_end = jinode->i_dirty_end,
529 };
530
531 return write_cache_pages(mapping, &wbc,
532 ext4_journalled_writepage_callback,
533 jinode->i_transaction);
534}
535
536static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
537{
538 int ret;
539
540 if (ext4_should_journal_data(jinode->i_vfs_inode))
541 ret = ext4_journalled_submit_inode_data_buffers(jinode);
542 else
543 ret = ext4_normal_submit_inode_data_buffers(jinode);
544 return ret;
545}
546
547static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
548{
549 int ret = 0;
550
551 if (!ext4_should_journal_data(jinode->i_vfs_inode))
552 ret = jbd2_journal_finish_inode_data_buffers(jinode);
553
554 return ret;
555}
556
557static bool system_going_down(void)
558{
559 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
560 || system_state == SYSTEM_RESTART;
561}
562
563struct ext4_err_translation {
564 int code;
565 int errno;
566};
567
568#define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
569
570static struct ext4_err_translation err_translation[] = {
571 EXT4_ERR_TRANSLATE(EIO),
572 EXT4_ERR_TRANSLATE(ENOMEM),
573 EXT4_ERR_TRANSLATE(EFSBADCRC),
574 EXT4_ERR_TRANSLATE(EFSCORRUPTED),
575 EXT4_ERR_TRANSLATE(ENOSPC),
576 EXT4_ERR_TRANSLATE(ENOKEY),
577 EXT4_ERR_TRANSLATE(EROFS),
578 EXT4_ERR_TRANSLATE(EFBIG),
579 EXT4_ERR_TRANSLATE(EEXIST),
580 EXT4_ERR_TRANSLATE(ERANGE),
581 EXT4_ERR_TRANSLATE(EOVERFLOW),
582 EXT4_ERR_TRANSLATE(EBUSY),
583 EXT4_ERR_TRANSLATE(ENOTDIR),
584 EXT4_ERR_TRANSLATE(ENOTEMPTY),
585 EXT4_ERR_TRANSLATE(ESHUTDOWN),
586 EXT4_ERR_TRANSLATE(EFAULT),
587};
588
589static int ext4_errno_to_code(int errno)
590{
591 int i;
592
593 for (i = 0; i < ARRAY_SIZE(err_translation); i++)
594 if (err_translation[i].errno == errno)
595 return err_translation[i].code;
596 return EXT4_ERR_UNKNOWN;
597}
598
599static void save_error_info(struct super_block *sb, int error,
600 __u32 ino, __u64 block,
601 const char *func, unsigned int line)
602{
603 struct ext4_sb_info *sbi = EXT4_SB(sb);
604
605 /* We default to EFSCORRUPTED error... */
606 if (error == 0)
607 error = EFSCORRUPTED;
608
609 spin_lock(&sbi->s_error_lock);
610 sbi->s_add_error_count++;
611 sbi->s_last_error_code = error;
612 sbi->s_last_error_line = line;
613 sbi->s_last_error_ino = ino;
614 sbi->s_last_error_block = block;
615 sbi->s_last_error_func = func;
616 sbi->s_last_error_time = ktime_get_real_seconds();
617 if (!sbi->s_first_error_time) {
618 sbi->s_first_error_code = error;
619 sbi->s_first_error_line = line;
620 sbi->s_first_error_ino = ino;
621 sbi->s_first_error_block = block;
622 sbi->s_first_error_func = func;
623 sbi->s_first_error_time = sbi->s_last_error_time;
624 }
625 spin_unlock(&sbi->s_error_lock);
626}
627
628/* Deal with the reporting of failure conditions on a filesystem such as
629 * inconsistencies detected or read IO failures.
630 *
631 * On ext2, we can store the error state of the filesystem in the
632 * superblock. That is not possible on ext4, because we may have other
633 * write ordering constraints on the superblock which prevent us from
634 * writing it out straight away; and given that the journal is about to
635 * be aborted, we can't rely on the current, or future, transactions to
636 * write out the superblock safely.
637 *
638 * We'll just use the jbd2_journal_abort() error code to record an error in
639 * the journal instead. On recovery, the journal will complain about
640 * that error until we've noted it down and cleared it.
641 *
642 * If force_ro is set, we unconditionally force the filesystem into an
643 * ABORT|READONLY state, unless the error response on the fs has been set to
644 * panic in which case we take the easy way out and panic immediately. This is
645 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
646 * at a critical moment in log management.
647 */
648static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
649 __u32 ino, __u64 block,
650 const char *func, unsigned int line)
651{
652 journal_t *journal = EXT4_SB(sb)->s_journal;
653 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
654
655 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
656 if (test_opt(sb, WARN_ON_ERROR))
657 WARN_ON_ONCE(1);
658
659 if (!continue_fs && !sb_rdonly(sb)) {
660 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
661 if (journal)
662 jbd2_journal_abort(journal, -EIO);
663 }
664
665 if (!bdev_read_only(sb->s_bdev)) {
666 save_error_info(sb, error, ino, block, func, line);
667 /*
668 * In case the fs should keep running, we need to writeout
669 * superblock through the journal. Due to lock ordering
670 * constraints, it may not be safe to do it right here so we
671 * defer superblock flushing to a workqueue.
672 */
673 if (continue_fs && journal)
674 schedule_work(&EXT4_SB(sb)->s_error_work);
675 else
676 ext4_commit_super(sb);
677 }
678
679 /*
680 * We force ERRORS_RO behavior when system is rebooting. Otherwise we
681 * could panic during 'reboot -f' as the underlying device got already
682 * disabled.
683 */
684 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
685 panic("EXT4-fs (device %s): panic forced after error\n",
686 sb->s_id);
687 }
688
689 if (sb_rdonly(sb) || continue_fs)
690 return;
691
692 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
693 /*
694 * Make sure updated value of ->s_mount_flags will be visible before
695 * ->s_flags update
696 */
697 smp_wmb();
698 sb->s_flags |= SB_RDONLY;
699}
700
701static void flush_stashed_error_work(struct work_struct *work)
702{
703 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
704 s_error_work);
705 journal_t *journal = sbi->s_journal;
706 handle_t *handle;
707
708 /*
709 * If the journal is still running, we have to write out superblock
710 * through the journal to avoid collisions of other journalled sb
711 * updates.
712 *
713 * We use directly jbd2 functions here to avoid recursing back into
714 * ext4 error handling code during handling of previous errors.
715 */
716 if (!sb_rdonly(sbi->s_sb) && journal) {
717 struct buffer_head *sbh = sbi->s_sbh;
718 handle = jbd2_journal_start(journal, 1);
719 if (IS_ERR(handle))
720 goto write_directly;
721 if (jbd2_journal_get_write_access(handle, sbh)) {
722 jbd2_journal_stop(handle);
723 goto write_directly;
724 }
725 ext4_update_super(sbi->s_sb);
726 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
727 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
728 "superblock detected");
729 clear_buffer_write_io_error(sbh);
730 set_buffer_uptodate(sbh);
731 }
732
733 if (jbd2_journal_dirty_metadata(handle, sbh)) {
734 jbd2_journal_stop(handle);
735 goto write_directly;
736 }
737 jbd2_journal_stop(handle);
738 ext4_notify_error_sysfs(sbi);
739 return;
740 }
741write_directly:
742 /*
743 * Write through journal failed. Write sb directly to get error info
744 * out and hope for the best.
745 */
746 ext4_commit_super(sbi->s_sb);
747 ext4_notify_error_sysfs(sbi);
748}
749
750#define ext4_error_ratelimit(sb) \
751 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \
752 "EXT4-fs error")
753
754void __ext4_error(struct super_block *sb, const char *function,
755 unsigned int line, bool force_ro, int error, __u64 block,
756 const char *fmt, ...)
757{
758 struct va_format vaf;
759 va_list args;
760
761 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
762 return;
763
764 trace_ext4_error(sb, function, line);
765 if (ext4_error_ratelimit(sb)) {
766 va_start(args, fmt);
767 vaf.fmt = fmt;
768 vaf.va = &args;
769 printk(KERN_CRIT
770 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
771 sb->s_id, function, line, current->comm, &vaf);
772 va_end(args);
773 }
774 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
775
776 ext4_handle_error(sb, force_ro, error, 0, block, function, line);
777}
778
779void __ext4_error_inode(struct inode *inode, const char *function,
780 unsigned int line, ext4_fsblk_t block, int error,
781 const char *fmt, ...)
782{
783 va_list args;
784 struct va_format vaf;
785
786 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
787 return;
788
789 trace_ext4_error(inode->i_sb, function, line);
790 if (ext4_error_ratelimit(inode->i_sb)) {
791 va_start(args, fmt);
792 vaf.fmt = fmt;
793 vaf.va = &args;
794 if (block)
795 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
796 "inode #%lu: block %llu: comm %s: %pV\n",
797 inode->i_sb->s_id, function, line, inode->i_ino,
798 block, current->comm, &vaf);
799 else
800 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
801 "inode #%lu: comm %s: %pV\n",
802 inode->i_sb->s_id, function, line, inode->i_ino,
803 current->comm, &vaf);
804 va_end(args);
805 }
806 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
807
808 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
809 function, line);
810}
811
812void __ext4_error_file(struct file *file, const char *function,
813 unsigned int line, ext4_fsblk_t block,
814 const char *fmt, ...)
815{
816 va_list args;
817 struct va_format vaf;
818 struct inode *inode = file_inode(file);
819 char pathname[80], *path;
820
821 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
822 return;
823
824 trace_ext4_error(inode->i_sb, function, line);
825 if (ext4_error_ratelimit(inode->i_sb)) {
826 path = file_path(file, pathname, sizeof(pathname));
827 if (IS_ERR(path))
828 path = "(unknown)";
829 va_start(args, fmt);
830 vaf.fmt = fmt;
831 vaf.va = &args;
832 if (block)
833 printk(KERN_CRIT
834 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
835 "block %llu: comm %s: path %s: %pV\n",
836 inode->i_sb->s_id, function, line, inode->i_ino,
837 block, current->comm, path, &vaf);
838 else
839 printk(KERN_CRIT
840 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
841 "comm %s: path %s: %pV\n",
842 inode->i_sb->s_id, function, line, inode->i_ino,
843 current->comm, path, &vaf);
844 va_end(args);
845 }
846 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
847
848 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
849 function, line);
850}
851
852const char *ext4_decode_error(struct super_block *sb, int errno,
853 char nbuf[16])
854{
855 char *errstr = NULL;
856
857 switch (errno) {
858 case -EFSCORRUPTED:
859 errstr = "Corrupt filesystem";
860 break;
861 case -EFSBADCRC:
862 errstr = "Filesystem failed CRC";
863 break;
864 case -EIO:
865 errstr = "IO failure";
866 break;
867 case -ENOMEM:
868 errstr = "Out of memory";
869 break;
870 case -EROFS:
871 if (!sb || (EXT4_SB(sb)->s_journal &&
872 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
873 errstr = "Journal has aborted";
874 else
875 errstr = "Readonly filesystem";
876 break;
877 default:
878 /* If the caller passed in an extra buffer for unknown
879 * errors, textualise them now. Else we just return
880 * NULL. */
881 if (nbuf) {
882 /* Check for truncated error codes... */
883 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
884 errstr = nbuf;
885 }
886 break;
887 }
888
889 return errstr;
890}
891
892/* __ext4_std_error decodes expected errors from journaling functions
893 * automatically and invokes the appropriate error response. */
894
895void __ext4_std_error(struct super_block *sb, const char *function,
896 unsigned int line, int errno)
897{
898 char nbuf[16];
899 const char *errstr;
900
901 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
902 return;
903
904 /* Special case: if the error is EROFS, and we're not already
905 * inside a transaction, then there's really no point in logging
906 * an error. */
907 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
908 return;
909
910 if (ext4_error_ratelimit(sb)) {
911 errstr = ext4_decode_error(sb, errno, nbuf);
912 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
913 sb->s_id, function, line, errstr);
914 }
915 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
916
917 ext4_handle_error(sb, false, -errno, 0, 0, function, line);
918}
919
920void __ext4_msg(struct super_block *sb,
921 const char *prefix, const char *fmt, ...)
922{
923 struct va_format vaf;
924 va_list args;
925
926 if (sb) {
927 atomic_inc(&EXT4_SB(sb)->s_msg_count);
928 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
929 "EXT4-fs"))
930 return;
931 }
932
933 va_start(args, fmt);
934 vaf.fmt = fmt;
935 vaf.va = &args;
936 if (sb)
937 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
938 else
939 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
940 va_end(args);
941}
942
943static int ext4_warning_ratelimit(struct super_block *sb)
944{
945 atomic_inc(&EXT4_SB(sb)->s_warning_count);
946 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
947 "EXT4-fs warning");
948}
949
950void __ext4_warning(struct super_block *sb, const char *function,
951 unsigned int line, const char *fmt, ...)
952{
953 struct va_format vaf;
954 va_list args;
955
956 if (!ext4_warning_ratelimit(sb))
957 return;
958
959 va_start(args, fmt);
960 vaf.fmt = fmt;
961 vaf.va = &args;
962 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
963 sb->s_id, function, line, &vaf);
964 va_end(args);
965}
966
967void __ext4_warning_inode(const struct inode *inode, const char *function,
968 unsigned int line, const char *fmt, ...)
969{
970 struct va_format vaf;
971 va_list args;
972
973 if (!ext4_warning_ratelimit(inode->i_sb))
974 return;
975
976 va_start(args, fmt);
977 vaf.fmt = fmt;
978 vaf.va = &args;
979 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
980 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
981 function, line, inode->i_ino, current->comm, &vaf);
982 va_end(args);
983}
984
985void __ext4_grp_locked_error(const char *function, unsigned int line,
986 struct super_block *sb, ext4_group_t grp,
987 unsigned long ino, ext4_fsblk_t block,
988 const char *fmt, ...)
989__releases(bitlock)
990__acquires(bitlock)
991{
992 struct va_format vaf;
993 va_list args;
994
995 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
996 return;
997
998 trace_ext4_error(sb, function, line);
999 if (ext4_error_ratelimit(sb)) {
1000 va_start(args, fmt);
1001 vaf.fmt = fmt;
1002 vaf.va = &args;
1003 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1004 sb->s_id, function, line, grp);
1005 if (ino)
1006 printk(KERN_CONT "inode %lu: ", ino);
1007 if (block)
1008 printk(KERN_CONT "block %llu:",
1009 (unsigned long long) block);
1010 printk(KERN_CONT "%pV\n", &vaf);
1011 va_end(args);
1012 }
1013
1014 if (test_opt(sb, ERRORS_CONT)) {
1015 if (test_opt(sb, WARN_ON_ERROR))
1016 WARN_ON_ONCE(1);
1017 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1018 if (!bdev_read_only(sb->s_bdev)) {
1019 save_error_info(sb, EFSCORRUPTED, ino, block, function,
1020 line);
1021 schedule_work(&EXT4_SB(sb)->s_error_work);
1022 }
1023 return;
1024 }
1025 ext4_unlock_group(sb, grp);
1026 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1027 /*
1028 * We only get here in the ERRORS_RO case; relocking the group
1029 * may be dangerous, but nothing bad will happen since the
1030 * filesystem will have already been marked read/only and the
1031 * journal has been aborted. We return 1 as a hint to callers
1032 * who might what to use the return value from
1033 * ext4_grp_locked_error() to distinguish between the
1034 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1035 * aggressively from the ext4 function in question, with a
1036 * more appropriate error code.
1037 */
1038 ext4_lock_group(sb, grp);
1039 return;
1040}
1041
1042void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1043 ext4_group_t group,
1044 unsigned int flags)
1045{
1046 struct ext4_sb_info *sbi = EXT4_SB(sb);
1047 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1048 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1049 int ret;
1050
1051 if (!grp || !gdp)
1052 return;
1053 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1054 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1055 &grp->bb_state);
1056 if (!ret)
1057 percpu_counter_sub(&sbi->s_freeclusters_counter,
1058 grp->bb_free);
1059 }
1060
1061 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1062 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1063 &grp->bb_state);
1064 if (!ret && gdp) {
1065 int count;
1066
1067 count = ext4_free_inodes_count(sb, gdp);
1068 percpu_counter_sub(&sbi->s_freeinodes_counter,
1069 count);
1070 }
1071 }
1072}
1073
1074void ext4_update_dynamic_rev(struct super_block *sb)
1075{
1076 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1077
1078 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1079 return;
1080
1081 ext4_warning(sb,
1082 "updating to rev %d because of new feature flag, "
1083 "running e2fsck is recommended",
1084 EXT4_DYNAMIC_REV);
1085
1086 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1087 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1088 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1089 /* leave es->s_feature_*compat flags alone */
1090 /* es->s_uuid will be set by e2fsck if empty */
1091
1092 /*
1093 * The rest of the superblock fields should be zero, and if not it
1094 * means they are likely already in use, so leave them alone. We
1095 * can leave it up to e2fsck to clean up any inconsistencies there.
1096 */
1097}
1098
1099/*
1100 * Open the external journal device
1101 */
1102static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1103{
1104 struct block_device *bdev;
1105
1106 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1107 if (IS_ERR(bdev))
1108 goto fail;
1109 return bdev;
1110
1111fail:
1112 ext4_msg(sb, KERN_ERR,
1113 "failed to open journal device unknown-block(%u,%u) %ld",
1114 MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1115 return NULL;
1116}
1117
1118/*
1119 * Release the journal device
1120 */
1121static void ext4_blkdev_put(struct block_device *bdev)
1122{
1123 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1124}
1125
1126static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1127{
1128 struct block_device *bdev;
1129 bdev = sbi->s_journal_bdev;
1130 if (bdev) {
1131 /*
1132 * Invalidate the journal device's buffers. We don't want them
1133 * floating about in memory - the physical journal device may
1134 * hotswapped, and it breaks the `ro-after' testing code.
1135 */
1136 invalidate_bdev(bdev);
1137 ext4_blkdev_put(bdev);
1138 sbi->s_journal_bdev = NULL;
1139 }
1140}
1141
1142static inline struct inode *orphan_list_entry(struct list_head *l)
1143{
1144 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1145}
1146
1147static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1148{
1149 struct list_head *l;
1150
1151 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1152 le32_to_cpu(sbi->s_es->s_last_orphan));
1153
1154 printk(KERN_ERR "sb_info orphan list:\n");
1155 list_for_each(l, &sbi->s_orphan) {
1156 struct inode *inode = orphan_list_entry(l);
1157 printk(KERN_ERR " "
1158 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1159 inode->i_sb->s_id, inode->i_ino, inode,
1160 inode->i_mode, inode->i_nlink,
1161 NEXT_ORPHAN(inode));
1162 }
1163}
1164
1165#ifdef CONFIG_QUOTA
1166static int ext4_quota_off(struct super_block *sb, int type);
1167
1168static inline void ext4_quotas_off(struct super_block *sb, int type)
1169{
1170 BUG_ON(type > EXT4_MAXQUOTAS);
1171
1172 /* Use our quota_off function to clear inode flags etc. */
1173 for (type--; type >= 0; type--)
1174 ext4_quota_off(sb, type);
1175}
1176
1177/*
1178 * This is a helper function which is used in the mount/remount
1179 * codepaths (which holds s_umount) to fetch the quota file name.
1180 */
1181static inline char *get_qf_name(struct super_block *sb,
1182 struct ext4_sb_info *sbi,
1183 int type)
1184{
1185 return rcu_dereference_protected(sbi->s_qf_names[type],
1186 lockdep_is_held(&sb->s_umount));
1187}
1188#else
1189static inline void ext4_quotas_off(struct super_block *sb, int type)
1190{
1191}
1192#endif
1193
1194static int ext4_percpu_param_init(struct ext4_sb_info *sbi)
1195{
1196 ext4_fsblk_t block;
1197 int err;
1198
1199 block = ext4_count_free_clusters(sbi->s_sb);
1200 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block));
1201 err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
1202 GFP_KERNEL);
1203 if (!err) {
1204 unsigned long freei = ext4_count_free_inodes(sbi->s_sb);
1205 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
1206 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
1207 GFP_KERNEL);
1208 }
1209 if (!err)
1210 err = percpu_counter_init(&sbi->s_dirs_counter,
1211 ext4_count_dirs(sbi->s_sb), GFP_KERNEL);
1212 if (!err)
1213 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
1214 GFP_KERNEL);
1215 if (!err)
1216 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
1217 GFP_KERNEL);
1218 if (!err)
1219 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
1220
1221 if (err)
1222 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory");
1223
1224 return err;
1225}
1226
1227static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi)
1228{
1229 percpu_counter_destroy(&sbi->s_freeclusters_counter);
1230 percpu_counter_destroy(&sbi->s_freeinodes_counter);
1231 percpu_counter_destroy(&sbi->s_dirs_counter);
1232 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1233 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1234 percpu_free_rwsem(&sbi->s_writepages_rwsem);
1235}
1236
1237static void ext4_group_desc_free(struct ext4_sb_info *sbi)
1238{
1239 struct buffer_head **group_desc;
1240 int i;
1241
1242 rcu_read_lock();
1243 group_desc = rcu_dereference(sbi->s_group_desc);
1244 for (i = 0; i < sbi->s_gdb_count; i++)
1245 brelse(group_desc[i]);
1246 kvfree(group_desc);
1247 rcu_read_unlock();
1248}
1249
1250static void ext4_flex_groups_free(struct ext4_sb_info *sbi)
1251{
1252 struct flex_groups **flex_groups;
1253 int i;
1254
1255 rcu_read_lock();
1256 flex_groups = rcu_dereference(sbi->s_flex_groups);
1257 if (flex_groups) {
1258 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1259 kvfree(flex_groups[i]);
1260 kvfree(flex_groups);
1261 }
1262 rcu_read_unlock();
1263}
1264
1265static void ext4_put_super(struct super_block *sb)
1266{
1267 struct ext4_sb_info *sbi = EXT4_SB(sb);
1268 struct ext4_super_block *es = sbi->s_es;
1269 int aborted = 0;
1270 int err;
1271
1272 /*
1273 * Unregister sysfs before destroying jbd2 journal.
1274 * Since we could still access attr_journal_task attribute via sysfs
1275 * path which could have sbi->s_journal->j_task as NULL
1276 * Unregister sysfs before flush sbi->s_error_work.
1277 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1278 * read metadata verify failed then will queue error work.
1279 * flush_stashed_error_work will call start_this_handle may trigger
1280 * BUG_ON.
1281 */
1282 ext4_unregister_sysfs(sb);
1283
1284 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1285 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.",
1286 &sb->s_uuid);
1287
1288 ext4_unregister_li_request(sb);
1289 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
1290
1291 flush_work(&sbi->s_error_work);
1292 destroy_workqueue(sbi->rsv_conversion_wq);
1293 ext4_release_orphan_info(sb);
1294
1295 if (sbi->s_journal) {
1296 aborted = is_journal_aborted(sbi->s_journal);
1297 err = jbd2_journal_destroy(sbi->s_journal);
1298 sbi->s_journal = NULL;
1299 if ((err < 0) && !aborted) {
1300 ext4_abort(sb, -err, "Couldn't clean up the journal");
1301 }
1302 }
1303
1304 ext4_es_unregister_shrinker(sbi);
1305 timer_shutdown_sync(&sbi->s_err_report);
1306 ext4_release_system_zone(sb);
1307 ext4_mb_release(sb);
1308 ext4_ext_release(sb);
1309
1310 if (!sb_rdonly(sb) && !aborted) {
1311 ext4_clear_feature_journal_needs_recovery(sb);
1312 ext4_clear_feature_orphan_present(sb);
1313 es->s_state = cpu_to_le16(sbi->s_mount_state);
1314 }
1315 if (!sb_rdonly(sb))
1316 ext4_commit_super(sb);
1317
1318 ext4_group_desc_free(sbi);
1319 ext4_flex_groups_free(sbi);
1320 ext4_percpu_param_destroy(sbi);
1321#ifdef CONFIG_QUOTA
1322 for (int i = 0; i < EXT4_MAXQUOTAS; i++)
1323 kfree(get_qf_name(sb, sbi, i));
1324#endif
1325
1326 /* Debugging code just in case the in-memory inode orphan list
1327 * isn't empty. The on-disk one can be non-empty if we've
1328 * detected an error and taken the fs readonly, but the
1329 * in-memory list had better be clean by this point. */
1330 if (!list_empty(&sbi->s_orphan))
1331 dump_orphan_list(sb, sbi);
1332 ASSERT(list_empty(&sbi->s_orphan));
1333
1334 sync_blockdev(sb->s_bdev);
1335 invalidate_bdev(sb->s_bdev);
1336 if (sbi->s_journal_bdev) {
1337 sync_blockdev(sbi->s_journal_bdev);
1338 ext4_blkdev_remove(sbi);
1339 }
1340
1341 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1342 sbi->s_ea_inode_cache = NULL;
1343
1344 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1345 sbi->s_ea_block_cache = NULL;
1346
1347 ext4_stop_mmpd(sbi);
1348
1349 brelse(sbi->s_sbh);
1350 sb->s_fs_info = NULL;
1351 /*
1352 * Now that we are completely done shutting down the
1353 * superblock, we need to actually destroy the kobject.
1354 */
1355 kobject_put(&sbi->s_kobj);
1356 wait_for_completion(&sbi->s_kobj_unregister);
1357 if (sbi->s_chksum_driver)
1358 crypto_free_shash(sbi->s_chksum_driver);
1359 kfree(sbi->s_blockgroup_lock);
1360 fs_put_dax(sbi->s_daxdev, NULL);
1361 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1362#if IS_ENABLED(CONFIG_UNICODE)
1363 utf8_unload(sb->s_encoding);
1364#endif
1365 kfree(sbi);
1366}
1367
1368static struct kmem_cache *ext4_inode_cachep;
1369
1370/*
1371 * Called inside transaction, so use GFP_NOFS
1372 */
1373static struct inode *ext4_alloc_inode(struct super_block *sb)
1374{
1375 struct ext4_inode_info *ei;
1376
1377 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1378 if (!ei)
1379 return NULL;
1380
1381 inode_set_iversion(&ei->vfs_inode, 1);
1382 ei->i_flags = 0;
1383 spin_lock_init(&ei->i_raw_lock);
1384 ei->i_prealloc_node = RB_ROOT;
1385 atomic_set(&ei->i_prealloc_active, 0);
1386 rwlock_init(&ei->i_prealloc_lock);
1387 ext4_es_init_tree(&ei->i_es_tree);
1388 rwlock_init(&ei->i_es_lock);
1389 INIT_LIST_HEAD(&ei->i_es_list);
1390 ei->i_es_all_nr = 0;
1391 ei->i_es_shk_nr = 0;
1392 ei->i_es_shrink_lblk = 0;
1393 ei->i_reserved_data_blocks = 0;
1394 spin_lock_init(&(ei->i_block_reservation_lock));
1395 ext4_init_pending_tree(&ei->i_pending_tree);
1396#ifdef CONFIG_QUOTA
1397 ei->i_reserved_quota = 0;
1398 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1399#endif
1400 ei->jinode = NULL;
1401 INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1402 spin_lock_init(&ei->i_completed_io_lock);
1403 ei->i_sync_tid = 0;
1404 ei->i_datasync_tid = 0;
1405 atomic_set(&ei->i_unwritten, 0);
1406 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1407 ext4_fc_init_inode(&ei->vfs_inode);
1408 mutex_init(&ei->i_fc_lock);
1409 return &ei->vfs_inode;
1410}
1411
1412static int ext4_drop_inode(struct inode *inode)
1413{
1414 int drop = generic_drop_inode(inode);
1415
1416 if (!drop)
1417 drop = fscrypt_drop_inode(inode);
1418
1419 trace_ext4_drop_inode(inode, drop);
1420 return drop;
1421}
1422
1423static void ext4_free_in_core_inode(struct inode *inode)
1424{
1425 fscrypt_free_inode(inode);
1426 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1427 pr_warn("%s: inode %ld still in fc list",
1428 __func__, inode->i_ino);
1429 }
1430 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1431}
1432
1433static void ext4_destroy_inode(struct inode *inode)
1434{
1435 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1436 ext4_msg(inode->i_sb, KERN_ERR,
1437 "Inode %lu (%p): orphan list check failed!",
1438 inode->i_ino, EXT4_I(inode));
1439 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1440 EXT4_I(inode), sizeof(struct ext4_inode_info),
1441 true);
1442 dump_stack();
1443 }
1444
1445 if (EXT4_I(inode)->i_reserved_data_blocks)
1446 ext4_msg(inode->i_sb, KERN_ERR,
1447 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1448 inode->i_ino, EXT4_I(inode),
1449 EXT4_I(inode)->i_reserved_data_blocks);
1450}
1451
1452static void init_once(void *foo)
1453{
1454 struct ext4_inode_info *ei = foo;
1455
1456 INIT_LIST_HEAD(&ei->i_orphan);
1457 init_rwsem(&ei->xattr_sem);
1458 init_rwsem(&ei->i_data_sem);
1459 inode_init_once(&ei->vfs_inode);
1460 ext4_fc_init_inode(&ei->vfs_inode);
1461}
1462
1463static int __init init_inodecache(void)
1464{
1465 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1466 sizeof(struct ext4_inode_info), 0,
1467 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1468 SLAB_ACCOUNT),
1469 offsetof(struct ext4_inode_info, i_data),
1470 sizeof_field(struct ext4_inode_info, i_data),
1471 init_once);
1472 if (ext4_inode_cachep == NULL)
1473 return -ENOMEM;
1474 return 0;
1475}
1476
1477static void destroy_inodecache(void)
1478{
1479 /*
1480 * Make sure all delayed rcu free inodes are flushed before we
1481 * destroy cache.
1482 */
1483 rcu_barrier();
1484 kmem_cache_destroy(ext4_inode_cachep);
1485}
1486
1487void ext4_clear_inode(struct inode *inode)
1488{
1489 ext4_fc_del(inode);
1490 invalidate_inode_buffers(inode);
1491 clear_inode(inode);
1492 ext4_discard_preallocations(inode, 0);
1493 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1494 dquot_drop(inode);
1495 if (EXT4_I(inode)->jinode) {
1496 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1497 EXT4_I(inode)->jinode);
1498 jbd2_free_inode(EXT4_I(inode)->jinode);
1499 EXT4_I(inode)->jinode = NULL;
1500 }
1501 fscrypt_put_encryption_info(inode);
1502 fsverity_cleanup_inode(inode);
1503}
1504
1505static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1506 u64 ino, u32 generation)
1507{
1508 struct inode *inode;
1509
1510 /*
1511 * Currently we don't know the generation for parent directory, so
1512 * a generation of 0 means "accept any"
1513 */
1514 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1515 if (IS_ERR(inode))
1516 return ERR_CAST(inode);
1517 if (generation && inode->i_generation != generation) {
1518 iput(inode);
1519 return ERR_PTR(-ESTALE);
1520 }
1521
1522 return inode;
1523}
1524
1525static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1526 int fh_len, int fh_type)
1527{
1528 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1529 ext4_nfs_get_inode);
1530}
1531
1532static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1533 int fh_len, int fh_type)
1534{
1535 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1536 ext4_nfs_get_inode);
1537}
1538
1539static int ext4_nfs_commit_metadata(struct inode *inode)
1540{
1541 struct writeback_control wbc = {
1542 .sync_mode = WB_SYNC_ALL
1543 };
1544
1545 trace_ext4_nfs_commit_metadata(inode);
1546 return ext4_write_inode(inode, &wbc);
1547}
1548
1549#ifdef CONFIG_QUOTA
1550static const char * const quotatypes[] = INITQFNAMES;
1551#define QTYPE2NAME(t) (quotatypes[t])
1552
1553static int ext4_write_dquot(struct dquot *dquot);
1554static int ext4_acquire_dquot(struct dquot *dquot);
1555static int ext4_release_dquot(struct dquot *dquot);
1556static int ext4_mark_dquot_dirty(struct dquot *dquot);
1557static int ext4_write_info(struct super_block *sb, int type);
1558static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1559 const struct path *path);
1560static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1561 size_t len, loff_t off);
1562static ssize_t ext4_quota_write(struct super_block *sb, int type,
1563 const char *data, size_t len, loff_t off);
1564static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1565 unsigned int flags);
1566
1567static struct dquot **ext4_get_dquots(struct inode *inode)
1568{
1569 return EXT4_I(inode)->i_dquot;
1570}
1571
1572static const struct dquot_operations ext4_quota_operations = {
1573 .get_reserved_space = ext4_get_reserved_space,
1574 .write_dquot = ext4_write_dquot,
1575 .acquire_dquot = ext4_acquire_dquot,
1576 .release_dquot = ext4_release_dquot,
1577 .mark_dirty = ext4_mark_dquot_dirty,
1578 .write_info = ext4_write_info,
1579 .alloc_dquot = dquot_alloc,
1580 .destroy_dquot = dquot_destroy,
1581 .get_projid = ext4_get_projid,
1582 .get_inode_usage = ext4_get_inode_usage,
1583 .get_next_id = dquot_get_next_id,
1584};
1585
1586static const struct quotactl_ops ext4_qctl_operations = {
1587 .quota_on = ext4_quota_on,
1588 .quota_off = ext4_quota_off,
1589 .quota_sync = dquot_quota_sync,
1590 .get_state = dquot_get_state,
1591 .set_info = dquot_set_dqinfo,
1592 .get_dqblk = dquot_get_dqblk,
1593 .set_dqblk = dquot_set_dqblk,
1594 .get_nextdqblk = dquot_get_next_dqblk,
1595};
1596#endif
1597
1598static const struct super_operations ext4_sops = {
1599 .alloc_inode = ext4_alloc_inode,
1600 .free_inode = ext4_free_in_core_inode,
1601 .destroy_inode = ext4_destroy_inode,
1602 .write_inode = ext4_write_inode,
1603 .dirty_inode = ext4_dirty_inode,
1604 .drop_inode = ext4_drop_inode,
1605 .evict_inode = ext4_evict_inode,
1606 .put_super = ext4_put_super,
1607 .sync_fs = ext4_sync_fs,
1608 .freeze_fs = ext4_freeze,
1609 .unfreeze_fs = ext4_unfreeze,
1610 .statfs = ext4_statfs,
1611 .show_options = ext4_show_options,
1612#ifdef CONFIG_QUOTA
1613 .quota_read = ext4_quota_read,
1614 .quota_write = ext4_quota_write,
1615 .get_dquots = ext4_get_dquots,
1616#endif
1617};
1618
1619static const struct export_operations ext4_export_ops = {
1620 .fh_to_dentry = ext4_fh_to_dentry,
1621 .fh_to_parent = ext4_fh_to_parent,
1622 .get_parent = ext4_get_parent,
1623 .commit_metadata = ext4_nfs_commit_metadata,
1624};
1625
1626enum {
1627 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1628 Opt_resgid, Opt_resuid, Opt_sb,
1629 Opt_nouid32, Opt_debug, Opt_removed,
1630 Opt_user_xattr, Opt_acl,
1631 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1632 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1633 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1634 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1635 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1636 Opt_inlinecrypt,
1637 Opt_usrjquota, Opt_grpjquota, Opt_quota,
1638 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1639 Opt_usrquota, Opt_grpquota, Opt_prjquota,
1640 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1641 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1642 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1643 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1644 Opt_inode_readahead_blks, Opt_journal_ioprio,
1645 Opt_dioread_nolock, Opt_dioread_lock,
1646 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1647 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1648 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1649 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1650#ifdef CONFIG_EXT4_DEBUG
1651 Opt_fc_debug_max_replay, Opt_fc_debug_force
1652#endif
1653};
1654
1655static const struct constant_table ext4_param_errors[] = {
1656 {"continue", EXT4_MOUNT_ERRORS_CONT},
1657 {"panic", EXT4_MOUNT_ERRORS_PANIC},
1658 {"remount-ro", EXT4_MOUNT_ERRORS_RO},
1659 {}
1660};
1661
1662static const struct constant_table ext4_param_data[] = {
1663 {"journal", EXT4_MOUNT_JOURNAL_DATA},
1664 {"ordered", EXT4_MOUNT_ORDERED_DATA},
1665 {"writeback", EXT4_MOUNT_WRITEBACK_DATA},
1666 {}
1667};
1668
1669static const struct constant_table ext4_param_data_err[] = {
1670 {"abort", Opt_data_err_abort},
1671 {"ignore", Opt_data_err_ignore},
1672 {}
1673};
1674
1675static const struct constant_table ext4_param_jqfmt[] = {
1676 {"vfsold", QFMT_VFS_OLD},
1677 {"vfsv0", QFMT_VFS_V0},
1678 {"vfsv1", QFMT_VFS_V1},
1679 {}
1680};
1681
1682static const struct constant_table ext4_param_dax[] = {
1683 {"always", Opt_dax_always},
1684 {"inode", Opt_dax_inode},
1685 {"never", Opt_dax_never},
1686 {}
1687};
1688
1689/* String parameter that allows empty argument */
1690#define fsparam_string_empty(NAME, OPT) \
1691 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1692
1693/*
1694 * Mount option specification
1695 * We don't use fsparam_flag_no because of the way we set the
1696 * options and the way we show them in _ext4_show_options(). To
1697 * keep the changes to a minimum, let's keep the negative options
1698 * separate for now.
1699 */
1700static const struct fs_parameter_spec ext4_param_specs[] = {
1701 fsparam_flag ("bsddf", Opt_bsd_df),
1702 fsparam_flag ("minixdf", Opt_minix_df),
1703 fsparam_flag ("grpid", Opt_grpid),
1704 fsparam_flag ("bsdgroups", Opt_grpid),
1705 fsparam_flag ("nogrpid", Opt_nogrpid),
1706 fsparam_flag ("sysvgroups", Opt_nogrpid),
1707 fsparam_u32 ("resgid", Opt_resgid),
1708 fsparam_u32 ("resuid", Opt_resuid),
1709 fsparam_u32 ("sb", Opt_sb),
1710 fsparam_enum ("errors", Opt_errors, ext4_param_errors),
1711 fsparam_flag ("nouid32", Opt_nouid32),
1712 fsparam_flag ("debug", Opt_debug),
1713 fsparam_flag ("oldalloc", Opt_removed),
1714 fsparam_flag ("orlov", Opt_removed),
1715 fsparam_flag ("user_xattr", Opt_user_xattr),
1716 fsparam_flag ("acl", Opt_acl),
1717 fsparam_flag ("norecovery", Opt_noload),
1718 fsparam_flag ("noload", Opt_noload),
1719 fsparam_flag ("bh", Opt_removed),
1720 fsparam_flag ("nobh", Opt_removed),
1721 fsparam_u32 ("commit", Opt_commit),
1722 fsparam_u32 ("min_batch_time", Opt_min_batch_time),
1723 fsparam_u32 ("max_batch_time", Opt_max_batch_time),
1724 fsparam_u32 ("journal_dev", Opt_journal_dev),
1725 fsparam_bdev ("journal_path", Opt_journal_path),
1726 fsparam_flag ("journal_checksum", Opt_journal_checksum),
1727 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum),
1728 fsparam_flag ("journal_async_commit",Opt_journal_async_commit),
1729 fsparam_flag ("abort", Opt_abort),
1730 fsparam_enum ("data", Opt_data, ext4_param_data),
1731 fsparam_enum ("data_err", Opt_data_err,
1732 ext4_param_data_err),
1733 fsparam_string_empty
1734 ("usrjquota", Opt_usrjquota),
1735 fsparam_string_empty
1736 ("grpjquota", Opt_grpjquota),
1737 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt),
1738 fsparam_flag ("grpquota", Opt_grpquota),
1739 fsparam_flag ("quota", Opt_quota),
1740 fsparam_flag ("noquota", Opt_noquota),
1741 fsparam_flag ("usrquota", Opt_usrquota),
1742 fsparam_flag ("prjquota", Opt_prjquota),
1743 fsparam_flag ("barrier", Opt_barrier),
1744 fsparam_u32 ("barrier", Opt_barrier),
1745 fsparam_flag ("nobarrier", Opt_nobarrier),
1746 fsparam_flag ("i_version", Opt_removed),
1747 fsparam_flag ("dax", Opt_dax),
1748 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax),
1749 fsparam_u32 ("stripe", Opt_stripe),
1750 fsparam_flag ("delalloc", Opt_delalloc),
1751 fsparam_flag ("nodelalloc", Opt_nodelalloc),
1752 fsparam_flag ("warn_on_error", Opt_warn_on_error),
1753 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error),
1754 fsparam_u32 ("debug_want_extra_isize",
1755 Opt_debug_want_extra_isize),
1756 fsparam_flag ("mblk_io_submit", Opt_removed),
1757 fsparam_flag ("nomblk_io_submit", Opt_removed),
1758 fsparam_flag ("block_validity", Opt_block_validity),
1759 fsparam_flag ("noblock_validity", Opt_noblock_validity),
1760 fsparam_u32 ("inode_readahead_blks",
1761 Opt_inode_readahead_blks),
1762 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio),
1763 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc),
1764 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc),
1765 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc),
1766 fsparam_flag ("dioread_nolock", Opt_dioread_nolock),
1767 fsparam_flag ("nodioread_nolock", Opt_dioread_lock),
1768 fsparam_flag ("dioread_lock", Opt_dioread_lock),
1769 fsparam_flag ("discard", Opt_discard),
1770 fsparam_flag ("nodiscard", Opt_nodiscard),
1771 fsparam_u32 ("init_itable", Opt_init_itable),
1772 fsparam_flag ("init_itable", Opt_init_itable),
1773 fsparam_flag ("noinit_itable", Opt_noinit_itable),
1774#ifdef CONFIG_EXT4_DEBUG
1775 fsparam_flag ("fc_debug_force", Opt_fc_debug_force),
1776 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1777#endif
1778 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb),
1779 fsparam_flag ("test_dummy_encryption",
1780 Opt_test_dummy_encryption),
1781 fsparam_string ("test_dummy_encryption",
1782 Opt_test_dummy_encryption),
1783 fsparam_flag ("inlinecrypt", Opt_inlinecrypt),
1784 fsparam_flag ("nombcache", Opt_nombcache),
1785 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */
1786 fsparam_flag ("prefetch_block_bitmaps",
1787 Opt_removed),
1788 fsparam_flag ("no_prefetch_block_bitmaps",
1789 Opt_no_prefetch_block_bitmaps),
1790 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan),
1791 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */
1792 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */
1793 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */
1794 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */
1795 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */
1796 {}
1797};
1798
1799#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1800
1801#define MOPT_SET 0x0001
1802#define MOPT_CLEAR 0x0002
1803#define MOPT_NOSUPPORT 0x0004
1804#define MOPT_EXPLICIT 0x0008
1805#ifdef CONFIG_QUOTA
1806#define MOPT_Q 0
1807#define MOPT_QFMT 0x0010
1808#else
1809#define MOPT_Q MOPT_NOSUPPORT
1810#define MOPT_QFMT MOPT_NOSUPPORT
1811#endif
1812#define MOPT_NO_EXT2 0x0020
1813#define MOPT_NO_EXT3 0x0040
1814#define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1815#define MOPT_SKIP 0x0080
1816#define MOPT_2 0x0100
1817
1818static const struct mount_opts {
1819 int token;
1820 int mount_opt;
1821 int flags;
1822} ext4_mount_opts[] = {
1823 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1824 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1825 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1826 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1827 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1828 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1829 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1830 MOPT_EXT4_ONLY | MOPT_SET},
1831 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1832 MOPT_EXT4_ONLY | MOPT_CLEAR},
1833 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1834 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1835 {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1836 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1837 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1838 MOPT_EXT4_ONLY | MOPT_CLEAR},
1839 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1840 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1841 {Opt_commit, 0, MOPT_NO_EXT2},
1842 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1843 MOPT_EXT4_ONLY | MOPT_CLEAR},
1844 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1845 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1846 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1847 EXT4_MOUNT_JOURNAL_CHECKSUM),
1848 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1849 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1850 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1851 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1852 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1853 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1854 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1855 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1856 {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1857 {Opt_journal_dev, 0, MOPT_NO_EXT2},
1858 {Opt_journal_path, 0, MOPT_NO_EXT2},
1859 {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1860 {Opt_data, 0, MOPT_NO_EXT2},
1861 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1862#ifdef CONFIG_EXT4_FS_POSIX_ACL
1863 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1864#else
1865 {Opt_acl, 0, MOPT_NOSUPPORT},
1866#endif
1867 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1868 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1869 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1870 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1871 MOPT_SET | MOPT_Q},
1872 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1873 MOPT_SET | MOPT_Q},
1874 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1875 MOPT_SET | MOPT_Q},
1876 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1877 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1878 MOPT_CLEAR | MOPT_Q},
1879 {Opt_usrjquota, 0, MOPT_Q},
1880 {Opt_grpjquota, 0, MOPT_Q},
1881 {Opt_jqfmt, 0, MOPT_QFMT},
1882 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1883 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1884 MOPT_SET},
1885#ifdef CONFIG_EXT4_DEBUG
1886 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1887 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1888#endif
1889 {Opt_err, 0, 0}
1890};
1891
1892#if IS_ENABLED(CONFIG_UNICODE)
1893static const struct ext4_sb_encodings {
1894 __u16 magic;
1895 char *name;
1896 unsigned int version;
1897} ext4_sb_encoding_map[] = {
1898 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1899};
1900
1901static const struct ext4_sb_encodings *
1902ext4_sb_read_encoding(const struct ext4_super_block *es)
1903{
1904 __u16 magic = le16_to_cpu(es->s_encoding);
1905 int i;
1906
1907 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1908 if (magic == ext4_sb_encoding_map[i].magic)
1909 return &ext4_sb_encoding_map[i];
1910
1911 return NULL;
1912}
1913#endif
1914
1915#define EXT4_SPEC_JQUOTA (1 << 0)
1916#define EXT4_SPEC_JQFMT (1 << 1)
1917#define EXT4_SPEC_DATAJ (1 << 2)
1918#define EXT4_SPEC_SB_BLOCK (1 << 3)
1919#define EXT4_SPEC_JOURNAL_DEV (1 << 4)
1920#define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5)
1921#define EXT4_SPEC_s_want_extra_isize (1 << 7)
1922#define EXT4_SPEC_s_max_batch_time (1 << 8)
1923#define EXT4_SPEC_s_min_batch_time (1 << 9)
1924#define EXT4_SPEC_s_inode_readahead_blks (1 << 10)
1925#define EXT4_SPEC_s_li_wait_mult (1 << 11)
1926#define EXT4_SPEC_s_max_dir_size_kb (1 << 12)
1927#define EXT4_SPEC_s_stripe (1 << 13)
1928#define EXT4_SPEC_s_resuid (1 << 14)
1929#define EXT4_SPEC_s_resgid (1 << 15)
1930#define EXT4_SPEC_s_commit_interval (1 << 16)
1931#define EXT4_SPEC_s_fc_debug_max_replay (1 << 17)
1932#define EXT4_SPEC_s_sb_block (1 << 18)
1933#define EXT4_SPEC_mb_optimize_scan (1 << 19)
1934
1935struct ext4_fs_context {
1936 char *s_qf_names[EXT4_MAXQUOTAS];
1937 struct fscrypt_dummy_policy dummy_enc_policy;
1938 int s_jquota_fmt; /* Format of quota to use */
1939#ifdef CONFIG_EXT4_DEBUG
1940 int s_fc_debug_max_replay;
1941#endif
1942 unsigned short qname_spec;
1943 unsigned long vals_s_flags; /* Bits to set in s_flags */
1944 unsigned long mask_s_flags; /* Bits changed in s_flags */
1945 unsigned long journal_devnum;
1946 unsigned long s_commit_interval;
1947 unsigned long s_stripe;
1948 unsigned int s_inode_readahead_blks;
1949 unsigned int s_want_extra_isize;
1950 unsigned int s_li_wait_mult;
1951 unsigned int s_max_dir_size_kb;
1952 unsigned int journal_ioprio;
1953 unsigned int vals_s_mount_opt;
1954 unsigned int mask_s_mount_opt;
1955 unsigned int vals_s_mount_opt2;
1956 unsigned int mask_s_mount_opt2;
1957 unsigned long vals_s_mount_flags;
1958 unsigned long mask_s_mount_flags;
1959 unsigned int opt_flags; /* MOPT flags */
1960 unsigned int spec;
1961 u32 s_max_batch_time;
1962 u32 s_min_batch_time;
1963 kuid_t s_resuid;
1964 kgid_t s_resgid;
1965 ext4_fsblk_t s_sb_block;
1966};
1967
1968static void ext4_fc_free(struct fs_context *fc)
1969{
1970 struct ext4_fs_context *ctx = fc->fs_private;
1971 int i;
1972
1973 if (!ctx)
1974 return;
1975
1976 for (i = 0; i < EXT4_MAXQUOTAS; i++)
1977 kfree(ctx->s_qf_names[i]);
1978
1979 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
1980 kfree(ctx);
1981}
1982
1983int ext4_init_fs_context(struct fs_context *fc)
1984{
1985 struct ext4_fs_context *ctx;
1986
1987 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
1988 if (!ctx)
1989 return -ENOMEM;
1990
1991 fc->fs_private = ctx;
1992 fc->ops = &ext4_context_ops;
1993
1994 return 0;
1995}
1996
1997#ifdef CONFIG_QUOTA
1998/*
1999 * Note the name of the specified quota file.
2000 */
2001static int note_qf_name(struct fs_context *fc, int qtype,
2002 struct fs_parameter *param)
2003{
2004 struct ext4_fs_context *ctx = fc->fs_private;
2005 char *qname;
2006
2007 if (param->size < 1) {
2008 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2009 return -EINVAL;
2010 }
2011 if (strchr(param->string, '/')) {
2012 ext4_msg(NULL, KERN_ERR,
2013 "quotafile must be on filesystem root");
2014 return -EINVAL;
2015 }
2016 if (ctx->s_qf_names[qtype]) {
2017 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2018 ext4_msg(NULL, KERN_ERR,
2019 "%s quota file already specified",
2020 QTYPE2NAME(qtype));
2021 return -EINVAL;
2022 }
2023 return 0;
2024 }
2025
2026 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2027 if (!qname) {
2028 ext4_msg(NULL, KERN_ERR,
2029 "Not enough memory for storing quotafile name");
2030 return -ENOMEM;
2031 }
2032 ctx->s_qf_names[qtype] = qname;
2033 ctx->qname_spec |= 1 << qtype;
2034 ctx->spec |= EXT4_SPEC_JQUOTA;
2035 return 0;
2036}
2037
2038/*
2039 * Clear the name of the specified quota file.
2040 */
2041static int unnote_qf_name(struct fs_context *fc, int qtype)
2042{
2043 struct ext4_fs_context *ctx = fc->fs_private;
2044
2045 if (ctx->s_qf_names[qtype])
2046 kfree(ctx->s_qf_names[qtype]);
2047
2048 ctx->s_qf_names[qtype] = NULL;
2049 ctx->qname_spec |= 1 << qtype;
2050 ctx->spec |= EXT4_SPEC_JQUOTA;
2051 return 0;
2052}
2053#endif
2054
2055static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2056 struct ext4_fs_context *ctx)
2057{
2058 int err;
2059
2060 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2061 ext4_msg(NULL, KERN_WARNING,
2062 "test_dummy_encryption option not supported");
2063 return -EINVAL;
2064 }
2065 err = fscrypt_parse_test_dummy_encryption(param,
2066 &ctx->dummy_enc_policy);
2067 if (err == -EINVAL) {
2068 ext4_msg(NULL, KERN_WARNING,
2069 "Value of option \"%s\" is unrecognized", param->key);
2070 } else if (err == -EEXIST) {
2071 ext4_msg(NULL, KERN_WARNING,
2072 "Conflicting test_dummy_encryption options");
2073 return -EINVAL;
2074 }
2075 return err;
2076}
2077
2078#define EXT4_SET_CTX(name) \
2079static inline void ctx_set_##name(struct ext4_fs_context *ctx, \
2080 unsigned long flag) \
2081{ \
2082 ctx->mask_s_##name |= flag; \
2083 ctx->vals_s_##name |= flag; \
2084}
2085
2086#define EXT4_CLEAR_CTX(name) \
2087static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \
2088 unsigned long flag) \
2089{ \
2090 ctx->mask_s_##name |= flag; \
2091 ctx->vals_s_##name &= ~flag; \
2092}
2093
2094#define EXT4_TEST_CTX(name) \
2095static inline unsigned long \
2096ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \
2097{ \
2098 return (ctx->vals_s_##name & flag); \
2099}
2100
2101EXT4_SET_CTX(flags); /* set only */
2102EXT4_SET_CTX(mount_opt);
2103EXT4_CLEAR_CTX(mount_opt);
2104EXT4_TEST_CTX(mount_opt);
2105EXT4_SET_CTX(mount_opt2);
2106EXT4_CLEAR_CTX(mount_opt2);
2107EXT4_TEST_CTX(mount_opt2);
2108
2109static inline void ctx_set_mount_flag(struct ext4_fs_context *ctx, int bit)
2110{
2111 set_bit(bit, &ctx->mask_s_mount_flags);
2112 set_bit(bit, &ctx->vals_s_mount_flags);
2113}
2114
2115static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2116{
2117 struct ext4_fs_context *ctx = fc->fs_private;
2118 struct fs_parse_result result;
2119 const struct mount_opts *m;
2120 int is_remount;
2121 kuid_t uid;
2122 kgid_t gid;
2123 int token;
2124
2125 token = fs_parse(fc, ext4_param_specs, param, &result);
2126 if (token < 0)
2127 return token;
2128 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2129
2130 for (m = ext4_mount_opts; m->token != Opt_err; m++)
2131 if (token == m->token)
2132 break;
2133
2134 ctx->opt_flags |= m->flags;
2135
2136 if (m->flags & MOPT_EXPLICIT) {
2137 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2138 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2139 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2140 ctx_set_mount_opt2(ctx,
2141 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2142 } else
2143 return -EINVAL;
2144 }
2145
2146 if (m->flags & MOPT_NOSUPPORT) {
2147 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2148 param->key);
2149 return 0;
2150 }
2151
2152 switch (token) {
2153#ifdef CONFIG_QUOTA
2154 case Opt_usrjquota:
2155 if (!*param->string)
2156 return unnote_qf_name(fc, USRQUOTA);
2157 else
2158 return note_qf_name(fc, USRQUOTA, param);
2159 case Opt_grpjquota:
2160 if (!*param->string)
2161 return unnote_qf_name(fc, GRPQUOTA);
2162 else
2163 return note_qf_name(fc, GRPQUOTA, param);
2164#endif
2165 case Opt_sb:
2166 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2167 ext4_msg(NULL, KERN_WARNING,
2168 "Ignoring %s option on remount", param->key);
2169 } else {
2170 ctx->s_sb_block = result.uint_32;
2171 ctx->spec |= EXT4_SPEC_s_sb_block;
2172 }
2173 return 0;
2174 case Opt_removed:
2175 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2176 param->key);
2177 return 0;
2178 case Opt_abort:
2179 ctx_set_mount_flag(ctx, EXT4_MF_FS_ABORTED);
2180 return 0;
2181 case Opt_inlinecrypt:
2182#ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2183 ctx_set_flags(ctx, SB_INLINECRYPT);
2184#else
2185 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2186#endif
2187 return 0;
2188 case Opt_errors:
2189 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2190 ctx_set_mount_opt(ctx, result.uint_32);
2191 return 0;
2192#ifdef CONFIG_QUOTA
2193 case Opt_jqfmt:
2194 ctx->s_jquota_fmt = result.uint_32;
2195 ctx->spec |= EXT4_SPEC_JQFMT;
2196 return 0;
2197#endif
2198 case Opt_data:
2199 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2200 ctx_set_mount_opt(ctx, result.uint_32);
2201 ctx->spec |= EXT4_SPEC_DATAJ;
2202 return 0;
2203 case Opt_commit:
2204 if (result.uint_32 == 0)
2205 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE;
2206 else if (result.uint_32 > INT_MAX / HZ) {
2207 ext4_msg(NULL, KERN_ERR,
2208 "Invalid commit interval %d, "
2209 "must be smaller than %d",
2210 result.uint_32, INT_MAX / HZ);
2211 return -EINVAL;
2212 }
2213 ctx->s_commit_interval = HZ * result.uint_32;
2214 ctx->spec |= EXT4_SPEC_s_commit_interval;
2215 return 0;
2216 case Opt_debug_want_extra_isize:
2217 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2218 ext4_msg(NULL, KERN_ERR,
2219 "Invalid want_extra_isize %d", result.uint_32);
2220 return -EINVAL;
2221 }
2222 ctx->s_want_extra_isize = result.uint_32;
2223 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2224 return 0;
2225 case Opt_max_batch_time:
2226 ctx->s_max_batch_time = result.uint_32;
2227 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2228 return 0;
2229 case Opt_min_batch_time:
2230 ctx->s_min_batch_time = result.uint_32;
2231 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2232 return 0;
2233 case Opt_inode_readahead_blks:
2234 if (result.uint_32 &&
2235 (result.uint_32 > (1 << 30) ||
2236 !is_power_of_2(result.uint_32))) {
2237 ext4_msg(NULL, KERN_ERR,
2238 "EXT4-fs: inode_readahead_blks must be "
2239 "0 or a power of 2 smaller than 2^31");
2240 return -EINVAL;
2241 }
2242 ctx->s_inode_readahead_blks = result.uint_32;
2243 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2244 return 0;
2245 case Opt_init_itable:
2246 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2247 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2248 if (param->type == fs_value_is_string)
2249 ctx->s_li_wait_mult = result.uint_32;
2250 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2251 return 0;
2252 case Opt_max_dir_size_kb:
2253 ctx->s_max_dir_size_kb = result.uint_32;
2254 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2255 return 0;
2256#ifdef CONFIG_EXT4_DEBUG
2257 case Opt_fc_debug_max_replay:
2258 ctx->s_fc_debug_max_replay = result.uint_32;
2259 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2260 return 0;
2261#endif
2262 case Opt_stripe:
2263 ctx->s_stripe = result.uint_32;
2264 ctx->spec |= EXT4_SPEC_s_stripe;
2265 return 0;
2266 case Opt_resuid:
2267 uid = make_kuid(current_user_ns(), result.uint_32);
2268 if (!uid_valid(uid)) {
2269 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2270 result.uint_32);
2271 return -EINVAL;
2272 }
2273 ctx->s_resuid = uid;
2274 ctx->spec |= EXT4_SPEC_s_resuid;
2275 return 0;
2276 case Opt_resgid:
2277 gid = make_kgid(current_user_ns(), result.uint_32);
2278 if (!gid_valid(gid)) {
2279 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2280 result.uint_32);
2281 return -EINVAL;
2282 }
2283 ctx->s_resgid = gid;
2284 ctx->spec |= EXT4_SPEC_s_resgid;
2285 return 0;
2286 case Opt_journal_dev:
2287 if (is_remount) {
2288 ext4_msg(NULL, KERN_ERR,
2289 "Cannot specify journal on remount");
2290 return -EINVAL;
2291 }
2292 ctx->journal_devnum = result.uint_32;
2293 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2294 return 0;
2295 case Opt_journal_path:
2296 {
2297 struct inode *journal_inode;
2298 struct path path;
2299 int error;
2300
2301 if (is_remount) {
2302 ext4_msg(NULL, KERN_ERR,
2303 "Cannot specify journal on remount");
2304 return -EINVAL;
2305 }
2306
2307 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path);
2308 if (error) {
2309 ext4_msg(NULL, KERN_ERR, "error: could not find "
2310 "journal device path");
2311 return -EINVAL;
2312 }
2313
2314 journal_inode = d_inode(path.dentry);
2315 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2316 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2317 path_put(&path);
2318 return 0;
2319 }
2320 case Opt_journal_ioprio:
2321 if (result.uint_32 > 7) {
2322 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2323 " (must be 0-7)");
2324 return -EINVAL;
2325 }
2326 ctx->journal_ioprio =
2327 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2328 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2329 return 0;
2330 case Opt_test_dummy_encryption:
2331 return ext4_parse_test_dummy_encryption(param, ctx);
2332 case Opt_dax:
2333 case Opt_dax_type:
2334#ifdef CONFIG_FS_DAX
2335 {
2336 int type = (token == Opt_dax) ?
2337 Opt_dax : result.uint_32;
2338
2339 switch (type) {
2340 case Opt_dax:
2341 case Opt_dax_always:
2342 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2343 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2344 break;
2345 case Opt_dax_never:
2346 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2347 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2348 break;
2349 case Opt_dax_inode:
2350 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2351 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2352 /* Strictly for printing options */
2353 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2354 break;
2355 }
2356 return 0;
2357 }
2358#else
2359 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2360 return -EINVAL;
2361#endif
2362 case Opt_data_err:
2363 if (result.uint_32 == Opt_data_err_abort)
2364 ctx_set_mount_opt(ctx, m->mount_opt);
2365 else if (result.uint_32 == Opt_data_err_ignore)
2366 ctx_clear_mount_opt(ctx, m->mount_opt);
2367 return 0;
2368 case Opt_mb_optimize_scan:
2369 if (result.int_32 == 1) {
2370 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2371 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2372 } else if (result.int_32 == 0) {
2373 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2374 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2375 } else {
2376 ext4_msg(NULL, KERN_WARNING,
2377 "mb_optimize_scan should be set to 0 or 1.");
2378 return -EINVAL;
2379 }
2380 return 0;
2381 }
2382
2383 /*
2384 * At this point we should only be getting options requiring MOPT_SET,
2385 * or MOPT_CLEAR. Anything else is a bug
2386 */
2387 if (m->token == Opt_err) {
2388 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2389 param->key);
2390 WARN_ON(1);
2391 return -EINVAL;
2392 }
2393
2394 else {
2395 unsigned int set = 0;
2396
2397 if ((param->type == fs_value_is_flag) ||
2398 result.uint_32 > 0)
2399 set = 1;
2400
2401 if (m->flags & MOPT_CLEAR)
2402 set = !set;
2403 else if (unlikely(!(m->flags & MOPT_SET))) {
2404 ext4_msg(NULL, KERN_WARNING,
2405 "buggy handling of option %s",
2406 param->key);
2407 WARN_ON(1);
2408 return -EINVAL;
2409 }
2410 if (m->flags & MOPT_2) {
2411 if (set != 0)
2412 ctx_set_mount_opt2(ctx, m->mount_opt);
2413 else
2414 ctx_clear_mount_opt2(ctx, m->mount_opt);
2415 } else {
2416 if (set != 0)
2417 ctx_set_mount_opt(ctx, m->mount_opt);
2418 else
2419 ctx_clear_mount_opt(ctx, m->mount_opt);
2420 }
2421 }
2422
2423 return 0;
2424}
2425
2426static int parse_options(struct fs_context *fc, char *options)
2427{
2428 struct fs_parameter param;
2429 int ret;
2430 char *key;
2431
2432 if (!options)
2433 return 0;
2434
2435 while ((key = strsep(&options, ",")) != NULL) {
2436 if (*key) {
2437 size_t v_len = 0;
2438 char *value = strchr(key, '=');
2439
2440 param.type = fs_value_is_flag;
2441 param.string = NULL;
2442
2443 if (value) {
2444 if (value == key)
2445 continue;
2446
2447 *value++ = 0;
2448 v_len = strlen(value);
2449 param.string = kmemdup_nul(value, v_len,
2450 GFP_KERNEL);
2451 if (!param.string)
2452 return -ENOMEM;
2453 param.type = fs_value_is_string;
2454 }
2455
2456 param.key = key;
2457 param.size = v_len;
2458
2459 ret = ext4_parse_param(fc, &param);
2460 if (param.string)
2461 kfree(param.string);
2462 if (ret < 0)
2463 return ret;
2464 }
2465 }
2466
2467 ret = ext4_validate_options(fc);
2468 if (ret < 0)
2469 return ret;
2470
2471 return 0;
2472}
2473
2474static int parse_apply_sb_mount_options(struct super_block *sb,
2475 struct ext4_fs_context *m_ctx)
2476{
2477 struct ext4_sb_info *sbi = EXT4_SB(sb);
2478 char *s_mount_opts = NULL;
2479 struct ext4_fs_context *s_ctx = NULL;
2480 struct fs_context *fc = NULL;
2481 int ret = -ENOMEM;
2482
2483 if (!sbi->s_es->s_mount_opts[0])
2484 return 0;
2485
2486 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2487 sizeof(sbi->s_es->s_mount_opts),
2488 GFP_KERNEL);
2489 if (!s_mount_opts)
2490 return ret;
2491
2492 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2493 if (!fc)
2494 goto out_free;
2495
2496 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2497 if (!s_ctx)
2498 goto out_free;
2499
2500 fc->fs_private = s_ctx;
2501 fc->s_fs_info = sbi;
2502
2503 ret = parse_options(fc, s_mount_opts);
2504 if (ret < 0)
2505 goto parse_failed;
2506
2507 ret = ext4_check_opt_consistency(fc, sb);
2508 if (ret < 0) {
2509parse_failed:
2510 ext4_msg(sb, KERN_WARNING,
2511 "failed to parse options in superblock: %s",
2512 s_mount_opts);
2513 ret = 0;
2514 goto out_free;
2515 }
2516
2517 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2518 m_ctx->journal_devnum = s_ctx->journal_devnum;
2519 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2520 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2521
2522 ext4_apply_options(fc, sb);
2523 ret = 0;
2524
2525out_free:
2526 if (fc) {
2527 ext4_fc_free(fc);
2528 kfree(fc);
2529 }
2530 kfree(s_mount_opts);
2531 return ret;
2532}
2533
2534static void ext4_apply_quota_options(struct fs_context *fc,
2535 struct super_block *sb)
2536{
2537#ifdef CONFIG_QUOTA
2538 bool quota_feature = ext4_has_feature_quota(sb);
2539 struct ext4_fs_context *ctx = fc->fs_private;
2540 struct ext4_sb_info *sbi = EXT4_SB(sb);
2541 char *qname;
2542 int i;
2543
2544 if (quota_feature)
2545 return;
2546
2547 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2548 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2549 if (!(ctx->qname_spec & (1 << i)))
2550 continue;
2551
2552 qname = ctx->s_qf_names[i]; /* May be NULL */
2553 if (qname)
2554 set_opt(sb, QUOTA);
2555 ctx->s_qf_names[i] = NULL;
2556 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2557 lockdep_is_held(&sb->s_umount));
2558 if (qname)
2559 kfree_rcu_mightsleep(qname);
2560 }
2561 }
2562
2563 if (ctx->spec & EXT4_SPEC_JQFMT)
2564 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2565#endif
2566}
2567
2568/*
2569 * Check quota settings consistency.
2570 */
2571static int ext4_check_quota_consistency(struct fs_context *fc,
2572 struct super_block *sb)
2573{
2574#ifdef CONFIG_QUOTA
2575 struct ext4_fs_context *ctx = fc->fs_private;
2576 struct ext4_sb_info *sbi = EXT4_SB(sb);
2577 bool quota_feature = ext4_has_feature_quota(sb);
2578 bool quota_loaded = sb_any_quota_loaded(sb);
2579 bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2580 int quota_flags, i;
2581
2582 /*
2583 * We do the test below only for project quotas. 'usrquota' and
2584 * 'grpquota' mount options are allowed even without quota feature
2585 * to support legacy quotas in quota files.
2586 */
2587 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2588 !ext4_has_feature_project(sb)) {
2589 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2590 "Cannot enable project quota enforcement.");
2591 return -EINVAL;
2592 }
2593
2594 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2595 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2596 if (quota_loaded &&
2597 ctx->mask_s_mount_opt & quota_flags &&
2598 !ctx_test_mount_opt(ctx, quota_flags))
2599 goto err_quota_change;
2600
2601 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2602
2603 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2604 if (!(ctx->qname_spec & (1 << i)))
2605 continue;
2606
2607 if (quota_loaded &&
2608 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2609 goto err_jquota_change;
2610
2611 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2612 strcmp(get_qf_name(sb, sbi, i),
2613 ctx->s_qf_names[i]) != 0)
2614 goto err_jquota_specified;
2615 }
2616
2617 if (quota_feature) {
2618 ext4_msg(NULL, KERN_INFO,
2619 "Journaled quota options ignored when "
2620 "QUOTA feature is enabled");
2621 return 0;
2622 }
2623 }
2624
2625 if (ctx->spec & EXT4_SPEC_JQFMT) {
2626 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2627 goto err_jquota_change;
2628 if (quota_feature) {
2629 ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2630 "ignored when QUOTA feature is enabled");
2631 return 0;
2632 }
2633 }
2634
2635 /* Make sure we don't mix old and new quota format */
2636 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2637 ctx->s_qf_names[USRQUOTA]);
2638 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2639 ctx->s_qf_names[GRPQUOTA]);
2640
2641 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2642 test_opt(sb, USRQUOTA));
2643
2644 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2645 test_opt(sb, GRPQUOTA));
2646
2647 if (usr_qf_name) {
2648 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2649 usrquota = false;
2650 }
2651 if (grp_qf_name) {
2652 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2653 grpquota = false;
2654 }
2655
2656 if (usr_qf_name || grp_qf_name) {
2657 if (usrquota || grpquota) {
2658 ext4_msg(NULL, KERN_ERR, "old and new quota "
2659 "format mixing");
2660 return -EINVAL;
2661 }
2662
2663 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2664 ext4_msg(NULL, KERN_ERR, "journaled quota format "
2665 "not specified");
2666 return -EINVAL;
2667 }
2668 }
2669
2670 return 0;
2671
2672err_quota_change:
2673 ext4_msg(NULL, KERN_ERR,
2674 "Cannot change quota options when quota turned on");
2675 return -EINVAL;
2676err_jquota_change:
2677 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2678 "options when quota turned on");
2679 return -EINVAL;
2680err_jquota_specified:
2681 ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2682 QTYPE2NAME(i));
2683 return -EINVAL;
2684#else
2685 return 0;
2686#endif
2687}
2688
2689static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2690 struct super_block *sb)
2691{
2692 const struct ext4_fs_context *ctx = fc->fs_private;
2693 const struct ext4_sb_info *sbi = EXT4_SB(sb);
2694
2695 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2696 return 0;
2697
2698 if (!ext4_has_feature_encrypt(sb)) {
2699 ext4_msg(NULL, KERN_WARNING,
2700 "test_dummy_encryption requires encrypt feature");
2701 return -EINVAL;
2702 }
2703 /*
2704 * This mount option is just for testing, and it's not worthwhile to
2705 * implement the extra complexity (e.g. RCU protection) that would be
2706 * needed to allow it to be set or changed during remount. We do allow
2707 * it to be specified during remount, but only if there is no change.
2708 */
2709 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2710 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2711 &ctx->dummy_enc_policy))
2712 return 0;
2713 ext4_msg(NULL, KERN_WARNING,
2714 "Can't set or change test_dummy_encryption on remount");
2715 return -EINVAL;
2716 }
2717 /* Also make sure s_mount_opts didn't contain a conflicting value. */
2718 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2719 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2720 &ctx->dummy_enc_policy))
2721 return 0;
2722 ext4_msg(NULL, KERN_WARNING,
2723 "Conflicting test_dummy_encryption options");
2724 return -EINVAL;
2725 }
2726 return 0;
2727}
2728
2729static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2730 struct super_block *sb)
2731{
2732 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2733 /* if already set, it was already verified to be the same */
2734 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2735 return;
2736 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2737 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2738 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2739}
2740
2741static int ext4_check_opt_consistency(struct fs_context *fc,
2742 struct super_block *sb)
2743{
2744 struct ext4_fs_context *ctx = fc->fs_private;
2745 struct ext4_sb_info *sbi = fc->s_fs_info;
2746 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2747 int err;
2748
2749 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2750 ext4_msg(NULL, KERN_ERR,
2751 "Mount option(s) incompatible with ext2");
2752 return -EINVAL;
2753 }
2754 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2755 ext4_msg(NULL, KERN_ERR,
2756 "Mount option(s) incompatible with ext3");
2757 return -EINVAL;
2758 }
2759
2760 if (ctx->s_want_extra_isize >
2761 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2762 ext4_msg(NULL, KERN_ERR,
2763 "Invalid want_extra_isize %d",
2764 ctx->s_want_extra_isize);
2765 return -EINVAL;
2766 }
2767
2768 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
2769 int blocksize =
2770 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2771 if (blocksize < PAGE_SIZE)
2772 ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
2773 "experimental mount option 'dioread_nolock' "
2774 "for blocksize < PAGE_SIZE");
2775 }
2776
2777 err = ext4_check_test_dummy_encryption(fc, sb);
2778 if (err)
2779 return err;
2780
2781 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2782 if (!sbi->s_journal) {
2783 ext4_msg(NULL, KERN_WARNING,
2784 "Remounting file system with no journal "
2785 "so ignoring journalled data option");
2786 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2787 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2788 test_opt(sb, DATA_FLAGS)) {
2789 ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2790 "on remount");
2791 return -EINVAL;
2792 }
2793 }
2794
2795 if (is_remount) {
2796 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2797 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2798 ext4_msg(NULL, KERN_ERR, "can't mount with "
2799 "both data=journal and dax");
2800 return -EINVAL;
2801 }
2802
2803 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2804 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2805 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2806fail_dax_change_remount:
2807 ext4_msg(NULL, KERN_ERR, "can't change "
2808 "dax mount option while remounting");
2809 return -EINVAL;
2810 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2811 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2812 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2813 goto fail_dax_change_remount;
2814 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2815 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2816 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2817 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2818 goto fail_dax_change_remount;
2819 }
2820 }
2821
2822 return ext4_check_quota_consistency(fc, sb);
2823}
2824
2825static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2826{
2827 struct ext4_fs_context *ctx = fc->fs_private;
2828 struct ext4_sb_info *sbi = fc->s_fs_info;
2829
2830 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2831 sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2832 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2833 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2834 sbi->s_mount_flags &= ~ctx->mask_s_mount_flags;
2835 sbi->s_mount_flags |= ctx->vals_s_mount_flags;
2836 sb->s_flags &= ~ctx->mask_s_flags;
2837 sb->s_flags |= ctx->vals_s_flags;
2838
2839#define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2840 APPLY(s_commit_interval);
2841 APPLY(s_stripe);
2842 APPLY(s_max_batch_time);
2843 APPLY(s_min_batch_time);
2844 APPLY(s_want_extra_isize);
2845 APPLY(s_inode_readahead_blks);
2846 APPLY(s_max_dir_size_kb);
2847 APPLY(s_li_wait_mult);
2848 APPLY(s_resgid);
2849 APPLY(s_resuid);
2850
2851#ifdef CONFIG_EXT4_DEBUG
2852 APPLY(s_fc_debug_max_replay);
2853#endif
2854
2855 ext4_apply_quota_options(fc, sb);
2856 ext4_apply_test_dummy_encryption(ctx, sb);
2857}
2858
2859
2860static int ext4_validate_options(struct fs_context *fc)
2861{
2862#ifdef CONFIG_QUOTA
2863 struct ext4_fs_context *ctx = fc->fs_private;
2864 char *usr_qf_name, *grp_qf_name;
2865
2866 usr_qf_name = ctx->s_qf_names[USRQUOTA];
2867 grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2868
2869 if (usr_qf_name || grp_qf_name) {
2870 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2871 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2872
2873 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2874 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2875
2876 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2877 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2878 ext4_msg(NULL, KERN_ERR, "old and new quota "
2879 "format mixing");
2880 return -EINVAL;
2881 }
2882 }
2883#endif
2884 return 1;
2885}
2886
2887static inline void ext4_show_quota_options(struct seq_file *seq,
2888 struct super_block *sb)
2889{
2890#if defined(CONFIG_QUOTA)
2891 struct ext4_sb_info *sbi = EXT4_SB(sb);
2892 char *usr_qf_name, *grp_qf_name;
2893
2894 if (sbi->s_jquota_fmt) {
2895 char *fmtname = "";
2896
2897 switch (sbi->s_jquota_fmt) {
2898 case QFMT_VFS_OLD:
2899 fmtname = "vfsold";
2900 break;
2901 case QFMT_VFS_V0:
2902 fmtname = "vfsv0";
2903 break;
2904 case QFMT_VFS_V1:
2905 fmtname = "vfsv1";
2906 break;
2907 }
2908 seq_printf(seq, ",jqfmt=%s", fmtname);
2909 }
2910
2911 rcu_read_lock();
2912 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2913 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2914 if (usr_qf_name)
2915 seq_show_option(seq, "usrjquota", usr_qf_name);
2916 if (grp_qf_name)
2917 seq_show_option(seq, "grpjquota", grp_qf_name);
2918 rcu_read_unlock();
2919#endif
2920}
2921
2922static const char *token2str(int token)
2923{
2924 const struct fs_parameter_spec *spec;
2925
2926 for (spec = ext4_param_specs; spec->name != NULL; spec++)
2927 if (spec->opt == token && !spec->type)
2928 break;
2929 return spec->name;
2930}
2931
2932/*
2933 * Show an option if
2934 * - it's set to a non-default value OR
2935 * - if the per-sb default is different from the global default
2936 */
2937static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2938 int nodefs)
2939{
2940 struct ext4_sb_info *sbi = EXT4_SB(sb);
2941 struct ext4_super_block *es = sbi->s_es;
2942 int def_errors;
2943 const struct mount_opts *m;
2944 char sep = nodefs ? '\n' : ',';
2945
2946#define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2947#define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2948
2949 if (sbi->s_sb_block != 1)
2950 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2951
2952 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2953 int want_set = m->flags & MOPT_SET;
2954 int opt_2 = m->flags & MOPT_2;
2955 unsigned int mount_opt, def_mount_opt;
2956
2957 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2958 m->flags & MOPT_SKIP)
2959 continue;
2960
2961 if (opt_2) {
2962 mount_opt = sbi->s_mount_opt2;
2963 def_mount_opt = sbi->s_def_mount_opt2;
2964 } else {
2965 mount_opt = sbi->s_mount_opt;
2966 def_mount_opt = sbi->s_def_mount_opt;
2967 }
2968 /* skip if same as the default */
2969 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt)))
2970 continue;
2971 /* select Opt_noFoo vs Opt_Foo */
2972 if ((want_set &&
2973 (mount_opt & m->mount_opt) != m->mount_opt) ||
2974 (!want_set && (mount_opt & m->mount_opt)))
2975 continue;
2976 SEQ_OPTS_PRINT("%s", token2str(m->token));
2977 }
2978
2979 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2980 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2981 SEQ_OPTS_PRINT("resuid=%u",
2982 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2983 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2984 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2985 SEQ_OPTS_PRINT("resgid=%u",
2986 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2987 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2988 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2989 SEQ_OPTS_PUTS("errors=remount-ro");
2990 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2991 SEQ_OPTS_PUTS("errors=continue");
2992 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2993 SEQ_OPTS_PUTS("errors=panic");
2994 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2995 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2996 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2997 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2998 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2999 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
3000 if (nodefs || sbi->s_stripe)
3001 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
3002 if (nodefs || EXT4_MOUNT_DATA_FLAGS &
3003 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3004 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3005 SEQ_OPTS_PUTS("data=journal");
3006 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3007 SEQ_OPTS_PUTS("data=ordered");
3008 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3009 SEQ_OPTS_PUTS("data=writeback");
3010 }
3011 if (nodefs ||
3012 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3013 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3014 sbi->s_inode_readahead_blks);
3015
3016 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3017 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3018 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3019 if (nodefs || sbi->s_max_dir_size_kb)
3020 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3021 if (test_opt(sb, DATA_ERR_ABORT))
3022 SEQ_OPTS_PUTS("data_err=abort");
3023
3024 fscrypt_show_test_dummy_encryption(seq, sep, sb);
3025
3026 if (sb->s_flags & SB_INLINECRYPT)
3027 SEQ_OPTS_PUTS("inlinecrypt");
3028
3029 if (test_opt(sb, DAX_ALWAYS)) {
3030 if (IS_EXT2_SB(sb))
3031 SEQ_OPTS_PUTS("dax");
3032 else
3033 SEQ_OPTS_PUTS("dax=always");
3034 } else if (test_opt2(sb, DAX_NEVER)) {
3035 SEQ_OPTS_PUTS("dax=never");
3036 } else if (test_opt2(sb, DAX_INODE)) {
3037 SEQ_OPTS_PUTS("dax=inode");
3038 }
3039
3040 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3041 !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3042 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3043 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3044 test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3045 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3046 }
3047
3048 ext4_show_quota_options(seq, sb);
3049 return 0;
3050}
3051
3052static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3053{
3054 return _ext4_show_options(seq, root->d_sb, 0);
3055}
3056
3057int ext4_seq_options_show(struct seq_file *seq, void *offset)
3058{
3059 struct super_block *sb = seq->private;
3060 int rc;
3061
3062 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3063 rc = _ext4_show_options(seq, sb, 1);
3064 seq_puts(seq, "\n");
3065 return rc;
3066}
3067
3068static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3069 int read_only)
3070{
3071 struct ext4_sb_info *sbi = EXT4_SB(sb);
3072 int err = 0;
3073
3074 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3075 ext4_msg(sb, KERN_ERR, "revision level too high, "
3076 "forcing read-only mode");
3077 err = -EROFS;
3078 goto done;
3079 }
3080 if (read_only)
3081 goto done;
3082 if (!(sbi->s_mount_state & EXT4_VALID_FS))
3083 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3084 "running e2fsck is recommended");
3085 else if (sbi->s_mount_state & EXT4_ERROR_FS)
3086 ext4_msg(sb, KERN_WARNING,
3087 "warning: mounting fs with errors, "
3088 "running e2fsck is recommended");
3089 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3090 le16_to_cpu(es->s_mnt_count) >=
3091 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3092 ext4_msg(sb, KERN_WARNING,
3093 "warning: maximal mount count reached, "
3094 "running e2fsck is recommended");
3095 else if (le32_to_cpu(es->s_checkinterval) &&
3096 (ext4_get_tstamp(es, s_lastcheck) +
3097 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3098 ext4_msg(sb, KERN_WARNING,
3099 "warning: checktime reached, "
3100 "running e2fsck is recommended");
3101 if (!sbi->s_journal)
3102 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3103 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3104 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3105 le16_add_cpu(&es->s_mnt_count, 1);
3106 ext4_update_tstamp(es, s_mtime);
3107 if (sbi->s_journal) {
3108 ext4_set_feature_journal_needs_recovery(sb);
3109 if (ext4_has_feature_orphan_file(sb))
3110 ext4_set_feature_orphan_present(sb);
3111 }
3112
3113 err = ext4_commit_super(sb);
3114done:
3115 if (test_opt(sb, DEBUG))
3116 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3117 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3118 sb->s_blocksize,
3119 sbi->s_groups_count,
3120 EXT4_BLOCKS_PER_GROUP(sb),
3121 EXT4_INODES_PER_GROUP(sb),
3122 sbi->s_mount_opt, sbi->s_mount_opt2);
3123 return err;
3124}
3125
3126int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3127{
3128 struct ext4_sb_info *sbi = EXT4_SB(sb);
3129 struct flex_groups **old_groups, **new_groups;
3130 int size, i, j;
3131
3132 if (!sbi->s_log_groups_per_flex)
3133 return 0;
3134
3135 size = ext4_flex_group(sbi, ngroup - 1) + 1;
3136 if (size <= sbi->s_flex_groups_allocated)
3137 return 0;
3138
3139 new_groups = kvzalloc(roundup_pow_of_two(size *
3140 sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3141 if (!new_groups) {
3142 ext4_msg(sb, KERN_ERR,
3143 "not enough memory for %d flex group pointers", size);
3144 return -ENOMEM;
3145 }
3146 for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3147 new_groups[i] = kvzalloc(roundup_pow_of_two(
3148 sizeof(struct flex_groups)),
3149 GFP_KERNEL);
3150 if (!new_groups[i]) {
3151 for (j = sbi->s_flex_groups_allocated; j < i; j++)
3152 kvfree(new_groups[j]);
3153 kvfree(new_groups);
3154 ext4_msg(sb, KERN_ERR,
3155 "not enough memory for %d flex groups", size);
3156 return -ENOMEM;
3157 }
3158 }
3159 rcu_read_lock();
3160 old_groups = rcu_dereference(sbi->s_flex_groups);
3161 if (old_groups)
3162 memcpy(new_groups, old_groups,
3163 (sbi->s_flex_groups_allocated *
3164 sizeof(struct flex_groups *)));
3165 rcu_read_unlock();
3166 rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3167 sbi->s_flex_groups_allocated = size;
3168 if (old_groups)
3169 ext4_kvfree_array_rcu(old_groups);
3170 return 0;
3171}
3172
3173static int ext4_fill_flex_info(struct super_block *sb)
3174{
3175 struct ext4_sb_info *sbi = EXT4_SB(sb);
3176 struct ext4_group_desc *gdp = NULL;
3177 struct flex_groups *fg;
3178 ext4_group_t flex_group;
3179 int i, err;
3180
3181 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3182 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3183 sbi->s_log_groups_per_flex = 0;
3184 return 1;
3185 }
3186
3187 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3188 if (err)
3189 goto failed;
3190
3191 for (i = 0; i < sbi->s_groups_count; i++) {
3192 gdp = ext4_get_group_desc(sb, i, NULL);
3193
3194 flex_group = ext4_flex_group(sbi, i);
3195 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3196 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3197 atomic64_add(ext4_free_group_clusters(sb, gdp),
3198 &fg->free_clusters);
3199 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3200 }
3201
3202 return 1;
3203failed:
3204 return 0;
3205}
3206
3207static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3208 struct ext4_group_desc *gdp)
3209{
3210 int offset = offsetof(struct ext4_group_desc, bg_checksum);
3211 __u16 crc = 0;
3212 __le32 le_group = cpu_to_le32(block_group);
3213 struct ext4_sb_info *sbi = EXT4_SB(sb);
3214
3215 if (ext4_has_metadata_csum(sbi->s_sb)) {
3216 /* Use new metadata_csum algorithm */
3217 __u32 csum32;
3218 __u16 dummy_csum = 0;
3219
3220 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3221 sizeof(le_group));
3222 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3223 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3224 sizeof(dummy_csum));
3225 offset += sizeof(dummy_csum);
3226 if (offset < sbi->s_desc_size)
3227 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3228 sbi->s_desc_size - offset);
3229
3230 crc = csum32 & 0xFFFF;
3231 goto out;
3232 }
3233
3234 /* old crc16 code */
3235 if (!ext4_has_feature_gdt_csum(sb))
3236 return 0;
3237
3238 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3239 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3240 crc = crc16(crc, (__u8 *)gdp, offset);
3241 offset += sizeof(gdp->bg_checksum); /* skip checksum */
3242 /* for checksum of struct ext4_group_desc do the rest...*/
3243 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size)
3244 crc = crc16(crc, (__u8 *)gdp + offset,
3245 sbi->s_desc_size - offset);
3246
3247out:
3248 return cpu_to_le16(crc);
3249}
3250
3251int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3252 struct ext4_group_desc *gdp)
3253{
3254 if (ext4_has_group_desc_csum(sb) &&
3255 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3256 return 0;
3257
3258 return 1;
3259}
3260
3261void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3262 struct ext4_group_desc *gdp)
3263{
3264 if (!ext4_has_group_desc_csum(sb))
3265 return;
3266 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3267}
3268
3269/* Called at mount-time, super-block is locked */
3270static int ext4_check_descriptors(struct super_block *sb,
3271 ext4_fsblk_t sb_block,
3272 ext4_group_t *first_not_zeroed)
3273{
3274 struct ext4_sb_info *sbi = EXT4_SB(sb);
3275 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3276 ext4_fsblk_t last_block;
3277 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3278 ext4_fsblk_t block_bitmap;
3279 ext4_fsblk_t inode_bitmap;
3280 ext4_fsblk_t inode_table;
3281 int flexbg_flag = 0;
3282 ext4_group_t i, grp = sbi->s_groups_count;
3283
3284 if (ext4_has_feature_flex_bg(sb))
3285 flexbg_flag = 1;
3286
3287 ext4_debug("Checking group descriptors");
3288
3289 for (i = 0; i < sbi->s_groups_count; i++) {
3290 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3291
3292 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3293 last_block = ext4_blocks_count(sbi->s_es) - 1;
3294 else
3295 last_block = first_block +
3296 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3297
3298 if ((grp == sbi->s_groups_count) &&
3299 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3300 grp = i;
3301
3302 block_bitmap = ext4_block_bitmap(sb, gdp);
3303 if (block_bitmap == sb_block) {
3304 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3305 "Block bitmap for group %u overlaps "
3306 "superblock", i);
3307 if (!sb_rdonly(sb))
3308 return 0;
3309 }
3310 if (block_bitmap >= sb_block + 1 &&
3311 block_bitmap <= last_bg_block) {
3312 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3313 "Block bitmap for group %u overlaps "
3314 "block group descriptors", i);
3315 if (!sb_rdonly(sb))
3316 return 0;
3317 }
3318 if (block_bitmap < first_block || block_bitmap > last_block) {
3319 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3320 "Block bitmap for group %u not in group "
3321 "(block %llu)!", i, block_bitmap);
3322 return 0;
3323 }
3324 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3325 if (inode_bitmap == sb_block) {
3326 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3327 "Inode bitmap for group %u overlaps "
3328 "superblock", i);
3329 if (!sb_rdonly(sb))
3330 return 0;
3331 }
3332 if (inode_bitmap >= sb_block + 1 &&
3333 inode_bitmap <= last_bg_block) {
3334 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3335 "Inode bitmap for group %u overlaps "
3336 "block group descriptors", i);
3337 if (!sb_rdonly(sb))
3338 return 0;
3339 }
3340 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3341 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3342 "Inode bitmap for group %u not in group "
3343 "(block %llu)!", i, inode_bitmap);
3344 return 0;
3345 }
3346 inode_table = ext4_inode_table(sb, gdp);
3347 if (inode_table == sb_block) {
3348 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3349 "Inode table for group %u overlaps "
3350 "superblock", i);
3351 if (!sb_rdonly(sb))
3352 return 0;
3353 }
3354 if (inode_table >= sb_block + 1 &&
3355 inode_table <= last_bg_block) {
3356 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3357 "Inode table for group %u overlaps "
3358 "block group descriptors", i);
3359 if (!sb_rdonly(sb))
3360 return 0;
3361 }
3362 if (inode_table < first_block ||
3363 inode_table + sbi->s_itb_per_group - 1 > last_block) {
3364 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3365 "Inode table for group %u not in group "
3366 "(block %llu)!", i, inode_table);
3367 return 0;
3368 }
3369 ext4_lock_group(sb, i);
3370 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3371 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3372 "Checksum for group %u failed (%u!=%u)",
3373 i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3374 gdp)), le16_to_cpu(gdp->bg_checksum));
3375 if (!sb_rdonly(sb)) {
3376 ext4_unlock_group(sb, i);
3377 return 0;
3378 }
3379 }
3380 ext4_unlock_group(sb, i);
3381 if (!flexbg_flag)
3382 first_block += EXT4_BLOCKS_PER_GROUP(sb);
3383 }
3384 if (NULL != first_not_zeroed)
3385 *first_not_zeroed = grp;
3386 return 1;
3387}
3388
3389/*
3390 * Maximal extent format file size.
3391 * Resulting logical blkno at s_maxbytes must fit in our on-disk
3392 * extent format containers, within a sector_t, and within i_blocks
3393 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
3394 * so that won't be a limiting factor.
3395 *
3396 * However there is other limiting factor. We do store extents in the form
3397 * of starting block and length, hence the resulting length of the extent
3398 * covering maximum file size must fit into on-disk format containers as
3399 * well. Given that length is always by 1 unit bigger than max unit (because
3400 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3401 *
3402 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3403 */
3404static loff_t ext4_max_size(int blkbits, int has_huge_files)
3405{
3406 loff_t res;
3407 loff_t upper_limit = MAX_LFS_FILESIZE;
3408
3409 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3410
3411 if (!has_huge_files) {
3412 upper_limit = (1LL << 32) - 1;
3413
3414 /* total blocks in file system block size */
3415 upper_limit >>= (blkbits - 9);
3416 upper_limit <<= blkbits;
3417 }
3418
3419 /*
3420 * 32-bit extent-start container, ee_block. We lower the maxbytes
3421 * by one fs block, so ee_len can cover the extent of maximum file
3422 * size
3423 */
3424 res = (1LL << 32) - 1;
3425 res <<= blkbits;
3426
3427 /* Sanity check against vm- & vfs- imposed limits */
3428 if (res > upper_limit)
3429 res = upper_limit;
3430
3431 return res;
3432}
3433
3434/*
3435 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
3436 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3437 * We need to be 1 filesystem block less than the 2^48 sector limit.
3438 */
3439static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3440{
3441 loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3442 int meta_blocks;
3443 unsigned int ppb = 1 << (bits - 2);
3444
3445 /*
3446 * This is calculated to be the largest file size for a dense, block
3447 * mapped file such that the file's total number of 512-byte sectors,
3448 * including data and all indirect blocks, does not exceed (2^48 - 1).
3449 *
3450 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3451 * number of 512-byte sectors of the file.
3452 */
3453 if (!has_huge_files) {
3454 /*
3455 * !has_huge_files or implies that the inode i_block field
3456 * represents total file blocks in 2^32 512-byte sectors ==
3457 * size of vfs inode i_blocks * 8
3458 */
3459 upper_limit = (1LL << 32) - 1;
3460
3461 /* total blocks in file system block size */
3462 upper_limit >>= (bits - 9);
3463
3464 } else {
3465 /*
3466 * We use 48 bit ext4_inode i_blocks
3467 * With EXT4_HUGE_FILE_FL set the i_blocks
3468 * represent total number of blocks in
3469 * file system block size
3470 */
3471 upper_limit = (1LL << 48) - 1;
3472
3473 }
3474
3475 /* Compute how many blocks we can address by block tree */
3476 res += ppb;
3477 res += ppb * ppb;
3478 res += ((loff_t)ppb) * ppb * ppb;
3479 /* Compute how many metadata blocks are needed */
3480 meta_blocks = 1;
3481 meta_blocks += 1 + ppb;
3482 meta_blocks += 1 + ppb + ppb * ppb;
3483 /* Does block tree limit file size? */
3484 if (res + meta_blocks <= upper_limit)
3485 goto check_lfs;
3486
3487 res = upper_limit;
3488 /* How many metadata blocks are needed for addressing upper_limit? */
3489 upper_limit -= EXT4_NDIR_BLOCKS;
3490 /* indirect blocks */
3491 meta_blocks = 1;
3492 upper_limit -= ppb;
3493 /* double indirect blocks */
3494 if (upper_limit < ppb * ppb) {
3495 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3496 res -= meta_blocks;
3497 goto check_lfs;
3498 }
3499 meta_blocks += 1 + ppb;
3500 upper_limit -= ppb * ppb;
3501 /* tripple indirect blocks for the rest */
3502 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3503 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3504 res -= meta_blocks;
3505check_lfs:
3506 res <<= bits;
3507 if (res > MAX_LFS_FILESIZE)
3508 res = MAX_LFS_FILESIZE;
3509
3510 return res;
3511}
3512
3513static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3514 ext4_fsblk_t logical_sb_block, int nr)
3515{
3516 struct ext4_sb_info *sbi = EXT4_SB(sb);
3517 ext4_group_t bg, first_meta_bg;
3518 int has_super = 0;
3519
3520 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3521
3522 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3523 return logical_sb_block + nr + 1;
3524 bg = sbi->s_desc_per_block * nr;
3525 if (ext4_bg_has_super(sb, bg))
3526 has_super = 1;
3527
3528 /*
3529 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3530 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled
3531 * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3532 * compensate.
3533 */
3534 if (sb->s_blocksize == 1024 && nr == 0 &&
3535 le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3536 has_super++;
3537
3538 return (has_super + ext4_group_first_block_no(sb, bg));
3539}
3540
3541/**
3542 * ext4_get_stripe_size: Get the stripe size.
3543 * @sbi: In memory super block info
3544 *
3545 * If we have specified it via mount option, then
3546 * use the mount option value. If the value specified at mount time is
3547 * greater than the blocks per group use the super block value.
3548 * If the super block value is greater than blocks per group return 0.
3549 * Allocator needs it be less than blocks per group.
3550 *
3551 */
3552static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3553{
3554 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3555 unsigned long stripe_width =
3556 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3557 int ret;
3558
3559 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3560 ret = sbi->s_stripe;
3561 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3562 ret = stripe_width;
3563 else if (stride && stride <= sbi->s_blocks_per_group)
3564 ret = stride;
3565 else
3566 ret = 0;
3567
3568 /*
3569 * If the stripe width is 1, this makes no sense and
3570 * we set it to 0 to turn off stripe handling code.
3571 */
3572 if (ret <= 1)
3573 ret = 0;
3574
3575 return ret;
3576}
3577
3578/*
3579 * Check whether this filesystem can be mounted based on
3580 * the features present and the RDONLY/RDWR mount requested.
3581 * Returns 1 if this filesystem can be mounted as requested,
3582 * 0 if it cannot be.
3583 */
3584int ext4_feature_set_ok(struct super_block *sb, int readonly)
3585{
3586 if (ext4_has_unknown_ext4_incompat_features(sb)) {
3587 ext4_msg(sb, KERN_ERR,
3588 "Couldn't mount because of "
3589 "unsupported optional features (%x)",
3590 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3591 ~EXT4_FEATURE_INCOMPAT_SUPP));
3592 return 0;
3593 }
3594
3595#if !IS_ENABLED(CONFIG_UNICODE)
3596 if (ext4_has_feature_casefold(sb)) {
3597 ext4_msg(sb, KERN_ERR,
3598 "Filesystem with casefold feature cannot be "
3599 "mounted without CONFIG_UNICODE");
3600 return 0;
3601 }
3602#endif
3603
3604 if (readonly)
3605 return 1;
3606
3607 if (ext4_has_feature_readonly(sb)) {
3608 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3609 sb->s_flags |= SB_RDONLY;
3610 return 1;
3611 }
3612
3613 /* Check that feature set is OK for a read-write mount */
3614 if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3615 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3616 "unsupported optional features (%x)",
3617 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3618 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3619 return 0;
3620 }
3621 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3622 ext4_msg(sb, KERN_ERR,
3623 "Can't support bigalloc feature without "
3624 "extents feature\n");
3625 return 0;
3626 }
3627
3628#if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3629 if (!readonly && (ext4_has_feature_quota(sb) ||
3630 ext4_has_feature_project(sb))) {
3631 ext4_msg(sb, KERN_ERR,
3632 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3633 return 0;
3634 }
3635#endif /* CONFIG_QUOTA */
3636 return 1;
3637}
3638
3639/*
3640 * This function is called once a day if we have errors logged
3641 * on the file system
3642 */
3643static void print_daily_error_info(struct timer_list *t)
3644{
3645 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3646 struct super_block *sb = sbi->s_sb;
3647 struct ext4_super_block *es = sbi->s_es;
3648
3649 if (es->s_error_count)
3650 /* fsck newer than v1.41.13 is needed to clean this condition. */
3651 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3652 le32_to_cpu(es->s_error_count));
3653 if (es->s_first_error_time) {
3654 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3655 sb->s_id,
3656 ext4_get_tstamp(es, s_first_error_time),
3657 (int) sizeof(es->s_first_error_func),
3658 es->s_first_error_func,
3659 le32_to_cpu(es->s_first_error_line));
3660 if (es->s_first_error_ino)
3661 printk(KERN_CONT ": inode %u",
3662 le32_to_cpu(es->s_first_error_ino));
3663 if (es->s_first_error_block)
3664 printk(KERN_CONT ": block %llu", (unsigned long long)
3665 le64_to_cpu(es->s_first_error_block));
3666 printk(KERN_CONT "\n");
3667 }
3668 if (es->s_last_error_time) {
3669 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3670 sb->s_id,
3671 ext4_get_tstamp(es, s_last_error_time),
3672 (int) sizeof(es->s_last_error_func),
3673 es->s_last_error_func,
3674 le32_to_cpu(es->s_last_error_line));
3675 if (es->s_last_error_ino)
3676 printk(KERN_CONT ": inode %u",
3677 le32_to_cpu(es->s_last_error_ino));
3678 if (es->s_last_error_block)
3679 printk(KERN_CONT ": block %llu", (unsigned long long)
3680 le64_to_cpu(es->s_last_error_block));
3681 printk(KERN_CONT "\n");
3682 }
3683 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
3684}
3685
3686/* Find next suitable group and run ext4_init_inode_table */
3687static int ext4_run_li_request(struct ext4_li_request *elr)
3688{
3689 struct ext4_group_desc *gdp = NULL;
3690 struct super_block *sb = elr->lr_super;
3691 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3692 ext4_group_t group = elr->lr_next_group;
3693 unsigned int prefetch_ios = 0;
3694 int ret = 0;
3695 int nr = EXT4_SB(sb)->s_mb_prefetch;
3696 u64 start_time;
3697
3698 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3699 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios);
3700 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr);
3701 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr);
3702 if (group >= elr->lr_next_group) {
3703 ret = 1;
3704 if (elr->lr_first_not_zeroed != ngroups &&
3705 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3706 elr->lr_next_group = elr->lr_first_not_zeroed;
3707 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3708 ret = 0;
3709 }
3710 }
3711 return ret;
3712 }
3713
3714 for (; group < ngroups; group++) {
3715 gdp = ext4_get_group_desc(sb, group, NULL);
3716 if (!gdp) {
3717 ret = 1;
3718 break;
3719 }
3720
3721 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3722 break;
3723 }
3724
3725 if (group >= ngroups)
3726 ret = 1;
3727
3728 if (!ret) {
3729 start_time = ktime_get_real_ns();
3730 ret = ext4_init_inode_table(sb, group,
3731 elr->lr_timeout ? 0 : 1);
3732 trace_ext4_lazy_itable_init(sb, group);
3733 if (elr->lr_timeout == 0) {
3734 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3735 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3736 }
3737 elr->lr_next_sched = jiffies + elr->lr_timeout;
3738 elr->lr_next_group = group + 1;
3739 }
3740 return ret;
3741}
3742
3743/*
3744 * Remove lr_request from the list_request and free the
3745 * request structure. Should be called with li_list_mtx held
3746 */
3747static void ext4_remove_li_request(struct ext4_li_request *elr)
3748{
3749 if (!elr)
3750 return;
3751
3752 list_del(&elr->lr_request);
3753 EXT4_SB(elr->lr_super)->s_li_request = NULL;
3754 kfree(elr);
3755}
3756
3757static void ext4_unregister_li_request(struct super_block *sb)
3758{
3759 mutex_lock(&ext4_li_mtx);
3760 if (!ext4_li_info) {
3761 mutex_unlock(&ext4_li_mtx);
3762 return;
3763 }
3764
3765 mutex_lock(&ext4_li_info->li_list_mtx);
3766 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3767 mutex_unlock(&ext4_li_info->li_list_mtx);
3768 mutex_unlock(&ext4_li_mtx);
3769}
3770
3771static struct task_struct *ext4_lazyinit_task;
3772
3773/*
3774 * This is the function where ext4lazyinit thread lives. It walks
3775 * through the request list searching for next scheduled filesystem.
3776 * When such a fs is found, run the lazy initialization request
3777 * (ext4_rn_li_request) and keep track of the time spend in this
3778 * function. Based on that time we compute next schedule time of
3779 * the request. When walking through the list is complete, compute
3780 * next waking time and put itself into sleep.
3781 */
3782static int ext4_lazyinit_thread(void *arg)
3783{
3784 struct ext4_lazy_init *eli = arg;
3785 struct list_head *pos, *n;
3786 struct ext4_li_request *elr;
3787 unsigned long next_wakeup, cur;
3788
3789 BUG_ON(NULL == eli);
3790 set_freezable();
3791
3792cont_thread:
3793 while (true) {
3794 next_wakeup = MAX_JIFFY_OFFSET;
3795
3796 mutex_lock(&eli->li_list_mtx);
3797 if (list_empty(&eli->li_request_list)) {
3798 mutex_unlock(&eli->li_list_mtx);
3799 goto exit_thread;
3800 }
3801 list_for_each_safe(pos, n, &eli->li_request_list) {
3802 int err = 0;
3803 int progress = 0;
3804 elr = list_entry(pos, struct ext4_li_request,
3805 lr_request);
3806
3807 if (time_before(jiffies, elr->lr_next_sched)) {
3808 if (time_before(elr->lr_next_sched, next_wakeup))
3809 next_wakeup = elr->lr_next_sched;
3810 continue;
3811 }
3812 if (down_read_trylock(&elr->lr_super->s_umount)) {
3813 if (sb_start_write_trylock(elr->lr_super)) {
3814 progress = 1;
3815 /*
3816 * We hold sb->s_umount, sb can not
3817 * be removed from the list, it is
3818 * now safe to drop li_list_mtx
3819 */
3820 mutex_unlock(&eli->li_list_mtx);
3821 err = ext4_run_li_request(elr);
3822 sb_end_write(elr->lr_super);
3823 mutex_lock(&eli->li_list_mtx);
3824 n = pos->next;
3825 }
3826 up_read((&elr->lr_super->s_umount));
3827 }
3828 /* error, remove the lazy_init job */
3829 if (err) {
3830 ext4_remove_li_request(elr);
3831 continue;
3832 }
3833 if (!progress) {
3834 elr->lr_next_sched = jiffies +
3835 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3836 }
3837 if (time_before(elr->lr_next_sched, next_wakeup))
3838 next_wakeup = elr->lr_next_sched;
3839 }
3840 mutex_unlock(&eli->li_list_mtx);
3841
3842 try_to_freeze();
3843
3844 cur = jiffies;
3845 if ((time_after_eq(cur, next_wakeup)) ||
3846 (MAX_JIFFY_OFFSET == next_wakeup)) {
3847 cond_resched();
3848 continue;
3849 }
3850
3851 schedule_timeout_interruptible(next_wakeup - cur);
3852
3853 if (kthread_should_stop()) {
3854 ext4_clear_request_list();
3855 goto exit_thread;
3856 }
3857 }
3858
3859exit_thread:
3860 /*
3861 * It looks like the request list is empty, but we need
3862 * to check it under the li_list_mtx lock, to prevent any
3863 * additions into it, and of course we should lock ext4_li_mtx
3864 * to atomically free the list and ext4_li_info, because at
3865 * this point another ext4 filesystem could be registering
3866 * new one.
3867 */
3868 mutex_lock(&ext4_li_mtx);
3869 mutex_lock(&eli->li_list_mtx);
3870 if (!list_empty(&eli->li_request_list)) {
3871 mutex_unlock(&eli->li_list_mtx);
3872 mutex_unlock(&ext4_li_mtx);
3873 goto cont_thread;
3874 }
3875 mutex_unlock(&eli->li_list_mtx);
3876 kfree(ext4_li_info);
3877 ext4_li_info = NULL;
3878 mutex_unlock(&ext4_li_mtx);
3879
3880 return 0;
3881}
3882
3883static void ext4_clear_request_list(void)
3884{
3885 struct list_head *pos, *n;
3886 struct ext4_li_request *elr;
3887
3888 mutex_lock(&ext4_li_info->li_list_mtx);
3889 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3890 elr = list_entry(pos, struct ext4_li_request,
3891 lr_request);
3892 ext4_remove_li_request(elr);
3893 }
3894 mutex_unlock(&ext4_li_info->li_list_mtx);
3895}
3896
3897static int ext4_run_lazyinit_thread(void)
3898{
3899 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3900 ext4_li_info, "ext4lazyinit");
3901 if (IS_ERR(ext4_lazyinit_task)) {
3902 int err = PTR_ERR(ext4_lazyinit_task);
3903 ext4_clear_request_list();
3904 kfree(ext4_li_info);
3905 ext4_li_info = NULL;
3906 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3907 "initialization thread\n",
3908 err);
3909 return err;
3910 }
3911 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3912 return 0;
3913}
3914
3915/*
3916 * Check whether it make sense to run itable init. thread or not.
3917 * If there is at least one uninitialized inode table, return
3918 * corresponding group number, else the loop goes through all
3919 * groups and return total number of groups.
3920 */
3921static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3922{
3923 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3924 struct ext4_group_desc *gdp = NULL;
3925
3926 if (!ext4_has_group_desc_csum(sb))
3927 return ngroups;
3928
3929 for (group = 0; group < ngroups; group++) {
3930 gdp = ext4_get_group_desc(sb, group, NULL);
3931 if (!gdp)
3932 continue;
3933
3934 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3935 break;
3936 }
3937
3938 return group;
3939}
3940
3941static int ext4_li_info_new(void)
3942{
3943 struct ext4_lazy_init *eli = NULL;
3944
3945 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3946 if (!eli)
3947 return -ENOMEM;
3948
3949 INIT_LIST_HEAD(&eli->li_request_list);
3950 mutex_init(&eli->li_list_mtx);
3951
3952 eli->li_state |= EXT4_LAZYINIT_QUIT;
3953
3954 ext4_li_info = eli;
3955
3956 return 0;
3957}
3958
3959static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3960 ext4_group_t start)
3961{
3962 struct ext4_li_request *elr;
3963
3964 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3965 if (!elr)
3966 return NULL;
3967
3968 elr->lr_super = sb;
3969 elr->lr_first_not_zeroed = start;
3970 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3971 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3972 elr->lr_next_group = start;
3973 } else {
3974 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3975 }
3976
3977 /*
3978 * Randomize first schedule time of the request to
3979 * spread the inode table initialization requests
3980 * better.
3981 */
3982 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3983 return elr;
3984}
3985
3986int ext4_register_li_request(struct super_block *sb,
3987 ext4_group_t first_not_zeroed)
3988{
3989 struct ext4_sb_info *sbi = EXT4_SB(sb);
3990 struct ext4_li_request *elr = NULL;
3991 ext4_group_t ngroups = sbi->s_groups_count;
3992 int ret = 0;
3993
3994 mutex_lock(&ext4_li_mtx);
3995 if (sbi->s_li_request != NULL) {
3996 /*
3997 * Reset timeout so it can be computed again, because
3998 * s_li_wait_mult might have changed.
3999 */
4000 sbi->s_li_request->lr_timeout = 0;
4001 goto out;
4002 }
4003
4004 if (sb_rdonly(sb) ||
4005 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4006 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE))))
4007 goto out;
4008
4009 elr = ext4_li_request_new(sb, first_not_zeroed);
4010 if (!elr) {
4011 ret = -ENOMEM;
4012 goto out;
4013 }
4014
4015 if (NULL == ext4_li_info) {
4016 ret = ext4_li_info_new();
4017 if (ret)
4018 goto out;
4019 }
4020
4021 mutex_lock(&ext4_li_info->li_list_mtx);
4022 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4023 mutex_unlock(&ext4_li_info->li_list_mtx);
4024
4025 sbi->s_li_request = elr;
4026 /*
4027 * set elr to NULL here since it has been inserted to
4028 * the request_list and the removal and free of it is
4029 * handled by ext4_clear_request_list from now on.
4030 */
4031 elr = NULL;
4032
4033 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4034 ret = ext4_run_lazyinit_thread();
4035 if (ret)
4036 goto out;
4037 }
4038out:
4039 mutex_unlock(&ext4_li_mtx);
4040 if (ret)
4041 kfree(elr);
4042 return ret;
4043}
4044
4045/*
4046 * We do not need to lock anything since this is called on
4047 * module unload.
4048 */
4049static void ext4_destroy_lazyinit_thread(void)
4050{
4051 /*
4052 * If thread exited earlier
4053 * there's nothing to be done.
4054 */
4055 if (!ext4_li_info || !ext4_lazyinit_task)
4056 return;
4057
4058 kthread_stop(ext4_lazyinit_task);
4059}
4060
4061static int set_journal_csum_feature_set(struct super_block *sb)
4062{
4063 int ret = 1;
4064 int compat, incompat;
4065 struct ext4_sb_info *sbi = EXT4_SB(sb);
4066
4067 if (ext4_has_metadata_csum(sb)) {
4068 /* journal checksum v3 */
4069 compat = 0;
4070 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4071 } else {
4072 /* journal checksum v1 */
4073 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4074 incompat = 0;
4075 }
4076
4077 jbd2_journal_clear_features(sbi->s_journal,
4078 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4079 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4080 JBD2_FEATURE_INCOMPAT_CSUM_V2);
4081 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4082 ret = jbd2_journal_set_features(sbi->s_journal,
4083 compat, 0,
4084 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4085 incompat);
4086 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4087 ret = jbd2_journal_set_features(sbi->s_journal,
4088 compat, 0,
4089 incompat);
4090 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4091 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4092 } else {
4093 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4094 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4095 }
4096
4097 return ret;
4098}
4099
4100/*
4101 * Note: calculating the overhead so we can be compatible with
4102 * historical BSD practice is quite difficult in the face of
4103 * clusters/bigalloc. This is because multiple metadata blocks from
4104 * different block group can end up in the same allocation cluster.
4105 * Calculating the exact overhead in the face of clustered allocation
4106 * requires either O(all block bitmaps) in memory or O(number of block
4107 * groups**2) in time. We will still calculate the superblock for
4108 * older file systems --- and if we come across with a bigalloc file
4109 * system with zero in s_overhead_clusters the estimate will be close to
4110 * correct especially for very large cluster sizes --- but for newer
4111 * file systems, it's better to calculate this figure once at mkfs
4112 * time, and store it in the superblock. If the superblock value is
4113 * present (even for non-bigalloc file systems), we will use it.
4114 */
4115static int count_overhead(struct super_block *sb, ext4_group_t grp,
4116 char *buf)
4117{
4118 struct ext4_sb_info *sbi = EXT4_SB(sb);
4119 struct ext4_group_desc *gdp;
4120 ext4_fsblk_t first_block, last_block, b;
4121 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4122 int s, j, count = 0;
4123 int has_super = ext4_bg_has_super(sb, grp);
4124
4125 if (!ext4_has_feature_bigalloc(sb))
4126 return (has_super + ext4_bg_num_gdb(sb, grp) +
4127 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4128 sbi->s_itb_per_group + 2);
4129
4130 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4131 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4132 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4133 for (i = 0; i < ngroups; i++) {
4134 gdp = ext4_get_group_desc(sb, i, NULL);
4135 b = ext4_block_bitmap(sb, gdp);
4136 if (b >= first_block && b <= last_block) {
4137 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4138 count++;
4139 }
4140 b = ext4_inode_bitmap(sb, gdp);
4141 if (b >= first_block && b <= last_block) {
4142 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4143 count++;
4144 }
4145 b = ext4_inode_table(sb, gdp);
4146 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4147 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4148 int c = EXT4_B2C(sbi, b - first_block);
4149 ext4_set_bit(c, buf);
4150 count++;
4151 }
4152 if (i != grp)
4153 continue;
4154 s = 0;
4155 if (ext4_bg_has_super(sb, grp)) {
4156 ext4_set_bit(s++, buf);
4157 count++;
4158 }
4159 j = ext4_bg_num_gdb(sb, grp);
4160 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4161 ext4_error(sb, "Invalid number of block group "
4162 "descriptor blocks: %d", j);
4163 j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4164 }
4165 count += j;
4166 for (; j > 0; j--)
4167 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4168 }
4169 if (!count)
4170 return 0;
4171 return EXT4_CLUSTERS_PER_GROUP(sb) -
4172 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4173}
4174
4175/*
4176 * Compute the overhead and stash it in sbi->s_overhead
4177 */
4178int ext4_calculate_overhead(struct super_block *sb)
4179{
4180 struct ext4_sb_info *sbi = EXT4_SB(sb);
4181 struct ext4_super_block *es = sbi->s_es;
4182 struct inode *j_inode;
4183 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4184 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4185 ext4_fsblk_t overhead = 0;
4186 char *buf = (char *) get_zeroed_page(GFP_NOFS);
4187
4188 if (!buf)
4189 return -ENOMEM;
4190
4191 /*
4192 * Compute the overhead (FS structures). This is constant
4193 * for a given filesystem unless the number of block groups
4194 * changes so we cache the previous value until it does.
4195 */
4196
4197 /*
4198 * All of the blocks before first_data_block are overhead
4199 */
4200 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4201
4202 /*
4203 * Add the overhead found in each block group
4204 */
4205 for (i = 0; i < ngroups; i++) {
4206 int blks;
4207
4208 blks = count_overhead(sb, i, buf);
4209 overhead += blks;
4210 if (blks)
4211 memset(buf, 0, PAGE_SIZE);
4212 cond_resched();
4213 }
4214
4215 /*
4216 * Add the internal journal blocks whether the journal has been
4217 * loaded or not
4218 */
4219 if (sbi->s_journal && !sbi->s_journal_bdev)
4220 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4221 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4222 /* j_inum for internal journal is non-zero */
4223 j_inode = ext4_get_journal_inode(sb, j_inum);
4224 if (j_inode) {
4225 j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4226 overhead += EXT4_NUM_B2C(sbi, j_blocks);
4227 iput(j_inode);
4228 } else {
4229 ext4_msg(sb, KERN_ERR, "can't get journal size");
4230 }
4231 }
4232 sbi->s_overhead = overhead;
4233 smp_wmb();
4234 free_page((unsigned long) buf);
4235 return 0;
4236}
4237
4238static void ext4_set_resv_clusters(struct super_block *sb)
4239{
4240 ext4_fsblk_t resv_clusters;
4241 struct ext4_sb_info *sbi = EXT4_SB(sb);
4242
4243 /*
4244 * There's no need to reserve anything when we aren't using extents.
4245 * The space estimates are exact, there are no unwritten extents,
4246 * hole punching doesn't need new metadata... This is needed especially
4247 * to keep ext2/3 backward compatibility.
4248 */
4249 if (!ext4_has_feature_extents(sb))
4250 return;
4251 /*
4252 * By default we reserve 2% or 4096 clusters, whichever is smaller.
4253 * This should cover the situations where we can not afford to run
4254 * out of space like for example punch hole, or converting
4255 * unwritten extents in delalloc path. In most cases such
4256 * allocation would require 1, or 2 blocks, higher numbers are
4257 * very rare.
4258 */
4259 resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4260 sbi->s_cluster_bits);
4261
4262 do_div(resv_clusters, 50);
4263 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4264
4265 atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4266}
4267
4268static const char *ext4_quota_mode(struct super_block *sb)
4269{
4270#ifdef CONFIG_QUOTA
4271 if (!ext4_quota_capable(sb))
4272 return "none";
4273
4274 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4275 return "journalled";
4276 else
4277 return "writeback";
4278#else
4279 return "disabled";
4280#endif
4281}
4282
4283static void ext4_setup_csum_trigger(struct super_block *sb,
4284 enum ext4_journal_trigger_type type,
4285 void (*trigger)(
4286 struct jbd2_buffer_trigger_type *type,
4287 struct buffer_head *bh,
4288 void *mapped_data,
4289 size_t size))
4290{
4291 struct ext4_sb_info *sbi = EXT4_SB(sb);
4292
4293 sbi->s_journal_triggers[type].sb = sb;
4294 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4295}
4296
4297static void ext4_free_sbi(struct ext4_sb_info *sbi)
4298{
4299 if (!sbi)
4300 return;
4301
4302 kfree(sbi->s_blockgroup_lock);
4303 fs_put_dax(sbi->s_daxdev, NULL);
4304 kfree(sbi);
4305}
4306
4307static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4308{
4309 struct ext4_sb_info *sbi;
4310
4311 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4312 if (!sbi)
4313 return NULL;
4314
4315 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off,
4316 NULL, NULL);
4317
4318 sbi->s_blockgroup_lock =
4319 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4320
4321 if (!sbi->s_blockgroup_lock)
4322 goto err_out;
4323
4324 sb->s_fs_info = sbi;
4325 sbi->s_sb = sb;
4326 return sbi;
4327err_out:
4328 fs_put_dax(sbi->s_daxdev, NULL);
4329 kfree(sbi);
4330 return NULL;
4331}
4332
4333static void ext4_set_def_opts(struct super_block *sb,
4334 struct ext4_super_block *es)
4335{
4336 unsigned long def_mount_opts;
4337
4338 /* Set defaults before we parse the mount options */
4339 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4340 set_opt(sb, INIT_INODE_TABLE);
4341 if (def_mount_opts & EXT4_DEFM_DEBUG)
4342 set_opt(sb, DEBUG);
4343 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4344 set_opt(sb, GRPID);
4345 if (def_mount_opts & EXT4_DEFM_UID16)
4346 set_opt(sb, NO_UID32);
4347 /* xattr user namespace & acls are now defaulted on */
4348 set_opt(sb, XATTR_USER);
4349#ifdef CONFIG_EXT4_FS_POSIX_ACL
4350 set_opt(sb, POSIX_ACL);
4351#endif
4352 if (ext4_has_feature_fast_commit(sb))
4353 set_opt2(sb, JOURNAL_FAST_COMMIT);
4354 /* don't forget to enable journal_csum when metadata_csum is enabled. */
4355 if (ext4_has_metadata_csum(sb))
4356 set_opt(sb, JOURNAL_CHECKSUM);
4357
4358 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4359 set_opt(sb, JOURNAL_DATA);
4360 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4361 set_opt(sb, ORDERED_DATA);
4362 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4363 set_opt(sb, WRITEBACK_DATA);
4364
4365 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC)
4366 set_opt(sb, ERRORS_PANIC);
4367 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE)
4368 set_opt(sb, ERRORS_CONT);
4369 else
4370 set_opt(sb, ERRORS_RO);
4371 /* block_validity enabled by default; disable with noblock_validity */
4372 set_opt(sb, BLOCK_VALIDITY);
4373 if (def_mount_opts & EXT4_DEFM_DISCARD)
4374 set_opt(sb, DISCARD);
4375
4376 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4377 set_opt(sb, BARRIER);
4378
4379 /*
4380 * enable delayed allocation by default
4381 * Use -o nodelalloc to turn it off
4382 */
4383 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4384 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4385 set_opt(sb, DELALLOC);
4386
4387 if (sb->s_blocksize == PAGE_SIZE)
4388 set_opt(sb, DIOREAD_NOLOCK);
4389}
4390
4391static int ext4_handle_clustersize(struct super_block *sb)
4392{
4393 struct ext4_sb_info *sbi = EXT4_SB(sb);
4394 struct ext4_super_block *es = sbi->s_es;
4395 int clustersize;
4396
4397 /* Handle clustersize */
4398 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4399 if (ext4_has_feature_bigalloc(sb)) {
4400 if (clustersize < sb->s_blocksize) {
4401 ext4_msg(sb, KERN_ERR,
4402 "cluster size (%d) smaller than "
4403 "block size (%lu)", clustersize, sb->s_blocksize);
4404 return -EINVAL;
4405 }
4406 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4407 le32_to_cpu(es->s_log_block_size);
4408 sbi->s_clusters_per_group =
4409 le32_to_cpu(es->s_clusters_per_group);
4410 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) {
4411 ext4_msg(sb, KERN_ERR,
4412 "#clusters per group too big: %lu",
4413 sbi->s_clusters_per_group);
4414 return -EINVAL;
4415 }
4416 if (sbi->s_blocks_per_group !=
4417 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) {
4418 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4419 "clusters per group (%lu) inconsistent",
4420 sbi->s_blocks_per_group,
4421 sbi->s_clusters_per_group);
4422 return -EINVAL;
4423 }
4424 } else {
4425 if (clustersize != sb->s_blocksize) {
4426 ext4_msg(sb, KERN_ERR,
4427 "fragment/cluster size (%d) != "
4428 "block size (%lu)", clustersize, sb->s_blocksize);
4429 return -EINVAL;
4430 }
4431 if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
4432 ext4_msg(sb, KERN_ERR,
4433 "#blocks per group too big: %lu",
4434 sbi->s_blocks_per_group);
4435 return -EINVAL;
4436 }
4437 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4438 sbi->s_cluster_bits = 0;
4439 }
4440 sbi->s_cluster_ratio = clustersize / sb->s_blocksize;
4441
4442 /* Do we have standard group size of clustersize * 8 blocks ? */
4443 if (sbi->s_blocks_per_group == clustersize << 3)
4444 set_opt2(sb, STD_GROUP_SIZE);
4445
4446 return 0;
4447}
4448
4449static void ext4_fast_commit_init(struct super_block *sb)
4450{
4451 struct ext4_sb_info *sbi = EXT4_SB(sb);
4452
4453 /* Initialize fast commit stuff */
4454 atomic_set(&sbi->s_fc_subtid, 0);
4455 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4456 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4457 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4458 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4459 sbi->s_fc_bytes = 0;
4460 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4461 sbi->s_fc_ineligible_tid = 0;
4462 spin_lock_init(&sbi->s_fc_lock);
4463 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4464 sbi->s_fc_replay_state.fc_regions = NULL;
4465 sbi->s_fc_replay_state.fc_regions_size = 0;
4466 sbi->s_fc_replay_state.fc_regions_used = 0;
4467 sbi->s_fc_replay_state.fc_regions_valid = 0;
4468 sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4469 sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4470 sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4471}
4472
4473static int ext4_inode_info_init(struct super_block *sb,
4474 struct ext4_super_block *es)
4475{
4476 struct ext4_sb_info *sbi = EXT4_SB(sb);
4477
4478 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4479 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4480 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4481 } else {
4482 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4483 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4484 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4485 ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4486 sbi->s_first_ino);
4487 return -EINVAL;
4488 }
4489 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4490 (!is_power_of_2(sbi->s_inode_size)) ||
4491 (sbi->s_inode_size > sb->s_blocksize)) {
4492 ext4_msg(sb, KERN_ERR,
4493 "unsupported inode size: %d",
4494 sbi->s_inode_size);
4495 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize);
4496 return -EINVAL;
4497 }
4498 /*
4499 * i_atime_extra is the last extra field available for
4500 * [acm]times in struct ext4_inode. Checking for that
4501 * field should suffice to ensure we have extra space
4502 * for all three.
4503 */
4504 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4505 sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4506 sb->s_time_gran = 1;
4507 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4508 } else {
4509 sb->s_time_gran = NSEC_PER_SEC;
4510 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4511 }
4512 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4513 }
4514
4515 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4516 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4517 EXT4_GOOD_OLD_INODE_SIZE;
4518 if (ext4_has_feature_extra_isize(sb)) {
4519 unsigned v, max = (sbi->s_inode_size -
4520 EXT4_GOOD_OLD_INODE_SIZE);
4521
4522 v = le16_to_cpu(es->s_want_extra_isize);
4523 if (v > max) {
4524 ext4_msg(sb, KERN_ERR,
4525 "bad s_want_extra_isize: %d", v);
4526 return -EINVAL;
4527 }
4528 if (sbi->s_want_extra_isize < v)
4529 sbi->s_want_extra_isize = v;
4530
4531 v = le16_to_cpu(es->s_min_extra_isize);
4532 if (v > max) {
4533 ext4_msg(sb, KERN_ERR,
4534 "bad s_min_extra_isize: %d", v);
4535 return -EINVAL;
4536 }
4537 if (sbi->s_want_extra_isize < v)
4538 sbi->s_want_extra_isize = v;
4539 }
4540 }
4541
4542 return 0;
4543}
4544
4545#if IS_ENABLED(CONFIG_UNICODE)
4546static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4547{
4548 const struct ext4_sb_encodings *encoding_info;
4549 struct unicode_map *encoding;
4550 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4551
4552 if (!ext4_has_feature_casefold(sb) || sb->s_encoding)
4553 return 0;
4554
4555 encoding_info = ext4_sb_read_encoding(es);
4556 if (!encoding_info) {
4557 ext4_msg(sb, KERN_ERR,
4558 "Encoding requested by superblock is unknown");
4559 return -EINVAL;
4560 }
4561
4562 encoding = utf8_load(encoding_info->version);
4563 if (IS_ERR(encoding)) {
4564 ext4_msg(sb, KERN_ERR,
4565 "can't mount with superblock charset: %s-%u.%u.%u "
4566 "not supported by the kernel. flags: 0x%x.",
4567 encoding_info->name,
4568 unicode_major(encoding_info->version),
4569 unicode_minor(encoding_info->version),
4570 unicode_rev(encoding_info->version),
4571 encoding_flags);
4572 return -EINVAL;
4573 }
4574 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4575 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4576 unicode_major(encoding_info->version),
4577 unicode_minor(encoding_info->version),
4578 unicode_rev(encoding_info->version),
4579 encoding_flags);
4580
4581 sb->s_encoding = encoding;
4582 sb->s_encoding_flags = encoding_flags;
4583
4584 return 0;
4585}
4586#else
4587static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4588{
4589 return 0;
4590}
4591#endif
4592
4593static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es)
4594{
4595 struct ext4_sb_info *sbi = EXT4_SB(sb);
4596
4597 /* Warn if metadata_csum and gdt_csum are both set. */
4598 if (ext4_has_feature_metadata_csum(sb) &&
4599 ext4_has_feature_gdt_csum(sb))
4600 ext4_warning(sb, "metadata_csum and uninit_bg are "
4601 "redundant flags; please run fsck.");
4602
4603 /* Check for a known checksum algorithm */
4604 if (!ext4_verify_csum_type(sb, es)) {
4605 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4606 "unknown checksum algorithm.");
4607 return -EINVAL;
4608 }
4609 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4610 ext4_orphan_file_block_trigger);
4611
4612 /* Load the checksum driver */
4613 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4614 if (IS_ERR(sbi->s_chksum_driver)) {
4615 int ret = PTR_ERR(sbi->s_chksum_driver);
4616 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4617 sbi->s_chksum_driver = NULL;
4618 return ret;
4619 }
4620
4621 /* Check superblock checksum */
4622 if (!ext4_superblock_csum_verify(sb, es)) {
4623 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4624 "invalid superblock checksum. Run e2fsck?");
4625 return -EFSBADCRC;
4626 }
4627
4628 /* Precompute checksum seed for all metadata */
4629 if (ext4_has_feature_csum_seed(sb))
4630 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4631 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4632 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4633 sizeof(es->s_uuid));
4634 return 0;
4635}
4636
4637static int ext4_check_feature_compatibility(struct super_block *sb,
4638 struct ext4_super_block *es,
4639 int silent)
4640{
4641 struct ext4_sb_info *sbi = EXT4_SB(sb);
4642
4643 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4644 (ext4_has_compat_features(sb) ||
4645 ext4_has_ro_compat_features(sb) ||
4646 ext4_has_incompat_features(sb)))
4647 ext4_msg(sb, KERN_WARNING,
4648 "feature flags set on rev 0 fs, "
4649 "running e2fsck is recommended");
4650
4651 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4652 set_opt2(sb, HURD_COMPAT);
4653 if (ext4_has_feature_64bit(sb)) {
4654 ext4_msg(sb, KERN_ERR,
4655 "The Hurd can't support 64-bit file systems");
4656 return -EINVAL;
4657 }
4658
4659 /*
4660 * ea_inode feature uses l_i_version field which is not
4661 * available in HURD_COMPAT mode.
4662 */
4663 if (ext4_has_feature_ea_inode(sb)) {
4664 ext4_msg(sb, KERN_ERR,
4665 "ea_inode feature is not supported for Hurd");
4666 return -EINVAL;
4667 }
4668 }
4669
4670 if (IS_EXT2_SB(sb)) {
4671 if (ext2_feature_set_ok(sb))
4672 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4673 "using the ext4 subsystem");
4674 else {
4675 /*
4676 * If we're probing be silent, if this looks like
4677 * it's actually an ext[34] filesystem.
4678 */
4679 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4680 return -EINVAL;
4681 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4682 "to feature incompatibilities");
4683 return -EINVAL;
4684 }
4685 }
4686
4687 if (IS_EXT3_SB(sb)) {
4688 if (ext3_feature_set_ok(sb))
4689 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4690 "using the ext4 subsystem");
4691 else {
4692 /*
4693 * If we're probing be silent, if this looks like
4694 * it's actually an ext4 filesystem.
4695 */
4696 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4697 return -EINVAL;
4698 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4699 "to feature incompatibilities");
4700 return -EINVAL;
4701 }
4702 }
4703
4704 /*
4705 * Check feature flags regardless of the revision level, since we
4706 * previously didn't change the revision level when setting the flags,
4707 * so there is a chance incompat flags are set on a rev 0 filesystem.
4708 */
4709 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4710 return -EINVAL;
4711
4712 if (sbi->s_daxdev) {
4713 if (sb->s_blocksize == PAGE_SIZE)
4714 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4715 else
4716 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4717 }
4718
4719 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4720 if (ext4_has_feature_inline_data(sb)) {
4721 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4722 " that may contain inline data");
4723 return -EINVAL;
4724 }
4725 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4726 ext4_msg(sb, KERN_ERR,
4727 "DAX unsupported by block device.");
4728 return -EINVAL;
4729 }
4730 }
4731
4732 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4733 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4734 es->s_encryption_level);
4735 return -EINVAL;
4736 }
4737
4738 return 0;
4739}
4740
4741static int ext4_check_geometry(struct super_block *sb,
4742 struct ext4_super_block *es)
4743{
4744 struct ext4_sb_info *sbi = EXT4_SB(sb);
4745 __u64 blocks_count;
4746 int err;
4747
4748 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) {
4749 ext4_msg(sb, KERN_ERR,
4750 "Number of reserved GDT blocks insanely large: %d",
4751 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4752 return -EINVAL;
4753 }
4754 /*
4755 * Test whether we have more sectors than will fit in sector_t,
4756 * and whether the max offset is addressable by the page cache.
4757 */
4758 err = generic_check_addressable(sb->s_blocksize_bits,
4759 ext4_blocks_count(es));
4760 if (err) {
4761 ext4_msg(sb, KERN_ERR, "filesystem"
4762 " too large to mount safely on this system");
4763 return err;
4764 }
4765
4766 /* check blocks count against device size */
4767 blocks_count = sb_bdev_nr_blocks(sb);
4768 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4769 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4770 "exceeds size of device (%llu blocks)",
4771 ext4_blocks_count(es), blocks_count);
4772 return -EINVAL;
4773 }
4774
4775 /*
4776 * It makes no sense for the first data block to be beyond the end
4777 * of the filesystem.
4778 */
4779 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4780 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4781 "block %u is beyond end of filesystem (%llu)",
4782 le32_to_cpu(es->s_first_data_block),
4783 ext4_blocks_count(es));
4784 return -EINVAL;
4785 }
4786 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4787 (sbi->s_cluster_ratio == 1)) {
4788 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4789 "block is 0 with a 1k block and cluster size");
4790 return -EINVAL;
4791 }
4792
4793 blocks_count = (ext4_blocks_count(es) -
4794 le32_to_cpu(es->s_first_data_block) +
4795 EXT4_BLOCKS_PER_GROUP(sb) - 1);
4796 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4797 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4798 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4799 "(block count %llu, first data block %u, "
4800 "blocks per group %lu)", blocks_count,
4801 ext4_blocks_count(es),
4802 le32_to_cpu(es->s_first_data_block),
4803 EXT4_BLOCKS_PER_GROUP(sb));
4804 return -EINVAL;
4805 }
4806 sbi->s_groups_count = blocks_count;
4807 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4808 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4809 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4810 le32_to_cpu(es->s_inodes_count)) {
4811 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4812 le32_to_cpu(es->s_inodes_count),
4813 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4814 return -EINVAL;
4815 }
4816
4817 return 0;
4818}
4819
4820static int ext4_group_desc_init(struct super_block *sb,
4821 struct ext4_super_block *es,
4822 ext4_fsblk_t logical_sb_block,
4823 ext4_group_t *first_not_zeroed)
4824{
4825 struct ext4_sb_info *sbi = EXT4_SB(sb);
4826 unsigned int db_count;
4827 ext4_fsblk_t block;
4828 int i;
4829
4830 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4831 EXT4_DESC_PER_BLOCK(sb);
4832 if (ext4_has_feature_meta_bg(sb)) {
4833 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4834 ext4_msg(sb, KERN_WARNING,
4835 "first meta block group too large: %u "
4836 "(group descriptor block count %u)",
4837 le32_to_cpu(es->s_first_meta_bg), db_count);
4838 return -EINVAL;
4839 }
4840 }
4841 rcu_assign_pointer(sbi->s_group_desc,
4842 kvmalloc_array(db_count,
4843 sizeof(struct buffer_head *),
4844 GFP_KERNEL));
4845 if (sbi->s_group_desc == NULL) {
4846 ext4_msg(sb, KERN_ERR, "not enough memory");
4847 return -ENOMEM;
4848 }
4849
4850 bgl_lock_init(sbi->s_blockgroup_lock);
4851
4852 /* Pre-read the descriptors into the buffer cache */
4853 for (i = 0; i < db_count; i++) {
4854 block = descriptor_loc(sb, logical_sb_block, i);
4855 ext4_sb_breadahead_unmovable(sb, block);
4856 }
4857
4858 for (i = 0; i < db_count; i++) {
4859 struct buffer_head *bh;
4860
4861 block = descriptor_loc(sb, logical_sb_block, i);
4862 bh = ext4_sb_bread_unmovable(sb, block);
4863 if (IS_ERR(bh)) {
4864 ext4_msg(sb, KERN_ERR,
4865 "can't read group descriptor %d", i);
4866 sbi->s_gdb_count = i;
4867 return PTR_ERR(bh);
4868 }
4869 rcu_read_lock();
4870 rcu_dereference(sbi->s_group_desc)[i] = bh;
4871 rcu_read_unlock();
4872 }
4873 sbi->s_gdb_count = db_count;
4874 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) {
4875 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4876 return -EFSCORRUPTED;
4877 }
4878
4879 return 0;
4880}
4881
4882static int ext4_load_and_init_journal(struct super_block *sb,
4883 struct ext4_super_block *es,
4884 struct ext4_fs_context *ctx)
4885{
4886 struct ext4_sb_info *sbi = EXT4_SB(sb);
4887 int err;
4888
4889 err = ext4_load_journal(sb, es, ctx->journal_devnum);
4890 if (err)
4891 return err;
4892
4893 if (ext4_has_feature_64bit(sb) &&
4894 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4895 JBD2_FEATURE_INCOMPAT_64BIT)) {
4896 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4897 goto out;
4898 }
4899
4900 if (!set_journal_csum_feature_set(sb)) {
4901 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4902 "feature set");
4903 goto out;
4904 }
4905
4906 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4907 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4908 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4909 ext4_msg(sb, KERN_ERR,
4910 "Failed to set fast commit journal feature");
4911 goto out;
4912 }
4913
4914 /* We have now updated the journal if required, so we can
4915 * validate the data journaling mode. */
4916 switch (test_opt(sb, DATA_FLAGS)) {
4917 case 0:
4918 /* No mode set, assume a default based on the journal
4919 * capabilities: ORDERED_DATA if the journal can
4920 * cope, else JOURNAL_DATA
4921 */
4922 if (jbd2_journal_check_available_features
4923 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4924 set_opt(sb, ORDERED_DATA);
4925 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4926 } else {
4927 set_opt(sb, JOURNAL_DATA);
4928 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4929 }
4930 break;
4931
4932 case EXT4_MOUNT_ORDERED_DATA:
4933 case EXT4_MOUNT_WRITEBACK_DATA:
4934 if (!jbd2_journal_check_available_features
4935 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4936 ext4_msg(sb, KERN_ERR, "Journal does not support "
4937 "requested data journaling mode");
4938 goto out;
4939 }
4940 break;
4941 default:
4942 break;
4943 }
4944
4945 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4946 test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4947 ext4_msg(sb, KERN_ERR, "can't mount with "
4948 "journal_async_commit in data=ordered mode");
4949 goto out;
4950 }
4951
4952 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
4953
4954 sbi->s_journal->j_submit_inode_data_buffers =
4955 ext4_journal_submit_inode_data_buffers;
4956 sbi->s_journal->j_finish_inode_data_buffers =
4957 ext4_journal_finish_inode_data_buffers;
4958
4959 return 0;
4960
4961out:
4962 /* flush s_error_work before journal destroy. */
4963 flush_work(&sbi->s_error_work);
4964 jbd2_journal_destroy(sbi->s_journal);
4965 sbi->s_journal = NULL;
4966 return -EINVAL;
4967}
4968
4969static int ext4_check_journal_data_mode(struct super_block *sb)
4970{
4971 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4972 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with "
4973 "data=journal disables delayed allocation, "
4974 "dioread_nolock, O_DIRECT and fast_commit support!\n");
4975 /* can't mount with both data=journal and dioread_nolock. */
4976 clear_opt(sb, DIOREAD_NOLOCK);
4977 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4978 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4979 ext4_msg(sb, KERN_ERR, "can't mount with "
4980 "both data=journal and delalloc");
4981 return -EINVAL;
4982 }
4983 if (test_opt(sb, DAX_ALWAYS)) {
4984 ext4_msg(sb, KERN_ERR, "can't mount with "
4985 "both data=journal and dax");
4986 return -EINVAL;
4987 }
4988 if (ext4_has_feature_encrypt(sb)) {
4989 ext4_msg(sb, KERN_WARNING,
4990 "encrypted files will use data=ordered "
4991 "instead of data journaling mode");
4992 }
4993 if (test_opt(sb, DELALLOC))
4994 clear_opt(sb, DELALLOC);
4995 } else {
4996 sb->s_iflags |= SB_I_CGROUPWB;
4997 }
4998
4999 return 0;
5000}
5001
5002static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb,
5003 int silent)
5004{
5005 struct ext4_sb_info *sbi = EXT4_SB(sb);
5006 struct ext4_super_block *es;
5007 ext4_fsblk_t logical_sb_block;
5008 unsigned long offset = 0;
5009 struct buffer_head *bh;
5010 int ret = -EINVAL;
5011 int blocksize;
5012
5013 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
5014 if (!blocksize) {
5015 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
5016 return -EINVAL;
5017 }
5018
5019 /*
5020 * The ext4 superblock will not be buffer aligned for other than 1kB
5021 * block sizes. We need to calculate the offset from buffer start.
5022 */
5023 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
5024 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5025 offset = do_div(logical_sb_block, blocksize);
5026 } else {
5027 logical_sb_block = sbi->s_sb_block;
5028 }
5029
5030 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5031 if (IS_ERR(bh)) {
5032 ext4_msg(sb, KERN_ERR, "unable to read superblock");
5033 return PTR_ERR(bh);
5034 }
5035 /*
5036 * Note: s_es must be initialized as soon as possible because
5037 * some ext4 macro-instructions depend on its value
5038 */
5039 es = (struct ext4_super_block *) (bh->b_data + offset);
5040 sbi->s_es = es;
5041 sb->s_magic = le16_to_cpu(es->s_magic);
5042 if (sb->s_magic != EXT4_SUPER_MAGIC) {
5043 if (!silent)
5044 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5045 goto out;
5046 }
5047
5048 if (le32_to_cpu(es->s_log_block_size) >
5049 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5050 ext4_msg(sb, KERN_ERR,
5051 "Invalid log block size: %u",
5052 le32_to_cpu(es->s_log_block_size));
5053 goto out;
5054 }
5055 if (le32_to_cpu(es->s_log_cluster_size) >
5056 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5057 ext4_msg(sb, KERN_ERR,
5058 "Invalid log cluster size: %u",
5059 le32_to_cpu(es->s_log_cluster_size));
5060 goto out;
5061 }
5062
5063 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
5064
5065 /*
5066 * If the default block size is not the same as the real block size,
5067 * we need to reload it.
5068 */
5069 if (sb->s_blocksize == blocksize) {
5070 *lsb = logical_sb_block;
5071 sbi->s_sbh = bh;
5072 return 0;
5073 }
5074
5075 /*
5076 * bh must be released before kill_bdev(), otherwise
5077 * it won't be freed and its page also. kill_bdev()
5078 * is called by sb_set_blocksize().
5079 */
5080 brelse(bh);
5081 /* Validate the filesystem blocksize */
5082 if (!sb_set_blocksize(sb, blocksize)) {
5083 ext4_msg(sb, KERN_ERR, "bad block size %d",
5084 blocksize);
5085 bh = NULL;
5086 goto out;
5087 }
5088
5089 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5090 offset = do_div(logical_sb_block, blocksize);
5091 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5092 if (IS_ERR(bh)) {
5093 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try");
5094 ret = PTR_ERR(bh);
5095 bh = NULL;
5096 goto out;
5097 }
5098 es = (struct ext4_super_block *)(bh->b_data + offset);
5099 sbi->s_es = es;
5100 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
5101 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!");
5102 goto out;
5103 }
5104 *lsb = logical_sb_block;
5105 sbi->s_sbh = bh;
5106 return 0;
5107out:
5108 brelse(bh);
5109 return ret;
5110}
5111
5112static void ext4_hash_info_init(struct super_block *sb)
5113{
5114 struct ext4_sb_info *sbi = EXT4_SB(sb);
5115 struct ext4_super_block *es = sbi->s_es;
5116 unsigned int i;
5117
5118 for (i = 0; i < 4; i++)
5119 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
5120
5121 sbi->s_def_hash_version = es->s_def_hash_version;
5122 if (ext4_has_feature_dir_index(sb)) {
5123 i = le32_to_cpu(es->s_flags);
5124 if (i & EXT2_FLAGS_UNSIGNED_HASH)
5125 sbi->s_hash_unsigned = 3;
5126 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
5127#ifdef __CHAR_UNSIGNED__
5128 if (!sb_rdonly(sb))
5129 es->s_flags |=
5130 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
5131 sbi->s_hash_unsigned = 3;
5132#else
5133 if (!sb_rdonly(sb))
5134 es->s_flags |=
5135 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
5136#endif
5137 }
5138 }
5139}
5140
5141static int ext4_block_group_meta_init(struct super_block *sb, int silent)
5142{
5143 struct ext4_sb_info *sbi = EXT4_SB(sb);
5144 struct ext4_super_block *es = sbi->s_es;
5145 int has_huge_files;
5146
5147 has_huge_files = ext4_has_feature_huge_file(sb);
5148 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
5149 has_huge_files);
5150 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
5151
5152 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
5153 if (ext4_has_feature_64bit(sb)) {
5154 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
5155 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
5156 !is_power_of_2(sbi->s_desc_size)) {
5157 ext4_msg(sb, KERN_ERR,
5158 "unsupported descriptor size %lu",
5159 sbi->s_desc_size);
5160 return -EINVAL;
5161 }
5162 } else
5163 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
5164
5165 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
5166 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
5167
5168 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb);
5169 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) {
5170 if (!silent)
5171 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5172 return -EINVAL;
5173 }
5174 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
5175 sbi->s_inodes_per_group > sb->s_blocksize * 8) {
5176 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
5177 sbi->s_inodes_per_group);
5178 return -EINVAL;
5179 }
5180 sbi->s_itb_per_group = sbi->s_inodes_per_group /
5181 sbi->s_inodes_per_block;
5182 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb);
5183 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
5184 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
5185 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
5186
5187 return 0;
5188}
5189
5190static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
5191{
5192 struct ext4_super_block *es = NULL;
5193 struct ext4_sb_info *sbi = EXT4_SB(sb);
5194 ext4_fsblk_t logical_sb_block;
5195 struct inode *root;
5196 int needs_recovery;
5197 int err;
5198 ext4_group_t first_not_zeroed;
5199 struct ext4_fs_context *ctx = fc->fs_private;
5200 int silent = fc->sb_flags & SB_SILENT;
5201
5202 /* Set defaults for the variables that will be set during parsing */
5203 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
5204 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5205
5206 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
5207 sbi->s_sectors_written_start =
5208 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
5209
5210 err = ext4_load_super(sb, &logical_sb_block, silent);
5211 if (err)
5212 goto out_fail;
5213
5214 es = sbi->s_es;
5215 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
5216
5217 err = ext4_init_metadata_csum(sb, es);
5218 if (err)
5219 goto failed_mount;
5220
5221 ext4_set_def_opts(sb, es);
5222
5223 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
5224 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
5225 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
5226 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
5227 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
5228
5229 /*
5230 * set default s_li_wait_mult for lazyinit, for the case there is
5231 * no mount option specified.
5232 */
5233 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
5234
5235 err = ext4_inode_info_init(sb, es);
5236 if (err)
5237 goto failed_mount;
5238
5239 err = parse_apply_sb_mount_options(sb, ctx);
5240 if (err < 0)
5241 goto failed_mount;
5242
5243 sbi->s_def_mount_opt = sbi->s_mount_opt;
5244 sbi->s_def_mount_opt2 = sbi->s_mount_opt2;
5245
5246 err = ext4_check_opt_consistency(fc, sb);
5247 if (err < 0)
5248 goto failed_mount;
5249
5250 ext4_apply_options(fc, sb);
5251
5252 err = ext4_encoding_init(sb, es);
5253 if (err)
5254 goto failed_mount;
5255
5256 err = ext4_check_journal_data_mode(sb);
5257 if (err)
5258 goto failed_mount;
5259
5260 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5261 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5262
5263 /* i_version is always enabled now */
5264 sb->s_flags |= SB_I_VERSION;
5265
5266 err = ext4_check_feature_compatibility(sb, es, silent);
5267 if (err)
5268 goto failed_mount;
5269
5270 err = ext4_block_group_meta_init(sb, silent);
5271 if (err)
5272 goto failed_mount;
5273
5274 ext4_hash_info_init(sb);
5275
5276 err = ext4_handle_clustersize(sb);
5277 if (err)
5278 goto failed_mount;
5279
5280 err = ext4_check_geometry(sb, es);
5281 if (err)
5282 goto failed_mount;
5283
5284 timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5285 spin_lock_init(&sbi->s_error_lock);
5286 INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
5287
5288 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed);
5289 if (err)
5290 goto failed_mount3;
5291
5292 err = ext4_es_register_shrinker(sbi);
5293 if (err)
5294 goto failed_mount3;
5295
5296 sbi->s_stripe = ext4_get_stripe_size(sbi);
5297 /*
5298 * It's hard to get stripe aligned blocks if stripe is not aligned with
5299 * cluster, just disable stripe and alert user to simpfy code and avoid
5300 * stripe aligned allocation which will rarely successes.
5301 */
5302 if (sbi->s_stripe > 0 && sbi->s_cluster_ratio > 1 &&
5303 sbi->s_stripe % sbi->s_cluster_ratio != 0) {
5304 ext4_msg(sb, KERN_WARNING,
5305 "stripe (%lu) is not aligned with cluster size (%u), "
5306 "stripe is disabled",
5307 sbi->s_stripe, sbi->s_cluster_ratio);
5308 sbi->s_stripe = 0;
5309 }
5310 sbi->s_extent_max_zeroout_kb = 32;
5311
5312 /*
5313 * set up enough so that it can read an inode
5314 */
5315 sb->s_op = &ext4_sops;
5316 sb->s_export_op = &ext4_export_ops;
5317 sb->s_xattr = ext4_xattr_handlers;
5318#ifdef CONFIG_FS_ENCRYPTION
5319 sb->s_cop = &ext4_cryptops;
5320#endif
5321#ifdef CONFIG_FS_VERITY
5322 sb->s_vop = &ext4_verityops;
5323#endif
5324#ifdef CONFIG_QUOTA
5325 sb->dq_op = &ext4_quota_operations;
5326 if (ext4_has_feature_quota(sb))
5327 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5328 else
5329 sb->s_qcop = &ext4_qctl_operations;
5330 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5331#endif
5332 memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
5333
5334 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5335 mutex_init(&sbi->s_orphan_lock);
5336
5337 ext4_fast_commit_init(sb);
5338
5339 sb->s_root = NULL;
5340
5341 needs_recovery = (es->s_last_orphan != 0 ||
5342 ext4_has_feature_orphan_present(sb) ||
5343 ext4_has_feature_journal_needs_recovery(sb));
5344
5345 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) {
5346 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block));
5347 if (err)
5348 goto failed_mount3a;
5349 }
5350
5351 err = -EINVAL;
5352 /*
5353 * The first inode we look at is the journal inode. Don't try
5354 * root first: it may be modified in the journal!
5355 */
5356 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5357 err = ext4_load_and_init_journal(sb, es, ctx);
5358 if (err)
5359 goto failed_mount3a;
5360 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5361 ext4_has_feature_journal_needs_recovery(sb)) {
5362 ext4_msg(sb, KERN_ERR, "required journal recovery "
5363 "suppressed and not mounted read-only");
5364 goto failed_mount3a;
5365 } else {
5366 /* Nojournal mode, all journal mount options are illegal */
5367 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5368 ext4_msg(sb, KERN_ERR, "can't mount with "
5369 "journal_async_commit, fs mounted w/o journal");
5370 goto failed_mount3a;
5371 }
5372
5373 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5374 ext4_msg(sb, KERN_ERR, "can't mount with "
5375 "journal_checksum, fs mounted w/o journal");
5376 goto failed_mount3a;
5377 }
5378 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5379 ext4_msg(sb, KERN_ERR, "can't mount with "
5380 "commit=%lu, fs mounted w/o journal",
5381 sbi->s_commit_interval / HZ);
5382 goto failed_mount3a;
5383 }
5384 if (EXT4_MOUNT_DATA_FLAGS &
5385 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5386 ext4_msg(sb, KERN_ERR, "can't mount with "
5387 "data=, fs mounted w/o journal");
5388 goto failed_mount3a;
5389 }
5390 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5391 clear_opt(sb, JOURNAL_CHECKSUM);
5392 clear_opt(sb, DATA_FLAGS);
5393 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5394 sbi->s_journal = NULL;
5395 needs_recovery = 0;
5396 }
5397
5398 if (!test_opt(sb, NO_MBCACHE)) {
5399 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5400 if (!sbi->s_ea_block_cache) {
5401 ext4_msg(sb, KERN_ERR,
5402 "Failed to create ea_block_cache");
5403 err = -EINVAL;
5404 goto failed_mount_wq;
5405 }
5406
5407 if (ext4_has_feature_ea_inode(sb)) {
5408 sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5409 if (!sbi->s_ea_inode_cache) {
5410 ext4_msg(sb, KERN_ERR,
5411 "Failed to create ea_inode_cache");
5412 err = -EINVAL;
5413 goto failed_mount_wq;
5414 }
5415 }
5416 }
5417
5418 /*
5419 * Get the # of file system overhead blocks from the
5420 * superblock if present.
5421 */
5422 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5423 /* ignore the precalculated value if it is ridiculous */
5424 if (sbi->s_overhead > ext4_blocks_count(es))
5425 sbi->s_overhead = 0;
5426 /*
5427 * If the bigalloc feature is not enabled recalculating the
5428 * overhead doesn't take long, so we might as well just redo
5429 * it to make sure we are using the correct value.
5430 */
5431 if (!ext4_has_feature_bigalloc(sb))
5432 sbi->s_overhead = 0;
5433 if (sbi->s_overhead == 0) {
5434 err = ext4_calculate_overhead(sb);
5435 if (err)
5436 goto failed_mount_wq;
5437 }
5438
5439 /*
5440 * The maximum number of concurrent works can be high and
5441 * concurrency isn't really necessary. Limit it to 1.
5442 */
5443 EXT4_SB(sb)->rsv_conversion_wq =
5444 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5445 if (!EXT4_SB(sb)->rsv_conversion_wq) {
5446 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5447 err = -ENOMEM;
5448 goto failed_mount4;
5449 }
5450
5451 /*
5452 * The jbd2_journal_load will have done any necessary log recovery,
5453 * so we can safely mount the rest of the filesystem now.
5454 */
5455
5456 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5457 if (IS_ERR(root)) {
5458 ext4_msg(sb, KERN_ERR, "get root inode failed");
5459 err = PTR_ERR(root);
5460 root = NULL;
5461 goto failed_mount4;
5462 }
5463 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5464 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5465 iput(root);
5466 err = -EFSCORRUPTED;
5467 goto failed_mount4;
5468 }
5469
5470 sb->s_root = d_make_root(root);
5471 if (!sb->s_root) {
5472 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5473 err = -ENOMEM;
5474 goto failed_mount4;
5475 }
5476
5477 err = ext4_setup_super(sb, es, sb_rdonly(sb));
5478 if (err == -EROFS) {
5479 sb->s_flags |= SB_RDONLY;
5480 } else if (err)
5481 goto failed_mount4a;
5482
5483 ext4_set_resv_clusters(sb);
5484
5485 if (test_opt(sb, BLOCK_VALIDITY)) {
5486 err = ext4_setup_system_zone(sb);
5487 if (err) {
5488 ext4_msg(sb, KERN_ERR, "failed to initialize system "
5489 "zone (%d)", err);
5490 goto failed_mount4a;
5491 }
5492 }
5493 ext4_fc_replay_cleanup(sb);
5494
5495 ext4_ext_init(sb);
5496
5497 /*
5498 * Enable optimize_scan if number of groups is > threshold. This can be
5499 * turned off by passing "mb_optimize_scan=0". This can also be
5500 * turned on forcefully by passing "mb_optimize_scan=1".
5501 */
5502 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5503 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5504 set_opt2(sb, MB_OPTIMIZE_SCAN);
5505 else
5506 clear_opt2(sb, MB_OPTIMIZE_SCAN);
5507 }
5508
5509 err = ext4_mb_init(sb);
5510 if (err) {
5511 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5512 err);
5513 goto failed_mount5;
5514 }
5515
5516 /*
5517 * We can only set up the journal commit callback once
5518 * mballoc is initialized
5519 */
5520 if (sbi->s_journal)
5521 sbi->s_journal->j_commit_callback =
5522 ext4_journal_commit_callback;
5523
5524 err = ext4_percpu_param_init(sbi);
5525 if (err)
5526 goto failed_mount6;
5527
5528 if (ext4_has_feature_flex_bg(sb))
5529 if (!ext4_fill_flex_info(sb)) {
5530 ext4_msg(sb, KERN_ERR,
5531 "unable to initialize "
5532 "flex_bg meta info!");
5533 err = -ENOMEM;
5534 goto failed_mount6;
5535 }
5536
5537 err = ext4_register_li_request(sb, first_not_zeroed);
5538 if (err)
5539 goto failed_mount6;
5540
5541 err = ext4_register_sysfs(sb);
5542 if (err)
5543 goto failed_mount7;
5544
5545 err = ext4_init_orphan_info(sb);
5546 if (err)
5547 goto failed_mount8;
5548#ifdef CONFIG_QUOTA
5549 /* Enable quota usage during mount. */
5550 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5551 err = ext4_enable_quotas(sb);
5552 if (err)
5553 goto failed_mount9;
5554 }
5555#endif /* CONFIG_QUOTA */
5556
5557 /*
5558 * Save the original bdev mapping's wb_err value which could be
5559 * used to detect the metadata async write error.
5560 */
5561 spin_lock_init(&sbi->s_bdev_wb_lock);
5562 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5563 &sbi->s_bdev_wb_err);
5564 sb->s_bdev->bd_super = sb;
5565 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5566 ext4_orphan_cleanup(sb, es);
5567 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5568 /*
5569 * Update the checksum after updating free space/inode counters and
5570 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5571 * checksum in the buffer cache until it is written out and
5572 * e2fsprogs programs trying to open a file system immediately
5573 * after it is mounted can fail.
5574 */
5575 ext4_superblock_csum_set(sb);
5576 if (needs_recovery) {
5577 ext4_msg(sb, KERN_INFO, "recovery complete");
5578 err = ext4_mark_recovery_complete(sb, es);
5579 if (err)
5580 goto failed_mount10;
5581 }
5582
5583 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5584 ext4_msg(sb, KERN_WARNING,
5585 "mounting with \"discard\" option, but the device does not support discard");
5586
5587 if (es->s_error_count)
5588 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5589
5590 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5591 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5592 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5593 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5594 atomic_set(&sbi->s_warning_count, 0);
5595 atomic_set(&sbi->s_msg_count, 0);
5596
5597 return 0;
5598
5599failed_mount10:
5600 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
5601failed_mount9: __maybe_unused
5602 ext4_release_orphan_info(sb);
5603failed_mount8:
5604 ext4_unregister_sysfs(sb);
5605 kobject_put(&sbi->s_kobj);
5606failed_mount7:
5607 ext4_unregister_li_request(sb);
5608failed_mount6:
5609 ext4_mb_release(sb);
5610 ext4_flex_groups_free(sbi);
5611 ext4_percpu_param_destroy(sbi);
5612failed_mount5:
5613 ext4_ext_release(sb);
5614 ext4_release_system_zone(sb);
5615failed_mount4a:
5616 dput(sb->s_root);
5617 sb->s_root = NULL;
5618failed_mount4:
5619 ext4_msg(sb, KERN_ERR, "mount failed");
5620 if (EXT4_SB(sb)->rsv_conversion_wq)
5621 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5622failed_mount_wq:
5623 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5624 sbi->s_ea_inode_cache = NULL;
5625
5626 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5627 sbi->s_ea_block_cache = NULL;
5628
5629 if (sbi->s_journal) {
5630 /* flush s_error_work before journal destroy. */
5631 flush_work(&sbi->s_error_work);
5632 jbd2_journal_destroy(sbi->s_journal);
5633 sbi->s_journal = NULL;
5634 }
5635failed_mount3a:
5636 ext4_es_unregister_shrinker(sbi);
5637failed_mount3:
5638 /* flush s_error_work before sbi destroy */
5639 flush_work(&sbi->s_error_work);
5640 del_timer_sync(&sbi->s_err_report);
5641 ext4_stop_mmpd(sbi);
5642 ext4_group_desc_free(sbi);
5643failed_mount:
5644 if (sbi->s_chksum_driver)
5645 crypto_free_shash(sbi->s_chksum_driver);
5646
5647#if IS_ENABLED(CONFIG_UNICODE)
5648 utf8_unload(sb->s_encoding);
5649#endif
5650
5651#ifdef CONFIG_QUOTA
5652 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++)
5653 kfree(get_qf_name(sb, sbi, i));
5654#endif
5655 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5656 /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5657 brelse(sbi->s_sbh);
5658 ext4_blkdev_remove(sbi);
5659out_fail:
5660 invalidate_bdev(sb->s_bdev);
5661 sb->s_fs_info = NULL;
5662 return err;
5663}
5664
5665static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5666{
5667 struct ext4_fs_context *ctx = fc->fs_private;
5668 struct ext4_sb_info *sbi;
5669 const char *descr;
5670 int ret;
5671
5672 sbi = ext4_alloc_sbi(sb);
5673 if (!sbi)
5674 return -ENOMEM;
5675
5676 fc->s_fs_info = sbi;
5677
5678 /* Cleanup superblock name */
5679 strreplace(sb->s_id, '/', '!');
5680
5681 sbi->s_sb_block = 1; /* Default super block location */
5682 if (ctx->spec & EXT4_SPEC_s_sb_block)
5683 sbi->s_sb_block = ctx->s_sb_block;
5684
5685 ret = __ext4_fill_super(fc, sb);
5686 if (ret < 0)
5687 goto free_sbi;
5688
5689 if (sbi->s_journal) {
5690 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5691 descr = " journalled data mode";
5692 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5693 descr = " ordered data mode";
5694 else
5695 descr = " writeback data mode";
5696 } else
5697 descr = "out journal";
5698
5699 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5700 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. "
5701 "Quota mode: %s.", &sb->s_uuid,
5702 sb_rdonly(sb) ? "ro" : "r/w", descr,
5703 ext4_quota_mode(sb));
5704
5705 /* Update the s_overhead_clusters if necessary */
5706 ext4_update_overhead(sb, false);
5707 return 0;
5708
5709free_sbi:
5710 ext4_free_sbi(sbi);
5711 fc->s_fs_info = NULL;
5712 return ret;
5713}
5714
5715static int ext4_get_tree(struct fs_context *fc)
5716{
5717 return get_tree_bdev(fc, ext4_fill_super);
5718}
5719
5720/*
5721 * Setup any per-fs journal parameters now. We'll do this both on
5722 * initial mount, once the journal has been initialised but before we've
5723 * done any recovery; and again on any subsequent remount.
5724 */
5725static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5726{
5727 struct ext4_sb_info *sbi = EXT4_SB(sb);
5728
5729 journal->j_commit_interval = sbi->s_commit_interval;
5730 journal->j_min_batch_time = sbi->s_min_batch_time;
5731 journal->j_max_batch_time = sbi->s_max_batch_time;
5732 ext4_fc_init(sb, journal);
5733
5734 write_lock(&journal->j_state_lock);
5735 if (test_opt(sb, BARRIER))
5736 journal->j_flags |= JBD2_BARRIER;
5737 else
5738 journal->j_flags &= ~JBD2_BARRIER;
5739 if (test_opt(sb, DATA_ERR_ABORT))
5740 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5741 else
5742 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5743 /*
5744 * Always enable journal cycle record option, letting the journal
5745 * records log transactions continuously between each mount.
5746 */
5747 journal->j_flags |= JBD2_CYCLE_RECORD;
5748 write_unlock(&journal->j_state_lock);
5749}
5750
5751static struct inode *ext4_get_journal_inode(struct super_block *sb,
5752 unsigned int journal_inum)
5753{
5754 struct inode *journal_inode;
5755
5756 /*
5757 * Test for the existence of a valid inode on disk. Bad things
5758 * happen if we iget() an unused inode, as the subsequent iput()
5759 * will try to delete it.
5760 */
5761 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5762 if (IS_ERR(journal_inode)) {
5763 ext4_msg(sb, KERN_ERR, "no journal found");
5764 return NULL;
5765 }
5766 if (!journal_inode->i_nlink) {
5767 make_bad_inode(journal_inode);
5768 iput(journal_inode);
5769 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5770 return NULL;
5771 }
5772
5773 ext4_debug("Journal inode found at %p: %lld bytes\n",
5774 journal_inode, journal_inode->i_size);
5775 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
5776 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5777 iput(journal_inode);
5778 return NULL;
5779 }
5780 return journal_inode;
5781}
5782
5783static int ext4_journal_bmap(journal_t *journal, sector_t *block)
5784{
5785 struct ext4_map_blocks map;
5786 int ret;
5787
5788 if (journal->j_inode == NULL)
5789 return 0;
5790
5791 map.m_lblk = *block;
5792 map.m_len = 1;
5793 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0);
5794 if (ret <= 0) {
5795 ext4_msg(journal->j_inode->i_sb, KERN_CRIT,
5796 "journal bmap failed: block %llu ret %d\n",
5797 *block, ret);
5798 jbd2_journal_abort(journal, ret ? ret : -EIO);
5799 return ret;
5800 }
5801 *block = map.m_pblk;
5802 return 0;
5803}
5804
5805static journal_t *ext4_get_journal(struct super_block *sb,
5806 unsigned int journal_inum)
5807{
5808 struct inode *journal_inode;
5809 journal_t *journal;
5810
5811 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5812 return NULL;
5813
5814 journal_inode = ext4_get_journal_inode(sb, journal_inum);
5815 if (!journal_inode)
5816 return NULL;
5817
5818 journal = jbd2_journal_init_inode(journal_inode);
5819 if (!journal) {
5820 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5821 iput(journal_inode);
5822 return NULL;
5823 }
5824 journal->j_private = sb;
5825 journal->j_bmap = ext4_journal_bmap;
5826 ext4_init_journal_params(sb, journal);
5827 return journal;
5828}
5829
5830static journal_t *ext4_get_dev_journal(struct super_block *sb,
5831 dev_t j_dev)
5832{
5833 struct buffer_head *bh;
5834 journal_t *journal;
5835 ext4_fsblk_t start;
5836 ext4_fsblk_t len;
5837 int hblock, blocksize;
5838 ext4_fsblk_t sb_block;
5839 unsigned long offset;
5840 struct ext4_super_block *es;
5841 struct block_device *bdev;
5842
5843 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5844 return NULL;
5845
5846 bdev = ext4_blkdev_get(j_dev, sb);
5847 if (bdev == NULL)
5848 return NULL;
5849
5850 blocksize = sb->s_blocksize;
5851 hblock = bdev_logical_block_size(bdev);
5852 if (blocksize < hblock) {
5853 ext4_msg(sb, KERN_ERR,
5854 "blocksize too small for journal device");
5855 goto out_bdev;
5856 }
5857
5858 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5859 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5860 set_blocksize(bdev, blocksize);
5861 if (!(bh = __bread(bdev, sb_block, blocksize))) {
5862 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5863 "external journal");
5864 goto out_bdev;
5865 }
5866
5867 es = (struct ext4_super_block *) (bh->b_data + offset);
5868 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5869 !(le32_to_cpu(es->s_feature_incompat) &
5870 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5871 ext4_msg(sb, KERN_ERR, "external journal has "
5872 "bad superblock");
5873 brelse(bh);
5874 goto out_bdev;
5875 }
5876
5877 if ((le32_to_cpu(es->s_feature_ro_compat) &
5878 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5879 es->s_checksum != ext4_superblock_csum(sb, es)) {
5880 ext4_msg(sb, KERN_ERR, "external journal has "
5881 "corrupt superblock");
5882 brelse(bh);
5883 goto out_bdev;
5884 }
5885
5886 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5887 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5888 brelse(bh);
5889 goto out_bdev;
5890 }
5891
5892 len = ext4_blocks_count(es);
5893 start = sb_block + 1;
5894 brelse(bh); /* we're done with the superblock */
5895
5896 journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5897 start, len, blocksize);
5898 if (!journal) {
5899 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5900 goto out_bdev;
5901 }
5902 journal->j_private = sb;
5903 if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5904 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5905 goto out_journal;
5906 }
5907 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5908 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5909 "user (unsupported) - %d",
5910 be32_to_cpu(journal->j_superblock->s_nr_users));
5911 goto out_journal;
5912 }
5913 EXT4_SB(sb)->s_journal_bdev = bdev;
5914 ext4_init_journal_params(sb, journal);
5915 return journal;
5916
5917out_journal:
5918 jbd2_journal_destroy(journal);
5919out_bdev:
5920 ext4_blkdev_put(bdev);
5921 return NULL;
5922}
5923
5924static int ext4_load_journal(struct super_block *sb,
5925 struct ext4_super_block *es,
5926 unsigned long journal_devnum)
5927{
5928 journal_t *journal;
5929 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5930 dev_t journal_dev;
5931 int err = 0;
5932 int really_read_only;
5933 int journal_dev_ro;
5934
5935 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5936 return -EFSCORRUPTED;
5937
5938 if (journal_devnum &&
5939 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5940 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5941 "numbers have changed");
5942 journal_dev = new_decode_dev(journal_devnum);
5943 } else
5944 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5945
5946 if (journal_inum && journal_dev) {
5947 ext4_msg(sb, KERN_ERR,
5948 "filesystem has both journal inode and journal device!");
5949 return -EINVAL;
5950 }
5951
5952 if (journal_inum) {
5953 journal = ext4_get_journal(sb, journal_inum);
5954 if (!journal)
5955 return -EINVAL;
5956 } else {
5957 journal = ext4_get_dev_journal(sb, journal_dev);
5958 if (!journal)
5959 return -EINVAL;
5960 }
5961
5962 journal_dev_ro = bdev_read_only(journal->j_dev);
5963 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5964
5965 if (journal_dev_ro && !sb_rdonly(sb)) {
5966 ext4_msg(sb, KERN_ERR,
5967 "journal device read-only, try mounting with '-o ro'");
5968 err = -EROFS;
5969 goto err_out;
5970 }
5971
5972 /*
5973 * Are we loading a blank journal or performing recovery after a
5974 * crash? For recovery, we need to check in advance whether we
5975 * can get read-write access to the device.
5976 */
5977 if (ext4_has_feature_journal_needs_recovery(sb)) {
5978 if (sb_rdonly(sb)) {
5979 ext4_msg(sb, KERN_INFO, "INFO: recovery "
5980 "required on readonly filesystem");
5981 if (really_read_only) {
5982 ext4_msg(sb, KERN_ERR, "write access "
5983 "unavailable, cannot proceed "
5984 "(try mounting with noload)");
5985 err = -EROFS;
5986 goto err_out;
5987 }
5988 ext4_msg(sb, KERN_INFO, "write access will "
5989 "be enabled during recovery");
5990 }
5991 }
5992
5993 if (!(journal->j_flags & JBD2_BARRIER))
5994 ext4_msg(sb, KERN_INFO, "barriers disabled");
5995
5996 if (!ext4_has_feature_journal_needs_recovery(sb))
5997 err = jbd2_journal_wipe(journal, !really_read_only);
5998 if (!err) {
5999 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
6000 __le16 orig_state;
6001 bool changed = false;
6002
6003 if (save)
6004 memcpy(save, ((char *) es) +
6005 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
6006 err = jbd2_journal_load(journal);
6007 if (save && memcmp(((char *) es) + EXT4_S_ERR_START,
6008 save, EXT4_S_ERR_LEN)) {
6009 memcpy(((char *) es) + EXT4_S_ERR_START,
6010 save, EXT4_S_ERR_LEN);
6011 changed = true;
6012 }
6013 kfree(save);
6014 orig_state = es->s_state;
6015 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state &
6016 EXT4_ERROR_FS);
6017 if (orig_state != es->s_state)
6018 changed = true;
6019 /* Write out restored error information to the superblock */
6020 if (changed && !really_read_only) {
6021 int err2;
6022 err2 = ext4_commit_super(sb);
6023 err = err ? : err2;
6024 }
6025 }
6026
6027 if (err) {
6028 ext4_msg(sb, KERN_ERR, "error loading journal");
6029 goto err_out;
6030 }
6031
6032 EXT4_SB(sb)->s_journal = journal;
6033 err = ext4_clear_journal_err(sb, es);
6034 if (err) {
6035 EXT4_SB(sb)->s_journal = NULL;
6036 jbd2_journal_destroy(journal);
6037 return err;
6038 }
6039
6040 if (!really_read_only && journal_devnum &&
6041 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
6042 es->s_journal_dev = cpu_to_le32(journal_devnum);
6043 ext4_commit_super(sb);
6044 }
6045 if (!really_read_only && journal_inum &&
6046 journal_inum != le32_to_cpu(es->s_journal_inum)) {
6047 es->s_journal_inum = cpu_to_le32(journal_inum);
6048 ext4_commit_super(sb);
6049 }
6050
6051 return 0;
6052
6053err_out:
6054 jbd2_journal_destroy(journal);
6055 return err;
6056}
6057
6058/* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
6059static void ext4_update_super(struct super_block *sb)
6060{
6061 struct ext4_sb_info *sbi = EXT4_SB(sb);
6062 struct ext4_super_block *es = sbi->s_es;
6063 struct buffer_head *sbh = sbi->s_sbh;
6064
6065 lock_buffer(sbh);
6066 /*
6067 * If the file system is mounted read-only, don't update the
6068 * superblock write time. This avoids updating the superblock
6069 * write time when we are mounting the root file system
6070 * read/only but we need to replay the journal; at that point,
6071 * for people who are east of GMT and who make their clock
6072 * tick in localtime for Windows bug-for-bug compatibility,
6073 * the clock is set in the future, and this will cause e2fsck
6074 * to complain and force a full file system check.
6075 */
6076 if (!(sb->s_flags & SB_RDONLY))
6077 ext4_update_tstamp(es, s_wtime);
6078 es->s_kbytes_written =
6079 cpu_to_le64(sbi->s_kbytes_written +
6080 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
6081 sbi->s_sectors_written_start) >> 1));
6082 if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
6083 ext4_free_blocks_count_set(es,
6084 EXT4_C2B(sbi, percpu_counter_sum_positive(
6085 &sbi->s_freeclusters_counter)));
6086 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
6087 es->s_free_inodes_count =
6088 cpu_to_le32(percpu_counter_sum_positive(
6089 &sbi->s_freeinodes_counter));
6090 /* Copy error information to the on-disk superblock */
6091 spin_lock(&sbi->s_error_lock);
6092 if (sbi->s_add_error_count > 0) {
6093 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6094 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
6095 __ext4_update_tstamp(&es->s_first_error_time,
6096 &es->s_first_error_time_hi,
6097 sbi->s_first_error_time);
6098 strncpy(es->s_first_error_func, sbi->s_first_error_func,
6099 sizeof(es->s_first_error_func));
6100 es->s_first_error_line =
6101 cpu_to_le32(sbi->s_first_error_line);
6102 es->s_first_error_ino =
6103 cpu_to_le32(sbi->s_first_error_ino);
6104 es->s_first_error_block =
6105 cpu_to_le64(sbi->s_first_error_block);
6106 es->s_first_error_errcode =
6107 ext4_errno_to_code(sbi->s_first_error_code);
6108 }
6109 __ext4_update_tstamp(&es->s_last_error_time,
6110 &es->s_last_error_time_hi,
6111 sbi->s_last_error_time);
6112 strncpy(es->s_last_error_func, sbi->s_last_error_func,
6113 sizeof(es->s_last_error_func));
6114 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
6115 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
6116 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
6117 es->s_last_error_errcode =
6118 ext4_errno_to_code(sbi->s_last_error_code);
6119 /*
6120 * Start the daily error reporting function if it hasn't been
6121 * started already
6122 */
6123 if (!es->s_error_count)
6124 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
6125 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
6126 sbi->s_add_error_count = 0;
6127 }
6128 spin_unlock(&sbi->s_error_lock);
6129
6130 ext4_superblock_csum_set(sb);
6131 unlock_buffer(sbh);
6132}
6133
6134static int ext4_commit_super(struct super_block *sb)
6135{
6136 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6137
6138 if (!sbh)
6139 return -EINVAL;
6140 if (block_device_ejected(sb))
6141 return -ENODEV;
6142
6143 ext4_update_super(sb);
6144
6145 lock_buffer(sbh);
6146 /* Buffer got discarded which means block device got invalidated */
6147 if (!buffer_mapped(sbh)) {
6148 unlock_buffer(sbh);
6149 return -EIO;
6150 }
6151
6152 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6153 /*
6154 * Oh, dear. A previous attempt to write the
6155 * superblock failed. This could happen because the
6156 * USB device was yanked out. Or it could happen to
6157 * be a transient write error and maybe the block will
6158 * be remapped. Nothing we can do but to retry the
6159 * write and hope for the best.
6160 */
6161 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6162 "superblock detected");
6163 clear_buffer_write_io_error(sbh);
6164 set_buffer_uptodate(sbh);
6165 }
6166 get_bh(sbh);
6167 /* Clear potential dirty bit if it was journalled update */
6168 clear_buffer_dirty(sbh);
6169 sbh->b_end_io = end_buffer_write_sync;
6170 submit_bh(REQ_OP_WRITE | REQ_SYNC |
6171 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
6172 wait_on_buffer(sbh);
6173 if (buffer_write_io_error(sbh)) {
6174 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6175 "superblock");
6176 clear_buffer_write_io_error(sbh);
6177 set_buffer_uptodate(sbh);
6178 return -EIO;
6179 }
6180 return 0;
6181}
6182
6183/*
6184 * Have we just finished recovery? If so, and if we are mounting (or
6185 * remounting) the filesystem readonly, then we will end up with a
6186 * consistent fs on disk. Record that fact.
6187 */
6188static int ext4_mark_recovery_complete(struct super_block *sb,
6189 struct ext4_super_block *es)
6190{
6191 int err;
6192 journal_t *journal = EXT4_SB(sb)->s_journal;
6193
6194 if (!ext4_has_feature_journal(sb)) {
6195 if (journal != NULL) {
6196 ext4_error(sb, "Journal got removed while the fs was "
6197 "mounted!");
6198 return -EFSCORRUPTED;
6199 }
6200 return 0;
6201 }
6202 jbd2_journal_lock_updates(journal);
6203 err = jbd2_journal_flush(journal, 0);
6204 if (err < 0)
6205 goto out;
6206
6207 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6208 ext4_has_feature_orphan_present(sb))) {
6209 if (!ext4_orphan_file_empty(sb)) {
6210 ext4_error(sb, "Orphan file not empty on read-only fs.");
6211 err = -EFSCORRUPTED;
6212 goto out;
6213 }
6214 ext4_clear_feature_journal_needs_recovery(sb);
6215 ext4_clear_feature_orphan_present(sb);
6216 ext4_commit_super(sb);
6217 }
6218out:
6219 jbd2_journal_unlock_updates(journal);
6220 return err;
6221}
6222
6223/*
6224 * If we are mounting (or read-write remounting) a filesystem whose journal
6225 * has recorded an error from a previous lifetime, move that error to the
6226 * main filesystem now.
6227 */
6228static int ext4_clear_journal_err(struct super_block *sb,
6229 struct ext4_super_block *es)
6230{
6231 journal_t *journal;
6232 int j_errno;
6233 const char *errstr;
6234
6235 if (!ext4_has_feature_journal(sb)) {
6236 ext4_error(sb, "Journal got removed while the fs was mounted!");
6237 return -EFSCORRUPTED;
6238 }
6239
6240 journal = EXT4_SB(sb)->s_journal;
6241
6242 /*
6243 * Now check for any error status which may have been recorded in the
6244 * journal by a prior ext4_error() or ext4_abort()
6245 */
6246
6247 j_errno = jbd2_journal_errno(journal);
6248 if (j_errno) {
6249 char nbuf[16];
6250
6251 errstr = ext4_decode_error(sb, j_errno, nbuf);
6252 ext4_warning(sb, "Filesystem error recorded "
6253 "from previous mount: %s", errstr);
6254
6255 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6256 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6257 j_errno = ext4_commit_super(sb);
6258 if (j_errno)
6259 return j_errno;
6260 ext4_warning(sb, "Marked fs in need of filesystem check.");
6261
6262 jbd2_journal_clear_err(journal);
6263 jbd2_journal_update_sb_errno(journal);
6264 }
6265 return 0;
6266}
6267
6268/*
6269 * Force the running and committing transactions to commit,
6270 * and wait on the commit.
6271 */
6272int ext4_force_commit(struct super_block *sb)
6273{
6274 journal_t *journal;
6275
6276 if (sb_rdonly(sb))
6277 return 0;
6278
6279 journal = EXT4_SB(sb)->s_journal;
6280 return ext4_journal_force_commit(journal);
6281}
6282
6283static int ext4_sync_fs(struct super_block *sb, int wait)
6284{
6285 int ret = 0;
6286 tid_t target;
6287 bool needs_barrier = false;
6288 struct ext4_sb_info *sbi = EXT4_SB(sb);
6289
6290 if (unlikely(ext4_forced_shutdown(sbi)))
6291 return 0;
6292
6293 trace_ext4_sync_fs(sb, wait);
6294 flush_workqueue(sbi->rsv_conversion_wq);
6295 /*
6296 * Writeback quota in non-journalled quota case - journalled quota has
6297 * no dirty dquots
6298 */
6299 dquot_writeback_dquots(sb, -1);
6300 /*
6301 * Data writeback is possible w/o journal transaction, so barrier must
6302 * being sent at the end of the function. But we can skip it if
6303 * transaction_commit will do it for us.
6304 */
6305 if (sbi->s_journal) {
6306 target = jbd2_get_latest_transaction(sbi->s_journal);
6307 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6308 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6309 needs_barrier = true;
6310
6311 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6312 if (wait)
6313 ret = jbd2_log_wait_commit(sbi->s_journal,
6314 target);
6315 }
6316 } else if (wait && test_opt(sb, BARRIER))
6317 needs_barrier = true;
6318 if (needs_barrier) {
6319 int err;
6320 err = blkdev_issue_flush(sb->s_bdev);
6321 if (!ret)
6322 ret = err;
6323 }
6324
6325 return ret;
6326}
6327
6328/*
6329 * LVM calls this function before a (read-only) snapshot is created. This
6330 * gives us a chance to flush the journal completely and mark the fs clean.
6331 *
6332 * Note that only this function cannot bring a filesystem to be in a clean
6333 * state independently. It relies on upper layer to stop all data & metadata
6334 * modifications.
6335 */
6336static int ext4_freeze(struct super_block *sb)
6337{
6338 int error = 0;
6339 journal_t *journal;
6340
6341 if (sb_rdonly(sb))
6342 return 0;
6343
6344 journal = EXT4_SB(sb)->s_journal;
6345
6346 if (journal) {
6347 /* Now we set up the journal barrier. */
6348 jbd2_journal_lock_updates(journal);
6349
6350 /*
6351 * Don't clear the needs_recovery flag if we failed to
6352 * flush the journal.
6353 */
6354 error = jbd2_journal_flush(journal, 0);
6355 if (error < 0)
6356 goto out;
6357
6358 /* Journal blocked and flushed, clear needs_recovery flag. */
6359 ext4_clear_feature_journal_needs_recovery(sb);
6360 if (ext4_orphan_file_empty(sb))
6361 ext4_clear_feature_orphan_present(sb);
6362 }
6363
6364 error = ext4_commit_super(sb);
6365out:
6366 if (journal)
6367 /* we rely on upper layer to stop further updates */
6368 jbd2_journal_unlock_updates(journal);
6369 return error;
6370}
6371
6372/*
6373 * Called by LVM after the snapshot is done. We need to reset the RECOVER
6374 * flag here, even though the filesystem is not technically dirty yet.
6375 */
6376static int ext4_unfreeze(struct super_block *sb)
6377{
6378 if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
6379 return 0;
6380
6381 if (EXT4_SB(sb)->s_journal) {
6382 /* Reset the needs_recovery flag before the fs is unlocked. */
6383 ext4_set_feature_journal_needs_recovery(sb);
6384 if (ext4_has_feature_orphan_file(sb))
6385 ext4_set_feature_orphan_present(sb);
6386 }
6387
6388 ext4_commit_super(sb);
6389 return 0;
6390}
6391
6392/*
6393 * Structure to save mount options for ext4_remount's benefit
6394 */
6395struct ext4_mount_options {
6396 unsigned long s_mount_opt;
6397 unsigned long s_mount_opt2;
6398 kuid_t s_resuid;
6399 kgid_t s_resgid;
6400 unsigned long s_commit_interval;
6401 u32 s_min_batch_time, s_max_batch_time;
6402#ifdef CONFIG_QUOTA
6403 int s_jquota_fmt;
6404 char *s_qf_names[EXT4_MAXQUOTAS];
6405#endif
6406};
6407
6408static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6409{
6410 struct ext4_fs_context *ctx = fc->fs_private;
6411 struct ext4_super_block *es;
6412 struct ext4_sb_info *sbi = EXT4_SB(sb);
6413 unsigned long old_sb_flags;
6414 struct ext4_mount_options old_opts;
6415 ext4_group_t g;
6416 int err = 0;
6417#ifdef CONFIG_QUOTA
6418 int enable_quota = 0;
6419 int i, j;
6420 char *to_free[EXT4_MAXQUOTAS];
6421#endif
6422
6423
6424 /* Store the original options */
6425 old_sb_flags = sb->s_flags;
6426 old_opts.s_mount_opt = sbi->s_mount_opt;
6427 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6428 old_opts.s_resuid = sbi->s_resuid;
6429 old_opts.s_resgid = sbi->s_resgid;
6430 old_opts.s_commit_interval = sbi->s_commit_interval;
6431 old_opts.s_min_batch_time = sbi->s_min_batch_time;
6432 old_opts.s_max_batch_time = sbi->s_max_batch_time;
6433#ifdef CONFIG_QUOTA
6434 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6435 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6436 if (sbi->s_qf_names[i]) {
6437 char *qf_name = get_qf_name(sb, sbi, i);
6438
6439 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6440 if (!old_opts.s_qf_names[i]) {
6441 for (j = 0; j < i; j++)
6442 kfree(old_opts.s_qf_names[j]);
6443 return -ENOMEM;
6444 }
6445 } else
6446 old_opts.s_qf_names[i] = NULL;
6447#endif
6448 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6449 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6450 ctx->journal_ioprio =
6451 sbi->s_journal->j_task->io_context->ioprio;
6452 else
6453 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6454
6455 }
6456
6457 ext4_apply_options(fc, sb);
6458
6459 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6460 test_opt(sb, JOURNAL_CHECKSUM)) {
6461 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6462 "during remount not supported; ignoring");
6463 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6464 }
6465
6466 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6467 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6468 ext4_msg(sb, KERN_ERR, "can't mount with "
6469 "both data=journal and delalloc");
6470 err = -EINVAL;
6471 goto restore_opts;
6472 }
6473 if (test_opt(sb, DIOREAD_NOLOCK)) {
6474 ext4_msg(sb, KERN_ERR, "can't mount with "
6475 "both data=journal and dioread_nolock");
6476 err = -EINVAL;
6477 goto restore_opts;
6478 }
6479 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6480 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6481 ext4_msg(sb, KERN_ERR, "can't mount with "
6482 "journal_async_commit in data=ordered mode");
6483 err = -EINVAL;
6484 goto restore_opts;
6485 }
6486 }
6487
6488 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6489 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6490 err = -EINVAL;
6491 goto restore_opts;
6492 }
6493
6494 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
6495 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6496
6497 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6498 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6499
6500 es = sbi->s_es;
6501
6502 if (sbi->s_journal) {
6503 ext4_init_journal_params(sb, sbi->s_journal);
6504 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6505 }
6506
6507 /* Flush outstanding errors before changing fs state */
6508 flush_work(&sbi->s_error_work);
6509
6510 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6511 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
6512 err = -EROFS;
6513 goto restore_opts;
6514 }
6515
6516 if (fc->sb_flags & SB_RDONLY) {
6517 err = sync_filesystem(sb);
6518 if (err < 0)
6519 goto restore_opts;
6520 err = dquot_suspend(sb, -1);
6521 if (err < 0)
6522 goto restore_opts;
6523
6524 /*
6525 * First of all, the unconditional stuff we have to do
6526 * to disable replay of the journal when we next remount
6527 */
6528 sb->s_flags |= SB_RDONLY;
6529
6530 /*
6531 * OK, test if we are remounting a valid rw partition
6532 * readonly, and if so set the rdonly flag and then
6533 * mark the partition as valid again.
6534 */
6535 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6536 (sbi->s_mount_state & EXT4_VALID_FS))
6537 es->s_state = cpu_to_le16(sbi->s_mount_state);
6538
6539 if (sbi->s_journal) {
6540 /*
6541 * We let remount-ro finish even if marking fs
6542 * as clean failed...
6543 */
6544 ext4_mark_recovery_complete(sb, es);
6545 }
6546 } else {
6547 /* Make sure we can mount this feature set readwrite */
6548 if (ext4_has_feature_readonly(sb) ||
6549 !ext4_feature_set_ok(sb, 0)) {
6550 err = -EROFS;
6551 goto restore_opts;
6552 }
6553 /*
6554 * Make sure the group descriptor checksums
6555 * are sane. If they aren't, refuse to remount r/w.
6556 */
6557 for (g = 0; g < sbi->s_groups_count; g++) {
6558 struct ext4_group_desc *gdp =
6559 ext4_get_group_desc(sb, g, NULL);
6560
6561 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6562 ext4_msg(sb, KERN_ERR,
6563 "ext4_remount: Checksum for group %u failed (%u!=%u)",
6564 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6565 le16_to_cpu(gdp->bg_checksum));
6566 err = -EFSBADCRC;
6567 goto restore_opts;
6568 }
6569 }
6570
6571 /*
6572 * If we have an unprocessed orphan list hanging
6573 * around from a previously readonly bdev mount,
6574 * require a full umount/remount for now.
6575 */
6576 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6577 ext4_msg(sb, KERN_WARNING, "Couldn't "
6578 "remount RDWR because of unprocessed "
6579 "orphan inode list. Please "
6580 "umount/remount instead");
6581 err = -EINVAL;
6582 goto restore_opts;
6583 }
6584
6585 /*
6586 * Mounting a RDONLY partition read-write, so reread
6587 * and store the current valid flag. (It may have
6588 * been changed by e2fsck since we originally mounted
6589 * the partition.)
6590 */
6591 if (sbi->s_journal) {
6592 err = ext4_clear_journal_err(sb, es);
6593 if (err)
6594 goto restore_opts;
6595 }
6596 sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6597 ~EXT4_FC_REPLAY);
6598
6599 err = ext4_setup_super(sb, es, 0);
6600 if (err)
6601 goto restore_opts;
6602
6603 sb->s_flags &= ~SB_RDONLY;
6604 if (ext4_has_feature_mmp(sb)) {
6605 err = ext4_multi_mount_protect(sb,
6606 le64_to_cpu(es->s_mmp_block));
6607 if (err)
6608 goto restore_opts;
6609 }
6610#ifdef CONFIG_QUOTA
6611 enable_quota = 1;
6612#endif
6613 }
6614 }
6615
6616 /*
6617 * Handle creation of system zone data early because it can fail.
6618 * Releasing of existing data is done when we are sure remount will
6619 * succeed.
6620 */
6621 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6622 err = ext4_setup_system_zone(sb);
6623 if (err)
6624 goto restore_opts;
6625 }
6626
6627 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6628 err = ext4_commit_super(sb);
6629 if (err)
6630 goto restore_opts;
6631 }
6632
6633#ifdef CONFIG_QUOTA
6634 if (enable_quota) {
6635 if (sb_any_quota_suspended(sb))
6636 dquot_resume(sb, -1);
6637 else if (ext4_has_feature_quota(sb)) {
6638 err = ext4_enable_quotas(sb);
6639 if (err)
6640 goto restore_opts;
6641 }
6642 }
6643 /* Release old quota file names */
6644 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6645 kfree(old_opts.s_qf_names[i]);
6646#endif
6647 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6648 ext4_release_system_zone(sb);
6649
6650 /*
6651 * Reinitialize lazy itable initialization thread based on
6652 * current settings
6653 */
6654 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6655 ext4_unregister_li_request(sb);
6656 else {
6657 ext4_group_t first_not_zeroed;
6658 first_not_zeroed = ext4_has_uninit_itable(sb);
6659 ext4_register_li_request(sb, first_not_zeroed);
6660 }
6661
6662 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6663 ext4_stop_mmpd(sbi);
6664
6665 return 0;
6666
6667restore_opts:
6668 /*
6669 * If there was a failing r/w to ro transition, we may need to
6670 * re-enable quota
6671 */
6672 if ((sb->s_flags & SB_RDONLY) && !(old_sb_flags & SB_RDONLY) &&
6673 sb_any_quota_suspended(sb))
6674 dquot_resume(sb, -1);
6675 sb->s_flags = old_sb_flags;
6676 sbi->s_mount_opt = old_opts.s_mount_opt;
6677 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6678 sbi->s_resuid = old_opts.s_resuid;
6679 sbi->s_resgid = old_opts.s_resgid;
6680 sbi->s_commit_interval = old_opts.s_commit_interval;
6681 sbi->s_min_batch_time = old_opts.s_min_batch_time;
6682 sbi->s_max_batch_time = old_opts.s_max_batch_time;
6683 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6684 ext4_release_system_zone(sb);
6685#ifdef CONFIG_QUOTA
6686 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6687 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6688 to_free[i] = get_qf_name(sb, sbi, i);
6689 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6690 }
6691 synchronize_rcu();
6692 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6693 kfree(to_free[i]);
6694#endif
6695 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6696 ext4_stop_mmpd(sbi);
6697 return err;
6698}
6699
6700static int ext4_reconfigure(struct fs_context *fc)
6701{
6702 struct super_block *sb = fc->root->d_sb;
6703 int ret;
6704
6705 fc->s_fs_info = EXT4_SB(sb);
6706
6707 ret = ext4_check_opt_consistency(fc, sb);
6708 if (ret < 0)
6709 return ret;
6710
6711 ret = __ext4_remount(fc, sb);
6712 if (ret < 0)
6713 return ret;
6714
6715 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.",
6716 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w",
6717 ext4_quota_mode(sb));
6718
6719 return 0;
6720}
6721
6722#ifdef CONFIG_QUOTA
6723static int ext4_statfs_project(struct super_block *sb,
6724 kprojid_t projid, struct kstatfs *buf)
6725{
6726 struct kqid qid;
6727 struct dquot *dquot;
6728 u64 limit;
6729 u64 curblock;
6730
6731 qid = make_kqid_projid(projid);
6732 dquot = dqget(sb, qid);
6733 if (IS_ERR(dquot))
6734 return PTR_ERR(dquot);
6735 spin_lock(&dquot->dq_dqb_lock);
6736
6737 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6738 dquot->dq_dqb.dqb_bhardlimit);
6739 limit >>= sb->s_blocksize_bits;
6740
6741 if (limit && buf->f_blocks > limit) {
6742 curblock = (dquot->dq_dqb.dqb_curspace +
6743 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6744 buf->f_blocks = limit;
6745 buf->f_bfree = buf->f_bavail =
6746 (buf->f_blocks > curblock) ?
6747 (buf->f_blocks - curblock) : 0;
6748 }
6749
6750 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6751 dquot->dq_dqb.dqb_ihardlimit);
6752 if (limit && buf->f_files > limit) {
6753 buf->f_files = limit;
6754 buf->f_ffree =
6755 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6756 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6757 }
6758
6759 spin_unlock(&dquot->dq_dqb_lock);
6760 dqput(dquot);
6761 return 0;
6762}
6763#endif
6764
6765static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6766{
6767 struct super_block *sb = dentry->d_sb;
6768 struct ext4_sb_info *sbi = EXT4_SB(sb);
6769 struct ext4_super_block *es = sbi->s_es;
6770 ext4_fsblk_t overhead = 0, resv_blocks;
6771 s64 bfree;
6772 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6773
6774 if (!test_opt(sb, MINIX_DF))
6775 overhead = sbi->s_overhead;
6776
6777 buf->f_type = EXT4_SUPER_MAGIC;
6778 buf->f_bsize = sb->s_blocksize;
6779 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6780 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6781 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6782 /* prevent underflow in case that few free space is available */
6783 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6784 buf->f_bavail = buf->f_bfree -
6785 (ext4_r_blocks_count(es) + resv_blocks);
6786 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6787 buf->f_bavail = 0;
6788 buf->f_files = le32_to_cpu(es->s_inodes_count);
6789 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6790 buf->f_namelen = EXT4_NAME_LEN;
6791 buf->f_fsid = uuid_to_fsid(es->s_uuid);
6792
6793#ifdef CONFIG_QUOTA
6794 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6795 sb_has_quota_limits_enabled(sb, PRJQUOTA))
6796 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6797#endif
6798 return 0;
6799}
6800
6801
6802#ifdef CONFIG_QUOTA
6803
6804/*
6805 * Helper functions so that transaction is started before we acquire dqio_sem
6806 * to keep correct lock ordering of transaction > dqio_sem
6807 */
6808static inline struct inode *dquot_to_inode(struct dquot *dquot)
6809{
6810 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6811}
6812
6813static int ext4_write_dquot(struct dquot *dquot)
6814{
6815 int ret, err;
6816 handle_t *handle;
6817 struct inode *inode;
6818
6819 inode = dquot_to_inode(dquot);
6820 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6821 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6822 if (IS_ERR(handle))
6823 return PTR_ERR(handle);
6824 ret = dquot_commit(dquot);
6825 err = ext4_journal_stop(handle);
6826 if (!ret)
6827 ret = err;
6828 return ret;
6829}
6830
6831static int ext4_acquire_dquot(struct dquot *dquot)
6832{
6833 int ret, err;
6834 handle_t *handle;
6835
6836 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6837 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6838 if (IS_ERR(handle))
6839 return PTR_ERR(handle);
6840 ret = dquot_acquire(dquot);
6841 err = ext4_journal_stop(handle);
6842 if (!ret)
6843 ret = err;
6844 return ret;
6845}
6846
6847static int ext4_release_dquot(struct dquot *dquot)
6848{
6849 int ret, err;
6850 handle_t *handle;
6851
6852 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6853 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6854 if (IS_ERR(handle)) {
6855 /* Release dquot anyway to avoid endless cycle in dqput() */
6856 dquot_release(dquot);
6857 return PTR_ERR(handle);
6858 }
6859 ret = dquot_release(dquot);
6860 err = ext4_journal_stop(handle);
6861 if (!ret)
6862 ret = err;
6863 return ret;
6864}
6865
6866static int ext4_mark_dquot_dirty(struct dquot *dquot)
6867{
6868 struct super_block *sb = dquot->dq_sb;
6869
6870 if (ext4_is_quota_journalled(sb)) {
6871 dquot_mark_dquot_dirty(dquot);
6872 return ext4_write_dquot(dquot);
6873 } else {
6874 return dquot_mark_dquot_dirty(dquot);
6875 }
6876}
6877
6878static int ext4_write_info(struct super_block *sb, int type)
6879{
6880 int ret, err;
6881 handle_t *handle;
6882
6883 /* Data block + inode block */
6884 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
6885 if (IS_ERR(handle))
6886 return PTR_ERR(handle);
6887 ret = dquot_commit_info(sb, type);
6888 err = ext4_journal_stop(handle);
6889 if (!ret)
6890 ret = err;
6891 return ret;
6892}
6893
6894static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6895{
6896 struct ext4_inode_info *ei = EXT4_I(inode);
6897
6898 /* The first argument of lockdep_set_subclass has to be
6899 * *exactly* the same as the argument to init_rwsem() --- in
6900 * this case, in init_once() --- or lockdep gets unhappy
6901 * because the name of the lock is set using the
6902 * stringification of the argument to init_rwsem().
6903 */
6904 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
6905 lockdep_set_subclass(&ei->i_data_sem, subclass);
6906}
6907
6908/*
6909 * Standard function to be called on quota_on
6910 */
6911static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6912 const struct path *path)
6913{
6914 int err;
6915
6916 if (!test_opt(sb, QUOTA))
6917 return -EINVAL;
6918
6919 /* Quotafile not on the same filesystem? */
6920 if (path->dentry->d_sb != sb)
6921 return -EXDEV;
6922
6923 /* Quota already enabled for this file? */
6924 if (IS_NOQUOTA(d_inode(path->dentry)))
6925 return -EBUSY;
6926
6927 /* Journaling quota? */
6928 if (EXT4_SB(sb)->s_qf_names[type]) {
6929 /* Quotafile not in fs root? */
6930 if (path->dentry->d_parent != sb->s_root)
6931 ext4_msg(sb, KERN_WARNING,
6932 "Quota file not on filesystem root. "
6933 "Journaled quota will not work");
6934 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6935 } else {
6936 /*
6937 * Clear the flag just in case mount options changed since
6938 * last time.
6939 */
6940 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6941 }
6942
6943 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6944 err = dquot_quota_on(sb, type, format_id, path);
6945 if (!err) {
6946 struct inode *inode = d_inode(path->dentry);
6947 handle_t *handle;
6948
6949 /*
6950 * Set inode flags to prevent userspace from messing with quota
6951 * files. If this fails, we return success anyway since quotas
6952 * are already enabled and this is not a hard failure.
6953 */
6954 inode_lock(inode);
6955 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6956 if (IS_ERR(handle))
6957 goto unlock_inode;
6958 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6959 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6960 S_NOATIME | S_IMMUTABLE);
6961 err = ext4_mark_inode_dirty(handle, inode);
6962 ext4_journal_stop(handle);
6963 unlock_inode:
6964 inode_unlock(inode);
6965 if (err)
6966 dquot_quota_off(sb, type);
6967 }
6968 if (err)
6969 lockdep_set_quota_inode(path->dentry->d_inode,
6970 I_DATA_SEM_NORMAL);
6971 return err;
6972}
6973
6974static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
6975{
6976 switch (type) {
6977 case USRQUOTA:
6978 return qf_inum == EXT4_USR_QUOTA_INO;
6979 case GRPQUOTA:
6980 return qf_inum == EXT4_GRP_QUOTA_INO;
6981 case PRJQUOTA:
6982 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
6983 default:
6984 BUG();
6985 }
6986}
6987
6988static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6989 unsigned int flags)
6990{
6991 int err;
6992 struct inode *qf_inode;
6993 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6994 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6995 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6996 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6997 };
6998
6999 BUG_ON(!ext4_has_feature_quota(sb));
7000
7001 if (!qf_inums[type])
7002 return -EPERM;
7003
7004 if (!ext4_check_quota_inum(type, qf_inums[type])) {
7005 ext4_error(sb, "Bad quota inum: %lu, type: %d",
7006 qf_inums[type], type);
7007 return -EUCLEAN;
7008 }
7009
7010 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
7011 if (IS_ERR(qf_inode)) {
7012 ext4_error(sb, "Bad quota inode: %lu, type: %d",
7013 qf_inums[type], type);
7014 return PTR_ERR(qf_inode);
7015 }
7016
7017 /* Don't account quota for quota files to avoid recursion */
7018 qf_inode->i_flags |= S_NOQUOTA;
7019 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
7020 err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
7021 if (err)
7022 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
7023 iput(qf_inode);
7024
7025 return err;
7026}
7027
7028/* Enable usage tracking for all quota types. */
7029int ext4_enable_quotas(struct super_block *sb)
7030{
7031 int type, err = 0;
7032 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
7033 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
7034 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
7035 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
7036 };
7037 bool quota_mopt[EXT4_MAXQUOTAS] = {
7038 test_opt(sb, USRQUOTA),
7039 test_opt(sb, GRPQUOTA),
7040 test_opt(sb, PRJQUOTA),
7041 };
7042
7043 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
7044 for (type = 0; type < EXT4_MAXQUOTAS; type++) {
7045 if (qf_inums[type]) {
7046 err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
7047 DQUOT_USAGE_ENABLED |
7048 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
7049 if (err) {
7050 ext4_warning(sb,
7051 "Failed to enable quota tracking "
7052 "(type=%d, err=%d, ino=%lu). "
7053 "Please run e2fsck to fix.", type,
7054 err, qf_inums[type]);
7055
7056 ext4_quotas_off(sb, type);
7057 return err;
7058 }
7059 }
7060 }
7061 return 0;
7062}
7063
7064static int ext4_quota_off(struct super_block *sb, int type)
7065{
7066 struct inode *inode = sb_dqopt(sb)->files[type];
7067 handle_t *handle;
7068 int err;
7069
7070 /* Force all delayed allocation blocks to be allocated.
7071 * Caller already holds s_umount sem */
7072 if (test_opt(sb, DELALLOC))
7073 sync_filesystem(sb);
7074
7075 if (!inode || !igrab(inode))
7076 goto out;
7077
7078 err = dquot_quota_off(sb, type);
7079 if (err || ext4_has_feature_quota(sb))
7080 goto out_put;
7081
7082 inode_lock(inode);
7083 /*
7084 * Update modification times of quota files when userspace can
7085 * start looking at them. If we fail, we return success anyway since
7086 * this is not a hard failure and quotas are already disabled.
7087 */
7088 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7089 if (IS_ERR(handle)) {
7090 err = PTR_ERR(handle);
7091 goto out_unlock;
7092 }
7093 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
7094 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
7095 inode->i_mtime = inode->i_ctime = current_time(inode);
7096 err = ext4_mark_inode_dirty(handle, inode);
7097 ext4_journal_stop(handle);
7098out_unlock:
7099 inode_unlock(inode);
7100out_put:
7101 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
7102 iput(inode);
7103 return err;
7104out:
7105 return dquot_quota_off(sb, type);
7106}
7107
7108/* Read data from quotafile - avoid pagecache and such because we cannot afford
7109 * acquiring the locks... As quota files are never truncated and quota code
7110 * itself serializes the operations (and no one else should touch the files)
7111 * we don't have to be afraid of races */
7112static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
7113 size_t len, loff_t off)
7114{
7115 struct inode *inode = sb_dqopt(sb)->files[type];
7116 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7117 int offset = off & (sb->s_blocksize - 1);
7118 int tocopy;
7119 size_t toread;
7120 struct buffer_head *bh;
7121 loff_t i_size = i_size_read(inode);
7122
7123 if (off > i_size)
7124 return 0;
7125 if (off+len > i_size)
7126 len = i_size-off;
7127 toread = len;
7128 while (toread > 0) {
7129 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread);
7130 bh = ext4_bread(NULL, inode, blk, 0);
7131 if (IS_ERR(bh))
7132 return PTR_ERR(bh);
7133 if (!bh) /* A hole? */
7134 memset(data, 0, tocopy);
7135 else
7136 memcpy(data, bh->b_data+offset, tocopy);
7137 brelse(bh);
7138 offset = 0;
7139 toread -= tocopy;
7140 data += tocopy;
7141 blk++;
7142 }
7143 return len;
7144}
7145
7146/* Write to quotafile (we know the transaction is already started and has
7147 * enough credits) */
7148static ssize_t ext4_quota_write(struct super_block *sb, int type,
7149 const char *data, size_t len, loff_t off)
7150{
7151 struct inode *inode = sb_dqopt(sb)->files[type];
7152 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7153 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7154 int retries = 0;
7155 struct buffer_head *bh;
7156 handle_t *handle = journal_current_handle();
7157
7158 if (!handle) {
7159 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7160 " cancelled because transaction is not started",
7161 (unsigned long long)off, (unsigned long long)len);
7162 return -EIO;
7163 }
7164 /*
7165 * Since we account only one data block in transaction credits,
7166 * then it is impossible to cross a block boundary.
7167 */
7168 if (sb->s_blocksize - offset < len) {
7169 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7170 " cancelled because not block aligned",
7171 (unsigned long long)off, (unsigned long long)len);
7172 return -EIO;
7173 }
7174
7175 do {
7176 bh = ext4_bread(handle, inode, blk,
7177 EXT4_GET_BLOCKS_CREATE |
7178 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7179 } while (PTR_ERR(bh) == -ENOSPC &&
7180 ext4_should_retry_alloc(inode->i_sb, &retries));
7181 if (IS_ERR(bh))
7182 return PTR_ERR(bh);
7183 if (!bh)
7184 goto out;
7185 BUFFER_TRACE(bh, "get write access");
7186 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7187 if (err) {
7188 brelse(bh);
7189 return err;
7190 }
7191 lock_buffer(bh);
7192 memcpy(bh->b_data+offset, data, len);
7193 flush_dcache_page(bh->b_page);
7194 unlock_buffer(bh);
7195 err = ext4_handle_dirty_metadata(handle, NULL, bh);
7196 brelse(bh);
7197out:
7198 if (inode->i_size < off + len) {
7199 i_size_write(inode, off + len);
7200 EXT4_I(inode)->i_disksize = inode->i_size;
7201 err2 = ext4_mark_inode_dirty(handle, inode);
7202 if (unlikely(err2 && !err))
7203 err = err2;
7204 }
7205 return err ? err : len;
7206}
7207#endif
7208
7209#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7210static inline void register_as_ext2(void)
7211{
7212 int err = register_filesystem(&ext2_fs_type);
7213 if (err)
7214 printk(KERN_WARNING
7215 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7216}
7217
7218static inline void unregister_as_ext2(void)
7219{
7220 unregister_filesystem(&ext2_fs_type);
7221}
7222
7223static inline int ext2_feature_set_ok(struct super_block *sb)
7224{
7225 if (ext4_has_unknown_ext2_incompat_features(sb))
7226 return 0;
7227 if (sb_rdonly(sb))
7228 return 1;
7229 if (ext4_has_unknown_ext2_ro_compat_features(sb))
7230 return 0;
7231 return 1;
7232}
7233#else
7234static inline void register_as_ext2(void) { }
7235static inline void unregister_as_ext2(void) { }
7236static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7237#endif
7238
7239static inline void register_as_ext3(void)
7240{
7241 int err = register_filesystem(&ext3_fs_type);
7242 if (err)
7243 printk(KERN_WARNING
7244 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7245}
7246
7247static inline void unregister_as_ext3(void)
7248{
7249 unregister_filesystem(&ext3_fs_type);
7250}
7251
7252static inline int ext3_feature_set_ok(struct super_block *sb)
7253{
7254 if (ext4_has_unknown_ext3_incompat_features(sb))
7255 return 0;
7256 if (!ext4_has_feature_journal(sb))
7257 return 0;
7258 if (sb_rdonly(sb))
7259 return 1;
7260 if (ext4_has_unknown_ext3_ro_compat_features(sb))
7261 return 0;
7262 return 1;
7263}
7264
7265static struct file_system_type ext4_fs_type = {
7266 .owner = THIS_MODULE,
7267 .name = "ext4",
7268 .init_fs_context = ext4_init_fs_context,
7269 .parameters = ext4_param_specs,
7270 .kill_sb = kill_block_super,
7271 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7272};
7273MODULE_ALIAS_FS("ext4");
7274
7275/* Shared across all ext4 file systems */
7276wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7277
7278static int __init ext4_init_fs(void)
7279{
7280 int i, err;
7281
7282 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7283 ext4_li_info = NULL;
7284
7285 /* Build-time check for flags consistency */
7286 ext4_check_flag_values();
7287
7288 for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7289 init_waitqueue_head(&ext4__ioend_wq[i]);
7290
7291 err = ext4_init_es();
7292 if (err)
7293 return err;
7294
7295 err = ext4_init_pending();
7296 if (err)
7297 goto out7;
7298
7299 err = ext4_init_post_read_processing();
7300 if (err)
7301 goto out6;
7302
7303 err = ext4_init_pageio();
7304 if (err)
7305 goto out5;
7306
7307 err = ext4_init_system_zone();
7308 if (err)
7309 goto out4;
7310
7311 err = ext4_init_sysfs();
7312 if (err)
7313 goto out3;
7314
7315 err = ext4_init_mballoc();
7316 if (err)
7317 goto out2;
7318 err = init_inodecache();
7319 if (err)
7320 goto out1;
7321
7322 err = ext4_fc_init_dentry_cache();
7323 if (err)
7324 goto out05;
7325
7326 register_as_ext3();
7327 register_as_ext2();
7328 err = register_filesystem(&ext4_fs_type);
7329 if (err)
7330 goto out;
7331
7332 return 0;
7333out:
7334 unregister_as_ext2();
7335 unregister_as_ext3();
7336 ext4_fc_destroy_dentry_cache();
7337out05:
7338 destroy_inodecache();
7339out1:
7340 ext4_exit_mballoc();
7341out2:
7342 ext4_exit_sysfs();
7343out3:
7344 ext4_exit_system_zone();
7345out4:
7346 ext4_exit_pageio();
7347out5:
7348 ext4_exit_post_read_processing();
7349out6:
7350 ext4_exit_pending();
7351out7:
7352 ext4_exit_es();
7353
7354 return err;
7355}
7356
7357static void __exit ext4_exit_fs(void)
7358{
7359 ext4_destroy_lazyinit_thread();
7360 unregister_as_ext2();
7361 unregister_as_ext3();
7362 unregister_filesystem(&ext4_fs_type);
7363 ext4_fc_destroy_dentry_cache();
7364 destroy_inodecache();
7365 ext4_exit_mballoc();
7366 ext4_exit_sysfs();
7367 ext4_exit_system_zone();
7368 ext4_exit_pageio();
7369 ext4_exit_post_read_processing();
7370 ext4_exit_es();
7371 ext4_exit_pending();
7372}
7373
7374MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7375MODULE_DESCRIPTION("Fourth Extended Filesystem");
7376MODULE_LICENSE("GPL");
7377MODULE_SOFTDEP("pre: crc32c");
7378module_init(ext4_init_fs)
7379module_exit(ext4_exit_fs)
Linux 6.x block layer and io_uring tree(s)