Loading...
1/*
2 * linux/fs/ext4/super.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 */
18
19#include <linux/module.h>
20#include <linux/string.h>
21#include <linux/fs.h>
22#include <linux/time.h>
23#include <linux/vmalloc.h>
24#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/parser.h>
29#include <linux/buffer_head.h>
30#include <linux/exportfs.h>
31#include <linux/vfs.h>
32#include <linux/random.h>
33#include <linux/mount.h>
34#include <linux/namei.h>
35#include <linux/quotaops.h>
36#include <linux/seq_file.h>
37#include <linux/ctype.h>
38#include <linux/log2.h>
39#include <linux/crc16.h>
40#include <linux/cleancache.h>
41#include <linux/uaccess.h>
42
43#include <linux/kthread.h>
44#include <linux/freezer.h>
45
46#include "ext4.h"
47#include "ext4_extents.h" /* Needed for trace points definition */
48#include "ext4_jbd2.h"
49#include "xattr.h"
50#include "acl.h"
51#include "mballoc.h"
52
53#define CREATE_TRACE_POINTS
54#include <trace/events/ext4.h>
55
56static struct ext4_lazy_init *ext4_li_info;
57static struct mutex ext4_li_mtx;
58static struct ratelimit_state ext4_mount_msg_ratelimit;
59
60static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
61 unsigned long journal_devnum);
62static int ext4_show_options(struct seq_file *seq, struct dentry *root);
63static int ext4_commit_super(struct super_block *sb, int sync);
64static void ext4_mark_recovery_complete(struct super_block *sb,
65 struct ext4_super_block *es);
66static void ext4_clear_journal_err(struct super_block *sb,
67 struct ext4_super_block *es);
68static int ext4_sync_fs(struct super_block *sb, int wait);
69static int ext4_remount(struct super_block *sb, int *flags, char *data);
70static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
71static int ext4_unfreeze(struct super_block *sb);
72static int ext4_freeze(struct super_block *sb);
73static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
74 const char *dev_name, void *data);
75static inline int ext2_feature_set_ok(struct super_block *sb);
76static inline int ext3_feature_set_ok(struct super_block *sb);
77static int ext4_feature_set_ok(struct super_block *sb, int readonly);
78static void ext4_destroy_lazyinit_thread(void);
79static void ext4_unregister_li_request(struct super_block *sb);
80static void ext4_clear_request_list(void);
81static struct inode *ext4_get_journal_inode(struct super_block *sb,
82 unsigned int journal_inum);
83
84/*
85 * Lock ordering
86 *
87 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
88 * i_mmap_rwsem (inode->i_mmap_rwsem)!
89 *
90 * page fault path:
91 * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
92 * page lock -> i_data_sem (rw)
93 *
94 * buffered write path:
95 * sb_start_write -> i_mutex -> mmap_sem
96 * sb_start_write -> i_mutex -> transaction start -> page lock ->
97 * i_data_sem (rw)
98 *
99 * truncate:
100 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
101 * i_mmap_rwsem (w) -> page lock
102 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
103 * transaction start -> i_data_sem (rw)
104 *
105 * direct IO:
106 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
107 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
108 * transaction start -> i_data_sem (rw)
109 *
110 * writepages:
111 * transaction start -> page lock(s) -> i_data_sem (rw)
112 */
113
114#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
115static struct file_system_type ext2_fs_type = {
116 .owner = THIS_MODULE,
117 .name = "ext2",
118 .mount = ext4_mount,
119 .kill_sb = kill_block_super,
120 .fs_flags = FS_REQUIRES_DEV,
121};
122MODULE_ALIAS_FS("ext2");
123MODULE_ALIAS("ext2");
124#define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
125#else
126#define IS_EXT2_SB(sb) (0)
127#endif
128
129
130static struct file_system_type ext3_fs_type = {
131 .owner = THIS_MODULE,
132 .name = "ext3",
133 .mount = ext4_mount,
134 .kill_sb = kill_block_super,
135 .fs_flags = FS_REQUIRES_DEV,
136};
137MODULE_ALIAS_FS("ext3");
138MODULE_ALIAS("ext3");
139#define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
140
141static int ext4_verify_csum_type(struct super_block *sb,
142 struct ext4_super_block *es)
143{
144 if (!ext4_has_feature_metadata_csum(sb))
145 return 1;
146
147 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
148}
149
150static __le32 ext4_superblock_csum(struct super_block *sb,
151 struct ext4_super_block *es)
152{
153 struct ext4_sb_info *sbi = EXT4_SB(sb);
154 int offset = offsetof(struct ext4_super_block, s_checksum);
155 __u32 csum;
156
157 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
158
159 return cpu_to_le32(csum);
160}
161
162static int ext4_superblock_csum_verify(struct super_block *sb,
163 struct ext4_super_block *es)
164{
165 if (!ext4_has_metadata_csum(sb))
166 return 1;
167
168 return es->s_checksum == ext4_superblock_csum(sb, es);
169}
170
171void ext4_superblock_csum_set(struct super_block *sb)
172{
173 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
174
175 if (!ext4_has_metadata_csum(sb))
176 return;
177
178 es->s_checksum = ext4_superblock_csum(sb, es);
179}
180
181void *ext4_kvmalloc(size_t size, gfp_t flags)
182{
183 void *ret;
184
185 ret = kmalloc(size, flags | __GFP_NOWARN);
186 if (!ret)
187 ret = __vmalloc(size, flags, PAGE_KERNEL);
188 return ret;
189}
190
191void *ext4_kvzalloc(size_t size, gfp_t flags)
192{
193 void *ret;
194
195 ret = kzalloc(size, flags | __GFP_NOWARN);
196 if (!ret)
197 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
198 return ret;
199}
200
201ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
202 struct ext4_group_desc *bg)
203{
204 return le32_to_cpu(bg->bg_block_bitmap_lo) |
205 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
206 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
207}
208
209ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
210 struct ext4_group_desc *bg)
211{
212 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
213 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
214 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
215}
216
217ext4_fsblk_t ext4_inode_table(struct super_block *sb,
218 struct ext4_group_desc *bg)
219{
220 return le32_to_cpu(bg->bg_inode_table_lo) |
221 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
222 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
223}
224
225__u32 ext4_free_group_clusters(struct super_block *sb,
226 struct ext4_group_desc *bg)
227{
228 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
229 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
230 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
231}
232
233__u32 ext4_free_inodes_count(struct super_block *sb,
234 struct ext4_group_desc *bg)
235{
236 return le16_to_cpu(bg->bg_free_inodes_count_lo) |
237 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
238 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
239}
240
241__u32 ext4_used_dirs_count(struct super_block *sb,
242 struct ext4_group_desc *bg)
243{
244 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
245 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
246 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
247}
248
249__u32 ext4_itable_unused_count(struct super_block *sb,
250 struct ext4_group_desc *bg)
251{
252 return le16_to_cpu(bg->bg_itable_unused_lo) |
253 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
254 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
255}
256
257void ext4_block_bitmap_set(struct super_block *sb,
258 struct ext4_group_desc *bg, ext4_fsblk_t blk)
259{
260 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
261 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
262 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
263}
264
265void ext4_inode_bitmap_set(struct super_block *sb,
266 struct ext4_group_desc *bg, ext4_fsblk_t blk)
267{
268 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
269 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
270 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
271}
272
273void ext4_inode_table_set(struct super_block *sb,
274 struct ext4_group_desc *bg, ext4_fsblk_t blk)
275{
276 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
277 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
278 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
279}
280
281void ext4_free_group_clusters_set(struct super_block *sb,
282 struct ext4_group_desc *bg, __u32 count)
283{
284 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
285 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
286 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
287}
288
289void ext4_free_inodes_set(struct super_block *sb,
290 struct ext4_group_desc *bg, __u32 count)
291{
292 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
293 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
294 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
295}
296
297void ext4_used_dirs_set(struct super_block *sb,
298 struct ext4_group_desc *bg, __u32 count)
299{
300 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
301 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
302 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
303}
304
305void ext4_itable_unused_set(struct super_block *sb,
306 struct ext4_group_desc *bg, __u32 count)
307{
308 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
309 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
310 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
311}
312
313
314static void __save_error_info(struct super_block *sb, const char *func,
315 unsigned int line)
316{
317 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
318
319 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
320 if (bdev_read_only(sb->s_bdev))
321 return;
322 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
323 es->s_last_error_time = cpu_to_le32(get_seconds());
324 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
325 es->s_last_error_line = cpu_to_le32(line);
326 if (!es->s_first_error_time) {
327 es->s_first_error_time = es->s_last_error_time;
328 strncpy(es->s_first_error_func, func,
329 sizeof(es->s_first_error_func));
330 es->s_first_error_line = cpu_to_le32(line);
331 es->s_first_error_ino = es->s_last_error_ino;
332 es->s_first_error_block = es->s_last_error_block;
333 }
334 /*
335 * Start the daily error reporting function if it hasn't been
336 * started already
337 */
338 if (!es->s_error_count)
339 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
340 le32_add_cpu(&es->s_error_count, 1);
341}
342
343static void save_error_info(struct super_block *sb, const char *func,
344 unsigned int line)
345{
346 __save_error_info(sb, func, line);
347 ext4_commit_super(sb, 1);
348}
349
350/*
351 * The del_gendisk() function uninitializes the disk-specific data
352 * structures, including the bdi structure, without telling anyone
353 * else. Once this happens, any attempt to call mark_buffer_dirty()
354 * (for example, by ext4_commit_super), will cause a kernel OOPS.
355 * This is a kludge to prevent these oops until we can put in a proper
356 * hook in del_gendisk() to inform the VFS and file system layers.
357 */
358static int block_device_ejected(struct super_block *sb)
359{
360 struct inode *bd_inode = sb->s_bdev->bd_inode;
361 struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
362
363 return bdi->dev == NULL;
364}
365
366static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
367{
368 struct super_block *sb = journal->j_private;
369 struct ext4_sb_info *sbi = EXT4_SB(sb);
370 int error = is_journal_aborted(journal);
371 struct ext4_journal_cb_entry *jce;
372
373 BUG_ON(txn->t_state == T_FINISHED);
374 spin_lock(&sbi->s_md_lock);
375 while (!list_empty(&txn->t_private_list)) {
376 jce = list_entry(txn->t_private_list.next,
377 struct ext4_journal_cb_entry, jce_list);
378 list_del_init(&jce->jce_list);
379 spin_unlock(&sbi->s_md_lock);
380 jce->jce_func(sb, jce, error);
381 spin_lock(&sbi->s_md_lock);
382 }
383 spin_unlock(&sbi->s_md_lock);
384}
385
386/* Deal with the reporting of failure conditions on a filesystem such as
387 * inconsistencies detected or read IO failures.
388 *
389 * On ext2, we can store the error state of the filesystem in the
390 * superblock. That is not possible on ext4, because we may have other
391 * write ordering constraints on the superblock which prevent us from
392 * writing it out straight away; and given that the journal is about to
393 * be aborted, we can't rely on the current, or future, transactions to
394 * write out the superblock safely.
395 *
396 * We'll just use the jbd2_journal_abort() error code to record an error in
397 * the journal instead. On recovery, the journal will complain about
398 * that error until we've noted it down and cleared it.
399 */
400
401static void ext4_handle_error(struct super_block *sb)
402{
403 if (sb->s_flags & MS_RDONLY)
404 return;
405
406 if (!test_opt(sb, ERRORS_CONT)) {
407 journal_t *journal = EXT4_SB(sb)->s_journal;
408
409 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
410 if (journal)
411 jbd2_journal_abort(journal, -EIO);
412 }
413 if (test_opt(sb, ERRORS_RO)) {
414 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
415 /*
416 * Make sure updated value of ->s_mount_flags will be visible
417 * before ->s_flags update
418 */
419 smp_wmb();
420 sb->s_flags |= MS_RDONLY;
421 }
422 if (test_opt(sb, ERRORS_PANIC)) {
423 if (EXT4_SB(sb)->s_journal &&
424 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
425 return;
426 panic("EXT4-fs (device %s): panic forced after error\n",
427 sb->s_id);
428 }
429}
430
431#define ext4_error_ratelimit(sb) \
432 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \
433 "EXT4-fs error")
434
435void __ext4_error(struct super_block *sb, const char *function,
436 unsigned int line, const char *fmt, ...)
437{
438 struct va_format vaf;
439 va_list args;
440
441 if (ext4_error_ratelimit(sb)) {
442 va_start(args, fmt);
443 vaf.fmt = fmt;
444 vaf.va = &args;
445 printk(KERN_CRIT
446 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
447 sb->s_id, function, line, current->comm, &vaf);
448 va_end(args);
449 }
450 save_error_info(sb, function, line);
451 ext4_handle_error(sb);
452}
453
454void __ext4_error_inode(struct inode *inode, const char *function,
455 unsigned int line, ext4_fsblk_t block,
456 const char *fmt, ...)
457{
458 va_list args;
459 struct va_format vaf;
460 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
461
462 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
463 es->s_last_error_block = cpu_to_le64(block);
464 if (ext4_error_ratelimit(inode->i_sb)) {
465 va_start(args, fmt);
466 vaf.fmt = fmt;
467 vaf.va = &args;
468 if (block)
469 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
470 "inode #%lu: block %llu: comm %s: %pV\n",
471 inode->i_sb->s_id, function, line, inode->i_ino,
472 block, current->comm, &vaf);
473 else
474 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
475 "inode #%lu: comm %s: %pV\n",
476 inode->i_sb->s_id, function, line, inode->i_ino,
477 current->comm, &vaf);
478 va_end(args);
479 }
480 save_error_info(inode->i_sb, function, line);
481 ext4_handle_error(inode->i_sb);
482}
483
484void __ext4_error_file(struct file *file, const char *function,
485 unsigned int line, ext4_fsblk_t block,
486 const char *fmt, ...)
487{
488 va_list args;
489 struct va_format vaf;
490 struct ext4_super_block *es;
491 struct inode *inode = file_inode(file);
492 char pathname[80], *path;
493
494 es = EXT4_SB(inode->i_sb)->s_es;
495 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
496 if (ext4_error_ratelimit(inode->i_sb)) {
497 path = file_path(file, pathname, sizeof(pathname));
498 if (IS_ERR(path))
499 path = "(unknown)";
500 va_start(args, fmt);
501 vaf.fmt = fmt;
502 vaf.va = &args;
503 if (block)
504 printk(KERN_CRIT
505 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
506 "block %llu: comm %s: path %s: %pV\n",
507 inode->i_sb->s_id, function, line, inode->i_ino,
508 block, current->comm, path, &vaf);
509 else
510 printk(KERN_CRIT
511 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
512 "comm %s: path %s: %pV\n",
513 inode->i_sb->s_id, function, line, inode->i_ino,
514 current->comm, path, &vaf);
515 va_end(args);
516 }
517 save_error_info(inode->i_sb, function, line);
518 ext4_handle_error(inode->i_sb);
519}
520
521const char *ext4_decode_error(struct super_block *sb, int errno,
522 char nbuf[16])
523{
524 char *errstr = NULL;
525
526 switch (errno) {
527 case -EFSCORRUPTED:
528 errstr = "Corrupt filesystem";
529 break;
530 case -EFSBADCRC:
531 errstr = "Filesystem failed CRC";
532 break;
533 case -EIO:
534 errstr = "IO failure";
535 break;
536 case -ENOMEM:
537 errstr = "Out of memory";
538 break;
539 case -EROFS:
540 if (!sb || (EXT4_SB(sb)->s_journal &&
541 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
542 errstr = "Journal has aborted";
543 else
544 errstr = "Readonly filesystem";
545 break;
546 default:
547 /* If the caller passed in an extra buffer for unknown
548 * errors, textualise them now. Else we just return
549 * NULL. */
550 if (nbuf) {
551 /* Check for truncated error codes... */
552 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
553 errstr = nbuf;
554 }
555 break;
556 }
557
558 return errstr;
559}
560
561/* __ext4_std_error decodes expected errors from journaling functions
562 * automatically and invokes the appropriate error response. */
563
564void __ext4_std_error(struct super_block *sb, const char *function,
565 unsigned int line, int errno)
566{
567 char nbuf[16];
568 const char *errstr;
569
570 /* Special case: if the error is EROFS, and we're not already
571 * inside a transaction, then there's really no point in logging
572 * an error. */
573 if (errno == -EROFS && journal_current_handle() == NULL &&
574 (sb->s_flags & MS_RDONLY))
575 return;
576
577 if (ext4_error_ratelimit(sb)) {
578 errstr = ext4_decode_error(sb, errno, nbuf);
579 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
580 sb->s_id, function, line, errstr);
581 }
582
583 save_error_info(sb, function, line);
584 ext4_handle_error(sb);
585}
586
587/*
588 * ext4_abort is a much stronger failure handler than ext4_error. The
589 * abort function may be used to deal with unrecoverable failures such
590 * as journal IO errors or ENOMEM at a critical moment in log management.
591 *
592 * We unconditionally force the filesystem into an ABORT|READONLY state,
593 * unless the error response on the fs has been set to panic in which
594 * case we take the easy way out and panic immediately.
595 */
596
597void __ext4_abort(struct super_block *sb, const char *function,
598 unsigned int line, const char *fmt, ...)
599{
600 struct va_format vaf;
601 va_list args;
602
603 save_error_info(sb, function, line);
604 va_start(args, fmt);
605 vaf.fmt = fmt;
606 vaf.va = &args;
607 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
608 sb->s_id, function, line, &vaf);
609 va_end(args);
610
611 if ((sb->s_flags & MS_RDONLY) == 0) {
612 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
613 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
614 /*
615 * Make sure updated value of ->s_mount_flags will be visible
616 * before ->s_flags update
617 */
618 smp_wmb();
619 sb->s_flags |= MS_RDONLY;
620 if (EXT4_SB(sb)->s_journal)
621 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
622 save_error_info(sb, function, line);
623 }
624 if (test_opt(sb, ERRORS_PANIC)) {
625 if (EXT4_SB(sb)->s_journal &&
626 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
627 return;
628 panic("EXT4-fs panic from previous error\n");
629 }
630}
631
632void __ext4_msg(struct super_block *sb,
633 const char *prefix, const char *fmt, ...)
634{
635 struct va_format vaf;
636 va_list args;
637
638 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
639 return;
640
641 va_start(args, fmt);
642 vaf.fmt = fmt;
643 vaf.va = &args;
644 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
645 va_end(args);
646}
647
648#define ext4_warning_ratelimit(sb) \
649 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
650 "EXT4-fs warning")
651
652void __ext4_warning(struct super_block *sb, const char *function,
653 unsigned int line, const char *fmt, ...)
654{
655 struct va_format vaf;
656 va_list args;
657
658 if (!ext4_warning_ratelimit(sb))
659 return;
660
661 va_start(args, fmt);
662 vaf.fmt = fmt;
663 vaf.va = &args;
664 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
665 sb->s_id, function, line, &vaf);
666 va_end(args);
667}
668
669void __ext4_warning_inode(const struct inode *inode, const char *function,
670 unsigned int line, const char *fmt, ...)
671{
672 struct va_format vaf;
673 va_list args;
674
675 if (!ext4_warning_ratelimit(inode->i_sb))
676 return;
677
678 va_start(args, fmt);
679 vaf.fmt = fmt;
680 vaf.va = &args;
681 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
682 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
683 function, line, inode->i_ino, current->comm, &vaf);
684 va_end(args);
685}
686
687void __ext4_grp_locked_error(const char *function, unsigned int line,
688 struct super_block *sb, ext4_group_t grp,
689 unsigned long ino, ext4_fsblk_t block,
690 const char *fmt, ...)
691__releases(bitlock)
692__acquires(bitlock)
693{
694 struct va_format vaf;
695 va_list args;
696 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
697
698 es->s_last_error_ino = cpu_to_le32(ino);
699 es->s_last_error_block = cpu_to_le64(block);
700 __save_error_info(sb, function, line);
701
702 if (ext4_error_ratelimit(sb)) {
703 va_start(args, fmt);
704 vaf.fmt = fmt;
705 vaf.va = &args;
706 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
707 sb->s_id, function, line, grp);
708 if (ino)
709 printk(KERN_CONT "inode %lu: ", ino);
710 if (block)
711 printk(KERN_CONT "block %llu:",
712 (unsigned long long) block);
713 printk(KERN_CONT "%pV\n", &vaf);
714 va_end(args);
715 }
716
717 if (test_opt(sb, ERRORS_CONT)) {
718 ext4_commit_super(sb, 0);
719 return;
720 }
721
722 ext4_unlock_group(sb, grp);
723 ext4_handle_error(sb);
724 /*
725 * We only get here in the ERRORS_RO case; relocking the group
726 * may be dangerous, but nothing bad will happen since the
727 * filesystem will have already been marked read/only and the
728 * journal has been aborted. We return 1 as a hint to callers
729 * who might what to use the return value from
730 * ext4_grp_locked_error() to distinguish between the
731 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
732 * aggressively from the ext4 function in question, with a
733 * more appropriate error code.
734 */
735 ext4_lock_group(sb, grp);
736 return;
737}
738
739void ext4_update_dynamic_rev(struct super_block *sb)
740{
741 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
742
743 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
744 return;
745
746 ext4_warning(sb,
747 "updating to rev %d because of new feature flag, "
748 "running e2fsck is recommended",
749 EXT4_DYNAMIC_REV);
750
751 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
752 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
753 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
754 /* leave es->s_feature_*compat flags alone */
755 /* es->s_uuid will be set by e2fsck if empty */
756
757 /*
758 * The rest of the superblock fields should be zero, and if not it
759 * means they are likely already in use, so leave them alone. We
760 * can leave it up to e2fsck to clean up any inconsistencies there.
761 */
762}
763
764/*
765 * Open the external journal device
766 */
767static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
768{
769 struct block_device *bdev;
770 char b[BDEVNAME_SIZE];
771
772 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
773 if (IS_ERR(bdev))
774 goto fail;
775 return bdev;
776
777fail:
778 ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
779 __bdevname(dev, b), PTR_ERR(bdev));
780 return NULL;
781}
782
783/*
784 * Release the journal device
785 */
786static void ext4_blkdev_put(struct block_device *bdev)
787{
788 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
789}
790
791static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
792{
793 struct block_device *bdev;
794 bdev = sbi->journal_bdev;
795 if (bdev) {
796 ext4_blkdev_put(bdev);
797 sbi->journal_bdev = NULL;
798 }
799}
800
801static inline struct inode *orphan_list_entry(struct list_head *l)
802{
803 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
804}
805
806static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
807{
808 struct list_head *l;
809
810 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
811 le32_to_cpu(sbi->s_es->s_last_orphan));
812
813 printk(KERN_ERR "sb_info orphan list:\n");
814 list_for_each(l, &sbi->s_orphan) {
815 struct inode *inode = orphan_list_entry(l);
816 printk(KERN_ERR " "
817 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
818 inode->i_sb->s_id, inode->i_ino, inode,
819 inode->i_mode, inode->i_nlink,
820 NEXT_ORPHAN(inode));
821 }
822}
823
824static void ext4_put_super(struct super_block *sb)
825{
826 struct ext4_sb_info *sbi = EXT4_SB(sb);
827 struct ext4_super_block *es = sbi->s_es;
828 int aborted = 0;
829 int i, err;
830
831 ext4_unregister_li_request(sb);
832 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
833
834 flush_workqueue(sbi->rsv_conversion_wq);
835 destroy_workqueue(sbi->rsv_conversion_wq);
836
837 if (sbi->s_journal) {
838 aborted = is_journal_aborted(sbi->s_journal);
839 err = jbd2_journal_destroy(sbi->s_journal);
840 sbi->s_journal = NULL;
841 if ((err < 0) && !aborted)
842 ext4_abort(sb, "Couldn't clean up the journal");
843 }
844
845 ext4_unregister_sysfs(sb);
846 ext4_es_unregister_shrinker(sbi);
847 del_timer_sync(&sbi->s_err_report);
848 ext4_release_system_zone(sb);
849 ext4_mb_release(sb);
850 ext4_ext_release(sb);
851
852 if (!(sb->s_flags & MS_RDONLY) && !aborted) {
853 ext4_clear_feature_journal_needs_recovery(sb);
854 es->s_state = cpu_to_le16(sbi->s_mount_state);
855 }
856 if (!(sb->s_flags & MS_RDONLY))
857 ext4_commit_super(sb, 1);
858
859 for (i = 0; i < sbi->s_gdb_count; i++)
860 brelse(sbi->s_group_desc[i]);
861 kvfree(sbi->s_group_desc);
862 kvfree(sbi->s_flex_groups);
863 percpu_counter_destroy(&sbi->s_freeclusters_counter);
864 percpu_counter_destroy(&sbi->s_freeinodes_counter);
865 percpu_counter_destroy(&sbi->s_dirs_counter);
866 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
867 percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
868#ifdef CONFIG_QUOTA
869 for (i = 0; i < EXT4_MAXQUOTAS; i++)
870 kfree(sbi->s_qf_names[i]);
871#endif
872
873 /* Debugging code just in case the in-memory inode orphan list
874 * isn't empty. The on-disk one can be non-empty if we've
875 * detected an error and taken the fs readonly, but the
876 * in-memory list had better be clean by this point. */
877 if (!list_empty(&sbi->s_orphan))
878 dump_orphan_list(sb, sbi);
879 J_ASSERT(list_empty(&sbi->s_orphan));
880
881 sync_blockdev(sb->s_bdev);
882 invalidate_bdev(sb->s_bdev);
883 if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
884 /*
885 * Invalidate the journal device's buffers. We don't want them
886 * floating about in memory - the physical journal device may
887 * hotswapped, and it breaks the `ro-after' testing code.
888 */
889 sync_blockdev(sbi->journal_bdev);
890 invalidate_bdev(sbi->journal_bdev);
891 ext4_blkdev_remove(sbi);
892 }
893 if (sbi->s_mb_cache) {
894 ext4_xattr_destroy_cache(sbi->s_mb_cache);
895 sbi->s_mb_cache = NULL;
896 }
897 if (sbi->s_mmp_tsk)
898 kthread_stop(sbi->s_mmp_tsk);
899 brelse(sbi->s_sbh);
900 sb->s_fs_info = NULL;
901 /*
902 * Now that we are completely done shutting down the
903 * superblock, we need to actually destroy the kobject.
904 */
905 kobject_put(&sbi->s_kobj);
906 wait_for_completion(&sbi->s_kobj_unregister);
907 if (sbi->s_chksum_driver)
908 crypto_free_shash(sbi->s_chksum_driver);
909 kfree(sbi->s_blockgroup_lock);
910 kfree(sbi);
911}
912
913static struct kmem_cache *ext4_inode_cachep;
914
915/*
916 * Called inside transaction, so use GFP_NOFS
917 */
918static struct inode *ext4_alloc_inode(struct super_block *sb)
919{
920 struct ext4_inode_info *ei;
921
922 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
923 if (!ei)
924 return NULL;
925
926 ei->vfs_inode.i_version = 1;
927 spin_lock_init(&ei->i_raw_lock);
928 INIT_LIST_HEAD(&ei->i_prealloc_list);
929 spin_lock_init(&ei->i_prealloc_lock);
930 ext4_es_init_tree(&ei->i_es_tree);
931 rwlock_init(&ei->i_es_lock);
932 INIT_LIST_HEAD(&ei->i_es_list);
933 ei->i_es_all_nr = 0;
934 ei->i_es_shk_nr = 0;
935 ei->i_es_shrink_lblk = 0;
936 ei->i_reserved_data_blocks = 0;
937 ei->i_reserved_meta_blocks = 0;
938 ei->i_allocated_meta_blocks = 0;
939 ei->i_da_metadata_calc_len = 0;
940 ei->i_da_metadata_calc_last_lblock = 0;
941 spin_lock_init(&(ei->i_block_reservation_lock));
942#ifdef CONFIG_QUOTA
943 ei->i_reserved_quota = 0;
944 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
945#endif
946 ei->jinode = NULL;
947 INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
948 spin_lock_init(&ei->i_completed_io_lock);
949 ei->i_sync_tid = 0;
950 ei->i_datasync_tid = 0;
951 atomic_set(&ei->i_unwritten, 0);
952 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
953 return &ei->vfs_inode;
954}
955
956static int ext4_drop_inode(struct inode *inode)
957{
958 int drop = generic_drop_inode(inode);
959
960 trace_ext4_drop_inode(inode, drop);
961 return drop;
962}
963
964static void ext4_i_callback(struct rcu_head *head)
965{
966 struct inode *inode = container_of(head, struct inode, i_rcu);
967 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
968}
969
970static void ext4_destroy_inode(struct inode *inode)
971{
972 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
973 ext4_msg(inode->i_sb, KERN_ERR,
974 "Inode %lu (%p): orphan list check failed!",
975 inode->i_ino, EXT4_I(inode));
976 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
977 EXT4_I(inode), sizeof(struct ext4_inode_info),
978 true);
979 dump_stack();
980 }
981 call_rcu(&inode->i_rcu, ext4_i_callback);
982}
983
984static void init_once(void *foo)
985{
986 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
987
988 INIT_LIST_HEAD(&ei->i_orphan);
989 init_rwsem(&ei->xattr_sem);
990 init_rwsem(&ei->i_data_sem);
991 init_rwsem(&ei->i_mmap_sem);
992 inode_init_once(&ei->vfs_inode);
993}
994
995static int __init init_inodecache(void)
996{
997 ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
998 sizeof(struct ext4_inode_info),
999 0, (SLAB_RECLAIM_ACCOUNT|
1000 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1001 init_once);
1002 if (ext4_inode_cachep == NULL)
1003 return -ENOMEM;
1004 return 0;
1005}
1006
1007static void destroy_inodecache(void)
1008{
1009 /*
1010 * Make sure all delayed rcu free inodes are flushed before we
1011 * destroy cache.
1012 */
1013 rcu_barrier();
1014 kmem_cache_destroy(ext4_inode_cachep);
1015}
1016
1017void ext4_clear_inode(struct inode *inode)
1018{
1019 invalidate_inode_buffers(inode);
1020 clear_inode(inode);
1021 dquot_drop(inode);
1022 ext4_discard_preallocations(inode);
1023 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1024 if (EXT4_I(inode)->jinode) {
1025 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1026 EXT4_I(inode)->jinode);
1027 jbd2_free_inode(EXT4_I(inode)->jinode);
1028 EXT4_I(inode)->jinode = NULL;
1029 }
1030#ifdef CONFIG_EXT4_FS_ENCRYPTION
1031 fscrypt_put_encryption_info(inode, NULL);
1032#endif
1033}
1034
1035static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1036 u64 ino, u32 generation)
1037{
1038 struct inode *inode;
1039
1040 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1041 return ERR_PTR(-ESTALE);
1042 if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1043 return ERR_PTR(-ESTALE);
1044
1045 /* iget isn't really right if the inode is currently unallocated!!
1046 *
1047 * ext4_read_inode will return a bad_inode if the inode had been
1048 * deleted, so we should be safe.
1049 *
1050 * Currently we don't know the generation for parent directory, so
1051 * a generation of 0 means "accept any"
1052 */
1053 inode = ext4_iget_normal(sb, ino);
1054 if (IS_ERR(inode))
1055 return ERR_CAST(inode);
1056 if (generation && inode->i_generation != generation) {
1057 iput(inode);
1058 return ERR_PTR(-ESTALE);
1059 }
1060
1061 return inode;
1062}
1063
1064static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1065 int fh_len, int fh_type)
1066{
1067 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1068 ext4_nfs_get_inode);
1069}
1070
1071static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1072 int fh_len, int fh_type)
1073{
1074 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1075 ext4_nfs_get_inode);
1076}
1077
1078/*
1079 * Try to release metadata pages (indirect blocks, directories) which are
1080 * mapped via the block device. Since these pages could have journal heads
1081 * which would prevent try_to_free_buffers() from freeing them, we must use
1082 * jbd2 layer's try_to_free_buffers() function to release them.
1083 */
1084static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1085 gfp_t wait)
1086{
1087 journal_t *journal = EXT4_SB(sb)->s_journal;
1088
1089 WARN_ON(PageChecked(page));
1090 if (!page_has_buffers(page))
1091 return 0;
1092 if (journal)
1093 return jbd2_journal_try_to_free_buffers(journal, page,
1094 wait & ~__GFP_DIRECT_RECLAIM);
1095 return try_to_free_buffers(page);
1096}
1097
1098#ifdef CONFIG_EXT4_FS_ENCRYPTION
1099static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1100{
1101 return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1102 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1103}
1104
1105static int ext4_key_prefix(struct inode *inode, u8 **key)
1106{
1107 *key = EXT4_SB(inode->i_sb)->key_prefix;
1108 return EXT4_SB(inode->i_sb)->key_prefix_size;
1109}
1110
1111static int ext4_prepare_context(struct inode *inode)
1112{
1113 return ext4_convert_inline_data(inode);
1114}
1115
1116static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1117 void *fs_data)
1118{
1119 handle_t *handle = fs_data;
1120 int res, res2, retries = 0;
1121
1122 /*
1123 * If a journal handle was specified, then the encryption context is
1124 * being set on a new inode via inheritance and is part of a larger
1125 * transaction to create the inode. Otherwise the encryption context is
1126 * being set on an existing inode in its own transaction. Only in the
1127 * latter case should the "retry on ENOSPC" logic be used.
1128 */
1129
1130 if (handle) {
1131 res = ext4_xattr_set_handle(handle, inode,
1132 EXT4_XATTR_INDEX_ENCRYPTION,
1133 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1134 ctx, len, 0);
1135 if (!res) {
1136 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1137 ext4_clear_inode_state(inode,
1138 EXT4_STATE_MAY_INLINE_DATA);
1139 /*
1140 * Update inode->i_flags - e.g. S_DAX may get disabled
1141 */
1142 ext4_set_inode_flags(inode);
1143 }
1144 return res;
1145 }
1146
1147retry:
1148 handle = ext4_journal_start(inode, EXT4_HT_MISC,
1149 ext4_jbd2_credits_xattr(inode));
1150 if (IS_ERR(handle))
1151 return PTR_ERR(handle);
1152
1153 res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1154 EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1155 ctx, len, 0);
1156 if (!res) {
1157 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1158 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1159 ext4_set_inode_flags(inode);
1160 res = ext4_mark_inode_dirty(handle, inode);
1161 if (res)
1162 EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1163 }
1164 res2 = ext4_journal_stop(handle);
1165
1166 if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1167 goto retry;
1168 if (!res)
1169 res = res2;
1170 return res;
1171}
1172
1173static int ext4_dummy_context(struct inode *inode)
1174{
1175 return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1176}
1177
1178static unsigned ext4_max_namelen(struct inode *inode)
1179{
1180 return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1181 EXT4_NAME_LEN;
1182}
1183
1184static struct fscrypt_operations ext4_cryptops = {
1185 .get_context = ext4_get_context,
1186 .key_prefix = ext4_key_prefix,
1187 .prepare_context = ext4_prepare_context,
1188 .set_context = ext4_set_context,
1189 .dummy_context = ext4_dummy_context,
1190 .is_encrypted = ext4_encrypted_inode,
1191 .empty_dir = ext4_empty_dir,
1192 .max_namelen = ext4_max_namelen,
1193};
1194#else
1195static struct fscrypt_operations ext4_cryptops = {
1196 .is_encrypted = ext4_encrypted_inode,
1197};
1198#endif
1199
1200#ifdef CONFIG_QUOTA
1201static char *quotatypes[] = INITQFNAMES;
1202#define QTYPE2NAME(t) (quotatypes[t])
1203
1204static int ext4_write_dquot(struct dquot *dquot);
1205static int ext4_acquire_dquot(struct dquot *dquot);
1206static int ext4_release_dquot(struct dquot *dquot);
1207static int ext4_mark_dquot_dirty(struct dquot *dquot);
1208static int ext4_write_info(struct super_block *sb, int type);
1209static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1210 const struct path *path);
1211static int ext4_quota_off(struct super_block *sb, int type);
1212static int ext4_quota_on_mount(struct super_block *sb, int type);
1213static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1214 size_t len, loff_t off);
1215static ssize_t ext4_quota_write(struct super_block *sb, int type,
1216 const char *data, size_t len, loff_t off);
1217static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1218 unsigned int flags);
1219static int ext4_enable_quotas(struct super_block *sb);
1220static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1221
1222static struct dquot **ext4_get_dquots(struct inode *inode)
1223{
1224 return EXT4_I(inode)->i_dquot;
1225}
1226
1227static const struct dquot_operations ext4_quota_operations = {
1228 .get_reserved_space = ext4_get_reserved_space,
1229 .write_dquot = ext4_write_dquot,
1230 .acquire_dquot = ext4_acquire_dquot,
1231 .release_dquot = ext4_release_dquot,
1232 .mark_dirty = ext4_mark_dquot_dirty,
1233 .write_info = ext4_write_info,
1234 .alloc_dquot = dquot_alloc,
1235 .destroy_dquot = dquot_destroy,
1236 .get_projid = ext4_get_projid,
1237 .get_next_id = ext4_get_next_id,
1238};
1239
1240static const struct quotactl_ops ext4_qctl_operations = {
1241 .quota_on = ext4_quota_on,
1242 .quota_off = ext4_quota_off,
1243 .quota_sync = dquot_quota_sync,
1244 .get_state = dquot_get_state,
1245 .set_info = dquot_set_dqinfo,
1246 .get_dqblk = dquot_get_dqblk,
1247 .set_dqblk = dquot_set_dqblk,
1248 .get_nextdqblk = dquot_get_next_dqblk,
1249};
1250#endif
1251
1252static const struct super_operations ext4_sops = {
1253 .alloc_inode = ext4_alloc_inode,
1254 .destroy_inode = ext4_destroy_inode,
1255 .write_inode = ext4_write_inode,
1256 .dirty_inode = ext4_dirty_inode,
1257 .drop_inode = ext4_drop_inode,
1258 .evict_inode = ext4_evict_inode,
1259 .put_super = ext4_put_super,
1260 .sync_fs = ext4_sync_fs,
1261 .freeze_fs = ext4_freeze,
1262 .unfreeze_fs = ext4_unfreeze,
1263 .statfs = ext4_statfs,
1264 .remount_fs = ext4_remount,
1265 .show_options = ext4_show_options,
1266#ifdef CONFIG_QUOTA
1267 .quota_read = ext4_quota_read,
1268 .quota_write = ext4_quota_write,
1269 .get_dquots = ext4_get_dquots,
1270#endif
1271 .bdev_try_to_free_page = bdev_try_to_free_page,
1272};
1273
1274static const struct export_operations ext4_export_ops = {
1275 .fh_to_dentry = ext4_fh_to_dentry,
1276 .fh_to_parent = ext4_fh_to_parent,
1277 .get_parent = ext4_get_parent,
1278};
1279
1280enum {
1281 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1282 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1283 Opt_nouid32, Opt_debug, Opt_removed,
1284 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1285 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1286 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1287 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1288 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1289 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1290 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1291 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1292 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1293 Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1294 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1295 Opt_lazytime, Opt_nolazytime,
1296 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1297 Opt_inode_readahead_blks, Opt_journal_ioprio,
1298 Opt_dioread_nolock, Opt_dioread_lock,
1299 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1300 Opt_max_dir_size_kb, Opt_nojournal_checksum,
1301};
1302
1303static const match_table_t tokens = {
1304 {Opt_bsd_df, "bsddf"},
1305 {Opt_minix_df, "minixdf"},
1306 {Opt_grpid, "grpid"},
1307 {Opt_grpid, "bsdgroups"},
1308 {Opt_nogrpid, "nogrpid"},
1309 {Opt_nogrpid, "sysvgroups"},
1310 {Opt_resgid, "resgid=%u"},
1311 {Opt_resuid, "resuid=%u"},
1312 {Opt_sb, "sb=%u"},
1313 {Opt_err_cont, "errors=continue"},
1314 {Opt_err_panic, "errors=panic"},
1315 {Opt_err_ro, "errors=remount-ro"},
1316 {Opt_nouid32, "nouid32"},
1317 {Opt_debug, "debug"},
1318 {Opt_removed, "oldalloc"},
1319 {Opt_removed, "orlov"},
1320 {Opt_user_xattr, "user_xattr"},
1321 {Opt_nouser_xattr, "nouser_xattr"},
1322 {Opt_acl, "acl"},
1323 {Opt_noacl, "noacl"},
1324 {Opt_noload, "norecovery"},
1325 {Opt_noload, "noload"},
1326 {Opt_removed, "nobh"},
1327 {Opt_removed, "bh"},
1328 {Opt_commit, "commit=%u"},
1329 {Opt_min_batch_time, "min_batch_time=%u"},
1330 {Opt_max_batch_time, "max_batch_time=%u"},
1331 {Opt_journal_dev, "journal_dev=%u"},
1332 {Opt_journal_path, "journal_path=%s"},
1333 {Opt_journal_checksum, "journal_checksum"},
1334 {Opt_nojournal_checksum, "nojournal_checksum"},
1335 {Opt_journal_async_commit, "journal_async_commit"},
1336 {Opt_abort, "abort"},
1337 {Opt_data_journal, "data=journal"},
1338 {Opt_data_ordered, "data=ordered"},
1339 {Opt_data_writeback, "data=writeback"},
1340 {Opt_data_err_abort, "data_err=abort"},
1341 {Opt_data_err_ignore, "data_err=ignore"},
1342 {Opt_offusrjquota, "usrjquota="},
1343 {Opt_usrjquota, "usrjquota=%s"},
1344 {Opt_offgrpjquota, "grpjquota="},
1345 {Opt_grpjquota, "grpjquota=%s"},
1346 {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1347 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1348 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1349 {Opt_grpquota, "grpquota"},
1350 {Opt_noquota, "noquota"},
1351 {Opt_quota, "quota"},
1352 {Opt_usrquota, "usrquota"},
1353 {Opt_prjquota, "prjquota"},
1354 {Opt_barrier, "barrier=%u"},
1355 {Opt_barrier, "barrier"},
1356 {Opt_nobarrier, "nobarrier"},
1357 {Opt_i_version, "i_version"},
1358 {Opt_dax, "dax"},
1359 {Opt_stripe, "stripe=%u"},
1360 {Opt_delalloc, "delalloc"},
1361 {Opt_lazytime, "lazytime"},
1362 {Opt_nolazytime, "nolazytime"},
1363 {Opt_nodelalloc, "nodelalloc"},
1364 {Opt_removed, "mblk_io_submit"},
1365 {Opt_removed, "nomblk_io_submit"},
1366 {Opt_block_validity, "block_validity"},
1367 {Opt_noblock_validity, "noblock_validity"},
1368 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1369 {Opt_journal_ioprio, "journal_ioprio=%u"},
1370 {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1371 {Opt_auto_da_alloc, "auto_da_alloc"},
1372 {Opt_noauto_da_alloc, "noauto_da_alloc"},
1373 {Opt_dioread_nolock, "dioread_nolock"},
1374 {Opt_dioread_lock, "dioread_lock"},
1375 {Opt_discard, "discard"},
1376 {Opt_nodiscard, "nodiscard"},
1377 {Opt_init_itable, "init_itable=%u"},
1378 {Opt_init_itable, "init_itable"},
1379 {Opt_noinit_itable, "noinit_itable"},
1380 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1381 {Opt_test_dummy_encryption, "test_dummy_encryption"},
1382 {Opt_removed, "check=none"}, /* mount option from ext2/3 */
1383 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */
1384 {Opt_removed, "reservation"}, /* mount option from ext2/3 */
1385 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1386 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */
1387 {Opt_err, NULL},
1388};
1389
1390static ext4_fsblk_t get_sb_block(void **data)
1391{
1392 ext4_fsblk_t sb_block;
1393 char *options = (char *) *data;
1394
1395 if (!options || strncmp(options, "sb=", 3) != 0)
1396 return 1; /* Default location */
1397
1398 options += 3;
1399 /* TODO: use simple_strtoll with >32bit ext4 */
1400 sb_block = simple_strtoul(options, &options, 0);
1401 if (*options && *options != ',') {
1402 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1403 (char *) *data);
1404 return 1;
1405 }
1406 if (*options == ',')
1407 options++;
1408 *data = (void *) options;
1409
1410 return sb_block;
1411}
1412
1413#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1414static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1415 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1416
1417#ifdef CONFIG_QUOTA
1418static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1419{
1420 struct ext4_sb_info *sbi = EXT4_SB(sb);
1421 char *qname;
1422 int ret = -1;
1423
1424 if (sb_any_quota_loaded(sb) &&
1425 !sbi->s_qf_names[qtype]) {
1426 ext4_msg(sb, KERN_ERR,
1427 "Cannot change journaled "
1428 "quota options when quota turned on");
1429 return -1;
1430 }
1431 if (ext4_has_feature_quota(sb)) {
1432 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1433 "ignored when QUOTA feature is enabled");
1434 return 1;
1435 }
1436 qname = match_strdup(args);
1437 if (!qname) {
1438 ext4_msg(sb, KERN_ERR,
1439 "Not enough memory for storing quotafile name");
1440 return -1;
1441 }
1442 if (sbi->s_qf_names[qtype]) {
1443 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1444 ret = 1;
1445 else
1446 ext4_msg(sb, KERN_ERR,
1447 "%s quota file already specified",
1448 QTYPE2NAME(qtype));
1449 goto errout;
1450 }
1451 if (strchr(qname, '/')) {
1452 ext4_msg(sb, KERN_ERR,
1453 "quotafile must be on filesystem root");
1454 goto errout;
1455 }
1456 sbi->s_qf_names[qtype] = qname;
1457 set_opt(sb, QUOTA);
1458 return 1;
1459errout:
1460 kfree(qname);
1461 return ret;
1462}
1463
1464static int clear_qf_name(struct super_block *sb, int qtype)
1465{
1466
1467 struct ext4_sb_info *sbi = EXT4_SB(sb);
1468
1469 if (sb_any_quota_loaded(sb) &&
1470 sbi->s_qf_names[qtype]) {
1471 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1472 " when quota turned on");
1473 return -1;
1474 }
1475 kfree(sbi->s_qf_names[qtype]);
1476 sbi->s_qf_names[qtype] = NULL;
1477 return 1;
1478}
1479#endif
1480
1481#define MOPT_SET 0x0001
1482#define MOPT_CLEAR 0x0002
1483#define MOPT_NOSUPPORT 0x0004
1484#define MOPT_EXPLICIT 0x0008
1485#define MOPT_CLEAR_ERR 0x0010
1486#define MOPT_GTE0 0x0020
1487#ifdef CONFIG_QUOTA
1488#define MOPT_Q 0
1489#define MOPT_QFMT 0x0040
1490#else
1491#define MOPT_Q MOPT_NOSUPPORT
1492#define MOPT_QFMT MOPT_NOSUPPORT
1493#endif
1494#define MOPT_DATAJ 0x0080
1495#define MOPT_NO_EXT2 0x0100
1496#define MOPT_NO_EXT3 0x0200
1497#define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1498#define MOPT_STRING 0x0400
1499
1500static const struct mount_opts {
1501 int token;
1502 int mount_opt;
1503 int flags;
1504} ext4_mount_opts[] = {
1505 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1506 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1507 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1508 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1509 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1510 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1511 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1512 MOPT_EXT4_ONLY | MOPT_SET},
1513 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1514 MOPT_EXT4_ONLY | MOPT_CLEAR},
1515 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1516 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1517 {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1518 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1519 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1520 MOPT_EXT4_ONLY | MOPT_CLEAR},
1521 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1522 MOPT_EXT4_ONLY | MOPT_CLEAR},
1523 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1524 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1525 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1526 EXT4_MOUNT_JOURNAL_CHECKSUM),
1527 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1528 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1529 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1530 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1531 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1532 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1533 MOPT_NO_EXT2},
1534 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1535 MOPT_NO_EXT2},
1536 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1537 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1538 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1539 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1540 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1541 {Opt_commit, 0, MOPT_GTE0},
1542 {Opt_max_batch_time, 0, MOPT_GTE0},
1543 {Opt_min_batch_time, 0, MOPT_GTE0},
1544 {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1545 {Opt_init_itable, 0, MOPT_GTE0},
1546 {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1547 {Opt_stripe, 0, MOPT_GTE0},
1548 {Opt_resuid, 0, MOPT_GTE0},
1549 {Opt_resgid, 0, MOPT_GTE0},
1550 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1551 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1552 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1553 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1554 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1555 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1556 MOPT_NO_EXT2 | MOPT_DATAJ},
1557 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1558 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1559#ifdef CONFIG_EXT4_FS_POSIX_ACL
1560 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1561 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1562#else
1563 {Opt_acl, 0, MOPT_NOSUPPORT},
1564 {Opt_noacl, 0, MOPT_NOSUPPORT},
1565#endif
1566 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1567 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1568 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1569 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1570 MOPT_SET | MOPT_Q},
1571 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1572 MOPT_SET | MOPT_Q},
1573 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1574 MOPT_SET | MOPT_Q},
1575 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1576 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1577 MOPT_CLEAR | MOPT_Q},
1578 {Opt_usrjquota, 0, MOPT_Q},
1579 {Opt_grpjquota, 0, MOPT_Q},
1580 {Opt_offusrjquota, 0, MOPT_Q},
1581 {Opt_offgrpjquota, 0, MOPT_Q},
1582 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1583 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1584 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1585 {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1586 {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1587 {Opt_err, 0, 0}
1588};
1589
1590static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1591 substring_t *args, unsigned long *journal_devnum,
1592 unsigned int *journal_ioprio, int is_remount)
1593{
1594 struct ext4_sb_info *sbi = EXT4_SB(sb);
1595 const struct mount_opts *m;
1596 kuid_t uid;
1597 kgid_t gid;
1598 int arg = 0;
1599
1600#ifdef CONFIG_QUOTA
1601 if (token == Opt_usrjquota)
1602 return set_qf_name(sb, USRQUOTA, &args[0]);
1603 else if (token == Opt_grpjquota)
1604 return set_qf_name(sb, GRPQUOTA, &args[0]);
1605 else if (token == Opt_offusrjquota)
1606 return clear_qf_name(sb, USRQUOTA);
1607 else if (token == Opt_offgrpjquota)
1608 return clear_qf_name(sb, GRPQUOTA);
1609#endif
1610 switch (token) {
1611 case Opt_noacl:
1612 case Opt_nouser_xattr:
1613 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1614 break;
1615 case Opt_sb:
1616 return 1; /* handled by get_sb_block() */
1617 case Opt_removed:
1618 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1619 return 1;
1620 case Opt_abort:
1621 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1622 return 1;
1623 case Opt_i_version:
1624 sb->s_flags |= MS_I_VERSION;
1625 return 1;
1626 case Opt_lazytime:
1627 sb->s_flags |= MS_LAZYTIME;
1628 return 1;
1629 case Opt_nolazytime:
1630 sb->s_flags &= ~MS_LAZYTIME;
1631 return 1;
1632 }
1633
1634 for (m = ext4_mount_opts; m->token != Opt_err; m++)
1635 if (token == m->token)
1636 break;
1637
1638 if (m->token == Opt_err) {
1639 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1640 "or missing value", opt);
1641 return -1;
1642 }
1643
1644 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1645 ext4_msg(sb, KERN_ERR,
1646 "Mount option \"%s\" incompatible with ext2", opt);
1647 return -1;
1648 }
1649 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1650 ext4_msg(sb, KERN_ERR,
1651 "Mount option \"%s\" incompatible with ext3", opt);
1652 return -1;
1653 }
1654
1655 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1656 return -1;
1657 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1658 return -1;
1659 if (m->flags & MOPT_EXPLICIT) {
1660 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1661 set_opt2(sb, EXPLICIT_DELALLOC);
1662 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1663 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1664 } else
1665 return -1;
1666 }
1667 if (m->flags & MOPT_CLEAR_ERR)
1668 clear_opt(sb, ERRORS_MASK);
1669 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1670 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1671 "options when quota turned on");
1672 return -1;
1673 }
1674
1675 if (m->flags & MOPT_NOSUPPORT) {
1676 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1677 } else if (token == Opt_commit) {
1678 if (arg == 0)
1679 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1680 sbi->s_commit_interval = HZ * arg;
1681 } else if (token == Opt_max_batch_time) {
1682 sbi->s_max_batch_time = arg;
1683 } else if (token == Opt_min_batch_time) {
1684 sbi->s_min_batch_time = arg;
1685 } else if (token == Opt_inode_readahead_blks) {
1686 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1687 ext4_msg(sb, KERN_ERR,
1688 "EXT4-fs: inode_readahead_blks must be "
1689 "0 or a power of 2 smaller than 2^31");
1690 return -1;
1691 }
1692 sbi->s_inode_readahead_blks = arg;
1693 } else if (token == Opt_init_itable) {
1694 set_opt(sb, INIT_INODE_TABLE);
1695 if (!args->from)
1696 arg = EXT4_DEF_LI_WAIT_MULT;
1697 sbi->s_li_wait_mult = arg;
1698 } else if (token == Opt_max_dir_size_kb) {
1699 sbi->s_max_dir_size_kb = arg;
1700 } else if (token == Opt_stripe) {
1701 sbi->s_stripe = arg;
1702 } else if (token == Opt_resuid) {
1703 uid = make_kuid(current_user_ns(), arg);
1704 if (!uid_valid(uid)) {
1705 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1706 return -1;
1707 }
1708 sbi->s_resuid = uid;
1709 } else if (token == Opt_resgid) {
1710 gid = make_kgid(current_user_ns(), arg);
1711 if (!gid_valid(gid)) {
1712 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1713 return -1;
1714 }
1715 sbi->s_resgid = gid;
1716 } else if (token == Opt_journal_dev) {
1717 if (is_remount) {
1718 ext4_msg(sb, KERN_ERR,
1719 "Cannot specify journal on remount");
1720 return -1;
1721 }
1722 *journal_devnum = arg;
1723 } else if (token == Opt_journal_path) {
1724 char *journal_path;
1725 struct inode *journal_inode;
1726 struct path path;
1727 int error;
1728
1729 if (is_remount) {
1730 ext4_msg(sb, KERN_ERR,
1731 "Cannot specify journal on remount");
1732 return -1;
1733 }
1734 journal_path = match_strdup(&args[0]);
1735 if (!journal_path) {
1736 ext4_msg(sb, KERN_ERR, "error: could not dup "
1737 "journal device string");
1738 return -1;
1739 }
1740
1741 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1742 if (error) {
1743 ext4_msg(sb, KERN_ERR, "error: could not find "
1744 "journal device path: error %d", error);
1745 kfree(journal_path);
1746 return -1;
1747 }
1748
1749 journal_inode = d_inode(path.dentry);
1750 if (!S_ISBLK(journal_inode->i_mode)) {
1751 ext4_msg(sb, KERN_ERR, "error: journal path %s "
1752 "is not a block device", journal_path);
1753 path_put(&path);
1754 kfree(journal_path);
1755 return -1;
1756 }
1757
1758 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1759 path_put(&path);
1760 kfree(journal_path);
1761 } else if (token == Opt_journal_ioprio) {
1762 if (arg > 7) {
1763 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1764 " (must be 0-7)");
1765 return -1;
1766 }
1767 *journal_ioprio =
1768 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1769 } else if (token == Opt_test_dummy_encryption) {
1770#ifdef CONFIG_EXT4_FS_ENCRYPTION
1771 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1772 ext4_msg(sb, KERN_WARNING,
1773 "Test dummy encryption mode enabled");
1774#else
1775 ext4_msg(sb, KERN_WARNING,
1776 "Test dummy encryption mount option ignored");
1777#endif
1778 } else if (m->flags & MOPT_DATAJ) {
1779 if (is_remount) {
1780 if (!sbi->s_journal)
1781 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1782 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1783 ext4_msg(sb, KERN_ERR,
1784 "Cannot change data mode on remount");
1785 return -1;
1786 }
1787 } else {
1788 clear_opt(sb, DATA_FLAGS);
1789 sbi->s_mount_opt |= m->mount_opt;
1790 }
1791#ifdef CONFIG_QUOTA
1792 } else if (m->flags & MOPT_QFMT) {
1793 if (sb_any_quota_loaded(sb) &&
1794 sbi->s_jquota_fmt != m->mount_opt) {
1795 ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1796 "quota options when quota turned on");
1797 return -1;
1798 }
1799 if (ext4_has_feature_quota(sb)) {
1800 ext4_msg(sb, KERN_INFO,
1801 "Quota format mount options ignored "
1802 "when QUOTA feature is enabled");
1803 return 1;
1804 }
1805 sbi->s_jquota_fmt = m->mount_opt;
1806#endif
1807 } else if (token == Opt_dax) {
1808#ifdef CONFIG_FS_DAX
1809 ext4_msg(sb, KERN_WARNING,
1810 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1811 sbi->s_mount_opt |= m->mount_opt;
1812#else
1813 ext4_msg(sb, KERN_INFO, "dax option not supported");
1814 return -1;
1815#endif
1816 } else if (token == Opt_data_err_abort) {
1817 sbi->s_mount_opt |= m->mount_opt;
1818 } else if (token == Opt_data_err_ignore) {
1819 sbi->s_mount_opt &= ~m->mount_opt;
1820 } else {
1821 if (!args->from)
1822 arg = 1;
1823 if (m->flags & MOPT_CLEAR)
1824 arg = !arg;
1825 else if (unlikely(!(m->flags & MOPT_SET))) {
1826 ext4_msg(sb, KERN_WARNING,
1827 "buggy handling of option %s", opt);
1828 WARN_ON(1);
1829 return -1;
1830 }
1831 if (arg != 0)
1832 sbi->s_mount_opt |= m->mount_opt;
1833 else
1834 sbi->s_mount_opt &= ~m->mount_opt;
1835 }
1836 return 1;
1837}
1838
1839static int parse_options(char *options, struct super_block *sb,
1840 unsigned long *journal_devnum,
1841 unsigned int *journal_ioprio,
1842 int is_remount)
1843{
1844 struct ext4_sb_info *sbi = EXT4_SB(sb);
1845 char *p;
1846 substring_t args[MAX_OPT_ARGS];
1847 int token;
1848
1849 if (!options)
1850 return 1;
1851
1852 while ((p = strsep(&options, ",")) != NULL) {
1853 if (!*p)
1854 continue;
1855 /*
1856 * Initialize args struct so we know whether arg was
1857 * found; some options take optional arguments.
1858 */
1859 args[0].to = args[0].from = NULL;
1860 token = match_token(p, tokens, args);
1861 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1862 journal_ioprio, is_remount) < 0)
1863 return 0;
1864 }
1865#ifdef CONFIG_QUOTA
1866 /*
1867 * We do the test below only for project quotas. 'usrquota' and
1868 * 'grpquota' mount options are allowed even without quota feature
1869 * to support legacy quotas in quota files.
1870 */
1871 if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1872 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1873 "Cannot enable project quota enforcement.");
1874 return 0;
1875 }
1876 if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1877 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1878 clear_opt(sb, USRQUOTA);
1879
1880 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1881 clear_opt(sb, GRPQUOTA);
1882
1883 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1884 ext4_msg(sb, KERN_ERR, "old and new quota "
1885 "format mixing");
1886 return 0;
1887 }
1888
1889 if (!sbi->s_jquota_fmt) {
1890 ext4_msg(sb, KERN_ERR, "journaled quota format "
1891 "not specified");
1892 return 0;
1893 }
1894 }
1895#endif
1896 if (test_opt(sb, DIOREAD_NOLOCK)) {
1897 int blocksize =
1898 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1899
1900 if (blocksize < PAGE_SIZE) {
1901 ext4_msg(sb, KERN_ERR, "can't mount with "
1902 "dioread_nolock if block size != PAGE_SIZE");
1903 return 0;
1904 }
1905 }
1906 return 1;
1907}
1908
1909static inline void ext4_show_quota_options(struct seq_file *seq,
1910 struct super_block *sb)
1911{
1912#if defined(CONFIG_QUOTA)
1913 struct ext4_sb_info *sbi = EXT4_SB(sb);
1914
1915 if (sbi->s_jquota_fmt) {
1916 char *fmtname = "";
1917
1918 switch (sbi->s_jquota_fmt) {
1919 case QFMT_VFS_OLD:
1920 fmtname = "vfsold";
1921 break;
1922 case QFMT_VFS_V0:
1923 fmtname = "vfsv0";
1924 break;
1925 case QFMT_VFS_V1:
1926 fmtname = "vfsv1";
1927 break;
1928 }
1929 seq_printf(seq, ",jqfmt=%s", fmtname);
1930 }
1931
1932 if (sbi->s_qf_names[USRQUOTA])
1933 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1934
1935 if (sbi->s_qf_names[GRPQUOTA])
1936 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1937#endif
1938}
1939
1940static const char *token2str(int token)
1941{
1942 const struct match_token *t;
1943
1944 for (t = tokens; t->token != Opt_err; t++)
1945 if (t->token == token && !strchr(t->pattern, '='))
1946 break;
1947 return t->pattern;
1948}
1949
1950/*
1951 * Show an option if
1952 * - it's set to a non-default value OR
1953 * - if the per-sb default is different from the global default
1954 */
1955static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1956 int nodefs)
1957{
1958 struct ext4_sb_info *sbi = EXT4_SB(sb);
1959 struct ext4_super_block *es = sbi->s_es;
1960 int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1961 const struct mount_opts *m;
1962 char sep = nodefs ? '\n' : ',';
1963
1964#define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1965#define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1966
1967 if (sbi->s_sb_block != 1)
1968 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1969
1970 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1971 int want_set = m->flags & MOPT_SET;
1972 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1973 (m->flags & MOPT_CLEAR_ERR))
1974 continue;
1975 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1976 continue; /* skip if same as the default */
1977 if ((want_set &&
1978 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1979 (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1980 continue; /* select Opt_noFoo vs Opt_Foo */
1981 SEQ_OPTS_PRINT("%s", token2str(m->token));
1982 }
1983
1984 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1985 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1986 SEQ_OPTS_PRINT("resuid=%u",
1987 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1988 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1989 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1990 SEQ_OPTS_PRINT("resgid=%u",
1991 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1992 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1993 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1994 SEQ_OPTS_PUTS("errors=remount-ro");
1995 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1996 SEQ_OPTS_PUTS("errors=continue");
1997 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1998 SEQ_OPTS_PUTS("errors=panic");
1999 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2000 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2001 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2002 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2003 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2004 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2005 if (sb->s_flags & MS_I_VERSION)
2006 SEQ_OPTS_PUTS("i_version");
2007 if (nodefs || sbi->s_stripe)
2008 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2009 if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2010 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2011 SEQ_OPTS_PUTS("data=journal");
2012 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2013 SEQ_OPTS_PUTS("data=ordered");
2014 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2015 SEQ_OPTS_PUTS("data=writeback");
2016 }
2017 if (nodefs ||
2018 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2019 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2020 sbi->s_inode_readahead_blks);
2021
2022 if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2023 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2024 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2025 if (nodefs || sbi->s_max_dir_size_kb)
2026 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2027 if (test_opt(sb, DATA_ERR_ABORT))
2028 SEQ_OPTS_PUTS("data_err=abort");
2029
2030 ext4_show_quota_options(seq, sb);
2031 return 0;
2032}
2033
2034static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2035{
2036 return _ext4_show_options(seq, root->d_sb, 0);
2037}
2038
2039int ext4_seq_options_show(struct seq_file *seq, void *offset)
2040{
2041 struct super_block *sb = seq->private;
2042 int rc;
2043
2044 seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2045 rc = _ext4_show_options(seq, sb, 1);
2046 seq_puts(seq, "\n");
2047 return rc;
2048}
2049
2050static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2051 int read_only)
2052{
2053 struct ext4_sb_info *sbi = EXT4_SB(sb);
2054 int res = 0;
2055
2056 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2057 ext4_msg(sb, KERN_ERR, "revision level too high, "
2058 "forcing read-only mode");
2059 res = MS_RDONLY;
2060 }
2061 if (read_only)
2062 goto done;
2063 if (!(sbi->s_mount_state & EXT4_VALID_FS))
2064 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2065 "running e2fsck is recommended");
2066 else if (sbi->s_mount_state & EXT4_ERROR_FS)
2067 ext4_msg(sb, KERN_WARNING,
2068 "warning: mounting fs with errors, "
2069 "running e2fsck is recommended");
2070 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2071 le16_to_cpu(es->s_mnt_count) >=
2072 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2073 ext4_msg(sb, KERN_WARNING,
2074 "warning: maximal mount count reached, "
2075 "running e2fsck is recommended");
2076 else if (le32_to_cpu(es->s_checkinterval) &&
2077 (le32_to_cpu(es->s_lastcheck) +
2078 le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2079 ext4_msg(sb, KERN_WARNING,
2080 "warning: checktime reached, "
2081 "running e2fsck is recommended");
2082 if (!sbi->s_journal)
2083 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2084 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2085 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2086 le16_add_cpu(&es->s_mnt_count, 1);
2087 es->s_mtime = cpu_to_le32(get_seconds());
2088 ext4_update_dynamic_rev(sb);
2089 if (sbi->s_journal)
2090 ext4_set_feature_journal_needs_recovery(sb);
2091
2092 ext4_commit_super(sb, 1);
2093done:
2094 if (test_opt(sb, DEBUG))
2095 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2096 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2097 sb->s_blocksize,
2098 sbi->s_groups_count,
2099 EXT4_BLOCKS_PER_GROUP(sb),
2100 EXT4_INODES_PER_GROUP(sb),
2101 sbi->s_mount_opt, sbi->s_mount_opt2);
2102
2103 cleancache_init_fs(sb);
2104 return res;
2105}
2106
2107int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2108{
2109 struct ext4_sb_info *sbi = EXT4_SB(sb);
2110 struct flex_groups *new_groups;
2111 int size;
2112
2113 if (!sbi->s_log_groups_per_flex)
2114 return 0;
2115
2116 size = ext4_flex_group(sbi, ngroup - 1) + 1;
2117 if (size <= sbi->s_flex_groups_allocated)
2118 return 0;
2119
2120 size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2121 new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2122 if (!new_groups) {
2123 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2124 size / (int) sizeof(struct flex_groups));
2125 return -ENOMEM;
2126 }
2127
2128 if (sbi->s_flex_groups) {
2129 memcpy(new_groups, sbi->s_flex_groups,
2130 (sbi->s_flex_groups_allocated *
2131 sizeof(struct flex_groups)));
2132 kvfree(sbi->s_flex_groups);
2133 }
2134 sbi->s_flex_groups = new_groups;
2135 sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2136 return 0;
2137}
2138
2139static int ext4_fill_flex_info(struct super_block *sb)
2140{
2141 struct ext4_sb_info *sbi = EXT4_SB(sb);
2142 struct ext4_group_desc *gdp = NULL;
2143 ext4_group_t flex_group;
2144 int i, err;
2145
2146 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2147 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2148 sbi->s_log_groups_per_flex = 0;
2149 return 1;
2150 }
2151
2152 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2153 if (err)
2154 goto failed;
2155
2156 for (i = 0; i < sbi->s_groups_count; i++) {
2157 gdp = ext4_get_group_desc(sb, i, NULL);
2158
2159 flex_group = ext4_flex_group(sbi, i);
2160 atomic_add(ext4_free_inodes_count(sb, gdp),
2161 &sbi->s_flex_groups[flex_group].free_inodes);
2162 atomic64_add(ext4_free_group_clusters(sb, gdp),
2163 &sbi->s_flex_groups[flex_group].free_clusters);
2164 atomic_add(ext4_used_dirs_count(sb, gdp),
2165 &sbi->s_flex_groups[flex_group].used_dirs);
2166 }
2167
2168 return 1;
2169failed:
2170 return 0;
2171}
2172
2173static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2174 struct ext4_group_desc *gdp)
2175{
2176 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2177 __u16 crc = 0;
2178 __le32 le_group = cpu_to_le32(block_group);
2179 struct ext4_sb_info *sbi = EXT4_SB(sb);
2180
2181 if (ext4_has_metadata_csum(sbi->s_sb)) {
2182 /* Use new metadata_csum algorithm */
2183 __u32 csum32;
2184 __u16 dummy_csum = 0;
2185
2186 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2187 sizeof(le_group));
2188 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2189 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2190 sizeof(dummy_csum));
2191 offset += sizeof(dummy_csum);
2192 if (offset < sbi->s_desc_size)
2193 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2194 sbi->s_desc_size - offset);
2195
2196 crc = csum32 & 0xFFFF;
2197 goto out;
2198 }
2199
2200 /* old crc16 code */
2201 if (!ext4_has_feature_gdt_csum(sb))
2202 return 0;
2203
2204 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2205 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2206 crc = crc16(crc, (__u8 *)gdp, offset);
2207 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2208 /* for checksum of struct ext4_group_desc do the rest...*/
2209 if (ext4_has_feature_64bit(sb) &&
2210 offset < le16_to_cpu(sbi->s_es->s_desc_size))
2211 crc = crc16(crc, (__u8 *)gdp + offset,
2212 le16_to_cpu(sbi->s_es->s_desc_size) -
2213 offset);
2214
2215out:
2216 return cpu_to_le16(crc);
2217}
2218
2219int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2220 struct ext4_group_desc *gdp)
2221{
2222 if (ext4_has_group_desc_csum(sb) &&
2223 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2224 return 0;
2225
2226 return 1;
2227}
2228
2229void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2230 struct ext4_group_desc *gdp)
2231{
2232 if (!ext4_has_group_desc_csum(sb))
2233 return;
2234 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2235}
2236
2237/* Called at mount-time, super-block is locked */
2238static int ext4_check_descriptors(struct super_block *sb,
2239 ext4_fsblk_t sb_block,
2240 ext4_group_t *first_not_zeroed)
2241{
2242 struct ext4_sb_info *sbi = EXT4_SB(sb);
2243 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2244 ext4_fsblk_t last_block;
2245 ext4_fsblk_t block_bitmap;
2246 ext4_fsblk_t inode_bitmap;
2247 ext4_fsblk_t inode_table;
2248 int flexbg_flag = 0;
2249 ext4_group_t i, grp = sbi->s_groups_count;
2250
2251 if (ext4_has_feature_flex_bg(sb))
2252 flexbg_flag = 1;
2253
2254 ext4_debug("Checking group descriptors");
2255
2256 for (i = 0; i < sbi->s_groups_count; i++) {
2257 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2258
2259 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2260 last_block = ext4_blocks_count(sbi->s_es) - 1;
2261 else
2262 last_block = first_block +
2263 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2264
2265 if ((grp == sbi->s_groups_count) &&
2266 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2267 grp = i;
2268
2269 block_bitmap = ext4_block_bitmap(sb, gdp);
2270 if (block_bitmap == sb_block) {
2271 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2272 "Block bitmap for group %u overlaps "
2273 "superblock", i);
2274 }
2275 if (block_bitmap < first_block || block_bitmap > last_block) {
2276 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2277 "Block bitmap for group %u not in group "
2278 "(block %llu)!", i, block_bitmap);
2279 return 0;
2280 }
2281 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2282 if (inode_bitmap == sb_block) {
2283 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2284 "Inode bitmap for group %u overlaps "
2285 "superblock", i);
2286 }
2287 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2288 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2289 "Inode bitmap for group %u not in group "
2290 "(block %llu)!", i, inode_bitmap);
2291 return 0;
2292 }
2293 inode_table = ext4_inode_table(sb, gdp);
2294 if (inode_table == sb_block) {
2295 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2296 "Inode table for group %u overlaps "
2297 "superblock", i);
2298 }
2299 if (inode_table < first_block ||
2300 inode_table + sbi->s_itb_per_group - 1 > last_block) {
2301 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2302 "Inode table for group %u not in group "
2303 "(block %llu)!", i, inode_table);
2304 return 0;
2305 }
2306 ext4_lock_group(sb, i);
2307 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2308 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2309 "Checksum for group %u failed (%u!=%u)",
2310 i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2311 gdp)), le16_to_cpu(gdp->bg_checksum));
2312 if (!(sb->s_flags & MS_RDONLY)) {
2313 ext4_unlock_group(sb, i);
2314 return 0;
2315 }
2316 }
2317 ext4_unlock_group(sb, i);
2318 if (!flexbg_flag)
2319 first_block += EXT4_BLOCKS_PER_GROUP(sb);
2320 }
2321 if (NULL != first_not_zeroed)
2322 *first_not_zeroed = grp;
2323 return 1;
2324}
2325
2326/* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2327 * the superblock) which were deleted from all directories, but held open by
2328 * a process at the time of a crash. We walk the list and try to delete these
2329 * inodes at recovery time (only with a read-write filesystem).
2330 *
2331 * In order to keep the orphan inode chain consistent during traversal (in
2332 * case of crash during recovery), we link each inode into the superblock
2333 * orphan list_head and handle it the same way as an inode deletion during
2334 * normal operation (which journals the operations for us).
2335 *
2336 * We only do an iget() and an iput() on each inode, which is very safe if we
2337 * accidentally point at an in-use or already deleted inode. The worst that
2338 * can happen in this case is that we get a "bit already cleared" message from
2339 * ext4_free_inode(). The only reason we would point at a wrong inode is if
2340 * e2fsck was run on this filesystem, and it must have already done the orphan
2341 * inode cleanup for us, so we can safely abort without any further action.
2342 */
2343static void ext4_orphan_cleanup(struct super_block *sb,
2344 struct ext4_super_block *es)
2345{
2346 unsigned int s_flags = sb->s_flags;
2347 int ret, nr_orphans = 0, nr_truncates = 0;
2348#ifdef CONFIG_QUOTA
2349 int i;
2350#endif
2351 if (!es->s_last_orphan) {
2352 jbd_debug(4, "no orphan inodes to clean up\n");
2353 return;
2354 }
2355
2356 if (bdev_read_only(sb->s_bdev)) {
2357 ext4_msg(sb, KERN_ERR, "write access "
2358 "unavailable, skipping orphan cleanup");
2359 return;
2360 }
2361
2362 /* Check if feature set would not allow a r/w mount */
2363 if (!ext4_feature_set_ok(sb, 0)) {
2364 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2365 "unknown ROCOMPAT features");
2366 return;
2367 }
2368
2369 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2370 /* don't clear list on RO mount w/ errors */
2371 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2372 ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2373 "clearing orphan list.\n");
2374 es->s_last_orphan = 0;
2375 }
2376 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2377 return;
2378 }
2379
2380 if (s_flags & MS_RDONLY) {
2381 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2382 sb->s_flags &= ~MS_RDONLY;
2383 }
2384#ifdef CONFIG_QUOTA
2385 /* Needed for iput() to work correctly and not trash data */
2386 sb->s_flags |= MS_ACTIVE;
2387 /* Turn on quotas so that they are updated correctly */
2388 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2389 if (EXT4_SB(sb)->s_qf_names[i]) {
2390 int ret = ext4_quota_on_mount(sb, i);
2391 if (ret < 0)
2392 ext4_msg(sb, KERN_ERR,
2393 "Cannot turn on journaled "
2394 "quota: error %d", ret);
2395 }
2396 }
2397#endif
2398
2399 while (es->s_last_orphan) {
2400 struct inode *inode;
2401
2402 /*
2403 * We may have encountered an error during cleanup; if
2404 * so, skip the rest.
2405 */
2406 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2407 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2408 es->s_last_orphan = 0;
2409 break;
2410 }
2411
2412 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2413 if (IS_ERR(inode)) {
2414 es->s_last_orphan = 0;
2415 break;
2416 }
2417
2418 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2419 dquot_initialize(inode);
2420 if (inode->i_nlink) {
2421 if (test_opt(sb, DEBUG))
2422 ext4_msg(sb, KERN_DEBUG,
2423 "%s: truncating inode %lu to %lld bytes",
2424 __func__, inode->i_ino, inode->i_size);
2425 jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2426 inode->i_ino, inode->i_size);
2427 inode_lock(inode);
2428 truncate_inode_pages(inode->i_mapping, inode->i_size);
2429 ret = ext4_truncate(inode);
2430 if (ret)
2431 ext4_std_error(inode->i_sb, ret);
2432 inode_unlock(inode);
2433 nr_truncates++;
2434 } else {
2435 if (test_opt(sb, DEBUG))
2436 ext4_msg(sb, KERN_DEBUG,
2437 "%s: deleting unreferenced inode %lu",
2438 __func__, inode->i_ino);
2439 jbd_debug(2, "deleting unreferenced inode %lu\n",
2440 inode->i_ino);
2441 nr_orphans++;
2442 }
2443 iput(inode); /* The delete magic happens here! */
2444 }
2445
2446#define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2447
2448 if (nr_orphans)
2449 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2450 PLURAL(nr_orphans));
2451 if (nr_truncates)
2452 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2453 PLURAL(nr_truncates));
2454#ifdef CONFIG_QUOTA
2455 /* Turn quotas off */
2456 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2457 if (sb_dqopt(sb)->files[i])
2458 dquot_quota_off(sb, i);
2459 }
2460#endif
2461 sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2462}
2463
2464/*
2465 * Maximal extent format file size.
2466 * Resulting logical blkno at s_maxbytes must fit in our on-disk
2467 * extent format containers, within a sector_t, and within i_blocks
2468 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
2469 * so that won't be a limiting factor.
2470 *
2471 * However there is other limiting factor. We do store extents in the form
2472 * of starting block and length, hence the resulting length of the extent
2473 * covering maximum file size must fit into on-disk format containers as
2474 * well. Given that length is always by 1 unit bigger than max unit (because
2475 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2476 *
2477 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2478 */
2479static loff_t ext4_max_size(int blkbits, int has_huge_files)
2480{
2481 loff_t res;
2482 loff_t upper_limit = MAX_LFS_FILESIZE;
2483
2484 /* small i_blocks in vfs inode? */
2485 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2486 /*
2487 * CONFIG_LBDAF is not enabled implies the inode
2488 * i_block represent total blocks in 512 bytes
2489 * 32 == size of vfs inode i_blocks * 8
2490 */
2491 upper_limit = (1LL << 32) - 1;
2492
2493 /* total blocks in file system block size */
2494 upper_limit >>= (blkbits - 9);
2495 upper_limit <<= blkbits;
2496 }
2497
2498 /*
2499 * 32-bit extent-start container, ee_block. We lower the maxbytes
2500 * by one fs block, so ee_len can cover the extent of maximum file
2501 * size
2502 */
2503 res = (1LL << 32) - 1;
2504 res <<= blkbits;
2505
2506 /* Sanity check against vm- & vfs- imposed limits */
2507 if (res > upper_limit)
2508 res = upper_limit;
2509
2510 return res;
2511}
2512
2513/*
2514 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
2515 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2516 * We need to be 1 filesystem block less than the 2^48 sector limit.
2517 */
2518static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2519{
2520 loff_t res = EXT4_NDIR_BLOCKS;
2521 int meta_blocks;
2522 loff_t upper_limit;
2523 /* This is calculated to be the largest file size for a dense, block
2524 * mapped file such that the file's total number of 512-byte sectors,
2525 * including data and all indirect blocks, does not exceed (2^48 - 1).
2526 *
2527 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2528 * number of 512-byte sectors of the file.
2529 */
2530
2531 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2532 /*
2533 * !has_huge_files or CONFIG_LBDAF not enabled implies that
2534 * the inode i_block field represents total file blocks in
2535 * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2536 */
2537 upper_limit = (1LL << 32) - 1;
2538
2539 /* total blocks in file system block size */
2540 upper_limit >>= (bits - 9);
2541
2542 } else {
2543 /*
2544 * We use 48 bit ext4_inode i_blocks
2545 * With EXT4_HUGE_FILE_FL set the i_blocks
2546 * represent total number of blocks in
2547 * file system block size
2548 */
2549 upper_limit = (1LL << 48) - 1;
2550
2551 }
2552
2553 /* indirect blocks */
2554 meta_blocks = 1;
2555 /* double indirect blocks */
2556 meta_blocks += 1 + (1LL << (bits-2));
2557 /* tripple indirect blocks */
2558 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2559
2560 upper_limit -= meta_blocks;
2561 upper_limit <<= bits;
2562
2563 res += 1LL << (bits-2);
2564 res += 1LL << (2*(bits-2));
2565 res += 1LL << (3*(bits-2));
2566 res <<= bits;
2567 if (res > upper_limit)
2568 res = upper_limit;
2569
2570 if (res > MAX_LFS_FILESIZE)
2571 res = MAX_LFS_FILESIZE;
2572
2573 return res;
2574}
2575
2576static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2577 ext4_fsblk_t logical_sb_block, int nr)
2578{
2579 struct ext4_sb_info *sbi = EXT4_SB(sb);
2580 ext4_group_t bg, first_meta_bg;
2581 int has_super = 0;
2582
2583 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2584
2585 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2586 return logical_sb_block + nr + 1;
2587 bg = sbi->s_desc_per_block * nr;
2588 if (ext4_bg_has_super(sb, bg))
2589 has_super = 1;
2590
2591 /*
2592 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2593 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled
2594 * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2595 * compensate.
2596 */
2597 if (sb->s_blocksize == 1024 && nr == 0 &&
2598 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2599 has_super++;
2600
2601 return (has_super + ext4_group_first_block_no(sb, bg));
2602}
2603
2604/**
2605 * ext4_get_stripe_size: Get the stripe size.
2606 * @sbi: In memory super block info
2607 *
2608 * If we have specified it via mount option, then
2609 * use the mount option value. If the value specified at mount time is
2610 * greater than the blocks per group use the super block value.
2611 * If the super block value is greater than blocks per group return 0.
2612 * Allocator needs it be less than blocks per group.
2613 *
2614 */
2615static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2616{
2617 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2618 unsigned long stripe_width =
2619 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2620 int ret;
2621
2622 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2623 ret = sbi->s_stripe;
2624 else if (stripe_width <= sbi->s_blocks_per_group)
2625 ret = stripe_width;
2626 else if (stride <= sbi->s_blocks_per_group)
2627 ret = stride;
2628 else
2629 ret = 0;
2630
2631 /*
2632 * If the stripe width is 1, this makes no sense and
2633 * we set it to 0 to turn off stripe handling code.
2634 */
2635 if (ret <= 1)
2636 ret = 0;
2637
2638 return ret;
2639}
2640
2641/*
2642 * Check whether this filesystem can be mounted based on
2643 * the features present and the RDONLY/RDWR mount requested.
2644 * Returns 1 if this filesystem can be mounted as requested,
2645 * 0 if it cannot be.
2646 */
2647static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2648{
2649 if (ext4_has_unknown_ext4_incompat_features(sb)) {
2650 ext4_msg(sb, KERN_ERR,
2651 "Couldn't mount because of "
2652 "unsupported optional features (%x)",
2653 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2654 ~EXT4_FEATURE_INCOMPAT_SUPP));
2655 return 0;
2656 }
2657
2658 if (readonly)
2659 return 1;
2660
2661 if (ext4_has_feature_readonly(sb)) {
2662 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2663 sb->s_flags |= MS_RDONLY;
2664 return 1;
2665 }
2666
2667 /* Check that feature set is OK for a read-write mount */
2668 if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2669 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2670 "unsupported optional features (%x)",
2671 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2672 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2673 return 0;
2674 }
2675 /*
2676 * Large file size enabled file system can only be mounted
2677 * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2678 */
2679 if (ext4_has_feature_huge_file(sb)) {
2680 if (sizeof(blkcnt_t) < sizeof(u64)) {
2681 ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2682 "cannot be mounted RDWR without "
2683 "CONFIG_LBDAF");
2684 return 0;
2685 }
2686 }
2687 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2688 ext4_msg(sb, KERN_ERR,
2689 "Can't support bigalloc feature without "
2690 "extents feature\n");
2691 return 0;
2692 }
2693
2694#ifndef CONFIG_QUOTA
2695 if (ext4_has_feature_quota(sb) && !readonly) {
2696 ext4_msg(sb, KERN_ERR,
2697 "Filesystem with quota feature cannot be mounted RDWR "
2698 "without CONFIG_QUOTA");
2699 return 0;
2700 }
2701 if (ext4_has_feature_project(sb) && !readonly) {
2702 ext4_msg(sb, KERN_ERR,
2703 "Filesystem with project quota feature cannot be mounted RDWR "
2704 "without CONFIG_QUOTA");
2705 return 0;
2706 }
2707#endif /* CONFIG_QUOTA */
2708 return 1;
2709}
2710
2711/*
2712 * This function is called once a day if we have errors logged
2713 * on the file system
2714 */
2715static void print_daily_error_info(unsigned long arg)
2716{
2717 struct super_block *sb = (struct super_block *) arg;
2718 struct ext4_sb_info *sbi;
2719 struct ext4_super_block *es;
2720
2721 sbi = EXT4_SB(sb);
2722 es = sbi->s_es;
2723
2724 if (es->s_error_count)
2725 /* fsck newer than v1.41.13 is needed to clean this condition. */
2726 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2727 le32_to_cpu(es->s_error_count));
2728 if (es->s_first_error_time) {
2729 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2730 sb->s_id, le32_to_cpu(es->s_first_error_time),
2731 (int) sizeof(es->s_first_error_func),
2732 es->s_first_error_func,
2733 le32_to_cpu(es->s_first_error_line));
2734 if (es->s_first_error_ino)
2735 printk(KERN_CONT ": inode %u",
2736 le32_to_cpu(es->s_first_error_ino));
2737 if (es->s_first_error_block)
2738 printk(KERN_CONT ": block %llu", (unsigned long long)
2739 le64_to_cpu(es->s_first_error_block));
2740 printk(KERN_CONT "\n");
2741 }
2742 if (es->s_last_error_time) {
2743 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2744 sb->s_id, le32_to_cpu(es->s_last_error_time),
2745 (int) sizeof(es->s_last_error_func),
2746 es->s_last_error_func,
2747 le32_to_cpu(es->s_last_error_line));
2748 if (es->s_last_error_ino)
2749 printk(KERN_CONT ": inode %u",
2750 le32_to_cpu(es->s_last_error_ino));
2751 if (es->s_last_error_block)
2752 printk(KERN_CONT ": block %llu", (unsigned long long)
2753 le64_to_cpu(es->s_last_error_block));
2754 printk(KERN_CONT "\n");
2755 }
2756 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
2757}
2758
2759/* Find next suitable group and run ext4_init_inode_table */
2760static int ext4_run_li_request(struct ext4_li_request *elr)
2761{
2762 struct ext4_group_desc *gdp = NULL;
2763 ext4_group_t group, ngroups;
2764 struct super_block *sb;
2765 unsigned long timeout = 0;
2766 int ret = 0;
2767
2768 sb = elr->lr_super;
2769 ngroups = EXT4_SB(sb)->s_groups_count;
2770
2771 for (group = elr->lr_next_group; group < ngroups; group++) {
2772 gdp = ext4_get_group_desc(sb, group, NULL);
2773 if (!gdp) {
2774 ret = 1;
2775 break;
2776 }
2777
2778 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2779 break;
2780 }
2781
2782 if (group >= ngroups)
2783 ret = 1;
2784
2785 if (!ret) {
2786 timeout = jiffies;
2787 ret = ext4_init_inode_table(sb, group,
2788 elr->lr_timeout ? 0 : 1);
2789 if (elr->lr_timeout == 0) {
2790 timeout = (jiffies - timeout) *
2791 elr->lr_sbi->s_li_wait_mult;
2792 elr->lr_timeout = timeout;
2793 }
2794 elr->lr_next_sched = jiffies + elr->lr_timeout;
2795 elr->lr_next_group = group + 1;
2796 }
2797 return ret;
2798}
2799
2800/*
2801 * Remove lr_request from the list_request and free the
2802 * request structure. Should be called with li_list_mtx held
2803 */
2804static void ext4_remove_li_request(struct ext4_li_request *elr)
2805{
2806 struct ext4_sb_info *sbi;
2807
2808 if (!elr)
2809 return;
2810
2811 sbi = elr->lr_sbi;
2812
2813 list_del(&elr->lr_request);
2814 sbi->s_li_request = NULL;
2815 kfree(elr);
2816}
2817
2818static void ext4_unregister_li_request(struct super_block *sb)
2819{
2820 mutex_lock(&ext4_li_mtx);
2821 if (!ext4_li_info) {
2822 mutex_unlock(&ext4_li_mtx);
2823 return;
2824 }
2825
2826 mutex_lock(&ext4_li_info->li_list_mtx);
2827 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2828 mutex_unlock(&ext4_li_info->li_list_mtx);
2829 mutex_unlock(&ext4_li_mtx);
2830}
2831
2832static struct task_struct *ext4_lazyinit_task;
2833
2834/*
2835 * This is the function where ext4lazyinit thread lives. It walks
2836 * through the request list searching for next scheduled filesystem.
2837 * When such a fs is found, run the lazy initialization request
2838 * (ext4_rn_li_request) and keep track of the time spend in this
2839 * function. Based on that time we compute next schedule time of
2840 * the request. When walking through the list is complete, compute
2841 * next waking time and put itself into sleep.
2842 */
2843static int ext4_lazyinit_thread(void *arg)
2844{
2845 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2846 struct list_head *pos, *n;
2847 struct ext4_li_request *elr;
2848 unsigned long next_wakeup, cur;
2849
2850 BUG_ON(NULL == eli);
2851
2852cont_thread:
2853 while (true) {
2854 next_wakeup = MAX_JIFFY_OFFSET;
2855
2856 mutex_lock(&eli->li_list_mtx);
2857 if (list_empty(&eli->li_request_list)) {
2858 mutex_unlock(&eli->li_list_mtx);
2859 goto exit_thread;
2860 }
2861 list_for_each_safe(pos, n, &eli->li_request_list) {
2862 int err = 0;
2863 int progress = 0;
2864 elr = list_entry(pos, struct ext4_li_request,
2865 lr_request);
2866
2867 if (time_before(jiffies, elr->lr_next_sched)) {
2868 if (time_before(elr->lr_next_sched, next_wakeup))
2869 next_wakeup = elr->lr_next_sched;
2870 continue;
2871 }
2872 if (down_read_trylock(&elr->lr_super->s_umount)) {
2873 if (sb_start_write_trylock(elr->lr_super)) {
2874 progress = 1;
2875 /*
2876 * We hold sb->s_umount, sb can not
2877 * be removed from the list, it is
2878 * now safe to drop li_list_mtx
2879 */
2880 mutex_unlock(&eli->li_list_mtx);
2881 err = ext4_run_li_request(elr);
2882 sb_end_write(elr->lr_super);
2883 mutex_lock(&eli->li_list_mtx);
2884 n = pos->next;
2885 }
2886 up_read((&elr->lr_super->s_umount));
2887 }
2888 /* error, remove the lazy_init job */
2889 if (err) {
2890 ext4_remove_li_request(elr);
2891 continue;
2892 }
2893 if (!progress) {
2894 elr->lr_next_sched = jiffies +
2895 (prandom_u32()
2896 % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2897 }
2898 if (time_before(elr->lr_next_sched, next_wakeup))
2899 next_wakeup = elr->lr_next_sched;
2900 }
2901 mutex_unlock(&eli->li_list_mtx);
2902
2903 try_to_freeze();
2904
2905 cur = jiffies;
2906 if ((time_after_eq(cur, next_wakeup)) ||
2907 (MAX_JIFFY_OFFSET == next_wakeup)) {
2908 cond_resched();
2909 continue;
2910 }
2911
2912 schedule_timeout_interruptible(next_wakeup - cur);
2913
2914 if (kthread_should_stop()) {
2915 ext4_clear_request_list();
2916 goto exit_thread;
2917 }
2918 }
2919
2920exit_thread:
2921 /*
2922 * It looks like the request list is empty, but we need
2923 * to check it under the li_list_mtx lock, to prevent any
2924 * additions into it, and of course we should lock ext4_li_mtx
2925 * to atomically free the list and ext4_li_info, because at
2926 * this point another ext4 filesystem could be registering
2927 * new one.
2928 */
2929 mutex_lock(&ext4_li_mtx);
2930 mutex_lock(&eli->li_list_mtx);
2931 if (!list_empty(&eli->li_request_list)) {
2932 mutex_unlock(&eli->li_list_mtx);
2933 mutex_unlock(&ext4_li_mtx);
2934 goto cont_thread;
2935 }
2936 mutex_unlock(&eli->li_list_mtx);
2937 kfree(ext4_li_info);
2938 ext4_li_info = NULL;
2939 mutex_unlock(&ext4_li_mtx);
2940
2941 return 0;
2942}
2943
2944static void ext4_clear_request_list(void)
2945{
2946 struct list_head *pos, *n;
2947 struct ext4_li_request *elr;
2948
2949 mutex_lock(&ext4_li_info->li_list_mtx);
2950 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2951 elr = list_entry(pos, struct ext4_li_request,
2952 lr_request);
2953 ext4_remove_li_request(elr);
2954 }
2955 mutex_unlock(&ext4_li_info->li_list_mtx);
2956}
2957
2958static int ext4_run_lazyinit_thread(void)
2959{
2960 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2961 ext4_li_info, "ext4lazyinit");
2962 if (IS_ERR(ext4_lazyinit_task)) {
2963 int err = PTR_ERR(ext4_lazyinit_task);
2964 ext4_clear_request_list();
2965 kfree(ext4_li_info);
2966 ext4_li_info = NULL;
2967 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2968 "initialization thread\n",
2969 err);
2970 return err;
2971 }
2972 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2973 return 0;
2974}
2975
2976/*
2977 * Check whether it make sense to run itable init. thread or not.
2978 * If there is at least one uninitialized inode table, return
2979 * corresponding group number, else the loop goes through all
2980 * groups and return total number of groups.
2981 */
2982static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2983{
2984 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2985 struct ext4_group_desc *gdp = NULL;
2986
2987 for (group = 0; group < ngroups; group++) {
2988 gdp = ext4_get_group_desc(sb, group, NULL);
2989 if (!gdp)
2990 continue;
2991
2992 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2993 break;
2994 }
2995
2996 return group;
2997}
2998
2999static int ext4_li_info_new(void)
3000{
3001 struct ext4_lazy_init *eli = NULL;
3002
3003 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3004 if (!eli)
3005 return -ENOMEM;
3006
3007 INIT_LIST_HEAD(&eli->li_request_list);
3008 mutex_init(&eli->li_list_mtx);
3009
3010 eli->li_state |= EXT4_LAZYINIT_QUIT;
3011
3012 ext4_li_info = eli;
3013
3014 return 0;
3015}
3016
3017static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3018 ext4_group_t start)
3019{
3020 struct ext4_sb_info *sbi = EXT4_SB(sb);
3021 struct ext4_li_request *elr;
3022
3023 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3024 if (!elr)
3025 return NULL;
3026
3027 elr->lr_super = sb;
3028 elr->lr_sbi = sbi;
3029 elr->lr_next_group = start;
3030
3031 /*
3032 * Randomize first schedule time of the request to
3033 * spread the inode table initialization requests
3034 * better.
3035 */
3036 elr->lr_next_sched = jiffies + (prandom_u32() %
3037 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3038 return elr;
3039}
3040
3041int ext4_register_li_request(struct super_block *sb,
3042 ext4_group_t first_not_zeroed)
3043{
3044 struct ext4_sb_info *sbi = EXT4_SB(sb);
3045 struct ext4_li_request *elr = NULL;
3046 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3047 int ret = 0;
3048
3049 mutex_lock(&ext4_li_mtx);
3050 if (sbi->s_li_request != NULL) {
3051 /*
3052 * Reset timeout so it can be computed again, because
3053 * s_li_wait_mult might have changed.
3054 */
3055 sbi->s_li_request->lr_timeout = 0;
3056 goto out;
3057 }
3058
3059 if (first_not_zeroed == ngroups ||
3060 (sb->s_flags & MS_RDONLY) ||
3061 !test_opt(sb, INIT_INODE_TABLE))
3062 goto out;
3063
3064 elr = ext4_li_request_new(sb, first_not_zeroed);
3065 if (!elr) {
3066 ret = -ENOMEM;
3067 goto out;
3068 }
3069
3070 if (NULL == ext4_li_info) {
3071 ret = ext4_li_info_new();
3072 if (ret)
3073 goto out;
3074 }
3075
3076 mutex_lock(&ext4_li_info->li_list_mtx);
3077 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3078 mutex_unlock(&ext4_li_info->li_list_mtx);
3079
3080 sbi->s_li_request = elr;
3081 /*
3082 * set elr to NULL here since it has been inserted to
3083 * the request_list and the removal and free of it is
3084 * handled by ext4_clear_request_list from now on.
3085 */
3086 elr = NULL;
3087
3088 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3089 ret = ext4_run_lazyinit_thread();
3090 if (ret)
3091 goto out;
3092 }
3093out:
3094 mutex_unlock(&ext4_li_mtx);
3095 if (ret)
3096 kfree(elr);
3097 return ret;
3098}
3099
3100/*
3101 * We do not need to lock anything since this is called on
3102 * module unload.
3103 */
3104static void ext4_destroy_lazyinit_thread(void)
3105{
3106 /*
3107 * If thread exited earlier
3108 * there's nothing to be done.
3109 */
3110 if (!ext4_li_info || !ext4_lazyinit_task)
3111 return;
3112
3113 kthread_stop(ext4_lazyinit_task);
3114}
3115
3116static int set_journal_csum_feature_set(struct super_block *sb)
3117{
3118 int ret = 1;
3119 int compat, incompat;
3120 struct ext4_sb_info *sbi = EXT4_SB(sb);
3121
3122 if (ext4_has_metadata_csum(sb)) {
3123 /* journal checksum v3 */
3124 compat = 0;
3125 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3126 } else {
3127 /* journal checksum v1 */
3128 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3129 incompat = 0;
3130 }
3131
3132 jbd2_journal_clear_features(sbi->s_journal,
3133 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3134 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3135 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3136 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3137 ret = jbd2_journal_set_features(sbi->s_journal,
3138 compat, 0,
3139 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3140 incompat);
3141 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3142 ret = jbd2_journal_set_features(sbi->s_journal,
3143 compat, 0,
3144 incompat);
3145 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3146 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3147 } else {
3148 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3149 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3150 }
3151
3152 return ret;
3153}
3154
3155/*
3156 * Note: calculating the overhead so we can be compatible with
3157 * historical BSD practice is quite difficult in the face of
3158 * clusters/bigalloc. This is because multiple metadata blocks from
3159 * different block group can end up in the same allocation cluster.
3160 * Calculating the exact overhead in the face of clustered allocation
3161 * requires either O(all block bitmaps) in memory or O(number of block
3162 * groups**2) in time. We will still calculate the superblock for
3163 * older file systems --- and if we come across with a bigalloc file
3164 * system with zero in s_overhead_clusters the estimate will be close to
3165 * correct especially for very large cluster sizes --- but for newer
3166 * file systems, it's better to calculate this figure once at mkfs
3167 * time, and store it in the superblock. If the superblock value is
3168 * present (even for non-bigalloc file systems), we will use it.
3169 */
3170static int count_overhead(struct super_block *sb, ext4_group_t grp,
3171 char *buf)
3172{
3173 struct ext4_sb_info *sbi = EXT4_SB(sb);
3174 struct ext4_group_desc *gdp;
3175 ext4_fsblk_t first_block, last_block, b;
3176 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3177 int s, j, count = 0;
3178
3179 if (!ext4_has_feature_bigalloc(sb))
3180 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3181 sbi->s_itb_per_group + 2);
3182
3183 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3184 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3185 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3186 for (i = 0; i < ngroups; i++) {
3187 gdp = ext4_get_group_desc(sb, i, NULL);
3188 b = ext4_block_bitmap(sb, gdp);
3189 if (b >= first_block && b <= last_block) {
3190 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3191 count++;
3192 }
3193 b = ext4_inode_bitmap(sb, gdp);
3194 if (b >= first_block && b <= last_block) {
3195 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3196 count++;
3197 }
3198 b = ext4_inode_table(sb, gdp);
3199 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3200 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3201 int c = EXT4_B2C(sbi, b - first_block);
3202 ext4_set_bit(c, buf);
3203 count++;
3204 }
3205 if (i != grp)
3206 continue;
3207 s = 0;
3208 if (ext4_bg_has_super(sb, grp)) {
3209 ext4_set_bit(s++, buf);
3210 count++;
3211 }
3212 j = ext4_bg_num_gdb(sb, grp);
3213 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3214 ext4_error(sb, "Invalid number of block group "
3215 "descriptor blocks: %d", j);
3216 j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3217 }
3218 count += j;
3219 for (; j > 0; j--)
3220 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3221 }
3222 if (!count)
3223 return 0;
3224 return EXT4_CLUSTERS_PER_GROUP(sb) -
3225 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3226}
3227
3228/*
3229 * Compute the overhead and stash it in sbi->s_overhead
3230 */
3231int ext4_calculate_overhead(struct super_block *sb)
3232{
3233 struct ext4_sb_info *sbi = EXT4_SB(sb);
3234 struct ext4_super_block *es = sbi->s_es;
3235 struct inode *j_inode;
3236 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3237 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3238 ext4_fsblk_t overhead = 0;
3239 char *buf = (char *) get_zeroed_page(GFP_NOFS);
3240
3241 if (!buf)
3242 return -ENOMEM;
3243
3244 /*
3245 * Compute the overhead (FS structures). This is constant
3246 * for a given filesystem unless the number of block groups
3247 * changes so we cache the previous value until it does.
3248 */
3249
3250 /*
3251 * All of the blocks before first_data_block are overhead
3252 */
3253 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3254
3255 /*
3256 * Add the overhead found in each block group
3257 */
3258 for (i = 0; i < ngroups; i++) {
3259 int blks;
3260
3261 blks = count_overhead(sb, i, buf);
3262 overhead += blks;
3263 if (blks)
3264 memset(buf, 0, PAGE_SIZE);
3265 cond_resched();
3266 }
3267
3268 /*
3269 * Add the internal journal blocks whether the journal has been
3270 * loaded or not
3271 */
3272 if (sbi->s_journal && !sbi->journal_bdev)
3273 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3274 else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3275 j_inode = ext4_get_journal_inode(sb, j_inum);
3276 if (j_inode) {
3277 j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3278 overhead += EXT4_NUM_B2C(sbi, j_blocks);
3279 iput(j_inode);
3280 } else {
3281 ext4_msg(sb, KERN_ERR, "can't get journal size");
3282 }
3283 }
3284 sbi->s_overhead = overhead;
3285 smp_wmb();
3286 free_page((unsigned long) buf);
3287 return 0;
3288}
3289
3290static void ext4_set_resv_clusters(struct super_block *sb)
3291{
3292 ext4_fsblk_t resv_clusters;
3293 struct ext4_sb_info *sbi = EXT4_SB(sb);
3294
3295 /*
3296 * There's no need to reserve anything when we aren't using extents.
3297 * The space estimates are exact, there are no unwritten extents,
3298 * hole punching doesn't need new metadata... This is needed especially
3299 * to keep ext2/3 backward compatibility.
3300 */
3301 if (!ext4_has_feature_extents(sb))
3302 return;
3303 /*
3304 * By default we reserve 2% or 4096 clusters, whichever is smaller.
3305 * This should cover the situations where we can not afford to run
3306 * out of space like for example punch hole, or converting
3307 * unwritten extents in delalloc path. In most cases such
3308 * allocation would require 1, or 2 blocks, higher numbers are
3309 * very rare.
3310 */
3311 resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3312 sbi->s_cluster_bits);
3313
3314 do_div(resv_clusters, 50);
3315 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3316
3317 atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3318}
3319
3320static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3321{
3322 char *orig_data = kstrdup(data, GFP_KERNEL);
3323 struct buffer_head *bh;
3324 struct ext4_super_block *es = NULL;
3325 struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3326 ext4_fsblk_t block;
3327 ext4_fsblk_t sb_block = get_sb_block(&data);
3328 ext4_fsblk_t logical_sb_block;
3329 unsigned long offset = 0;
3330 unsigned long journal_devnum = 0;
3331 unsigned long def_mount_opts;
3332 struct inode *root;
3333 const char *descr;
3334 int ret = -ENOMEM;
3335 int blocksize, clustersize;
3336 unsigned int db_count;
3337 unsigned int i;
3338 int needs_recovery, has_huge_files, has_bigalloc;
3339 __u64 blocks_count;
3340 int err = 0;
3341 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3342 ext4_group_t first_not_zeroed;
3343
3344 if ((data && !orig_data) || !sbi)
3345 goto out_free_base;
3346
3347 sbi->s_blockgroup_lock =
3348 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3349 if (!sbi->s_blockgroup_lock)
3350 goto out_free_base;
3351
3352 sb->s_fs_info = sbi;
3353 sbi->s_sb = sb;
3354 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3355 sbi->s_sb_block = sb_block;
3356 if (sb->s_bdev->bd_part)
3357 sbi->s_sectors_written_start =
3358 part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3359
3360 /* Cleanup superblock name */
3361 strreplace(sb->s_id, '/', '!');
3362
3363 /* -EINVAL is default */
3364 ret = -EINVAL;
3365 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3366 if (!blocksize) {
3367 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3368 goto out_fail;
3369 }
3370
3371 /*
3372 * The ext4 superblock will not be buffer aligned for other than 1kB
3373 * block sizes. We need to calculate the offset from buffer start.
3374 */
3375 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3376 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3377 offset = do_div(logical_sb_block, blocksize);
3378 } else {
3379 logical_sb_block = sb_block;
3380 }
3381
3382 if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3383 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3384 goto out_fail;
3385 }
3386 /*
3387 * Note: s_es must be initialized as soon as possible because
3388 * some ext4 macro-instructions depend on its value
3389 */
3390 es = (struct ext4_super_block *) (bh->b_data + offset);
3391 sbi->s_es = es;
3392 sb->s_magic = le16_to_cpu(es->s_magic);
3393 if (sb->s_magic != EXT4_SUPER_MAGIC)
3394 goto cantfind_ext4;
3395 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3396
3397 /* Warn if metadata_csum and gdt_csum are both set. */
3398 if (ext4_has_feature_metadata_csum(sb) &&
3399 ext4_has_feature_gdt_csum(sb))
3400 ext4_warning(sb, "metadata_csum and uninit_bg are "
3401 "redundant flags; please run fsck.");
3402
3403 /* Check for a known checksum algorithm */
3404 if (!ext4_verify_csum_type(sb, es)) {
3405 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3406 "unknown checksum algorithm.");
3407 silent = 1;
3408 goto cantfind_ext4;
3409 }
3410
3411 /* Load the checksum driver */
3412 if (ext4_has_feature_metadata_csum(sb)) {
3413 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3414 if (IS_ERR(sbi->s_chksum_driver)) {
3415 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3416 ret = PTR_ERR(sbi->s_chksum_driver);
3417 sbi->s_chksum_driver = NULL;
3418 goto failed_mount;
3419 }
3420 }
3421
3422 /* Check superblock checksum */
3423 if (!ext4_superblock_csum_verify(sb, es)) {
3424 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3425 "invalid superblock checksum. Run e2fsck?");
3426 silent = 1;
3427 ret = -EFSBADCRC;
3428 goto cantfind_ext4;
3429 }
3430
3431 /* Precompute checksum seed for all metadata */
3432 if (ext4_has_feature_csum_seed(sb))
3433 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3434 else if (ext4_has_metadata_csum(sb))
3435 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3436 sizeof(es->s_uuid));
3437
3438 /* Set defaults before we parse the mount options */
3439 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3440 set_opt(sb, INIT_INODE_TABLE);
3441 if (def_mount_opts & EXT4_DEFM_DEBUG)
3442 set_opt(sb, DEBUG);
3443 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3444 set_opt(sb, GRPID);
3445 if (def_mount_opts & EXT4_DEFM_UID16)
3446 set_opt(sb, NO_UID32);
3447 /* xattr user namespace & acls are now defaulted on */
3448 set_opt(sb, XATTR_USER);
3449#ifdef CONFIG_EXT4_FS_POSIX_ACL
3450 set_opt(sb, POSIX_ACL);
3451#endif
3452 /* don't forget to enable journal_csum when metadata_csum is enabled. */
3453 if (ext4_has_metadata_csum(sb))
3454 set_opt(sb, JOURNAL_CHECKSUM);
3455
3456 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3457 set_opt(sb, JOURNAL_DATA);
3458 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3459 set_opt(sb, ORDERED_DATA);
3460 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3461 set_opt(sb, WRITEBACK_DATA);
3462
3463 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3464 set_opt(sb, ERRORS_PANIC);
3465 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3466 set_opt(sb, ERRORS_CONT);
3467 else
3468 set_opt(sb, ERRORS_RO);
3469 /* block_validity enabled by default; disable with noblock_validity */
3470 set_opt(sb, BLOCK_VALIDITY);
3471 if (def_mount_opts & EXT4_DEFM_DISCARD)
3472 set_opt(sb, DISCARD);
3473
3474 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3475 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3476 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3477 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3478 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3479
3480 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3481 set_opt(sb, BARRIER);
3482
3483 /*
3484 * enable delayed allocation by default
3485 * Use -o nodelalloc to turn it off
3486 */
3487 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3488 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3489 set_opt(sb, DELALLOC);
3490
3491 /*
3492 * set default s_li_wait_mult for lazyinit, for the case there is
3493 * no mount option specified.
3494 */
3495 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3496
3497 if (sbi->s_es->s_mount_opts[0]) {
3498 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3499 sizeof(sbi->s_es->s_mount_opts),
3500 GFP_KERNEL);
3501 if (!s_mount_opts)
3502 goto failed_mount;
3503 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3504 &journal_ioprio, 0)) {
3505 ext4_msg(sb, KERN_WARNING,
3506 "failed to parse options in superblock: %s",
3507 s_mount_opts);
3508 }
3509 kfree(s_mount_opts);
3510 }
3511 sbi->s_def_mount_opt = sbi->s_mount_opt;
3512 if (!parse_options((char *) data, sb, &journal_devnum,
3513 &journal_ioprio, 0))
3514 goto failed_mount;
3515
3516 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3517 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3518 "with data=journal disables delayed "
3519 "allocation and O_DIRECT support!\n");
3520 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3521 ext4_msg(sb, KERN_ERR, "can't mount with "
3522 "both data=journal and delalloc");
3523 goto failed_mount;
3524 }
3525 if (test_opt(sb, DIOREAD_NOLOCK)) {
3526 ext4_msg(sb, KERN_ERR, "can't mount with "
3527 "both data=journal and dioread_nolock");
3528 goto failed_mount;
3529 }
3530 if (test_opt(sb, DAX)) {
3531 ext4_msg(sb, KERN_ERR, "can't mount with "
3532 "both data=journal and dax");
3533 goto failed_mount;
3534 }
3535 if (ext4_has_feature_encrypt(sb)) {
3536 ext4_msg(sb, KERN_WARNING,
3537 "encrypted files will use data=ordered "
3538 "instead of data journaling mode");
3539 }
3540 if (test_opt(sb, DELALLOC))
3541 clear_opt(sb, DELALLOC);
3542 } else {
3543 sb->s_iflags |= SB_I_CGROUPWB;
3544 }
3545
3546 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3547 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3548
3549 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3550 (ext4_has_compat_features(sb) ||
3551 ext4_has_ro_compat_features(sb) ||
3552 ext4_has_incompat_features(sb)))
3553 ext4_msg(sb, KERN_WARNING,
3554 "feature flags set on rev 0 fs, "
3555 "running e2fsck is recommended");
3556
3557 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3558 set_opt2(sb, HURD_COMPAT);
3559 if (ext4_has_feature_64bit(sb)) {
3560 ext4_msg(sb, KERN_ERR,
3561 "The Hurd can't support 64-bit file systems");
3562 goto failed_mount;
3563 }
3564 }
3565
3566 if (IS_EXT2_SB(sb)) {
3567 if (ext2_feature_set_ok(sb))
3568 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3569 "using the ext4 subsystem");
3570 else {
3571 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3572 "to feature incompatibilities");
3573 goto failed_mount;
3574 }
3575 }
3576
3577 if (IS_EXT3_SB(sb)) {
3578 if (ext3_feature_set_ok(sb))
3579 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3580 "using the ext4 subsystem");
3581 else {
3582 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3583 "to feature incompatibilities");
3584 goto failed_mount;
3585 }
3586 }
3587
3588 /*
3589 * Check feature flags regardless of the revision level, since we
3590 * previously didn't change the revision level when setting the flags,
3591 * so there is a chance incompat flags are set on a rev 0 filesystem.
3592 */
3593 if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3594 goto failed_mount;
3595
3596 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3597 if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3598 blocksize > EXT4_MAX_BLOCK_SIZE) {
3599 ext4_msg(sb, KERN_ERR,
3600 "Unsupported filesystem blocksize %d (%d log_block_size)",
3601 blocksize, le32_to_cpu(es->s_log_block_size));
3602 goto failed_mount;
3603 }
3604 if (le32_to_cpu(es->s_log_block_size) >
3605 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3606 ext4_msg(sb, KERN_ERR,
3607 "Invalid log block size: %u",
3608 le32_to_cpu(es->s_log_block_size));
3609 goto failed_mount;
3610 }
3611
3612 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3613 ext4_msg(sb, KERN_ERR,
3614 "Number of reserved GDT blocks insanely large: %d",
3615 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3616 goto failed_mount;
3617 }
3618
3619 if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3620 err = bdev_dax_supported(sb, blocksize);
3621 if (err)
3622 goto failed_mount;
3623 }
3624
3625 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3626 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3627 es->s_encryption_level);
3628 goto failed_mount;
3629 }
3630
3631 if (sb->s_blocksize != blocksize) {
3632 /* Validate the filesystem blocksize */
3633 if (!sb_set_blocksize(sb, blocksize)) {
3634 ext4_msg(sb, KERN_ERR, "bad block size %d",
3635 blocksize);
3636 goto failed_mount;
3637 }
3638
3639 brelse(bh);
3640 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3641 offset = do_div(logical_sb_block, blocksize);
3642 bh = sb_bread_unmovable(sb, logical_sb_block);
3643 if (!bh) {
3644 ext4_msg(sb, KERN_ERR,
3645 "Can't read superblock on 2nd try");
3646 goto failed_mount;
3647 }
3648 es = (struct ext4_super_block *)(bh->b_data + offset);
3649 sbi->s_es = es;
3650 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3651 ext4_msg(sb, KERN_ERR,
3652 "Magic mismatch, very weird!");
3653 goto failed_mount;
3654 }
3655 }
3656
3657 has_huge_files = ext4_has_feature_huge_file(sb);
3658 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3659 has_huge_files);
3660 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3661
3662 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3663 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3664 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3665 } else {
3666 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3667 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3668 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3669 (!is_power_of_2(sbi->s_inode_size)) ||
3670 (sbi->s_inode_size > blocksize)) {
3671 ext4_msg(sb, KERN_ERR,
3672 "unsupported inode size: %d",
3673 sbi->s_inode_size);
3674 goto failed_mount;
3675 }
3676 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3677 sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3678 }
3679
3680 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3681 if (ext4_has_feature_64bit(sb)) {
3682 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3683 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3684 !is_power_of_2(sbi->s_desc_size)) {
3685 ext4_msg(sb, KERN_ERR,
3686 "unsupported descriptor size %lu",
3687 sbi->s_desc_size);
3688 goto failed_mount;
3689 }
3690 } else
3691 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3692
3693 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3694 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3695
3696 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3697 if (sbi->s_inodes_per_block == 0)
3698 goto cantfind_ext4;
3699 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3700 sbi->s_inodes_per_group > blocksize * 8) {
3701 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3702 sbi->s_blocks_per_group);
3703 goto failed_mount;
3704 }
3705 sbi->s_itb_per_group = sbi->s_inodes_per_group /
3706 sbi->s_inodes_per_block;
3707 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3708 sbi->s_sbh = bh;
3709 sbi->s_mount_state = le16_to_cpu(es->s_state);
3710 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3711 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3712
3713 for (i = 0; i < 4; i++)
3714 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3715 sbi->s_def_hash_version = es->s_def_hash_version;
3716 if (ext4_has_feature_dir_index(sb)) {
3717 i = le32_to_cpu(es->s_flags);
3718 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3719 sbi->s_hash_unsigned = 3;
3720 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3721#ifdef __CHAR_UNSIGNED__
3722 if (!(sb->s_flags & MS_RDONLY))
3723 es->s_flags |=
3724 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3725 sbi->s_hash_unsigned = 3;
3726#else
3727 if (!(sb->s_flags & MS_RDONLY))
3728 es->s_flags |=
3729 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3730#endif
3731 }
3732 }
3733
3734 /* Handle clustersize */
3735 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3736 has_bigalloc = ext4_has_feature_bigalloc(sb);
3737 if (has_bigalloc) {
3738 if (clustersize < blocksize) {
3739 ext4_msg(sb, KERN_ERR,
3740 "cluster size (%d) smaller than "
3741 "block size (%d)", clustersize, blocksize);
3742 goto failed_mount;
3743 }
3744 if (le32_to_cpu(es->s_log_cluster_size) >
3745 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3746 ext4_msg(sb, KERN_ERR,
3747 "Invalid log cluster size: %u",
3748 le32_to_cpu(es->s_log_cluster_size));
3749 goto failed_mount;
3750 }
3751 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3752 le32_to_cpu(es->s_log_block_size);
3753 sbi->s_clusters_per_group =
3754 le32_to_cpu(es->s_clusters_per_group);
3755 if (sbi->s_clusters_per_group > blocksize * 8) {
3756 ext4_msg(sb, KERN_ERR,
3757 "#clusters per group too big: %lu",
3758 sbi->s_clusters_per_group);
3759 goto failed_mount;
3760 }
3761 if (sbi->s_blocks_per_group !=
3762 (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3763 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3764 "clusters per group (%lu) inconsistent",
3765 sbi->s_blocks_per_group,
3766 sbi->s_clusters_per_group);
3767 goto failed_mount;
3768 }
3769 } else {
3770 if (clustersize != blocksize) {
3771 ext4_warning(sb, "fragment/cluster size (%d) != "
3772 "block size (%d)", clustersize,
3773 blocksize);
3774 clustersize = blocksize;
3775 }
3776 if (sbi->s_blocks_per_group > blocksize * 8) {
3777 ext4_msg(sb, KERN_ERR,
3778 "#blocks per group too big: %lu",
3779 sbi->s_blocks_per_group);
3780 goto failed_mount;
3781 }
3782 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3783 sbi->s_cluster_bits = 0;
3784 }
3785 sbi->s_cluster_ratio = clustersize / blocksize;
3786
3787 /* Do we have standard group size of clustersize * 8 blocks ? */
3788 if (sbi->s_blocks_per_group == clustersize << 3)
3789 set_opt2(sb, STD_GROUP_SIZE);
3790
3791 /*
3792 * Test whether we have more sectors than will fit in sector_t,
3793 * and whether the max offset is addressable by the page cache.
3794 */
3795 err = generic_check_addressable(sb->s_blocksize_bits,
3796 ext4_blocks_count(es));
3797 if (err) {
3798 ext4_msg(sb, KERN_ERR, "filesystem"
3799 " too large to mount safely on this system");
3800 if (sizeof(sector_t) < 8)
3801 ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3802 goto failed_mount;
3803 }
3804
3805 if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3806 goto cantfind_ext4;
3807
3808 /* check blocks count against device size */
3809 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3810 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3811 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3812 "exceeds size of device (%llu blocks)",
3813 ext4_blocks_count(es), blocks_count);
3814 goto failed_mount;
3815 }
3816
3817 /*
3818 * It makes no sense for the first data block to be beyond the end
3819 * of the filesystem.
3820 */
3821 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3822 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3823 "block %u is beyond end of filesystem (%llu)",
3824 le32_to_cpu(es->s_first_data_block),
3825 ext4_blocks_count(es));
3826 goto failed_mount;
3827 }
3828 blocks_count = (ext4_blocks_count(es) -
3829 le32_to_cpu(es->s_first_data_block) +
3830 EXT4_BLOCKS_PER_GROUP(sb) - 1);
3831 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3832 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3833 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3834 "(block count %llu, first data block %u, "
3835 "blocks per group %lu)", sbi->s_groups_count,
3836 ext4_blocks_count(es),
3837 le32_to_cpu(es->s_first_data_block),
3838 EXT4_BLOCKS_PER_GROUP(sb));
3839 goto failed_mount;
3840 }
3841 sbi->s_groups_count = blocks_count;
3842 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3843 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3844 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3845 EXT4_DESC_PER_BLOCK(sb);
3846 if (ext4_has_feature_meta_bg(sb)) {
3847 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3848 ext4_msg(sb, KERN_WARNING,
3849 "first meta block group too large: %u "
3850 "(group descriptor block count %u)",
3851 le32_to_cpu(es->s_first_meta_bg), db_count);
3852 goto failed_mount;
3853 }
3854 }
3855 sbi->s_group_desc = ext4_kvmalloc(db_count *
3856 sizeof(struct buffer_head *),
3857 GFP_KERNEL);
3858 if (sbi->s_group_desc == NULL) {
3859 ext4_msg(sb, KERN_ERR, "not enough memory");
3860 ret = -ENOMEM;
3861 goto failed_mount;
3862 }
3863
3864 bgl_lock_init(sbi->s_blockgroup_lock);
3865
3866 for (i = 0; i < db_count; i++) {
3867 block = descriptor_loc(sb, logical_sb_block, i);
3868 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3869 if (!sbi->s_group_desc[i]) {
3870 ext4_msg(sb, KERN_ERR,
3871 "can't read group descriptor %d", i);
3872 db_count = i;
3873 goto failed_mount2;
3874 }
3875 }
3876 if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3877 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3878 ret = -EFSCORRUPTED;
3879 goto failed_mount2;
3880 }
3881
3882 sbi->s_gdb_count = db_count;
3883 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3884 spin_lock_init(&sbi->s_next_gen_lock);
3885
3886 setup_timer(&sbi->s_err_report, print_daily_error_info,
3887 (unsigned long) sb);
3888
3889 /* Register extent status tree shrinker */
3890 if (ext4_es_register_shrinker(sbi))
3891 goto failed_mount3;
3892
3893 sbi->s_stripe = ext4_get_stripe_size(sbi);
3894 sbi->s_extent_max_zeroout_kb = 32;
3895
3896 /*
3897 * set up enough so that it can read an inode
3898 */
3899 sb->s_op = &ext4_sops;
3900 sb->s_export_op = &ext4_export_ops;
3901 sb->s_xattr = ext4_xattr_handlers;
3902 sb->s_cop = &ext4_cryptops;
3903#ifdef CONFIG_QUOTA
3904 sb->dq_op = &ext4_quota_operations;
3905 if (ext4_has_feature_quota(sb))
3906 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3907 else
3908 sb->s_qcop = &ext4_qctl_operations;
3909 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3910#endif
3911 memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3912
3913 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3914 mutex_init(&sbi->s_orphan_lock);
3915
3916 sb->s_root = NULL;
3917
3918 needs_recovery = (es->s_last_orphan != 0 ||
3919 ext4_has_feature_journal_needs_recovery(sb));
3920
3921 if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3922 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3923 goto failed_mount3a;
3924
3925 /*
3926 * The first inode we look at is the journal inode. Don't try
3927 * root first: it may be modified in the journal!
3928 */
3929 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3930 err = ext4_load_journal(sb, es, journal_devnum);
3931 if (err)
3932 goto failed_mount3a;
3933 } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3934 ext4_has_feature_journal_needs_recovery(sb)) {
3935 ext4_msg(sb, KERN_ERR, "required journal recovery "
3936 "suppressed and not mounted read-only");
3937 goto failed_mount_wq;
3938 } else {
3939 /* Nojournal mode, all journal mount options are illegal */
3940 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3941 ext4_msg(sb, KERN_ERR, "can't mount with "
3942 "journal_checksum, fs mounted w/o journal");
3943 goto failed_mount_wq;
3944 }
3945 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3946 ext4_msg(sb, KERN_ERR, "can't mount with "
3947 "journal_async_commit, fs mounted w/o journal");
3948 goto failed_mount_wq;
3949 }
3950 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3951 ext4_msg(sb, KERN_ERR, "can't mount with "
3952 "commit=%lu, fs mounted w/o journal",
3953 sbi->s_commit_interval / HZ);
3954 goto failed_mount_wq;
3955 }
3956 if (EXT4_MOUNT_DATA_FLAGS &
3957 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3958 ext4_msg(sb, KERN_ERR, "can't mount with "
3959 "data=, fs mounted w/o journal");
3960 goto failed_mount_wq;
3961 }
3962 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3963 clear_opt(sb, JOURNAL_CHECKSUM);
3964 clear_opt(sb, DATA_FLAGS);
3965 sbi->s_journal = NULL;
3966 needs_recovery = 0;
3967 goto no_journal;
3968 }
3969
3970 if (ext4_has_feature_64bit(sb) &&
3971 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3972 JBD2_FEATURE_INCOMPAT_64BIT)) {
3973 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3974 goto failed_mount_wq;
3975 }
3976
3977 if (!set_journal_csum_feature_set(sb)) {
3978 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3979 "feature set");
3980 goto failed_mount_wq;
3981 }
3982
3983 /* We have now updated the journal if required, so we can
3984 * validate the data journaling mode. */
3985 switch (test_opt(sb, DATA_FLAGS)) {
3986 case 0:
3987 /* No mode set, assume a default based on the journal
3988 * capabilities: ORDERED_DATA if the journal can
3989 * cope, else JOURNAL_DATA
3990 */
3991 if (jbd2_journal_check_available_features
3992 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3993 set_opt(sb, ORDERED_DATA);
3994 else
3995 set_opt(sb, JOURNAL_DATA);
3996 break;
3997
3998 case EXT4_MOUNT_ORDERED_DATA:
3999 case EXT4_MOUNT_WRITEBACK_DATA:
4000 if (!jbd2_journal_check_available_features
4001 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4002 ext4_msg(sb, KERN_ERR, "Journal does not support "
4003 "requested data journaling mode");
4004 goto failed_mount_wq;
4005 }
4006 default:
4007 break;
4008 }
4009
4010 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4011 test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4012 ext4_msg(sb, KERN_ERR, "can't mount with "
4013 "journal_async_commit in data=ordered mode");
4014 goto failed_mount_wq;
4015 }
4016
4017 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4018
4019 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4020
4021no_journal:
4022 sbi->s_mb_cache = ext4_xattr_create_cache();
4023 if (!sbi->s_mb_cache) {
4024 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4025 goto failed_mount_wq;
4026 }
4027
4028 if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4029 (blocksize != PAGE_SIZE)) {
4030 ext4_msg(sb, KERN_ERR,
4031 "Unsupported blocksize for fs encryption");
4032 goto failed_mount_wq;
4033 }
4034
4035 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4036 !ext4_has_feature_encrypt(sb)) {
4037 ext4_set_feature_encrypt(sb);
4038 ext4_commit_super(sb, 1);
4039 }
4040
4041 /*
4042 * Get the # of file system overhead blocks from the
4043 * superblock if present.
4044 */
4045 if (es->s_overhead_clusters)
4046 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4047 else {
4048 err = ext4_calculate_overhead(sb);
4049 if (err)
4050 goto failed_mount_wq;
4051 }
4052
4053 /*
4054 * The maximum number of concurrent works can be high and
4055 * concurrency isn't really necessary. Limit it to 1.
4056 */
4057 EXT4_SB(sb)->rsv_conversion_wq =
4058 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4059 if (!EXT4_SB(sb)->rsv_conversion_wq) {
4060 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4061 ret = -ENOMEM;
4062 goto failed_mount4;
4063 }
4064
4065 /*
4066 * The jbd2_journal_load will have done any necessary log recovery,
4067 * so we can safely mount the rest of the filesystem now.
4068 */
4069
4070 root = ext4_iget(sb, EXT4_ROOT_INO);
4071 if (IS_ERR(root)) {
4072 ext4_msg(sb, KERN_ERR, "get root inode failed");
4073 ret = PTR_ERR(root);
4074 root = NULL;
4075 goto failed_mount4;
4076 }
4077 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4078 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4079 iput(root);
4080 goto failed_mount4;
4081 }
4082 sb->s_root = d_make_root(root);
4083 if (!sb->s_root) {
4084 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4085 ret = -ENOMEM;
4086 goto failed_mount4;
4087 }
4088
4089 if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4090 sb->s_flags |= MS_RDONLY;
4091
4092 /* determine the minimum size of new large inodes, if present */
4093 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4094 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4095 EXT4_GOOD_OLD_INODE_SIZE;
4096 if (ext4_has_feature_extra_isize(sb)) {
4097 if (sbi->s_want_extra_isize <
4098 le16_to_cpu(es->s_want_extra_isize))
4099 sbi->s_want_extra_isize =
4100 le16_to_cpu(es->s_want_extra_isize);
4101 if (sbi->s_want_extra_isize <
4102 le16_to_cpu(es->s_min_extra_isize))
4103 sbi->s_want_extra_isize =
4104 le16_to_cpu(es->s_min_extra_isize);
4105 }
4106 }
4107 /* Check if enough inode space is available */
4108 if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4109 sbi->s_inode_size) {
4110 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4111 EXT4_GOOD_OLD_INODE_SIZE;
4112 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4113 "available");
4114 }
4115
4116 ext4_set_resv_clusters(sb);
4117
4118 err = ext4_setup_system_zone(sb);
4119 if (err) {
4120 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4121 "zone (%d)", err);
4122 goto failed_mount4a;
4123 }
4124
4125 ext4_ext_init(sb);
4126 err = ext4_mb_init(sb);
4127 if (err) {
4128 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4129 err);
4130 goto failed_mount5;
4131 }
4132
4133 block = ext4_count_free_clusters(sb);
4134 ext4_free_blocks_count_set(sbi->s_es,
4135 EXT4_C2B(sbi, block));
4136 err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4137 GFP_KERNEL);
4138 if (!err) {
4139 unsigned long freei = ext4_count_free_inodes(sb);
4140 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4141 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4142 GFP_KERNEL);
4143 }
4144 if (!err)
4145 err = percpu_counter_init(&sbi->s_dirs_counter,
4146 ext4_count_dirs(sb), GFP_KERNEL);
4147 if (!err)
4148 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4149 GFP_KERNEL);
4150 if (!err)
4151 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4152
4153 if (err) {
4154 ext4_msg(sb, KERN_ERR, "insufficient memory");
4155 goto failed_mount6;
4156 }
4157
4158 if (ext4_has_feature_flex_bg(sb))
4159 if (!ext4_fill_flex_info(sb)) {
4160 ext4_msg(sb, KERN_ERR,
4161 "unable to initialize "
4162 "flex_bg meta info!");
4163 goto failed_mount6;
4164 }
4165
4166 err = ext4_register_li_request(sb, first_not_zeroed);
4167 if (err)
4168 goto failed_mount6;
4169
4170 err = ext4_register_sysfs(sb);
4171 if (err)
4172 goto failed_mount7;
4173
4174#ifdef CONFIG_QUOTA
4175 /* Enable quota usage during mount. */
4176 if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4177 err = ext4_enable_quotas(sb);
4178 if (err)
4179 goto failed_mount8;
4180 }
4181#endif /* CONFIG_QUOTA */
4182
4183 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4184 ext4_orphan_cleanup(sb, es);
4185 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4186 if (needs_recovery) {
4187 ext4_msg(sb, KERN_INFO, "recovery complete");
4188 ext4_mark_recovery_complete(sb, es);
4189 }
4190 if (EXT4_SB(sb)->s_journal) {
4191 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4192 descr = " journalled data mode";
4193 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4194 descr = " ordered data mode";
4195 else
4196 descr = " writeback data mode";
4197 } else
4198 descr = "out journal";
4199
4200 if (test_opt(sb, DISCARD)) {
4201 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4202 if (!blk_queue_discard(q))
4203 ext4_msg(sb, KERN_WARNING,
4204 "mounting with \"discard\" option, but "
4205 "the device does not support discard");
4206 }
4207
4208 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4209 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4210 "Opts: %.*s%s%s", descr,
4211 (int) sizeof(sbi->s_es->s_mount_opts),
4212 sbi->s_es->s_mount_opts,
4213 *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4214
4215 if (es->s_error_count)
4216 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4217
4218 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4219 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4220 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4221 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4222
4223 kfree(orig_data);
4224#ifdef CONFIG_EXT4_FS_ENCRYPTION
4225 memcpy(sbi->key_prefix, EXT4_KEY_DESC_PREFIX,
4226 EXT4_KEY_DESC_PREFIX_SIZE);
4227 sbi->key_prefix_size = EXT4_KEY_DESC_PREFIX_SIZE;
4228#endif
4229 return 0;
4230
4231cantfind_ext4:
4232 if (!silent)
4233 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4234 goto failed_mount;
4235
4236#ifdef CONFIG_QUOTA
4237failed_mount8:
4238 ext4_unregister_sysfs(sb);
4239#endif
4240failed_mount7:
4241 ext4_unregister_li_request(sb);
4242failed_mount6:
4243 ext4_mb_release(sb);
4244 if (sbi->s_flex_groups)
4245 kvfree(sbi->s_flex_groups);
4246 percpu_counter_destroy(&sbi->s_freeclusters_counter);
4247 percpu_counter_destroy(&sbi->s_freeinodes_counter);
4248 percpu_counter_destroy(&sbi->s_dirs_counter);
4249 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4250failed_mount5:
4251 ext4_ext_release(sb);
4252 ext4_release_system_zone(sb);
4253failed_mount4a:
4254 dput(sb->s_root);
4255 sb->s_root = NULL;
4256failed_mount4:
4257 ext4_msg(sb, KERN_ERR, "mount failed");
4258 if (EXT4_SB(sb)->rsv_conversion_wq)
4259 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4260failed_mount_wq:
4261 if (sbi->s_mb_cache) {
4262 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4263 sbi->s_mb_cache = NULL;
4264 }
4265 if (sbi->s_journal) {
4266 jbd2_journal_destroy(sbi->s_journal);
4267 sbi->s_journal = NULL;
4268 }
4269failed_mount3a:
4270 ext4_es_unregister_shrinker(sbi);
4271failed_mount3:
4272 del_timer_sync(&sbi->s_err_report);
4273 if (sbi->s_mmp_tsk)
4274 kthread_stop(sbi->s_mmp_tsk);
4275failed_mount2:
4276 for (i = 0; i < db_count; i++)
4277 brelse(sbi->s_group_desc[i]);
4278 kvfree(sbi->s_group_desc);
4279failed_mount:
4280 if (sbi->s_chksum_driver)
4281 crypto_free_shash(sbi->s_chksum_driver);
4282#ifdef CONFIG_QUOTA
4283 for (i = 0; i < EXT4_MAXQUOTAS; i++)
4284 kfree(sbi->s_qf_names[i]);
4285#endif
4286 ext4_blkdev_remove(sbi);
4287 brelse(bh);
4288out_fail:
4289 sb->s_fs_info = NULL;
4290 kfree(sbi->s_blockgroup_lock);
4291out_free_base:
4292 kfree(sbi);
4293 kfree(orig_data);
4294 return err ? err : ret;
4295}
4296
4297/*
4298 * Setup any per-fs journal parameters now. We'll do this both on
4299 * initial mount, once the journal has been initialised but before we've
4300 * done any recovery; and again on any subsequent remount.
4301 */
4302static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4303{
4304 struct ext4_sb_info *sbi = EXT4_SB(sb);
4305
4306 journal->j_commit_interval = sbi->s_commit_interval;
4307 journal->j_min_batch_time = sbi->s_min_batch_time;
4308 journal->j_max_batch_time = sbi->s_max_batch_time;
4309
4310 write_lock(&journal->j_state_lock);
4311 if (test_opt(sb, BARRIER))
4312 journal->j_flags |= JBD2_BARRIER;
4313 else
4314 journal->j_flags &= ~JBD2_BARRIER;
4315 if (test_opt(sb, DATA_ERR_ABORT))
4316 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4317 else
4318 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4319 write_unlock(&journal->j_state_lock);
4320}
4321
4322static struct inode *ext4_get_journal_inode(struct super_block *sb,
4323 unsigned int journal_inum)
4324{
4325 struct inode *journal_inode;
4326
4327 /*
4328 * Test for the existence of a valid inode on disk. Bad things
4329 * happen if we iget() an unused inode, as the subsequent iput()
4330 * will try to delete it.
4331 */
4332 journal_inode = ext4_iget(sb, journal_inum);
4333 if (IS_ERR(journal_inode)) {
4334 ext4_msg(sb, KERN_ERR, "no journal found");
4335 return NULL;
4336 }
4337 if (!journal_inode->i_nlink) {
4338 make_bad_inode(journal_inode);
4339 iput(journal_inode);
4340 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4341 return NULL;
4342 }
4343
4344 jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4345 journal_inode, journal_inode->i_size);
4346 if (!S_ISREG(journal_inode->i_mode)) {
4347 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4348 iput(journal_inode);
4349 return NULL;
4350 }
4351 return journal_inode;
4352}
4353
4354static journal_t *ext4_get_journal(struct super_block *sb,
4355 unsigned int journal_inum)
4356{
4357 struct inode *journal_inode;
4358 journal_t *journal;
4359
4360 BUG_ON(!ext4_has_feature_journal(sb));
4361
4362 journal_inode = ext4_get_journal_inode(sb, journal_inum);
4363 if (!journal_inode)
4364 return NULL;
4365
4366 journal = jbd2_journal_init_inode(journal_inode);
4367 if (!journal) {
4368 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4369 iput(journal_inode);
4370 return NULL;
4371 }
4372 journal->j_private = sb;
4373 ext4_init_journal_params(sb, journal);
4374 return journal;
4375}
4376
4377static journal_t *ext4_get_dev_journal(struct super_block *sb,
4378 dev_t j_dev)
4379{
4380 struct buffer_head *bh;
4381 journal_t *journal;
4382 ext4_fsblk_t start;
4383 ext4_fsblk_t len;
4384 int hblock, blocksize;
4385 ext4_fsblk_t sb_block;
4386 unsigned long offset;
4387 struct ext4_super_block *es;
4388 struct block_device *bdev;
4389
4390 BUG_ON(!ext4_has_feature_journal(sb));
4391
4392 bdev = ext4_blkdev_get(j_dev, sb);
4393 if (bdev == NULL)
4394 return NULL;
4395
4396 blocksize = sb->s_blocksize;
4397 hblock = bdev_logical_block_size(bdev);
4398 if (blocksize < hblock) {
4399 ext4_msg(sb, KERN_ERR,
4400 "blocksize too small for journal device");
4401 goto out_bdev;
4402 }
4403
4404 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4405 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4406 set_blocksize(bdev, blocksize);
4407 if (!(bh = __bread(bdev, sb_block, blocksize))) {
4408 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4409 "external journal");
4410 goto out_bdev;
4411 }
4412
4413 es = (struct ext4_super_block *) (bh->b_data + offset);
4414 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4415 !(le32_to_cpu(es->s_feature_incompat) &
4416 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4417 ext4_msg(sb, KERN_ERR, "external journal has "
4418 "bad superblock");
4419 brelse(bh);
4420 goto out_bdev;
4421 }
4422
4423 if ((le32_to_cpu(es->s_feature_ro_compat) &
4424 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4425 es->s_checksum != ext4_superblock_csum(sb, es)) {
4426 ext4_msg(sb, KERN_ERR, "external journal has "
4427 "corrupt superblock");
4428 brelse(bh);
4429 goto out_bdev;
4430 }
4431
4432 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4433 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4434 brelse(bh);
4435 goto out_bdev;
4436 }
4437
4438 len = ext4_blocks_count(es);
4439 start = sb_block + 1;
4440 brelse(bh); /* we're done with the superblock */
4441
4442 journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4443 start, len, blocksize);
4444 if (!journal) {
4445 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4446 goto out_bdev;
4447 }
4448 journal->j_private = sb;
4449 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4450 wait_on_buffer(journal->j_sb_buffer);
4451 if (!buffer_uptodate(journal->j_sb_buffer)) {
4452 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4453 goto out_journal;
4454 }
4455 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4456 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4457 "user (unsupported) - %d",
4458 be32_to_cpu(journal->j_superblock->s_nr_users));
4459 goto out_journal;
4460 }
4461 EXT4_SB(sb)->journal_bdev = bdev;
4462 ext4_init_journal_params(sb, journal);
4463 return journal;
4464
4465out_journal:
4466 jbd2_journal_destroy(journal);
4467out_bdev:
4468 ext4_blkdev_put(bdev);
4469 return NULL;
4470}
4471
4472static int ext4_load_journal(struct super_block *sb,
4473 struct ext4_super_block *es,
4474 unsigned long journal_devnum)
4475{
4476 journal_t *journal;
4477 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4478 dev_t journal_dev;
4479 int err = 0;
4480 int really_read_only;
4481
4482 BUG_ON(!ext4_has_feature_journal(sb));
4483
4484 if (journal_devnum &&
4485 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4486 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4487 "numbers have changed");
4488 journal_dev = new_decode_dev(journal_devnum);
4489 } else
4490 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4491
4492 really_read_only = bdev_read_only(sb->s_bdev);
4493
4494 /*
4495 * Are we loading a blank journal or performing recovery after a
4496 * crash? For recovery, we need to check in advance whether we
4497 * can get read-write access to the device.
4498 */
4499 if (ext4_has_feature_journal_needs_recovery(sb)) {
4500 if (sb->s_flags & MS_RDONLY) {
4501 ext4_msg(sb, KERN_INFO, "INFO: recovery "
4502 "required on readonly filesystem");
4503 if (really_read_only) {
4504 ext4_msg(sb, KERN_ERR, "write access "
4505 "unavailable, cannot proceed");
4506 return -EROFS;
4507 }
4508 ext4_msg(sb, KERN_INFO, "write access will "
4509 "be enabled during recovery");
4510 }
4511 }
4512
4513 if (journal_inum && journal_dev) {
4514 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4515 "and inode journals!");
4516 return -EINVAL;
4517 }
4518
4519 if (journal_inum) {
4520 if (!(journal = ext4_get_journal(sb, journal_inum)))
4521 return -EINVAL;
4522 } else {
4523 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4524 return -EINVAL;
4525 }
4526
4527 if (!(journal->j_flags & JBD2_BARRIER))
4528 ext4_msg(sb, KERN_INFO, "barriers disabled");
4529
4530 if (!ext4_has_feature_journal_needs_recovery(sb))
4531 err = jbd2_journal_wipe(journal, !really_read_only);
4532 if (!err) {
4533 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4534 if (save)
4535 memcpy(save, ((char *) es) +
4536 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4537 err = jbd2_journal_load(journal);
4538 if (save)
4539 memcpy(((char *) es) + EXT4_S_ERR_START,
4540 save, EXT4_S_ERR_LEN);
4541 kfree(save);
4542 }
4543
4544 if (err) {
4545 ext4_msg(sb, KERN_ERR, "error loading journal");
4546 jbd2_journal_destroy(journal);
4547 return err;
4548 }
4549
4550 EXT4_SB(sb)->s_journal = journal;
4551 ext4_clear_journal_err(sb, es);
4552
4553 if (!really_read_only && journal_devnum &&
4554 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4555 es->s_journal_dev = cpu_to_le32(journal_devnum);
4556
4557 /* Make sure we flush the recovery flag to disk. */
4558 ext4_commit_super(sb, 1);
4559 }
4560
4561 return 0;
4562}
4563
4564static int ext4_commit_super(struct super_block *sb, int sync)
4565{
4566 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4567 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4568 int error = 0;
4569
4570 if (!sbh || block_device_ejected(sb))
4571 return error;
4572 /*
4573 * If the file system is mounted read-only, don't update the
4574 * superblock write time. This avoids updating the superblock
4575 * write time when we are mounting the root file system
4576 * read/only but we need to replay the journal; at that point,
4577 * for people who are east of GMT and who make their clock
4578 * tick in localtime for Windows bug-for-bug compatibility,
4579 * the clock is set in the future, and this will cause e2fsck
4580 * to complain and force a full file system check.
4581 */
4582 if (!(sb->s_flags & MS_RDONLY))
4583 es->s_wtime = cpu_to_le32(get_seconds());
4584 if (sb->s_bdev->bd_part)
4585 es->s_kbytes_written =
4586 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4587 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4588 EXT4_SB(sb)->s_sectors_written_start) >> 1));
4589 else
4590 es->s_kbytes_written =
4591 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4592 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4593 ext4_free_blocks_count_set(es,
4594 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4595 &EXT4_SB(sb)->s_freeclusters_counter)));
4596 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4597 es->s_free_inodes_count =
4598 cpu_to_le32(percpu_counter_sum_positive(
4599 &EXT4_SB(sb)->s_freeinodes_counter));
4600 BUFFER_TRACE(sbh, "marking dirty");
4601 ext4_superblock_csum_set(sb);
4602 if (sync)
4603 lock_buffer(sbh);
4604 if (buffer_write_io_error(sbh)) {
4605 /*
4606 * Oh, dear. A previous attempt to write the
4607 * superblock failed. This could happen because the
4608 * USB device was yanked out. Or it could happen to
4609 * be a transient write error and maybe the block will
4610 * be remapped. Nothing we can do but to retry the
4611 * write and hope for the best.
4612 */
4613 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4614 "superblock detected");
4615 clear_buffer_write_io_error(sbh);
4616 set_buffer_uptodate(sbh);
4617 }
4618 mark_buffer_dirty(sbh);
4619 if (sync) {
4620 unlock_buffer(sbh);
4621 error = __sync_dirty_buffer(sbh,
4622 test_opt(sb, BARRIER) ? REQ_FUA : REQ_SYNC);
4623 if (error)
4624 return error;
4625
4626 error = buffer_write_io_error(sbh);
4627 if (error) {
4628 ext4_msg(sb, KERN_ERR, "I/O error while writing "
4629 "superblock");
4630 clear_buffer_write_io_error(sbh);
4631 set_buffer_uptodate(sbh);
4632 }
4633 }
4634 return error;
4635}
4636
4637/*
4638 * Have we just finished recovery? If so, and if we are mounting (or
4639 * remounting) the filesystem readonly, then we will end up with a
4640 * consistent fs on disk. Record that fact.
4641 */
4642static void ext4_mark_recovery_complete(struct super_block *sb,
4643 struct ext4_super_block *es)
4644{
4645 journal_t *journal = EXT4_SB(sb)->s_journal;
4646
4647 if (!ext4_has_feature_journal(sb)) {
4648 BUG_ON(journal != NULL);
4649 return;
4650 }
4651 jbd2_journal_lock_updates(journal);
4652 if (jbd2_journal_flush(journal) < 0)
4653 goto out;
4654
4655 if (ext4_has_feature_journal_needs_recovery(sb) &&
4656 sb->s_flags & MS_RDONLY) {
4657 ext4_clear_feature_journal_needs_recovery(sb);
4658 ext4_commit_super(sb, 1);
4659 }
4660
4661out:
4662 jbd2_journal_unlock_updates(journal);
4663}
4664
4665/*
4666 * If we are mounting (or read-write remounting) a filesystem whose journal
4667 * has recorded an error from a previous lifetime, move that error to the
4668 * main filesystem now.
4669 */
4670static void ext4_clear_journal_err(struct super_block *sb,
4671 struct ext4_super_block *es)
4672{
4673 journal_t *journal;
4674 int j_errno;
4675 const char *errstr;
4676
4677 BUG_ON(!ext4_has_feature_journal(sb));
4678
4679 journal = EXT4_SB(sb)->s_journal;
4680
4681 /*
4682 * Now check for any error status which may have been recorded in the
4683 * journal by a prior ext4_error() or ext4_abort()
4684 */
4685
4686 j_errno = jbd2_journal_errno(journal);
4687 if (j_errno) {
4688 char nbuf[16];
4689
4690 errstr = ext4_decode_error(sb, j_errno, nbuf);
4691 ext4_warning(sb, "Filesystem error recorded "
4692 "from previous mount: %s", errstr);
4693 ext4_warning(sb, "Marking fs in need of filesystem check.");
4694
4695 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4696 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4697 ext4_commit_super(sb, 1);
4698
4699 jbd2_journal_clear_err(journal);
4700 jbd2_journal_update_sb_errno(journal);
4701 }
4702}
4703
4704/*
4705 * Force the running and committing transactions to commit,
4706 * and wait on the commit.
4707 */
4708int ext4_force_commit(struct super_block *sb)
4709{
4710 journal_t *journal;
4711
4712 if (sb->s_flags & MS_RDONLY)
4713 return 0;
4714
4715 journal = EXT4_SB(sb)->s_journal;
4716 return ext4_journal_force_commit(journal);
4717}
4718
4719static int ext4_sync_fs(struct super_block *sb, int wait)
4720{
4721 int ret = 0;
4722 tid_t target;
4723 bool needs_barrier = false;
4724 struct ext4_sb_info *sbi = EXT4_SB(sb);
4725
4726 trace_ext4_sync_fs(sb, wait);
4727 flush_workqueue(sbi->rsv_conversion_wq);
4728 /*
4729 * Writeback quota in non-journalled quota case - journalled quota has
4730 * no dirty dquots
4731 */
4732 dquot_writeback_dquots(sb, -1);
4733 /*
4734 * Data writeback is possible w/o journal transaction, so barrier must
4735 * being sent at the end of the function. But we can skip it if
4736 * transaction_commit will do it for us.
4737 */
4738 if (sbi->s_journal) {
4739 target = jbd2_get_latest_transaction(sbi->s_journal);
4740 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4741 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4742 needs_barrier = true;
4743
4744 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4745 if (wait)
4746 ret = jbd2_log_wait_commit(sbi->s_journal,
4747 target);
4748 }
4749 } else if (wait && test_opt(sb, BARRIER))
4750 needs_barrier = true;
4751 if (needs_barrier) {
4752 int err;
4753 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4754 if (!ret)
4755 ret = err;
4756 }
4757
4758 return ret;
4759}
4760
4761/*
4762 * LVM calls this function before a (read-only) snapshot is created. This
4763 * gives us a chance to flush the journal completely and mark the fs clean.
4764 *
4765 * Note that only this function cannot bring a filesystem to be in a clean
4766 * state independently. It relies on upper layer to stop all data & metadata
4767 * modifications.
4768 */
4769static int ext4_freeze(struct super_block *sb)
4770{
4771 int error = 0;
4772 journal_t *journal;
4773
4774 if (sb->s_flags & MS_RDONLY)
4775 return 0;
4776
4777 journal = EXT4_SB(sb)->s_journal;
4778
4779 if (journal) {
4780 /* Now we set up the journal barrier. */
4781 jbd2_journal_lock_updates(journal);
4782
4783 /*
4784 * Don't clear the needs_recovery flag if we failed to
4785 * flush the journal.
4786 */
4787 error = jbd2_journal_flush(journal);
4788 if (error < 0)
4789 goto out;
4790
4791 /* Journal blocked and flushed, clear needs_recovery flag. */
4792 ext4_clear_feature_journal_needs_recovery(sb);
4793 }
4794
4795 error = ext4_commit_super(sb, 1);
4796out:
4797 if (journal)
4798 /* we rely on upper layer to stop further updates */
4799 jbd2_journal_unlock_updates(journal);
4800 return error;
4801}
4802
4803/*
4804 * Called by LVM after the snapshot is done. We need to reset the RECOVER
4805 * flag here, even though the filesystem is not technically dirty yet.
4806 */
4807static int ext4_unfreeze(struct super_block *sb)
4808{
4809 if (sb->s_flags & MS_RDONLY)
4810 return 0;
4811
4812 if (EXT4_SB(sb)->s_journal) {
4813 /* Reset the needs_recovery flag before the fs is unlocked. */
4814 ext4_set_feature_journal_needs_recovery(sb);
4815 }
4816
4817 ext4_commit_super(sb, 1);
4818 return 0;
4819}
4820
4821/*
4822 * Structure to save mount options for ext4_remount's benefit
4823 */
4824struct ext4_mount_options {
4825 unsigned long s_mount_opt;
4826 unsigned long s_mount_opt2;
4827 kuid_t s_resuid;
4828 kgid_t s_resgid;
4829 unsigned long s_commit_interval;
4830 u32 s_min_batch_time, s_max_batch_time;
4831#ifdef CONFIG_QUOTA
4832 int s_jquota_fmt;
4833 char *s_qf_names[EXT4_MAXQUOTAS];
4834#endif
4835};
4836
4837static int ext4_remount(struct super_block *sb, int *flags, char *data)
4838{
4839 struct ext4_super_block *es;
4840 struct ext4_sb_info *sbi = EXT4_SB(sb);
4841 unsigned long old_sb_flags;
4842 struct ext4_mount_options old_opts;
4843 int enable_quota = 0;
4844 ext4_group_t g;
4845 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4846 int err = 0;
4847#ifdef CONFIG_QUOTA
4848 int i, j;
4849#endif
4850 char *orig_data = kstrdup(data, GFP_KERNEL);
4851
4852 /* Store the original options */
4853 old_sb_flags = sb->s_flags;
4854 old_opts.s_mount_opt = sbi->s_mount_opt;
4855 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4856 old_opts.s_resuid = sbi->s_resuid;
4857 old_opts.s_resgid = sbi->s_resgid;
4858 old_opts.s_commit_interval = sbi->s_commit_interval;
4859 old_opts.s_min_batch_time = sbi->s_min_batch_time;
4860 old_opts.s_max_batch_time = sbi->s_max_batch_time;
4861#ifdef CONFIG_QUOTA
4862 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4863 for (i = 0; i < EXT4_MAXQUOTAS; i++)
4864 if (sbi->s_qf_names[i]) {
4865 old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4866 GFP_KERNEL);
4867 if (!old_opts.s_qf_names[i]) {
4868 for (j = 0; j < i; j++)
4869 kfree(old_opts.s_qf_names[j]);
4870 kfree(orig_data);
4871 return -ENOMEM;
4872 }
4873 } else
4874 old_opts.s_qf_names[i] = NULL;
4875#endif
4876 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4877 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4878
4879 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4880 err = -EINVAL;
4881 goto restore_opts;
4882 }
4883
4884 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4885 test_opt(sb, JOURNAL_CHECKSUM)) {
4886 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4887 "during remount not supported; ignoring");
4888 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4889 }
4890
4891 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4892 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4893 ext4_msg(sb, KERN_ERR, "can't mount with "
4894 "both data=journal and delalloc");
4895 err = -EINVAL;
4896 goto restore_opts;
4897 }
4898 if (test_opt(sb, DIOREAD_NOLOCK)) {
4899 ext4_msg(sb, KERN_ERR, "can't mount with "
4900 "both data=journal and dioread_nolock");
4901 err = -EINVAL;
4902 goto restore_opts;
4903 }
4904 if (test_opt(sb, DAX)) {
4905 ext4_msg(sb, KERN_ERR, "can't mount with "
4906 "both data=journal and dax");
4907 err = -EINVAL;
4908 goto restore_opts;
4909 }
4910 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
4911 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4912 ext4_msg(sb, KERN_ERR, "can't mount with "
4913 "journal_async_commit in data=ordered mode");
4914 err = -EINVAL;
4915 goto restore_opts;
4916 }
4917 }
4918
4919 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4920 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4921 "dax flag with busy inodes while remounting");
4922 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4923 }
4924
4925 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4926 ext4_abort(sb, "Abort forced by user");
4927
4928 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4929 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4930
4931 es = sbi->s_es;
4932
4933 if (sbi->s_journal) {
4934 ext4_init_journal_params(sb, sbi->s_journal);
4935 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4936 }
4937
4938 if (*flags & MS_LAZYTIME)
4939 sb->s_flags |= MS_LAZYTIME;
4940
4941 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4942 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4943 err = -EROFS;
4944 goto restore_opts;
4945 }
4946
4947 if (*flags & MS_RDONLY) {
4948 err = sync_filesystem(sb);
4949 if (err < 0)
4950 goto restore_opts;
4951 err = dquot_suspend(sb, -1);
4952 if (err < 0)
4953 goto restore_opts;
4954
4955 /*
4956 * First of all, the unconditional stuff we have to do
4957 * to disable replay of the journal when we next remount
4958 */
4959 sb->s_flags |= MS_RDONLY;
4960
4961 /*
4962 * OK, test if we are remounting a valid rw partition
4963 * readonly, and if so set the rdonly flag and then
4964 * mark the partition as valid again.
4965 */
4966 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4967 (sbi->s_mount_state & EXT4_VALID_FS))
4968 es->s_state = cpu_to_le16(sbi->s_mount_state);
4969
4970 if (sbi->s_journal)
4971 ext4_mark_recovery_complete(sb, es);
4972 } else {
4973 /* Make sure we can mount this feature set readwrite */
4974 if (ext4_has_feature_readonly(sb) ||
4975 !ext4_feature_set_ok(sb, 0)) {
4976 err = -EROFS;
4977 goto restore_opts;
4978 }
4979 /*
4980 * Make sure the group descriptor checksums
4981 * are sane. If they aren't, refuse to remount r/w.
4982 */
4983 for (g = 0; g < sbi->s_groups_count; g++) {
4984 struct ext4_group_desc *gdp =
4985 ext4_get_group_desc(sb, g, NULL);
4986
4987 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4988 ext4_msg(sb, KERN_ERR,
4989 "ext4_remount: Checksum for group %u failed (%u!=%u)",
4990 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4991 le16_to_cpu(gdp->bg_checksum));
4992 err = -EFSBADCRC;
4993 goto restore_opts;
4994 }
4995 }
4996
4997 /*
4998 * If we have an unprocessed orphan list hanging
4999 * around from a previously readonly bdev mount,
5000 * require a full umount/remount for now.
5001 */
5002 if (es->s_last_orphan) {
5003 ext4_msg(sb, KERN_WARNING, "Couldn't "
5004 "remount RDWR because of unprocessed "
5005 "orphan inode list. Please "
5006 "umount/remount instead");
5007 err = -EINVAL;
5008 goto restore_opts;
5009 }
5010
5011 /*
5012 * Mounting a RDONLY partition read-write, so reread
5013 * and store the current valid flag. (It may have
5014 * been changed by e2fsck since we originally mounted
5015 * the partition.)
5016 */
5017 if (sbi->s_journal)
5018 ext4_clear_journal_err(sb, es);
5019 sbi->s_mount_state = le16_to_cpu(es->s_state);
5020 if (!ext4_setup_super(sb, es, 0))
5021 sb->s_flags &= ~MS_RDONLY;
5022 if (ext4_has_feature_mmp(sb))
5023 if (ext4_multi_mount_protect(sb,
5024 le64_to_cpu(es->s_mmp_block))) {
5025 err = -EROFS;
5026 goto restore_opts;
5027 }
5028 enable_quota = 1;
5029 }
5030 }
5031
5032 /*
5033 * Reinitialize lazy itable initialization thread based on
5034 * current settings
5035 */
5036 if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5037 ext4_unregister_li_request(sb);
5038 else {
5039 ext4_group_t first_not_zeroed;
5040 first_not_zeroed = ext4_has_uninit_itable(sb);
5041 ext4_register_li_request(sb, first_not_zeroed);
5042 }
5043
5044 ext4_setup_system_zone(sb);
5045 if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5046 ext4_commit_super(sb, 1);
5047
5048#ifdef CONFIG_QUOTA
5049 /* Release old quota file names */
5050 for (i = 0; i < EXT4_MAXQUOTAS; i++)
5051 kfree(old_opts.s_qf_names[i]);
5052 if (enable_quota) {
5053 if (sb_any_quota_suspended(sb))
5054 dquot_resume(sb, -1);
5055 else if (ext4_has_feature_quota(sb)) {
5056 err = ext4_enable_quotas(sb);
5057 if (err)
5058 goto restore_opts;
5059 }
5060 }
5061#endif
5062
5063 *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5064 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5065 kfree(orig_data);
5066 return 0;
5067
5068restore_opts:
5069 sb->s_flags = old_sb_flags;
5070 sbi->s_mount_opt = old_opts.s_mount_opt;
5071 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5072 sbi->s_resuid = old_opts.s_resuid;
5073 sbi->s_resgid = old_opts.s_resgid;
5074 sbi->s_commit_interval = old_opts.s_commit_interval;
5075 sbi->s_min_batch_time = old_opts.s_min_batch_time;
5076 sbi->s_max_batch_time = old_opts.s_max_batch_time;
5077#ifdef CONFIG_QUOTA
5078 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5079 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5080 kfree(sbi->s_qf_names[i]);
5081 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5082 }
5083#endif
5084 kfree(orig_data);
5085 return err;
5086}
5087
5088#ifdef CONFIG_QUOTA
5089static int ext4_statfs_project(struct super_block *sb,
5090 kprojid_t projid, struct kstatfs *buf)
5091{
5092 struct kqid qid;
5093 struct dquot *dquot;
5094 u64 limit;
5095 u64 curblock;
5096
5097 qid = make_kqid_projid(projid);
5098 dquot = dqget(sb, qid);
5099 if (IS_ERR(dquot))
5100 return PTR_ERR(dquot);
5101 spin_lock(&dq_data_lock);
5102
5103 limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5104 dquot->dq_dqb.dqb_bsoftlimit :
5105 dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5106 if (limit && buf->f_blocks > limit) {
5107 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5108 buf->f_blocks = limit;
5109 buf->f_bfree = buf->f_bavail =
5110 (buf->f_blocks > curblock) ?
5111 (buf->f_blocks - curblock) : 0;
5112 }
5113
5114 limit = dquot->dq_dqb.dqb_isoftlimit ?
5115 dquot->dq_dqb.dqb_isoftlimit :
5116 dquot->dq_dqb.dqb_ihardlimit;
5117 if (limit && buf->f_files > limit) {
5118 buf->f_files = limit;
5119 buf->f_ffree =
5120 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5121 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5122 }
5123
5124 spin_unlock(&dq_data_lock);
5125 dqput(dquot);
5126 return 0;
5127}
5128#endif
5129
5130static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5131{
5132 struct super_block *sb = dentry->d_sb;
5133 struct ext4_sb_info *sbi = EXT4_SB(sb);
5134 struct ext4_super_block *es = sbi->s_es;
5135 ext4_fsblk_t overhead = 0, resv_blocks;
5136 u64 fsid;
5137 s64 bfree;
5138 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5139
5140 if (!test_opt(sb, MINIX_DF))
5141 overhead = sbi->s_overhead;
5142
5143 buf->f_type = EXT4_SUPER_MAGIC;
5144 buf->f_bsize = sb->s_blocksize;
5145 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5146 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5147 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5148 /* prevent underflow in case that few free space is available */
5149 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5150 buf->f_bavail = buf->f_bfree -
5151 (ext4_r_blocks_count(es) + resv_blocks);
5152 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5153 buf->f_bavail = 0;
5154 buf->f_files = le32_to_cpu(es->s_inodes_count);
5155 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5156 buf->f_namelen = EXT4_NAME_LEN;
5157 fsid = le64_to_cpup((void *)es->s_uuid) ^
5158 le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5159 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5160 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5161
5162#ifdef CONFIG_QUOTA
5163 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5164 sb_has_quota_limits_enabled(sb, PRJQUOTA))
5165 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5166#endif
5167 return 0;
5168}
5169
5170/* Helper function for writing quotas on sync - we need to start transaction
5171 * before quota file is locked for write. Otherwise the are possible deadlocks:
5172 * Process 1 Process 2
5173 * ext4_create() quota_sync()
5174 * jbd2_journal_start() write_dquot()
5175 * dquot_initialize() down(dqio_mutex)
5176 * down(dqio_mutex) jbd2_journal_start()
5177 *
5178 */
5179
5180#ifdef CONFIG_QUOTA
5181
5182static inline struct inode *dquot_to_inode(struct dquot *dquot)
5183{
5184 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5185}
5186
5187static int ext4_write_dquot(struct dquot *dquot)
5188{
5189 int ret, err;
5190 handle_t *handle;
5191 struct inode *inode;
5192
5193 inode = dquot_to_inode(dquot);
5194 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5195 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5196 if (IS_ERR(handle))
5197 return PTR_ERR(handle);
5198 ret = dquot_commit(dquot);
5199 err = ext4_journal_stop(handle);
5200 if (!ret)
5201 ret = err;
5202 return ret;
5203}
5204
5205static int ext4_acquire_dquot(struct dquot *dquot)
5206{
5207 int ret, err;
5208 handle_t *handle;
5209
5210 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5211 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5212 if (IS_ERR(handle))
5213 return PTR_ERR(handle);
5214 ret = dquot_acquire(dquot);
5215 err = ext4_journal_stop(handle);
5216 if (!ret)
5217 ret = err;
5218 return ret;
5219}
5220
5221static int ext4_release_dquot(struct dquot *dquot)
5222{
5223 int ret, err;
5224 handle_t *handle;
5225
5226 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5227 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5228 if (IS_ERR(handle)) {
5229 /* Release dquot anyway to avoid endless cycle in dqput() */
5230 dquot_release(dquot);
5231 return PTR_ERR(handle);
5232 }
5233 ret = dquot_release(dquot);
5234 err = ext4_journal_stop(handle);
5235 if (!ret)
5236 ret = err;
5237 return ret;
5238}
5239
5240static int ext4_mark_dquot_dirty(struct dquot *dquot)
5241{
5242 struct super_block *sb = dquot->dq_sb;
5243 struct ext4_sb_info *sbi = EXT4_SB(sb);
5244
5245 /* Are we journaling quotas? */
5246 if (ext4_has_feature_quota(sb) ||
5247 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5248 dquot_mark_dquot_dirty(dquot);
5249 return ext4_write_dquot(dquot);
5250 } else {
5251 return dquot_mark_dquot_dirty(dquot);
5252 }
5253}
5254
5255static int ext4_write_info(struct super_block *sb, int type)
5256{
5257 int ret, err;
5258 handle_t *handle;
5259
5260 /* Data block + inode block */
5261 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5262 if (IS_ERR(handle))
5263 return PTR_ERR(handle);
5264 ret = dquot_commit_info(sb, type);
5265 err = ext4_journal_stop(handle);
5266 if (!ret)
5267 ret = err;
5268 return ret;
5269}
5270
5271/*
5272 * Turn on quotas during mount time - we need to find
5273 * the quota file and such...
5274 */
5275static int ext4_quota_on_mount(struct super_block *sb, int type)
5276{
5277 return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5278 EXT4_SB(sb)->s_jquota_fmt, type);
5279}
5280
5281static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5282{
5283 struct ext4_inode_info *ei = EXT4_I(inode);
5284
5285 /* The first argument of lockdep_set_subclass has to be
5286 * *exactly* the same as the argument to init_rwsem() --- in
5287 * this case, in init_once() --- or lockdep gets unhappy
5288 * because the name of the lock is set using the
5289 * stringification of the argument to init_rwsem().
5290 */
5291 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
5292 lockdep_set_subclass(&ei->i_data_sem, subclass);
5293}
5294
5295/*
5296 * Standard function to be called on quota_on
5297 */
5298static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5299 const struct path *path)
5300{
5301 int err;
5302
5303 if (!test_opt(sb, QUOTA))
5304 return -EINVAL;
5305
5306 /* Quotafile not on the same filesystem? */
5307 if (path->dentry->d_sb != sb)
5308 return -EXDEV;
5309 /* Journaling quota? */
5310 if (EXT4_SB(sb)->s_qf_names[type]) {
5311 /* Quotafile not in fs root? */
5312 if (path->dentry->d_parent != sb->s_root)
5313 ext4_msg(sb, KERN_WARNING,
5314 "Quota file not on filesystem root. "
5315 "Journaled quota will not work");
5316 }
5317
5318 /*
5319 * When we journal data on quota file, we have to flush journal to see
5320 * all updates to the file when we bypass pagecache...
5321 */
5322 if (EXT4_SB(sb)->s_journal &&
5323 ext4_should_journal_data(d_inode(path->dentry))) {
5324 /*
5325 * We don't need to lock updates but journal_flush() could
5326 * otherwise be livelocked...
5327 */
5328 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5329 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5330 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5331 if (err)
5332 return err;
5333 }
5334 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5335 err = dquot_quota_on(sb, type, format_id, path);
5336 if (err)
5337 lockdep_set_quota_inode(path->dentry->d_inode,
5338 I_DATA_SEM_NORMAL);
5339 return err;
5340}
5341
5342static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5343 unsigned int flags)
5344{
5345 int err;
5346 struct inode *qf_inode;
5347 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5348 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5349 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5350 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5351 };
5352
5353 BUG_ON(!ext4_has_feature_quota(sb));
5354
5355 if (!qf_inums[type])
5356 return -EPERM;
5357
5358 qf_inode = ext4_iget(sb, qf_inums[type]);
5359 if (IS_ERR(qf_inode)) {
5360 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5361 return PTR_ERR(qf_inode);
5362 }
5363
5364 /* Don't account quota for quota files to avoid recursion */
5365 qf_inode->i_flags |= S_NOQUOTA;
5366 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5367 err = dquot_enable(qf_inode, type, format_id, flags);
5368 iput(qf_inode);
5369 if (err)
5370 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5371
5372 return err;
5373}
5374
5375/* Enable usage tracking for all quota types. */
5376static int ext4_enable_quotas(struct super_block *sb)
5377{
5378 int type, err = 0;
5379 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5380 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5381 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5382 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5383 };
5384 bool quota_mopt[EXT4_MAXQUOTAS] = {
5385 test_opt(sb, USRQUOTA),
5386 test_opt(sb, GRPQUOTA),
5387 test_opt(sb, PRJQUOTA),
5388 };
5389
5390 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5391 for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5392 if (qf_inums[type]) {
5393 err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5394 DQUOT_USAGE_ENABLED |
5395 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5396 if (err) {
5397 ext4_warning(sb,
5398 "Failed to enable quota tracking "
5399 "(type=%d, err=%d). Please run "
5400 "e2fsck to fix.", type, err);
5401 return err;
5402 }
5403 }
5404 }
5405 return 0;
5406}
5407
5408static int ext4_quota_off(struct super_block *sb, int type)
5409{
5410 struct inode *inode = sb_dqopt(sb)->files[type];
5411 handle_t *handle;
5412
5413 /* Force all delayed allocation blocks to be allocated.
5414 * Caller already holds s_umount sem */
5415 if (test_opt(sb, DELALLOC))
5416 sync_filesystem(sb);
5417
5418 if (!inode)
5419 goto out;
5420
5421 /* Update modification times of quota files when userspace can
5422 * start looking at them */
5423 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5424 if (IS_ERR(handle))
5425 goto out;
5426 inode->i_mtime = inode->i_ctime = current_time(inode);
5427 ext4_mark_inode_dirty(handle, inode);
5428 ext4_journal_stop(handle);
5429
5430out:
5431 return dquot_quota_off(sb, type);
5432}
5433
5434/* Read data from quotafile - avoid pagecache and such because we cannot afford
5435 * acquiring the locks... As quota files are never truncated and quota code
5436 * itself serializes the operations (and no one else should touch the files)
5437 * we don't have to be afraid of races */
5438static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5439 size_t len, loff_t off)
5440{
5441 struct inode *inode = sb_dqopt(sb)->files[type];
5442 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5443 int offset = off & (sb->s_blocksize - 1);
5444 int tocopy;
5445 size_t toread;
5446 struct buffer_head *bh;
5447 loff_t i_size = i_size_read(inode);
5448
5449 if (off > i_size)
5450 return 0;
5451 if (off+len > i_size)
5452 len = i_size-off;
5453 toread = len;
5454 while (toread > 0) {
5455 tocopy = sb->s_blocksize - offset < toread ?
5456 sb->s_blocksize - offset : toread;
5457 bh = ext4_bread(NULL, inode, blk, 0);
5458 if (IS_ERR(bh))
5459 return PTR_ERR(bh);
5460 if (!bh) /* A hole? */
5461 memset(data, 0, tocopy);
5462 else
5463 memcpy(data, bh->b_data+offset, tocopy);
5464 brelse(bh);
5465 offset = 0;
5466 toread -= tocopy;
5467 data += tocopy;
5468 blk++;
5469 }
5470 return len;
5471}
5472
5473/* Write to quotafile (we know the transaction is already started and has
5474 * enough credits) */
5475static ssize_t ext4_quota_write(struct super_block *sb, int type,
5476 const char *data, size_t len, loff_t off)
5477{
5478 struct inode *inode = sb_dqopt(sb)->files[type];
5479 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5480 int err, offset = off & (sb->s_blocksize - 1);
5481 int retries = 0;
5482 struct buffer_head *bh;
5483 handle_t *handle = journal_current_handle();
5484
5485 if (EXT4_SB(sb)->s_journal && !handle) {
5486 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5487 " cancelled because transaction is not started",
5488 (unsigned long long)off, (unsigned long long)len);
5489 return -EIO;
5490 }
5491 /*
5492 * Since we account only one data block in transaction credits,
5493 * then it is impossible to cross a block boundary.
5494 */
5495 if (sb->s_blocksize - offset < len) {
5496 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5497 " cancelled because not block aligned",
5498 (unsigned long long)off, (unsigned long long)len);
5499 return -EIO;
5500 }
5501
5502 do {
5503 bh = ext4_bread(handle, inode, blk,
5504 EXT4_GET_BLOCKS_CREATE |
5505 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5506 } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5507 ext4_should_retry_alloc(inode->i_sb, &retries));
5508 if (IS_ERR(bh))
5509 return PTR_ERR(bh);
5510 if (!bh)
5511 goto out;
5512 BUFFER_TRACE(bh, "get write access");
5513 err = ext4_journal_get_write_access(handle, bh);
5514 if (err) {
5515 brelse(bh);
5516 return err;
5517 }
5518 lock_buffer(bh);
5519 memcpy(bh->b_data+offset, data, len);
5520 flush_dcache_page(bh->b_page);
5521 unlock_buffer(bh);
5522 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5523 brelse(bh);
5524out:
5525 if (inode->i_size < off + len) {
5526 i_size_write(inode, off + len);
5527 EXT4_I(inode)->i_disksize = inode->i_size;
5528 ext4_mark_inode_dirty(handle, inode);
5529 }
5530 return len;
5531}
5532
5533static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5534{
5535 const struct quota_format_ops *ops;
5536
5537 if (!sb_has_quota_loaded(sb, qid->type))
5538 return -ESRCH;
5539 ops = sb_dqopt(sb)->ops[qid->type];
5540 if (!ops || !ops->get_next_id)
5541 return -ENOSYS;
5542 return dquot_get_next_id(sb, qid);
5543}
5544#endif
5545
5546static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5547 const char *dev_name, void *data)
5548{
5549 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5550}
5551
5552#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5553static inline void register_as_ext2(void)
5554{
5555 int err = register_filesystem(&ext2_fs_type);
5556 if (err)
5557 printk(KERN_WARNING
5558 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5559}
5560
5561static inline void unregister_as_ext2(void)
5562{
5563 unregister_filesystem(&ext2_fs_type);
5564}
5565
5566static inline int ext2_feature_set_ok(struct super_block *sb)
5567{
5568 if (ext4_has_unknown_ext2_incompat_features(sb))
5569 return 0;
5570 if (sb->s_flags & MS_RDONLY)
5571 return 1;
5572 if (ext4_has_unknown_ext2_ro_compat_features(sb))
5573 return 0;
5574 return 1;
5575}
5576#else
5577static inline void register_as_ext2(void) { }
5578static inline void unregister_as_ext2(void) { }
5579static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5580#endif
5581
5582static inline void register_as_ext3(void)
5583{
5584 int err = register_filesystem(&ext3_fs_type);
5585 if (err)
5586 printk(KERN_WARNING
5587 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5588}
5589
5590static inline void unregister_as_ext3(void)
5591{
5592 unregister_filesystem(&ext3_fs_type);
5593}
5594
5595static inline int ext3_feature_set_ok(struct super_block *sb)
5596{
5597 if (ext4_has_unknown_ext3_incompat_features(sb))
5598 return 0;
5599 if (!ext4_has_feature_journal(sb))
5600 return 0;
5601 if (sb->s_flags & MS_RDONLY)
5602 return 1;
5603 if (ext4_has_unknown_ext3_ro_compat_features(sb))
5604 return 0;
5605 return 1;
5606}
5607
5608static struct file_system_type ext4_fs_type = {
5609 .owner = THIS_MODULE,
5610 .name = "ext4",
5611 .mount = ext4_mount,
5612 .kill_sb = kill_block_super,
5613 .fs_flags = FS_REQUIRES_DEV,
5614};
5615MODULE_ALIAS_FS("ext4");
5616
5617/* Shared across all ext4 file systems */
5618wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5619
5620static int __init ext4_init_fs(void)
5621{
5622 int i, err;
5623
5624 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5625 ext4_li_info = NULL;
5626 mutex_init(&ext4_li_mtx);
5627
5628 /* Build-time check for flags consistency */
5629 ext4_check_flag_values();
5630
5631 for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5632 init_waitqueue_head(&ext4__ioend_wq[i]);
5633
5634 err = ext4_init_es();
5635 if (err)
5636 return err;
5637
5638 err = ext4_init_pageio();
5639 if (err)
5640 goto out5;
5641
5642 err = ext4_init_system_zone();
5643 if (err)
5644 goto out4;
5645
5646 err = ext4_init_sysfs();
5647 if (err)
5648 goto out3;
5649
5650 err = ext4_init_mballoc();
5651 if (err)
5652 goto out2;
5653 err = init_inodecache();
5654 if (err)
5655 goto out1;
5656 register_as_ext3();
5657 register_as_ext2();
5658 err = register_filesystem(&ext4_fs_type);
5659 if (err)
5660 goto out;
5661
5662 return 0;
5663out:
5664 unregister_as_ext2();
5665 unregister_as_ext3();
5666 destroy_inodecache();
5667out1:
5668 ext4_exit_mballoc();
5669out2:
5670 ext4_exit_sysfs();
5671out3:
5672 ext4_exit_system_zone();
5673out4:
5674 ext4_exit_pageio();
5675out5:
5676 ext4_exit_es();
5677
5678 return err;
5679}
5680
5681static void __exit ext4_exit_fs(void)
5682{
5683 ext4_destroy_lazyinit_thread();
5684 unregister_as_ext2();
5685 unregister_as_ext3();
5686 unregister_filesystem(&ext4_fs_type);
5687 destroy_inodecache();
5688 ext4_exit_mballoc();
5689 ext4_exit_sysfs();
5690 ext4_exit_system_zone();
5691 ext4_exit_pageio();
5692 ext4_exit_es();
5693}
5694
5695MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5696MODULE_DESCRIPTION("Fourth Extended Filesystem");
5697MODULE_LICENSE("GPL");
5698module_init(ext4_init_fs)
5699module_exit(ext4_exit_fs)
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 void ext4_kill_sb(struct super_block *sb);
97static const struct fs_parameter_spec ext4_param_specs[];
98
99/*
100 * Lock ordering
101 *
102 * page fault path:
103 * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
104 * -> page lock -> i_data_sem (rw)
105 *
106 * buffered write path:
107 * sb_start_write -> i_mutex -> mmap_lock
108 * sb_start_write -> i_mutex -> transaction start -> page lock ->
109 * i_data_sem (rw)
110 *
111 * truncate:
112 * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
113 * page lock
114 * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
115 * i_data_sem (rw)
116 *
117 * direct IO:
118 * sb_start_write -> i_mutex -> mmap_lock
119 * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
120 *
121 * writepages:
122 * transaction start -> page lock(s) -> i_data_sem (rw)
123 */
124
125static const struct fs_context_operations ext4_context_ops = {
126 .parse_param = ext4_parse_param,
127 .get_tree = ext4_get_tree,
128 .reconfigure = ext4_reconfigure,
129 .free = ext4_fc_free,
130};
131
132
133#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
134static struct file_system_type ext2_fs_type = {
135 .owner = THIS_MODULE,
136 .name = "ext2",
137 .init_fs_context = ext4_init_fs_context,
138 .parameters = ext4_param_specs,
139 .kill_sb = ext4_kill_sb,
140 .fs_flags = FS_REQUIRES_DEV,
141};
142MODULE_ALIAS_FS("ext2");
143MODULE_ALIAS("ext2");
144#define IS_EXT2_SB(sb) ((sb)->s_type == &ext2_fs_type)
145#else
146#define IS_EXT2_SB(sb) (0)
147#endif
148
149
150static struct file_system_type ext3_fs_type = {
151 .owner = THIS_MODULE,
152 .name = "ext3",
153 .init_fs_context = ext4_init_fs_context,
154 .parameters = ext4_param_specs,
155 .kill_sb = ext4_kill_sb,
156 .fs_flags = FS_REQUIRES_DEV,
157};
158MODULE_ALIAS_FS("ext3");
159MODULE_ALIAS("ext3");
160#define IS_EXT3_SB(sb) ((sb)->s_type == &ext3_fs_type)
161
162
163static inline void __ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags,
164 bh_end_io_t *end_io)
165{
166 /*
167 * buffer's verified bit is no longer valid after reading from
168 * disk again due to write out error, clear it to make sure we
169 * recheck the buffer contents.
170 */
171 clear_buffer_verified(bh);
172
173 bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
174 get_bh(bh);
175 submit_bh(REQ_OP_READ | op_flags, bh);
176}
177
178void ext4_read_bh_nowait(struct buffer_head *bh, blk_opf_t op_flags,
179 bh_end_io_t *end_io)
180{
181 BUG_ON(!buffer_locked(bh));
182
183 if (ext4_buffer_uptodate(bh)) {
184 unlock_buffer(bh);
185 return;
186 }
187 __ext4_read_bh(bh, op_flags, end_io);
188}
189
190int ext4_read_bh(struct buffer_head *bh, blk_opf_t op_flags, bh_end_io_t *end_io)
191{
192 BUG_ON(!buffer_locked(bh));
193
194 if (ext4_buffer_uptodate(bh)) {
195 unlock_buffer(bh);
196 return 0;
197 }
198
199 __ext4_read_bh(bh, op_flags, end_io);
200
201 wait_on_buffer(bh);
202 if (buffer_uptodate(bh))
203 return 0;
204 return -EIO;
205}
206
207int ext4_read_bh_lock(struct buffer_head *bh, blk_opf_t op_flags, bool wait)
208{
209 lock_buffer(bh);
210 if (!wait) {
211 ext4_read_bh_nowait(bh, op_flags, NULL);
212 return 0;
213 }
214 return ext4_read_bh(bh, op_flags, NULL);
215}
216
217/*
218 * This works like __bread_gfp() except it uses ERR_PTR for error
219 * returns. Currently with sb_bread it's impossible to distinguish
220 * between ENOMEM and EIO situations (since both result in a NULL
221 * return.
222 */
223static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
224 sector_t block,
225 blk_opf_t op_flags, gfp_t gfp)
226{
227 struct buffer_head *bh;
228 int ret;
229
230 bh = sb_getblk_gfp(sb, block, gfp);
231 if (bh == NULL)
232 return ERR_PTR(-ENOMEM);
233 if (ext4_buffer_uptodate(bh))
234 return bh;
235
236 ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
237 if (ret) {
238 put_bh(bh);
239 return ERR_PTR(ret);
240 }
241 return bh;
242}
243
244struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
245 blk_opf_t op_flags)
246{
247 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_inode->i_mapping,
248 ~__GFP_FS) | __GFP_MOVABLE;
249
250 return __ext4_sb_bread_gfp(sb, block, op_flags, gfp);
251}
252
253struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
254 sector_t block)
255{
256 gfp_t gfp = mapping_gfp_constraint(sb->s_bdev->bd_inode->i_mapping,
257 ~__GFP_FS);
258
259 return __ext4_sb_bread_gfp(sb, block, 0, gfp);
260}
261
262void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
263{
264 struct buffer_head *bh = bdev_getblk(sb->s_bdev, block,
265 sb->s_blocksize, GFP_NOWAIT | __GFP_NOWARN);
266
267 if (likely(bh)) {
268 if (trylock_buffer(bh))
269 ext4_read_bh_nowait(bh, REQ_RAHEAD, NULL);
270 brelse(bh);
271 }
272}
273
274static int ext4_verify_csum_type(struct super_block *sb,
275 struct ext4_super_block *es)
276{
277 if (!ext4_has_feature_metadata_csum(sb))
278 return 1;
279
280 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
281}
282
283__le32 ext4_superblock_csum(struct super_block *sb,
284 struct ext4_super_block *es)
285{
286 struct ext4_sb_info *sbi = EXT4_SB(sb);
287 int offset = offsetof(struct ext4_super_block, s_checksum);
288 __u32 csum;
289
290 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
291
292 return cpu_to_le32(csum);
293}
294
295static int ext4_superblock_csum_verify(struct super_block *sb,
296 struct ext4_super_block *es)
297{
298 if (!ext4_has_metadata_csum(sb))
299 return 1;
300
301 return es->s_checksum == ext4_superblock_csum(sb, es);
302}
303
304void ext4_superblock_csum_set(struct super_block *sb)
305{
306 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
307
308 if (!ext4_has_metadata_csum(sb))
309 return;
310
311 es->s_checksum = ext4_superblock_csum(sb, es);
312}
313
314ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
315 struct ext4_group_desc *bg)
316{
317 return le32_to_cpu(bg->bg_block_bitmap_lo) |
318 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
319 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
320}
321
322ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
323 struct ext4_group_desc *bg)
324{
325 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
326 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
327 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
328}
329
330ext4_fsblk_t ext4_inode_table(struct super_block *sb,
331 struct ext4_group_desc *bg)
332{
333 return le32_to_cpu(bg->bg_inode_table_lo) |
334 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
335 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
336}
337
338__u32 ext4_free_group_clusters(struct super_block *sb,
339 struct ext4_group_desc *bg)
340{
341 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
342 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
343 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
344}
345
346__u32 ext4_free_inodes_count(struct super_block *sb,
347 struct ext4_group_desc *bg)
348{
349 return le16_to_cpu(bg->bg_free_inodes_count_lo) |
350 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
351 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
352}
353
354__u32 ext4_used_dirs_count(struct super_block *sb,
355 struct ext4_group_desc *bg)
356{
357 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
358 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
359 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
360}
361
362__u32 ext4_itable_unused_count(struct super_block *sb,
363 struct ext4_group_desc *bg)
364{
365 return le16_to_cpu(bg->bg_itable_unused_lo) |
366 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
367 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
368}
369
370void ext4_block_bitmap_set(struct super_block *sb,
371 struct ext4_group_desc *bg, ext4_fsblk_t blk)
372{
373 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
374 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
375 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
376}
377
378void ext4_inode_bitmap_set(struct super_block *sb,
379 struct ext4_group_desc *bg, ext4_fsblk_t blk)
380{
381 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
382 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
383 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
384}
385
386void ext4_inode_table_set(struct super_block *sb,
387 struct ext4_group_desc *bg, ext4_fsblk_t blk)
388{
389 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
390 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
391 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
392}
393
394void ext4_free_group_clusters_set(struct super_block *sb,
395 struct ext4_group_desc *bg, __u32 count)
396{
397 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
398 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
399 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
400}
401
402void ext4_free_inodes_set(struct super_block *sb,
403 struct ext4_group_desc *bg, __u32 count)
404{
405 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
406 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
407 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
408}
409
410void ext4_used_dirs_set(struct super_block *sb,
411 struct ext4_group_desc *bg, __u32 count)
412{
413 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
414 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
415 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
416}
417
418void ext4_itable_unused_set(struct super_block *sb,
419 struct ext4_group_desc *bg, __u32 count)
420{
421 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
422 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
423 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
424}
425
426static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
427{
428 now = clamp_val(now, 0, (1ull << 40) - 1);
429
430 *lo = cpu_to_le32(lower_32_bits(now));
431 *hi = upper_32_bits(now);
432}
433
434static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
435{
436 return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
437}
438#define ext4_update_tstamp(es, tstamp) \
439 __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
440 ktime_get_real_seconds())
441#define ext4_get_tstamp(es, tstamp) \
442 __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
443
444#define EXT4_SB_REFRESH_INTERVAL_SEC (3600) /* seconds (1 hour) */
445#define EXT4_SB_REFRESH_INTERVAL_KB (16384) /* kilobytes (16MB) */
446
447/*
448 * The ext4_maybe_update_superblock() function checks and updates the
449 * superblock if needed.
450 *
451 * This function is designed to update the on-disk superblock only under
452 * certain conditions to prevent excessive disk writes and unnecessary
453 * waking of the disk from sleep. The superblock will be updated if:
454 * 1. More than an hour has passed since the last superblock update, and
455 * 2. More than 16MB have been written since the last superblock update.
456 *
457 * @sb: The superblock
458 */
459static void ext4_maybe_update_superblock(struct super_block *sb)
460{
461 struct ext4_sb_info *sbi = EXT4_SB(sb);
462 struct ext4_super_block *es = sbi->s_es;
463 journal_t *journal = sbi->s_journal;
464 time64_t now;
465 __u64 last_update;
466 __u64 lifetime_write_kbytes;
467 __u64 diff_size;
468
469 if (sb_rdonly(sb) || !(sb->s_flags & SB_ACTIVE) ||
470 !journal || (journal->j_flags & JBD2_UNMOUNT))
471 return;
472
473 now = ktime_get_real_seconds();
474 last_update = ext4_get_tstamp(es, s_wtime);
475
476 if (likely(now - last_update < EXT4_SB_REFRESH_INTERVAL_SEC))
477 return;
478
479 lifetime_write_kbytes = sbi->s_kbytes_written +
480 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
481 sbi->s_sectors_written_start) >> 1);
482
483 /* Get the number of kilobytes not written to disk to account
484 * for statistics and compare with a multiple of 16 MB. This
485 * is used to determine when the next superblock commit should
486 * occur (i.e. not more often than once per 16MB if there was
487 * less written in an hour).
488 */
489 diff_size = lifetime_write_kbytes - le64_to_cpu(es->s_kbytes_written);
490
491 if (diff_size > EXT4_SB_REFRESH_INTERVAL_KB)
492 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
493}
494
495/*
496 * The del_gendisk() function uninitializes the disk-specific data
497 * structures, including the bdi structure, without telling anyone
498 * else. Once this happens, any attempt to call mark_buffer_dirty()
499 * (for example, by ext4_commit_super), will cause a kernel OOPS.
500 * This is a kludge to prevent these oops until we can put in a proper
501 * hook in del_gendisk() to inform the VFS and file system layers.
502 */
503static int block_device_ejected(struct super_block *sb)
504{
505 struct inode *bd_inode = sb->s_bdev->bd_inode;
506 struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
507
508 return bdi->dev == NULL;
509}
510
511static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
512{
513 struct super_block *sb = journal->j_private;
514 struct ext4_sb_info *sbi = EXT4_SB(sb);
515 int error = is_journal_aborted(journal);
516 struct ext4_journal_cb_entry *jce;
517
518 BUG_ON(txn->t_state == T_FINISHED);
519
520 ext4_process_freed_data(sb, txn->t_tid);
521 ext4_maybe_update_superblock(sb);
522
523 spin_lock(&sbi->s_md_lock);
524 while (!list_empty(&txn->t_private_list)) {
525 jce = list_entry(txn->t_private_list.next,
526 struct ext4_journal_cb_entry, jce_list);
527 list_del_init(&jce->jce_list);
528 spin_unlock(&sbi->s_md_lock);
529 jce->jce_func(sb, jce, error);
530 spin_lock(&sbi->s_md_lock);
531 }
532 spin_unlock(&sbi->s_md_lock);
533}
534
535/*
536 * This writepage callback for write_cache_pages()
537 * takes care of a few cases after page cleaning.
538 *
539 * write_cache_pages() already checks for dirty pages
540 * and calls clear_page_dirty_for_io(), which we want,
541 * to write protect the pages.
542 *
543 * However, we may have to redirty a page (see below.)
544 */
545static int ext4_journalled_writepage_callback(struct folio *folio,
546 struct writeback_control *wbc,
547 void *data)
548{
549 transaction_t *transaction = (transaction_t *) data;
550 struct buffer_head *bh, *head;
551 struct journal_head *jh;
552
553 bh = head = folio_buffers(folio);
554 do {
555 /*
556 * We have to redirty a page in these cases:
557 * 1) If buffer is dirty, it means the page was dirty because it
558 * contains a buffer that needs checkpointing. So the dirty bit
559 * needs to be preserved so that checkpointing writes the buffer
560 * properly.
561 * 2) If buffer is not part of the committing transaction
562 * (we may have just accidentally come across this buffer because
563 * inode range tracking is not exact) or if the currently running
564 * transaction already contains this buffer as well, dirty bit
565 * needs to be preserved so that the buffer gets writeprotected
566 * properly on running transaction's commit.
567 */
568 jh = bh2jh(bh);
569 if (buffer_dirty(bh) ||
570 (jh && (jh->b_transaction != transaction ||
571 jh->b_next_transaction))) {
572 folio_redirty_for_writepage(wbc, folio);
573 goto out;
574 }
575 } while ((bh = bh->b_this_page) != head);
576
577out:
578 return AOP_WRITEPAGE_ACTIVATE;
579}
580
581static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
582{
583 struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
584 struct writeback_control wbc = {
585 .sync_mode = WB_SYNC_ALL,
586 .nr_to_write = LONG_MAX,
587 .range_start = jinode->i_dirty_start,
588 .range_end = jinode->i_dirty_end,
589 };
590
591 return write_cache_pages(mapping, &wbc,
592 ext4_journalled_writepage_callback,
593 jinode->i_transaction);
594}
595
596static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
597{
598 int ret;
599
600 if (ext4_should_journal_data(jinode->i_vfs_inode))
601 ret = ext4_journalled_submit_inode_data_buffers(jinode);
602 else
603 ret = ext4_normal_submit_inode_data_buffers(jinode);
604 return ret;
605}
606
607static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
608{
609 int ret = 0;
610
611 if (!ext4_should_journal_data(jinode->i_vfs_inode))
612 ret = jbd2_journal_finish_inode_data_buffers(jinode);
613
614 return ret;
615}
616
617static bool system_going_down(void)
618{
619 return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
620 || system_state == SYSTEM_RESTART;
621}
622
623struct ext4_err_translation {
624 int code;
625 int errno;
626};
627
628#define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
629
630static struct ext4_err_translation err_translation[] = {
631 EXT4_ERR_TRANSLATE(EIO),
632 EXT4_ERR_TRANSLATE(ENOMEM),
633 EXT4_ERR_TRANSLATE(EFSBADCRC),
634 EXT4_ERR_TRANSLATE(EFSCORRUPTED),
635 EXT4_ERR_TRANSLATE(ENOSPC),
636 EXT4_ERR_TRANSLATE(ENOKEY),
637 EXT4_ERR_TRANSLATE(EROFS),
638 EXT4_ERR_TRANSLATE(EFBIG),
639 EXT4_ERR_TRANSLATE(EEXIST),
640 EXT4_ERR_TRANSLATE(ERANGE),
641 EXT4_ERR_TRANSLATE(EOVERFLOW),
642 EXT4_ERR_TRANSLATE(EBUSY),
643 EXT4_ERR_TRANSLATE(ENOTDIR),
644 EXT4_ERR_TRANSLATE(ENOTEMPTY),
645 EXT4_ERR_TRANSLATE(ESHUTDOWN),
646 EXT4_ERR_TRANSLATE(EFAULT),
647};
648
649static int ext4_errno_to_code(int errno)
650{
651 int i;
652
653 for (i = 0; i < ARRAY_SIZE(err_translation); i++)
654 if (err_translation[i].errno == errno)
655 return err_translation[i].code;
656 return EXT4_ERR_UNKNOWN;
657}
658
659static void save_error_info(struct super_block *sb, int error,
660 __u32 ino, __u64 block,
661 const char *func, unsigned int line)
662{
663 struct ext4_sb_info *sbi = EXT4_SB(sb);
664
665 /* We default to EFSCORRUPTED error... */
666 if (error == 0)
667 error = EFSCORRUPTED;
668
669 spin_lock(&sbi->s_error_lock);
670 sbi->s_add_error_count++;
671 sbi->s_last_error_code = error;
672 sbi->s_last_error_line = line;
673 sbi->s_last_error_ino = ino;
674 sbi->s_last_error_block = block;
675 sbi->s_last_error_func = func;
676 sbi->s_last_error_time = ktime_get_real_seconds();
677 if (!sbi->s_first_error_time) {
678 sbi->s_first_error_code = error;
679 sbi->s_first_error_line = line;
680 sbi->s_first_error_ino = ino;
681 sbi->s_first_error_block = block;
682 sbi->s_first_error_func = func;
683 sbi->s_first_error_time = sbi->s_last_error_time;
684 }
685 spin_unlock(&sbi->s_error_lock);
686}
687
688/* Deal with the reporting of failure conditions on a filesystem such as
689 * inconsistencies detected or read IO failures.
690 *
691 * On ext2, we can store the error state of the filesystem in the
692 * superblock. That is not possible on ext4, because we may have other
693 * write ordering constraints on the superblock which prevent us from
694 * writing it out straight away; and given that the journal is about to
695 * be aborted, we can't rely on the current, or future, transactions to
696 * write out the superblock safely.
697 *
698 * We'll just use the jbd2_journal_abort() error code to record an error in
699 * the journal instead. On recovery, the journal will complain about
700 * that error until we've noted it down and cleared it.
701 *
702 * If force_ro is set, we unconditionally force the filesystem into an
703 * ABORT|READONLY state, unless the error response on the fs has been set to
704 * panic in which case we take the easy way out and panic immediately. This is
705 * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
706 * at a critical moment in log management.
707 */
708static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
709 __u32 ino, __u64 block,
710 const char *func, unsigned int line)
711{
712 journal_t *journal = EXT4_SB(sb)->s_journal;
713 bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
714
715 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
716 if (test_opt(sb, WARN_ON_ERROR))
717 WARN_ON_ONCE(1);
718
719 if (!continue_fs && !sb_rdonly(sb)) {
720 set_bit(EXT4_FLAGS_SHUTDOWN, &EXT4_SB(sb)->s_ext4_flags);
721 if (journal)
722 jbd2_journal_abort(journal, -EIO);
723 }
724
725 if (!bdev_read_only(sb->s_bdev)) {
726 save_error_info(sb, error, ino, block, func, line);
727 /*
728 * In case the fs should keep running, we need to writeout
729 * superblock through the journal. Due to lock ordering
730 * constraints, it may not be safe to do it right here so we
731 * defer superblock flushing to a workqueue.
732 */
733 if (continue_fs && journal)
734 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
735 else
736 ext4_commit_super(sb);
737 }
738
739 /*
740 * We force ERRORS_RO behavior when system is rebooting. Otherwise we
741 * could panic during 'reboot -f' as the underlying device got already
742 * disabled.
743 */
744 if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
745 panic("EXT4-fs (device %s): panic forced after error\n",
746 sb->s_id);
747 }
748
749 if (sb_rdonly(sb) || continue_fs)
750 return;
751
752 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
753 /*
754 * Make sure updated value of ->s_mount_flags will be visible before
755 * ->s_flags update
756 */
757 smp_wmb();
758 sb->s_flags |= SB_RDONLY;
759}
760
761static void update_super_work(struct work_struct *work)
762{
763 struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
764 s_sb_upd_work);
765 journal_t *journal = sbi->s_journal;
766 handle_t *handle;
767
768 /*
769 * If the journal is still running, we have to write out superblock
770 * through the journal to avoid collisions of other journalled sb
771 * updates.
772 *
773 * We use directly jbd2 functions here to avoid recursing back into
774 * ext4 error handling code during handling of previous errors.
775 */
776 if (!sb_rdonly(sbi->s_sb) && journal) {
777 struct buffer_head *sbh = sbi->s_sbh;
778 bool call_notify_err = false;
779
780 handle = jbd2_journal_start(journal, 1);
781 if (IS_ERR(handle))
782 goto write_directly;
783 if (jbd2_journal_get_write_access(handle, sbh)) {
784 jbd2_journal_stop(handle);
785 goto write_directly;
786 }
787
788 if (sbi->s_add_error_count > 0)
789 call_notify_err = true;
790
791 ext4_update_super(sbi->s_sb);
792 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
793 ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
794 "superblock detected");
795 clear_buffer_write_io_error(sbh);
796 set_buffer_uptodate(sbh);
797 }
798
799 if (jbd2_journal_dirty_metadata(handle, sbh)) {
800 jbd2_journal_stop(handle);
801 goto write_directly;
802 }
803 jbd2_journal_stop(handle);
804
805 if (call_notify_err)
806 ext4_notify_error_sysfs(sbi);
807
808 return;
809 }
810write_directly:
811 /*
812 * Write through journal failed. Write sb directly to get error info
813 * out and hope for the best.
814 */
815 ext4_commit_super(sbi->s_sb);
816 ext4_notify_error_sysfs(sbi);
817}
818
819#define ext4_error_ratelimit(sb) \
820 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \
821 "EXT4-fs error")
822
823void __ext4_error(struct super_block *sb, const char *function,
824 unsigned int line, bool force_ro, int error, __u64 block,
825 const char *fmt, ...)
826{
827 struct va_format vaf;
828 va_list args;
829
830 if (unlikely(ext4_forced_shutdown(sb)))
831 return;
832
833 trace_ext4_error(sb, function, line);
834 if (ext4_error_ratelimit(sb)) {
835 va_start(args, fmt);
836 vaf.fmt = fmt;
837 vaf.va = &args;
838 printk(KERN_CRIT
839 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
840 sb->s_id, function, line, current->comm, &vaf);
841 va_end(args);
842 }
843 fsnotify_sb_error(sb, NULL, error ? error : EFSCORRUPTED);
844
845 ext4_handle_error(sb, force_ro, error, 0, block, function, line);
846}
847
848void __ext4_error_inode(struct inode *inode, const char *function,
849 unsigned int line, ext4_fsblk_t block, int error,
850 const char *fmt, ...)
851{
852 va_list args;
853 struct va_format vaf;
854
855 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
856 return;
857
858 trace_ext4_error(inode->i_sb, function, line);
859 if (ext4_error_ratelimit(inode->i_sb)) {
860 va_start(args, fmt);
861 vaf.fmt = fmt;
862 vaf.va = &args;
863 if (block)
864 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
865 "inode #%lu: block %llu: comm %s: %pV\n",
866 inode->i_sb->s_id, function, line, inode->i_ino,
867 block, current->comm, &vaf);
868 else
869 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
870 "inode #%lu: comm %s: %pV\n",
871 inode->i_sb->s_id, function, line, inode->i_ino,
872 current->comm, &vaf);
873 va_end(args);
874 }
875 fsnotify_sb_error(inode->i_sb, inode, error ? error : EFSCORRUPTED);
876
877 ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
878 function, line);
879}
880
881void __ext4_error_file(struct file *file, const char *function,
882 unsigned int line, ext4_fsblk_t block,
883 const char *fmt, ...)
884{
885 va_list args;
886 struct va_format vaf;
887 struct inode *inode = file_inode(file);
888 char pathname[80], *path;
889
890 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
891 return;
892
893 trace_ext4_error(inode->i_sb, function, line);
894 if (ext4_error_ratelimit(inode->i_sb)) {
895 path = file_path(file, pathname, sizeof(pathname));
896 if (IS_ERR(path))
897 path = "(unknown)";
898 va_start(args, fmt);
899 vaf.fmt = fmt;
900 vaf.va = &args;
901 if (block)
902 printk(KERN_CRIT
903 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
904 "block %llu: comm %s: path %s: %pV\n",
905 inode->i_sb->s_id, function, line, inode->i_ino,
906 block, current->comm, path, &vaf);
907 else
908 printk(KERN_CRIT
909 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
910 "comm %s: path %s: %pV\n",
911 inode->i_sb->s_id, function, line, inode->i_ino,
912 current->comm, path, &vaf);
913 va_end(args);
914 }
915 fsnotify_sb_error(inode->i_sb, inode, EFSCORRUPTED);
916
917 ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
918 function, line);
919}
920
921const char *ext4_decode_error(struct super_block *sb, int errno,
922 char nbuf[16])
923{
924 char *errstr = NULL;
925
926 switch (errno) {
927 case -EFSCORRUPTED:
928 errstr = "Corrupt filesystem";
929 break;
930 case -EFSBADCRC:
931 errstr = "Filesystem failed CRC";
932 break;
933 case -EIO:
934 errstr = "IO failure";
935 break;
936 case -ENOMEM:
937 errstr = "Out of memory";
938 break;
939 case -EROFS:
940 if (!sb || (EXT4_SB(sb)->s_journal &&
941 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
942 errstr = "Journal has aborted";
943 else
944 errstr = "Readonly filesystem";
945 break;
946 default:
947 /* If the caller passed in an extra buffer for unknown
948 * errors, textualise them now. Else we just return
949 * NULL. */
950 if (nbuf) {
951 /* Check for truncated error codes... */
952 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
953 errstr = nbuf;
954 }
955 break;
956 }
957
958 return errstr;
959}
960
961/* __ext4_std_error decodes expected errors from journaling functions
962 * automatically and invokes the appropriate error response. */
963
964void __ext4_std_error(struct super_block *sb, const char *function,
965 unsigned int line, int errno)
966{
967 char nbuf[16];
968 const char *errstr;
969
970 if (unlikely(ext4_forced_shutdown(sb)))
971 return;
972
973 /* Special case: if the error is EROFS, and we're not already
974 * inside a transaction, then there's really no point in logging
975 * an error. */
976 if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
977 return;
978
979 if (ext4_error_ratelimit(sb)) {
980 errstr = ext4_decode_error(sb, errno, nbuf);
981 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
982 sb->s_id, function, line, errstr);
983 }
984 fsnotify_sb_error(sb, NULL, errno ? errno : EFSCORRUPTED);
985
986 ext4_handle_error(sb, false, -errno, 0, 0, function, line);
987}
988
989void __ext4_msg(struct super_block *sb,
990 const char *prefix, const char *fmt, ...)
991{
992 struct va_format vaf;
993 va_list args;
994
995 if (sb) {
996 atomic_inc(&EXT4_SB(sb)->s_msg_count);
997 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state),
998 "EXT4-fs"))
999 return;
1000 }
1001
1002 va_start(args, fmt);
1003 vaf.fmt = fmt;
1004 vaf.va = &args;
1005 if (sb)
1006 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
1007 else
1008 printk("%sEXT4-fs: %pV\n", prefix, &vaf);
1009 va_end(args);
1010}
1011
1012static int ext4_warning_ratelimit(struct super_block *sb)
1013{
1014 atomic_inc(&EXT4_SB(sb)->s_warning_count);
1015 return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
1016 "EXT4-fs warning");
1017}
1018
1019void __ext4_warning(struct super_block *sb, const char *function,
1020 unsigned int line, const char *fmt, ...)
1021{
1022 struct va_format vaf;
1023 va_list args;
1024
1025 if (!ext4_warning_ratelimit(sb))
1026 return;
1027
1028 va_start(args, fmt);
1029 vaf.fmt = fmt;
1030 vaf.va = &args;
1031 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
1032 sb->s_id, function, line, &vaf);
1033 va_end(args);
1034}
1035
1036void __ext4_warning_inode(const struct inode *inode, const char *function,
1037 unsigned int line, const char *fmt, ...)
1038{
1039 struct va_format vaf;
1040 va_list args;
1041
1042 if (!ext4_warning_ratelimit(inode->i_sb))
1043 return;
1044
1045 va_start(args, fmt);
1046 vaf.fmt = fmt;
1047 vaf.va = &args;
1048 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
1049 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
1050 function, line, inode->i_ino, current->comm, &vaf);
1051 va_end(args);
1052}
1053
1054void __ext4_grp_locked_error(const char *function, unsigned int line,
1055 struct super_block *sb, ext4_group_t grp,
1056 unsigned long ino, ext4_fsblk_t block,
1057 const char *fmt, ...)
1058__releases(bitlock)
1059__acquires(bitlock)
1060{
1061 struct va_format vaf;
1062 va_list args;
1063
1064 if (unlikely(ext4_forced_shutdown(sb)))
1065 return;
1066
1067 trace_ext4_error(sb, function, line);
1068 if (ext4_error_ratelimit(sb)) {
1069 va_start(args, fmt);
1070 vaf.fmt = fmt;
1071 vaf.va = &args;
1072 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
1073 sb->s_id, function, line, grp);
1074 if (ino)
1075 printk(KERN_CONT "inode %lu: ", ino);
1076 if (block)
1077 printk(KERN_CONT "block %llu:",
1078 (unsigned long long) block);
1079 printk(KERN_CONT "%pV\n", &vaf);
1080 va_end(args);
1081 }
1082
1083 if (test_opt(sb, ERRORS_CONT)) {
1084 if (test_opt(sb, WARN_ON_ERROR))
1085 WARN_ON_ONCE(1);
1086 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
1087 if (!bdev_read_only(sb->s_bdev)) {
1088 save_error_info(sb, EFSCORRUPTED, ino, block, function,
1089 line);
1090 schedule_work(&EXT4_SB(sb)->s_sb_upd_work);
1091 }
1092 return;
1093 }
1094 ext4_unlock_group(sb, grp);
1095 ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1096 /*
1097 * We only get here in the ERRORS_RO case; relocking the group
1098 * may be dangerous, but nothing bad will happen since the
1099 * filesystem will have already been marked read/only and the
1100 * journal has been aborted. We return 1 as a hint to callers
1101 * who might what to use the return value from
1102 * ext4_grp_locked_error() to distinguish between the
1103 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1104 * aggressively from the ext4 function in question, with a
1105 * more appropriate error code.
1106 */
1107 ext4_lock_group(sb, grp);
1108 return;
1109}
1110
1111void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1112 ext4_group_t group,
1113 unsigned int flags)
1114{
1115 struct ext4_sb_info *sbi = EXT4_SB(sb);
1116 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1117 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1118 int ret;
1119
1120 if (!grp || !gdp)
1121 return;
1122 if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1123 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1124 &grp->bb_state);
1125 if (!ret)
1126 percpu_counter_sub(&sbi->s_freeclusters_counter,
1127 grp->bb_free);
1128 }
1129
1130 if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1131 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1132 &grp->bb_state);
1133 if (!ret && gdp) {
1134 int count;
1135
1136 count = ext4_free_inodes_count(sb, gdp);
1137 percpu_counter_sub(&sbi->s_freeinodes_counter,
1138 count);
1139 }
1140 }
1141}
1142
1143void ext4_update_dynamic_rev(struct super_block *sb)
1144{
1145 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1146
1147 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1148 return;
1149
1150 ext4_warning(sb,
1151 "updating to rev %d because of new feature flag, "
1152 "running e2fsck is recommended",
1153 EXT4_DYNAMIC_REV);
1154
1155 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1156 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1157 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1158 /* leave es->s_feature_*compat flags alone */
1159 /* es->s_uuid will be set by e2fsck if empty */
1160
1161 /*
1162 * The rest of the superblock fields should be zero, and if not it
1163 * means they are likely already in use, so leave them alone. We
1164 * can leave it up to e2fsck to clean up any inconsistencies there.
1165 */
1166}
1167
1168static inline struct inode *orphan_list_entry(struct list_head *l)
1169{
1170 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1171}
1172
1173static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1174{
1175 struct list_head *l;
1176
1177 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1178 le32_to_cpu(sbi->s_es->s_last_orphan));
1179
1180 printk(KERN_ERR "sb_info orphan list:\n");
1181 list_for_each(l, &sbi->s_orphan) {
1182 struct inode *inode = orphan_list_entry(l);
1183 printk(KERN_ERR " "
1184 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1185 inode->i_sb->s_id, inode->i_ino, inode,
1186 inode->i_mode, inode->i_nlink,
1187 NEXT_ORPHAN(inode));
1188 }
1189}
1190
1191#ifdef CONFIG_QUOTA
1192static int ext4_quota_off(struct super_block *sb, int type);
1193
1194static inline void ext4_quotas_off(struct super_block *sb, int type)
1195{
1196 BUG_ON(type > EXT4_MAXQUOTAS);
1197
1198 /* Use our quota_off function to clear inode flags etc. */
1199 for (type--; type >= 0; type--)
1200 ext4_quota_off(sb, type);
1201}
1202
1203/*
1204 * This is a helper function which is used in the mount/remount
1205 * codepaths (which holds s_umount) to fetch the quota file name.
1206 */
1207static inline char *get_qf_name(struct super_block *sb,
1208 struct ext4_sb_info *sbi,
1209 int type)
1210{
1211 return rcu_dereference_protected(sbi->s_qf_names[type],
1212 lockdep_is_held(&sb->s_umount));
1213}
1214#else
1215static inline void ext4_quotas_off(struct super_block *sb, int type)
1216{
1217}
1218#endif
1219
1220static int ext4_percpu_param_init(struct ext4_sb_info *sbi)
1221{
1222 ext4_fsblk_t block;
1223 int err;
1224
1225 block = ext4_count_free_clusters(sbi->s_sb);
1226 ext4_free_blocks_count_set(sbi->s_es, EXT4_C2B(sbi, block));
1227 err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
1228 GFP_KERNEL);
1229 if (!err) {
1230 unsigned long freei = ext4_count_free_inodes(sbi->s_sb);
1231 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
1232 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
1233 GFP_KERNEL);
1234 }
1235 if (!err)
1236 err = percpu_counter_init(&sbi->s_dirs_counter,
1237 ext4_count_dirs(sbi->s_sb), GFP_KERNEL);
1238 if (!err)
1239 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
1240 GFP_KERNEL);
1241 if (!err)
1242 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
1243 GFP_KERNEL);
1244 if (!err)
1245 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
1246
1247 if (err)
1248 ext4_msg(sbi->s_sb, KERN_ERR, "insufficient memory");
1249
1250 return err;
1251}
1252
1253static void ext4_percpu_param_destroy(struct ext4_sb_info *sbi)
1254{
1255 percpu_counter_destroy(&sbi->s_freeclusters_counter);
1256 percpu_counter_destroy(&sbi->s_freeinodes_counter);
1257 percpu_counter_destroy(&sbi->s_dirs_counter);
1258 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1259 percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1260 percpu_free_rwsem(&sbi->s_writepages_rwsem);
1261}
1262
1263static void ext4_group_desc_free(struct ext4_sb_info *sbi)
1264{
1265 struct buffer_head **group_desc;
1266 int i;
1267
1268 rcu_read_lock();
1269 group_desc = rcu_dereference(sbi->s_group_desc);
1270 for (i = 0; i < sbi->s_gdb_count; i++)
1271 brelse(group_desc[i]);
1272 kvfree(group_desc);
1273 rcu_read_unlock();
1274}
1275
1276static void ext4_flex_groups_free(struct ext4_sb_info *sbi)
1277{
1278 struct flex_groups **flex_groups;
1279 int i;
1280
1281 rcu_read_lock();
1282 flex_groups = rcu_dereference(sbi->s_flex_groups);
1283 if (flex_groups) {
1284 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1285 kvfree(flex_groups[i]);
1286 kvfree(flex_groups);
1287 }
1288 rcu_read_unlock();
1289}
1290
1291static void ext4_put_super(struct super_block *sb)
1292{
1293 struct ext4_sb_info *sbi = EXT4_SB(sb);
1294 struct ext4_super_block *es = sbi->s_es;
1295 int aborted = 0;
1296 int err;
1297
1298 /*
1299 * Unregister sysfs before destroying jbd2 journal.
1300 * Since we could still access attr_journal_task attribute via sysfs
1301 * path which could have sbi->s_journal->j_task as NULL
1302 * Unregister sysfs before flush sbi->s_sb_upd_work.
1303 * Since user may read /proc/fs/ext4/xx/mb_groups during umount, If
1304 * read metadata verify failed then will queue error work.
1305 * update_super_work will call start_this_handle may trigger
1306 * BUG_ON.
1307 */
1308 ext4_unregister_sysfs(sb);
1309
1310 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs unmount"))
1311 ext4_msg(sb, KERN_INFO, "unmounting filesystem %pU.",
1312 &sb->s_uuid);
1313
1314 ext4_unregister_li_request(sb);
1315 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
1316
1317 flush_work(&sbi->s_sb_upd_work);
1318 destroy_workqueue(sbi->rsv_conversion_wq);
1319 ext4_release_orphan_info(sb);
1320
1321 if (sbi->s_journal) {
1322 aborted = is_journal_aborted(sbi->s_journal);
1323 err = jbd2_journal_destroy(sbi->s_journal);
1324 sbi->s_journal = NULL;
1325 if ((err < 0) && !aborted) {
1326 ext4_abort(sb, -err, "Couldn't clean up the journal");
1327 }
1328 }
1329
1330 ext4_es_unregister_shrinker(sbi);
1331 timer_shutdown_sync(&sbi->s_err_report);
1332 ext4_release_system_zone(sb);
1333 ext4_mb_release(sb);
1334 ext4_ext_release(sb);
1335
1336 if (!sb_rdonly(sb) && !aborted) {
1337 ext4_clear_feature_journal_needs_recovery(sb);
1338 ext4_clear_feature_orphan_present(sb);
1339 es->s_state = cpu_to_le16(sbi->s_mount_state);
1340 }
1341 if (!sb_rdonly(sb))
1342 ext4_commit_super(sb);
1343
1344 ext4_group_desc_free(sbi);
1345 ext4_flex_groups_free(sbi);
1346 ext4_percpu_param_destroy(sbi);
1347#ifdef CONFIG_QUOTA
1348 for (int i = 0; i < EXT4_MAXQUOTAS; i++)
1349 kfree(get_qf_name(sb, sbi, i));
1350#endif
1351
1352 /* Debugging code just in case the in-memory inode orphan list
1353 * isn't empty. The on-disk one can be non-empty if we've
1354 * detected an error and taken the fs readonly, but the
1355 * in-memory list had better be clean by this point. */
1356 if (!list_empty(&sbi->s_orphan))
1357 dump_orphan_list(sb, sbi);
1358 ASSERT(list_empty(&sbi->s_orphan));
1359
1360 sync_blockdev(sb->s_bdev);
1361 invalidate_bdev(sb->s_bdev);
1362 if (sbi->s_journal_bdev_file) {
1363 /*
1364 * Invalidate the journal device's buffers. We don't want them
1365 * floating about in memory - the physical journal device may
1366 * hotswapped, and it breaks the `ro-after' testing code.
1367 */
1368 sync_blockdev(file_bdev(sbi->s_journal_bdev_file));
1369 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
1370 }
1371
1372 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1373 sbi->s_ea_inode_cache = NULL;
1374
1375 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1376 sbi->s_ea_block_cache = NULL;
1377
1378 ext4_stop_mmpd(sbi);
1379
1380 brelse(sbi->s_sbh);
1381 sb->s_fs_info = NULL;
1382 /*
1383 * Now that we are completely done shutting down the
1384 * superblock, we need to actually destroy the kobject.
1385 */
1386 kobject_put(&sbi->s_kobj);
1387 wait_for_completion(&sbi->s_kobj_unregister);
1388 if (sbi->s_chksum_driver)
1389 crypto_free_shash(sbi->s_chksum_driver);
1390 kfree(sbi->s_blockgroup_lock);
1391 fs_put_dax(sbi->s_daxdev, NULL);
1392 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1393#if IS_ENABLED(CONFIG_UNICODE)
1394 utf8_unload(sb->s_encoding);
1395#endif
1396 kfree(sbi);
1397}
1398
1399static struct kmem_cache *ext4_inode_cachep;
1400
1401/*
1402 * Called inside transaction, so use GFP_NOFS
1403 */
1404static struct inode *ext4_alloc_inode(struct super_block *sb)
1405{
1406 struct ext4_inode_info *ei;
1407
1408 ei = alloc_inode_sb(sb, ext4_inode_cachep, GFP_NOFS);
1409 if (!ei)
1410 return NULL;
1411
1412 inode_set_iversion(&ei->vfs_inode, 1);
1413 ei->i_flags = 0;
1414 spin_lock_init(&ei->i_raw_lock);
1415 ei->i_prealloc_node = RB_ROOT;
1416 atomic_set(&ei->i_prealloc_active, 0);
1417 rwlock_init(&ei->i_prealloc_lock);
1418 ext4_es_init_tree(&ei->i_es_tree);
1419 rwlock_init(&ei->i_es_lock);
1420 INIT_LIST_HEAD(&ei->i_es_list);
1421 ei->i_es_all_nr = 0;
1422 ei->i_es_shk_nr = 0;
1423 ei->i_es_shrink_lblk = 0;
1424 ei->i_reserved_data_blocks = 0;
1425 spin_lock_init(&(ei->i_block_reservation_lock));
1426 ext4_init_pending_tree(&ei->i_pending_tree);
1427#ifdef CONFIG_QUOTA
1428 ei->i_reserved_quota = 0;
1429 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1430#endif
1431 ei->jinode = NULL;
1432 INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1433 spin_lock_init(&ei->i_completed_io_lock);
1434 ei->i_sync_tid = 0;
1435 ei->i_datasync_tid = 0;
1436 atomic_set(&ei->i_unwritten, 0);
1437 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1438 ext4_fc_init_inode(&ei->vfs_inode);
1439 mutex_init(&ei->i_fc_lock);
1440 return &ei->vfs_inode;
1441}
1442
1443static int ext4_drop_inode(struct inode *inode)
1444{
1445 int drop = generic_drop_inode(inode);
1446
1447 if (!drop)
1448 drop = fscrypt_drop_inode(inode);
1449
1450 trace_ext4_drop_inode(inode, drop);
1451 return drop;
1452}
1453
1454static void ext4_free_in_core_inode(struct inode *inode)
1455{
1456 fscrypt_free_inode(inode);
1457 if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1458 pr_warn("%s: inode %ld still in fc list",
1459 __func__, inode->i_ino);
1460 }
1461 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1462}
1463
1464static void ext4_destroy_inode(struct inode *inode)
1465{
1466 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1467 ext4_msg(inode->i_sb, KERN_ERR,
1468 "Inode %lu (%p): orphan list check failed!",
1469 inode->i_ino, EXT4_I(inode));
1470 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1471 EXT4_I(inode), sizeof(struct ext4_inode_info),
1472 true);
1473 dump_stack();
1474 }
1475
1476 if (EXT4_I(inode)->i_reserved_data_blocks)
1477 ext4_msg(inode->i_sb, KERN_ERR,
1478 "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1479 inode->i_ino, EXT4_I(inode),
1480 EXT4_I(inode)->i_reserved_data_blocks);
1481}
1482
1483static void ext4_shutdown(struct super_block *sb)
1484{
1485 ext4_force_shutdown(sb, EXT4_GOING_FLAGS_NOLOGFLUSH);
1486}
1487
1488static void init_once(void *foo)
1489{
1490 struct ext4_inode_info *ei = foo;
1491
1492 INIT_LIST_HEAD(&ei->i_orphan);
1493 init_rwsem(&ei->xattr_sem);
1494 init_rwsem(&ei->i_data_sem);
1495 inode_init_once(&ei->vfs_inode);
1496 ext4_fc_init_inode(&ei->vfs_inode);
1497}
1498
1499static int __init init_inodecache(void)
1500{
1501 ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1502 sizeof(struct ext4_inode_info), 0,
1503 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
1504 offsetof(struct ext4_inode_info, i_data),
1505 sizeof_field(struct ext4_inode_info, i_data),
1506 init_once);
1507 if (ext4_inode_cachep == NULL)
1508 return -ENOMEM;
1509 return 0;
1510}
1511
1512static void destroy_inodecache(void)
1513{
1514 /*
1515 * Make sure all delayed rcu free inodes are flushed before we
1516 * destroy cache.
1517 */
1518 rcu_barrier();
1519 kmem_cache_destroy(ext4_inode_cachep);
1520}
1521
1522void ext4_clear_inode(struct inode *inode)
1523{
1524 ext4_fc_del(inode);
1525 invalidate_inode_buffers(inode);
1526 clear_inode(inode);
1527 ext4_discard_preallocations(inode);
1528 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1529 dquot_drop(inode);
1530 if (EXT4_I(inode)->jinode) {
1531 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1532 EXT4_I(inode)->jinode);
1533 jbd2_free_inode(EXT4_I(inode)->jinode);
1534 EXT4_I(inode)->jinode = NULL;
1535 }
1536 fscrypt_put_encryption_info(inode);
1537 fsverity_cleanup_inode(inode);
1538}
1539
1540static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1541 u64 ino, u32 generation)
1542{
1543 struct inode *inode;
1544
1545 /*
1546 * Currently we don't know the generation for parent directory, so
1547 * a generation of 0 means "accept any"
1548 */
1549 inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1550 if (IS_ERR(inode))
1551 return ERR_CAST(inode);
1552 if (generation && inode->i_generation != generation) {
1553 iput(inode);
1554 return ERR_PTR(-ESTALE);
1555 }
1556
1557 return inode;
1558}
1559
1560static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1561 int fh_len, int fh_type)
1562{
1563 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1564 ext4_nfs_get_inode);
1565}
1566
1567static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1568 int fh_len, int fh_type)
1569{
1570 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1571 ext4_nfs_get_inode);
1572}
1573
1574static int ext4_nfs_commit_metadata(struct inode *inode)
1575{
1576 struct writeback_control wbc = {
1577 .sync_mode = WB_SYNC_ALL
1578 };
1579
1580 trace_ext4_nfs_commit_metadata(inode);
1581 return ext4_write_inode(inode, &wbc);
1582}
1583
1584#ifdef CONFIG_QUOTA
1585static const char * const quotatypes[] = INITQFNAMES;
1586#define QTYPE2NAME(t) (quotatypes[t])
1587
1588static int ext4_write_dquot(struct dquot *dquot);
1589static int ext4_acquire_dquot(struct dquot *dquot);
1590static int ext4_release_dquot(struct dquot *dquot);
1591static int ext4_mark_dquot_dirty(struct dquot *dquot);
1592static int ext4_write_info(struct super_block *sb, int type);
1593static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1594 const struct path *path);
1595static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1596 size_t len, loff_t off);
1597static ssize_t ext4_quota_write(struct super_block *sb, int type,
1598 const char *data, size_t len, loff_t off);
1599static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1600 unsigned int flags);
1601
1602static struct dquot __rcu **ext4_get_dquots(struct inode *inode)
1603{
1604 return EXT4_I(inode)->i_dquot;
1605}
1606
1607static const struct dquot_operations ext4_quota_operations = {
1608 .get_reserved_space = ext4_get_reserved_space,
1609 .write_dquot = ext4_write_dquot,
1610 .acquire_dquot = ext4_acquire_dquot,
1611 .release_dquot = ext4_release_dquot,
1612 .mark_dirty = ext4_mark_dquot_dirty,
1613 .write_info = ext4_write_info,
1614 .alloc_dquot = dquot_alloc,
1615 .destroy_dquot = dquot_destroy,
1616 .get_projid = ext4_get_projid,
1617 .get_inode_usage = ext4_get_inode_usage,
1618 .get_next_id = dquot_get_next_id,
1619};
1620
1621static const struct quotactl_ops ext4_qctl_operations = {
1622 .quota_on = ext4_quota_on,
1623 .quota_off = ext4_quota_off,
1624 .quota_sync = dquot_quota_sync,
1625 .get_state = dquot_get_state,
1626 .set_info = dquot_set_dqinfo,
1627 .get_dqblk = dquot_get_dqblk,
1628 .set_dqblk = dquot_set_dqblk,
1629 .get_nextdqblk = dquot_get_next_dqblk,
1630};
1631#endif
1632
1633static const struct super_operations ext4_sops = {
1634 .alloc_inode = ext4_alloc_inode,
1635 .free_inode = ext4_free_in_core_inode,
1636 .destroy_inode = ext4_destroy_inode,
1637 .write_inode = ext4_write_inode,
1638 .dirty_inode = ext4_dirty_inode,
1639 .drop_inode = ext4_drop_inode,
1640 .evict_inode = ext4_evict_inode,
1641 .put_super = ext4_put_super,
1642 .sync_fs = ext4_sync_fs,
1643 .freeze_fs = ext4_freeze,
1644 .unfreeze_fs = ext4_unfreeze,
1645 .statfs = ext4_statfs,
1646 .show_options = ext4_show_options,
1647 .shutdown = ext4_shutdown,
1648#ifdef CONFIG_QUOTA
1649 .quota_read = ext4_quota_read,
1650 .quota_write = ext4_quota_write,
1651 .get_dquots = ext4_get_dquots,
1652#endif
1653};
1654
1655static const struct export_operations ext4_export_ops = {
1656 .encode_fh = generic_encode_ino32_fh,
1657 .fh_to_dentry = ext4_fh_to_dentry,
1658 .fh_to_parent = ext4_fh_to_parent,
1659 .get_parent = ext4_get_parent,
1660 .commit_metadata = ext4_nfs_commit_metadata,
1661};
1662
1663enum {
1664 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1665 Opt_resgid, Opt_resuid, Opt_sb,
1666 Opt_nouid32, Opt_debug, Opt_removed,
1667 Opt_user_xattr, Opt_acl,
1668 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1669 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1670 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1671 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1672 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1673 Opt_inlinecrypt,
1674 Opt_usrjquota, Opt_grpjquota, Opt_quota,
1675 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1676 Opt_usrquota, Opt_grpquota, Opt_prjquota,
1677 Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1678 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1679 Opt_nowarn_on_error, Opt_mblk_io_submit, Opt_debug_want_extra_isize,
1680 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1681 Opt_inode_readahead_blks, Opt_journal_ioprio,
1682 Opt_dioread_nolock, Opt_dioread_lock,
1683 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1684 Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1685 Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1686 Opt_errors, Opt_data, Opt_data_err, Opt_jqfmt, Opt_dax_type,
1687#ifdef CONFIG_EXT4_DEBUG
1688 Opt_fc_debug_max_replay, Opt_fc_debug_force
1689#endif
1690};
1691
1692static const struct constant_table ext4_param_errors[] = {
1693 {"continue", EXT4_MOUNT_ERRORS_CONT},
1694 {"panic", EXT4_MOUNT_ERRORS_PANIC},
1695 {"remount-ro", EXT4_MOUNT_ERRORS_RO},
1696 {}
1697};
1698
1699static const struct constant_table ext4_param_data[] = {
1700 {"journal", EXT4_MOUNT_JOURNAL_DATA},
1701 {"ordered", EXT4_MOUNT_ORDERED_DATA},
1702 {"writeback", EXT4_MOUNT_WRITEBACK_DATA},
1703 {}
1704};
1705
1706static const struct constant_table ext4_param_data_err[] = {
1707 {"abort", Opt_data_err_abort},
1708 {"ignore", Opt_data_err_ignore},
1709 {}
1710};
1711
1712static const struct constant_table ext4_param_jqfmt[] = {
1713 {"vfsold", QFMT_VFS_OLD},
1714 {"vfsv0", QFMT_VFS_V0},
1715 {"vfsv1", QFMT_VFS_V1},
1716 {}
1717};
1718
1719static const struct constant_table ext4_param_dax[] = {
1720 {"always", Opt_dax_always},
1721 {"inode", Opt_dax_inode},
1722 {"never", Opt_dax_never},
1723 {}
1724};
1725
1726/* String parameter that allows empty argument */
1727#define fsparam_string_empty(NAME, OPT) \
1728 __fsparam(fs_param_is_string, NAME, OPT, fs_param_can_be_empty, NULL)
1729
1730/*
1731 * Mount option specification
1732 * We don't use fsparam_flag_no because of the way we set the
1733 * options and the way we show them in _ext4_show_options(). To
1734 * keep the changes to a minimum, let's keep the negative options
1735 * separate for now.
1736 */
1737static const struct fs_parameter_spec ext4_param_specs[] = {
1738 fsparam_flag ("bsddf", Opt_bsd_df),
1739 fsparam_flag ("minixdf", Opt_minix_df),
1740 fsparam_flag ("grpid", Opt_grpid),
1741 fsparam_flag ("bsdgroups", Opt_grpid),
1742 fsparam_flag ("nogrpid", Opt_nogrpid),
1743 fsparam_flag ("sysvgroups", Opt_nogrpid),
1744 fsparam_u32 ("resgid", Opt_resgid),
1745 fsparam_u32 ("resuid", Opt_resuid),
1746 fsparam_u32 ("sb", Opt_sb),
1747 fsparam_enum ("errors", Opt_errors, ext4_param_errors),
1748 fsparam_flag ("nouid32", Opt_nouid32),
1749 fsparam_flag ("debug", Opt_debug),
1750 fsparam_flag ("oldalloc", Opt_removed),
1751 fsparam_flag ("orlov", Opt_removed),
1752 fsparam_flag ("user_xattr", Opt_user_xattr),
1753 fsparam_flag ("acl", Opt_acl),
1754 fsparam_flag ("norecovery", Opt_noload),
1755 fsparam_flag ("noload", Opt_noload),
1756 fsparam_flag ("bh", Opt_removed),
1757 fsparam_flag ("nobh", Opt_removed),
1758 fsparam_u32 ("commit", Opt_commit),
1759 fsparam_u32 ("min_batch_time", Opt_min_batch_time),
1760 fsparam_u32 ("max_batch_time", Opt_max_batch_time),
1761 fsparam_u32 ("journal_dev", Opt_journal_dev),
1762 fsparam_bdev ("journal_path", Opt_journal_path),
1763 fsparam_flag ("journal_checksum", Opt_journal_checksum),
1764 fsparam_flag ("nojournal_checksum", Opt_nojournal_checksum),
1765 fsparam_flag ("journal_async_commit",Opt_journal_async_commit),
1766 fsparam_flag ("abort", Opt_abort),
1767 fsparam_enum ("data", Opt_data, ext4_param_data),
1768 fsparam_enum ("data_err", Opt_data_err,
1769 ext4_param_data_err),
1770 fsparam_string_empty
1771 ("usrjquota", Opt_usrjquota),
1772 fsparam_string_empty
1773 ("grpjquota", Opt_grpjquota),
1774 fsparam_enum ("jqfmt", Opt_jqfmt, ext4_param_jqfmt),
1775 fsparam_flag ("grpquota", Opt_grpquota),
1776 fsparam_flag ("quota", Opt_quota),
1777 fsparam_flag ("noquota", Opt_noquota),
1778 fsparam_flag ("usrquota", Opt_usrquota),
1779 fsparam_flag ("prjquota", Opt_prjquota),
1780 fsparam_flag ("barrier", Opt_barrier),
1781 fsparam_u32 ("barrier", Opt_barrier),
1782 fsparam_flag ("nobarrier", Opt_nobarrier),
1783 fsparam_flag ("i_version", Opt_removed),
1784 fsparam_flag ("dax", Opt_dax),
1785 fsparam_enum ("dax", Opt_dax_type, ext4_param_dax),
1786 fsparam_u32 ("stripe", Opt_stripe),
1787 fsparam_flag ("delalloc", Opt_delalloc),
1788 fsparam_flag ("nodelalloc", Opt_nodelalloc),
1789 fsparam_flag ("warn_on_error", Opt_warn_on_error),
1790 fsparam_flag ("nowarn_on_error", Opt_nowarn_on_error),
1791 fsparam_u32 ("debug_want_extra_isize",
1792 Opt_debug_want_extra_isize),
1793 fsparam_flag ("mblk_io_submit", Opt_removed),
1794 fsparam_flag ("nomblk_io_submit", Opt_removed),
1795 fsparam_flag ("block_validity", Opt_block_validity),
1796 fsparam_flag ("noblock_validity", Opt_noblock_validity),
1797 fsparam_u32 ("inode_readahead_blks",
1798 Opt_inode_readahead_blks),
1799 fsparam_u32 ("journal_ioprio", Opt_journal_ioprio),
1800 fsparam_u32 ("auto_da_alloc", Opt_auto_da_alloc),
1801 fsparam_flag ("auto_da_alloc", Opt_auto_da_alloc),
1802 fsparam_flag ("noauto_da_alloc", Opt_noauto_da_alloc),
1803 fsparam_flag ("dioread_nolock", Opt_dioread_nolock),
1804 fsparam_flag ("nodioread_nolock", Opt_dioread_lock),
1805 fsparam_flag ("dioread_lock", Opt_dioread_lock),
1806 fsparam_flag ("discard", Opt_discard),
1807 fsparam_flag ("nodiscard", Opt_nodiscard),
1808 fsparam_u32 ("init_itable", Opt_init_itable),
1809 fsparam_flag ("init_itable", Opt_init_itable),
1810 fsparam_flag ("noinit_itable", Opt_noinit_itable),
1811#ifdef CONFIG_EXT4_DEBUG
1812 fsparam_flag ("fc_debug_force", Opt_fc_debug_force),
1813 fsparam_u32 ("fc_debug_max_replay", Opt_fc_debug_max_replay),
1814#endif
1815 fsparam_u32 ("max_dir_size_kb", Opt_max_dir_size_kb),
1816 fsparam_flag ("test_dummy_encryption",
1817 Opt_test_dummy_encryption),
1818 fsparam_string ("test_dummy_encryption",
1819 Opt_test_dummy_encryption),
1820 fsparam_flag ("inlinecrypt", Opt_inlinecrypt),
1821 fsparam_flag ("nombcache", Opt_nombcache),
1822 fsparam_flag ("no_mbcache", Opt_nombcache), /* for backward compatibility */
1823 fsparam_flag ("prefetch_block_bitmaps",
1824 Opt_removed),
1825 fsparam_flag ("no_prefetch_block_bitmaps",
1826 Opt_no_prefetch_block_bitmaps),
1827 fsparam_s32 ("mb_optimize_scan", Opt_mb_optimize_scan),
1828 fsparam_string ("check", Opt_removed), /* mount option from ext2/3 */
1829 fsparam_flag ("nocheck", Opt_removed), /* mount option from ext2/3 */
1830 fsparam_flag ("reservation", Opt_removed), /* mount option from ext2/3 */
1831 fsparam_flag ("noreservation", Opt_removed), /* mount option from ext2/3 */
1832 fsparam_u32 ("journal", Opt_removed), /* mount option from ext2/3 */
1833 {}
1834};
1835
1836#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1837
1838#define MOPT_SET 0x0001
1839#define MOPT_CLEAR 0x0002
1840#define MOPT_NOSUPPORT 0x0004
1841#define MOPT_EXPLICIT 0x0008
1842#ifdef CONFIG_QUOTA
1843#define MOPT_Q 0
1844#define MOPT_QFMT 0x0010
1845#else
1846#define MOPT_Q MOPT_NOSUPPORT
1847#define MOPT_QFMT MOPT_NOSUPPORT
1848#endif
1849#define MOPT_NO_EXT2 0x0020
1850#define MOPT_NO_EXT3 0x0040
1851#define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1852#define MOPT_SKIP 0x0080
1853#define MOPT_2 0x0100
1854
1855static const struct mount_opts {
1856 int token;
1857 int mount_opt;
1858 int flags;
1859} ext4_mount_opts[] = {
1860 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1861 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1862 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1863 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1864 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1865 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1866 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1867 MOPT_EXT4_ONLY | MOPT_SET},
1868 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1869 MOPT_EXT4_ONLY | MOPT_CLEAR},
1870 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1871 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1872 {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1873 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1874 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1875 MOPT_EXT4_ONLY | MOPT_CLEAR},
1876 {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1877 {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1878 {Opt_commit, 0, MOPT_NO_EXT2},
1879 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1880 MOPT_EXT4_ONLY | MOPT_CLEAR},
1881 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1882 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1883 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1884 EXT4_MOUNT_JOURNAL_CHECKSUM),
1885 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1886 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1887 {Opt_data_err, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_NO_EXT2},
1888 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1889 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1890 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1891 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1892 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1893 {Opt_dax_type, 0, MOPT_EXT4_ONLY},
1894 {Opt_journal_dev, 0, MOPT_NO_EXT2},
1895 {Opt_journal_path, 0, MOPT_NO_EXT2},
1896 {Opt_journal_ioprio, 0, MOPT_NO_EXT2},
1897 {Opt_data, 0, MOPT_NO_EXT2},
1898 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1899#ifdef CONFIG_EXT4_FS_POSIX_ACL
1900 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1901#else
1902 {Opt_acl, 0, MOPT_NOSUPPORT},
1903#endif
1904 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1905 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1906 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1907 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1908 MOPT_SET | MOPT_Q},
1909 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1910 MOPT_SET | MOPT_Q},
1911 {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1912 MOPT_SET | MOPT_Q},
1913 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1914 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1915 MOPT_CLEAR | MOPT_Q},
1916 {Opt_usrjquota, 0, MOPT_Q},
1917 {Opt_grpjquota, 0, MOPT_Q},
1918 {Opt_jqfmt, 0, MOPT_QFMT},
1919 {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1920 {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
1921 MOPT_SET},
1922#ifdef CONFIG_EXT4_DEBUG
1923 {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
1924 MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
1925#endif
1926 {Opt_abort, EXT4_MOUNT2_ABORT, MOPT_SET | MOPT_2},
1927 {Opt_err, 0, 0}
1928};
1929
1930#if IS_ENABLED(CONFIG_UNICODE)
1931static const struct ext4_sb_encodings {
1932 __u16 magic;
1933 char *name;
1934 unsigned int version;
1935} ext4_sb_encoding_map[] = {
1936 {EXT4_ENC_UTF8_12_1, "utf8", UNICODE_AGE(12, 1, 0)},
1937};
1938
1939static const struct ext4_sb_encodings *
1940ext4_sb_read_encoding(const struct ext4_super_block *es)
1941{
1942 __u16 magic = le16_to_cpu(es->s_encoding);
1943 int i;
1944
1945 for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1946 if (magic == ext4_sb_encoding_map[i].magic)
1947 return &ext4_sb_encoding_map[i];
1948
1949 return NULL;
1950}
1951#endif
1952
1953#define EXT4_SPEC_JQUOTA (1 << 0)
1954#define EXT4_SPEC_JQFMT (1 << 1)
1955#define EXT4_SPEC_DATAJ (1 << 2)
1956#define EXT4_SPEC_SB_BLOCK (1 << 3)
1957#define EXT4_SPEC_JOURNAL_DEV (1 << 4)
1958#define EXT4_SPEC_JOURNAL_IOPRIO (1 << 5)
1959#define EXT4_SPEC_s_want_extra_isize (1 << 7)
1960#define EXT4_SPEC_s_max_batch_time (1 << 8)
1961#define EXT4_SPEC_s_min_batch_time (1 << 9)
1962#define EXT4_SPEC_s_inode_readahead_blks (1 << 10)
1963#define EXT4_SPEC_s_li_wait_mult (1 << 11)
1964#define EXT4_SPEC_s_max_dir_size_kb (1 << 12)
1965#define EXT4_SPEC_s_stripe (1 << 13)
1966#define EXT4_SPEC_s_resuid (1 << 14)
1967#define EXT4_SPEC_s_resgid (1 << 15)
1968#define EXT4_SPEC_s_commit_interval (1 << 16)
1969#define EXT4_SPEC_s_fc_debug_max_replay (1 << 17)
1970#define EXT4_SPEC_s_sb_block (1 << 18)
1971#define EXT4_SPEC_mb_optimize_scan (1 << 19)
1972
1973struct ext4_fs_context {
1974 char *s_qf_names[EXT4_MAXQUOTAS];
1975 struct fscrypt_dummy_policy dummy_enc_policy;
1976 int s_jquota_fmt; /* Format of quota to use */
1977#ifdef CONFIG_EXT4_DEBUG
1978 int s_fc_debug_max_replay;
1979#endif
1980 unsigned short qname_spec;
1981 unsigned long vals_s_flags; /* Bits to set in s_flags */
1982 unsigned long mask_s_flags; /* Bits changed in s_flags */
1983 unsigned long journal_devnum;
1984 unsigned long s_commit_interval;
1985 unsigned long s_stripe;
1986 unsigned int s_inode_readahead_blks;
1987 unsigned int s_want_extra_isize;
1988 unsigned int s_li_wait_mult;
1989 unsigned int s_max_dir_size_kb;
1990 unsigned int journal_ioprio;
1991 unsigned int vals_s_mount_opt;
1992 unsigned int mask_s_mount_opt;
1993 unsigned int vals_s_mount_opt2;
1994 unsigned int mask_s_mount_opt2;
1995 unsigned int opt_flags; /* MOPT flags */
1996 unsigned int spec;
1997 u32 s_max_batch_time;
1998 u32 s_min_batch_time;
1999 kuid_t s_resuid;
2000 kgid_t s_resgid;
2001 ext4_fsblk_t s_sb_block;
2002};
2003
2004static void ext4_fc_free(struct fs_context *fc)
2005{
2006 struct ext4_fs_context *ctx = fc->fs_private;
2007 int i;
2008
2009 if (!ctx)
2010 return;
2011
2012 for (i = 0; i < EXT4_MAXQUOTAS; i++)
2013 kfree(ctx->s_qf_names[i]);
2014
2015 fscrypt_free_dummy_policy(&ctx->dummy_enc_policy);
2016 kfree(ctx);
2017}
2018
2019int ext4_init_fs_context(struct fs_context *fc)
2020{
2021 struct ext4_fs_context *ctx;
2022
2023 ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2024 if (!ctx)
2025 return -ENOMEM;
2026
2027 fc->fs_private = ctx;
2028 fc->ops = &ext4_context_ops;
2029
2030 return 0;
2031}
2032
2033#ifdef CONFIG_QUOTA
2034/*
2035 * Note the name of the specified quota file.
2036 */
2037static int note_qf_name(struct fs_context *fc, int qtype,
2038 struct fs_parameter *param)
2039{
2040 struct ext4_fs_context *ctx = fc->fs_private;
2041 char *qname;
2042
2043 if (param->size < 1) {
2044 ext4_msg(NULL, KERN_ERR, "Missing quota name");
2045 return -EINVAL;
2046 }
2047 if (strchr(param->string, '/')) {
2048 ext4_msg(NULL, KERN_ERR,
2049 "quotafile must be on filesystem root");
2050 return -EINVAL;
2051 }
2052 if (ctx->s_qf_names[qtype]) {
2053 if (strcmp(ctx->s_qf_names[qtype], param->string) != 0) {
2054 ext4_msg(NULL, KERN_ERR,
2055 "%s quota file already specified",
2056 QTYPE2NAME(qtype));
2057 return -EINVAL;
2058 }
2059 return 0;
2060 }
2061
2062 qname = kmemdup_nul(param->string, param->size, GFP_KERNEL);
2063 if (!qname) {
2064 ext4_msg(NULL, KERN_ERR,
2065 "Not enough memory for storing quotafile name");
2066 return -ENOMEM;
2067 }
2068 ctx->s_qf_names[qtype] = qname;
2069 ctx->qname_spec |= 1 << qtype;
2070 ctx->spec |= EXT4_SPEC_JQUOTA;
2071 return 0;
2072}
2073
2074/*
2075 * Clear the name of the specified quota file.
2076 */
2077static int unnote_qf_name(struct fs_context *fc, int qtype)
2078{
2079 struct ext4_fs_context *ctx = fc->fs_private;
2080
2081 if (ctx->s_qf_names[qtype])
2082 kfree(ctx->s_qf_names[qtype]);
2083
2084 ctx->s_qf_names[qtype] = NULL;
2085 ctx->qname_spec |= 1 << qtype;
2086 ctx->spec |= EXT4_SPEC_JQUOTA;
2087 return 0;
2088}
2089#endif
2090
2091static int ext4_parse_test_dummy_encryption(const struct fs_parameter *param,
2092 struct ext4_fs_context *ctx)
2093{
2094 int err;
2095
2096 if (!IS_ENABLED(CONFIG_FS_ENCRYPTION)) {
2097 ext4_msg(NULL, KERN_WARNING,
2098 "test_dummy_encryption option not supported");
2099 return -EINVAL;
2100 }
2101 err = fscrypt_parse_test_dummy_encryption(param,
2102 &ctx->dummy_enc_policy);
2103 if (err == -EINVAL) {
2104 ext4_msg(NULL, KERN_WARNING,
2105 "Value of option \"%s\" is unrecognized", param->key);
2106 } else if (err == -EEXIST) {
2107 ext4_msg(NULL, KERN_WARNING,
2108 "Conflicting test_dummy_encryption options");
2109 return -EINVAL;
2110 }
2111 return err;
2112}
2113
2114#define EXT4_SET_CTX(name) \
2115static inline void ctx_set_##name(struct ext4_fs_context *ctx, \
2116 unsigned long flag) \
2117{ \
2118 ctx->mask_s_##name |= flag; \
2119 ctx->vals_s_##name |= flag; \
2120}
2121
2122#define EXT4_CLEAR_CTX(name) \
2123static inline void ctx_clear_##name(struct ext4_fs_context *ctx, \
2124 unsigned long flag) \
2125{ \
2126 ctx->mask_s_##name |= flag; \
2127 ctx->vals_s_##name &= ~flag; \
2128}
2129
2130#define EXT4_TEST_CTX(name) \
2131static inline unsigned long \
2132ctx_test_##name(struct ext4_fs_context *ctx, unsigned long flag) \
2133{ \
2134 return (ctx->vals_s_##name & flag); \
2135}
2136
2137EXT4_SET_CTX(flags); /* set only */
2138EXT4_SET_CTX(mount_opt);
2139EXT4_CLEAR_CTX(mount_opt);
2140EXT4_TEST_CTX(mount_opt);
2141EXT4_SET_CTX(mount_opt2);
2142EXT4_CLEAR_CTX(mount_opt2);
2143EXT4_TEST_CTX(mount_opt2);
2144
2145static int ext4_parse_param(struct fs_context *fc, struct fs_parameter *param)
2146{
2147 struct ext4_fs_context *ctx = fc->fs_private;
2148 struct fs_parse_result result;
2149 const struct mount_opts *m;
2150 int is_remount;
2151 kuid_t uid;
2152 kgid_t gid;
2153 int token;
2154
2155 token = fs_parse(fc, ext4_param_specs, param, &result);
2156 if (token < 0)
2157 return token;
2158 is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2159
2160 for (m = ext4_mount_opts; m->token != Opt_err; m++)
2161 if (token == m->token)
2162 break;
2163
2164 ctx->opt_flags |= m->flags;
2165
2166 if (m->flags & MOPT_EXPLICIT) {
2167 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2168 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_EXPLICIT_DELALLOC);
2169 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2170 ctx_set_mount_opt2(ctx,
2171 EXT4_MOUNT2_EXPLICIT_JOURNAL_CHECKSUM);
2172 } else
2173 return -EINVAL;
2174 }
2175
2176 if (m->flags & MOPT_NOSUPPORT) {
2177 ext4_msg(NULL, KERN_ERR, "%s option not supported",
2178 param->key);
2179 return 0;
2180 }
2181
2182 switch (token) {
2183#ifdef CONFIG_QUOTA
2184 case Opt_usrjquota:
2185 if (!*param->string)
2186 return unnote_qf_name(fc, USRQUOTA);
2187 else
2188 return note_qf_name(fc, USRQUOTA, param);
2189 case Opt_grpjquota:
2190 if (!*param->string)
2191 return unnote_qf_name(fc, GRPQUOTA);
2192 else
2193 return note_qf_name(fc, GRPQUOTA, param);
2194#endif
2195 case Opt_sb:
2196 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2197 ext4_msg(NULL, KERN_WARNING,
2198 "Ignoring %s option on remount", param->key);
2199 } else {
2200 ctx->s_sb_block = result.uint_32;
2201 ctx->spec |= EXT4_SPEC_s_sb_block;
2202 }
2203 return 0;
2204 case Opt_removed:
2205 ext4_msg(NULL, KERN_WARNING, "Ignoring removed %s option",
2206 param->key);
2207 return 0;
2208 case Opt_inlinecrypt:
2209#ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2210 ctx_set_flags(ctx, SB_INLINECRYPT);
2211#else
2212 ext4_msg(NULL, KERN_ERR, "inline encryption not supported");
2213#endif
2214 return 0;
2215 case Opt_errors:
2216 ctx_clear_mount_opt(ctx, EXT4_MOUNT_ERRORS_MASK);
2217 ctx_set_mount_opt(ctx, result.uint_32);
2218 return 0;
2219#ifdef CONFIG_QUOTA
2220 case Opt_jqfmt:
2221 ctx->s_jquota_fmt = result.uint_32;
2222 ctx->spec |= EXT4_SPEC_JQFMT;
2223 return 0;
2224#endif
2225 case Opt_data:
2226 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2227 ctx_set_mount_opt(ctx, result.uint_32);
2228 ctx->spec |= EXT4_SPEC_DATAJ;
2229 return 0;
2230 case Opt_commit:
2231 if (result.uint_32 == 0)
2232 result.uint_32 = JBD2_DEFAULT_MAX_COMMIT_AGE;
2233 else if (result.uint_32 > INT_MAX / HZ) {
2234 ext4_msg(NULL, KERN_ERR,
2235 "Invalid commit interval %d, "
2236 "must be smaller than %d",
2237 result.uint_32, INT_MAX / HZ);
2238 return -EINVAL;
2239 }
2240 ctx->s_commit_interval = HZ * result.uint_32;
2241 ctx->spec |= EXT4_SPEC_s_commit_interval;
2242 return 0;
2243 case Opt_debug_want_extra_isize:
2244 if ((result.uint_32 & 1) || (result.uint_32 < 4)) {
2245 ext4_msg(NULL, KERN_ERR,
2246 "Invalid want_extra_isize %d", result.uint_32);
2247 return -EINVAL;
2248 }
2249 ctx->s_want_extra_isize = result.uint_32;
2250 ctx->spec |= EXT4_SPEC_s_want_extra_isize;
2251 return 0;
2252 case Opt_max_batch_time:
2253 ctx->s_max_batch_time = result.uint_32;
2254 ctx->spec |= EXT4_SPEC_s_max_batch_time;
2255 return 0;
2256 case Opt_min_batch_time:
2257 ctx->s_min_batch_time = result.uint_32;
2258 ctx->spec |= EXT4_SPEC_s_min_batch_time;
2259 return 0;
2260 case Opt_inode_readahead_blks:
2261 if (result.uint_32 &&
2262 (result.uint_32 > (1 << 30) ||
2263 !is_power_of_2(result.uint_32))) {
2264 ext4_msg(NULL, KERN_ERR,
2265 "EXT4-fs: inode_readahead_blks must be "
2266 "0 or a power of 2 smaller than 2^31");
2267 return -EINVAL;
2268 }
2269 ctx->s_inode_readahead_blks = result.uint_32;
2270 ctx->spec |= EXT4_SPEC_s_inode_readahead_blks;
2271 return 0;
2272 case Opt_init_itable:
2273 ctx_set_mount_opt(ctx, EXT4_MOUNT_INIT_INODE_TABLE);
2274 ctx->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
2275 if (param->type == fs_value_is_string)
2276 ctx->s_li_wait_mult = result.uint_32;
2277 ctx->spec |= EXT4_SPEC_s_li_wait_mult;
2278 return 0;
2279 case Opt_max_dir_size_kb:
2280 ctx->s_max_dir_size_kb = result.uint_32;
2281 ctx->spec |= EXT4_SPEC_s_max_dir_size_kb;
2282 return 0;
2283#ifdef CONFIG_EXT4_DEBUG
2284 case Opt_fc_debug_max_replay:
2285 ctx->s_fc_debug_max_replay = result.uint_32;
2286 ctx->spec |= EXT4_SPEC_s_fc_debug_max_replay;
2287 return 0;
2288#endif
2289 case Opt_stripe:
2290 ctx->s_stripe = result.uint_32;
2291 ctx->spec |= EXT4_SPEC_s_stripe;
2292 return 0;
2293 case Opt_resuid:
2294 uid = make_kuid(current_user_ns(), result.uint_32);
2295 if (!uid_valid(uid)) {
2296 ext4_msg(NULL, KERN_ERR, "Invalid uid value %d",
2297 result.uint_32);
2298 return -EINVAL;
2299 }
2300 ctx->s_resuid = uid;
2301 ctx->spec |= EXT4_SPEC_s_resuid;
2302 return 0;
2303 case Opt_resgid:
2304 gid = make_kgid(current_user_ns(), result.uint_32);
2305 if (!gid_valid(gid)) {
2306 ext4_msg(NULL, KERN_ERR, "Invalid gid value %d",
2307 result.uint_32);
2308 return -EINVAL;
2309 }
2310 ctx->s_resgid = gid;
2311 ctx->spec |= EXT4_SPEC_s_resgid;
2312 return 0;
2313 case Opt_journal_dev:
2314 if (is_remount) {
2315 ext4_msg(NULL, KERN_ERR,
2316 "Cannot specify journal on remount");
2317 return -EINVAL;
2318 }
2319 ctx->journal_devnum = result.uint_32;
2320 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2321 return 0;
2322 case Opt_journal_path:
2323 {
2324 struct inode *journal_inode;
2325 struct path path;
2326 int error;
2327
2328 if (is_remount) {
2329 ext4_msg(NULL, KERN_ERR,
2330 "Cannot specify journal on remount");
2331 return -EINVAL;
2332 }
2333
2334 error = fs_lookup_param(fc, param, 1, LOOKUP_FOLLOW, &path);
2335 if (error) {
2336 ext4_msg(NULL, KERN_ERR, "error: could not find "
2337 "journal device path");
2338 return -EINVAL;
2339 }
2340
2341 journal_inode = d_inode(path.dentry);
2342 ctx->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2343 ctx->spec |= EXT4_SPEC_JOURNAL_DEV;
2344 path_put(&path);
2345 return 0;
2346 }
2347 case Opt_journal_ioprio:
2348 if (result.uint_32 > 7) {
2349 ext4_msg(NULL, KERN_ERR, "Invalid journal IO priority"
2350 " (must be 0-7)");
2351 return -EINVAL;
2352 }
2353 ctx->journal_ioprio =
2354 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, result.uint_32);
2355 ctx->spec |= EXT4_SPEC_JOURNAL_IOPRIO;
2356 return 0;
2357 case Opt_test_dummy_encryption:
2358 return ext4_parse_test_dummy_encryption(param, ctx);
2359 case Opt_dax:
2360 case Opt_dax_type:
2361#ifdef CONFIG_FS_DAX
2362 {
2363 int type = (token == Opt_dax) ?
2364 Opt_dax : result.uint_32;
2365
2366 switch (type) {
2367 case Opt_dax:
2368 case Opt_dax_always:
2369 ctx_set_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2370 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2371 break;
2372 case Opt_dax_never:
2373 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2374 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2375 break;
2376 case Opt_dax_inode:
2377 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS);
2378 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER);
2379 /* Strictly for printing options */
2380 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE);
2381 break;
2382 }
2383 return 0;
2384 }
2385#else
2386 ext4_msg(NULL, KERN_INFO, "dax option not supported");
2387 return -EINVAL;
2388#endif
2389 case Opt_data_err:
2390 if (result.uint_32 == Opt_data_err_abort)
2391 ctx_set_mount_opt(ctx, m->mount_opt);
2392 else if (result.uint_32 == Opt_data_err_ignore)
2393 ctx_clear_mount_opt(ctx, m->mount_opt);
2394 return 0;
2395 case Opt_mb_optimize_scan:
2396 if (result.int_32 == 1) {
2397 ctx_set_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2398 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2399 } else if (result.int_32 == 0) {
2400 ctx_clear_mount_opt2(ctx, EXT4_MOUNT2_MB_OPTIMIZE_SCAN);
2401 ctx->spec |= EXT4_SPEC_mb_optimize_scan;
2402 } else {
2403 ext4_msg(NULL, KERN_WARNING,
2404 "mb_optimize_scan should be set to 0 or 1.");
2405 return -EINVAL;
2406 }
2407 return 0;
2408 }
2409
2410 /*
2411 * At this point we should only be getting options requiring MOPT_SET,
2412 * or MOPT_CLEAR. Anything else is a bug
2413 */
2414 if (m->token == Opt_err) {
2415 ext4_msg(NULL, KERN_WARNING, "buggy handling of option %s",
2416 param->key);
2417 WARN_ON(1);
2418 return -EINVAL;
2419 }
2420
2421 else {
2422 unsigned int set = 0;
2423
2424 if ((param->type == fs_value_is_flag) ||
2425 result.uint_32 > 0)
2426 set = 1;
2427
2428 if (m->flags & MOPT_CLEAR)
2429 set = !set;
2430 else if (unlikely(!(m->flags & MOPT_SET))) {
2431 ext4_msg(NULL, KERN_WARNING,
2432 "buggy handling of option %s",
2433 param->key);
2434 WARN_ON(1);
2435 return -EINVAL;
2436 }
2437 if (m->flags & MOPT_2) {
2438 if (set != 0)
2439 ctx_set_mount_opt2(ctx, m->mount_opt);
2440 else
2441 ctx_clear_mount_opt2(ctx, m->mount_opt);
2442 } else {
2443 if (set != 0)
2444 ctx_set_mount_opt(ctx, m->mount_opt);
2445 else
2446 ctx_clear_mount_opt(ctx, m->mount_opt);
2447 }
2448 }
2449
2450 return 0;
2451}
2452
2453static int parse_options(struct fs_context *fc, char *options)
2454{
2455 struct fs_parameter param;
2456 int ret;
2457 char *key;
2458
2459 if (!options)
2460 return 0;
2461
2462 while ((key = strsep(&options, ",")) != NULL) {
2463 if (*key) {
2464 size_t v_len = 0;
2465 char *value = strchr(key, '=');
2466
2467 param.type = fs_value_is_flag;
2468 param.string = NULL;
2469
2470 if (value) {
2471 if (value == key)
2472 continue;
2473
2474 *value++ = 0;
2475 v_len = strlen(value);
2476 param.string = kmemdup_nul(value, v_len,
2477 GFP_KERNEL);
2478 if (!param.string)
2479 return -ENOMEM;
2480 param.type = fs_value_is_string;
2481 }
2482
2483 param.key = key;
2484 param.size = v_len;
2485
2486 ret = ext4_parse_param(fc, ¶m);
2487 if (param.string)
2488 kfree(param.string);
2489 if (ret < 0)
2490 return ret;
2491 }
2492 }
2493
2494 ret = ext4_validate_options(fc);
2495 if (ret < 0)
2496 return ret;
2497
2498 return 0;
2499}
2500
2501static int parse_apply_sb_mount_options(struct super_block *sb,
2502 struct ext4_fs_context *m_ctx)
2503{
2504 struct ext4_sb_info *sbi = EXT4_SB(sb);
2505 char *s_mount_opts = NULL;
2506 struct ext4_fs_context *s_ctx = NULL;
2507 struct fs_context *fc = NULL;
2508 int ret = -ENOMEM;
2509
2510 if (!sbi->s_es->s_mount_opts[0])
2511 return 0;
2512
2513 s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
2514 sizeof(sbi->s_es->s_mount_opts),
2515 GFP_KERNEL);
2516 if (!s_mount_opts)
2517 return ret;
2518
2519 fc = kzalloc(sizeof(struct fs_context), GFP_KERNEL);
2520 if (!fc)
2521 goto out_free;
2522
2523 s_ctx = kzalloc(sizeof(struct ext4_fs_context), GFP_KERNEL);
2524 if (!s_ctx)
2525 goto out_free;
2526
2527 fc->fs_private = s_ctx;
2528 fc->s_fs_info = sbi;
2529
2530 ret = parse_options(fc, s_mount_opts);
2531 if (ret < 0)
2532 goto parse_failed;
2533
2534 ret = ext4_check_opt_consistency(fc, sb);
2535 if (ret < 0) {
2536parse_failed:
2537 ext4_msg(sb, KERN_WARNING,
2538 "failed to parse options in superblock: %s",
2539 s_mount_opts);
2540 ret = 0;
2541 goto out_free;
2542 }
2543
2544 if (s_ctx->spec & EXT4_SPEC_JOURNAL_DEV)
2545 m_ctx->journal_devnum = s_ctx->journal_devnum;
2546 if (s_ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)
2547 m_ctx->journal_ioprio = s_ctx->journal_ioprio;
2548
2549 ext4_apply_options(fc, sb);
2550 ret = 0;
2551
2552out_free:
2553 if (fc) {
2554 ext4_fc_free(fc);
2555 kfree(fc);
2556 }
2557 kfree(s_mount_opts);
2558 return ret;
2559}
2560
2561static void ext4_apply_quota_options(struct fs_context *fc,
2562 struct super_block *sb)
2563{
2564#ifdef CONFIG_QUOTA
2565 bool quota_feature = ext4_has_feature_quota(sb);
2566 struct ext4_fs_context *ctx = fc->fs_private;
2567 struct ext4_sb_info *sbi = EXT4_SB(sb);
2568 char *qname;
2569 int i;
2570
2571 if (quota_feature)
2572 return;
2573
2574 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2575 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2576 if (!(ctx->qname_spec & (1 << i)))
2577 continue;
2578
2579 qname = ctx->s_qf_names[i]; /* May be NULL */
2580 if (qname)
2581 set_opt(sb, QUOTA);
2582 ctx->s_qf_names[i] = NULL;
2583 qname = rcu_replace_pointer(sbi->s_qf_names[i], qname,
2584 lockdep_is_held(&sb->s_umount));
2585 if (qname)
2586 kfree_rcu_mightsleep(qname);
2587 }
2588 }
2589
2590 if (ctx->spec & EXT4_SPEC_JQFMT)
2591 sbi->s_jquota_fmt = ctx->s_jquota_fmt;
2592#endif
2593}
2594
2595/*
2596 * Check quota settings consistency.
2597 */
2598static int ext4_check_quota_consistency(struct fs_context *fc,
2599 struct super_block *sb)
2600{
2601#ifdef CONFIG_QUOTA
2602 struct ext4_fs_context *ctx = fc->fs_private;
2603 struct ext4_sb_info *sbi = EXT4_SB(sb);
2604 bool quota_feature = ext4_has_feature_quota(sb);
2605 bool quota_loaded = sb_any_quota_loaded(sb);
2606 bool usr_qf_name, grp_qf_name, usrquota, grpquota;
2607 int quota_flags, i;
2608
2609 /*
2610 * We do the test below only for project quotas. 'usrquota' and
2611 * 'grpquota' mount options are allowed even without quota feature
2612 * to support legacy quotas in quota files.
2613 */
2614 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_PRJQUOTA) &&
2615 !ext4_has_feature_project(sb)) {
2616 ext4_msg(NULL, KERN_ERR, "Project quota feature not enabled. "
2617 "Cannot enable project quota enforcement.");
2618 return -EINVAL;
2619 }
2620
2621 quota_flags = EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
2622 EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA;
2623 if (quota_loaded &&
2624 ctx->mask_s_mount_opt & quota_flags &&
2625 !ctx_test_mount_opt(ctx, quota_flags))
2626 goto err_quota_change;
2627
2628 if (ctx->spec & EXT4_SPEC_JQUOTA) {
2629
2630 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2631 if (!(ctx->qname_spec & (1 << i)))
2632 continue;
2633
2634 if (quota_loaded &&
2635 !!sbi->s_qf_names[i] != !!ctx->s_qf_names[i])
2636 goto err_jquota_change;
2637
2638 if (sbi->s_qf_names[i] && ctx->s_qf_names[i] &&
2639 strcmp(get_qf_name(sb, sbi, i),
2640 ctx->s_qf_names[i]) != 0)
2641 goto err_jquota_specified;
2642 }
2643
2644 if (quota_feature) {
2645 ext4_msg(NULL, KERN_INFO,
2646 "Journaled quota options ignored when "
2647 "QUOTA feature is enabled");
2648 return 0;
2649 }
2650 }
2651
2652 if (ctx->spec & EXT4_SPEC_JQFMT) {
2653 if (sbi->s_jquota_fmt != ctx->s_jquota_fmt && quota_loaded)
2654 goto err_jquota_change;
2655 if (quota_feature) {
2656 ext4_msg(NULL, KERN_INFO, "Quota format mount options "
2657 "ignored when QUOTA feature is enabled");
2658 return 0;
2659 }
2660 }
2661
2662 /* Make sure we don't mix old and new quota format */
2663 usr_qf_name = (get_qf_name(sb, sbi, USRQUOTA) ||
2664 ctx->s_qf_names[USRQUOTA]);
2665 grp_qf_name = (get_qf_name(sb, sbi, GRPQUOTA) ||
2666 ctx->s_qf_names[GRPQUOTA]);
2667
2668 usrquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2669 test_opt(sb, USRQUOTA));
2670
2671 grpquota = (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) ||
2672 test_opt(sb, GRPQUOTA));
2673
2674 if (usr_qf_name) {
2675 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2676 usrquota = false;
2677 }
2678 if (grp_qf_name) {
2679 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2680 grpquota = false;
2681 }
2682
2683 if (usr_qf_name || grp_qf_name) {
2684 if (usrquota || grpquota) {
2685 ext4_msg(NULL, KERN_ERR, "old and new quota "
2686 "format mixing");
2687 return -EINVAL;
2688 }
2689
2690 if (!(ctx->spec & EXT4_SPEC_JQFMT || sbi->s_jquota_fmt)) {
2691 ext4_msg(NULL, KERN_ERR, "journaled quota format "
2692 "not specified");
2693 return -EINVAL;
2694 }
2695 }
2696
2697 return 0;
2698
2699err_quota_change:
2700 ext4_msg(NULL, KERN_ERR,
2701 "Cannot change quota options when quota turned on");
2702 return -EINVAL;
2703err_jquota_change:
2704 ext4_msg(NULL, KERN_ERR, "Cannot change journaled quota "
2705 "options when quota turned on");
2706 return -EINVAL;
2707err_jquota_specified:
2708 ext4_msg(NULL, KERN_ERR, "%s quota file already specified",
2709 QTYPE2NAME(i));
2710 return -EINVAL;
2711#else
2712 return 0;
2713#endif
2714}
2715
2716static int ext4_check_test_dummy_encryption(const struct fs_context *fc,
2717 struct super_block *sb)
2718{
2719 const struct ext4_fs_context *ctx = fc->fs_private;
2720 const struct ext4_sb_info *sbi = EXT4_SB(sb);
2721
2722 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy))
2723 return 0;
2724
2725 if (!ext4_has_feature_encrypt(sb)) {
2726 ext4_msg(NULL, KERN_WARNING,
2727 "test_dummy_encryption requires encrypt feature");
2728 return -EINVAL;
2729 }
2730 /*
2731 * This mount option is just for testing, and it's not worthwhile to
2732 * implement the extra complexity (e.g. RCU protection) that would be
2733 * needed to allow it to be set or changed during remount. We do allow
2734 * it to be specified during remount, but only if there is no change.
2735 */
2736 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
2737 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2738 &ctx->dummy_enc_policy))
2739 return 0;
2740 ext4_msg(NULL, KERN_WARNING,
2741 "Can't set or change test_dummy_encryption on remount");
2742 return -EINVAL;
2743 }
2744 /* Also make sure s_mount_opts didn't contain a conflicting value. */
2745 if (fscrypt_is_dummy_policy_set(&sbi->s_dummy_enc_policy)) {
2746 if (fscrypt_dummy_policies_equal(&sbi->s_dummy_enc_policy,
2747 &ctx->dummy_enc_policy))
2748 return 0;
2749 ext4_msg(NULL, KERN_WARNING,
2750 "Conflicting test_dummy_encryption options");
2751 return -EINVAL;
2752 }
2753 return 0;
2754}
2755
2756static void ext4_apply_test_dummy_encryption(struct ext4_fs_context *ctx,
2757 struct super_block *sb)
2758{
2759 if (!fscrypt_is_dummy_policy_set(&ctx->dummy_enc_policy) ||
2760 /* if already set, it was already verified to be the same */
2761 fscrypt_is_dummy_policy_set(&EXT4_SB(sb)->s_dummy_enc_policy))
2762 return;
2763 EXT4_SB(sb)->s_dummy_enc_policy = ctx->dummy_enc_policy;
2764 memset(&ctx->dummy_enc_policy, 0, sizeof(ctx->dummy_enc_policy));
2765 ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2766}
2767
2768static int ext4_check_opt_consistency(struct fs_context *fc,
2769 struct super_block *sb)
2770{
2771 struct ext4_fs_context *ctx = fc->fs_private;
2772 struct ext4_sb_info *sbi = fc->s_fs_info;
2773 int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE;
2774 int err;
2775
2776 if ((ctx->opt_flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2777 ext4_msg(NULL, KERN_ERR,
2778 "Mount option(s) incompatible with ext2");
2779 return -EINVAL;
2780 }
2781 if ((ctx->opt_flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2782 ext4_msg(NULL, KERN_ERR,
2783 "Mount option(s) incompatible with ext3");
2784 return -EINVAL;
2785 }
2786
2787 if (ctx->s_want_extra_isize >
2788 (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE)) {
2789 ext4_msg(NULL, KERN_ERR,
2790 "Invalid want_extra_isize %d",
2791 ctx->s_want_extra_isize);
2792 return -EINVAL;
2793 }
2794
2795 err = ext4_check_test_dummy_encryption(fc, sb);
2796 if (err)
2797 return err;
2798
2799 if ((ctx->spec & EXT4_SPEC_DATAJ) && is_remount) {
2800 if (!sbi->s_journal) {
2801 ext4_msg(NULL, KERN_WARNING,
2802 "Remounting file system with no journal "
2803 "so ignoring journalled data option");
2804 ctx_clear_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS);
2805 } else if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DATA_FLAGS) !=
2806 test_opt(sb, DATA_FLAGS)) {
2807 ext4_msg(NULL, KERN_ERR, "Cannot change data mode "
2808 "on remount");
2809 return -EINVAL;
2810 }
2811 }
2812
2813 if (is_remount) {
2814 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2815 (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2816 ext4_msg(NULL, KERN_ERR, "can't mount with "
2817 "both data=journal and dax");
2818 return -EINVAL;
2819 }
2820
2821 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DAX_ALWAYS) &&
2822 (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2823 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2824fail_dax_change_remount:
2825 ext4_msg(NULL, KERN_ERR, "can't change "
2826 "dax mount option while remounting");
2827 return -EINVAL;
2828 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_NEVER) &&
2829 (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2830 (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS))) {
2831 goto fail_dax_change_remount;
2832 } else if (ctx_test_mount_opt2(ctx, EXT4_MOUNT2_DAX_INODE) &&
2833 ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2834 (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2835 !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE))) {
2836 goto fail_dax_change_remount;
2837 }
2838 }
2839
2840 return ext4_check_quota_consistency(fc, sb);
2841}
2842
2843static void ext4_apply_options(struct fs_context *fc, struct super_block *sb)
2844{
2845 struct ext4_fs_context *ctx = fc->fs_private;
2846 struct ext4_sb_info *sbi = fc->s_fs_info;
2847
2848 sbi->s_mount_opt &= ~ctx->mask_s_mount_opt;
2849 sbi->s_mount_opt |= ctx->vals_s_mount_opt;
2850 sbi->s_mount_opt2 &= ~ctx->mask_s_mount_opt2;
2851 sbi->s_mount_opt2 |= ctx->vals_s_mount_opt2;
2852 sb->s_flags &= ~ctx->mask_s_flags;
2853 sb->s_flags |= ctx->vals_s_flags;
2854
2855#define APPLY(X) ({ if (ctx->spec & EXT4_SPEC_##X) sbi->X = ctx->X; })
2856 APPLY(s_commit_interval);
2857 APPLY(s_stripe);
2858 APPLY(s_max_batch_time);
2859 APPLY(s_min_batch_time);
2860 APPLY(s_want_extra_isize);
2861 APPLY(s_inode_readahead_blks);
2862 APPLY(s_max_dir_size_kb);
2863 APPLY(s_li_wait_mult);
2864 APPLY(s_resgid);
2865 APPLY(s_resuid);
2866
2867#ifdef CONFIG_EXT4_DEBUG
2868 APPLY(s_fc_debug_max_replay);
2869#endif
2870
2871 ext4_apply_quota_options(fc, sb);
2872 ext4_apply_test_dummy_encryption(ctx, sb);
2873}
2874
2875
2876static int ext4_validate_options(struct fs_context *fc)
2877{
2878#ifdef CONFIG_QUOTA
2879 struct ext4_fs_context *ctx = fc->fs_private;
2880 char *usr_qf_name, *grp_qf_name;
2881
2882 usr_qf_name = ctx->s_qf_names[USRQUOTA];
2883 grp_qf_name = ctx->s_qf_names[GRPQUOTA];
2884
2885 if (usr_qf_name || grp_qf_name) {
2886 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) && usr_qf_name)
2887 ctx_clear_mount_opt(ctx, EXT4_MOUNT_USRQUOTA);
2888
2889 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA) && grp_qf_name)
2890 ctx_clear_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA);
2891
2892 if (ctx_test_mount_opt(ctx, EXT4_MOUNT_USRQUOTA) ||
2893 ctx_test_mount_opt(ctx, EXT4_MOUNT_GRPQUOTA)) {
2894 ext4_msg(NULL, KERN_ERR, "old and new quota "
2895 "format mixing");
2896 return -EINVAL;
2897 }
2898 }
2899#endif
2900 return 1;
2901}
2902
2903static inline void ext4_show_quota_options(struct seq_file *seq,
2904 struct super_block *sb)
2905{
2906#if defined(CONFIG_QUOTA)
2907 struct ext4_sb_info *sbi = EXT4_SB(sb);
2908 char *usr_qf_name, *grp_qf_name;
2909
2910 if (sbi->s_jquota_fmt) {
2911 char *fmtname = "";
2912
2913 switch (sbi->s_jquota_fmt) {
2914 case QFMT_VFS_OLD:
2915 fmtname = "vfsold";
2916 break;
2917 case QFMT_VFS_V0:
2918 fmtname = "vfsv0";
2919 break;
2920 case QFMT_VFS_V1:
2921 fmtname = "vfsv1";
2922 break;
2923 }
2924 seq_printf(seq, ",jqfmt=%s", fmtname);
2925 }
2926
2927 rcu_read_lock();
2928 usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2929 grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2930 if (usr_qf_name)
2931 seq_show_option(seq, "usrjquota", usr_qf_name);
2932 if (grp_qf_name)
2933 seq_show_option(seq, "grpjquota", grp_qf_name);
2934 rcu_read_unlock();
2935#endif
2936}
2937
2938static const char *token2str(int token)
2939{
2940 const struct fs_parameter_spec *spec;
2941
2942 for (spec = ext4_param_specs; spec->name != NULL; spec++)
2943 if (spec->opt == token && !spec->type)
2944 break;
2945 return spec->name;
2946}
2947
2948/*
2949 * Show an option if
2950 * - it's set to a non-default value OR
2951 * - if the per-sb default is different from the global default
2952 */
2953static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2954 int nodefs)
2955{
2956 struct ext4_sb_info *sbi = EXT4_SB(sb);
2957 struct ext4_super_block *es = sbi->s_es;
2958 int def_errors;
2959 const struct mount_opts *m;
2960 char sep = nodefs ? '\n' : ',';
2961
2962#define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2963#define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2964
2965 if (sbi->s_sb_block != 1)
2966 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2967
2968 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2969 int want_set = m->flags & MOPT_SET;
2970 int opt_2 = m->flags & MOPT_2;
2971 unsigned int mount_opt, def_mount_opt;
2972
2973 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2974 m->flags & MOPT_SKIP)
2975 continue;
2976
2977 if (opt_2) {
2978 mount_opt = sbi->s_mount_opt2;
2979 def_mount_opt = sbi->s_def_mount_opt2;
2980 } else {
2981 mount_opt = sbi->s_mount_opt;
2982 def_mount_opt = sbi->s_def_mount_opt;
2983 }
2984 /* skip if same as the default */
2985 if (!nodefs && !(m->mount_opt & (mount_opt ^ def_mount_opt)))
2986 continue;
2987 /* select Opt_noFoo vs Opt_Foo */
2988 if ((want_set &&
2989 (mount_opt & m->mount_opt) != m->mount_opt) ||
2990 (!want_set && (mount_opt & m->mount_opt)))
2991 continue;
2992 SEQ_OPTS_PRINT("%s", token2str(m->token));
2993 }
2994
2995 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2996 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2997 SEQ_OPTS_PRINT("resuid=%u",
2998 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2999 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
3000 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
3001 SEQ_OPTS_PRINT("resgid=%u",
3002 from_kgid_munged(&init_user_ns, sbi->s_resgid));
3003 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
3004 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
3005 SEQ_OPTS_PUTS("errors=remount-ro");
3006 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
3007 SEQ_OPTS_PUTS("errors=continue");
3008 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
3009 SEQ_OPTS_PUTS("errors=panic");
3010 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
3011 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
3012 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
3013 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
3014 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
3015 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
3016 if (nodefs || sbi->s_stripe)
3017 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
3018 if (nodefs || EXT4_MOUNT_DATA_FLAGS &
3019 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3020 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3021 SEQ_OPTS_PUTS("data=journal");
3022 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3023 SEQ_OPTS_PUTS("data=ordered");
3024 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
3025 SEQ_OPTS_PUTS("data=writeback");
3026 }
3027 if (nodefs ||
3028 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
3029 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
3030 sbi->s_inode_readahead_blks);
3031
3032 if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
3033 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
3034 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
3035 if (nodefs || sbi->s_max_dir_size_kb)
3036 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
3037 if (test_opt(sb, DATA_ERR_ABORT))
3038 SEQ_OPTS_PUTS("data_err=abort");
3039
3040 fscrypt_show_test_dummy_encryption(seq, sep, sb);
3041
3042 if (sb->s_flags & SB_INLINECRYPT)
3043 SEQ_OPTS_PUTS("inlinecrypt");
3044
3045 if (test_opt(sb, DAX_ALWAYS)) {
3046 if (IS_EXT2_SB(sb))
3047 SEQ_OPTS_PUTS("dax");
3048 else
3049 SEQ_OPTS_PUTS("dax=always");
3050 } else if (test_opt2(sb, DAX_NEVER)) {
3051 SEQ_OPTS_PUTS("dax=never");
3052 } else if (test_opt2(sb, DAX_INODE)) {
3053 SEQ_OPTS_PUTS("dax=inode");
3054 }
3055
3056 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3057 !test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3058 SEQ_OPTS_PUTS("mb_optimize_scan=0");
3059 } else if (sbi->s_groups_count < MB_DEFAULT_LINEAR_SCAN_THRESHOLD &&
3060 test_opt2(sb, MB_OPTIMIZE_SCAN)) {
3061 SEQ_OPTS_PUTS("mb_optimize_scan=1");
3062 }
3063
3064 ext4_show_quota_options(seq, sb);
3065 return 0;
3066}
3067
3068static int ext4_show_options(struct seq_file *seq, struct dentry *root)
3069{
3070 return _ext4_show_options(seq, root->d_sb, 0);
3071}
3072
3073int ext4_seq_options_show(struct seq_file *seq, void *offset)
3074{
3075 struct super_block *sb = seq->private;
3076 int rc;
3077
3078 seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
3079 rc = _ext4_show_options(seq, sb, 1);
3080 seq_puts(seq, "\n");
3081 return rc;
3082}
3083
3084static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
3085 int read_only)
3086{
3087 struct ext4_sb_info *sbi = EXT4_SB(sb);
3088 int err = 0;
3089
3090 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
3091 ext4_msg(sb, KERN_ERR, "revision level too high, "
3092 "forcing read-only mode");
3093 err = -EROFS;
3094 goto done;
3095 }
3096 if (read_only)
3097 goto done;
3098 if (!(sbi->s_mount_state & EXT4_VALID_FS))
3099 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
3100 "running e2fsck is recommended");
3101 else if (sbi->s_mount_state & EXT4_ERROR_FS)
3102 ext4_msg(sb, KERN_WARNING,
3103 "warning: mounting fs with errors, "
3104 "running e2fsck is recommended");
3105 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
3106 le16_to_cpu(es->s_mnt_count) >=
3107 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
3108 ext4_msg(sb, KERN_WARNING,
3109 "warning: maximal mount count reached, "
3110 "running e2fsck is recommended");
3111 else if (le32_to_cpu(es->s_checkinterval) &&
3112 (ext4_get_tstamp(es, s_lastcheck) +
3113 le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
3114 ext4_msg(sb, KERN_WARNING,
3115 "warning: checktime reached, "
3116 "running e2fsck is recommended");
3117 if (!sbi->s_journal)
3118 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
3119 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
3120 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
3121 le16_add_cpu(&es->s_mnt_count, 1);
3122 ext4_update_tstamp(es, s_mtime);
3123 if (sbi->s_journal) {
3124 ext4_set_feature_journal_needs_recovery(sb);
3125 if (ext4_has_feature_orphan_file(sb))
3126 ext4_set_feature_orphan_present(sb);
3127 }
3128
3129 err = ext4_commit_super(sb);
3130done:
3131 if (test_opt(sb, DEBUG))
3132 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
3133 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
3134 sb->s_blocksize,
3135 sbi->s_groups_count,
3136 EXT4_BLOCKS_PER_GROUP(sb),
3137 EXT4_INODES_PER_GROUP(sb),
3138 sbi->s_mount_opt, sbi->s_mount_opt2);
3139 return err;
3140}
3141
3142int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
3143{
3144 struct ext4_sb_info *sbi = EXT4_SB(sb);
3145 struct flex_groups **old_groups, **new_groups;
3146 int size, i, j;
3147
3148 if (!sbi->s_log_groups_per_flex)
3149 return 0;
3150
3151 size = ext4_flex_group(sbi, ngroup - 1) + 1;
3152 if (size <= sbi->s_flex_groups_allocated)
3153 return 0;
3154
3155 new_groups = kvzalloc(roundup_pow_of_two(size *
3156 sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
3157 if (!new_groups) {
3158 ext4_msg(sb, KERN_ERR,
3159 "not enough memory for %d flex group pointers", size);
3160 return -ENOMEM;
3161 }
3162 for (i = sbi->s_flex_groups_allocated; i < size; i++) {
3163 new_groups[i] = kvzalloc(roundup_pow_of_two(
3164 sizeof(struct flex_groups)),
3165 GFP_KERNEL);
3166 if (!new_groups[i]) {
3167 for (j = sbi->s_flex_groups_allocated; j < i; j++)
3168 kvfree(new_groups[j]);
3169 kvfree(new_groups);
3170 ext4_msg(sb, KERN_ERR,
3171 "not enough memory for %d flex groups", size);
3172 return -ENOMEM;
3173 }
3174 }
3175 rcu_read_lock();
3176 old_groups = rcu_dereference(sbi->s_flex_groups);
3177 if (old_groups)
3178 memcpy(new_groups, old_groups,
3179 (sbi->s_flex_groups_allocated *
3180 sizeof(struct flex_groups *)));
3181 rcu_read_unlock();
3182 rcu_assign_pointer(sbi->s_flex_groups, new_groups);
3183 sbi->s_flex_groups_allocated = size;
3184 if (old_groups)
3185 ext4_kvfree_array_rcu(old_groups);
3186 return 0;
3187}
3188
3189static int ext4_fill_flex_info(struct super_block *sb)
3190{
3191 struct ext4_sb_info *sbi = EXT4_SB(sb);
3192 struct ext4_group_desc *gdp = NULL;
3193 struct flex_groups *fg;
3194 ext4_group_t flex_group;
3195 int i, err;
3196
3197 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
3198 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
3199 sbi->s_log_groups_per_flex = 0;
3200 return 1;
3201 }
3202
3203 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
3204 if (err)
3205 goto failed;
3206
3207 for (i = 0; i < sbi->s_groups_count; i++) {
3208 gdp = ext4_get_group_desc(sb, i, NULL);
3209
3210 flex_group = ext4_flex_group(sbi, i);
3211 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
3212 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
3213 atomic64_add(ext4_free_group_clusters(sb, gdp),
3214 &fg->free_clusters);
3215 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
3216 }
3217
3218 return 1;
3219failed:
3220 return 0;
3221}
3222
3223static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
3224 struct ext4_group_desc *gdp)
3225{
3226 int offset = offsetof(struct ext4_group_desc, bg_checksum);
3227 __u16 crc = 0;
3228 __le32 le_group = cpu_to_le32(block_group);
3229 struct ext4_sb_info *sbi = EXT4_SB(sb);
3230
3231 if (ext4_has_metadata_csum(sbi->s_sb)) {
3232 /* Use new metadata_csum algorithm */
3233 __u32 csum32;
3234 __u16 dummy_csum = 0;
3235
3236 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
3237 sizeof(le_group));
3238 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
3239 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
3240 sizeof(dummy_csum));
3241 offset += sizeof(dummy_csum);
3242 if (offset < sbi->s_desc_size)
3243 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
3244 sbi->s_desc_size - offset);
3245
3246 crc = csum32 & 0xFFFF;
3247 goto out;
3248 }
3249
3250 /* old crc16 code */
3251 if (!ext4_has_feature_gdt_csum(sb))
3252 return 0;
3253
3254 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
3255 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
3256 crc = crc16(crc, (__u8 *)gdp, offset);
3257 offset += sizeof(gdp->bg_checksum); /* skip checksum */
3258 /* for checksum of struct ext4_group_desc do the rest...*/
3259 if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size)
3260 crc = crc16(crc, (__u8 *)gdp + offset,
3261 sbi->s_desc_size - offset);
3262
3263out:
3264 return cpu_to_le16(crc);
3265}
3266
3267int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
3268 struct ext4_group_desc *gdp)
3269{
3270 if (ext4_has_group_desc_csum(sb) &&
3271 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
3272 return 0;
3273
3274 return 1;
3275}
3276
3277void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
3278 struct ext4_group_desc *gdp)
3279{
3280 if (!ext4_has_group_desc_csum(sb))
3281 return;
3282 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
3283}
3284
3285/* Called at mount-time, super-block is locked */
3286static int ext4_check_descriptors(struct super_block *sb,
3287 ext4_fsblk_t sb_block,
3288 ext4_group_t *first_not_zeroed)
3289{
3290 struct ext4_sb_info *sbi = EXT4_SB(sb);
3291 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
3292 ext4_fsblk_t last_block;
3293 ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
3294 ext4_fsblk_t block_bitmap;
3295 ext4_fsblk_t inode_bitmap;
3296 ext4_fsblk_t inode_table;
3297 int flexbg_flag = 0;
3298 ext4_group_t i, grp = sbi->s_groups_count;
3299
3300 if (ext4_has_feature_flex_bg(sb))
3301 flexbg_flag = 1;
3302
3303 ext4_debug("Checking group descriptors");
3304
3305 for (i = 0; i < sbi->s_groups_count; i++) {
3306 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
3307
3308 if (i == sbi->s_groups_count - 1 || flexbg_flag)
3309 last_block = ext4_blocks_count(sbi->s_es) - 1;
3310 else
3311 last_block = first_block +
3312 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
3313
3314 if ((grp == sbi->s_groups_count) &&
3315 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3316 grp = i;
3317
3318 block_bitmap = ext4_block_bitmap(sb, gdp);
3319 if (block_bitmap == sb_block) {
3320 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3321 "Block bitmap for group %u overlaps "
3322 "superblock", i);
3323 if (!sb_rdonly(sb))
3324 return 0;
3325 }
3326 if (block_bitmap >= sb_block + 1 &&
3327 block_bitmap <= last_bg_block) {
3328 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3329 "Block bitmap for group %u overlaps "
3330 "block group descriptors", i);
3331 if (!sb_rdonly(sb))
3332 return 0;
3333 }
3334 if (block_bitmap < first_block || block_bitmap > last_block) {
3335 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3336 "Block bitmap for group %u not in group "
3337 "(block %llu)!", i, block_bitmap);
3338 return 0;
3339 }
3340 inode_bitmap = ext4_inode_bitmap(sb, gdp);
3341 if (inode_bitmap == sb_block) {
3342 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3343 "Inode bitmap for group %u overlaps "
3344 "superblock", i);
3345 if (!sb_rdonly(sb))
3346 return 0;
3347 }
3348 if (inode_bitmap >= sb_block + 1 &&
3349 inode_bitmap <= last_bg_block) {
3350 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3351 "Inode bitmap for group %u overlaps "
3352 "block group descriptors", i);
3353 if (!sb_rdonly(sb))
3354 return 0;
3355 }
3356 if (inode_bitmap < first_block || inode_bitmap > last_block) {
3357 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3358 "Inode bitmap for group %u not in group "
3359 "(block %llu)!", i, inode_bitmap);
3360 return 0;
3361 }
3362 inode_table = ext4_inode_table(sb, gdp);
3363 if (inode_table == sb_block) {
3364 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3365 "Inode table for group %u overlaps "
3366 "superblock", i);
3367 if (!sb_rdonly(sb))
3368 return 0;
3369 }
3370 if (inode_table >= sb_block + 1 &&
3371 inode_table <= last_bg_block) {
3372 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3373 "Inode table for group %u overlaps "
3374 "block group descriptors", i);
3375 if (!sb_rdonly(sb))
3376 return 0;
3377 }
3378 if (inode_table < first_block ||
3379 inode_table + sbi->s_itb_per_group - 1 > last_block) {
3380 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3381 "Inode table for group %u not in group "
3382 "(block %llu)!", i, inode_table);
3383 return 0;
3384 }
3385 ext4_lock_group(sb, i);
3386 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
3387 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
3388 "Checksum for group %u failed (%u!=%u)",
3389 i, le16_to_cpu(ext4_group_desc_csum(sb, i,
3390 gdp)), le16_to_cpu(gdp->bg_checksum));
3391 if (!sb_rdonly(sb)) {
3392 ext4_unlock_group(sb, i);
3393 return 0;
3394 }
3395 }
3396 ext4_unlock_group(sb, i);
3397 if (!flexbg_flag)
3398 first_block += EXT4_BLOCKS_PER_GROUP(sb);
3399 }
3400 if (NULL != first_not_zeroed)
3401 *first_not_zeroed = grp;
3402 return 1;
3403}
3404
3405/*
3406 * Maximal extent format file size.
3407 * Resulting logical blkno at s_maxbytes must fit in our on-disk
3408 * extent format containers, within a sector_t, and within i_blocks
3409 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
3410 * so that won't be a limiting factor.
3411 *
3412 * However there is other limiting factor. We do store extents in the form
3413 * of starting block and length, hence the resulting length of the extent
3414 * covering maximum file size must fit into on-disk format containers as
3415 * well. Given that length is always by 1 unit bigger than max unit (because
3416 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
3417 *
3418 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
3419 */
3420static loff_t ext4_max_size(int blkbits, int has_huge_files)
3421{
3422 loff_t res;
3423 loff_t upper_limit = MAX_LFS_FILESIZE;
3424
3425 BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
3426
3427 if (!has_huge_files) {
3428 upper_limit = (1LL << 32) - 1;
3429
3430 /* total blocks in file system block size */
3431 upper_limit >>= (blkbits - 9);
3432 upper_limit <<= blkbits;
3433 }
3434
3435 /*
3436 * 32-bit extent-start container, ee_block. We lower the maxbytes
3437 * by one fs block, so ee_len can cover the extent of maximum file
3438 * size
3439 */
3440 res = (1LL << 32) - 1;
3441 res <<= blkbits;
3442
3443 /* Sanity check against vm- & vfs- imposed limits */
3444 if (res > upper_limit)
3445 res = upper_limit;
3446
3447 return res;
3448}
3449
3450/*
3451 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
3452 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3453 * We need to be 1 filesystem block less than the 2^48 sector limit.
3454 */
3455static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3456{
3457 loff_t upper_limit, res = EXT4_NDIR_BLOCKS;
3458 int meta_blocks;
3459 unsigned int ppb = 1 << (bits - 2);
3460
3461 /*
3462 * This is calculated to be the largest file size for a dense, block
3463 * mapped file such that the file's total number of 512-byte sectors,
3464 * including data and all indirect blocks, does not exceed (2^48 - 1).
3465 *
3466 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3467 * number of 512-byte sectors of the file.
3468 */
3469 if (!has_huge_files) {
3470 /*
3471 * !has_huge_files or implies that the inode i_block field
3472 * represents total file blocks in 2^32 512-byte sectors ==
3473 * size of vfs inode i_blocks * 8
3474 */
3475 upper_limit = (1LL << 32) - 1;
3476
3477 /* total blocks in file system block size */
3478 upper_limit >>= (bits - 9);
3479
3480 } else {
3481 /*
3482 * We use 48 bit ext4_inode i_blocks
3483 * With EXT4_HUGE_FILE_FL set the i_blocks
3484 * represent total number of blocks in
3485 * file system block size
3486 */
3487 upper_limit = (1LL << 48) - 1;
3488
3489 }
3490
3491 /* Compute how many blocks we can address by block tree */
3492 res += ppb;
3493 res += ppb * ppb;
3494 res += ((loff_t)ppb) * ppb * ppb;
3495 /* Compute how many metadata blocks are needed */
3496 meta_blocks = 1;
3497 meta_blocks += 1 + ppb;
3498 meta_blocks += 1 + ppb + ppb * ppb;
3499 /* Does block tree limit file size? */
3500 if (res + meta_blocks <= upper_limit)
3501 goto check_lfs;
3502
3503 res = upper_limit;
3504 /* How many metadata blocks are needed for addressing upper_limit? */
3505 upper_limit -= EXT4_NDIR_BLOCKS;
3506 /* indirect blocks */
3507 meta_blocks = 1;
3508 upper_limit -= ppb;
3509 /* double indirect blocks */
3510 if (upper_limit < ppb * ppb) {
3511 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb);
3512 res -= meta_blocks;
3513 goto check_lfs;
3514 }
3515 meta_blocks += 1 + ppb;
3516 upper_limit -= ppb * ppb;
3517 /* tripple indirect blocks for the rest */
3518 meta_blocks += 1 + DIV_ROUND_UP_ULL(upper_limit, ppb) +
3519 DIV_ROUND_UP_ULL(upper_limit, ppb*ppb);
3520 res -= meta_blocks;
3521check_lfs:
3522 res <<= bits;
3523 if (res > MAX_LFS_FILESIZE)
3524 res = MAX_LFS_FILESIZE;
3525
3526 return res;
3527}
3528
3529static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3530 ext4_fsblk_t logical_sb_block, int nr)
3531{
3532 struct ext4_sb_info *sbi = EXT4_SB(sb);
3533 ext4_group_t bg, first_meta_bg;
3534 int has_super = 0;
3535
3536 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3537
3538 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3539 return logical_sb_block + nr + 1;
3540 bg = sbi->s_desc_per_block * nr;
3541 if (ext4_bg_has_super(sb, bg))
3542 has_super = 1;
3543
3544 /*
3545 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3546 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled
3547 * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3548 * compensate.
3549 */
3550 if (sb->s_blocksize == 1024 && nr == 0 &&
3551 le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3552 has_super++;
3553
3554 return (has_super + ext4_group_first_block_no(sb, bg));
3555}
3556
3557/**
3558 * ext4_get_stripe_size: Get the stripe size.
3559 * @sbi: In memory super block info
3560 *
3561 * If we have specified it via mount option, then
3562 * use the mount option value. If the value specified at mount time is
3563 * greater than the blocks per group use the super block value.
3564 * If the super block value is greater than blocks per group return 0.
3565 * Allocator needs it be less than blocks per group.
3566 *
3567 */
3568static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3569{
3570 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3571 unsigned long stripe_width =
3572 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3573 int ret;
3574
3575 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3576 ret = sbi->s_stripe;
3577 else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3578 ret = stripe_width;
3579 else if (stride && stride <= sbi->s_blocks_per_group)
3580 ret = stride;
3581 else
3582 ret = 0;
3583
3584 /*
3585 * If the stripe width is 1, this makes no sense and
3586 * we set it to 0 to turn off stripe handling code.
3587 */
3588 if (ret <= 1)
3589 ret = 0;
3590
3591 return ret;
3592}
3593
3594/*
3595 * Check whether this filesystem can be mounted based on
3596 * the features present and the RDONLY/RDWR mount requested.
3597 * Returns 1 if this filesystem can be mounted as requested,
3598 * 0 if it cannot be.
3599 */
3600int ext4_feature_set_ok(struct super_block *sb, int readonly)
3601{
3602 if (ext4_has_unknown_ext4_incompat_features(sb)) {
3603 ext4_msg(sb, KERN_ERR,
3604 "Couldn't mount because of "
3605 "unsupported optional features (%x)",
3606 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3607 ~EXT4_FEATURE_INCOMPAT_SUPP));
3608 return 0;
3609 }
3610
3611#if !IS_ENABLED(CONFIG_UNICODE)
3612 if (ext4_has_feature_casefold(sb)) {
3613 ext4_msg(sb, KERN_ERR,
3614 "Filesystem with casefold feature cannot be "
3615 "mounted without CONFIG_UNICODE");
3616 return 0;
3617 }
3618#endif
3619
3620 if (readonly)
3621 return 1;
3622
3623 if (ext4_has_feature_readonly(sb)) {
3624 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3625 sb->s_flags |= SB_RDONLY;
3626 return 1;
3627 }
3628
3629 /* Check that feature set is OK for a read-write mount */
3630 if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3631 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3632 "unsupported optional features (%x)",
3633 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3634 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3635 return 0;
3636 }
3637 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3638 ext4_msg(sb, KERN_ERR,
3639 "Can't support bigalloc feature without "
3640 "extents feature\n");
3641 return 0;
3642 }
3643
3644#if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3645 if (!readonly && (ext4_has_feature_quota(sb) ||
3646 ext4_has_feature_project(sb))) {
3647 ext4_msg(sb, KERN_ERR,
3648 "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3649 return 0;
3650 }
3651#endif /* CONFIG_QUOTA */
3652 return 1;
3653}
3654
3655/*
3656 * This function is called once a day if we have errors logged
3657 * on the file system
3658 */
3659static void print_daily_error_info(struct timer_list *t)
3660{
3661 struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3662 struct super_block *sb = sbi->s_sb;
3663 struct ext4_super_block *es = sbi->s_es;
3664
3665 if (es->s_error_count)
3666 /* fsck newer than v1.41.13 is needed to clean this condition. */
3667 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3668 le32_to_cpu(es->s_error_count));
3669 if (es->s_first_error_time) {
3670 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3671 sb->s_id,
3672 ext4_get_tstamp(es, s_first_error_time),
3673 (int) sizeof(es->s_first_error_func),
3674 es->s_first_error_func,
3675 le32_to_cpu(es->s_first_error_line));
3676 if (es->s_first_error_ino)
3677 printk(KERN_CONT ": inode %u",
3678 le32_to_cpu(es->s_first_error_ino));
3679 if (es->s_first_error_block)
3680 printk(KERN_CONT ": block %llu", (unsigned long long)
3681 le64_to_cpu(es->s_first_error_block));
3682 printk(KERN_CONT "\n");
3683 }
3684 if (es->s_last_error_time) {
3685 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3686 sb->s_id,
3687 ext4_get_tstamp(es, s_last_error_time),
3688 (int) sizeof(es->s_last_error_func),
3689 es->s_last_error_func,
3690 le32_to_cpu(es->s_last_error_line));
3691 if (es->s_last_error_ino)
3692 printk(KERN_CONT ": inode %u",
3693 le32_to_cpu(es->s_last_error_ino));
3694 if (es->s_last_error_block)
3695 printk(KERN_CONT ": block %llu", (unsigned long long)
3696 le64_to_cpu(es->s_last_error_block));
3697 printk(KERN_CONT "\n");
3698 }
3699 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
3700}
3701
3702/* Find next suitable group and run ext4_init_inode_table */
3703static int ext4_run_li_request(struct ext4_li_request *elr)
3704{
3705 struct ext4_group_desc *gdp = NULL;
3706 struct super_block *sb = elr->lr_super;
3707 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3708 ext4_group_t group = elr->lr_next_group;
3709 unsigned int prefetch_ios = 0;
3710 int ret = 0;
3711 int nr = EXT4_SB(sb)->s_mb_prefetch;
3712 u64 start_time;
3713
3714 if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3715 elr->lr_next_group = ext4_mb_prefetch(sb, group, nr, &prefetch_ios);
3716 ext4_mb_prefetch_fini(sb, elr->lr_next_group, nr);
3717 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group, nr);
3718 if (group >= elr->lr_next_group) {
3719 ret = 1;
3720 if (elr->lr_first_not_zeroed != ngroups &&
3721 !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3722 elr->lr_next_group = elr->lr_first_not_zeroed;
3723 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3724 ret = 0;
3725 }
3726 }
3727 return ret;
3728 }
3729
3730 for (; group < ngroups; group++) {
3731 gdp = ext4_get_group_desc(sb, group, NULL);
3732 if (!gdp) {
3733 ret = 1;
3734 break;
3735 }
3736
3737 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3738 break;
3739 }
3740
3741 if (group >= ngroups)
3742 ret = 1;
3743
3744 if (!ret) {
3745 start_time = ktime_get_real_ns();
3746 ret = ext4_init_inode_table(sb, group,
3747 elr->lr_timeout ? 0 : 1);
3748 trace_ext4_lazy_itable_init(sb, group);
3749 if (elr->lr_timeout == 0) {
3750 elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3751 EXT4_SB(elr->lr_super)->s_li_wait_mult);
3752 }
3753 elr->lr_next_sched = jiffies + elr->lr_timeout;
3754 elr->lr_next_group = group + 1;
3755 }
3756 return ret;
3757}
3758
3759/*
3760 * Remove lr_request from the list_request and free the
3761 * request structure. Should be called with li_list_mtx held
3762 */
3763static void ext4_remove_li_request(struct ext4_li_request *elr)
3764{
3765 if (!elr)
3766 return;
3767
3768 list_del(&elr->lr_request);
3769 EXT4_SB(elr->lr_super)->s_li_request = NULL;
3770 kfree(elr);
3771}
3772
3773static void ext4_unregister_li_request(struct super_block *sb)
3774{
3775 mutex_lock(&ext4_li_mtx);
3776 if (!ext4_li_info) {
3777 mutex_unlock(&ext4_li_mtx);
3778 return;
3779 }
3780
3781 mutex_lock(&ext4_li_info->li_list_mtx);
3782 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3783 mutex_unlock(&ext4_li_info->li_list_mtx);
3784 mutex_unlock(&ext4_li_mtx);
3785}
3786
3787static struct task_struct *ext4_lazyinit_task;
3788
3789/*
3790 * This is the function where ext4lazyinit thread lives. It walks
3791 * through the request list searching for next scheduled filesystem.
3792 * When such a fs is found, run the lazy initialization request
3793 * (ext4_rn_li_request) and keep track of the time spend in this
3794 * function. Based on that time we compute next schedule time of
3795 * the request. When walking through the list is complete, compute
3796 * next waking time and put itself into sleep.
3797 */
3798static int ext4_lazyinit_thread(void *arg)
3799{
3800 struct ext4_lazy_init *eli = arg;
3801 struct list_head *pos, *n;
3802 struct ext4_li_request *elr;
3803 unsigned long next_wakeup, cur;
3804
3805 BUG_ON(NULL == eli);
3806 set_freezable();
3807
3808cont_thread:
3809 while (true) {
3810 next_wakeup = MAX_JIFFY_OFFSET;
3811
3812 mutex_lock(&eli->li_list_mtx);
3813 if (list_empty(&eli->li_request_list)) {
3814 mutex_unlock(&eli->li_list_mtx);
3815 goto exit_thread;
3816 }
3817 list_for_each_safe(pos, n, &eli->li_request_list) {
3818 int err = 0;
3819 int progress = 0;
3820 elr = list_entry(pos, struct ext4_li_request,
3821 lr_request);
3822
3823 if (time_before(jiffies, elr->lr_next_sched)) {
3824 if (time_before(elr->lr_next_sched, next_wakeup))
3825 next_wakeup = elr->lr_next_sched;
3826 continue;
3827 }
3828 if (down_read_trylock(&elr->lr_super->s_umount)) {
3829 if (sb_start_write_trylock(elr->lr_super)) {
3830 progress = 1;
3831 /*
3832 * We hold sb->s_umount, sb can not
3833 * be removed from the list, it is
3834 * now safe to drop li_list_mtx
3835 */
3836 mutex_unlock(&eli->li_list_mtx);
3837 err = ext4_run_li_request(elr);
3838 sb_end_write(elr->lr_super);
3839 mutex_lock(&eli->li_list_mtx);
3840 n = pos->next;
3841 }
3842 up_read((&elr->lr_super->s_umount));
3843 }
3844 /* error, remove the lazy_init job */
3845 if (err) {
3846 ext4_remove_li_request(elr);
3847 continue;
3848 }
3849 if (!progress) {
3850 elr->lr_next_sched = jiffies +
3851 get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3852 }
3853 if (time_before(elr->lr_next_sched, next_wakeup))
3854 next_wakeup = elr->lr_next_sched;
3855 }
3856 mutex_unlock(&eli->li_list_mtx);
3857
3858 try_to_freeze();
3859
3860 cur = jiffies;
3861 if ((time_after_eq(cur, next_wakeup)) ||
3862 (MAX_JIFFY_OFFSET == next_wakeup)) {
3863 cond_resched();
3864 continue;
3865 }
3866
3867 schedule_timeout_interruptible(next_wakeup - cur);
3868
3869 if (kthread_should_stop()) {
3870 ext4_clear_request_list();
3871 goto exit_thread;
3872 }
3873 }
3874
3875exit_thread:
3876 /*
3877 * It looks like the request list is empty, but we need
3878 * to check it under the li_list_mtx lock, to prevent any
3879 * additions into it, and of course we should lock ext4_li_mtx
3880 * to atomically free the list and ext4_li_info, because at
3881 * this point another ext4 filesystem could be registering
3882 * new one.
3883 */
3884 mutex_lock(&ext4_li_mtx);
3885 mutex_lock(&eli->li_list_mtx);
3886 if (!list_empty(&eli->li_request_list)) {
3887 mutex_unlock(&eli->li_list_mtx);
3888 mutex_unlock(&ext4_li_mtx);
3889 goto cont_thread;
3890 }
3891 mutex_unlock(&eli->li_list_mtx);
3892 kfree(ext4_li_info);
3893 ext4_li_info = NULL;
3894 mutex_unlock(&ext4_li_mtx);
3895
3896 return 0;
3897}
3898
3899static void ext4_clear_request_list(void)
3900{
3901 struct list_head *pos, *n;
3902 struct ext4_li_request *elr;
3903
3904 mutex_lock(&ext4_li_info->li_list_mtx);
3905 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3906 elr = list_entry(pos, struct ext4_li_request,
3907 lr_request);
3908 ext4_remove_li_request(elr);
3909 }
3910 mutex_unlock(&ext4_li_info->li_list_mtx);
3911}
3912
3913static int ext4_run_lazyinit_thread(void)
3914{
3915 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3916 ext4_li_info, "ext4lazyinit");
3917 if (IS_ERR(ext4_lazyinit_task)) {
3918 int err = PTR_ERR(ext4_lazyinit_task);
3919 ext4_clear_request_list();
3920 kfree(ext4_li_info);
3921 ext4_li_info = NULL;
3922 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3923 "initialization thread\n",
3924 err);
3925 return err;
3926 }
3927 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3928 return 0;
3929}
3930
3931/*
3932 * Check whether it make sense to run itable init. thread or not.
3933 * If there is at least one uninitialized inode table, return
3934 * corresponding group number, else the loop goes through all
3935 * groups and return total number of groups.
3936 */
3937static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3938{
3939 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3940 struct ext4_group_desc *gdp = NULL;
3941
3942 if (!ext4_has_group_desc_csum(sb))
3943 return ngroups;
3944
3945 for (group = 0; group < ngroups; group++) {
3946 gdp = ext4_get_group_desc(sb, group, NULL);
3947 if (!gdp)
3948 continue;
3949
3950 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3951 break;
3952 }
3953
3954 return group;
3955}
3956
3957static int ext4_li_info_new(void)
3958{
3959 struct ext4_lazy_init *eli = NULL;
3960
3961 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3962 if (!eli)
3963 return -ENOMEM;
3964
3965 INIT_LIST_HEAD(&eli->li_request_list);
3966 mutex_init(&eli->li_list_mtx);
3967
3968 eli->li_state |= EXT4_LAZYINIT_QUIT;
3969
3970 ext4_li_info = eli;
3971
3972 return 0;
3973}
3974
3975static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3976 ext4_group_t start)
3977{
3978 struct ext4_li_request *elr;
3979
3980 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3981 if (!elr)
3982 return NULL;
3983
3984 elr->lr_super = sb;
3985 elr->lr_first_not_zeroed = start;
3986 if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3987 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3988 elr->lr_next_group = start;
3989 } else {
3990 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3991 }
3992
3993 /*
3994 * Randomize first schedule time of the request to
3995 * spread the inode table initialization requests
3996 * better.
3997 */
3998 elr->lr_next_sched = jiffies + get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
3999 return elr;
4000}
4001
4002int ext4_register_li_request(struct super_block *sb,
4003 ext4_group_t first_not_zeroed)
4004{
4005 struct ext4_sb_info *sbi = EXT4_SB(sb);
4006 struct ext4_li_request *elr = NULL;
4007 ext4_group_t ngroups = sbi->s_groups_count;
4008 int ret = 0;
4009
4010 mutex_lock(&ext4_li_mtx);
4011 if (sbi->s_li_request != NULL) {
4012 /*
4013 * Reset timeout so it can be computed again, because
4014 * s_li_wait_mult might have changed.
4015 */
4016 sbi->s_li_request->lr_timeout = 0;
4017 goto out;
4018 }
4019
4020 if (sb_rdonly(sb) ||
4021 (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
4022 (first_not_zeroed == ngroups || !test_opt(sb, INIT_INODE_TABLE))))
4023 goto out;
4024
4025 elr = ext4_li_request_new(sb, first_not_zeroed);
4026 if (!elr) {
4027 ret = -ENOMEM;
4028 goto out;
4029 }
4030
4031 if (NULL == ext4_li_info) {
4032 ret = ext4_li_info_new();
4033 if (ret)
4034 goto out;
4035 }
4036
4037 mutex_lock(&ext4_li_info->li_list_mtx);
4038 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
4039 mutex_unlock(&ext4_li_info->li_list_mtx);
4040
4041 sbi->s_li_request = elr;
4042 /*
4043 * set elr to NULL here since it has been inserted to
4044 * the request_list and the removal and free of it is
4045 * handled by ext4_clear_request_list from now on.
4046 */
4047 elr = NULL;
4048
4049 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
4050 ret = ext4_run_lazyinit_thread();
4051 if (ret)
4052 goto out;
4053 }
4054out:
4055 mutex_unlock(&ext4_li_mtx);
4056 if (ret)
4057 kfree(elr);
4058 return ret;
4059}
4060
4061/*
4062 * We do not need to lock anything since this is called on
4063 * module unload.
4064 */
4065static void ext4_destroy_lazyinit_thread(void)
4066{
4067 /*
4068 * If thread exited earlier
4069 * there's nothing to be done.
4070 */
4071 if (!ext4_li_info || !ext4_lazyinit_task)
4072 return;
4073
4074 kthread_stop(ext4_lazyinit_task);
4075}
4076
4077static int set_journal_csum_feature_set(struct super_block *sb)
4078{
4079 int ret = 1;
4080 int compat, incompat;
4081 struct ext4_sb_info *sbi = EXT4_SB(sb);
4082
4083 if (ext4_has_metadata_csum(sb)) {
4084 /* journal checksum v3 */
4085 compat = 0;
4086 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
4087 } else {
4088 /* journal checksum v1 */
4089 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
4090 incompat = 0;
4091 }
4092
4093 jbd2_journal_clear_features(sbi->s_journal,
4094 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
4095 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
4096 JBD2_FEATURE_INCOMPAT_CSUM_V2);
4097 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4098 ret = jbd2_journal_set_features(sbi->s_journal,
4099 compat, 0,
4100 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
4101 incompat);
4102 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
4103 ret = jbd2_journal_set_features(sbi->s_journal,
4104 compat, 0,
4105 incompat);
4106 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4107 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4108 } else {
4109 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
4110 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
4111 }
4112
4113 return ret;
4114}
4115
4116/*
4117 * Note: calculating the overhead so we can be compatible with
4118 * historical BSD practice is quite difficult in the face of
4119 * clusters/bigalloc. This is because multiple metadata blocks from
4120 * different block group can end up in the same allocation cluster.
4121 * Calculating the exact overhead in the face of clustered allocation
4122 * requires either O(all block bitmaps) in memory or O(number of block
4123 * groups**2) in time. We will still calculate the superblock for
4124 * older file systems --- and if we come across with a bigalloc file
4125 * system with zero in s_overhead_clusters the estimate will be close to
4126 * correct especially for very large cluster sizes --- but for newer
4127 * file systems, it's better to calculate this figure once at mkfs
4128 * time, and store it in the superblock. If the superblock value is
4129 * present (even for non-bigalloc file systems), we will use it.
4130 */
4131static int count_overhead(struct super_block *sb, ext4_group_t grp,
4132 char *buf)
4133{
4134 struct ext4_sb_info *sbi = EXT4_SB(sb);
4135 struct ext4_group_desc *gdp;
4136 ext4_fsblk_t first_block, last_block, b;
4137 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4138 int s, j, count = 0;
4139 int has_super = ext4_bg_has_super(sb, grp);
4140
4141 if (!ext4_has_feature_bigalloc(sb))
4142 return (has_super + ext4_bg_num_gdb(sb, grp) +
4143 (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
4144 sbi->s_itb_per_group + 2);
4145
4146 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
4147 (grp * EXT4_BLOCKS_PER_GROUP(sb));
4148 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
4149 for (i = 0; i < ngroups; i++) {
4150 gdp = ext4_get_group_desc(sb, i, NULL);
4151 b = ext4_block_bitmap(sb, gdp);
4152 if (b >= first_block && b <= last_block) {
4153 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4154 count++;
4155 }
4156 b = ext4_inode_bitmap(sb, gdp);
4157 if (b >= first_block && b <= last_block) {
4158 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
4159 count++;
4160 }
4161 b = ext4_inode_table(sb, gdp);
4162 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
4163 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
4164 int c = EXT4_B2C(sbi, b - first_block);
4165 ext4_set_bit(c, buf);
4166 count++;
4167 }
4168 if (i != grp)
4169 continue;
4170 s = 0;
4171 if (ext4_bg_has_super(sb, grp)) {
4172 ext4_set_bit(s++, buf);
4173 count++;
4174 }
4175 j = ext4_bg_num_gdb(sb, grp);
4176 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
4177 ext4_error(sb, "Invalid number of block group "
4178 "descriptor blocks: %d", j);
4179 j = EXT4_BLOCKS_PER_GROUP(sb) - s;
4180 }
4181 count += j;
4182 for (; j > 0; j--)
4183 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
4184 }
4185 if (!count)
4186 return 0;
4187 return EXT4_CLUSTERS_PER_GROUP(sb) -
4188 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
4189}
4190
4191/*
4192 * Compute the overhead and stash it in sbi->s_overhead
4193 */
4194int ext4_calculate_overhead(struct super_block *sb)
4195{
4196 struct ext4_sb_info *sbi = EXT4_SB(sb);
4197 struct ext4_super_block *es = sbi->s_es;
4198 struct inode *j_inode;
4199 unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
4200 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4201 ext4_fsblk_t overhead = 0;
4202 char *buf = (char *) get_zeroed_page(GFP_NOFS);
4203
4204 if (!buf)
4205 return -ENOMEM;
4206
4207 /*
4208 * Compute the overhead (FS structures). This is constant
4209 * for a given filesystem unless the number of block groups
4210 * changes so we cache the previous value until it does.
4211 */
4212
4213 /*
4214 * All of the blocks before first_data_block are overhead
4215 */
4216 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4217
4218 /*
4219 * Add the overhead found in each block group
4220 */
4221 for (i = 0; i < ngroups; i++) {
4222 int blks;
4223
4224 blks = count_overhead(sb, i, buf);
4225 overhead += blks;
4226 if (blks)
4227 memset(buf, 0, PAGE_SIZE);
4228 cond_resched();
4229 }
4230
4231 /*
4232 * Add the internal journal blocks whether the journal has been
4233 * loaded or not
4234 */
4235 if (sbi->s_journal && !sbi->s_journal_bdev_file)
4236 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
4237 else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
4238 /* j_inum for internal journal is non-zero */
4239 j_inode = ext4_get_journal_inode(sb, j_inum);
4240 if (!IS_ERR(j_inode)) {
4241 j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
4242 overhead += EXT4_NUM_B2C(sbi, j_blocks);
4243 iput(j_inode);
4244 } else {
4245 ext4_msg(sb, KERN_ERR, "can't get journal size");
4246 }
4247 }
4248 sbi->s_overhead = overhead;
4249 smp_wmb();
4250 free_page((unsigned long) buf);
4251 return 0;
4252}
4253
4254static void ext4_set_resv_clusters(struct super_block *sb)
4255{
4256 ext4_fsblk_t resv_clusters;
4257 struct ext4_sb_info *sbi = EXT4_SB(sb);
4258
4259 /*
4260 * There's no need to reserve anything when we aren't using extents.
4261 * The space estimates are exact, there are no unwritten extents,
4262 * hole punching doesn't need new metadata... This is needed especially
4263 * to keep ext2/3 backward compatibility.
4264 */
4265 if (!ext4_has_feature_extents(sb))
4266 return;
4267 /*
4268 * By default we reserve 2% or 4096 clusters, whichever is smaller.
4269 * This should cover the situations where we can not afford to run
4270 * out of space like for example punch hole, or converting
4271 * unwritten extents in delalloc path. In most cases such
4272 * allocation would require 1, or 2 blocks, higher numbers are
4273 * very rare.
4274 */
4275 resv_clusters = (ext4_blocks_count(sbi->s_es) >>
4276 sbi->s_cluster_bits);
4277
4278 do_div(resv_clusters, 50);
4279 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
4280
4281 atomic64_set(&sbi->s_resv_clusters, resv_clusters);
4282}
4283
4284static const char *ext4_quota_mode(struct super_block *sb)
4285{
4286#ifdef CONFIG_QUOTA
4287 if (!ext4_quota_capable(sb))
4288 return "none";
4289
4290 if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
4291 return "journalled";
4292 else
4293 return "writeback";
4294#else
4295 return "disabled";
4296#endif
4297}
4298
4299static void ext4_setup_csum_trigger(struct super_block *sb,
4300 enum ext4_journal_trigger_type type,
4301 void (*trigger)(
4302 struct jbd2_buffer_trigger_type *type,
4303 struct buffer_head *bh,
4304 void *mapped_data,
4305 size_t size))
4306{
4307 struct ext4_sb_info *sbi = EXT4_SB(sb);
4308
4309 sbi->s_journal_triggers[type].sb = sb;
4310 sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
4311}
4312
4313static void ext4_free_sbi(struct ext4_sb_info *sbi)
4314{
4315 if (!sbi)
4316 return;
4317
4318 kfree(sbi->s_blockgroup_lock);
4319 fs_put_dax(sbi->s_daxdev, NULL);
4320 kfree(sbi);
4321}
4322
4323static struct ext4_sb_info *ext4_alloc_sbi(struct super_block *sb)
4324{
4325 struct ext4_sb_info *sbi;
4326
4327 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
4328 if (!sbi)
4329 return NULL;
4330
4331 sbi->s_daxdev = fs_dax_get_by_bdev(sb->s_bdev, &sbi->s_dax_part_off,
4332 NULL, NULL);
4333
4334 sbi->s_blockgroup_lock =
4335 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
4336
4337 if (!sbi->s_blockgroup_lock)
4338 goto err_out;
4339
4340 sb->s_fs_info = sbi;
4341 sbi->s_sb = sb;
4342 return sbi;
4343err_out:
4344 fs_put_dax(sbi->s_daxdev, NULL);
4345 kfree(sbi);
4346 return NULL;
4347}
4348
4349static void ext4_set_def_opts(struct super_block *sb,
4350 struct ext4_super_block *es)
4351{
4352 unsigned long def_mount_opts;
4353
4354 /* Set defaults before we parse the mount options */
4355 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4356 set_opt(sb, INIT_INODE_TABLE);
4357 if (def_mount_opts & EXT4_DEFM_DEBUG)
4358 set_opt(sb, DEBUG);
4359 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4360 set_opt(sb, GRPID);
4361 if (def_mount_opts & EXT4_DEFM_UID16)
4362 set_opt(sb, NO_UID32);
4363 /* xattr user namespace & acls are now defaulted on */
4364 set_opt(sb, XATTR_USER);
4365#ifdef CONFIG_EXT4_FS_POSIX_ACL
4366 set_opt(sb, POSIX_ACL);
4367#endif
4368 if (ext4_has_feature_fast_commit(sb))
4369 set_opt2(sb, JOURNAL_FAST_COMMIT);
4370 /* don't forget to enable journal_csum when metadata_csum is enabled. */
4371 if (ext4_has_metadata_csum(sb))
4372 set_opt(sb, JOURNAL_CHECKSUM);
4373
4374 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4375 set_opt(sb, JOURNAL_DATA);
4376 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4377 set_opt(sb, ORDERED_DATA);
4378 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4379 set_opt(sb, WRITEBACK_DATA);
4380
4381 if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_PANIC)
4382 set_opt(sb, ERRORS_PANIC);
4383 else if (le16_to_cpu(es->s_errors) == EXT4_ERRORS_CONTINUE)
4384 set_opt(sb, ERRORS_CONT);
4385 else
4386 set_opt(sb, ERRORS_RO);
4387 /* block_validity enabled by default; disable with noblock_validity */
4388 set_opt(sb, BLOCK_VALIDITY);
4389 if (def_mount_opts & EXT4_DEFM_DISCARD)
4390 set_opt(sb, DISCARD);
4391
4392 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4393 set_opt(sb, BARRIER);
4394
4395 /*
4396 * enable delayed allocation by default
4397 * Use -o nodelalloc to turn it off
4398 */
4399 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4400 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4401 set_opt(sb, DELALLOC);
4402
4403 if (sb->s_blocksize <= PAGE_SIZE)
4404 set_opt(sb, DIOREAD_NOLOCK);
4405}
4406
4407static int ext4_handle_clustersize(struct super_block *sb)
4408{
4409 struct ext4_sb_info *sbi = EXT4_SB(sb);
4410 struct ext4_super_block *es = sbi->s_es;
4411 int clustersize;
4412
4413 /* Handle clustersize */
4414 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4415 if (ext4_has_feature_bigalloc(sb)) {
4416 if (clustersize < sb->s_blocksize) {
4417 ext4_msg(sb, KERN_ERR,
4418 "cluster size (%d) smaller than "
4419 "block size (%lu)", clustersize, sb->s_blocksize);
4420 return -EINVAL;
4421 }
4422 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4423 le32_to_cpu(es->s_log_block_size);
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_cluster_bits = 0;
4438 }
4439 sbi->s_clusters_per_group = le32_to_cpu(es->s_clusters_per_group);
4440 if (sbi->s_clusters_per_group > sb->s_blocksize * 8) {
4441 ext4_msg(sb, KERN_ERR, "#clusters per group too big: %lu",
4442 sbi->s_clusters_per_group);
4443 return -EINVAL;
4444 }
4445 if (sbi->s_blocks_per_group !=
4446 (sbi->s_clusters_per_group * (clustersize / sb->s_blocksize))) {
4447 ext4_msg(sb, KERN_ERR,
4448 "blocks per group (%lu) and clusters per group (%lu) inconsistent",
4449 sbi->s_blocks_per_group, sbi->s_clusters_per_group);
4450 return -EINVAL;
4451 }
4452 sbi->s_cluster_ratio = clustersize / sb->s_blocksize;
4453
4454 /* Do we have standard group size of clustersize * 8 blocks ? */
4455 if (sbi->s_blocks_per_group == clustersize << 3)
4456 set_opt2(sb, STD_GROUP_SIZE);
4457
4458 return 0;
4459}
4460
4461static void ext4_fast_commit_init(struct super_block *sb)
4462{
4463 struct ext4_sb_info *sbi = EXT4_SB(sb);
4464
4465 /* Initialize fast commit stuff */
4466 atomic_set(&sbi->s_fc_subtid, 0);
4467 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4468 INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4469 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4470 INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4471 sbi->s_fc_bytes = 0;
4472 ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4473 sbi->s_fc_ineligible_tid = 0;
4474 spin_lock_init(&sbi->s_fc_lock);
4475 memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4476 sbi->s_fc_replay_state.fc_regions = NULL;
4477 sbi->s_fc_replay_state.fc_regions_size = 0;
4478 sbi->s_fc_replay_state.fc_regions_used = 0;
4479 sbi->s_fc_replay_state.fc_regions_valid = 0;
4480 sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4481 sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4482 sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4483}
4484
4485static int ext4_inode_info_init(struct super_block *sb,
4486 struct ext4_super_block *es)
4487{
4488 struct ext4_sb_info *sbi = EXT4_SB(sb);
4489
4490 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4491 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4492 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4493 } else {
4494 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4495 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4496 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4497 ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4498 sbi->s_first_ino);
4499 return -EINVAL;
4500 }
4501 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4502 (!is_power_of_2(sbi->s_inode_size)) ||
4503 (sbi->s_inode_size > sb->s_blocksize)) {
4504 ext4_msg(sb, KERN_ERR,
4505 "unsupported inode size: %d",
4506 sbi->s_inode_size);
4507 ext4_msg(sb, KERN_ERR, "blocksize: %lu", sb->s_blocksize);
4508 return -EINVAL;
4509 }
4510 /*
4511 * i_atime_extra is the last extra field available for
4512 * [acm]times in struct ext4_inode. Checking for that
4513 * field should suffice to ensure we have extra space
4514 * for all three.
4515 */
4516 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4517 sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4518 sb->s_time_gran = 1;
4519 sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4520 } else {
4521 sb->s_time_gran = NSEC_PER_SEC;
4522 sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4523 }
4524 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4525 }
4526
4527 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4528 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4529 EXT4_GOOD_OLD_INODE_SIZE;
4530 if (ext4_has_feature_extra_isize(sb)) {
4531 unsigned v, max = (sbi->s_inode_size -
4532 EXT4_GOOD_OLD_INODE_SIZE);
4533
4534 v = le16_to_cpu(es->s_want_extra_isize);
4535 if (v > max) {
4536 ext4_msg(sb, KERN_ERR,
4537 "bad s_want_extra_isize: %d", v);
4538 return -EINVAL;
4539 }
4540 if (sbi->s_want_extra_isize < v)
4541 sbi->s_want_extra_isize = v;
4542
4543 v = le16_to_cpu(es->s_min_extra_isize);
4544 if (v > max) {
4545 ext4_msg(sb, KERN_ERR,
4546 "bad s_min_extra_isize: %d", v);
4547 return -EINVAL;
4548 }
4549 if (sbi->s_want_extra_isize < v)
4550 sbi->s_want_extra_isize = v;
4551 }
4552 }
4553
4554 return 0;
4555}
4556
4557#if IS_ENABLED(CONFIG_UNICODE)
4558static int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4559{
4560 const struct ext4_sb_encodings *encoding_info;
4561 struct unicode_map *encoding;
4562 __u16 encoding_flags = le16_to_cpu(es->s_encoding_flags);
4563
4564 if (!ext4_has_feature_casefold(sb) || sb->s_encoding)
4565 return 0;
4566
4567 encoding_info = ext4_sb_read_encoding(es);
4568 if (!encoding_info) {
4569 ext4_msg(sb, KERN_ERR,
4570 "Encoding requested by superblock is unknown");
4571 return -EINVAL;
4572 }
4573
4574 encoding = utf8_load(encoding_info->version);
4575 if (IS_ERR(encoding)) {
4576 ext4_msg(sb, KERN_ERR,
4577 "can't mount with superblock charset: %s-%u.%u.%u "
4578 "not supported by the kernel. flags: 0x%x.",
4579 encoding_info->name,
4580 unicode_major(encoding_info->version),
4581 unicode_minor(encoding_info->version),
4582 unicode_rev(encoding_info->version),
4583 encoding_flags);
4584 return -EINVAL;
4585 }
4586 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4587 "%s-%u.%u.%u with flags 0x%hx", encoding_info->name,
4588 unicode_major(encoding_info->version),
4589 unicode_minor(encoding_info->version),
4590 unicode_rev(encoding_info->version),
4591 encoding_flags);
4592
4593 sb->s_encoding = encoding;
4594 sb->s_encoding_flags = encoding_flags;
4595
4596 return 0;
4597}
4598#else
4599static inline int ext4_encoding_init(struct super_block *sb, struct ext4_super_block *es)
4600{
4601 return 0;
4602}
4603#endif
4604
4605static int ext4_init_metadata_csum(struct super_block *sb, struct ext4_super_block *es)
4606{
4607 struct ext4_sb_info *sbi = EXT4_SB(sb);
4608
4609 /* Warn if metadata_csum and gdt_csum are both set. */
4610 if (ext4_has_feature_metadata_csum(sb) &&
4611 ext4_has_feature_gdt_csum(sb))
4612 ext4_warning(sb, "metadata_csum and uninit_bg are "
4613 "redundant flags; please run fsck.");
4614
4615 /* Check for a known checksum algorithm */
4616 if (!ext4_verify_csum_type(sb, es)) {
4617 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4618 "unknown checksum algorithm.");
4619 return -EINVAL;
4620 }
4621 ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
4622 ext4_orphan_file_block_trigger);
4623
4624 /* Load the checksum driver */
4625 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
4626 if (IS_ERR(sbi->s_chksum_driver)) {
4627 int ret = PTR_ERR(sbi->s_chksum_driver);
4628 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
4629 sbi->s_chksum_driver = NULL;
4630 return ret;
4631 }
4632
4633 /* Check superblock checksum */
4634 if (!ext4_superblock_csum_verify(sb, es)) {
4635 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
4636 "invalid superblock checksum. Run e2fsck?");
4637 return -EFSBADCRC;
4638 }
4639
4640 /* Precompute checksum seed for all metadata */
4641 if (ext4_has_feature_csum_seed(sb))
4642 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
4643 else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
4644 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
4645 sizeof(es->s_uuid));
4646 return 0;
4647}
4648
4649static int ext4_check_feature_compatibility(struct super_block *sb,
4650 struct ext4_super_block *es,
4651 int silent)
4652{
4653 struct ext4_sb_info *sbi = EXT4_SB(sb);
4654
4655 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4656 (ext4_has_compat_features(sb) ||
4657 ext4_has_ro_compat_features(sb) ||
4658 ext4_has_incompat_features(sb)))
4659 ext4_msg(sb, KERN_WARNING,
4660 "feature flags set on rev 0 fs, "
4661 "running e2fsck is recommended");
4662
4663 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4664 set_opt2(sb, HURD_COMPAT);
4665 if (ext4_has_feature_64bit(sb)) {
4666 ext4_msg(sb, KERN_ERR,
4667 "The Hurd can't support 64-bit file systems");
4668 return -EINVAL;
4669 }
4670
4671 /*
4672 * ea_inode feature uses l_i_version field which is not
4673 * available in HURD_COMPAT mode.
4674 */
4675 if (ext4_has_feature_ea_inode(sb)) {
4676 ext4_msg(sb, KERN_ERR,
4677 "ea_inode feature is not supported for Hurd");
4678 return -EINVAL;
4679 }
4680 }
4681
4682 if (IS_EXT2_SB(sb)) {
4683 if (ext2_feature_set_ok(sb))
4684 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4685 "using the ext4 subsystem");
4686 else {
4687 /*
4688 * If we're probing be silent, if this looks like
4689 * it's actually an ext[34] filesystem.
4690 */
4691 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4692 return -EINVAL;
4693 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4694 "to feature incompatibilities");
4695 return -EINVAL;
4696 }
4697 }
4698
4699 if (IS_EXT3_SB(sb)) {
4700 if (ext3_feature_set_ok(sb))
4701 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4702 "using the ext4 subsystem");
4703 else {
4704 /*
4705 * If we're probing be silent, if this looks like
4706 * it's actually an ext4 filesystem.
4707 */
4708 if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4709 return -EINVAL;
4710 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4711 "to feature incompatibilities");
4712 return -EINVAL;
4713 }
4714 }
4715
4716 /*
4717 * Check feature flags regardless of the revision level, since we
4718 * previously didn't change the revision level when setting the flags,
4719 * so there is a chance incompat flags are set on a rev 0 filesystem.
4720 */
4721 if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4722 return -EINVAL;
4723
4724 if (sbi->s_daxdev) {
4725 if (sb->s_blocksize == PAGE_SIZE)
4726 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4727 else
4728 ext4_msg(sb, KERN_ERR, "unsupported blocksize for DAX\n");
4729 }
4730
4731 if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4732 if (ext4_has_feature_inline_data(sb)) {
4733 ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4734 " that may contain inline data");
4735 return -EINVAL;
4736 }
4737 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4738 ext4_msg(sb, KERN_ERR,
4739 "DAX unsupported by block device.");
4740 return -EINVAL;
4741 }
4742 }
4743
4744 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4745 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4746 es->s_encryption_level);
4747 return -EINVAL;
4748 }
4749
4750 return 0;
4751}
4752
4753static int ext4_check_geometry(struct super_block *sb,
4754 struct ext4_super_block *es)
4755{
4756 struct ext4_sb_info *sbi = EXT4_SB(sb);
4757 __u64 blocks_count;
4758 int err;
4759
4760 if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (sb->s_blocksize / 4)) {
4761 ext4_msg(sb, KERN_ERR,
4762 "Number of reserved GDT blocks insanely large: %d",
4763 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4764 return -EINVAL;
4765 }
4766 /*
4767 * Test whether we have more sectors than will fit in sector_t,
4768 * and whether the max offset is addressable by the page cache.
4769 */
4770 err = generic_check_addressable(sb->s_blocksize_bits,
4771 ext4_blocks_count(es));
4772 if (err) {
4773 ext4_msg(sb, KERN_ERR, "filesystem"
4774 " too large to mount safely on this system");
4775 return err;
4776 }
4777
4778 /* check blocks count against device size */
4779 blocks_count = sb_bdev_nr_blocks(sb);
4780 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4781 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4782 "exceeds size of device (%llu blocks)",
4783 ext4_blocks_count(es), blocks_count);
4784 return -EINVAL;
4785 }
4786
4787 /*
4788 * It makes no sense for the first data block to be beyond the end
4789 * of the filesystem.
4790 */
4791 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4792 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4793 "block %u is beyond end of filesystem (%llu)",
4794 le32_to_cpu(es->s_first_data_block),
4795 ext4_blocks_count(es));
4796 return -EINVAL;
4797 }
4798 if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4799 (sbi->s_cluster_ratio == 1)) {
4800 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4801 "block is 0 with a 1k block and cluster size");
4802 return -EINVAL;
4803 }
4804
4805 blocks_count = (ext4_blocks_count(es) -
4806 le32_to_cpu(es->s_first_data_block) +
4807 EXT4_BLOCKS_PER_GROUP(sb) - 1);
4808 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4809 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4810 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4811 "(block count %llu, first data block %u, "
4812 "blocks per group %lu)", blocks_count,
4813 ext4_blocks_count(es),
4814 le32_to_cpu(es->s_first_data_block),
4815 EXT4_BLOCKS_PER_GROUP(sb));
4816 return -EINVAL;
4817 }
4818 sbi->s_groups_count = blocks_count;
4819 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4820 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4821 if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4822 le32_to_cpu(es->s_inodes_count)) {
4823 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4824 le32_to_cpu(es->s_inodes_count),
4825 ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4826 return -EINVAL;
4827 }
4828
4829 return 0;
4830}
4831
4832static int ext4_group_desc_init(struct super_block *sb,
4833 struct ext4_super_block *es,
4834 ext4_fsblk_t logical_sb_block,
4835 ext4_group_t *first_not_zeroed)
4836{
4837 struct ext4_sb_info *sbi = EXT4_SB(sb);
4838 unsigned int db_count;
4839 ext4_fsblk_t block;
4840 int i;
4841
4842 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4843 EXT4_DESC_PER_BLOCK(sb);
4844 if (ext4_has_feature_meta_bg(sb)) {
4845 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4846 ext4_msg(sb, KERN_WARNING,
4847 "first meta block group too large: %u "
4848 "(group descriptor block count %u)",
4849 le32_to_cpu(es->s_first_meta_bg), db_count);
4850 return -EINVAL;
4851 }
4852 }
4853 rcu_assign_pointer(sbi->s_group_desc,
4854 kvmalloc_array(db_count,
4855 sizeof(struct buffer_head *),
4856 GFP_KERNEL));
4857 if (sbi->s_group_desc == NULL) {
4858 ext4_msg(sb, KERN_ERR, "not enough memory");
4859 return -ENOMEM;
4860 }
4861
4862 bgl_lock_init(sbi->s_blockgroup_lock);
4863
4864 /* Pre-read the descriptors into the buffer cache */
4865 for (i = 0; i < db_count; i++) {
4866 block = descriptor_loc(sb, logical_sb_block, i);
4867 ext4_sb_breadahead_unmovable(sb, block);
4868 }
4869
4870 for (i = 0; i < db_count; i++) {
4871 struct buffer_head *bh;
4872
4873 block = descriptor_loc(sb, logical_sb_block, i);
4874 bh = ext4_sb_bread_unmovable(sb, block);
4875 if (IS_ERR(bh)) {
4876 ext4_msg(sb, KERN_ERR,
4877 "can't read group descriptor %d", i);
4878 sbi->s_gdb_count = i;
4879 return PTR_ERR(bh);
4880 }
4881 rcu_read_lock();
4882 rcu_dereference(sbi->s_group_desc)[i] = bh;
4883 rcu_read_unlock();
4884 }
4885 sbi->s_gdb_count = db_count;
4886 if (!ext4_check_descriptors(sb, logical_sb_block, first_not_zeroed)) {
4887 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4888 return -EFSCORRUPTED;
4889 }
4890
4891 return 0;
4892}
4893
4894static int ext4_load_and_init_journal(struct super_block *sb,
4895 struct ext4_super_block *es,
4896 struct ext4_fs_context *ctx)
4897{
4898 struct ext4_sb_info *sbi = EXT4_SB(sb);
4899 int err;
4900
4901 err = ext4_load_journal(sb, es, ctx->journal_devnum);
4902 if (err)
4903 return err;
4904
4905 if (ext4_has_feature_64bit(sb) &&
4906 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4907 JBD2_FEATURE_INCOMPAT_64BIT)) {
4908 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4909 goto out;
4910 }
4911
4912 if (!set_journal_csum_feature_set(sb)) {
4913 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4914 "feature set");
4915 goto out;
4916 }
4917
4918 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4919 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4920 JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4921 ext4_msg(sb, KERN_ERR,
4922 "Failed to set fast commit journal feature");
4923 goto out;
4924 }
4925
4926 /* We have now updated the journal if required, so we can
4927 * validate the data journaling mode. */
4928 switch (test_opt(sb, DATA_FLAGS)) {
4929 case 0:
4930 /* No mode set, assume a default based on the journal
4931 * capabilities: ORDERED_DATA if the journal can
4932 * cope, else JOURNAL_DATA
4933 */
4934 if (jbd2_journal_check_available_features
4935 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4936 set_opt(sb, ORDERED_DATA);
4937 sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4938 } else {
4939 set_opt(sb, JOURNAL_DATA);
4940 sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4941 }
4942 break;
4943
4944 case EXT4_MOUNT_ORDERED_DATA:
4945 case EXT4_MOUNT_WRITEBACK_DATA:
4946 if (!jbd2_journal_check_available_features
4947 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4948 ext4_msg(sb, KERN_ERR, "Journal does not support "
4949 "requested data journaling mode");
4950 goto out;
4951 }
4952 break;
4953 default:
4954 break;
4955 }
4956
4957 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4958 test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4959 ext4_msg(sb, KERN_ERR, "can't mount with "
4960 "journal_async_commit in data=ordered mode");
4961 goto out;
4962 }
4963
4964 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
4965
4966 sbi->s_journal->j_submit_inode_data_buffers =
4967 ext4_journal_submit_inode_data_buffers;
4968 sbi->s_journal->j_finish_inode_data_buffers =
4969 ext4_journal_finish_inode_data_buffers;
4970
4971 return 0;
4972
4973out:
4974 /* flush s_sb_upd_work before destroying the journal. */
4975 flush_work(&sbi->s_sb_upd_work);
4976 jbd2_journal_destroy(sbi->s_journal);
4977 sbi->s_journal = NULL;
4978 return -EINVAL;
4979}
4980
4981static int ext4_check_journal_data_mode(struct super_block *sb)
4982{
4983 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4984 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with "
4985 "data=journal disables delayed allocation, "
4986 "dioread_nolock, O_DIRECT and fast_commit support!\n");
4987 /* can't mount with both data=journal and dioread_nolock. */
4988 clear_opt(sb, DIOREAD_NOLOCK);
4989 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4990 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4991 ext4_msg(sb, KERN_ERR, "can't mount with "
4992 "both data=journal and delalloc");
4993 return -EINVAL;
4994 }
4995 if (test_opt(sb, DAX_ALWAYS)) {
4996 ext4_msg(sb, KERN_ERR, "can't mount with "
4997 "both data=journal and dax");
4998 return -EINVAL;
4999 }
5000 if (ext4_has_feature_encrypt(sb)) {
5001 ext4_msg(sb, KERN_WARNING,
5002 "encrypted files will use data=ordered "
5003 "instead of data journaling mode");
5004 }
5005 if (test_opt(sb, DELALLOC))
5006 clear_opt(sb, DELALLOC);
5007 } else {
5008 sb->s_iflags |= SB_I_CGROUPWB;
5009 }
5010
5011 return 0;
5012}
5013
5014static int ext4_load_super(struct super_block *sb, ext4_fsblk_t *lsb,
5015 int silent)
5016{
5017 struct ext4_sb_info *sbi = EXT4_SB(sb);
5018 struct ext4_super_block *es;
5019 ext4_fsblk_t logical_sb_block;
5020 unsigned long offset = 0;
5021 struct buffer_head *bh;
5022 int ret = -EINVAL;
5023 int blocksize;
5024
5025 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
5026 if (!blocksize) {
5027 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
5028 return -EINVAL;
5029 }
5030
5031 /*
5032 * The ext4 superblock will not be buffer aligned for other than 1kB
5033 * block sizes. We need to calculate the offset from buffer start.
5034 */
5035 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
5036 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5037 offset = do_div(logical_sb_block, blocksize);
5038 } else {
5039 logical_sb_block = sbi->s_sb_block;
5040 }
5041
5042 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5043 if (IS_ERR(bh)) {
5044 ext4_msg(sb, KERN_ERR, "unable to read superblock");
5045 return PTR_ERR(bh);
5046 }
5047 /*
5048 * Note: s_es must be initialized as soon as possible because
5049 * some ext4 macro-instructions depend on its value
5050 */
5051 es = (struct ext4_super_block *) (bh->b_data + offset);
5052 sbi->s_es = es;
5053 sb->s_magic = le16_to_cpu(es->s_magic);
5054 if (sb->s_magic != EXT4_SUPER_MAGIC) {
5055 if (!silent)
5056 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5057 goto out;
5058 }
5059
5060 if (le32_to_cpu(es->s_log_block_size) >
5061 (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5062 ext4_msg(sb, KERN_ERR,
5063 "Invalid log block size: %u",
5064 le32_to_cpu(es->s_log_block_size));
5065 goto out;
5066 }
5067 if (le32_to_cpu(es->s_log_cluster_size) >
5068 (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
5069 ext4_msg(sb, KERN_ERR,
5070 "Invalid log cluster size: %u",
5071 le32_to_cpu(es->s_log_cluster_size));
5072 goto out;
5073 }
5074
5075 blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
5076
5077 /*
5078 * If the default block size is not the same as the real block size,
5079 * we need to reload it.
5080 */
5081 if (sb->s_blocksize == blocksize) {
5082 *lsb = logical_sb_block;
5083 sbi->s_sbh = bh;
5084 return 0;
5085 }
5086
5087 /*
5088 * bh must be released before kill_bdev(), otherwise
5089 * it won't be freed and its page also. kill_bdev()
5090 * is called by sb_set_blocksize().
5091 */
5092 brelse(bh);
5093 /* Validate the filesystem blocksize */
5094 if (!sb_set_blocksize(sb, blocksize)) {
5095 ext4_msg(sb, KERN_ERR, "bad block size %d",
5096 blocksize);
5097 bh = NULL;
5098 goto out;
5099 }
5100
5101 logical_sb_block = sbi->s_sb_block * EXT4_MIN_BLOCK_SIZE;
5102 offset = do_div(logical_sb_block, blocksize);
5103 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
5104 if (IS_ERR(bh)) {
5105 ext4_msg(sb, KERN_ERR, "Can't read superblock on 2nd try");
5106 ret = PTR_ERR(bh);
5107 bh = NULL;
5108 goto out;
5109 }
5110 es = (struct ext4_super_block *)(bh->b_data + offset);
5111 sbi->s_es = es;
5112 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
5113 ext4_msg(sb, KERN_ERR, "Magic mismatch, very weird!");
5114 goto out;
5115 }
5116 *lsb = logical_sb_block;
5117 sbi->s_sbh = bh;
5118 return 0;
5119out:
5120 brelse(bh);
5121 return ret;
5122}
5123
5124static void ext4_hash_info_init(struct super_block *sb)
5125{
5126 struct ext4_sb_info *sbi = EXT4_SB(sb);
5127 struct ext4_super_block *es = sbi->s_es;
5128 unsigned int i;
5129
5130 for (i = 0; i < 4; i++)
5131 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
5132
5133 sbi->s_def_hash_version = es->s_def_hash_version;
5134 if (ext4_has_feature_dir_index(sb)) {
5135 i = le32_to_cpu(es->s_flags);
5136 if (i & EXT2_FLAGS_UNSIGNED_HASH)
5137 sbi->s_hash_unsigned = 3;
5138 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
5139#ifdef __CHAR_UNSIGNED__
5140 if (!sb_rdonly(sb))
5141 es->s_flags |=
5142 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
5143 sbi->s_hash_unsigned = 3;
5144#else
5145 if (!sb_rdonly(sb))
5146 es->s_flags |=
5147 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
5148#endif
5149 }
5150 }
5151}
5152
5153static int ext4_block_group_meta_init(struct super_block *sb, int silent)
5154{
5155 struct ext4_sb_info *sbi = EXT4_SB(sb);
5156 struct ext4_super_block *es = sbi->s_es;
5157 int has_huge_files;
5158
5159 has_huge_files = ext4_has_feature_huge_file(sb);
5160 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
5161 has_huge_files);
5162 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
5163
5164 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
5165 if (ext4_has_feature_64bit(sb)) {
5166 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
5167 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
5168 !is_power_of_2(sbi->s_desc_size)) {
5169 ext4_msg(sb, KERN_ERR,
5170 "unsupported descriptor size %lu",
5171 sbi->s_desc_size);
5172 return -EINVAL;
5173 }
5174 } else
5175 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
5176
5177 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
5178 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
5179
5180 sbi->s_inodes_per_block = sb->s_blocksize / EXT4_INODE_SIZE(sb);
5181 if (sbi->s_inodes_per_block == 0 || sbi->s_blocks_per_group == 0) {
5182 if (!silent)
5183 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5184 return -EINVAL;
5185 }
5186 if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
5187 sbi->s_inodes_per_group > sb->s_blocksize * 8) {
5188 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
5189 sbi->s_inodes_per_group);
5190 return -EINVAL;
5191 }
5192 sbi->s_itb_per_group = sbi->s_inodes_per_group /
5193 sbi->s_inodes_per_block;
5194 sbi->s_desc_per_block = sb->s_blocksize / EXT4_DESC_SIZE(sb);
5195 sbi->s_mount_state = le16_to_cpu(es->s_state) & ~EXT4_FC_REPLAY;
5196 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
5197 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
5198
5199 return 0;
5200}
5201
5202static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
5203{
5204 struct ext4_super_block *es = NULL;
5205 struct ext4_sb_info *sbi = EXT4_SB(sb);
5206 ext4_fsblk_t logical_sb_block;
5207 struct inode *root;
5208 int needs_recovery;
5209 int err;
5210 ext4_group_t first_not_zeroed;
5211 struct ext4_fs_context *ctx = fc->fs_private;
5212 int silent = fc->sb_flags & SB_SILENT;
5213
5214 /* Set defaults for the variables that will be set during parsing */
5215 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
5216 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5217
5218 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
5219 sbi->s_sectors_written_start =
5220 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
5221
5222 err = ext4_load_super(sb, &logical_sb_block, silent);
5223 if (err)
5224 goto out_fail;
5225
5226 es = sbi->s_es;
5227 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
5228
5229 err = ext4_init_metadata_csum(sb, es);
5230 if (err)
5231 goto failed_mount;
5232
5233 ext4_set_def_opts(sb, es);
5234
5235 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
5236 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
5237 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
5238 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
5239 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
5240
5241 /*
5242 * set default s_li_wait_mult for lazyinit, for the case there is
5243 * no mount option specified.
5244 */
5245 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
5246
5247 err = ext4_inode_info_init(sb, es);
5248 if (err)
5249 goto failed_mount;
5250
5251 err = parse_apply_sb_mount_options(sb, ctx);
5252 if (err < 0)
5253 goto failed_mount;
5254
5255 sbi->s_def_mount_opt = sbi->s_mount_opt;
5256 sbi->s_def_mount_opt2 = sbi->s_mount_opt2;
5257
5258 err = ext4_check_opt_consistency(fc, sb);
5259 if (err < 0)
5260 goto failed_mount;
5261
5262 ext4_apply_options(fc, sb);
5263
5264 err = ext4_encoding_init(sb, es);
5265 if (err)
5266 goto failed_mount;
5267
5268 err = ext4_check_journal_data_mode(sb);
5269 if (err)
5270 goto failed_mount;
5271
5272 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5273 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5274
5275 /* i_version is always enabled now */
5276 sb->s_flags |= SB_I_VERSION;
5277
5278 err = ext4_check_feature_compatibility(sb, es, silent);
5279 if (err)
5280 goto failed_mount;
5281
5282 err = ext4_block_group_meta_init(sb, silent);
5283 if (err)
5284 goto failed_mount;
5285
5286 ext4_hash_info_init(sb);
5287
5288 err = ext4_handle_clustersize(sb);
5289 if (err)
5290 goto failed_mount;
5291
5292 err = ext4_check_geometry(sb, es);
5293 if (err)
5294 goto failed_mount;
5295
5296 timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
5297 spin_lock_init(&sbi->s_error_lock);
5298 INIT_WORK(&sbi->s_sb_upd_work, update_super_work);
5299
5300 err = ext4_group_desc_init(sb, es, logical_sb_block, &first_not_zeroed);
5301 if (err)
5302 goto failed_mount3;
5303
5304 err = ext4_es_register_shrinker(sbi);
5305 if (err)
5306 goto failed_mount3;
5307
5308 sbi->s_stripe = ext4_get_stripe_size(sbi);
5309 /*
5310 * It's hard to get stripe aligned blocks if stripe is not aligned with
5311 * cluster, just disable stripe and alert user to simpfy code and avoid
5312 * stripe aligned allocation which will rarely successes.
5313 */
5314 if (sbi->s_stripe > 0 && sbi->s_cluster_ratio > 1 &&
5315 sbi->s_stripe % sbi->s_cluster_ratio != 0) {
5316 ext4_msg(sb, KERN_WARNING,
5317 "stripe (%lu) is not aligned with cluster size (%u), "
5318 "stripe is disabled",
5319 sbi->s_stripe, sbi->s_cluster_ratio);
5320 sbi->s_stripe = 0;
5321 }
5322 sbi->s_extent_max_zeroout_kb = 32;
5323
5324 /*
5325 * set up enough so that it can read an inode
5326 */
5327 sb->s_op = &ext4_sops;
5328 sb->s_export_op = &ext4_export_ops;
5329 sb->s_xattr = ext4_xattr_handlers;
5330#ifdef CONFIG_FS_ENCRYPTION
5331 sb->s_cop = &ext4_cryptops;
5332#endif
5333#ifdef CONFIG_FS_VERITY
5334 sb->s_vop = &ext4_verityops;
5335#endif
5336#ifdef CONFIG_QUOTA
5337 sb->dq_op = &ext4_quota_operations;
5338 if (ext4_has_feature_quota(sb))
5339 sb->s_qcop = &dquot_quotactl_sysfile_ops;
5340 else
5341 sb->s_qcop = &ext4_qctl_operations;
5342 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
5343#endif
5344 super_set_uuid(sb, es->s_uuid, sizeof(es->s_uuid));
5345
5346 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
5347 mutex_init(&sbi->s_orphan_lock);
5348
5349 ext4_fast_commit_init(sb);
5350
5351 sb->s_root = NULL;
5352
5353 needs_recovery = (es->s_last_orphan != 0 ||
5354 ext4_has_feature_orphan_present(sb) ||
5355 ext4_has_feature_journal_needs_recovery(sb));
5356
5357 if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb)) {
5358 err = ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block));
5359 if (err)
5360 goto failed_mount3a;
5361 }
5362
5363 err = -EINVAL;
5364 /*
5365 * The first inode we look at is the journal inode. Don't try
5366 * root first: it may be modified in the journal!
5367 */
5368 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
5369 err = ext4_load_and_init_journal(sb, es, ctx);
5370 if (err)
5371 goto failed_mount3a;
5372 } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
5373 ext4_has_feature_journal_needs_recovery(sb)) {
5374 ext4_msg(sb, KERN_ERR, "required journal recovery "
5375 "suppressed and not mounted read-only");
5376 goto failed_mount3a;
5377 } else {
5378 /* Nojournal mode, all journal mount options are illegal */
5379 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5380 ext4_msg(sb, KERN_ERR, "can't mount with "
5381 "journal_async_commit, fs mounted w/o journal");
5382 goto failed_mount3a;
5383 }
5384
5385 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
5386 ext4_msg(sb, KERN_ERR, "can't mount with "
5387 "journal_checksum, fs mounted w/o journal");
5388 goto failed_mount3a;
5389 }
5390 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
5391 ext4_msg(sb, KERN_ERR, "can't mount with "
5392 "commit=%lu, fs mounted w/o journal",
5393 sbi->s_commit_interval / HZ);
5394 goto failed_mount3a;
5395 }
5396 if (EXT4_MOUNT_DATA_FLAGS &
5397 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
5398 ext4_msg(sb, KERN_ERR, "can't mount with "
5399 "data=, fs mounted w/o journal");
5400 goto failed_mount3a;
5401 }
5402 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
5403 clear_opt(sb, JOURNAL_CHECKSUM);
5404 clear_opt(sb, DATA_FLAGS);
5405 clear_opt2(sb, JOURNAL_FAST_COMMIT);
5406 sbi->s_journal = NULL;
5407 needs_recovery = 0;
5408 }
5409
5410 if (!test_opt(sb, NO_MBCACHE)) {
5411 sbi->s_ea_block_cache = ext4_xattr_create_cache();
5412 if (!sbi->s_ea_block_cache) {
5413 ext4_msg(sb, KERN_ERR,
5414 "Failed to create ea_block_cache");
5415 err = -EINVAL;
5416 goto failed_mount_wq;
5417 }
5418
5419 if (ext4_has_feature_ea_inode(sb)) {
5420 sbi->s_ea_inode_cache = ext4_xattr_create_cache();
5421 if (!sbi->s_ea_inode_cache) {
5422 ext4_msg(sb, KERN_ERR,
5423 "Failed to create ea_inode_cache");
5424 err = -EINVAL;
5425 goto failed_mount_wq;
5426 }
5427 }
5428 }
5429
5430 /*
5431 * Get the # of file system overhead blocks from the
5432 * superblock if present.
5433 */
5434 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
5435 /* ignore the precalculated value if it is ridiculous */
5436 if (sbi->s_overhead > ext4_blocks_count(es))
5437 sbi->s_overhead = 0;
5438 /*
5439 * If the bigalloc feature is not enabled recalculating the
5440 * overhead doesn't take long, so we might as well just redo
5441 * it to make sure we are using the correct value.
5442 */
5443 if (!ext4_has_feature_bigalloc(sb))
5444 sbi->s_overhead = 0;
5445 if (sbi->s_overhead == 0) {
5446 err = ext4_calculate_overhead(sb);
5447 if (err)
5448 goto failed_mount_wq;
5449 }
5450
5451 /*
5452 * The maximum number of concurrent works can be high and
5453 * concurrency isn't really necessary. Limit it to 1.
5454 */
5455 EXT4_SB(sb)->rsv_conversion_wq =
5456 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
5457 if (!EXT4_SB(sb)->rsv_conversion_wq) {
5458 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
5459 err = -ENOMEM;
5460 goto failed_mount4;
5461 }
5462
5463 /*
5464 * The jbd2_journal_load will have done any necessary log recovery,
5465 * so we can safely mount the rest of the filesystem now.
5466 */
5467
5468 root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
5469 if (IS_ERR(root)) {
5470 ext4_msg(sb, KERN_ERR, "get root inode failed");
5471 err = PTR_ERR(root);
5472 root = NULL;
5473 goto failed_mount4;
5474 }
5475 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
5476 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
5477 iput(root);
5478 err = -EFSCORRUPTED;
5479 goto failed_mount4;
5480 }
5481
5482 generic_set_sb_d_ops(sb);
5483 sb->s_root = d_make_root(root);
5484 if (!sb->s_root) {
5485 ext4_msg(sb, KERN_ERR, "get root dentry failed");
5486 err = -ENOMEM;
5487 goto failed_mount4;
5488 }
5489
5490 err = ext4_setup_super(sb, es, sb_rdonly(sb));
5491 if (err == -EROFS) {
5492 sb->s_flags |= SB_RDONLY;
5493 } else if (err)
5494 goto failed_mount4a;
5495
5496 ext4_set_resv_clusters(sb);
5497
5498 if (test_opt(sb, BLOCK_VALIDITY)) {
5499 err = ext4_setup_system_zone(sb);
5500 if (err) {
5501 ext4_msg(sb, KERN_ERR, "failed to initialize system "
5502 "zone (%d)", err);
5503 goto failed_mount4a;
5504 }
5505 }
5506 ext4_fc_replay_cleanup(sb);
5507
5508 ext4_ext_init(sb);
5509
5510 /*
5511 * Enable optimize_scan if number of groups is > threshold. This can be
5512 * turned off by passing "mb_optimize_scan=0". This can also be
5513 * turned on forcefully by passing "mb_optimize_scan=1".
5514 */
5515 if (!(ctx->spec & EXT4_SPEC_mb_optimize_scan)) {
5516 if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
5517 set_opt2(sb, MB_OPTIMIZE_SCAN);
5518 else
5519 clear_opt2(sb, MB_OPTIMIZE_SCAN);
5520 }
5521
5522 err = ext4_mb_init(sb);
5523 if (err) {
5524 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
5525 err);
5526 goto failed_mount5;
5527 }
5528
5529 /*
5530 * We can only set up the journal commit callback once
5531 * mballoc is initialized
5532 */
5533 if (sbi->s_journal)
5534 sbi->s_journal->j_commit_callback =
5535 ext4_journal_commit_callback;
5536
5537 err = ext4_percpu_param_init(sbi);
5538 if (err)
5539 goto failed_mount6;
5540
5541 if (ext4_has_feature_flex_bg(sb))
5542 if (!ext4_fill_flex_info(sb)) {
5543 ext4_msg(sb, KERN_ERR,
5544 "unable to initialize "
5545 "flex_bg meta info!");
5546 err = -ENOMEM;
5547 goto failed_mount6;
5548 }
5549
5550 err = ext4_register_li_request(sb, first_not_zeroed);
5551 if (err)
5552 goto failed_mount6;
5553
5554 err = ext4_register_sysfs(sb);
5555 if (err)
5556 goto failed_mount7;
5557
5558 err = ext4_init_orphan_info(sb);
5559 if (err)
5560 goto failed_mount8;
5561#ifdef CONFIG_QUOTA
5562 /* Enable quota usage during mount. */
5563 if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
5564 err = ext4_enable_quotas(sb);
5565 if (err)
5566 goto failed_mount9;
5567 }
5568#endif /* CONFIG_QUOTA */
5569
5570 /*
5571 * Save the original bdev mapping's wb_err value which could be
5572 * used to detect the metadata async write error.
5573 */
5574 spin_lock_init(&sbi->s_bdev_wb_lock);
5575 errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
5576 &sbi->s_bdev_wb_err);
5577 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
5578 ext4_orphan_cleanup(sb, es);
5579 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
5580 /*
5581 * Update the checksum after updating free space/inode counters and
5582 * ext4_orphan_cleanup. Otherwise the superblock can have an incorrect
5583 * checksum in the buffer cache until it is written out and
5584 * e2fsprogs programs trying to open a file system immediately
5585 * after it is mounted can fail.
5586 */
5587 ext4_superblock_csum_set(sb);
5588 if (needs_recovery) {
5589 ext4_msg(sb, KERN_INFO, "recovery complete");
5590 err = ext4_mark_recovery_complete(sb, es);
5591 if (err)
5592 goto failed_mount10;
5593 }
5594
5595 if (test_opt(sb, DISCARD) && !bdev_max_discard_sectors(sb->s_bdev))
5596 ext4_msg(sb, KERN_WARNING,
5597 "mounting with \"discard\" option, but the device does not support discard");
5598
5599 if (es->s_error_count)
5600 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
5601
5602 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
5603 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
5604 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
5605 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
5606 atomic_set(&sbi->s_warning_count, 0);
5607 atomic_set(&sbi->s_msg_count, 0);
5608
5609 return 0;
5610
5611failed_mount10:
5612 ext4_quotas_off(sb, EXT4_MAXQUOTAS);
5613failed_mount9: __maybe_unused
5614 ext4_release_orphan_info(sb);
5615failed_mount8:
5616 ext4_unregister_sysfs(sb);
5617 kobject_put(&sbi->s_kobj);
5618failed_mount7:
5619 ext4_unregister_li_request(sb);
5620failed_mount6:
5621 ext4_mb_release(sb);
5622 ext4_flex_groups_free(sbi);
5623 ext4_percpu_param_destroy(sbi);
5624failed_mount5:
5625 ext4_ext_release(sb);
5626 ext4_release_system_zone(sb);
5627failed_mount4a:
5628 dput(sb->s_root);
5629 sb->s_root = NULL;
5630failed_mount4:
5631 ext4_msg(sb, KERN_ERR, "mount failed");
5632 if (EXT4_SB(sb)->rsv_conversion_wq)
5633 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5634failed_mount_wq:
5635 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5636 sbi->s_ea_inode_cache = NULL;
5637
5638 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5639 sbi->s_ea_block_cache = NULL;
5640
5641 if (sbi->s_journal) {
5642 /* flush s_sb_upd_work before journal destroy. */
5643 flush_work(&sbi->s_sb_upd_work);
5644 jbd2_journal_destroy(sbi->s_journal);
5645 sbi->s_journal = NULL;
5646 }
5647failed_mount3a:
5648 ext4_es_unregister_shrinker(sbi);
5649failed_mount3:
5650 /* flush s_sb_upd_work before sbi destroy */
5651 flush_work(&sbi->s_sb_upd_work);
5652 del_timer_sync(&sbi->s_err_report);
5653 ext4_stop_mmpd(sbi);
5654 ext4_group_desc_free(sbi);
5655failed_mount:
5656 if (sbi->s_chksum_driver)
5657 crypto_free_shash(sbi->s_chksum_driver);
5658
5659#if IS_ENABLED(CONFIG_UNICODE)
5660 utf8_unload(sb->s_encoding);
5661#endif
5662
5663#ifdef CONFIG_QUOTA
5664 for (unsigned int i = 0; i < EXT4_MAXQUOTAS; i++)
5665 kfree(get_qf_name(sb, sbi, i));
5666#endif
5667 fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5668 brelse(sbi->s_sbh);
5669 if (sbi->s_journal_bdev_file) {
5670 invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
5671 bdev_fput(sbi->s_journal_bdev_file);
5672 }
5673out_fail:
5674 invalidate_bdev(sb->s_bdev);
5675 sb->s_fs_info = NULL;
5676 return err;
5677}
5678
5679static int ext4_fill_super(struct super_block *sb, struct fs_context *fc)
5680{
5681 struct ext4_fs_context *ctx = fc->fs_private;
5682 struct ext4_sb_info *sbi;
5683 const char *descr;
5684 int ret;
5685
5686 sbi = ext4_alloc_sbi(sb);
5687 if (!sbi)
5688 return -ENOMEM;
5689
5690 fc->s_fs_info = sbi;
5691
5692 /* Cleanup superblock name */
5693 strreplace(sb->s_id, '/', '!');
5694
5695 sbi->s_sb_block = 1; /* Default super block location */
5696 if (ctx->spec & EXT4_SPEC_s_sb_block)
5697 sbi->s_sb_block = ctx->s_sb_block;
5698
5699 ret = __ext4_fill_super(fc, sb);
5700 if (ret < 0)
5701 goto free_sbi;
5702
5703 if (sbi->s_journal) {
5704 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
5705 descr = " journalled data mode";
5706 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
5707 descr = " ordered data mode";
5708 else
5709 descr = " writeback data mode";
5710 } else
5711 descr = "out journal";
5712
5713 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
5714 ext4_msg(sb, KERN_INFO, "mounted filesystem %pU %s with%s. "
5715 "Quota mode: %s.", &sb->s_uuid,
5716 sb_rdonly(sb) ? "ro" : "r/w", descr,
5717 ext4_quota_mode(sb));
5718
5719 /* Update the s_overhead_clusters if necessary */
5720 ext4_update_overhead(sb, false);
5721 return 0;
5722
5723free_sbi:
5724 ext4_free_sbi(sbi);
5725 fc->s_fs_info = NULL;
5726 return ret;
5727}
5728
5729static int ext4_get_tree(struct fs_context *fc)
5730{
5731 return get_tree_bdev(fc, ext4_fill_super);
5732}
5733
5734/*
5735 * Setup any per-fs journal parameters now. We'll do this both on
5736 * initial mount, once the journal has been initialised but before we've
5737 * done any recovery; and again on any subsequent remount.
5738 */
5739static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5740{
5741 struct ext4_sb_info *sbi = EXT4_SB(sb);
5742
5743 journal->j_commit_interval = sbi->s_commit_interval;
5744 journal->j_min_batch_time = sbi->s_min_batch_time;
5745 journal->j_max_batch_time = sbi->s_max_batch_time;
5746 ext4_fc_init(sb, journal);
5747
5748 write_lock(&journal->j_state_lock);
5749 if (test_opt(sb, BARRIER))
5750 journal->j_flags |= JBD2_BARRIER;
5751 else
5752 journal->j_flags &= ~JBD2_BARRIER;
5753 if (test_opt(sb, DATA_ERR_ABORT))
5754 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5755 else
5756 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5757 /*
5758 * Always enable journal cycle record option, letting the journal
5759 * records log transactions continuously between each mount.
5760 */
5761 journal->j_flags |= JBD2_CYCLE_RECORD;
5762 write_unlock(&journal->j_state_lock);
5763}
5764
5765static struct inode *ext4_get_journal_inode(struct super_block *sb,
5766 unsigned int journal_inum)
5767{
5768 struct inode *journal_inode;
5769
5770 /*
5771 * Test for the existence of a valid inode on disk. Bad things
5772 * happen if we iget() an unused inode, as the subsequent iput()
5773 * will try to delete it.
5774 */
5775 journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5776 if (IS_ERR(journal_inode)) {
5777 ext4_msg(sb, KERN_ERR, "no journal found");
5778 return ERR_CAST(journal_inode);
5779 }
5780 if (!journal_inode->i_nlink) {
5781 make_bad_inode(journal_inode);
5782 iput(journal_inode);
5783 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5784 return ERR_PTR(-EFSCORRUPTED);
5785 }
5786 if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
5787 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5788 iput(journal_inode);
5789 return ERR_PTR(-EFSCORRUPTED);
5790 }
5791
5792 ext4_debug("Journal inode found at %p: %lld bytes\n",
5793 journal_inode, journal_inode->i_size);
5794 return journal_inode;
5795}
5796
5797static int ext4_journal_bmap(journal_t *journal, sector_t *block)
5798{
5799 struct ext4_map_blocks map;
5800 int ret;
5801
5802 if (journal->j_inode == NULL)
5803 return 0;
5804
5805 map.m_lblk = *block;
5806 map.m_len = 1;
5807 ret = ext4_map_blocks(NULL, journal->j_inode, &map, 0);
5808 if (ret <= 0) {
5809 ext4_msg(journal->j_inode->i_sb, KERN_CRIT,
5810 "journal bmap failed: block %llu ret %d\n",
5811 *block, ret);
5812 jbd2_journal_abort(journal, ret ? ret : -EIO);
5813 return ret;
5814 }
5815 *block = map.m_pblk;
5816 return 0;
5817}
5818
5819static journal_t *ext4_open_inode_journal(struct super_block *sb,
5820 unsigned int journal_inum)
5821{
5822 struct inode *journal_inode;
5823 journal_t *journal;
5824
5825 journal_inode = ext4_get_journal_inode(sb, journal_inum);
5826 if (IS_ERR(journal_inode))
5827 return ERR_CAST(journal_inode);
5828
5829 journal = jbd2_journal_init_inode(journal_inode);
5830 if (IS_ERR(journal)) {
5831 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5832 iput(journal_inode);
5833 return ERR_CAST(journal);
5834 }
5835 journal->j_private = sb;
5836 journal->j_bmap = ext4_journal_bmap;
5837 ext4_init_journal_params(sb, journal);
5838 return journal;
5839}
5840
5841static struct file *ext4_get_journal_blkdev(struct super_block *sb,
5842 dev_t j_dev, ext4_fsblk_t *j_start,
5843 ext4_fsblk_t *j_len)
5844{
5845 struct buffer_head *bh;
5846 struct block_device *bdev;
5847 struct file *bdev_file;
5848 int hblock, blocksize;
5849 ext4_fsblk_t sb_block;
5850 unsigned long offset;
5851 struct ext4_super_block *es;
5852 int errno;
5853
5854 bdev_file = bdev_file_open_by_dev(j_dev,
5855 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES,
5856 sb, &fs_holder_ops);
5857 if (IS_ERR(bdev_file)) {
5858 ext4_msg(sb, KERN_ERR,
5859 "failed to open journal device unknown-block(%u,%u) %ld",
5860 MAJOR(j_dev), MINOR(j_dev), PTR_ERR(bdev_file));
5861 return bdev_file;
5862 }
5863
5864 bdev = file_bdev(bdev_file);
5865 blocksize = sb->s_blocksize;
5866 hblock = bdev_logical_block_size(bdev);
5867 if (blocksize < hblock) {
5868 ext4_msg(sb, KERN_ERR,
5869 "blocksize too small for journal device");
5870 errno = -EINVAL;
5871 goto out_bdev;
5872 }
5873
5874 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5875 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5876 set_blocksize(bdev, blocksize);
5877 bh = __bread(bdev, sb_block, blocksize);
5878 if (!bh) {
5879 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5880 "external journal");
5881 errno = -EINVAL;
5882 goto out_bdev;
5883 }
5884
5885 es = (struct ext4_super_block *) (bh->b_data + offset);
5886 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5887 !(le32_to_cpu(es->s_feature_incompat) &
5888 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5889 ext4_msg(sb, KERN_ERR, "external journal has bad superblock");
5890 errno = -EFSCORRUPTED;
5891 goto out_bh;
5892 }
5893
5894 if ((le32_to_cpu(es->s_feature_ro_compat) &
5895 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5896 es->s_checksum != ext4_superblock_csum(sb, es)) {
5897 ext4_msg(sb, KERN_ERR, "external journal has corrupt superblock");
5898 errno = -EFSCORRUPTED;
5899 goto out_bh;
5900 }
5901
5902 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5903 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5904 errno = -EFSCORRUPTED;
5905 goto out_bh;
5906 }
5907
5908 *j_start = sb_block + 1;
5909 *j_len = ext4_blocks_count(es);
5910 brelse(bh);
5911 return bdev_file;
5912
5913out_bh:
5914 brelse(bh);
5915out_bdev:
5916 bdev_fput(bdev_file);
5917 return ERR_PTR(errno);
5918}
5919
5920static journal_t *ext4_open_dev_journal(struct super_block *sb,
5921 dev_t j_dev)
5922{
5923 journal_t *journal;
5924 ext4_fsblk_t j_start;
5925 ext4_fsblk_t j_len;
5926 struct file *bdev_file;
5927 int errno = 0;
5928
5929 bdev_file = ext4_get_journal_blkdev(sb, j_dev, &j_start, &j_len);
5930 if (IS_ERR(bdev_file))
5931 return ERR_CAST(bdev_file);
5932
5933 journal = jbd2_journal_init_dev(file_bdev(bdev_file), sb->s_bdev, j_start,
5934 j_len, sb->s_blocksize);
5935 if (IS_ERR(journal)) {
5936 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5937 errno = PTR_ERR(journal);
5938 goto out_bdev;
5939 }
5940 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5941 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5942 "user (unsupported) - %d",
5943 be32_to_cpu(journal->j_superblock->s_nr_users));
5944 errno = -EINVAL;
5945 goto out_journal;
5946 }
5947 journal->j_private = sb;
5948 EXT4_SB(sb)->s_journal_bdev_file = bdev_file;
5949 ext4_init_journal_params(sb, journal);
5950 return journal;
5951
5952out_journal:
5953 jbd2_journal_destroy(journal);
5954out_bdev:
5955 bdev_fput(bdev_file);
5956 return ERR_PTR(errno);
5957}
5958
5959static int ext4_load_journal(struct super_block *sb,
5960 struct ext4_super_block *es,
5961 unsigned long journal_devnum)
5962{
5963 journal_t *journal;
5964 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5965 dev_t journal_dev;
5966 int err = 0;
5967 int really_read_only;
5968 int journal_dev_ro;
5969
5970 if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5971 return -EFSCORRUPTED;
5972
5973 if (journal_devnum &&
5974 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5975 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5976 "numbers have changed");
5977 journal_dev = new_decode_dev(journal_devnum);
5978 } else
5979 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5980
5981 if (journal_inum && journal_dev) {
5982 ext4_msg(sb, KERN_ERR,
5983 "filesystem has both journal inode and journal device!");
5984 return -EINVAL;
5985 }
5986
5987 if (journal_inum) {
5988 journal = ext4_open_inode_journal(sb, journal_inum);
5989 if (IS_ERR(journal))
5990 return PTR_ERR(journal);
5991 } else {
5992 journal = ext4_open_dev_journal(sb, journal_dev);
5993 if (IS_ERR(journal))
5994 return PTR_ERR(journal);
5995 }
5996
5997 journal_dev_ro = bdev_read_only(journal->j_dev);
5998 really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5999
6000 if (journal_dev_ro && !sb_rdonly(sb)) {
6001 ext4_msg(sb, KERN_ERR,
6002 "journal device read-only, try mounting with '-o ro'");
6003 err = -EROFS;
6004 goto err_out;
6005 }
6006
6007 /*
6008 * Are we loading a blank journal or performing recovery after a
6009 * crash? For recovery, we need to check in advance whether we
6010 * can get read-write access to the device.
6011 */
6012 if (ext4_has_feature_journal_needs_recovery(sb)) {
6013 if (sb_rdonly(sb)) {
6014 ext4_msg(sb, KERN_INFO, "INFO: recovery "
6015 "required on readonly filesystem");
6016 if (really_read_only) {
6017 ext4_msg(sb, KERN_ERR, "write access "
6018 "unavailable, cannot proceed "
6019 "(try mounting with noload)");
6020 err = -EROFS;
6021 goto err_out;
6022 }
6023 ext4_msg(sb, KERN_INFO, "write access will "
6024 "be enabled during recovery");
6025 }
6026 }
6027
6028 if (!(journal->j_flags & JBD2_BARRIER))
6029 ext4_msg(sb, KERN_INFO, "barriers disabled");
6030
6031 if (!ext4_has_feature_journal_needs_recovery(sb))
6032 err = jbd2_journal_wipe(journal, !really_read_only);
6033 if (!err) {
6034 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
6035 __le16 orig_state;
6036 bool changed = false;
6037
6038 if (save)
6039 memcpy(save, ((char *) es) +
6040 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
6041 err = jbd2_journal_load(journal);
6042 if (save && memcmp(((char *) es) + EXT4_S_ERR_START,
6043 save, EXT4_S_ERR_LEN)) {
6044 memcpy(((char *) es) + EXT4_S_ERR_START,
6045 save, EXT4_S_ERR_LEN);
6046 changed = true;
6047 }
6048 kfree(save);
6049 orig_state = es->s_state;
6050 es->s_state |= cpu_to_le16(EXT4_SB(sb)->s_mount_state &
6051 EXT4_ERROR_FS);
6052 if (orig_state != es->s_state)
6053 changed = true;
6054 /* Write out restored error information to the superblock */
6055 if (changed && !really_read_only) {
6056 int err2;
6057 err2 = ext4_commit_super(sb);
6058 err = err ? : err2;
6059 }
6060 }
6061
6062 if (err) {
6063 ext4_msg(sb, KERN_ERR, "error loading journal");
6064 goto err_out;
6065 }
6066
6067 EXT4_SB(sb)->s_journal = journal;
6068 err = ext4_clear_journal_err(sb, es);
6069 if (err) {
6070 EXT4_SB(sb)->s_journal = NULL;
6071 jbd2_journal_destroy(journal);
6072 return err;
6073 }
6074
6075 if (!really_read_only && journal_devnum &&
6076 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
6077 es->s_journal_dev = cpu_to_le32(journal_devnum);
6078 ext4_commit_super(sb);
6079 }
6080 if (!really_read_only && journal_inum &&
6081 journal_inum != le32_to_cpu(es->s_journal_inum)) {
6082 es->s_journal_inum = cpu_to_le32(journal_inum);
6083 ext4_commit_super(sb);
6084 }
6085
6086 return 0;
6087
6088err_out:
6089 jbd2_journal_destroy(journal);
6090 return err;
6091}
6092
6093/* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
6094static void ext4_update_super(struct super_block *sb)
6095{
6096 struct ext4_sb_info *sbi = EXT4_SB(sb);
6097 struct ext4_super_block *es = sbi->s_es;
6098 struct buffer_head *sbh = sbi->s_sbh;
6099
6100 lock_buffer(sbh);
6101 /*
6102 * If the file system is mounted read-only, don't update the
6103 * superblock write time. This avoids updating the superblock
6104 * write time when we are mounting the root file system
6105 * read/only but we need to replay the journal; at that point,
6106 * for people who are east of GMT and who make their clock
6107 * tick in localtime for Windows bug-for-bug compatibility,
6108 * the clock is set in the future, and this will cause e2fsck
6109 * to complain and force a full file system check.
6110 */
6111 if (!sb_rdonly(sb))
6112 ext4_update_tstamp(es, s_wtime);
6113 es->s_kbytes_written =
6114 cpu_to_le64(sbi->s_kbytes_written +
6115 ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
6116 sbi->s_sectors_written_start) >> 1));
6117 if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
6118 ext4_free_blocks_count_set(es,
6119 EXT4_C2B(sbi, percpu_counter_sum_positive(
6120 &sbi->s_freeclusters_counter)));
6121 if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
6122 es->s_free_inodes_count =
6123 cpu_to_le32(percpu_counter_sum_positive(
6124 &sbi->s_freeinodes_counter));
6125 /* Copy error information to the on-disk superblock */
6126 spin_lock(&sbi->s_error_lock);
6127 if (sbi->s_add_error_count > 0) {
6128 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6129 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
6130 __ext4_update_tstamp(&es->s_first_error_time,
6131 &es->s_first_error_time_hi,
6132 sbi->s_first_error_time);
6133 strncpy(es->s_first_error_func, sbi->s_first_error_func,
6134 sizeof(es->s_first_error_func));
6135 es->s_first_error_line =
6136 cpu_to_le32(sbi->s_first_error_line);
6137 es->s_first_error_ino =
6138 cpu_to_le32(sbi->s_first_error_ino);
6139 es->s_first_error_block =
6140 cpu_to_le64(sbi->s_first_error_block);
6141 es->s_first_error_errcode =
6142 ext4_errno_to_code(sbi->s_first_error_code);
6143 }
6144 __ext4_update_tstamp(&es->s_last_error_time,
6145 &es->s_last_error_time_hi,
6146 sbi->s_last_error_time);
6147 strncpy(es->s_last_error_func, sbi->s_last_error_func,
6148 sizeof(es->s_last_error_func));
6149 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
6150 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
6151 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
6152 es->s_last_error_errcode =
6153 ext4_errno_to_code(sbi->s_last_error_code);
6154 /*
6155 * Start the daily error reporting function if it hasn't been
6156 * started already
6157 */
6158 if (!es->s_error_count)
6159 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
6160 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
6161 sbi->s_add_error_count = 0;
6162 }
6163 spin_unlock(&sbi->s_error_lock);
6164
6165 ext4_superblock_csum_set(sb);
6166 unlock_buffer(sbh);
6167}
6168
6169static int ext4_commit_super(struct super_block *sb)
6170{
6171 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
6172
6173 if (!sbh)
6174 return -EINVAL;
6175 if (block_device_ejected(sb))
6176 return -ENODEV;
6177
6178 ext4_update_super(sb);
6179
6180 lock_buffer(sbh);
6181 /* Buffer got discarded which means block device got invalidated */
6182 if (!buffer_mapped(sbh)) {
6183 unlock_buffer(sbh);
6184 return -EIO;
6185 }
6186
6187 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
6188 /*
6189 * Oh, dear. A previous attempt to write the
6190 * superblock failed. This could happen because the
6191 * USB device was yanked out. Or it could happen to
6192 * be a transient write error and maybe the block will
6193 * be remapped. Nothing we can do but to retry the
6194 * write and hope for the best.
6195 */
6196 ext4_msg(sb, KERN_ERR, "previous I/O error to "
6197 "superblock detected");
6198 clear_buffer_write_io_error(sbh);
6199 set_buffer_uptodate(sbh);
6200 }
6201 get_bh(sbh);
6202 /* Clear potential dirty bit if it was journalled update */
6203 clear_buffer_dirty(sbh);
6204 sbh->b_end_io = end_buffer_write_sync;
6205 submit_bh(REQ_OP_WRITE | REQ_SYNC |
6206 (test_opt(sb, BARRIER) ? REQ_FUA : 0), sbh);
6207 wait_on_buffer(sbh);
6208 if (buffer_write_io_error(sbh)) {
6209 ext4_msg(sb, KERN_ERR, "I/O error while writing "
6210 "superblock");
6211 clear_buffer_write_io_error(sbh);
6212 set_buffer_uptodate(sbh);
6213 return -EIO;
6214 }
6215 return 0;
6216}
6217
6218/*
6219 * Have we just finished recovery? If so, and if we are mounting (or
6220 * remounting) the filesystem readonly, then we will end up with a
6221 * consistent fs on disk. Record that fact.
6222 */
6223static int ext4_mark_recovery_complete(struct super_block *sb,
6224 struct ext4_super_block *es)
6225{
6226 int err;
6227 journal_t *journal = EXT4_SB(sb)->s_journal;
6228
6229 if (!ext4_has_feature_journal(sb)) {
6230 if (journal != NULL) {
6231 ext4_error(sb, "Journal got removed while the fs was "
6232 "mounted!");
6233 return -EFSCORRUPTED;
6234 }
6235 return 0;
6236 }
6237 jbd2_journal_lock_updates(journal);
6238 err = jbd2_journal_flush(journal, 0);
6239 if (err < 0)
6240 goto out;
6241
6242 if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
6243 ext4_has_feature_orphan_present(sb))) {
6244 if (!ext4_orphan_file_empty(sb)) {
6245 ext4_error(sb, "Orphan file not empty on read-only fs.");
6246 err = -EFSCORRUPTED;
6247 goto out;
6248 }
6249 ext4_clear_feature_journal_needs_recovery(sb);
6250 ext4_clear_feature_orphan_present(sb);
6251 ext4_commit_super(sb);
6252 }
6253out:
6254 jbd2_journal_unlock_updates(journal);
6255 return err;
6256}
6257
6258/*
6259 * If we are mounting (or read-write remounting) a filesystem whose journal
6260 * has recorded an error from a previous lifetime, move that error to the
6261 * main filesystem now.
6262 */
6263static int ext4_clear_journal_err(struct super_block *sb,
6264 struct ext4_super_block *es)
6265{
6266 journal_t *journal;
6267 int j_errno;
6268 const char *errstr;
6269
6270 if (!ext4_has_feature_journal(sb)) {
6271 ext4_error(sb, "Journal got removed while the fs was mounted!");
6272 return -EFSCORRUPTED;
6273 }
6274
6275 journal = EXT4_SB(sb)->s_journal;
6276
6277 /*
6278 * Now check for any error status which may have been recorded in the
6279 * journal by a prior ext4_error() or ext4_abort()
6280 */
6281
6282 j_errno = jbd2_journal_errno(journal);
6283 if (j_errno) {
6284 char nbuf[16];
6285
6286 errstr = ext4_decode_error(sb, j_errno, nbuf);
6287 ext4_warning(sb, "Filesystem error recorded "
6288 "from previous mount: %s", errstr);
6289
6290 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
6291 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
6292 j_errno = ext4_commit_super(sb);
6293 if (j_errno)
6294 return j_errno;
6295 ext4_warning(sb, "Marked fs in need of filesystem check.");
6296
6297 jbd2_journal_clear_err(journal);
6298 jbd2_journal_update_sb_errno(journal);
6299 }
6300 return 0;
6301}
6302
6303/*
6304 * Force the running and committing transactions to commit,
6305 * and wait on the commit.
6306 */
6307int ext4_force_commit(struct super_block *sb)
6308{
6309 return ext4_journal_force_commit(EXT4_SB(sb)->s_journal);
6310}
6311
6312static int ext4_sync_fs(struct super_block *sb, int wait)
6313{
6314 int ret = 0;
6315 tid_t target;
6316 bool needs_barrier = false;
6317 struct ext4_sb_info *sbi = EXT4_SB(sb);
6318
6319 if (unlikely(ext4_forced_shutdown(sb)))
6320 return 0;
6321
6322 trace_ext4_sync_fs(sb, wait);
6323 flush_workqueue(sbi->rsv_conversion_wq);
6324 /*
6325 * Writeback quota in non-journalled quota case - journalled quota has
6326 * no dirty dquots
6327 */
6328 dquot_writeback_dquots(sb, -1);
6329 /*
6330 * Data writeback is possible w/o journal transaction, so barrier must
6331 * being sent at the end of the function. But we can skip it if
6332 * transaction_commit will do it for us.
6333 */
6334 if (sbi->s_journal) {
6335 target = jbd2_get_latest_transaction(sbi->s_journal);
6336 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
6337 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
6338 needs_barrier = true;
6339
6340 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
6341 if (wait)
6342 ret = jbd2_log_wait_commit(sbi->s_journal,
6343 target);
6344 }
6345 } else if (wait && test_opt(sb, BARRIER))
6346 needs_barrier = true;
6347 if (needs_barrier) {
6348 int err;
6349 err = blkdev_issue_flush(sb->s_bdev);
6350 if (!ret)
6351 ret = err;
6352 }
6353
6354 return ret;
6355}
6356
6357/*
6358 * LVM calls this function before a (read-only) snapshot is created. This
6359 * gives us a chance to flush the journal completely and mark the fs clean.
6360 *
6361 * Note that only this function cannot bring a filesystem to be in a clean
6362 * state independently. It relies on upper layer to stop all data & metadata
6363 * modifications.
6364 */
6365static int ext4_freeze(struct super_block *sb)
6366{
6367 int error = 0;
6368 journal_t *journal = EXT4_SB(sb)->s_journal;
6369
6370 if (journal) {
6371 /* Now we set up the journal barrier. */
6372 jbd2_journal_lock_updates(journal);
6373
6374 /*
6375 * Don't clear the needs_recovery flag if we failed to
6376 * flush the journal.
6377 */
6378 error = jbd2_journal_flush(journal, 0);
6379 if (error < 0)
6380 goto out;
6381
6382 /* Journal blocked and flushed, clear needs_recovery flag. */
6383 ext4_clear_feature_journal_needs_recovery(sb);
6384 if (ext4_orphan_file_empty(sb))
6385 ext4_clear_feature_orphan_present(sb);
6386 }
6387
6388 error = ext4_commit_super(sb);
6389out:
6390 if (journal)
6391 /* we rely on upper layer to stop further updates */
6392 jbd2_journal_unlock_updates(journal);
6393 return error;
6394}
6395
6396/*
6397 * Called by LVM after the snapshot is done. We need to reset the RECOVER
6398 * flag here, even though the filesystem is not technically dirty yet.
6399 */
6400static int ext4_unfreeze(struct super_block *sb)
6401{
6402 if (ext4_forced_shutdown(sb))
6403 return 0;
6404
6405 if (EXT4_SB(sb)->s_journal) {
6406 /* Reset the needs_recovery flag before the fs is unlocked. */
6407 ext4_set_feature_journal_needs_recovery(sb);
6408 if (ext4_has_feature_orphan_file(sb))
6409 ext4_set_feature_orphan_present(sb);
6410 }
6411
6412 ext4_commit_super(sb);
6413 return 0;
6414}
6415
6416/*
6417 * Structure to save mount options for ext4_remount's benefit
6418 */
6419struct ext4_mount_options {
6420 unsigned long s_mount_opt;
6421 unsigned long s_mount_opt2;
6422 kuid_t s_resuid;
6423 kgid_t s_resgid;
6424 unsigned long s_commit_interval;
6425 u32 s_min_batch_time, s_max_batch_time;
6426#ifdef CONFIG_QUOTA
6427 int s_jquota_fmt;
6428 char *s_qf_names[EXT4_MAXQUOTAS];
6429#endif
6430};
6431
6432static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
6433{
6434 struct ext4_fs_context *ctx = fc->fs_private;
6435 struct ext4_super_block *es;
6436 struct ext4_sb_info *sbi = EXT4_SB(sb);
6437 unsigned long old_sb_flags;
6438 struct ext4_mount_options old_opts;
6439 ext4_group_t g;
6440 int err = 0;
6441 int alloc_ctx;
6442#ifdef CONFIG_QUOTA
6443 int enable_quota = 0;
6444 int i, j;
6445 char *to_free[EXT4_MAXQUOTAS];
6446#endif
6447
6448
6449 /* Store the original options */
6450 old_sb_flags = sb->s_flags;
6451 old_opts.s_mount_opt = sbi->s_mount_opt;
6452 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
6453 old_opts.s_resuid = sbi->s_resuid;
6454 old_opts.s_resgid = sbi->s_resgid;
6455 old_opts.s_commit_interval = sbi->s_commit_interval;
6456 old_opts.s_min_batch_time = sbi->s_min_batch_time;
6457 old_opts.s_max_batch_time = sbi->s_max_batch_time;
6458#ifdef CONFIG_QUOTA
6459 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
6460 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6461 if (sbi->s_qf_names[i]) {
6462 char *qf_name = get_qf_name(sb, sbi, i);
6463
6464 old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
6465 if (!old_opts.s_qf_names[i]) {
6466 for (j = 0; j < i; j++)
6467 kfree(old_opts.s_qf_names[j]);
6468 return -ENOMEM;
6469 }
6470 } else
6471 old_opts.s_qf_names[i] = NULL;
6472#endif
6473 if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO)) {
6474 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
6475 ctx->journal_ioprio =
6476 sbi->s_journal->j_task->io_context->ioprio;
6477 else
6478 ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
6479
6480 }
6481
6482 /*
6483 * Changing the DIOREAD_NOLOCK or DELALLOC mount options may cause
6484 * two calls to ext4_should_dioread_nolock() to return inconsistent
6485 * values, triggering WARN_ON in ext4_add_complete_io(). we grab
6486 * here s_writepages_rwsem to avoid race between writepages ops and
6487 * remount.
6488 */
6489 alloc_ctx = ext4_writepages_down_write(sb);
6490 ext4_apply_options(fc, sb);
6491 ext4_writepages_up_write(sb, alloc_ctx);
6492
6493 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
6494 test_opt(sb, JOURNAL_CHECKSUM)) {
6495 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
6496 "during remount not supported; ignoring");
6497 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
6498 }
6499
6500 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
6501 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
6502 ext4_msg(sb, KERN_ERR, "can't mount with "
6503 "both data=journal and delalloc");
6504 err = -EINVAL;
6505 goto restore_opts;
6506 }
6507 if (test_opt(sb, DIOREAD_NOLOCK)) {
6508 ext4_msg(sb, KERN_ERR, "can't mount with "
6509 "both data=journal and dioread_nolock");
6510 err = -EINVAL;
6511 goto restore_opts;
6512 }
6513 } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
6514 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
6515 ext4_msg(sb, KERN_ERR, "can't mount with "
6516 "journal_async_commit in data=ordered mode");
6517 err = -EINVAL;
6518 goto restore_opts;
6519 }
6520 }
6521
6522 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
6523 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
6524 err = -EINVAL;
6525 goto restore_opts;
6526 }
6527
6528 if (test_opt2(sb, ABORT))
6529 ext4_abort(sb, ESHUTDOWN, "Abort forced by user");
6530
6531 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
6532 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
6533
6534 es = sbi->s_es;
6535
6536 if (sbi->s_journal) {
6537 ext4_init_journal_params(sb, sbi->s_journal);
6538 set_task_ioprio(sbi->s_journal->j_task, ctx->journal_ioprio);
6539 }
6540
6541 /* Flush outstanding errors before changing fs state */
6542 flush_work(&sbi->s_sb_upd_work);
6543
6544 if ((bool)(fc->sb_flags & SB_RDONLY) != sb_rdonly(sb)) {
6545 if (ext4_forced_shutdown(sb)) {
6546 err = -EROFS;
6547 goto restore_opts;
6548 }
6549
6550 if (fc->sb_flags & SB_RDONLY) {
6551 err = sync_filesystem(sb);
6552 if (err < 0)
6553 goto restore_opts;
6554 err = dquot_suspend(sb, -1);
6555 if (err < 0)
6556 goto restore_opts;
6557
6558 /*
6559 * First of all, the unconditional stuff we have to do
6560 * to disable replay of the journal when we next remount
6561 */
6562 sb->s_flags |= SB_RDONLY;
6563
6564 /*
6565 * OK, test if we are remounting a valid rw partition
6566 * readonly, and if so set the rdonly flag and then
6567 * mark the partition as valid again.
6568 */
6569 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
6570 (sbi->s_mount_state & EXT4_VALID_FS))
6571 es->s_state = cpu_to_le16(sbi->s_mount_state);
6572
6573 if (sbi->s_journal) {
6574 /*
6575 * We let remount-ro finish even if marking fs
6576 * as clean failed...
6577 */
6578 ext4_mark_recovery_complete(sb, es);
6579 }
6580 } else {
6581 /* Make sure we can mount this feature set readwrite */
6582 if (ext4_has_feature_readonly(sb) ||
6583 !ext4_feature_set_ok(sb, 0)) {
6584 err = -EROFS;
6585 goto restore_opts;
6586 }
6587 /*
6588 * Make sure the group descriptor checksums
6589 * are sane. If they aren't, refuse to remount r/w.
6590 */
6591 for (g = 0; g < sbi->s_groups_count; g++) {
6592 struct ext4_group_desc *gdp =
6593 ext4_get_group_desc(sb, g, NULL);
6594
6595 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
6596 ext4_msg(sb, KERN_ERR,
6597 "ext4_remount: Checksum for group %u failed (%u!=%u)",
6598 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
6599 le16_to_cpu(gdp->bg_checksum));
6600 err = -EFSBADCRC;
6601 goto restore_opts;
6602 }
6603 }
6604
6605 /*
6606 * If we have an unprocessed orphan list hanging
6607 * around from a previously readonly bdev mount,
6608 * require a full umount/remount for now.
6609 */
6610 if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
6611 ext4_msg(sb, KERN_WARNING, "Couldn't "
6612 "remount RDWR because of unprocessed "
6613 "orphan inode list. Please "
6614 "umount/remount instead");
6615 err = -EINVAL;
6616 goto restore_opts;
6617 }
6618
6619 /*
6620 * Mounting a RDONLY partition read-write, so reread
6621 * and store the current valid flag. (It may have
6622 * been changed by e2fsck since we originally mounted
6623 * the partition.)
6624 */
6625 if (sbi->s_journal) {
6626 err = ext4_clear_journal_err(sb, es);
6627 if (err)
6628 goto restore_opts;
6629 }
6630 sbi->s_mount_state = (le16_to_cpu(es->s_state) &
6631 ~EXT4_FC_REPLAY);
6632
6633 err = ext4_setup_super(sb, es, 0);
6634 if (err)
6635 goto restore_opts;
6636
6637 sb->s_flags &= ~SB_RDONLY;
6638 if (ext4_has_feature_mmp(sb)) {
6639 err = ext4_multi_mount_protect(sb,
6640 le64_to_cpu(es->s_mmp_block));
6641 if (err)
6642 goto restore_opts;
6643 }
6644#ifdef CONFIG_QUOTA
6645 enable_quota = 1;
6646#endif
6647 }
6648 }
6649
6650 /*
6651 * Handle creation of system zone data early because it can fail.
6652 * Releasing of existing data is done when we are sure remount will
6653 * succeed.
6654 */
6655 if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
6656 err = ext4_setup_system_zone(sb);
6657 if (err)
6658 goto restore_opts;
6659 }
6660
6661 if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
6662 err = ext4_commit_super(sb);
6663 if (err)
6664 goto restore_opts;
6665 }
6666
6667#ifdef CONFIG_QUOTA
6668 if (enable_quota) {
6669 if (sb_any_quota_suspended(sb))
6670 dquot_resume(sb, -1);
6671 else if (ext4_has_feature_quota(sb)) {
6672 err = ext4_enable_quotas(sb);
6673 if (err)
6674 goto restore_opts;
6675 }
6676 }
6677 /* Release old quota file names */
6678 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6679 kfree(old_opts.s_qf_names[i]);
6680#endif
6681 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6682 ext4_release_system_zone(sb);
6683
6684 /*
6685 * Reinitialize lazy itable initialization thread based on
6686 * current settings
6687 */
6688 if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
6689 ext4_unregister_li_request(sb);
6690 else {
6691 ext4_group_t first_not_zeroed;
6692 first_not_zeroed = ext4_has_uninit_itable(sb);
6693 ext4_register_li_request(sb, first_not_zeroed);
6694 }
6695
6696 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6697 ext4_stop_mmpd(sbi);
6698
6699 return 0;
6700
6701restore_opts:
6702 /*
6703 * If there was a failing r/w to ro transition, we may need to
6704 * re-enable quota
6705 */
6706 if (sb_rdonly(sb) && !(old_sb_flags & SB_RDONLY) &&
6707 sb_any_quota_suspended(sb))
6708 dquot_resume(sb, -1);
6709
6710 alloc_ctx = ext4_writepages_down_write(sb);
6711 sb->s_flags = old_sb_flags;
6712 sbi->s_mount_opt = old_opts.s_mount_opt;
6713 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6714 sbi->s_resuid = old_opts.s_resuid;
6715 sbi->s_resgid = old_opts.s_resgid;
6716 sbi->s_commit_interval = old_opts.s_commit_interval;
6717 sbi->s_min_batch_time = old_opts.s_min_batch_time;
6718 sbi->s_max_batch_time = old_opts.s_max_batch_time;
6719 ext4_writepages_up_write(sb, alloc_ctx);
6720
6721 if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6722 ext4_release_system_zone(sb);
6723#ifdef CONFIG_QUOTA
6724 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6725 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6726 to_free[i] = get_qf_name(sb, sbi, i);
6727 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6728 }
6729 synchronize_rcu();
6730 for (i = 0; i < EXT4_MAXQUOTAS; i++)
6731 kfree(to_free[i]);
6732#endif
6733 if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6734 ext4_stop_mmpd(sbi);
6735 return err;
6736}
6737
6738static int ext4_reconfigure(struct fs_context *fc)
6739{
6740 struct super_block *sb = fc->root->d_sb;
6741 int ret;
6742
6743 fc->s_fs_info = EXT4_SB(sb);
6744
6745 ret = ext4_check_opt_consistency(fc, sb);
6746 if (ret < 0)
6747 return ret;
6748
6749 ret = __ext4_remount(fc, sb);
6750 if (ret < 0)
6751 return ret;
6752
6753 ext4_msg(sb, KERN_INFO, "re-mounted %pU %s. Quota mode: %s.",
6754 &sb->s_uuid, sb_rdonly(sb) ? "ro" : "r/w",
6755 ext4_quota_mode(sb));
6756
6757 return 0;
6758}
6759
6760#ifdef CONFIG_QUOTA
6761static int ext4_statfs_project(struct super_block *sb,
6762 kprojid_t projid, struct kstatfs *buf)
6763{
6764 struct kqid qid;
6765 struct dquot *dquot;
6766 u64 limit;
6767 u64 curblock;
6768
6769 qid = make_kqid_projid(projid);
6770 dquot = dqget(sb, qid);
6771 if (IS_ERR(dquot))
6772 return PTR_ERR(dquot);
6773 spin_lock(&dquot->dq_dqb_lock);
6774
6775 limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6776 dquot->dq_dqb.dqb_bhardlimit);
6777 limit >>= sb->s_blocksize_bits;
6778
6779 if (limit && buf->f_blocks > limit) {
6780 curblock = (dquot->dq_dqb.dqb_curspace +
6781 dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6782 buf->f_blocks = limit;
6783 buf->f_bfree = buf->f_bavail =
6784 (buf->f_blocks > curblock) ?
6785 (buf->f_blocks - curblock) : 0;
6786 }
6787
6788 limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6789 dquot->dq_dqb.dqb_ihardlimit);
6790 if (limit && buf->f_files > limit) {
6791 buf->f_files = limit;
6792 buf->f_ffree =
6793 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6794 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6795 }
6796
6797 spin_unlock(&dquot->dq_dqb_lock);
6798 dqput(dquot);
6799 return 0;
6800}
6801#endif
6802
6803static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6804{
6805 struct super_block *sb = dentry->d_sb;
6806 struct ext4_sb_info *sbi = EXT4_SB(sb);
6807 struct ext4_super_block *es = sbi->s_es;
6808 ext4_fsblk_t overhead = 0, resv_blocks;
6809 s64 bfree;
6810 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6811
6812 if (!test_opt(sb, MINIX_DF))
6813 overhead = sbi->s_overhead;
6814
6815 buf->f_type = EXT4_SUPER_MAGIC;
6816 buf->f_bsize = sb->s_blocksize;
6817 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6818 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6819 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6820 /* prevent underflow in case that few free space is available */
6821 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6822 buf->f_bavail = buf->f_bfree -
6823 (ext4_r_blocks_count(es) + resv_blocks);
6824 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6825 buf->f_bavail = 0;
6826 buf->f_files = le32_to_cpu(es->s_inodes_count);
6827 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6828 buf->f_namelen = EXT4_NAME_LEN;
6829 buf->f_fsid = uuid_to_fsid(es->s_uuid);
6830
6831#ifdef CONFIG_QUOTA
6832 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6833 sb_has_quota_limits_enabled(sb, PRJQUOTA))
6834 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6835#endif
6836 return 0;
6837}
6838
6839
6840#ifdef CONFIG_QUOTA
6841
6842/*
6843 * Helper functions so that transaction is started before we acquire dqio_sem
6844 * to keep correct lock ordering of transaction > dqio_sem
6845 */
6846static inline struct inode *dquot_to_inode(struct dquot *dquot)
6847{
6848 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6849}
6850
6851static int ext4_write_dquot(struct dquot *dquot)
6852{
6853 int ret, err;
6854 handle_t *handle;
6855 struct inode *inode;
6856
6857 inode = dquot_to_inode(dquot);
6858 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6859 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6860 if (IS_ERR(handle))
6861 return PTR_ERR(handle);
6862 ret = dquot_commit(dquot);
6863 if (ret < 0)
6864 ext4_error_err(dquot->dq_sb, -ret,
6865 "Failed to commit dquot type %d",
6866 dquot->dq_id.type);
6867 err = ext4_journal_stop(handle);
6868 if (!ret)
6869 ret = err;
6870 return ret;
6871}
6872
6873static int ext4_acquire_dquot(struct dquot *dquot)
6874{
6875 int ret, err;
6876 handle_t *handle;
6877
6878 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6879 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6880 if (IS_ERR(handle))
6881 return PTR_ERR(handle);
6882 ret = dquot_acquire(dquot);
6883 if (ret < 0)
6884 ext4_error_err(dquot->dq_sb, -ret,
6885 "Failed to acquire dquot type %d",
6886 dquot->dq_id.type);
6887 err = ext4_journal_stop(handle);
6888 if (!ret)
6889 ret = err;
6890 return ret;
6891}
6892
6893static int ext4_release_dquot(struct dquot *dquot)
6894{
6895 int ret, err;
6896 handle_t *handle;
6897
6898 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6899 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6900 if (IS_ERR(handle)) {
6901 /* Release dquot anyway to avoid endless cycle in dqput() */
6902 dquot_release(dquot);
6903 return PTR_ERR(handle);
6904 }
6905 ret = dquot_release(dquot);
6906 if (ret < 0)
6907 ext4_error_err(dquot->dq_sb, -ret,
6908 "Failed to release dquot type %d",
6909 dquot->dq_id.type);
6910 err = ext4_journal_stop(handle);
6911 if (!ret)
6912 ret = err;
6913 return ret;
6914}
6915
6916static int ext4_mark_dquot_dirty(struct dquot *dquot)
6917{
6918 struct super_block *sb = dquot->dq_sb;
6919
6920 if (ext4_is_quota_journalled(sb)) {
6921 dquot_mark_dquot_dirty(dquot);
6922 return ext4_write_dquot(dquot);
6923 } else {
6924 return dquot_mark_dquot_dirty(dquot);
6925 }
6926}
6927
6928static int ext4_write_info(struct super_block *sb, int type)
6929{
6930 int ret, err;
6931 handle_t *handle;
6932
6933 /* Data block + inode block */
6934 handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
6935 if (IS_ERR(handle))
6936 return PTR_ERR(handle);
6937 ret = dquot_commit_info(sb, type);
6938 err = ext4_journal_stop(handle);
6939 if (!ret)
6940 ret = err;
6941 return ret;
6942}
6943
6944static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6945{
6946 struct ext4_inode_info *ei = EXT4_I(inode);
6947
6948 /* The first argument of lockdep_set_subclass has to be
6949 * *exactly* the same as the argument to init_rwsem() --- in
6950 * this case, in init_once() --- or lockdep gets unhappy
6951 * because the name of the lock is set using the
6952 * stringification of the argument to init_rwsem().
6953 */
6954 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
6955 lockdep_set_subclass(&ei->i_data_sem, subclass);
6956}
6957
6958/*
6959 * Standard function to be called on quota_on
6960 */
6961static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6962 const struct path *path)
6963{
6964 int err;
6965
6966 if (!test_opt(sb, QUOTA))
6967 return -EINVAL;
6968
6969 /* Quotafile not on the same filesystem? */
6970 if (path->dentry->d_sb != sb)
6971 return -EXDEV;
6972
6973 /* Quota already enabled for this file? */
6974 if (IS_NOQUOTA(d_inode(path->dentry)))
6975 return -EBUSY;
6976
6977 /* Journaling quota? */
6978 if (EXT4_SB(sb)->s_qf_names[type]) {
6979 /* Quotafile not in fs root? */
6980 if (path->dentry->d_parent != sb->s_root)
6981 ext4_msg(sb, KERN_WARNING,
6982 "Quota file not on filesystem root. "
6983 "Journaled quota will not work");
6984 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6985 } else {
6986 /*
6987 * Clear the flag just in case mount options changed since
6988 * last time.
6989 */
6990 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6991 }
6992
6993 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6994 err = dquot_quota_on(sb, type, format_id, path);
6995 if (!err) {
6996 struct inode *inode = d_inode(path->dentry);
6997 handle_t *handle;
6998
6999 /*
7000 * Set inode flags to prevent userspace from messing with quota
7001 * files. If this fails, we return success anyway since quotas
7002 * are already enabled and this is not a hard failure.
7003 */
7004 inode_lock(inode);
7005 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7006 if (IS_ERR(handle))
7007 goto unlock_inode;
7008 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
7009 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
7010 S_NOATIME | S_IMMUTABLE);
7011 err = ext4_mark_inode_dirty(handle, inode);
7012 ext4_journal_stop(handle);
7013 unlock_inode:
7014 inode_unlock(inode);
7015 if (err)
7016 dquot_quota_off(sb, type);
7017 }
7018 if (err)
7019 lockdep_set_quota_inode(path->dentry->d_inode,
7020 I_DATA_SEM_NORMAL);
7021 return err;
7022}
7023
7024static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
7025{
7026 switch (type) {
7027 case USRQUOTA:
7028 return qf_inum == EXT4_USR_QUOTA_INO;
7029 case GRPQUOTA:
7030 return qf_inum == EXT4_GRP_QUOTA_INO;
7031 case PRJQUOTA:
7032 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
7033 default:
7034 BUG();
7035 }
7036}
7037
7038static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
7039 unsigned int flags)
7040{
7041 int err;
7042 struct inode *qf_inode;
7043 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
7044 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
7045 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
7046 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
7047 };
7048
7049 BUG_ON(!ext4_has_feature_quota(sb));
7050
7051 if (!qf_inums[type])
7052 return -EPERM;
7053
7054 if (!ext4_check_quota_inum(type, qf_inums[type])) {
7055 ext4_error(sb, "Bad quota inum: %lu, type: %d",
7056 qf_inums[type], type);
7057 return -EUCLEAN;
7058 }
7059
7060 qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
7061 if (IS_ERR(qf_inode)) {
7062 ext4_error(sb, "Bad quota inode: %lu, type: %d",
7063 qf_inums[type], type);
7064 return PTR_ERR(qf_inode);
7065 }
7066
7067 /* Don't account quota for quota files to avoid recursion */
7068 qf_inode->i_flags |= S_NOQUOTA;
7069 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
7070 err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
7071 if (err)
7072 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
7073 iput(qf_inode);
7074
7075 return err;
7076}
7077
7078/* Enable usage tracking for all quota types. */
7079int ext4_enable_quotas(struct super_block *sb)
7080{
7081 int type, err = 0;
7082 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
7083 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
7084 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
7085 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
7086 };
7087 bool quota_mopt[EXT4_MAXQUOTAS] = {
7088 test_opt(sb, USRQUOTA),
7089 test_opt(sb, GRPQUOTA),
7090 test_opt(sb, PRJQUOTA),
7091 };
7092
7093 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
7094 for (type = 0; type < EXT4_MAXQUOTAS; type++) {
7095 if (qf_inums[type]) {
7096 err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
7097 DQUOT_USAGE_ENABLED |
7098 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
7099 if (err) {
7100 ext4_warning(sb,
7101 "Failed to enable quota tracking "
7102 "(type=%d, err=%d, ino=%lu). "
7103 "Please run e2fsck to fix.", type,
7104 err, qf_inums[type]);
7105
7106 ext4_quotas_off(sb, type);
7107 return err;
7108 }
7109 }
7110 }
7111 return 0;
7112}
7113
7114static int ext4_quota_off(struct super_block *sb, int type)
7115{
7116 struct inode *inode = sb_dqopt(sb)->files[type];
7117 handle_t *handle;
7118 int err;
7119
7120 /* Force all delayed allocation blocks to be allocated.
7121 * Caller already holds s_umount sem */
7122 if (test_opt(sb, DELALLOC))
7123 sync_filesystem(sb);
7124
7125 if (!inode || !igrab(inode))
7126 goto out;
7127
7128 err = dquot_quota_off(sb, type);
7129 if (err || ext4_has_feature_quota(sb))
7130 goto out_put;
7131 /*
7132 * When the filesystem was remounted read-only first, we cannot cleanup
7133 * inode flags here. Bad luck but people should be using QUOTA feature
7134 * these days anyway.
7135 */
7136 if (sb_rdonly(sb))
7137 goto out_put;
7138
7139 inode_lock(inode);
7140 /*
7141 * Update modification times of quota files when userspace can
7142 * start looking at them. If we fail, we return success anyway since
7143 * this is not a hard failure and quotas are already disabled.
7144 */
7145 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
7146 if (IS_ERR(handle)) {
7147 err = PTR_ERR(handle);
7148 goto out_unlock;
7149 }
7150 EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
7151 inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
7152 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
7153 err = ext4_mark_inode_dirty(handle, inode);
7154 ext4_journal_stop(handle);
7155out_unlock:
7156 inode_unlock(inode);
7157out_put:
7158 lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
7159 iput(inode);
7160 return err;
7161out:
7162 return dquot_quota_off(sb, type);
7163}
7164
7165/* Read data from quotafile - avoid pagecache and such because we cannot afford
7166 * acquiring the locks... As quota files are never truncated and quota code
7167 * itself serializes the operations (and no one else should touch the files)
7168 * we don't have to be afraid of races */
7169static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
7170 size_t len, loff_t off)
7171{
7172 struct inode *inode = sb_dqopt(sb)->files[type];
7173 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7174 int offset = off & (sb->s_blocksize - 1);
7175 int tocopy;
7176 size_t toread;
7177 struct buffer_head *bh;
7178 loff_t i_size = i_size_read(inode);
7179
7180 if (off > i_size)
7181 return 0;
7182 if (off+len > i_size)
7183 len = i_size-off;
7184 toread = len;
7185 while (toread > 0) {
7186 tocopy = min_t(unsigned long, sb->s_blocksize - offset, toread);
7187 bh = ext4_bread(NULL, inode, blk, 0);
7188 if (IS_ERR(bh))
7189 return PTR_ERR(bh);
7190 if (!bh) /* A hole? */
7191 memset(data, 0, tocopy);
7192 else
7193 memcpy(data, bh->b_data+offset, tocopy);
7194 brelse(bh);
7195 offset = 0;
7196 toread -= tocopy;
7197 data += tocopy;
7198 blk++;
7199 }
7200 return len;
7201}
7202
7203/* Write to quotafile (we know the transaction is already started and has
7204 * enough credits) */
7205static ssize_t ext4_quota_write(struct super_block *sb, int type,
7206 const char *data, size_t len, loff_t off)
7207{
7208 struct inode *inode = sb_dqopt(sb)->files[type];
7209 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
7210 int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
7211 int retries = 0;
7212 struct buffer_head *bh;
7213 handle_t *handle = journal_current_handle();
7214
7215 if (!handle) {
7216 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7217 " cancelled because transaction is not started",
7218 (unsigned long long)off, (unsigned long long)len);
7219 return -EIO;
7220 }
7221 /*
7222 * Since we account only one data block in transaction credits,
7223 * then it is impossible to cross a block boundary.
7224 */
7225 if (sb->s_blocksize - offset < len) {
7226 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
7227 " cancelled because not block aligned",
7228 (unsigned long long)off, (unsigned long long)len);
7229 return -EIO;
7230 }
7231
7232 do {
7233 bh = ext4_bread(handle, inode, blk,
7234 EXT4_GET_BLOCKS_CREATE |
7235 EXT4_GET_BLOCKS_METADATA_NOFAIL);
7236 } while (PTR_ERR(bh) == -ENOSPC &&
7237 ext4_should_retry_alloc(inode->i_sb, &retries));
7238 if (IS_ERR(bh))
7239 return PTR_ERR(bh);
7240 if (!bh)
7241 goto out;
7242 BUFFER_TRACE(bh, "get write access");
7243 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
7244 if (err) {
7245 brelse(bh);
7246 return err;
7247 }
7248 lock_buffer(bh);
7249 memcpy(bh->b_data+offset, data, len);
7250 flush_dcache_page(bh->b_page);
7251 unlock_buffer(bh);
7252 err = ext4_handle_dirty_metadata(handle, NULL, bh);
7253 brelse(bh);
7254out:
7255 if (inode->i_size < off + len) {
7256 i_size_write(inode, off + len);
7257 EXT4_I(inode)->i_disksize = inode->i_size;
7258 err2 = ext4_mark_inode_dirty(handle, inode);
7259 if (unlikely(err2 && !err))
7260 err = err2;
7261 }
7262 return err ? err : len;
7263}
7264#endif
7265
7266#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
7267static inline void register_as_ext2(void)
7268{
7269 int err = register_filesystem(&ext2_fs_type);
7270 if (err)
7271 printk(KERN_WARNING
7272 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
7273}
7274
7275static inline void unregister_as_ext2(void)
7276{
7277 unregister_filesystem(&ext2_fs_type);
7278}
7279
7280static inline int ext2_feature_set_ok(struct super_block *sb)
7281{
7282 if (ext4_has_unknown_ext2_incompat_features(sb))
7283 return 0;
7284 if (sb_rdonly(sb))
7285 return 1;
7286 if (ext4_has_unknown_ext2_ro_compat_features(sb))
7287 return 0;
7288 return 1;
7289}
7290#else
7291static inline void register_as_ext2(void) { }
7292static inline void unregister_as_ext2(void) { }
7293static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
7294#endif
7295
7296static inline void register_as_ext3(void)
7297{
7298 int err = register_filesystem(&ext3_fs_type);
7299 if (err)
7300 printk(KERN_WARNING
7301 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
7302}
7303
7304static inline void unregister_as_ext3(void)
7305{
7306 unregister_filesystem(&ext3_fs_type);
7307}
7308
7309static inline int ext3_feature_set_ok(struct super_block *sb)
7310{
7311 if (ext4_has_unknown_ext3_incompat_features(sb))
7312 return 0;
7313 if (!ext4_has_feature_journal(sb))
7314 return 0;
7315 if (sb_rdonly(sb))
7316 return 1;
7317 if (ext4_has_unknown_ext3_ro_compat_features(sb))
7318 return 0;
7319 return 1;
7320}
7321
7322static void ext4_kill_sb(struct super_block *sb)
7323{
7324 struct ext4_sb_info *sbi = EXT4_SB(sb);
7325 struct file *bdev_file = sbi ? sbi->s_journal_bdev_file : NULL;
7326
7327 kill_block_super(sb);
7328
7329 if (bdev_file)
7330 bdev_fput(bdev_file);
7331}
7332
7333static struct file_system_type ext4_fs_type = {
7334 .owner = THIS_MODULE,
7335 .name = "ext4",
7336 .init_fs_context = ext4_init_fs_context,
7337 .parameters = ext4_param_specs,
7338 .kill_sb = ext4_kill_sb,
7339 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
7340};
7341MODULE_ALIAS_FS("ext4");
7342
7343/* Shared across all ext4 file systems */
7344wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
7345
7346static int __init ext4_init_fs(void)
7347{
7348 int i, err;
7349
7350 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
7351 ext4_li_info = NULL;
7352
7353 /* Build-time check for flags consistency */
7354 ext4_check_flag_values();
7355
7356 for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
7357 init_waitqueue_head(&ext4__ioend_wq[i]);
7358
7359 err = ext4_init_es();
7360 if (err)
7361 return err;
7362
7363 err = ext4_init_pending();
7364 if (err)
7365 goto out7;
7366
7367 err = ext4_init_post_read_processing();
7368 if (err)
7369 goto out6;
7370
7371 err = ext4_init_pageio();
7372 if (err)
7373 goto out5;
7374
7375 err = ext4_init_system_zone();
7376 if (err)
7377 goto out4;
7378
7379 err = ext4_init_sysfs();
7380 if (err)
7381 goto out3;
7382
7383 err = ext4_init_mballoc();
7384 if (err)
7385 goto out2;
7386 err = init_inodecache();
7387 if (err)
7388 goto out1;
7389
7390 err = ext4_fc_init_dentry_cache();
7391 if (err)
7392 goto out05;
7393
7394 register_as_ext3();
7395 register_as_ext2();
7396 err = register_filesystem(&ext4_fs_type);
7397 if (err)
7398 goto out;
7399
7400 return 0;
7401out:
7402 unregister_as_ext2();
7403 unregister_as_ext3();
7404 ext4_fc_destroy_dentry_cache();
7405out05:
7406 destroy_inodecache();
7407out1:
7408 ext4_exit_mballoc();
7409out2:
7410 ext4_exit_sysfs();
7411out3:
7412 ext4_exit_system_zone();
7413out4:
7414 ext4_exit_pageio();
7415out5:
7416 ext4_exit_post_read_processing();
7417out6:
7418 ext4_exit_pending();
7419out7:
7420 ext4_exit_es();
7421
7422 return err;
7423}
7424
7425static void __exit ext4_exit_fs(void)
7426{
7427 ext4_destroy_lazyinit_thread();
7428 unregister_as_ext2();
7429 unregister_as_ext3();
7430 unregister_filesystem(&ext4_fs_type);
7431 ext4_fc_destroy_dentry_cache();
7432 destroy_inodecache();
7433 ext4_exit_mballoc();
7434 ext4_exit_sysfs();
7435 ext4_exit_system_zone();
7436 ext4_exit_pageio();
7437 ext4_exit_post_read_processing();
7438 ext4_exit_es();
7439 ext4_exit_pending();
7440}
7441
7442MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
7443MODULE_DESCRIPTION("Fourth Extended Filesystem");
7444MODULE_LICENSE("GPL");
7445MODULE_SOFTDEP("pre: crc32c");
7446module_init(ext4_init_fs)
7447module_exit(ext4_exit_fs)