Merge tag 'mips-fixes_5.15_1' of git://git.kernel.org/pub/scm/linux/kernel/git/mips...
[linux-2.6-microblaze.git] / fs / ext4 / super.c
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/cleancache.h>
43 #include <linux/uaccess.h>
44 #include <linux/iversion.h>
45 #include <linux/unicode.h>
46 #include <linux/part_stat.h>
47 #include <linux/kthread.h>
48 #include <linux/freezer.h>
49
50 #include "ext4.h"
51 #include "ext4_extents.h"       /* Needed for trace points definition */
52 #include "ext4_jbd2.h"
53 #include "xattr.h"
54 #include "acl.h"
55 #include "mballoc.h"
56 #include "fsmap.h"
57
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/ext4.h>
60
61 static struct ext4_lazy_init *ext4_li_info;
62 static DEFINE_MUTEX(ext4_li_mtx);
63 static struct ratelimit_state ext4_mount_msg_ratelimit;
64
65 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
66                              unsigned long journal_devnum);
67 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
68 static void ext4_update_super(struct super_block *sb);
69 static int ext4_commit_super(struct super_block *sb);
70 static int ext4_mark_recovery_complete(struct super_block *sb,
71                                         struct ext4_super_block *es);
72 static int ext4_clear_journal_err(struct super_block *sb,
73                                   struct ext4_super_block *es);
74 static int ext4_sync_fs(struct super_block *sb, int wait);
75 static int ext4_remount(struct super_block *sb, int *flags, char *data);
76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
77 static int ext4_unfreeze(struct super_block *sb);
78 static int ext4_freeze(struct super_block *sb);
79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
80                        const char *dev_name, void *data);
81 static inline int ext2_feature_set_ok(struct super_block *sb);
82 static inline int ext3_feature_set_ok(struct super_block *sb);
83 static void ext4_destroy_lazyinit_thread(void);
84 static void ext4_unregister_li_request(struct super_block *sb);
85 static void ext4_clear_request_list(void);
86 static struct inode *ext4_get_journal_inode(struct super_block *sb,
87                                             unsigned int journal_inum);
88
89 /*
90  * Lock ordering
91  *
92  * page fault path:
93  * mmap_lock -> sb_start_pagefault -> invalidate_lock (r) -> transaction start
94  *   -> page lock -> i_data_sem (rw)
95  *
96  * buffered write path:
97  * sb_start_write -> i_mutex -> mmap_lock
98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
99  *   i_data_sem (rw)
100  *
101  * truncate:
102  * sb_start_write -> i_mutex -> invalidate_lock (w) -> i_mmap_rwsem (w) ->
103  *   page lock
104  * sb_start_write -> i_mutex -> invalidate_lock (w) -> transaction start ->
105  *   i_data_sem (rw)
106  *
107  * direct IO:
108  * sb_start_write -> i_mutex -> mmap_lock
109  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
110  *
111  * writepages:
112  * transaction start -> page lock(s) -> i_data_sem (rw)
113  */
114
115 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
116 static struct file_system_type ext2_fs_type = {
117         .owner          = THIS_MODULE,
118         .name           = "ext2",
119         .mount          = ext4_mount,
120         .kill_sb        = kill_block_super,
121         .fs_flags       = FS_REQUIRES_DEV,
122 };
123 MODULE_ALIAS_FS("ext2");
124 MODULE_ALIAS("ext2");
125 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
126 #else
127 #define IS_EXT2_SB(sb) (0)
128 #endif
129
130
131 static struct file_system_type ext3_fs_type = {
132         .owner          = THIS_MODULE,
133         .name           = "ext3",
134         .mount          = ext4_mount,
135         .kill_sb        = kill_block_super,
136         .fs_flags       = FS_REQUIRES_DEV,
137 };
138 MODULE_ALIAS_FS("ext3");
139 MODULE_ALIAS("ext3");
140 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
141
142
143 static inline void __ext4_read_bh(struct buffer_head *bh, int op_flags,
144                                   bh_end_io_t *end_io)
145 {
146         /*
147          * buffer's verified bit is no longer valid after reading from
148          * disk again due to write out error, clear it to make sure we
149          * recheck the buffer contents.
150          */
151         clear_buffer_verified(bh);
152
153         bh->b_end_io = end_io ? end_io : end_buffer_read_sync;
154         get_bh(bh);
155         submit_bh(REQ_OP_READ, op_flags, bh);
156 }
157
158 void ext4_read_bh_nowait(struct buffer_head *bh, int op_flags,
159                          bh_end_io_t *end_io)
160 {
161         BUG_ON(!buffer_locked(bh));
162
163         if (ext4_buffer_uptodate(bh)) {
164                 unlock_buffer(bh);
165                 return;
166         }
167         __ext4_read_bh(bh, op_flags, end_io);
168 }
169
170 int ext4_read_bh(struct buffer_head *bh, int op_flags, bh_end_io_t *end_io)
171 {
172         BUG_ON(!buffer_locked(bh));
173
174         if (ext4_buffer_uptodate(bh)) {
175                 unlock_buffer(bh);
176                 return 0;
177         }
178
179         __ext4_read_bh(bh, op_flags, end_io);
180
181         wait_on_buffer(bh);
182         if (buffer_uptodate(bh))
183                 return 0;
184         return -EIO;
185 }
186
187 int ext4_read_bh_lock(struct buffer_head *bh, int op_flags, bool wait)
188 {
189         if (trylock_buffer(bh)) {
190                 if (wait)
191                         return ext4_read_bh(bh, op_flags, NULL);
192                 ext4_read_bh_nowait(bh, op_flags, NULL);
193                 return 0;
194         }
195         if (wait) {
196                 wait_on_buffer(bh);
197                 if (buffer_uptodate(bh))
198                         return 0;
199                 return -EIO;
200         }
201         return 0;
202 }
203
204 /*
205  * This works like __bread_gfp() except it uses ERR_PTR for error
206  * returns.  Currently with sb_bread it's impossible to distinguish
207  * between ENOMEM and EIO situations (since both result in a NULL
208  * return.
209  */
210 static struct buffer_head *__ext4_sb_bread_gfp(struct super_block *sb,
211                                                sector_t block, int op_flags,
212                                                gfp_t gfp)
213 {
214         struct buffer_head *bh;
215         int ret;
216
217         bh = sb_getblk_gfp(sb, block, gfp);
218         if (bh == NULL)
219                 return ERR_PTR(-ENOMEM);
220         if (ext4_buffer_uptodate(bh))
221                 return bh;
222
223         ret = ext4_read_bh_lock(bh, REQ_META | op_flags, true);
224         if (ret) {
225                 put_bh(bh);
226                 return ERR_PTR(ret);
227         }
228         return bh;
229 }
230
231 struct buffer_head *ext4_sb_bread(struct super_block *sb, sector_t block,
232                                    int op_flags)
233 {
234         return __ext4_sb_bread_gfp(sb, block, op_flags, __GFP_MOVABLE);
235 }
236
237 struct buffer_head *ext4_sb_bread_unmovable(struct super_block *sb,
238                                             sector_t block)
239 {
240         return __ext4_sb_bread_gfp(sb, block, 0, 0);
241 }
242
243 void ext4_sb_breadahead_unmovable(struct super_block *sb, sector_t block)
244 {
245         struct buffer_head *bh = sb_getblk_gfp(sb, block, 0);
246
247         if (likely(bh)) {
248                 ext4_read_bh_lock(bh, REQ_RAHEAD, false);
249                 brelse(bh);
250         }
251 }
252
253 static int ext4_verify_csum_type(struct super_block *sb,
254                                  struct ext4_super_block *es)
255 {
256         if (!ext4_has_feature_metadata_csum(sb))
257                 return 1;
258
259         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
260 }
261
262 static __le32 ext4_superblock_csum(struct super_block *sb,
263                                    struct ext4_super_block *es)
264 {
265         struct ext4_sb_info *sbi = EXT4_SB(sb);
266         int offset = offsetof(struct ext4_super_block, s_checksum);
267         __u32 csum;
268
269         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
270
271         return cpu_to_le32(csum);
272 }
273
274 static int ext4_superblock_csum_verify(struct super_block *sb,
275                                        struct ext4_super_block *es)
276 {
277         if (!ext4_has_metadata_csum(sb))
278                 return 1;
279
280         return es->s_checksum == ext4_superblock_csum(sb, es);
281 }
282
283 void ext4_superblock_csum_set(struct super_block *sb)
284 {
285         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
286
287         if (!ext4_has_metadata_csum(sb))
288                 return;
289
290         es->s_checksum = ext4_superblock_csum(sb, es);
291 }
292
293 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
294                                struct ext4_group_desc *bg)
295 {
296         return le32_to_cpu(bg->bg_block_bitmap_lo) |
297                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
298                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
299 }
300
301 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
302                                struct ext4_group_desc *bg)
303 {
304         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
305                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
306                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
307 }
308
309 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
310                               struct ext4_group_desc *bg)
311 {
312         return le32_to_cpu(bg->bg_inode_table_lo) |
313                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
314                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
315 }
316
317 __u32 ext4_free_group_clusters(struct super_block *sb,
318                                struct ext4_group_desc *bg)
319 {
320         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
321                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
322                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
323 }
324
325 __u32 ext4_free_inodes_count(struct super_block *sb,
326                               struct ext4_group_desc *bg)
327 {
328         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
329                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
330                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
331 }
332
333 __u32 ext4_used_dirs_count(struct super_block *sb,
334                               struct ext4_group_desc *bg)
335 {
336         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
337                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
338                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
339 }
340
341 __u32 ext4_itable_unused_count(struct super_block *sb,
342                               struct ext4_group_desc *bg)
343 {
344         return le16_to_cpu(bg->bg_itable_unused_lo) |
345                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
346                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
347 }
348
349 void ext4_block_bitmap_set(struct super_block *sb,
350                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
351 {
352         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
353         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
354                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
355 }
356
357 void ext4_inode_bitmap_set(struct super_block *sb,
358                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
359 {
360         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
361         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
362                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
363 }
364
365 void ext4_inode_table_set(struct super_block *sb,
366                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
367 {
368         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
369         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
370                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
371 }
372
373 void ext4_free_group_clusters_set(struct super_block *sb,
374                                   struct ext4_group_desc *bg, __u32 count)
375 {
376         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
377         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
378                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
379 }
380
381 void ext4_free_inodes_set(struct super_block *sb,
382                           struct ext4_group_desc *bg, __u32 count)
383 {
384         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
385         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
386                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
387 }
388
389 void ext4_used_dirs_set(struct super_block *sb,
390                           struct ext4_group_desc *bg, __u32 count)
391 {
392         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
393         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
394                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
395 }
396
397 void ext4_itable_unused_set(struct super_block *sb,
398                           struct ext4_group_desc *bg, __u32 count)
399 {
400         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
401         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
402                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
403 }
404
405 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi, time64_t now)
406 {
407         now = clamp_val(now, 0, (1ull << 40) - 1);
408
409         *lo = cpu_to_le32(lower_32_bits(now));
410         *hi = upper_32_bits(now);
411 }
412
413 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
414 {
415         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
416 }
417 #define ext4_update_tstamp(es, tstamp) \
418         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi, \
419                              ktime_get_real_seconds())
420 #define ext4_get_tstamp(es, tstamp) \
421         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
422
423 /*
424  * The del_gendisk() function uninitializes the disk-specific data
425  * structures, including the bdi structure, without telling anyone
426  * else.  Once this happens, any attempt to call mark_buffer_dirty()
427  * (for example, by ext4_commit_super), will cause a kernel OOPS.
428  * This is a kludge to prevent these oops until we can put in a proper
429  * hook in del_gendisk() to inform the VFS and file system layers.
430  */
431 static int block_device_ejected(struct super_block *sb)
432 {
433         struct inode *bd_inode = sb->s_bdev->bd_inode;
434         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
435
436         return bdi->dev == NULL;
437 }
438
439 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
440 {
441         struct super_block              *sb = journal->j_private;
442         struct ext4_sb_info             *sbi = EXT4_SB(sb);
443         int                             error = is_journal_aborted(journal);
444         struct ext4_journal_cb_entry    *jce;
445
446         BUG_ON(txn->t_state == T_FINISHED);
447
448         ext4_process_freed_data(sb, txn->t_tid);
449
450         spin_lock(&sbi->s_md_lock);
451         while (!list_empty(&txn->t_private_list)) {
452                 jce = list_entry(txn->t_private_list.next,
453                                  struct ext4_journal_cb_entry, jce_list);
454                 list_del_init(&jce->jce_list);
455                 spin_unlock(&sbi->s_md_lock);
456                 jce->jce_func(sb, jce, error);
457                 spin_lock(&sbi->s_md_lock);
458         }
459         spin_unlock(&sbi->s_md_lock);
460 }
461
462 /*
463  * This writepage callback for write_cache_pages()
464  * takes care of a few cases after page cleaning.
465  *
466  * write_cache_pages() already checks for dirty pages
467  * and calls clear_page_dirty_for_io(), which we want,
468  * to write protect the pages.
469  *
470  * However, we may have to redirty a page (see below.)
471  */
472 static int ext4_journalled_writepage_callback(struct page *page,
473                                               struct writeback_control *wbc,
474                                               void *data)
475 {
476         transaction_t *transaction = (transaction_t *) data;
477         struct buffer_head *bh, *head;
478         struct journal_head *jh;
479
480         bh = head = page_buffers(page);
481         do {
482                 /*
483                  * We have to redirty a page in these cases:
484                  * 1) If buffer is dirty, it means the page was dirty because it
485                  * contains a buffer that needs checkpointing. So the dirty bit
486                  * needs to be preserved so that checkpointing writes the buffer
487                  * properly.
488                  * 2) If buffer is not part of the committing transaction
489                  * (we may have just accidentally come across this buffer because
490                  * inode range tracking is not exact) or if the currently running
491                  * transaction already contains this buffer as well, dirty bit
492                  * needs to be preserved so that the buffer gets writeprotected
493                  * properly on running transaction's commit.
494                  */
495                 jh = bh2jh(bh);
496                 if (buffer_dirty(bh) ||
497                     (jh && (jh->b_transaction != transaction ||
498                             jh->b_next_transaction))) {
499                         redirty_page_for_writepage(wbc, page);
500                         goto out;
501                 }
502         } while ((bh = bh->b_this_page) != head);
503
504 out:
505         return AOP_WRITEPAGE_ACTIVATE;
506 }
507
508 static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
509 {
510         struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
511         struct writeback_control wbc = {
512                 .sync_mode =  WB_SYNC_ALL,
513                 .nr_to_write = LONG_MAX,
514                 .range_start = jinode->i_dirty_start,
515                 .range_end = jinode->i_dirty_end,
516         };
517
518         return write_cache_pages(mapping, &wbc,
519                                  ext4_journalled_writepage_callback,
520                                  jinode->i_transaction);
521 }
522
523 static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
524 {
525         int ret;
526
527         if (ext4_should_journal_data(jinode->i_vfs_inode))
528                 ret = ext4_journalled_submit_inode_data_buffers(jinode);
529         else
530                 ret = jbd2_journal_submit_inode_data_buffers(jinode);
531
532         return ret;
533 }
534
535 static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
536 {
537         int ret = 0;
538
539         if (!ext4_should_journal_data(jinode->i_vfs_inode))
540                 ret = jbd2_journal_finish_inode_data_buffers(jinode);
541
542         return ret;
543 }
544
545 static bool system_going_down(void)
546 {
547         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
548                 || system_state == SYSTEM_RESTART;
549 }
550
551 struct ext4_err_translation {
552         int code;
553         int errno;
554 };
555
556 #define EXT4_ERR_TRANSLATE(err) { .code = EXT4_ERR_##err, .errno = err }
557
558 static struct ext4_err_translation err_translation[] = {
559         EXT4_ERR_TRANSLATE(EIO),
560         EXT4_ERR_TRANSLATE(ENOMEM),
561         EXT4_ERR_TRANSLATE(EFSBADCRC),
562         EXT4_ERR_TRANSLATE(EFSCORRUPTED),
563         EXT4_ERR_TRANSLATE(ENOSPC),
564         EXT4_ERR_TRANSLATE(ENOKEY),
565         EXT4_ERR_TRANSLATE(EROFS),
566         EXT4_ERR_TRANSLATE(EFBIG),
567         EXT4_ERR_TRANSLATE(EEXIST),
568         EXT4_ERR_TRANSLATE(ERANGE),
569         EXT4_ERR_TRANSLATE(EOVERFLOW),
570         EXT4_ERR_TRANSLATE(EBUSY),
571         EXT4_ERR_TRANSLATE(ENOTDIR),
572         EXT4_ERR_TRANSLATE(ENOTEMPTY),
573         EXT4_ERR_TRANSLATE(ESHUTDOWN),
574         EXT4_ERR_TRANSLATE(EFAULT),
575 };
576
577 static int ext4_errno_to_code(int errno)
578 {
579         int i;
580
581         for (i = 0; i < ARRAY_SIZE(err_translation); i++)
582                 if (err_translation[i].errno == errno)
583                         return err_translation[i].code;
584         return EXT4_ERR_UNKNOWN;
585 }
586
587 static void save_error_info(struct super_block *sb, int error,
588                             __u32 ino, __u64 block,
589                             const char *func, unsigned int line)
590 {
591         struct ext4_sb_info *sbi = EXT4_SB(sb);
592
593         /* We default to EFSCORRUPTED error... */
594         if (error == 0)
595                 error = EFSCORRUPTED;
596
597         spin_lock(&sbi->s_error_lock);
598         sbi->s_add_error_count++;
599         sbi->s_last_error_code = error;
600         sbi->s_last_error_line = line;
601         sbi->s_last_error_ino = ino;
602         sbi->s_last_error_block = block;
603         sbi->s_last_error_func = func;
604         sbi->s_last_error_time = ktime_get_real_seconds();
605         if (!sbi->s_first_error_time) {
606                 sbi->s_first_error_code = error;
607                 sbi->s_first_error_line = line;
608                 sbi->s_first_error_ino = ino;
609                 sbi->s_first_error_block = block;
610                 sbi->s_first_error_func = func;
611                 sbi->s_first_error_time = sbi->s_last_error_time;
612         }
613         spin_unlock(&sbi->s_error_lock);
614 }
615
616 /* Deal with the reporting of failure conditions on a filesystem such as
617  * inconsistencies detected or read IO failures.
618  *
619  * On ext2, we can store the error state of the filesystem in the
620  * superblock.  That is not possible on ext4, because we may have other
621  * write ordering constraints on the superblock which prevent us from
622  * writing it out straight away; and given that the journal is about to
623  * be aborted, we can't rely on the current, or future, transactions to
624  * write out the superblock safely.
625  *
626  * We'll just use the jbd2_journal_abort() error code to record an error in
627  * the journal instead.  On recovery, the journal will complain about
628  * that error until we've noted it down and cleared it.
629  *
630  * If force_ro is set, we unconditionally force the filesystem into an
631  * ABORT|READONLY state, unless the error response on the fs has been set to
632  * panic in which case we take the easy way out and panic immediately. This is
633  * used to deal with unrecoverable failures such as journal IO errors or ENOMEM
634  * at a critical moment in log management.
635  */
636 static void ext4_handle_error(struct super_block *sb, bool force_ro, int error,
637                               __u32 ino, __u64 block,
638                               const char *func, unsigned int line)
639 {
640         journal_t *journal = EXT4_SB(sb)->s_journal;
641         bool continue_fs = !force_ro && test_opt(sb, ERRORS_CONT);
642
643         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
644         if (test_opt(sb, WARN_ON_ERROR))
645                 WARN_ON_ONCE(1);
646
647         if (!continue_fs && !sb_rdonly(sb)) {
648                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
649                 if (journal)
650                         jbd2_journal_abort(journal, -EIO);
651         }
652
653         if (!bdev_read_only(sb->s_bdev)) {
654                 save_error_info(sb, error, ino, block, func, line);
655                 /*
656                  * In case the fs should keep running, we need to writeout
657                  * superblock through the journal. Due to lock ordering
658                  * constraints, it may not be safe to do it right here so we
659                  * defer superblock flushing to a workqueue.
660                  */
661                 if (continue_fs && journal)
662                         schedule_work(&EXT4_SB(sb)->s_error_work);
663                 else
664                         ext4_commit_super(sb);
665         }
666
667         /*
668          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
669          * could panic during 'reboot -f' as the underlying device got already
670          * disabled.
671          */
672         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
673                 panic("EXT4-fs (device %s): panic forced after error\n",
674                         sb->s_id);
675         }
676
677         if (sb_rdonly(sb) || continue_fs)
678                 return;
679
680         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
681         /*
682          * Make sure updated value of ->s_mount_flags will be visible before
683          * ->s_flags update
684          */
685         smp_wmb();
686         sb->s_flags |= SB_RDONLY;
687 }
688
689 static void flush_stashed_error_work(struct work_struct *work)
690 {
691         struct ext4_sb_info *sbi = container_of(work, struct ext4_sb_info,
692                                                 s_error_work);
693         journal_t *journal = sbi->s_journal;
694         handle_t *handle;
695
696         /*
697          * If the journal is still running, we have to write out superblock
698          * through the journal to avoid collisions of other journalled sb
699          * updates.
700          *
701          * We use directly jbd2 functions here to avoid recursing back into
702          * ext4 error handling code during handling of previous errors.
703          */
704         if (!sb_rdonly(sbi->s_sb) && journal) {
705                 struct buffer_head *sbh = sbi->s_sbh;
706                 handle = jbd2_journal_start(journal, 1);
707                 if (IS_ERR(handle))
708                         goto write_directly;
709                 if (jbd2_journal_get_write_access(handle, sbh)) {
710                         jbd2_journal_stop(handle);
711                         goto write_directly;
712                 }
713                 ext4_update_super(sbi->s_sb);
714                 if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
715                         ext4_msg(sbi->s_sb, KERN_ERR, "previous I/O error to "
716                                  "superblock detected");
717                         clear_buffer_write_io_error(sbh);
718                         set_buffer_uptodate(sbh);
719                 }
720
721                 if (jbd2_journal_dirty_metadata(handle, sbh)) {
722                         jbd2_journal_stop(handle);
723                         goto write_directly;
724                 }
725                 jbd2_journal_stop(handle);
726                 ext4_notify_error_sysfs(sbi);
727                 return;
728         }
729 write_directly:
730         /*
731          * Write through journal failed. Write sb directly to get error info
732          * out and hope for the best.
733          */
734         ext4_commit_super(sbi->s_sb);
735         ext4_notify_error_sysfs(sbi);
736 }
737
738 #define ext4_error_ratelimit(sb)                                        \
739                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
740                              "EXT4-fs error")
741
742 void __ext4_error(struct super_block *sb, const char *function,
743                   unsigned int line, bool force_ro, int error, __u64 block,
744                   const char *fmt, ...)
745 {
746         struct va_format vaf;
747         va_list args;
748
749         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
750                 return;
751
752         trace_ext4_error(sb, function, line);
753         if (ext4_error_ratelimit(sb)) {
754                 va_start(args, fmt);
755                 vaf.fmt = fmt;
756                 vaf.va = &args;
757                 printk(KERN_CRIT
758                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
759                        sb->s_id, function, line, current->comm, &vaf);
760                 va_end(args);
761         }
762         ext4_handle_error(sb, force_ro, error, 0, block, function, line);
763 }
764
765 void __ext4_error_inode(struct inode *inode, const char *function,
766                         unsigned int line, ext4_fsblk_t block, int error,
767                         const char *fmt, ...)
768 {
769         va_list args;
770         struct va_format vaf;
771
772         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
773                 return;
774
775         trace_ext4_error(inode->i_sb, function, line);
776         if (ext4_error_ratelimit(inode->i_sb)) {
777                 va_start(args, fmt);
778                 vaf.fmt = fmt;
779                 vaf.va = &args;
780                 if (block)
781                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
782                                "inode #%lu: block %llu: comm %s: %pV\n",
783                                inode->i_sb->s_id, function, line, inode->i_ino,
784                                block, current->comm, &vaf);
785                 else
786                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
787                                "inode #%lu: comm %s: %pV\n",
788                                inode->i_sb->s_id, function, line, inode->i_ino,
789                                current->comm, &vaf);
790                 va_end(args);
791         }
792         ext4_handle_error(inode->i_sb, false, error, inode->i_ino, block,
793                           function, line);
794 }
795
796 void __ext4_error_file(struct file *file, const char *function,
797                        unsigned int line, ext4_fsblk_t block,
798                        const char *fmt, ...)
799 {
800         va_list args;
801         struct va_format vaf;
802         struct inode *inode = file_inode(file);
803         char pathname[80], *path;
804
805         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
806                 return;
807
808         trace_ext4_error(inode->i_sb, function, line);
809         if (ext4_error_ratelimit(inode->i_sb)) {
810                 path = file_path(file, pathname, sizeof(pathname));
811                 if (IS_ERR(path))
812                         path = "(unknown)";
813                 va_start(args, fmt);
814                 vaf.fmt = fmt;
815                 vaf.va = &args;
816                 if (block)
817                         printk(KERN_CRIT
818                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
819                                "block %llu: comm %s: path %s: %pV\n",
820                                inode->i_sb->s_id, function, line, inode->i_ino,
821                                block, current->comm, path, &vaf);
822                 else
823                         printk(KERN_CRIT
824                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
825                                "comm %s: path %s: %pV\n",
826                                inode->i_sb->s_id, function, line, inode->i_ino,
827                                current->comm, path, &vaf);
828                 va_end(args);
829         }
830         ext4_handle_error(inode->i_sb, false, EFSCORRUPTED, inode->i_ino, block,
831                           function, line);
832 }
833
834 const char *ext4_decode_error(struct super_block *sb, int errno,
835                               char nbuf[16])
836 {
837         char *errstr = NULL;
838
839         switch (errno) {
840         case -EFSCORRUPTED:
841                 errstr = "Corrupt filesystem";
842                 break;
843         case -EFSBADCRC:
844                 errstr = "Filesystem failed CRC";
845                 break;
846         case -EIO:
847                 errstr = "IO failure";
848                 break;
849         case -ENOMEM:
850                 errstr = "Out of memory";
851                 break;
852         case -EROFS:
853                 if (!sb || (EXT4_SB(sb)->s_journal &&
854                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
855                         errstr = "Journal has aborted";
856                 else
857                         errstr = "Readonly filesystem";
858                 break;
859         default:
860                 /* If the caller passed in an extra buffer for unknown
861                  * errors, textualise them now.  Else we just return
862                  * NULL. */
863                 if (nbuf) {
864                         /* Check for truncated error codes... */
865                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
866                                 errstr = nbuf;
867                 }
868                 break;
869         }
870
871         return errstr;
872 }
873
874 /* __ext4_std_error decodes expected errors from journaling functions
875  * automatically and invokes the appropriate error response.  */
876
877 void __ext4_std_error(struct super_block *sb, const char *function,
878                       unsigned int line, int errno)
879 {
880         char nbuf[16];
881         const char *errstr;
882
883         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
884                 return;
885
886         /* Special case: if the error is EROFS, and we're not already
887          * inside a transaction, then there's really no point in logging
888          * an error. */
889         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
890                 return;
891
892         if (ext4_error_ratelimit(sb)) {
893                 errstr = ext4_decode_error(sb, errno, nbuf);
894                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
895                        sb->s_id, function, line, errstr);
896         }
897
898         ext4_handle_error(sb, false, -errno, 0, 0, function, line);
899 }
900
901 void __ext4_msg(struct super_block *sb,
902                 const char *prefix, const char *fmt, ...)
903 {
904         struct va_format vaf;
905         va_list args;
906
907         atomic_inc(&EXT4_SB(sb)->s_msg_count);
908         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
909                 return;
910
911         va_start(args, fmt);
912         vaf.fmt = fmt;
913         vaf.va = &args;
914         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
915         va_end(args);
916 }
917
918 static int ext4_warning_ratelimit(struct super_block *sb)
919 {
920         atomic_inc(&EXT4_SB(sb)->s_warning_count);
921         return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
922                             "EXT4-fs warning");
923 }
924
925 void __ext4_warning(struct super_block *sb, const char *function,
926                     unsigned int line, const char *fmt, ...)
927 {
928         struct va_format vaf;
929         va_list args;
930
931         if (!ext4_warning_ratelimit(sb))
932                 return;
933
934         va_start(args, fmt);
935         vaf.fmt = fmt;
936         vaf.va = &args;
937         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
938                sb->s_id, function, line, &vaf);
939         va_end(args);
940 }
941
942 void __ext4_warning_inode(const struct inode *inode, const char *function,
943                           unsigned int line, const char *fmt, ...)
944 {
945         struct va_format vaf;
946         va_list args;
947
948         if (!ext4_warning_ratelimit(inode->i_sb))
949                 return;
950
951         va_start(args, fmt);
952         vaf.fmt = fmt;
953         vaf.va = &args;
954         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
955                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
956                function, line, inode->i_ino, current->comm, &vaf);
957         va_end(args);
958 }
959
960 void __ext4_grp_locked_error(const char *function, unsigned int line,
961                              struct super_block *sb, ext4_group_t grp,
962                              unsigned long ino, ext4_fsblk_t block,
963                              const char *fmt, ...)
964 __releases(bitlock)
965 __acquires(bitlock)
966 {
967         struct va_format vaf;
968         va_list args;
969
970         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
971                 return;
972
973         trace_ext4_error(sb, function, line);
974         if (ext4_error_ratelimit(sb)) {
975                 va_start(args, fmt);
976                 vaf.fmt = fmt;
977                 vaf.va = &args;
978                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
979                        sb->s_id, function, line, grp);
980                 if (ino)
981                         printk(KERN_CONT "inode %lu: ", ino);
982                 if (block)
983                         printk(KERN_CONT "block %llu:",
984                                (unsigned long long) block);
985                 printk(KERN_CONT "%pV\n", &vaf);
986                 va_end(args);
987         }
988
989         if (test_opt(sb, ERRORS_CONT)) {
990                 if (test_opt(sb, WARN_ON_ERROR))
991                         WARN_ON_ONCE(1);
992                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
993                 if (!bdev_read_only(sb->s_bdev)) {
994                         save_error_info(sb, EFSCORRUPTED, ino, block, function,
995                                         line);
996                         schedule_work(&EXT4_SB(sb)->s_error_work);
997                 }
998                 return;
999         }
1000         ext4_unlock_group(sb, grp);
1001         ext4_handle_error(sb, false, EFSCORRUPTED, ino, block, function, line);
1002         /*
1003          * We only get here in the ERRORS_RO case; relocking the group
1004          * may be dangerous, but nothing bad will happen since the
1005          * filesystem will have already been marked read/only and the
1006          * journal has been aborted.  We return 1 as a hint to callers
1007          * who might what to use the return value from
1008          * ext4_grp_locked_error() to distinguish between the
1009          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
1010          * aggressively from the ext4 function in question, with a
1011          * more appropriate error code.
1012          */
1013         ext4_lock_group(sb, grp);
1014         return;
1015 }
1016
1017 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
1018                                      ext4_group_t group,
1019                                      unsigned int flags)
1020 {
1021         struct ext4_sb_info *sbi = EXT4_SB(sb);
1022         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1023         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
1024         int ret;
1025
1026         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
1027                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
1028                                             &grp->bb_state);
1029                 if (!ret)
1030                         percpu_counter_sub(&sbi->s_freeclusters_counter,
1031                                            grp->bb_free);
1032         }
1033
1034         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
1035                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
1036                                             &grp->bb_state);
1037                 if (!ret && gdp) {
1038                         int count;
1039
1040                         count = ext4_free_inodes_count(sb, gdp);
1041                         percpu_counter_sub(&sbi->s_freeinodes_counter,
1042                                            count);
1043                 }
1044         }
1045 }
1046
1047 void ext4_update_dynamic_rev(struct super_block *sb)
1048 {
1049         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1050
1051         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
1052                 return;
1053
1054         ext4_warning(sb,
1055                      "updating to rev %d because of new feature flag, "
1056                      "running e2fsck is recommended",
1057                      EXT4_DYNAMIC_REV);
1058
1059         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
1060         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
1061         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
1062         /* leave es->s_feature_*compat flags alone */
1063         /* es->s_uuid will be set by e2fsck if empty */
1064
1065         /*
1066          * The rest of the superblock fields should be zero, and if not it
1067          * means they are likely already in use, so leave them alone.  We
1068          * can leave it up to e2fsck to clean up any inconsistencies there.
1069          */
1070 }
1071
1072 /*
1073  * Open the external journal device
1074  */
1075 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
1076 {
1077         struct block_device *bdev;
1078
1079         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
1080         if (IS_ERR(bdev))
1081                 goto fail;
1082         return bdev;
1083
1084 fail:
1085         ext4_msg(sb, KERN_ERR,
1086                  "failed to open journal device unknown-block(%u,%u) %ld",
1087                  MAJOR(dev), MINOR(dev), PTR_ERR(bdev));
1088         return NULL;
1089 }
1090
1091 /*
1092  * Release the journal device
1093  */
1094 static void ext4_blkdev_put(struct block_device *bdev)
1095 {
1096         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1097 }
1098
1099 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
1100 {
1101         struct block_device *bdev;
1102         bdev = sbi->s_journal_bdev;
1103         if (bdev) {
1104                 ext4_blkdev_put(bdev);
1105                 sbi->s_journal_bdev = NULL;
1106         }
1107 }
1108
1109 static inline struct inode *orphan_list_entry(struct list_head *l)
1110 {
1111         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
1112 }
1113
1114 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
1115 {
1116         struct list_head *l;
1117
1118         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
1119                  le32_to_cpu(sbi->s_es->s_last_orphan));
1120
1121         printk(KERN_ERR "sb_info orphan list:\n");
1122         list_for_each(l, &sbi->s_orphan) {
1123                 struct inode *inode = orphan_list_entry(l);
1124                 printk(KERN_ERR "  "
1125                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
1126                        inode->i_sb->s_id, inode->i_ino, inode,
1127                        inode->i_mode, inode->i_nlink,
1128                        NEXT_ORPHAN(inode));
1129         }
1130 }
1131
1132 #ifdef CONFIG_QUOTA
1133 static int ext4_quota_off(struct super_block *sb, int type);
1134
1135 static inline void ext4_quota_off_umount(struct super_block *sb)
1136 {
1137         int type;
1138
1139         /* Use our quota_off function to clear inode flags etc. */
1140         for (type = 0; type < EXT4_MAXQUOTAS; type++)
1141                 ext4_quota_off(sb, type);
1142 }
1143
1144 /*
1145  * This is a helper function which is used in the mount/remount
1146  * codepaths (which holds s_umount) to fetch the quota file name.
1147  */
1148 static inline char *get_qf_name(struct super_block *sb,
1149                                 struct ext4_sb_info *sbi,
1150                                 int type)
1151 {
1152         return rcu_dereference_protected(sbi->s_qf_names[type],
1153                                          lockdep_is_held(&sb->s_umount));
1154 }
1155 #else
1156 static inline void ext4_quota_off_umount(struct super_block *sb)
1157 {
1158 }
1159 #endif
1160
1161 static void ext4_put_super(struct super_block *sb)
1162 {
1163         struct ext4_sb_info *sbi = EXT4_SB(sb);
1164         struct ext4_super_block *es = sbi->s_es;
1165         struct buffer_head **group_desc;
1166         struct flex_groups **flex_groups;
1167         int aborted = 0;
1168         int i, err;
1169
1170         ext4_unregister_li_request(sb);
1171         ext4_quota_off_umount(sb);
1172
1173         flush_work(&sbi->s_error_work);
1174         destroy_workqueue(sbi->rsv_conversion_wq);
1175         ext4_release_orphan_info(sb);
1176
1177         /*
1178          * Unregister sysfs before destroying jbd2 journal.
1179          * Since we could still access attr_journal_task attribute via sysfs
1180          * path which could have sbi->s_journal->j_task as NULL
1181          */
1182         ext4_unregister_sysfs(sb);
1183
1184         if (sbi->s_journal) {
1185                 aborted = is_journal_aborted(sbi->s_journal);
1186                 err = jbd2_journal_destroy(sbi->s_journal);
1187                 sbi->s_journal = NULL;
1188                 if ((err < 0) && !aborted) {
1189                         ext4_abort(sb, -err, "Couldn't clean up the journal");
1190                 }
1191         }
1192
1193         ext4_es_unregister_shrinker(sbi);
1194         del_timer_sync(&sbi->s_err_report);
1195         ext4_release_system_zone(sb);
1196         ext4_mb_release(sb);
1197         ext4_ext_release(sb);
1198
1199         if (!sb_rdonly(sb) && !aborted) {
1200                 ext4_clear_feature_journal_needs_recovery(sb);
1201                 ext4_clear_feature_orphan_present(sb);
1202                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1203         }
1204         if (!sb_rdonly(sb))
1205                 ext4_commit_super(sb);
1206
1207         rcu_read_lock();
1208         group_desc = rcu_dereference(sbi->s_group_desc);
1209         for (i = 0; i < sbi->s_gdb_count; i++)
1210                 brelse(group_desc[i]);
1211         kvfree(group_desc);
1212         flex_groups = rcu_dereference(sbi->s_flex_groups);
1213         if (flex_groups) {
1214                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1215                         kvfree(flex_groups[i]);
1216                 kvfree(flex_groups);
1217         }
1218         rcu_read_unlock();
1219         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1220         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1221         percpu_counter_destroy(&sbi->s_dirs_counter);
1222         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1223         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1224         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1225 #ifdef CONFIG_QUOTA
1226         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1227                 kfree(get_qf_name(sb, sbi, i));
1228 #endif
1229
1230         /* Debugging code just in case the in-memory inode orphan list
1231          * isn't empty.  The on-disk one can be non-empty if we've
1232          * detected an error and taken the fs readonly, but the
1233          * in-memory list had better be clean by this point. */
1234         if (!list_empty(&sbi->s_orphan))
1235                 dump_orphan_list(sb, sbi);
1236         ASSERT(list_empty(&sbi->s_orphan));
1237
1238         sync_blockdev(sb->s_bdev);
1239         invalidate_bdev(sb->s_bdev);
1240         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1241                 /*
1242                  * Invalidate the journal device's buffers.  We don't want them
1243                  * floating about in memory - the physical journal device may
1244                  * hotswapped, and it breaks the `ro-after' testing code.
1245                  */
1246                 sync_blockdev(sbi->s_journal_bdev);
1247                 invalidate_bdev(sbi->s_journal_bdev);
1248                 ext4_blkdev_remove(sbi);
1249         }
1250
1251         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1252         sbi->s_ea_inode_cache = NULL;
1253
1254         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1255         sbi->s_ea_block_cache = NULL;
1256
1257         ext4_stop_mmpd(sbi);
1258
1259         brelse(sbi->s_sbh);
1260         sb->s_fs_info = NULL;
1261         /*
1262          * Now that we are completely done shutting down the
1263          * superblock, we need to actually destroy the kobject.
1264          */
1265         kobject_put(&sbi->s_kobj);
1266         wait_for_completion(&sbi->s_kobj_unregister);
1267         if (sbi->s_chksum_driver)
1268                 crypto_free_shash(sbi->s_chksum_driver);
1269         kfree(sbi->s_blockgroup_lock);
1270         fs_put_dax(sbi->s_daxdev);
1271         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
1272 #ifdef CONFIG_UNICODE
1273         utf8_unload(sb->s_encoding);
1274 #endif
1275         kfree(sbi);
1276 }
1277
1278 static struct kmem_cache *ext4_inode_cachep;
1279
1280 /*
1281  * Called inside transaction, so use GFP_NOFS
1282  */
1283 static struct inode *ext4_alloc_inode(struct super_block *sb)
1284 {
1285         struct ext4_inode_info *ei;
1286
1287         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1288         if (!ei)
1289                 return NULL;
1290
1291         inode_set_iversion(&ei->vfs_inode, 1);
1292         spin_lock_init(&ei->i_raw_lock);
1293         INIT_LIST_HEAD(&ei->i_prealloc_list);
1294         atomic_set(&ei->i_prealloc_active, 0);
1295         spin_lock_init(&ei->i_prealloc_lock);
1296         ext4_es_init_tree(&ei->i_es_tree);
1297         rwlock_init(&ei->i_es_lock);
1298         INIT_LIST_HEAD(&ei->i_es_list);
1299         ei->i_es_all_nr = 0;
1300         ei->i_es_shk_nr = 0;
1301         ei->i_es_shrink_lblk = 0;
1302         ei->i_reserved_data_blocks = 0;
1303         spin_lock_init(&(ei->i_block_reservation_lock));
1304         ext4_init_pending_tree(&ei->i_pending_tree);
1305 #ifdef CONFIG_QUOTA
1306         ei->i_reserved_quota = 0;
1307         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1308 #endif
1309         ei->jinode = NULL;
1310         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1311         spin_lock_init(&ei->i_completed_io_lock);
1312         ei->i_sync_tid = 0;
1313         ei->i_datasync_tid = 0;
1314         atomic_set(&ei->i_unwritten, 0);
1315         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1316         ext4_fc_init_inode(&ei->vfs_inode);
1317         mutex_init(&ei->i_fc_lock);
1318         return &ei->vfs_inode;
1319 }
1320
1321 static int ext4_drop_inode(struct inode *inode)
1322 {
1323         int drop = generic_drop_inode(inode);
1324
1325         if (!drop)
1326                 drop = fscrypt_drop_inode(inode);
1327
1328         trace_ext4_drop_inode(inode, drop);
1329         return drop;
1330 }
1331
1332 static void ext4_free_in_core_inode(struct inode *inode)
1333 {
1334         fscrypt_free_inode(inode);
1335         if (!list_empty(&(EXT4_I(inode)->i_fc_list))) {
1336                 pr_warn("%s: inode %ld still in fc list",
1337                         __func__, inode->i_ino);
1338         }
1339         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1340 }
1341
1342 static void ext4_destroy_inode(struct inode *inode)
1343 {
1344         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1345                 ext4_msg(inode->i_sb, KERN_ERR,
1346                          "Inode %lu (%p): orphan list check failed!",
1347                          inode->i_ino, EXT4_I(inode));
1348                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1349                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1350                                 true);
1351                 dump_stack();
1352         }
1353
1354         if (EXT4_I(inode)->i_reserved_data_blocks)
1355                 ext4_msg(inode->i_sb, KERN_ERR,
1356                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1357                          inode->i_ino, EXT4_I(inode),
1358                          EXT4_I(inode)->i_reserved_data_blocks);
1359 }
1360
1361 static void init_once(void *foo)
1362 {
1363         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1364
1365         INIT_LIST_HEAD(&ei->i_orphan);
1366         init_rwsem(&ei->xattr_sem);
1367         init_rwsem(&ei->i_data_sem);
1368         inode_init_once(&ei->vfs_inode);
1369         ext4_fc_init_inode(&ei->vfs_inode);
1370 }
1371
1372 static int __init init_inodecache(void)
1373 {
1374         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1375                                 sizeof(struct ext4_inode_info), 0,
1376                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1377                                         SLAB_ACCOUNT),
1378                                 offsetof(struct ext4_inode_info, i_data),
1379                                 sizeof_field(struct ext4_inode_info, i_data),
1380                                 init_once);
1381         if (ext4_inode_cachep == NULL)
1382                 return -ENOMEM;
1383         return 0;
1384 }
1385
1386 static void destroy_inodecache(void)
1387 {
1388         /*
1389          * Make sure all delayed rcu free inodes are flushed before we
1390          * destroy cache.
1391          */
1392         rcu_barrier();
1393         kmem_cache_destroy(ext4_inode_cachep);
1394 }
1395
1396 void ext4_clear_inode(struct inode *inode)
1397 {
1398         ext4_fc_del(inode);
1399         invalidate_inode_buffers(inode);
1400         clear_inode(inode);
1401         ext4_discard_preallocations(inode, 0);
1402         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1403         dquot_drop(inode);
1404         if (EXT4_I(inode)->jinode) {
1405                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1406                                                EXT4_I(inode)->jinode);
1407                 jbd2_free_inode(EXT4_I(inode)->jinode);
1408                 EXT4_I(inode)->jinode = NULL;
1409         }
1410         fscrypt_put_encryption_info(inode);
1411         fsverity_cleanup_inode(inode);
1412 }
1413
1414 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1415                                         u64 ino, u32 generation)
1416 {
1417         struct inode *inode;
1418
1419         /*
1420          * Currently we don't know the generation for parent directory, so
1421          * a generation of 0 means "accept any"
1422          */
1423         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1424         if (IS_ERR(inode))
1425                 return ERR_CAST(inode);
1426         if (generation && inode->i_generation != generation) {
1427                 iput(inode);
1428                 return ERR_PTR(-ESTALE);
1429         }
1430
1431         return inode;
1432 }
1433
1434 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1435                                         int fh_len, int fh_type)
1436 {
1437         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1438                                     ext4_nfs_get_inode);
1439 }
1440
1441 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1442                                         int fh_len, int fh_type)
1443 {
1444         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1445                                     ext4_nfs_get_inode);
1446 }
1447
1448 static int ext4_nfs_commit_metadata(struct inode *inode)
1449 {
1450         struct writeback_control wbc = {
1451                 .sync_mode = WB_SYNC_ALL
1452         };
1453
1454         trace_ext4_nfs_commit_metadata(inode);
1455         return ext4_write_inode(inode, &wbc);
1456 }
1457
1458 #ifdef CONFIG_FS_ENCRYPTION
1459 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1460 {
1461         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1462                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1463 }
1464
1465 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1466                                                         void *fs_data)
1467 {
1468         handle_t *handle = fs_data;
1469         int res, res2, credits, retries = 0;
1470
1471         /*
1472          * Encrypting the root directory is not allowed because e2fsck expects
1473          * lost+found to exist and be unencrypted, and encrypting the root
1474          * directory would imply encrypting the lost+found directory as well as
1475          * the filename "lost+found" itself.
1476          */
1477         if (inode->i_ino == EXT4_ROOT_INO)
1478                 return -EPERM;
1479
1480         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1481                 return -EINVAL;
1482
1483         if (ext4_test_inode_flag(inode, EXT4_INODE_DAX))
1484                 return -EOPNOTSUPP;
1485
1486         res = ext4_convert_inline_data(inode);
1487         if (res)
1488                 return res;
1489
1490         /*
1491          * If a journal handle was specified, then the encryption context is
1492          * being set on a new inode via inheritance and is part of a larger
1493          * transaction to create the inode.  Otherwise the encryption context is
1494          * being set on an existing inode in its own transaction.  Only in the
1495          * latter case should the "retry on ENOSPC" logic be used.
1496          */
1497
1498         if (handle) {
1499                 res = ext4_xattr_set_handle(handle, inode,
1500                                             EXT4_XATTR_INDEX_ENCRYPTION,
1501                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1502                                             ctx, len, 0);
1503                 if (!res) {
1504                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1505                         ext4_clear_inode_state(inode,
1506                                         EXT4_STATE_MAY_INLINE_DATA);
1507                         /*
1508                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1509                          * S_DAX may be disabled
1510                          */
1511                         ext4_set_inode_flags(inode, false);
1512                 }
1513                 return res;
1514         }
1515
1516         res = dquot_initialize(inode);
1517         if (res)
1518                 return res;
1519 retry:
1520         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1521                                      &credits);
1522         if (res)
1523                 return res;
1524
1525         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1526         if (IS_ERR(handle))
1527                 return PTR_ERR(handle);
1528
1529         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1530                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1531                                     ctx, len, 0);
1532         if (!res) {
1533                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1534                 /*
1535                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1536                  * S_DAX may be disabled
1537                  */
1538                 ext4_set_inode_flags(inode, false);
1539                 res = ext4_mark_inode_dirty(handle, inode);
1540                 if (res)
1541                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1542         }
1543         res2 = ext4_journal_stop(handle);
1544
1545         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1546                 goto retry;
1547         if (!res)
1548                 res = res2;
1549         return res;
1550 }
1551
1552 static const union fscrypt_policy *ext4_get_dummy_policy(struct super_block *sb)
1553 {
1554         return EXT4_SB(sb)->s_dummy_enc_policy.policy;
1555 }
1556
1557 static bool ext4_has_stable_inodes(struct super_block *sb)
1558 {
1559         return ext4_has_feature_stable_inodes(sb);
1560 }
1561
1562 static void ext4_get_ino_and_lblk_bits(struct super_block *sb,
1563                                        int *ino_bits_ret, int *lblk_bits_ret)
1564 {
1565         *ino_bits_ret = 8 * sizeof(EXT4_SB(sb)->s_es->s_inodes_count);
1566         *lblk_bits_ret = 8 * sizeof(ext4_lblk_t);
1567 }
1568
1569 static const struct fscrypt_operations ext4_cryptops = {
1570         .key_prefix             = "ext4:",
1571         .get_context            = ext4_get_context,
1572         .set_context            = ext4_set_context,
1573         .get_dummy_policy       = ext4_get_dummy_policy,
1574         .empty_dir              = ext4_empty_dir,
1575         .max_namelen            = EXT4_NAME_LEN,
1576         .has_stable_inodes      = ext4_has_stable_inodes,
1577         .get_ino_and_lblk_bits  = ext4_get_ino_and_lblk_bits,
1578 };
1579 #endif
1580
1581 #ifdef CONFIG_QUOTA
1582 static const char * const quotatypes[] = INITQFNAMES;
1583 #define QTYPE2NAME(t) (quotatypes[t])
1584
1585 static int ext4_write_dquot(struct dquot *dquot);
1586 static int ext4_acquire_dquot(struct dquot *dquot);
1587 static int ext4_release_dquot(struct dquot *dquot);
1588 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1589 static int ext4_write_info(struct super_block *sb, int type);
1590 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1591                          const struct path *path);
1592 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1593                                size_t len, loff_t off);
1594 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1595                                 const char *data, size_t len, loff_t off);
1596 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1597                              unsigned int flags);
1598
1599 static struct dquot **ext4_get_dquots(struct inode *inode)
1600 {
1601         return EXT4_I(inode)->i_dquot;
1602 }
1603
1604 static const struct dquot_operations ext4_quota_operations = {
1605         .get_reserved_space     = ext4_get_reserved_space,
1606         .write_dquot            = ext4_write_dquot,
1607         .acquire_dquot          = ext4_acquire_dquot,
1608         .release_dquot          = ext4_release_dquot,
1609         .mark_dirty             = ext4_mark_dquot_dirty,
1610         .write_info             = ext4_write_info,
1611         .alloc_dquot            = dquot_alloc,
1612         .destroy_dquot          = dquot_destroy,
1613         .get_projid             = ext4_get_projid,
1614         .get_inode_usage        = ext4_get_inode_usage,
1615         .get_next_id            = dquot_get_next_id,
1616 };
1617
1618 static const struct quotactl_ops ext4_qctl_operations = {
1619         .quota_on       = ext4_quota_on,
1620         .quota_off      = ext4_quota_off,
1621         .quota_sync     = dquot_quota_sync,
1622         .get_state      = dquot_get_state,
1623         .set_info       = dquot_set_dqinfo,
1624         .get_dqblk      = dquot_get_dqblk,
1625         .set_dqblk      = dquot_set_dqblk,
1626         .get_nextdqblk  = dquot_get_next_dqblk,
1627 };
1628 #endif
1629
1630 static const struct super_operations ext4_sops = {
1631         .alloc_inode    = ext4_alloc_inode,
1632         .free_inode     = ext4_free_in_core_inode,
1633         .destroy_inode  = ext4_destroy_inode,
1634         .write_inode    = ext4_write_inode,
1635         .dirty_inode    = ext4_dirty_inode,
1636         .drop_inode     = ext4_drop_inode,
1637         .evict_inode    = ext4_evict_inode,
1638         .put_super      = ext4_put_super,
1639         .sync_fs        = ext4_sync_fs,
1640         .freeze_fs      = ext4_freeze,
1641         .unfreeze_fs    = ext4_unfreeze,
1642         .statfs         = ext4_statfs,
1643         .remount_fs     = ext4_remount,
1644         .show_options   = ext4_show_options,
1645 #ifdef CONFIG_QUOTA
1646         .quota_read     = ext4_quota_read,
1647         .quota_write    = ext4_quota_write,
1648         .get_dquots     = ext4_get_dquots,
1649 #endif
1650 };
1651
1652 static const struct export_operations ext4_export_ops = {
1653         .fh_to_dentry = ext4_fh_to_dentry,
1654         .fh_to_parent = ext4_fh_to_parent,
1655         .get_parent = ext4_get_parent,
1656         .commit_metadata = ext4_nfs_commit_metadata,
1657 };
1658
1659 enum {
1660         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1661         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1662         Opt_nouid32, Opt_debug, Opt_removed,
1663         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1664         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1665         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1666         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1667         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1668         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1669         Opt_inlinecrypt,
1670         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1671         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1672         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1673         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
1674         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
1675         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1676         Opt_nowarn_on_error, Opt_mblk_io_submit,
1677         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1678         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1679         Opt_inode_readahead_blks, Opt_journal_ioprio,
1680         Opt_dioread_nolock, Opt_dioread_lock,
1681         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1682         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1683         Opt_no_prefetch_block_bitmaps, Opt_mb_optimize_scan,
1684 #ifdef CONFIG_EXT4_DEBUG
1685         Opt_fc_debug_max_replay, Opt_fc_debug_force
1686 #endif
1687 };
1688
1689 static const match_table_t tokens = {
1690         {Opt_bsd_df, "bsddf"},
1691         {Opt_minix_df, "minixdf"},
1692         {Opt_grpid, "grpid"},
1693         {Opt_grpid, "bsdgroups"},
1694         {Opt_nogrpid, "nogrpid"},
1695         {Opt_nogrpid, "sysvgroups"},
1696         {Opt_resgid, "resgid=%u"},
1697         {Opt_resuid, "resuid=%u"},
1698         {Opt_sb, "sb=%u"},
1699         {Opt_err_cont, "errors=continue"},
1700         {Opt_err_panic, "errors=panic"},
1701         {Opt_err_ro, "errors=remount-ro"},
1702         {Opt_nouid32, "nouid32"},
1703         {Opt_debug, "debug"},
1704         {Opt_removed, "oldalloc"},
1705         {Opt_removed, "orlov"},
1706         {Opt_user_xattr, "user_xattr"},
1707         {Opt_nouser_xattr, "nouser_xattr"},
1708         {Opt_acl, "acl"},
1709         {Opt_noacl, "noacl"},
1710         {Opt_noload, "norecovery"},
1711         {Opt_noload, "noload"},
1712         {Opt_removed, "nobh"},
1713         {Opt_removed, "bh"},
1714         {Opt_commit, "commit=%u"},
1715         {Opt_min_batch_time, "min_batch_time=%u"},
1716         {Opt_max_batch_time, "max_batch_time=%u"},
1717         {Opt_journal_dev, "journal_dev=%u"},
1718         {Opt_journal_path, "journal_path=%s"},
1719         {Opt_journal_checksum, "journal_checksum"},
1720         {Opt_nojournal_checksum, "nojournal_checksum"},
1721         {Opt_journal_async_commit, "journal_async_commit"},
1722         {Opt_abort, "abort"},
1723         {Opt_data_journal, "data=journal"},
1724         {Opt_data_ordered, "data=ordered"},
1725         {Opt_data_writeback, "data=writeback"},
1726         {Opt_data_err_abort, "data_err=abort"},
1727         {Opt_data_err_ignore, "data_err=ignore"},
1728         {Opt_offusrjquota, "usrjquota="},
1729         {Opt_usrjquota, "usrjquota=%s"},
1730         {Opt_offgrpjquota, "grpjquota="},
1731         {Opt_grpjquota, "grpjquota=%s"},
1732         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1733         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1734         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1735         {Opt_grpquota, "grpquota"},
1736         {Opt_noquota, "noquota"},
1737         {Opt_quota, "quota"},
1738         {Opt_usrquota, "usrquota"},
1739         {Opt_prjquota, "prjquota"},
1740         {Opt_barrier, "barrier=%u"},
1741         {Opt_barrier, "barrier"},
1742         {Opt_nobarrier, "nobarrier"},
1743         {Opt_i_version, "i_version"},
1744         {Opt_dax, "dax"},
1745         {Opt_dax_always, "dax=always"},
1746         {Opt_dax_inode, "dax=inode"},
1747         {Opt_dax_never, "dax=never"},
1748         {Opt_stripe, "stripe=%u"},
1749         {Opt_delalloc, "delalloc"},
1750         {Opt_warn_on_error, "warn_on_error"},
1751         {Opt_nowarn_on_error, "nowarn_on_error"},
1752         {Opt_lazytime, "lazytime"},
1753         {Opt_nolazytime, "nolazytime"},
1754         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1755         {Opt_nodelalloc, "nodelalloc"},
1756         {Opt_removed, "mblk_io_submit"},
1757         {Opt_removed, "nomblk_io_submit"},
1758         {Opt_block_validity, "block_validity"},
1759         {Opt_noblock_validity, "noblock_validity"},
1760         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1761         {Opt_journal_ioprio, "journal_ioprio=%u"},
1762         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1763         {Opt_auto_da_alloc, "auto_da_alloc"},
1764         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1765         {Opt_dioread_nolock, "dioread_nolock"},
1766         {Opt_dioread_lock, "nodioread_nolock"},
1767         {Opt_dioread_lock, "dioread_lock"},
1768         {Opt_discard, "discard"},
1769         {Opt_nodiscard, "nodiscard"},
1770         {Opt_init_itable, "init_itable=%u"},
1771         {Opt_init_itable, "init_itable"},
1772         {Opt_noinit_itable, "noinit_itable"},
1773 #ifdef CONFIG_EXT4_DEBUG
1774         {Opt_fc_debug_force, "fc_debug_force"},
1775         {Opt_fc_debug_max_replay, "fc_debug_max_replay=%u"},
1776 #endif
1777         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1778         {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
1779         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1780         {Opt_inlinecrypt, "inlinecrypt"},
1781         {Opt_nombcache, "nombcache"},
1782         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1783         {Opt_removed, "prefetch_block_bitmaps"},
1784         {Opt_no_prefetch_block_bitmaps, "no_prefetch_block_bitmaps"},
1785         {Opt_mb_optimize_scan, "mb_optimize_scan=%d"},
1786         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1787         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1788         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1789         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1790         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1791         {Opt_err, NULL},
1792 };
1793
1794 static ext4_fsblk_t get_sb_block(void **data)
1795 {
1796         ext4_fsblk_t    sb_block;
1797         char            *options = (char *) *data;
1798
1799         if (!options || strncmp(options, "sb=", 3) != 0)
1800                 return 1;       /* Default location */
1801
1802         options += 3;
1803         /* TODO: use simple_strtoll with >32bit ext4 */
1804         sb_block = simple_strtoul(options, &options, 0);
1805         if (*options && *options != ',') {
1806                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1807                        (char *) *data);
1808                 return 1;
1809         }
1810         if (*options == ',')
1811                 options++;
1812         *data = (void *) options;
1813
1814         return sb_block;
1815 }
1816
1817 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1818 #define DEFAULT_MB_OPTIMIZE_SCAN        (-1)
1819
1820 static const char deprecated_msg[] =
1821         "Mount option \"%s\" will be removed by %s\n"
1822         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1823
1824 #ifdef CONFIG_QUOTA
1825 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1826 {
1827         struct ext4_sb_info *sbi = EXT4_SB(sb);
1828         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1829         int ret = -1;
1830
1831         if (sb_any_quota_loaded(sb) && !old_qname) {
1832                 ext4_msg(sb, KERN_ERR,
1833                         "Cannot change journaled "
1834                         "quota options when quota turned on");
1835                 return -1;
1836         }
1837         if (ext4_has_feature_quota(sb)) {
1838                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1839                          "ignored when QUOTA feature is enabled");
1840                 return 1;
1841         }
1842         qname = match_strdup(args);
1843         if (!qname) {
1844                 ext4_msg(sb, KERN_ERR,
1845                         "Not enough memory for storing quotafile name");
1846                 return -1;
1847         }
1848         if (old_qname) {
1849                 if (strcmp(old_qname, qname) == 0)
1850                         ret = 1;
1851                 else
1852                         ext4_msg(sb, KERN_ERR,
1853                                  "%s quota file already specified",
1854                                  QTYPE2NAME(qtype));
1855                 goto errout;
1856         }
1857         if (strchr(qname, '/')) {
1858                 ext4_msg(sb, KERN_ERR,
1859                         "quotafile must be on filesystem root");
1860                 goto errout;
1861         }
1862         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1863         set_opt(sb, QUOTA);
1864         return 1;
1865 errout:
1866         kfree(qname);
1867         return ret;
1868 }
1869
1870 static int clear_qf_name(struct super_block *sb, int qtype)
1871 {
1872
1873         struct ext4_sb_info *sbi = EXT4_SB(sb);
1874         char *old_qname = get_qf_name(sb, sbi, qtype);
1875
1876         if (sb_any_quota_loaded(sb) && old_qname) {
1877                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1878                         " when quota turned on");
1879                 return -1;
1880         }
1881         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1882         synchronize_rcu();
1883         kfree(old_qname);
1884         return 1;
1885 }
1886 #endif
1887
1888 #define MOPT_SET        0x0001
1889 #define MOPT_CLEAR      0x0002
1890 #define MOPT_NOSUPPORT  0x0004
1891 #define MOPT_EXPLICIT   0x0008
1892 #define MOPT_CLEAR_ERR  0x0010
1893 #define MOPT_GTE0       0x0020
1894 #ifdef CONFIG_QUOTA
1895 #define MOPT_Q          0
1896 #define MOPT_QFMT       0x0040
1897 #else
1898 #define MOPT_Q          MOPT_NOSUPPORT
1899 #define MOPT_QFMT       MOPT_NOSUPPORT
1900 #endif
1901 #define MOPT_DATAJ      0x0080
1902 #define MOPT_NO_EXT2    0x0100
1903 #define MOPT_NO_EXT3    0x0200
1904 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1905 #define MOPT_STRING     0x0400
1906 #define MOPT_SKIP       0x0800
1907 #define MOPT_2          0x1000
1908
1909 static const struct mount_opts {
1910         int     token;
1911         int     mount_opt;
1912         int     flags;
1913 } ext4_mount_opts[] = {
1914         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1915         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1916         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1917         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1918         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1919         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1920         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1921          MOPT_EXT4_ONLY | MOPT_SET},
1922         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1923          MOPT_EXT4_ONLY | MOPT_CLEAR},
1924         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1925         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1926         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1927          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1928         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1929          MOPT_EXT4_ONLY | MOPT_CLEAR},
1930         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1931         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1932         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1933          MOPT_EXT4_ONLY | MOPT_CLEAR},
1934         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1935          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1936         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1937                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1938          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1939         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1940         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1941         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1942         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1943         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1944          MOPT_NO_EXT2},
1945         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1946          MOPT_NO_EXT2},
1947         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1948         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1949         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1950         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1951         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1952         {Opt_commit, 0, MOPT_GTE0},
1953         {Opt_max_batch_time, 0, MOPT_GTE0},
1954         {Opt_min_batch_time, 0, MOPT_GTE0},
1955         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1956         {Opt_init_itable, 0, MOPT_GTE0},
1957         {Opt_dax, EXT4_MOUNT_DAX_ALWAYS, MOPT_SET | MOPT_SKIP},
1958         {Opt_dax_always, EXT4_MOUNT_DAX_ALWAYS,
1959                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1960         {Opt_dax_inode, EXT4_MOUNT2_DAX_INODE,
1961                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1962         {Opt_dax_never, EXT4_MOUNT2_DAX_NEVER,
1963                 MOPT_EXT4_ONLY | MOPT_SET | MOPT_SKIP},
1964         {Opt_stripe, 0, MOPT_GTE0},
1965         {Opt_resuid, 0, MOPT_GTE0},
1966         {Opt_resgid, 0, MOPT_GTE0},
1967         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1968         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1969         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1970         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1971         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1972         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1973          MOPT_NO_EXT2 | MOPT_DATAJ},
1974         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1975         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1976 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1977         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1978         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1979 #else
1980         {Opt_acl, 0, MOPT_NOSUPPORT},
1981         {Opt_noacl, 0, MOPT_NOSUPPORT},
1982 #endif
1983         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1984         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1985         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1986         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1987         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1988                                                         MOPT_SET | MOPT_Q},
1989         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1990                                                         MOPT_SET | MOPT_Q},
1991         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1992                                                         MOPT_SET | MOPT_Q},
1993         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1994                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1995                                                         MOPT_CLEAR | MOPT_Q},
1996         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
1997         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
1998         {Opt_offusrjquota, 0, MOPT_Q},
1999         {Opt_offgrpjquota, 0, MOPT_Q},
2000         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
2001         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
2002         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
2003         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
2004         {Opt_test_dummy_encryption, 0, MOPT_STRING},
2005         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
2006         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
2007          MOPT_SET},
2008         {Opt_mb_optimize_scan, EXT4_MOUNT2_MB_OPTIMIZE_SCAN, MOPT_GTE0},
2009 #ifdef CONFIG_EXT4_DEBUG
2010         {Opt_fc_debug_force, EXT4_MOUNT2_JOURNAL_FAST_COMMIT,
2011          MOPT_SET | MOPT_2 | MOPT_EXT4_ONLY},
2012         {Opt_fc_debug_max_replay, 0, MOPT_GTE0},
2013 #endif
2014         {Opt_err, 0, 0}
2015 };
2016
2017 #ifdef CONFIG_UNICODE
2018 static const struct ext4_sb_encodings {
2019         __u16 magic;
2020         char *name;
2021         char *version;
2022 } ext4_sb_encoding_map[] = {
2023         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
2024 };
2025
2026 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
2027                                  const struct ext4_sb_encodings **encoding,
2028                                  __u16 *flags)
2029 {
2030         __u16 magic = le16_to_cpu(es->s_encoding);
2031         int i;
2032
2033         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
2034                 if (magic == ext4_sb_encoding_map[i].magic)
2035                         break;
2036
2037         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
2038                 return -EINVAL;
2039
2040         *encoding = &ext4_sb_encoding_map[i];
2041         *flags = le16_to_cpu(es->s_encoding_flags);
2042
2043         return 0;
2044 }
2045 #endif
2046
2047 static int ext4_set_test_dummy_encryption(struct super_block *sb,
2048                                           const char *opt,
2049                                           const substring_t *arg,
2050                                           bool is_remount)
2051 {
2052 #ifdef CONFIG_FS_ENCRYPTION
2053         struct ext4_sb_info *sbi = EXT4_SB(sb);
2054         int err;
2055
2056         /*
2057          * This mount option is just for testing, and it's not worthwhile to
2058          * implement the extra complexity (e.g. RCU protection) that would be
2059          * needed to allow it to be set or changed during remount.  We do allow
2060          * it to be specified during remount, but only if there is no change.
2061          */
2062         if (is_remount && !sbi->s_dummy_enc_policy.policy) {
2063                 ext4_msg(sb, KERN_WARNING,
2064                          "Can't set test_dummy_encryption on remount");
2065                 return -1;
2066         }
2067         err = fscrypt_set_test_dummy_encryption(sb, arg->from,
2068                                                 &sbi->s_dummy_enc_policy);
2069         if (err) {
2070                 if (err == -EEXIST)
2071                         ext4_msg(sb, KERN_WARNING,
2072                                  "Can't change test_dummy_encryption on remount");
2073                 else if (err == -EINVAL)
2074                         ext4_msg(sb, KERN_WARNING,
2075                                  "Value of option \"%s\" is unrecognized", opt);
2076                 else
2077                         ext4_msg(sb, KERN_WARNING,
2078                                  "Error processing option \"%s\" [%d]",
2079                                  opt, err);
2080                 return -1;
2081         }
2082         ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
2083 #else
2084         ext4_msg(sb, KERN_WARNING,
2085                  "Test dummy encryption mount option ignored");
2086 #endif
2087         return 1;
2088 }
2089
2090 struct ext4_parsed_options {
2091         unsigned long journal_devnum;
2092         unsigned int journal_ioprio;
2093         int mb_optimize_scan;
2094 };
2095
2096 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
2097                             substring_t *args, struct ext4_parsed_options *parsed_opts,
2098                             int is_remount)
2099 {
2100         struct ext4_sb_info *sbi = EXT4_SB(sb);
2101         const struct mount_opts *m;
2102         kuid_t uid;
2103         kgid_t gid;
2104         int arg = 0;
2105
2106 #ifdef CONFIG_QUOTA
2107         if (token == Opt_usrjquota)
2108                 return set_qf_name(sb, USRQUOTA, &args[0]);
2109         else if (token == Opt_grpjquota)
2110                 return set_qf_name(sb, GRPQUOTA, &args[0]);
2111         else if (token == Opt_offusrjquota)
2112                 return clear_qf_name(sb, USRQUOTA);
2113         else if (token == Opt_offgrpjquota)
2114                 return clear_qf_name(sb, GRPQUOTA);
2115 #endif
2116         switch (token) {
2117         case Opt_noacl:
2118         case Opt_nouser_xattr:
2119                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
2120                 break;
2121         case Opt_sb:
2122                 return 1;       /* handled by get_sb_block() */
2123         case Opt_removed:
2124                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
2125                 return 1;
2126         case Opt_abort:
2127                 ext4_set_mount_flag(sb, EXT4_MF_FS_ABORTED);
2128                 return 1;
2129         case Opt_i_version:
2130                 sb->s_flags |= SB_I_VERSION;
2131                 return 1;
2132         case Opt_lazytime:
2133                 sb->s_flags |= SB_LAZYTIME;
2134                 return 1;
2135         case Opt_nolazytime:
2136                 sb->s_flags &= ~SB_LAZYTIME;
2137                 return 1;
2138         case Opt_inlinecrypt:
2139 #ifdef CONFIG_FS_ENCRYPTION_INLINE_CRYPT
2140                 sb->s_flags |= SB_INLINECRYPT;
2141 #else
2142                 ext4_msg(sb, KERN_ERR, "inline encryption not supported");
2143 #endif
2144                 return 1;
2145         }
2146
2147         for (m = ext4_mount_opts; m->token != Opt_err; m++)
2148                 if (token == m->token)
2149                         break;
2150
2151         if (m->token == Opt_err) {
2152                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
2153                          "or missing value", opt);
2154                 return -1;
2155         }
2156
2157         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
2158                 ext4_msg(sb, KERN_ERR,
2159                          "Mount option \"%s\" incompatible with ext2", opt);
2160                 return -1;
2161         }
2162         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
2163                 ext4_msg(sb, KERN_ERR,
2164                          "Mount option \"%s\" incompatible with ext3", opt);
2165                 return -1;
2166         }
2167
2168         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
2169                 return -1;
2170         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
2171                 return -1;
2172         if (m->flags & MOPT_EXPLICIT) {
2173                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
2174                         set_opt2(sb, EXPLICIT_DELALLOC);
2175                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
2176                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
2177                 } else
2178                         return -1;
2179         }
2180         if (m->flags & MOPT_CLEAR_ERR)
2181                 clear_opt(sb, ERRORS_MASK);
2182         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
2183                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
2184                          "options when quota turned on");
2185                 return -1;
2186         }
2187
2188         if (m->flags & MOPT_NOSUPPORT) {
2189                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
2190         } else if (token == Opt_commit) {
2191                 if (arg == 0)
2192                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
2193                 else if (arg > INT_MAX / HZ) {
2194                         ext4_msg(sb, KERN_ERR,
2195                                  "Invalid commit interval %d, "
2196                                  "must be smaller than %d",
2197                                  arg, INT_MAX / HZ);
2198                         return -1;
2199                 }
2200                 sbi->s_commit_interval = HZ * arg;
2201         } else if (token == Opt_debug_want_extra_isize) {
2202                 if ((arg & 1) ||
2203                     (arg < 4) ||
2204                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
2205                         ext4_msg(sb, KERN_ERR,
2206                                  "Invalid want_extra_isize %d", arg);
2207                         return -1;
2208                 }
2209                 sbi->s_want_extra_isize = arg;
2210         } else if (token == Opt_max_batch_time) {
2211                 sbi->s_max_batch_time = arg;
2212         } else if (token == Opt_min_batch_time) {
2213                 sbi->s_min_batch_time = arg;
2214         } else if (token == Opt_inode_readahead_blks) {
2215                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
2216                         ext4_msg(sb, KERN_ERR,
2217                                  "EXT4-fs: inode_readahead_blks must be "
2218                                  "0 or a power of 2 smaller than 2^31");
2219                         return -1;
2220                 }
2221                 sbi->s_inode_readahead_blks = arg;
2222         } else if (token == Opt_init_itable) {
2223                 set_opt(sb, INIT_INODE_TABLE);
2224                 if (!args->from)
2225                         arg = EXT4_DEF_LI_WAIT_MULT;
2226                 sbi->s_li_wait_mult = arg;
2227         } else if (token == Opt_max_dir_size_kb) {
2228                 sbi->s_max_dir_size_kb = arg;
2229 #ifdef CONFIG_EXT4_DEBUG
2230         } else if (token == Opt_fc_debug_max_replay) {
2231                 sbi->s_fc_debug_max_replay = arg;
2232 #endif
2233         } else if (token == Opt_stripe) {
2234                 sbi->s_stripe = arg;
2235         } else if (token == Opt_resuid) {
2236                 uid = make_kuid(current_user_ns(), arg);
2237                 if (!uid_valid(uid)) {
2238                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
2239                         return -1;
2240                 }
2241                 sbi->s_resuid = uid;
2242         } else if (token == Opt_resgid) {
2243                 gid = make_kgid(current_user_ns(), arg);
2244                 if (!gid_valid(gid)) {
2245                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
2246                         return -1;
2247                 }
2248                 sbi->s_resgid = gid;
2249         } else if (token == Opt_journal_dev) {
2250                 if (is_remount) {
2251                         ext4_msg(sb, KERN_ERR,
2252                                  "Cannot specify journal on remount");
2253                         return -1;
2254                 }
2255                 parsed_opts->journal_devnum = arg;
2256         } else if (token == Opt_journal_path) {
2257                 char *journal_path;
2258                 struct inode *journal_inode;
2259                 struct path path;
2260                 int error;
2261
2262                 if (is_remount) {
2263                         ext4_msg(sb, KERN_ERR,
2264                                  "Cannot specify journal on remount");
2265                         return -1;
2266                 }
2267                 journal_path = match_strdup(&args[0]);
2268                 if (!journal_path) {
2269                         ext4_msg(sb, KERN_ERR, "error: could not dup "
2270                                 "journal device string");
2271                         return -1;
2272                 }
2273
2274                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
2275                 if (error) {
2276                         ext4_msg(sb, KERN_ERR, "error: could not find "
2277                                 "journal device path: error %d", error);
2278                         kfree(journal_path);
2279                         return -1;
2280                 }
2281
2282                 journal_inode = d_inode(path.dentry);
2283                 if (!S_ISBLK(journal_inode->i_mode)) {
2284                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
2285                                 "is not a block device", journal_path);
2286                         path_put(&path);
2287                         kfree(journal_path);
2288                         return -1;
2289                 }
2290
2291                 parsed_opts->journal_devnum = new_encode_dev(journal_inode->i_rdev);
2292                 path_put(&path);
2293                 kfree(journal_path);
2294         } else if (token == Opt_journal_ioprio) {
2295                 if (arg > 7) {
2296                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
2297                                  " (must be 0-7)");
2298                         return -1;
2299                 }
2300                 parsed_opts->journal_ioprio =
2301                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2302         } else if (token == Opt_test_dummy_encryption) {
2303                 return ext4_set_test_dummy_encryption(sb, opt, &args[0],
2304                                                       is_remount);
2305         } else if (m->flags & MOPT_DATAJ) {
2306                 if (is_remount) {
2307                         if (!sbi->s_journal)
2308                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2309                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2310                                 ext4_msg(sb, KERN_ERR,
2311                                          "Cannot change data mode on remount");
2312                                 return -1;
2313                         }
2314                 } else {
2315                         clear_opt(sb, DATA_FLAGS);
2316                         sbi->s_mount_opt |= m->mount_opt;
2317                 }
2318 #ifdef CONFIG_QUOTA
2319         } else if (m->flags & MOPT_QFMT) {
2320                 if (sb_any_quota_loaded(sb) &&
2321                     sbi->s_jquota_fmt != m->mount_opt) {
2322                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2323                                  "quota options when quota turned on");
2324                         return -1;
2325                 }
2326                 if (ext4_has_feature_quota(sb)) {
2327                         ext4_msg(sb, KERN_INFO,
2328                                  "Quota format mount options ignored "
2329                                  "when QUOTA feature is enabled");
2330                         return 1;
2331                 }
2332                 sbi->s_jquota_fmt = m->mount_opt;
2333 #endif
2334         } else if (token == Opt_dax || token == Opt_dax_always ||
2335                    token == Opt_dax_inode || token == Opt_dax_never) {
2336 #ifdef CONFIG_FS_DAX
2337                 switch (token) {
2338                 case Opt_dax:
2339                 case Opt_dax_always:
2340                         if (is_remount &&
2341                             (!(sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2342                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER))) {
2343                         fail_dax_change_remount:
2344                                 ext4_msg(sb, KERN_ERR, "can't change "
2345                                          "dax mount option while remounting");
2346                                 return -1;
2347                         }
2348                         if (is_remount &&
2349                             (test_opt(sb, DATA_FLAGS) ==
2350                              EXT4_MOUNT_JOURNAL_DATA)) {
2351                                     ext4_msg(sb, KERN_ERR, "can't mount with "
2352                                              "both data=journal and dax");
2353                                     return -1;
2354                         }
2355                         ext4_msg(sb, KERN_WARNING,
2356                                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2357                         sbi->s_mount_opt |= EXT4_MOUNT_DAX_ALWAYS;
2358                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2359                         break;
2360                 case Opt_dax_never:
2361                         if (is_remount &&
2362                             (!(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2363                              (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS)))
2364                                 goto fail_dax_change_remount;
2365                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2366                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2367                         break;
2368                 case Opt_dax_inode:
2369                         if (is_remount &&
2370                             ((sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) ||
2371                              (sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_NEVER) ||
2372                              !(sbi->s_mount_opt2 & EXT4_MOUNT2_DAX_INODE)))
2373                                 goto fail_dax_change_remount;
2374                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2375                         sbi->s_mount_opt2 &= ~EXT4_MOUNT2_DAX_NEVER;
2376                         /* Strictly for printing options */
2377                         sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_INODE;
2378                         break;
2379                 }
2380 #else
2381                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2382                 sbi->s_mount_opt2 |= EXT4_MOUNT2_DAX_NEVER;
2383                 sbi->s_mount_opt &= ~EXT4_MOUNT_DAX_ALWAYS;
2384                 return -1;
2385 #endif
2386         } else if (token == Opt_data_err_abort) {
2387                 sbi->s_mount_opt |= m->mount_opt;
2388         } else if (token == Opt_data_err_ignore) {
2389                 sbi->s_mount_opt &= ~m->mount_opt;
2390         } else if (token == Opt_mb_optimize_scan) {
2391                 if (arg != 0 && arg != 1) {
2392                         ext4_msg(sb, KERN_WARNING,
2393                                  "mb_optimize_scan should be set to 0 or 1.");
2394                         return -1;
2395                 }
2396                 parsed_opts->mb_optimize_scan = arg;
2397         } else {
2398                 if (!args->from)
2399                         arg = 1;
2400                 if (m->flags & MOPT_CLEAR)
2401                         arg = !arg;
2402                 else if (unlikely(!(m->flags & MOPT_SET))) {
2403                         ext4_msg(sb, KERN_WARNING,
2404                                  "buggy handling of option %s", opt);
2405                         WARN_ON(1);
2406                         return -1;
2407                 }
2408                 if (m->flags & MOPT_2) {
2409                         if (arg != 0)
2410                                 sbi->s_mount_opt2 |= m->mount_opt;
2411                         else
2412                                 sbi->s_mount_opt2 &= ~m->mount_opt;
2413                 } else {
2414                         if (arg != 0)
2415                                 sbi->s_mount_opt |= m->mount_opt;
2416                         else
2417                                 sbi->s_mount_opt &= ~m->mount_opt;
2418                 }
2419         }
2420         return 1;
2421 }
2422
2423 static int parse_options(char *options, struct super_block *sb,
2424                          struct ext4_parsed_options *ret_opts,
2425                          int is_remount)
2426 {
2427         struct ext4_sb_info __maybe_unused *sbi = EXT4_SB(sb);
2428         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2429         substring_t args[MAX_OPT_ARGS];
2430         int token;
2431
2432         if (!options)
2433                 return 1;
2434
2435         while ((p = strsep(&options, ",")) != NULL) {
2436                 if (!*p)
2437                         continue;
2438                 /*
2439                  * Initialize args struct so we know whether arg was
2440                  * found; some options take optional arguments.
2441                  */
2442                 args[0].to = args[0].from = NULL;
2443                 token = match_token(p, tokens, args);
2444                 if (handle_mount_opt(sb, p, token, args, ret_opts,
2445                                      is_remount) < 0)
2446                         return 0;
2447         }
2448 #ifdef CONFIG_QUOTA
2449         /*
2450          * We do the test below only for project quotas. 'usrquota' and
2451          * 'grpquota' mount options are allowed even without quota feature
2452          * to support legacy quotas in quota files.
2453          */
2454         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2455                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2456                          "Cannot enable project quota enforcement.");
2457                 return 0;
2458         }
2459         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2460         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2461         if (usr_qf_name || grp_qf_name) {
2462                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2463                         clear_opt(sb, USRQUOTA);
2464
2465                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2466                         clear_opt(sb, GRPQUOTA);
2467
2468                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2469                         ext4_msg(sb, KERN_ERR, "old and new quota "
2470                                         "format mixing");
2471                         return 0;
2472                 }
2473
2474                 if (!sbi->s_jquota_fmt) {
2475                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2476                                         "not specified");
2477                         return 0;
2478                 }
2479         }
2480 #endif
2481         if (test_opt(sb, DIOREAD_NOLOCK)) {
2482                 int blocksize =
2483                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2484                 if (blocksize < PAGE_SIZE)
2485                         ext4_msg(sb, KERN_WARNING, "Warning: mounting with an "
2486                                  "experimental mount option 'dioread_nolock' "
2487                                  "for blocksize < PAGE_SIZE");
2488         }
2489         return 1;
2490 }
2491
2492 static inline void ext4_show_quota_options(struct seq_file *seq,
2493                                            struct super_block *sb)
2494 {
2495 #if defined(CONFIG_QUOTA)
2496         struct ext4_sb_info *sbi = EXT4_SB(sb);
2497         char *usr_qf_name, *grp_qf_name;
2498
2499         if (sbi->s_jquota_fmt) {
2500                 char *fmtname = "";
2501
2502                 switch (sbi->s_jquota_fmt) {
2503                 case QFMT_VFS_OLD:
2504                         fmtname = "vfsold";
2505                         break;
2506                 case QFMT_VFS_V0:
2507                         fmtname = "vfsv0";
2508                         break;
2509                 case QFMT_VFS_V1:
2510                         fmtname = "vfsv1";
2511                         break;
2512                 }
2513                 seq_printf(seq, ",jqfmt=%s", fmtname);
2514         }
2515
2516         rcu_read_lock();
2517         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2518         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2519         if (usr_qf_name)
2520                 seq_show_option(seq, "usrjquota", usr_qf_name);
2521         if (grp_qf_name)
2522                 seq_show_option(seq, "grpjquota", grp_qf_name);
2523         rcu_read_unlock();
2524 #endif
2525 }
2526
2527 static const char *token2str(int token)
2528 {
2529         const struct match_token *t;
2530
2531         for (t = tokens; t->token != Opt_err; t++)
2532                 if (t->token == token && !strchr(t->pattern, '='))
2533                         break;
2534         return t->pattern;
2535 }
2536
2537 /*
2538  * Show an option if
2539  *  - it's set to a non-default value OR
2540  *  - if the per-sb default is different from the global default
2541  */
2542 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2543                               int nodefs)
2544 {
2545         struct ext4_sb_info *sbi = EXT4_SB(sb);
2546         struct ext4_super_block *es = sbi->s_es;
2547         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2548         const struct mount_opts *m;
2549         char sep = nodefs ? '\n' : ',';
2550
2551 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2552 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2553
2554         if (sbi->s_sb_block != 1)
2555                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2556
2557         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2558                 int want_set = m->flags & MOPT_SET;
2559                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2560                     (m->flags & MOPT_CLEAR_ERR) || m->flags & MOPT_SKIP)
2561                         continue;
2562                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2563                         continue; /* skip if same as the default */
2564                 if ((want_set &&
2565                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2566                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2567                         continue; /* select Opt_noFoo vs Opt_Foo */
2568                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2569         }
2570
2571         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2572             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2573                 SEQ_OPTS_PRINT("resuid=%u",
2574                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2575         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2576             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2577                 SEQ_OPTS_PRINT("resgid=%u",
2578                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2579         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2580         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2581                 SEQ_OPTS_PUTS("errors=remount-ro");
2582         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2583                 SEQ_OPTS_PUTS("errors=continue");
2584         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2585                 SEQ_OPTS_PUTS("errors=panic");
2586         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2587                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2588         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2589                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2590         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2591                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2592         if (sb->s_flags & SB_I_VERSION)
2593                 SEQ_OPTS_PUTS("i_version");
2594         if (nodefs || sbi->s_stripe)
2595                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2596         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2597                         (sbi->s_mount_opt ^ def_mount_opt)) {
2598                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2599                         SEQ_OPTS_PUTS("data=journal");
2600                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2601                         SEQ_OPTS_PUTS("data=ordered");
2602                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2603                         SEQ_OPTS_PUTS("data=writeback");
2604         }
2605         if (nodefs ||
2606             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2607                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2608                                sbi->s_inode_readahead_blks);
2609
2610         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2611                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2612                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2613         if (nodefs || sbi->s_max_dir_size_kb)
2614                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2615         if (test_opt(sb, DATA_ERR_ABORT))
2616                 SEQ_OPTS_PUTS("data_err=abort");
2617
2618         fscrypt_show_test_dummy_encryption(seq, sep, sb);
2619
2620         if (sb->s_flags & SB_INLINECRYPT)
2621                 SEQ_OPTS_PUTS("inlinecrypt");
2622
2623         if (test_opt(sb, DAX_ALWAYS)) {
2624                 if (IS_EXT2_SB(sb))
2625                         SEQ_OPTS_PUTS("dax");
2626                 else
2627                         SEQ_OPTS_PUTS("dax=always");
2628         } else if (test_opt2(sb, DAX_NEVER)) {
2629                 SEQ_OPTS_PUTS("dax=never");
2630         } else if (test_opt2(sb, DAX_INODE)) {
2631                 SEQ_OPTS_PUTS("dax=inode");
2632         }
2633         ext4_show_quota_options(seq, sb);
2634         return 0;
2635 }
2636
2637 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2638 {
2639         return _ext4_show_options(seq, root->d_sb, 0);
2640 }
2641
2642 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2643 {
2644         struct super_block *sb = seq->private;
2645         int rc;
2646
2647         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2648         rc = _ext4_show_options(seq, sb, 1);
2649         seq_puts(seq, "\n");
2650         return rc;
2651 }
2652
2653 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2654                             int read_only)
2655 {
2656         struct ext4_sb_info *sbi = EXT4_SB(sb);
2657         int err = 0;
2658
2659         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2660                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2661                          "forcing read-only mode");
2662                 err = -EROFS;
2663                 goto done;
2664         }
2665         if (read_only)
2666                 goto done;
2667         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2668                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2669                          "running e2fsck is recommended");
2670         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2671                 ext4_msg(sb, KERN_WARNING,
2672                          "warning: mounting fs with errors, "
2673                          "running e2fsck is recommended");
2674         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2675                  le16_to_cpu(es->s_mnt_count) >=
2676                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2677                 ext4_msg(sb, KERN_WARNING,
2678                          "warning: maximal mount count reached, "
2679                          "running e2fsck is recommended");
2680         else if (le32_to_cpu(es->s_checkinterval) &&
2681                  (ext4_get_tstamp(es, s_lastcheck) +
2682                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2683                 ext4_msg(sb, KERN_WARNING,
2684                          "warning: checktime reached, "
2685                          "running e2fsck is recommended");
2686         if (!sbi->s_journal)
2687                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2688         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2689                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2690         le16_add_cpu(&es->s_mnt_count, 1);
2691         ext4_update_tstamp(es, s_mtime);
2692         if (sbi->s_journal) {
2693                 ext4_set_feature_journal_needs_recovery(sb);
2694                 if (ext4_has_feature_orphan_file(sb))
2695                         ext4_set_feature_orphan_present(sb);
2696         }
2697
2698         err = ext4_commit_super(sb);
2699 done:
2700         if (test_opt(sb, DEBUG))
2701                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2702                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2703                         sb->s_blocksize,
2704                         sbi->s_groups_count,
2705                         EXT4_BLOCKS_PER_GROUP(sb),
2706                         EXT4_INODES_PER_GROUP(sb),
2707                         sbi->s_mount_opt, sbi->s_mount_opt2);
2708
2709         cleancache_init_fs(sb);
2710         return err;
2711 }
2712
2713 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2714 {
2715         struct ext4_sb_info *sbi = EXT4_SB(sb);
2716         struct flex_groups **old_groups, **new_groups;
2717         int size, i, j;
2718
2719         if (!sbi->s_log_groups_per_flex)
2720                 return 0;
2721
2722         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2723         if (size <= sbi->s_flex_groups_allocated)
2724                 return 0;
2725
2726         new_groups = kvzalloc(roundup_pow_of_two(size *
2727                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2728         if (!new_groups) {
2729                 ext4_msg(sb, KERN_ERR,
2730                          "not enough memory for %d flex group pointers", size);
2731                 return -ENOMEM;
2732         }
2733         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2734                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2735                                          sizeof(struct flex_groups)),
2736                                          GFP_KERNEL);
2737                 if (!new_groups[i]) {
2738                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2739                                 kvfree(new_groups[j]);
2740                         kvfree(new_groups);
2741                         ext4_msg(sb, KERN_ERR,
2742                                  "not enough memory for %d flex groups", size);
2743                         return -ENOMEM;
2744                 }
2745         }
2746         rcu_read_lock();
2747         old_groups = rcu_dereference(sbi->s_flex_groups);
2748         if (old_groups)
2749                 memcpy(new_groups, old_groups,
2750                        (sbi->s_flex_groups_allocated *
2751                         sizeof(struct flex_groups *)));
2752         rcu_read_unlock();
2753         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2754         sbi->s_flex_groups_allocated = size;
2755         if (old_groups)
2756                 ext4_kvfree_array_rcu(old_groups);
2757         return 0;
2758 }
2759
2760 static int ext4_fill_flex_info(struct super_block *sb)
2761 {
2762         struct ext4_sb_info *sbi = EXT4_SB(sb);
2763         struct ext4_group_desc *gdp = NULL;
2764         struct flex_groups *fg;
2765         ext4_group_t flex_group;
2766         int i, err;
2767
2768         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2769         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2770                 sbi->s_log_groups_per_flex = 0;
2771                 return 1;
2772         }
2773
2774         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2775         if (err)
2776                 goto failed;
2777
2778         for (i = 0; i < sbi->s_groups_count; i++) {
2779                 gdp = ext4_get_group_desc(sb, i, NULL);
2780
2781                 flex_group = ext4_flex_group(sbi, i);
2782                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2783                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2784                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2785                              &fg->free_clusters);
2786                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2787         }
2788
2789         return 1;
2790 failed:
2791         return 0;
2792 }
2793
2794 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2795                                    struct ext4_group_desc *gdp)
2796 {
2797         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2798         __u16 crc = 0;
2799         __le32 le_group = cpu_to_le32(block_group);
2800         struct ext4_sb_info *sbi = EXT4_SB(sb);
2801
2802         if (ext4_has_metadata_csum(sbi->s_sb)) {
2803                 /* Use new metadata_csum algorithm */
2804                 __u32 csum32;
2805                 __u16 dummy_csum = 0;
2806
2807                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2808                                      sizeof(le_group));
2809                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2810                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2811                                      sizeof(dummy_csum));
2812                 offset += sizeof(dummy_csum);
2813                 if (offset < sbi->s_desc_size)
2814                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2815                                              sbi->s_desc_size - offset);
2816
2817                 crc = csum32 & 0xFFFF;
2818                 goto out;
2819         }
2820
2821         /* old crc16 code */
2822         if (!ext4_has_feature_gdt_csum(sb))
2823                 return 0;
2824
2825         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2826         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2827         crc = crc16(crc, (__u8 *)gdp, offset);
2828         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2829         /* for checksum of struct ext4_group_desc do the rest...*/
2830         if (ext4_has_feature_64bit(sb) &&
2831             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2832                 crc = crc16(crc, (__u8 *)gdp + offset,
2833                             le16_to_cpu(sbi->s_es->s_desc_size) -
2834                                 offset);
2835
2836 out:
2837         return cpu_to_le16(crc);
2838 }
2839
2840 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2841                                 struct ext4_group_desc *gdp)
2842 {
2843         if (ext4_has_group_desc_csum(sb) &&
2844             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2845                 return 0;
2846
2847         return 1;
2848 }
2849
2850 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2851                               struct ext4_group_desc *gdp)
2852 {
2853         if (!ext4_has_group_desc_csum(sb))
2854                 return;
2855         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2856 }
2857
2858 /* Called at mount-time, super-block is locked */
2859 static int ext4_check_descriptors(struct super_block *sb,
2860                                   ext4_fsblk_t sb_block,
2861                                   ext4_group_t *first_not_zeroed)
2862 {
2863         struct ext4_sb_info *sbi = EXT4_SB(sb);
2864         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2865         ext4_fsblk_t last_block;
2866         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2867         ext4_fsblk_t block_bitmap;
2868         ext4_fsblk_t inode_bitmap;
2869         ext4_fsblk_t inode_table;
2870         int flexbg_flag = 0;
2871         ext4_group_t i, grp = sbi->s_groups_count;
2872
2873         if (ext4_has_feature_flex_bg(sb))
2874                 flexbg_flag = 1;
2875
2876         ext4_debug("Checking group descriptors");
2877
2878         for (i = 0; i < sbi->s_groups_count; i++) {
2879                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2880
2881                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2882                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2883                 else
2884                         last_block = first_block +
2885                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2886
2887                 if ((grp == sbi->s_groups_count) &&
2888                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2889                         grp = i;
2890
2891                 block_bitmap = ext4_block_bitmap(sb, gdp);
2892                 if (block_bitmap == sb_block) {
2893                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2894                                  "Block bitmap for group %u overlaps "
2895                                  "superblock", i);
2896                         if (!sb_rdonly(sb))
2897                                 return 0;
2898                 }
2899                 if (block_bitmap >= sb_block + 1 &&
2900                     block_bitmap <= last_bg_block) {
2901                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2902                                  "Block bitmap for group %u overlaps "
2903                                  "block group descriptors", i);
2904                         if (!sb_rdonly(sb))
2905                                 return 0;
2906                 }
2907                 if (block_bitmap < first_block || block_bitmap > last_block) {
2908                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2909                                "Block bitmap for group %u not in group "
2910                                "(block %llu)!", i, block_bitmap);
2911                         return 0;
2912                 }
2913                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2914                 if (inode_bitmap == sb_block) {
2915                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2916                                  "Inode bitmap for group %u overlaps "
2917                                  "superblock", i);
2918                         if (!sb_rdonly(sb))
2919                                 return 0;
2920                 }
2921                 if (inode_bitmap >= sb_block + 1 &&
2922                     inode_bitmap <= last_bg_block) {
2923                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2924                                  "Inode bitmap for group %u overlaps "
2925                                  "block group descriptors", i);
2926                         if (!sb_rdonly(sb))
2927                                 return 0;
2928                 }
2929                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2930                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2931                                "Inode bitmap for group %u not in group "
2932                                "(block %llu)!", i, inode_bitmap);
2933                         return 0;
2934                 }
2935                 inode_table = ext4_inode_table(sb, gdp);
2936                 if (inode_table == sb_block) {
2937                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2938                                  "Inode table for group %u overlaps "
2939                                  "superblock", i);
2940                         if (!sb_rdonly(sb))
2941                                 return 0;
2942                 }
2943                 if (inode_table >= sb_block + 1 &&
2944                     inode_table <= last_bg_block) {
2945                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2946                                  "Inode table for group %u overlaps "
2947                                  "block group descriptors", i);
2948                         if (!sb_rdonly(sb))
2949                                 return 0;
2950                 }
2951                 if (inode_table < first_block ||
2952                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2953                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2954                                "Inode table for group %u not in group "
2955                                "(block %llu)!", i, inode_table);
2956                         return 0;
2957                 }
2958                 ext4_lock_group(sb, i);
2959                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2960                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2961                                  "Checksum for group %u failed (%u!=%u)",
2962                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2963                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2964                         if (!sb_rdonly(sb)) {
2965                                 ext4_unlock_group(sb, i);
2966                                 return 0;
2967                         }
2968                 }
2969                 ext4_unlock_group(sb, i);
2970                 if (!flexbg_flag)
2971                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2972         }
2973         if (NULL != first_not_zeroed)
2974                 *first_not_zeroed = grp;
2975         return 1;
2976 }
2977
2978 /*
2979  * Maximal extent format file size.
2980  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2981  * extent format containers, within a sector_t, and within i_blocks
2982  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2983  * so that won't be a limiting factor.
2984  *
2985  * However there is other limiting factor. We do store extents in the form
2986  * of starting block and length, hence the resulting length of the extent
2987  * covering maximum file size must fit into on-disk format containers as
2988  * well. Given that length is always by 1 unit bigger than max unit (because
2989  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2990  *
2991  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2992  */
2993 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2994 {
2995         loff_t res;
2996         loff_t upper_limit = MAX_LFS_FILESIZE;
2997
2998         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2999
3000         if (!has_huge_files) {
3001                 upper_limit = (1LL << 32) - 1;
3002
3003                 /* total blocks in file system block size */
3004                 upper_limit >>= (blkbits - 9);
3005                 upper_limit <<= blkbits;
3006         }
3007
3008         /*
3009          * 32-bit extent-start container, ee_block. We lower the maxbytes
3010          * by one fs block, so ee_len can cover the extent of maximum file
3011          * size
3012          */
3013         res = (1LL << 32) - 1;
3014         res <<= blkbits;
3015
3016         /* Sanity check against vm- & vfs- imposed limits */
3017         if (res > upper_limit)
3018                 res = upper_limit;
3019
3020         return res;
3021 }
3022
3023 /*
3024  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
3025  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
3026  * We need to be 1 filesystem block less than the 2^48 sector limit.
3027  */
3028 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
3029 {
3030         unsigned long long upper_limit, res = EXT4_NDIR_BLOCKS;
3031         int meta_blocks;
3032
3033         /*
3034          * This is calculated to be the largest file size for a dense, block
3035          * mapped file such that the file's total number of 512-byte sectors,
3036          * including data and all indirect blocks, does not exceed (2^48 - 1).
3037          *
3038          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
3039          * number of 512-byte sectors of the file.
3040          */
3041         if (!has_huge_files) {
3042                 /*
3043                  * !has_huge_files or implies that the inode i_block field
3044                  * represents total file blocks in 2^32 512-byte sectors ==
3045                  * size of vfs inode i_blocks * 8
3046                  */
3047                 upper_limit = (1LL << 32) - 1;
3048
3049                 /* total blocks in file system block size */
3050                 upper_limit >>= (bits - 9);
3051
3052         } else {
3053                 /*
3054                  * We use 48 bit ext4_inode i_blocks
3055                  * With EXT4_HUGE_FILE_FL set the i_blocks
3056                  * represent total number of blocks in
3057                  * file system block size
3058                  */
3059                 upper_limit = (1LL << 48) - 1;
3060
3061         }
3062
3063         /* indirect blocks */
3064         meta_blocks = 1;
3065         /* double indirect blocks */
3066         meta_blocks += 1 + (1LL << (bits-2));
3067         /* tripple indirect blocks */
3068         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
3069
3070         upper_limit -= meta_blocks;
3071         upper_limit <<= bits;
3072
3073         res += 1LL << (bits-2);
3074         res += 1LL << (2*(bits-2));
3075         res += 1LL << (3*(bits-2));
3076         res <<= bits;
3077         if (res > upper_limit)
3078                 res = upper_limit;
3079
3080         if (res > MAX_LFS_FILESIZE)
3081                 res = MAX_LFS_FILESIZE;
3082
3083         return (loff_t)res;
3084 }
3085
3086 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
3087                                    ext4_fsblk_t logical_sb_block, int nr)
3088 {
3089         struct ext4_sb_info *sbi = EXT4_SB(sb);
3090         ext4_group_t bg, first_meta_bg;
3091         int has_super = 0;
3092
3093         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
3094
3095         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
3096                 return logical_sb_block + nr + 1;
3097         bg = sbi->s_desc_per_block * nr;
3098         if (ext4_bg_has_super(sb, bg))
3099                 has_super = 1;
3100
3101         /*
3102          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
3103          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
3104          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
3105          * compensate.
3106          */
3107         if (sb->s_blocksize == 1024 && nr == 0 &&
3108             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
3109                 has_super++;
3110
3111         return (has_super + ext4_group_first_block_no(sb, bg));
3112 }
3113
3114 /**
3115  * ext4_get_stripe_size: Get the stripe size.
3116  * @sbi: In memory super block info
3117  *
3118  * If we have specified it via mount option, then
3119  * use the mount option value. If the value specified at mount time is
3120  * greater than the blocks per group use the super block value.
3121  * If the super block value is greater than blocks per group return 0.
3122  * Allocator needs it be less than blocks per group.
3123  *
3124  */
3125 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
3126 {
3127         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
3128         unsigned long stripe_width =
3129                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
3130         int ret;
3131
3132         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
3133                 ret = sbi->s_stripe;
3134         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
3135                 ret = stripe_width;
3136         else if (stride && stride <= sbi->s_blocks_per_group)
3137                 ret = stride;
3138         else
3139                 ret = 0;
3140
3141         /*
3142          * If the stripe width is 1, this makes no sense and
3143          * we set it to 0 to turn off stripe handling code.
3144          */
3145         if (ret <= 1)
3146                 ret = 0;
3147
3148         return ret;
3149 }
3150
3151 /*
3152  * Check whether this filesystem can be mounted based on
3153  * the features present and the RDONLY/RDWR mount requested.
3154  * Returns 1 if this filesystem can be mounted as requested,
3155  * 0 if it cannot be.
3156  */
3157 int ext4_feature_set_ok(struct super_block *sb, int readonly)
3158 {
3159         if (ext4_has_unknown_ext4_incompat_features(sb)) {
3160                 ext4_msg(sb, KERN_ERR,
3161                         "Couldn't mount because of "
3162                         "unsupported optional features (%x)",
3163                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
3164                         ~EXT4_FEATURE_INCOMPAT_SUPP));
3165                 return 0;
3166         }
3167
3168 #ifndef CONFIG_UNICODE
3169         if (ext4_has_feature_casefold(sb)) {
3170                 ext4_msg(sb, KERN_ERR,
3171                          "Filesystem with casefold feature cannot be "
3172                          "mounted without CONFIG_UNICODE");
3173                 return 0;
3174         }
3175 #endif
3176
3177         if (readonly)
3178                 return 1;
3179
3180         if (ext4_has_feature_readonly(sb)) {
3181                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
3182                 sb->s_flags |= SB_RDONLY;
3183                 return 1;
3184         }
3185
3186         /* Check that feature set is OK for a read-write mount */
3187         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
3188                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
3189                          "unsupported optional features (%x)",
3190                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3191                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3192                 return 0;
3193         }
3194         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3195                 ext4_msg(sb, KERN_ERR,
3196                          "Can't support bigalloc feature without "
3197                          "extents feature\n");
3198                 return 0;
3199         }
3200
3201 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3202         if (!readonly && (ext4_has_feature_quota(sb) ||
3203                           ext4_has_feature_project(sb))) {
3204                 ext4_msg(sb, KERN_ERR,
3205                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3206                 return 0;
3207         }
3208 #endif  /* CONFIG_QUOTA */
3209         return 1;
3210 }
3211
3212 /*
3213  * This function is called once a day if we have errors logged
3214  * on the file system
3215  */
3216 static void print_daily_error_info(struct timer_list *t)
3217 {
3218         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3219         struct super_block *sb = sbi->s_sb;
3220         struct ext4_super_block *es = sbi->s_es;
3221
3222         if (es->s_error_count)
3223                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3224                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3225                          le32_to_cpu(es->s_error_count));
3226         if (es->s_first_error_time) {
3227                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3228                        sb->s_id,
3229                        ext4_get_tstamp(es, s_first_error_time),
3230                        (int) sizeof(es->s_first_error_func),
3231                        es->s_first_error_func,
3232                        le32_to_cpu(es->s_first_error_line));
3233                 if (es->s_first_error_ino)
3234                         printk(KERN_CONT ": inode %u",
3235                                le32_to_cpu(es->s_first_error_ino));
3236                 if (es->s_first_error_block)
3237                         printk(KERN_CONT ": block %llu", (unsigned long long)
3238                                le64_to_cpu(es->s_first_error_block));
3239                 printk(KERN_CONT "\n");
3240         }
3241         if (es->s_last_error_time) {
3242                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3243                        sb->s_id,
3244                        ext4_get_tstamp(es, s_last_error_time),
3245                        (int) sizeof(es->s_last_error_func),
3246                        es->s_last_error_func,
3247                        le32_to_cpu(es->s_last_error_line));
3248                 if (es->s_last_error_ino)
3249                         printk(KERN_CONT ": inode %u",
3250                                le32_to_cpu(es->s_last_error_ino));
3251                 if (es->s_last_error_block)
3252                         printk(KERN_CONT ": block %llu", (unsigned long long)
3253                                le64_to_cpu(es->s_last_error_block));
3254                 printk(KERN_CONT "\n");
3255         }
3256         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3257 }
3258
3259 /* Find next suitable group and run ext4_init_inode_table */
3260 static int ext4_run_li_request(struct ext4_li_request *elr)
3261 {
3262         struct ext4_group_desc *gdp = NULL;
3263         struct super_block *sb = elr->lr_super;
3264         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3265         ext4_group_t group = elr->lr_next_group;
3266         unsigned long timeout = 0;
3267         unsigned int prefetch_ios = 0;
3268         int ret = 0;
3269
3270         if (elr->lr_mode == EXT4_LI_MODE_PREFETCH_BBITMAP) {
3271                 elr->lr_next_group = ext4_mb_prefetch(sb, group,
3272                                 EXT4_SB(sb)->s_mb_prefetch, &prefetch_ios);
3273                 if (prefetch_ios)
3274                         ext4_mb_prefetch_fini(sb, elr->lr_next_group,
3275                                               prefetch_ios);
3276                 trace_ext4_prefetch_bitmaps(sb, group, elr->lr_next_group,
3277                                             prefetch_ios);
3278                 if (group >= elr->lr_next_group) {
3279                         ret = 1;
3280                         if (elr->lr_first_not_zeroed != ngroups &&
3281                             !sb_rdonly(sb) && test_opt(sb, INIT_INODE_TABLE)) {
3282                                 elr->lr_next_group = elr->lr_first_not_zeroed;
3283                                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3284                                 ret = 0;
3285                         }
3286                 }
3287                 return ret;
3288         }
3289
3290         for (; group < ngroups; group++) {
3291                 gdp = ext4_get_group_desc(sb, group, NULL);
3292                 if (!gdp) {
3293                         ret = 1;
3294                         break;
3295                 }
3296
3297                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3298                         break;
3299         }
3300
3301         if (group >= ngroups)
3302                 ret = 1;
3303
3304         if (!ret) {
3305                 timeout = jiffies;
3306                 ret = ext4_init_inode_table(sb, group,
3307                                             elr->lr_timeout ? 0 : 1);
3308                 trace_ext4_lazy_itable_init(sb, group);
3309                 if (elr->lr_timeout == 0) {
3310                         timeout = (jiffies - timeout) *
3311                                 EXT4_SB(elr->lr_super)->s_li_wait_mult;
3312                         elr->lr_timeout = timeout;
3313                 }
3314                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3315                 elr->lr_next_group = group + 1;
3316         }
3317         return ret;
3318 }
3319
3320 /*
3321  * Remove lr_request from the list_request and free the
3322  * request structure. Should be called with li_list_mtx held
3323  */
3324 static void ext4_remove_li_request(struct ext4_li_request *elr)
3325 {
3326         if (!elr)
3327                 return;
3328
3329         list_del(&elr->lr_request);
3330         EXT4_SB(elr->lr_super)->s_li_request = NULL;
3331         kfree(elr);
3332 }
3333
3334 static void ext4_unregister_li_request(struct super_block *sb)
3335 {
3336         mutex_lock(&ext4_li_mtx);
3337         if (!ext4_li_info) {
3338                 mutex_unlock(&ext4_li_mtx);
3339                 return;
3340         }
3341
3342         mutex_lock(&ext4_li_info->li_list_mtx);
3343         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3344         mutex_unlock(&ext4_li_info->li_list_mtx);
3345         mutex_unlock(&ext4_li_mtx);
3346 }
3347
3348 static struct task_struct *ext4_lazyinit_task;
3349
3350 /*
3351  * This is the function where ext4lazyinit thread lives. It walks
3352  * through the request list searching for next scheduled filesystem.
3353  * When such a fs is found, run the lazy initialization request
3354  * (ext4_rn_li_request) and keep track of the time spend in this
3355  * function. Based on that time we compute next schedule time of
3356  * the request. When walking through the list is complete, compute
3357  * next waking time and put itself into sleep.
3358  */
3359 static int ext4_lazyinit_thread(void *arg)
3360 {
3361         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3362         struct list_head *pos, *n;
3363         struct ext4_li_request *elr;
3364         unsigned long next_wakeup, cur;
3365
3366         BUG_ON(NULL == eli);
3367
3368 cont_thread:
3369         while (true) {
3370                 next_wakeup = MAX_JIFFY_OFFSET;
3371
3372                 mutex_lock(&eli->li_list_mtx);
3373                 if (list_empty(&eli->li_request_list)) {
3374                         mutex_unlock(&eli->li_list_mtx);
3375                         goto exit_thread;
3376                 }
3377                 list_for_each_safe(pos, n, &eli->li_request_list) {
3378                         int err = 0;
3379                         int progress = 0;
3380                         elr = list_entry(pos, struct ext4_li_request,
3381                                          lr_request);
3382
3383                         if (time_before(jiffies, elr->lr_next_sched)) {
3384                                 if (time_before(elr->lr_next_sched, next_wakeup))
3385                                         next_wakeup = elr->lr_next_sched;
3386                                 continue;
3387                         }
3388                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3389                                 if (sb_start_write_trylock(elr->lr_super)) {
3390                                         progress = 1;
3391                                         /*
3392                                          * We hold sb->s_umount, sb can not
3393                                          * be removed from the list, it is
3394                                          * now safe to drop li_list_mtx
3395                                          */
3396                                         mutex_unlock(&eli->li_list_mtx);
3397                                         err = ext4_run_li_request(elr);
3398                                         sb_end_write(elr->lr_super);
3399                                         mutex_lock(&eli->li_list_mtx);
3400                                         n = pos->next;
3401                                 }
3402                                 up_read((&elr->lr_super->s_umount));
3403                         }
3404                         /* error, remove the lazy_init job */
3405                         if (err) {
3406                                 ext4_remove_li_request(elr);
3407                                 continue;
3408                         }
3409                         if (!progress) {
3410                                 elr->lr_next_sched = jiffies +
3411                                         (prandom_u32()
3412                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3413                         }
3414                         if (time_before(elr->lr_next_sched, next_wakeup))
3415                                 next_wakeup = elr->lr_next_sched;
3416                 }
3417                 mutex_unlock(&eli->li_list_mtx);
3418
3419                 try_to_freeze();
3420
3421                 cur = jiffies;
3422                 if ((time_after_eq(cur, next_wakeup)) ||
3423                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3424                         cond_resched();
3425                         continue;
3426                 }
3427
3428                 schedule_timeout_interruptible(next_wakeup - cur);
3429
3430                 if (kthread_should_stop()) {
3431                         ext4_clear_request_list();
3432                         goto exit_thread;
3433                 }
3434         }
3435
3436 exit_thread:
3437         /*
3438          * It looks like the request list is empty, but we need
3439          * to check it under the li_list_mtx lock, to prevent any
3440          * additions into it, and of course we should lock ext4_li_mtx
3441          * to atomically free the list and ext4_li_info, because at
3442          * this point another ext4 filesystem could be registering
3443          * new one.
3444          */
3445         mutex_lock(&ext4_li_mtx);
3446         mutex_lock(&eli->li_list_mtx);
3447         if (!list_empty(&eli->li_request_list)) {
3448                 mutex_unlock(&eli->li_list_mtx);
3449                 mutex_unlock(&ext4_li_mtx);
3450                 goto cont_thread;
3451         }
3452         mutex_unlock(&eli->li_list_mtx);
3453         kfree(ext4_li_info);
3454         ext4_li_info = NULL;
3455         mutex_unlock(&ext4_li_mtx);
3456
3457         return 0;
3458 }
3459
3460 static void ext4_clear_request_list(void)
3461 {
3462         struct list_head *pos, *n;
3463         struct ext4_li_request *elr;
3464
3465         mutex_lock(&ext4_li_info->li_list_mtx);
3466         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3467                 elr = list_entry(pos, struct ext4_li_request,
3468                                  lr_request);
3469                 ext4_remove_li_request(elr);
3470         }
3471         mutex_unlock(&ext4_li_info->li_list_mtx);
3472 }
3473
3474 static int ext4_run_lazyinit_thread(void)
3475 {
3476         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3477                                          ext4_li_info, "ext4lazyinit");
3478         if (IS_ERR(ext4_lazyinit_task)) {
3479                 int err = PTR_ERR(ext4_lazyinit_task);
3480                 ext4_clear_request_list();
3481                 kfree(ext4_li_info);
3482                 ext4_li_info = NULL;
3483                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3484                                  "initialization thread\n",
3485                                  err);
3486                 return err;
3487         }
3488         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3489         return 0;
3490 }
3491
3492 /*
3493  * Check whether it make sense to run itable init. thread or not.
3494  * If there is at least one uninitialized inode table, return
3495  * corresponding group number, else the loop goes through all
3496  * groups and return total number of groups.
3497  */
3498 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3499 {
3500         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3501         struct ext4_group_desc *gdp = NULL;
3502
3503         if (!ext4_has_group_desc_csum(sb))
3504                 return ngroups;
3505
3506         for (group = 0; group < ngroups; group++) {
3507                 gdp = ext4_get_group_desc(sb, group, NULL);
3508                 if (!gdp)
3509                         continue;
3510
3511                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3512                         break;
3513         }
3514
3515         return group;
3516 }
3517
3518 static int ext4_li_info_new(void)
3519 {
3520         struct ext4_lazy_init *eli = NULL;
3521
3522         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3523         if (!eli)
3524                 return -ENOMEM;
3525
3526         INIT_LIST_HEAD(&eli->li_request_list);
3527         mutex_init(&eli->li_list_mtx);
3528
3529         eli->li_state |= EXT4_LAZYINIT_QUIT;
3530
3531         ext4_li_info = eli;
3532
3533         return 0;
3534 }
3535
3536 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3537                                             ext4_group_t start)
3538 {
3539         struct ext4_li_request *elr;
3540
3541         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3542         if (!elr)
3543                 return NULL;
3544
3545         elr->lr_super = sb;
3546         elr->lr_first_not_zeroed = start;
3547         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS)) {
3548                 elr->lr_mode = EXT4_LI_MODE_ITABLE;
3549                 elr->lr_next_group = start;
3550         } else {
3551                 elr->lr_mode = EXT4_LI_MODE_PREFETCH_BBITMAP;
3552         }
3553
3554         /*
3555          * Randomize first schedule time of the request to
3556          * spread the inode table initialization requests
3557          * better.
3558          */
3559         elr->lr_next_sched = jiffies + (prandom_u32() %
3560                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3561         return elr;
3562 }
3563
3564 int ext4_register_li_request(struct super_block *sb,
3565                              ext4_group_t first_not_zeroed)
3566 {
3567         struct ext4_sb_info *sbi = EXT4_SB(sb);
3568         struct ext4_li_request *elr = NULL;
3569         ext4_group_t ngroups = sbi->s_groups_count;
3570         int ret = 0;
3571
3572         mutex_lock(&ext4_li_mtx);
3573         if (sbi->s_li_request != NULL) {
3574                 /*
3575                  * Reset timeout so it can be computed again, because
3576                  * s_li_wait_mult might have changed.
3577                  */
3578                 sbi->s_li_request->lr_timeout = 0;
3579                 goto out;
3580         }
3581
3582         if (test_opt(sb, NO_PREFETCH_BLOCK_BITMAPS) &&
3583             (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3584              !test_opt(sb, INIT_INODE_TABLE)))
3585                 goto out;
3586
3587         elr = ext4_li_request_new(sb, first_not_zeroed);
3588         if (!elr) {
3589                 ret = -ENOMEM;
3590                 goto out;
3591         }
3592
3593         if (NULL == ext4_li_info) {
3594                 ret = ext4_li_info_new();
3595                 if (ret)
3596                         goto out;
3597         }
3598
3599         mutex_lock(&ext4_li_info->li_list_mtx);
3600         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3601         mutex_unlock(&ext4_li_info->li_list_mtx);
3602
3603         sbi->s_li_request = elr;
3604         /*
3605          * set elr to NULL here since it has been inserted to
3606          * the request_list and the removal and free of it is
3607          * handled by ext4_clear_request_list from now on.
3608          */
3609         elr = NULL;
3610
3611         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3612                 ret = ext4_run_lazyinit_thread();
3613                 if (ret)
3614                         goto out;
3615         }
3616 out:
3617         mutex_unlock(&ext4_li_mtx);
3618         if (ret)
3619                 kfree(elr);
3620         return ret;
3621 }
3622
3623 /*
3624  * We do not need to lock anything since this is called on
3625  * module unload.
3626  */
3627 static void ext4_destroy_lazyinit_thread(void)
3628 {
3629         /*
3630          * If thread exited earlier
3631          * there's nothing to be done.
3632          */
3633         if (!ext4_li_info || !ext4_lazyinit_task)
3634                 return;
3635
3636         kthread_stop(ext4_lazyinit_task);
3637 }
3638
3639 static int set_journal_csum_feature_set(struct super_block *sb)
3640 {
3641         int ret = 1;
3642         int compat, incompat;
3643         struct ext4_sb_info *sbi = EXT4_SB(sb);
3644
3645         if (ext4_has_metadata_csum(sb)) {
3646                 /* journal checksum v3 */
3647                 compat = 0;
3648                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3649         } else {
3650                 /* journal checksum v1 */
3651                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3652                 incompat = 0;
3653         }
3654
3655         jbd2_journal_clear_features(sbi->s_journal,
3656                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3657                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3658                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3659         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3660                 ret = jbd2_journal_set_features(sbi->s_journal,
3661                                 compat, 0,
3662                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3663                                 incompat);
3664         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3665                 ret = jbd2_journal_set_features(sbi->s_journal,
3666                                 compat, 0,
3667                                 incompat);
3668                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3669                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3670         } else {
3671                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3672                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3673         }
3674
3675         return ret;
3676 }
3677
3678 /*
3679  * Note: calculating the overhead so we can be compatible with
3680  * historical BSD practice is quite difficult in the face of
3681  * clusters/bigalloc.  This is because multiple metadata blocks from
3682  * different block group can end up in the same allocation cluster.
3683  * Calculating the exact overhead in the face of clustered allocation
3684  * requires either O(all block bitmaps) in memory or O(number of block
3685  * groups**2) in time.  We will still calculate the superblock for
3686  * older file systems --- and if we come across with a bigalloc file
3687  * system with zero in s_overhead_clusters the estimate will be close to
3688  * correct especially for very large cluster sizes --- but for newer
3689  * file systems, it's better to calculate this figure once at mkfs
3690  * time, and store it in the superblock.  If the superblock value is
3691  * present (even for non-bigalloc file systems), we will use it.
3692  */
3693 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3694                           char *buf)
3695 {
3696         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3697         struct ext4_group_desc  *gdp;
3698         ext4_fsblk_t            first_block, last_block, b;
3699         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3700         int                     s, j, count = 0;
3701
3702         if (!ext4_has_feature_bigalloc(sb))
3703                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3704                         sbi->s_itb_per_group + 2);
3705
3706         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3707                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3708         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3709         for (i = 0; i < ngroups; i++) {
3710                 gdp = ext4_get_group_desc(sb, i, NULL);
3711                 b = ext4_block_bitmap(sb, gdp);
3712                 if (b >= first_block && b <= last_block) {
3713                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3714                         count++;
3715                 }
3716                 b = ext4_inode_bitmap(sb, gdp);
3717                 if (b >= first_block && b <= last_block) {
3718                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3719                         count++;
3720                 }
3721                 b = ext4_inode_table(sb, gdp);
3722                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3723                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3724                                 int c = EXT4_B2C(sbi, b - first_block);
3725                                 ext4_set_bit(c, buf);
3726                                 count++;
3727                         }
3728                 if (i != grp)
3729                         continue;
3730                 s = 0;
3731                 if (ext4_bg_has_super(sb, grp)) {
3732                         ext4_set_bit(s++, buf);
3733                         count++;
3734                 }
3735                 j = ext4_bg_num_gdb(sb, grp);
3736                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3737                         ext4_error(sb, "Invalid number of block group "
3738                                    "descriptor blocks: %d", j);
3739                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3740                 }
3741                 count += j;
3742                 for (; j > 0; j--)
3743                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3744         }
3745         if (!count)
3746                 return 0;
3747         return EXT4_CLUSTERS_PER_GROUP(sb) -
3748                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3749 }
3750
3751 /*
3752  * Compute the overhead and stash it in sbi->s_overhead
3753  */
3754 int ext4_calculate_overhead(struct super_block *sb)
3755 {
3756         struct ext4_sb_info *sbi = EXT4_SB(sb);
3757         struct ext4_super_block *es = sbi->s_es;
3758         struct inode *j_inode;
3759         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3760         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3761         ext4_fsblk_t overhead = 0;
3762         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3763
3764         if (!buf)
3765                 return -ENOMEM;
3766
3767         /*
3768          * Compute the overhead (FS structures).  This is constant
3769          * for a given filesystem unless the number of block groups
3770          * changes so we cache the previous value until it does.
3771          */
3772
3773         /*
3774          * All of the blocks before first_data_block are overhead
3775          */
3776         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3777
3778         /*
3779          * Add the overhead found in each block group
3780          */
3781         for (i = 0; i < ngroups; i++) {
3782                 int blks;
3783
3784                 blks = count_overhead(sb, i, buf);
3785                 overhead += blks;
3786                 if (blks)
3787                         memset(buf, 0, PAGE_SIZE);
3788                 cond_resched();
3789         }
3790
3791         /*
3792          * Add the internal journal blocks whether the journal has been
3793          * loaded or not
3794          */
3795         if (sbi->s_journal && !sbi->s_journal_bdev)
3796                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_total_len);
3797         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3798                 /* j_inum for internal journal is non-zero */
3799                 j_inode = ext4_get_journal_inode(sb, j_inum);
3800                 if (j_inode) {
3801                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3802                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3803                         iput(j_inode);
3804                 } else {
3805                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3806                 }
3807         }
3808         sbi->s_overhead = overhead;
3809         smp_wmb();
3810         free_page((unsigned long) buf);
3811         return 0;
3812 }
3813
3814 static void ext4_set_resv_clusters(struct super_block *sb)
3815 {
3816         ext4_fsblk_t resv_clusters;
3817         struct ext4_sb_info *sbi = EXT4_SB(sb);
3818
3819         /*
3820          * There's no need to reserve anything when we aren't using extents.
3821          * The space estimates are exact, there are no unwritten extents,
3822          * hole punching doesn't need new metadata... This is needed especially
3823          * to keep ext2/3 backward compatibility.
3824          */
3825         if (!ext4_has_feature_extents(sb))
3826                 return;
3827         /*
3828          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3829          * This should cover the situations where we can not afford to run
3830          * out of space like for example punch hole, or converting
3831          * unwritten extents in delalloc path. In most cases such
3832          * allocation would require 1, or 2 blocks, higher numbers are
3833          * very rare.
3834          */
3835         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3836                          sbi->s_cluster_bits);
3837
3838         do_div(resv_clusters, 50);
3839         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3840
3841         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3842 }
3843
3844 static const char *ext4_quota_mode(struct super_block *sb)
3845 {
3846 #ifdef CONFIG_QUOTA
3847         if (!ext4_quota_capable(sb))
3848                 return "none";
3849
3850         if (EXT4_SB(sb)->s_journal && ext4_is_quota_journalled(sb))
3851                 return "journalled";
3852         else
3853                 return "writeback";
3854 #else
3855         return "disabled";
3856 #endif
3857 }
3858
3859 static void ext4_setup_csum_trigger(struct super_block *sb,
3860                                     enum ext4_journal_trigger_type type,
3861                                     void (*trigger)(
3862                                         struct jbd2_buffer_trigger_type *type,
3863                                         struct buffer_head *bh,
3864                                         void *mapped_data,
3865                                         size_t size))
3866 {
3867         struct ext4_sb_info *sbi = EXT4_SB(sb);
3868
3869         sbi->s_journal_triggers[type].sb = sb;
3870         sbi->s_journal_triggers[type].tr_triggers.t_frozen = trigger;
3871 }
3872
3873 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3874 {
3875         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3876         char *orig_data = kstrdup(data, GFP_KERNEL);
3877         struct buffer_head *bh, **group_desc;
3878         struct ext4_super_block *es = NULL;
3879         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3880         struct flex_groups **flex_groups;
3881         ext4_fsblk_t block;
3882         ext4_fsblk_t sb_block = get_sb_block(&data);
3883         ext4_fsblk_t logical_sb_block;
3884         unsigned long offset = 0;
3885         unsigned long def_mount_opts;
3886         struct inode *root;
3887         const char *descr;
3888         int ret = -ENOMEM;
3889         int blocksize, clustersize;
3890         unsigned int db_count;
3891         unsigned int i;
3892         int needs_recovery, has_huge_files;
3893         __u64 blocks_count;
3894         int err = 0;
3895         ext4_group_t first_not_zeroed;
3896         struct ext4_parsed_options parsed_opts;
3897
3898         /* Set defaults for the variables that will be set during parsing */
3899         parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3900         parsed_opts.journal_devnum = 0;
3901         parsed_opts.mb_optimize_scan = DEFAULT_MB_OPTIMIZE_SCAN;
3902
3903         if ((data && !orig_data) || !sbi)
3904                 goto out_free_base;
3905
3906         sbi->s_daxdev = dax_dev;
3907         sbi->s_blockgroup_lock =
3908                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3909         if (!sbi->s_blockgroup_lock)
3910                 goto out_free_base;
3911
3912         sb->s_fs_info = sbi;
3913         sbi->s_sb = sb;
3914         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3915         sbi->s_sb_block = sb_block;
3916         sbi->s_sectors_written_start =
3917                 part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
3918
3919         /* Cleanup superblock name */
3920         strreplace(sb->s_id, '/', '!');
3921
3922         /* -EINVAL is default */
3923         ret = -EINVAL;
3924         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3925         if (!blocksize) {
3926                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3927                 goto out_fail;
3928         }
3929
3930         /*
3931          * The ext4 superblock will not be buffer aligned for other than 1kB
3932          * block sizes.  We need to calculate the offset from buffer start.
3933          */
3934         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3935                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3936                 offset = do_div(logical_sb_block, blocksize);
3937         } else {
3938                 logical_sb_block = sb_block;
3939         }
3940
3941         bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
3942         if (IS_ERR(bh)) {
3943                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3944                 ret = PTR_ERR(bh);
3945                 goto out_fail;
3946         }
3947         /*
3948          * Note: s_es must be initialized as soon as possible because
3949          *       some ext4 macro-instructions depend on its value
3950          */
3951         es = (struct ext4_super_block *) (bh->b_data + offset);
3952         sbi->s_es = es;
3953         sb->s_magic = le16_to_cpu(es->s_magic);
3954         if (sb->s_magic != EXT4_SUPER_MAGIC)
3955                 goto cantfind_ext4;
3956         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3957
3958         /* Warn if metadata_csum and gdt_csum are both set. */
3959         if (ext4_has_feature_metadata_csum(sb) &&
3960             ext4_has_feature_gdt_csum(sb))
3961                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3962                              "redundant flags; please run fsck.");
3963
3964         /* Check for a known checksum algorithm */
3965         if (!ext4_verify_csum_type(sb, es)) {
3966                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3967                          "unknown checksum algorithm.");
3968                 silent = 1;
3969                 goto cantfind_ext4;
3970         }
3971         ext4_setup_csum_trigger(sb, EXT4_JTR_ORPHAN_FILE,
3972                                 ext4_orphan_file_block_trigger);
3973
3974         /* Load the checksum driver */
3975         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3976         if (IS_ERR(sbi->s_chksum_driver)) {
3977                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3978                 ret = PTR_ERR(sbi->s_chksum_driver);
3979                 sbi->s_chksum_driver = NULL;
3980                 goto failed_mount;
3981         }
3982
3983         /* Check superblock checksum */
3984         if (!ext4_superblock_csum_verify(sb, es)) {
3985                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3986                          "invalid superblock checksum.  Run e2fsck?");
3987                 silent = 1;
3988                 ret = -EFSBADCRC;
3989                 goto cantfind_ext4;
3990         }
3991
3992         /* Precompute checksum seed for all metadata */
3993         if (ext4_has_feature_csum_seed(sb))
3994                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3995         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3996                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3997                                                sizeof(es->s_uuid));
3998
3999         /* Set defaults before we parse the mount options */
4000         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
4001         set_opt(sb, INIT_INODE_TABLE);
4002         if (def_mount_opts & EXT4_DEFM_DEBUG)
4003                 set_opt(sb, DEBUG);
4004         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
4005                 set_opt(sb, GRPID);
4006         if (def_mount_opts & EXT4_DEFM_UID16)
4007                 set_opt(sb, NO_UID32);
4008         /* xattr user namespace & acls are now defaulted on */
4009         set_opt(sb, XATTR_USER);
4010 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4011         set_opt(sb, POSIX_ACL);
4012 #endif
4013         if (ext4_has_feature_fast_commit(sb))
4014                 set_opt2(sb, JOURNAL_FAST_COMMIT);
4015         /* don't forget to enable journal_csum when metadata_csum is enabled. */
4016         if (ext4_has_metadata_csum(sb))
4017                 set_opt(sb, JOURNAL_CHECKSUM);
4018
4019         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
4020                 set_opt(sb, JOURNAL_DATA);
4021         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
4022                 set_opt(sb, ORDERED_DATA);
4023         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
4024                 set_opt(sb, WRITEBACK_DATA);
4025
4026         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
4027                 set_opt(sb, ERRORS_PANIC);
4028         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
4029                 set_opt(sb, ERRORS_CONT);
4030         else
4031                 set_opt(sb, ERRORS_RO);
4032         /* block_validity enabled by default; disable with noblock_validity */
4033         set_opt(sb, BLOCK_VALIDITY);
4034         if (def_mount_opts & EXT4_DEFM_DISCARD)
4035                 set_opt(sb, DISCARD);
4036
4037         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
4038         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
4039         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
4040         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
4041         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
4042
4043         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
4044                 set_opt(sb, BARRIER);
4045
4046         /*
4047          * enable delayed allocation by default
4048          * Use -o nodelalloc to turn it off
4049          */
4050         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
4051             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
4052                 set_opt(sb, DELALLOC);
4053
4054         /*
4055          * set default s_li_wait_mult for lazyinit, for the case there is
4056          * no mount option specified.
4057          */
4058         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
4059
4060         if (le32_to_cpu(es->s_log_block_size) >
4061             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4062                 ext4_msg(sb, KERN_ERR,
4063                          "Invalid log block size: %u",
4064                          le32_to_cpu(es->s_log_block_size));
4065                 goto failed_mount;
4066         }
4067         if (le32_to_cpu(es->s_log_cluster_size) >
4068             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4069                 ext4_msg(sb, KERN_ERR,
4070                          "Invalid log cluster size: %u",
4071                          le32_to_cpu(es->s_log_cluster_size));
4072                 goto failed_mount;
4073         }
4074
4075         blocksize = EXT4_MIN_BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
4076
4077         if (blocksize == PAGE_SIZE)
4078                 set_opt(sb, DIOREAD_NOLOCK);
4079
4080         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
4081                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
4082                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
4083         } else {
4084                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
4085                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
4086                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
4087                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
4088                                  sbi->s_first_ino);
4089                         goto failed_mount;
4090                 }
4091                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
4092                     (!is_power_of_2(sbi->s_inode_size)) ||
4093                     (sbi->s_inode_size > blocksize)) {
4094                         ext4_msg(sb, KERN_ERR,
4095                                "unsupported inode size: %d",
4096                                sbi->s_inode_size);
4097                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
4098                         goto failed_mount;
4099                 }
4100                 /*
4101                  * i_atime_extra is the last extra field available for
4102                  * [acm]times in struct ext4_inode. Checking for that
4103                  * field should suffice to ensure we have extra space
4104                  * for all three.
4105                  */
4106                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
4107                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
4108                         sb->s_time_gran = 1;
4109                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
4110                 } else {
4111                         sb->s_time_gran = NSEC_PER_SEC;
4112                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
4113                 }
4114                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
4115         }
4116         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4117                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4118                         EXT4_GOOD_OLD_INODE_SIZE;
4119                 if (ext4_has_feature_extra_isize(sb)) {
4120                         unsigned v, max = (sbi->s_inode_size -
4121                                            EXT4_GOOD_OLD_INODE_SIZE);
4122
4123                         v = le16_to_cpu(es->s_want_extra_isize);
4124                         if (v > max) {
4125                                 ext4_msg(sb, KERN_ERR,
4126                                          "bad s_want_extra_isize: %d", v);
4127                                 goto failed_mount;
4128                         }
4129                         if (sbi->s_want_extra_isize < v)
4130                                 sbi->s_want_extra_isize = v;
4131
4132                         v = le16_to_cpu(es->s_min_extra_isize);
4133                         if (v > max) {
4134                                 ext4_msg(sb, KERN_ERR,
4135                                          "bad s_min_extra_isize: %d", v);
4136                                 goto failed_mount;
4137                         }
4138                         if (sbi->s_want_extra_isize < v)
4139                                 sbi->s_want_extra_isize = v;
4140                 }
4141         }
4142
4143         if (sbi->s_es->s_mount_opts[0]) {
4144                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
4145                                               sizeof(sbi->s_es->s_mount_opts),
4146                                               GFP_KERNEL);
4147                 if (!s_mount_opts)
4148                         goto failed_mount;
4149                 if (!parse_options(s_mount_opts, sb, &parsed_opts, 0)) {
4150                         ext4_msg(sb, KERN_WARNING,
4151                                  "failed to parse options in superblock: %s",
4152                                  s_mount_opts);
4153                 }
4154                 kfree(s_mount_opts);
4155         }
4156         sbi->s_def_mount_opt = sbi->s_mount_opt;
4157         if (!parse_options((char *) data, sb, &parsed_opts, 0))
4158                 goto failed_mount;
4159
4160 #ifdef CONFIG_UNICODE
4161         if (ext4_has_feature_casefold(sb) && !sb->s_encoding) {
4162                 const struct ext4_sb_encodings *encoding_info;
4163                 struct unicode_map *encoding;
4164                 __u16 encoding_flags;
4165
4166                 if (ext4_sb_read_encoding(es, &encoding_info,
4167                                           &encoding_flags)) {
4168                         ext4_msg(sb, KERN_ERR,
4169                                  "Encoding requested by superblock is unknown");
4170                         goto failed_mount;
4171                 }
4172
4173                 encoding = utf8_load(encoding_info->version);
4174                 if (IS_ERR(encoding)) {
4175                         ext4_msg(sb, KERN_ERR,
4176                                  "can't mount with superblock charset: %s-%s "
4177                                  "not supported by the kernel. flags: 0x%x.",
4178                                  encoding_info->name, encoding_info->version,
4179                                  encoding_flags);
4180                         goto failed_mount;
4181                 }
4182                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
4183                          "%s-%s with flags 0x%hx", encoding_info->name,
4184                          encoding_info->version?:"\b", encoding_flags);
4185
4186                 sb->s_encoding = encoding;
4187                 sb->s_encoding_flags = encoding_flags;
4188         }
4189 #endif
4190
4191         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4192                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting with data=journal disables delayed allocation, dioread_nolock, O_DIRECT and fast_commit support!\n");
4193                 /* can't mount with both data=journal and dioread_nolock. */
4194                 clear_opt(sb, DIOREAD_NOLOCK);
4195                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4196                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4197                         ext4_msg(sb, KERN_ERR, "can't mount with "
4198                                  "both data=journal and delalloc");
4199                         goto failed_mount;
4200                 }
4201                 if (test_opt(sb, DAX_ALWAYS)) {
4202                         ext4_msg(sb, KERN_ERR, "can't mount with "
4203                                  "both data=journal and dax");
4204                         goto failed_mount;
4205                 }
4206                 if (ext4_has_feature_encrypt(sb)) {
4207                         ext4_msg(sb, KERN_WARNING,
4208                                  "encrypted files will use data=ordered "
4209                                  "instead of data journaling mode");
4210                 }
4211                 if (test_opt(sb, DELALLOC))
4212                         clear_opt(sb, DELALLOC);
4213         } else {
4214                 sb->s_iflags |= SB_I_CGROUPWB;
4215         }
4216
4217         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
4218                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
4219
4220         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
4221             (ext4_has_compat_features(sb) ||
4222              ext4_has_ro_compat_features(sb) ||
4223              ext4_has_incompat_features(sb)))
4224                 ext4_msg(sb, KERN_WARNING,
4225                        "feature flags set on rev 0 fs, "
4226                        "running e2fsck is recommended");
4227
4228         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
4229                 set_opt2(sb, HURD_COMPAT);
4230                 if (ext4_has_feature_64bit(sb)) {
4231                         ext4_msg(sb, KERN_ERR,
4232                                  "The Hurd can't support 64-bit file systems");
4233                         goto failed_mount;
4234                 }
4235
4236                 /*
4237                  * ea_inode feature uses l_i_version field which is not
4238                  * available in HURD_COMPAT mode.
4239                  */
4240                 if (ext4_has_feature_ea_inode(sb)) {
4241                         ext4_msg(sb, KERN_ERR,
4242                                  "ea_inode feature is not supported for Hurd");
4243                         goto failed_mount;
4244                 }
4245         }
4246
4247         if (IS_EXT2_SB(sb)) {
4248                 if (ext2_feature_set_ok(sb))
4249                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4250                                  "using the ext4 subsystem");
4251                 else {
4252                         /*
4253                          * If we're probing be silent, if this looks like
4254                          * it's actually an ext[34] filesystem.
4255                          */
4256                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4257                                 goto failed_mount;
4258                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4259                                  "to feature incompatibilities");
4260                         goto failed_mount;
4261                 }
4262         }
4263
4264         if (IS_EXT3_SB(sb)) {
4265                 if (ext3_feature_set_ok(sb))
4266                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4267                                  "using the ext4 subsystem");
4268                 else {
4269                         /*
4270                          * If we're probing be silent, if this looks like
4271                          * it's actually an ext4 filesystem.
4272                          */
4273                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4274                                 goto failed_mount;
4275                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4276                                  "to feature incompatibilities");
4277                         goto failed_mount;
4278                 }
4279         }
4280
4281         /*
4282          * Check feature flags regardless of the revision level, since we
4283          * previously didn't change the revision level when setting the flags,
4284          * so there is a chance incompat flags are set on a rev 0 filesystem.
4285          */
4286         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4287                 goto failed_mount;
4288
4289         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4290                 ext4_msg(sb, KERN_ERR,
4291                          "Number of reserved GDT blocks insanely large: %d",
4292                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4293                 goto failed_mount;
4294         }
4295
4296         if (dax_supported(dax_dev, sb->s_bdev, blocksize, 0,
4297                         bdev_nr_sectors(sb->s_bdev)))
4298                 set_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags);
4299
4300         if (sbi->s_mount_opt & EXT4_MOUNT_DAX_ALWAYS) {
4301                 if (ext4_has_feature_inline_data(sb)) {
4302                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4303                                         " that may contain inline data");
4304                         goto failed_mount;
4305                 }
4306                 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) {
4307                         ext4_msg(sb, KERN_ERR,
4308                                 "DAX unsupported by block device.");
4309                         goto failed_mount;
4310                 }
4311         }
4312
4313         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4314                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4315                          es->s_encryption_level);
4316                 goto failed_mount;
4317         }
4318
4319         if (sb->s_blocksize != blocksize) {
4320                 /*
4321                  * bh must be released before kill_bdev(), otherwise
4322                  * it won't be freed and its page also. kill_bdev()
4323                  * is called by sb_set_blocksize().
4324                  */
4325                 brelse(bh);
4326                 /* Validate the filesystem blocksize */
4327                 if (!sb_set_blocksize(sb, blocksize)) {
4328                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4329                                         blocksize);
4330                         bh = NULL;
4331                         goto failed_mount;
4332                 }
4333
4334                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4335                 offset = do_div(logical_sb_block, blocksize);
4336                 bh = ext4_sb_bread_unmovable(sb, logical_sb_block);
4337                 if (IS_ERR(bh)) {
4338                         ext4_msg(sb, KERN_ERR,
4339                                "Can't read superblock on 2nd try");
4340                         ret = PTR_ERR(bh);
4341                         bh = NULL;
4342                         goto failed_mount;
4343                 }
4344                 es = (struct ext4_super_block *)(bh->b_data + offset);
4345                 sbi->s_es = es;
4346                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4347                         ext4_msg(sb, KERN_ERR,
4348                                "Magic mismatch, very weird!");
4349                         goto failed_mount;
4350                 }
4351         }
4352
4353         has_huge_files = ext4_has_feature_huge_file(sb);
4354         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4355                                                       has_huge_files);
4356         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4357
4358         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4359         if (ext4_has_feature_64bit(sb)) {
4360                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4361                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4362                     !is_power_of_2(sbi->s_desc_size)) {
4363                         ext4_msg(sb, KERN_ERR,
4364                                "unsupported descriptor size %lu",
4365                                sbi->s_desc_size);
4366                         goto failed_mount;
4367                 }
4368         } else
4369                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4370
4371         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4372         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4373
4374         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4375         if (sbi->s_inodes_per_block == 0)
4376                 goto cantfind_ext4;
4377         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4378             sbi->s_inodes_per_group > blocksize * 8) {
4379                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4380                          sbi->s_inodes_per_group);
4381                 goto failed_mount;
4382         }
4383         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4384                                         sbi->s_inodes_per_block;
4385         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4386         sbi->s_sbh = bh;
4387         sbi->s_mount_state = le16_to_cpu(es->s_state);
4388         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4389         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4390
4391         for (i = 0; i < 4; i++)
4392                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4393         sbi->s_def_hash_version = es->s_def_hash_version;
4394         if (ext4_has_feature_dir_index(sb)) {
4395                 i = le32_to_cpu(es->s_flags);
4396                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4397                         sbi->s_hash_unsigned = 3;
4398                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4399 #ifdef __CHAR_UNSIGNED__
4400                         if (!sb_rdonly(sb))
4401                                 es->s_flags |=
4402                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4403                         sbi->s_hash_unsigned = 3;
4404 #else
4405                         if (!sb_rdonly(sb))
4406                                 es->s_flags |=
4407                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4408 #endif
4409                 }
4410         }
4411
4412         /* Handle clustersize */
4413         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4414         if (ext4_has_feature_bigalloc(sb)) {
4415                 if (clustersize < blocksize) {
4416                         ext4_msg(sb, KERN_ERR,
4417                                  "cluster size (%d) smaller than "
4418                                  "block size (%d)", clustersize, blocksize);
4419                         goto failed_mount;
4420                 }
4421                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4422                         le32_to_cpu(es->s_log_block_size);
4423                 sbi->s_clusters_per_group =
4424                         le32_to_cpu(es->s_clusters_per_group);
4425                 if (sbi->s_clusters_per_group > blocksize * 8) {
4426                         ext4_msg(sb, KERN_ERR,
4427                                  "#clusters per group too big: %lu",
4428                                  sbi->s_clusters_per_group);
4429                         goto failed_mount;
4430                 }
4431                 if (sbi->s_blocks_per_group !=
4432                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4433                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4434                                  "clusters per group (%lu) inconsistent",
4435                                  sbi->s_blocks_per_group,
4436                                  sbi->s_clusters_per_group);
4437                         goto failed_mount;
4438                 }
4439         } else {
4440                 if (clustersize != blocksize) {
4441                         ext4_msg(sb, KERN_ERR,
4442                                  "fragment/cluster size (%d) != "
4443                                  "block size (%d)", clustersize, blocksize);
4444                         goto failed_mount;
4445                 }
4446                 if (sbi->s_blocks_per_group > blocksize * 8) {
4447                         ext4_msg(sb, KERN_ERR,
4448                                  "#blocks per group too big: %lu",
4449                                  sbi->s_blocks_per_group);
4450                         goto failed_mount;
4451                 }
4452                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4453                 sbi->s_cluster_bits = 0;
4454         }
4455         sbi->s_cluster_ratio = clustersize / blocksize;
4456
4457         /* Do we have standard group size of clustersize * 8 blocks ? */
4458         if (sbi->s_blocks_per_group == clustersize << 3)
4459                 set_opt2(sb, STD_GROUP_SIZE);
4460
4461         /*
4462          * Test whether we have more sectors than will fit in sector_t,
4463          * and whether the max offset is addressable by the page cache.
4464          */
4465         err = generic_check_addressable(sb->s_blocksize_bits,
4466                                         ext4_blocks_count(es));
4467         if (err) {
4468                 ext4_msg(sb, KERN_ERR, "filesystem"
4469                          " too large to mount safely on this system");
4470                 goto failed_mount;
4471         }
4472
4473         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4474                 goto cantfind_ext4;
4475
4476         /* check blocks count against device size */
4477         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4478         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4479                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4480                        "exceeds size of device (%llu blocks)",
4481                        ext4_blocks_count(es), blocks_count);
4482                 goto failed_mount;
4483         }
4484
4485         /*
4486          * It makes no sense for the first data block to be beyond the end
4487          * of the filesystem.
4488          */
4489         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4490                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4491                          "block %u is beyond end of filesystem (%llu)",
4492                          le32_to_cpu(es->s_first_data_block),
4493                          ext4_blocks_count(es));
4494                 goto failed_mount;
4495         }
4496         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4497             (sbi->s_cluster_ratio == 1)) {
4498                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4499                          "block is 0 with a 1k block and cluster size");
4500                 goto failed_mount;
4501         }
4502
4503         blocks_count = (ext4_blocks_count(es) -
4504                         le32_to_cpu(es->s_first_data_block) +
4505                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4506         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4507         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4508                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4509                        "(block count %llu, first data block %u, "
4510                        "blocks per group %lu)", blocks_count,
4511                        ext4_blocks_count(es),
4512                        le32_to_cpu(es->s_first_data_block),
4513                        EXT4_BLOCKS_PER_GROUP(sb));
4514                 goto failed_mount;
4515         }
4516         sbi->s_groups_count = blocks_count;
4517         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4518                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4519         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4520             le32_to_cpu(es->s_inodes_count)) {
4521                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4522                          le32_to_cpu(es->s_inodes_count),
4523                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4524                 ret = -EINVAL;
4525                 goto failed_mount;
4526         }
4527         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4528                    EXT4_DESC_PER_BLOCK(sb);
4529         if (ext4_has_feature_meta_bg(sb)) {
4530                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4531                         ext4_msg(sb, KERN_WARNING,
4532                                  "first meta block group too large: %u "
4533                                  "(group descriptor block count %u)",
4534                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4535                         goto failed_mount;
4536                 }
4537         }
4538         rcu_assign_pointer(sbi->s_group_desc,
4539                            kvmalloc_array(db_count,
4540                                           sizeof(struct buffer_head *),
4541                                           GFP_KERNEL));
4542         if (sbi->s_group_desc == NULL) {
4543                 ext4_msg(sb, KERN_ERR, "not enough memory");
4544                 ret = -ENOMEM;
4545                 goto failed_mount;
4546         }
4547
4548         bgl_lock_init(sbi->s_blockgroup_lock);
4549
4550         /* Pre-read the descriptors into the buffer cache */
4551         for (i = 0; i < db_count; i++) {
4552                 block = descriptor_loc(sb, logical_sb_block, i);
4553                 ext4_sb_breadahead_unmovable(sb, block);
4554         }
4555
4556         for (i = 0; i < db_count; i++) {
4557                 struct buffer_head *bh;
4558
4559                 block = descriptor_loc(sb, logical_sb_block, i);
4560                 bh = ext4_sb_bread_unmovable(sb, block);
4561                 if (IS_ERR(bh)) {
4562                         ext4_msg(sb, KERN_ERR,
4563                                "can't read group descriptor %d", i);
4564                         db_count = i;
4565                         ret = PTR_ERR(bh);
4566                         goto failed_mount2;
4567                 }
4568                 rcu_read_lock();
4569                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4570                 rcu_read_unlock();
4571         }
4572         sbi->s_gdb_count = db_count;
4573         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4574                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4575                 ret = -EFSCORRUPTED;
4576                 goto failed_mount2;
4577         }
4578
4579         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4580         spin_lock_init(&sbi->s_error_lock);
4581         INIT_WORK(&sbi->s_error_work, flush_stashed_error_work);
4582
4583         /* Register extent status tree shrinker */
4584         if (ext4_es_register_shrinker(sbi))
4585                 goto failed_mount3;
4586
4587         sbi->s_stripe = ext4_get_stripe_size(sbi);
4588         sbi->s_extent_max_zeroout_kb = 32;
4589
4590         /*
4591          * set up enough so that it can read an inode
4592          */
4593         sb->s_op = &ext4_sops;
4594         sb->s_export_op = &ext4_export_ops;
4595         sb->s_xattr = ext4_xattr_handlers;
4596 #ifdef CONFIG_FS_ENCRYPTION
4597         sb->s_cop = &ext4_cryptops;
4598 #endif
4599 #ifdef CONFIG_FS_VERITY
4600         sb->s_vop = &ext4_verityops;
4601 #endif
4602 #ifdef CONFIG_QUOTA
4603         sb->dq_op = &ext4_quota_operations;
4604         if (ext4_has_feature_quota(sb))
4605                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4606         else
4607                 sb->s_qcop = &ext4_qctl_operations;
4608         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4609 #endif
4610         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4611
4612         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4613         mutex_init(&sbi->s_orphan_lock);
4614
4615         /* Initialize fast commit stuff */
4616         atomic_set(&sbi->s_fc_subtid, 0);
4617         atomic_set(&sbi->s_fc_ineligible_updates, 0);
4618         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_MAIN]);
4619         INIT_LIST_HEAD(&sbi->s_fc_q[FC_Q_STAGING]);
4620         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_MAIN]);
4621         INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
4622         sbi->s_fc_bytes = 0;
4623         ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
4624         ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
4625         spin_lock_init(&sbi->s_fc_lock);
4626         memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));
4627         sbi->s_fc_replay_state.fc_regions = NULL;
4628         sbi->s_fc_replay_state.fc_regions_size = 0;
4629         sbi->s_fc_replay_state.fc_regions_used = 0;
4630         sbi->s_fc_replay_state.fc_regions_valid = 0;
4631         sbi->s_fc_replay_state.fc_modified_inodes = NULL;
4632         sbi->s_fc_replay_state.fc_modified_inodes_size = 0;
4633         sbi->s_fc_replay_state.fc_modified_inodes_used = 0;
4634
4635         sb->s_root = NULL;
4636
4637         needs_recovery = (es->s_last_orphan != 0 ||
4638                           ext4_has_feature_orphan_present(sb) ||
4639                           ext4_has_feature_journal_needs_recovery(sb));
4640
4641         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4642                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4643                         goto failed_mount3a;
4644
4645         /*
4646          * The first inode we look at is the journal inode.  Don't try
4647          * root first: it may be modified in the journal!
4648          */
4649         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4650                 err = ext4_load_journal(sb, es, parsed_opts.journal_devnum);
4651                 if (err)
4652                         goto failed_mount3a;
4653         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4654                    ext4_has_feature_journal_needs_recovery(sb)) {
4655                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4656                        "suppressed and not mounted read-only");
4657                 goto failed_mount_wq;
4658         } else {
4659                 /* Nojournal mode, all journal mount options are illegal */
4660                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4661                         ext4_msg(sb, KERN_ERR, "can't mount with "
4662                                  "journal_checksum, fs mounted w/o journal");
4663                         goto failed_mount_wq;
4664                 }
4665                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4666                         ext4_msg(sb, KERN_ERR, "can't mount with "
4667                                  "journal_async_commit, fs mounted w/o journal");
4668                         goto failed_mount_wq;
4669                 }
4670                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4671                         ext4_msg(sb, KERN_ERR, "can't mount with "
4672                                  "commit=%lu, fs mounted w/o journal",
4673                                  sbi->s_commit_interval / HZ);
4674                         goto failed_mount_wq;
4675                 }
4676                 if (EXT4_MOUNT_DATA_FLAGS &
4677                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4678                         ext4_msg(sb, KERN_ERR, "can't mount with "
4679                                  "data=, fs mounted w/o journal");
4680                         goto failed_mount_wq;
4681                 }
4682                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4683                 clear_opt(sb, JOURNAL_CHECKSUM);
4684                 clear_opt(sb, DATA_FLAGS);
4685                 clear_opt2(sb, JOURNAL_FAST_COMMIT);
4686                 sbi->s_journal = NULL;
4687                 needs_recovery = 0;
4688                 goto no_journal;
4689         }
4690
4691         if (ext4_has_feature_64bit(sb) &&
4692             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4693                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4694                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4695                 goto failed_mount_wq;
4696         }
4697
4698         if (!set_journal_csum_feature_set(sb)) {
4699                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4700                          "feature set");
4701                 goto failed_mount_wq;
4702         }
4703
4704         if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4705                 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4706                                           JBD2_FEATURE_INCOMPAT_FAST_COMMIT)) {
4707                 ext4_msg(sb, KERN_ERR,
4708                         "Failed to set fast commit journal feature");
4709                 goto failed_mount_wq;
4710         }
4711
4712         /* We have now updated the journal if required, so we can
4713          * validate the data journaling mode. */
4714         switch (test_opt(sb, DATA_FLAGS)) {
4715         case 0:
4716                 /* No mode set, assume a default based on the journal
4717                  * capabilities: ORDERED_DATA if the journal can
4718                  * cope, else JOURNAL_DATA
4719                  */
4720                 if (jbd2_journal_check_available_features
4721                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4722                         set_opt(sb, ORDERED_DATA);
4723                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4724                 } else {
4725                         set_opt(sb, JOURNAL_DATA);
4726                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4727                 }
4728                 break;
4729
4730         case EXT4_MOUNT_ORDERED_DATA:
4731         case EXT4_MOUNT_WRITEBACK_DATA:
4732                 if (!jbd2_journal_check_available_features
4733                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4734                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4735                                "requested data journaling mode");
4736                         goto failed_mount_wq;
4737                 }
4738                 break;
4739         default:
4740                 break;
4741         }
4742
4743         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4744             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4745                 ext4_msg(sb, KERN_ERR, "can't mount with "
4746                         "journal_async_commit in data=ordered mode");
4747                 goto failed_mount_wq;
4748         }
4749
4750         set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio);
4751
4752         sbi->s_journal->j_submit_inode_data_buffers =
4753                 ext4_journal_submit_inode_data_buffers;
4754         sbi->s_journal->j_finish_inode_data_buffers =
4755                 ext4_journal_finish_inode_data_buffers;
4756
4757 no_journal:
4758         if (!test_opt(sb, NO_MBCACHE)) {
4759                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4760                 if (!sbi->s_ea_block_cache) {
4761                         ext4_msg(sb, KERN_ERR,
4762                                  "Failed to create ea_block_cache");
4763                         goto failed_mount_wq;
4764                 }
4765
4766                 if (ext4_has_feature_ea_inode(sb)) {
4767                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4768                         if (!sbi->s_ea_inode_cache) {
4769                                 ext4_msg(sb, KERN_ERR,
4770                                          "Failed to create ea_inode_cache");
4771                                 goto failed_mount_wq;
4772                         }
4773                 }
4774         }
4775
4776         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4777                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4778                 goto failed_mount_wq;
4779         }
4780
4781         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4782             !ext4_has_feature_encrypt(sb)) {
4783                 ext4_set_feature_encrypt(sb);
4784                 ext4_commit_super(sb);
4785         }
4786
4787         /*
4788          * Get the # of file system overhead blocks from the
4789          * superblock if present.
4790          */
4791         if (es->s_overhead_clusters)
4792                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4793         else {
4794                 err = ext4_calculate_overhead(sb);
4795                 if (err)
4796                         goto failed_mount_wq;
4797         }
4798
4799         /*
4800          * The maximum number of concurrent works can be high and
4801          * concurrency isn't really necessary.  Limit it to 1.
4802          */
4803         EXT4_SB(sb)->rsv_conversion_wq =
4804                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4805         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4806                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4807                 ret = -ENOMEM;
4808                 goto failed_mount4;
4809         }
4810
4811         /*
4812          * The jbd2_journal_load will have done any necessary log recovery,
4813          * so we can safely mount the rest of the filesystem now.
4814          */
4815
4816         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4817         if (IS_ERR(root)) {
4818                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4819                 ret = PTR_ERR(root);
4820                 root = NULL;
4821                 goto failed_mount4;
4822         }
4823         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4824                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4825                 iput(root);
4826                 goto failed_mount4;
4827         }
4828
4829         sb->s_root = d_make_root(root);
4830         if (!sb->s_root) {
4831                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4832                 ret = -ENOMEM;
4833                 goto failed_mount4;
4834         }
4835
4836         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4837         if (ret == -EROFS) {
4838                 sb->s_flags |= SB_RDONLY;
4839                 ret = 0;
4840         } else if (ret)
4841                 goto failed_mount4a;
4842
4843         ext4_set_resv_clusters(sb);
4844
4845         if (test_opt(sb, BLOCK_VALIDITY)) {
4846                 err = ext4_setup_system_zone(sb);
4847                 if (err) {
4848                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4849                                  "zone (%d)", err);
4850                         goto failed_mount4a;
4851                 }
4852         }
4853         ext4_fc_replay_cleanup(sb);
4854
4855         ext4_ext_init(sb);
4856
4857         /*
4858          * Enable optimize_scan if number of groups is > threshold. This can be
4859          * turned off by passing "mb_optimize_scan=0". This can also be
4860          * turned on forcefully by passing "mb_optimize_scan=1".
4861          */
4862         if (parsed_opts.mb_optimize_scan == 1)
4863                 set_opt2(sb, MB_OPTIMIZE_SCAN);
4864         else if (parsed_opts.mb_optimize_scan == 0)
4865                 clear_opt2(sb, MB_OPTIMIZE_SCAN);
4866         else if (sbi->s_groups_count >= MB_DEFAULT_LINEAR_SCAN_THRESHOLD)
4867                 set_opt2(sb, MB_OPTIMIZE_SCAN);
4868
4869         err = ext4_mb_init(sb);
4870         if (err) {
4871                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4872                          err);
4873                 goto failed_mount5;
4874         }
4875
4876         /*
4877          * We can only set up the journal commit callback once
4878          * mballoc is initialized
4879          */
4880         if (sbi->s_journal)
4881                 sbi->s_journal->j_commit_callback =
4882                         ext4_journal_commit_callback;
4883
4884         block = ext4_count_free_clusters(sb);
4885         ext4_free_blocks_count_set(sbi->s_es,
4886                                    EXT4_C2B(sbi, block));
4887         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4888                                   GFP_KERNEL);
4889         if (!err) {
4890                 unsigned long freei = ext4_count_free_inodes(sb);
4891                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4892                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4893                                           GFP_KERNEL);
4894         }
4895         /*
4896          * Update the checksum after updating free space/inode
4897          * counters.  Otherwise the superblock can have an incorrect
4898          * checksum in the buffer cache until it is written out and
4899          * e2fsprogs programs trying to open a file system immediately
4900          * after it is mounted can fail.
4901          */
4902         ext4_superblock_csum_set(sb);
4903         if (!err)
4904                 err = percpu_counter_init(&sbi->s_dirs_counter,
4905                                           ext4_count_dirs(sb), GFP_KERNEL);
4906         if (!err)
4907                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4908                                           GFP_KERNEL);
4909         if (!err)
4910                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
4911                                           GFP_KERNEL);
4912         if (!err)
4913                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4914
4915         if (err) {
4916                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4917                 goto failed_mount6;
4918         }
4919
4920         if (ext4_has_feature_flex_bg(sb))
4921                 if (!ext4_fill_flex_info(sb)) {
4922                         ext4_msg(sb, KERN_ERR,
4923                                "unable to initialize "
4924                                "flex_bg meta info!");
4925                         ret = -ENOMEM;
4926                         goto failed_mount6;
4927                 }
4928
4929         err = ext4_register_li_request(sb, first_not_zeroed);
4930         if (err)
4931                 goto failed_mount6;
4932
4933         err = ext4_register_sysfs(sb);
4934         if (err)
4935                 goto failed_mount7;
4936
4937         err = ext4_init_orphan_info(sb);
4938         if (err)
4939                 goto failed_mount8;
4940 #ifdef CONFIG_QUOTA
4941         /* Enable quota usage during mount. */
4942         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4943                 err = ext4_enable_quotas(sb);
4944                 if (err)
4945                         goto failed_mount9;
4946         }
4947 #endif  /* CONFIG_QUOTA */
4948
4949         /*
4950          * Save the original bdev mapping's wb_err value which could be
4951          * used to detect the metadata async write error.
4952          */
4953         spin_lock_init(&sbi->s_bdev_wb_lock);
4954         errseq_check_and_advance(&sb->s_bdev->bd_inode->i_mapping->wb_err,
4955                                  &sbi->s_bdev_wb_err);
4956         sb->s_bdev->bd_super = sb;
4957         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4958         ext4_orphan_cleanup(sb, es);
4959         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4960         if (needs_recovery) {
4961                 ext4_msg(sb, KERN_INFO, "recovery complete");
4962                 err = ext4_mark_recovery_complete(sb, es);
4963                 if (err)
4964                         goto failed_mount9;
4965         }
4966         if (EXT4_SB(sb)->s_journal) {
4967                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4968                         descr = " journalled data mode";
4969                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4970                         descr = " ordered data mode";
4971                 else
4972                         descr = " writeback data mode";
4973         } else
4974                 descr = "out journal";
4975
4976         if (test_opt(sb, DISCARD)) {
4977                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4978                 if (!blk_queue_discard(q))
4979                         ext4_msg(sb, KERN_WARNING,
4980                                  "mounting with \"discard\" option, but "
4981                                  "the device does not support discard");
4982         }
4983
4984         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4985                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4986                          "Opts: %.*s%s%s. Quota mode: %s.", descr,
4987                          (int) sizeof(sbi->s_es->s_mount_opts),
4988                          sbi->s_es->s_mount_opts,
4989                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data,
4990                          ext4_quota_mode(sb));
4991
4992         if (es->s_error_count)
4993                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4994
4995         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4996         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4997         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4998         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4999         atomic_set(&sbi->s_warning_count, 0);
5000         atomic_set(&sbi->s_msg_count, 0);
5001
5002         kfree(orig_data);
5003         return 0;
5004
5005 cantfind_ext4:
5006         if (!silent)
5007                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
5008         goto failed_mount;
5009
5010 failed_mount9:
5011         ext4_release_orphan_info(sb);
5012 failed_mount8:
5013         ext4_unregister_sysfs(sb);
5014         kobject_put(&sbi->s_kobj);
5015 failed_mount7:
5016         ext4_unregister_li_request(sb);
5017 failed_mount6:
5018         ext4_mb_release(sb);
5019         rcu_read_lock();
5020         flex_groups = rcu_dereference(sbi->s_flex_groups);
5021         if (flex_groups) {
5022                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
5023                         kvfree(flex_groups[i]);
5024                 kvfree(flex_groups);
5025         }
5026         rcu_read_unlock();
5027         percpu_counter_destroy(&sbi->s_freeclusters_counter);
5028         percpu_counter_destroy(&sbi->s_freeinodes_counter);
5029         percpu_counter_destroy(&sbi->s_dirs_counter);
5030         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
5031         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
5032         percpu_free_rwsem(&sbi->s_writepages_rwsem);
5033 failed_mount5:
5034         ext4_ext_release(sb);
5035         ext4_release_system_zone(sb);
5036 failed_mount4a:
5037         dput(sb->s_root);
5038         sb->s_root = NULL;
5039 failed_mount4:
5040         ext4_msg(sb, KERN_ERR, "mount failed");
5041         if (EXT4_SB(sb)->rsv_conversion_wq)
5042                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
5043 failed_mount_wq:
5044         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
5045         sbi->s_ea_inode_cache = NULL;
5046
5047         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
5048         sbi->s_ea_block_cache = NULL;
5049
5050         if (sbi->s_journal) {
5051                 /* flush s_error_work before journal destroy. */
5052                 flush_work(&sbi->s_error_work);
5053                 jbd2_journal_destroy(sbi->s_journal);
5054                 sbi->s_journal = NULL;
5055         }
5056 failed_mount3a:
5057         ext4_es_unregister_shrinker(sbi);
5058 failed_mount3:
5059         /* flush s_error_work before sbi destroy */
5060         flush_work(&sbi->s_error_work);
5061         del_timer_sync(&sbi->s_err_report);
5062         ext4_stop_mmpd(sbi);
5063 failed_mount2:
5064         rcu_read_lock();
5065         group_desc = rcu_dereference(sbi->s_group_desc);
5066         for (i = 0; i < db_count; i++)
5067                 brelse(group_desc[i]);
5068         kvfree(group_desc);
5069         rcu_read_unlock();
5070 failed_mount:
5071         if (sbi->s_chksum_driver)
5072                 crypto_free_shash(sbi->s_chksum_driver);
5073
5074 #ifdef CONFIG_UNICODE
5075         utf8_unload(sb->s_encoding);
5076 #endif
5077
5078 #ifdef CONFIG_QUOTA
5079         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5080                 kfree(get_qf_name(sb, sbi, i));
5081 #endif
5082         fscrypt_free_dummy_policy(&sbi->s_dummy_enc_policy);
5083         /* ext4_blkdev_remove() calls kill_bdev(), release bh before it. */
5084         brelse(bh);
5085         ext4_blkdev_remove(sbi);
5086 out_fail:
5087         sb->s_fs_info = NULL;
5088         kfree(sbi->s_blockgroup_lock);
5089 out_free_base:
5090         kfree(sbi);
5091         kfree(orig_data);
5092         fs_put_dax(dax_dev);
5093         return err ? err : ret;
5094 }
5095
5096 /*
5097  * Setup any per-fs journal parameters now.  We'll do this both on
5098  * initial mount, once the journal has been initialised but before we've
5099  * done any recovery; and again on any subsequent remount.
5100  */
5101 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
5102 {
5103         struct ext4_sb_info *sbi = EXT4_SB(sb);
5104
5105         journal->j_commit_interval = sbi->s_commit_interval;
5106         journal->j_min_batch_time = sbi->s_min_batch_time;
5107         journal->j_max_batch_time = sbi->s_max_batch_time;
5108         ext4_fc_init(sb, journal);
5109
5110         write_lock(&journal->j_state_lock);
5111         if (test_opt(sb, BARRIER))
5112                 journal->j_flags |= JBD2_BARRIER;
5113         else
5114                 journal->j_flags &= ~JBD2_BARRIER;
5115         if (test_opt(sb, DATA_ERR_ABORT))
5116                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
5117         else
5118                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
5119         write_unlock(&journal->j_state_lock);
5120 }
5121
5122 static struct inode *ext4_get_journal_inode(struct super_block *sb,
5123                                              unsigned int journal_inum)
5124 {
5125         struct inode *journal_inode;
5126
5127         /*
5128          * Test for the existence of a valid inode on disk.  Bad things
5129          * happen if we iget() an unused inode, as the subsequent iput()
5130          * will try to delete it.
5131          */
5132         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
5133         if (IS_ERR(journal_inode)) {
5134                 ext4_msg(sb, KERN_ERR, "no journal found");
5135                 return NULL;
5136         }
5137         if (!journal_inode->i_nlink) {
5138                 make_bad_inode(journal_inode);
5139                 iput(journal_inode);
5140                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
5141                 return NULL;
5142         }
5143
5144         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
5145                   journal_inode, journal_inode->i_size);
5146         if (!S_ISREG(journal_inode->i_mode)) {
5147                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
5148                 iput(journal_inode);
5149                 return NULL;
5150         }
5151         return journal_inode;
5152 }
5153
5154 static journal_t *ext4_get_journal(struct super_block *sb,
5155                                    unsigned int journal_inum)
5156 {
5157         struct inode *journal_inode;
5158         journal_t *journal;
5159
5160         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5161                 return NULL;
5162
5163         journal_inode = ext4_get_journal_inode(sb, journal_inum);
5164         if (!journal_inode)
5165                 return NULL;
5166
5167         journal = jbd2_journal_init_inode(journal_inode);
5168         if (!journal) {
5169                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
5170                 iput(journal_inode);
5171                 return NULL;
5172         }
5173         journal->j_private = sb;
5174         ext4_init_journal_params(sb, journal);
5175         return journal;
5176 }
5177
5178 static journal_t *ext4_get_dev_journal(struct super_block *sb,
5179                                        dev_t j_dev)
5180 {
5181         struct buffer_head *bh;
5182         journal_t *journal;
5183         ext4_fsblk_t start;
5184         ext4_fsblk_t len;
5185         int hblock, blocksize;
5186         ext4_fsblk_t sb_block;
5187         unsigned long offset;
5188         struct ext4_super_block *es;
5189         struct block_device *bdev;
5190
5191         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5192                 return NULL;
5193
5194         bdev = ext4_blkdev_get(j_dev, sb);
5195         if (bdev == NULL)
5196                 return NULL;
5197
5198         blocksize = sb->s_blocksize;
5199         hblock = bdev_logical_block_size(bdev);
5200         if (blocksize < hblock) {
5201                 ext4_msg(sb, KERN_ERR,
5202                         "blocksize too small for journal device");
5203                 goto out_bdev;
5204         }
5205
5206         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
5207         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
5208         set_blocksize(bdev, blocksize);
5209         if (!(bh = __bread(bdev, sb_block, blocksize))) {
5210                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
5211                        "external journal");
5212                 goto out_bdev;
5213         }
5214
5215         es = (struct ext4_super_block *) (bh->b_data + offset);
5216         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
5217             !(le32_to_cpu(es->s_feature_incompat) &
5218               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
5219                 ext4_msg(sb, KERN_ERR, "external journal has "
5220                                         "bad superblock");
5221                 brelse(bh);
5222                 goto out_bdev;
5223         }
5224
5225         if ((le32_to_cpu(es->s_feature_ro_compat) &
5226              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
5227             es->s_checksum != ext4_superblock_csum(sb, es)) {
5228                 ext4_msg(sb, KERN_ERR, "external journal has "
5229                                        "corrupt superblock");
5230                 brelse(bh);
5231                 goto out_bdev;
5232         }
5233
5234         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
5235                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
5236                 brelse(bh);
5237                 goto out_bdev;
5238         }
5239
5240         len = ext4_blocks_count(es);
5241         start = sb_block + 1;
5242         brelse(bh);     /* we're done with the superblock */
5243
5244         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
5245                                         start, len, blocksize);
5246         if (!journal) {
5247                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
5248                 goto out_bdev;
5249         }
5250         journal->j_private = sb;
5251         if (ext4_read_bh_lock(journal->j_sb_buffer, REQ_META | REQ_PRIO, true)) {
5252                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
5253                 goto out_journal;
5254         }
5255         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
5256                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
5257                                         "user (unsupported) - %d",
5258                         be32_to_cpu(journal->j_superblock->s_nr_users));
5259                 goto out_journal;
5260         }
5261         EXT4_SB(sb)->s_journal_bdev = bdev;
5262         ext4_init_journal_params(sb, journal);
5263         return journal;
5264
5265 out_journal:
5266         jbd2_journal_destroy(journal);
5267 out_bdev:
5268         ext4_blkdev_put(bdev);
5269         return NULL;
5270 }
5271
5272 static int ext4_load_journal(struct super_block *sb,
5273                              struct ext4_super_block *es,
5274                              unsigned long journal_devnum)
5275 {
5276         journal_t *journal;
5277         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
5278         dev_t journal_dev;
5279         int err = 0;
5280         int really_read_only;
5281         int journal_dev_ro;
5282
5283         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
5284                 return -EFSCORRUPTED;
5285
5286         if (journal_devnum &&
5287             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5288                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
5289                         "numbers have changed");
5290                 journal_dev = new_decode_dev(journal_devnum);
5291         } else
5292                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
5293
5294         if (journal_inum && journal_dev) {
5295                 ext4_msg(sb, KERN_ERR,
5296                          "filesystem has both journal inode and journal device!");
5297                 return -EINVAL;
5298         }
5299
5300         if (journal_inum) {
5301                 journal = ext4_get_journal(sb, journal_inum);
5302                 if (!journal)
5303                         return -EINVAL;
5304         } else {
5305                 journal = ext4_get_dev_journal(sb, journal_dev);
5306                 if (!journal)
5307                         return -EINVAL;
5308         }
5309
5310         journal_dev_ro = bdev_read_only(journal->j_dev);
5311         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5312
5313         if (journal_dev_ro && !sb_rdonly(sb)) {
5314                 ext4_msg(sb, KERN_ERR,
5315                          "journal device read-only, try mounting with '-o ro'");
5316                 err = -EROFS;
5317                 goto err_out;
5318         }
5319
5320         /*
5321          * Are we loading a blank journal or performing recovery after a
5322          * crash?  For recovery, we need to check in advance whether we
5323          * can get read-write access to the device.
5324          */
5325         if (ext4_has_feature_journal_needs_recovery(sb)) {
5326                 if (sb_rdonly(sb)) {
5327                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5328                                         "required on readonly filesystem");
5329                         if (really_read_only) {
5330                                 ext4_msg(sb, KERN_ERR, "write access "
5331                                         "unavailable, cannot proceed "
5332                                         "(try mounting with noload)");
5333                                 err = -EROFS;
5334                                 goto err_out;
5335                         }
5336                         ext4_msg(sb, KERN_INFO, "write access will "
5337                                "be enabled during recovery");
5338                 }
5339         }
5340
5341         if (!(journal->j_flags & JBD2_BARRIER))
5342                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5343
5344         if (!ext4_has_feature_journal_needs_recovery(sb))
5345                 err = jbd2_journal_wipe(journal, !really_read_only);
5346         if (!err) {
5347                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5348                 if (save)
5349                         memcpy(save, ((char *) es) +
5350                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5351                 err = jbd2_journal_load(journal);
5352                 if (save)
5353                         memcpy(((char *) es) + EXT4_S_ERR_START,
5354                                save, EXT4_S_ERR_LEN);
5355                 kfree(save);
5356         }
5357
5358         if (err) {
5359                 ext4_msg(sb, KERN_ERR, "error loading journal");
5360                 goto err_out;
5361         }
5362
5363         EXT4_SB(sb)->s_journal = journal;
5364         err = ext4_clear_journal_err(sb, es);
5365         if (err) {
5366                 EXT4_SB(sb)->s_journal = NULL;
5367                 jbd2_journal_destroy(journal);
5368                 return err;
5369         }
5370
5371         if (!really_read_only && journal_devnum &&
5372             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5373                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5374
5375                 /* Make sure we flush the recovery flag to disk. */
5376                 ext4_commit_super(sb);
5377         }
5378
5379         return 0;
5380
5381 err_out:
5382         jbd2_journal_destroy(journal);
5383         return err;
5384 }
5385
5386 /* Copy state of EXT4_SB(sb) into buffer for on-disk superblock */
5387 static void ext4_update_super(struct super_block *sb)
5388 {
5389         struct ext4_sb_info *sbi = EXT4_SB(sb);
5390         struct ext4_super_block *es = sbi->s_es;
5391         struct buffer_head *sbh = sbi->s_sbh;
5392
5393         lock_buffer(sbh);
5394         /*
5395          * If the file system is mounted read-only, don't update the
5396          * superblock write time.  This avoids updating the superblock
5397          * write time when we are mounting the root file system
5398          * read/only but we need to replay the journal; at that point,
5399          * for people who are east of GMT and who make their clock
5400          * tick in localtime for Windows bug-for-bug compatibility,
5401          * the clock is set in the future, and this will cause e2fsck
5402          * to complain and force a full file system check.
5403          */
5404         if (!(sb->s_flags & SB_RDONLY))
5405                 ext4_update_tstamp(es, s_wtime);
5406         es->s_kbytes_written =
5407                 cpu_to_le64(sbi->s_kbytes_written +
5408                     ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
5409                       sbi->s_sectors_written_start) >> 1));
5410         if (percpu_counter_initialized(&sbi->s_freeclusters_counter))
5411                 ext4_free_blocks_count_set(es,
5412                         EXT4_C2B(sbi, percpu_counter_sum_positive(
5413                                 &sbi->s_freeclusters_counter)));
5414         if (percpu_counter_initialized(&sbi->s_freeinodes_counter))
5415                 es->s_free_inodes_count =
5416                         cpu_to_le32(percpu_counter_sum_positive(
5417                                 &sbi->s_freeinodes_counter));
5418         /* Copy error information to the on-disk superblock */
5419         spin_lock(&sbi->s_error_lock);
5420         if (sbi->s_add_error_count > 0) {
5421                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5422                 if (!es->s_first_error_time && !es->s_first_error_time_hi) {
5423                         __ext4_update_tstamp(&es->s_first_error_time,
5424                                              &es->s_first_error_time_hi,
5425                                              sbi->s_first_error_time);
5426                         strncpy(es->s_first_error_func, sbi->s_first_error_func,
5427                                 sizeof(es->s_first_error_func));
5428                         es->s_first_error_line =
5429                                 cpu_to_le32(sbi->s_first_error_line);
5430                         es->s_first_error_ino =
5431                                 cpu_to_le32(sbi->s_first_error_ino);
5432                         es->s_first_error_block =
5433                                 cpu_to_le64(sbi->s_first_error_block);
5434                         es->s_first_error_errcode =
5435                                 ext4_errno_to_code(sbi->s_first_error_code);
5436                 }
5437                 __ext4_update_tstamp(&es->s_last_error_time,
5438                                      &es->s_last_error_time_hi,
5439                                      sbi->s_last_error_time);
5440                 strncpy(es->s_last_error_func, sbi->s_last_error_func,
5441                         sizeof(es->s_last_error_func));
5442                 es->s_last_error_line = cpu_to_le32(sbi->s_last_error_line);
5443                 es->s_last_error_ino = cpu_to_le32(sbi->s_last_error_ino);
5444                 es->s_last_error_block = cpu_to_le64(sbi->s_last_error_block);
5445                 es->s_last_error_errcode =
5446                                 ext4_errno_to_code(sbi->s_last_error_code);
5447                 /*
5448                  * Start the daily error reporting function if it hasn't been
5449                  * started already
5450                  */
5451                 if (!es->s_error_count)
5452                         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);
5453                 le32_add_cpu(&es->s_error_count, sbi->s_add_error_count);
5454                 sbi->s_add_error_count = 0;
5455         }
5456         spin_unlock(&sbi->s_error_lock);
5457
5458         ext4_superblock_csum_set(sb);
5459         unlock_buffer(sbh);
5460 }
5461
5462 static int ext4_commit_super(struct super_block *sb)
5463 {
5464         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5465         int error = 0;
5466
5467         if (!sbh)
5468                 return -EINVAL;
5469         if (block_device_ejected(sb))
5470                 return -ENODEV;
5471
5472         ext4_update_super(sb);
5473
5474         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5475                 /*
5476                  * Oh, dear.  A previous attempt to write the
5477                  * superblock failed.  This could happen because the
5478                  * USB device was yanked out.  Or it could happen to
5479                  * be a transient write error and maybe the block will
5480                  * be remapped.  Nothing we can do but to retry the
5481                  * write and hope for the best.
5482                  */
5483                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5484                        "superblock detected");
5485                 clear_buffer_write_io_error(sbh);
5486                 set_buffer_uptodate(sbh);
5487         }
5488         BUFFER_TRACE(sbh, "marking dirty");
5489         mark_buffer_dirty(sbh);
5490         error = __sync_dirty_buffer(sbh,
5491                 REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5492         if (buffer_write_io_error(sbh)) {
5493                 ext4_msg(sb, KERN_ERR, "I/O error while writing "
5494                        "superblock");
5495                 clear_buffer_write_io_error(sbh);
5496                 set_buffer_uptodate(sbh);
5497         }
5498         return error;
5499 }
5500
5501 /*
5502  * Have we just finished recovery?  If so, and if we are mounting (or
5503  * remounting) the filesystem readonly, then we will end up with a
5504  * consistent fs on disk.  Record that fact.
5505  */
5506 static int ext4_mark_recovery_complete(struct super_block *sb,
5507                                        struct ext4_super_block *es)
5508 {
5509         int err;
5510         journal_t *journal = EXT4_SB(sb)->s_journal;
5511
5512         if (!ext4_has_feature_journal(sb)) {
5513                 if (journal != NULL) {
5514                         ext4_error(sb, "Journal got removed while the fs was "
5515                                    "mounted!");
5516                         return -EFSCORRUPTED;
5517                 }
5518                 return 0;
5519         }
5520         jbd2_journal_lock_updates(journal);
5521         err = jbd2_journal_flush(journal, 0);
5522         if (err < 0)
5523                 goto out;
5524
5525         if (sb_rdonly(sb) && (ext4_has_feature_journal_needs_recovery(sb) ||
5526             ext4_has_feature_orphan_present(sb))) {
5527                 if (!ext4_orphan_file_empty(sb)) {
5528                         ext4_error(sb, "Orphan file not empty on read-only fs.");
5529                         err = -EFSCORRUPTED;
5530                         goto out;
5531                 }
5532                 ext4_clear_feature_journal_needs_recovery(sb);
5533                 ext4_clear_feature_orphan_present(sb);
5534                 ext4_commit_super(sb);
5535         }
5536 out:
5537         jbd2_journal_unlock_updates(journal);
5538         return err;
5539 }
5540
5541 /*
5542  * If we are mounting (or read-write remounting) a filesystem whose journal
5543  * has recorded an error from a previous lifetime, move that error to the
5544  * main filesystem now.
5545  */
5546 static int ext4_clear_journal_err(struct super_block *sb,
5547                                    struct ext4_super_block *es)
5548 {
5549         journal_t *journal;
5550         int j_errno;
5551         const char *errstr;
5552
5553         if (!ext4_has_feature_journal(sb)) {
5554                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5555                 return -EFSCORRUPTED;
5556         }
5557
5558         journal = EXT4_SB(sb)->s_journal;
5559
5560         /*
5561          * Now check for any error status which may have been recorded in the
5562          * journal by a prior ext4_error() or ext4_abort()
5563          */
5564
5565         j_errno = jbd2_journal_errno(journal);
5566         if (j_errno) {
5567                 char nbuf[16];
5568
5569                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5570                 ext4_warning(sb, "Filesystem error recorded "
5571                              "from previous mount: %s", errstr);
5572                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5573
5574                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5575                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5576                 ext4_commit_super(sb);
5577
5578                 jbd2_journal_clear_err(journal);
5579                 jbd2_journal_update_sb_errno(journal);
5580         }
5581         return 0;
5582 }
5583
5584 /*
5585  * Force the running and committing transactions to commit,
5586  * and wait on the commit.
5587  */
5588 int ext4_force_commit(struct super_block *sb)
5589 {
5590         journal_t *journal;
5591
5592         if (sb_rdonly(sb))
5593                 return 0;
5594
5595         journal = EXT4_SB(sb)->s_journal;
5596         return ext4_journal_force_commit(journal);
5597 }
5598
5599 static int ext4_sync_fs(struct super_block *sb, int wait)
5600 {
5601         int ret = 0;
5602         tid_t target;
5603         bool needs_barrier = false;
5604         struct ext4_sb_info *sbi = EXT4_SB(sb);
5605
5606         if (unlikely(ext4_forced_shutdown(sbi)))
5607                 return 0;
5608
5609         trace_ext4_sync_fs(sb, wait);
5610         flush_workqueue(sbi->rsv_conversion_wq);
5611         /*
5612          * Writeback quota in non-journalled quota case - journalled quota has
5613          * no dirty dquots
5614          */
5615         dquot_writeback_dquots(sb, -1);
5616         /*
5617          * Data writeback is possible w/o journal transaction, so barrier must
5618          * being sent at the end of the function. But we can skip it if
5619          * transaction_commit will do it for us.
5620          */
5621         if (sbi->s_journal) {
5622                 target = jbd2_get_latest_transaction(sbi->s_journal);
5623                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5624                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5625                         needs_barrier = true;
5626
5627                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5628                         if (wait)
5629                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5630                                                            target);
5631                 }
5632         } else if (wait && test_opt(sb, BARRIER))
5633                 needs_barrier = true;
5634         if (needs_barrier) {
5635                 int err;
5636                 err = blkdev_issue_flush(sb->s_bdev);
5637                 if (!ret)
5638                         ret = err;
5639         }
5640
5641         return ret;
5642 }
5643
5644 /*
5645  * LVM calls this function before a (read-only) snapshot is created.  This
5646  * gives us a chance to flush the journal completely and mark the fs clean.
5647  *
5648  * Note that only this function cannot bring a filesystem to be in a clean
5649  * state independently. It relies on upper layer to stop all data & metadata
5650  * modifications.
5651  */
5652 static int ext4_freeze(struct super_block *sb)
5653 {
5654         int error = 0;
5655         journal_t *journal;
5656
5657         if (sb_rdonly(sb))
5658                 return 0;
5659
5660         journal = EXT4_SB(sb)->s_journal;
5661
5662         if (journal) {
5663                 /* Now we set up the journal barrier. */
5664                 jbd2_journal_lock_updates(journal);
5665
5666                 /*
5667                  * Don't clear the needs_recovery flag if we failed to
5668                  * flush the journal.
5669                  */
5670                 error = jbd2_journal_flush(journal, 0);
5671                 if (error < 0)
5672                         goto out;
5673
5674                 /* Journal blocked and flushed, clear needs_recovery flag. */
5675                 ext4_clear_feature_journal_needs_recovery(sb);
5676                 if (ext4_orphan_file_empty(sb))
5677                         ext4_clear_feature_orphan_present(sb);
5678         }
5679
5680         error = ext4_commit_super(sb);
5681 out:
5682         if (journal)
5683                 /* we rely on upper layer to stop further updates */
5684                 jbd2_journal_unlock_updates(journal);
5685         return error;
5686 }
5687
5688 /*
5689  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5690  * flag here, even though the filesystem is not technically dirty yet.
5691  */
5692 static int ext4_unfreeze(struct super_block *sb)
5693 {
5694         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5695                 return 0;
5696
5697         if (EXT4_SB(sb)->s_journal) {
5698                 /* Reset the needs_recovery flag before the fs is unlocked. */
5699                 ext4_set_feature_journal_needs_recovery(sb);
5700                 if (ext4_has_feature_orphan_file(sb))
5701                         ext4_set_feature_orphan_present(sb);
5702         }
5703
5704         ext4_commit_super(sb);
5705         return 0;
5706 }
5707
5708 /*
5709  * Structure to save mount options for ext4_remount's benefit
5710  */
5711 struct ext4_mount_options {
5712         unsigned long s_mount_opt;
5713         unsigned long s_mount_opt2;
5714         kuid_t s_resuid;
5715         kgid_t s_resgid;
5716         unsigned long s_commit_interval;
5717         u32 s_min_batch_time, s_max_batch_time;
5718 #ifdef CONFIG_QUOTA
5719         int s_jquota_fmt;
5720         char *s_qf_names[EXT4_MAXQUOTAS];
5721 #endif
5722 };
5723
5724 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5725 {
5726         struct ext4_super_block *es;
5727         struct ext4_sb_info *sbi = EXT4_SB(sb);
5728         unsigned long old_sb_flags, vfs_flags;
5729         struct ext4_mount_options old_opts;
5730         int enable_quota = 0;
5731         ext4_group_t g;
5732         int err = 0;
5733 #ifdef CONFIG_QUOTA
5734         int i, j;
5735         char *to_free[EXT4_MAXQUOTAS];
5736 #endif
5737         char *orig_data = kstrdup(data, GFP_KERNEL);
5738         struct ext4_parsed_options parsed_opts;
5739
5740         parsed_opts.journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5741         parsed_opts.journal_devnum = 0;
5742
5743         if (data && !orig_data)
5744                 return -ENOMEM;
5745
5746         /* Store the original options */
5747         old_sb_flags = sb->s_flags;
5748         old_opts.s_mount_opt = sbi->s_mount_opt;
5749         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5750         old_opts.s_resuid = sbi->s_resuid;
5751         old_opts.s_resgid = sbi->s_resgid;
5752         old_opts.s_commit_interval = sbi->s_commit_interval;
5753         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5754         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5755 #ifdef CONFIG_QUOTA
5756         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5757         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5758                 if (sbi->s_qf_names[i]) {
5759                         char *qf_name = get_qf_name(sb, sbi, i);
5760
5761                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5762                         if (!old_opts.s_qf_names[i]) {
5763                                 for (j = 0; j < i; j++)
5764                                         kfree(old_opts.s_qf_names[j]);
5765                                 kfree(orig_data);
5766                                 return -ENOMEM;
5767                         }
5768                 } else
5769                         old_opts.s_qf_names[i] = NULL;
5770 #endif
5771         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5772                 parsed_opts.journal_ioprio =
5773                         sbi->s_journal->j_task->io_context->ioprio;
5774
5775         /*
5776          * Some options can be enabled by ext4 and/or by VFS mount flag
5777          * either way we need to make sure it matches in both *flags and
5778          * s_flags. Copy those selected flags from *flags to s_flags
5779          */
5780         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5781         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5782
5783         if (!parse_options(data, sb, &parsed_opts, 1)) {
5784                 err = -EINVAL;
5785                 goto restore_opts;
5786         }
5787
5788         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5789             test_opt(sb, JOURNAL_CHECKSUM)) {
5790                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5791                          "during remount not supported; ignoring");
5792                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5793         }
5794
5795         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5796                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5797                         ext4_msg(sb, KERN_ERR, "can't mount with "
5798                                  "both data=journal and delalloc");
5799                         err = -EINVAL;
5800                         goto restore_opts;
5801                 }
5802                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5803                         ext4_msg(sb, KERN_ERR, "can't mount with "
5804                                  "both data=journal and dioread_nolock");
5805                         err = -EINVAL;
5806                         goto restore_opts;
5807                 }
5808         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5809                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5810                         ext4_msg(sb, KERN_ERR, "can't mount with "
5811                                 "journal_async_commit in data=ordered mode");
5812                         err = -EINVAL;
5813                         goto restore_opts;
5814                 }
5815         }
5816
5817         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5818                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5819                 err = -EINVAL;
5820                 goto restore_opts;
5821         }
5822
5823         if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
5824                 ext4_abort(sb, EXT4_ERR_ESHUTDOWN, "Abort forced by user");
5825
5826         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5827                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5828
5829         es = sbi->s_es;
5830
5831         if (sbi->s_journal) {
5832                 ext4_init_journal_params(sb, sbi->s_journal);
5833                 set_task_ioprio(sbi->s_journal->j_task, parsed_opts.journal_ioprio);
5834         }
5835
5836         /* Flush outstanding errors before changing fs state */
5837         flush_work(&sbi->s_error_work);
5838
5839         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5840                 if (ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) {
5841                         err = -EROFS;
5842                         goto restore_opts;
5843                 }
5844
5845                 if (*flags & SB_RDONLY) {
5846                         err = sync_filesystem(sb);
5847                         if (err < 0)
5848                                 goto restore_opts;
5849                         err = dquot_suspend(sb, -1);
5850                         if (err < 0)
5851                                 goto restore_opts;
5852
5853                         /*
5854                          * First of all, the unconditional stuff we have to do
5855                          * to disable replay of the journal when we next remount
5856                          */
5857                         sb->s_flags |= SB_RDONLY;
5858
5859                         /*
5860                          * OK, test if we are remounting a valid rw partition
5861                          * readonly, and if so set the rdonly flag and then
5862                          * mark the partition as valid again.
5863                          */
5864                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5865                             (sbi->s_mount_state & EXT4_VALID_FS))
5866                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5867
5868                         if (sbi->s_journal) {
5869                                 /*
5870                                  * We let remount-ro finish even if marking fs
5871                                  * as clean failed...
5872                                  */
5873                                 ext4_mark_recovery_complete(sb, es);
5874                         }
5875                 } else {
5876                         /* Make sure we can mount this feature set readwrite */
5877                         if (ext4_has_feature_readonly(sb) ||
5878                             !ext4_feature_set_ok(sb, 0)) {
5879                                 err = -EROFS;
5880                                 goto restore_opts;
5881                         }
5882                         /*
5883                          * Make sure the group descriptor checksums
5884                          * are sane.  If they aren't, refuse to remount r/w.
5885                          */
5886                         for (g = 0; g < sbi->s_groups_count; g++) {
5887                                 struct ext4_group_desc *gdp =
5888                                         ext4_get_group_desc(sb, g, NULL);
5889
5890                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5891                                         ext4_msg(sb, KERN_ERR,
5892                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5893                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5894                                                le16_to_cpu(gdp->bg_checksum));
5895                                         err = -EFSBADCRC;
5896                                         goto restore_opts;
5897                                 }
5898                         }
5899
5900                         /*
5901                          * If we have an unprocessed orphan list hanging
5902                          * around from a previously readonly bdev mount,
5903                          * require a full umount/remount for now.
5904                          */
5905                         if (es->s_last_orphan || !ext4_orphan_file_empty(sb)) {
5906                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5907                                        "remount RDWR because of unprocessed "
5908                                        "orphan inode list.  Please "
5909                                        "umount/remount instead");
5910                                 err = -EINVAL;
5911                                 goto restore_opts;
5912                         }
5913
5914                         /*
5915                          * Mounting a RDONLY partition read-write, so reread
5916                          * and store the current valid flag.  (It may have
5917                          * been changed by e2fsck since we originally mounted
5918                          * the partition.)
5919                          */
5920                         if (sbi->s_journal) {
5921                                 err = ext4_clear_journal_err(sb, es);
5922                                 if (err)
5923                                         goto restore_opts;
5924                         }
5925                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5926
5927                         err = ext4_setup_super(sb, es, 0);
5928                         if (err)
5929                                 goto restore_opts;
5930
5931                         sb->s_flags &= ~SB_RDONLY;
5932                         if (ext4_has_feature_mmp(sb))
5933                                 if (ext4_multi_mount_protect(sb,
5934                                                 le64_to_cpu(es->s_mmp_block))) {
5935                                         err = -EROFS;
5936                                         goto restore_opts;
5937                                 }
5938                         enable_quota = 1;
5939                 }
5940         }
5941
5942         /*
5943          * Reinitialize lazy itable initialization thread based on
5944          * current settings
5945          */
5946         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5947                 ext4_unregister_li_request(sb);
5948         else {
5949                 ext4_group_t first_not_zeroed;
5950                 first_not_zeroed = ext4_has_uninit_itable(sb);
5951                 ext4_register_li_request(sb, first_not_zeroed);
5952         }
5953
5954         /*
5955          * Handle creation of system zone data early because it can fail.
5956          * Releasing of existing data is done when we are sure remount will
5957          * succeed.
5958          */
5959         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->s_system_blks) {
5960                 err = ext4_setup_system_zone(sb);
5961                 if (err)
5962                         goto restore_opts;
5963         }
5964
5965         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5966                 err = ext4_commit_super(sb);
5967                 if (err)
5968                         goto restore_opts;
5969         }
5970
5971 #ifdef CONFIG_QUOTA
5972         /* Release old quota file names */
5973         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5974                 kfree(old_opts.s_qf_names[i]);
5975         if (enable_quota) {
5976                 if (sb_any_quota_suspended(sb))
5977                         dquot_resume(sb, -1);
5978                 else if (ext4_has_feature_quota(sb)) {
5979                         err = ext4_enable_quotas(sb);
5980                         if (err)
5981                                 goto restore_opts;
5982                 }
5983         }
5984 #endif
5985         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
5986                 ext4_release_system_zone(sb);
5987
5988         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
5989                 ext4_stop_mmpd(sbi);
5990
5991         /*
5992          * Some options can be enabled by ext4 and/or by VFS mount flag
5993          * either way we need to make sure it matches in both *flags and
5994          * s_flags. Copy those selected flags from s_flags to *flags
5995          */
5996         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
5997
5998         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s. Quota mode: %s.",
5999                  orig_data, ext4_quota_mode(sb));
6000         kfree(orig_data);
6001         return 0;
6002
6003 restore_opts:
6004         sb->s_flags = old_sb_flags;
6005         sbi->s_mount_opt = old_opts.s_mount_opt;
6006         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
6007         sbi->s_resuid = old_opts.s_resuid;
6008         sbi->s_resgid = old_opts.s_resgid;
6009         sbi->s_commit_interval = old_opts.s_commit_interval;
6010         sbi->s_min_batch_time = old_opts.s_min_batch_time;
6011         sbi->s_max_batch_time = old_opts.s_max_batch_time;
6012         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->s_system_blks)
6013                 ext4_release_system_zone(sb);
6014 #ifdef CONFIG_QUOTA
6015         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
6016         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
6017                 to_free[i] = get_qf_name(sb, sbi, i);
6018                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
6019         }
6020         synchronize_rcu();
6021         for (i = 0; i < EXT4_MAXQUOTAS; i++)
6022                 kfree(to_free[i]);
6023 #endif
6024         if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb))
6025                 ext4_stop_mmpd(sbi);
6026         kfree(orig_data);
6027         return err;
6028 }
6029
6030 #ifdef CONFIG_QUOTA
6031 static int ext4_statfs_project(struct super_block *sb,
6032                                kprojid_t projid, struct kstatfs *buf)
6033 {
6034         struct kqid qid;
6035         struct dquot *dquot;
6036         u64 limit;
6037         u64 curblock;
6038
6039         qid = make_kqid_projid(projid);
6040         dquot = dqget(sb, qid);
6041         if (IS_ERR(dquot))
6042                 return PTR_ERR(dquot);
6043         spin_lock(&dquot->dq_dqb_lock);
6044
6045         limit = min_not_zero(dquot->dq_dqb.dqb_bsoftlimit,
6046                              dquot->dq_dqb.dqb_bhardlimit);
6047         limit >>= sb->s_blocksize_bits;
6048
6049         if (limit && buf->f_blocks > limit) {
6050                 curblock = (dquot->dq_dqb.dqb_curspace +
6051                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
6052                 buf->f_blocks = limit;
6053                 buf->f_bfree = buf->f_bavail =
6054                         (buf->f_blocks > curblock) ?
6055                          (buf->f_blocks - curblock) : 0;
6056         }
6057
6058         limit = min_not_zero(dquot->dq_dqb.dqb_isoftlimit,
6059                              dquot->dq_dqb.dqb_ihardlimit);
6060         if (limit && buf->f_files > limit) {
6061                 buf->f_files = limit;
6062                 buf->f_ffree =
6063                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
6064                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
6065         }
6066
6067         spin_unlock(&dquot->dq_dqb_lock);
6068         dqput(dquot);
6069         return 0;
6070 }
6071 #endif
6072
6073 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
6074 {
6075         struct super_block *sb = dentry->d_sb;
6076         struct ext4_sb_info *sbi = EXT4_SB(sb);
6077         struct ext4_super_block *es = sbi->s_es;
6078         ext4_fsblk_t overhead = 0, resv_blocks;
6079         s64 bfree;
6080         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
6081
6082         if (!test_opt(sb, MINIX_DF))
6083                 overhead = sbi->s_overhead;
6084
6085         buf->f_type = EXT4_SUPER_MAGIC;
6086         buf->f_bsize = sb->s_blocksize;
6087         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
6088         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
6089                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
6090         /* prevent underflow in case that few free space is available */
6091         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
6092         buf->f_bavail = buf->f_bfree -
6093                         (ext4_r_blocks_count(es) + resv_blocks);
6094         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
6095                 buf->f_bavail = 0;
6096         buf->f_files = le32_to_cpu(es->s_inodes_count);
6097         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
6098         buf->f_namelen = EXT4_NAME_LEN;
6099         buf->f_fsid = uuid_to_fsid(es->s_uuid);
6100
6101 #ifdef CONFIG_QUOTA
6102         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
6103             sb_has_quota_limits_enabled(sb, PRJQUOTA))
6104                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
6105 #endif
6106         return 0;
6107 }
6108
6109
6110 #ifdef CONFIG_QUOTA
6111
6112 /*
6113  * Helper functions so that transaction is started before we acquire dqio_sem
6114  * to keep correct lock ordering of transaction > dqio_sem
6115  */
6116 static inline struct inode *dquot_to_inode(struct dquot *dquot)
6117 {
6118         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
6119 }
6120
6121 static int ext4_write_dquot(struct dquot *dquot)
6122 {
6123         int ret, err;
6124         handle_t *handle;
6125         struct inode *inode;
6126
6127         inode = dquot_to_inode(dquot);
6128         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
6129                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
6130         if (IS_ERR(handle))
6131                 return PTR_ERR(handle);
6132         ret = dquot_commit(dquot);
6133         err = ext4_journal_stop(handle);
6134         if (!ret)
6135                 ret = err;
6136         return ret;
6137 }
6138
6139 static int ext4_acquire_dquot(struct dquot *dquot)
6140 {
6141         int ret, err;
6142         handle_t *handle;
6143
6144         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6145                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
6146         if (IS_ERR(handle))
6147                 return PTR_ERR(handle);
6148         ret = dquot_acquire(dquot);
6149         err = ext4_journal_stop(handle);
6150         if (!ret)
6151                 ret = err;
6152         return ret;
6153 }
6154
6155 static int ext4_release_dquot(struct dquot *dquot)
6156 {
6157         int ret, err;
6158         handle_t *handle;
6159
6160         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
6161                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
6162         if (IS_ERR(handle)) {
6163                 /* Release dquot anyway to avoid endless cycle in dqput() */
6164                 dquot_release(dquot);
6165                 return PTR_ERR(handle);
6166         }
6167         ret = dquot_release(dquot);
6168         err = ext4_journal_stop(handle);
6169         if (!ret)
6170                 ret = err;
6171         return ret;
6172 }
6173
6174 static int ext4_mark_dquot_dirty(struct dquot *dquot)
6175 {
6176         struct super_block *sb = dquot->dq_sb;
6177
6178         if (ext4_is_quota_journalled(sb)) {
6179                 dquot_mark_dquot_dirty(dquot);
6180                 return ext4_write_dquot(dquot);
6181         } else {
6182                 return dquot_mark_dquot_dirty(dquot);
6183         }
6184 }
6185
6186 static int ext4_write_info(struct super_block *sb, int type)
6187 {
6188         int ret, err;
6189         handle_t *handle;
6190
6191         /* Data block + inode block */
6192         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
6193         if (IS_ERR(handle))
6194                 return PTR_ERR(handle);
6195         ret = dquot_commit_info(sb, type);
6196         err = ext4_journal_stop(handle);
6197         if (!ret)
6198                 ret = err;
6199         return ret;
6200 }
6201
6202 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
6203 {
6204         struct ext4_inode_info *ei = EXT4_I(inode);
6205
6206         /* The first argument of lockdep_set_subclass has to be
6207          * *exactly* the same as the argument to init_rwsem() --- in
6208          * this case, in init_once() --- or lockdep gets unhappy
6209          * because the name of the lock is set using the
6210          * stringification of the argument to init_rwsem().
6211          */
6212         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
6213         lockdep_set_subclass(&ei->i_data_sem, subclass);
6214 }
6215
6216 /*
6217  * Standard function to be called on quota_on
6218  */
6219 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
6220                          const struct path *path)
6221 {
6222         int err;
6223
6224         if (!test_opt(sb, QUOTA))
6225                 return -EINVAL;
6226
6227         /* Quotafile not on the same filesystem? */
6228         if (path->dentry->d_sb != sb)
6229                 return -EXDEV;
6230
6231         /* Quota already enabled for this file? */
6232         if (IS_NOQUOTA(d_inode(path->dentry)))
6233                 return -EBUSY;
6234
6235         /* Journaling quota? */
6236         if (EXT4_SB(sb)->s_qf_names[type]) {
6237                 /* Quotafile not in fs root? */
6238                 if (path->dentry->d_parent != sb->s_root)
6239                         ext4_msg(sb, KERN_WARNING,
6240                                 "Quota file not on filesystem root. "
6241                                 "Journaled quota will not work");
6242                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
6243         } else {
6244                 /*
6245                  * Clear the flag just in case mount options changed since
6246                  * last time.
6247                  */
6248                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
6249         }
6250
6251         /*
6252          * When we journal data on quota file, we have to flush journal to see
6253          * all updates to the file when we bypass pagecache...
6254          */
6255         if (EXT4_SB(sb)->s_journal &&
6256             ext4_should_journal_data(d_inode(path->dentry))) {
6257                 /*
6258                  * We don't need to lock updates but journal_flush() could
6259                  * otherwise be livelocked...
6260                  */
6261                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
6262                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal, 0);
6263                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
6264                 if (err)
6265                         return err;
6266         }
6267
6268         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
6269         err = dquot_quota_on(sb, type, format_id, path);
6270         if (err) {
6271                 lockdep_set_quota_inode(path->dentry->d_inode,
6272                                              I_DATA_SEM_NORMAL);
6273         } else {
6274                 struct inode *inode = d_inode(path->dentry);
6275                 handle_t *handle;
6276
6277                 /*
6278                  * Set inode flags to prevent userspace from messing with quota
6279                  * files. If this fails, we return success anyway since quotas
6280                  * are already enabled and this is not a hard failure.
6281                  */
6282                 inode_lock(inode);
6283                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6284                 if (IS_ERR(handle))
6285                         goto unlock_inode;
6286                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
6287                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
6288                                 S_NOATIME | S_IMMUTABLE);
6289                 err = ext4_mark_inode_dirty(handle, inode);
6290                 ext4_journal_stop(handle);
6291         unlock_inode:
6292                 inode_unlock(inode);
6293         }
6294         return err;
6295 }
6296
6297 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
6298                              unsigned int flags)
6299 {
6300         int err;
6301         struct inode *qf_inode;
6302         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6303                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6304                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6305                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6306         };
6307
6308         BUG_ON(!ext4_has_feature_quota(sb));
6309
6310         if (!qf_inums[type])
6311                 return -EPERM;
6312
6313         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6314         if (IS_ERR(qf_inode)) {
6315                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
6316                 return PTR_ERR(qf_inode);
6317         }
6318
6319         /* Don't account quota for quota files to avoid recursion */
6320         qf_inode->i_flags |= S_NOQUOTA;
6321         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6322         err = dquot_load_quota_inode(qf_inode, type, format_id, flags);
6323         if (err)
6324                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6325         iput(qf_inode);
6326
6327         return err;
6328 }
6329
6330 /* Enable usage tracking for all quota types. */
6331 int ext4_enable_quotas(struct super_block *sb)
6332 {
6333         int type, err = 0;
6334         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6335                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6336                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6337                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6338         };
6339         bool quota_mopt[EXT4_MAXQUOTAS] = {
6340                 test_opt(sb, USRQUOTA),
6341                 test_opt(sb, GRPQUOTA),
6342                 test_opt(sb, PRJQUOTA),
6343         };
6344
6345         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6346         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6347                 if (qf_inums[type]) {
6348                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6349                                 DQUOT_USAGE_ENABLED |
6350                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6351                         if (err) {
6352                                 ext4_warning(sb,
6353                                         "Failed to enable quota tracking "
6354                                         "(type=%d, err=%d). Please run "
6355                                         "e2fsck to fix.", type, err);
6356                                 for (type--; type >= 0; type--)
6357                                         dquot_quota_off(sb, type);
6358
6359                                 return err;
6360                         }
6361                 }
6362         }
6363         return 0;
6364 }
6365
6366 static int ext4_quota_off(struct super_block *sb, int type)
6367 {
6368         struct inode *inode = sb_dqopt(sb)->files[type];
6369         handle_t *handle;
6370         int err;
6371
6372         /* Force all delayed allocation blocks to be allocated.
6373          * Caller already holds s_umount sem */
6374         if (test_opt(sb, DELALLOC))
6375                 sync_filesystem(sb);
6376
6377         if (!inode || !igrab(inode))
6378                 goto out;
6379
6380         err = dquot_quota_off(sb, type);
6381         if (err || ext4_has_feature_quota(sb))
6382                 goto out_put;
6383
6384         inode_lock(inode);
6385         /*
6386          * Update modification times of quota files when userspace can
6387          * start looking at them. If we fail, we return success anyway since
6388          * this is not a hard failure and quotas are already disabled.
6389          */
6390         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6391         if (IS_ERR(handle)) {
6392                 err = PTR_ERR(handle);
6393                 goto out_unlock;
6394         }
6395         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6396         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6397         inode->i_mtime = inode->i_ctime = current_time(inode);
6398         err = ext4_mark_inode_dirty(handle, inode);
6399         ext4_journal_stop(handle);
6400 out_unlock:
6401         inode_unlock(inode);
6402 out_put:
6403         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6404         iput(inode);
6405         return err;
6406 out:
6407         return dquot_quota_off(sb, type);
6408 }
6409
6410 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6411  * acquiring the locks... As quota files are never truncated and quota code
6412  * itself serializes the operations (and no one else should touch the files)
6413  * we don't have to be afraid of races */
6414 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6415                                size_t len, loff_t off)
6416 {
6417         struct inode *inode = sb_dqopt(sb)->files[type];
6418         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6419         int offset = off & (sb->s_blocksize - 1);
6420         int tocopy;
6421         size_t toread;
6422         struct buffer_head *bh;
6423         loff_t i_size = i_size_read(inode);
6424
6425         if (off > i_size)
6426                 return 0;
6427         if (off+len > i_size)
6428                 len = i_size-off;
6429         toread = len;
6430         while (toread > 0) {
6431                 tocopy = sb->s_blocksize - offset < toread ?
6432                                 sb->s_blocksize - offset : toread;
6433                 bh = ext4_bread(NULL, inode, blk, 0);
6434                 if (IS_ERR(bh))
6435                         return PTR_ERR(bh);
6436                 if (!bh)        /* A hole? */
6437                         memset(data, 0, tocopy);
6438                 else
6439                         memcpy(data, bh->b_data+offset, tocopy);
6440                 brelse(bh);
6441                 offset = 0;
6442                 toread -= tocopy;
6443                 data += tocopy;
6444                 blk++;
6445         }
6446         return len;
6447 }
6448
6449 /* Write to quotafile (we know the transaction is already started and has
6450  * enough credits) */
6451 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6452                                 const char *data, size_t len, loff_t off)
6453 {
6454         struct inode *inode = sb_dqopt(sb)->files[type];
6455         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6456         int err = 0, err2 = 0, offset = off & (sb->s_blocksize - 1);
6457         int retries = 0;
6458         struct buffer_head *bh;
6459         handle_t *handle = journal_current_handle();
6460
6461         if (EXT4_SB(sb)->s_journal && !handle) {
6462                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6463                         " cancelled because transaction is not started",
6464                         (unsigned long long)off, (unsigned long long)len);
6465                 return -EIO;
6466         }
6467         /*
6468          * Since we account only one data block in transaction credits,
6469          * then it is impossible to cross a block boundary.
6470          */
6471         if (sb->s_blocksize - offset < len) {
6472                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6473                         " cancelled because not block aligned",
6474                         (unsigned long long)off, (unsigned long long)len);
6475                 return -EIO;
6476         }
6477
6478         do {
6479                 bh = ext4_bread(handle, inode, blk,
6480                                 EXT4_GET_BLOCKS_CREATE |
6481                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6482         } while (PTR_ERR(bh) == -ENOSPC &&
6483                  ext4_should_retry_alloc(inode->i_sb, &retries));
6484         if (IS_ERR(bh))
6485                 return PTR_ERR(bh);
6486         if (!bh)
6487                 goto out;
6488         BUFFER_TRACE(bh, "get write access");
6489         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
6490         if (err) {
6491                 brelse(bh);
6492                 return err;
6493         }
6494         lock_buffer(bh);
6495         memcpy(bh->b_data+offset, data, len);
6496         flush_dcache_page(bh->b_page);
6497         unlock_buffer(bh);
6498         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6499         brelse(bh);
6500 out:
6501         if (inode->i_size < off + len) {
6502                 i_size_write(inode, off + len);
6503                 EXT4_I(inode)->i_disksize = inode->i_size;
6504                 err2 = ext4_mark_inode_dirty(handle, inode);
6505                 if (unlikely(err2 && !err))
6506                         err = err2;
6507         }
6508         return err ? err : len;
6509 }
6510 #endif
6511
6512 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6513                        const char *dev_name, void *data)
6514 {
6515         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6516 }
6517
6518 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6519 static inline void register_as_ext2(void)
6520 {
6521         int err = register_filesystem(&ext2_fs_type);
6522         if (err)
6523                 printk(KERN_WARNING
6524                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6525 }
6526
6527 static inline void unregister_as_ext2(void)
6528 {
6529         unregister_filesystem(&ext2_fs_type);
6530 }
6531
6532 static inline int ext2_feature_set_ok(struct super_block *sb)
6533 {
6534         if (ext4_has_unknown_ext2_incompat_features(sb))
6535                 return 0;
6536         if (sb_rdonly(sb))
6537                 return 1;
6538         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6539                 return 0;
6540         return 1;
6541 }
6542 #else
6543 static inline void register_as_ext2(void) { }
6544 static inline void unregister_as_ext2(void) { }
6545 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6546 #endif
6547
6548 static inline void register_as_ext3(void)
6549 {
6550         int err = register_filesystem(&ext3_fs_type);
6551         if (err)
6552                 printk(KERN_WARNING
6553                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6554 }
6555
6556 static inline void unregister_as_ext3(void)
6557 {
6558         unregister_filesystem(&ext3_fs_type);
6559 }
6560
6561 static inline int ext3_feature_set_ok(struct super_block *sb)
6562 {
6563         if (ext4_has_unknown_ext3_incompat_features(sb))
6564                 return 0;
6565         if (!ext4_has_feature_journal(sb))
6566                 return 0;
6567         if (sb_rdonly(sb))
6568                 return 1;
6569         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6570                 return 0;
6571         return 1;
6572 }
6573
6574 static struct file_system_type ext4_fs_type = {
6575         .owner          = THIS_MODULE,
6576         .name           = "ext4",
6577         .mount          = ext4_mount,
6578         .kill_sb        = kill_block_super,
6579         .fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
6580 };
6581 MODULE_ALIAS_FS("ext4");
6582
6583 /* Shared across all ext4 file systems */
6584 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6585
6586 static int __init ext4_init_fs(void)
6587 {
6588         int i, err;
6589
6590         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6591         ext4_li_info = NULL;
6592
6593         /* Build-time check for flags consistency */
6594         ext4_check_flag_values();
6595
6596         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6597                 init_waitqueue_head(&ext4__ioend_wq[i]);
6598
6599         err = ext4_init_es();
6600         if (err)
6601                 return err;
6602
6603         err = ext4_init_pending();
6604         if (err)
6605                 goto out7;
6606
6607         err = ext4_init_post_read_processing();
6608         if (err)
6609                 goto out6;
6610
6611         err = ext4_init_pageio();
6612         if (err)
6613                 goto out5;
6614
6615         err = ext4_init_system_zone();
6616         if (err)
6617                 goto out4;
6618
6619         err = ext4_init_sysfs();
6620         if (err)
6621                 goto out3;
6622
6623         err = ext4_init_mballoc();
6624         if (err)
6625                 goto out2;
6626         err = init_inodecache();
6627         if (err)
6628                 goto out1;
6629
6630         err = ext4_fc_init_dentry_cache();
6631         if (err)
6632                 goto out05;
6633
6634         register_as_ext3();
6635         register_as_ext2();
6636         err = register_filesystem(&ext4_fs_type);
6637         if (err)
6638                 goto out;
6639
6640         return 0;
6641 out:
6642         unregister_as_ext2();
6643         unregister_as_ext3();
6644 out05:
6645         destroy_inodecache();
6646 out1:
6647         ext4_exit_mballoc();
6648 out2:
6649         ext4_exit_sysfs();
6650 out3:
6651         ext4_exit_system_zone();
6652 out4:
6653         ext4_exit_pageio();
6654 out5:
6655         ext4_exit_post_read_processing();
6656 out6:
6657         ext4_exit_pending();
6658 out7:
6659         ext4_exit_es();
6660
6661         return err;
6662 }
6663
6664 static void __exit ext4_exit_fs(void)
6665 {
6666         ext4_destroy_lazyinit_thread();
6667         unregister_as_ext2();
6668         unregister_as_ext3();
6669         unregister_filesystem(&ext4_fs_type);
6670         destroy_inodecache();
6671         ext4_exit_mballoc();
6672         ext4_exit_sysfs();
6673         ext4_exit_system_zone();
6674         ext4_exit_pageio();
6675         ext4_exit_post_read_processing();
6676         ext4_exit_es();
6677         ext4_exit_pending();
6678 }
6679
6680 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6681 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6682 MODULE_LICENSE("GPL");
6683 MODULE_SOFTDEP("pre: crc32c");
6684 module_init(ext4_init_fs)
6685 module_exit(ext4_exit_fs)