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
  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 struct 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 int ext4_commit_super(struct super_block *sb, int sync);
  69 static int ext4_mark_recovery_complete(struct super_block *sb,
  70                                         struct ext4_super_block *es);
  71 static int ext4_clear_journal_err(struct super_block *sb,
  72                                   struct ext4_super_block *es);
  73 static int ext4_sync_fs(struct super_block *sb, int wait);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  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  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  93  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  94  *
  95  * page fault path:
  96  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  97  *   page lock -> i_data_sem (rw)
  98  *
  99  * buffered write path:
 100  * sb_start_write -> i_mutex -> mmap_sem
 101  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 102  *   i_data_sem (rw)
 103  *
 104  * truncate:
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 106  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 107  *   i_data_sem (rw)
 108  *
 109  * direct IO:
 110  * sb_start_write -> i_mutex -> mmap_sem
 111  * sb_start_write -> i_mutex -> transaction start -> i_data_sem (rw)
 112  *
 113  * writepages:
 114  * transaction start -> page lock(s) -> i_data_sem (rw)
 115  */
 116
 117 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 118 static struct file_system_type ext2_fs_type = {
 119         .owner          = THIS_MODULE,
 120         .name           = "ext2",
 121         .mount          = ext4_mount,
 122         .kill_sb        = kill_block_super,
 123         .fs_flags       = FS_REQUIRES_DEV,
 124 };
 125 MODULE_ALIAS_FS("ext2");
 126 MODULE_ALIAS("ext2");
 127 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 128 #else
 129 #define IS_EXT2_SB(sb) (0)
 130 #endif
 131
 132
 133 static struct file_system_type ext3_fs_type = {
 134         .owner          = THIS_MODULE,
 135         .name           = "ext3",
 136         .mount          = ext4_mount,
 137         .kill_sb        = kill_block_super,
 138         .fs_flags       = FS_REQUIRES_DEV,
 139 };
 140 MODULE_ALIAS_FS("ext3");
 141 MODULE_ALIAS("ext3");
 142 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 143
 144 /*
 145  * This works like sb_bread() except it uses ERR_PTR for error
 146  * returns.  Currently with sb_bread it's impossible to distinguish
 147  * between ENOMEM and EIO situations (since both result in a NULL
 148  * return.
 149  */
 150 struct buffer_head *
 151 ext4_sb_bread(struct super_block *sb, sector_t block, int op_flags)
 152 {
 153         struct buffer_head *bh = sb_getblk(sb, block);
 154
 155         if (bh == NULL)
 156                 return ERR_PTR(-ENOMEM);
 157         if (buffer_uptodate(bh))
 158                 return bh;
 159         ll_rw_block(REQ_OP_READ, REQ_META | op_flags, 1, &bh);
 160         wait_on_buffer(bh);
 161         if (buffer_uptodate(bh))
 162                 return bh;
 163         put_bh(bh);
 164         return ERR_PTR(-EIO);
 165 }
 166
 167 static int ext4_verify_csum_type(struct super_block *sb,
 168                                  struct ext4_super_block *es)
 169 {
 170         if (!ext4_has_feature_metadata_csum(sb))
 171                 return 1;
 172
 173         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 174 }
 175
 176 static __le32 ext4_superblock_csum(struct super_block *sb,
 177                                    struct ext4_super_block *es)
 178 {
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         int offset = offsetof(struct ext4_super_block, s_checksum);
 181         __u32 csum;
 182
 183         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 184
 185         return cpu_to_le32(csum);
 186 }
 187
 188 static int ext4_superblock_csum_verify(struct super_block *sb,
 189                                        struct ext4_super_block *es)
 190 {
 191         if (!ext4_has_metadata_csum(sb))
 192                 return 1;
 193
 194         return es->s_checksum == ext4_superblock_csum(sb, es);
 195 }
 196
 197 void ext4_superblock_csum_set(struct super_block *sb)
 198 {
 199         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 200
 201         if (!ext4_has_metadata_csum(sb))
 202                 return;
 203
 204         es->s_checksum = ext4_superblock_csum(sb, es);
 205 }
 206
 207 void *ext4_kvmalloc(size_t size, gfp_t flags)
 208 {
 209         void *ret;
 210
 211         ret = kmalloc(size, flags | __GFP_NOWARN);
 212         if (!ret)
 213                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 214         return ret;
 215 }
 216
 217 void *ext4_kvzalloc(size_t size, gfp_t flags)
 218 {
 219         void *ret;
 220
 221         ret = kzalloc(size, flags | __GFP_NOWARN);
 222         if (!ret)
 223                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 224         return ret;
 225 }
 226
 227 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 233 }
 234
 235 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 236                                struct ext4_group_desc *bg)
 237 {
 238         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 241 }
 242
 243 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le32_to_cpu(bg->bg_inode_table_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 249 }
 250
 251 __u32 ext4_free_group_clusters(struct super_block *sb,
 252                                struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 257 }
 258
 259 __u32 ext4_free_inodes_count(struct super_block *sb,
 260                               struct ext4_group_desc *bg)
 261 {
 262         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 263                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 264                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 265 }
 266
 267 __u32 ext4_used_dirs_count(struct super_block *sb,
 268                               struct ext4_group_desc *bg)
 269 {
 270         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 271                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 272                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 273 }
 274
 275 __u32 ext4_itable_unused_count(struct super_block *sb,
 276                               struct ext4_group_desc *bg)
 277 {
 278         return le16_to_cpu(bg->bg_itable_unused_lo) |
 279                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 280                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 281 }
 282
 283 void ext4_block_bitmap_set(struct super_block *sb,
 284                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 285 {
 286         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 289 }
 290
 291 void ext4_inode_bitmap_set(struct super_block *sb,
 292                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 293 {
 294         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 297 }
 298
 299 void ext4_inode_table_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 301 {
 302         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 305 }
 306
 307 void ext4_free_group_clusters_set(struct super_block *sb,
 308                                   struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315 void ext4_free_inodes_set(struct super_block *sb,
 316                           struct ext4_group_desc *bg, __u32 count)
 317 {
 318         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 319         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 320                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 321 }
 322
 323 void ext4_used_dirs_set(struct super_block *sb,
 324                           struct ext4_group_desc *bg, __u32 count)
 325 {
 326         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 327         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 328                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 329 }
 330
 331 void ext4_itable_unused_set(struct super_block *sb,
 332                           struct ext4_group_desc *bg, __u32 count)
 333 {
 334         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 335         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 336                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 337 }
 338
 339 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 340 {
 341         time64_t now = ktime_get_real_seconds();
 342
 343         now = clamp_val(now, 0, (1ull << 40) - 1);
 344
 345         *lo = cpu_to_le32(lower_32_bits(now));
 346         *hi = upper_32_bits(now);
 347 }
 348
 349 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 350 {
 351         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 352 }
 353 #define ext4_update_tstamp(es, tstamp) \
 354         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 355 #define ext4_get_tstamp(es, tstamp) \
 356         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 357
 358 static void __save_error_info(struct super_block *sb, const char *func,
 359                             unsigned int line)
 360 {
 361         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 362
 363         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 364         if (bdev_read_only(sb->s_bdev))
 365                 return;
 366         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 367         ext4_update_tstamp(es, s_last_error_time);
 368         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 369         es->s_last_error_line = cpu_to_le32(line);
 370         if (!es->s_first_error_time) {
 371                 es->s_first_error_time = es->s_last_error_time;
 372                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 373                 strncpy(es->s_first_error_func, func,
 374                         sizeof(es->s_first_error_func));
 375                 es->s_first_error_line = cpu_to_le32(line);
 376                 es->s_first_error_ino = es->s_last_error_ino;
 377                 es->s_first_error_block = es->s_last_error_block;
 378         }
 379         /*
 380          * Start the daily error reporting function if it hasn't been
 381          * started already
 382          */
 383         if (!es->s_error_count)
 384                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 385         le32_add_cpu(&es->s_error_count, 1);
 386 }
 387
 388 static void save_error_info(struct super_block *sb, const char *func,
 389                             unsigned int line)
 390 {
 391         __save_error_info(sb, func, line);
 392         if (!bdev_read_only(sb->s_bdev))
 393                 ext4_commit_super(sb, 1);
 394 }
 395
 396 /*
 397  * The del_gendisk() function uninitializes the disk-specific data
 398  * structures, including the bdi structure, without telling anyone
 399  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 400  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 401  * This is a kludge to prevent these oops until we can put in a proper
 402  * hook in del_gendisk() to inform the VFS and file system layers.
 403  */
 404 static int block_device_ejected(struct super_block *sb)
 405 {
 406         struct inode *bd_inode = sb->s_bdev->bd_inode;
 407         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 408
 409         return bdi->dev == NULL;
 410 }
 411
 412 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 413 {
 414         struct super_block              *sb = journal->j_private;
 415         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 416         int                             error = is_journal_aborted(journal);
 417         struct ext4_journal_cb_entry    *jce;
 418
 419         BUG_ON(txn->t_state == T_FINISHED);
 420
 421         ext4_process_freed_data(sb, txn->t_tid);
 422
 423         spin_lock(&sbi->s_md_lock);
 424         while (!list_empty(&txn->t_private_list)) {
 425                 jce = list_entry(txn->t_private_list.next,
 426                                  struct ext4_journal_cb_entry, jce_list);
 427                 list_del_init(&jce->jce_list);
 428                 spin_unlock(&sbi->s_md_lock);
 429                 jce->jce_func(sb, jce, error);
 430                 spin_lock(&sbi->s_md_lock);
 431         }
 432         spin_unlock(&sbi->s_md_lock);
 433 }
 434
 435 static bool system_going_down(void)
 436 {
 437         return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
 438                 || system_state == SYSTEM_RESTART;
 439 }
 440
 441 /* Deal with the reporting of failure conditions on a filesystem such as
 442  * inconsistencies detected or read IO failures.
 443  *
 444  * On ext2, we can store the error state of the filesystem in the
 445  * superblock.  That is not possible on ext4, because we may have other
 446  * write ordering constraints on the superblock which prevent us from
 447  * writing it out straight away; and given that the journal is about to
 448  * be aborted, we can't rely on the current, or future, transactions to
 449  * write out the superblock safely.
 450  *
 451  * We'll just use the jbd2_journal_abort() error code to record an error in
 452  * the journal instead.  On recovery, the journal will complain about
 453  * that error until we've noted it down and cleared it.
 454  */
 455
 456 static void ext4_handle_error(struct super_block *sb)
 457 {
 458         journal_t *journal = EXT4_SB(sb)->s_journal;
 459
 460         if (test_opt(sb, WARN_ON_ERROR))
 461                 WARN_ON_ONCE(1);
 462
 463         if (sb_rdonly(sb) || test_opt(sb, ERRORS_CONT))
 464                 return;
 465
 466         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 467         if (journal)
 468                 jbd2_journal_abort(journal, -EIO);
 469         /*
 470          * We force ERRORS_RO behavior when system is rebooting. Otherwise we
 471          * could panic during 'reboot -f' as the underlying device got already
 472          * disabled.
 473          */
 474         if (test_opt(sb, ERRORS_RO) || system_going_down()) {
 475                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 476                 /*
 477                  * Make sure updated value of ->s_mount_flags will be visible
 478                  * before ->s_flags update
 479                  */
 480                 smp_wmb();
 481                 sb->s_flags |= SB_RDONLY;
 482         } else if (test_opt(sb, ERRORS_PANIC)) {
 483                 if (EXT4_SB(sb)->s_journal &&
 484                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 485                         return;
 486                 panic("EXT4-fs (device %s): panic forced after error\n",
 487                         sb->s_id);
 488         }
 489 }
 490
 491 #define ext4_error_ratelimit(sb)                                        \
 492                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 493                              "EXT4-fs error")
 494
 495 void __ext4_error(struct super_block *sb, const char *function,
 496                   unsigned int line, const char *fmt, ...)
 497 {
 498         struct va_format vaf;
 499         va_list args;
 500
 501         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 502                 return;
 503
 504         trace_ext4_error(sb, function, line);
 505         if (ext4_error_ratelimit(sb)) {
 506                 va_start(args, fmt);
 507                 vaf.fmt = fmt;
 508                 vaf.va = &args;
 509                 printk(KERN_CRIT
 510                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 511                        sb->s_id, function, line, current->comm, &vaf);
 512                 va_end(args);
 513         }
 514         save_error_info(sb, function, line);
 515         ext4_handle_error(sb);
 516 }
 517
 518 void __ext4_error_inode(struct inode *inode, const char *function,
 519                         unsigned int line, ext4_fsblk_t block,
 520                         const char *fmt, ...)
 521 {
 522         va_list args;
 523         struct va_format vaf;
 524         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 525
 526         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 527                 return;
 528
 529         trace_ext4_error(inode->i_sb, function, line);
 530         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 531         es->s_last_error_block = cpu_to_le64(block);
 532         if (ext4_error_ratelimit(inode->i_sb)) {
 533                 va_start(args, fmt);
 534                 vaf.fmt = fmt;
 535                 vaf.va = &args;
 536                 if (block)
 537                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 538                                "inode #%lu: block %llu: comm %s: %pV\n",
 539                                inode->i_sb->s_id, function, line, inode->i_ino,
 540                                block, current->comm, &vaf);
 541                 else
 542                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 543                                "inode #%lu: comm %s: %pV\n",
 544                                inode->i_sb->s_id, function, line, inode->i_ino,
 545                                current->comm, &vaf);
 546                 va_end(args);
 547         }
 548         save_error_info(inode->i_sb, function, line);
 549         ext4_handle_error(inode->i_sb);
 550 }
 551
 552 void __ext4_error_file(struct file *file, const char *function,
 553                        unsigned int line, ext4_fsblk_t block,
 554                        const char *fmt, ...)
 555 {
 556         va_list args;
 557         struct va_format vaf;
 558         struct ext4_super_block *es;
 559         struct inode *inode = file_inode(file);
 560         char pathname[80], *path;
 561
 562         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 563                 return;
 564
 565         trace_ext4_error(inode->i_sb, function, line);
 566         es = EXT4_SB(inode->i_sb)->s_es;
 567         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 568         if (ext4_error_ratelimit(inode->i_sb)) {
 569                 path = file_path(file, pathname, sizeof(pathname));
 570                 if (IS_ERR(path))
 571                         path = "(unknown)";
 572                 va_start(args, fmt);
 573                 vaf.fmt = fmt;
 574                 vaf.va = &args;
 575                 if (block)
 576                         printk(KERN_CRIT
 577                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 578                                "block %llu: comm %s: path %s: %pV\n",
 579                                inode->i_sb->s_id, function, line, inode->i_ino,
 580                                block, current->comm, path, &vaf);
 581                 else
 582                         printk(KERN_CRIT
 583                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 584                                "comm %s: path %s: %pV\n",
 585                                inode->i_sb->s_id, function, line, inode->i_ino,
 586                                current->comm, path, &vaf);
 587                 va_end(args);
 588         }
 589         save_error_info(inode->i_sb, function, line);
 590         ext4_handle_error(inode->i_sb);
 591 }
 592
 593 const char *ext4_decode_error(struct super_block *sb, int errno,
 594                               char nbuf[16])
 595 {
 596         char *errstr = NULL;
 597
 598         switch (errno) {
 599         case -EFSCORRUPTED:
 600                 errstr = "Corrupt filesystem";
 601                 break;
 602         case -EFSBADCRC:
 603                 errstr = "Filesystem failed CRC";
 604                 break;
 605         case -EIO:
 606                 errstr = "IO failure";
 607                 break;
 608         case -ENOMEM:
 609                 errstr = "Out of memory";
 610                 break;
 611         case -EROFS:
 612                 if (!sb || (EXT4_SB(sb)->s_journal &&
 613                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 614                         errstr = "Journal has aborted";
 615                 else
 616                         errstr = "Readonly filesystem";
 617                 break;
 618         default:
 619                 /* If the caller passed in an extra buffer for unknown
 620                  * errors, textualise them now.  Else we just return
 621                  * NULL. */
 622                 if (nbuf) {
 623                         /* Check for truncated error codes... */
 624                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 625                                 errstr = nbuf;
 626                 }
 627                 break;
 628         }
 629
 630         return errstr;
 631 }
 632
 633 /* __ext4_std_error decodes expected errors from journaling functions
 634  * automatically and invokes the appropriate error response.  */
 635
 636 void __ext4_std_error(struct super_block *sb, const char *function,
 637                       unsigned int line, int errno)
 638 {
 639         char nbuf[16];
 640         const char *errstr;
 641
 642         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 643                 return;
 644
 645         /* Special case: if the error is EROFS, and we're not already
 646          * inside a transaction, then there's really no point in logging
 647          * an error. */
 648         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 649                 return;
 650
 651         if (ext4_error_ratelimit(sb)) {
 652                 errstr = ext4_decode_error(sb, errno, nbuf);
 653                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 654                        sb->s_id, function, line, errstr);
 655         }
 656
 657         save_error_info(sb, function, line);
 658         ext4_handle_error(sb);
 659 }
 660
 661 /*
 662  * ext4_abort is a much stronger failure handler than ext4_error.  The
 663  * abort function may be used to deal with unrecoverable failures such
 664  * as journal IO errors or ENOMEM at a critical moment in log management.
 665  *
 666  * We unconditionally force the filesystem into an ABORT|READONLY state,
 667  * unless the error response on the fs has been set to panic in which
 668  * case we take the easy way out and panic immediately.
 669  */
 670
 671 void __ext4_abort(struct super_block *sb, const char *function,
 672                 unsigned int line, const char *fmt, ...)
 673 {
 674         struct va_format vaf;
 675         va_list args;
 676
 677         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 678                 return;
 679
 680         save_error_info(sb, function, line);
 681         va_start(args, fmt);
 682         vaf.fmt = fmt;
 683         vaf.va = &args;
 684         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 685                sb->s_id, function, line, &vaf);
 686         va_end(args);
 687
 688         if (sb_rdonly(sb) == 0) {
 689                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 690                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 691                 /*
 692                  * Make sure updated value of ->s_mount_flags will be visible
 693                  * before ->s_flags update
 694                  */
 695                 smp_wmb();
 696                 sb->s_flags |= SB_RDONLY;
 697                 if (EXT4_SB(sb)->s_journal)
 698                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 699                 save_error_info(sb, function, line);
 700         }
 701         if (test_opt(sb, ERRORS_PANIC) && !system_going_down()) {
 702                 if (EXT4_SB(sb)->s_journal &&
 703                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 704                         return;
 705                 panic("EXT4-fs panic from previous error\n");
 706         }
 707 }
 708
 709 void __ext4_msg(struct super_block *sb,
 710                 const char *prefix, const char *fmt, ...)
 711 {
 712         struct va_format vaf;
 713         va_list args;
 714
 715         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 716                 return;
 717
 718         va_start(args, fmt);
 719         vaf.fmt = fmt;
 720         vaf.va = &args;
 721         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 722         va_end(args);
 723 }
 724
 725 #define ext4_warning_ratelimit(sb)                                      \
 726                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 727                              "EXT4-fs warning")
 728
 729 void __ext4_warning(struct super_block *sb, const char *function,
 730                     unsigned int line, const char *fmt, ...)
 731 {
 732         struct va_format vaf;
 733         va_list args;
 734
 735         if (!ext4_warning_ratelimit(sb))
 736                 return;
 737
 738         va_start(args, fmt);
 739         vaf.fmt = fmt;
 740         vaf.va = &args;
 741         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 742                sb->s_id, function, line, &vaf);
 743         va_end(args);
 744 }
 745
 746 void __ext4_warning_inode(const struct inode *inode, const char *function,
 747                           unsigned int line, const char *fmt, ...)
 748 {
 749         struct va_format vaf;
 750         va_list args;
 751
 752         if (!ext4_warning_ratelimit(inode->i_sb))
 753                 return;
 754
 755         va_start(args, fmt);
 756         vaf.fmt = fmt;
 757         vaf.va = &args;
 758         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 759                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 760                function, line, inode->i_ino, current->comm, &vaf);
 761         va_end(args);
 762 }
 763
 764 void __ext4_grp_locked_error(const char *function, unsigned int line,
 765                              struct super_block *sb, ext4_group_t grp,
 766                              unsigned long ino, ext4_fsblk_t block,
 767                              const char *fmt, ...)
 768 __releases(bitlock)
 769 __acquires(bitlock)
 770 {
 771         struct va_format vaf;
 772         va_list args;
 773         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 774
 775         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 776                 return;
 777
 778         trace_ext4_error(sb, function, line);
 779         es->s_last_error_ino = cpu_to_le32(ino);
 780         es->s_last_error_block = cpu_to_le64(block);
 781         __save_error_info(sb, function, line);
 782
 783         if (ext4_error_ratelimit(sb)) {
 784                 va_start(args, fmt);
 785                 vaf.fmt = fmt;
 786                 vaf.va = &args;
 787                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 788                        sb->s_id, function, line, grp);
 789                 if (ino)
 790                         printk(KERN_CONT "inode %lu: ", ino);
 791                 if (block)
 792                         printk(KERN_CONT "block %llu:",
 793                                (unsigned long long) block);
 794                 printk(KERN_CONT "%pV\n", &vaf);
 795                 va_end(args);
 796         }
 797
 798         if (test_opt(sb, WARN_ON_ERROR))
 799                 WARN_ON_ONCE(1);
 800
 801         if (test_opt(sb, ERRORS_CONT)) {
 802                 ext4_commit_super(sb, 0);
 803                 return;
 804         }
 805
 806         ext4_unlock_group(sb, grp);
 807         ext4_commit_super(sb, 1);
 808         ext4_handle_error(sb);
 809         /*
 810          * We only get here in the ERRORS_RO case; relocking the group
 811          * may be dangerous, but nothing bad will happen since the
 812          * filesystem will have already been marked read/only and the
 813          * journal has been aborted.  We return 1 as a hint to callers
 814          * who might what to use the return value from
 815          * ext4_grp_locked_error() to distinguish between the
 816          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 817          * aggressively from the ext4 function in question, with a
 818          * more appropriate error code.
 819          */
 820         ext4_lock_group(sb, grp);
 821         return;
 822 }
 823
 824 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 825                                      ext4_group_t group,
 826                                      unsigned int flags)
 827 {
 828         struct ext4_sb_info *sbi = EXT4_SB(sb);
 829         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 830         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 831         int ret;
 832
 833         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 834                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 835                                             &grp->bb_state);
 836                 if (!ret)
 837                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 838                                            grp->bb_free);
 839         }
 840
 841         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 842                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 843                                             &grp->bb_state);
 844                 if (!ret && gdp) {
 845                         int count;
 846
 847                         count = ext4_free_inodes_count(sb, gdp);
 848                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 849                                            count);
 850                 }
 851         }
 852 }
 853
 854 void ext4_update_dynamic_rev(struct super_block *sb)
 855 {
 856         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 857
 858         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 859                 return;
 860
 861         ext4_warning(sb,
 862                      "updating to rev %d because of new feature flag, "
 863                      "running e2fsck is recommended",
 864                      EXT4_DYNAMIC_REV);
 865
 866         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 867         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 868         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 869         /* leave es->s_feature_*compat flags alone */
 870         /* es->s_uuid will be set by e2fsck if empty */
 871
 872         /*
 873          * The rest of the superblock fields should be zero, and if not it
 874          * means they are likely already in use, so leave them alone.  We
 875          * can leave it up to e2fsck to clean up any inconsistencies there.
 876          */
 877 }
 878
 879 /*
 880  * Open the external journal device
 881  */
 882 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 883 {
 884         struct block_device *bdev;
 885         char b[BDEVNAME_SIZE];
 886
 887         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 888         if (IS_ERR(bdev))
 889                 goto fail;
 890         return bdev;
 891
 892 fail:
 893         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 894                         __bdevname(dev, b), PTR_ERR(bdev));
 895         return NULL;
 896 }
 897
 898 /*
 899  * Release the journal device
 900  */
 901 static void ext4_blkdev_put(struct block_device *bdev)
 902 {
 903         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 904 }
 905
 906 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 907 {
 908         struct block_device *bdev;
 909         bdev = sbi->s_journal_bdev;
 910         if (bdev) {
 911                 /*
 912                  * Invalidate the journal device's buffers.  We don't want them
 913                  * floating about in memory - the physical journal device may
 914                  * hotswapped, and it breaks the `ro-after' testing code.
 915                  */
 916                 invalidate_bdev(bdev);
 917                 ext4_blkdev_put(bdev);
 918                 sbi->s_journal_bdev = NULL;
 919         }
 920 }
 921
 922 static inline struct inode *orphan_list_entry(struct list_head *l)
 923 {
 924         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 925 }
 926
 927 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 928 {
 929         struct list_head *l;
 930
 931         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 932                  le32_to_cpu(sbi->s_es->s_last_orphan));
 933
 934         printk(KERN_ERR "sb_info orphan list:\n");
 935         list_for_each(l, &sbi->s_orphan) {
 936                 struct inode *inode = orphan_list_entry(l);
 937                 printk(KERN_ERR "  "
 938                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 939                        inode->i_sb->s_id, inode->i_ino, inode,
 940                        inode->i_mode, inode->i_nlink,
 941                        NEXT_ORPHAN(inode));
 942         }
 943 }
 944
 945 #ifdef CONFIG_QUOTA
 946 static int ext4_quota_off(struct super_block *sb, int type);
 947
 948 static inline void ext4_quota_off_umount(struct super_block *sb)
 949 {
 950         int type;
 951
 952         /* Use our quota_off function to clear inode flags etc. */
 953         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 954                 ext4_quota_off(sb, type);
 955 }
 956
 957 /*
 958  * This is a helper function which is used in the mount/remount
 959  * codepaths (which holds s_umount) to fetch the quota file name.
 960  */
 961 static inline char *get_qf_name(struct super_block *sb,
 962                                 struct ext4_sb_info *sbi,
 963                                 int type)
 964 {
 965         return rcu_dereference_protected(sbi->s_qf_names[type],
 966                                          lockdep_is_held(&sb->s_umount));
 967 }
 968 #else
 969 static inline void ext4_quota_off_umount(struct super_block *sb)
 970 {
 971 }
 972 #endif
 973
 974 static void ext4_put_super(struct super_block *sb)
 975 {
 976         struct ext4_sb_info *sbi = EXT4_SB(sb);
 977         struct ext4_super_block *es = sbi->s_es;
 978         struct buffer_head **group_desc;
 979         struct flex_groups **flex_groups;
 980         int aborted = 0;
 981         int i, err;
 982
 983         ext4_unregister_li_request(sb);
 984         ext4_quota_off_umount(sb);
 985
 986         destroy_workqueue(sbi->rsv_conversion_wq);
 987
 988         if (sbi->s_journal) {
 989                 aborted = is_journal_aborted(sbi->s_journal);
 990                 err = jbd2_journal_destroy(sbi->s_journal);
 991                 sbi->s_journal = NULL;
 992                 if ((err < 0) && !aborted)
 993                         ext4_abort(sb, "Couldn't clean up the journal");
 994         }
 995
 996         ext4_unregister_sysfs(sb);
 997         ext4_es_unregister_shrinker(sbi);
 998         del_timer_sync(&sbi->s_err_report);
 999         ext4_release_system_zone(sb);
1000         ext4_mb_release(sb);
1001         ext4_ext_release(sb);
1002
1003         if (!sb_rdonly(sb) && !aborted) {
1004                 ext4_clear_feature_journal_needs_recovery(sb);
1005                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1006         }
1007         if (!sb_rdonly(sb))
1008                 ext4_commit_super(sb, 1);
1009
1010         rcu_read_lock();
1011         group_desc = rcu_dereference(sbi->s_group_desc);
1012         for (i = 0; i < sbi->s_gdb_count; i++)
1013                 brelse(group_desc[i]);
1014         kvfree(group_desc);
1015         flex_groups = rcu_dereference(sbi->s_flex_groups);
1016         if (flex_groups) {
1017                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
1018                         kvfree(flex_groups[i]);
1019                 kvfree(flex_groups);
1020         }
1021         rcu_read_unlock();
1022         percpu_counter_destroy(&sbi->s_freeclusters_counter);
1023         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1024         percpu_counter_destroy(&sbi->s_dirs_counter);
1025         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
1026         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
1027         percpu_free_rwsem(&sbi->s_writepages_rwsem);
1028 #ifdef CONFIG_QUOTA
1029         for (i = 0; i < EXT4_MAXQUOTAS; i++)
1030                 kfree(get_qf_name(sb, sbi, i));
1031 #endif
1032
1033         /* Debugging code just in case the in-memory inode orphan list
1034          * isn't empty.  The on-disk one can be non-empty if we've
1035          * detected an error and taken the fs readonly, but the
1036          * in-memory list had better be clean by this point. */
1037         if (!list_empty(&sbi->s_orphan))
1038                 dump_orphan_list(sb, sbi);
1039         J_ASSERT(list_empty(&sbi->s_orphan));
1040
1041         sync_blockdev(sb->s_bdev);
1042         invalidate_bdev(sb->s_bdev);
1043         if (sbi->s_journal_bdev && sbi->s_journal_bdev != sb->s_bdev) {
1044                 sync_blockdev(sbi->s_journal_bdev);
1045                 ext4_blkdev_remove(sbi);
1046         }
1047
1048         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1049         sbi->s_ea_inode_cache = NULL;
1050
1051         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1052         sbi->s_ea_block_cache = NULL;
1053
1054         if (sbi->s_mmp_tsk)
1055                 kthread_stop(sbi->s_mmp_tsk);
1056         brelse(sbi->s_sbh);
1057         sb->s_fs_info = NULL;
1058         /*
1059          * Now that we are completely done shutting down the
1060          * superblock, we need to actually destroy the kobject.
1061          */
1062         kobject_put(&sbi->s_kobj);
1063         wait_for_completion(&sbi->s_kobj_unregister);
1064         if (sbi->s_chksum_driver)
1065                 crypto_free_shash(sbi->s_chksum_driver);
1066         kfree(sbi->s_blockgroup_lock);
1067         fs_put_dax(sbi->s_daxdev);
1068 #ifdef CONFIG_UNICODE
1069         utf8_unload(sbi->s_encoding);
1070 #endif
1071         kfree(sbi);
1072 }
1073
1074 static struct kmem_cache *ext4_inode_cachep;
1075
1076 /*
1077  * Called inside transaction, so use GFP_NOFS
1078  */
1079 static struct inode *ext4_alloc_inode(struct super_block *sb)
1080 {
1081         struct ext4_inode_info *ei;
1082
1083         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1084         if (!ei)
1085                 return NULL;
1086
1087         inode_set_iversion(&ei->vfs_inode, 1);
1088         ei->i_flags = 0;
1089         spin_lock_init(&ei->i_raw_lock);
1090         INIT_LIST_HEAD(&ei->i_prealloc_list);
1091         spin_lock_init(&ei->i_prealloc_lock);
1092         ext4_es_init_tree(&ei->i_es_tree);
1093         rwlock_init(&ei->i_es_lock);
1094         INIT_LIST_HEAD(&ei->i_es_list);
1095         ei->i_es_all_nr = 0;
1096         ei->i_es_shk_nr = 0;
1097         ei->i_es_shrink_lblk = 0;
1098         ei->i_reserved_data_blocks = 0;
1099         ei->i_da_metadata_calc_len = 0;
1100         ei->i_da_metadata_calc_last_lblock = 0;
1101         spin_lock_init(&(ei->i_block_reservation_lock));
1102         ext4_init_pending_tree(&ei->i_pending_tree);
1103 #ifdef CONFIG_QUOTA
1104         ei->i_reserved_quota = 0;
1105         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1106 #endif
1107         ei->jinode = NULL;
1108         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1109         spin_lock_init(&ei->i_completed_io_lock);
1110         ei->i_sync_tid = 0;
1111         ei->i_datasync_tid = 0;
1112         atomic_set(&ei->i_unwritten, 0);
1113         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1114         return &ei->vfs_inode;
1115 }
1116
1117 static int ext4_drop_inode(struct inode *inode)
1118 {
1119         int drop = generic_drop_inode(inode);
1120
1121         if (!drop)
1122                 drop = fscrypt_drop_inode(inode);
1123
1124         trace_ext4_drop_inode(inode, drop);
1125         return drop;
1126 }
1127
1128 static void ext4_free_in_core_inode(struct inode *inode)
1129 {
1130         fscrypt_free_inode(inode);
1131         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1132 }
1133
1134 static void ext4_destroy_inode(struct inode *inode)
1135 {
1136         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1137                 ext4_msg(inode->i_sb, KERN_ERR,
1138                          "Inode %lu (%p): orphan list check failed!",
1139                          inode->i_ino, EXT4_I(inode));
1140                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1141                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1142                                 true);
1143                 dump_stack();
1144         }
1145
1146         if (EXT4_I(inode)->i_reserved_data_blocks)
1147                 ext4_msg(inode->i_sb, KERN_ERR,
1148                          "Inode %lu (%p): i_reserved_data_blocks (%u) not cleared!",
1149                          inode->i_ino, EXT4_I(inode),
1150                          EXT4_I(inode)->i_reserved_data_blocks);
1151 }
1152
1153 static void init_once(void *foo)
1154 {
1155         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1156
1157         INIT_LIST_HEAD(&ei->i_orphan);
1158         init_rwsem(&ei->xattr_sem);
1159         init_rwsem(&ei->i_data_sem);
1160         init_rwsem(&ei->i_mmap_sem);
1161         inode_init_once(&ei->vfs_inode);
1162 }
1163
1164 static int __init init_inodecache(void)
1165 {
1166         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1167                                 sizeof(struct ext4_inode_info), 0,
1168                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1169                                         SLAB_ACCOUNT),
1170                                 offsetof(struct ext4_inode_info, i_data),
1171                                 sizeof_field(struct ext4_inode_info, i_data),
1172                                 init_once);
1173         if (ext4_inode_cachep == NULL)
1174                 return -ENOMEM;
1175         return 0;
1176 }
1177
1178 static void destroy_inodecache(void)
1179 {
1180         /*
1181          * Make sure all delayed rcu free inodes are flushed before we
1182          * destroy cache.
1183          */
1184         rcu_barrier();
1185         kmem_cache_destroy(ext4_inode_cachep);
1186 }
1187
1188 void ext4_clear_inode(struct inode *inode)
1189 {
1190         invalidate_inode_buffers(inode);
1191         clear_inode(inode);
1192         ext4_discard_preallocations(inode);
1193         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1194         dquot_drop(inode);
1195         if (EXT4_I(inode)->jinode) {
1196                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1197                                                EXT4_I(inode)->jinode);
1198                 jbd2_free_inode(EXT4_I(inode)->jinode);
1199                 EXT4_I(inode)->jinode = NULL;
1200         }
1201         fscrypt_put_encryption_info(inode);
1202         fsverity_cleanup_inode(inode);
1203 }
1204
1205 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1206                                         u64 ino, u32 generation)
1207 {
1208         struct inode *inode;
1209
1210         /*
1211          * Currently we don't know the generation for parent directory, so
1212          * a generation of 0 means "accept any"
1213          */
1214         inode = ext4_iget(sb, ino, EXT4_IGET_HANDLE);
1215         if (IS_ERR(inode))
1216                 return ERR_CAST(inode);
1217         if (generation && inode->i_generation != generation) {
1218                 iput(inode);
1219                 return ERR_PTR(-ESTALE);
1220         }
1221
1222         return inode;
1223 }
1224
1225 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1226                                         int fh_len, int fh_type)
1227 {
1228         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1229                                     ext4_nfs_get_inode);
1230 }
1231
1232 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1233                                         int fh_len, int fh_type)
1234 {
1235         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1236                                     ext4_nfs_get_inode);
1237 }
1238
1239 static int ext4_nfs_commit_metadata(struct inode *inode)
1240 {
1241         struct writeback_control wbc = {
1242                 .sync_mode = WB_SYNC_ALL
1243         };
1244
1245         trace_ext4_nfs_commit_metadata(inode);
1246         return ext4_write_inode(inode, &wbc);
1247 }
1248
1249 /*
1250  * Try to release metadata pages (indirect blocks, directories) which are
1251  * mapped via the block device.  Since these pages could have journal heads
1252  * which would prevent try_to_free_buffers() from freeing them, we must use
1253  * jbd2 layer's try_to_free_buffers() function to release them.
1254  */
1255 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1256                                  gfp_t wait)
1257 {
1258         journal_t *journal = EXT4_SB(sb)->s_journal;
1259
1260         WARN_ON(PageChecked(page));
1261         if (!page_has_buffers(page))
1262                 return 0;
1263         if (journal)
1264                 return jbd2_journal_try_to_free_buffers(journal, page,
1265                                                 wait & ~__GFP_DIRECT_RECLAIM);
1266         return try_to_free_buffers(page);
1267 }
1268
1269 #ifdef CONFIG_FS_ENCRYPTION
1270 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1271 {
1272         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1273                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1274 }
1275
1276 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1277                                                         void *fs_data)
1278 {
1279         handle_t *handle = fs_data;
1280         int res, res2, credits, retries = 0;
1281
1282         /*
1283          * Encrypting the root directory is not allowed because e2fsck expects
1284          * lost+found to exist and be unencrypted, and encrypting the root
1285          * directory would imply encrypting the lost+found directory as well as
1286          * the filename "lost+found" itself.
1287          */
1288         if (inode->i_ino == EXT4_ROOT_INO)
1289                 return -EPERM;
1290
1291         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1292                 return -EINVAL;
1293
1294         res = ext4_convert_inline_data(inode);
1295         if (res)
1296                 return res;
1297
1298         /*
1299          * If a journal handle was specified, then the encryption context is
1300          * being set on a new inode via inheritance and is part of a larger
1301          * transaction to create the inode.  Otherwise the encryption context is
1302          * being set on an existing inode in its own transaction.  Only in the
1303          * latter case should the "retry on ENOSPC" logic be used.
1304          */
1305
1306         if (handle) {
1307                 res = ext4_xattr_set_handle(handle, inode,
1308                                             EXT4_XATTR_INDEX_ENCRYPTION,
1309                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1310                                             ctx, len, 0);
1311                 if (!res) {
1312                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1313                         ext4_clear_inode_state(inode,
1314                                         EXT4_STATE_MAY_INLINE_DATA);
1315                         /*
1316                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1317                          * S_DAX may be disabled
1318                          */
1319                         ext4_set_inode_flags(inode);
1320                 }
1321                 return res;
1322         }
1323
1324         res = dquot_initialize(inode);
1325         if (res)
1326                 return res;
1327 retry:
1328         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1329                                      &credits);
1330         if (res)
1331                 return res;
1332
1333         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1334         if (IS_ERR(handle))
1335                 return PTR_ERR(handle);
1336
1337         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1338                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1339                                     ctx, len, 0);
1340         if (!res) {
1341                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1342                 /*
1343                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1344                  * S_DAX may be disabled
1345                  */
1346                 ext4_set_inode_flags(inode);
1347                 res = ext4_mark_inode_dirty(handle, inode);
1348                 if (res)
1349                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1350         }
1351         res2 = ext4_journal_stop(handle);
1352
1353         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1354                 goto retry;
1355         if (!res)
1356                 res = res2;
1357         return res;
1358 }
1359
1360 static bool ext4_dummy_context(struct inode *inode)
1361 {
1362         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1363 }
1364
1365 static const struct fscrypt_operations ext4_cryptops = {
1366         .key_prefix             = "ext4:",
1367         .get_context            = ext4_get_context,
1368         .set_context            = ext4_set_context,
1369         .dummy_context          = ext4_dummy_context,
1370         .empty_dir              = ext4_empty_dir,
1371         .max_namelen            = EXT4_NAME_LEN,
1372 };
1373 #endif
1374
1375 #ifdef CONFIG_QUOTA
1376 static const char * const quotatypes[] = INITQFNAMES;
1377 #define QTYPE2NAME(t) (quotatypes[t])
1378
1379 static int ext4_write_dquot(struct dquot *dquot);
1380 static int ext4_acquire_dquot(struct dquot *dquot);
1381 static int ext4_release_dquot(struct dquot *dquot);
1382 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1383 static int ext4_write_info(struct super_block *sb, int type);
1384 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1385                          const struct path *path);
1386 static int ext4_quota_on_mount(struct super_block *sb, int type);
1387 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1388                                size_t len, loff_t off);
1389 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1390                                 const char *data, size_t len, loff_t off);
1391 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1392                              unsigned int flags);
1393 static int ext4_enable_quotas(struct super_block *sb);
1394 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1395
1396 static struct dquot **ext4_get_dquots(struct inode *inode)
1397 {
1398         return EXT4_I(inode)->i_dquot;
1399 }
1400
1401 static const struct dquot_operations ext4_quota_operations = {
1402         .get_reserved_space     = ext4_get_reserved_space,
1403         .write_dquot            = ext4_write_dquot,
1404         .acquire_dquot          = ext4_acquire_dquot,
1405         .release_dquot          = ext4_release_dquot,
1406         .mark_dirty             = ext4_mark_dquot_dirty,
1407         .write_info             = ext4_write_info,
1408         .alloc_dquot            = dquot_alloc,
1409         .destroy_dquot          = dquot_destroy,
1410         .get_projid             = ext4_get_projid,
1411         .get_inode_usage        = ext4_get_inode_usage,
1412         .get_next_id            = ext4_get_next_id,
1413 };
1414
1415 static const struct quotactl_ops ext4_qctl_operations = {
1416         .quota_on       = ext4_quota_on,
1417         .quota_off      = ext4_quota_off,
1418         .quota_sync     = dquot_quota_sync,
1419         .get_state      = dquot_get_state,
1420         .set_info       = dquot_set_dqinfo,
1421         .get_dqblk      = dquot_get_dqblk,
1422         .set_dqblk      = dquot_set_dqblk,
1423         .get_nextdqblk  = dquot_get_next_dqblk,
1424 };
1425 #endif
1426
1427 static const struct super_operations ext4_sops = {
1428         .alloc_inode    = ext4_alloc_inode,
1429         .free_inode     = ext4_free_in_core_inode,
1430         .destroy_inode  = ext4_destroy_inode,
1431         .write_inode    = ext4_write_inode,
1432         .dirty_inode    = ext4_dirty_inode,
1433         .drop_inode     = ext4_drop_inode,
1434         .evict_inode    = ext4_evict_inode,
1435         .put_super      = ext4_put_super,
1436         .sync_fs        = ext4_sync_fs,
1437         .freeze_fs      = ext4_freeze,
1438         .unfreeze_fs    = ext4_unfreeze,
1439         .statfs         = ext4_statfs,
1440         .remount_fs     = ext4_remount,
1441         .show_options   = ext4_show_options,
1442 #ifdef CONFIG_QUOTA
1443         .quota_read     = ext4_quota_read,
1444         .quota_write    = ext4_quota_write,
1445         .get_dquots     = ext4_get_dquots,
1446 #endif
1447         .bdev_try_to_free_page = bdev_try_to_free_page,
1448 };
1449
1450 static const struct export_operations ext4_export_ops = {
1451         .fh_to_dentry = ext4_fh_to_dentry,
1452         .fh_to_parent = ext4_fh_to_parent,
1453         .get_parent = ext4_get_parent,
1454         .commit_metadata = ext4_nfs_commit_metadata,
1455 };
1456
1457 enum {
1458         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1459         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1460         Opt_nouid32, Opt_debug, Opt_removed,
1461         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1462         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1463         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1464         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1465         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1466         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1467         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1468         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1469         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1470         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1471         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1472         Opt_nowarn_on_error, Opt_mblk_io_submit,
1473         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1474         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1475         Opt_inode_readahead_blks, Opt_journal_ioprio,
1476         Opt_dioread_nolock, Opt_dioread_lock,
1477         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1478         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1479 };
1480
1481 static const match_table_t tokens = {
1482         {Opt_bsd_df, "bsddf"},
1483         {Opt_minix_df, "minixdf"},
1484         {Opt_grpid, "grpid"},
1485         {Opt_grpid, "bsdgroups"},
1486         {Opt_nogrpid, "nogrpid"},
1487         {Opt_nogrpid, "sysvgroups"},
1488         {Opt_resgid, "resgid=%u"},
1489         {Opt_resuid, "resuid=%u"},
1490         {Opt_sb, "sb=%u"},
1491         {Opt_err_cont, "errors=continue"},
1492         {Opt_err_panic, "errors=panic"},
1493         {Opt_err_ro, "errors=remount-ro"},
1494         {Opt_nouid32, "nouid32"},
1495         {Opt_debug, "debug"},
1496         {Opt_removed, "oldalloc"},
1497         {Opt_removed, "orlov"},
1498         {Opt_user_xattr, "user_xattr"},
1499         {Opt_nouser_xattr, "nouser_xattr"},
1500         {Opt_acl, "acl"},
1501         {Opt_noacl, "noacl"},
1502         {Opt_noload, "norecovery"},
1503         {Opt_noload, "noload"},
1504         {Opt_removed, "nobh"},
1505         {Opt_removed, "bh"},
1506         {Opt_commit, "commit=%u"},
1507         {Opt_min_batch_time, "min_batch_time=%u"},
1508         {Opt_max_batch_time, "max_batch_time=%u"},
1509         {Opt_journal_dev, "journal_dev=%u"},
1510         {Opt_journal_path, "journal_path=%s"},
1511         {Opt_journal_checksum, "journal_checksum"},
1512         {Opt_nojournal_checksum, "nojournal_checksum"},
1513         {Opt_journal_async_commit, "journal_async_commit"},
1514         {Opt_abort, "abort"},
1515         {Opt_data_journal, "data=journal"},
1516         {Opt_data_ordered, "data=ordered"},
1517         {Opt_data_writeback, "data=writeback"},
1518         {Opt_data_err_abort, "data_err=abort"},
1519         {Opt_data_err_ignore, "data_err=ignore"},
1520         {Opt_offusrjquota, "usrjquota="},
1521         {Opt_usrjquota, "usrjquota=%s"},
1522         {Opt_offgrpjquota, "grpjquota="},
1523         {Opt_grpjquota, "grpjquota=%s"},
1524         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1525         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1526         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1527         {Opt_grpquota, "grpquota"},
1528         {Opt_noquota, "noquota"},
1529         {Opt_quota, "quota"},
1530         {Opt_usrquota, "usrquota"},
1531         {Opt_prjquota, "prjquota"},
1532         {Opt_barrier, "barrier=%u"},
1533         {Opt_barrier, "barrier"},
1534         {Opt_nobarrier, "nobarrier"},
1535         {Opt_i_version, "i_version"},
1536         {Opt_dax, "dax"},
1537         {Opt_stripe, "stripe=%u"},
1538         {Opt_delalloc, "delalloc"},
1539         {Opt_warn_on_error, "warn_on_error"},
1540         {Opt_nowarn_on_error, "nowarn_on_error"},
1541         {Opt_lazytime, "lazytime"},
1542         {Opt_nolazytime, "nolazytime"},
1543         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1544         {Opt_nodelalloc, "nodelalloc"},
1545         {Opt_removed, "mblk_io_submit"},
1546         {Opt_removed, "nomblk_io_submit"},
1547         {Opt_block_validity, "block_validity"},
1548         {Opt_noblock_validity, "noblock_validity"},
1549         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1550         {Opt_journal_ioprio, "journal_ioprio=%u"},
1551         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1552         {Opt_auto_da_alloc, "auto_da_alloc"},
1553         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1554         {Opt_dioread_nolock, "dioread_nolock"},
1555         {Opt_dioread_lock, "dioread_lock"},
1556         {Opt_discard, "discard"},
1557         {Opt_nodiscard, "nodiscard"},
1558         {Opt_init_itable, "init_itable=%u"},
1559         {Opt_init_itable, "init_itable"},
1560         {Opt_noinit_itable, "noinit_itable"},
1561         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1562         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1563         {Opt_nombcache, "nombcache"},
1564         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1565         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1566         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1567         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1568         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1569         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1570         {Opt_err, NULL},
1571 };
1572
1573 static ext4_fsblk_t get_sb_block(void **data)
1574 {
1575         ext4_fsblk_t    sb_block;
1576         char            *options = (char *) *data;
1577
1578         if (!options || strncmp(options, "sb=", 3) != 0)
1579                 return 1;       /* Default location */
1580
1581         options += 3;
1582         /* TODO: use simple_strtoll with >32bit ext4 */
1583         sb_block = simple_strtoul(options, &options, 0);
1584         if (*options && *options != ',') {
1585                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1586                        (char *) *data);
1587                 return 1;
1588         }
1589         if (*options == ',')
1590                 options++;
1591         *data = (void *) options;
1592
1593         return sb_block;
1594 }
1595
1596 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1597 static const char deprecated_msg[] =
1598         "Mount option \"%s\" will be removed by %s\n"
1599         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1600
1601 #ifdef CONFIG_QUOTA
1602 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1603 {
1604         struct ext4_sb_info *sbi = EXT4_SB(sb);
1605         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1606         int ret = -1;
1607
1608         if (sb_any_quota_loaded(sb) && !old_qname) {
1609                 ext4_msg(sb, KERN_ERR,
1610                         "Cannot change journaled "
1611                         "quota options when quota turned on");
1612                 return -1;
1613         }
1614         if (ext4_has_feature_quota(sb)) {
1615                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1616                          "ignored when QUOTA feature is enabled");
1617                 return 1;
1618         }
1619         qname = match_strdup(args);
1620         if (!qname) {
1621                 ext4_msg(sb, KERN_ERR,
1622                         "Not enough memory for storing quotafile name");
1623                 return -1;
1624         }
1625         if (old_qname) {
1626                 if (strcmp(old_qname, qname) == 0)
1627                         ret = 1;
1628                 else
1629                         ext4_msg(sb, KERN_ERR,
1630                                  "%s quota file already specified",
1631                                  QTYPE2NAME(qtype));
1632                 goto errout;
1633         }
1634         if (strchr(qname, '/')) {
1635                 ext4_msg(sb, KERN_ERR,
1636                         "quotafile must be on filesystem root");
1637                 goto errout;
1638         }
1639         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1640         set_opt(sb, QUOTA);
1641         return 1;
1642 errout:
1643         kfree(qname);
1644         return ret;
1645 }
1646
1647 static int clear_qf_name(struct super_block *sb, int qtype)
1648 {
1649
1650         struct ext4_sb_info *sbi = EXT4_SB(sb);
1651         char *old_qname = get_qf_name(sb, sbi, qtype);
1652
1653         if (sb_any_quota_loaded(sb) && old_qname) {
1654                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1655                         " when quota turned on");
1656                 return -1;
1657         }
1658         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1659         synchronize_rcu();
1660         kfree(old_qname);
1661         return 1;
1662 }
1663 #endif
1664
1665 #define MOPT_SET        0x0001
1666 #define MOPT_CLEAR      0x0002
1667 #define MOPT_NOSUPPORT  0x0004
1668 #define MOPT_EXPLICIT   0x0008
1669 #define MOPT_CLEAR_ERR  0x0010
1670 #define MOPT_GTE0       0x0020
1671 #ifdef CONFIG_QUOTA
1672 #define MOPT_Q          0
1673 #define MOPT_QFMT       0x0040
1674 #else
1675 #define MOPT_Q          MOPT_NOSUPPORT
1676 #define MOPT_QFMT       MOPT_NOSUPPORT
1677 #endif
1678 #define MOPT_DATAJ      0x0080
1679 #define MOPT_NO_EXT2    0x0100
1680 #define MOPT_NO_EXT3    0x0200
1681 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1682 #define MOPT_STRING     0x0400
1683
1684 static const struct mount_opts {
1685         int     token;
1686         int     mount_opt;
1687         int     flags;
1688 } ext4_mount_opts[] = {
1689         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1690         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1691         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1692         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1693         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1694         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1695         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1696          MOPT_EXT4_ONLY | MOPT_SET},
1697         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1698          MOPT_EXT4_ONLY | MOPT_CLEAR},
1699         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1700         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1701         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1702          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1703         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1704          MOPT_EXT4_ONLY | MOPT_CLEAR},
1705         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1706         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1707         {Opt_commit, 0, MOPT_NO_EXT2},
1708         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1709          MOPT_EXT4_ONLY | MOPT_CLEAR},
1710         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1711          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1712         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1713                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1714          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1715         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1716         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1717         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1718         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1719         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1720          MOPT_NO_EXT2},
1721         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1722          MOPT_NO_EXT2},
1723         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1724         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1725         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1726         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1727         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1728         {Opt_commit, 0, MOPT_GTE0},
1729         {Opt_max_batch_time, 0, MOPT_GTE0},
1730         {Opt_min_batch_time, 0, MOPT_GTE0},
1731         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1732         {Opt_init_itable, 0, MOPT_GTE0},
1733         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1734         {Opt_stripe, 0, MOPT_GTE0},
1735         {Opt_resuid, 0, MOPT_GTE0},
1736         {Opt_resgid, 0, MOPT_GTE0},
1737         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1738         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1739         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1740         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1741         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1742         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1743          MOPT_NO_EXT2 | MOPT_DATAJ},
1744         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1745         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1746 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1747         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1748         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1749 #else
1750         {Opt_acl, 0, MOPT_NOSUPPORT},
1751         {Opt_noacl, 0, MOPT_NOSUPPORT},
1752 #endif
1753         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1754         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1755         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1756         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1757         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1758                                                         MOPT_SET | MOPT_Q},
1759         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1760                                                         MOPT_SET | MOPT_Q},
1761         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1762                                                         MOPT_SET | MOPT_Q},
1763         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1764                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1765                                                         MOPT_CLEAR | MOPT_Q},
1766         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
1767         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
1768         {Opt_offusrjquota, 0, MOPT_Q},
1769         {Opt_offgrpjquota, 0, MOPT_Q},
1770         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1771         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1772         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1773         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1774         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1775         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1776         {Opt_err, 0, 0}
1777 };
1778
1779 #ifdef CONFIG_UNICODE
1780 static const struct ext4_sb_encodings {
1781         __u16 magic;
1782         char *name;
1783         char *version;
1784 } ext4_sb_encoding_map[] = {
1785         {EXT4_ENC_UTF8_12_1, "utf8", "12.1.0"},
1786 };
1787
1788 static int ext4_sb_read_encoding(const struct ext4_super_block *es,
1789                                  const struct ext4_sb_encodings **encoding,
1790                                  __u16 *flags)
1791 {
1792         __u16 magic = le16_to_cpu(es->s_encoding);
1793         int i;
1794
1795         for (i = 0; i < ARRAY_SIZE(ext4_sb_encoding_map); i++)
1796                 if (magic == ext4_sb_encoding_map[i].magic)
1797                         break;
1798
1799         if (i >= ARRAY_SIZE(ext4_sb_encoding_map))
1800                 return -EINVAL;
1801
1802         *encoding = &ext4_sb_encoding_map[i];
1803         *flags = le16_to_cpu(es->s_encoding_flags);
1804
1805         return 0;
1806 }
1807 #endif
1808
1809 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1810                             substring_t *args, unsigned long *journal_devnum,
1811                             unsigned int *journal_ioprio, int is_remount)
1812 {
1813         struct ext4_sb_info *sbi = EXT4_SB(sb);
1814         const struct mount_opts *m;
1815         kuid_t uid;
1816         kgid_t gid;
1817         int arg = 0;
1818
1819 #ifdef CONFIG_QUOTA
1820         if (token == Opt_usrjquota)
1821                 return set_qf_name(sb, USRQUOTA, &args[0]);
1822         else if (token == Opt_grpjquota)
1823                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1824         else if (token == Opt_offusrjquota)
1825                 return clear_qf_name(sb, USRQUOTA);
1826         else if (token == Opt_offgrpjquota)
1827                 return clear_qf_name(sb, GRPQUOTA);
1828 #endif
1829         switch (token) {
1830         case Opt_noacl:
1831         case Opt_nouser_xattr:
1832                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1833                 break;
1834         case Opt_sb:
1835                 return 1;       /* handled by get_sb_block() */
1836         case Opt_removed:
1837                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1838                 return 1;
1839         case Opt_abort:
1840                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1841                 return 1;
1842         case Opt_i_version:
1843                 sb->s_flags |= SB_I_VERSION;
1844                 return 1;
1845         case Opt_lazytime:
1846                 sb->s_flags |= SB_LAZYTIME;
1847                 return 1;
1848         case Opt_nolazytime:
1849                 sb->s_flags &= ~SB_LAZYTIME;
1850                 return 1;
1851         }
1852
1853         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1854                 if (token == m->token)
1855                         break;
1856
1857         if (m->token == Opt_err) {
1858                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1859                          "or missing value", opt);
1860                 return -1;
1861         }
1862
1863         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1864                 ext4_msg(sb, KERN_ERR,
1865                          "Mount option \"%s\" incompatible with ext2", opt);
1866                 return -1;
1867         }
1868         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1869                 ext4_msg(sb, KERN_ERR,
1870                          "Mount option \"%s\" incompatible with ext3", opt);
1871                 return -1;
1872         }
1873
1874         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1875                 return -1;
1876         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1877                 return -1;
1878         if (m->flags & MOPT_EXPLICIT) {
1879                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1880                         set_opt2(sb, EXPLICIT_DELALLOC);
1881                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1882                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1883                 } else
1884                         return -1;
1885         }
1886         if (m->flags & MOPT_CLEAR_ERR)
1887                 clear_opt(sb, ERRORS_MASK);
1888         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1889                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1890                          "options when quota turned on");
1891                 return -1;
1892         }
1893
1894         if (m->flags & MOPT_NOSUPPORT) {
1895                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1896         } else if (token == Opt_commit) {
1897                 if (arg == 0)
1898                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1899                 else if (arg > INT_MAX / HZ) {
1900                         ext4_msg(sb, KERN_ERR,
1901                                  "Invalid commit interval %d, "
1902                                  "must be smaller than %d",
1903                                  arg, INT_MAX / HZ);
1904                         return -1;
1905                 }
1906                 sbi->s_commit_interval = HZ * arg;
1907         } else if (token == Opt_debug_want_extra_isize) {
1908                 if ((arg & 1) ||
1909                     (arg < 4) ||
1910                     (arg > (sbi->s_inode_size - EXT4_GOOD_OLD_INODE_SIZE))) {
1911                         ext4_msg(sb, KERN_ERR,
1912                                  "Invalid want_extra_isize %d", arg);
1913                         return -1;
1914                 }
1915                 sbi->s_want_extra_isize = arg;
1916         } else if (token == Opt_max_batch_time) {
1917                 sbi->s_max_batch_time = arg;
1918         } else if (token == Opt_min_batch_time) {
1919                 sbi->s_min_batch_time = arg;
1920         } else if (token == Opt_inode_readahead_blks) {
1921                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1922                         ext4_msg(sb, KERN_ERR,
1923                                  "EXT4-fs: inode_readahead_blks must be "
1924                                  "0 or a power of 2 smaller than 2^31");
1925                         return -1;
1926                 }
1927                 sbi->s_inode_readahead_blks = arg;
1928         } else if (token == Opt_init_itable) {
1929                 set_opt(sb, INIT_INODE_TABLE);
1930                 if (!args->from)
1931                         arg = EXT4_DEF_LI_WAIT_MULT;
1932                 sbi->s_li_wait_mult = arg;
1933         } else if (token == Opt_max_dir_size_kb) {
1934                 sbi->s_max_dir_size_kb = arg;
1935         } else if (token == Opt_stripe) {
1936                 sbi->s_stripe = arg;
1937         } else if (token == Opt_resuid) {
1938                 uid = make_kuid(current_user_ns(), arg);
1939                 if (!uid_valid(uid)) {
1940                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1941                         return -1;
1942                 }
1943                 sbi->s_resuid = uid;
1944         } else if (token == Opt_resgid) {
1945                 gid = make_kgid(current_user_ns(), arg);
1946                 if (!gid_valid(gid)) {
1947                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1948                         return -1;
1949                 }
1950                 sbi->s_resgid = gid;
1951         } else if (token == Opt_journal_dev) {
1952                 if (is_remount) {
1953                         ext4_msg(sb, KERN_ERR,
1954                                  "Cannot specify journal on remount");
1955                         return -1;
1956                 }
1957                 *journal_devnum = arg;
1958         } else if (token == Opt_journal_path) {
1959                 char *journal_path;
1960                 struct inode *journal_inode;
1961                 struct path path;
1962                 int error;
1963
1964                 if (is_remount) {
1965                         ext4_msg(sb, KERN_ERR,
1966                                  "Cannot specify journal on remount");
1967                         return -1;
1968                 }
1969                 journal_path = match_strdup(&args[0]);
1970                 if (!journal_path) {
1971                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1972                                 "journal device string");
1973                         return -1;
1974                 }
1975
1976                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1977                 if (error) {
1978                         ext4_msg(sb, KERN_ERR, "error: could not find "
1979                                 "journal device path: error %d", error);
1980                         kfree(journal_path);
1981                         return -1;
1982                 }
1983
1984                 journal_inode = d_inode(path.dentry);
1985                 if (!S_ISBLK(journal_inode->i_mode)) {
1986                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1987                                 "is not a block device", journal_path);
1988                         path_put(&path);
1989                         kfree(journal_path);
1990                         return -1;
1991                 }
1992
1993                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1994                 path_put(&path);
1995                 kfree(journal_path);
1996         } else if (token == Opt_journal_ioprio) {
1997                 if (arg > 7) {
1998                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1999                                  " (must be 0-7)");
2000                         return -1;
2001                 }
2002                 *journal_ioprio =
2003                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
2004         } else if (token == Opt_test_dummy_encryption) {
2005 #ifdef CONFIG_FS_ENCRYPTION
2006                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
2007                 ext4_msg(sb, KERN_WARNING,
2008                          "Test dummy encryption mode enabled");
2009 #else
2010                 ext4_msg(sb, KERN_WARNING,
2011                          "Test dummy encryption mount option ignored");
2012 #endif
2013         } else if (m->flags & MOPT_DATAJ) {
2014                 if (is_remount) {
2015                         if (!sbi->s_journal)
2016                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
2017                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
2018                                 ext4_msg(sb, KERN_ERR,
2019                                          "Cannot change data mode on remount");
2020                                 return -1;
2021                         }
2022                 } else {
2023                         clear_opt(sb, DATA_FLAGS);
2024                         sbi->s_mount_opt |= m->mount_opt;
2025                 }
2026 #ifdef CONFIG_QUOTA
2027         } else if (m->flags & MOPT_QFMT) {
2028                 if (sb_any_quota_loaded(sb) &&
2029                     sbi->s_jquota_fmt != m->mount_opt) {
2030                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
2031                                  "quota options when quota turned on");
2032                         return -1;
2033                 }
2034                 if (ext4_has_feature_quota(sb)) {
2035                         ext4_msg(sb, KERN_INFO,
2036                                  "Quota format mount options ignored "
2037                                  "when QUOTA feature is enabled");
2038                         return 1;
2039                 }
2040                 sbi->s_jquota_fmt = m->mount_opt;
2041 #endif
2042         } else if (token == Opt_dax) {
2043 #ifdef CONFIG_FS_DAX
2044                 if (is_remount && test_opt(sb, DAX)) {
2045                         ext4_msg(sb, KERN_ERR, "can't mount with "
2046                                 "both data=journal and dax");
2047                         return -1;
2048                 }
2049                 if (is_remount && !(sbi->s_mount_opt & EXT4_MOUNT_DAX)) {
2050                         ext4_msg(sb, KERN_ERR, "can't change "
2051                                         "dax mount option while remounting");
2052                         return -1;
2053                 }
2054                 ext4_msg(sb, KERN_WARNING,
2055                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
2056                 sbi->s_mount_opt |= m->mount_opt;
2057 #else
2058                 ext4_msg(sb, KERN_INFO, "dax option not supported");
2059                 return -1;
2060 #endif
2061         } else if (token == Opt_data_err_abort) {
2062                 sbi->s_mount_opt |= m->mount_opt;
2063         } else if (token == Opt_data_err_ignore) {
2064                 sbi->s_mount_opt &= ~m->mount_opt;
2065         } else {
2066                 if (!args->from)
2067                         arg = 1;
2068                 if (m->flags & MOPT_CLEAR)
2069                         arg = !arg;
2070                 else if (unlikely(!(m->flags & MOPT_SET))) {
2071                         ext4_msg(sb, KERN_WARNING,
2072                                  "buggy handling of option %s", opt);
2073                         WARN_ON(1);
2074                         return -1;
2075                 }
2076                 if (arg != 0)
2077                         sbi->s_mount_opt |= m->mount_opt;
2078                 else
2079                         sbi->s_mount_opt &= ~m->mount_opt;
2080         }
2081         return 1;
2082 }
2083
2084 static int parse_options(char *options, struct super_block *sb,
2085                          unsigned long *journal_devnum,
2086                          unsigned int *journal_ioprio,
2087                          int is_remount)
2088 {
2089         struct ext4_sb_info *sbi = EXT4_SB(sb);
2090         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
2091         substring_t args[MAX_OPT_ARGS];
2092         int token;
2093
2094         if (!options)
2095                 return 1;
2096
2097         while ((p = strsep(&options, ",")) != NULL) {
2098                 if (!*p)
2099                         continue;
2100                 /*
2101                  * Initialize args struct so we know whether arg was
2102                  * found; some options take optional arguments.
2103                  */
2104                 args[0].to = args[0].from = NULL;
2105                 token = match_token(p, tokens, args);
2106                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
2107                                      journal_ioprio, is_remount) < 0)
2108                         return 0;
2109         }
2110 #ifdef CONFIG_QUOTA
2111         /*
2112          * We do the test below only for project quotas. 'usrquota' and
2113          * 'grpquota' mount options are allowed even without quota feature
2114          * to support legacy quotas in quota files.
2115          */
2116         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2117                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2118                          "Cannot enable project quota enforcement.");
2119                 return 0;
2120         }
2121         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2122         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2123         if (usr_qf_name || grp_qf_name) {
2124                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2125                         clear_opt(sb, USRQUOTA);
2126
2127                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2128                         clear_opt(sb, GRPQUOTA);
2129
2130                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2131                         ext4_msg(sb, KERN_ERR, "old and new quota "
2132                                         "format mixing");
2133                         return 0;
2134                 }
2135
2136                 if (!sbi->s_jquota_fmt) {
2137                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2138                                         "not specified");
2139                         return 0;
2140                 }
2141         }
2142 #endif
2143         if (test_opt(sb, DIOREAD_NOLOCK)) {
2144                 int blocksize =
2145                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2146
2147                 if (blocksize < PAGE_SIZE) {
2148                         ext4_msg(sb, KERN_ERR, "can't mount with "
2149                                  "dioread_nolock if block size != PAGE_SIZE");
2150                         return 0;
2151                 }
2152         }
2153         return 1;
2154 }
2155
2156 static inline void ext4_show_quota_options(struct seq_file *seq,
2157                                            struct super_block *sb)
2158 {
2159 #if defined(CONFIG_QUOTA)
2160         struct ext4_sb_info *sbi = EXT4_SB(sb);
2161         char *usr_qf_name, *grp_qf_name;
2162
2163         if (sbi->s_jquota_fmt) {
2164                 char *fmtname = "";
2165
2166                 switch (sbi->s_jquota_fmt) {
2167                 case QFMT_VFS_OLD:
2168                         fmtname = "vfsold";
2169                         break;
2170                 case QFMT_VFS_V0:
2171                         fmtname = "vfsv0";
2172                         break;
2173                 case QFMT_VFS_V1:
2174                         fmtname = "vfsv1";
2175                         break;
2176                 }
2177                 seq_printf(seq, ",jqfmt=%s", fmtname);
2178         }
2179
2180         rcu_read_lock();
2181         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2182         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2183         if (usr_qf_name)
2184                 seq_show_option(seq, "usrjquota", usr_qf_name);
2185         if (grp_qf_name)
2186                 seq_show_option(seq, "grpjquota", grp_qf_name);
2187         rcu_read_unlock();
2188 #endif
2189 }
2190
2191 static const char *token2str(int token)
2192 {
2193         const struct match_token *t;
2194
2195         for (t = tokens; t->token != Opt_err; t++)
2196                 if (t->token == token && !strchr(t->pattern, '='))
2197                         break;
2198         return t->pattern;
2199 }
2200
2201 /*
2202  * Show an option if
2203  *  - it's set to a non-default value OR
2204  *  - if the per-sb default is different from the global default
2205  */
2206 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2207                               int nodefs)
2208 {
2209         struct ext4_sb_info *sbi = EXT4_SB(sb);
2210         struct ext4_super_block *es = sbi->s_es;
2211         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2212         const struct mount_opts *m;
2213         char sep = nodefs ? '\n' : ',';
2214
2215 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2216 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2217
2218         if (sbi->s_sb_block != 1)
2219                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2220
2221         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2222                 int want_set = m->flags & MOPT_SET;
2223                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2224                     (m->flags & MOPT_CLEAR_ERR))
2225                         continue;
2226                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2227                         continue; /* skip if same as the default */
2228                 if ((want_set &&
2229                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2230                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2231                         continue; /* select Opt_noFoo vs Opt_Foo */
2232                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2233         }
2234
2235         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2236             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2237                 SEQ_OPTS_PRINT("resuid=%u",
2238                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2239         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2240             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2241                 SEQ_OPTS_PRINT("resgid=%u",
2242                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2243         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2244         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2245                 SEQ_OPTS_PUTS("errors=remount-ro");
2246         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2247                 SEQ_OPTS_PUTS("errors=continue");
2248         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2249                 SEQ_OPTS_PUTS("errors=panic");
2250         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2251                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2252         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2253                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2254         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2255                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2256         if (sb->s_flags & SB_I_VERSION)
2257                 SEQ_OPTS_PUTS("i_version");
2258         if (nodefs || sbi->s_stripe)
2259                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2260         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2261                         (sbi->s_mount_opt ^ def_mount_opt)) {
2262                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2263                         SEQ_OPTS_PUTS("data=journal");
2264                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2265                         SEQ_OPTS_PUTS("data=ordered");
2266                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2267                         SEQ_OPTS_PUTS("data=writeback");
2268         }
2269         if (nodefs ||
2270             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2271                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2272                                sbi->s_inode_readahead_blks);
2273
2274         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2275                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2276                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2277         if (nodefs || sbi->s_max_dir_size_kb)
2278                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2279         if (test_opt(sb, DATA_ERR_ABORT))
2280                 SEQ_OPTS_PUTS("data_err=abort");
2281         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2282                 SEQ_OPTS_PUTS("test_dummy_encryption");
2283
2284         ext4_show_quota_options(seq, sb);
2285         return 0;
2286 }
2287
2288 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2289 {
2290         return _ext4_show_options(seq, root->d_sb, 0);
2291 }
2292
2293 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2294 {
2295         struct super_block *sb = seq->private;
2296         int rc;
2297
2298         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2299         rc = _ext4_show_options(seq, sb, 1);
2300         seq_puts(seq, "\n");
2301         return rc;
2302 }
2303
2304 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2305                             int read_only)
2306 {
2307         struct ext4_sb_info *sbi = EXT4_SB(sb);
2308         int err = 0;
2309
2310         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2311                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2312                          "forcing read-only mode");
2313                 err = -EROFS;
2314                 goto done;
2315         }
2316         if (read_only)
2317                 goto done;
2318         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2319                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2320                          "running e2fsck is recommended");
2321         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2322                 ext4_msg(sb, KERN_WARNING,
2323                          "warning: mounting fs with errors, "
2324                          "running e2fsck is recommended");
2325         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2326                  le16_to_cpu(es->s_mnt_count) >=
2327                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2328                 ext4_msg(sb, KERN_WARNING,
2329                          "warning: maximal mount count reached, "
2330                          "running e2fsck is recommended");
2331         else if (le32_to_cpu(es->s_checkinterval) &&
2332                  (ext4_get_tstamp(es, s_lastcheck) +
2333                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2334                 ext4_msg(sb, KERN_WARNING,
2335                          "warning: checktime reached, "
2336                          "running e2fsck is recommended");
2337         if (!sbi->s_journal)
2338                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2339         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2340                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2341         le16_add_cpu(&es->s_mnt_count, 1);
2342         ext4_update_tstamp(es, s_mtime);
2343         if (sbi->s_journal)
2344                 ext4_set_feature_journal_needs_recovery(sb);
2345
2346         err = ext4_commit_super(sb, 1);
2347 done:
2348         if (test_opt(sb, DEBUG))
2349                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2350                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2351                         sb->s_blocksize,
2352                         sbi->s_groups_count,
2353                         EXT4_BLOCKS_PER_GROUP(sb),
2354                         EXT4_INODES_PER_GROUP(sb),
2355                         sbi->s_mount_opt, sbi->s_mount_opt2);
2356
2357         cleancache_init_fs(sb);
2358         return err;
2359 }
2360
2361 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2362 {
2363         struct ext4_sb_info *sbi = EXT4_SB(sb);
2364         struct flex_groups **old_groups, **new_groups;
2365         int size, i, j;
2366
2367         if (!sbi->s_log_groups_per_flex)
2368                 return 0;
2369
2370         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2371         if (size <= sbi->s_flex_groups_allocated)
2372                 return 0;
2373
2374         new_groups = kvzalloc(roundup_pow_of_two(size *
2375                               sizeof(*sbi->s_flex_groups)), GFP_KERNEL);
2376         if (!new_groups) {
2377                 ext4_msg(sb, KERN_ERR,
2378                          "not enough memory for %d flex group pointers", size);
2379                 return -ENOMEM;
2380         }
2381         for (i = sbi->s_flex_groups_allocated; i < size; i++) {
2382                 new_groups[i] = kvzalloc(roundup_pow_of_two(
2383                                          sizeof(struct flex_groups)),
2384                                          GFP_KERNEL);
2385                 if (!new_groups[i]) {
2386                         for (j = sbi->s_flex_groups_allocated; j < i; j++)
2387                                 kvfree(new_groups[j]);
2388                         kvfree(new_groups);
2389                         ext4_msg(sb, KERN_ERR,
2390                                  "not enough memory for %d flex groups", size);
2391                         return -ENOMEM;
2392                 }
2393         }
2394         rcu_read_lock();
2395         old_groups = rcu_dereference(sbi->s_flex_groups);
2396         if (old_groups)
2397                 memcpy(new_groups, old_groups,
2398                        (sbi->s_flex_groups_allocated *
2399                         sizeof(struct flex_groups *)));
2400         rcu_read_unlock();
2401         rcu_assign_pointer(sbi->s_flex_groups, new_groups);
2402         sbi->s_flex_groups_allocated = size;
2403         if (old_groups)
2404                 ext4_kvfree_array_rcu(old_groups);
2405         return 0;
2406 }
2407
2408 static int ext4_fill_flex_info(struct super_block *sb)
2409 {
2410         struct ext4_sb_info *sbi = EXT4_SB(sb);
2411         struct ext4_group_desc *gdp = NULL;
2412         struct flex_groups *fg;
2413         ext4_group_t flex_group;
2414         int i, err;
2415
2416         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2417         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2418                 sbi->s_log_groups_per_flex = 0;
2419                 return 1;
2420         }
2421
2422         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2423         if (err)
2424                 goto failed;
2425
2426         for (i = 0; i < sbi->s_groups_count; i++) {
2427                 gdp = ext4_get_group_desc(sb, i, NULL);
2428
2429                 flex_group = ext4_flex_group(sbi, i);
2430                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
2431                 atomic_add(ext4_free_inodes_count(sb, gdp), &fg->free_inodes);
2432                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2433                              &fg->free_clusters);
2434                 atomic_add(ext4_used_dirs_count(sb, gdp), &fg->used_dirs);
2435         }
2436
2437         return 1;
2438 failed:
2439         return 0;
2440 }
2441
2442 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2443                                    struct ext4_group_desc *gdp)
2444 {
2445         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2446         __u16 crc = 0;
2447         __le32 le_group = cpu_to_le32(block_group);
2448         struct ext4_sb_info *sbi = EXT4_SB(sb);
2449
2450         if (ext4_has_metadata_csum(sbi->s_sb)) {
2451                 /* Use new metadata_csum algorithm */
2452                 __u32 csum32;
2453                 __u16 dummy_csum = 0;
2454
2455                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2456                                      sizeof(le_group));
2457                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2458                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2459                                      sizeof(dummy_csum));
2460                 offset += sizeof(dummy_csum);
2461                 if (offset < sbi->s_desc_size)
2462                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2463                                              sbi->s_desc_size - offset);
2464
2465                 crc = csum32 & 0xFFFF;
2466                 goto out;
2467         }
2468
2469         /* old crc16 code */
2470         if (!ext4_has_feature_gdt_csum(sb))
2471                 return 0;
2472
2473         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2474         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2475         crc = crc16(crc, (__u8 *)gdp, offset);
2476         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2477         /* for checksum of struct ext4_group_desc do the rest...*/
2478         if (ext4_has_feature_64bit(sb) && offset < sbi->s_desc_size)
2479                 crc = crc16(crc, (__u8 *)gdp + offset,
2480                             sbi->s_desc_size - offset);
2481
2482 out:
2483         return cpu_to_le16(crc);
2484 }
2485
2486 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2487                                 struct ext4_group_desc *gdp)
2488 {
2489         if (ext4_has_group_desc_csum(sb) &&
2490             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2491                 return 0;
2492
2493         return 1;
2494 }
2495
2496 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2497                               struct ext4_group_desc *gdp)
2498 {
2499         if (!ext4_has_group_desc_csum(sb))
2500                 return;
2501         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2502 }
2503
2504 /* Called at mount-time, super-block is locked */
2505 static int ext4_check_descriptors(struct super_block *sb,
2506                                   ext4_fsblk_t sb_block,
2507                                   ext4_group_t *first_not_zeroed)
2508 {
2509         struct ext4_sb_info *sbi = EXT4_SB(sb);
2510         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2511         ext4_fsblk_t last_block;
2512         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2513         ext4_fsblk_t block_bitmap;
2514         ext4_fsblk_t inode_bitmap;
2515         ext4_fsblk_t inode_table;
2516         int flexbg_flag = 0;
2517         ext4_group_t i, grp = sbi->s_groups_count;
2518
2519         if (ext4_has_feature_flex_bg(sb))
2520                 flexbg_flag = 1;
2521
2522         ext4_debug("Checking group descriptors");
2523
2524         for (i = 0; i < sbi->s_groups_count; i++) {
2525                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2526
2527                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2528                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2529                 else
2530                         last_block = first_block +
2531                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2532
2533                 if ((grp == sbi->s_groups_count) &&
2534                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2535                         grp = i;
2536
2537                 block_bitmap = ext4_block_bitmap(sb, gdp);
2538                 if (block_bitmap == sb_block) {
2539                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2540                                  "Block bitmap for group %u overlaps "
2541                                  "superblock", i);
2542                         if (!sb_rdonly(sb))
2543                                 return 0;
2544                 }
2545                 if (block_bitmap >= sb_block + 1 &&
2546                     block_bitmap <= last_bg_block) {
2547                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2548                                  "Block bitmap for group %u overlaps "
2549                                  "block group descriptors", i);
2550                         if (!sb_rdonly(sb))
2551                                 return 0;
2552                 }
2553                 if (block_bitmap < first_block || block_bitmap > last_block) {
2554                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2555                                "Block bitmap for group %u not in group "
2556                                "(block %llu)!", i, block_bitmap);
2557                         return 0;
2558                 }
2559                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2560                 if (inode_bitmap == sb_block) {
2561                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2562                                  "Inode bitmap for group %u overlaps "
2563                                  "superblock", i);
2564                         if (!sb_rdonly(sb))
2565                                 return 0;
2566                 }
2567                 if (inode_bitmap >= sb_block + 1 &&
2568                     inode_bitmap <= last_bg_block) {
2569                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2570                                  "Inode bitmap for group %u overlaps "
2571                                  "block group descriptors", i);
2572                         if (!sb_rdonly(sb))
2573                                 return 0;
2574                 }
2575                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2576                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2577                                "Inode bitmap for group %u not in group "
2578                                "(block %llu)!", i, inode_bitmap);
2579                         return 0;
2580                 }
2581                 inode_table = ext4_inode_table(sb, gdp);
2582                 if (inode_table == sb_block) {
2583                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2584                                  "Inode table for group %u overlaps "
2585                                  "superblock", i);
2586                         if (!sb_rdonly(sb))
2587                                 return 0;
2588                 }
2589                 if (inode_table >= sb_block + 1 &&
2590                     inode_table <= last_bg_block) {
2591                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2592                                  "Inode table for group %u overlaps "
2593                                  "block group descriptors", i);
2594                         if (!sb_rdonly(sb))
2595                                 return 0;
2596                 }
2597                 if (inode_table < first_block ||
2598                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2599                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2600                                "Inode table for group %u not in group "
2601                                "(block %llu)!", i, inode_table);
2602                         return 0;
2603                 }
2604                 ext4_lock_group(sb, i);
2605                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2606                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2607                                  "Checksum for group %u failed (%u!=%u)",
2608                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2609                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2610                         if (!sb_rdonly(sb)) {
2611                                 ext4_unlock_group(sb, i);
2612                                 return 0;
2613                         }
2614                 }
2615                 ext4_unlock_group(sb, i);
2616                 if (!flexbg_flag)
2617                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2618         }
2619         if (NULL != first_not_zeroed)
2620                 *first_not_zeroed = grp;
2621         return 1;
2622 }
2623
2624 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2625  * the superblock) which were deleted from all directories, but held open by
2626  * a process at the time of a crash.  We walk the list and try to delete these
2627  * inodes at recovery time (only with a read-write filesystem).
2628  *
2629  * In order to keep the orphan inode chain consistent during traversal (in
2630  * case of crash during recovery), we link each inode into the superblock
2631  * orphan list_head and handle it the same way as an inode deletion during
2632  * normal operation (which journals the operations for us).
2633  *
2634  * We only do an iget() and an iput() on each inode, which is very safe if we
2635  * accidentally point at an in-use or already deleted inode.  The worst that
2636  * can happen in this case is that we get a "bit already cleared" message from
2637  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2638  * e2fsck was run on this filesystem, and it must have already done the orphan
2639  * inode cleanup for us, so we can safely abort without any further action.
2640  */
2641 static void ext4_orphan_cleanup(struct super_block *sb,
2642                                 struct ext4_super_block *es)
2643 {
2644         unsigned int s_flags = sb->s_flags;
2645         int ret, nr_orphans = 0, nr_truncates = 0;
2646 #ifdef CONFIG_QUOTA
2647         int quota_update = 0;
2648         int i;
2649 #endif
2650         if (!es->s_last_orphan) {
2651                 jbd_debug(4, "no orphan inodes to clean up\n");
2652                 return;
2653         }
2654
2655         if (bdev_read_only(sb->s_bdev)) {
2656                 ext4_msg(sb, KERN_ERR, "write access "
2657                         "unavailable, skipping orphan cleanup");
2658                 return;
2659         }
2660
2661         /* Check if feature set would not allow a r/w mount */
2662         if (!ext4_feature_set_ok(sb, 0)) {
2663                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2664                          "unknown ROCOMPAT features");
2665                 return;
2666         }
2667
2668         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2669                 /* don't clear list on RO mount w/ errors */
2670                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2671                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2672                                   "clearing orphan list.\n");
2673                         es->s_last_orphan = 0;
2674                 }
2675                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2676                 return;
2677         }
2678
2679         if (s_flags & SB_RDONLY) {
2680                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2681                 sb->s_flags &= ~SB_RDONLY;
2682         }
2683 #ifdef CONFIG_QUOTA
2684         /*
2685          * Turn on quotas which were not enabled for read-only mounts if
2686          * filesystem has quota feature, so that they are updated correctly.
2687          */
2688         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2689                 int ret = ext4_enable_quotas(sb);
2690
2691                 if (!ret)
2692                         quota_update = 1;
2693                 else
2694                         ext4_msg(sb, KERN_ERR,
2695                                 "Cannot turn on quotas: error %d", ret);
2696         }
2697
2698         /* Turn on journaled quotas used for old sytle */
2699         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2700                 if (EXT4_SB(sb)->s_qf_names[i]) {
2701                         int ret = ext4_quota_on_mount(sb, i);
2702
2703                         if (!ret)
2704                                 quota_update = 1;
2705                         else
2706                                 ext4_msg(sb, KERN_ERR,
2707                                         "Cannot turn on journaled "
2708                                         "quota: type %d: error %d", i, ret);
2709                 }
2710         }
2711 #endif
2712
2713         while (es->s_last_orphan) {
2714                 struct inode *inode;
2715
2716                 /*
2717                  * We may have encountered an error during cleanup; if
2718                  * so, skip the rest.
2719                  */
2720                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2721                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2722                         es->s_last_orphan = 0;
2723                         break;
2724                 }
2725
2726                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2727                 if (IS_ERR(inode)) {
2728                         es->s_last_orphan = 0;
2729                         break;
2730                 }
2731
2732                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2733                 dquot_initialize(inode);
2734                 if (inode->i_nlink) {
2735                         if (test_opt(sb, DEBUG))
2736                                 ext4_msg(sb, KERN_DEBUG,
2737                                         "%s: truncating inode %lu to %lld bytes",
2738                                         __func__, inode->i_ino, inode->i_size);
2739                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2740                                   inode->i_ino, inode->i_size);
2741                         inode_lock(inode);
2742                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2743                         ret = ext4_truncate(inode);
2744                         if (ret) {
2745                                 /*
2746                                  * We need to clean up the in-core orphan list
2747                                  * manually if ext4_truncate() failed to get a
2748                                  * transaction handle.
2749                                  */
2750                                 ext4_orphan_del(NULL, inode);
2751                                 ext4_std_error(inode->i_sb, ret);
2752                         }
2753                         inode_unlock(inode);
2754                         nr_truncates++;
2755                 } else {
2756                         if (test_opt(sb, DEBUG))
2757                                 ext4_msg(sb, KERN_DEBUG,
2758                                         "%s: deleting unreferenced inode %lu",
2759                                         __func__, inode->i_ino);
2760                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2761                                   inode->i_ino);
2762                         nr_orphans++;
2763                 }
2764                 iput(inode);  /* The delete magic happens here! */
2765         }
2766
2767 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2768
2769         if (nr_orphans)
2770                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2771                        PLURAL(nr_orphans));
2772         if (nr_truncates)
2773                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2774                        PLURAL(nr_truncates));
2775 #ifdef CONFIG_QUOTA
2776         /* Turn off quotas if they were enabled for orphan cleanup */
2777         if (quota_update) {
2778                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2779                         if (sb_dqopt(sb)->files[i])
2780                                 dquot_quota_off(sb, i);
2781                 }
2782         }
2783 #endif
2784         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2785 }
2786
2787 /*
2788  * Maximal extent format file size.
2789  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2790  * extent format containers, within a sector_t, and within i_blocks
2791  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2792  * so that won't be a limiting factor.
2793  *
2794  * However there is other limiting factor. We do store extents in the form
2795  * of starting block and length, hence the resulting length of the extent
2796  * covering maximum file size must fit into on-disk format containers as
2797  * well. Given that length is always by 1 unit bigger than max unit (because
2798  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2799  *
2800  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2801  */
2802 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2803 {
2804         loff_t res;
2805         loff_t upper_limit = MAX_LFS_FILESIZE;
2806
2807         BUILD_BUG_ON(sizeof(blkcnt_t) < sizeof(u64));
2808
2809         if (!has_huge_files) {
2810                 upper_limit = (1LL << 32) - 1;
2811
2812                 /* total blocks in file system block size */
2813                 upper_limit >>= (blkbits - 9);
2814                 upper_limit <<= blkbits;
2815         }
2816
2817         /*
2818          * 32-bit extent-start container, ee_block. We lower the maxbytes
2819          * by one fs block, so ee_len can cover the extent of maximum file
2820          * size
2821          */
2822         res = (1LL << 32) - 1;
2823         res <<= blkbits;
2824
2825         /* Sanity check against vm- & vfs- imposed limits */
2826         if (res > upper_limit)
2827                 res = upper_limit;
2828
2829         return res;
2830 }
2831
2832 /*
2833  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2834  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2835  * We need to be 1 filesystem block less than the 2^48 sector limit.
2836  */
2837 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2838 {
2839         unsigned long long upper_limit, res = EXT4_NDIR_BLOCKS;
2840         int meta_blocks;
2841
2842         /*
2843          * This is calculated to be the largest file size for a dense, block
2844          * mapped file such that the file's total number of 512-byte sectors,
2845          * including data and all indirect blocks, does not exceed (2^48 - 1).
2846          *
2847          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2848          * number of 512-byte sectors of the file.
2849          */
2850         if (!has_huge_files) {
2851                 /*
2852                  * !has_huge_files or implies that the inode i_block field
2853                  * represents total file blocks in 2^32 512-byte sectors ==
2854                  * size of vfs inode i_blocks * 8
2855                  */
2856                 upper_limit = (1LL << 32) - 1;
2857
2858                 /* total blocks in file system block size */
2859                 upper_limit >>= (bits - 9);
2860
2861         } else {
2862                 /*
2863                  * We use 48 bit ext4_inode i_blocks
2864                  * With EXT4_HUGE_FILE_FL set the i_blocks
2865                  * represent total number of blocks in
2866                  * file system block size
2867                  */
2868                 upper_limit = (1LL << 48) - 1;
2869
2870         }
2871
2872         /* indirect blocks */
2873         meta_blocks = 1;
2874         /* double indirect blocks */
2875         meta_blocks += 1 + (1LL << (bits-2));
2876         /* tripple indirect blocks */
2877         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2878
2879         upper_limit -= meta_blocks;
2880         upper_limit <<= bits;
2881
2882         res += 1LL << (bits-2);
2883         res += 1LL << (2*(bits-2));
2884         res += 1LL << (3*(bits-2));
2885         res <<= bits;
2886         if (res > upper_limit)
2887                 res = upper_limit;
2888
2889         if (res > MAX_LFS_FILESIZE)
2890                 res = MAX_LFS_FILESIZE;
2891
2892         return (loff_t)res;
2893 }
2894
2895 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2896                                    ext4_fsblk_t logical_sb_block, int nr)
2897 {
2898         struct ext4_sb_info *sbi = EXT4_SB(sb);
2899         ext4_group_t bg, first_meta_bg;
2900         int has_super = 0;
2901
2902         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2903
2904         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2905                 return logical_sb_block + nr + 1;
2906         bg = sbi->s_desc_per_block * nr;
2907         if (ext4_bg_has_super(sb, bg))
2908                 has_super = 1;
2909
2910         /*
2911          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2912          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2913          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2914          * compensate.
2915          */
2916         if (sb->s_blocksize == 1024 && nr == 0 &&
2917             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2918                 has_super++;
2919
2920         return (has_super + ext4_group_first_block_no(sb, bg));
2921 }
2922
2923 /**
2924  * ext4_get_stripe_size: Get the stripe size.
2925  * @sbi: In memory super block info
2926  *
2927  * If we have specified it via mount option, then
2928  * use the mount option value. If the value specified at mount time is
2929  * greater than the blocks per group use the super block value.
2930  * If the super block value is greater than blocks per group return 0.
2931  * Allocator needs it be less than blocks per group.
2932  *
2933  */
2934 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2935 {
2936         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2937         unsigned long stripe_width =
2938                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2939         int ret;
2940
2941         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2942                 ret = sbi->s_stripe;
2943         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2944                 ret = stripe_width;
2945         else if (stride && stride <= sbi->s_blocks_per_group)
2946                 ret = stride;
2947         else
2948                 ret = 0;
2949
2950         /*
2951          * If the stripe width is 1, this makes no sense and
2952          * we set it to 0 to turn off stripe handling code.
2953          */
2954         if (ret <= 1)
2955                 ret = 0;
2956
2957         return ret;
2958 }
2959
2960 /*
2961  * Check whether this filesystem can be mounted based on
2962  * the features present and the RDONLY/RDWR mount requested.
2963  * Returns 1 if this filesystem can be mounted as requested,
2964  * 0 if it cannot be.
2965  */
2966 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2967 {
2968         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2969                 ext4_msg(sb, KERN_ERR,
2970                         "Couldn't mount because of "
2971                         "unsupported optional features (%x)",
2972                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2973                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2974                 return 0;
2975         }
2976
2977 #ifndef CONFIG_UNICODE
2978         if (ext4_has_feature_casefold(sb)) {
2979                 ext4_msg(sb, KERN_ERR,
2980                          "Filesystem with casefold feature cannot be "
2981                          "mounted without CONFIG_UNICODE");
2982                 return 0;
2983         }
2984 #endif
2985
2986         if (readonly)
2987                 return 1;
2988
2989         if (ext4_has_feature_readonly(sb)) {
2990                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2991                 sb->s_flags |= SB_RDONLY;
2992                 return 1;
2993         }
2994
2995         /* Check that feature set is OK for a read-write mount */
2996         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2997                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2998                          "unsupported optional features (%x)",
2999                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
3000                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
3001                 return 0;
3002         }
3003         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
3004                 ext4_msg(sb, KERN_ERR,
3005                          "Can't support bigalloc feature without "
3006                          "extents feature\n");
3007                 return 0;
3008         }
3009
3010 #if !IS_ENABLED(CONFIG_QUOTA) || !IS_ENABLED(CONFIG_QFMT_V2)
3011         if (!readonly && (ext4_has_feature_quota(sb) ||
3012                           ext4_has_feature_project(sb))) {
3013                 ext4_msg(sb, KERN_ERR,
3014                          "The kernel was not built with CONFIG_QUOTA and CONFIG_QFMT_V2");
3015                 return 0;
3016         }
3017 #endif  /* CONFIG_QUOTA */
3018         return 1;
3019 }
3020
3021 /*
3022  * This function is called once a day if we have errors logged
3023  * on the file system
3024  */
3025 static void print_daily_error_info(struct timer_list *t)
3026 {
3027         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
3028         struct super_block *sb = sbi->s_sb;
3029         struct ext4_super_block *es = sbi->s_es;
3030
3031         if (es->s_error_count)
3032                 /* fsck newer than v1.41.13 is needed to clean this condition. */
3033                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
3034                          le32_to_cpu(es->s_error_count));
3035         if (es->s_first_error_time) {
3036                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
3037                        sb->s_id,
3038                        ext4_get_tstamp(es, s_first_error_time),
3039                        (int) sizeof(es->s_first_error_func),
3040                        es->s_first_error_func,
3041                        le32_to_cpu(es->s_first_error_line));
3042                 if (es->s_first_error_ino)
3043                         printk(KERN_CONT ": inode %u",
3044                                le32_to_cpu(es->s_first_error_ino));
3045                 if (es->s_first_error_block)
3046                         printk(KERN_CONT ": block %llu", (unsigned long long)
3047                                le64_to_cpu(es->s_first_error_block));
3048                 printk(KERN_CONT "\n");
3049         }
3050         if (es->s_last_error_time) {
3051                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
3052                        sb->s_id,
3053                        ext4_get_tstamp(es, s_last_error_time),
3054                        (int) sizeof(es->s_last_error_func),
3055                        es->s_last_error_func,
3056                        le32_to_cpu(es->s_last_error_line));
3057                 if (es->s_last_error_ino)
3058                         printk(KERN_CONT ": inode %u",
3059                                le32_to_cpu(es->s_last_error_ino));
3060                 if (es->s_last_error_block)
3061                         printk(KERN_CONT ": block %llu", (unsigned long long)
3062                                le64_to_cpu(es->s_last_error_block));
3063                 printk(KERN_CONT "\n");
3064         }
3065         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
3066 }
3067
3068 /* Find next suitable group and run ext4_init_inode_table */
3069 static int ext4_run_li_request(struct ext4_li_request *elr)
3070 {
3071         struct ext4_group_desc *gdp = NULL;
3072         ext4_group_t group, ngroups;
3073         struct super_block *sb;
3074         int ret = 0;
3075         u64 start_time;
3076
3077         sb = elr->lr_super;
3078         ngroups = EXT4_SB(sb)->s_groups_count;
3079
3080         for (group = elr->lr_next_group; group < ngroups; group++) {
3081                 gdp = ext4_get_group_desc(sb, group, NULL);
3082                 if (!gdp) {
3083                         ret = 1;
3084                         break;
3085                 }
3086
3087                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3088                         break;
3089         }
3090
3091         if (group >= ngroups)
3092                 ret = 1;
3093
3094         if (!ret) {
3095                 start_time = ktime_get_real_ns();
3096                 ret = ext4_init_inode_table(sb, group,
3097                                             elr->lr_timeout ? 0 : 1);
3098                 if (elr->lr_timeout == 0) {
3099                         elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
3100                                   elr->lr_sbi->s_li_wait_mult);
3101                 }
3102                 elr->lr_next_sched = jiffies + elr->lr_timeout;
3103                 elr->lr_next_group = group + 1;
3104         }
3105         return ret;
3106 }
3107
3108 /*
3109  * Remove lr_request from the list_request and free the
3110  * request structure. Should be called with li_list_mtx held
3111  */
3112 static void ext4_remove_li_request(struct ext4_li_request *elr)
3113 {
3114         struct ext4_sb_info *sbi;
3115
3116         if (!elr)
3117                 return;
3118
3119         sbi = elr->lr_sbi;
3120
3121         list_del(&elr->lr_request);
3122         sbi->s_li_request = NULL;
3123         kfree(elr);
3124 }
3125
3126 static void ext4_unregister_li_request(struct super_block *sb)
3127 {
3128         mutex_lock(&ext4_li_mtx);
3129         if (!ext4_li_info) {
3130                 mutex_unlock(&ext4_li_mtx);
3131                 return;
3132         }
3133
3134         mutex_lock(&ext4_li_info->li_list_mtx);
3135         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3136         mutex_unlock(&ext4_li_info->li_list_mtx);
3137         mutex_unlock(&ext4_li_mtx);
3138 }
3139
3140 static struct task_struct *ext4_lazyinit_task;
3141
3142 /*
3143  * This is the function where ext4lazyinit thread lives. It walks
3144  * through the request list searching for next scheduled filesystem.
3145  * When such a fs is found, run the lazy initialization request
3146  * (ext4_rn_li_request) and keep track of the time spend in this
3147  * function. Based on that time we compute next schedule time of
3148  * the request. When walking through the list is complete, compute
3149  * next waking time and put itself into sleep.
3150  */
3151 static int ext4_lazyinit_thread(void *arg)
3152 {
3153         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3154         struct list_head *pos, *n;
3155         struct ext4_li_request *elr;
3156         unsigned long next_wakeup, cur;
3157
3158         BUG_ON(NULL == eli);
3159         set_freezable();
3160
3161 cont_thread:
3162         while (true) {
3163                 next_wakeup = MAX_JIFFY_OFFSET;
3164
3165                 mutex_lock(&eli->li_list_mtx);
3166                 if (list_empty(&eli->li_request_list)) {
3167                         mutex_unlock(&eli->li_list_mtx);
3168                         goto exit_thread;
3169                 }
3170                 list_for_each_safe(pos, n, &eli->li_request_list) {
3171                         int err = 0;
3172                         int progress = 0;
3173                         elr = list_entry(pos, struct ext4_li_request,
3174                                          lr_request);
3175
3176                         if (time_before(jiffies, elr->lr_next_sched)) {
3177                                 if (time_before(elr->lr_next_sched, next_wakeup))
3178                                         next_wakeup = elr->lr_next_sched;
3179                                 continue;
3180                         }
3181                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3182                                 if (sb_start_write_trylock(elr->lr_super)) {
3183                                         progress = 1;
3184                                         /*
3185                                          * We hold sb->s_umount, sb can not
3186                                          * be removed from the list, it is
3187                                          * now safe to drop li_list_mtx
3188                                          */
3189                                         mutex_unlock(&eli->li_list_mtx);
3190                                         err = ext4_run_li_request(elr);
3191                                         sb_end_write(elr->lr_super);
3192                                         mutex_lock(&eli->li_list_mtx);
3193                                         n = pos->next;
3194                                 }
3195                                 up_read((&elr->lr_super->s_umount));
3196                         }
3197                         /* error, remove the lazy_init job */
3198                         if (err) {
3199                                 ext4_remove_li_request(elr);
3200                                 continue;
3201                         }
3202                         if (!progress) {
3203                                 elr->lr_next_sched = jiffies +
3204                                         (prandom_u32()
3205                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3206                         }
3207                         if (time_before(elr->lr_next_sched, next_wakeup))
3208                                 next_wakeup = elr->lr_next_sched;
3209                 }
3210                 mutex_unlock(&eli->li_list_mtx);
3211
3212                 try_to_freeze();
3213
3214                 cur = jiffies;
3215                 if ((time_after_eq(cur, next_wakeup)) ||
3216                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3217                         cond_resched();
3218                         continue;
3219                 }
3220
3221                 schedule_timeout_interruptible(next_wakeup - cur);
3222
3223                 if (kthread_should_stop()) {
3224                         ext4_clear_request_list();
3225                         goto exit_thread;
3226                 }
3227         }
3228
3229 exit_thread:
3230         /*
3231          * It looks like the request list is empty, but we need
3232          * to check it under the li_list_mtx lock, to prevent any
3233          * additions into it, and of course we should lock ext4_li_mtx
3234          * to atomically free the list and ext4_li_info, because at
3235          * this point another ext4 filesystem could be registering
3236          * new one.
3237          */
3238         mutex_lock(&ext4_li_mtx);
3239         mutex_lock(&eli->li_list_mtx);
3240         if (!list_empty(&eli->li_request_list)) {
3241                 mutex_unlock(&eli->li_list_mtx);
3242                 mutex_unlock(&ext4_li_mtx);
3243                 goto cont_thread;
3244         }
3245         mutex_unlock(&eli->li_list_mtx);
3246         kfree(ext4_li_info);
3247         ext4_li_info = NULL;
3248         mutex_unlock(&ext4_li_mtx);
3249
3250         return 0;
3251 }
3252
3253 static void ext4_clear_request_list(void)
3254 {
3255         struct list_head *pos, *n;
3256         struct ext4_li_request *elr;
3257
3258         mutex_lock(&ext4_li_info->li_list_mtx);
3259         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3260                 elr = list_entry(pos, struct ext4_li_request,
3261                                  lr_request);
3262                 ext4_remove_li_request(elr);
3263         }
3264         mutex_unlock(&ext4_li_info->li_list_mtx);
3265 }
3266
3267 static int ext4_run_lazyinit_thread(void)
3268 {
3269         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3270                                          ext4_li_info, "ext4lazyinit");
3271         if (IS_ERR(ext4_lazyinit_task)) {
3272                 int err = PTR_ERR(ext4_lazyinit_task);
3273                 ext4_clear_request_list();
3274                 kfree(ext4_li_info);
3275                 ext4_li_info = NULL;
3276                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3277                                  "initialization thread\n",
3278                                  err);
3279                 return err;
3280         }
3281         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3282         return 0;
3283 }
3284
3285 /*
3286  * Check whether it make sense to run itable init. thread or not.
3287  * If there is at least one uninitialized inode table, return
3288  * corresponding group number, else the loop goes through all
3289  * groups and return total number of groups.
3290  */
3291 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3292 {
3293         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3294         struct ext4_group_desc *gdp = NULL;
3295
3296         if (!ext4_has_group_desc_csum(sb))
3297                 return ngroups;
3298
3299         for (group = 0; group < ngroups; group++) {
3300                 gdp = ext4_get_group_desc(sb, group, NULL);
3301                 if (!gdp)
3302                         continue;
3303
3304                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3305                         break;
3306         }
3307
3308         return group;
3309 }
3310
3311 static int ext4_li_info_new(void)
3312 {
3313         struct ext4_lazy_init *eli = NULL;
3314
3315         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3316         if (!eli)
3317                 return -ENOMEM;
3318
3319         INIT_LIST_HEAD(&eli->li_request_list);
3320         mutex_init(&eli->li_list_mtx);
3321
3322         eli->li_state |= EXT4_LAZYINIT_QUIT;
3323
3324         ext4_li_info = eli;
3325
3326         return 0;
3327 }
3328
3329 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3330                                             ext4_group_t start)
3331 {
3332         struct ext4_sb_info *sbi = EXT4_SB(sb);
3333         struct ext4_li_request *elr;
3334
3335         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3336         if (!elr)
3337                 return NULL;
3338
3339         elr->lr_super = sb;
3340         elr->lr_sbi = sbi;
3341         elr->lr_next_group = start;
3342
3343         /*
3344          * Randomize first schedule time of the request to
3345          * spread the inode table initialization requests
3346          * better.
3347          */
3348         elr->lr_next_sched = jiffies + (prandom_u32() %
3349                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3350         return elr;
3351 }
3352
3353 int ext4_register_li_request(struct super_block *sb,
3354                              ext4_group_t first_not_zeroed)
3355 {
3356         struct ext4_sb_info *sbi = EXT4_SB(sb);
3357         struct ext4_li_request *elr = NULL;
3358         ext4_group_t ngroups = sbi->s_groups_count;
3359         int ret = 0;
3360
3361         mutex_lock(&ext4_li_mtx);
3362         if (sbi->s_li_request != NULL) {
3363                 /*
3364                  * Reset timeout so it can be computed again, because
3365                  * s_li_wait_mult might have changed.
3366                  */
3367                 sbi->s_li_request->lr_timeout = 0;
3368                 goto out;
3369         }
3370
3371         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3372             !test_opt(sb, INIT_INODE_TABLE))
3373                 goto out;
3374
3375         elr = ext4_li_request_new(sb, first_not_zeroed);
3376         if (!elr) {
3377                 ret = -ENOMEM;
3378                 goto out;
3379         }
3380
3381         if (NULL == ext4_li_info) {
3382                 ret = ext4_li_info_new();
3383                 if (ret)
3384                         goto out;
3385         }
3386
3387         mutex_lock(&ext4_li_info->li_list_mtx);
3388         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3389         mutex_unlock(&ext4_li_info->li_list_mtx);
3390
3391         sbi->s_li_request = elr;
3392         /*
3393          * set elr to NULL here since it has been inserted to
3394          * the request_list and the removal and free of it is
3395          * handled by ext4_clear_request_list from now on.
3396          */
3397         elr = NULL;
3398
3399         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3400                 ret = ext4_run_lazyinit_thread();
3401                 if (ret)
3402                         goto out;
3403         }
3404 out:
3405         mutex_unlock(&ext4_li_mtx);
3406         if (ret)
3407                 kfree(elr);
3408         return ret;
3409 }
3410
3411 /*
3412  * We do not need to lock anything since this is called on
3413  * module unload.
3414  */
3415 static void ext4_destroy_lazyinit_thread(void)
3416 {
3417         /*
3418          * If thread exited earlier
3419          * there's nothing to be done.
3420          */
3421         if (!ext4_li_info || !ext4_lazyinit_task)
3422                 return;
3423
3424         kthread_stop(ext4_lazyinit_task);
3425 }
3426
3427 static int set_journal_csum_feature_set(struct super_block *sb)
3428 {
3429         int ret = 1;
3430         int compat, incompat;
3431         struct ext4_sb_info *sbi = EXT4_SB(sb);
3432
3433         if (ext4_has_metadata_csum(sb)) {
3434                 /* journal checksum v3 */
3435                 compat = 0;
3436                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3437         } else {
3438                 /* journal checksum v1 */
3439                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3440                 incompat = 0;
3441         }
3442
3443         jbd2_journal_clear_features(sbi->s_journal,
3444                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3445                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3446                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3447         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3448                 ret = jbd2_journal_set_features(sbi->s_journal,
3449                                 compat, 0,
3450                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3451                                 incompat);
3452         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3453                 ret = jbd2_journal_set_features(sbi->s_journal,
3454                                 compat, 0,
3455                                 incompat);
3456                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3457                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3458         } else {
3459                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3460                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3461         }
3462
3463         return ret;
3464 }
3465
3466 /*
3467  * Note: calculating the overhead so we can be compatible with
3468  * historical BSD practice is quite difficult in the face of
3469  * clusters/bigalloc.  This is because multiple metadata blocks from
3470  * different block group can end up in the same allocation cluster.
3471  * Calculating the exact overhead in the face of clustered allocation
3472  * requires either O(all block bitmaps) in memory or O(number of block
3473  * groups**2) in time.  We will still calculate the superblock for
3474  * older file systems --- and if we come across with a bigalloc file
3475  * system with zero in s_overhead_clusters the estimate will be close to
3476  * correct especially for very large cluster sizes --- but for newer
3477  * file systems, it's better to calculate this figure once at mkfs
3478  * time, and store it in the superblock.  If the superblock value is
3479  * present (even for non-bigalloc file systems), we will use it.
3480  */
3481 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3482                           char *buf)
3483 {
3484         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3485         struct ext4_group_desc  *gdp;
3486         ext4_fsblk_t            first_block, last_block, b;
3487         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3488         int                     s, j, count = 0;
3489         int                     has_super = ext4_bg_has_super(sb, grp);
3490
3491         if (!ext4_has_feature_bigalloc(sb))
3492                 return (has_super + ext4_bg_num_gdb(sb, grp) +
3493                         (has_super ? le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0) +
3494                         sbi->s_itb_per_group + 2);
3495
3496         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3497                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3498         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3499         for (i = 0; i < ngroups; i++) {
3500                 gdp = ext4_get_group_desc(sb, i, NULL);
3501                 b = ext4_block_bitmap(sb, gdp);
3502                 if (b >= first_block && b <= last_block) {
3503                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3504                         count++;
3505                 }
3506                 b = ext4_inode_bitmap(sb, gdp);
3507                 if (b >= first_block && b <= last_block) {
3508                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3509                         count++;
3510                 }
3511                 b = ext4_inode_table(sb, gdp);
3512                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3513                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3514                                 int c = EXT4_B2C(sbi, b - first_block);
3515                                 ext4_set_bit(c, buf);
3516                                 count++;
3517                         }
3518                 if (i != grp)
3519                         continue;
3520                 s = 0;
3521                 if (ext4_bg_has_super(sb, grp)) {
3522                         ext4_set_bit(s++, buf);
3523                         count++;
3524                 }
3525                 j = ext4_bg_num_gdb(sb, grp);
3526                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3527                         ext4_error(sb, "Invalid number of block group "
3528                                    "descriptor blocks: %d", j);
3529                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3530                 }
3531                 count += j;
3532                 for (; j > 0; j--)
3533                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3534         }
3535         if (!count)
3536                 return 0;
3537         return EXT4_CLUSTERS_PER_GROUP(sb) -
3538                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3539 }
3540
3541 /*
3542  * Compute the overhead and stash it in sbi->s_overhead
3543  */
3544 int ext4_calculate_overhead(struct super_block *sb)
3545 {
3546         struct ext4_sb_info *sbi = EXT4_SB(sb);
3547         struct ext4_super_block *es = sbi->s_es;
3548         struct inode *j_inode;
3549         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3550         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3551         ext4_fsblk_t overhead = 0;
3552         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3553
3554         if (!buf)
3555                 return -ENOMEM;
3556
3557         /*
3558          * Compute the overhead (FS structures).  This is constant
3559          * for a given filesystem unless the number of block groups
3560          * changes so we cache the previous value until it does.
3561          */
3562
3563         /*
3564          * All of the blocks before first_data_block are overhead
3565          */
3566         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3567
3568         /*
3569          * Add the overhead found in each block group
3570          */
3571         for (i = 0; i < ngroups; i++) {
3572                 int blks;
3573
3574                 blks = count_overhead(sb, i, buf);
3575                 overhead += blks;
3576                 if (blks)
3577                         memset(buf, 0, PAGE_SIZE);
3578                 cond_resched();
3579         }
3580
3581         /*
3582          * Add the internal journal blocks whether the journal has been
3583          * loaded or not
3584          */
3585         if (sbi->s_journal && !sbi->s_journal_bdev)
3586                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3587         else if (ext4_has_feature_journal(sb) && !sbi->s_journal && j_inum) {
3588                 /* j_inum for internal journal is non-zero */
3589                 j_inode = ext4_get_journal_inode(sb, j_inum);
3590                 if (j_inode) {
3591                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3592                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3593                         iput(j_inode);
3594                 } else {
3595                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3596                 }
3597         }
3598         sbi->s_overhead = overhead;
3599         smp_wmb();
3600         free_page((unsigned long) buf);
3601         return 0;
3602 }
3603
3604 static void ext4_set_resv_clusters(struct super_block *sb)
3605 {
3606         ext4_fsblk_t resv_clusters;
3607         struct ext4_sb_info *sbi = EXT4_SB(sb);
3608
3609         /*
3610          * There's no need to reserve anything when we aren't using extents.
3611          * The space estimates are exact, there are no unwritten extents,
3612          * hole punching doesn't need new metadata... This is needed especially
3613          * to keep ext2/3 backward compatibility.
3614          */
3615         if (!ext4_has_feature_extents(sb))
3616                 return;
3617         /*
3618          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3619          * This should cover the situations where we can not afford to run
3620          * out of space like for example punch hole, or converting
3621          * unwritten extents in delalloc path. In most cases such
3622          * allocation would require 1, or 2 blocks, higher numbers are
3623          * very rare.
3624          */
3625         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3626                          sbi->s_cluster_bits);
3627
3628         do_div(resv_clusters, 50);
3629         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3630
3631         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3632 }
3633
3634 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3635 {
3636         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3637         char *orig_data = kstrdup(data, GFP_KERNEL);
3638         struct buffer_head *bh, **group_desc;
3639         struct ext4_super_block *es = NULL;
3640         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3641         struct flex_groups **flex_groups;
3642         ext4_fsblk_t block;
3643         ext4_fsblk_t sb_block = get_sb_block(&data);
3644         ext4_fsblk_t logical_sb_block;
3645         unsigned long offset = 0;
3646         unsigned long journal_devnum = 0;
3647         unsigned long def_mount_opts;
3648         struct inode *root;
3649         const char *descr;
3650         int ret = -ENOMEM;
3651         int blocksize, clustersize;
3652         unsigned int db_count;
3653         unsigned int i;
3654         int needs_recovery, has_huge_files, has_bigalloc;
3655         __u64 blocks_count;
3656         int err = 0;
3657         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3658         ext4_group_t first_not_zeroed;
3659
3660         if ((data && !orig_data) || !sbi)
3661                 goto out_free_base;
3662
3663         sbi->s_daxdev = dax_dev;
3664         sbi->s_blockgroup_lock =
3665                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3666         if (!sbi->s_blockgroup_lock)
3667                 goto out_free_base;
3668
3669         sb->s_fs_info = sbi;
3670         sbi->s_sb = sb;
3671         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3672         sbi->s_sb_block = sb_block;
3673         if (sb->s_bdev->bd_part)
3674                 sbi->s_sectors_written_start =
3675                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3676
3677         /* Cleanup superblock name */
3678         strreplace(sb->s_id, '/', '!');
3679
3680         /* -EINVAL is default */
3681         ret = -EINVAL;
3682         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3683         if (!blocksize) {
3684                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3685                 goto out_fail;
3686         }
3687
3688         /*
3689          * The ext4 superblock will not be buffer aligned for other than 1kB
3690          * block sizes.  We need to calculate the offset from buffer start.
3691          */
3692         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3693                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3694                 offset = do_div(logical_sb_block, blocksize);
3695         } else {
3696                 logical_sb_block = sb_block;
3697         }
3698
3699         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3700                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3701                 goto out_fail;
3702         }
3703         /*
3704          * Note: s_es must be initialized as soon as possible because
3705          *       some ext4 macro-instructions depend on its value
3706          */
3707         es = (struct ext4_super_block *) (bh->b_data + offset);
3708         sbi->s_es = es;
3709         sb->s_magic = le16_to_cpu(es->s_magic);
3710         if (sb->s_magic != EXT4_SUPER_MAGIC)
3711                 goto cantfind_ext4;
3712         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3713
3714         /* Warn if metadata_csum and gdt_csum are both set. */
3715         if (ext4_has_feature_metadata_csum(sb) &&
3716             ext4_has_feature_gdt_csum(sb))
3717                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3718                              "redundant flags; please run fsck.");
3719
3720         /* Check for a known checksum algorithm */
3721         if (!ext4_verify_csum_type(sb, es)) {
3722                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3723                          "unknown checksum algorithm.");
3724                 silent = 1;
3725                 goto cantfind_ext4;
3726         }
3727
3728         /* Load the checksum driver */
3729         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3730         if (IS_ERR(sbi->s_chksum_driver)) {
3731                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3732                 ret = PTR_ERR(sbi->s_chksum_driver);
3733                 sbi->s_chksum_driver = NULL;
3734                 goto failed_mount;
3735         }
3736
3737         /* Check superblock checksum */
3738         if (!ext4_superblock_csum_verify(sb, es)) {
3739                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3740                          "invalid superblock checksum.  Run e2fsck?");
3741                 silent = 1;
3742                 ret = -EFSBADCRC;
3743                 goto cantfind_ext4;
3744         }
3745
3746         /* Precompute checksum seed for all metadata */
3747         if (ext4_has_feature_csum_seed(sb))
3748                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3749         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3750                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3751                                                sizeof(es->s_uuid));
3752
3753         /* Set defaults before we parse the mount options */
3754         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3755         set_opt(sb, INIT_INODE_TABLE);
3756         if (def_mount_opts & EXT4_DEFM_DEBUG)
3757                 set_opt(sb, DEBUG);
3758         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3759                 set_opt(sb, GRPID);
3760         if (def_mount_opts & EXT4_DEFM_UID16)
3761                 set_opt(sb, NO_UID32);
3762         /* xattr user namespace & acls are now defaulted on */
3763         set_opt(sb, XATTR_USER);
3764 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3765         set_opt(sb, POSIX_ACL);
3766 #endif
3767         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3768         if (ext4_has_metadata_csum(sb))
3769                 set_opt(sb, JOURNAL_CHECKSUM);
3770
3771         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3772                 set_opt(sb, JOURNAL_DATA);
3773         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3774                 set_opt(sb, ORDERED_DATA);
3775         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3776                 set_opt(sb, WRITEBACK_DATA);
3777
3778         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3779                 set_opt(sb, ERRORS_PANIC);
3780         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3781                 set_opt(sb, ERRORS_CONT);
3782         else
3783                 set_opt(sb, ERRORS_RO);
3784         /* block_validity enabled by default; disable with noblock_validity */
3785         set_opt(sb, BLOCK_VALIDITY);
3786         if (def_mount_opts & EXT4_DEFM_DISCARD)
3787                 set_opt(sb, DISCARD);
3788
3789         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3790         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3791         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3792         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3793         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3794
3795         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3796                 set_opt(sb, BARRIER);
3797
3798         /*
3799          * enable delayed allocation by default
3800          * Use -o nodelalloc to turn it off
3801          */
3802         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3803             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3804                 set_opt(sb, DELALLOC);
3805
3806         /*
3807          * set default s_li_wait_mult for lazyinit, for the case there is
3808          * no mount option specified.
3809          */
3810         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3811
3812         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3813         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3814             blocksize > EXT4_MAX_BLOCK_SIZE) {
3815                 ext4_msg(sb, KERN_ERR,
3816                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3817                          blocksize, le32_to_cpu(es->s_log_block_size));
3818                 goto failed_mount;
3819         }
3820
3821         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3822                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3823                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3824         } else {
3825                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3826                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3827                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3828                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3829                                  sbi->s_first_ino);
3830                         goto failed_mount;
3831                 }
3832                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3833                     (!is_power_of_2(sbi->s_inode_size)) ||
3834                     (sbi->s_inode_size > blocksize)) {
3835                         ext4_msg(sb, KERN_ERR,
3836                                "unsupported inode size: %d",
3837                                sbi->s_inode_size);
3838                         ext4_msg(sb, KERN_ERR, "blocksize: %d", blocksize);
3839                         goto failed_mount;
3840                 }
3841                 /*
3842                  * i_atime_extra is the last extra field available for
3843                  * [acm]times in struct ext4_inode. Checking for that
3844                  * field should suffice to ensure we have extra space
3845                  * for all three.
3846                  */
3847                 if (sbi->s_inode_size >= offsetof(struct ext4_inode, i_atime_extra) +
3848                         sizeof(((struct ext4_inode *)0)->i_atime_extra)) {
3849                         sb->s_time_gran = 1;
3850                         sb->s_time_max = EXT4_EXTRA_TIMESTAMP_MAX;
3851                 } else {
3852                         sb->s_time_gran = NSEC_PER_SEC;
3853                         sb->s_time_max = EXT4_NON_EXTRA_TIMESTAMP_MAX;
3854                 }
3855                 sb->s_time_min = EXT4_TIMESTAMP_MIN;
3856         }
3857         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3858                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3859                         EXT4_GOOD_OLD_INODE_SIZE;
3860                 if (ext4_has_feature_extra_isize(sb)) {
3861                         unsigned v, max = (sbi->s_inode_size -
3862                                            EXT4_GOOD_OLD_INODE_SIZE);
3863
3864                         v = le16_to_cpu(es->s_want_extra_isize);
3865                         if (v > max) {
3866                                 ext4_msg(sb, KERN_ERR,
3867                                          "bad s_want_extra_isize: %d", v);
3868                                 goto failed_mount;
3869                         }
3870                         if (sbi->s_want_extra_isize < v)
3871                                 sbi->s_want_extra_isize = v;
3872
3873                         v = le16_to_cpu(es->s_min_extra_isize);
3874                         if (v > max) {
3875                                 ext4_msg(sb, KERN_ERR,
3876                                          "bad s_min_extra_isize: %d", v);
3877                                 goto failed_mount;
3878                         }
3879                         if (sbi->s_want_extra_isize < v)
3880                                 sbi->s_want_extra_isize = v;
3881                 }
3882         }
3883
3884         if (sbi->s_es->s_mount_opts[0]) {
3885                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3886                                               sizeof(sbi->s_es->s_mount_opts),
3887                                               GFP_KERNEL);
3888                 if (!s_mount_opts)
3889                         goto failed_mount;
3890                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3891                                    &journal_ioprio, 0)) {
3892                         ext4_msg(sb, KERN_WARNING,
3893                                  "failed to parse options in superblock: %s",
3894                                  s_mount_opts);
3895                 }
3896                 kfree(s_mount_opts);
3897         }
3898         sbi->s_def_mount_opt = sbi->s_mount_opt;
3899         if (!parse_options((char *) data, sb, &journal_devnum,
3900                            &journal_ioprio, 0))
3901                 goto failed_mount;
3902
3903 #ifdef CONFIG_UNICODE
3904         if (ext4_has_feature_casefold(sb) && !sbi->s_encoding) {
3905                 const struct ext4_sb_encodings *encoding_info;
3906                 struct unicode_map *encoding;
3907                 __u16 encoding_flags;
3908
3909                 if (ext4_has_feature_encrypt(sb)) {
3910                         ext4_msg(sb, KERN_ERR,
3911                                  "Can't mount with encoding and encryption");
3912                         goto failed_mount;
3913                 }
3914
3915                 if (ext4_sb_read_encoding(es, &encoding_info,
3916                                           &encoding_flags)) {
3917                         ext4_msg(sb, KERN_ERR,
3918                                  "Encoding requested by superblock is unknown");
3919                         goto failed_mount;
3920                 }
3921
3922                 encoding = utf8_load(encoding_info->version);
3923                 if (IS_ERR(encoding)) {
3924                         ext4_msg(sb, KERN_ERR,
3925                                  "can't mount with superblock charset: %s-%s "
3926                                  "not supported by the kernel. flags: 0x%x.",
3927                                  encoding_info->name, encoding_info->version,
3928                                  encoding_flags);
3929                         goto failed_mount;
3930                 }
3931                 ext4_msg(sb, KERN_INFO,"Using encoding defined by superblock: "
3932                          "%s-%s with flags 0x%hx", encoding_info->name,
3933                          encoding_info->version?:"\b", encoding_flags);
3934
3935                 sbi->s_encoding = encoding;
3936                 sbi->s_encoding_flags = encoding_flags;
3937         }
3938 #endif
3939
3940         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3941                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3942                             "with data=journal disables delayed "
3943                             "allocation and O_DIRECT support!\n");
3944                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3945                         ext4_msg(sb, KERN_ERR, "can't mount with "
3946                                  "both data=journal and delalloc");
3947                         goto failed_mount;
3948                 }
3949                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3950                         ext4_msg(sb, KERN_ERR, "can't mount with "
3951                                  "both data=journal and dioread_nolock");
3952                         goto failed_mount;
3953                 }
3954                 if (test_opt(sb, DAX)) {
3955                         ext4_msg(sb, KERN_ERR, "can't mount with "
3956                                  "both data=journal and dax");
3957                         goto failed_mount;
3958                 }
3959                 if (ext4_has_feature_encrypt(sb)) {
3960                         ext4_msg(sb, KERN_WARNING,
3961                                  "encrypted files will use data=ordered "
3962                                  "instead of data journaling mode");
3963                 }
3964                 if (test_opt(sb, DELALLOC))
3965                         clear_opt(sb, DELALLOC);
3966         } else {
3967                 sb->s_iflags |= SB_I_CGROUPWB;
3968         }
3969
3970         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3971                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3972
3973         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3974             (ext4_has_compat_features(sb) ||
3975              ext4_has_ro_compat_features(sb) ||
3976              ext4_has_incompat_features(sb)))
3977                 ext4_msg(sb, KERN_WARNING,
3978                        "feature flags set on rev 0 fs, "
3979                        "running e2fsck is recommended");
3980
3981         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3982                 set_opt2(sb, HURD_COMPAT);
3983                 if (ext4_has_feature_64bit(sb)) {
3984                         ext4_msg(sb, KERN_ERR,
3985                                  "The Hurd can't support 64-bit file systems");
3986                         goto failed_mount;
3987                 }
3988
3989                 /*
3990                  * ea_inode feature uses l_i_version field which is not
3991                  * available in HURD_COMPAT mode.
3992                  */
3993                 if (ext4_has_feature_ea_inode(sb)) {
3994                         ext4_msg(sb, KERN_ERR,
3995                                  "ea_inode feature is not supported for Hurd");
3996                         goto failed_mount;
3997                 }
3998         }
3999
4000         if (IS_EXT2_SB(sb)) {
4001                 if (ext2_feature_set_ok(sb))
4002                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
4003                                  "using the ext4 subsystem");
4004                 else {
4005                         /*
4006                          * If we're probing be silent, if this looks like
4007                          * it's actually an ext[34] filesystem.
4008                          */
4009                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4010                                 goto failed_mount;
4011                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
4012                                  "to feature incompatibilities");
4013                         goto failed_mount;
4014                 }
4015         }
4016
4017         if (IS_EXT3_SB(sb)) {
4018                 if (ext3_feature_set_ok(sb))
4019                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
4020                                  "using the ext4 subsystem");
4021                 else {
4022                         /*
4023                          * If we're probing be silent, if this looks like
4024                          * it's actually an ext4 filesystem.
4025                          */
4026                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
4027                                 goto failed_mount;
4028                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
4029                                  "to feature incompatibilities");
4030                         goto failed_mount;
4031                 }
4032         }
4033
4034         /*
4035          * Check feature flags regardless of the revision level, since we
4036          * previously didn't change the revision level when setting the flags,
4037          * so there is a chance incompat flags are set on a rev 0 filesystem.
4038          */
4039         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
4040                 goto failed_mount;
4041
4042         if (le32_to_cpu(es->s_log_block_size) >
4043             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4044                 ext4_msg(sb, KERN_ERR,
4045                          "Invalid log block size: %u",
4046                          le32_to_cpu(es->s_log_block_size));
4047                 goto failed_mount;
4048         }
4049         if (le32_to_cpu(es->s_log_cluster_size) >
4050             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
4051                 ext4_msg(sb, KERN_ERR,
4052                          "Invalid log cluster size: %u",
4053                          le32_to_cpu(es->s_log_cluster_size));
4054                 goto failed_mount;
4055         }
4056
4057         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
4058                 ext4_msg(sb, KERN_ERR,
4059                          "Number of reserved GDT blocks insanely large: %d",
4060                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
4061                 goto failed_mount;
4062         }
4063
4064         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
4065                 if (ext4_has_feature_inline_data(sb)) {
4066                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
4067                                         " that may contain inline data");
4068                         goto failed_mount;
4069                 }
4070                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
4071                         ext4_msg(sb, KERN_ERR,
4072                                 "DAX unsupported by block device.");
4073                         goto failed_mount;
4074                 }
4075         }
4076
4077         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
4078                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
4079                          es->s_encryption_level);
4080                 goto failed_mount;
4081         }
4082
4083         if (sb->s_blocksize != blocksize) {
4084                 /* Validate the filesystem blocksize */
4085                 if (!sb_set_blocksize(sb, blocksize)) {
4086                         ext4_msg(sb, KERN_ERR, "bad block size %d",
4087                                         blocksize);
4088                         goto failed_mount;
4089                 }
4090
4091                 brelse(bh);
4092                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
4093                 offset = do_div(logical_sb_block, blocksize);
4094                 bh = sb_bread_unmovable(sb, logical_sb_block);
4095                 if (!bh) {
4096                         ext4_msg(sb, KERN_ERR,
4097                                "Can't read superblock on 2nd try");
4098                         goto failed_mount;
4099                 }
4100                 es = (struct ext4_super_block *)(bh->b_data + offset);
4101                 sbi->s_es = es;
4102                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
4103                         ext4_msg(sb, KERN_ERR,
4104                                "Magic mismatch, very weird!");
4105                         goto failed_mount;
4106                 }
4107         }
4108
4109         has_huge_files = ext4_has_feature_huge_file(sb);
4110         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
4111                                                       has_huge_files);
4112         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
4113
4114         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
4115         if (ext4_has_feature_64bit(sb)) {
4116                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
4117                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
4118                     !is_power_of_2(sbi->s_desc_size)) {
4119                         ext4_msg(sb, KERN_ERR,
4120                                "unsupported descriptor size %lu",
4121                                sbi->s_desc_size);
4122                         goto failed_mount;
4123                 }
4124         } else
4125                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
4126
4127         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
4128         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
4129
4130         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
4131         if (sbi->s_inodes_per_block == 0)
4132                 goto cantfind_ext4;
4133         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
4134             sbi->s_inodes_per_group > blocksize * 8) {
4135                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
4136                          sbi->s_inodes_per_group);
4137                 goto failed_mount;
4138         }
4139         sbi->s_itb_per_group = sbi->s_inodes_per_group /
4140                                         sbi->s_inodes_per_block;
4141         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
4142         sbi->s_sbh = bh;
4143         sbi->s_mount_state = le16_to_cpu(es->s_state);
4144         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
4145         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
4146
4147         for (i = 0; i < 4; i++)
4148                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
4149         sbi->s_def_hash_version = es->s_def_hash_version;
4150         if (ext4_has_feature_dir_index(sb)) {
4151                 i = le32_to_cpu(es->s_flags);
4152                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
4153                         sbi->s_hash_unsigned = 3;
4154                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
4155 #ifdef __CHAR_UNSIGNED__
4156                         if (!sb_rdonly(sb))
4157                                 es->s_flags |=
4158                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
4159                         sbi->s_hash_unsigned = 3;
4160 #else
4161                         if (!sb_rdonly(sb))
4162                                 es->s_flags |=
4163                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
4164 #endif
4165                 }
4166         }
4167
4168         /* Handle clustersize */
4169         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
4170         has_bigalloc = ext4_has_feature_bigalloc(sb);
4171         if (has_bigalloc) {
4172                 if (clustersize < blocksize) {
4173                         ext4_msg(sb, KERN_ERR,
4174                                  "cluster size (%d) smaller than "
4175                                  "block size (%d)", clustersize, blocksize);
4176                         goto failed_mount;
4177                 }
4178                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
4179                         le32_to_cpu(es->s_log_block_size);
4180                 sbi->s_clusters_per_group =
4181                         le32_to_cpu(es->s_clusters_per_group);
4182                 if (sbi->s_clusters_per_group > blocksize * 8) {
4183                         ext4_msg(sb, KERN_ERR,
4184                                  "#clusters per group too big: %lu",
4185                                  sbi->s_clusters_per_group);
4186                         goto failed_mount;
4187                 }
4188                 if (sbi->s_blocks_per_group !=
4189                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
4190                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
4191                                  "clusters per group (%lu) inconsistent",
4192                                  sbi->s_blocks_per_group,
4193                                  sbi->s_clusters_per_group);
4194                         goto failed_mount;
4195                 }
4196         } else {
4197                 if (clustersize != blocksize) {
4198                         ext4_msg(sb, KERN_ERR,
4199                                  "fragment/cluster size (%d) != "
4200                                  "block size (%d)", clustersize, blocksize);
4201                         goto failed_mount;
4202                 }
4203                 if (sbi->s_blocks_per_group > blocksize * 8) {
4204                         ext4_msg(sb, KERN_ERR,
4205                                  "#blocks per group too big: %lu",
4206                                  sbi->s_blocks_per_group);
4207                         goto failed_mount;
4208                 }
4209                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4210                 sbi->s_cluster_bits = 0;
4211         }
4212         sbi->s_cluster_ratio = clustersize / blocksize;
4213
4214         /* Do we have standard group size of clustersize * 8 blocks ? */
4215         if (sbi->s_blocks_per_group == clustersize << 3)
4216                 set_opt2(sb, STD_GROUP_SIZE);
4217
4218         /*
4219          * Test whether we have more sectors than will fit in sector_t,
4220          * and whether the max offset is addressable by the page cache.
4221          */
4222         err = generic_check_addressable(sb->s_blocksize_bits,
4223                                         ext4_blocks_count(es));
4224         if (err) {
4225                 ext4_msg(sb, KERN_ERR, "filesystem"
4226                          " too large to mount safely on this system");
4227                 goto failed_mount;
4228         }
4229
4230         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4231                 goto cantfind_ext4;
4232
4233         /* check blocks count against device size */
4234         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4235         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4236                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4237                        "exceeds size of device (%llu blocks)",
4238                        ext4_blocks_count(es), blocks_count);
4239                 goto failed_mount;
4240         }
4241
4242         /*
4243          * It makes no sense for the first data block to be beyond the end
4244          * of the filesystem.
4245          */
4246         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4247                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4248                          "block %u is beyond end of filesystem (%llu)",
4249                          le32_to_cpu(es->s_first_data_block),
4250                          ext4_blocks_count(es));
4251                 goto failed_mount;
4252         }
4253         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4254             (sbi->s_cluster_ratio == 1)) {
4255                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4256                          "block is 0 with a 1k block and cluster size");
4257                 goto failed_mount;
4258         }
4259
4260         blocks_count = (ext4_blocks_count(es) -
4261                         le32_to_cpu(es->s_first_data_block) +
4262                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4263         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4264         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4265                 ext4_msg(sb, KERN_WARNING, "groups count too large: %llu "
4266                        "(block count %llu, first data block %u, "
4267                        "blocks per group %lu)", blocks_count,
4268                        ext4_blocks_count(es),
4269                        le32_to_cpu(es->s_first_data_block),
4270                        EXT4_BLOCKS_PER_GROUP(sb));
4271                 goto failed_mount;
4272         }
4273         sbi->s_groups_count = blocks_count;
4274         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4275                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4276         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4277             le32_to_cpu(es->s_inodes_count)) {
4278                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4279                          le32_to_cpu(es->s_inodes_count),
4280                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4281                 ret = -EINVAL;
4282                 goto failed_mount;
4283         }
4284         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4285                    EXT4_DESC_PER_BLOCK(sb);
4286         if (ext4_has_feature_meta_bg(sb)) {
4287                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4288                         ext4_msg(sb, KERN_WARNING,
4289                                  "first meta block group too large: %u "
4290                                  "(group descriptor block count %u)",
4291                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4292                         goto failed_mount;
4293                 }
4294         }
4295         rcu_assign_pointer(sbi->s_group_desc,
4296                            kvmalloc_array(db_count,
4297                                           sizeof(struct buffer_head *),
4298                                           GFP_KERNEL));
4299         if (sbi->s_group_desc == NULL) {
4300                 ext4_msg(sb, KERN_ERR, "not enough memory");
4301                 ret = -ENOMEM;
4302                 goto failed_mount;
4303         }
4304
4305         bgl_lock_init(sbi->s_blockgroup_lock);
4306
4307         /* Pre-read the descriptors into the buffer cache */
4308         for (i = 0; i < db_count; i++) {
4309                 block = descriptor_loc(sb, logical_sb_block, i);
4310                 sb_breadahead_unmovable(sb, block);
4311         }
4312
4313         for (i = 0; i < db_count; i++) {
4314                 struct buffer_head *bh;
4315
4316                 block = descriptor_loc(sb, logical_sb_block, i);
4317                 bh = sb_bread_unmovable(sb, block);
4318                 if (!bh) {
4319                         ext4_msg(sb, KERN_ERR,
4320                                "can't read group descriptor %d", i);
4321                         db_count = i;
4322                         goto failed_mount2;
4323                 }
4324                 rcu_read_lock();
4325                 rcu_dereference(sbi->s_group_desc)[i] = bh;
4326                 rcu_read_unlock();
4327         }
4328         sbi->s_gdb_count = db_count;
4329         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4330                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4331                 ret = -EFSCORRUPTED;
4332                 goto failed_mount2;
4333         }
4334
4335         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4336
4337         /* Register extent status tree shrinker */
4338         if (ext4_es_register_shrinker(sbi))
4339                 goto failed_mount3;
4340
4341         sbi->s_stripe = ext4_get_stripe_size(sbi);
4342         sbi->s_extent_max_zeroout_kb = 32;
4343
4344         /*
4345          * set up enough so that it can read an inode
4346          */
4347         sb->s_op = &ext4_sops;
4348         sb->s_export_op = &ext4_export_ops;
4349         sb->s_xattr = ext4_xattr_handlers;
4350 #ifdef CONFIG_FS_ENCRYPTION
4351         sb->s_cop = &ext4_cryptops;
4352 #endif
4353 #ifdef CONFIG_FS_VERITY
4354         sb->s_vop = &ext4_verityops;
4355 #endif
4356 #ifdef CONFIG_QUOTA
4357         sb->dq_op = &ext4_quota_operations;
4358         if (ext4_has_feature_quota(sb))
4359                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4360         else
4361                 sb->s_qcop = &ext4_qctl_operations;
4362         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4363 #endif
4364         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4365
4366         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4367         mutex_init(&sbi->s_orphan_lock);
4368
4369         sb->s_root = NULL;
4370
4371         needs_recovery = (es->s_last_orphan != 0 ||
4372                           ext4_has_feature_journal_needs_recovery(sb));
4373
4374         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4375                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4376                         goto failed_mount3a;
4377
4378         /*
4379          * The first inode we look at is the journal inode.  Don't try
4380          * root first: it may be modified in the journal!
4381          */
4382         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4383                 err = ext4_load_journal(sb, es, journal_devnum);
4384                 if (err)
4385                         goto failed_mount3a;
4386         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4387                    ext4_has_feature_journal_needs_recovery(sb)) {
4388                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4389                        "suppressed and not mounted read-only");
4390                 goto failed_mount3a;
4391         } else {
4392                 /* Nojournal mode, all journal mount options are illegal */
4393                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4394                         ext4_msg(sb, KERN_ERR, "can't mount with "
4395                                  "journal_async_commit, fs mounted w/o journal");
4396                         goto failed_mount3a;
4397                 }
4398
4399                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4400                         ext4_msg(sb, KERN_ERR, "can't mount with "
4401                                  "journal_checksum, fs mounted w/o journal");
4402                         goto failed_mount3a;
4403                 }
4404                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4405                         ext4_msg(sb, KERN_ERR, "can't mount with "
4406                                  "commit=%lu, fs mounted w/o journal",
4407                                  sbi->s_commit_interval / HZ);
4408                         goto failed_mount3a;
4409                 }
4410                 if (EXT4_MOUNT_DATA_FLAGS &
4411                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4412                         ext4_msg(sb, KERN_ERR, "can't mount with "
4413                                  "data=, fs mounted w/o journal");
4414                         goto failed_mount3a;
4415                 }
4416                 sbi->s_def_mount_opt &= ~EXT4_MOUNT_JOURNAL_CHECKSUM;
4417                 clear_opt(sb, JOURNAL_CHECKSUM);
4418                 clear_opt(sb, DATA_FLAGS);
4419                 sbi->s_journal = NULL;
4420                 needs_recovery = 0;
4421                 goto no_journal;
4422         }
4423
4424         if (ext4_has_feature_64bit(sb) &&
4425             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4426                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4427                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4428                 goto failed_mount_wq;
4429         }
4430
4431         if (!set_journal_csum_feature_set(sb)) {
4432                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4433                          "feature set");
4434                 goto failed_mount_wq;
4435         }
4436
4437         /* We have now updated the journal if required, so we can
4438          * validate the data journaling mode. */
4439         switch (test_opt(sb, DATA_FLAGS)) {
4440         case 0:
4441                 /* No mode set, assume a default based on the journal
4442                  * capabilities: ORDERED_DATA if the journal can
4443                  * cope, else JOURNAL_DATA
4444                  */
4445                 if (jbd2_journal_check_available_features
4446                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4447                         set_opt(sb, ORDERED_DATA);
4448                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4449                 } else {
4450                         set_opt(sb, JOURNAL_DATA);
4451                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4452                 }
4453                 break;
4454
4455         case EXT4_MOUNT_ORDERED_DATA:
4456         case EXT4_MOUNT_WRITEBACK_DATA:
4457                 if (!jbd2_journal_check_available_features
4458                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4459                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4460                                "requested data journaling mode");
4461                         goto failed_mount_wq;
4462                 }
4463         default:
4464                 break;
4465         }
4466
4467         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4468             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4469                 ext4_msg(sb, KERN_ERR, "can't mount with "
4470                         "journal_async_commit in data=ordered mode");
4471                 goto failed_mount_wq;
4472         }
4473
4474         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4475
4476         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4477
4478 no_journal:
4479         if (!test_opt(sb, NO_MBCACHE)) {
4480                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4481                 if (!sbi->s_ea_block_cache) {
4482                         ext4_msg(sb, KERN_ERR,
4483                                  "Failed to create ea_block_cache");
4484                         goto failed_mount_wq;
4485                 }
4486
4487                 if (ext4_has_feature_ea_inode(sb)) {
4488                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4489                         if (!sbi->s_ea_inode_cache) {
4490                                 ext4_msg(sb, KERN_ERR,
4491                                          "Failed to create ea_inode_cache");
4492                                 goto failed_mount_wq;
4493                         }
4494                 }
4495         }
4496
4497         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4498             (blocksize != PAGE_SIZE)) {
4499                 ext4_msg(sb, KERN_ERR,
4500                          "Unsupported blocksize for fs encryption");
4501                 goto failed_mount_wq;
4502         }
4503
4504         if (ext4_has_feature_verity(sb) && blocksize != PAGE_SIZE) {
4505                 ext4_msg(sb, KERN_ERR, "Unsupported blocksize for fs-verity");
4506                 goto failed_mount_wq;
4507         }
4508
4509         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4510             !ext4_has_feature_encrypt(sb)) {
4511                 ext4_set_feature_encrypt(sb);
4512                 ext4_commit_super(sb, 1);
4513         }
4514
4515         /*
4516          * Get the # of file system overhead blocks from the
4517          * superblock if present.
4518          */
4519         sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4520         /* ignore the precalculated value if it is ridiculous */
4521         if (sbi->s_overhead > ext4_blocks_count(es))
4522                 sbi->s_overhead = 0;
4523         /*
4524          * If the bigalloc feature is not enabled recalculating the
4525          * overhead doesn't take long, so we might as well just redo
4526          * it to make sure we are using the correct value.
4527          */
4528         if (!ext4_has_feature_bigalloc(sb))
4529                 sbi->s_overhead = 0;
4530         if (sbi->s_overhead == 0) {
4531                 err = ext4_calculate_overhead(sb);
4532                 if (err)
4533                         goto failed_mount_wq;
4534         }
4535
4536         /*
4537          * The maximum number of concurrent works can be high and
4538          * concurrency isn't really necessary.  Limit it to 1.
4539          */
4540         EXT4_SB(sb)->rsv_conversion_wq =
4541                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4542         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4543                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4544                 ret = -ENOMEM;
4545                 goto failed_mount4;
4546         }
4547
4548         /*
4549          * The jbd2_journal_load will have done any necessary log recovery,
4550          * so we can safely mount the rest of the filesystem now.
4551          */
4552
4553         root = ext4_iget(sb, EXT4_ROOT_INO, EXT4_IGET_SPECIAL);
4554         if (IS_ERR(root)) {
4555                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4556                 ret = PTR_ERR(root);
4557                 root = NULL;
4558                 goto failed_mount4;
4559         }
4560         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4561                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4562                 iput(root);
4563                 goto failed_mount4;
4564         }
4565
4566 #ifdef CONFIG_UNICODE
4567         if (sbi->s_encoding)
4568                 sb->s_d_op = &ext4_dentry_ops;
4569 #endif
4570
4571         sb->s_root = d_make_root(root);
4572         if (!sb->s_root) {
4573                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4574                 ret = -ENOMEM;
4575                 goto failed_mount4;
4576         }
4577
4578         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4579         if (ret == -EROFS) {
4580                 sb->s_flags |= SB_RDONLY;
4581                 ret = 0;
4582         } else if (ret)
4583                 goto failed_mount4a;
4584
4585         ext4_set_resv_clusters(sb);
4586
4587         if (test_opt(sb, BLOCK_VALIDITY)) {
4588                 err = ext4_setup_system_zone(sb);
4589                 if (err) {
4590                         ext4_msg(sb, KERN_ERR, "failed to initialize system "
4591                                  "zone (%d)", err);
4592                         goto failed_mount4a;
4593                 }
4594         }
4595
4596         ext4_ext_init(sb);
4597         err = ext4_mb_init(sb);
4598         if (err) {
4599                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4600                          err);
4601                 goto failed_mount5;
4602         }
4603
4604         block = ext4_count_free_clusters(sb);
4605         ext4_free_blocks_count_set(sbi->s_es,
4606                                    EXT4_C2B(sbi, block));
4607         ext4_superblock_csum_set(sb);
4608         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4609                                   GFP_KERNEL);
4610         if (!err) {
4611                 unsigned long freei = ext4_count_free_inodes(sb);
4612                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4613                 ext4_superblock_csum_set(sb);
4614                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4615                                           GFP_KERNEL);
4616         }
4617         if (!err)
4618                 err = percpu_counter_init(&sbi->s_dirs_counter,
4619                                           ext4_count_dirs(sb), GFP_KERNEL);
4620         if (!err)
4621                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4622                                           GFP_KERNEL);
4623         if (!err)
4624                 err = percpu_counter_init(&sbi->s_sra_exceeded_retry_limit, 0,
4625                                           GFP_KERNEL);
4626         if (!err)
4627                 err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
4628
4629         if (err) {
4630                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4631                 goto failed_mount6;
4632         }
4633
4634         if (ext4_has_feature_flex_bg(sb))
4635                 if (!ext4_fill_flex_info(sb)) {
4636                         ext4_msg(sb, KERN_ERR,
4637                                "unable to initialize "
4638                                "flex_bg meta info!");
4639                         ret = -ENOMEM;
4640                         goto failed_mount6;
4641                 }
4642
4643         err = ext4_register_li_request(sb, first_not_zeroed);
4644         if (err)
4645                 goto failed_mount6;
4646
4647         err = ext4_register_sysfs(sb);
4648         if (err)
4649                 goto failed_mount7;
4650
4651 #ifdef CONFIG_QUOTA
4652         /* Enable quota usage during mount. */
4653         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4654                 err = ext4_enable_quotas(sb);
4655                 if (err)
4656                         goto failed_mount8;
4657         }
4658 #endif  /* CONFIG_QUOTA */
4659
4660         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4661         ext4_orphan_cleanup(sb, es);
4662         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4663         if (needs_recovery) {
4664                 ext4_msg(sb, KERN_INFO, "recovery complete");
4665                 err = ext4_mark_recovery_complete(sb, es);
4666                 if (err)
4667                         goto failed_mount8;
4668         }
4669         if (EXT4_SB(sb)->s_journal) {
4670                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4671                         descr = " journalled data mode";
4672                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4673                         descr = " ordered data mode";
4674                 else
4675                         descr = " writeback data mode";
4676         } else
4677                 descr = "out journal";
4678
4679         if (test_opt(sb, DISCARD)) {
4680                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4681                 if (!blk_queue_discard(q))
4682                         ext4_msg(sb, KERN_WARNING,
4683                                  "mounting with \"discard\" option, but "
4684                                  "the device does not support discard");
4685         }
4686
4687         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4688                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4689                          "Opts: %.*s%s%s", descr,
4690                          (int) sizeof(sbi->s_es->s_mount_opts),
4691                          sbi->s_es->s_mount_opts,
4692                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4693
4694         if (es->s_error_count)
4695                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4696
4697         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4698         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4699         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4700         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4701
4702         kfree(orig_data);
4703         return 0;
4704
4705 cantfind_ext4:
4706         if (!silent)
4707                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4708         goto failed_mount;
4709
4710 failed_mount8:
4711         ext4_unregister_sysfs(sb);
4712         kobject_put(&sbi->s_kobj);
4713 failed_mount7:
4714         ext4_unregister_li_request(sb);
4715 failed_mount6:
4716         ext4_mb_release(sb);
4717         rcu_read_lock();
4718         flex_groups = rcu_dereference(sbi->s_flex_groups);
4719         if (flex_groups) {
4720                 for (i = 0; i < sbi->s_flex_groups_allocated; i++)
4721                         kvfree(flex_groups[i]);
4722                 kvfree(flex_groups);
4723         }
4724         rcu_read_unlock();
4725         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4726         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4727         percpu_counter_destroy(&sbi->s_dirs_counter);
4728         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4729         percpu_counter_destroy(&sbi->s_sra_exceeded_retry_limit);
4730         percpu_free_rwsem(&sbi->s_writepages_rwsem);
4731 failed_mount5:
4732         ext4_ext_release(sb);
4733         ext4_release_system_zone(sb);
4734 failed_mount4a:
4735         dput(sb->s_root);
4736         sb->s_root = NULL;
4737 failed_mount4:
4738         ext4_msg(sb, KERN_ERR, "mount failed");
4739         if (EXT4_SB(sb)->rsv_conversion_wq)
4740                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4741 failed_mount_wq:
4742         ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4743         sbi->s_ea_inode_cache = NULL;
4744
4745         ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4746         sbi->s_ea_block_cache = NULL;
4747
4748         if (sbi->s_journal) {
4749                 jbd2_journal_destroy(sbi->s_journal);
4750                 sbi->s_journal = NULL;
4751         }
4752 failed_mount3a:
4753         ext4_es_unregister_shrinker(sbi);
4754 failed_mount3:
4755         del_timer_sync(&sbi->s_err_report);
4756         if (sbi->s_mmp_tsk)
4757                 kthread_stop(sbi->s_mmp_tsk);
4758 failed_mount2:
4759         rcu_read_lock();
4760         group_desc = rcu_dereference(sbi->s_group_desc);
4761         for (i = 0; i < db_count; i++)
4762                 brelse(group_desc[i]);
4763         kvfree(group_desc);
4764         rcu_read_unlock();
4765 failed_mount:
4766         if (sbi->s_chksum_driver)
4767                 crypto_free_shash(sbi->s_chksum_driver);
4768
4769 #ifdef CONFIG_UNICODE
4770         utf8_unload(sbi->s_encoding);
4771 #endif
4772
4773 #ifdef CONFIG_QUOTA
4774         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4775                 kfree(get_qf_name(sb, sbi, i));
4776 #endif
4777         ext4_blkdev_remove(sbi);
4778         brelse(bh);
4779 out_fail:
4780         invalidate_bdev(sb->s_bdev);
4781         sb->s_fs_info = NULL;
4782         kfree(sbi->s_blockgroup_lock);
4783 out_free_base:
4784         kfree(sbi);
4785         kfree(orig_data);
4786         fs_put_dax(dax_dev);
4787         return err ? err : ret;
4788 }
4789
4790 /*
4791  * Setup any per-fs journal parameters now.  We'll do this both on
4792  * initial mount, once the journal has been initialised but before we've
4793  * done any recovery; and again on any subsequent remount.
4794  */
4795 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4796 {
4797         struct ext4_sb_info *sbi = EXT4_SB(sb);
4798
4799         journal->j_commit_interval = sbi->s_commit_interval;
4800         journal->j_min_batch_time = sbi->s_min_batch_time;
4801         journal->j_max_batch_time = sbi->s_max_batch_time;
4802
4803         write_lock(&journal->j_state_lock);
4804         if (test_opt(sb, BARRIER))
4805                 journal->j_flags |= JBD2_BARRIER;
4806         else
4807                 journal->j_flags &= ~JBD2_BARRIER;
4808         if (test_opt(sb, DATA_ERR_ABORT))
4809                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4810         else
4811                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4812         write_unlock(&journal->j_state_lock);
4813 }
4814
4815 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4816                                              unsigned int journal_inum)
4817 {
4818         struct inode *journal_inode;
4819
4820         /*
4821          * Test for the existence of a valid inode on disk.  Bad things
4822          * happen if we iget() an unused inode, as the subsequent iput()
4823          * will try to delete it.
4824          */
4825         journal_inode = ext4_iget(sb, journal_inum, EXT4_IGET_SPECIAL);
4826         if (IS_ERR(journal_inode)) {
4827                 ext4_msg(sb, KERN_ERR, "no journal found");
4828                 return NULL;
4829         }
4830         if (!journal_inode->i_nlink) {
4831                 make_bad_inode(journal_inode);
4832                 iput(journal_inode);
4833                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4834                 return NULL;
4835         }
4836
4837         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4838                   journal_inode, journal_inode->i_size);
4839         if (!S_ISREG(journal_inode->i_mode) || IS_ENCRYPTED(journal_inode)) {
4840                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4841                 iput(journal_inode);
4842                 return NULL;
4843         }
4844         return journal_inode;
4845 }
4846
4847 static journal_t *ext4_get_journal(struct super_block *sb,
4848                                    unsigned int journal_inum)
4849 {
4850         struct inode *journal_inode;
4851         journal_t *journal;
4852
4853         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4854                 return NULL;
4855
4856         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4857         if (!journal_inode)
4858                 return NULL;
4859
4860         journal = jbd2_journal_init_inode(journal_inode);
4861         if (!journal) {
4862                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4863                 iput(journal_inode);
4864                 return NULL;
4865         }
4866         journal->j_private = sb;
4867         ext4_init_journal_params(sb, journal);
4868         return journal;
4869 }
4870
4871 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4872                                        dev_t j_dev)
4873 {
4874         struct buffer_head *bh;
4875         journal_t *journal;
4876         ext4_fsblk_t start;
4877         ext4_fsblk_t len;
4878         int hblock, blocksize;
4879         ext4_fsblk_t sb_block;
4880         unsigned long offset;
4881         struct ext4_super_block *es;
4882         struct block_device *bdev;
4883
4884         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4885                 return NULL;
4886
4887         bdev = ext4_blkdev_get(j_dev, sb);
4888         if (bdev == NULL)
4889                 return NULL;
4890
4891         blocksize = sb->s_blocksize;
4892         hblock = bdev_logical_block_size(bdev);
4893         if (blocksize < hblock) {
4894                 ext4_msg(sb, KERN_ERR,
4895                         "blocksize too small for journal device");
4896                 goto out_bdev;
4897         }
4898
4899         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4900         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4901         set_blocksize(bdev, blocksize);
4902         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4903                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4904                        "external journal");
4905                 goto out_bdev;
4906         }
4907
4908         es = (struct ext4_super_block *) (bh->b_data + offset);
4909         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4910             !(le32_to_cpu(es->s_feature_incompat) &
4911               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4912                 ext4_msg(sb, KERN_ERR, "external journal has "
4913                                         "bad superblock");
4914                 brelse(bh);
4915                 goto out_bdev;
4916         }
4917
4918         if ((le32_to_cpu(es->s_feature_ro_compat) &
4919              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4920             es->s_checksum != ext4_superblock_csum(sb, es)) {
4921                 ext4_msg(sb, KERN_ERR, "external journal has "
4922                                        "corrupt superblock");
4923                 brelse(bh);
4924                 goto out_bdev;
4925         }
4926
4927         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4928                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4929                 brelse(bh);
4930                 goto out_bdev;
4931         }
4932
4933         len = ext4_blocks_count(es);
4934         start = sb_block + 1;
4935         brelse(bh);     /* we're done with the superblock */
4936
4937         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4938                                         start, len, blocksize);
4939         if (!journal) {
4940                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4941                 goto out_bdev;
4942         }
4943         journal->j_private = sb;
4944         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4945         wait_on_buffer(journal->j_sb_buffer);
4946         if (!buffer_uptodate(journal->j_sb_buffer)) {
4947                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4948                 goto out_journal;
4949         }
4950         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4951                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4952                                         "user (unsupported) - %d",
4953                         be32_to_cpu(journal->j_superblock->s_nr_users));
4954                 goto out_journal;
4955         }
4956         EXT4_SB(sb)->s_journal_bdev = bdev;
4957         ext4_init_journal_params(sb, journal);
4958         return journal;
4959
4960 out_journal:
4961         jbd2_journal_destroy(journal);
4962 out_bdev:
4963         ext4_blkdev_put(bdev);
4964         return NULL;
4965 }
4966
4967 static int ext4_load_journal(struct super_block *sb,
4968                              struct ext4_super_block *es,
4969                              unsigned long journal_devnum)
4970 {
4971         journal_t *journal;
4972         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4973         dev_t journal_dev;
4974         int err = 0;
4975         int really_read_only;
4976         int journal_dev_ro;
4977
4978         if (WARN_ON_ONCE(!ext4_has_feature_journal(sb)))
4979                 return -EFSCORRUPTED;
4980
4981         if (journal_devnum &&
4982             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4983                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4984                         "numbers have changed");
4985                 journal_dev = new_decode_dev(journal_devnum);
4986         } else
4987                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4988
4989         if (journal_inum && journal_dev) {
4990                 ext4_msg(sb, KERN_ERR,
4991                          "filesystem has both journal inode and journal device!");
4992                 return -EINVAL;
4993         }
4994
4995         if (journal_inum) {
4996                 journal = ext4_get_journal(sb, journal_inum);
4997                 if (!journal)
4998                         return -EINVAL;
4999         } else {
5000                 journal = ext4_get_dev_journal(sb, journal_dev);
5001                 if (!journal)
5002                         return -EINVAL;
5003         }
5004
5005         journal_dev_ro = bdev_read_only(journal->j_dev);
5006         really_read_only = bdev_read_only(sb->s_bdev) | journal_dev_ro;
5007
5008         if (journal_dev_ro && !sb_rdonly(sb)) {
5009                 ext4_msg(sb, KERN_ERR,
5010                          "journal device read-only, try mounting with '-o ro'");
5011                 err = -EROFS;
5012                 goto err_out;
5013         }
5014
5015         /*
5016          * Are we loading a blank journal or performing recovery after a
5017          * crash?  For recovery, we need to check in advance whether we
5018          * can get read-write access to the device.
5019          */
5020         if (ext4_has_feature_journal_needs_recovery(sb)) {
5021                 if (sb_rdonly(sb)) {
5022                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
5023                                         "required on readonly filesystem");
5024                         if (really_read_only) {
5025                                 ext4_msg(sb, KERN_ERR, "write access "
5026                                         "unavailable, cannot proceed "
5027                                         "(try mounting with noload)");
5028                                 err = -EROFS;
5029                                 goto err_out;
5030                         }
5031                         ext4_msg(sb, KERN_INFO, "write access will "
5032                                "be enabled during recovery");
5033                 }
5034         }
5035
5036         if (!(journal->j_flags & JBD2_BARRIER))
5037                 ext4_msg(sb, KERN_INFO, "barriers disabled");
5038
5039         if (!ext4_has_feature_journal_needs_recovery(sb))
5040                 err = jbd2_journal_wipe(journal, !really_read_only);
5041         if (!err) {
5042                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
5043                 if (save)
5044                         memcpy(save, ((char *) es) +
5045                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
5046                 err = jbd2_journal_load(journal);
5047                 if (save)
5048                         memcpy(((char *) es) + EXT4_S_ERR_START,
5049                                save, EXT4_S_ERR_LEN);
5050                 kfree(save);
5051         }
5052
5053         if (err) {
5054                 ext4_msg(sb, KERN_ERR, "error loading journal");
5055                 goto err_out;
5056         }
5057
5058         EXT4_SB(sb)->s_journal = journal;
5059         err = ext4_clear_journal_err(sb, es);
5060         if (err) {
5061                 EXT4_SB(sb)->s_journal = NULL;
5062                 jbd2_journal_destroy(journal);
5063                 return err;
5064         }
5065
5066         if (!really_read_only && journal_devnum &&
5067             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
5068                 es->s_journal_dev = cpu_to_le32(journal_devnum);
5069
5070                 /* Make sure we flush the recovery flag to disk. */
5071                 ext4_commit_super(sb, 1);
5072         }
5073
5074         return 0;
5075
5076 err_out:
5077         jbd2_journal_destroy(journal);
5078         return err;
5079 }
5080
5081 static int ext4_commit_super(struct super_block *sb, int sync)
5082 {
5083         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5084         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
5085         int error = 0;
5086
5087         if (!sbh)
5088                 return -EINVAL;
5089         if (block_device_ejected(sb))
5090                 return -ENODEV;
5091
5092         /*
5093          * If the file system is mounted read-only, don't update the
5094          * superblock write time.  This avoids updating the superblock
5095          * write time when we are mounting the root file system
5096          * read/only but we need to replay the journal; at that point,
5097          * for people who are east of GMT and who make their clock
5098          * tick in localtime for Windows bug-for-bug compatibility,
5099          * the clock is set in the future, and this will cause e2fsck
5100          * to complain and force a full file system check.
5101          */
5102         if (!(sb->s_flags & SB_RDONLY))
5103                 ext4_update_tstamp(es, s_wtime);
5104         if (sb->s_bdev->bd_part)
5105                 es->s_kbytes_written =
5106                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
5107                             ((part_stat_read(sb->s_bdev->bd_part,
5108                                              sectors[STAT_WRITE]) -
5109                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
5110         else
5111                 es->s_kbytes_written =
5112                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
5113         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
5114                 ext4_free_blocks_count_set(es,
5115                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
5116                                 &EXT4_SB(sb)->s_freeclusters_counter)));
5117         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
5118                 es->s_free_inodes_count =
5119                         cpu_to_le32(percpu_counter_sum_positive(
5120                                 &EXT4_SB(sb)->s_freeinodes_counter));
5121         BUFFER_TRACE(sbh, "marking dirty");
5122         ext4_superblock_csum_set(sb);
5123         if (sync)
5124                 lock_buffer(sbh);
5125         if (buffer_write_io_error(sbh) || !buffer_uptodate(sbh)) {
5126                 /*
5127                  * Oh, dear.  A previous attempt to write the
5128                  * superblock failed.  This could happen because the
5129                  * USB device was yanked out.  Or it could happen to
5130                  * be a transient write error and maybe the block will
5131                  * be remapped.  Nothing we can do but to retry the
5132                  * write and hope for the best.
5133                  */
5134                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
5135                        "superblock detected");
5136                 clear_buffer_write_io_error(sbh);
5137                 set_buffer_uptodate(sbh);
5138         }
5139         mark_buffer_dirty(sbh);
5140         if (sync) {
5141                 unlock_buffer(sbh);
5142                 error = __sync_dirty_buffer(sbh,
5143                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
5144                 if (buffer_write_io_error(sbh)) {
5145                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
5146                                "superblock");
5147                         clear_buffer_write_io_error(sbh);
5148                         set_buffer_uptodate(sbh);
5149                 }
5150         }
5151         return error;
5152 }
5153
5154 /*
5155  * Have we just finished recovery?  If so, and if we are mounting (or
5156  * remounting) the filesystem readonly, then we will end up with a
5157  * consistent fs on disk.  Record that fact.
5158  */
5159 static int ext4_mark_recovery_complete(struct super_block *sb,
5160                                        struct ext4_super_block *es)
5161 {
5162         int err;
5163         journal_t *journal = EXT4_SB(sb)->s_journal;
5164
5165         if (!ext4_has_feature_journal(sb)) {
5166                 if (journal != NULL) {
5167                         ext4_error(sb, "Journal got removed while the fs was "
5168                                    "mounted!");
5169                         return -EFSCORRUPTED;
5170                 }
5171                 return 0;
5172         }
5173         jbd2_journal_lock_updates(journal);
5174         err = jbd2_journal_flush(journal);
5175         if (err < 0)
5176                 goto out;
5177
5178         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
5179                 ext4_clear_feature_journal_needs_recovery(sb);
5180                 ext4_commit_super(sb, 1);
5181         }
5182 out:
5183         jbd2_journal_unlock_updates(journal);
5184         return err;
5185 }
5186
5187 /*
5188  * If we are mounting (or read-write remounting) a filesystem whose journal
5189  * has recorded an error from a previous lifetime, move that error to the
5190  * main filesystem now.
5191  */
5192 static int ext4_clear_journal_err(struct super_block *sb,
5193                                    struct ext4_super_block *es)
5194 {
5195         journal_t *journal;
5196         int j_errno;
5197         const char *errstr;
5198
5199         if (!ext4_has_feature_journal(sb)) {
5200                 ext4_error(sb, "Journal got removed while the fs was mounted!");
5201                 return -EFSCORRUPTED;
5202         }
5203
5204         journal = EXT4_SB(sb)->s_journal;
5205
5206         /*
5207          * Now check for any error status which may have been recorded in the
5208          * journal by a prior ext4_error() or ext4_abort()
5209          */
5210
5211         j_errno = jbd2_journal_errno(journal);
5212         if (j_errno) {
5213                 char nbuf[16];
5214
5215                 errstr = ext4_decode_error(sb, j_errno, nbuf);
5216                 ext4_warning(sb, "Filesystem error recorded "
5217                              "from previous mount: %s", errstr);
5218                 ext4_warning(sb, "Marking fs in need of filesystem check.");
5219
5220                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
5221                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
5222                 ext4_commit_super(sb, 1);
5223
5224                 jbd2_journal_clear_err(journal);
5225                 jbd2_journal_update_sb_errno(journal);
5226         }
5227         return 0;
5228 }
5229
5230 /*
5231  * Force the running and committing transactions to commit,
5232  * and wait on the commit.
5233  */
5234 int ext4_force_commit(struct super_block *sb)
5235 {
5236         journal_t *journal;
5237
5238         if (sb_rdonly(sb))
5239                 return 0;
5240
5241         journal = EXT4_SB(sb)->s_journal;
5242         return ext4_journal_force_commit(journal);
5243 }
5244
5245 static int ext4_sync_fs(struct super_block *sb, int wait)
5246 {
5247         int ret = 0;
5248         tid_t target;
5249         bool needs_barrier = false;
5250         struct ext4_sb_info *sbi = EXT4_SB(sb);
5251
5252         if (unlikely(ext4_forced_shutdown(sbi)))
5253                 return 0;
5254
5255         trace_ext4_sync_fs(sb, wait);
5256         flush_workqueue(sbi->rsv_conversion_wq);
5257         /*
5258          * Writeback quota in non-journalled quota case - journalled quota has
5259          * no dirty dquots
5260          */
5261         dquot_writeback_dquots(sb, -1);
5262         /*
5263          * Data writeback is possible w/o journal transaction, so barrier must
5264          * being sent at the end of the function. But we can skip it if
5265          * transaction_commit will do it for us.
5266          */
5267         if (sbi->s_journal) {
5268                 target = jbd2_get_latest_transaction(sbi->s_journal);
5269                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5270                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5271                         needs_barrier = true;
5272
5273                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5274                         if (wait)
5275                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5276                                                            target);
5277                 }
5278         } else if (wait && test_opt(sb, BARRIER))
5279                 needs_barrier = true;
5280         if (needs_barrier) {
5281                 int err;
5282                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5283                 if (!ret)
5284                         ret = err;
5285         }
5286
5287         return ret;
5288 }
5289
5290 /*
5291  * LVM calls this function before a (read-only) snapshot is created.  This
5292  * gives us a chance to flush the journal completely and mark the fs clean.
5293  *
5294  * Note that only this function cannot bring a filesystem to be in a clean
5295  * state independently. It relies on upper layer to stop all data & metadata
5296  * modifications.
5297  */
5298 static int ext4_freeze(struct super_block *sb)
5299 {
5300         int error = 0;
5301         journal_t *journal;
5302
5303         if (sb_rdonly(sb))
5304                 return 0;
5305
5306         journal = EXT4_SB(sb)->s_journal;
5307
5308         if (journal) {
5309                 /* Now we set up the journal barrier. */
5310                 jbd2_journal_lock_updates(journal);
5311
5312                 /*
5313                  * Don't clear the needs_recovery flag if we failed to
5314                  * flush the journal.
5315                  */
5316                 error = jbd2_journal_flush(journal);
5317                 if (error < 0)
5318                         goto out;
5319
5320                 /* Journal blocked and flushed, clear needs_recovery flag. */
5321                 ext4_clear_feature_journal_needs_recovery(sb);
5322         }
5323
5324         error = ext4_commit_super(sb, 1);
5325 out:
5326         if (journal)
5327                 /* we rely on upper layer to stop further updates */
5328                 jbd2_journal_unlock_updates(journal);
5329         return error;
5330 }
5331
5332 /*
5333  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5334  * flag here, even though the filesystem is not technically dirty yet.
5335  */
5336 static int ext4_unfreeze(struct super_block *sb)
5337 {
5338         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5339                 return 0;
5340
5341         if (EXT4_SB(sb)->s_journal) {
5342                 /* Reset the needs_recovery flag before the fs is unlocked. */
5343                 ext4_set_feature_journal_needs_recovery(sb);
5344         }
5345
5346         ext4_commit_super(sb, 1);
5347         return 0;
5348 }
5349
5350 /*
5351  * Structure to save mount options for ext4_remount's benefit
5352  */
5353 struct ext4_mount_options {
5354         unsigned long s_mount_opt;
5355         unsigned long s_mount_opt2;
5356         kuid_t s_resuid;
5357         kgid_t s_resgid;
5358         unsigned long s_commit_interval;
5359         u32 s_min_batch_time, s_max_batch_time;
5360 #ifdef CONFIG_QUOTA
5361         int s_jquota_fmt;
5362         char *s_qf_names[EXT4_MAXQUOTAS];
5363 #endif
5364 };
5365
5366 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5367 {
5368         struct ext4_super_block *es;
5369         struct ext4_sb_info *sbi = EXT4_SB(sb);
5370         unsigned long old_sb_flags, vfs_flags;
5371         struct ext4_mount_options old_opts;
5372         int enable_quota = 0;
5373         ext4_group_t g;
5374         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5375         int err = 0;
5376 #ifdef CONFIG_QUOTA
5377         int i, j;
5378         char *to_free[EXT4_MAXQUOTAS];
5379 #endif
5380         char *orig_data = kstrdup(data, GFP_KERNEL);
5381
5382         if (data && !orig_data)
5383                 return -ENOMEM;
5384
5385         /* Store the original options */
5386         old_sb_flags = sb->s_flags;
5387         old_opts.s_mount_opt = sbi->s_mount_opt;
5388         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5389         old_opts.s_resuid = sbi->s_resuid;
5390         old_opts.s_resgid = sbi->s_resgid;
5391         old_opts.s_commit_interval = sbi->s_commit_interval;
5392         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5393         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5394 #ifdef CONFIG_QUOTA
5395         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5396         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5397                 if (sbi->s_qf_names[i]) {
5398                         char *qf_name = get_qf_name(sb, sbi, i);
5399
5400                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5401                         if (!old_opts.s_qf_names[i]) {
5402                                 for (j = 0; j < i; j++)
5403                                         kfree(old_opts.s_qf_names[j]);
5404                                 kfree(orig_data);
5405                                 return -ENOMEM;
5406                         }
5407                 } else
5408                         old_opts.s_qf_names[i] = NULL;
5409 #endif
5410         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5411                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5412
5413         /*
5414          * Some options can be enabled by ext4 and/or by VFS mount flag
5415          * either way we need to make sure it matches in both *flags and
5416          * s_flags. Copy those selected flags from *flags to s_flags
5417          */
5418         vfs_flags = SB_LAZYTIME | SB_I_VERSION;
5419         sb->s_flags = (sb->s_flags & ~vfs_flags) | (*flags & vfs_flags);
5420
5421         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5422                 err = -EINVAL;
5423                 goto restore_opts;
5424         }
5425
5426         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5427             test_opt(sb, JOURNAL_CHECKSUM)) {
5428                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5429                          "during remount not supported; ignoring");
5430                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5431         }
5432
5433         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5434                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5435                         ext4_msg(sb, KERN_ERR, "can't mount with "
5436                                  "both data=journal and delalloc");
5437                         err = -EINVAL;
5438                         goto restore_opts;
5439                 }
5440                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5441                         ext4_msg(sb, KERN_ERR, "can't mount with "
5442                                  "both data=journal and dioread_nolock");
5443                         err = -EINVAL;
5444                         goto restore_opts;
5445                 }
5446         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5447                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5448                         ext4_msg(sb, KERN_ERR, "can't mount with "
5449                                 "journal_async_commit in data=ordered mode");
5450                         err = -EINVAL;
5451                         goto restore_opts;
5452                 }
5453         }
5454
5455         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5456                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5457                 err = -EINVAL;
5458                 goto restore_opts;
5459         }
5460
5461         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5462                 ext4_abort(sb, "Abort forced by user");
5463
5464         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5465                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5466
5467         es = sbi->s_es;
5468
5469         if (sbi->s_journal) {
5470                 ext4_init_journal_params(sb, sbi->s_journal);
5471                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5472         }
5473
5474         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5475                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5476                         err = -EROFS;
5477                         goto restore_opts;
5478                 }
5479
5480                 if (*flags & SB_RDONLY) {
5481                         err = sync_filesystem(sb);
5482                         if (err < 0)
5483                                 goto restore_opts;
5484                         err = dquot_suspend(sb, -1);
5485                         if (err < 0)
5486                                 goto restore_opts;
5487
5488                         /*
5489                          * First of all, the unconditional stuff we have to do
5490                          * to disable replay of the journal when we next remount
5491                          */
5492                         sb->s_flags |= SB_RDONLY;
5493
5494                         /*
5495                          * OK, test if we are remounting a valid rw partition
5496                          * readonly, and if so set the rdonly flag and then
5497                          * mark the partition as valid again.
5498                          */
5499                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5500                             (sbi->s_mount_state & EXT4_VALID_FS))
5501                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5502
5503                         if (sbi->s_journal) {
5504                                 /*
5505                                  * We let remount-ro finish even if marking fs
5506                                  * as clean failed...
5507                                  */
5508                                 ext4_mark_recovery_complete(sb, es);
5509                         }
5510                         if (sbi->s_mmp_tsk)
5511                                 kthread_stop(sbi->s_mmp_tsk);
5512                 } else {
5513                         /* Make sure we can mount this feature set readwrite */
5514                         if (ext4_has_feature_readonly(sb) ||
5515                             !ext4_feature_set_ok(sb, 0)) {
5516                                 err = -EROFS;
5517                                 goto restore_opts;
5518                         }
5519                         /*
5520                          * Make sure the group descriptor checksums
5521                          * are sane.  If they aren't, refuse to remount r/w.
5522                          */
5523                         for (g = 0; g < sbi->s_groups_count; g++) {
5524                                 struct ext4_group_desc *gdp =
5525                                         ext4_get_group_desc(sb, g, NULL);
5526
5527                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5528                                         ext4_msg(sb, KERN_ERR,
5529                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5530                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5531                                                le16_to_cpu(gdp->bg_checksum));
5532                                         err = -EFSBADCRC;
5533                                         goto restore_opts;
5534                                 }
5535                         }
5536
5537                         /*
5538                          * If we have an unprocessed orphan list hanging
5539                          * around from a previously readonly bdev mount,
5540                          * require a full umount/remount for now.
5541                          */
5542                         if (es->s_last_orphan) {
5543                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5544                                        "remount RDWR because of unprocessed "
5545                                        "orphan inode list.  Please "
5546                                        "umount/remount instead");
5547                                 err = -EINVAL;
5548                                 goto restore_opts;
5549                         }
5550
5551                         /*
5552                          * Mounting a RDONLY partition read-write, so reread
5553                          * and store the current valid flag.  (It may have
5554                          * been changed by e2fsck since we originally mounted
5555                          * the partition.)
5556                          */
5557                         if (sbi->s_journal) {
5558                                 err = ext4_clear_journal_err(sb, es);
5559                                 if (err)
5560                                         goto restore_opts;
5561                         }
5562                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5563
5564                         err = ext4_setup_super(sb, es, 0);
5565                         if (err)
5566                                 goto restore_opts;
5567
5568                         sb->s_flags &= ~SB_RDONLY;
5569                         if (ext4_has_feature_mmp(sb))
5570                                 if (ext4_multi_mount_protect(sb,
5571                                                 le64_to_cpu(es->s_mmp_block))) {
5572                                         err = -EROFS;
5573                                         goto restore_opts;
5574                                 }
5575                         enable_quota = 1;
5576                 }
5577         }
5578
5579         /*
5580          * Reinitialize lazy itable initialization thread based on
5581          * current settings
5582          */
5583         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5584                 ext4_unregister_li_request(sb);
5585         else {
5586                 ext4_group_t first_not_zeroed;
5587                 first_not_zeroed = ext4_has_uninit_itable(sb);
5588                 ext4_register_li_request(sb, first_not_zeroed);
5589         }
5590
5591         /*
5592          * Handle creation of system zone data early because it can fail.
5593          * Releasing of existing data is done when we are sure remount will
5594          * succeed.
5595          */
5596         if (test_opt(sb, BLOCK_VALIDITY) && !sbi->system_blks) {
5597                 err = ext4_setup_system_zone(sb);
5598                 if (err)
5599                         goto restore_opts;
5600         }
5601
5602         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5603                 err = ext4_commit_super(sb, 1);
5604                 if (err)
5605                         goto restore_opts;
5606         }
5607
5608 #ifdef CONFIG_QUOTA
5609         if (enable_quota) {
5610                 if (sb_any_quota_suspended(sb))
5611                         dquot_resume(sb, -1);
5612                 else if (ext4_has_feature_quota(sb)) {
5613                         err = ext4_enable_quotas(sb);
5614                         if (err)
5615                                 goto restore_opts;
5616                 }
5617         }
5618         /* Release old quota file names */
5619         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5620                 kfree(old_opts.s_qf_names[i]);
5621 #endif
5622         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->system_blks)
5623                 ext4_release_system_zone(sb);
5624
5625         /*
5626          * Some options can be enabled by ext4 and/or by VFS mount flag
5627          * either way we need to make sure it matches in both *flags and
5628          * s_flags. Copy those selected flags from s_flags to *flags
5629          */
5630         *flags = (*flags & ~vfs_flags) | (sb->s_flags & vfs_flags);
5631
5632         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5633         kfree(orig_data);
5634         return 0;
5635
5636 restore_opts:
5637         /*
5638          * If there was a failing r/w to ro transition, we may need to
5639          * re-enable quota
5640          */
5641         if ((sb->s_flags & SB_RDONLY) && !(old_sb_flags & SB_RDONLY) &&
5642             sb_any_quota_suspended(sb))
5643                 dquot_resume(sb, -1);
5644         sb->s_flags = old_sb_flags;
5645         sbi->s_mount_opt = old_opts.s_mount_opt;
5646         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5647         sbi->s_resuid = old_opts.s_resuid;
5648         sbi->s_resgid = old_opts.s_resgid;
5649         sbi->s_commit_interval = old_opts.s_commit_interval;
5650         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5651         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5652         if (!test_opt(sb, BLOCK_VALIDITY) && sbi->system_blks)
5653                 ext4_release_system_zone(sb);
5654 #ifdef CONFIG_QUOTA
5655         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5656         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5657                 to_free[i] = get_qf_name(sb, sbi, i);
5658                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5659         }
5660         synchronize_rcu();
5661         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5662                 kfree(to_free[i]);
5663 #endif
5664         kfree(orig_data);
5665         return err;
5666 }
5667
5668 #ifdef CONFIG_QUOTA
5669 static int ext4_statfs_project(struct super_block *sb,
5670                                kprojid_t projid, struct kstatfs *buf)
5671 {
5672         struct kqid qid;
5673         struct dquot *dquot;
5674         u64 limit;
5675         u64 curblock;
5676
5677         qid = make_kqid_projid(projid);
5678         dquot = dqget(sb, qid);
5679         if (IS_ERR(dquot))
5680                 return PTR_ERR(dquot);
5681         spin_lock(&dquot->dq_dqb_lock);
5682
5683         limit = 0;
5684         if (dquot->dq_dqb.dqb_bsoftlimit &&
5685             (!limit || dquot->dq_dqb.dqb_bsoftlimit < limit))
5686                 limit = dquot->dq_dqb.dqb_bsoftlimit;
5687         if (dquot->dq_dqb.dqb_bhardlimit &&
5688             (!limit || dquot->dq_dqb.dqb_bhardlimit < limit))
5689                 limit = dquot->dq_dqb.dqb_bhardlimit;
5690         limit >>= sb->s_blocksize_bits;
5691
5692         if (limit && buf->f_blocks > limit) {
5693                 curblock = (dquot->dq_dqb.dqb_curspace +
5694                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5695                 buf->f_blocks = limit;
5696                 buf->f_bfree = buf->f_bavail =
5697                         (buf->f_blocks > curblock) ?
5698                          (buf->f_blocks - curblock) : 0;
5699         }
5700
5701         limit = 0;
5702         if (dquot->dq_dqb.dqb_isoftlimit &&
5703             (!limit || dquot->dq_dqb.dqb_isoftlimit < limit))
5704                 limit = dquot->dq_dqb.dqb_isoftlimit;
5705         if (dquot->dq_dqb.dqb_ihardlimit &&
5706             (!limit || dquot->dq_dqb.dqb_ihardlimit < limit))
5707                 limit = dquot->dq_dqb.dqb_ihardlimit;
5708
5709         if (limit && buf->f_files > limit) {
5710                 buf->f_files = limit;
5711                 buf->f_ffree =
5712                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5713                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5714         }
5715
5716         spin_unlock(&dquot->dq_dqb_lock);
5717         dqput(dquot);
5718         return 0;
5719 }
5720 #endif
5721
5722 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5723 {
5724         struct super_block *sb = dentry->d_sb;
5725         struct ext4_sb_info *sbi = EXT4_SB(sb);
5726         struct ext4_super_block *es = sbi->s_es;
5727         ext4_fsblk_t overhead = 0, resv_blocks;
5728         u64 fsid;
5729         s64 bfree;
5730         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5731
5732         if (!test_opt(sb, MINIX_DF))
5733                 overhead = sbi->s_overhead;
5734
5735         buf->f_type = EXT4_SUPER_MAGIC;
5736         buf->f_bsize = sb->s_blocksize;
5737         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5738         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5739                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5740         /* prevent underflow in case that few free space is available */
5741         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5742         buf->f_bavail = buf->f_bfree -
5743                         (ext4_r_blocks_count(es) + resv_blocks);
5744         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5745                 buf->f_bavail = 0;
5746         buf->f_files = le32_to_cpu(es->s_inodes_count);
5747         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5748         buf->f_namelen = EXT4_NAME_LEN;
5749         fsid = le64_to_cpup((void *)es->s_uuid) ^
5750                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5751         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5752         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5753
5754 #ifdef CONFIG_QUOTA
5755         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5756             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5757                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5758 #endif
5759         return 0;
5760 }
5761
5762
5763 #ifdef CONFIG_QUOTA
5764
5765 /*
5766  * Helper functions so that transaction is started before we acquire dqio_sem
5767  * to keep correct lock ordering of transaction > dqio_sem
5768  */
5769 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5770 {
5771         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5772 }
5773
5774 static int ext4_write_dquot(struct dquot *dquot)
5775 {
5776         int ret, err;
5777         handle_t *handle;
5778         struct inode *inode;
5779
5780         inode = dquot_to_inode(dquot);
5781         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5782                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5783         if (IS_ERR(handle))
5784                 return PTR_ERR(handle);
5785         ret = dquot_commit(dquot);
5786         err = ext4_journal_stop(handle);
5787         if (!ret)
5788                 ret = err;
5789         return ret;
5790 }
5791
5792 static int ext4_acquire_dquot(struct dquot *dquot)
5793 {
5794         int ret, err;
5795         handle_t *handle;
5796
5797         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5798                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5799         if (IS_ERR(handle))
5800                 return PTR_ERR(handle);
5801         ret = dquot_acquire(dquot);
5802         err = ext4_journal_stop(handle);
5803         if (!ret)
5804                 ret = err;
5805         return ret;
5806 }
5807
5808 static int ext4_release_dquot(struct dquot *dquot)
5809 {
5810         int ret, err;
5811         handle_t *handle;
5812
5813         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5814                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5815         if (IS_ERR(handle)) {
5816                 /* Release dquot anyway to avoid endless cycle in dqput() */
5817                 dquot_release(dquot);
5818                 return PTR_ERR(handle);
5819         }
5820         ret = dquot_release(dquot);
5821         err = ext4_journal_stop(handle);
5822         if (!ret)
5823                 ret = err;
5824         return ret;
5825 }
5826
5827 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5828 {
5829         struct super_block *sb = dquot->dq_sb;
5830         struct ext4_sb_info *sbi = EXT4_SB(sb);
5831
5832         /* Are we journaling quotas? */
5833         if (ext4_has_feature_quota(sb) ||
5834             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5835                 dquot_mark_dquot_dirty(dquot);
5836                 return ext4_write_dquot(dquot);
5837         } else {
5838                 return dquot_mark_dquot_dirty(dquot);
5839         }
5840 }
5841
5842 static int ext4_write_info(struct super_block *sb, int type)
5843 {
5844         int ret, err;
5845         handle_t *handle;
5846
5847         /* Data block + inode block */
5848         handle = ext4_journal_start_sb(sb, EXT4_HT_QUOTA, 2);
5849         if (IS_ERR(handle))
5850                 return PTR_ERR(handle);
5851         ret = dquot_commit_info(sb, type);
5852         err = ext4_journal_stop(handle);
5853         if (!ret)
5854                 ret = err;
5855         return ret;
5856 }
5857
5858 /*
5859  * Turn on quotas during mount time - we need to find
5860  * the quota file and such...
5861  */
5862 static int ext4_quota_on_mount(struct super_block *sb, int type)
5863 {
5864         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5865                                         EXT4_SB(sb)->s_jquota_fmt, type);
5866 }
5867
5868 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5869 {
5870         struct ext4_inode_info *ei = EXT4_I(inode);
5871
5872         /* The first argument of lockdep_set_subclass has to be
5873          * *exactly* the same as the argument to init_rwsem() --- in
5874          * this case, in init_once() --- or lockdep gets unhappy
5875          * because the name of the lock is set using the
5876          * stringification of the argument to init_rwsem().
5877          */
5878         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5879         lockdep_set_subclass(&ei->i_data_sem, subclass);
5880 }
5881
5882 /*
5883  * Standard function to be called on quota_on
5884  */
5885 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5886                          const struct path *path)
5887 {
5888         int err;
5889
5890         if (!test_opt(sb, QUOTA))
5891                 return -EINVAL;
5892
5893         /* Quotafile not on the same filesystem? */
5894         if (path->dentry->d_sb != sb)
5895                 return -EXDEV;
5896
5897         /* Quota already enabled for this file? */
5898         if (IS_NOQUOTA(d_inode(path->dentry)))
5899                 return -EBUSY;
5900
5901         /* Journaling quota? */
5902         if (EXT4_SB(sb)->s_qf_names[type]) {
5903                 /* Quotafile not in fs root? */
5904                 if (path->dentry->d_parent != sb->s_root)
5905                         ext4_msg(sb, KERN_WARNING,
5906                                 "Quota file not on filesystem root. "
5907                                 "Journaled quota will not work");
5908                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5909         } else {
5910                 /*
5911                  * Clear the flag just in case mount options changed since
5912                  * last time.
5913                  */
5914                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5915         }
5916
5917         /*
5918          * When we journal data on quota file, we have to flush journal to see
5919          * all updates to the file when we bypass pagecache...
5920          */
5921         if (EXT4_SB(sb)->s_journal &&
5922             ext4_should_journal_data(d_inode(path->dentry))) {
5923                 /*
5924                  * We don't need to lock updates but journal_flush() could
5925                  * otherwise be livelocked...
5926                  */
5927                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5928                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5929                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5930                 if (err)
5931                         return err;
5932         }
5933
5934         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5935         err = dquot_quota_on(sb, type, format_id, path);
5936         if (!err) {
5937                 struct inode *inode = d_inode(path->dentry);
5938                 handle_t *handle;
5939
5940                 /*
5941                  * Set inode flags to prevent userspace from messing with quota
5942                  * files. If this fails, we return success anyway since quotas
5943                  * are already enabled and this is not a hard failure.
5944                  */
5945                 inode_lock(inode);
5946                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5947                 if (IS_ERR(handle))
5948                         goto unlock_inode;
5949                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5950                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5951                                 S_NOATIME | S_IMMUTABLE);
5952                 ext4_mark_inode_dirty(handle, inode);
5953                 ext4_journal_stop(handle);
5954         unlock_inode:
5955                 inode_unlock(inode);
5956                 if (err)
5957                         dquot_quota_off(sb, type);
5958         }
5959         if (err)
5960                 lockdep_set_quota_inode(path->dentry->d_inode,
5961                                              I_DATA_SEM_NORMAL);
5962         return err;
5963 }
5964
5965 static inline bool ext4_check_quota_inum(int type, unsigned long qf_inum)
5966 {
5967         switch (type) {
5968         case USRQUOTA:
5969                 return qf_inum == EXT4_USR_QUOTA_INO;
5970         case GRPQUOTA:
5971                 return qf_inum == EXT4_GRP_QUOTA_INO;
5972         case PRJQUOTA:
5973                 return qf_inum >= EXT4_GOOD_OLD_FIRST_INO;
5974         default:
5975                 BUG();
5976         }
5977 }
5978
5979 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5980                              unsigned int flags)
5981 {
5982         int err;
5983         struct inode *qf_inode;
5984         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5985                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5986                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5987                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5988         };
5989
5990         BUG_ON(!ext4_has_feature_quota(sb));
5991
5992         if (!qf_inums[type])
5993                 return -EPERM;
5994
5995         if (!ext4_check_quota_inum(type, qf_inums[type])) {
5996                 ext4_error(sb, "Bad quota inum: %lu, type: %d",
5997                                 qf_inums[type], type);
5998                 return -EUCLEAN;
5999         }
6000
6001         qf_inode = ext4_iget(sb, qf_inums[type], EXT4_IGET_SPECIAL);
6002         if (IS_ERR(qf_inode)) {
6003                 ext4_error(sb, "Bad quota inode: %lu, type: %d",
6004                                 qf_inums[type], type);
6005                 return PTR_ERR(qf_inode);
6006         }
6007
6008         /* Don't account quota for quota files to avoid recursion */
6009         qf_inode->i_flags |= S_NOQUOTA;
6010         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
6011         err = dquot_enable(qf_inode, type, format_id, flags);
6012         if (err)
6013                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
6014         iput(qf_inode);
6015
6016         return err;
6017 }
6018
6019 /* Enable usage tracking for all quota types. */
6020 static int ext4_enable_quotas(struct super_block *sb)
6021 {
6022         int type, err = 0;
6023         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
6024                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
6025                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
6026                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
6027         };
6028         bool quota_mopt[EXT4_MAXQUOTAS] = {
6029                 test_opt(sb, USRQUOTA),
6030                 test_opt(sb, GRPQUOTA),
6031                 test_opt(sb, PRJQUOTA),
6032         };
6033
6034         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
6035         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
6036                 if (qf_inums[type]) {
6037                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
6038                                 DQUOT_USAGE_ENABLED |
6039                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
6040                         if (err) {
6041                                 ext4_warning(sb,
6042                                         "Failed to enable quota tracking "
6043                                         "(type=%d, err=%d, ino=%lu). "
6044                                         "Please run e2fsck to fix.", type,
6045                                         err, qf_inums[type]);
6046                                 for (type--; type >= 0; type--) {
6047                                         struct inode *inode;
6048
6049                                         inode = sb_dqopt(sb)->files[type];
6050                                         if (inode)
6051                                                 inode = igrab(inode);
6052                                         dquot_quota_off(sb, type);
6053                                         if (inode) {
6054                                                 lockdep_set_quota_inode(inode,
6055                                                         I_DATA_SEM_NORMAL);
6056                                                 iput(inode);
6057                                         }
6058                                 }
6059
6060                                 return err;
6061                         }
6062                 }
6063         }
6064         return 0;
6065 }
6066
6067 static int ext4_quota_off(struct super_block *sb, int type)
6068 {
6069         struct inode *inode = sb_dqopt(sb)->files[type];
6070         handle_t *handle;
6071         int err;
6072
6073         /* Force all delayed allocation blocks to be allocated.
6074          * Caller already holds s_umount sem */
6075         if (test_opt(sb, DELALLOC))
6076                 sync_filesystem(sb);
6077
6078         if (!inode || !igrab(inode))
6079                 goto out;
6080
6081         err = dquot_quota_off(sb, type);
6082         if (err || ext4_has_feature_quota(sb))
6083                 goto out_put;
6084
6085         inode_lock(inode);
6086         /*
6087          * Update modification times of quota files when userspace can
6088          * start looking at them. If we fail, we return success anyway since
6089          * this is not a hard failure and quotas are already disabled.
6090          */
6091         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
6092         if (IS_ERR(handle))
6093                 goto out_unlock;
6094         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
6095         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
6096         inode->i_mtime = inode->i_ctime = current_time(inode);
6097         ext4_mark_inode_dirty(handle, inode);
6098         ext4_journal_stop(handle);
6099 out_unlock:
6100         inode_unlock(inode);
6101 out_put:
6102         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
6103         iput(inode);
6104         return err;
6105 out:
6106         return dquot_quota_off(sb, type);
6107 }
6108
6109 /* Read data from quotafile - avoid pagecache and such because we cannot afford
6110  * acquiring the locks... As quota files are never truncated and quota code
6111  * itself serializes the operations (and no one else should touch the files)
6112  * we don't have to be afraid of races */
6113 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
6114                                size_t len, loff_t off)
6115 {
6116         struct inode *inode = sb_dqopt(sb)->files[type];
6117         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6118         int offset = off & (sb->s_blocksize - 1);
6119         int tocopy;
6120         size_t toread;
6121         struct buffer_head *bh;
6122         loff_t i_size = i_size_read(inode);
6123
6124         if (off > i_size)
6125                 return 0;
6126         if (off+len > i_size)
6127                 len = i_size-off;
6128         toread = len;
6129         while (toread > 0) {
6130                 tocopy = sb->s_blocksize - offset < toread ?
6131                                 sb->s_blocksize - offset : toread;
6132                 bh = ext4_bread(NULL, inode, blk, 0);
6133                 if (IS_ERR(bh))
6134                         return PTR_ERR(bh);
6135                 if (!bh)        /* A hole? */
6136                         memset(data, 0, tocopy);
6137                 else
6138                         memcpy(data, bh->b_data+offset, tocopy);
6139                 brelse(bh);
6140                 offset = 0;
6141                 toread -= tocopy;
6142                 data += tocopy;
6143                 blk++;
6144         }
6145         return len;
6146 }
6147
6148 /* Write to quotafile (we know the transaction is already started and has
6149  * enough credits) */
6150 static ssize_t ext4_quota_write(struct super_block *sb, int type,
6151                                 const char *data, size_t len, loff_t off)
6152 {
6153         struct inode *inode = sb_dqopt(sb)->files[type];
6154         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
6155         int err, offset = off & (sb->s_blocksize - 1);
6156         int retries = 0;
6157         struct buffer_head *bh;
6158         handle_t *handle = journal_current_handle();
6159
6160         if (!handle) {
6161                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6162                         " cancelled because transaction is not started",
6163                         (unsigned long long)off, (unsigned long long)len);
6164                 return -EIO;
6165         }
6166         /*
6167          * Since we account only one data block in transaction credits,
6168          * then it is impossible to cross a block boundary.
6169          */
6170         if (sb->s_blocksize - offset < len) {
6171                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
6172                         " cancelled because not block aligned",
6173                         (unsigned long long)off, (unsigned long long)len);
6174                 return -EIO;
6175         }
6176
6177         do {
6178                 bh = ext4_bread(handle, inode, blk,
6179                                 EXT4_GET_BLOCKS_CREATE |
6180                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
6181         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
6182                  ext4_should_retry_alloc(inode->i_sb, &retries));
6183         if (IS_ERR(bh))
6184                 return PTR_ERR(bh);
6185         if (!bh)
6186                 goto out;
6187         BUFFER_TRACE(bh, "get write access");
6188         err = ext4_journal_get_write_access(handle, bh);
6189         if (err) {
6190                 brelse(bh);
6191                 return err;
6192         }
6193         lock_buffer(bh);
6194         memcpy(bh->b_data+offset, data, len);
6195         flush_dcache_page(bh->b_page);
6196         unlock_buffer(bh);
6197         err = ext4_handle_dirty_metadata(handle, NULL, bh);
6198         brelse(bh);
6199 out:
6200         if (inode->i_size < off + len) {
6201                 i_size_write(inode, off + len);
6202                 EXT4_I(inode)->i_disksize = inode->i_size;
6203                 ext4_mark_inode_dirty(handle, inode);
6204         }
6205         return len;
6206 }
6207
6208 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
6209 {
6210         const struct quota_format_ops   *ops;
6211
6212         if (!sb_has_quota_loaded(sb, qid->type))
6213                 return -ESRCH;
6214         ops = sb_dqopt(sb)->ops[qid->type];
6215         if (!ops || !ops->get_next_id)
6216                 return -ENOSYS;
6217         return dquot_get_next_id(sb, qid);
6218 }
6219 #endif
6220
6221 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
6222                        const char *dev_name, void *data)
6223 {
6224         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
6225 }
6226
6227 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
6228 static inline void register_as_ext2(void)
6229 {
6230         int err = register_filesystem(&ext2_fs_type);
6231         if (err)
6232                 printk(KERN_WARNING
6233                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
6234 }
6235
6236 static inline void unregister_as_ext2(void)
6237 {
6238         unregister_filesystem(&ext2_fs_type);
6239 }
6240
6241 static inline int ext2_feature_set_ok(struct super_block *sb)
6242 {
6243         if (ext4_has_unknown_ext2_incompat_features(sb))
6244                 return 0;
6245         if (sb_rdonly(sb))
6246                 return 1;
6247         if (ext4_has_unknown_ext2_ro_compat_features(sb))
6248                 return 0;
6249         return 1;
6250 }
6251 #else
6252 static inline void register_as_ext2(void) { }
6253 static inline void unregister_as_ext2(void) { }
6254 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
6255 #endif
6256
6257 static inline void register_as_ext3(void)
6258 {
6259         int err = register_filesystem(&ext3_fs_type);
6260         if (err)
6261                 printk(KERN_WARNING
6262                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
6263 }
6264
6265 static inline void unregister_as_ext3(void)
6266 {
6267         unregister_filesystem(&ext3_fs_type);
6268 }
6269
6270 static inline int ext3_feature_set_ok(struct super_block *sb)
6271 {
6272         if (ext4_has_unknown_ext3_incompat_features(sb))
6273                 return 0;
6274         if (!ext4_has_feature_journal(sb))
6275                 return 0;
6276         if (sb_rdonly(sb))
6277                 return 1;
6278         if (ext4_has_unknown_ext3_ro_compat_features(sb))
6279                 return 0;
6280         return 1;
6281 }
6282
6283 static struct file_system_type ext4_fs_type = {
6284         .owner          = THIS_MODULE,
6285         .name           = "ext4",
6286         .mount          = ext4_mount,
6287         .kill_sb        = kill_block_super,
6288         .fs_flags       = FS_REQUIRES_DEV,
6289 };
6290 MODULE_ALIAS_FS("ext4");
6291
6292 /* Shared across all ext4 file systems */
6293 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
6294
6295 static int __init ext4_init_fs(void)
6296 {
6297         int i, err;
6298
6299         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
6300         ext4_li_info = NULL;
6301         mutex_init(&ext4_li_mtx);
6302
6303         /* Build-time check for flags consistency */
6304         ext4_check_flag_values();
6305
6306         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
6307                 init_waitqueue_head(&ext4__ioend_wq[i]);
6308
6309         err = ext4_init_es();
6310         if (err)
6311                 return err;
6312
6313         err = ext4_init_pending();
6314         if (err)
6315                 goto out7;
6316
6317         err = ext4_init_post_read_processing();
6318         if (err)
6319                 goto out6;
6320
6321         err = ext4_init_pageio();
6322         if (err)
6323                 goto out5;
6324
6325         err = ext4_init_system_zone();
6326         if (err)
6327                 goto out4;
6328
6329         err = ext4_init_sysfs();
6330         if (err)
6331                 goto out3;
6332
6333         err = ext4_init_mballoc();
6334         if (err)
6335                 goto out2;
6336         err = init_inodecache();
6337         if (err)
6338                 goto out1;
6339         register_as_ext3();
6340         register_as_ext2();
6341         err = register_filesystem(&ext4_fs_type);
6342         if (err)
6343                 goto out;
6344
6345         return 0;
6346 out:
6347         unregister_as_ext2();
6348         unregister_as_ext3();
6349         destroy_inodecache();
6350 out1:
6351         ext4_exit_mballoc();
6352 out2:
6353         ext4_exit_sysfs();
6354 out3:
6355         ext4_exit_system_zone();
6356 out4:
6357         ext4_exit_pageio();
6358 out5:
6359         ext4_exit_post_read_processing();
6360 out6:
6361         ext4_exit_pending();
6362 out7:
6363         ext4_exit_es();
6364
6365         return err;
6366 }
6367
6368 static void __exit ext4_exit_fs(void)
6369 {
6370         ext4_destroy_lazyinit_thread();
6371         unregister_as_ext2();
6372         unregister_as_ext3();
6373         unregister_filesystem(&ext4_fs_type);
6374         destroy_inodecache();
6375         ext4_exit_mballoc();
6376         ext4_exit_sysfs();
6377         ext4_exit_system_zone();
6378         ext4_exit_pageio();
6379         ext4_exit_post_read_processing();
6380         ext4_exit_es();
6381         ext4_exit_pending();
6382 }
6383
6384 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6385 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6386 MODULE_LICENSE("GPL");
6387 MODULE_SOFTDEP("pre: crc32c");
6388 module_init(ext4_init_fs)
6389 module_exit(ext4_exit_fs)