fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  84 static struct file_system_type ext2_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext2",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 MODULE_ALIAS_FS("ext2");
  92 MODULE_ALIAS("ext2");
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 static struct file_system_type ext3_fs_type = {
 100         .owner          = THIS_MODULE,
 101         .name           = "ext3",
 102         .mount          = ext4_mount,
 103         .kill_sb        = kill_block_super,
 104         .fs_flags       = FS_REQUIRES_DEV,
 105 };
 106 MODULE_ALIAS_FS("ext3");
 107 MODULE_ALIAS("ext3");
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109
 110 static int ext4_verify_csum_type(struct super_block *sb,
 111                                  struct ext4_super_block *es)
 112 {
 113         if (!ext4_has_feature_metadata_csum(sb))
 114                 return 1;
 115
 116         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 117 }
 118
 119 static __le32 ext4_superblock_csum(struct super_block *sb,
 120                                    struct ext4_super_block *es)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         int offset = offsetof(struct ext4_super_block, s_checksum);
 124         __u32 csum;
 125
 126         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_superblock_csum_verify(struct super_block *sb,
 132                                        struct ext4_super_block *es)
 133 {
 134         if (!ext4_has_metadata_csum(sb))
 135                 return 1;
 136
 137         return es->s_checksum == ext4_superblock_csum(sb, es);
 138 }
 139
 140 void ext4_superblock_csum_set(struct super_block *sb)
 141 {
 142         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 143
 144         if (!ext4_has_metadata_csum(sb))
 145                 return;
 146
 147         es->s_checksum = ext4_superblock_csum(sb, es);
 148 }
 149
 150 void *ext4_kvmalloc(size_t size, gfp_t flags)
 151 {
 152         void *ret;
 153
 154         ret = kmalloc(size, flags | __GFP_NOWARN);
 155         if (!ret)
 156                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 157         return ret;
 158 }
 159
 160 void *ext4_kvzalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kzalloc(size, flags | __GFP_NOWARN);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 171                                struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 176 }
 177
 178 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_table_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 192 }
 193
 194 __u32 ext4_free_group_clusters(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_free_inodes_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_used_dirs_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_itable_unused_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_itable_unused_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 224 }
 225
 226 void ext4_block_bitmap_set(struct super_block *sb,
 227                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_inode_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_table_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_free_group_clusters_set(struct super_block *sb,
 251                                   struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_free_inodes_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_used_dirs_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_itable_unused_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282
 283 static void __save_error_info(struct super_block *sb, const char *func,
 284                             unsigned int line)
 285 {
 286         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 287
 288         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 289         if (bdev_read_only(sb->s_bdev))
 290                 return;
 291         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 292         es->s_last_error_time = cpu_to_le32(get_seconds());
 293         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 294         es->s_last_error_line = cpu_to_le32(line);
 295         if (!es->s_first_error_time) {
 296                 es->s_first_error_time = es->s_last_error_time;
 297                 strncpy(es->s_first_error_func, func,
 298                         sizeof(es->s_first_error_func));
 299                 es->s_first_error_line = cpu_to_le32(line);
 300                 es->s_first_error_ino = es->s_last_error_ino;
 301                 es->s_first_error_block = es->s_last_error_block;
 302         }
 303         /*
 304          * Start the daily error reporting function if it hasn't been
 305          * started already
 306          */
 307         if (!es->s_error_count)
 308                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 309         le32_add_cpu(&es->s_error_count, 1);
 310 }
 311
 312 static void save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         __save_error_info(sb, func, line);
 316         ext4_commit_super(sb, 1);
 317 }
 318
 319 /*
 320  * The del_gendisk() function uninitializes the disk-specific data
 321  * structures, including the bdi structure, without telling anyone
 322  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 323  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 324  * This is a kludge to prevent these oops until we can put in a proper
 325  * hook in del_gendisk() to inform the VFS and file system layers.
 326  */
 327 static int block_device_ejected(struct super_block *sb)
 328 {
 329         struct inode *bd_inode = sb->s_bdev->bd_inode;
 330         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 331
 332         return bdi->dev == NULL;
 333 }
 334
 335 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 336 {
 337         struct super_block              *sb = journal->j_private;
 338         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 339         int                             error = is_journal_aborted(journal);
 340         struct ext4_journal_cb_entry    *jce;
 341
 342         BUG_ON(txn->t_state == T_FINISHED);
 343         spin_lock(&sbi->s_md_lock);
 344         while (!list_empty(&txn->t_private_list)) {
 345                 jce = list_entry(txn->t_private_list.next,
 346                                  struct ext4_journal_cb_entry, jce_list);
 347                 list_del_init(&jce->jce_list);
 348                 spin_unlock(&sbi->s_md_lock);
 349                 jce->jce_func(sb, jce, error);
 350                 spin_lock(&sbi->s_md_lock);
 351         }
 352         spin_unlock(&sbi->s_md_lock);
 353 }
 354
 355 /* Deal with the reporting of failure conditions on a filesystem such as
 356  * inconsistencies detected or read IO failures.
 357  *
 358  * On ext2, we can store the error state of the filesystem in the
 359  * superblock.  That is not possible on ext4, because we may have other
 360  * write ordering constraints on the superblock which prevent us from
 361  * writing it out straight away; and given that the journal is about to
 362  * be aborted, we can't rely on the current, or future, transactions to
 363  * write out the superblock safely.
 364  *
 365  * We'll just use the jbd2_journal_abort() error code to record an error in
 366  * the journal instead.  On recovery, the journal will complain about
 367  * that error until we've noted it down and cleared it.
 368  */
 369
 370 static void ext4_handle_error(struct super_block *sb)
 371 {
 372         if (sb->s_flags & MS_RDONLY)
 373                 return;
 374
 375         if (!test_opt(sb, ERRORS_CONT)) {
 376                 journal_t *journal = EXT4_SB(sb)->s_journal;
 377
 378                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 379                 if (journal)
 380                         jbd2_journal_abort(journal, -EIO);
 381         }
 382         if (test_opt(sb, ERRORS_RO)) {
 383                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 384                 /*
 385                  * Make sure updated value of ->s_mount_flags will be visible
 386                  * before ->s_flags update
 387                  */
 388                 smp_wmb();
 389                 sb->s_flags |= MS_RDONLY;
 390         }
 391         if (test_opt(sb, ERRORS_PANIC)) {
 392                 if (EXT4_SB(sb)->s_journal &&
 393                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 394                         return;
 395                 panic("EXT4-fs (device %s): panic forced after error\n",
 396                         sb->s_id);
 397         }
 398 }
 399
 400 #define ext4_error_ratelimit(sb)                                        \
 401                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 402                              "EXT4-fs error")
 403
 404 void __ext4_error(struct super_block *sb, const char *function,
 405                   unsigned int line, const char *fmt, ...)
 406 {
 407         struct va_format vaf;
 408         va_list args;
 409
 410         if (ext4_error_ratelimit(sb)) {
 411                 va_start(args, fmt);
 412                 vaf.fmt = fmt;
 413                 vaf.va = &args;
 414                 printk(KERN_CRIT
 415                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 416                        sb->s_id, function, line, current->comm, &vaf);
 417                 va_end(args);
 418         }
 419         save_error_info(sb, function, line);
 420         ext4_handle_error(sb);
 421 }
 422
 423 void __ext4_error_inode(struct inode *inode, const char *function,
 424                         unsigned int line, ext4_fsblk_t block,
 425                         const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         if (ext4_error_ratelimit(inode->i_sb)) {
 434                 va_start(args, fmt);
 435                 vaf.fmt = fmt;
 436                 vaf.va = &args;
 437                 if (block)
 438                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 439                                "inode #%lu: block %llu: comm %s: %pV\n",
 440                                inode->i_sb->s_id, function, line, inode->i_ino,
 441                                block, current->comm, &vaf);
 442                 else
 443                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 444                                "inode #%lu: comm %s: %pV\n",
 445                                inode->i_sb->s_id, function, line, inode->i_ino,
 446                                current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(inode->i_sb, function, line);
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 void __ext4_error_file(struct file *file, const char *function,
 454                        unsigned int line, ext4_fsblk_t block,
 455                        const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es;
 460         struct inode *inode = file_inode(file);
 461         char pathname[80], *path;
 462
 463         es = EXT4_SB(inode->i_sb)->s_es;
 464         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 465         if (ext4_error_ratelimit(inode->i_sb)) {
 466                 path = file_path(file, pathname, sizeof(pathname));
 467                 if (IS_ERR(path))
 468                         path = "(unknown)";
 469                 va_start(args, fmt);
 470                 vaf.fmt = fmt;
 471                 vaf.va = &args;
 472                 if (block)
 473                         printk(KERN_CRIT
 474                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 475                                "block %llu: comm %s: path %s: %pV\n",
 476                                inode->i_sb->s_id, function, line, inode->i_ino,
 477                                block, current->comm, path, &vaf);
 478                 else
 479                         printk(KERN_CRIT
 480                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 481                                "comm %s: path %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                current->comm, path, &vaf);
 484                 va_end(args);
 485         }
 486         save_error_info(inode->i_sb, function, line);
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EFSCORRUPTED:
 497                 errstr = "Corrupt filesystem";
 498                 break;
 499         case -EFSBADCRC:
 500                 errstr = "Filesystem failed CRC";
 501                 break;
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         if (ext4_error_ratelimit(sb)) {
 547                 errstr = ext4_decode_error(sb, errno, nbuf);
 548                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 549                        sb->s_id, function, line, errstr);
 550         }
 551
 552         save_error_info(sb, function, line);
 553         ext4_handle_error(sb);
 554 }
 555
 556 /*
 557  * ext4_abort is a much stronger failure handler than ext4_error.  The
 558  * abort function may be used to deal with unrecoverable failures such
 559  * as journal IO errors or ENOMEM at a critical moment in log management.
 560  *
 561  * We unconditionally force the filesystem into an ABORT|READONLY state,
 562  * unless the error response on the fs has been set to panic in which
 563  * case we take the easy way out and panic immediately.
 564  */
 565
 566 void __ext4_abort(struct super_block *sb, const char *function,
 567                 unsigned int line, const char *fmt, ...)
 568 {
 569         va_list args;
 570
 571         save_error_info(sb, function, line);
 572         va_start(args, fmt);
 573         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 574                function, line);
 575         vprintk(fmt, args);
 576         printk("\n");
 577         va_end(args);
 578
 579         if ((sb->s_flags & MS_RDONLY) == 0) {
 580                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 /*
 583                  * Make sure updated value of ->s_mount_flags will be visible
 584                  * before ->s_flags update
 585                  */
 586                 smp_wmb();
 587                 sb->s_flags |= MS_RDONLY;
 588                 if (EXT4_SB(sb)->s_journal)
 589                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 590                 save_error_info(sb, function, line);
 591         }
 592         if (test_opt(sb, ERRORS_PANIC)) {
 593                 if (EXT4_SB(sb)->s_journal &&
 594                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 595                         return;
 596                 panic("EXT4-fs panic from previous error\n");
 597         }
 598 }
 599
 600 void __ext4_msg(struct super_block *sb,
 601                 const char *prefix, const char *fmt, ...)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605
 606         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 607                 return;
 608
 609         va_start(args, fmt);
 610         vaf.fmt = fmt;
 611         vaf.va = &args;
 612         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 613         va_end(args);
 614 }
 615
 616 #define ext4_warning_ratelimit(sb)                                      \
 617                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 618                              "EXT4-fs warning")
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!ext4_warning_ratelimit(sb))
 627                 return;
 628
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635 }
 636
 637 void __ext4_warning_inode(const struct inode *inode, const char *function,
 638                           unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (!ext4_warning_ratelimit(inode->i_sb))
 644                 return;
 645
 646         va_start(args, fmt);
 647         vaf.fmt = fmt;
 648         vaf.va = &args;
 649         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 650                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 651                function, line, inode->i_ino, current->comm, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_grp_locked_error(const char *function, unsigned int line,
 656                              struct super_block *sb, ext4_group_t grp,
 657                              unsigned long ino, ext4_fsblk_t block,
 658                              const char *fmt, ...)
 659 __releases(bitlock)
 660 __acquires(bitlock)
 661 {
 662         struct va_format vaf;
 663         va_list args;
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         es->s_last_error_ino = cpu_to_le32(ino);
 667         es->s_last_error_block = cpu_to_le64(block);
 668         __save_error_info(sb, function, line);
 669
 670         if (ext4_error_ratelimit(sb)) {
 671                 va_start(args, fmt);
 672                 vaf.fmt = fmt;
 673                 vaf.va = &args;
 674                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 675                        sb->s_id, function, line, grp);
 676                 if (ino)
 677                         printk(KERN_CONT "inode %lu: ", ino);
 678                 if (block)
 679                         printk(KERN_CONT "block %llu:",
 680                                (unsigned long long) block);
 681                 printk(KERN_CONT "%pV\n", &vaf);
 682                 va_end(args);
 683         }
 684
 685         if (test_opt(sb, ERRORS_CONT)) {
 686                 ext4_commit_super(sb, 0);
 687                 return;
 688         }
 689
 690         ext4_unlock_group(sb, grp);
 691         ext4_handle_error(sb);
 692         /*
 693          * We only get here in the ERRORS_RO case; relocking the group
 694          * may be dangerous, but nothing bad will happen since the
 695          * filesystem will have already been marked read/only and the
 696          * journal has been aborted.  We return 1 as a hint to callers
 697          * who might what to use the return value from
 698          * ext4_grp_locked_error() to distinguish between the
 699          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 700          * aggressively from the ext4 function in question, with a
 701          * more appropriate error code.
 702          */
 703         ext4_lock_group(sb, grp);
 704         return;
 705 }
 706
 707 void ext4_update_dynamic_rev(struct super_block *sb)
 708 {
 709         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 710
 711         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 712                 return;
 713
 714         ext4_warning(sb,
 715                      "updating to rev %d because of new feature flag, "
 716                      "running e2fsck is recommended",
 717                      EXT4_DYNAMIC_REV);
 718
 719         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 720         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 721         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 722         /* leave es->s_feature_*compat flags alone */
 723         /* es->s_uuid will be set by e2fsck if empty */
 724
 725         /*
 726          * The rest of the superblock fields should be zero, and if not it
 727          * means they are likely already in use, so leave them alone.  We
 728          * can leave it up to e2fsck to clean up any inconsistencies there.
 729          */
 730 }
 731
 732 /*
 733  * Open the external journal device
 734  */
 735 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 736 {
 737         struct block_device *bdev;
 738         char b[BDEVNAME_SIZE];
 739
 740         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 741         if (IS_ERR(bdev))
 742                 goto fail;
 743         return bdev;
 744
 745 fail:
 746         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 747                         __bdevname(dev, b), PTR_ERR(bdev));
 748         return NULL;
 749 }
 750
 751 /*
 752  * Release the journal device
 753  */
 754 static void ext4_blkdev_put(struct block_device *bdev)
 755 {
 756         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 757 }
 758
 759 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 760 {
 761         struct block_device *bdev;
 762         bdev = sbi->journal_bdev;
 763         if (bdev) {
 764                 ext4_blkdev_put(bdev);
 765                 sbi->journal_bdev = NULL;
 766         }
 767 }
 768
 769 static inline struct inode *orphan_list_entry(struct list_head *l)
 770 {
 771         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 772 }
 773
 774 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 775 {
 776         struct list_head *l;
 777
 778         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 779                  le32_to_cpu(sbi->s_es->s_last_orphan));
 780
 781         printk(KERN_ERR "sb_info orphan list:\n");
 782         list_for_each(l, &sbi->s_orphan) {
 783                 struct inode *inode = orphan_list_entry(l);
 784                 printk(KERN_ERR "  "
 785                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 786                        inode->i_sb->s_id, inode->i_ino, inode,
 787                        inode->i_mode, inode->i_nlink,
 788                        NEXT_ORPHAN(inode));
 789         }
 790 }
 791
 792 static void ext4_put_super(struct super_block *sb)
 793 {
 794         struct ext4_sb_info *sbi = EXT4_SB(sb);
 795         struct ext4_super_block *es = sbi->s_es;
 796         int aborted = 0;
 797         int i, err;
 798
 799         ext4_unregister_li_request(sb);
 800         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 801
 802         flush_workqueue(sbi->rsv_conversion_wq);
 803         destroy_workqueue(sbi->rsv_conversion_wq);
 804
 805         if (sbi->s_journal) {
 806                 aborted = is_journal_aborted(sbi->s_journal);
 807                 err = jbd2_journal_destroy(sbi->s_journal);
 808                 sbi->s_journal = NULL;
 809                 if ((err < 0) && !aborted)
 810                         ext4_abort(sb, "Couldn't clean up the journal");
 811         }
 812
 813         ext4_unregister_sysfs(sb);
 814         ext4_es_unregister_shrinker(sbi);
 815         del_timer_sync(&sbi->s_err_report);
 816         ext4_release_system_zone(sb);
 817         ext4_mb_release(sb);
 818         ext4_ext_release(sb);
 819
 820         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 821                 ext4_clear_feature_journal_needs_recovery(sb);
 822                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 823         }
 824         if (!(sb->s_flags & MS_RDONLY))
 825                 ext4_commit_super(sb, 1);
 826
 827         for (i = 0; i < sbi->s_gdb_count; i++)
 828                 brelse(sbi->s_group_desc[i]);
 829         kvfree(sbi->s_group_desc);
 830         kvfree(sbi->s_flex_groups);
 831         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 832         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 833         percpu_counter_destroy(&sbi->s_dirs_counter);
 834         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 835         brelse(sbi->s_sbh);
 836 #ifdef CONFIG_QUOTA
 837         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 838                 kfree(sbi->s_qf_names[i]);
 839 #endif
 840
 841         /* Debugging code just in case the in-memory inode orphan list
 842          * isn't empty.  The on-disk one can be non-empty if we've
 843          * detected an error and taken the fs readonly, but the
 844          * in-memory list had better be clean by this point. */
 845         if (!list_empty(&sbi->s_orphan))
 846                 dump_orphan_list(sb, sbi);
 847         J_ASSERT(list_empty(&sbi->s_orphan));
 848
 849         sync_blockdev(sb->s_bdev);
 850         invalidate_bdev(sb->s_bdev);
 851         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 852                 /*
 853                  * Invalidate the journal device's buffers.  We don't want them
 854                  * floating about in memory - the physical journal device may
 855                  * hotswapped, and it breaks the `ro-after' testing code.
 856                  */
 857                 sync_blockdev(sbi->journal_bdev);
 858                 invalidate_bdev(sbi->journal_bdev);
 859                 ext4_blkdev_remove(sbi);
 860         }
 861         if (sbi->s_mb_cache) {
 862                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 863                 sbi->s_mb_cache = NULL;
 864         }
 865         if (sbi->s_mmp_tsk)
 866                 kthread_stop(sbi->s_mmp_tsk);
 867         sb->s_fs_info = NULL;
 868         /*
 869          * Now that we are completely done shutting down the
 870          * superblock, we need to actually destroy the kobject.
 871          */
 872         kobject_put(&sbi->s_kobj);
 873         wait_for_completion(&sbi->s_kobj_unregister);
 874         if (sbi->s_chksum_driver)
 875                 crypto_free_shash(sbi->s_chksum_driver);
 876         kfree(sbi->s_blockgroup_lock);
 877         kfree(sbi);
 878 }
 879
 880 static struct kmem_cache *ext4_inode_cachep;
 881
 882 /*
 883  * Called inside transaction, so use GFP_NOFS
 884  */
 885 static struct inode *ext4_alloc_inode(struct super_block *sb)
 886 {
 887         struct ext4_inode_info *ei;
 888
 889         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 890         if (!ei)
 891                 return NULL;
 892
 893         ei->vfs_inode.i_version = 1;
 894         spin_lock_init(&ei->i_raw_lock);
 895         INIT_LIST_HEAD(&ei->i_prealloc_list);
 896         spin_lock_init(&ei->i_prealloc_lock);
 897         ext4_es_init_tree(&ei->i_es_tree);
 898         rwlock_init(&ei->i_es_lock);
 899         INIT_LIST_HEAD(&ei->i_es_list);
 900         ei->i_es_all_nr = 0;
 901         ei->i_es_shk_nr = 0;
 902         ei->i_es_shrink_lblk = 0;
 903         ei->i_reserved_data_blocks = 0;
 904         ei->i_reserved_meta_blocks = 0;
 905         ei->i_allocated_meta_blocks = 0;
 906         ei->i_da_metadata_calc_len = 0;
 907         ei->i_da_metadata_calc_last_lblock = 0;
 908         spin_lock_init(&(ei->i_block_reservation_lock));
 909 #ifdef CONFIG_QUOTA
 910         ei->i_reserved_quota = 0;
 911         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 912 #endif
 913         ei->jinode = NULL;
 914         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 915         spin_lock_init(&ei->i_completed_io_lock);
 916         ei->i_sync_tid = 0;
 917         ei->i_datasync_tid = 0;
 918         atomic_set(&ei->i_ioend_count, 0);
 919         atomic_set(&ei->i_unwritten, 0);
 920         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 921 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 922         ei->i_crypt_info = NULL;
 923 #endif
 924         return &ei->vfs_inode;
 925 }
 926
 927 static int ext4_drop_inode(struct inode *inode)
 928 {
 929         int drop = generic_drop_inode(inode);
 930
 931         trace_ext4_drop_inode(inode, drop);
 932         return drop;
 933 }
 934
 935 static void ext4_i_callback(struct rcu_head *head)
 936 {
 937         struct inode *inode = container_of(head, struct inode, i_rcu);
 938         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 939 }
 940
 941 static void ext4_destroy_inode(struct inode *inode)
 942 {
 943         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 944                 ext4_msg(inode->i_sb, KERN_ERR,
 945                          "Inode %lu (%p): orphan list check failed!",
 946                          inode->i_ino, EXT4_I(inode));
 947                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 948                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 949                                 true);
 950                 dump_stack();
 951         }
 952         call_rcu(&inode->i_rcu, ext4_i_callback);
 953 }
 954
 955 static void init_once(void *foo)
 956 {
 957         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 958
 959         INIT_LIST_HEAD(&ei->i_orphan);
 960         init_rwsem(&ei->xattr_sem);
 961         init_rwsem(&ei->i_data_sem);
 962         init_rwsem(&ei->i_mmap_sem);
 963         inode_init_once(&ei->vfs_inode);
 964 }
 965
 966 static int __init init_inodecache(void)
 967 {
 968         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 969                                              sizeof(struct ext4_inode_info),
 970                                              0, (SLAB_RECLAIM_ACCOUNT|
 971                                                 SLAB_MEM_SPREAD),
 972                                              init_once);
 973         if (ext4_inode_cachep == NULL)
 974                 return -ENOMEM;
 975         return 0;
 976 }
 977
 978 static void destroy_inodecache(void)
 979 {
 980         /*
 981          * Make sure all delayed rcu free inodes are flushed before we
 982          * destroy cache.
 983          */
 984         rcu_barrier();
 985         kmem_cache_destroy(ext4_inode_cachep);
 986 }
 987
 988 void ext4_clear_inode(struct inode *inode)
 989 {
 990         invalidate_inode_buffers(inode);
 991         clear_inode(inode);
 992         dquot_drop(inode);
 993         ext4_discard_preallocations(inode);
 994         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 995         if (EXT4_I(inode)->jinode) {
 996                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 997                                                EXT4_I(inode)->jinode);
 998                 jbd2_free_inode(EXT4_I(inode)->jinode);
 999                 EXT4_I(inode)->jinode = NULL;
1000         }
1001 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1002         if (EXT4_I(inode)->i_crypt_info)
1003                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1004 #endif
1005 }
1006
1007 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1008                                         u64 ino, u32 generation)
1009 {
1010         struct inode *inode;
1011
1012         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1013                 return ERR_PTR(-ESTALE);
1014         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1015                 return ERR_PTR(-ESTALE);
1016
1017         /* iget isn't really right if the inode is currently unallocated!!
1018          *
1019          * ext4_read_inode will return a bad_inode if the inode had been
1020          * deleted, so we should be safe.
1021          *
1022          * Currently we don't know the generation for parent directory, so
1023          * a generation of 0 means "accept any"
1024          */
1025         inode = ext4_iget_normal(sb, ino);
1026         if (IS_ERR(inode))
1027                 return ERR_CAST(inode);
1028         if (generation && inode->i_generation != generation) {
1029                 iput(inode);
1030                 return ERR_PTR(-ESTALE);
1031         }
1032
1033         return inode;
1034 }
1035
1036 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1037                                         int fh_len, int fh_type)
1038 {
1039         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1040                                     ext4_nfs_get_inode);
1041 }
1042
1043 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1044                                         int fh_len, int fh_type)
1045 {
1046         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1047                                     ext4_nfs_get_inode);
1048 }
1049
1050 /*
1051  * Try to release metadata pages (indirect blocks, directories) which are
1052  * mapped via the block device.  Since these pages could have journal heads
1053  * which would prevent try_to_free_buffers() from freeing them, we must use
1054  * jbd2 layer's try_to_free_buffers() function to release them.
1055  */
1056 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1057                                  gfp_t wait)
1058 {
1059         journal_t *journal = EXT4_SB(sb)->s_journal;
1060
1061         WARN_ON(PageChecked(page));
1062         if (!page_has_buffers(page))
1063                 return 0;
1064         if (journal)
1065                 return jbd2_journal_try_to_free_buffers(journal, page,
1066                                                 wait & ~__GFP_DIRECT_RECLAIM);
1067         return try_to_free_buffers(page);
1068 }
1069
1070 #ifdef CONFIG_QUOTA
1071 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1072 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1073
1074 static int ext4_write_dquot(struct dquot *dquot);
1075 static int ext4_acquire_dquot(struct dquot *dquot);
1076 static int ext4_release_dquot(struct dquot *dquot);
1077 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1078 static int ext4_write_info(struct super_block *sb, int type);
1079 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1080                          struct path *path);
1081 static int ext4_quota_off(struct super_block *sb, int type);
1082 static int ext4_quota_on_mount(struct super_block *sb, int type);
1083 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1084                                size_t len, loff_t off);
1085 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1086                                 const char *data, size_t len, loff_t off);
1087 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1088                              unsigned int flags);
1089 static int ext4_enable_quotas(struct super_block *sb);
1090
1091 static struct dquot **ext4_get_dquots(struct inode *inode)
1092 {
1093         return EXT4_I(inode)->i_dquot;
1094 }
1095
1096 static const struct dquot_operations ext4_quota_operations = {
1097         .get_reserved_space = ext4_get_reserved_space,
1098         .write_dquot    = ext4_write_dquot,
1099         .acquire_dquot  = ext4_acquire_dquot,
1100         .release_dquot  = ext4_release_dquot,
1101         .mark_dirty     = ext4_mark_dquot_dirty,
1102         .write_info     = ext4_write_info,
1103         .alloc_dquot    = dquot_alloc,
1104         .destroy_dquot  = dquot_destroy,
1105 };
1106
1107 static const struct quotactl_ops ext4_qctl_operations = {
1108         .quota_on       = ext4_quota_on,
1109         .quota_off      = ext4_quota_off,
1110         .quota_sync     = dquot_quota_sync,
1111         .get_state      = dquot_get_state,
1112         .set_info       = dquot_set_dqinfo,
1113         .get_dqblk      = dquot_get_dqblk,
1114         .set_dqblk      = dquot_set_dqblk
1115 };
1116 #endif
1117
1118 static const struct super_operations ext4_sops = {
1119         .alloc_inode    = ext4_alloc_inode,
1120         .destroy_inode  = ext4_destroy_inode,
1121         .write_inode    = ext4_write_inode,
1122         .dirty_inode    = ext4_dirty_inode,
1123         .drop_inode     = ext4_drop_inode,
1124         .evict_inode    = ext4_evict_inode,
1125         .put_super      = ext4_put_super,
1126         .sync_fs        = ext4_sync_fs,
1127         .freeze_fs      = ext4_freeze,
1128         .unfreeze_fs    = ext4_unfreeze,
1129         .statfs         = ext4_statfs,
1130         .remount_fs     = ext4_remount,
1131         .show_options   = ext4_show_options,
1132 #ifdef CONFIG_QUOTA
1133         .quota_read     = ext4_quota_read,
1134         .quota_write    = ext4_quota_write,
1135         .get_dquots     = ext4_get_dquots,
1136 #endif
1137         .bdev_try_to_free_page = bdev_try_to_free_page,
1138 };
1139
1140 static const struct export_operations ext4_export_ops = {
1141         .fh_to_dentry = ext4_fh_to_dentry,
1142         .fh_to_parent = ext4_fh_to_parent,
1143         .get_parent = ext4_get_parent,
1144 };
1145
1146 enum {
1147         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1148         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1149         Opt_nouid32, Opt_debug, Opt_removed,
1150         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1151         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1152         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1153         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1154         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1155         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1156         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1157         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1158         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1159         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1160         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1161         Opt_lazytime, Opt_nolazytime,
1162         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1163         Opt_inode_readahead_blks, Opt_journal_ioprio,
1164         Opt_dioread_nolock, Opt_dioread_lock,
1165         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1166         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1167 };
1168
1169 static const match_table_t tokens = {
1170         {Opt_bsd_df, "bsddf"},
1171         {Opt_minix_df, "minixdf"},
1172         {Opt_grpid, "grpid"},
1173         {Opt_grpid, "bsdgroups"},
1174         {Opt_nogrpid, "nogrpid"},
1175         {Opt_nogrpid, "sysvgroups"},
1176         {Opt_resgid, "resgid=%u"},
1177         {Opt_resuid, "resuid=%u"},
1178         {Opt_sb, "sb=%u"},
1179         {Opt_err_cont, "errors=continue"},
1180         {Opt_err_panic, "errors=panic"},
1181         {Opt_err_ro, "errors=remount-ro"},
1182         {Opt_nouid32, "nouid32"},
1183         {Opt_debug, "debug"},
1184         {Opt_removed, "oldalloc"},
1185         {Opt_removed, "orlov"},
1186         {Opt_user_xattr, "user_xattr"},
1187         {Opt_nouser_xattr, "nouser_xattr"},
1188         {Opt_acl, "acl"},
1189         {Opt_noacl, "noacl"},
1190         {Opt_noload, "norecovery"},
1191         {Opt_noload, "noload"},
1192         {Opt_removed, "nobh"},
1193         {Opt_removed, "bh"},
1194         {Opt_commit, "commit=%u"},
1195         {Opt_min_batch_time, "min_batch_time=%u"},
1196         {Opt_max_batch_time, "max_batch_time=%u"},
1197         {Opt_journal_dev, "journal_dev=%u"},
1198         {Opt_journal_path, "journal_path=%s"},
1199         {Opt_journal_checksum, "journal_checksum"},
1200         {Opt_nojournal_checksum, "nojournal_checksum"},
1201         {Opt_journal_async_commit, "journal_async_commit"},
1202         {Opt_abort, "abort"},
1203         {Opt_data_journal, "data=journal"},
1204         {Opt_data_ordered, "data=ordered"},
1205         {Opt_data_writeback, "data=writeback"},
1206         {Opt_data_err_abort, "data_err=abort"},
1207         {Opt_data_err_ignore, "data_err=ignore"},
1208         {Opt_offusrjquota, "usrjquota="},
1209         {Opt_usrjquota, "usrjquota=%s"},
1210         {Opt_offgrpjquota, "grpjquota="},
1211         {Opt_grpjquota, "grpjquota=%s"},
1212         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1213         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1214         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1215         {Opt_grpquota, "grpquota"},
1216         {Opt_noquota, "noquota"},
1217         {Opt_quota, "quota"},
1218         {Opt_usrquota, "usrquota"},
1219         {Opt_barrier, "barrier=%u"},
1220         {Opt_barrier, "barrier"},
1221         {Opt_nobarrier, "nobarrier"},
1222         {Opt_i_version, "i_version"},
1223         {Opt_dax, "dax"},
1224         {Opt_stripe, "stripe=%u"},
1225         {Opt_delalloc, "delalloc"},
1226         {Opt_lazytime, "lazytime"},
1227         {Opt_nolazytime, "nolazytime"},
1228         {Opt_nodelalloc, "nodelalloc"},
1229         {Opt_removed, "mblk_io_submit"},
1230         {Opt_removed, "nomblk_io_submit"},
1231         {Opt_block_validity, "block_validity"},
1232         {Opt_noblock_validity, "noblock_validity"},
1233         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1234         {Opt_journal_ioprio, "journal_ioprio=%u"},
1235         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc"},
1237         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1238         {Opt_dioread_nolock, "dioread_nolock"},
1239         {Opt_dioread_lock, "dioread_lock"},
1240         {Opt_discard, "discard"},
1241         {Opt_nodiscard, "nodiscard"},
1242         {Opt_init_itable, "init_itable=%u"},
1243         {Opt_init_itable, "init_itable"},
1244         {Opt_noinit_itable, "noinit_itable"},
1245         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1246         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1247         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1248         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1249         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1250         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1251         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1252         {Opt_err, NULL},
1253 };
1254
1255 static ext4_fsblk_t get_sb_block(void **data)
1256 {
1257         ext4_fsblk_t    sb_block;
1258         char            *options = (char *) *data;
1259
1260         if (!options || strncmp(options, "sb=", 3) != 0)
1261                 return 1;       /* Default location */
1262
1263         options += 3;
1264         /* TODO: use simple_strtoll with >32bit ext4 */
1265         sb_block = simple_strtoul(options, &options, 0);
1266         if (*options && *options != ',') {
1267                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1268                        (char *) *data);
1269                 return 1;
1270         }
1271         if (*options == ',')
1272                 options++;
1273         *data = (void *) options;
1274
1275         return sb_block;
1276 }
1277
1278 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1279 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1280         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1281
1282 #ifdef CONFIG_QUOTA
1283 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1284 {
1285         struct ext4_sb_info *sbi = EXT4_SB(sb);
1286         char *qname;
1287         int ret = -1;
1288
1289         if (sb_any_quota_loaded(sb) &&
1290                 !sbi->s_qf_names[qtype]) {
1291                 ext4_msg(sb, KERN_ERR,
1292                         "Cannot change journaled "
1293                         "quota options when quota turned on");
1294                 return -1;
1295         }
1296         if (ext4_has_feature_quota(sb)) {
1297                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1298                          "ignored when QUOTA feature is enabled");
1299                 return 1;
1300         }
1301         qname = match_strdup(args);
1302         if (!qname) {
1303                 ext4_msg(sb, KERN_ERR,
1304                         "Not enough memory for storing quotafile name");
1305                 return -1;
1306         }
1307         if (sbi->s_qf_names[qtype]) {
1308                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1309                         ret = 1;
1310                 else
1311                         ext4_msg(sb, KERN_ERR,
1312                                  "%s quota file already specified",
1313                                  QTYPE2NAME(qtype));
1314                 goto errout;
1315         }
1316         if (strchr(qname, '/')) {
1317                 ext4_msg(sb, KERN_ERR,
1318                         "quotafile must be on filesystem root");
1319                 goto errout;
1320         }
1321         sbi->s_qf_names[qtype] = qname;
1322         set_opt(sb, QUOTA);
1323         return 1;
1324 errout:
1325         kfree(qname);
1326         return ret;
1327 }
1328
1329 static int clear_qf_name(struct super_block *sb, int qtype)
1330 {
1331
1332         struct ext4_sb_info *sbi = EXT4_SB(sb);
1333
1334         if (sb_any_quota_loaded(sb) &&
1335                 sbi->s_qf_names[qtype]) {
1336                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1337                         " when quota turned on");
1338                 return -1;
1339         }
1340         kfree(sbi->s_qf_names[qtype]);
1341         sbi->s_qf_names[qtype] = NULL;
1342         return 1;
1343 }
1344 #endif
1345
1346 #define MOPT_SET        0x0001
1347 #define MOPT_CLEAR      0x0002
1348 #define MOPT_NOSUPPORT  0x0004
1349 #define MOPT_EXPLICIT   0x0008
1350 #define MOPT_CLEAR_ERR  0x0010
1351 #define MOPT_GTE0       0x0020
1352 #ifdef CONFIG_QUOTA
1353 #define MOPT_Q          0
1354 #define MOPT_QFMT       0x0040
1355 #else
1356 #define MOPT_Q          MOPT_NOSUPPORT
1357 #define MOPT_QFMT       MOPT_NOSUPPORT
1358 #endif
1359 #define MOPT_DATAJ      0x0080
1360 #define MOPT_NO_EXT2    0x0100
1361 #define MOPT_NO_EXT3    0x0200
1362 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1363 #define MOPT_STRING     0x0400
1364
1365 static const struct mount_opts {
1366         int     token;
1367         int     mount_opt;
1368         int     flags;
1369 } ext4_mount_opts[] = {
1370         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1371         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1372         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1373         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1374         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1375         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1376         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1377          MOPT_EXT4_ONLY | MOPT_SET},
1378         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1379          MOPT_EXT4_ONLY | MOPT_CLEAR},
1380         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1381         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1382         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1383          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1384         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1385          MOPT_EXT4_ONLY | MOPT_CLEAR},
1386         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1387          MOPT_EXT4_ONLY | MOPT_CLEAR},
1388         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1389          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1390         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1391                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1392          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1393         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1394         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1397         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1398          MOPT_NO_EXT2 | MOPT_SET},
1399         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1400          MOPT_NO_EXT2 | MOPT_CLEAR},
1401         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1402         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1403         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1404         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1405         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1406         {Opt_commit, 0, MOPT_GTE0},
1407         {Opt_max_batch_time, 0, MOPT_GTE0},
1408         {Opt_min_batch_time, 0, MOPT_GTE0},
1409         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1410         {Opt_init_itable, 0, MOPT_GTE0},
1411         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1412         {Opt_stripe, 0, MOPT_GTE0},
1413         {Opt_resuid, 0, MOPT_GTE0},
1414         {Opt_resgid, 0, MOPT_GTE0},
1415         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1416         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1417         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1418         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1419         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1420         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1421          MOPT_NO_EXT2 | MOPT_DATAJ},
1422         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1423         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1424 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1425         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1426         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1427 #else
1428         {Opt_acl, 0, MOPT_NOSUPPORT},
1429         {Opt_noacl, 0, MOPT_NOSUPPORT},
1430 #endif
1431         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1432         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1433         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1434         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1435                                                         MOPT_SET | MOPT_Q},
1436         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1437                                                         MOPT_SET | MOPT_Q},
1438         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1439                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1440         {Opt_usrjquota, 0, MOPT_Q},
1441         {Opt_grpjquota, 0, MOPT_Q},
1442         {Opt_offusrjquota, 0, MOPT_Q},
1443         {Opt_offgrpjquota, 0, MOPT_Q},
1444         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1445         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1446         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1447         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1448         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1449         {Opt_err, 0, 0}
1450 };
1451
1452 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1453                             substring_t *args, unsigned long *journal_devnum,
1454                             unsigned int *journal_ioprio, int is_remount)
1455 {
1456         struct ext4_sb_info *sbi = EXT4_SB(sb);
1457         const struct mount_opts *m;
1458         kuid_t uid;
1459         kgid_t gid;
1460         int arg = 0;
1461
1462 #ifdef CONFIG_QUOTA
1463         if (token == Opt_usrjquota)
1464                 return set_qf_name(sb, USRQUOTA, &args[0]);
1465         else if (token == Opt_grpjquota)
1466                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1467         else if (token == Opt_offusrjquota)
1468                 return clear_qf_name(sb, USRQUOTA);
1469         else if (token == Opt_offgrpjquota)
1470                 return clear_qf_name(sb, GRPQUOTA);
1471 #endif
1472         switch (token) {
1473         case Opt_noacl:
1474         case Opt_nouser_xattr:
1475                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1476                 break;
1477         case Opt_sb:
1478                 return 1;       /* handled by get_sb_block() */
1479         case Opt_removed:
1480                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1481                 return 1;
1482         case Opt_abort:
1483                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1484                 return 1;
1485         case Opt_i_version:
1486                 sb->s_flags |= MS_I_VERSION;
1487                 return 1;
1488         case Opt_lazytime:
1489                 sb->s_flags |= MS_LAZYTIME;
1490                 return 1;
1491         case Opt_nolazytime:
1492                 sb->s_flags &= ~MS_LAZYTIME;
1493                 return 1;
1494         }
1495
1496         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1497                 if (token == m->token)
1498                         break;
1499
1500         if (m->token == Opt_err) {
1501                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1502                          "or missing value", opt);
1503                 return -1;
1504         }
1505
1506         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1507                 ext4_msg(sb, KERN_ERR,
1508                          "Mount option \"%s\" incompatible with ext2", opt);
1509                 return -1;
1510         }
1511         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1512                 ext4_msg(sb, KERN_ERR,
1513                          "Mount option \"%s\" incompatible with ext3", opt);
1514                 return -1;
1515         }
1516
1517         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1518                 return -1;
1519         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1520                 return -1;
1521         if (m->flags & MOPT_EXPLICIT) {
1522                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1523                         set_opt2(sb, EXPLICIT_DELALLOC);
1524                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1525                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1526                 } else
1527                         return -1;
1528         }
1529         if (m->flags & MOPT_CLEAR_ERR)
1530                 clear_opt(sb, ERRORS_MASK);
1531         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1532                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1533                          "options when quota turned on");
1534                 return -1;
1535         }
1536
1537         if (m->flags & MOPT_NOSUPPORT) {
1538                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1539         } else if (token == Opt_commit) {
1540                 if (arg == 0)
1541                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1542                 sbi->s_commit_interval = HZ * arg;
1543         } else if (token == Opt_max_batch_time) {
1544                 sbi->s_max_batch_time = arg;
1545         } else if (token == Opt_min_batch_time) {
1546                 sbi->s_min_batch_time = arg;
1547         } else if (token == Opt_inode_readahead_blks) {
1548                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1549                         ext4_msg(sb, KERN_ERR,
1550                                  "EXT4-fs: inode_readahead_blks must be "
1551                                  "0 or a power of 2 smaller than 2^31");
1552                         return -1;
1553                 }
1554                 sbi->s_inode_readahead_blks = arg;
1555         } else if (token == Opt_init_itable) {
1556                 set_opt(sb, INIT_INODE_TABLE);
1557                 if (!args->from)
1558                         arg = EXT4_DEF_LI_WAIT_MULT;
1559                 sbi->s_li_wait_mult = arg;
1560         } else if (token == Opt_max_dir_size_kb) {
1561                 sbi->s_max_dir_size_kb = arg;
1562         } else if (token == Opt_stripe) {
1563                 sbi->s_stripe = arg;
1564         } else if (token == Opt_resuid) {
1565                 uid = make_kuid(current_user_ns(), arg);
1566                 if (!uid_valid(uid)) {
1567                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1568                         return -1;
1569                 }
1570                 sbi->s_resuid = uid;
1571         } else if (token == Opt_resgid) {
1572                 gid = make_kgid(current_user_ns(), arg);
1573                 if (!gid_valid(gid)) {
1574                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1575                         return -1;
1576                 }
1577                 sbi->s_resgid = gid;
1578         } else if (token == Opt_journal_dev) {
1579                 if (is_remount) {
1580                         ext4_msg(sb, KERN_ERR,
1581                                  "Cannot specify journal on remount");
1582                         return -1;
1583                 }
1584                 *journal_devnum = arg;
1585         } else if (token == Opt_journal_path) {
1586                 char *journal_path;
1587                 struct inode *journal_inode;
1588                 struct path path;
1589                 int error;
1590
1591                 if (is_remount) {
1592                         ext4_msg(sb, KERN_ERR,
1593                                  "Cannot specify journal on remount");
1594                         return -1;
1595                 }
1596                 journal_path = match_strdup(&args[0]);
1597                 if (!journal_path) {
1598                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1599                                 "journal device string");
1600                         return -1;
1601                 }
1602
1603                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1604                 if (error) {
1605                         ext4_msg(sb, KERN_ERR, "error: could not find "
1606                                 "journal device path: error %d", error);
1607                         kfree(journal_path);
1608                         return -1;
1609                 }
1610
1611                 journal_inode = d_inode(path.dentry);
1612                 if (!S_ISBLK(journal_inode->i_mode)) {
1613                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1614                                 "is not a block device", journal_path);
1615                         path_put(&path);
1616                         kfree(journal_path);
1617                         return -1;
1618                 }
1619
1620                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1621                 path_put(&path);
1622                 kfree(journal_path);
1623         } else if (token == Opt_journal_ioprio) {
1624                 if (arg > 7) {
1625                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1626                                  " (must be 0-7)");
1627                         return -1;
1628                 }
1629                 *journal_ioprio =
1630                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1631         } else if (token == Opt_test_dummy_encryption) {
1632 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1633                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1634                 ext4_msg(sb, KERN_WARNING,
1635                          "Test dummy encryption mode enabled");
1636 #else
1637                 ext4_msg(sb, KERN_WARNING,
1638                          "Test dummy encryption mount option ignored");
1639 #endif
1640         } else if (m->flags & MOPT_DATAJ) {
1641                 if (is_remount) {
1642                         if (!sbi->s_journal)
1643                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1644                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1645                                 ext4_msg(sb, KERN_ERR,
1646                                          "Cannot change data mode on remount");
1647                                 return -1;
1648                         }
1649                 } else {
1650                         clear_opt(sb, DATA_FLAGS);
1651                         sbi->s_mount_opt |= m->mount_opt;
1652                 }
1653 #ifdef CONFIG_QUOTA
1654         } else if (m->flags & MOPT_QFMT) {
1655                 if (sb_any_quota_loaded(sb) &&
1656                     sbi->s_jquota_fmt != m->mount_opt) {
1657                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1658                                  "quota options when quota turned on");
1659                         return -1;
1660                 }
1661                 if (ext4_has_feature_quota(sb)) {
1662                         ext4_msg(sb, KERN_INFO,
1663                                  "Quota format mount options ignored "
1664                                  "when QUOTA feature is enabled");
1665                         return 1;
1666                 }
1667                 sbi->s_jquota_fmt = m->mount_opt;
1668 #endif
1669         } else if (token == Opt_dax) {
1670 #ifdef CONFIG_FS_DAX
1671                 ext4_msg(sb, KERN_WARNING,
1672                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1673                         sbi->s_mount_opt |= m->mount_opt;
1674 #else
1675                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1676                 return -1;
1677 #endif
1678         } else {
1679                 if (!args->from)
1680                         arg = 1;
1681                 if (m->flags & MOPT_CLEAR)
1682                         arg = !arg;
1683                 else if (unlikely(!(m->flags & MOPT_SET))) {
1684                         ext4_msg(sb, KERN_WARNING,
1685                                  "buggy handling of option %s", opt);
1686                         WARN_ON(1);
1687                         return -1;
1688                 }
1689                 if (arg != 0)
1690                         sbi->s_mount_opt |= m->mount_opt;
1691                 else
1692                         sbi->s_mount_opt &= ~m->mount_opt;
1693         }
1694         return 1;
1695 }
1696
1697 static int parse_options(char *options, struct super_block *sb,
1698                          unsigned long *journal_devnum,
1699                          unsigned int *journal_ioprio,
1700                          int is_remount)
1701 {
1702         struct ext4_sb_info *sbi = EXT4_SB(sb);
1703         char *p;
1704         substring_t args[MAX_OPT_ARGS];
1705         int token;
1706
1707         if (!options)
1708                 return 1;
1709
1710         while ((p = strsep(&options, ",")) != NULL) {
1711                 if (!*p)
1712                         continue;
1713                 /*
1714                  * Initialize args struct so we know whether arg was
1715                  * found; some options take optional arguments.
1716                  */
1717                 args[0].to = args[0].from = NULL;
1718                 token = match_token(p, tokens, args);
1719                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1720                                      journal_ioprio, is_remount) < 0)
1721                         return 0;
1722         }
1723 #ifdef CONFIG_QUOTA
1724         if (ext4_has_feature_quota(sb) &&
1725             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1726                 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1727                          "mount options ignored.");
1728                 clear_opt(sb, USRQUOTA);
1729                 clear_opt(sb, GRPQUOTA);
1730         } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1731                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1732                         clear_opt(sb, USRQUOTA);
1733
1734                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1735                         clear_opt(sb, GRPQUOTA);
1736
1737                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1738                         ext4_msg(sb, KERN_ERR, "old and new quota "
1739                                         "format mixing");
1740                         return 0;
1741                 }
1742
1743                 if (!sbi->s_jquota_fmt) {
1744                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1745                                         "not specified");
1746                         return 0;
1747                 }
1748         }
1749 #endif
1750         if (test_opt(sb, DIOREAD_NOLOCK)) {
1751                 int blocksize =
1752                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1753
1754                 if (blocksize < PAGE_CACHE_SIZE) {
1755                         ext4_msg(sb, KERN_ERR, "can't mount with "
1756                                  "dioread_nolock if block size != PAGE_SIZE");
1757                         return 0;
1758                 }
1759         }
1760         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1761             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1762                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1763                          "in data=ordered mode");
1764                 return 0;
1765         }
1766         return 1;
1767 }
1768
1769 static inline void ext4_show_quota_options(struct seq_file *seq,
1770                                            struct super_block *sb)
1771 {
1772 #if defined(CONFIG_QUOTA)
1773         struct ext4_sb_info *sbi = EXT4_SB(sb);
1774
1775         if (sbi->s_jquota_fmt) {
1776                 char *fmtname = "";
1777
1778                 switch (sbi->s_jquota_fmt) {
1779                 case QFMT_VFS_OLD:
1780                         fmtname = "vfsold";
1781                         break;
1782                 case QFMT_VFS_V0:
1783                         fmtname = "vfsv0";
1784                         break;
1785                 case QFMT_VFS_V1:
1786                         fmtname = "vfsv1";
1787                         break;
1788                 }
1789                 seq_printf(seq, ",jqfmt=%s", fmtname);
1790         }
1791
1792         if (sbi->s_qf_names[USRQUOTA])
1793                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1794
1795         if (sbi->s_qf_names[GRPQUOTA])
1796                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1797 #endif
1798 }
1799
1800 static const char *token2str(int token)
1801 {
1802         const struct match_token *t;
1803
1804         for (t = tokens; t->token != Opt_err; t++)
1805                 if (t->token == token && !strchr(t->pattern, '='))
1806                         break;
1807         return t->pattern;
1808 }
1809
1810 /*
1811  * Show an option if
1812  *  - it's set to a non-default value OR
1813  *  - if the per-sb default is different from the global default
1814  */
1815 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1816                               int nodefs)
1817 {
1818         struct ext4_sb_info *sbi = EXT4_SB(sb);
1819         struct ext4_super_block *es = sbi->s_es;
1820         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1821         const struct mount_opts *m;
1822         char sep = nodefs ? '\n' : ',';
1823
1824 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1825 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1826
1827         if (sbi->s_sb_block != 1)
1828                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1829
1830         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1831                 int want_set = m->flags & MOPT_SET;
1832                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1833                     (m->flags & MOPT_CLEAR_ERR))
1834                         continue;
1835                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1836                         continue; /* skip if same as the default */
1837                 if ((want_set &&
1838                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1839                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1840                         continue; /* select Opt_noFoo vs Opt_Foo */
1841                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1842         }
1843
1844         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1845             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1846                 SEQ_OPTS_PRINT("resuid=%u",
1847                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1848         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1849             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1850                 SEQ_OPTS_PRINT("resgid=%u",
1851                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1852         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1853         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1854                 SEQ_OPTS_PUTS("errors=remount-ro");
1855         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1856                 SEQ_OPTS_PUTS("errors=continue");
1857         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1858                 SEQ_OPTS_PUTS("errors=panic");
1859         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1860                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1861         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1862                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1863         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1864                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1865         if (sb->s_flags & MS_I_VERSION)
1866                 SEQ_OPTS_PUTS("i_version");
1867         if (nodefs || sbi->s_stripe)
1868                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1869         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1870                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1871                         SEQ_OPTS_PUTS("data=journal");
1872                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1873                         SEQ_OPTS_PUTS("data=ordered");
1874                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1875                         SEQ_OPTS_PUTS("data=writeback");
1876         }
1877         if (nodefs ||
1878             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1879                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1880                                sbi->s_inode_readahead_blks);
1881
1882         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1883                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1884                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1885         if (nodefs || sbi->s_max_dir_size_kb)
1886                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1887
1888         ext4_show_quota_options(seq, sb);
1889         return 0;
1890 }
1891
1892 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1893 {
1894         return _ext4_show_options(seq, root->d_sb, 0);
1895 }
1896
1897 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1898 {
1899         struct super_block *sb = seq->private;
1900         int rc;
1901
1902         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1903         rc = _ext4_show_options(seq, sb, 1);
1904         seq_puts(seq, "\n");
1905         return rc;
1906 }
1907
1908 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1909                             int read_only)
1910 {
1911         struct ext4_sb_info *sbi = EXT4_SB(sb);
1912         int res = 0;
1913
1914         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1915                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1916                          "forcing read-only mode");
1917                 res = MS_RDONLY;
1918         }
1919         if (read_only)
1920                 goto done;
1921         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1922                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1923                          "running e2fsck is recommended");
1924         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1925                 ext4_msg(sb, KERN_WARNING,
1926                          "warning: mounting fs with errors, "
1927                          "running e2fsck is recommended");
1928         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1929                  le16_to_cpu(es->s_mnt_count) >=
1930                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1931                 ext4_msg(sb, KERN_WARNING,
1932                          "warning: maximal mount count reached, "
1933                          "running e2fsck is recommended");
1934         else if (le32_to_cpu(es->s_checkinterval) &&
1935                 (le32_to_cpu(es->s_lastcheck) +
1936                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1937                 ext4_msg(sb, KERN_WARNING,
1938                          "warning: checktime reached, "
1939                          "running e2fsck is recommended");
1940         if (!sbi->s_journal)
1941                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1942         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1943                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1944         le16_add_cpu(&es->s_mnt_count, 1);
1945         es->s_mtime = cpu_to_le32(get_seconds());
1946         ext4_update_dynamic_rev(sb);
1947         if (sbi->s_journal)
1948                 ext4_set_feature_journal_needs_recovery(sb);
1949
1950         ext4_commit_super(sb, 1);
1951 done:
1952         if (test_opt(sb, DEBUG))
1953                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1954                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1955                         sb->s_blocksize,
1956                         sbi->s_groups_count,
1957                         EXT4_BLOCKS_PER_GROUP(sb),
1958                         EXT4_INODES_PER_GROUP(sb),
1959                         sbi->s_mount_opt, sbi->s_mount_opt2);
1960
1961         cleancache_init_fs(sb);
1962         return res;
1963 }
1964
1965 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1966 {
1967         struct ext4_sb_info *sbi = EXT4_SB(sb);
1968         struct flex_groups *new_groups;
1969         int size;
1970
1971         if (!sbi->s_log_groups_per_flex)
1972                 return 0;
1973
1974         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1975         if (size <= sbi->s_flex_groups_allocated)
1976                 return 0;
1977
1978         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1979         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1980         if (!new_groups) {
1981                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1982                          size / (int) sizeof(struct flex_groups));
1983                 return -ENOMEM;
1984         }
1985
1986         if (sbi->s_flex_groups) {
1987                 memcpy(new_groups, sbi->s_flex_groups,
1988                        (sbi->s_flex_groups_allocated *
1989                         sizeof(struct flex_groups)));
1990                 kvfree(sbi->s_flex_groups);
1991         }
1992         sbi->s_flex_groups = new_groups;
1993         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1994         return 0;
1995 }
1996
1997 static int ext4_fill_flex_info(struct super_block *sb)
1998 {
1999         struct ext4_sb_info *sbi = EXT4_SB(sb);
2000         struct ext4_group_desc *gdp = NULL;
2001         ext4_group_t flex_group;
2002         int i, err;
2003
2004         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2005         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2006                 sbi->s_log_groups_per_flex = 0;
2007                 return 1;
2008         }
2009
2010         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2011         if (err)
2012                 goto failed;
2013
2014         for (i = 0; i < sbi->s_groups_count; i++) {
2015                 gdp = ext4_get_group_desc(sb, i, NULL);
2016
2017                 flex_group = ext4_flex_group(sbi, i);
2018                 atomic_add(ext4_free_inodes_count(sb, gdp),
2019                            &sbi->s_flex_groups[flex_group].free_inodes);
2020                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2021                              &sbi->s_flex_groups[flex_group].free_clusters);
2022                 atomic_add(ext4_used_dirs_count(sb, gdp),
2023                            &sbi->s_flex_groups[flex_group].used_dirs);
2024         }
2025
2026         return 1;
2027 failed:
2028         return 0;
2029 }
2030
2031 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2032                                    struct ext4_group_desc *gdp)
2033 {
2034         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2035         __u16 crc = 0;
2036         __le32 le_group = cpu_to_le32(block_group);
2037         struct ext4_sb_info *sbi = EXT4_SB(sb);
2038
2039         if (ext4_has_metadata_csum(sbi->s_sb)) {
2040                 /* Use new metadata_csum algorithm */
2041                 __u32 csum32;
2042                 __u16 dummy_csum = 0;
2043
2044                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2045                                      sizeof(le_group));
2046                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2047                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2048                                      sizeof(dummy_csum));
2049                 offset += sizeof(dummy_csum);
2050                 if (offset < sbi->s_desc_size)
2051                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2052                                              sbi->s_desc_size - offset);
2053
2054                 crc = csum32 & 0xFFFF;
2055                 goto out;
2056         }
2057
2058         /* old crc16 code */
2059         if (!ext4_has_feature_gdt_csum(sb))
2060                 return 0;
2061
2062         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2063         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2064         crc = crc16(crc, (__u8 *)gdp, offset);
2065         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2066         /* for checksum of struct ext4_group_desc do the rest...*/
2067         if (ext4_has_feature_64bit(sb) &&
2068             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2069                 crc = crc16(crc, (__u8 *)gdp + offset,
2070                             le16_to_cpu(sbi->s_es->s_desc_size) -
2071                                 offset);
2072
2073 out:
2074         return cpu_to_le16(crc);
2075 }
2076
2077 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2078                                 struct ext4_group_desc *gdp)
2079 {
2080         if (ext4_has_group_desc_csum(sb) &&
2081             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2082                 return 0;
2083
2084         return 1;
2085 }
2086
2087 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2088                               struct ext4_group_desc *gdp)
2089 {
2090         if (!ext4_has_group_desc_csum(sb))
2091                 return;
2092         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2093 }
2094
2095 /* Called at mount-time, super-block is locked */
2096 static int ext4_check_descriptors(struct super_block *sb,
2097                                   ext4_fsblk_t sb_block,
2098                                   ext4_group_t *first_not_zeroed)
2099 {
2100         struct ext4_sb_info *sbi = EXT4_SB(sb);
2101         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2102         ext4_fsblk_t last_block;
2103         ext4_fsblk_t block_bitmap;
2104         ext4_fsblk_t inode_bitmap;
2105         ext4_fsblk_t inode_table;
2106         int flexbg_flag = 0;
2107         ext4_group_t i, grp = sbi->s_groups_count;
2108
2109         if (ext4_has_feature_flex_bg(sb))
2110                 flexbg_flag = 1;
2111
2112         ext4_debug("Checking group descriptors");
2113
2114         for (i = 0; i < sbi->s_groups_count; i++) {
2115                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2116
2117                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2118                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2119                 else
2120                         last_block = first_block +
2121                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2122
2123                 if ((grp == sbi->s_groups_count) &&
2124                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2125                         grp = i;
2126
2127                 block_bitmap = ext4_block_bitmap(sb, gdp);
2128                 if (block_bitmap == sb_block) {
2129                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2130                                  "Block bitmap for group %u overlaps "
2131                                  "superblock", i);
2132                 }
2133                 if (block_bitmap < first_block || block_bitmap > last_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                "Block bitmap for group %u not in group "
2136                                "(block %llu)!", i, block_bitmap);
2137                         return 0;
2138                 }
2139                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2140                 if (inode_bitmap == sb_block) {
2141                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2142                                  "Inode bitmap for group %u overlaps "
2143                                  "superblock", i);
2144                 }
2145                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                "Inode bitmap for group %u not in group "
2148                                "(block %llu)!", i, inode_bitmap);
2149                         return 0;
2150                 }
2151                 inode_table = ext4_inode_table(sb, gdp);
2152                 if (inode_table == sb_block) {
2153                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2154                                  "Inode table for group %u overlaps "
2155                                  "superblock", i);
2156                 }
2157                 if (inode_table < first_block ||
2158                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2159                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2160                                "Inode table for group %u not in group "
2161                                "(block %llu)!", i, inode_table);
2162                         return 0;
2163                 }
2164                 ext4_lock_group(sb, i);
2165                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2166                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2167                                  "Checksum for group %u failed (%u!=%u)",
2168                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2169                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2170                         if (!(sb->s_flags & MS_RDONLY)) {
2171                                 ext4_unlock_group(sb, i);
2172                                 return 0;
2173                         }
2174                 }
2175                 ext4_unlock_group(sb, i);
2176                 if (!flexbg_flag)
2177                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2178         }
2179         if (NULL != first_not_zeroed)
2180                 *first_not_zeroed = grp;
2181         return 1;
2182 }
2183
2184 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2185  * the superblock) which were deleted from all directories, but held open by
2186  * a process at the time of a crash.  We walk the list and try to delete these
2187  * inodes at recovery time (only with a read-write filesystem).
2188  *
2189  * In order to keep the orphan inode chain consistent during traversal (in
2190  * case of crash during recovery), we link each inode into the superblock
2191  * orphan list_head and handle it the same way as an inode deletion during
2192  * normal operation (which journals the operations for us).
2193  *
2194  * We only do an iget() and an iput() on each inode, which is very safe if we
2195  * accidentally point at an in-use or already deleted inode.  The worst that
2196  * can happen in this case is that we get a "bit already cleared" message from
2197  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2198  * e2fsck was run on this filesystem, and it must have already done the orphan
2199  * inode cleanup for us, so we can safely abort without any further action.
2200  */
2201 static void ext4_orphan_cleanup(struct super_block *sb,
2202                                 struct ext4_super_block *es)
2203 {
2204         unsigned int s_flags = sb->s_flags;
2205         int nr_orphans = 0, nr_truncates = 0;
2206 #ifdef CONFIG_QUOTA
2207         int i;
2208 #endif
2209         if (!es->s_last_orphan) {
2210                 jbd_debug(4, "no orphan inodes to clean up\n");
2211                 return;
2212         }
2213
2214         if (bdev_read_only(sb->s_bdev)) {
2215                 ext4_msg(sb, KERN_ERR, "write access "
2216                         "unavailable, skipping orphan cleanup");
2217                 return;
2218         }
2219
2220         /* Check if feature set would not allow a r/w mount */
2221         if (!ext4_feature_set_ok(sb, 0)) {
2222                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2223                          "unknown ROCOMPAT features");
2224                 return;
2225         }
2226
2227         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2228                 /* don't clear list on RO mount w/ errors */
2229                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2230                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2231                                   "clearing orphan list.\n");
2232                         es->s_last_orphan = 0;
2233                 }
2234                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2235                 return;
2236         }
2237
2238         if (s_flags & MS_RDONLY) {
2239                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2240                 sb->s_flags &= ~MS_RDONLY;
2241         }
2242 #ifdef CONFIG_QUOTA
2243         /* Needed for iput() to work correctly and not trash data */
2244         sb->s_flags |= MS_ACTIVE;
2245         /* Turn on quotas so that they are updated correctly */
2246         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2247                 if (EXT4_SB(sb)->s_qf_names[i]) {
2248                         int ret = ext4_quota_on_mount(sb, i);
2249                         if (ret < 0)
2250                                 ext4_msg(sb, KERN_ERR,
2251                                         "Cannot turn on journaled "
2252                                         "quota: error %d", ret);
2253                 }
2254         }
2255 #endif
2256
2257         while (es->s_last_orphan) {
2258                 struct inode *inode;
2259
2260                 /*
2261                  * We may have encountered an error during cleanup; if
2262                  * so, skip the rest.
2263                  */
2264                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2265                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2266                         es->s_last_orphan = 0;
2267                         break;
2268                 }
2269
2270                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2271                 if (IS_ERR(inode)) {
2272                         es->s_last_orphan = 0;
2273                         break;
2274                 }
2275
2276                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2277                 dquot_initialize(inode);
2278                 if (inode->i_nlink) {
2279                         if (test_opt(sb, DEBUG))
2280                                 ext4_msg(sb, KERN_DEBUG,
2281                                         "%s: truncating inode %lu to %lld bytes",
2282                                         __func__, inode->i_ino, inode->i_size);
2283                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2284                                   inode->i_ino, inode->i_size);
2285                         mutex_lock(&inode->i_mutex);
2286                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2287                         ext4_truncate(inode);
2288                         mutex_unlock(&inode->i_mutex);
2289                         nr_truncates++;
2290                 } else {
2291                         if (test_opt(sb, DEBUG))
2292                                 ext4_msg(sb, KERN_DEBUG,
2293                                         "%s: deleting unreferenced inode %lu",
2294                                         __func__, inode->i_ino);
2295                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2296                                   inode->i_ino);
2297                         nr_orphans++;
2298                 }
2299                 iput(inode);  /* The delete magic happens here! */
2300         }
2301
2302 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2303
2304         if (nr_orphans)
2305                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2306                        PLURAL(nr_orphans));
2307         if (nr_truncates)
2308                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2309                        PLURAL(nr_truncates));
2310 #ifdef CONFIG_QUOTA
2311         /* Turn quotas off */
2312         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2313                 if (sb_dqopt(sb)->files[i])
2314                         dquot_quota_off(sb, i);
2315         }
2316 #endif
2317         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2318 }
2319
2320 /*
2321  * Maximal extent format file size.
2322  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2323  * extent format containers, within a sector_t, and within i_blocks
2324  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2325  * so that won't be a limiting factor.
2326  *
2327  * However there is other limiting factor. We do store extents in the form
2328  * of starting block and length, hence the resulting length of the extent
2329  * covering maximum file size must fit into on-disk format containers as
2330  * well. Given that length is always by 1 unit bigger than max unit (because
2331  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2332  *
2333  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2334  */
2335 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2336 {
2337         loff_t res;
2338         loff_t upper_limit = MAX_LFS_FILESIZE;
2339
2340         /* small i_blocks in vfs inode? */
2341         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2342                 /*
2343                  * CONFIG_LBDAF is not enabled implies the inode
2344                  * i_block represent total blocks in 512 bytes
2345                  * 32 == size of vfs inode i_blocks * 8
2346                  */
2347                 upper_limit = (1LL << 32) - 1;
2348
2349                 /* total blocks in file system block size */
2350                 upper_limit >>= (blkbits - 9);
2351                 upper_limit <<= blkbits;
2352         }
2353
2354         /*
2355          * 32-bit extent-start container, ee_block. We lower the maxbytes
2356          * by one fs block, so ee_len can cover the extent of maximum file
2357          * size
2358          */
2359         res = (1LL << 32) - 1;
2360         res <<= blkbits;
2361
2362         /* Sanity check against vm- & vfs- imposed limits */
2363         if (res > upper_limit)
2364                 res = upper_limit;
2365
2366         return res;
2367 }
2368
2369 /*
2370  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2371  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2372  * We need to be 1 filesystem block less than the 2^48 sector limit.
2373  */
2374 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2375 {
2376         loff_t res = EXT4_NDIR_BLOCKS;
2377         int meta_blocks;
2378         loff_t upper_limit;
2379         /* This is calculated to be the largest file size for a dense, block
2380          * mapped file such that the file's total number of 512-byte sectors,
2381          * including data and all indirect blocks, does not exceed (2^48 - 1).
2382          *
2383          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2384          * number of 512-byte sectors of the file.
2385          */
2386
2387         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2388                 /*
2389                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2390                  * the inode i_block field represents total file blocks in
2391                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2392                  */
2393                 upper_limit = (1LL << 32) - 1;
2394
2395                 /* total blocks in file system block size */
2396                 upper_limit >>= (bits - 9);
2397
2398         } else {
2399                 /*
2400                  * We use 48 bit ext4_inode i_blocks
2401                  * With EXT4_HUGE_FILE_FL set the i_blocks
2402                  * represent total number of blocks in
2403                  * file system block size
2404                  */
2405                 upper_limit = (1LL << 48) - 1;
2406
2407         }
2408
2409         /* indirect blocks */
2410         meta_blocks = 1;
2411         /* double indirect blocks */
2412         meta_blocks += 1 + (1LL << (bits-2));
2413         /* tripple indirect blocks */
2414         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2415
2416         upper_limit -= meta_blocks;
2417         upper_limit <<= bits;
2418
2419         res += 1LL << (bits-2);
2420         res += 1LL << (2*(bits-2));
2421         res += 1LL << (3*(bits-2));
2422         res <<= bits;
2423         if (res > upper_limit)
2424                 res = upper_limit;
2425
2426         if (res > MAX_LFS_FILESIZE)
2427                 res = MAX_LFS_FILESIZE;
2428
2429         return res;
2430 }
2431
2432 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2433                                    ext4_fsblk_t logical_sb_block, int nr)
2434 {
2435         struct ext4_sb_info *sbi = EXT4_SB(sb);
2436         ext4_group_t bg, first_meta_bg;
2437         int has_super = 0;
2438
2439         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2440
2441         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2442                 return logical_sb_block + nr + 1;
2443         bg = sbi->s_desc_per_block * nr;
2444         if (ext4_bg_has_super(sb, bg))
2445                 has_super = 1;
2446
2447         /*
2448          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2449          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2450          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2451          * compensate.
2452          */
2453         if (sb->s_blocksize == 1024 && nr == 0 &&
2454             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2455                 has_super++;
2456
2457         return (has_super + ext4_group_first_block_no(sb, bg));
2458 }
2459
2460 /**
2461  * ext4_get_stripe_size: Get the stripe size.
2462  * @sbi: In memory super block info
2463  *
2464  * If we have specified it via mount option, then
2465  * use the mount option value. If the value specified at mount time is
2466  * greater than the blocks per group use the super block value.
2467  * If the super block value is greater than blocks per group return 0.
2468  * Allocator needs it be less than blocks per group.
2469  *
2470  */
2471 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2472 {
2473         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2474         unsigned long stripe_width =
2475                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2476         int ret;
2477
2478         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2479                 ret = sbi->s_stripe;
2480         else if (stripe_width <= sbi->s_blocks_per_group)
2481                 ret = stripe_width;
2482         else if (stride <= sbi->s_blocks_per_group)
2483                 ret = stride;
2484         else
2485                 ret = 0;
2486
2487         /*
2488          * If the stripe width is 1, this makes no sense and
2489          * we set it to 0 to turn off stripe handling code.
2490          */
2491         if (ret <= 1)
2492                 ret = 0;
2493
2494         return ret;
2495 }
2496
2497 /*
2498  * Check whether this filesystem can be mounted based on
2499  * the features present and the RDONLY/RDWR mount requested.
2500  * Returns 1 if this filesystem can be mounted as requested,
2501  * 0 if it cannot be.
2502  */
2503 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2504 {
2505         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2506                 ext4_msg(sb, KERN_ERR,
2507                         "Couldn't mount because of "
2508                         "unsupported optional features (%x)",
2509                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2510                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2511                 return 0;
2512         }
2513
2514         if (readonly)
2515                 return 1;
2516
2517         if (ext4_has_feature_readonly(sb)) {
2518                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2519                 sb->s_flags |= MS_RDONLY;
2520                 return 1;
2521         }
2522
2523         /* Check that feature set is OK for a read-write mount */
2524         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2525                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2526                          "unsupported optional features (%x)",
2527                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2528                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2529                 return 0;
2530         }
2531         /*
2532          * Large file size enabled file system can only be mounted
2533          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2534          */
2535         if (ext4_has_feature_huge_file(sb)) {
2536                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2537                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2538                                  "cannot be mounted RDWR without "
2539                                  "CONFIG_LBDAF");
2540                         return 0;
2541                 }
2542         }
2543         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2544                 ext4_msg(sb, KERN_ERR,
2545                          "Can't support bigalloc feature without "
2546                          "extents feature\n");
2547                 return 0;
2548         }
2549
2550 #ifndef CONFIG_QUOTA
2551         if (ext4_has_feature_quota(sb) && !readonly) {
2552                 ext4_msg(sb, KERN_ERR,
2553                          "Filesystem with quota feature cannot be mounted RDWR "
2554                          "without CONFIG_QUOTA");
2555                 return 0;
2556         }
2557 #endif  /* CONFIG_QUOTA */
2558         return 1;
2559 }
2560
2561 /*
2562  * This function is called once a day if we have errors logged
2563  * on the file system
2564  */
2565 static void print_daily_error_info(unsigned long arg)
2566 {
2567         struct super_block *sb = (struct super_block *) arg;
2568         struct ext4_sb_info *sbi;
2569         struct ext4_super_block *es;
2570
2571         sbi = EXT4_SB(sb);
2572         es = sbi->s_es;
2573
2574         if (es->s_error_count)
2575                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2576                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2577                          le32_to_cpu(es->s_error_count));
2578         if (es->s_first_error_time) {
2579                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2580                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2581                        (int) sizeof(es->s_first_error_func),
2582                        es->s_first_error_func,
2583                        le32_to_cpu(es->s_first_error_line));
2584                 if (es->s_first_error_ino)
2585                         printk(": inode %u",
2586                                le32_to_cpu(es->s_first_error_ino));
2587                 if (es->s_first_error_block)
2588                         printk(": block %llu", (unsigned long long)
2589                                le64_to_cpu(es->s_first_error_block));
2590                 printk("\n");
2591         }
2592         if (es->s_last_error_time) {
2593                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2594                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2595                        (int) sizeof(es->s_last_error_func),
2596                        es->s_last_error_func,
2597                        le32_to_cpu(es->s_last_error_line));
2598                 if (es->s_last_error_ino)
2599                         printk(": inode %u",
2600                                le32_to_cpu(es->s_last_error_ino));
2601                 if (es->s_last_error_block)
2602                         printk(": block %llu", (unsigned long long)
2603                                le64_to_cpu(es->s_last_error_block));
2604                 printk("\n");
2605         }
2606         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2607 }
2608
2609 /* Find next suitable group and run ext4_init_inode_table */
2610 static int ext4_run_li_request(struct ext4_li_request *elr)
2611 {
2612         struct ext4_group_desc *gdp = NULL;
2613         ext4_group_t group, ngroups;
2614         struct super_block *sb;
2615         unsigned long timeout = 0;
2616         int ret = 0;
2617
2618         sb = elr->lr_super;
2619         ngroups = EXT4_SB(sb)->s_groups_count;
2620
2621         sb_start_write(sb);
2622         for (group = elr->lr_next_group; group < ngroups; group++) {
2623                 gdp = ext4_get_group_desc(sb, group, NULL);
2624                 if (!gdp) {
2625                         ret = 1;
2626                         break;
2627                 }
2628
2629                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2630                         break;
2631         }
2632
2633         if (group >= ngroups)
2634                 ret = 1;
2635
2636         if (!ret) {
2637                 timeout = jiffies;
2638                 ret = ext4_init_inode_table(sb, group,
2639                                             elr->lr_timeout ? 0 : 1);
2640                 if (elr->lr_timeout == 0) {
2641                         timeout = (jiffies - timeout) *
2642                                   elr->lr_sbi->s_li_wait_mult;
2643                         elr->lr_timeout = timeout;
2644                 }
2645                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2646                 elr->lr_next_group = group + 1;
2647         }
2648         sb_end_write(sb);
2649
2650         return ret;
2651 }
2652
2653 /*
2654  * Remove lr_request from the list_request and free the
2655  * request structure. Should be called with li_list_mtx held
2656  */
2657 static void ext4_remove_li_request(struct ext4_li_request *elr)
2658 {
2659         struct ext4_sb_info *sbi;
2660
2661         if (!elr)
2662                 return;
2663
2664         sbi = elr->lr_sbi;
2665
2666         list_del(&elr->lr_request);
2667         sbi->s_li_request = NULL;
2668         kfree(elr);
2669 }
2670
2671 static void ext4_unregister_li_request(struct super_block *sb)
2672 {
2673         mutex_lock(&ext4_li_mtx);
2674         if (!ext4_li_info) {
2675                 mutex_unlock(&ext4_li_mtx);
2676                 return;
2677         }
2678
2679         mutex_lock(&ext4_li_info->li_list_mtx);
2680         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2681         mutex_unlock(&ext4_li_info->li_list_mtx);
2682         mutex_unlock(&ext4_li_mtx);
2683 }
2684
2685 static struct task_struct *ext4_lazyinit_task;
2686
2687 /*
2688  * This is the function where ext4lazyinit thread lives. It walks
2689  * through the request list searching for next scheduled filesystem.
2690  * When such a fs is found, run the lazy initialization request
2691  * (ext4_rn_li_request) and keep track of the time spend in this
2692  * function. Based on that time we compute next schedule time of
2693  * the request. When walking through the list is complete, compute
2694  * next waking time and put itself into sleep.
2695  */
2696 static int ext4_lazyinit_thread(void *arg)
2697 {
2698         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2699         struct list_head *pos, *n;
2700         struct ext4_li_request *elr;
2701         unsigned long next_wakeup, cur;
2702
2703         BUG_ON(NULL == eli);
2704
2705 cont_thread:
2706         while (true) {
2707                 next_wakeup = MAX_JIFFY_OFFSET;
2708
2709                 mutex_lock(&eli->li_list_mtx);
2710                 if (list_empty(&eli->li_request_list)) {
2711                         mutex_unlock(&eli->li_list_mtx);
2712                         goto exit_thread;
2713                 }
2714
2715                 list_for_each_safe(pos, n, &eli->li_request_list) {
2716                         elr = list_entry(pos, struct ext4_li_request,
2717                                          lr_request);
2718
2719                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2720                                 if (ext4_run_li_request(elr) != 0) {
2721                                         /* error, remove the lazy_init job */
2722                                         ext4_remove_li_request(elr);
2723                                         continue;
2724                                 }
2725                         }
2726
2727                         if (time_before(elr->lr_next_sched, next_wakeup))
2728                                 next_wakeup = elr->lr_next_sched;
2729                 }
2730                 mutex_unlock(&eli->li_list_mtx);
2731
2732                 try_to_freeze();
2733
2734                 cur = jiffies;
2735                 if ((time_after_eq(cur, next_wakeup)) ||
2736                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2737                         cond_resched();
2738                         continue;
2739                 }
2740
2741                 schedule_timeout_interruptible(next_wakeup - cur);
2742
2743                 if (kthread_should_stop()) {
2744                         ext4_clear_request_list();
2745                         goto exit_thread;
2746                 }
2747         }
2748
2749 exit_thread:
2750         /*
2751          * It looks like the request list is empty, but we need
2752          * to check it under the li_list_mtx lock, to prevent any
2753          * additions into it, and of course we should lock ext4_li_mtx
2754          * to atomically free the list and ext4_li_info, because at
2755          * this point another ext4 filesystem could be registering
2756          * new one.
2757          */
2758         mutex_lock(&ext4_li_mtx);
2759         mutex_lock(&eli->li_list_mtx);
2760         if (!list_empty(&eli->li_request_list)) {
2761                 mutex_unlock(&eli->li_list_mtx);
2762                 mutex_unlock(&ext4_li_mtx);
2763                 goto cont_thread;
2764         }
2765         mutex_unlock(&eli->li_list_mtx);
2766         kfree(ext4_li_info);
2767         ext4_li_info = NULL;
2768         mutex_unlock(&ext4_li_mtx);
2769
2770         return 0;
2771 }
2772
2773 static void ext4_clear_request_list(void)
2774 {
2775         struct list_head *pos, *n;
2776         struct ext4_li_request *elr;
2777
2778         mutex_lock(&ext4_li_info->li_list_mtx);
2779         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2780                 elr = list_entry(pos, struct ext4_li_request,
2781                                  lr_request);
2782                 ext4_remove_li_request(elr);
2783         }
2784         mutex_unlock(&ext4_li_info->li_list_mtx);
2785 }
2786
2787 static int ext4_run_lazyinit_thread(void)
2788 {
2789         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2790                                          ext4_li_info, "ext4lazyinit");
2791         if (IS_ERR(ext4_lazyinit_task)) {
2792                 int err = PTR_ERR(ext4_lazyinit_task);
2793                 ext4_clear_request_list();
2794                 kfree(ext4_li_info);
2795                 ext4_li_info = NULL;
2796                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2797                                  "initialization thread\n",
2798                                  err);
2799                 return err;
2800         }
2801         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2802         return 0;
2803 }
2804
2805 /*
2806  * Check whether it make sense to run itable init. thread or not.
2807  * If there is at least one uninitialized inode table, return
2808  * corresponding group number, else the loop goes through all
2809  * groups and return total number of groups.
2810  */
2811 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2812 {
2813         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2814         struct ext4_group_desc *gdp = NULL;
2815
2816         for (group = 0; group < ngroups; group++) {
2817                 gdp = ext4_get_group_desc(sb, group, NULL);
2818                 if (!gdp)
2819                         continue;
2820
2821                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2822                         break;
2823         }
2824
2825         return group;
2826 }
2827
2828 static int ext4_li_info_new(void)
2829 {
2830         struct ext4_lazy_init *eli = NULL;
2831
2832         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2833         if (!eli)
2834                 return -ENOMEM;
2835
2836         INIT_LIST_HEAD(&eli->li_request_list);
2837         mutex_init(&eli->li_list_mtx);
2838
2839         eli->li_state |= EXT4_LAZYINIT_QUIT;
2840
2841         ext4_li_info = eli;
2842
2843         return 0;
2844 }
2845
2846 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2847                                             ext4_group_t start)
2848 {
2849         struct ext4_sb_info *sbi = EXT4_SB(sb);
2850         struct ext4_li_request *elr;
2851
2852         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2853         if (!elr)
2854                 return NULL;
2855
2856         elr->lr_super = sb;
2857         elr->lr_sbi = sbi;
2858         elr->lr_next_group = start;
2859
2860         /*
2861          * Randomize first schedule time of the request to
2862          * spread the inode table initialization requests
2863          * better.
2864          */
2865         elr->lr_next_sched = jiffies + (prandom_u32() %
2866                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2867         return elr;
2868 }
2869
2870 int ext4_register_li_request(struct super_block *sb,
2871                              ext4_group_t first_not_zeroed)
2872 {
2873         struct ext4_sb_info *sbi = EXT4_SB(sb);
2874         struct ext4_li_request *elr = NULL;
2875         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2876         int ret = 0;
2877
2878         mutex_lock(&ext4_li_mtx);
2879         if (sbi->s_li_request != NULL) {
2880                 /*
2881                  * Reset timeout so it can be computed again, because
2882                  * s_li_wait_mult might have changed.
2883                  */
2884                 sbi->s_li_request->lr_timeout = 0;
2885                 goto out;
2886         }
2887
2888         if (first_not_zeroed == ngroups ||
2889             (sb->s_flags & MS_RDONLY) ||
2890             !test_opt(sb, INIT_INODE_TABLE))
2891                 goto out;
2892
2893         elr = ext4_li_request_new(sb, first_not_zeroed);
2894         if (!elr) {
2895                 ret = -ENOMEM;
2896                 goto out;
2897         }
2898
2899         if (NULL == ext4_li_info) {
2900                 ret = ext4_li_info_new();
2901                 if (ret)
2902                         goto out;
2903         }
2904
2905         mutex_lock(&ext4_li_info->li_list_mtx);
2906         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2907         mutex_unlock(&ext4_li_info->li_list_mtx);
2908
2909         sbi->s_li_request = elr;
2910         /*
2911          * set elr to NULL here since it has been inserted to
2912          * the request_list and the removal and free of it is
2913          * handled by ext4_clear_request_list from now on.
2914          */
2915         elr = NULL;
2916
2917         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2918                 ret = ext4_run_lazyinit_thread();
2919                 if (ret)
2920                         goto out;
2921         }
2922 out:
2923         mutex_unlock(&ext4_li_mtx);
2924         if (ret)
2925                 kfree(elr);
2926         return ret;
2927 }
2928
2929 /*
2930  * We do not need to lock anything since this is called on
2931  * module unload.
2932  */
2933 static void ext4_destroy_lazyinit_thread(void)
2934 {
2935         /*
2936          * If thread exited earlier
2937          * there's nothing to be done.
2938          */
2939         if (!ext4_li_info || !ext4_lazyinit_task)
2940                 return;
2941
2942         kthread_stop(ext4_lazyinit_task);
2943 }
2944
2945 static int set_journal_csum_feature_set(struct super_block *sb)
2946 {
2947         int ret = 1;
2948         int compat, incompat;
2949         struct ext4_sb_info *sbi = EXT4_SB(sb);
2950
2951         if (ext4_has_metadata_csum(sb)) {
2952                 /* journal checksum v3 */
2953                 compat = 0;
2954                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2955         } else {
2956                 /* journal checksum v1 */
2957                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2958                 incompat = 0;
2959         }
2960
2961         jbd2_journal_clear_features(sbi->s_journal,
2962                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2963                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2964                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2965         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2966                 ret = jbd2_journal_set_features(sbi->s_journal,
2967                                 compat, 0,
2968                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2969                                 incompat);
2970         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2971                 ret = jbd2_journal_set_features(sbi->s_journal,
2972                                 compat, 0,
2973                                 incompat);
2974                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2975                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2976         } else {
2977                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2978                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2979         }
2980
2981         return ret;
2982 }
2983
2984 /*
2985  * Note: calculating the overhead so we can be compatible with
2986  * historical BSD practice is quite difficult in the face of
2987  * clusters/bigalloc.  This is because multiple metadata blocks from
2988  * different block group can end up in the same allocation cluster.
2989  * Calculating the exact overhead in the face of clustered allocation
2990  * requires either O(all block bitmaps) in memory or O(number of block
2991  * groups**2) in time.  We will still calculate the superblock for
2992  * older file systems --- and if we come across with a bigalloc file
2993  * system with zero in s_overhead_clusters the estimate will be close to
2994  * correct especially for very large cluster sizes --- but for newer
2995  * file systems, it's better to calculate this figure once at mkfs
2996  * time, and store it in the superblock.  If the superblock value is
2997  * present (even for non-bigalloc file systems), we will use it.
2998  */
2999 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3000                           char *buf)
3001 {
3002         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3003         struct ext4_group_desc  *gdp;
3004         ext4_fsblk_t            first_block, last_block, b;
3005         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3006         int                     s, j, count = 0;
3007
3008         if (!ext4_has_feature_bigalloc(sb))
3009                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3010                         sbi->s_itb_per_group + 2);
3011
3012         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3013                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3014         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3015         for (i = 0; i < ngroups; i++) {
3016                 gdp = ext4_get_group_desc(sb, i, NULL);
3017                 b = ext4_block_bitmap(sb, gdp);
3018                 if (b >= first_block && b <= last_block) {
3019                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3020                         count++;
3021                 }
3022                 b = ext4_inode_bitmap(sb, gdp);
3023                 if (b >= first_block && b <= last_block) {
3024                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3025                         count++;
3026                 }
3027                 b = ext4_inode_table(sb, gdp);
3028                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3029                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3030                                 int c = EXT4_B2C(sbi, b - first_block);
3031                                 ext4_set_bit(c, buf);
3032                                 count++;
3033                         }
3034                 if (i != grp)
3035                         continue;
3036                 s = 0;
3037                 if (ext4_bg_has_super(sb, grp)) {
3038                         ext4_set_bit(s++, buf);
3039                         count++;
3040                 }
3041                 j = ext4_bg_num_gdb(sb, grp);
3042                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3043                         ext4_error(sb, "Invalid number of block group "
3044                                    "descriptor blocks: %d", j);
3045                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3046                 }
3047                 count += j;
3048                 for (; j > 0; j--)
3049                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3050         }
3051         if (!count)
3052                 return 0;
3053         return EXT4_CLUSTERS_PER_GROUP(sb) -
3054                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3055 }
3056
3057 /*
3058  * Compute the overhead and stash it in sbi->s_overhead
3059  */
3060 int ext4_calculate_overhead(struct super_block *sb)
3061 {
3062         struct ext4_sb_info *sbi = EXT4_SB(sb);
3063         struct ext4_super_block *es = sbi->s_es;
3064         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3065         ext4_fsblk_t overhead = 0;
3066         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3067
3068         if (!buf)
3069                 return -ENOMEM;
3070
3071         /*
3072          * Compute the overhead (FS structures).  This is constant
3073          * for a given filesystem unless the number of block groups
3074          * changes so we cache the previous value until it does.
3075          */
3076
3077         /*
3078          * All of the blocks before first_data_block are overhead
3079          */
3080         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3081
3082         /*
3083          * Add the overhead found in each block group
3084          */
3085         for (i = 0; i < ngroups; i++) {
3086                 int blks;
3087
3088                 blks = count_overhead(sb, i, buf);
3089                 overhead += blks;
3090                 if (blks)
3091                         memset(buf, 0, PAGE_SIZE);
3092                 cond_resched();
3093         }
3094         /* Add the internal journal blocks as well */
3095         if (sbi->s_journal && !sbi->journal_bdev)
3096                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3097
3098         sbi->s_overhead = overhead;
3099         smp_wmb();
3100         free_page((unsigned long) buf);
3101         return 0;
3102 }
3103
3104 static void ext4_set_resv_clusters(struct super_block *sb)
3105 {
3106         ext4_fsblk_t resv_clusters;
3107         struct ext4_sb_info *sbi = EXT4_SB(sb);
3108
3109         /*
3110          * There's no need to reserve anything when we aren't using extents.
3111          * The space estimates are exact, there are no unwritten extents,
3112          * hole punching doesn't need new metadata... This is needed especially
3113          * to keep ext2/3 backward compatibility.
3114          */
3115         if (!ext4_has_feature_extents(sb))
3116                 return;
3117         /*
3118          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3119          * This should cover the situations where we can not afford to run
3120          * out of space like for example punch hole, or converting
3121          * unwritten extents in delalloc path. In most cases such
3122          * allocation would require 1, or 2 blocks, higher numbers are
3123          * very rare.
3124          */
3125         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3126                          sbi->s_cluster_bits);
3127
3128         do_div(resv_clusters, 50);
3129         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3130
3131         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3132 }
3133
3134 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3135 {
3136         char *orig_data = kstrdup(data, GFP_KERNEL);
3137         struct buffer_head *bh;
3138         struct ext4_super_block *es = NULL;
3139         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3140         ext4_fsblk_t block;
3141         ext4_fsblk_t sb_block = get_sb_block(&data);
3142         ext4_fsblk_t logical_sb_block;
3143         unsigned long offset = 0;
3144         unsigned long journal_devnum = 0;
3145         unsigned long def_mount_opts;
3146         struct inode *root;
3147         const char *descr;
3148         int ret = -ENOMEM;
3149         int blocksize, clustersize;
3150         unsigned int db_count;
3151         unsigned int i;
3152         int needs_recovery, has_huge_files, has_bigalloc;
3153         __u64 blocks_count;
3154         int err = 0;
3155         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3156         ext4_group_t first_not_zeroed;
3157
3158         if ((data && !orig_data) || !sbi)
3159                 goto out_free_base;
3160
3161         sbi->s_blockgroup_lock =
3162                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3163         if (!sbi->s_blockgroup_lock)
3164                 goto out_free_base;
3165
3166         sb->s_fs_info = sbi;
3167         sbi->s_sb = sb;
3168         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3169         sbi->s_sb_block = sb_block;
3170         if (sb->s_bdev->bd_part)
3171                 sbi->s_sectors_written_start =
3172                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3173
3174         /* Cleanup superblock name */
3175         strreplace(sb->s_id, '/', '!');
3176
3177         /* -EINVAL is default */
3178         ret = -EINVAL;
3179         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3180         if (!blocksize) {
3181                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3182                 goto out_fail;
3183         }
3184
3185         /*
3186          * The ext4 superblock will not be buffer aligned for other than 1kB
3187          * block sizes.  We need to calculate the offset from buffer start.
3188          */
3189         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3190                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3191                 offset = do_div(logical_sb_block, blocksize);
3192         } else {
3193                 logical_sb_block = sb_block;
3194         }
3195
3196         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3197                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3198                 goto out_fail;
3199         }
3200         /*
3201          * Note: s_es must be initialized as soon as possible because
3202          *       some ext4 macro-instructions depend on its value
3203          */
3204         es = (struct ext4_super_block *) (bh->b_data + offset);
3205         sbi->s_es = es;
3206         sb->s_magic = le16_to_cpu(es->s_magic);
3207         if (sb->s_magic != EXT4_SUPER_MAGIC)
3208                 goto cantfind_ext4;
3209         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3210
3211         /* Warn if metadata_csum and gdt_csum are both set. */
3212         if (ext4_has_feature_metadata_csum(sb) &&
3213             ext4_has_feature_gdt_csum(sb))
3214                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3215                              "redundant flags; please run fsck.");
3216
3217         /* Check for a known checksum algorithm */
3218         if (!ext4_verify_csum_type(sb, es)) {
3219                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3220                          "unknown checksum algorithm.");
3221                 silent = 1;
3222                 goto cantfind_ext4;
3223         }
3224
3225         /* Load the checksum driver */
3226         if (ext4_has_feature_metadata_csum(sb)) {
3227                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3228                 if (IS_ERR(sbi->s_chksum_driver)) {
3229                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3230                         ret = PTR_ERR(sbi->s_chksum_driver);
3231                         sbi->s_chksum_driver = NULL;
3232                         goto failed_mount;
3233                 }
3234         }
3235
3236         /* Check superblock checksum */
3237         if (!ext4_superblock_csum_verify(sb, es)) {
3238                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3239                          "invalid superblock checksum.  Run e2fsck?");
3240                 silent = 1;
3241                 ret = -EFSBADCRC;
3242                 goto cantfind_ext4;
3243         }
3244
3245         /* Precompute checksum seed for all metadata */
3246         if (ext4_has_feature_csum_seed(sb))
3247                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3248         else if (ext4_has_metadata_csum(sb))
3249                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3250                                                sizeof(es->s_uuid));
3251
3252         /* Set defaults before we parse the mount options */
3253         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3254         set_opt(sb, INIT_INODE_TABLE);
3255         if (def_mount_opts & EXT4_DEFM_DEBUG)
3256                 set_opt(sb, DEBUG);
3257         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3258                 set_opt(sb, GRPID);
3259         if (def_mount_opts & EXT4_DEFM_UID16)
3260                 set_opt(sb, NO_UID32);
3261         /* xattr user namespace & acls are now defaulted on */
3262         set_opt(sb, XATTR_USER);
3263 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3264         set_opt(sb, POSIX_ACL);
3265 #endif
3266         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3267         if (ext4_has_metadata_csum(sb))
3268                 set_opt(sb, JOURNAL_CHECKSUM);
3269
3270         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3271                 set_opt(sb, JOURNAL_DATA);
3272         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3273                 set_opt(sb, ORDERED_DATA);
3274         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3275                 set_opt(sb, WRITEBACK_DATA);
3276
3277         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3278                 set_opt(sb, ERRORS_PANIC);
3279         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3280                 set_opt(sb, ERRORS_CONT);
3281         else
3282                 set_opt(sb, ERRORS_RO);
3283         /* block_validity enabled by default; disable with noblock_validity */
3284         set_opt(sb, BLOCK_VALIDITY);
3285         if (def_mount_opts & EXT4_DEFM_DISCARD)
3286                 set_opt(sb, DISCARD);
3287
3288         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3289         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3290         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3291         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3292         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3293
3294         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3295                 set_opt(sb, BARRIER);
3296
3297         /*
3298          * enable delayed allocation by default
3299          * Use -o nodelalloc to turn it off
3300          */
3301         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3302             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3303                 set_opt(sb, DELALLOC);
3304
3305         /*
3306          * set default s_li_wait_mult for lazyinit, for the case there is
3307          * no mount option specified.
3308          */
3309         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3310
3311         if (sbi->s_es->s_mount_opts[0]) {
3312                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3313                                               sizeof(sbi->s_es->s_mount_opts),
3314                                               GFP_KERNEL);
3315                 if (!s_mount_opts)
3316                         goto failed_mount;
3317                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3318                                    &journal_ioprio, 0)) {
3319                         ext4_msg(sb, KERN_WARNING,
3320                                  "failed to parse options in superblock: %s",
3321                                  s_mount_opts);
3322                 }
3323                 kfree(s_mount_opts);
3324         }
3325         sbi->s_def_mount_opt = sbi->s_mount_opt;
3326         if (!parse_options((char *) data, sb, &journal_devnum,
3327                            &journal_ioprio, 0))
3328                 goto failed_mount;
3329
3330         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3331                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3332                             "with data=journal disables delayed "
3333                             "allocation and O_DIRECT support!\n");
3334                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3335                         ext4_msg(sb, KERN_ERR, "can't mount with "
3336                                  "both data=journal and delalloc");
3337                         goto failed_mount;
3338                 }
3339                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3340                         ext4_msg(sb, KERN_ERR, "can't mount with "
3341                                  "both data=journal and dioread_nolock");
3342                         goto failed_mount;
3343                 }
3344                 if (test_opt(sb, DAX)) {
3345                         ext4_msg(sb, KERN_ERR, "can't mount with "
3346                                  "both data=journal and dax");
3347                         goto failed_mount;
3348                 }
3349                 if (ext4_has_feature_encrypt(sb)) {
3350                         ext4_msg(sb, KERN_WARNING,
3351                                  "encrypted files will use data=ordered "
3352                                  "instead of data journaling mode");
3353                 }
3354                 if (test_opt(sb, DELALLOC))
3355                         clear_opt(sb, DELALLOC);
3356         } else {
3357                 sb->s_iflags |= SB_I_CGROUPWB;
3358         }
3359
3360         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3361                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3362
3363         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3364             (ext4_has_compat_features(sb) ||
3365              ext4_has_ro_compat_features(sb) ||
3366              ext4_has_incompat_features(sb)))
3367                 ext4_msg(sb, KERN_WARNING,
3368                        "feature flags set on rev 0 fs, "
3369                        "running e2fsck is recommended");
3370
3371         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3372                 set_opt2(sb, HURD_COMPAT);
3373                 if (ext4_has_feature_64bit(sb)) {
3374                         ext4_msg(sb, KERN_ERR,
3375                                  "The Hurd can't support 64-bit file systems");
3376                         goto failed_mount;
3377                 }
3378         }
3379
3380         if (IS_EXT2_SB(sb)) {
3381                 if (ext2_feature_set_ok(sb))
3382                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3383                                  "using the ext4 subsystem");
3384                 else {
3385                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3386                                  "to feature incompatibilities");
3387                         goto failed_mount;
3388                 }
3389         }
3390
3391         if (IS_EXT3_SB(sb)) {
3392                 if (ext3_feature_set_ok(sb))
3393                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3394                                  "using the ext4 subsystem");
3395                 else {
3396                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3397                                  "to feature incompatibilities");
3398                         goto failed_mount;
3399                 }
3400         }
3401
3402         /*
3403          * Check feature flags regardless of the revision level, since we
3404          * previously didn't change the revision level when setting the flags,
3405          * so there is a chance incompat flags are set on a rev 0 filesystem.
3406          */
3407         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3408                 goto failed_mount;
3409
3410         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3411         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3412             blocksize > EXT4_MAX_BLOCK_SIZE) {
3413                 ext4_msg(sb, KERN_ERR,
3414                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3415                          blocksize, le32_to_cpu(es->s_log_block_size));
3416                 goto failed_mount;
3417         }
3418         if (le32_to_cpu(es->s_log_block_size) >
3419             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3420                 ext4_msg(sb, KERN_ERR,
3421                          "Invalid log block size: %u",
3422                          le32_to_cpu(es->s_log_block_size));
3423                 goto failed_mount;
3424         }
3425
3426         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3427                 ext4_msg(sb, KERN_ERR,
3428                          "Number of reserved GDT blocks insanely large: %d",
3429                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3430                 goto failed_mount;
3431         }
3432
3433         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3434                 if (blocksize != PAGE_SIZE) {
3435                         ext4_msg(sb, KERN_ERR,
3436                                         "error: unsupported blocksize for dax");
3437                         goto failed_mount;
3438                 }
3439                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3440                         ext4_msg(sb, KERN_ERR,
3441                                         "error: device does not support dax");
3442                         goto failed_mount;
3443                 }
3444         }
3445
3446         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3447                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3448                          es->s_encryption_level);
3449                 goto failed_mount;
3450         }
3451
3452         if (sb->s_blocksize != blocksize) {
3453                 /* Validate the filesystem blocksize */
3454                 if (!sb_set_blocksize(sb, blocksize)) {
3455                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3456                                         blocksize);
3457                         goto failed_mount;
3458                 }
3459
3460                 brelse(bh);
3461                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3462                 offset = do_div(logical_sb_block, blocksize);
3463                 bh = sb_bread_unmovable(sb, logical_sb_block);
3464                 if (!bh) {
3465                         ext4_msg(sb, KERN_ERR,
3466                                "Can't read superblock on 2nd try");
3467                         goto failed_mount;
3468                 }
3469                 es = (struct ext4_super_block *)(bh->b_data + offset);
3470                 sbi->s_es = es;
3471                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3472                         ext4_msg(sb, KERN_ERR,
3473                                "Magic mismatch, very weird!");
3474                         goto failed_mount;
3475                 }
3476         }
3477
3478         has_huge_files = ext4_has_feature_huge_file(sb);
3479         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3480                                                       has_huge_files);
3481         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3482
3483         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3484                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3485                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3486         } else {
3487                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3488                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3489                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3490                     (!is_power_of_2(sbi->s_inode_size)) ||
3491                     (sbi->s_inode_size > blocksize)) {
3492                         ext4_msg(sb, KERN_ERR,
3493                                "unsupported inode size: %d",
3494                                sbi->s_inode_size);
3495                         goto failed_mount;
3496                 }
3497                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3498                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3499         }
3500
3501         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3502         if (ext4_has_feature_64bit(sb)) {
3503                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3504                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3505                     !is_power_of_2(sbi->s_desc_size)) {
3506                         ext4_msg(sb, KERN_ERR,
3507                                "unsupported descriptor size %lu",
3508                                sbi->s_desc_size);
3509                         goto failed_mount;
3510                 }
3511         } else
3512                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3513
3514         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3515         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3516
3517         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3518         if (sbi->s_inodes_per_block == 0)
3519                 goto cantfind_ext4;
3520         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3521             sbi->s_inodes_per_group > blocksize * 8) {
3522                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3523                          sbi->s_blocks_per_group);
3524                 goto failed_mount;
3525         }
3526         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3527                                         sbi->s_inodes_per_block;
3528         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3529         sbi->s_sbh = bh;
3530         sbi->s_mount_state = le16_to_cpu(es->s_state);
3531         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3532         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3533
3534         for (i = 0; i < 4; i++)
3535                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3536         sbi->s_def_hash_version = es->s_def_hash_version;
3537         if (ext4_has_feature_dir_index(sb)) {
3538                 i = le32_to_cpu(es->s_flags);
3539                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3540                         sbi->s_hash_unsigned = 3;
3541                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3542 #ifdef __CHAR_UNSIGNED__
3543                         if (!(sb->s_flags & MS_RDONLY))
3544                                 es->s_flags |=
3545                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3546                         sbi->s_hash_unsigned = 3;
3547 #else
3548                         if (!(sb->s_flags & MS_RDONLY))
3549                                 es->s_flags |=
3550                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3551 #endif
3552                 }
3553         }
3554
3555         /* Handle clustersize */
3556         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3557         has_bigalloc = ext4_has_feature_bigalloc(sb);
3558         if (has_bigalloc) {
3559                 if (clustersize < blocksize) {
3560                         ext4_msg(sb, KERN_ERR,
3561                                  "cluster size (%d) smaller than "
3562                                  "block size (%d)", clustersize, blocksize);
3563                         goto failed_mount;
3564                 }
3565                 if (le32_to_cpu(es->s_log_cluster_size) >
3566                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3567                         ext4_msg(sb, KERN_ERR,
3568                                  "Invalid log cluster size: %u",
3569                                  le32_to_cpu(es->s_log_cluster_size));
3570                         goto failed_mount;
3571                 }
3572                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3573                         le32_to_cpu(es->s_log_block_size);
3574                 sbi->s_clusters_per_group =
3575                         le32_to_cpu(es->s_clusters_per_group);
3576                 if (sbi->s_clusters_per_group > blocksize * 8) {
3577                         ext4_msg(sb, KERN_ERR,
3578                                  "#clusters per group too big: %lu",
3579                                  sbi->s_clusters_per_group);
3580                         goto failed_mount;
3581                 }
3582                 if (sbi->s_blocks_per_group !=
3583                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3584                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3585                                  "clusters per group (%lu) inconsistent",
3586                                  sbi->s_blocks_per_group,
3587                                  sbi->s_clusters_per_group);
3588                         goto failed_mount;
3589                 }
3590         } else {
3591                 if (clustersize != blocksize) {
3592                         ext4_warning(sb, "fragment/cluster size (%d) != "
3593                                      "block size (%d)", clustersize,
3594                                      blocksize);
3595                         clustersize = blocksize;
3596                 }
3597                 if (sbi->s_blocks_per_group > blocksize * 8) {
3598                         ext4_msg(sb, KERN_ERR,
3599                                  "#blocks per group too big: %lu",
3600                                  sbi->s_blocks_per_group);
3601                         goto failed_mount;
3602                 }
3603                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3604                 sbi->s_cluster_bits = 0;
3605         }
3606         sbi->s_cluster_ratio = clustersize / blocksize;
3607
3608         /* Do we have standard group size of clustersize * 8 blocks ? */
3609         if (sbi->s_blocks_per_group == clustersize << 3)
3610                 set_opt2(sb, STD_GROUP_SIZE);
3611
3612         /*
3613          * Test whether we have more sectors than will fit in sector_t,
3614          * and whether the max offset is addressable by the page cache.
3615          */
3616         err = generic_check_addressable(sb->s_blocksize_bits,
3617                                         ext4_blocks_count(es));
3618         if (err) {
3619                 ext4_msg(sb, KERN_ERR, "filesystem"
3620                          " too large to mount safely on this system");
3621                 if (sizeof(sector_t) < 8)
3622                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3623                 goto failed_mount;
3624         }
3625
3626         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3627                 goto cantfind_ext4;
3628
3629         /* check blocks count against device size */
3630         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3631         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3632                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3633                        "exceeds size of device (%llu blocks)",
3634                        ext4_blocks_count(es), blocks_count);
3635                 goto failed_mount;
3636         }
3637
3638         /*
3639          * It makes no sense for the first data block to be beyond the end
3640          * of the filesystem.
3641          */
3642         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3643                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3644                          "block %u is beyond end of filesystem (%llu)",
3645                          le32_to_cpu(es->s_first_data_block),
3646                          ext4_blocks_count(es));
3647                 goto failed_mount;
3648         }
3649         blocks_count = (ext4_blocks_count(es) -
3650                         le32_to_cpu(es->s_first_data_block) +
3651                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3652         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3653         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3654                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3655                        "(block count %llu, first data block %u, "
3656                        "blocks per group %lu)", sbi->s_groups_count,
3657                        ext4_blocks_count(es),
3658                        le32_to_cpu(es->s_first_data_block),
3659                        EXT4_BLOCKS_PER_GROUP(sb));
3660                 goto failed_mount;
3661         }
3662         sbi->s_groups_count = blocks_count;
3663         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3664                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3665         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3666                    EXT4_DESC_PER_BLOCK(sb);
3667         if (ext4_has_feature_meta_bg(sb)) {
3668                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3669                         ext4_msg(sb, KERN_WARNING,
3670                                  "first meta block group too large: %u "
3671                                  "(group descriptor block count %u)",
3672                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3673                         goto failed_mount;
3674                 }
3675         }
3676         sbi->s_group_desc = ext4_kvmalloc(db_count *
3677                                           sizeof(struct buffer_head *),
3678                                           GFP_KERNEL);
3679         if (sbi->s_group_desc == NULL) {
3680                 ext4_msg(sb, KERN_ERR, "not enough memory");
3681                 ret = -ENOMEM;
3682                 goto failed_mount;
3683         }
3684
3685         bgl_lock_init(sbi->s_blockgroup_lock);
3686
3687         for (i = 0; i < db_count; i++) {
3688                 block = descriptor_loc(sb, logical_sb_block, i);
3689                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3690                 if (!sbi->s_group_desc[i]) {
3691                         ext4_msg(sb, KERN_ERR,
3692                                "can't read group descriptor %d", i);
3693                         db_count = i;
3694                         goto failed_mount2;
3695                 }
3696         }
3697         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3698                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3699                 ret = -EFSCORRUPTED;
3700                 goto failed_mount2;
3701         }
3702
3703         sbi->s_gdb_count = db_count;
3704         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3705         spin_lock_init(&sbi->s_next_gen_lock);
3706
3707         setup_timer(&sbi->s_err_report, print_daily_error_info,
3708                 (unsigned long) sb);
3709
3710         /* Register extent status tree shrinker */
3711         if (ext4_es_register_shrinker(sbi))
3712                 goto failed_mount3;
3713
3714         sbi->s_stripe = ext4_get_stripe_size(sbi);
3715         sbi->s_extent_max_zeroout_kb = 32;
3716
3717         /*
3718          * set up enough so that it can read an inode
3719          */
3720         sb->s_op = &ext4_sops;
3721         sb->s_export_op = &ext4_export_ops;
3722         sb->s_xattr = ext4_xattr_handlers;
3723 #ifdef CONFIG_QUOTA
3724         sb->dq_op = &ext4_quota_operations;
3725         if (ext4_has_feature_quota(sb))
3726                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3727         else
3728                 sb->s_qcop = &ext4_qctl_operations;
3729         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3730 #endif
3731         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3732
3733         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3734         mutex_init(&sbi->s_orphan_lock);
3735
3736         sb->s_root = NULL;
3737
3738         needs_recovery = (es->s_last_orphan != 0 ||
3739                           ext4_has_feature_journal_needs_recovery(sb));
3740
3741         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3742                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3743                         goto failed_mount3a;
3744
3745         /*
3746          * The first inode we look at is the journal inode.  Don't try
3747          * root first: it may be modified in the journal!
3748          */
3749         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3750                 err = ext4_load_journal(sb, es, journal_devnum);
3751                 if (err)
3752                         goto failed_mount3a;
3753         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3754                    ext4_has_feature_journal_needs_recovery(sb)) {
3755                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3756                        "suppressed and not mounted read-only");
3757                 goto failed_mount_wq;
3758         } else {
3759                 /* Nojournal mode, all journal mount options are illegal */
3760                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3761                         ext4_msg(sb, KERN_ERR, "can't mount with "
3762                                  "journal_checksum, fs mounted w/o journal");
3763                         goto failed_mount_wq;
3764                 }
3765                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3766                         ext4_msg(sb, KERN_ERR, "can't mount with "
3767                                  "journal_async_commit, fs mounted w/o journal");
3768                         goto failed_mount_wq;
3769                 }
3770                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3771                         ext4_msg(sb, KERN_ERR, "can't mount with "
3772                                  "commit=%lu, fs mounted w/o journal",
3773                                  sbi->s_commit_interval / HZ);
3774                         goto failed_mount_wq;
3775                 }
3776                 if (EXT4_MOUNT_DATA_FLAGS &
3777                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3778                         ext4_msg(sb, KERN_ERR, "can't mount with "
3779                                  "data=, fs mounted w/o journal");
3780                         goto failed_mount_wq;
3781                 }
3782                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3783                 clear_opt(sb, JOURNAL_CHECKSUM);
3784                 clear_opt(sb, DATA_FLAGS);
3785                 sbi->s_journal = NULL;
3786                 needs_recovery = 0;
3787                 goto no_journal;
3788         }
3789
3790         if (ext4_has_feature_64bit(sb) &&
3791             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3792                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3793                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3794                 goto failed_mount_wq;
3795         }
3796
3797         if (!set_journal_csum_feature_set(sb)) {
3798                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3799                          "feature set");
3800                 goto failed_mount_wq;
3801         }
3802
3803         /* We have now updated the journal if required, so we can
3804          * validate the data journaling mode. */
3805         switch (test_opt(sb, DATA_FLAGS)) {
3806         case 0:
3807                 /* No mode set, assume a default based on the journal
3808                  * capabilities: ORDERED_DATA if the journal can
3809                  * cope, else JOURNAL_DATA
3810                  */
3811                 if (jbd2_journal_check_available_features
3812                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3813                         set_opt(sb, ORDERED_DATA);
3814                 else
3815                         set_opt(sb, JOURNAL_DATA);
3816                 break;
3817
3818         case EXT4_MOUNT_ORDERED_DATA:
3819         case EXT4_MOUNT_WRITEBACK_DATA:
3820                 if (!jbd2_journal_check_available_features
3821                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3822                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3823                                "requested data journaling mode");
3824                         goto failed_mount_wq;
3825                 }
3826         default:
3827                 break;
3828         }
3829         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3830
3831         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3832
3833 no_journal:
3834         if (ext4_mballoc_ready) {
3835                 sbi->s_mb_cache = ext4_xattr_create_cache();
3836                 if (!sbi->s_mb_cache) {
3837                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3838                         goto failed_mount_wq;
3839                 }
3840         }
3841
3842         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3843             (blocksize != PAGE_CACHE_SIZE)) {
3844                 ext4_msg(sb, KERN_ERR,
3845                          "Unsupported blocksize for fs encryption");
3846                 goto failed_mount_wq;
3847         }
3848
3849         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3850             !ext4_has_feature_encrypt(sb)) {
3851                 ext4_set_feature_encrypt(sb);
3852                 ext4_commit_super(sb, 1);
3853         }
3854
3855         /*
3856          * Get the # of file system overhead blocks from the
3857          * superblock if present.
3858          */
3859         if (es->s_overhead_clusters)
3860                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3861         else {
3862                 err = ext4_calculate_overhead(sb);
3863                 if (err)
3864                         goto failed_mount_wq;
3865         }
3866
3867         /*
3868          * The maximum number of concurrent works can be high and
3869          * concurrency isn't really necessary.  Limit it to 1.
3870          */
3871         EXT4_SB(sb)->rsv_conversion_wq =
3872                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3873         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3874                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3875                 ret = -ENOMEM;
3876                 goto failed_mount4;
3877         }
3878
3879         /*
3880          * The jbd2_journal_load will have done any necessary log recovery,
3881          * so we can safely mount the rest of the filesystem now.
3882          */
3883
3884         root = ext4_iget(sb, EXT4_ROOT_INO);
3885         if (IS_ERR(root)) {
3886                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3887                 ret = PTR_ERR(root);
3888                 root = NULL;
3889                 goto failed_mount4;
3890         }
3891         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3892                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3893                 iput(root);
3894                 goto failed_mount4;
3895         }
3896         sb->s_root = d_make_root(root);
3897         if (!sb->s_root) {
3898                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3899                 ret = -ENOMEM;
3900                 goto failed_mount4;
3901         }
3902
3903         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3904                 sb->s_flags |= MS_RDONLY;
3905
3906         /* determine the minimum size of new large inodes, if present */
3907         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3908                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3909                                                      EXT4_GOOD_OLD_INODE_SIZE;
3910                 if (ext4_has_feature_extra_isize(sb)) {
3911                         if (sbi->s_want_extra_isize <
3912                             le16_to_cpu(es->s_want_extra_isize))
3913                                 sbi->s_want_extra_isize =
3914                                         le16_to_cpu(es->s_want_extra_isize);
3915                         if (sbi->s_want_extra_isize <
3916                             le16_to_cpu(es->s_min_extra_isize))
3917                                 sbi->s_want_extra_isize =
3918                                         le16_to_cpu(es->s_min_extra_isize);
3919                 }
3920         }
3921         /* Check if enough inode space is available */
3922         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3923                                                         sbi->s_inode_size) {
3924                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3925                                                        EXT4_GOOD_OLD_INODE_SIZE;
3926                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3927                          "available");
3928         }
3929
3930         ext4_set_resv_clusters(sb);
3931
3932         err = ext4_setup_system_zone(sb);
3933         if (err) {
3934                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3935                          "zone (%d)", err);
3936                 goto failed_mount4a;
3937         }
3938
3939         ext4_ext_init(sb);
3940         err = ext4_mb_init(sb);
3941         if (err) {
3942                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3943                          err);
3944                 goto failed_mount5;
3945         }
3946
3947         block = ext4_count_free_clusters(sb);
3948         ext4_free_blocks_count_set(sbi->s_es,
3949                                    EXT4_C2B(sbi, block));
3950         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3951                                   GFP_KERNEL);
3952         if (!err) {
3953                 unsigned long freei = ext4_count_free_inodes(sb);
3954                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3955                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3956                                           GFP_KERNEL);
3957         }
3958         if (!err)
3959                 err = percpu_counter_init(&sbi->s_dirs_counter,
3960                                           ext4_count_dirs(sb), GFP_KERNEL);
3961         if (!err)
3962                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3963                                           GFP_KERNEL);
3964         if (err) {
3965                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3966                 goto failed_mount6;
3967         }
3968
3969         if (ext4_has_feature_flex_bg(sb))
3970                 if (!ext4_fill_flex_info(sb)) {
3971                         ext4_msg(sb, KERN_ERR,
3972                                "unable to initialize "
3973                                "flex_bg meta info!");
3974                         goto failed_mount6;
3975                 }
3976
3977         err = ext4_register_li_request(sb, first_not_zeroed);
3978         if (err)
3979                 goto failed_mount6;
3980
3981         err = ext4_register_sysfs(sb);
3982         if (err)
3983                 goto failed_mount7;
3984
3985 #ifdef CONFIG_QUOTA
3986         /* Enable quota usage during mount. */
3987         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3988                 err = ext4_enable_quotas(sb);
3989                 if (err)
3990                         goto failed_mount8;
3991         }
3992 #endif  /* CONFIG_QUOTA */
3993
3994         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3995         ext4_orphan_cleanup(sb, es);
3996         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3997         if (needs_recovery) {
3998                 ext4_msg(sb, KERN_INFO, "recovery complete");
3999                 ext4_mark_recovery_complete(sb, es);
4000         }
4001         if (EXT4_SB(sb)->s_journal) {
4002                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4003                         descr = " journalled data mode";
4004                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4005                         descr = " ordered data mode";
4006                 else
4007                         descr = " writeback data mode";
4008         } else
4009                 descr = "out journal";
4010
4011         if (test_opt(sb, DISCARD)) {
4012                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4013                 if (!blk_queue_discard(q))
4014                         ext4_msg(sb, KERN_WARNING,
4015                                  "mounting with \"discard\" option, but "
4016                                  "the device does not support discard");
4017         }
4018
4019         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4020                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4021                          "Opts: %.*s%s%s", descr,
4022                          (int) sizeof(sbi->s_es->s_mount_opts),
4023                          sbi->s_es->s_mount_opts,
4024                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4025
4026         if (es->s_error_count)
4027                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4028
4029         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4030         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4031         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4032         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4033
4034         kfree(orig_data);
4035         return 0;
4036
4037 cantfind_ext4:
4038         if (!silent)
4039                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4040         goto failed_mount;
4041
4042 #ifdef CONFIG_QUOTA
4043 failed_mount8:
4044         ext4_unregister_sysfs(sb);
4045 #endif
4046 failed_mount7:
4047         ext4_unregister_li_request(sb);
4048 failed_mount6:
4049         ext4_mb_release(sb);
4050         if (sbi->s_flex_groups)
4051                 kvfree(sbi->s_flex_groups);
4052         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4053         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4054         percpu_counter_destroy(&sbi->s_dirs_counter);
4055         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4056 failed_mount5:
4057         ext4_ext_release(sb);
4058         ext4_release_system_zone(sb);
4059 failed_mount4a:
4060         dput(sb->s_root);
4061         sb->s_root = NULL;
4062 failed_mount4:
4063         ext4_msg(sb, KERN_ERR, "mount failed");
4064         if (EXT4_SB(sb)->rsv_conversion_wq)
4065                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4066 failed_mount_wq:
4067         if (sbi->s_mb_cache) {
4068                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4069                 sbi->s_mb_cache = NULL;
4070         }
4071         if (sbi->s_journal) {
4072                 jbd2_journal_destroy(sbi->s_journal);
4073                 sbi->s_journal = NULL;
4074         }
4075 failed_mount3a:
4076         ext4_es_unregister_shrinker(sbi);
4077 failed_mount3:
4078         del_timer_sync(&sbi->s_err_report);
4079         if (sbi->s_mmp_tsk)
4080                 kthread_stop(sbi->s_mmp_tsk);
4081 failed_mount2:
4082         for (i = 0; i < db_count; i++)
4083                 brelse(sbi->s_group_desc[i]);
4084         kvfree(sbi->s_group_desc);
4085 failed_mount:
4086         if (sbi->s_chksum_driver)
4087                 crypto_free_shash(sbi->s_chksum_driver);
4088 #ifdef CONFIG_QUOTA
4089         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4090                 kfree(sbi->s_qf_names[i]);
4091 #endif
4092         ext4_blkdev_remove(sbi);
4093         brelse(bh);
4094 out_fail:
4095         sb->s_fs_info = NULL;
4096         kfree(sbi->s_blockgroup_lock);
4097 out_free_base:
4098         kfree(sbi);
4099         kfree(orig_data);
4100         return err ? err : ret;
4101 }
4102
4103 /*
4104  * Setup any per-fs journal parameters now.  We'll do this both on
4105  * initial mount, once the journal has been initialised but before we've
4106  * done any recovery; and again on any subsequent remount.
4107  */
4108 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4109 {
4110         struct ext4_sb_info *sbi = EXT4_SB(sb);
4111
4112         journal->j_commit_interval = sbi->s_commit_interval;
4113         journal->j_min_batch_time = sbi->s_min_batch_time;
4114         journal->j_max_batch_time = sbi->s_max_batch_time;
4115
4116         write_lock(&journal->j_state_lock);
4117         if (test_opt(sb, BARRIER))
4118                 journal->j_flags |= JBD2_BARRIER;
4119         else
4120                 journal->j_flags &= ~JBD2_BARRIER;
4121         if (test_opt(sb, DATA_ERR_ABORT))
4122                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4123         else
4124                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4125         write_unlock(&journal->j_state_lock);
4126 }
4127
4128 static journal_t *ext4_get_journal(struct super_block *sb,
4129                                    unsigned int journal_inum)
4130 {
4131         struct inode *journal_inode;
4132         journal_t *journal;
4133
4134         BUG_ON(!ext4_has_feature_journal(sb));
4135
4136         /* First, test for the existence of a valid inode on disk.  Bad
4137          * things happen if we iget() an unused inode, as the subsequent
4138          * iput() will try to delete it. */
4139
4140         journal_inode = ext4_iget(sb, journal_inum);
4141         if (IS_ERR(journal_inode)) {
4142                 ext4_msg(sb, KERN_ERR, "no journal found");
4143                 return NULL;
4144         }
4145         if (!journal_inode->i_nlink) {
4146                 make_bad_inode(journal_inode);
4147                 iput(journal_inode);
4148                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4149                 return NULL;
4150         }
4151
4152         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4153                   journal_inode, journal_inode->i_size);
4154         if (!S_ISREG(journal_inode->i_mode)) {
4155                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4156                 iput(journal_inode);
4157                 return NULL;
4158         }
4159
4160         journal = jbd2_journal_init_inode(journal_inode);
4161         if (!journal) {
4162                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4163                 iput(journal_inode);
4164                 return NULL;
4165         }
4166         journal->j_private = sb;
4167         ext4_init_journal_params(sb, journal);
4168         return journal;
4169 }
4170
4171 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4172                                        dev_t j_dev)
4173 {
4174         struct buffer_head *bh;
4175         journal_t *journal;
4176         ext4_fsblk_t start;
4177         ext4_fsblk_t len;
4178         int hblock, blocksize;
4179         ext4_fsblk_t sb_block;
4180         unsigned long offset;
4181         struct ext4_super_block *es;
4182         struct block_device *bdev;
4183
4184         BUG_ON(!ext4_has_feature_journal(sb));
4185
4186         bdev = ext4_blkdev_get(j_dev, sb);
4187         if (bdev == NULL)
4188                 return NULL;
4189
4190         blocksize = sb->s_blocksize;
4191         hblock = bdev_logical_block_size(bdev);
4192         if (blocksize < hblock) {
4193                 ext4_msg(sb, KERN_ERR,
4194                         "blocksize too small for journal device");
4195                 goto out_bdev;
4196         }
4197
4198         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4199         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4200         set_blocksize(bdev, blocksize);
4201         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4202                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4203                        "external journal");
4204                 goto out_bdev;
4205         }
4206
4207         es = (struct ext4_super_block *) (bh->b_data + offset);
4208         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4209             !(le32_to_cpu(es->s_feature_incompat) &
4210               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4211                 ext4_msg(sb, KERN_ERR, "external journal has "
4212                                         "bad superblock");
4213                 brelse(bh);
4214                 goto out_bdev;
4215         }
4216
4217         if ((le32_to_cpu(es->s_feature_ro_compat) &
4218              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4219             es->s_checksum != ext4_superblock_csum(sb, es)) {
4220                 ext4_msg(sb, KERN_ERR, "external journal has "
4221                                        "corrupt superblock");
4222                 brelse(bh);
4223                 goto out_bdev;
4224         }
4225
4226         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4227                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4228                 brelse(bh);
4229                 goto out_bdev;
4230         }
4231
4232         len = ext4_blocks_count(es);
4233         start = sb_block + 1;
4234         brelse(bh);     /* we're done with the superblock */
4235
4236         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4237                                         start, len, blocksize);
4238         if (!journal) {
4239                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4240                 goto out_bdev;
4241         }
4242         journal->j_private = sb;
4243         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4244         wait_on_buffer(journal->j_sb_buffer);
4245         if (!buffer_uptodate(journal->j_sb_buffer)) {
4246                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4247                 goto out_journal;
4248         }
4249         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4250                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4251                                         "user (unsupported) - %d",
4252                         be32_to_cpu(journal->j_superblock->s_nr_users));
4253                 goto out_journal;
4254         }
4255         EXT4_SB(sb)->journal_bdev = bdev;
4256         ext4_init_journal_params(sb, journal);
4257         return journal;
4258
4259 out_journal:
4260         jbd2_journal_destroy(journal);
4261 out_bdev:
4262         ext4_blkdev_put(bdev);
4263         return NULL;
4264 }
4265
4266 static int ext4_load_journal(struct super_block *sb,
4267                              struct ext4_super_block *es,
4268                              unsigned long journal_devnum)
4269 {
4270         journal_t *journal;
4271         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4272         dev_t journal_dev;
4273         int err = 0;
4274         int really_read_only;
4275
4276         BUG_ON(!ext4_has_feature_journal(sb));
4277
4278         if (journal_devnum &&
4279             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4280                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4281                         "numbers have changed");
4282                 journal_dev = new_decode_dev(journal_devnum);
4283         } else
4284                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4285
4286         really_read_only = bdev_read_only(sb->s_bdev);
4287
4288         /*
4289          * Are we loading a blank journal or performing recovery after a
4290          * crash?  For recovery, we need to check in advance whether we
4291          * can get read-write access to the device.
4292          */
4293         if (ext4_has_feature_journal_needs_recovery(sb)) {
4294                 if (sb->s_flags & MS_RDONLY) {
4295                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4296                                         "required on readonly filesystem");
4297                         if (really_read_only) {
4298                                 ext4_msg(sb, KERN_ERR, "write access "
4299                                         "unavailable, cannot proceed");
4300                                 return -EROFS;
4301                         }
4302                         ext4_msg(sb, KERN_INFO, "write access will "
4303                                "be enabled during recovery");
4304                 }
4305         }
4306
4307         if (journal_inum && journal_dev) {
4308                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4309                        "and inode journals!");
4310                 return -EINVAL;
4311         }
4312
4313         if (journal_inum) {
4314                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4315                         return -EINVAL;
4316         } else {
4317                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4318                         return -EINVAL;
4319         }
4320
4321         if (!(journal->j_flags & JBD2_BARRIER))
4322                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4323
4324         if (!ext4_has_feature_journal_needs_recovery(sb))
4325                 err = jbd2_journal_wipe(journal, !really_read_only);
4326         if (!err) {
4327                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4328                 if (save)
4329                         memcpy(save, ((char *) es) +
4330                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4331                 err = jbd2_journal_load(journal);
4332                 if (save)
4333                         memcpy(((char *) es) + EXT4_S_ERR_START,
4334                                save, EXT4_S_ERR_LEN);
4335                 kfree(save);
4336         }
4337
4338         if (err) {
4339                 ext4_msg(sb, KERN_ERR, "error loading journal");
4340                 jbd2_journal_destroy(journal);
4341                 return err;
4342         }
4343
4344         EXT4_SB(sb)->s_journal = journal;
4345         ext4_clear_journal_err(sb, es);
4346
4347         if (!really_read_only && journal_devnum &&
4348             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4349                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4350
4351                 /* Make sure we flush the recovery flag to disk. */
4352                 ext4_commit_super(sb, 1);
4353         }
4354
4355         return 0;
4356 }
4357
4358 static int ext4_commit_super(struct super_block *sb, int sync)
4359 {
4360         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4361         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4362         int error = 0;
4363
4364         if (!sbh || block_device_ejected(sb))
4365                 return error;
4366         if (buffer_write_io_error(sbh)) {
4367                 /*
4368                  * Oh, dear.  A previous attempt to write the
4369                  * superblock failed.  This could happen because the
4370                  * USB device was yanked out.  Or it could happen to
4371                  * be a transient write error and maybe the block will
4372                  * be remapped.  Nothing we can do but to retry the
4373                  * write and hope for the best.
4374                  */
4375                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4376                        "superblock detected");
4377                 clear_buffer_write_io_error(sbh);
4378                 set_buffer_uptodate(sbh);
4379         }
4380         /*
4381          * If the file system is mounted read-only, don't update the
4382          * superblock write time.  This avoids updating the superblock
4383          * write time when we are mounting the root file system
4384          * read/only but we need to replay the journal; at that point,
4385          * for people who are east of GMT and who make their clock
4386          * tick in localtime for Windows bug-for-bug compatibility,
4387          * the clock is set in the future, and this will cause e2fsck
4388          * to complain and force a full file system check.
4389          */
4390         if (!(sb->s_flags & MS_RDONLY))
4391                 es->s_wtime = cpu_to_le32(get_seconds());
4392         if (sb->s_bdev->bd_part)
4393                 es->s_kbytes_written =
4394                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4395                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4396                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4397         else
4398                 es->s_kbytes_written =
4399                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4400         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4401                 ext4_free_blocks_count_set(es,
4402                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4403                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4404         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4405                 es->s_free_inodes_count =
4406                         cpu_to_le32(percpu_counter_sum_positive(
4407                                 &EXT4_SB(sb)->s_freeinodes_counter));
4408         BUFFER_TRACE(sbh, "marking dirty");
4409         ext4_superblock_csum_set(sb);
4410         mark_buffer_dirty(sbh);
4411         if (sync) {
4412                 error = __sync_dirty_buffer(sbh,
4413                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4414                 if (error)
4415                         return error;
4416
4417                 error = buffer_write_io_error(sbh);
4418                 if (error) {
4419                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4420                                "superblock");
4421                         clear_buffer_write_io_error(sbh);
4422                         set_buffer_uptodate(sbh);
4423                 }
4424         }
4425         return error;
4426 }
4427
4428 /*
4429  * Have we just finished recovery?  If so, and if we are mounting (or
4430  * remounting) the filesystem readonly, then we will end up with a
4431  * consistent fs on disk.  Record that fact.
4432  */
4433 static void ext4_mark_recovery_complete(struct super_block *sb,
4434                                         struct ext4_super_block *es)
4435 {
4436         journal_t *journal = EXT4_SB(sb)->s_journal;
4437
4438         if (!ext4_has_feature_journal(sb)) {
4439                 BUG_ON(journal != NULL);
4440                 return;
4441         }
4442         jbd2_journal_lock_updates(journal);
4443         if (jbd2_journal_flush(journal) < 0)
4444                 goto out;
4445
4446         if (ext4_has_feature_journal_needs_recovery(sb) &&
4447             sb->s_flags & MS_RDONLY) {
4448                 ext4_clear_feature_journal_needs_recovery(sb);
4449                 ext4_commit_super(sb, 1);
4450         }
4451
4452 out:
4453         jbd2_journal_unlock_updates(journal);
4454 }
4455
4456 /*
4457  * If we are mounting (or read-write remounting) a filesystem whose journal
4458  * has recorded an error from a previous lifetime, move that error to the
4459  * main filesystem now.
4460  */
4461 static void ext4_clear_journal_err(struct super_block *sb,
4462                                    struct ext4_super_block *es)
4463 {
4464         journal_t *journal;
4465         int j_errno;
4466         const char *errstr;
4467
4468         BUG_ON(!ext4_has_feature_journal(sb));
4469
4470         journal = EXT4_SB(sb)->s_journal;
4471
4472         /*
4473          * Now check for any error status which may have been recorded in the
4474          * journal by a prior ext4_error() or ext4_abort()
4475          */
4476
4477         j_errno = jbd2_journal_errno(journal);
4478         if (j_errno) {
4479                 char nbuf[16];
4480
4481                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4482                 ext4_warning(sb, "Filesystem error recorded "
4483                              "from previous mount: %s", errstr);
4484                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4485
4486                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4487                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4488                 ext4_commit_super(sb, 1);
4489
4490                 jbd2_journal_clear_err(journal);
4491                 jbd2_journal_update_sb_errno(journal);
4492         }
4493 }
4494
4495 /*
4496  * Force the running and committing transactions to commit,
4497  * and wait on the commit.
4498  */
4499 int ext4_force_commit(struct super_block *sb)
4500 {
4501         journal_t *journal;
4502
4503         if (sb->s_flags & MS_RDONLY)
4504                 return 0;
4505
4506         journal = EXT4_SB(sb)->s_journal;
4507         return ext4_journal_force_commit(journal);
4508 }
4509
4510 static int ext4_sync_fs(struct super_block *sb, int wait)
4511 {
4512         int ret = 0;
4513         tid_t target;
4514         bool needs_barrier = false;
4515         struct ext4_sb_info *sbi = EXT4_SB(sb);
4516
4517         trace_ext4_sync_fs(sb, wait);
4518         flush_workqueue(sbi->rsv_conversion_wq);
4519         /*
4520          * Writeback quota in non-journalled quota case - journalled quota has
4521          * no dirty dquots
4522          */
4523         dquot_writeback_dquots(sb, -1);
4524         /*
4525          * Data writeback is possible w/o journal transaction, so barrier must
4526          * being sent at the end of the function. But we can skip it if
4527          * transaction_commit will do it for us.
4528          */
4529         if (sbi->s_journal) {
4530                 target = jbd2_get_latest_transaction(sbi->s_journal);
4531                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4532                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4533                         needs_barrier = true;
4534
4535                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4536                         if (wait)
4537                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4538                                                            target);
4539                 }
4540         } else if (wait && test_opt(sb, BARRIER))
4541                 needs_barrier = true;
4542         if (needs_barrier) {
4543                 int err;
4544                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4545                 if (!ret)
4546                         ret = err;
4547         }
4548
4549         return ret;
4550 }
4551
4552 /*
4553  * LVM calls this function before a (read-only) snapshot is created.  This
4554  * gives us a chance to flush the journal completely and mark the fs clean.
4555  *
4556  * Note that only this function cannot bring a filesystem to be in a clean
4557  * state independently. It relies on upper layer to stop all data & metadata
4558  * modifications.
4559  */
4560 static int ext4_freeze(struct super_block *sb)
4561 {
4562         int error = 0;
4563         journal_t *journal;
4564
4565         if (sb->s_flags & MS_RDONLY)
4566                 return 0;
4567
4568         journal = EXT4_SB(sb)->s_journal;
4569
4570         if (journal) {
4571                 /* Now we set up the journal barrier. */
4572                 jbd2_journal_lock_updates(journal);
4573
4574                 /*
4575                  * Don't clear the needs_recovery flag if we failed to
4576                  * flush the journal.
4577                  */
4578                 error = jbd2_journal_flush(journal);
4579                 if (error < 0)
4580                         goto out;
4581
4582                 /* Journal blocked and flushed, clear needs_recovery flag. */
4583                 ext4_clear_feature_journal_needs_recovery(sb);
4584         }
4585
4586         error = ext4_commit_super(sb, 1);
4587 out:
4588         if (journal)
4589                 /* we rely on upper layer to stop further updates */
4590                 jbd2_journal_unlock_updates(journal);
4591         return error;
4592 }
4593
4594 /*
4595  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4596  * flag here, even though the filesystem is not technically dirty yet.
4597  */
4598 static int ext4_unfreeze(struct super_block *sb)
4599 {
4600         if (sb->s_flags & MS_RDONLY)
4601                 return 0;
4602
4603         if (EXT4_SB(sb)->s_journal) {
4604                 /* Reset the needs_recovery flag before the fs is unlocked. */
4605                 ext4_set_feature_journal_needs_recovery(sb);
4606         }
4607
4608         ext4_commit_super(sb, 1);
4609         return 0;
4610 }
4611
4612 /*
4613  * Structure to save mount options for ext4_remount's benefit
4614  */
4615 struct ext4_mount_options {
4616         unsigned long s_mount_opt;
4617         unsigned long s_mount_opt2;
4618         kuid_t s_resuid;
4619         kgid_t s_resgid;
4620         unsigned long s_commit_interval;
4621         u32 s_min_batch_time, s_max_batch_time;
4622 #ifdef CONFIG_QUOTA
4623         int s_jquota_fmt;
4624         char *s_qf_names[EXT4_MAXQUOTAS];
4625 #endif
4626 };
4627
4628 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4629 {
4630         struct ext4_super_block *es;
4631         struct ext4_sb_info *sbi = EXT4_SB(sb);
4632         unsigned long old_sb_flags;
4633         struct ext4_mount_options old_opts;
4634         int enable_quota = 0;
4635         ext4_group_t g;
4636         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4637         int err = 0;
4638 #ifdef CONFIG_QUOTA
4639         int i, j;
4640 #endif
4641         char *orig_data = kstrdup(data, GFP_KERNEL);
4642
4643         /* Store the original options */
4644         old_sb_flags = sb->s_flags;
4645         old_opts.s_mount_opt = sbi->s_mount_opt;
4646         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4647         old_opts.s_resuid = sbi->s_resuid;
4648         old_opts.s_resgid = sbi->s_resgid;
4649         old_opts.s_commit_interval = sbi->s_commit_interval;
4650         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4651         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4652 #ifdef CONFIG_QUOTA
4653         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4654         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4655                 if (sbi->s_qf_names[i]) {
4656                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4657                                                          GFP_KERNEL);
4658                         if (!old_opts.s_qf_names[i]) {
4659                                 for (j = 0; j < i; j++)
4660                                         kfree(old_opts.s_qf_names[j]);
4661                                 kfree(orig_data);
4662                                 return -ENOMEM;
4663                         }
4664                 } else
4665                         old_opts.s_qf_names[i] = NULL;
4666 #endif
4667         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4668                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4669
4670         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4671                 err = -EINVAL;
4672                 goto restore_opts;
4673         }
4674
4675         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4676             test_opt(sb, JOURNAL_CHECKSUM)) {
4677                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4678                          "during remount not supported; ignoring");
4679                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4680         }
4681
4682         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4683                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4684                         ext4_msg(sb, KERN_ERR, "can't mount with "
4685                                  "both data=journal and delalloc");
4686                         err = -EINVAL;
4687                         goto restore_opts;
4688                 }
4689                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4690                         ext4_msg(sb, KERN_ERR, "can't mount with "
4691                                  "both data=journal and dioread_nolock");
4692                         err = -EINVAL;
4693                         goto restore_opts;
4694                 }
4695                 if (test_opt(sb, DAX)) {
4696                         ext4_msg(sb, KERN_ERR, "can't mount with "
4697                                  "both data=journal and dax");
4698                         err = -EINVAL;
4699                         goto restore_opts;
4700                 }
4701         }
4702
4703         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4704                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4705                         "dax flag with busy inodes while remounting");
4706                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4707         }
4708
4709         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4710                 ext4_abort(sb, "Abort forced by user");
4711
4712         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4713                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4714
4715         es = sbi->s_es;
4716
4717         if (sbi->s_journal) {
4718                 ext4_init_journal_params(sb, sbi->s_journal);
4719                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4720         }
4721
4722         if (*flags & MS_LAZYTIME)
4723                 sb->s_flags |= MS_LAZYTIME;
4724
4725         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4726                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4727                         err = -EROFS;
4728                         goto restore_opts;
4729                 }
4730
4731                 if (*flags & MS_RDONLY) {
4732                         err = sync_filesystem(sb);
4733                         if (err < 0)
4734                                 goto restore_opts;
4735                         err = dquot_suspend(sb, -1);
4736                         if (err < 0)
4737                                 goto restore_opts;
4738
4739                         /*
4740                          * First of all, the unconditional stuff we have to do
4741                          * to disable replay of the journal when we next remount
4742                          */
4743                         sb->s_flags |= MS_RDONLY;
4744
4745                         /*
4746                          * OK, test if we are remounting a valid rw partition
4747                          * readonly, and if so set the rdonly flag and then
4748                          * mark the partition as valid again.
4749                          */
4750                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4751                             (sbi->s_mount_state & EXT4_VALID_FS))
4752                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4753
4754                         if (sbi->s_journal)
4755                                 ext4_mark_recovery_complete(sb, es);
4756                 } else {
4757                         /* Make sure we can mount this feature set readwrite */
4758                         if (ext4_has_feature_readonly(sb) ||
4759                             !ext4_feature_set_ok(sb, 0)) {
4760                                 err = -EROFS;
4761                                 goto restore_opts;
4762                         }
4763                         /*
4764                          * Make sure the group descriptor checksums
4765                          * are sane.  If they aren't, refuse to remount r/w.
4766                          */
4767                         for (g = 0; g < sbi->s_groups_count; g++) {
4768                                 struct ext4_group_desc *gdp =
4769                                         ext4_get_group_desc(sb, g, NULL);
4770
4771                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4772                                         ext4_msg(sb, KERN_ERR,
4773                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4774                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4775                                                le16_to_cpu(gdp->bg_checksum));
4776                                         err = -EFSBADCRC;
4777                                         goto restore_opts;
4778                                 }
4779                         }
4780
4781                         /*
4782                          * If we have an unprocessed orphan list hanging
4783                          * around from a previously readonly bdev mount,
4784                          * require a full umount/remount for now.
4785                          */
4786                         if (es->s_last_orphan) {
4787                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4788                                        "remount RDWR because of unprocessed "
4789                                        "orphan inode list.  Please "
4790                                        "umount/remount instead");
4791                                 err = -EINVAL;
4792                                 goto restore_opts;
4793                         }
4794
4795                         /*
4796                          * Mounting a RDONLY partition read-write, so reread
4797                          * and store the current valid flag.  (It may have
4798                          * been changed by e2fsck since we originally mounted
4799                          * the partition.)
4800                          */
4801                         if (sbi->s_journal)
4802                                 ext4_clear_journal_err(sb, es);
4803                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4804                         if (!ext4_setup_super(sb, es, 0))
4805                                 sb->s_flags &= ~MS_RDONLY;
4806                         if (ext4_has_feature_mmp(sb))
4807                                 if (ext4_multi_mount_protect(sb,
4808                                                 le64_to_cpu(es->s_mmp_block))) {
4809                                         err = -EROFS;
4810                                         goto restore_opts;
4811                                 }
4812                         enable_quota = 1;
4813                 }
4814         }
4815
4816         /*
4817          * Reinitialize lazy itable initialization thread based on
4818          * current settings
4819          */
4820         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4821                 ext4_unregister_li_request(sb);
4822         else {
4823                 ext4_group_t first_not_zeroed;
4824                 first_not_zeroed = ext4_has_uninit_itable(sb);
4825                 ext4_register_li_request(sb, first_not_zeroed);
4826         }
4827
4828         ext4_setup_system_zone(sb);
4829         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4830                 ext4_commit_super(sb, 1);
4831
4832 #ifdef CONFIG_QUOTA
4833         /* Release old quota file names */
4834         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4835                 kfree(old_opts.s_qf_names[i]);
4836         if (enable_quota) {
4837                 if (sb_any_quota_suspended(sb))
4838                         dquot_resume(sb, -1);
4839                 else if (ext4_has_feature_quota(sb)) {
4840                         err = ext4_enable_quotas(sb);
4841                         if (err)
4842                                 goto restore_opts;
4843                 }
4844         }
4845 #endif
4846
4847         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4848         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4849         kfree(orig_data);
4850         return 0;
4851
4852 restore_opts:
4853         sb->s_flags = old_sb_flags;
4854         sbi->s_mount_opt = old_opts.s_mount_opt;
4855         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4856         sbi->s_resuid = old_opts.s_resuid;
4857         sbi->s_resgid = old_opts.s_resgid;
4858         sbi->s_commit_interval = old_opts.s_commit_interval;
4859         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4860         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4861 #ifdef CONFIG_QUOTA
4862         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4863         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4864                 kfree(sbi->s_qf_names[i]);
4865                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4866         }
4867 #endif
4868         kfree(orig_data);
4869         return err;
4870 }
4871
4872 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4873 {
4874         struct super_block *sb = dentry->d_sb;
4875         struct ext4_sb_info *sbi = EXT4_SB(sb);
4876         struct ext4_super_block *es = sbi->s_es;
4877         ext4_fsblk_t overhead = 0, resv_blocks;
4878         u64 fsid;
4879         s64 bfree;
4880         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4881
4882         if (!test_opt(sb, MINIX_DF))
4883                 overhead = sbi->s_overhead;
4884
4885         buf->f_type = EXT4_SUPER_MAGIC;
4886         buf->f_bsize = sb->s_blocksize;
4887         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4888         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4889                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4890         /* prevent underflow in case that few free space is available */
4891         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4892         buf->f_bavail = buf->f_bfree -
4893                         (ext4_r_blocks_count(es) + resv_blocks);
4894         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4895                 buf->f_bavail = 0;
4896         buf->f_files = le32_to_cpu(es->s_inodes_count);
4897         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4898         buf->f_namelen = EXT4_NAME_LEN;
4899         fsid = le64_to_cpup((void *)es->s_uuid) ^
4900                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4901         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4902         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4903
4904         return 0;
4905 }
4906
4907 /* Helper function for writing quotas on sync - we need to start transaction
4908  * before quota file is locked for write. Otherwise the are possible deadlocks:
4909  * Process 1                         Process 2
4910  * ext4_create()                     quota_sync()
4911  *   jbd2_journal_start()                  write_dquot()
4912  *   dquot_initialize()                         down(dqio_mutex)
4913  *     down(dqio_mutex)                    jbd2_journal_start()
4914  *
4915  */
4916
4917 #ifdef CONFIG_QUOTA
4918
4919 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4920 {
4921         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4922 }
4923
4924 static int ext4_write_dquot(struct dquot *dquot)
4925 {
4926         int ret, err;
4927         handle_t *handle;
4928         struct inode *inode;
4929
4930         inode = dquot_to_inode(dquot);
4931         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4932                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4933         if (IS_ERR(handle))
4934                 return PTR_ERR(handle);
4935         ret = dquot_commit(dquot);
4936         err = ext4_journal_stop(handle);
4937         if (!ret)
4938                 ret = err;
4939         return ret;
4940 }
4941
4942 static int ext4_acquire_dquot(struct dquot *dquot)
4943 {
4944         int ret, err;
4945         handle_t *handle;
4946
4947         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4948                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4949         if (IS_ERR(handle))
4950                 return PTR_ERR(handle);
4951         ret = dquot_acquire(dquot);
4952         err = ext4_journal_stop(handle);
4953         if (!ret)
4954                 ret = err;
4955         return ret;
4956 }
4957
4958 static int ext4_release_dquot(struct dquot *dquot)
4959 {
4960         int ret, err;
4961         handle_t *handle;
4962
4963         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4964                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4965         if (IS_ERR(handle)) {
4966                 /* Release dquot anyway to avoid endless cycle in dqput() */
4967                 dquot_release(dquot);
4968                 return PTR_ERR(handle);
4969         }
4970         ret = dquot_release(dquot);
4971         err = ext4_journal_stop(handle);
4972         if (!ret)
4973                 ret = err;
4974         return ret;
4975 }
4976
4977 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4978 {
4979         struct super_block *sb = dquot->dq_sb;
4980         struct ext4_sb_info *sbi = EXT4_SB(sb);
4981
4982         /* Are we journaling quotas? */
4983         if (ext4_has_feature_quota(sb) ||
4984             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4985                 dquot_mark_dquot_dirty(dquot);
4986                 return ext4_write_dquot(dquot);
4987         } else {
4988                 return dquot_mark_dquot_dirty(dquot);
4989         }
4990 }
4991
4992 static int ext4_write_info(struct super_block *sb, int type)
4993 {
4994         int ret, err;
4995         handle_t *handle;
4996
4997         /* Data block + inode block */
4998         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
4999         if (IS_ERR(handle))
5000                 return PTR_ERR(handle);
5001         ret = dquot_commit_info(sb, type);
5002         err = ext4_journal_stop(handle);
5003         if (!ret)
5004                 ret = err;
5005         return ret;
5006 }
5007
5008 /*
5009  * Turn on quotas during mount time - we need to find
5010  * the quota file and such...
5011  */
5012 static int ext4_quota_on_mount(struct super_block *sb, int type)
5013 {
5014         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5015                                         EXT4_SB(sb)->s_jquota_fmt, type);
5016 }
5017
5018 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5019 {
5020         struct ext4_inode_info *ei = EXT4_I(inode);
5021
5022         /* The first argument of lockdep_set_subclass has to be
5023          * *exactly* the same as the argument to init_rwsem() --- in
5024          * this case, in init_once() --- or lockdep gets unhappy
5025          * because the name of the lock is set using the
5026          * stringification of the argument to init_rwsem().
5027          */
5028         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5029         lockdep_set_subclass(&ei->i_data_sem, subclass);
5030 }
5031
5032 /*
5033  * Standard function to be called on quota_on
5034  */
5035 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5036                          struct path *path)
5037 {
5038         int err;
5039
5040         if (!test_opt(sb, QUOTA))
5041                 return -EINVAL;
5042
5043         /* Quotafile not on the same filesystem? */
5044         if (path->dentry->d_sb != sb)
5045                 return -EXDEV;
5046         /* Journaling quota? */
5047         if (EXT4_SB(sb)->s_qf_names[type]) {
5048                 /* Quotafile not in fs root? */
5049                 if (path->dentry->d_parent != sb->s_root)
5050                         ext4_msg(sb, KERN_WARNING,
5051                                 "Quota file not on filesystem root. "
5052                                 "Journaled quota will not work");
5053         }
5054
5055         /*
5056          * When we journal data on quota file, we have to flush journal to see
5057          * all updates to the file when we bypass pagecache...
5058          */
5059         if (EXT4_SB(sb)->s_journal &&
5060             ext4_should_journal_data(d_inode(path->dentry))) {
5061                 /*
5062                  * We don't need to lock updates but journal_flush() could
5063                  * otherwise be livelocked...
5064                  */
5065                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5066                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5067                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5068                 if (err)
5069                         return err;
5070         }
5071         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5072         err = dquot_quota_on(sb, type, format_id, path);
5073         if (err)
5074                 lockdep_set_quota_inode(path->dentry->d_inode,
5075                                              I_DATA_SEM_NORMAL);
5076         return err;
5077 }
5078
5079 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5080                              unsigned int flags)
5081 {
5082         int err;
5083         struct inode *qf_inode;
5084         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5085                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5086                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5087         };
5088
5089         BUG_ON(!ext4_has_feature_quota(sb));
5090
5091         if (!qf_inums[type])
5092                 return -EPERM;
5093
5094         qf_inode = ext4_iget(sb, qf_inums[type]);
5095         if (IS_ERR(qf_inode)) {
5096                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5097                 return PTR_ERR(qf_inode);
5098         }
5099
5100         /* Don't account quota for quota files to avoid recursion */
5101         qf_inode->i_flags |= S_NOQUOTA;
5102         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5103         err = dquot_enable(qf_inode, type, format_id, flags);
5104         iput(qf_inode);
5105         if (err)
5106                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5107
5108         return err;
5109 }
5110
5111 /* Enable usage tracking for all quota types. */
5112 static int ext4_enable_quotas(struct super_block *sb)
5113 {
5114         int type, err = 0;
5115         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5116                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5117                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5118         };
5119
5120         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5121         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5122                 if (qf_inums[type]) {
5123                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5124                                                 DQUOT_USAGE_ENABLED);
5125                         if (err) {
5126                                 ext4_warning(sb,
5127                                         "Failed to enable quota tracking "
5128                                         "(type=%d, err=%d). Please run "
5129                                         "e2fsck to fix.", type, err);
5130                                 return err;
5131                         }
5132                 }
5133         }
5134         return 0;
5135 }
5136
5137 static int ext4_quota_off(struct super_block *sb, int type)
5138 {
5139         struct inode *inode = sb_dqopt(sb)->files[type];
5140         handle_t *handle;
5141
5142         /* Force all delayed allocation blocks to be allocated.
5143          * Caller already holds s_umount sem */
5144         if (test_opt(sb, DELALLOC))
5145                 sync_filesystem(sb);
5146
5147         if (!inode)
5148                 goto out;
5149
5150         /* Update modification times of quota files when userspace can
5151          * start looking at them */
5152         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5153         if (IS_ERR(handle))
5154                 goto out;
5155         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5156         ext4_mark_inode_dirty(handle, inode);
5157         ext4_journal_stop(handle);
5158
5159 out:
5160         return dquot_quota_off(sb, type);
5161 }
5162
5163 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5164  * acquiring the locks... As quota files are never truncated and quota code
5165  * itself serializes the operations (and no one else should touch the files)
5166  * we don't have to be afraid of races */
5167 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5168                                size_t len, loff_t off)
5169 {
5170         struct inode *inode = sb_dqopt(sb)->files[type];
5171         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5172         int offset = off & (sb->s_blocksize - 1);
5173         int tocopy;
5174         size_t toread;
5175         struct buffer_head *bh;
5176         loff_t i_size = i_size_read(inode);
5177
5178         if (off > i_size)
5179                 return 0;
5180         if (off+len > i_size)
5181                 len = i_size-off;
5182         toread = len;
5183         while (toread > 0) {
5184                 tocopy = sb->s_blocksize - offset < toread ?
5185                                 sb->s_blocksize - offset : toread;
5186                 bh = ext4_bread(NULL, inode, blk, 0);
5187                 if (IS_ERR(bh))
5188                         return PTR_ERR(bh);
5189                 if (!bh)        /* A hole? */
5190                         memset(data, 0, tocopy);
5191                 else
5192                         memcpy(data, bh->b_data+offset, tocopy);
5193                 brelse(bh);
5194                 offset = 0;
5195                 toread -= tocopy;
5196                 data += tocopy;
5197                 blk++;
5198         }
5199         return len;
5200 }
5201
5202 /* Write to quotafile (we know the transaction is already started and has
5203  * enough credits) */
5204 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5205                                 const char *data, size_t len, loff_t off)
5206 {
5207         struct inode *inode = sb_dqopt(sb)->files[type];
5208         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5209         int err, offset = off & (sb->s_blocksize - 1);
5210         int retries = 0;
5211         struct buffer_head *bh;
5212         handle_t *handle = journal_current_handle();
5213
5214         if (EXT4_SB(sb)->s_journal && !handle) {
5215                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5216                         " cancelled because transaction is not started",
5217                         (unsigned long long)off, (unsigned long long)len);
5218                 return -EIO;
5219         }
5220         /*
5221          * Since we account only one data block in transaction credits,
5222          * then it is impossible to cross a block boundary.
5223          */
5224         if (sb->s_blocksize - offset < len) {
5225                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5226                         " cancelled because not block aligned",
5227                         (unsigned long long)off, (unsigned long long)len);
5228                 return -EIO;
5229         }
5230
5231         do {
5232                 bh = ext4_bread(handle, inode, blk,
5233                                 EXT4_GET_BLOCKS_CREATE |
5234                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5235         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5236                  ext4_should_retry_alloc(inode->i_sb, &retries));
5237         if (IS_ERR(bh))
5238                 return PTR_ERR(bh);
5239         if (!bh)
5240                 goto out;
5241         BUFFER_TRACE(bh, "get write access");
5242         err = ext4_journal_get_write_access(handle, bh);
5243         if (err) {
5244                 brelse(bh);
5245                 return err;
5246         }
5247         lock_buffer(bh);
5248         memcpy(bh->b_data+offset, data, len);
5249         flush_dcache_page(bh->b_page);
5250         unlock_buffer(bh);
5251         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5252         brelse(bh);
5253 out:
5254         if (inode->i_size < off + len) {
5255                 i_size_write(inode, off + len);
5256                 EXT4_I(inode)->i_disksize = inode->i_size;
5257                 ext4_mark_inode_dirty(handle, inode);
5258         }
5259         return len;
5260 }
5261
5262 #endif
5263
5264 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5265                        const char *dev_name, void *data)
5266 {
5267         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5268 }
5269
5270 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5271 static inline void register_as_ext2(void)
5272 {
5273         int err = register_filesystem(&ext2_fs_type);
5274         if (err)
5275                 printk(KERN_WARNING
5276                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5277 }
5278
5279 static inline void unregister_as_ext2(void)
5280 {
5281         unregister_filesystem(&ext2_fs_type);
5282 }
5283
5284 static inline int ext2_feature_set_ok(struct super_block *sb)
5285 {
5286         if (ext4_has_unknown_ext2_incompat_features(sb))
5287                 return 0;
5288         if (sb->s_flags & MS_RDONLY)
5289                 return 1;
5290         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5291                 return 0;
5292         return 1;
5293 }
5294 #else
5295 static inline void register_as_ext2(void) { }
5296 static inline void unregister_as_ext2(void) { }
5297 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5298 #endif
5299
5300 static inline void register_as_ext3(void)
5301 {
5302         int err = register_filesystem(&ext3_fs_type);
5303         if (err)
5304                 printk(KERN_WARNING
5305                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5306 }
5307
5308 static inline void unregister_as_ext3(void)
5309 {
5310         unregister_filesystem(&ext3_fs_type);
5311 }
5312
5313 static inline int ext3_feature_set_ok(struct super_block *sb)
5314 {
5315         if (ext4_has_unknown_ext3_incompat_features(sb))
5316                 return 0;
5317         if (!ext4_has_feature_journal(sb))
5318                 return 0;
5319         if (sb->s_flags & MS_RDONLY)
5320                 return 1;
5321         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5322                 return 0;
5323         return 1;
5324 }
5325
5326 static struct file_system_type ext4_fs_type = {
5327         .owner          = THIS_MODULE,
5328         .name           = "ext4",
5329         .mount          = ext4_mount,
5330         .kill_sb        = kill_block_super,
5331         .fs_flags       = FS_REQUIRES_DEV,
5332 };
5333 MODULE_ALIAS_FS("ext4");
5334
5335 /* Shared across all ext4 file systems */
5336 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5337 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5338
5339 static int __init ext4_init_fs(void)
5340 {
5341         int i, err;
5342
5343         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5344         ext4_li_info = NULL;
5345         mutex_init(&ext4_li_mtx);
5346
5347         /* Build-time check for flags consistency */
5348         ext4_check_flag_values();
5349
5350         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5351                 mutex_init(&ext4__aio_mutex[i]);
5352                 init_waitqueue_head(&ext4__ioend_wq[i]);
5353         }
5354
5355         err = ext4_init_es();
5356         if (err)
5357                 return err;
5358
5359         err = ext4_init_pageio();
5360         if (err)
5361                 goto out5;
5362
5363         err = ext4_init_system_zone();
5364         if (err)
5365                 goto out4;
5366
5367         err = ext4_init_sysfs();
5368         if (err)
5369                 goto out3;
5370
5371         err = ext4_init_mballoc();
5372         if (err)
5373                 goto out2;
5374         else
5375                 ext4_mballoc_ready = 1;
5376         err = init_inodecache();
5377         if (err)
5378                 goto out1;
5379         register_as_ext3();
5380         register_as_ext2();
5381         err = register_filesystem(&ext4_fs_type);
5382         if (err)
5383                 goto out;
5384
5385         return 0;
5386 out:
5387         unregister_as_ext2();
5388         unregister_as_ext3();
5389         destroy_inodecache();
5390 out1:
5391         ext4_mballoc_ready = 0;
5392         ext4_exit_mballoc();
5393 out2:
5394         ext4_exit_sysfs();
5395 out3:
5396         ext4_exit_system_zone();
5397 out4:
5398         ext4_exit_pageio();
5399 out5:
5400         ext4_exit_es();
5401
5402         return err;
5403 }
5404
5405 static void __exit ext4_exit_fs(void)
5406 {
5407         ext4_exit_crypto();
5408         ext4_destroy_lazyinit_thread();
5409         unregister_as_ext2();
5410         unregister_as_ext3();
5411         unregister_filesystem(&ext4_fs_type);
5412         destroy_inodecache();
5413         ext4_exit_mballoc();
5414         ext4_exit_sysfs();
5415         ext4_exit_system_zone();
5416         ext4_exit_pageio();
5417         ext4_exit_es();
5418 }
5419
5420 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5421 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5422 MODULE_LICENSE("GPL");
5423 module_init(ext4_init_fs)
5424 module_exit(ext4_exit_fs)