fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 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/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 static int ext4_verify_csum_type(struct super_block *sb,
 144                                  struct ext4_super_block *es)
 145 {
 146         if (!ext4_has_feature_metadata_csum(sb))
 147                 return 1;
 148
 149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 150 }
 151
 152 static __le32 ext4_superblock_csum(struct super_block *sb,
 153                                    struct ext4_super_block *es)
 154 {
 155         struct ext4_sb_info *sbi = EXT4_SB(sb);
 156         int offset = offsetof(struct ext4_super_block, s_checksum);
 157         __u32 csum;
 158
 159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 160
 161         return cpu_to_le32(csum);
 162 }
 163
 164 static int ext4_superblock_csum_verify(struct super_block *sb,
 165                                        struct ext4_super_block *es)
 166 {
 167         if (!ext4_has_metadata_csum(sb))
 168                 return 1;
 169
 170         return es->s_checksum == ext4_superblock_csum(sb, es);
 171 }
 172
 173 void ext4_superblock_csum_set(struct super_block *sb)
 174 {
 175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 176
 177         if (!ext4_has_metadata_csum(sb))
 178                 return;
 179
 180         es->s_checksum = ext4_superblock_csum(sb, es);
 181 }
 182
 183 void *ext4_kvmalloc(size_t size, gfp_t flags)
 184 {
 185         void *ret;
 186
 187         ret = kmalloc(size, flags | __GFP_NOWARN);
 188         if (!ret)
 189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 190         return ret;
 191 }
 192
 193 void *ext4_kvzalloc(size_t size, gfp_t flags)
 194 {
 195         void *ret;
 196
 197         ret = kzalloc(size, flags | __GFP_NOWARN);
 198         if (!ret)
 199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 200         return ret;
 201 }
 202
 203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 209 }
 210
 211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 212                                struct ext4_group_desc *bg)
 213 {
 214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 217 }
 218
 219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le32_to_cpu(bg->bg_inode_table_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 225 }
 226
 227 __u32 ext4_free_group_clusters(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 233 }
 234
 235 __u32 ext4_free_inodes_count(struct super_block *sb,
 236                               struct ext4_group_desc *bg)
 237 {
 238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 241 }
 242
 243 __u32 ext4_used_dirs_count(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 249 }
 250
 251 __u32 ext4_itable_unused_count(struct super_block *sb,
 252                               struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_itable_unused_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 257 }
 258
 259 void ext4_block_bitmap_set(struct super_block *sb,
 260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 261 {
 262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 265 }
 266
 267 void ext4_inode_bitmap_set(struct super_block *sb,
 268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 269 {
 270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 273 }
 274
 275 void ext4_inode_table_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 277 {
 278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 281 }
 282
 283 void ext4_free_group_clusters_set(struct super_block *sb,
 284                                   struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291 void ext4_free_inodes_set(struct super_block *sb,
 292                           struct ext4_group_desc *bg, __u32 count)
 293 {
 294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 297 }
 298
 299 void ext4_used_dirs_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, __u32 count)
 301 {
 302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 305 }
 306
 307 void ext4_itable_unused_set(struct super_block *sb,
 308                           struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315
 316 static void __save_error_info(struct super_block *sb, const char *func,
 317                             unsigned int line)
 318 {
 319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 320
 321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 322         if (bdev_read_only(sb->s_bdev))
 323                 return;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         le32_add_cpu(&es->s_error_count, 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352 /*
 353  * The del_gendisk() function uninitializes the disk-specific data
 354  * structures, including the bdi structure, without telling anyone
 355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 357  * This is a kludge to prevent these oops until we can put in a proper
 358  * hook in del_gendisk() to inform the VFS and file system layers.
 359  */
 360 static int block_device_ejected(struct super_block *sb)
 361 {
 362         struct inode *bd_inode = sb->s_bdev->bd_inode;
 363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 364
 365         return bdi->dev == NULL;
 366 }
 367
 368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 369 {
 370         struct super_block              *sb = journal->j_private;
 371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 372         int                             error = is_journal_aborted(journal);
 373         struct ext4_journal_cb_entry    *jce;
 374
 375         BUG_ON(txn->t_state == T_FINISHED);
 376
 377         ext4_process_freed_data(sb, txn->t_tid);
 378
 379         spin_lock(&sbi->s_md_lock);
 380         while (!list_empty(&txn->t_private_list)) {
 381                 jce = list_entry(txn->t_private_list.next,
 382                                  struct ext4_journal_cb_entry, jce_list);
 383                 list_del_init(&jce->jce_list);
 384                 spin_unlock(&sbi->s_md_lock);
 385                 jce->jce_func(sb, jce, error);
 386                 spin_lock(&sbi->s_md_lock);
 387         }
 388         spin_unlock(&sbi->s_md_lock);
 389 }
 390
 391 /* Deal with the reporting of failure conditions on a filesystem such as
 392  * inconsistencies detected or read IO failures.
 393  *
 394  * On ext2, we can store the error state of the filesystem in the
 395  * superblock.  That is not possible on ext4, because we may have other
 396  * write ordering constraints on the superblock which prevent us from
 397  * writing it out straight away; and given that the journal is about to
 398  * be aborted, we can't rely on the current, or future, transactions to
 399  * write out the superblock safely.
 400  *
 401  * We'll just use the jbd2_journal_abort() error code to record an error in
 402  * the journal instead.  On recovery, the journal will complain about
 403  * that error until we've noted it down and cleared it.
 404  */
 405
 406 static void ext4_handle_error(struct super_block *sb)
 407 {
 408         if (sb->s_flags & MS_RDONLY)
 409                 return;
 410
 411         if (!test_opt(sb, ERRORS_CONT)) {
 412                 journal_t *journal = EXT4_SB(sb)->s_journal;
 413
 414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 415                 if (journal)
 416                         jbd2_journal_abort(journal, -EIO);
 417         }
 418         if (test_opt(sb, ERRORS_RO)) {
 419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 420                 /*
 421                  * Make sure updated value of ->s_mount_flags will be visible
 422                  * before ->s_flags update
 423                  */
 424                 smp_wmb();
 425                 sb->s_flags |= MS_RDONLY;
 426         }
 427         if (test_opt(sb, ERRORS_PANIC)) {
 428                 if (EXT4_SB(sb)->s_journal &&
 429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 430                         return;
 431                 panic("EXT4-fs (device %s): panic forced after error\n",
 432                         sb->s_id);
 433         }
 434 }
 435
 436 #define ext4_error_ratelimit(sb)                                        \
 437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 438                              "EXT4-fs error")
 439
 440 void __ext4_error(struct super_block *sb, const char *function,
 441                   unsigned int line, const char *fmt, ...)
 442 {
 443         struct va_format vaf;
 444         va_list args;
 445
 446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 447                 return;
 448
 449         if (ext4_error_ratelimit(sb)) {
 450                 va_start(args, fmt);
 451                 vaf.fmt = fmt;
 452                 vaf.va = &args;
 453                 printk(KERN_CRIT
 454                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 455                        sb->s_id, function, line, current->comm, &vaf);
 456                 va_end(args);
 457         }
 458         save_error_info(sb, function, line);
 459         ext4_handle_error(sb);
 460 }
 461
 462 void __ext4_error_inode(struct inode *inode, const char *function,
 463                         unsigned int line, ext4_fsblk_t block,
 464                         const char *fmt, ...)
 465 {
 466         va_list args;
 467         struct va_format vaf;
 468         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 469
 470         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 471                 return;
 472
 473         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 474         es->s_last_error_block = cpu_to_le64(block);
 475         if (ext4_error_ratelimit(inode->i_sb)) {
 476                 va_start(args, fmt);
 477                 vaf.fmt = fmt;
 478                 vaf.va = &args;
 479                 if (block)
 480                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 481                                "inode #%lu: block %llu: comm %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                block, current->comm, &vaf);
 484                 else
 485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 486                                "inode #%lu: comm %s: %pV\n",
 487                                inode->i_sb->s_id, function, line, inode->i_ino,
 488                                current->comm, &vaf);
 489                 va_end(args);
 490         }
 491         save_error_info(inode->i_sb, function, line);
 492         ext4_handle_error(inode->i_sb);
 493 }
 494
 495 void __ext4_error_file(struct file *file, const char *function,
 496                        unsigned int line, ext4_fsblk_t block,
 497                        const char *fmt, ...)
 498 {
 499         va_list args;
 500         struct va_format vaf;
 501         struct ext4_super_block *es;
 502         struct inode *inode = file_inode(file);
 503         char pathname[80], *path;
 504
 505         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 506                 return;
 507
 508         es = EXT4_SB(inode->i_sb)->s_es;
 509         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 510         if (ext4_error_ratelimit(inode->i_sb)) {
 511                 path = file_path(file, pathname, sizeof(pathname));
 512                 if (IS_ERR(path))
 513                         path = "(unknown)";
 514                 va_start(args, fmt);
 515                 vaf.fmt = fmt;
 516                 vaf.va = &args;
 517                 if (block)
 518                         printk(KERN_CRIT
 519                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 520                                "block %llu: comm %s: path %s: %pV\n",
 521                                inode->i_sb->s_id, function, line, inode->i_ino,
 522                                block, current->comm, path, &vaf);
 523                 else
 524                         printk(KERN_CRIT
 525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 526                                "comm %s: path %s: %pV\n",
 527                                inode->i_sb->s_id, function, line, inode->i_ino,
 528                                current->comm, path, &vaf);
 529                 va_end(args);
 530         }
 531         save_error_info(inode->i_sb, function, line);
 532         ext4_handle_error(inode->i_sb);
 533 }
 534
 535 const char *ext4_decode_error(struct super_block *sb, int errno,
 536                               char nbuf[16])
 537 {
 538         char *errstr = NULL;
 539
 540         switch (errno) {
 541         case -EFSCORRUPTED:
 542                 errstr = "Corrupt filesystem";
 543                 break;
 544         case -EFSBADCRC:
 545                 errstr = "Filesystem failed CRC";
 546                 break;
 547         case -EIO:
 548                 errstr = "IO failure";
 549                 break;
 550         case -ENOMEM:
 551                 errstr = "Out of memory";
 552                 break;
 553         case -EROFS:
 554                 if (!sb || (EXT4_SB(sb)->s_journal &&
 555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 556                         errstr = "Journal has aborted";
 557                 else
 558                         errstr = "Readonly filesystem";
 559                 break;
 560         default:
 561                 /* If the caller passed in an extra buffer for unknown
 562                  * errors, textualise them now.  Else we just return
 563                  * NULL. */
 564                 if (nbuf) {
 565                         /* Check for truncated error codes... */
 566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 567                                 errstr = nbuf;
 568                 }
 569                 break;
 570         }
 571
 572         return errstr;
 573 }
 574
 575 /* __ext4_std_error decodes expected errors from journaling functions
 576  * automatically and invokes the appropriate error response.  */
 577
 578 void __ext4_std_error(struct super_block *sb, const char *function,
 579                       unsigned int line, int errno)
 580 {
 581         char nbuf[16];
 582         const char *errstr;
 583
 584         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 585                 return;
 586
 587         /* Special case: if the error is EROFS, and we're not already
 588          * inside a transaction, then there's really no point in logging
 589          * an error. */
 590         if (errno == -EROFS && journal_current_handle() == NULL &&
 591             (sb->s_flags & MS_RDONLY))
 592                 return;
 593
 594         if (ext4_error_ratelimit(sb)) {
 595                 errstr = ext4_decode_error(sb, errno, nbuf);
 596                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 597                        sb->s_id, function, line, errstr);
 598         }
 599
 600         save_error_info(sb, function, line);
 601         ext4_handle_error(sb);
 602 }
 603
 604 /*
 605  * ext4_abort is a much stronger failure handler than ext4_error.  The
 606  * abort function may be used to deal with unrecoverable failures such
 607  * as journal IO errors or ENOMEM at a critical moment in log management.
 608  *
 609  * We unconditionally force the filesystem into an ABORT|READONLY state,
 610  * unless the error response on the fs has been set to panic in which
 611  * case we take the easy way out and panic immediately.
 612  */
 613
 614 void __ext4_abort(struct super_block *sb, const char *function,
 615                 unsigned int line, const char *fmt, ...)
 616 {
 617         struct va_format vaf;
 618         va_list args;
 619
 620         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 621                 return;
 622
 623         save_error_info(sb, function, line);
 624         va_start(args, fmt);
 625         vaf.fmt = fmt;
 626         vaf.va = &args;
 627         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 628                sb->s_id, function, line, &vaf);
 629         va_end(args);
 630
 631         if ((sb->s_flags & MS_RDONLY) == 0) {
 632                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 633                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 634                 /*
 635                  * Make sure updated value of ->s_mount_flags will be visible
 636                  * before ->s_flags update
 637                  */
 638                 smp_wmb();
 639                 sb->s_flags |= MS_RDONLY;
 640                 if (EXT4_SB(sb)->s_journal)
 641                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 642                 save_error_info(sb, function, line);
 643         }
 644         if (test_opt(sb, ERRORS_PANIC)) {
 645                 if (EXT4_SB(sb)->s_journal &&
 646                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 647                         return;
 648                 panic("EXT4-fs panic from previous error\n");
 649         }
 650 }
 651
 652 void __ext4_msg(struct super_block *sb,
 653                 const char *prefix, const char *fmt, ...)
 654 {
 655         struct va_format vaf;
 656         va_list args;
 657
 658         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 659                 return;
 660
 661         va_start(args, fmt);
 662         vaf.fmt = fmt;
 663         vaf.va = &args;
 664         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 665         va_end(args);
 666 }
 667
 668 #define ext4_warning_ratelimit(sb)                                      \
 669                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 670                              "EXT4-fs warning")
 671
 672 void __ext4_warning(struct super_block *sb, const char *function,
 673                     unsigned int line, const char *fmt, ...)
 674 {
 675         struct va_format vaf;
 676         va_list args;
 677
 678         if (!ext4_warning_ratelimit(sb))
 679                 return;
 680
 681         va_start(args, fmt);
 682         vaf.fmt = fmt;
 683         vaf.va = &args;
 684         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 685                sb->s_id, function, line, &vaf);
 686         va_end(args);
 687 }
 688
 689 void __ext4_warning_inode(const struct inode *inode, const char *function,
 690                           unsigned int line, const char *fmt, ...)
 691 {
 692         struct va_format vaf;
 693         va_list args;
 694
 695         if (!ext4_warning_ratelimit(inode->i_sb))
 696                 return;
 697
 698         va_start(args, fmt);
 699         vaf.fmt = fmt;
 700         vaf.va = &args;
 701         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 702                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 703                function, line, inode->i_ino, current->comm, &vaf);
 704         va_end(args);
 705 }
 706
 707 void __ext4_grp_locked_error(const char *function, unsigned int line,
 708                              struct super_block *sb, ext4_group_t grp,
 709                              unsigned long ino, ext4_fsblk_t block,
 710                              const char *fmt, ...)
 711 __releases(bitlock)
 712 __acquires(bitlock)
 713 {
 714         struct va_format vaf;
 715         va_list args;
 716         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 717
 718         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 719                 return;
 720
 721         es->s_last_error_ino = cpu_to_le32(ino);
 722         es->s_last_error_block = cpu_to_le64(block);
 723         __save_error_info(sb, function, line);
 724
 725         if (ext4_error_ratelimit(sb)) {
 726                 va_start(args, fmt);
 727                 vaf.fmt = fmt;
 728                 vaf.va = &args;
 729                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 730                        sb->s_id, function, line, grp);
 731                 if (ino)
 732                         printk(KERN_CONT "inode %lu: ", ino);
 733                 if (block)
 734                         printk(KERN_CONT "block %llu:",
 735                                (unsigned long long) block);
 736                 printk(KERN_CONT "%pV\n", &vaf);
 737                 va_end(args);
 738         }
 739
 740         if (test_opt(sb, ERRORS_CONT)) {
 741                 ext4_commit_super(sb, 0);
 742                 return;
 743         }
 744
 745         ext4_unlock_group(sb, grp);
 746         ext4_handle_error(sb);
 747         /*
 748          * We only get here in the ERRORS_RO case; relocking the group
 749          * may be dangerous, but nothing bad will happen since the
 750          * filesystem will have already been marked read/only and the
 751          * journal has been aborted.  We return 1 as a hint to callers
 752          * who might what to use the return value from
 753          * ext4_grp_locked_error() to distinguish between the
 754          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 755          * aggressively from the ext4 function in question, with a
 756          * more appropriate error code.
 757          */
 758         ext4_lock_group(sb, grp);
 759         return;
 760 }
 761
 762 void ext4_update_dynamic_rev(struct super_block *sb)
 763 {
 764         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 765
 766         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 767                 return;
 768
 769         ext4_warning(sb,
 770                      "updating to rev %d because of new feature flag, "
 771                      "running e2fsck is recommended",
 772                      EXT4_DYNAMIC_REV);
 773
 774         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 775         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 776         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 777         /* leave es->s_feature_*compat flags alone */
 778         /* es->s_uuid will be set by e2fsck if empty */
 779
 780         /*
 781          * The rest of the superblock fields should be zero, and if not it
 782          * means they are likely already in use, so leave them alone.  We
 783          * can leave it up to e2fsck to clean up any inconsistencies there.
 784          */
 785 }
 786
 787 /*
 788  * Open the external journal device
 789  */
 790 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 791 {
 792         struct block_device *bdev;
 793         char b[BDEVNAME_SIZE];
 794
 795         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 796         if (IS_ERR(bdev))
 797                 goto fail;
 798         return bdev;
 799
 800 fail:
 801         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 802                         __bdevname(dev, b), PTR_ERR(bdev));
 803         return NULL;
 804 }
 805
 806 /*
 807  * Release the journal device
 808  */
 809 static void ext4_blkdev_put(struct block_device *bdev)
 810 {
 811         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 812 }
 813
 814 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 815 {
 816         struct block_device *bdev;
 817         bdev = sbi->journal_bdev;
 818         if (bdev) {
 819                 ext4_blkdev_put(bdev);
 820                 sbi->journal_bdev = NULL;
 821         }
 822 }
 823
 824 static inline struct inode *orphan_list_entry(struct list_head *l)
 825 {
 826         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 827 }
 828
 829 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 830 {
 831         struct list_head *l;
 832
 833         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 834                  le32_to_cpu(sbi->s_es->s_last_orphan));
 835
 836         printk(KERN_ERR "sb_info orphan list:\n");
 837         list_for_each(l, &sbi->s_orphan) {
 838                 struct inode *inode = orphan_list_entry(l);
 839                 printk(KERN_ERR "  "
 840                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 841                        inode->i_sb->s_id, inode->i_ino, inode,
 842                        inode->i_mode, inode->i_nlink,
 843                        NEXT_ORPHAN(inode));
 844         }
 845 }
 846
 847 #ifdef CONFIG_QUOTA
 848 static int ext4_quota_off(struct super_block *sb, int type);
 849
 850 static inline void ext4_quota_off_umount(struct super_block *sb)
 851 {
 852         int type;
 853
 854         /* Use our quota_off function to clear inode flags etc. */
 855         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 856                 ext4_quota_off(sb, type);
 857 }
 858 #else
 859 static inline void ext4_quota_off_umount(struct super_block *sb)
 860 {
 861 }
 862 #endif
 863
 864 static void ext4_put_super(struct super_block *sb)
 865 {
 866         struct ext4_sb_info *sbi = EXT4_SB(sb);
 867         struct ext4_super_block *es = sbi->s_es;
 868         int aborted = 0;
 869         int i, err;
 870
 871         ext4_unregister_li_request(sb);
 872         ext4_quota_off_umount(sb);
 873
 874         flush_workqueue(sbi->rsv_conversion_wq);
 875         destroy_workqueue(sbi->rsv_conversion_wq);
 876
 877         if (sbi->s_journal) {
 878                 aborted = is_journal_aborted(sbi->s_journal);
 879                 err = jbd2_journal_destroy(sbi->s_journal);
 880                 sbi->s_journal = NULL;
 881                 if ((err < 0) && !aborted)
 882                         ext4_abort(sb, "Couldn't clean up the journal");
 883         }
 884
 885         ext4_unregister_sysfs(sb);
 886         ext4_es_unregister_shrinker(sbi);
 887         del_timer_sync(&sbi->s_err_report);
 888         ext4_release_system_zone(sb);
 889         ext4_mb_release(sb);
 890         ext4_ext_release(sb);
 891
 892         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 893                 ext4_clear_feature_journal_needs_recovery(sb);
 894                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 895         }
 896         if (!(sb->s_flags & MS_RDONLY))
 897                 ext4_commit_super(sb, 1);
 898
 899         for (i = 0; i < sbi->s_gdb_count; i++)
 900                 brelse(sbi->s_group_desc[i]);
 901         kvfree(sbi->s_group_desc);
 902         kvfree(sbi->s_flex_groups);
 903         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 904         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 905         percpu_counter_destroy(&sbi->s_dirs_counter);
 906         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 907         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 908 #ifdef CONFIG_QUOTA
 909         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 910                 kfree(sbi->s_qf_names[i]);
 911 #endif
 912
 913         /* Debugging code just in case the in-memory inode orphan list
 914          * isn't empty.  The on-disk one can be non-empty if we've
 915          * detected an error and taken the fs readonly, but the
 916          * in-memory list had better be clean by this point. */
 917         if (!list_empty(&sbi->s_orphan))
 918                 dump_orphan_list(sb, sbi);
 919         J_ASSERT(list_empty(&sbi->s_orphan));
 920
 921         sync_blockdev(sb->s_bdev);
 922         invalidate_bdev(sb->s_bdev);
 923         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 924                 /*
 925                  * Invalidate the journal device's buffers.  We don't want them
 926                  * floating about in memory - the physical journal device may
 927                  * hotswapped, and it breaks the `ro-after' testing code.
 928                  */
 929                 sync_blockdev(sbi->journal_bdev);
 930                 invalidate_bdev(sbi->journal_bdev);
 931                 ext4_blkdev_remove(sbi);
 932         }
 933         if (sbi->s_ea_inode_cache) {
 934                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 935                 sbi->s_ea_inode_cache = NULL;
 936         }
 937         if (sbi->s_ea_block_cache) {
 938                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 939                 sbi->s_ea_block_cache = NULL;
 940         }
 941         if (sbi->s_mmp_tsk)
 942                 kthread_stop(sbi->s_mmp_tsk);
 943         brelse(sbi->s_sbh);
 944         sb->s_fs_info = NULL;
 945         /*
 946          * Now that we are completely done shutting down the
 947          * superblock, we need to actually destroy the kobject.
 948          */
 949         kobject_put(&sbi->s_kobj);
 950         wait_for_completion(&sbi->s_kobj_unregister);
 951         if (sbi->s_chksum_driver)
 952                 crypto_free_shash(sbi->s_chksum_driver);
 953         kfree(sbi->s_blockgroup_lock);
 954         kfree(sbi);
 955 }
 956
 957 static struct kmem_cache *ext4_inode_cachep;
 958
 959 /*
 960  * Called inside transaction, so use GFP_NOFS
 961  */
 962 static struct inode *ext4_alloc_inode(struct super_block *sb)
 963 {
 964         struct ext4_inode_info *ei;
 965
 966         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 967         if (!ei)
 968                 return NULL;
 969
 970         ei->vfs_inode.i_version = 1;
 971         spin_lock_init(&ei->i_raw_lock);
 972         INIT_LIST_HEAD(&ei->i_prealloc_list);
 973         spin_lock_init(&ei->i_prealloc_lock);
 974         ext4_es_init_tree(&ei->i_es_tree);
 975         rwlock_init(&ei->i_es_lock);
 976         INIT_LIST_HEAD(&ei->i_es_list);
 977         ei->i_es_all_nr = 0;
 978         ei->i_es_shk_nr = 0;
 979         ei->i_es_shrink_lblk = 0;
 980         ei->i_reserved_data_blocks = 0;
 981         ei->i_reserved_meta_blocks = 0;
 982         ei->i_allocated_meta_blocks = 0;
 983         ei->i_da_metadata_calc_len = 0;
 984         ei->i_da_metadata_calc_last_lblock = 0;
 985         spin_lock_init(&(ei->i_block_reservation_lock));
 986 #ifdef CONFIG_QUOTA
 987         ei->i_reserved_quota = 0;
 988         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 989 #endif
 990         ei->jinode = NULL;
 991         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 992         spin_lock_init(&ei->i_completed_io_lock);
 993         ei->i_sync_tid = 0;
 994         ei->i_datasync_tid = 0;
 995         atomic_set(&ei->i_unwritten, 0);
 996         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 997         return &ei->vfs_inode;
 998 }
 999
1000 static int ext4_drop_inode(struct inode *inode)
1001 {
1002         int drop = generic_drop_inode(inode);
1003
1004         trace_ext4_drop_inode(inode, drop);
1005         return drop;
1006 }
1007
1008 static void ext4_i_callback(struct rcu_head *head)
1009 {
1010         struct inode *inode = container_of(head, struct inode, i_rcu);
1011         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1012 }
1013
1014 static void ext4_destroy_inode(struct inode *inode)
1015 {
1016         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1017                 ext4_msg(inode->i_sb, KERN_ERR,
1018                          "Inode %lu (%p): orphan list check failed!",
1019                          inode->i_ino, EXT4_I(inode));
1020                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1021                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1022                                 true);
1023                 dump_stack();
1024         }
1025         call_rcu(&inode->i_rcu, ext4_i_callback);
1026 }
1027
1028 static void init_once(void *foo)
1029 {
1030         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1031
1032         INIT_LIST_HEAD(&ei->i_orphan);
1033         init_rwsem(&ei->xattr_sem);
1034         init_rwsem(&ei->i_data_sem);
1035         init_rwsem(&ei->i_mmap_sem);
1036         inode_init_once(&ei->vfs_inode);
1037 }
1038
1039 static int __init init_inodecache(void)
1040 {
1041         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1042                                              sizeof(struct ext4_inode_info),
1043                                              0, (SLAB_RECLAIM_ACCOUNT|
1044                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1045                                              init_once);
1046         if (ext4_inode_cachep == NULL)
1047                 return -ENOMEM;
1048         return 0;
1049 }
1050
1051 static void destroy_inodecache(void)
1052 {
1053         /*
1054          * Make sure all delayed rcu free inodes are flushed before we
1055          * destroy cache.
1056          */
1057         rcu_barrier();
1058         kmem_cache_destroy(ext4_inode_cachep);
1059 }
1060
1061 void ext4_clear_inode(struct inode *inode)
1062 {
1063         invalidate_inode_buffers(inode);
1064         clear_inode(inode);
1065         dquot_drop(inode);
1066         ext4_discard_preallocations(inode);
1067         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1068         if (EXT4_I(inode)->jinode) {
1069                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1070                                                EXT4_I(inode)->jinode);
1071                 jbd2_free_inode(EXT4_I(inode)->jinode);
1072                 EXT4_I(inode)->jinode = NULL;
1073         }
1074 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1075         fscrypt_put_encryption_info(inode, NULL);
1076 #endif
1077 }
1078
1079 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1080                                         u64 ino, u32 generation)
1081 {
1082         struct inode *inode;
1083
1084         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1085                 return ERR_PTR(-ESTALE);
1086         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1087                 return ERR_PTR(-ESTALE);
1088
1089         /* iget isn't really right if the inode is currently unallocated!!
1090          *
1091          * ext4_read_inode will return a bad_inode if the inode had been
1092          * deleted, so we should be safe.
1093          *
1094          * Currently we don't know the generation for parent directory, so
1095          * a generation of 0 means "accept any"
1096          */
1097         inode = ext4_iget_normal(sb, ino);
1098         if (IS_ERR(inode))
1099                 return ERR_CAST(inode);
1100         if (generation && inode->i_generation != generation) {
1101                 iput(inode);
1102                 return ERR_PTR(-ESTALE);
1103         }
1104
1105         return inode;
1106 }
1107
1108 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1109                                         int fh_len, int fh_type)
1110 {
1111         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1112                                     ext4_nfs_get_inode);
1113 }
1114
1115 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1116                                         int fh_len, int fh_type)
1117 {
1118         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1119                                     ext4_nfs_get_inode);
1120 }
1121
1122 /*
1123  * Try to release metadata pages (indirect blocks, directories) which are
1124  * mapped via the block device.  Since these pages could have journal heads
1125  * which would prevent try_to_free_buffers() from freeing them, we must use
1126  * jbd2 layer's try_to_free_buffers() function to release them.
1127  */
1128 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1129                                  gfp_t wait)
1130 {
1131         journal_t *journal = EXT4_SB(sb)->s_journal;
1132
1133         WARN_ON(PageChecked(page));
1134         if (!page_has_buffers(page))
1135                 return 0;
1136         if (journal)
1137                 return jbd2_journal_try_to_free_buffers(journal, page,
1138                                                 wait & ~__GFP_DIRECT_RECLAIM);
1139         return try_to_free_buffers(page);
1140 }
1141
1142 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1143 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1144 {
1145         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1146                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1147 }
1148
1149 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1150                                                         void *fs_data)
1151 {
1152         handle_t *handle = fs_data;
1153         int res, res2, credits, retries = 0;
1154
1155         res = ext4_convert_inline_data(inode);
1156         if (res)
1157                 return res;
1158
1159         /*
1160          * If a journal handle was specified, then the encryption context is
1161          * being set on a new inode via inheritance and is part of a larger
1162          * transaction to create the inode.  Otherwise the encryption context is
1163          * being set on an existing inode in its own transaction.  Only in the
1164          * latter case should the "retry on ENOSPC" logic be used.
1165          */
1166
1167         if (handle) {
1168                 res = ext4_xattr_set_handle(handle, inode,
1169                                             EXT4_XATTR_INDEX_ENCRYPTION,
1170                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1171                                             ctx, len, 0);
1172                 if (!res) {
1173                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1174                         ext4_clear_inode_state(inode,
1175                                         EXT4_STATE_MAY_INLINE_DATA);
1176                         /*
1177                          * Update inode->i_flags - e.g. S_DAX may get disabled
1178                          */
1179                         ext4_set_inode_flags(inode);
1180                 }
1181                 return res;
1182         }
1183
1184         res = dquot_initialize(inode);
1185         if (res)
1186                 return res;
1187 retry:
1188         res = ext4_xattr_set_credits(inode, len, &credits);
1189         if (res)
1190                 return res;
1191
1192         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1193         if (IS_ERR(handle))
1194                 return PTR_ERR(handle);
1195
1196         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1197                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1198                                     ctx, len, 0);
1199         if (!res) {
1200                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1201                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1202                 ext4_set_inode_flags(inode);
1203                 res = ext4_mark_inode_dirty(handle, inode);
1204                 if (res)
1205                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1206         }
1207         res2 = ext4_journal_stop(handle);
1208
1209         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1210                 goto retry;
1211         if (!res)
1212                 res = res2;
1213         return res;
1214 }
1215
1216 static int ext4_dummy_context(struct inode *inode)
1217 {
1218         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1219 }
1220
1221 static unsigned ext4_max_namelen(struct inode *inode)
1222 {
1223         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1224                 EXT4_NAME_LEN;
1225 }
1226
1227 static const struct fscrypt_operations ext4_cryptops = {
1228         .key_prefix             = "ext4:",
1229         .get_context            = ext4_get_context,
1230         .set_context            = ext4_set_context,
1231         .dummy_context          = ext4_dummy_context,
1232         .is_encrypted           = ext4_encrypted_inode,
1233         .empty_dir              = ext4_empty_dir,
1234         .max_namelen            = ext4_max_namelen,
1235 };
1236 #else
1237 static const struct fscrypt_operations ext4_cryptops = {
1238         .is_encrypted           = ext4_encrypted_inode,
1239 };
1240 #endif
1241
1242 #ifdef CONFIG_QUOTA
1243 static const char * const quotatypes[] = INITQFNAMES;
1244 #define QTYPE2NAME(t) (quotatypes[t])
1245
1246 static int ext4_write_dquot(struct dquot *dquot);
1247 static int ext4_acquire_dquot(struct dquot *dquot);
1248 static int ext4_release_dquot(struct dquot *dquot);
1249 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1250 static int ext4_write_info(struct super_block *sb, int type);
1251 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1252                          const struct path *path);
1253 static int ext4_quota_on_mount(struct super_block *sb, int type);
1254 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1255                                size_t len, loff_t off);
1256 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1257                                 const char *data, size_t len, loff_t off);
1258 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1259                              unsigned int flags);
1260 static int ext4_enable_quotas(struct super_block *sb);
1261 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1262
1263 static struct dquot **ext4_get_dquots(struct inode *inode)
1264 {
1265         return EXT4_I(inode)->i_dquot;
1266 }
1267
1268 static const struct dquot_operations ext4_quota_operations = {
1269         .get_reserved_space     = ext4_get_reserved_space,
1270         .write_dquot            = ext4_write_dquot,
1271         .acquire_dquot          = ext4_acquire_dquot,
1272         .release_dquot          = ext4_release_dquot,
1273         .mark_dirty             = ext4_mark_dquot_dirty,
1274         .write_info             = ext4_write_info,
1275         .alloc_dquot            = dquot_alloc,
1276         .destroy_dquot          = dquot_destroy,
1277         .get_projid             = ext4_get_projid,
1278         .get_inode_usage        = ext4_get_inode_usage,
1279         .get_next_id            = ext4_get_next_id,
1280 };
1281
1282 static const struct quotactl_ops ext4_qctl_operations = {
1283         .quota_on       = ext4_quota_on,
1284         .quota_off      = ext4_quota_off,
1285         .quota_sync     = dquot_quota_sync,
1286         .get_state      = dquot_get_state,
1287         .set_info       = dquot_set_dqinfo,
1288         .get_dqblk      = dquot_get_dqblk,
1289         .set_dqblk      = dquot_set_dqblk,
1290         .get_nextdqblk  = dquot_get_next_dqblk,
1291 };
1292 #endif
1293
1294 static const struct super_operations ext4_sops = {
1295         .alloc_inode    = ext4_alloc_inode,
1296         .destroy_inode  = ext4_destroy_inode,
1297         .write_inode    = ext4_write_inode,
1298         .dirty_inode    = ext4_dirty_inode,
1299         .drop_inode     = ext4_drop_inode,
1300         .evict_inode    = ext4_evict_inode,
1301         .put_super      = ext4_put_super,
1302         .sync_fs        = ext4_sync_fs,
1303         .freeze_fs      = ext4_freeze,
1304         .unfreeze_fs    = ext4_unfreeze,
1305         .statfs         = ext4_statfs,
1306         .remount_fs     = ext4_remount,
1307         .show_options   = ext4_show_options,
1308 #ifdef CONFIG_QUOTA
1309         .quota_read     = ext4_quota_read,
1310         .quota_write    = ext4_quota_write,
1311         .get_dquots     = ext4_get_dquots,
1312 #endif
1313         .bdev_try_to_free_page = bdev_try_to_free_page,
1314 };
1315
1316 static const struct export_operations ext4_export_ops = {
1317         .fh_to_dentry = ext4_fh_to_dentry,
1318         .fh_to_parent = ext4_fh_to_parent,
1319         .get_parent = ext4_get_parent,
1320 };
1321
1322 enum {
1323         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1324         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1325         Opt_nouid32, Opt_debug, Opt_removed,
1326         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1327         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1328         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1329         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1330         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1331         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1332         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1333         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1334         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1335         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1336         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1337         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1338         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1339         Opt_inode_readahead_blks, Opt_journal_ioprio,
1340         Opt_dioread_nolock, Opt_dioread_lock,
1341         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1342         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1343 };
1344
1345 static const match_table_t tokens = {
1346         {Opt_bsd_df, "bsddf"},
1347         {Opt_minix_df, "minixdf"},
1348         {Opt_grpid, "grpid"},
1349         {Opt_grpid, "bsdgroups"},
1350         {Opt_nogrpid, "nogrpid"},
1351         {Opt_nogrpid, "sysvgroups"},
1352         {Opt_resgid, "resgid=%u"},
1353         {Opt_resuid, "resuid=%u"},
1354         {Opt_sb, "sb=%u"},
1355         {Opt_err_cont, "errors=continue"},
1356         {Opt_err_panic, "errors=panic"},
1357         {Opt_err_ro, "errors=remount-ro"},
1358         {Opt_nouid32, "nouid32"},
1359         {Opt_debug, "debug"},
1360         {Opt_removed, "oldalloc"},
1361         {Opt_removed, "orlov"},
1362         {Opt_user_xattr, "user_xattr"},
1363         {Opt_nouser_xattr, "nouser_xattr"},
1364         {Opt_acl, "acl"},
1365         {Opt_noacl, "noacl"},
1366         {Opt_noload, "norecovery"},
1367         {Opt_noload, "noload"},
1368         {Opt_removed, "nobh"},
1369         {Opt_removed, "bh"},
1370         {Opt_commit, "commit=%u"},
1371         {Opt_min_batch_time, "min_batch_time=%u"},
1372         {Opt_max_batch_time, "max_batch_time=%u"},
1373         {Opt_journal_dev, "journal_dev=%u"},
1374         {Opt_journal_path, "journal_path=%s"},
1375         {Opt_journal_checksum, "journal_checksum"},
1376         {Opt_nojournal_checksum, "nojournal_checksum"},
1377         {Opt_journal_async_commit, "journal_async_commit"},
1378         {Opt_abort, "abort"},
1379         {Opt_data_journal, "data=journal"},
1380         {Opt_data_ordered, "data=ordered"},
1381         {Opt_data_writeback, "data=writeback"},
1382         {Opt_data_err_abort, "data_err=abort"},
1383         {Opt_data_err_ignore, "data_err=ignore"},
1384         {Opt_offusrjquota, "usrjquota="},
1385         {Opt_usrjquota, "usrjquota=%s"},
1386         {Opt_offgrpjquota, "grpjquota="},
1387         {Opt_grpjquota, "grpjquota=%s"},
1388         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1389         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1390         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1391         {Opt_grpquota, "grpquota"},
1392         {Opt_noquota, "noquota"},
1393         {Opt_quota, "quota"},
1394         {Opt_usrquota, "usrquota"},
1395         {Opt_prjquota, "prjquota"},
1396         {Opt_barrier, "barrier=%u"},
1397         {Opt_barrier, "barrier"},
1398         {Opt_nobarrier, "nobarrier"},
1399         {Opt_i_version, "i_version"},
1400         {Opt_dax, "dax"},
1401         {Opt_stripe, "stripe=%u"},
1402         {Opt_delalloc, "delalloc"},
1403         {Opt_lazytime, "lazytime"},
1404         {Opt_nolazytime, "nolazytime"},
1405         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1406         {Opt_nodelalloc, "nodelalloc"},
1407         {Opt_removed, "mblk_io_submit"},
1408         {Opt_removed, "nomblk_io_submit"},
1409         {Opt_block_validity, "block_validity"},
1410         {Opt_noblock_validity, "noblock_validity"},
1411         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1412         {Opt_journal_ioprio, "journal_ioprio=%u"},
1413         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1414         {Opt_auto_da_alloc, "auto_da_alloc"},
1415         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1416         {Opt_dioread_nolock, "dioread_nolock"},
1417         {Opt_dioread_lock, "dioread_lock"},
1418         {Opt_discard, "discard"},
1419         {Opt_nodiscard, "nodiscard"},
1420         {Opt_init_itable, "init_itable=%u"},
1421         {Opt_init_itable, "init_itable"},
1422         {Opt_noinit_itable, "noinit_itable"},
1423         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1424         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1425         {Opt_nombcache, "nombcache"},
1426         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1427         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1428         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1429         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1430         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1431         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1432         {Opt_err, NULL},
1433 };
1434
1435 static ext4_fsblk_t get_sb_block(void **data)
1436 {
1437         ext4_fsblk_t    sb_block;
1438         char            *options = (char *) *data;
1439
1440         if (!options || strncmp(options, "sb=", 3) != 0)
1441                 return 1;       /* Default location */
1442
1443         options += 3;
1444         /* TODO: use simple_strtoll with >32bit ext4 */
1445         sb_block = simple_strtoul(options, &options, 0);
1446         if (*options && *options != ',') {
1447                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1448                        (char *) *data);
1449                 return 1;
1450         }
1451         if (*options == ',')
1452                 options++;
1453         *data = (void *) options;
1454
1455         return sb_block;
1456 }
1457
1458 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1459 static const char deprecated_msg[] =
1460         "Mount option \"%s\" will be removed by %s\n"
1461         "Contact [email protected] if you think we should keep it.\n";
1462
1463 #ifdef CONFIG_QUOTA
1464 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1465 {
1466         struct ext4_sb_info *sbi = EXT4_SB(sb);
1467         char *qname;
1468         int ret = -1;
1469
1470         if (sb_any_quota_loaded(sb) &&
1471                 !sbi->s_qf_names[qtype]) {
1472                 ext4_msg(sb, KERN_ERR,
1473                         "Cannot change journaled "
1474                         "quota options when quota turned on");
1475                 return -1;
1476         }
1477         if (ext4_has_feature_quota(sb)) {
1478                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1479                          "ignored when QUOTA feature is enabled");
1480                 return 1;
1481         }
1482         qname = match_strdup(args);
1483         if (!qname) {
1484                 ext4_msg(sb, KERN_ERR,
1485                         "Not enough memory for storing quotafile name");
1486                 return -1;
1487         }
1488         if (sbi->s_qf_names[qtype]) {
1489                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1490                         ret = 1;
1491                 else
1492                         ext4_msg(sb, KERN_ERR,
1493                                  "%s quota file already specified",
1494                                  QTYPE2NAME(qtype));
1495                 goto errout;
1496         }
1497         if (strchr(qname, '/')) {
1498                 ext4_msg(sb, KERN_ERR,
1499                         "quotafile must be on filesystem root");
1500                 goto errout;
1501         }
1502         sbi->s_qf_names[qtype] = qname;
1503         set_opt(sb, QUOTA);
1504         return 1;
1505 errout:
1506         kfree(qname);
1507         return ret;
1508 }
1509
1510 static int clear_qf_name(struct super_block *sb, int qtype)
1511 {
1512
1513         struct ext4_sb_info *sbi = EXT4_SB(sb);
1514
1515         if (sb_any_quota_loaded(sb) &&
1516                 sbi->s_qf_names[qtype]) {
1517                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1518                         " when quota turned on");
1519                 return -1;
1520         }
1521         kfree(sbi->s_qf_names[qtype]);
1522         sbi->s_qf_names[qtype] = NULL;
1523         return 1;
1524 }
1525 #endif
1526
1527 #define MOPT_SET        0x0001
1528 #define MOPT_CLEAR      0x0002
1529 #define MOPT_NOSUPPORT  0x0004
1530 #define MOPT_EXPLICIT   0x0008
1531 #define MOPT_CLEAR_ERR  0x0010
1532 #define MOPT_GTE0       0x0020
1533 #ifdef CONFIG_QUOTA
1534 #define MOPT_Q          0
1535 #define MOPT_QFMT       0x0040
1536 #else
1537 #define MOPT_Q          MOPT_NOSUPPORT
1538 #define MOPT_QFMT       MOPT_NOSUPPORT
1539 #endif
1540 #define MOPT_DATAJ      0x0080
1541 #define MOPT_NO_EXT2    0x0100
1542 #define MOPT_NO_EXT3    0x0200
1543 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1544 #define MOPT_STRING     0x0400
1545
1546 static const struct mount_opts {
1547         int     token;
1548         int     mount_opt;
1549         int     flags;
1550 } ext4_mount_opts[] = {
1551         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1552         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1553         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1554         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1555         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1556         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1557         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1558          MOPT_EXT4_ONLY | MOPT_SET},
1559         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1560          MOPT_EXT4_ONLY | MOPT_CLEAR},
1561         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1562         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1563         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1564          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1565         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1566          MOPT_EXT4_ONLY | MOPT_CLEAR},
1567         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1568          MOPT_EXT4_ONLY | MOPT_CLEAR},
1569         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1570          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1571         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1572                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1573          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1574         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1575         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1576         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1577         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1578         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1579          MOPT_NO_EXT2},
1580         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1581          MOPT_NO_EXT2},
1582         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1583         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1584         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1585         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1586         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1587         {Opt_commit, 0, MOPT_GTE0},
1588         {Opt_max_batch_time, 0, MOPT_GTE0},
1589         {Opt_min_batch_time, 0, MOPT_GTE0},
1590         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1591         {Opt_init_itable, 0, MOPT_GTE0},
1592         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1593         {Opt_stripe, 0, MOPT_GTE0},
1594         {Opt_resuid, 0, MOPT_GTE0},
1595         {Opt_resgid, 0, MOPT_GTE0},
1596         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1597         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1598         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1599         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1600         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1601         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1602          MOPT_NO_EXT2 | MOPT_DATAJ},
1603         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1604         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1605 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1606         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1607         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1608 #else
1609         {Opt_acl, 0, MOPT_NOSUPPORT},
1610         {Opt_noacl, 0, MOPT_NOSUPPORT},
1611 #endif
1612         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1613         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1614         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1615         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1616         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1617                                                         MOPT_SET | MOPT_Q},
1618         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1619                                                         MOPT_SET | MOPT_Q},
1620         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1621                                                         MOPT_SET | MOPT_Q},
1622         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1623                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1624                                                         MOPT_CLEAR | MOPT_Q},
1625         {Opt_usrjquota, 0, MOPT_Q},
1626         {Opt_grpjquota, 0, MOPT_Q},
1627         {Opt_offusrjquota, 0, MOPT_Q},
1628         {Opt_offgrpjquota, 0, MOPT_Q},
1629         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1630         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1631         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1632         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1633         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1634         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1635         {Opt_err, 0, 0}
1636 };
1637
1638 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1639                             substring_t *args, unsigned long *journal_devnum,
1640                             unsigned int *journal_ioprio, int is_remount)
1641 {
1642         struct ext4_sb_info *sbi = EXT4_SB(sb);
1643         const struct mount_opts *m;
1644         kuid_t uid;
1645         kgid_t gid;
1646         int arg = 0;
1647
1648 #ifdef CONFIG_QUOTA
1649         if (token == Opt_usrjquota)
1650                 return set_qf_name(sb, USRQUOTA, &args[0]);
1651         else if (token == Opt_grpjquota)
1652                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1653         else if (token == Opt_offusrjquota)
1654                 return clear_qf_name(sb, USRQUOTA);
1655         else if (token == Opt_offgrpjquota)
1656                 return clear_qf_name(sb, GRPQUOTA);
1657 #endif
1658         switch (token) {
1659         case Opt_noacl:
1660         case Opt_nouser_xattr:
1661                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1662                 break;
1663         case Opt_sb:
1664                 return 1;       /* handled by get_sb_block() */
1665         case Opt_removed:
1666                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1667                 return 1;
1668         case Opt_abort:
1669                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1670                 return 1;
1671         case Opt_i_version:
1672                 sb->s_flags |= MS_I_VERSION;
1673                 return 1;
1674         case Opt_lazytime:
1675                 sb->s_flags |= MS_LAZYTIME;
1676                 return 1;
1677         case Opt_nolazytime:
1678                 sb->s_flags &= ~MS_LAZYTIME;
1679                 return 1;
1680         }
1681
1682         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1683                 if (token == m->token)
1684                         break;
1685
1686         if (m->token == Opt_err) {
1687                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1688                          "or missing value", opt);
1689                 return -1;
1690         }
1691
1692         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1693                 ext4_msg(sb, KERN_ERR,
1694                          "Mount option \"%s\" incompatible with ext2", opt);
1695                 return -1;
1696         }
1697         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1698                 ext4_msg(sb, KERN_ERR,
1699                          "Mount option \"%s\" incompatible with ext3", opt);
1700                 return -1;
1701         }
1702
1703         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1704                 return -1;
1705         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1706                 return -1;
1707         if (m->flags & MOPT_EXPLICIT) {
1708                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1709                         set_opt2(sb, EXPLICIT_DELALLOC);
1710                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1711                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1712                 } else
1713                         return -1;
1714         }
1715         if (m->flags & MOPT_CLEAR_ERR)
1716                 clear_opt(sb, ERRORS_MASK);
1717         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1718                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1719                          "options when quota turned on");
1720                 return -1;
1721         }
1722
1723         if (m->flags & MOPT_NOSUPPORT) {
1724                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1725         } else if (token == Opt_commit) {
1726                 if (arg == 0)
1727                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1728                 sbi->s_commit_interval = HZ * arg;
1729         } else if (token == Opt_debug_want_extra_isize) {
1730                 sbi->s_want_extra_isize = arg;
1731         } else if (token == Opt_max_batch_time) {
1732                 sbi->s_max_batch_time = arg;
1733         } else if (token == Opt_min_batch_time) {
1734                 sbi->s_min_batch_time = arg;
1735         } else if (token == Opt_inode_readahead_blks) {
1736                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1737                         ext4_msg(sb, KERN_ERR,
1738                                  "EXT4-fs: inode_readahead_blks must be "
1739                                  "0 or a power of 2 smaller than 2^31");
1740                         return -1;
1741                 }
1742                 sbi->s_inode_readahead_blks = arg;
1743         } else if (token == Opt_init_itable) {
1744                 set_opt(sb, INIT_INODE_TABLE);
1745                 if (!args->from)
1746                         arg = EXT4_DEF_LI_WAIT_MULT;
1747                 sbi->s_li_wait_mult = arg;
1748         } else if (token == Opt_max_dir_size_kb) {
1749                 sbi->s_max_dir_size_kb = arg;
1750         } else if (token == Opt_stripe) {
1751                 sbi->s_stripe = arg;
1752         } else if (token == Opt_resuid) {
1753                 uid = make_kuid(current_user_ns(), arg);
1754                 if (!uid_valid(uid)) {
1755                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1756                         return -1;
1757                 }
1758                 sbi->s_resuid = uid;
1759         } else if (token == Opt_resgid) {
1760                 gid = make_kgid(current_user_ns(), arg);
1761                 if (!gid_valid(gid)) {
1762                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1763                         return -1;
1764                 }
1765                 sbi->s_resgid = gid;
1766         } else if (token == Opt_journal_dev) {
1767                 if (is_remount) {
1768                         ext4_msg(sb, KERN_ERR,
1769                                  "Cannot specify journal on remount");
1770                         return -1;
1771                 }
1772                 *journal_devnum = arg;
1773         } else if (token == Opt_journal_path) {
1774                 char *journal_path;
1775                 struct inode *journal_inode;
1776                 struct path path;
1777                 int error;
1778
1779                 if (is_remount) {
1780                         ext4_msg(sb, KERN_ERR,
1781                                  "Cannot specify journal on remount");
1782                         return -1;
1783                 }
1784                 journal_path = match_strdup(&args[0]);
1785                 if (!journal_path) {
1786                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1787                                 "journal device string");
1788                         return -1;
1789                 }
1790
1791                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1792                 if (error) {
1793                         ext4_msg(sb, KERN_ERR, "error: could not find "
1794                                 "journal device path: error %d", error);
1795                         kfree(journal_path);
1796                         return -1;
1797                 }
1798
1799                 journal_inode = d_inode(path.dentry);
1800                 if (!S_ISBLK(journal_inode->i_mode)) {
1801                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1802                                 "is not a block device", journal_path);
1803                         path_put(&path);
1804                         kfree(journal_path);
1805                         return -1;
1806                 }
1807
1808                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1809                 path_put(&path);
1810                 kfree(journal_path);
1811         } else if (token == Opt_journal_ioprio) {
1812                 if (arg > 7) {
1813                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1814                                  " (must be 0-7)");
1815                         return -1;
1816                 }
1817                 *journal_ioprio =
1818                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1819         } else if (token == Opt_test_dummy_encryption) {
1820 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1821                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1822                 ext4_msg(sb, KERN_WARNING,
1823                          "Test dummy encryption mode enabled");
1824 #else
1825                 ext4_msg(sb, KERN_WARNING,
1826                          "Test dummy encryption mount option ignored");
1827 #endif
1828         } else if (m->flags & MOPT_DATAJ) {
1829                 if (is_remount) {
1830                         if (!sbi->s_journal)
1831                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1832                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1833                                 ext4_msg(sb, KERN_ERR,
1834                                          "Cannot change data mode on remount");
1835                                 return -1;
1836                         }
1837                 } else {
1838                         clear_opt(sb, DATA_FLAGS);
1839                         sbi->s_mount_opt |= m->mount_opt;
1840                 }
1841 #ifdef CONFIG_QUOTA
1842         } else if (m->flags & MOPT_QFMT) {
1843                 if (sb_any_quota_loaded(sb) &&
1844                     sbi->s_jquota_fmt != m->mount_opt) {
1845                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1846                                  "quota options when quota turned on");
1847                         return -1;
1848                 }
1849                 if (ext4_has_feature_quota(sb)) {
1850                         ext4_msg(sb, KERN_INFO,
1851                                  "Quota format mount options ignored "
1852                                  "when QUOTA feature is enabled");
1853                         return 1;
1854                 }
1855                 sbi->s_jquota_fmt = m->mount_opt;
1856 #endif
1857         } else if (token == Opt_dax) {
1858 #ifdef CONFIG_FS_DAX
1859                 ext4_msg(sb, KERN_WARNING,
1860                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1861                         sbi->s_mount_opt |= m->mount_opt;
1862 #else
1863                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1864                 return -1;
1865 #endif
1866         } else if (token == Opt_data_err_abort) {
1867                 sbi->s_mount_opt |= m->mount_opt;
1868         } else if (token == Opt_data_err_ignore) {
1869                 sbi->s_mount_opt &= ~m->mount_opt;
1870         } else {
1871                 if (!args->from)
1872                         arg = 1;
1873                 if (m->flags & MOPT_CLEAR)
1874                         arg = !arg;
1875                 else if (unlikely(!(m->flags & MOPT_SET))) {
1876                         ext4_msg(sb, KERN_WARNING,
1877                                  "buggy handling of option %s", opt);
1878                         WARN_ON(1);
1879                         return -1;
1880                 }
1881                 if (arg != 0)
1882                         sbi->s_mount_opt |= m->mount_opt;
1883                 else
1884                         sbi->s_mount_opt &= ~m->mount_opt;
1885         }
1886         return 1;
1887 }
1888
1889 static int parse_options(char *options, struct super_block *sb,
1890                          unsigned long *journal_devnum,
1891                          unsigned int *journal_ioprio,
1892                          int is_remount)
1893 {
1894         struct ext4_sb_info *sbi = EXT4_SB(sb);
1895         char *p;
1896         substring_t args[MAX_OPT_ARGS];
1897         int token;
1898
1899         if (!options)
1900                 return 1;
1901
1902         while ((p = strsep(&options, ",")) != NULL) {
1903                 if (!*p)
1904                         continue;
1905                 /*
1906                  * Initialize args struct so we know whether arg was
1907                  * found; some options take optional arguments.
1908                  */
1909                 args[0].to = args[0].from = NULL;
1910                 token = match_token(p, tokens, args);
1911                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1912                                      journal_ioprio, is_remount) < 0)
1913                         return 0;
1914         }
1915 #ifdef CONFIG_QUOTA
1916         /*
1917          * We do the test below only for project quotas. 'usrquota' and
1918          * 'grpquota' mount options are allowed even without quota feature
1919          * to support legacy quotas in quota files.
1920          */
1921         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1922                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1923                          "Cannot enable project quota enforcement.");
1924                 return 0;
1925         }
1926         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1927                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1928                         clear_opt(sb, USRQUOTA);
1929
1930                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1931                         clear_opt(sb, GRPQUOTA);
1932
1933                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1934                         ext4_msg(sb, KERN_ERR, "old and new quota "
1935                                         "format mixing");
1936                         return 0;
1937                 }
1938
1939                 if (!sbi->s_jquota_fmt) {
1940                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1941                                         "not specified");
1942                         return 0;
1943                 }
1944         }
1945 #endif
1946         if (test_opt(sb, DIOREAD_NOLOCK)) {
1947                 int blocksize =
1948                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1949
1950                 if (blocksize < PAGE_SIZE) {
1951                         ext4_msg(sb, KERN_ERR, "can't mount with "
1952                                  "dioread_nolock if block size != PAGE_SIZE");
1953                         return 0;
1954                 }
1955         }
1956         return 1;
1957 }
1958
1959 static inline void ext4_show_quota_options(struct seq_file *seq,
1960                                            struct super_block *sb)
1961 {
1962 #if defined(CONFIG_QUOTA)
1963         struct ext4_sb_info *sbi = EXT4_SB(sb);
1964
1965         if (sbi->s_jquota_fmt) {
1966                 char *fmtname = "";
1967
1968                 switch (sbi->s_jquota_fmt) {
1969                 case QFMT_VFS_OLD:
1970                         fmtname = "vfsold";
1971                         break;
1972                 case QFMT_VFS_V0:
1973                         fmtname = "vfsv0";
1974                         break;
1975                 case QFMT_VFS_V1:
1976                         fmtname = "vfsv1";
1977                         break;
1978                 }
1979                 seq_printf(seq, ",jqfmt=%s", fmtname);
1980         }
1981
1982         if (sbi->s_qf_names[USRQUOTA])
1983                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1984
1985         if (sbi->s_qf_names[GRPQUOTA])
1986                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1987 #endif
1988 }
1989
1990 static const char *token2str(int token)
1991 {
1992         const struct match_token *t;
1993
1994         for (t = tokens; t->token != Opt_err; t++)
1995                 if (t->token == token && !strchr(t->pattern, '='))
1996                         break;
1997         return t->pattern;
1998 }
1999
2000 /*
2001  * Show an option if
2002  *  - it's set to a non-default value OR
2003  *  - if the per-sb default is different from the global default
2004  */
2005 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2006                               int nodefs)
2007 {
2008         struct ext4_sb_info *sbi = EXT4_SB(sb);
2009         struct ext4_super_block *es = sbi->s_es;
2010         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2011         const struct mount_opts *m;
2012         char sep = nodefs ? '\n' : ',';
2013
2014 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2015 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2016
2017         if (sbi->s_sb_block != 1)
2018                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2019
2020         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2021                 int want_set = m->flags & MOPT_SET;
2022                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2023                     (m->flags & MOPT_CLEAR_ERR))
2024                         continue;
2025                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2026                         continue; /* skip if same as the default */
2027                 if ((want_set &&
2028                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2029                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2030                         continue; /* select Opt_noFoo vs Opt_Foo */
2031                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2032         }
2033
2034         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2035             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2036                 SEQ_OPTS_PRINT("resuid=%u",
2037                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2038         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2039             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2040                 SEQ_OPTS_PRINT("resgid=%u",
2041                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2042         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2043         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2044                 SEQ_OPTS_PUTS("errors=remount-ro");
2045         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2046                 SEQ_OPTS_PUTS("errors=continue");
2047         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2048                 SEQ_OPTS_PUTS("errors=panic");
2049         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2050                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2051         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2052                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2053         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2054                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2055         if (sb->s_flags & MS_I_VERSION)
2056                 SEQ_OPTS_PUTS("i_version");
2057         if (nodefs || sbi->s_stripe)
2058                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2059         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2060                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2061                         SEQ_OPTS_PUTS("data=journal");
2062                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2063                         SEQ_OPTS_PUTS("data=ordered");
2064                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2065                         SEQ_OPTS_PUTS("data=writeback");
2066         }
2067         if (nodefs ||
2068             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2069                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2070                                sbi->s_inode_readahead_blks);
2071
2072         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2073                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2074                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2075         if (nodefs || sbi->s_max_dir_size_kb)
2076                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2077         if (test_opt(sb, DATA_ERR_ABORT))
2078                 SEQ_OPTS_PUTS("data_err=abort");
2079
2080         ext4_show_quota_options(seq, sb);
2081         return 0;
2082 }
2083
2084 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2085 {
2086         return _ext4_show_options(seq, root->d_sb, 0);
2087 }
2088
2089 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2090 {
2091         struct super_block *sb = seq->private;
2092         int rc;
2093
2094         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2095         rc = _ext4_show_options(seq, sb, 1);
2096         seq_puts(seq, "\n");
2097         return rc;
2098 }
2099
2100 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2101                             int read_only)
2102 {
2103         struct ext4_sb_info *sbi = EXT4_SB(sb);
2104         int res = 0;
2105
2106         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2107                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2108                          "forcing read-only mode");
2109                 res = MS_RDONLY;
2110         }
2111         if (read_only)
2112                 goto done;
2113         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2114                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2115                          "running e2fsck is recommended");
2116         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2117                 ext4_msg(sb, KERN_WARNING,
2118                          "warning: mounting fs with errors, "
2119                          "running e2fsck is recommended");
2120         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2121                  le16_to_cpu(es->s_mnt_count) >=
2122                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2123                 ext4_msg(sb, KERN_WARNING,
2124                          "warning: maximal mount count reached, "
2125                          "running e2fsck is recommended");
2126         else if (le32_to_cpu(es->s_checkinterval) &&
2127                 (le32_to_cpu(es->s_lastcheck) +
2128                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2129                 ext4_msg(sb, KERN_WARNING,
2130                          "warning: checktime reached, "
2131                          "running e2fsck is recommended");
2132         if (!sbi->s_journal)
2133                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2134         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2135                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2136         le16_add_cpu(&es->s_mnt_count, 1);
2137         es->s_mtime = cpu_to_le32(get_seconds());
2138         ext4_update_dynamic_rev(sb);
2139         if (sbi->s_journal)
2140                 ext4_set_feature_journal_needs_recovery(sb);
2141
2142         ext4_commit_super(sb, 1);
2143 done:
2144         if (test_opt(sb, DEBUG))
2145                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2146                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2147                         sb->s_blocksize,
2148                         sbi->s_groups_count,
2149                         EXT4_BLOCKS_PER_GROUP(sb),
2150                         EXT4_INODES_PER_GROUP(sb),
2151                         sbi->s_mount_opt, sbi->s_mount_opt2);
2152
2153         cleancache_init_fs(sb);
2154         return res;
2155 }
2156
2157 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2158 {
2159         struct ext4_sb_info *sbi = EXT4_SB(sb);
2160         struct flex_groups *new_groups;
2161         int size;
2162
2163         if (!sbi->s_log_groups_per_flex)
2164                 return 0;
2165
2166         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2167         if (size <= sbi->s_flex_groups_allocated)
2168                 return 0;
2169
2170         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2171         new_groups = kvzalloc(size, GFP_KERNEL);
2172         if (!new_groups) {
2173                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2174                          size / (int) sizeof(struct flex_groups));
2175                 return -ENOMEM;
2176         }
2177
2178         if (sbi->s_flex_groups) {
2179                 memcpy(new_groups, sbi->s_flex_groups,
2180                        (sbi->s_flex_groups_allocated *
2181                         sizeof(struct flex_groups)));
2182                 kvfree(sbi->s_flex_groups);
2183         }
2184         sbi->s_flex_groups = new_groups;
2185         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2186         return 0;
2187 }
2188
2189 static int ext4_fill_flex_info(struct super_block *sb)
2190 {
2191         struct ext4_sb_info *sbi = EXT4_SB(sb);
2192         struct ext4_group_desc *gdp = NULL;
2193         ext4_group_t flex_group;
2194         int i, err;
2195
2196         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2197         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2198                 sbi->s_log_groups_per_flex = 0;
2199                 return 1;
2200         }
2201
2202         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2203         if (err)
2204                 goto failed;
2205
2206         for (i = 0; i < sbi->s_groups_count; i++) {
2207                 gdp = ext4_get_group_desc(sb, i, NULL);
2208
2209                 flex_group = ext4_flex_group(sbi, i);
2210                 atomic_add(ext4_free_inodes_count(sb, gdp),
2211                            &sbi->s_flex_groups[flex_group].free_inodes);
2212                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2213                              &sbi->s_flex_groups[flex_group].free_clusters);
2214                 atomic_add(ext4_used_dirs_count(sb, gdp),
2215                            &sbi->s_flex_groups[flex_group].used_dirs);
2216         }
2217
2218         return 1;
2219 failed:
2220         return 0;
2221 }
2222
2223 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2224                                    struct ext4_group_desc *gdp)
2225 {
2226         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2227         __u16 crc = 0;
2228         __le32 le_group = cpu_to_le32(block_group);
2229         struct ext4_sb_info *sbi = EXT4_SB(sb);
2230
2231         if (ext4_has_metadata_csum(sbi->s_sb)) {
2232                 /* Use new metadata_csum algorithm */
2233                 __u32 csum32;
2234                 __u16 dummy_csum = 0;
2235
2236                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2237                                      sizeof(le_group));
2238                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2239                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2240                                      sizeof(dummy_csum));
2241                 offset += sizeof(dummy_csum);
2242                 if (offset < sbi->s_desc_size)
2243                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2244                                              sbi->s_desc_size - offset);
2245
2246                 crc = csum32 & 0xFFFF;
2247                 goto out;
2248         }
2249
2250         /* old crc16 code */
2251         if (!ext4_has_feature_gdt_csum(sb))
2252                 return 0;
2253
2254         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2255         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2256         crc = crc16(crc, (__u8 *)gdp, offset);
2257         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2258         /* for checksum of struct ext4_group_desc do the rest...*/
2259         if (ext4_has_feature_64bit(sb) &&
2260             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2261                 crc = crc16(crc, (__u8 *)gdp + offset,
2262                             le16_to_cpu(sbi->s_es->s_desc_size) -
2263                                 offset);
2264
2265 out:
2266         return cpu_to_le16(crc);
2267 }
2268
2269 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2270                                 struct ext4_group_desc *gdp)
2271 {
2272         if (ext4_has_group_desc_csum(sb) &&
2273             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2274                 return 0;
2275
2276         return 1;
2277 }
2278
2279 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2280                               struct ext4_group_desc *gdp)
2281 {
2282         if (!ext4_has_group_desc_csum(sb))
2283                 return;
2284         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2285 }
2286
2287 /* Called at mount-time, super-block is locked */
2288 static int ext4_check_descriptors(struct super_block *sb,
2289                                   ext4_fsblk_t sb_block,
2290                                   ext4_group_t *first_not_zeroed)
2291 {
2292         struct ext4_sb_info *sbi = EXT4_SB(sb);
2293         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2294         ext4_fsblk_t last_block;
2295         ext4_fsblk_t block_bitmap;
2296         ext4_fsblk_t inode_bitmap;
2297         ext4_fsblk_t inode_table;
2298         int flexbg_flag = 0;
2299         ext4_group_t i, grp = sbi->s_groups_count;
2300
2301         if (ext4_has_feature_flex_bg(sb))
2302                 flexbg_flag = 1;
2303
2304         ext4_debug("Checking group descriptors");
2305
2306         for (i = 0; i < sbi->s_groups_count; i++) {
2307                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2308
2309                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2310                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2311                 else
2312                         last_block = first_block +
2313                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2314
2315                 if ((grp == sbi->s_groups_count) &&
2316                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2317                         grp = i;
2318
2319                 block_bitmap = ext4_block_bitmap(sb, gdp);
2320                 if (block_bitmap == sb_block) {
2321                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2322                                  "Block bitmap for group %u overlaps "
2323                                  "superblock", i);
2324                 }
2325                 if (block_bitmap < first_block || block_bitmap > last_block) {
2326                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2327                                "Block bitmap for group %u not in group "
2328                                "(block %llu)!", i, block_bitmap);
2329                         return 0;
2330                 }
2331                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2332                 if (inode_bitmap == sb_block) {
2333                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2334                                  "Inode bitmap for group %u overlaps "
2335                                  "superblock", i);
2336                 }
2337                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2338                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2339                                "Inode bitmap for group %u not in group "
2340                                "(block %llu)!", i, inode_bitmap);
2341                         return 0;
2342                 }
2343                 inode_table = ext4_inode_table(sb, gdp);
2344                 if (inode_table == sb_block) {
2345                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2346                                  "Inode table for group %u overlaps "
2347                                  "superblock", i);
2348                 }
2349                 if (inode_table < first_block ||
2350                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2351                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2352                                "Inode table for group %u not in group "
2353                                "(block %llu)!", i, inode_table);
2354                         return 0;
2355                 }
2356                 ext4_lock_group(sb, i);
2357                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2358                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2359                                  "Checksum for group %u failed (%u!=%u)",
2360                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2361                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2362                         if (!(sb->s_flags & MS_RDONLY)) {
2363                                 ext4_unlock_group(sb, i);
2364                                 return 0;
2365                         }
2366                 }
2367                 ext4_unlock_group(sb, i);
2368                 if (!flexbg_flag)
2369                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2370         }
2371         if (NULL != first_not_zeroed)
2372                 *first_not_zeroed = grp;
2373         return 1;
2374 }
2375
2376 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2377  * the superblock) which were deleted from all directories, but held open by
2378  * a process at the time of a crash.  We walk the list and try to delete these
2379  * inodes at recovery time (only with a read-write filesystem).
2380  *
2381  * In order to keep the orphan inode chain consistent during traversal (in
2382  * case of crash during recovery), we link each inode into the superblock
2383  * orphan list_head and handle it the same way as an inode deletion during
2384  * normal operation (which journals the operations for us).
2385  *
2386  * We only do an iget() and an iput() on each inode, which is very safe if we
2387  * accidentally point at an in-use or already deleted inode.  The worst that
2388  * can happen in this case is that we get a "bit already cleared" message from
2389  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2390  * e2fsck was run on this filesystem, and it must have already done the orphan
2391  * inode cleanup for us, so we can safely abort without any further action.
2392  */
2393 static void ext4_orphan_cleanup(struct super_block *sb,
2394                                 struct ext4_super_block *es)
2395 {
2396         unsigned int s_flags = sb->s_flags;
2397         int ret, nr_orphans = 0, nr_truncates = 0;
2398 #ifdef CONFIG_QUOTA
2399         int i;
2400 #endif
2401         if (!es->s_last_orphan) {
2402                 jbd_debug(4, "no orphan inodes to clean up\n");
2403                 return;
2404         }
2405
2406         if (bdev_read_only(sb->s_bdev)) {
2407                 ext4_msg(sb, KERN_ERR, "write access "
2408                         "unavailable, skipping orphan cleanup");
2409                 return;
2410         }
2411
2412         /* Check if feature set would not allow a r/w mount */
2413         if (!ext4_feature_set_ok(sb, 0)) {
2414                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2415                          "unknown ROCOMPAT features");
2416                 return;
2417         }
2418
2419         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2420                 /* don't clear list on RO mount w/ errors */
2421                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2422                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2423                                   "clearing orphan list.\n");
2424                         es->s_last_orphan = 0;
2425                 }
2426                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2427                 return;
2428         }
2429
2430         if (s_flags & MS_RDONLY) {
2431                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2432                 sb->s_flags &= ~MS_RDONLY;
2433         }
2434 #ifdef CONFIG_QUOTA
2435         /* Needed for iput() to work correctly and not trash data */
2436         sb->s_flags |= MS_ACTIVE;
2437         /* Turn on quotas so that they are updated correctly */
2438         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2439                 if (EXT4_SB(sb)->s_qf_names[i]) {
2440                         int ret = ext4_quota_on_mount(sb, i);
2441                         if (ret < 0)
2442                                 ext4_msg(sb, KERN_ERR,
2443                                         "Cannot turn on journaled "
2444                                         "quota: error %d", ret);
2445                 }
2446         }
2447 #endif
2448
2449         while (es->s_last_orphan) {
2450                 struct inode *inode;
2451
2452                 /*
2453                  * We may have encountered an error during cleanup; if
2454                  * so, skip the rest.
2455                  */
2456                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2457                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2458                         es->s_last_orphan = 0;
2459                         break;
2460                 }
2461
2462                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2463                 if (IS_ERR(inode)) {
2464                         es->s_last_orphan = 0;
2465                         break;
2466                 }
2467
2468                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2469                 dquot_initialize(inode);
2470                 if (inode->i_nlink) {
2471                         if (test_opt(sb, DEBUG))
2472                                 ext4_msg(sb, KERN_DEBUG,
2473                                         "%s: truncating inode %lu to %lld bytes",
2474                                         __func__, inode->i_ino, inode->i_size);
2475                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2476                                   inode->i_ino, inode->i_size);
2477                         inode_lock(inode);
2478                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2479                         ret = ext4_truncate(inode);
2480                         if (ret)
2481                                 ext4_std_error(inode->i_sb, ret);
2482                         inode_unlock(inode);
2483                         nr_truncates++;
2484                 } else {
2485                         if (test_opt(sb, DEBUG))
2486                                 ext4_msg(sb, KERN_DEBUG,
2487                                         "%s: deleting unreferenced inode %lu",
2488                                         __func__, inode->i_ino);
2489                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2490                                   inode->i_ino);
2491                         nr_orphans++;
2492                 }
2493                 iput(inode);  /* The delete magic happens here! */
2494         }
2495
2496 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2497
2498         if (nr_orphans)
2499                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2500                        PLURAL(nr_orphans));
2501         if (nr_truncates)
2502                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2503                        PLURAL(nr_truncates));
2504 #ifdef CONFIG_QUOTA
2505         /* Turn quotas off */
2506         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2507                 if (sb_dqopt(sb)->files[i])
2508                         dquot_quota_off(sb, i);
2509         }
2510 #endif
2511         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2512 }
2513
2514 /*
2515  * Maximal extent format file size.
2516  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2517  * extent format containers, within a sector_t, and within i_blocks
2518  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2519  * so that won't be a limiting factor.
2520  *
2521  * However there is other limiting factor. We do store extents in the form
2522  * of starting block and length, hence the resulting length of the extent
2523  * covering maximum file size must fit into on-disk format containers as
2524  * well. Given that length is always by 1 unit bigger than max unit (because
2525  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2526  *
2527  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2528  */
2529 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2530 {
2531         loff_t res;
2532         loff_t upper_limit = MAX_LFS_FILESIZE;
2533
2534         /* small i_blocks in vfs inode? */
2535         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2536                 /*
2537                  * CONFIG_LBDAF is not enabled implies the inode
2538                  * i_block represent total blocks in 512 bytes
2539                  * 32 == size of vfs inode i_blocks * 8
2540                  */
2541                 upper_limit = (1LL << 32) - 1;
2542
2543                 /* total blocks in file system block size */
2544                 upper_limit >>= (blkbits - 9);
2545                 upper_limit <<= blkbits;
2546         }
2547
2548         /*
2549          * 32-bit extent-start container, ee_block. We lower the maxbytes
2550          * by one fs block, so ee_len can cover the extent of maximum file
2551          * size
2552          */
2553         res = (1LL << 32) - 1;
2554         res <<= blkbits;
2555
2556         /* Sanity check against vm- & vfs- imposed limits */
2557         if (res > upper_limit)
2558                 res = upper_limit;
2559
2560         return res;
2561 }
2562
2563 /*
2564  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2565  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2566  * We need to be 1 filesystem block less than the 2^48 sector limit.
2567  */
2568 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2569 {
2570         loff_t res = EXT4_NDIR_BLOCKS;
2571         int meta_blocks;
2572         loff_t upper_limit;
2573         /* This is calculated to be the largest file size for a dense, block
2574          * mapped file such that the file's total number of 512-byte sectors,
2575          * including data and all indirect blocks, does not exceed (2^48 - 1).
2576          *
2577          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2578          * number of 512-byte sectors of the file.
2579          */
2580
2581         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2582                 /*
2583                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2584                  * the inode i_block field represents total file blocks in
2585                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2586                  */
2587                 upper_limit = (1LL << 32) - 1;
2588
2589                 /* total blocks in file system block size */
2590                 upper_limit >>= (bits - 9);
2591
2592         } else {
2593                 /*
2594                  * We use 48 bit ext4_inode i_blocks
2595                  * With EXT4_HUGE_FILE_FL set the i_blocks
2596                  * represent total number of blocks in
2597                  * file system block size
2598                  */
2599                 upper_limit = (1LL << 48) - 1;
2600
2601         }
2602
2603         /* indirect blocks */
2604         meta_blocks = 1;
2605         /* double indirect blocks */
2606         meta_blocks += 1 + (1LL << (bits-2));
2607         /* tripple indirect blocks */
2608         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2609
2610         upper_limit -= meta_blocks;
2611         upper_limit <<= bits;
2612
2613         res += 1LL << (bits-2);
2614         res += 1LL << (2*(bits-2));
2615         res += 1LL << (3*(bits-2));
2616         res <<= bits;
2617         if (res > upper_limit)
2618                 res = upper_limit;
2619
2620         if (res > MAX_LFS_FILESIZE)
2621                 res = MAX_LFS_FILESIZE;
2622
2623         return res;
2624 }
2625
2626 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2627                                    ext4_fsblk_t logical_sb_block, int nr)
2628 {
2629         struct ext4_sb_info *sbi = EXT4_SB(sb);
2630         ext4_group_t bg, first_meta_bg;
2631         int has_super = 0;
2632
2633         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2634
2635         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2636                 return logical_sb_block + nr + 1;
2637         bg = sbi->s_desc_per_block * nr;
2638         if (ext4_bg_has_super(sb, bg))
2639                 has_super = 1;
2640
2641         /*
2642          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2643          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2644          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2645          * compensate.
2646          */
2647         if (sb->s_blocksize == 1024 && nr == 0 &&
2648             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2649                 has_super++;
2650
2651         return (has_super + ext4_group_first_block_no(sb, bg));
2652 }
2653
2654 /**
2655  * ext4_get_stripe_size: Get the stripe size.
2656  * @sbi: In memory super block info
2657  *
2658  * If we have specified it via mount option, then
2659  * use the mount option value. If the value specified at mount time is
2660  * greater than the blocks per group use the super block value.
2661  * If the super block value is greater than blocks per group return 0.
2662  * Allocator needs it be less than blocks per group.
2663  *
2664  */
2665 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2666 {
2667         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2668         unsigned long stripe_width =
2669                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2670         int ret;
2671
2672         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2673                 ret = sbi->s_stripe;
2674         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2675                 ret = stripe_width;
2676         else if (stride && stride <= sbi->s_blocks_per_group)
2677                 ret = stride;
2678         else
2679                 ret = 0;
2680
2681         /*
2682          * If the stripe width is 1, this makes no sense and
2683          * we set it to 0 to turn off stripe handling code.
2684          */
2685         if (ret <= 1)
2686                 ret = 0;
2687
2688         return ret;
2689 }
2690
2691 /*
2692  * Check whether this filesystem can be mounted based on
2693  * the features present and the RDONLY/RDWR mount requested.
2694  * Returns 1 if this filesystem can be mounted as requested,
2695  * 0 if it cannot be.
2696  */
2697 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2698 {
2699         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2700                 ext4_msg(sb, KERN_ERR,
2701                         "Couldn't mount because of "
2702                         "unsupported optional features (%x)",
2703                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2704                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2705                 return 0;
2706         }
2707
2708         if (readonly)
2709                 return 1;
2710
2711         if (ext4_has_feature_readonly(sb)) {
2712                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2713                 sb->s_flags |= MS_RDONLY;
2714                 return 1;
2715         }
2716
2717         /* Check that feature set is OK for a read-write mount */
2718         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2719                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2720                          "unsupported optional features (%x)",
2721                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2722                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2723                 return 0;
2724         }
2725         /*
2726          * Large file size enabled file system can only be mounted
2727          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2728          */
2729         if (ext4_has_feature_huge_file(sb)) {
2730                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2731                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2732                                  "cannot be mounted RDWR without "
2733                                  "CONFIG_LBDAF");
2734                         return 0;
2735                 }
2736         }
2737         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2738                 ext4_msg(sb, KERN_ERR,
2739                          "Can't support bigalloc feature without "
2740                          "extents feature\n");
2741                 return 0;
2742         }
2743
2744 #ifndef CONFIG_QUOTA
2745         if (ext4_has_feature_quota(sb) && !readonly) {
2746                 ext4_msg(sb, KERN_ERR,
2747                          "Filesystem with quota feature cannot be mounted RDWR "
2748                          "without CONFIG_QUOTA");
2749                 return 0;
2750         }
2751         if (ext4_has_feature_project(sb) && !readonly) {
2752                 ext4_msg(sb, KERN_ERR,
2753                          "Filesystem with project quota feature cannot be mounted RDWR "
2754                          "without CONFIG_QUOTA");
2755                 return 0;
2756         }
2757 #endif  /* CONFIG_QUOTA */
2758         return 1;
2759 }
2760
2761 /*
2762  * This function is called once a day if we have errors logged
2763  * on the file system
2764  */
2765 static void print_daily_error_info(unsigned long arg)
2766 {
2767         struct super_block *sb = (struct super_block *) arg;
2768         struct ext4_sb_info *sbi;
2769         struct ext4_super_block *es;
2770
2771         sbi = EXT4_SB(sb);
2772         es = sbi->s_es;
2773
2774         if (es->s_error_count)
2775                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2776                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2777                          le32_to_cpu(es->s_error_count));
2778         if (es->s_first_error_time) {
2779                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2780                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2781                        (int) sizeof(es->s_first_error_func),
2782                        es->s_first_error_func,
2783                        le32_to_cpu(es->s_first_error_line));
2784                 if (es->s_first_error_ino)
2785                         printk(KERN_CONT ": inode %u",
2786                                le32_to_cpu(es->s_first_error_ino));
2787                 if (es->s_first_error_block)
2788                         printk(KERN_CONT ": block %llu", (unsigned long long)
2789                                le64_to_cpu(es->s_first_error_block));
2790                 printk(KERN_CONT "\n");
2791         }
2792         if (es->s_last_error_time) {
2793                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2794                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2795                        (int) sizeof(es->s_last_error_func),
2796                        es->s_last_error_func,
2797                        le32_to_cpu(es->s_last_error_line));
2798                 if (es->s_last_error_ino)
2799                         printk(KERN_CONT ": inode %u",
2800                                le32_to_cpu(es->s_last_error_ino));
2801                 if (es->s_last_error_block)
2802                         printk(KERN_CONT ": block %llu", (unsigned long long)
2803                                le64_to_cpu(es->s_last_error_block));
2804                 printk(KERN_CONT "\n");
2805         }
2806         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2807 }
2808
2809 /* Find next suitable group and run ext4_init_inode_table */
2810 static int ext4_run_li_request(struct ext4_li_request *elr)
2811 {
2812         struct ext4_group_desc *gdp = NULL;
2813         ext4_group_t group, ngroups;
2814         struct super_block *sb;
2815         unsigned long timeout = 0;
2816         int ret = 0;
2817
2818         sb = elr->lr_super;
2819         ngroups = EXT4_SB(sb)->s_groups_count;
2820
2821         for (group = elr->lr_next_group; group < ngroups; group++) {
2822                 gdp = ext4_get_group_desc(sb, group, NULL);
2823                 if (!gdp) {
2824                         ret = 1;
2825                         break;
2826                 }
2827
2828                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2829                         break;
2830         }
2831
2832         if (group >= ngroups)
2833                 ret = 1;
2834
2835         if (!ret) {
2836                 timeout = jiffies;
2837                 ret = ext4_init_inode_table(sb, group,
2838                                             elr->lr_timeout ? 0 : 1);
2839                 if (elr->lr_timeout == 0) {
2840                         timeout = (jiffies - timeout) *
2841                                   elr->lr_sbi->s_li_wait_mult;
2842                         elr->lr_timeout = timeout;
2843                 }
2844                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2845                 elr->lr_next_group = group + 1;
2846         }
2847         return ret;
2848 }
2849
2850 /*
2851  * Remove lr_request from the list_request and free the
2852  * request structure. Should be called with li_list_mtx held
2853  */
2854 static void ext4_remove_li_request(struct ext4_li_request *elr)
2855 {
2856         struct ext4_sb_info *sbi;
2857
2858         if (!elr)
2859                 return;
2860
2861         sbi = elr->lr_sbi;
2862
2863         list_del(&elr->lr_request);
2864         sbi->s_li_request = NULL;
2865         kfree(elr);
2866 }
2867
2868 static void ext4_unregister_li_request(struct super_block *sb)
2869 {
2870         mutex_lock(&ext4_li_mtx);
2871         if (!ext4_li_info) {
2872                 mutex_unlock(&ext4_li_mtx);
2873                 return;
2874         }
2875
2876         mutex_lock(&ext4_li_info->li_list_mtx);
2877         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2878         mutex_unlock(&ext4_li_info->li_list_mtx);
2879         mutex_unlock(&ext4_li_mtx);
2880 }
2881
2882 static struct task_struct *ext4_lazyinit_task;
2883
2884 /*
2885  * This is the function where ext4lazyinit thread lives. It walks
2886  * through the request list searching for next scheduled filesystem.
2887  * When such a fs is found, run the lazy initialization request
2888  * (ext4_rn_li_request) and keep track of the time spend in this
2889  * function. Based on that time we compute next schedule time of
2890  * the request. When walking through the list is complete, compute
2891  * next waking time and put itself into sleep.
2892  */
2893 static int ext4_lazyinit_thread(void *arg)
2894 {
2895         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2896         struct list_head *pos, *n;
2897         struct ext4_li_request *elr;
2898         unsigned long next_wakeup, cur;
2899
2900         BUG_ON(NULL == eli);
2901
2902 cont_thread:
2903         while (true) {
2904                 next_wakeup = MAX_JIFFY_OFFSET;
2905
2906                 mutex_lock(&eli->li_list_mtx);
2907                 if (list_empty(&eli->li_request_list)) {
2908                         mutex_unlock(&eli->li_list_mtx);
2909                         goto exit_thread;
2910                 }
2911                 list_for_each_safe(pos, n, &eli->li_request_list) {
2912                         int err = 0;
2913                         int progress = 0;
2914                         elr = list_entry(pos, struct ext4_li_request,
2915                                          lr_request);
2916
2917                         if (time_before(jiffies, elr->lr_next_sched)) {
2918                                 if (time_before(elr->lr_next_sched, next_wakeup))
2919                                         next_wakeup = elr->lr_next_sched;
2920                                 continue;
2921                         }
2922                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2923                                 if (sb_start_write_trylock(elr->lr_super)) {
2924                                         progress = 1;
2925                                         /*
2926                                          * We hold sb->s_umount, sb can not
2927                                          * be removed from the list, it is
2928                                          * now safe to drop li_list_mtx
2929                                          */
2930                                         mutex_unlock(&eli->li_list_mtx);
2931                                         err = ext4_run_li_request(elr);
2932                                         sb_end_write(elr->lr_super);
2933                                         mutex_lock(&eli->li_list_mtx);
2934                                         n = pos->next;
2935                                 }
2936                                 up_read((&elr->lr_super->s_umount));
2937                         }
2938                         /* error, remove the lazy_init job */
2939                         if (err) {
2940                                 ext4_remove_li_request(elr);
2941                                 continue;
2942                         }
2943                         if (!progress) {
2944                                 elr->lr_next_sched = jiffies +
2945                                         (prandom_u32()
2946                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2947                         }
2948                         if (time_before(elr->lr_next_sched, next_wakeup))
2949                                 next_wakeup = elr->lr_next_sched;
2950                 }
2951                 mutex_unlock(&eli->li_list_mtx);
2952
2953                 try_to_freeze();
2954
2955                 cur = jiffies;
2956                 if ((time_after_eq(cur, next_wakeup)) ||
2957                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2958                         cond_resched();
2959                         continue;
2960                 }
2961
2962                 schedule_timeout_interruptible(next_wakeup - cur);
2963
2964                 if (kthread_should_stop()) {
2965                         ext4_clear_request_list();
2966                         goto exit_thread;
2967                 }
2968         }
2969
2970 exit_thread:
2971         /*
2972          * It looks like the request list is empty, but we need
2973          * to check it under the li_list_mtx lock, to prevent any
2974          * additions into it, and of course we should lock ext4_li_mtx
2975          * to atomically free the list and ext4_li_info, because at
2976          * this point another ext4 filesystem could be registering
2977          * new one.
2978          */
2979         mutex_lock(&ext4_li_mtx);
2980         mutex_lock(&eli->li_list_mtx);
2981         if (!list_empty(&eli->li_request_list)) {
2982                 mutex_unlock(&eli->li_list_mtx);
2983                 mutex_unlock(&ext4_li_mtx);
2984                 goto cont_thread;
2985         }
2986         mutex_unlock(&eli->li_list_mtx);
2987         kfree(ext4_li_info);
2988         ext4_li_info = NULL;
2989         mutex_unlock(&ext4_li_mtx);
2990
2991         return 0;
2992 }
2993
2994 static void ext4_clear_request_list(void)
2995 {
2996         struct list_head *pos, *n;
2997         struct ext4_li_request *elr;
2998
2999         mutex_lock(&ext4_li_info->li_list_mtx);
3000         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3001                 elr = list_entry(pos, struct ext4_li_request,
3002                                  lr_request);
3003                 ext4_remove_li_request(elr);
3004         }
3005         mutex_unlock(&ext4_li_info->li_list_mtx);
3006 }
3007
3008 static int ext4_run_lazyinit_thread(void)
3009 {
3010         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3011                                          ext4_li_info, "ext4lazyinit");
3012         if (IS_ERR(ext4_lazyinit_task)) {
3013                 int err = PTR_ERR(ext4_lazyinit_task);
3014                 ext4_clear_request_list();
3015                 kfree(ext4_li_info);
3016                 ext4_li_info = NULL;
3017                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3018                                  "initialization thread\n",
3019                                  err);
3020                 return err;
3021         }
3022         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3023         return 0;
3024 }
3025
3026 /*
3027  * Check whether it make sense to run itable init. thread or not.
3028  * If there is at least one uninitialized inode table, return
3029  * corresponding group number, else the loop goes through all
3030  * groups and return total number of groups.
3031  */
3032 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3033 {
3034         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3035         struct ext4_group_desc *gdp = NULL;
3036
3037         for (group = 0; group < ngroups; group++) {
3038                 gdp = ext4_get_group_desc(sb, group, NULL);
3039                 if (!gdp)
3040                         continue;
3041
3042                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3043                         break;
3044         }
3045
3046         return group;
3047 }
3048
3049 static int ext4_li_info_new(void)
3050 {
3051         struct ext4_lazy_init *eli = NULL;
3052
3053         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3054         if (!eli)
3055                 return -ENOMEM;
3056
3057         INIT_LIST_HEAD(&eli->li_request_list);
3058         mutex_init(&eli->li_list_mtx);
3059
3060         eli->li_state |= EXT4_LAZYINIT_QUIT;
3061
3062         ext4_li_info = eli;
3063
3064         return 0;
3065 }
3066
3067 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3068                                             ext4_group_t start)
3069 {
3070         struct ext4_sb_info *sbi = EXT4_SB(sb);
3071         struct ext4_li_request *elr;
3072
3073         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3074         if (!elr)
3075                 return NULL;
3076
3077         elr->lr_super = sb;
3078         elr->lr_sbi = sbi;
3079         elr->lr_next_group = start;
3080
3081         /*
3082          * Randomize first schedule time of the request to
3083          * spread the inode table initialization requests
3084          * better.
3085          */
3086         elr->lr_next_sched = jiffies + (prandom_u32() %
3087                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3088         return elr;
3089 }
3090
3091 int ext4_register_li_request(struct super_block *sb,
3092                              ext4_group_t first_not_zeroed)
3093 {
3094         struct ext4_sb_info *sbi = EXT4_SB(sb);
3095         struct ext4_li_request *elr = NULL;
3096         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3097         int ret = 0;
3098
3099         mutex_lock(&ext4_li_mtx);
3100         if (sbi->s_li_request != NULL) {
3101                 /*
3102                  * Reset timeout so it can be computed again, because
3103                  * s_li_wait_mult might have changed.
3104                  */
3105                 sbi->s_li_request->lr_timeout = 0;
3106                 goto out;
3107         }
3108
3109         if (first_not_zeroed == ngroups ||
3110             (sb->s_flags & MS_RDONLY) ||
3111             !test_opt(sb, INIT_INODE_TABLE))
3112                 goto out;
3113
3114         elr = ext4_li_request_new(sb, first_not_zeroed);
3115         if (!elr) {
3116                 ret = -ENOMEM;
3117                 goto out;
3118         }
3119
3120         if (NULL == ext4_li_info) {
3121                 ret = ext4_li_info_new();
3122                 if (ret)
3123                         goto out;
3124         }
3125
3126         mutex_lock(&ext4_li_info->li_list_mtx);
3127         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3128         mutex_unlock(&ext4_li_info->li_list_mtx);
3129
3130         sbi->s_li_request = elr;
3131         /*
3132          * set elr to NULL here since it has been inserted to
3133          * the request_list and the removal and free of it is
3134          * handled by ext4_clear_request_list from now on.
3135          */
3136         elr = NULL;
3137
3138         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3139                 ret = ext4_run_lazyinit_thread();
3140                 if (ret)
3141                         goto out;
3142         }
3143 out:
3144         mutex_unlock(&ext4_li_mtx);
3145         if (ret)
3146                 kfree(elr);
3147         return ret;
3148 }
3149
3150 /*
3151  * We do not need to lock anything since this is called on
3152  * module unload.
3153  */
3154 static void ext4_destroy_lazyinit_thread(void)
3155 {
3156         /*
3157          * If thread exited earlier
3158          * there's nothing to be done.
3159          */
3160         if (!ext4_li_info || !ext4_lazyinit_task)
3161                 return;
3162
3163         kthread_stop(ext4_lazyinit_task);
3164 }
3165
3166 static int set_journal_csum_feature_set(struct super_block *sb)
3167 {
3168         int ret = 1;
3169         int compat, incompat;
3170         struct ext4_sb_info *sbi = EXT4_SB(sb);
3171
3172         if (ext4_has_metadata_csum(sb)) {
3173                 /* journal checksum v3 */
3174                 compat = 0;
3175                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3176         } else {
3177                 /* journal checksum v1 */
3178                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3179                 incompat = 0;
3180         }
3181
3182         jbd2_journal_clear_features(sbi->s_journal,
3183                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3184                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3185                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3186         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3187                 ret = jbd2_journal_set_features(sbi->s_journal,
3188                                 compat, 0,
3189                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3190                                 incompat);
3191         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3192                 ret = jbd2_journal_set_features(sbi->s_journal,
3193                                 compat, 0,
3194                                 incompat);
3195                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3196                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3197         } else {
3198                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3199                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3200         }
3201
3202         return ret;
3203 }
3204
3205 /*
3206  * Note: calculating the overhead so we can be compatible with
3207  * historical BSD practice is quite difficult in the face of
3208  * clusters/bigalloc.  This is because multiple metadata blocks from
3209  * different block group can end up in the same allocation cluster.
3210  * Calculating the exact overhead in the face of clustered allocation
3211  * requires either O(all block bitmaps) in memory or O(number of block
3212  * groups**2) in time.  We will still calculate the superblock for
3213  * older file systems --- and if we come across with a bigalloc file
3214  * system with zero in s_overhead_clusters the estimate will be close to
3215  * correct especially for very large cluster sizes --- but for newer
3216  * file systems, it's better to calculate this figure once at mkfs
3217  * time, and store it in the superblock.  If the superblock value is
3218  * present (even for non-bigalloc file systems), we will use it.
3219  */
3220 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3221                           char *buf)
3222 {
3223         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3224         struct ext4_group_desc  *gdp;
3225         ext4_fsblk_t            first_block, last_block, b;
3226         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3227         int                     s, j, count = 0;
3228
3229         if (!ext4_has_feature_bigalloc(sb))
3230                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3231                         sbi->s_itb_per_group + 2);
3232
3233         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3234                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3235         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3236         for (i = 0; i < ngroups; i++) {
3237                 gdp = ext4_get_group_desc(sb, i, NULL);
3238                 b = ext4_block_bitmap(sb, gdp);
3239                 if (b >= first_block && b <= last_block) {
3240                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3241                         count++;
3242                 }
3243                 b = ext4_inode_bitmap(sb, gdp);
3244                 if (b >= first_block && b <= last_block) {
3245                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3246                         count++;
3247                 }
3248                 b = ext4_inode_table(sb, gdp);
3249                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3250                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3251                                 int c = EXT4_B2C(sbi, b - first_block);
3252                                 ext4_set_bit(c, buf);
3253                                 count++;
3254                         }
3255                 if (i != grp)
3256                         continue;
3257                 s = 0;
3258                 if (ext4_bg_has_super(sb, grp)) {
3259                         ext4_set_bit(s++, buf);
3260                         count++;
3261                 }
3262                 j = ext4_bg_num_gdb(sb, grp);
3263                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3264                         ext4_error(sb, "Invalid number of block group "
3265                                    "descriptor blocks: %d", j);
3266                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3267                 }
3268                 count += j;
3269                 for (; j > 0; j--)
3270                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3271         }
3272         if (!count)
3273                 return 0;
3274         return EXT4_CLUSTERS_PER_GROUP(sb) -
3275                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3276 }
3277
3278 /*
3279  * Compute the overhead and stash it in sbi->s_overhead
3280  */
3281 int ext4_calculate_overhead(struct super_block *sb)
3282 {
3283         struct ext4_sb_info *sbi = EXT4_SB(sb);
3284         struct ext4_super_block *es = sbi->s_es;
3285         struct inode *j_inode;
3286         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3287         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3288         ext4_fsblk_t overhead = 0;
3289         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3290
3291         if (!buf)
3292                 return -ENOMEM;
3293
3294         /*
3295          * Compute the overhead (FS structures).  This is constant
3296          * for a given filesystem unless the number of block groups
3297          * changes so we cache the previous value until it does.
3298          */
3299
3300         /*
3301          * All of the blocks before first_data_block are overhead
3302          */
3303         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3304
3305         /*
3306          * Add the overhead found in each block group
3307          */
3308         for (i = 0; i < ngroups; i++) {
3309                 int blks;
3310
3311                 blks = count_overhead(sb, i, buf);
3312                 overhead += blks;
3313                 if (blks)
3314                         memset(buf, 0, PAGE_SIZE);
3315                 cond_resched();
3316         }
3317
3318         /*
3319          * Add the internal journal blocks whether the journal has been
3320          * loaded or not
3321          */
3322         if (sbi->s_journal && !sbi->journal_bdev)
3323                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3324         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3325                 j_inode = ext4_get_journal_inode(sb, j_inum);
3326                 if (j_inode) {
3327                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3328                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3329                         iput(j_inode);
3330                 } else {
3331                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3332                 }
3333         }
3334         sbi->s_overhead = overhead;
3335         smp_wmb();
3336         free_page((unsigned long) buf);
3337         return 0;
3338 }
3339
3340 static void ext4_set_resv_clusters(struct super_block *sb)
3341 {
3342         ext4_fsblk_t resv_clusters;
3343         struct ext4_sb_info *sbi = EXT4_SB(sb);
3344
3345         /*
3346          * There's no need to reserve anything when we aren't using extents.
3347          * The space estimates are exact, there are no unwritten extents,
3348          * hole punching doesn't need new metadata... This is needed especially
3349          * to keep ext2/3 backward compatibility.
3350          */
3351         if (!ext4_has_feature_extents(sb))
3352                 return;
3353         /*
3354          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3355          * This should cover the situations where we can not afford to run
3356          * out of space like for example punch hole, or converting
3357          * unwritten extents in delalloc path. In most cases such
3358          * allocation would require 1, or 2 blocks, higher numbers are
3359          * very rare.
3360          */
3361         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3362                          sbi->s_cluster_bits);
3363
3364         do_div(resv_clusters, 50);
3365         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3366
3367         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3368 }
3369
3370 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3371 {
3372         char *orig_data = kstrdup(data, GFP_KERNEL);
3373         struct buffer_head *bh;
3374         struct ext4_super_block *es = NULL;
3375         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3376         ext4_fsblk_t block;
3377         ext4_fsblk_t sb_block = get_sb_block(&data);
3378         ext4_fsblk_t logical_sb_block;
3379         unsigned long offset = 0;
3380         unsigned long journal_devnum = 0;
3381         unsigned long def_mount_opts;
3382         struct inode *root;
3383         const char *descr;
3384         int ret = -ENOMEM;
3385         int blocksize, clustersize;
3386         unsigned int db_count;
3387         unsigned int i;
3388         int needs_recovery, has_huge_files, has_bigalloc;
3389         __u64 blocks_count;
3390         int err = 0;
3391         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3392         ext4_group_t first_not_zeroed;
3393
3394         if ((data && !orig_data) || !sbi)
3395                 goto out_free_base;
3396
3397         sbi->s_blockgroup_lock =
3398                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3399         if (!sbi->s_blockgroup_lock)
3400                 goto out_free_base;
3401
3402         sb->s_fs_info = sbi;
3403         sbi->s_sb = sb;
3404         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3405         sbi->s_sb_block = sb_block;
3406         if (sb->s_bdev->bd_part)
3407                 sbi->s_sectors_written_start =
3408                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3409
3410         /* Cleanup superblock name */
3411         strreplace(sb->s_id, '/', '!');
3412
3413         /* -EINVAL is default */
3414         ret = -EINVAL;
3415         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3416         if (!blocksize) {
3417                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3418                 goto out_fail;
3419         }
3420
3421         /*
3422          * The ext4 superblock will not be buffer aligned for other than 1kB
3423          * block sizes.  We need to calculate the offset from buffer start.
3424          */
3425         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3426                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3427                 offset = do_div(logical_sb_block, blocksize);
3428         } else {
3429                 logical_sb_block = sb_block;
3430         }
3431
3432         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3433                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3434                 goto out_fail;
3435         }
3436         /*
3437          * Note: s_es must be initialized as soon as possible because
3438          *       some ext4 macro-instructions depend on its value
3439          */
3440         es = (struct ext4_super_block *) (bh->b_data + offset);
3441         sbi->s_es = es;
3442         sb->s_magic = le16_to_cpu(es->s_magic);
3443         if (sb->s_magic != EXT4_SUPER_MAGIC)
3444                 goto cantfind_ext4;
3445         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3446
3447         /* Warn if metadata_csum and gdt_csum are both set. */
3448         if (ext4_has_feature_metadata_csum(sb) &&
3449             ext4_has_feature_gdt_csum(sb))
3450                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3451                              "redundant flags; please run fsck.");
3452
3453         /* Check for a known checksum algorithm */
3454         if (!ext4_verify_csum_type(sb, es)) {
3455                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3456                          "unknown checksum algorithm.");
3457                 silent = 1;
3458                 goto cantfind_ext4;
3459         }
3460
3461         /* Load the checksum driver */
3462         if (ext4_has_feature_metadata_csum(sb) ||
3463             ext4_has_feature_ea_inode(sb)) {
3464                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3465                 if (IS_ERR(sbi->s_chksum_driver)) {
3466                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3467                         ret = PTR_ERR(sbi->s_chksum_driver);
3468                         sbi->s_chksum_driver = NULL;
3469                         goto failed_mount;
3470                 }
3471         }
3472
3473         /* Check superblock checksum */
3474         if (!ext4_superblock_csum_verify(sb, es)) {
3475                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3476                          "invalid superblock checksum.  Run e2fsck?");
3477                 silent = 1;
3478                 ret = -EFSBADCRC;
3479                 goto cantfind_ext4;
3480         }
3481
3482         /* Precompute checksum seed for all metadata */
3483         if (ext4_has_feature_csum_seed(sb))
3484                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3485         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3486                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3487                                                sizeof(es->s_uuid));
3488
3489         /* Set defaults before we parse the mount options */
3490         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3491         set_opt(sb, INIT_INODE_TABLE);
3492         if (def_mount_opts & EXT4_DEFM_DEBUG)
3493                 set_opt(sb, DEBUG);
3494         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3495                 set_opt(sb, GRPID);
3496         if (def_mount_opts & EXT4_DEFM_UID16)
3497                 set_opt(sb, NO_UID32);
3498         /* xattr user namespace & acls are now defaulted on */
3499         set_opt(sb, XATTR_USER);
3500 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3501         set_opt(sb, POSIX_ACL);
3502 #endif
3503         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3504         if (ext4_has_metadata_csum(sb))
3505                 set_opt(sb, JOURNAL_CHECKSUM);
3506
3507         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3508                 set_opt(sb, JOURNAL_DATA);
3509         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3510                 set_opt(sb, ORDERED_DATA);
3511         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3512                 set_opt(sb, WRITEBACK_DATA);
3513
3514         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3515                 set_opt(sb, ERRORS_PANIC);
3516         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3517                 set_opt(sb, ERRORS_CONT);
3518         else
3519                 set_opt(sb, ERRORS_RO);
3520         /* block_validity enabled by default; disable with noblock_validity */
3521         set_opt(sb, BLOCK_VALIDITY);
3522         if (def_mount_opts & EXT4_DEFM_DISCARD)
3523                 set_opt(sb, DISCARD);
3524
3525         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3526         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3527         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3528         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3529         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3530
3531         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3532                 set_opt(sb, BARRIER);
3533
3534         /*
3535          * enable delayed allocation by default
3536          * Use -o nodelalloc to turn it off
3537          */
3538         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3539             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3540                 set_opt(sb, DELALLOC);
3541
3542         /*
3543          * set default s_li_wait_mult for lazyinit, for the case there is
3544          * no mount option specified.
3545          */
3546         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3547
3548         if (sbi->s_es->s_mount_opts[0]) {
3549                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3550                                               sizeof(sbi->s_es->s_mount_opts),
3551                                               GFP_KERNEL);
3552                 if (!s_mount_opts)
3553                         goto failed_mount;
3554                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3555                                    &journal_ioprio, 0)) {
3556                         ext4_msg(sb, KERN_WARNING,
3557                                  "failed to parse options in superblock: %s",
3558                                  s_mount_opts);
3559                 }
3560                 kfree(s_mount_opts);
3561         }
3562         sbi->s_def_mount_opt = sbi->s_mount_opt;
3563         if (!parse_options((char *) data, sb, &journal_devnum,
3564                            &journal_ioprio, 0))
3565                 goto failed_mount;
3566
3567         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3568                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3569                             "with data=journal disables delayed "
3570                             "allocation and O_DIRECT support!\n");
3571                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3572                         ext4_msg(sb, KERN_ERR, "can't mount with "
3573                                  "both data=journal and delalloc");
3574                         goto failed_mount;
3575                 }
3576                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3577                         ext4_msg(sb, KERN_ERR, "can't mount with "
3578                                  "both data=journal and dioread_nolock");
3579                         goto failed_mount;
3580                 }
3581                 if (test_opt(sb, DAX)) {
3582                         ext4_msg(sb, KERN_ERR, "can't mount with "
3583                                  "both data=journal and dax");
3584                         goto failed_mount;
3585                 }
3586                 if (ext4_has_feature_encrypt(sb)) {
3587                         ext4_msg(sb, KERN_WARNING,
3588                                  "encrypted files will use data=ordered "
3589                                  "instead of data journaling mode");
3590                 }
3591                 if (test_opt(sb, DELALLOC))
3592                         clear_opt(sb, DELALLOC);
3593         } else {
3594                 sb->s_iflags |= SB_I_CGROUPWB;
3595         }
3596
3597         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3598                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3599
3600         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3601             (ext4_has_compat_features(sb) ||
3602              ext4_has_ro_compat_features(sb) ||
3603              ext4_has_incompat_features(sb)))
3604                 ext4_msg(sb, KERN_WARNING,
3605                        "feature flags set on rev 0 fs, "
3606                        "running e2fsck is recommended");
3607
3608         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3609                 set_opt2(sb, HURD_COMPAT);
3610                 if (ext4_has_feature_64bit(sb)) {
3611                         ext4_msg(sb, KERN_ERR,
3612                                  "The Hurd can't support 64-bit file systems");
3613                         goto failed_mount;
3614                 }
3615
3616                 /*
3617                  * ea_inode feature uses l_i_version field which is not
3618                  * available in HURD_COMPAT mode.
3619                  */
3620                 if (ext4_has_feature_ea_inode(sb)) {
3621                         ext4_msg(sb, KERN_ERR,
3622                                  "ea_inode feature is not supported for Hurd");
3623                         goto failed_mount;
3624                 }
3625         }
3626
3627         if (IS_EXT2_SB(sb)) {
3628                 if (ext2_feature_set_ok(sb))
3629                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3630                                  "using the ext4 subsystem");
3631                 else {
3632                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3633                                  "to feature incompatibilities");
3634                         goto failed_mount;
3635                 }
3636         }
3637
3638         if (IS_EXT3_SB(sb)) {
3639                 if (ext3_feature_set_ok(sb))
3640                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3641                                  "using the ext4 subsystem");
3642                 else {
3643                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3644                                  "to feature incompatibilities");
3645                         goto failed_mount;
3646                 }
3647         }
3648
3649         /*
3650          * Check feature flags regardless of the revision level, since we
3651          * previously didn't change the revision level when setting the flags,
3652          * so there is a chance incompat flags are set on a rev 0 filesystem.
3653          */
3654         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3655                 goto failed_mount;
3656
3657         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3658         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3659             blocksize > EXT4_MAX_BLOCK_SIZE) {
3660                 ext4_msg(sb, KERN_ERR,
3661                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3662                          blocksize, le32_to_cpu(es->s_log_block_size));
3663                 goto failed_mount;
3664         }
3665         if (le32_to_cpu(es->s_log_block_size) >
3666             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3667                 ext4_msg(sb, KERN_ERR,
3668                          "Invalid log block size: %u",
3669                          le32_to_cpu(es->s_log_block_size));
3670                 goto failed_mount;
3671         }
3672
3673         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3674                 ext4_msg(sb, KERN_ERR,
3675                          "Number of reserved GDT blocks insanely large: %d",
3676                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3677                 goto failed_mount;
3678         }
3679
3680         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3681                 err = bdev_dax_supported(sb, blocksize);
3682                 if (err)
3683                         goto failed_mount;
3684         }
3685
3686         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3687                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3688                          es->s_encryption_level);
3689                 goto failed_mount;
3690         }
3691
3692         if (sb->s_blocksize != blocksize) {
3693                 /* Validate the filesystem blocksize */
3694                 if (!sb_set_blocksize(sb, blocksize)) {
3695                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3696                                         blocksize);
3697                         goto failed_mount;
3698                 }
3699
3700                 brelse(bh);
3701                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3702                 offset = do_div(logical_sb_block, blocksize);
3703                 bh = sb_bread_unmovable(sb, logical_sb_block);
3704                 if (!bh) {
3705                         ext4_msg(sb, KERN_ERR,
3706                                "Can't read superblock on 2nd try");
3707                         goto failed_mount;
3708                 }
3709                 es = (struct ext4_super_block *)(bh->b_data + offset);
3710                 sbi->s_es = es;
3711                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                "Magic mismatch, very weird!");
3714                         goto failed_mount;
3715                 }
3716         }
3717
3718         has_huge_files = ext4_has_feature_huge_file(sb);
3719         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3720                                                       has_huge_files);
3721         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3722
3723         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3724                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3725                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3726         } else {
3727                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3728                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3729                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3730                     (!is_power_of_2(sbi->s_inode_size)) ||
3731                     (sbi->s_inode_size > blocksize)) {
3732                         ext4_msg(sb, KERN_ERR,
3733                                "unsupported inode size: %d",
3734                                sbi->s_inode_size);
3735                         goto failed_mount;
3736                 }
3737                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3738                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3739         }
3740
3741         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3742         if (ext4_has_feature_64bit(sb)) {
3743                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3744                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3745                     !is_power_of_2(sbi->s_desc_size)) {
3746                         ext4_msg(sb, KERN_ERR,
3747                                "unsupported descriptor size %lu",
3748                                sbi->s_desc_size);
3749                         goto failed_mount;
3750                 }
3751         } else
3752                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3753
3754         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3755         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3756
3757         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3758         if (sbi->s_inodes_per_block == 0)
3759                 goto cantfind_ext4;
3760         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3761             sbi->s_inodes_per_group > blocksize * 8) {
3762                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3763                          sbi->s_blocks_per_group);
3764                 goto failed_mount;
3765         }
3766         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3767                                         sbi->s_inodes_per_block;
3768         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3769         sbi->s_sbh = bh;
3770         sbi->s_mount_state = le16_to_cpu(es->s_state);
3771         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3772         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3773
3774         for (i = 0; i < 4; i++)
3775                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3776         sbi->s_def_hash_version = es->s_def_hash_version;
3777         if (ext4_has_feature_dir_index(sb)) {
3778                 i = le32_to_cpu(es->s_flags);
3779                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3780                         sbi->s_hash_unsigned = 3;
3781                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3782 #ifdef __CHAR_UNSIGNED__
3783                         if (!(sb->s_flags & MS_RDONLY))
3784                                 es->s_flags |=
3785                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3786                         sbi->s_hash_unsigned = 3;
3787 #else
3788                         if (!(sb->s_flags & MS_RDONLY))
3789                                 es->s_flags |=
3790                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3791 #endif
3792                 }
3793         }
3794
3795         /* Handle clustersize */
3796         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3797         has_bigalloc = ext4_has_feature_bigalloc(sb);
3798         if (has_bigalloc) {
3799                 if (clustersize < blocksize) {
3800                         ext4_msg(sb, KERN_ERR,
3801                                  "cluster size (%d) smaller than "
3802                                  "block size (%d)", clustersize, blocksize);
3803                         goto failed_mount;
3804                 }
3805                 if (le32_to_cpu(es->s_log_cluster_size) >
3806                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3807                         ext4_msg(sb, KERN_ERR,
3808                                  "Invalid log cluster size: %u",
3809                                  le32_to_cpu(es->s_log_cluster_size));
3810                         goto failed_mount;
3811                 }
3812                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3813                         le32_to_cpu(es->s_log_block_size);
3814                 sbi->s_clusters_per_group =
3815                         le32_to_cpu(es->s_clusters_per_group);
3816                 if (sbi->s_clusters_per_group > blocksize * 8) {
3817                         ext4_msg(sb, KERN_ERR,
3818                                  "#clusters per group too big: %lu",
3819                                  sbi->s_clusters_per_group);
3820                         goto failed_mount;
3821                 }
3822                 if (sbi->s_blocks_per_group !=
3823                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3824                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3825                                  "clusters per group (%lu) inconsistent",
3826                                  sbi->s_blocks_per_group,
3827                                  sbi->s_clusters_per_group);
3828                         goto failed_mount;
3829                 }
3830         } else {
3831                 if (clustersize != blocksize) {
3832                         ext4_warning(sb, "fragment/cluster size (%d) != "
3833                                      "block size (%d)", clustersize,
3834                                      blocksize);
3835                         clustersize = blocksize;
3836                 }
3837                 if (sbi->s_blocks_per_group > blocksize * 8) {
3838                         ext4_msg(sb, KERN_ERR,
3839                                  "#blocks per group too big: %lu",
3840                                  sbi->s_blocks_per_group);
3841                         goto failed_mount;
3842                 }
3843                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3844                 sbi->s_cluster_bits = 0;
3845         }
3846         sbi->s_cluster_ratio = clustersize / blocksize;
3847
3848         /* Do we have standard group size of clustersize * 8 blocks ? */
3849         if (sbi->s_blocks_per_group == clustersize << 3)
3850                 set_opt2(sb, STD_GROUP_SIZE);
3851
3852         /*
3853          * Test whether we have more sectors than will fit in sector_t,
3854          * and whether the max offset is addressable by the page cache.
3855          */
3856         err = generic_check_addressable(sb->s_blocksize_bits,
3857                                         ext4_blocks_count(es));
3858         if (err) {
3859                 ext4_msg(sb, KERN_ERR, "filesystem"
3860                          " too large to mount safely on this system");
3861                 if (sizeof(sector_t) < 8)
3862                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3863                 goto failed_mount;
3864         }
3865
3866         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3867                 goto cantfind_ext4;
3868
3869         /* check blocks count against device size */
3870         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3871         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3872                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3873                        "exceeds size of device (%llu blocks)",
3874                        ext4_blocks_count(es), blocks_count);
3875                 goto failed_mount;
3876         }
3877
3878         /*
3879          * It makes no sense for the first data block to be beyond the end
3880          * of the filesystem.
3881          */
3882         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3883                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3884                          "block %u is beyond end of filesystem (%llu)",
3885                          le32_to_cpu(es->s_first_data_block),
3886                          ext4_blocks_count(es));
3887                 goto failed_mount;
3888         }
3889         blocks_count = (ext4_blocks_count(es) -
3890                         le32_to_cpu(es->s_first_data_block) +
3891                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3892         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3893         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3894                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3895                        "(block count %llu, first data block %u, "
3896                        "blocks per group %lu)", sbi->s_groups_count,
3897                        ext4_blocks_count(es),
3898                        le32_to_cpu(es->s_first_data_block),
3899                        EXT4_BLOCKS_PER_GROUP(sb));
3900                 goto failed_mount;
3901         }
3902         sbi->s_groups_count = blocks_count;
3903         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3904                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3905         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3906                    EXT4_DESC_PER_BLOCK(sb);
3907         if (ext4_has_feature_meta_bg(sb)) {
3908                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3909                         ext4_msg(sb, KERN_WARNING,
3910                                  "first meta block group too large: %u "
3911                                  "(group descriptor block count %u)",
3912                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3913                         goto failed_mount;
3914                 }
3915         }
3916         sbi->s_group_desc = kvmalloc(db_count *
3917                                           sizeof(struct buffer_head *),
3918                                           GFP_KERNEL);
3919         if (sbi->s_group_desc == NULL) {
3920                 ext4_msg(sb, KERN_ERR, "not enough memory");
3921                 ret = -ENOMEM;
3922                 goto failed_mount;
3923         }
3924
3925         bgl_lock_init(sbi->s_blockgroup_lock);
3926
3927         /* Pre-read the descriptors into the buffer cache */
3928         for (i = 0; i < db_count; i++) {
3929                 block = descriptor_loc(sb, logical_sb_block, i);
3930                 sb_breadahead(sb, block);
3931         }
3932
3933         for (i = 0; i < db_count; i++) {
3934                 block = descriptor_loc(sb, logical_sb_block, i);
3935                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3936                 if (!sbi->s_group_desc[i]) {
3937                         ext4_msg(sb, KERN_ERR,
3938                                "can't read group descriptor %d", i);
3939                         db_count = i;
3940                         goto failed_mount2;
3941                 }
3942         }
3943         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3944                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3945                 ret = -EFSCORRUPTED;
3946                 goto failed_mount2;
3947         }
3948
3949         sbi->s_gdb_count = db_count;
3950         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3951         spin_lock_init(&sbi->s_next_gen_lock);
3952
3953         setup_timer(&sbi->s_err_report, print_daily_error_info,
3954                 (unsigned long) sb);
3955
3956         /* Register extent status tree shrinker */
3957         if (ext4_es_register_shrinker(sbi))
3958                 goto failed_mount3;
3959
3960         sbi->s_stripe = ext4_get_stripe_size(sbi);
3961         sbi->s_extent_max_zeroout_kb = 32;
3962
3963         /*
3964          * set up enough so that it can read an inode
3965          */
3966         sb->s_op = &ext4_sops;
3967         sb->s_export_op = &ext4_export_ops;
3968         sb->s_xattr = ext4_xattr_handlers;
3969         sb->s_cop = &ext4_cryptops;
3970 #ifdef CONFIG_QUOTA
3971         sb->dq_op = &ext4_quota_operations;
3972         if (ext4_has_feature_quota(sb))
3973                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3974         else
3975                 sb->s_qcop = &ext4_qctl_operations;
3976         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3977 #endif
3978         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3979
3980         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3981         mutex_init(&sbi->s_orphan_lock);
3982
3983         sb->s_root = NULL;
3984
3985         needs_recovery = (es->s_last_orphan != 0 ||
3986                           ext4_has_feature_journal_needs_recovery(sb));
3987
3988         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3989                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3990                         goto failed_mount3a;
3991
3992         /*
3993          * The first inode we look at is the journal inode.  Don't try
3994          * root first: it may be modified in the journal!
3995          */
3996         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3997                 err = ext4_load_journal(sb, es, journal_devnum);
3998                 if (err)
3999                         goto failed_mount3a;
4000         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
4001                    ext4_has_feature_journal_needs_recovery(sb)) {
4002                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4003                        "suppressed and not mounted read-only");
4004                 goto failed_mount_wq;
4005         } else {
4006                 /* Nojournal mode, all journal mount options are illegal */
4007                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4008                         ext4_msg(sb, KERN_ERR, "can't mount with "
4009                                  "journal_checksum, fs mounted w/o journal");
4010                         goto failed_mount_wq;
4011                 }
4012                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4013                         ext4_msg(sb, KERN_ERR, "can't mount with "
4014                                  "journal_async_commit, fs mounted w/o journal");
4015                         goto failed_mount_wq;
4016                 }
4017                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4018                         ext4_msg(sb, KERN_ERR, "can't mount with "
4019                                  "commit=%lu, fs mounted w/o journal",
4020                                  sbi->s_commit_interval / HZ);
4021                         goto failed_mount_wq;
4022                 }
4023                 if (EXT4_MOUNT_DATA_FLAGS &
4024                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4025                         ext4_msg(sb, KERN_ERR, "can't mount with "
4026                                  "data=, fs mounted w/o journal");
4027                         goto failed_mount_wq;
4028                 }
4029                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4030                 clear_opt(sb, JOURNAL_CHECKSUM);
4031                 clear_opt(sb, DATA_FLAGS);
4032                 sbi->s_journal = NULL;
4033                 needs_recovery = 0;
4034                 goto no_journal;
4035         }
4036
4037         if (ext4_has_feature_64bit(sb) &&
4038             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4039                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4040                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4041                 goto failed_mount_wq;
4042         }
4043
4044         if (!set_journal_csum_feature_set(sb)) {
4045                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4046                          "feature set");
4047                 goto failed_mount_wq;
4048         }
4049
4050         /* We have now updated the journal if required, so we can
4051          * validate the data journaling mode. */
4052         switch (test_opt(sb, DATA_FLAGS)) {
4053         case 0:
4054                 /* No mode set, assume a default based on the journal
4055                  * capabilities: ORDERED_DATA if the journal can
4056                  * cope, else JOURNAL_DATA
4057                  */
4058                 if (jbd2_journal_check_available_features
4059                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4060                         set_opt(sb, ORDERED_DATA);
4061                 else
4062                         set_opt(sb, JOURNAL_DATA);
4063                 break;
4064
4065         case EXT4_MOUNT_ORDERED_DATA:
4066         case EXT4_MOUNT_WRITEBACK_DATA:
4067                 if (!jbd2_journal_check_available_features
4068                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4069                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4070                                "requested data journaling mode");
4071                         goto failed_mount_wq;
4072                 }
4073         default:
4074                 break;
4075         }
4076
4077         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4078             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4079                 ext4_msg(sb, KERN_ERR, "can't mount with "
4080                         "journal_async_commit in data=ordered mode");
4081                 goto failed_mount_wq;
4082         }
4083
4084         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4085
4086         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4087
4088 no_journal:
4089         if (!test_opt(sb, NO_MBCACHE)) {
4090                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4091                 if (!sbi->s_ea_block_cache) {
4092                         ext4_msg(sb, KERN_ERR,
4093                                  "Failed to create ea_block_cache");
4094                         goto failed_mount_wq;
4095                 }
4096
4097                 if (ext4_has_feature_ea_inode(sb)) {
4098                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4099                         if (!sbi->s_ea_inode_cache) {
4100                                 ext4_msg(sb, KERN_ERR,
4101                                          "Failed to create ea_inode_cache");
4102                                 goto failed_mount_wq;
4103                         }
4104                 }
4105         }
4106
4107         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4108             (blocksize != PAGE_SIZE)) {
4109                 ext4_msg(sb, KERN_ERR,
4110                          "Unsupported blocksize for fs encryption");
4111                 goto failed_mount_wq;
4112         }
4113
4114         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4115             !ext4_has_feature_encrypt(sb)) {
4116                 ext4_set_feature_encrypt(sb);
4117                 ext4_commit_super(sb, 1);
4118         }
4119
4120         /*
4121          * Get the # of file system overhead blocks from the
4122          * superblock if present.
4123          */
4124         if (es->s_overhead_clusters)
4125                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4126         else {
4127                 err = ext4_calculate_overhead(sb);
4128                 if (err)
4129                         goto failed_mount_wq;
4130         }
4131
4132         /*
4133          * The maximum number of concurrent works can be high and
4134          * concurrency isn't really necessary.  Limit it to 1.
4135          */
4136         EXT4_SB(sb)->rsv_conversion_wq =
4137                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4138         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4139                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4140                 ret = -ENOMEM;
4141                 goto failed_mount4;
4142         }
4143
4144         /*
4145          * The jbd2_journal_load will have done any necessary log recovery,
4146          * so we can safely mount the rest of the filesystem now.
4147          */
4148
4149         root = ext4_iget(sb, EXT4_ROOT_INO);
4150         if (IS_ERR(root)) {
4151                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4152                 ret = PTR_ERR(root);
4153                 root = NULL;
4154                 goto failed_mount4;
4155         }
4156         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4157                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4158                 iput(root);
4159                 goto failed_mount4;
4160         }
4161         sb->s_root = d_make_root(root);
4162         if (!sb->s_root) {
4163                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4164                 ret = -ENOMEM;
4165                 goto failed_mount4;
4166         }
4167
4168         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4169                 sb->s_flags |= MS_RDONLY;
4170
4171         /* determine the minimum size of new large inodes, if present */
4172         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4173             sbi->s_want_extra_isize == 0) {
4174                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4175                                                      EXT4_GOOD_OLD_INODE_SIZE;
4176                 if (ext4_has_feature_extra_isize(sb)) {
4177                         if (sbi->s_want_extra_isize <
4178                             le16_to_cpu(es->s_want_extra_isize))
4179                                 sbi->s_want_extra_isize =
4180                                         le16_to_cpu(es->s_want_extra_isize);
4181                         if (sbi->s_want_extra_isize <
4182                             le16_to_cpu(es->s_min_extra_isize))
4183                                 sbi->s_want_extra_isize =
4184                                         le16_to_cpu(es->s_min_extra_isize);
4185                 }
4186         }
4187         /* Check if enough inode space is available */
4188         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4189                                                         sbi->s_inode_size) {
4190                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4191                                                        EXT4_GOOD_OLD_INODE_SIZE;
4192                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4193                          "available");
4194         }
4195
4196         ext4_set_resv_clusters(sb);
4197
4198         err = ext4_setup_system_zone(sb);
4199         if (err) {
4200                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4201                          "zone (%d)", err);
4202                 goto failed_mount4a;
4203         }
4204
4205         ext4_ext_init(sb);
4206         err = ext4_mb_init(sb);
4207         if (err) {
4208                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4209                          err);
4210                 goto failed_mount5;
4211         }
4212
4213         block = ext4_count_free_clusters(sb);
4214         ext4_free_blocks_count_set(sbi->s_es,
4215                                    EXT4_C2B(sbi, block));
4216         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4217                                   GFP_KERNEL);
4218         if (!err) {
4219                 unsigned long freei = ext4_count_free_inodes(sb);
4220                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4221                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4222                                           GFP_KERNEL);
4223         }
4224         if (!err)
4225                 err = percpu_counter_init(&sbi->s_dirs_counter,
4226                                           ext4_count_dirs(sb), GFP_KERNEL);
4227         if (!err)
4228                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4229                                           GFP_KERNEL);
4230         if (!err)
4231                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4232
4233         if (err) {
4234                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4235                 goto failed_mount6;
4236         }
4237
4238         if (ext4_has_feature_flex_bg(sb))
4239                 if (!ext4_fill_flex_info(sb)) {
4240                         ext4_msg(sb, KERN_ERR,
4241                                "unable to initialize "
4242                                "flex_bg meta info!");
4243                         goto failed_mount6;
4244                 }
4245
4246         err = ext4_register_li_request(sb, first_not_zeroed);
4247         if (err)
4248                 goto failed_mount6;
4249
4250         err = ext4_register_sysfs(sb);
4251         if (err)
4252                 goto failed_mount7;
4253
4254 #ifdef CONFIG_QUOTA
4255         /* Enable quota usage during mount. */
4256         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4257                 err = ext4_enable_quotas(sb);
4258                 if (err)
4259                         goto failed_mount8;
4260         }
4261 #endif  /* CONFIG_QUOTA */
4262
4263         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4264         ext4_orphan_cleanup(sb, es);
4265         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4266         if (needs_recovery) {
4267                 ext4_msg(sb, KERN_INFO, "recovery complete");
4268                 ext4_mark_recovery_complete(sb, es);
4269         }
4270         if (EXT4_SB(sb)->s_journal) {
4271                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4272                         descr = " journalled data mode";
4273                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4274                         descr = " ordered data mode";
4275                 else
4276                         descr = " writeback data mode";
4277         } else
4278                 descr = "out journal";
4279
4280         if (test_opt(sb, DISCARD)) {
4281                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4282                 if (!blk_queue_discard(q))
4283                         ext4_msg(sb, KERN_WARNING,
4284                                  "mounting with \"discard\" option, but "
4285                                  "the device does not support discard");
4286         }
4287
4288         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4289                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4290                          "Opts: %.*s%s%s", descr,
4291                          (int) sizeof(sbi->s_es->s_mount_opts),
4292                          sbi->s_es->s_mount_opts,
4293                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4294
4295         if (es->s_error_count)
4296                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4297
4298         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4299         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4300         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4301         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4302
4303         kfree(orig_data);
4304         return 0;
4305
4306 cantfind_ext4:
4307         if (!silent)
4308                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4309         goto failed_mount;
4310
4311 #ifdef CONFIG_QUOTA
4312 failed_mount8:
4313         ext4_unregister_sysfs(sb);
4314 #endif
4315 failed_mount7:
4316         ext4_unregister_li_request(sb);
4317 failed_mount6:
4318         ext4_mb_release(sb);
4319         if (sbi->s_flex_groups)
4320                 kvfree(sbi->s_flex_groups);
4321         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4322         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4323         percpu_counter_destroy(&sbi->s_dirs_counter);
4324         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4325 failed_mount5:
4326         ext4_ext_release(sb);
4327         ext4_release_system_zone(sb);
4328 failed_mount4a:
4329         dput(sb->s_root);
4330         sb->s_root = NULL;
4331 failed_mount4:
4332         ext4_msg(sb, KERN_ERR, "mount failed");
4333         if (EXT4_SB(sb)->rsv_conversion_wq)
4334                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4335 failed_mount_wq:
4336         if (sbi->s_ea_inode_cache) {
4337                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4338                 sbi->s_ea_inode_cache = NULL;
4339         }
4340         if (sbi->s_ea_block_cache) {
4341                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4342                 sbi->s_ea_block_cache = NULL;
4343         }
4344         if (sbi->s_journal) {
4345                 jbd2_journal_destroy(sbi->s_journal);
4346                 sbi->s_journal = NULL;
4347         }
4348 failed_mount3a:
4349         ext4_es_unregister_shrinker(sbi);
4350 failed_mount3:
4351         del_timer_sync(&sbi->s_err_report);
4352         if (sbi->s_mmp_tsk)
4353                 kthread_stop(sbi->s_mmp_tsk);
4354 failed_mount2:
4355         for (i = 0; i < db_count; i++)
4356                 brelse(sbi->s_group_desc[i]);
4357         kvfree(sbi->s_group_desc);
4358 failed_mount:
4359         if (sbi->s_chksum_driver)
4360                 crypto_free_shash(sbi->s_chksum_driver);
4361 #ifdef CONFIG_QUOTA
4362         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4363                 kfree(sbi->s_qf_names[i]);
4364 #endif
4365         ext4_blkdev_remove(sbi);
4366         brelse(bh);
4367 out_fail:
4368         sb->s_fs_info = NULL;
4369         kfree(sbi->s_blockgroup_lock);
4370 out_free_base:
4371         kfree(sbi);
4372         kfree(orig_data);
4373         return err ? err : ret;
4374 }
4375
4376 /*
4377  * Setup any per-fs journal parameters now.  We'll do this both on
4378  * initial mount, once the journal has been initialised but before we've
4379  * done any recovery; and again on any subsequent remount.
4380  */
4381 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4382 {
4383         struct ext4_sb_info *sbi = EXT4_SB(sb);
4384
4385         journal->j_commit_interval = sbi->s_commit_interval;
4386         journal->j_min_batch_time = sbi->s_min_batch_time;
4387         journal->j_max_batch_time = sbi->s_max_batch_time;
4388
4389         write_lock(&journal->j_state_lock);
4390         if (test_opt(sb, BARRIER))
4391                 journal->j_flags |= JBD2_BARRIER;
4392         else
4393                 journal->j_flags &= ~JBD2_BARRIER;
4394         if (test_opt(sb, DATA_ERR_ABORT))
4395                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4396         else
4397                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4398         write_unlock(&journal->j_state_lock);
4399 }
4400
4401 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4402                                              unsigned int journal_inum)
4403 {
4404         struct inode *journal_inode;
4405
4406         /*
4407          * Test for the existence of a valid inode on disk.  Bad things
4408          * happen if we iget() an unused inode, as the subsequent iput()
4409          * will try to delete it.
4410          */
4411         journal_inode = ext4_iget(sb, journal_inum);
4412         if (IS_ERR(journal_inode)) {
4413                 ext4_msg(sb, KERN_ERR, "no journal found");
4414                 return NULL;
4415         }
4416         if (!journal_inode->i_nlink) {
4417                 make_bad_inode(journal_inode);
4418                 iput(journal_inode);
4419                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4420                 return NULL;
4421         }
4422
4423         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4424                   journal_inode, journal_inode->i_size);
4425         if (!S_ISREG(journal_inode->i_mode)) {
4426                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4427                 iput(journal_inode);
4428                 return NULL;
4429         }
4430         return journal_inode;
4431 }
4432
4433 static journal_t *ext4_get_journal(struct super_block *sb,
4434                                    unsigned int journal_inum)
4435 {
4436         struct inode *journal_inode;
4437         journal_t *journal;
4438
4439         BUG_ON(!ext4_has_feature_journal(sb));
4440
4441         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4442         if (!journal_inode)
4443                 return NULL;
4444
4445         journal = jbd2_journal_init_inode(journal_inode);
4446         if (!journal) {
4447                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4448                 iput(journal_inode);
4449                 return NULL;
4450         }
4451         journal->j_private = sb;
4452         ext4_init_journal_params(sb, journal);
4453         return journal;
4454 }
4455
4456 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4457                                        dev_t j_dev)
4458 {
4459         struct buffer_head *bh;
4460         journal_t *journal;
4461         ext4_fsblk_t start;
4462         ext4_fsblk_t len;
4463         int hblock, blocksize;
4464         ext4_fsblk_t sb_block;
4465         unsigned long offset;
4466         struct ext4_super_block *es;
4467         struct block_device *bdev;
4468
4469         BUG_ON(!ext4_has_feature_journal(sb));
4470
4471         bdev = ext4_blkdev_get(j_dev, sb);
4472         if (bdev == NULL)
4473                 return NULL;
4474
4475         blocksize = sb->s_blocksize;
4476         hblock = bdev_logical_block_size(bdev);
4477         if (blocksize < hblock) {
4478                 ext4_msg(sb, KERN_ERR,
4479                         "blocksize too small for journal device");
4480                 goto out_bdev;
4481         }
4482
4483         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4484         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4485         set_blocksize(bdev, blocksize);
4486         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4487                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4488                        "external journal");
4489                 goto out_bdev;
4490         }
4491
4492         es = (struct ext4_super_block *) (bh->b_data + offset);
4493         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4494             !(le32_to_cpu(es->s_feature_incompat) &
4495               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4496                 ext4_msg(sb, KERN_ERR, "external journal has "
4497                                         "bad superblock");
4498                 brelse(bh);
4499                 goto out_bdev;
4500         }
4501
4502         if ((le32_to_cpu(es->s_feature_ro_compat) &
4503              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4504             es->s_checksum != ext4_superblock_csum(sb, es)) {
4505                 ext4_msg(sb, KERN_ERR, "external journal has "
4506                                        "corrupt superblock");
4507                 brelse(bh);
4508                 goto out_bdev;
4509         }
4510
4511         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4512                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4513                 brelse(bh);
4514                 goto out_bdev;
4515         }
4516
4517         len = ext4_blocks_count(es);
4518         start = sb_block + 1;
4519         brelse(bh);     /* we're done with the superblock */
4520
4521         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4522                                         start, len, blocksize);
4523         if (!journal) {
4524                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4525                 goto out_bdev;
4526         }
4527         journal->j_private = sb;
4528         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4529         wait_on_buffer(journal->j_sb_buffer);
4530         if (!buffer_uptodate(journal->j_sb_buffer)) {
4531                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4532                 goto out_journal;
4533         }
4534         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4535                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4536                                         "user (unsupported) - %d",
4537                         be32_to_cpu(journal->j_superblock->s_nr_users));
4538                 goto out_journal;
4539         }
4540         EXT4_SB(sb)->journal_bdev = bdev;
4541         ext4_init_journal_params(sb, journal);
4542         return journal;
4543
4544 out_journal:
4545         jbd2_journal_destroy(journal);
4546 out_bdev:
4547         ext4_blkdev_put(bdev);
4548         return NULL;
4549 }
4550
4551 static int ext4_load_journal(struct super_block *sb,
4552                              struct ext4_super_block *es,
4553                              unsigned long journal_devnum)
4554 {
4555         journal_t *journal;
4556         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4557         dev_t journal_dev;
4558         int err = 0;
4559         int really_read_only;
4560
4561         BUG_ON(!ext4_has_feature_journal(sb));
4562
4563         if (journal_devnum &&
4564             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4565                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4566                         "numbers have changed");
4567                 journal_dev = new_decode_dev(journal_devnum);
4568         } else
4569                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4570
4571         really_read_only = bdev_read_only(sb->s_bdev);
4572
4573         /*
4574          * Are we loading a blank journal or performing recovery after a
4575          * crash?  For recovery, we need to check in advance whether we
4576          * can get read-write access to the device.
4577          */
4578         if (ext4_has_feature_journal_needs_recovery(sb)) {
4579                 if (sb->s_flags & MS_RDONLY) {
4580                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4581                                         "required on readonly filesystem");
4582                         if (really_read_only) {
4583                                 ext4_msg(sb, KERN_ERR, "write access "
4584                                         "unavailable, cannot proceed");
4585                                 return -EROFS;
4586                         }
4587                         ext4_msg(sb, KERN_INFO, "write access will "
4588                                "be enabled during recovery");
4589                 }
4590         }
4591
4592         if (journal_inum && journal_dev) {
4593                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4594                        "and inode journals!");
4595                 return -EINVAL;
4596         }
4597
4598         if (journal_inum) {
4599                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4600                         return -EINVAL;
4601         } else {
4602                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4603                         return -EINVAL;
4604         }
4605
4606         if (!(journal->j_flags & JBD2_BARRIER))
4607                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4608
4609         if (!ext4_has_feature_journal_needs_recovery(sb))
4610                 err = jbd2_journal_wipe(journal, !really_read_only);
4611         if (!err) {
4612                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4613                 if (save)
4614                         memcpy(save, ((char *) es) +
4615                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4616                 err = jbd2_journal_load(journal);
4617                 if (save)
4618                         memcpy(((char *) es) + EXT4_S_ERR_START,
4619                                save, EXT4_S_ERR_LEN);
4620                 kfree(save);
4621         }
4622
4623         if (err) {
4624                 ext4_msg(sb, KERN_ERR, "error loading journal");
4625                 jbd2_journal_destroy(journal);
4626                 return err;
4627         }
4628
4629         EXT4_SB(sb)->s_journal = journal;
4630         ext4_clear_journal_err(sb, es);
4631
4632         if (!really_read_only && journal_devnum &&
4633             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4634                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4635
4636                 /* Make sure we flush the recovery flag to disk. */
4637                 ext4_commit_super(sb, 1);
4638         }
4639
4640         return 0;
4641 }
4642
4643 static int ext4_commit_super(struct super_block *sb, int sync)
4644 {
4645         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4646         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4647         int error = 0;
4648
4649         if (!sbh || block_device_ejected(sb))
4650                 return error;
4651         /*
4652          * If the file system is mounted read-only, don't update the
4653          * superblock write time.  This avoids updating the superblock
4654          * write time when we are mounting the root file system
4655          * read/only but we need to replay the journal; at that point,
4656          * for people who are east of GMT and who make their clock
4657          * tick in localtime for Windows bug-for-bug compatibility,
4658          * the clock is set in the future, and this will cause e2fsck
4659          * to complain and force a full file system check.
4660          */
4661         if (!(sb->s_flags & MS_RDONLY))
4662                 es->s_wtime = cpu_to_le32(get_seconds());
4663         if (sb->s_bdev->bd_part)
4664                 es->s_kbytes_written =
4665                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4666                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4667                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4668         else
4669                 es->s_kbytes_written =
4670                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4671         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4672                 ext4_free_blocks_count_set(es,
4673                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4674                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4675         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4676                 es->s_free_inodes_count =
4677                         cpu_to_le32(percpu_counter_sum_positive(
4678                                 &EXT4_SB(sb)->s_freeinodes_counter));
4679         BUFFER_TRACE(sbh, "marking dirty");
4680         ext4_superblock_csum_set(sb);
4681         if (sync)
4682                 lock_buffer(sbh);
4683         if (buffer_write_io_error(sbh)) {
4684                 /*
4685                  * Oh, dear.  A previous attempt to write the
4686                  * superblock failed.  This could happen because the
4687                  * USB device was yanked out.  Or it could happen to
4688                  * be a transient write error and maybe the block will
4689                  * be remapped.  Nothing we can do but to retry the
4690                  * write and hope for the best.
4691                  */
4692                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4693                        "superblock detected");
4694                 clear_buffer_write_io_error(sbh);
4695                 set_buffer_uptodate(sbh);
4696         }
4697         mark_buffer_dirty(sbh);
4698         if (sync) {
4699                 unlock_buffer(sbh);
4700                 error = __sync_dirty_buffer(sbh,
4701                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4702                 if (error)
4703                         return error;
4704
4705                 error = buffer_write_io_error(sbh);
4706                 if (error) {
4707                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4708                                "superblock");
4709                         clear_buffer_write_io_error(sbh);
4710                         set_buffer_uptodate(sbh);
4711                 }
4712         }
4713         return error;
4714 }
4715
4716 /*
4717  * Have we just finished recovery?  If so, and if we are mounting (or
4718  * remounting) the filesystem readonly, then we will end up with a
4719  * consistent fs on disk.  Record that fact.
4720  */
4721 static void ext4_mark_recovery_complete(struct super_block *sb,
4722                                         struct ext4_super_block *es)
4723 {
4724         journal_t *journal = EXT4_SB(sb)->s_journal;
4725
4726         if (!ext4_has_feature_journal(sb)) {
4727                 BUG_ON(journal != NULL);
4728                 return;
4729         }
4730         jbd2_journal_lock_updates(journal);
4731         if (jbd2_journal_flush(journal) < 0)
4732                 goto out;
4733
4734         if (ext4_has_feature_journal_needs_recovery(sb) &&
4735             sb->s_flags & MS_RDONLY) {
4736                 ext4_clear_feature_journal_needs_recovery(sb);
4737                 ext4_commit_super(sb, 1);
4738         }
4739
4740 out:
4741         jbd2_journal_unlock_updates(journal);
4742 }
4743
4744 /*
4745  * If we are mounting (or read-write remounting) a filesystem whose journal
4746  * has recorded an error from a previous lifetime, move that error to the
4747  * main filesystem now.
4748  */
4749 static void ext4_clear_journal_err(struct super_block *sb,
4750                                    struct ext4_super_block *es)
4751 {
4752         journal_t *journal;
4753         int j_errno;
4754         const char *errstr;
4755
4756         BUG_ON(!ext4_has_feature_journal(sb));
4757
4758         journal = EXT4_SB(sb)->s_journal;
4759
4760         /*
4761          * Now check for any error status which may have been recorded in the
4762          * journal by a prior ext4_error() or ext4_abort()
4763          */
4764
4765         j_errno = jbd2_journal_errno(journal);
4766         if (j_errno) {
4767                 char nbuf[16];
4768
4769                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4770                 ext4_warning(sb, "Filesystem error recorded "
4771                              "from previous mount: %s", errstr);
4772                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4773
4774                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4775                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4776                 ext4_commit_super(sb, 1);
4777
4778                 jbd2_journal_clear_err(journal);
4779                 jbd2_journal_update_sb_errno(journal);
4780         }
4781 }
4782
4783 /*
4784  * Force the running and committing transactions to commit,
4785  * and wait on the commit.
4786  */
4787 int ext4_force_commit(struct super_block *sb)
4788 {
4789         journal_t *journal;
4790
4791         if (sb->s_flags & MS_RDONLY)
4792                 return 0;
4793
4794         journal = EXT4_SB(sb)->s_journal;
4795         return ext4_journal_force_commit(journal);
4796 }
4797
4798 static int ext4_sync_fs(struct super_block *sb, int wait)
4799 {
4800         int ret = 0;
4801         tid_t target;
4802         bool needs_barrier = false;
4803         struct ext4_sb_info *sbi = EXT4_SB(sb);
4804
4805         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4806                 return 0;
4807
4808         trace_ext4_sync_fs(sb, wait);
4809         flush_workqueue(sbi->rsv_conversion_wq);
4810         /*
4811          * Writeback quota in non-journalled quota case - journalled quota has
4812          * no dirty dquots
4813          */
4814         dquot_writeback_dquots(sb, -1);
4815         /*
4816          * Data writeback is possible w/o journal transaction, so barrier must
4817          * being sent at the end of the function. But we can skip it if
4818          * transaction_commit will do it for us.
4819          */
4820         if (sbi->s_journal) {
4821                 target = jbd2_get_latest_transaction(sbi->s_journal);
4822                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4823                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4824                         needs_barrier = true;
4825
4826                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4827                         if (wait)
4828                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4829                                                            target);
4830                 }
4831         } else if (wait && test_opt(sb, BARRIER))
4832                 needs_barrier = true;
4833         if (needs_barrier) {
4834                 int err;
4835                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4836                 if (!ret)
4837                         ret = err;
4838         }
4839
4840         return ret;
4841 }
4842
4843 /*
4844  * LVM calls this function before a (read-only) snapshot is created.  This
4845  * gives us a chance to flush the journal completely and mark the fs clean.
4846  *
4847  * Note that only this function cannot bring a filesystem to be in a clean
4848  * state independently. It relies on upper layer to stop all data & metadata
4849  * modifications.
4850  */
4851 static int ext4_freeze(struct super_block *sb)
4852 {
4853         int error = 0;
4854         journal_t *journal;
4855
4856         if (sb->s_flags & MS_RDONLY)
4857                 return 0;
4858
4859         journal = EXT4_SB(sb)->s_journal;
4860
4861         if (journal) {
4862                 /* Now we set up the journal barrier. */
4863                 jbd2_journal_lock_updates(journal);
4864
4865                 /*
4866                  * Don't clear the needs_recovery flag if we failed to
4867                  * flush the journal.
4868                  */
4869                 error = jbd2_journal_flush(journal);
4870                 if (error < 0)
4871                         goto out;
4872
4873                 /* Journal blocked and flushed, clear needs_recovery flag. */
4874                 ext4_clear_feature_journal_needs_recovery(sb);
4875         }
4876
4877         error = ext4_commit_super(sb, 1);
4878 out:
4879         if (journal)
4880                 /* we rely on upper layer to stop further updates */
4881                 jbd2_journal_unlock_updates(journal);
4882         return error;
4883 }
4884
4885 /*
4886  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4887  * flag here, even though the filesystem is not technically dirty yet.
4888  */
4889 static int ext4_unfreeze(struct super_block *sb)
4890 {
4891         if ((sb->s_flags & MS_RDONLY) || ext4_forced_shutdown(EXT4_SB(sb)))
4892                 return 0;
4893
4894         if (EXT4_SB(sb)->s_journal) {
4895                 /* Reset the needs_recovery flag before the fs is unlocked. */
4896                 ext4_set_feature_journal_needs_recovery(sb);
4897         }
4898
4899         ext4_commit_super(sb, 1);
4900         return 0;
4901 }
4902
4903 /*
4904  * Structure to save mount options for ext4_remount's benefit
4905  */
4906 struct ext4_mount_options {
4907         unsigned long s_mount_opt;
4908         unsigned long s_mount_opt2;
4909         kuid_t s_resuid;
4910         kgid_t s_resgid;
4911         unsigned long s_commit_interval;
4912         u32 s_min_batch_time, s_max_batch_time;
4913 #ifdef CONFIG_QUOTA
4914         int s_jquota_fmt;
4915         char *s_qf_names[EXT4_MAXQUOTAS];
4916 #endif
4917 };
4918
4919 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4920 {
4921         struct ext4_super_block *es;
4922         struct ext4_sb_info *sbi = EXT4_SB(sb);
4923         unsigned long old_sb_flags;
4924         struct ext4_mount_options old_opts;
4925         int enable_quota = 0;
4926         ext4_group_t g;
4927         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4928         int err = 0;
4929 #ifdef CONFIG_QUOTA
4930         int i, j;
4931 #endif
4932         char *orig_data = kstrdup(data, GFP_KERNEL);
4933
4934         /* Store the original options */
4935         old_sb_flags = sb->s_flags;
4936         old_opts.s_mount_opt = sbi->s_mount_opt;
4937         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4938         old_opts.s_resuid = sbi->s_resuid;
4939         old_opts.s_resgid = sbi->s_resgid;
4940         old_opts.s_commit_interval = sbi->s_commit_interval;
4941         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4942         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4943 #ifdef CONFIG_QUOTA
4944         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4945         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4946                 if (sbi->s_qf_names[i]) {
4947                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4948                                                          GFP_KERNEL);
4949                         if (!old_opts.s_qf_names[i]) {
4950                                 for (j = 0; j < i; j++)
4951                                         kfree(old_opts.s_qf_names[j]);
4952                                 kfree(orig_data);
4953                                 return -ENOMEM;
4954                         }
4955                 } else
4956                         old_opts.s_qf_names[i] = NULL;
4957 #endif
4958         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4959                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4960
4961         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4962                 err = -EINVAL;
4963                 goto restore_opts;
4964         }
4965
4966         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4967             test_opt(sb, JOURNAL_CHECKSUM)) {
4968                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4969                          "during remount not supported; ignoring");
4970                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4971         }
4972
4973         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4974                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4975                         ext4_msg(sb, KERN_ERR, "can't mount with "
4976                                  "both data=journal and delalloc");
4977                         err = -EINVAL;
4978                         goto restore_opts;
4979                 }
4980                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4981                         ext4_msg(sb, KERN_ERR, "can't mount with "
4982                                  "both data=journal and dioread_nolock");
4983                         err = -EINVAL;
4984                         goto restore_opts;
4985                 }
4986                 if (test_opt(sb, DAX)) {
4987                         ext4_msg(sb, KERN_ERR, "can't mount with "
4988                                  "both data=journal and dax");
4989                         err = -EINVAL;
4990                         goto restore_opts;
4991                 }
4992         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
4993                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4994                         ext4_msg(sb, KERN_ERR, "can't mount with "
4995                                 "journal_async_commit in data=ordered mode");
4996                         err = -EINVAL;
4997                         goto restore_opts;
4998                 }
4999         }
5000
5001         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5002                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5003                 err = -EINVAL;
5004                 goto restore_opts;
5005         }
5006
5007         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5008                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5009                         "dax flag with busy inodes while remounting");
5010                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5011         }
5012
5013         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5014                 ext4_abort(sb, "Abort forced by user");
5015
5016         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
5017                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
5018
5019         es = sbi->s_es;
5020
5021         if (sbi->s_journal) {
5022                 ext4_init_journal_params(sb, sbi->s_journal);
5023                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5024         }
5025
5026         if (*flags & MS_LAZYTIME)
5027                 sb->s_flags |= MS_LAZYTIME;
5028
5029         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
5030                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5031                         err = -EROFS;
5032                         goto restore_opts;
5033                 }
5034
5035                 if (*flags & MS_RDONLY) {
5036                         err = sync_filesystem(sb);
5037                         if (err < 0)
5038                                 goto restore_opts;
5039                         err = dquot_suspend(sb, -1);
5040                         if (err < 0)
5041                                 goto restore_opts;
5042
5043                         /*
5044                          * First of all, the unconditional stuff we have to do
5045                          * to disable replay of the journal when we next remount
5046                          */
5047                         sb->s_flags |= MS_RDONLY;
5048
5049                         /*
5050                          * OK, test if we are remounting a valid rw partition
5051                          * readonly, and if so set the rdonly flag and then
5052                          * mark the partition as valid again.
5053                          */
5054                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5055                             (sbi->s_mount_state & EXT4_VALID_FS))
5056                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5057
5058                         if (sbi->s_journal)
5059                                 ext4_mark_recovery_complete(sb, es);
5060                 } else {
5061                         /* Make sure we can mount this feature set readwrite */
5062                         if (ext4_has_feature_readonly(sb) ||
5063                             !ext4_feature_set_ok(sb, 0)) {
5064                                 err = -EROFS;
5065                                 goto restore_opts;
5066                         }
5067                         /*
5068                          * Make sure the group descriptor checksums
5069                          * are sane.  If they aren't, refuse to remount r/w.
5070                          */
5071                         for (g = 0; g < sbi->s_groups_count; g++) {
5072                                 struct ext4_group_desc *gdp =
5073                                         ext4_get_group_desc(sb, g, NULL);
5074
5075                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5076                                         ext4_msg(sb, KERN_ERR,
5077                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5078                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5079                                                le16_to_cpu(gdp->bg_checksum));
5080                                         err = -EFSBADCRC;
5081                                         goto restore_opts;
5082                                 }
5083                         }
5084
5085                         /*
5086                          * If we have an unprocessed orphan list hanging
5087                          * around from a previously readonly bdev mount,
5088                          * require a full umount/remount for now.
5089                          */
5090                         if (es->s_last_orphan) {
5091                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5092                                        "remount RDWR because of unprocessed "
5093                                        "orphan inode list.  Please "
5094                                        "umount/remount instead");
5095                                 err = -EINVAL;
5096                                 goto restore_opts;
5097                         }
5098
5099                         /*
5100                          * Mounting a RDONLY partition read-write, so reread
5101                          * and store the current valid flag.  (It may have
5102                          * been changed by e2fsck since we originally mounted
5103                          * the partition.)
5104                          */
5105                         if (sbi->s_journal)
5106                                 ext4_clear_journal_err(sb, es);
5107                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5108                         if (!ext4_setup_super(sb, es, 0))
5109                                 sb->s_flags &= ~MS_RDONLY;
5110                         if (ext4_has_feature_mmp(sb))
5111                                 if (ext4_multi_mount_protect(sb,
5112                                                 le64_to_cpu(es->s_mmp_block))) {
5113                                         err = -EROFS;
5114                                         goto restore_opts;
5115                                 }
5116                         enable_quota = 1;
5117                 }
5118         }
5119
5120         /*
5121          * Reinitialize lazy itable initialization thread based on
5122          * current settings
5123          */
5124         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5125                 ext4_unregister_li_request(sb);
5126         else {
5127                 ext4_group_t first_not_zeroed;
5128                 first_not_zeroed = ext4_has_uninit_itable(sb);
5129                 ext4_register_li_request(sb, first_not_zeroed);
5130         }
5131
5132         ext4_setup_system_zone(sb);
5133         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5134                 ext4_commit_super(sb, 1);
5135
5136 #ifdef CONFIG_QUOTA
5137         /* Release old quota file names */
5138         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5139                 kfree(old_opts.s_qf_names[i]);
5140         if (enable_quota) {
5141                 if (sb_any_quota_suspended(sb))
5142                         dquot_resume(sb, -1);
5143                 else if (ext4_has_feature_quota(sb)) {
5144                         err = ext4_enable_quotas(sb);
5145                         if (err)
5146                                 goto restore_opts;
5147                 }
5148         }
5149 #endif
5150
5151         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5152         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5153         kfree(orig_data);
5154         return 0;
5155
5156 restore_opts:
5157         sb->s_flags = old_sb_flags;
5158         sbi->s_mount_opt = old_opts.s_mount_opt;
5159         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5160         sbi->s_resuid = old_opts.s_resuid;
5161         sbi->s_resgid = old_opts.s_resgid;
5162         sbi->s_commit_interval = old_opts.s_commit_interval;
5163         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5164         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5165 #ifdef CONFIG_QUOTA
5166         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5167         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5168                 kfree(sbi->s_qf_names[i]);
5169                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5170         }
5171 #endif
5172         kfree(orig_data);
5173         return err;
5174 }
5175
5176 #ifdef CONFIG_QUOTA
5177 static int ext4_statfs_project(struct super_block *sb,
5178                                kprojid_t projid, struct kstatfs *buf)
5179 {
5180         struct kqid qid;
5181         struct dquot *dquot;
5182         u64 limit;
5183         u64 curblock;
5184
5185         qid = make_kqid_projid(projid);
5186         dquot = dqget(sb, qid);
5187         if (IS_ERR(dquot))
5188                 return PTR_ERR(dquot);
5189         spin_lock(&dq_data_lock);
5190
5191         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5192                  dquot->dq_dqb.dqb_bsoftlimit :
5193                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5194         if (limit && buf->f_blocks > limit) {
5195                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5196                 buf->f_blocks = limit;
5197                 buf->f_bfree = buf->f_bavail =
5198                         (buf->f_blocks > curblock) ?
5199                          (buf->f_blocks - curblock) : 0;
5200         }
5201
5202         limit = dquot->dq_dqb.dqb_isoftlimit ?
5203                 dquot->dq_dqb.dqb_isoftlimit :
5204                 dquot->dq_dqb.dqb_ihardlimit;
5205         if (limit && buf->f_files > limit) {
5206                 buf->f_files = limit;
5207                 buf->f_ffree =
5208                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5209                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5210         }
5211
5212         spin_unlock(&dq_data_lock);
5213         dqput(dquot);
5214         return 0;
5215 }
5216 #endif
5217
5218 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5219 {
5220         struct super_block *sb = dentry->d_sb;
5221         struct ext4_sb_info *sbi = EXT4_SB(sb);
5222         struct ext4_super_block *es = sbi->s_es;
5223         ext4_fsblk_t overhead = 0, resv_blocks;
5224         u64 fsid;
5225         s64 bfree;
5226         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5227
5228         if (!test_opt(sb, MINIX_DF))
5229                 overhead = sbi->s_overhead;
5230
5231         buf->f_type = EXT4_SUPER_MAGIC;
5232         buf->f_bsize = sb->s_blocksize;
5233         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5234         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5235                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5236         /* prevent underflow in case that few free space is available */
5237         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5238         buf->f_bavail = buf->f_bfree -
5239                         (ext4_r_blocks_count(es) + resv_blocks);
5240         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5241                 buf->f_bavail = 0;
5242         buf->f_files = le32_to_cpu(es->s_inodes_count);
5243         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5244         buf->f_namelen = EXT4_NAME_LEN;
5245         fsid = le64_to_cpup((void *)es->s_uuid) ^
5246                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5247         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5248         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5249
5250 #ifdef CONFIG_QUOTA
5251         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5252             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5253                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5254 #endif
5255         return 0;
5256 }
5257
5258 /* Helper function for writing quotas on sync - we need to start transaction
5259  * before quota file is locked for write. Otherwise the are possible deadlocks:
5260  * Process 1                         Process 2
5261  * ext4_create()                     quota_sync()
5262  *   jbd2_journal_start()                  write_dquot()
5263  *   dquot_initialize()                         down(dqio_mutex)
5264  *     down(dqio_mutex)                    jbd2_journal_start()
5265  *
5266  */
5267
5268 #ifdef CONFIG_QUOTA
5269
5270 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5271 {
5272         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5273 }
5274
5275 static int ext4_write_dquot(struct dquot *dquot)
5276 {
5277         int ret, err;
5278         handle_t *handle;
5279         struct inode *inode;
5280
5281         inode = dquot_to_inode(dquot);
5282         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5283                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5284         if (IS_ERR(handle))
5285                 return PTR_ERR(handle);
5286         ret = dquot_commit(dquot);
5287         err = ext4_journal_stop(handle);
5288         if (!ret)
5289                 ret = err;
5290         return ret;
5291 }
5292
5293 static int ext4_acquire_dquot(struct dquot *dquot)
5294 {
5295         int ret, err;
5296         handle_t *handle;
5297
5298         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5299                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5300         if (IS_ERR(handle))
5301                 return PTR_ERR(handle);
5302         ret = dquot_acquire(dquot);
5303         err = ext4_journal_stop(handle);
5304         if (!ret)
5305                 ret = err;
5306         return ret;
5307 }
5308
5309 static int ext4_release_dquot(struct dquot *dquot)
5310 {
5311         int ret, err;
5312         handle_t *handle;
5313
5314         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5315                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5316         if (IS_ERR(handle)) {
5317                 /* Release dquot anyway to avoid endless cycle in dqput() */
5318                 dquot_release(dquot);
5319                 return PTR_ERR(handle);
5320         }
5321         ret = dquot_release(dquot);
5322         err = ext4_journal_stop(handle);
5323         if (!ret)
5324                 ret = err;
5325         return ret;
5326 }
5327
5328 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5329 {
5330         struct super_block *sb = dquot->dq_sb;
5331         struct ext4_sb_info *sbi = EXT4_SB(sb);
5332
5333         /* Are we journaling quotas? */
5334         if (ext4_has_feature_quota(sb) ||
5335             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5336                 dquot_mark_dquot_dirty(dquot);
5337                 return ext4_write_dquot(dquot);
5338         } else {
5339                 return dquot_mark_dquot_dirty(dquot);
5340         }
5341 }
5342
5343 static int ext4_write_info(struct super_block *sb, int type)
5344 {
5345         int ret, err;
5346         handle_t *handle;
5347
5348         /* Data block + inode block */
5349         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5350         if (IS_ERR(handle))
5351                 return PTR_ERR(handle);
5352         ret = dquot_commit_info(sb, type);
5353         err = ext4_journal_stop(handle);
5354         if (!ret)
5355                 ret = err;
5356         return ret;
5357 }
5358
5359 /*
5360  * Turn on quotas during mount time - we need to find
5361  * the quota file and such...
5362  */
5363 static int ext4_quota_on_mount(struct super_block *sb, int type)
5364 {
5365         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5366                                         EXT4_SB(sb)->s_jquota_fmt, type);
5367 }
5368
5369 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5370 {
5371         struct ext4_inode_info *ei = EXT4_I(inode);
5372
5373         /* The first argument of lockdep_set_subclass has to be
5374          * *exactly* the same as the argument to init_rwsem() --- in
5375          * this case, in init_once() --- or lockdep gets unhappy
5376          * because the name of the lock is set using the
5377          * stringification of the argument to init_rwsem().
5378          */
5379         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5380         lockdep_set_subclass(&ei->i_data_sem, subclass);
5381 }
5382
5383 /*
5384  * Standard function to be called on quota_on
5385  */
5386 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5387                          const struct path *path)
5388 {
5389         int err;
5390
5391         if (!test_opt(sb, QUOTA))
5392                 return -EINVAL;
5393
5394         /* Quotafile not on the same filesystem? */
5395         if (path->dentry->d_sb != sb)
5396                 return -EXDEV;
5397         /* Journaling quota? */
5398         if (EXT4_SB(sb)->s_qf_names[type]) {
5399                 /* Quotafile not in fs root? */
5400                 if (path->dentry->d_parent != sb->s_root)
5401                         ext4_msg(sb, KERN_WARNING,
5402                                 "Quota file not on filesystem root. "
5403                                 "Journaled quota will not work");
5404         }
5405
5406         /*
5407          * When we journal data on quota file, we have to flush journal to see
5408          * all updates to the file when we bypass pagecache...
5409          */
5410         if (EXT4_SB(sb)->s_journal &&
5411             ext4_should_journal_data(d_inode(path->dentry))) {
5412                 /*
5413                  * We don't need to lock updates but journal_flush() could
5414                  * otherwise be livelocked...
5415                  */
5416                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5417                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5418                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5419                 if (err)
5420                         return err;
5421         }
5422
5423         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5424         err = dquot_quota_on(sb, type, format_id, path);
5425         if (err) {
5426                 lockdep_set_quota_inode(path->dentry->d_inode,
5427                                              I_DATA_SEM_NORMAL);
5428         } else {
5429                 struct inode *inode = d_inode(path->dentry);
5430                 handle_t *handle;
5431
5432                 /*
5433                  * Set inode flags to prevent userspace from messing with quota
5434                  * files. If this fails, we return success anyway since quotas
5435                  * are already enabled and this is not a hard failure.
5436                  */
5437                 inode_lock(inode);
5438                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5439                 if (IS_ERR(handle))
5440                         goto unlock_inode;
5441                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5442                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5443                                 S_NOATIME | S_IMMUTABLE);
5444                 ext4_mark_inode_dirty(handle, inode);
5445                 ext4_journal_stop(handle);
5446         unlock_inode:
5447                 inode_unlock(inode);
5448         }
5449         return err;
5450 }
5451
5452 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5453                              unsigned int flags)
5454 {
5455         int err;
5456         struct inode *qf_inode;
5457         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5458                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5459                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5460                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5461         };
5462
5463         BUG_ON(!ext4_has_feature_quota(sb));
5464
5465         if (!qf_inums[type])
5466                 return -EPERM;
5467
5468         qf_inode = ext4_iget(sb, qf_inums[type]);
5469         if (IS_ERR(qf_inode)) {
5470                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5471                 return PTR_ERR(qf_inode);
5472         }
5473
5474         /* Don't account quota for quota files to avoid recursion */
5475         qf_inode->i_flags |= S_NOQUOTA;
5476         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5477         err = dquot_enable(qf_inode, type, format_id, flags);
5478         iput(qf_inode);
5479         if (err)
5480                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5481
5482         return err;
5483 }
5484
5485 /* Enable usage tracking for all quota types. */
5486 static int ext4_enable_quotas(struct super_block *sb)
5487 {
5488         int type, err = 0;
5489         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5490                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5491                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5492                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5493         };
5494         bool quota_mopt[EXT4_MAXQUOTAS] = {
5495                 test_opt(sb, USRQUOTA),
5496                 test_opt(sb, GRPQUOTA),
5497                 test_opt(sb, PRJQUOTA),
5498         };
5499
5500         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5501         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5502                 if (qf_inums[type]) {
5503                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5504                                 DQUOT_USAGE_ENABLED |
5505                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5506                         if (err) {
5507                                 ext4_warning(sb,
5508                                         "Failed to enable quota tracking "
5509                                         "(type=%d, err=%d). Please run "
5510                                         "e2fsck to fix.", type, err);
5511                                 return err;
5512                         }
5513                 }
5514         }
5515         return 0;
5516 }
5517
5518 static int ext4_quota_off(struct super_block *sb, int type)
5519 {
5520         struct inode *inode = sb_dqopt(sb)->files[type];
5521         handle_t *handle;
5522         int err;
5523
5524         /* Force all delayed allocation blocks to be allocated.
5525          * Caller already holds s_umount sem */
5526         if (test_opt(sb, DELALLOC))
5527                 sync_filesystem(sb);
5528
5529         if (!inode || !igrab(inode))
5530                 goto out;
5531
5532         err = dquot_quota_off(sb, type);
5533         if (err || ext4_has_feature_quota(sb))
5534                 goto out_put;
5535
5536         inode_lock(inode);
5537         /*
5538          * Update modification times of quota files when userspace can
5539          * start looking at them. If we fail, we return success anyway since
5540          * this is not a hard failure and quotas are already disabled.
5541          */
5542         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5543         if (IS_ERR(handle))
5544                 goto out_unlock;
5545         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5546         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5547         inode->i_mtime = inode->i_ctime = current_time(inode);
5548         ext4_mark_inode_dirty(handle, inode);
5549         ext4_journal_stop(handle);
5550 out_unlock:
5551         inode_unlock(inode);
5552 out_put:
5553         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5554         iput(inode);
5555         return err;
5556 out:
5557         return dquot_quota_off(sb, type);
5558 }
5559
5560 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5561  * acquiring the locks... As quota files are never truncated and quota code
5562  * itself serializes the operations (and no one else should touch the files)
5563  * we don't have to be afraid of races */
5564 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5565                                size_t len, loff_t off)
5566 {
5567         struct inode *inode = sb_dqopt(sb)->files[type];
5568         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5569         int offset = off & (sb->s_blocksize - 1);
5570         int tocopy;
5571         size_t toread;
5572         struct buffer_head *bh;
5573         loff_t i_size = i_size_read(inode);
5574
5575         if (off > i_size)
5576                 return 0;
5577         if (off+len > i_size)
5578                 len = i_size-off;
5579         toread = len;
5580         while (toread > 0) {
5581                 tocopy = sb->s_blocksize - offset < toread ?
5582                                 sb->s_blocksize - offset : toread;
5583                 bh = ext4_bread(NULL, inode, blk, 0);
5584                 if (IS_ERR(bh))
5585                         return PTR_ERR(bh);
5586                 if (!bh)        /* A hole? */
5587                         memset(data, 0, tocopy);
5588                 else
5589                         memcpy(data, bh->b_data+offset, tocopy);
5590                 brelse(bh);
5591                 offset = 0;
5592                 toread -= tocopy;
5593                 data += tocopy;
5594                 blk++;
5595         }
5596         return len;
5597 }
5598
5599 /* Write to quotafile (we know the transaction is already started and has
5600  * enough credits) */
5601 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5602                                 const char *data, size_t len, loff_t off)
5603 {
5604         struct inode *inode = sb_dqopt(sb)->files[type];
5605         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5606         int err, offset = off & (sb->s_blocksize - 1);
5607         int retries = 0;
5608         struct buffer_head *bh;
5609         handle_t *handle = journal_current_handle();
5610
5611         if (EXT4_SB(sb)->s_journal && !handle) {
5612                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5613                         " cancelled because transaction is not started",
5614                         (unsigned long long)off, (unsigned long long)len);
5615                 return -EIO;
5616         }
5617         /*
5618          * Since we account only one data block in transaction credits,
5619          * then it is impossible to cross a block boundary.
5620          */
5621         if (sb->s_blocksize - offset < len) {
5622                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5623                         " cancelled because not block aligned",
5624                         (unsigned long long)off, (unsigned long long)len);
5625                 return -EIO;
5626         }
5627
5628         do {
5629                 bh = ext4_bread(handle, inode, blk,
5630                                 EXT4_GET_BLOCKS_CREATE |
5631                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5632         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5633                  ext4_should_retry_alloc(inode->i_sb, &retries));
5634         if (IS_ERR(bh))
5635                 return PTR_ERR(bh);
5636         if (!bh)
5637                 goto out;
5638         BUFFER_TRACE(bh, "get write access");
5639         err = ext4_journal_get_write_access(handle, bh);
5640         if (err) {
5641                 brelse(bh);
5642                 return err;
5643         }
5644         lock_buffer(bh);
5645         memcpy(bh->b_data+offset, data, len);
5646         flush_dcache_page(bh->b_page);
5647         unlock_buffer(bh);
5648         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5649         brelse(bh);
5650 out:
5651         if (inode->i_size < off + len) {
5652                 i_size_write(inode, off + len);
5653                 EXT4_I(inode)->i_disksize = inode->i_size;
5654                 ext4_mark_inode_dirty(handle, inode);
5655         }
5656         return len;
5657 }
5658
5659 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5660 {
5661         const struct quota_format_ops   *ops;
5662
5663         if (!sb_has_quota_loaded(sb, qid->type))
5664                 return -ESRCH;
5665         ops = sb_dqopt(sb)->ops[qid->type];
5666         if (!ops || !ops->get_next_id)
5667                 return -ENOSYS;
5668         return dquot_get_next_id(sb, qid);
5669 }
5670 #endif
5671
5672 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5673                        const char *dev_name, void *data)
5674 {
5675         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5676 }
5677
5678 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5679 static inline void register_as_ext2(void)
5680 {
5681         int err = register_filesystem(&ext2_fs_type);
5682         if (err)
5683                 printk(KERN_WARNING
5684                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5685 }
5686
5687 static inline void unregister_as_ext2(void)
5688 {
5689         unregister_filesystem(&ext2_fs_type);
5690 }
5691
5692 static inline int ext2_feature_set_ok(struct super_block *sb)
5693 {
5694         if (ext4_has_unknown_ext2_incompat_features(sb))
5695                 return 0;
5696         if (sb->s_flags & MS_RDONLY)
5697                 return 1;
5698         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5699                 return 0;
5700         return 1;
5701 }
5702 #else
5703 static inline void register_as_ext2(void) { }
5704 static inline void unregister_as_ext2(void) { }
5705 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5706 #endif
5707
5708 static inline void register_as_ext3(void)
5709 {
5710         int err = register_filesystem(&ext3_fs_type);
5711         if (err)
5712                 printk(KERN_WARNING
5713                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5714 }
5715
5716 static inline void unregister_as_ext3(void)
5717 {
5718         unregister_filesystem(&ext3_fs_type);
5719 }
5720
5721 static inline int ext3_feature_set_ok(struct super_block *sb)
5722 {
5723         if (ext4_has_unknown_ext3_incompat_features(sb))
5724                 return 0;
5725         if (!ext4_has_feature_journal(sb))
5726                 return 0;
5727         if (sb->s_flags & MS_RDONLY)
5728                 return 1;
5729         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5730                 return 0;
5731         return 1;
5732 }
5733
5734 static struct file_system_type ext4_fs_type = {
5735         .owner          = THIS_MODULE,
5736         .name           = "ext4",
5737         .mount          = ext4_mount,
5738         .kill_sb        = kill_block_super,
5739         .fs_flags       = FS_REQUIRES_DEV,
5740 };
5741 MODULE_ALIAS_FS("ext4");
5742
5743 /* Shared across all ext4 file systems */
5744 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5745
5746 static int __init ext4_init_fs(void)
5747 {
5748         int i, err;
5749
5750         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5751         ext4_li_info = NULL;
5752         mutex_init(&ext4_li_mtx);
5753
5754         /* Build-time check for flags consistency */
5755         ext4_check_flag_values();
5756
5757         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5758                 init_waitqueue_head(&ext4__ioend_wq[i]);
5759
5760         err = ext4_init_es();
5761         if (err)
5762                 return err;
5763
5764         err = ext4_init_pageio();
5765         if (err)
5766                 goto out5;
5767
5768         err = ext4_init_system_zone();
5769         if (err)
5770                 goto out4;
5771
5772         err = ext4_init_sysfs();
5773         if (err)
5774                 goto out3;
5775
5776         err = ext4_init_mballoc();
5777         if (err)
5778                 goto out2;
5779         err = init_inodecache();
5780         if (err)
5781                 goto out1;
5782         register_as_ext3();
5783         register_as_ext2();
5784         err = register_filesystem(&ext4_fs_type);
5785         if (err)
5786                 goto out;
5787
5788         return 0;
5789 out:
5790         unregister_as_ext2();
5791         unregister_as_ext3();
5792         destroy_inodecache();
5793 out1:
5794         ext4_exit_mballoc();
5795 out2:
5796         ext4_exit_sysfs();
5797 out3:
5798         ext4_exit_system_zone();
5799 out4:
5800         ext4_exit_pageio();
5801 out5:
5802         ext4_exit_es();
5803
5804         return err;
5805 }
5806
5807 static void __exit ext4_exit_fs(void)
5808 {
5809         ext4_destroy_lazyinit_thread();
5810         unregister_as_ext2();
5811         unregister_as_ext3();
5812         unregister_filesystem(&ext4_fs_type);
5813         destroy_inodecache();
5814         ext4_exit_mballoc();
5815         ext4_exit_sysfs();
5816         ext4_exit_system_zone();
5817         ext4_exit_pageio();
5818         ext4_exit_es();
5819 }
5820
5821 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5822 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5823 MODULE_LICENSE("GPL");
5824 module_init(ext4_init_fs)
5825 module_exit(ext4_exit_fs)