fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.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/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct proc_dir_entry *ext4_proc_root;
  57 static struct kset *ext4_kset;
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ext4_features *ext4_feat;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static const char *ext4_decode_error(struct super_block *sb, int errno,
  71                                      char nbuf[16]);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static void ext4_write_super(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  95 #else
  96 #define IS_EXT2_SB(sb) (0)
  97 #endif
  98
  99
 100 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 101 static struct file_system_type ext3_fs_type = {
 102         .owner          = THIS_MODULE,
 103         .name           = "ext3",
 104         .mount          = ext4_mount,
 105         .kill_sb        = kill_block_super,
 106         .fs_flags       = FS_REQUIRES_DEV,
 107 };
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109 #else
 110 #define IS_EXT3_SB(sb) (0)
 111 #endif
 112
 113 void *ext4_kvmalloc(size_t size, gfp_t flags)
 114 {
 115         void *ret;
 116
 117         ret = kmalloc(size, flags);
 118         if (!ret)
 119                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 120         return ret;
 121 }
 122
 123 void *ext4_kvzalloc(size_t size, gfp_t flags)
 124 {
 125         void *ret;
 126
 127         ret = kzalloc(size, flags);
 128         if (!ret)
 129                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 130         return ret;
 131 }
 132
 133 void ext4_kvfree(void *ptr)
 134 {
 135         if (is_vmalloc_addr(ptr))
 136                 vfree(ptr);
 137         else
 138                 kfree(ptr);
 139
 140 }
 141
 142 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 143                                struct ext4_group_desc *bg)
 144 {
 145         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 148 }
 149
 150 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 151                                struct ext4_group_desc *bg)
 152 {
 153         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 154                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 155                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 156 }
 157
 158 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 159                               struct ext4_group_desc *bg)
 160 {
 161         return le32_to_cpu(bg->bg_inode_table_lo) |
 162                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 163                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 164 }
 165
 166 __u32 ext4_free_blks_count(struct super_block *sb,
 167                               struct ext4_group_desc *bg)
 168 {
 169         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 170                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 171                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 172 }
 173
 174 __u32 ext4_free_inodes_count(struct super_block *sb,
 175                               struct ext4_group_desc *bg)
 176 {
 177         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 178                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 179                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 180 }
 181
 182 __u32 ext4_used_dirs_count(struct super_block *sb,
 183                               struct ext4_group_desc *bg)
 184 {
 185         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 186                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 187                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 188 }
 189
 190 __u32 ext4_itable_unused_count(struct super_block *sb,
 191                               struct ext4_group_desc *bg)
 192 {
 193         return le16_to_cpu(bg->bg_itable_unused_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 196 }
 197
 198 void ext4_block_bitmap_set(struct super_block *sb,
 199                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 200 {
 201         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 204 }
 205
 206 void ext4_inode_bitmap_set(struct super_block *sb,
 207                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 208 {
 209         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 210         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 211                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 212 }
 213
 214 void ext4_inode_table_set(struct super_block *sb,
 215                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 216 {
 217         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 218         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 219                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 220 }
 221
 222 void ext4_free_blks_set(struct super_block *sb,
 223                           struct ext4_group_desc *bg, __u32 count)
 224 {
 225         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 226         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 227                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 228 }
 229
 230 void ext4_free_inodes_set(struct super_block *sb,
 231                           struct ext4_group_desc *bg, __u32 count)
 232 {
 233         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 234         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 235                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 236 }
 237
 238 void ext4_used_dirs_set(struct super_block *sb,
 239                           struct ext4_group_desc *bg, __u32 count)
 240 {
 241         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 242         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 243                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 244 }
 245
 246 void ext4_itable_unused_set(struct super_block *sb,
 247                           struct ext4_group_desc *bg, __u32 count)
 248 {
 249         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 252 }
 253
 254
 255 /* Just increment the non-pointer handle value */
 256 static handle_t *ext4_get_nojournal(void)
 257 {
 258         handle_t *handle = current->journal_info;
 259         unsigned long ref_cnt = (unsigned long)handle;
 260
 261         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 262
 263         ref_cnt++;
 264         handle = (handle_t *)ref_cnt;
 265
 266         current->journal_info = handle;
 267         return handle;
 268 }
 269
 270
 271 /* Decrement the non-pointer handle value */
 272 static void ext4_put_nojournal(handle_t *handle)
 273 {
 274         unsigned long ref_cnt = (unsigned long)handle;
 275
 276         BUG_ON(ref_cnt == 0);
 277
 278         ref_cnt--;
 279         handle = (handle_t *)ref_cnt;
 280
 281         current->journal_info = handle;
 282 }
 283
 284 /*
 285  * Wrappers for jbd2_journal_start/end.
 286  *
 287  * The only special thing we need to do here is to make sure that all
 288  * journal_end calls result in the superblock being marked dirty, so
 289  * that sync() will call the filesystem's write_super callback if
 290  * appropriate.
 291  *
 292  * To avoid j_barrier hold in userspace when a user calls freeze(),
 293  * ext4 prevents a new handle from being started by s_frozen, which
 294  * is in an upper layer.
 295  */
 296 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 297 {
 298         journal_t *journal;
 299         handle_t  *handle;
 300
 301         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 302         if (sb->s_flags & MS_RDONLY)
 303                 return ERR_PTR(-EROFS);
 304
 305         journal = EXT4_SB(sb)->s_journal;
 306         handle = ext4_journal_current_handle();
 307
 308         /*
 309          * If a handle has been started, it should be allowed to
 310          * finish, otherwise deadlock could happen between freeze
 311          * and others(e.g. truncate) due to the restart of the
 312          * journal handle if the filesystem is forzen and active
 313          * handles are not stopped.
 314          */
 315         if (!handle)
 316                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 317
 318         if (!journal)
 319                 return ext4_get_nojournal();
 320         /*
 321          * Special case here: if the journal has aborted behind our
 322          * backs (eg. EIO in the commit thread), then we still need to
 323          * take the FS itself readonly cleanly.
 324          */
 325         if (is_journal_aborted(journal)) {
 326                 ext4_abort(sb, "Detected aborted journal");
 327                 return ERR_PTR(-EROFS);
 328         }
 329         return jbd2_journal_start(journal, nblocks);
 330 }
 331
 332 /*
 333  * The only special thing we need to do here is to make sure that all
 334  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 335  * that sync() will call the filesystem's write_super callback if
 336  * appropriate.
 337  */
 338 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 339 {
 340         struct super_block *sb;
 341         int err;
 342         int rc;
 343
 344         if (!ext4_handle_valid(handle)) {
 345                 ext4_put_nojournal(handle);
 346                 return 0;
 347         }
 348         sb = handle->h_transaction->t_journal->j_private;
 349         err = handle->h_err;
 350         rc = jbd2_journal_stop(handle);
 351
 352         if (!err)
 353                 err = rc;
 354         if (err)
 355                 __ext4_std_error(sb, where, line, err);
 356         return err;
 357 }
 358
 359 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 360                                const char *err_fn, struct buffer_head *bh,
 361                                handle_t *handle, int err)
 362 {
 363         char nbuf[16];
 364         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 365
 366         BUG_ON(!ext4_handle_valid(handle));
 367
 368         if (bh)
 369                 BUFFER_TRACE(bh, "abort");
 370
 371         if (!handle->h_err)
 372                 handle->h_err = err;
 373
 374         if (is_handle_aborted(handle))
 375                 return;
 376
 377         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 378                caller, line, errstr, err_fn);
 379
 380         jbd2_journal_abort_handle(handle);
 381 }
 382
 383 static void __save_error_info(struct super_block *sb, const char *func,
 384                             unsigned int line)
 385 {
 386         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 387
 388         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 389         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 390         es->s_last_error_time = cpu_to_le32(get_seconds());
 391         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 392         es->s_last_error_line = cpu_to_le32(line);
 393         if (!es->s_first_error_time) {
 394                 es->s_first_error_time = es->s_last_error_time;
 395                 strncpy(es->s_first_error_func, func,
 396                         sizeof(es->s_first_error_func));
 397                 es->s_first_error_line = cpu_to_le32(line);
 398                 es->s_first_error_ino = es->s_last_error_ino;
 399                 es->s_first_error_block = es->s_last_error_block;
 400         }
 401         /*
 402          * Start the daily error reporting function if it hasn't been
 403          * started already
 404          */
 405         if (!es->s_error_count)
 406                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 407         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 408 }
 409
 410 static void save_error_info(struct super_block *sb, const char *func,
 411                             unsigned int line)
 412 {
 413         __save_error_info(sb, func, line);
 414         ext4_commit_super(sb, 1);
 415 }
 416
 417
 418 /* Deal with the reporting of failure conditions on a filesystem such as
 419  * inconsistencies detected or read IO failures.
 420  *
 421  * On ext2, we can store the error state of the filesystem in the
 422  * superblock.  That is not possible on ext4, because we may have other
 423  * write ordering constraints on the superblock which prevent us from
 424  * writing it out straight away; and given that the journal is about to
 425  * be aborted, we can't rely on the current, or future, transactions to
 426  * write out the superblock safely.
 427  *
 428  * We'll just use the jbd2_journal_abort() error code to record an error in
 429  * the journal instead.  On recovery, the journal will complain about
 430  * that error until we've noted it down and cleared it.
 431  */
 432
 433 static void ext4_handle_error(struct super_block *sb)
 434 {
 435         if (sb->s_flags & MS_RDONLY)
 436                 return;
 437
 438         if (!test_opt(sb, ERRORS_CONT)) {
 439                 journal_t *journal = EXT4_SB(sb)->s_journal;
 440
 441                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 442                 if (journal)
 443                         jbd2_journal_abort(journal, -EIO);
 444         }
 445         if (test_opt(sb, ERRORS_RO)) {
 446                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 447                 sb->s_flags |= MS_RDONLY;
 448         }
 449         if (test_opt(sb, ERRORS_PANIC))
 450                 panic("EXT4-fs (device %s): panic forced after error\n",
 451                         sb->s_id);
 452 }
 453
 454 void __ext4_error(struct super_block *sb, const char *function,
 455                   unsigned int line, const char *fmt, ...)
 456 {
 457         struct va_format vaf;
 458         va_list args;
 459
 460         va_start(args, fmt);
 461         vaf.fmt = fmt;
 462         vaf.va = &args;
 463         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 464                sb->s_id, function, line, current->comm, &vaf);
 465         va_end(args);
 466
 467         ext4_handle_error(sb);
 468 }
 469
 470 void ext4_error_inode(struct inode *inode, const char *function,
 471                       unsigned int line, ext4_fsblk_t block,
 472                       const char *fmt, ...)
 473 {
 474         va_list args;
 475         struct va_format vaf;
 476         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 477
 478         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 479         es->s_last_error_block = cpu_to_le64(block);
 480         save_error_info(inode->i_sb, function, line);
 481         va_start(args, fmt);
 482         vaf.fmt = fmt;
 483         vaf.va = &args;
 484         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 485                inode->i_sb->s_id, function, line, inode->i_ino);
 486         if (block)
 487                 printk(KERN_CONT "block %llu: ", block);
 488         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 489         va_end(args);
 490
 491         ext4_handle_error(inode->i_sb);
 492 }
 493
 494 void ext4_error_file(struct file *file, const char *function,
 495                      unsigned int line, ext4_fsblk_t block,
 496                      const char *fmt, ...)
 497 {
 498         va_list args;
 499         struct va_format vaf;
 500         struct ext4_super_block *es;
 501         struct inode *inode = file->f_dentry->d_inode;
 502         char pathname[80], *path;
 503
 504         es = EXT4_SB(inode->i_sb)->s_es;
 505         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 506         save_error_info(inode->i_sb, function, line);
 507         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 508         if (IS_ERR(path))
 509                 path = "(unknown)";
 510         printk(KERN_CRIT
 511                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 512                inode->i_sb->s_id, function, line, inode->i_ino);
 513         if (block)
 514                 printk(KERN_CONT "block %llu: ", block);
 515         va_start(args, fmt);
 516         vaf.fmt = fmt;
 517         vaf.va = &args;
 518         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 519         va_end(args);
 520
 521         ext4_handle_error(inode->i_sb);
 522 }
 523
 524 static const char *ext4_decode_error(struct super_block *sb, int errno,
 525                                      char nbuf[16])
 526 {
 527         char *errstr = NULL;
 528
 529         switch (errno) {
 530         case -EIO:
 531                 errstr = "IO failure";
 532                 break;
 533         case -ENOMEM:
 534                 errstr = "Out of memory";
 535                 break;
 536         case -EROFS:
 537                 if (!sb || (EXT4_SB(sb)->s_journal &&
 538                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 539                         errstr = "Journal has aborted";
 540                 else
 541                         errstr = "Readonly filesystem";
 542                 break;
 543         default:
 544                 /* If the caller passed in an extra buffer for unknown
 545                  * errors, textualise them now.  Else we just return
 546                  * NULL. */
 547                 if (nbuf) {
 548                         /* Check for truncated error codes... */
 549                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 550                                 errstr = nbuf;
 551                 }
 552                 break;
 553         }
 554
 555         return errstr;
 556 }
 557
 558 /* __ext4_std_error decodes expected errors from journaling functions
 559  * automatically and invokes the appropriate error response.  */
 560
 561 void __ext4_std_error(struct super_block *sb, const char *function,
 562                       unsigned int line, int errno)
 563 {
 564         char nbuf[16];
 565         const char *errstr;
 566
 567         /* Special case: if the error is EROFS, and we're not already
 568          * inside a transaction, then there's really no point in logging
 569          * an error. */
 570         if (errno == -EROFS && journal_current_handle() == NULL &&
 571             (sb->s_flags & MS_RDONLY))
 572                 return;
 573
 574         errstr = ext4_decode_error(sb, errno, nbuf);
 575         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 576                sb->s_id, function, line, errstr);
 577         save_error_info(sb, function, line);
 578
 579         ext4_handle_error(sb);
 580 }
 581
 582 /*
 583  * ext4_abort is a much stronger failure handler than ext4_error.  The
 584  * abort function may be used to deal with unrecoverable failures such
 585  * as journal IO errors or ENOMEM at a critical moment in log management.
 586  *
 587  * We unconditionally force the filesystem into an ABORT|READONLY state,
 588  * unless the error response on the fs has been set to panic in which
 589  * case we take the easy way out and panic immediately.
 590  */
 591
 592 void __ext4_abort(struct super_block *sb, const char *function,
 593                 unsigned int line, const char *fmt, ...)
 594 {
 595         va_list args;
 596
 597         save_error_info(sb, function, line);
 598         va_start(args, fmt);
 599         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 600                function, line);
 601         vprintk(fmt, args);
 602         printk("\n");
 603         va_end(args);
 604
 605         if ((sb->s_flags & MS_RDONLY) == 0) {
 606                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 607                 sb->s_flags |= MS_RDONLY;
 608                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 609                 if (EXT4_SB(sb)->s_journal)
 610                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 611                 save_error_info(sb, function, line);
 612         }
 613         if (test_opt(sb, ERRORS_PANIC))
 614                 panic("EXT4-fs panic from previous error\n");
 615 }
 616
 617 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 618 {
 619         struct va_format vaf;
 620         va_list args;
 621
 622         va_start(args, fmt);
 623         vaf.fmt = fmt;
 624         vaf.va = &args;
 625         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 626         va_end(args);
 627 }
 628
 629 void __ext4_warning(struct super_block *sb, const char *function,
 630                     unsigned int line, const char *fmt, ...)
 631 {
 632         struct va_format vaf;
 633         va_list args;
 634
 635         va_start(args, fmt);
 636         vaf.fmt = fmt;
 637         vaf.va = &args;
 638         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 639                sb->s_id, function, line, &vaf);
 640         va_end(args);
 641 }
 642
 643 void __ext4_grp_locked_error(const char *function, unsigned int line,
 644                              struct super_block *sb, ext4_group_t grp,
 645                              unsigned long ino, ext4_fsblk_t block,
 646                              const char *fmt, ...)
 647 __releases(bitlock)
 648 __acquires(bitlock)
 649 {
 650         struct va_format vaf;
 651         va_list args;
 652         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 653
 654         es->s_last_error_ino = cpu_to_le32(ino);
 655         es->s_last_error_block = cpu_to_le64(block);
 656         __save_error_info(sb, function, line);
 657
 658         va_start(args, fmt);
 659
 660         vaf.fmt = fmt;
 661         vaf.va = &args;
 662         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 663                sb->s_id, function, line, grp);
 664         if (ino)
 665                 printk(KERN_CONT "inode %lu: ", ino);
 666         if (block)
 667                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 668         printk(KERN_CONT "%pV\n", &vaf);
 669         va_end(args);
 670
 671         if (test_opt(sb, ERRORS_CONT)) {
 672                 ext4_commit_super(sb, 0);
 673                 return;
 674         }
 675
 676         ext4_unlock_group(sb, grp);
 677         ext4_handle_error(sb);
 678         /*
 679          * We only get here in the ERRORS_RO case; relocking the group
 680          * may be dangerous, but nothing bad will happen since the
 681          * filesystem will have already been marked read/only and the
 682          * journal has been aborted.  We return 1 as a hint to callers
 683          * who might what to use the return value from
 684          * ext4_grp_locked_error() to distinguish between the
 685          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 686          * aggressively from the ext4 function in question, with a
 687          * more appropriate error code.
 688          */
 689         ext4_lock_group(sb, grp);
 690         return;
 691 }
 692
 693 void ext4_update_dynamic_rev(struct super_block *sb)
 694 {
 695         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 696
 697         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 698                 return;
 699
 700         ext4_warning(sb,
 701                      "updating to rev %d because of new feature flag, "
 702                      "running e2fsck is recommended",
 703                      EXT4_DYNAMIC_REV);
 704
 705         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 706         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 707         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 708         /* leave es->s_feature_*compat flags alone */
 709         /* es->s_uuid will be set by e2fsck if empty */
 710
 711         /*
 712          * The rest of the superblock fields should be zero, and if not it
 713          * means they are likely already in use, so leave them alone.  We
 714          * can leave it up to e2fsck to clean up any inconsistencies there.
 715          */
 716 }
 717
 718 /*
 719  * Open the external journal device
 720  */
 721 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 722 {
 723         struct block_device *bdev;
 724         char b[BDEVNAME_SIZE];
 725
 726         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 727         if (IS_ERR(bdev))
 728                 goto fail;
 729         return bdev;
 730
 731 fail:
 732         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 733                         __bdevname(dev, b), PTR_ERR(bdev));
 734         return NULL;
 735 }
 736
 737 /*
 738  * Release the journal device
 739  */
 740 static int ext4_blkdev_put(struct block_device *bdev)
 741 {
 742         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 743 }
 744
 745 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 746 {
 747         struct block_device *bdev;
 748         int ret = -ENODEV;
 749
 750         bdev = sbi->journal_bdev;
 751         if (bdev) {
 752                 ret = ext4_blkdev_put(bdev);
 753                 sbi->journal_bdev = NULL;
 754         }
 755         return ret;
 756 }
 757
 758 static inline struct inode *orphan_list_entry(struct list_head *l)
 759 {
 760         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 761 }
 762
 763 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 764 {
 765         struct list_head *l;
 766
 767         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 768                  le32_to_cpu(sbi->s_es->s_last_orphan));
 769
 770         printk(KERN_ERR "sb_info orphan list:\n");
 771         list_for_each(l, &sbi->s_orphan) {
 772                 struct inode *inode = orphan_list_entry(l);
 773                 printk(KERN_ERR "  "
 774                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 775                        inode->i_sb->s_id, inode->i_ino, inode,
 776                        inode->i_mode, inode->i_nlink,
 777                        NEXT_ORPHAN(inode));
 778         }
 779 }
 780
 781 static void ext4_put_super(struct super_block *sb)
 782 {
 783         struct ext4_sb_info *sbi = EXT4_SB(sb);
 784         struct ext4_super_block *es = sbi->s_es;
 785         int i, err;
 786
 787         ext4_unregister_li_request(sb);
 788         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 789
 790         flush_workqueue(sbi->dio_unwritten_wq);
 791         destroy_workqueue(sbi->dio_unwritten_wq);
 792
 793         lock_super(sb);
 794         if (sb->s_dirt)
 795                 ext4_commit_super(sb, 1);
 796
 797         if (sbi->s_journal) {
 798                 err = jbd2_journal_destroy(sbi->s_journal);
 799                 sbi->s_journal = NULL;
 800                 if (err < 0)
 801                         ext4_abort(sb, "Couldn't clean up the journal");
 802         }
 803
 804         del_timer(&sbi->s_err_report);
 805         ext4_release_system_zone(sb);
 806         ext4_mb_release(sb);
 807         ext4_ext_release(sb);
 808         ext4_xattr_put_super(sb);
 809
 810         if (!(sb->s_flags & MS_RDONLY)) {
 811                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 812                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 813                 ext4_commit_super(sb, 1);
 814         }
 815         if (sbi->s_proc) {
 816                 remove_proc_entry(sb->s_id, ext4_proc_root);
 817         }
 818         kobject_del(&sbi->s_kobj);
 819
 820         for (i = 0; i < sbi->s_gdb_count; i++)
 821                 brelse(sbi->s_group_desc[i]);
 822         ext4_kvfree(sbi->s_group_desc);
 823         ext4_kvfree(sbi->s_flex_groups);
 824         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 825         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 826         percpu_counter_destroy(&sbi->s_dirs_counter);
 827         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 828         brelse(sbi->s_sbh);
 829 #ifdef CONFIG_QUOTA
 830         for (i = 0; i < MAXQUOTAS; i++)
 831                 kfree(sbi->s_qf_names[i]);
 832 #endif
 833
 834         /* Debugging code just in case the in-memory inode orphan list
 835          * isn't empty.  The on-disk one can be non-empty if we've
 836          * detected an error and taken the fs readonly, but the
 837          * in-memory list had better be clean by this point. */
 838         if (!list_empty(&sbi->s_orphan))
 839                 dump_orphan_list(sb, sbi);
 840         J_ASSERT(list_empty(&sbi->s_orphan));
 841
 842         invalidate_bdev(sb->s_bdev);
 843         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 844                 /*
 845                  * Invalidate the journal device's buffers.  We don't want them
 846                  * floating about in memory - the physical journal device may
 847                  * hotswapped, and it breaks the `ro-after' testing code.
 848                  */
 849                 sync_blockdev(sbi->journal_bdev);
 850                 invalidate_bdev(sbi->journal_bdev);
 851                 ext4_blkdev_remove(sbi);
 852         }
 853         if (sbi->s_mmp_tsk)
 854                 kthread_stop(sbi->s_mmp_tsk);
 855         sb->s_fs_info = NULL;
 856         /*
 857          * Now that we are completely done shutting down the
 858          * superblock, we need to actually destroy the kobject.
 859          */
 860         unlock_super(sb);
 861         kobject_put(&sbi->s_kobj);
 862         wait_for_completion(&sbi->s_kobj_unregister);
 863         kfree(sbi->s_blockgroup_lock);
 864         kfree(sbi);
 865 }
 866
 867 static struct kmem_cache *ext4_inode_cachep;
 868
 869 /*
 870  * Called inside transaction, so use GFP_NOFS
 871  */
 872 static struct inode *ext4_alloc_inode(struct super_block *sb)
 873 {
 874         struct ext4_inode_info *ei;
 875
 876         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 877         if (!ei)
 878                 return NULL;
 879
 880         ei->vfs_inode.i_version = 1;
 881         ei->vfs_inode.i_data.writeback_index = 0;
 882         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 883         INIT_LIST_HEAD(&ei->i_prealloc_list);
 884         spin_lock_init(&ei->i_prealloc_lock);
 885         ei->i_reserved_data_blocks = 0;
 886         ei->i_reserved_meta_blocks = 0;
 887         ei->i_allocated_meta_blocks = 0;
 888         ei->i_da_metadata_calc_len = 0;
 889         spin_lock_init(&(ei->i_block_reservation_lock));
 890 #ifdef CONFIG_QUOTA
 891         ei->i_reserved_quota = 0;
 892 #endif
 893         ei->jinode = NULL;
 894         INIT_LIST_HEAD(&ei->i_completed_io_list);
 895         spin_lock_init(&ei->i_completed_io_lock);
 896         ei->cur_aio_dio = NULL;
 897         ei->i_sync_tid = 0;
 898         ei->i_datasync_tid = 0;
 899         atomic_set(&ei->i_ioend_count, 0);
 900         atomic_set(&ei->i_aiodio_unwritten, 0);
 901
 902         return &ei->vfs_inode;
 903 }
 904
 905 static int ext4_drop_inode(struct inode *inode)
 906 {
 907         int drop = generic_drop_inode(inode);
 908
 909         trace_ext4_drop_inode(inode, drop);
 910         return drop;
 911 }
 912
 913 static void ext4_i_callback(struct rcu_head *head)
 914 {
 915         struct inode *inode = container_of(head, struct inode, i_rcu);
 916         INIT_LIST_HEAD(&inode->i_dentry);
 917         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 918 }
 919
 920 static void ext4_destroy_inode(struct inode *inode)
 921 {
 922         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 923                 ext4_msg(inode->i_sb, KERN_ERR,
 924                          "Inode %lu (%p): orphan list check failed!",
 925                          inode->i_ino, EXT4_I(inode));
 926                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 927                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 928                                 true);
 929                 dump_stack();
 930         }
 931         call_rcu(&inode->i_rcu, ext4_i_callback);
 932 }
 933
 934 static void init_once(void *foo)
 935 {
 936         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 937
 938         INIT_LIST_HEAD(&ei->i_orphan);
 939 #ifdef CONFIG_EXT4_FS_XATTR
 940         init_rwsem(&ei->xattr_sem);
 941 #endif
 942         init_rwsem(&ei->i_data_sem);
 943         inode_init_once(&ei->vfs_inode);
 944 }
 945
 946 static int init_inodecache(void)
 947 {
 948         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 949                                              sizeof(struct ext4_inode_info),
 950                                              0, (SLAB_RECLAIM_ACCOUNT|
 951                                                 SLAB_MEM_SPREAD),
 952                                              init_once);
 953         if (ext4_inode_cachep == NULL)
 954                 return -ENOMEM;
 955         return 0;
 956 }
 957
 958 static void destroy_inodecache(void)
 959 {
 960         kmem_cache_destroy(ext4_inode_cachep);
 961 }
 962
 963 void ext4_clear_inode(struct inode *inode)
 964 {
 965         invalidate_inode_buffers(inode);
 966         end_writeback(inode);
 967         dquot_drop(inode);
 968         ext4_discard_preallocations(inode);
 969         if (EXT4_I(inode)->jinode) {
 970                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 971                                                EXT4_I(inode)->jinode);
 972                 jbd2_free_inode(EXT4_I(inode)->jinode);
 973                 EXT4_I(inode)->jinode = NULL;
 974         }
 975 }
 976
 977 static inline void ext4_show_quota_options(struct seq_file *seq,
 978                                            struct super_block *sb)
 979 {
 980 #if defined(CONFIG_QUOTA)
 981         struct ext4_sb_info *sbi = EXT4_SB(sb);
 982
 983         if (sbi->s_jquota_fmt) {
 984                 char *fmtname = "";
 985
 986                 switch (sbi->s_jquota_fmt) {
 987                 case QFMT_VFS_OLD:
 988                         fmtname = "vfsold";
 989                         break;
 990                 case QFMT_VFS_V0:
 991                         fmtname = "vfsv0";
 992                         break;
 993                 case QFMT_VFS_V1:
 994                         fmtname = "vfsv1";
 995                         break;
 996                 }
 997                 seq_printf(seq, ",jqfmt=%s", fmtname);
 998         }
 999
1000         if (sbi->s_qf_names[USRQUOTA])
1001                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1002
1003         if (sbi->s_qf_names[GRPQUOTA])
1004                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1005
1006         if (test_opt(sb, USRQUOTA))
1007                 seq_puts(seq, ",usrquota");
1008
1009         if (test_opt(sb, GRPQUOTA))
1010                 seq_puts(seq, ",grpquota");
1011 #endif
1012 }
1013
1014 /*
1015  * Show an option if
1016  *  - it's set to a non-default value OR
1017  *  - if the per-sb default is different from the global default
1018  */
1019 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1020 {
1021         int def_errors;
1022         unsigned long def_mount_opts;
1023         struct super_block *sb = vfs->mnt_sb;
1024         struct ext4_sb_info *sbi = EXT4_SB(sb);
1025         struct ext4_super_block *es = sbi->s_es;
1026
1027         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1028         def_errors     = le16_to_cpu(es->s_errors);
1029
1030         if (sbi->s_sb_block != 1)
1031                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1032         if (test_opt(sb, MINIX_DF))
1033                 seq_puts(seq, ",minixdf");
1034         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1035                 seq_puts(seq, ",grpid");
1036         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1037                 seq_puts(seq, ",nogrpid");
1038         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1039             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1040                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1041         }
1042         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1043             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1044                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1045         }
1046         if (test_opt(sb, ERRORS_RO)) {
1047                 if (def_errors == EXT4_ERRORS_PANIC ||
1048                     def_errors == EXT4_ERRORS_CONTINUE) {
1049                         seq_puts(seq, ",errors=remount-ro");
1050                 }
1051         }
1052         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1053                 seq_puts(seq, ",errors=continue");
1054         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1055                 seq_puts(seq, ",errors=panic");
1056         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1057                 seq_puts(seq, ",nouid32");
1058         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1059                 seq_puts(seq, ",debug");
1060         if (test_opt(sb, OLDALLOC))
1061                 seq_puts(seq, ",oldalloc");
1062 #ifdef CONFIG_EXT4_FS_XATTR
1063         if (test_opt(sb, XATTR_USER))
1064                 seq_puts(seq, ",user_xattr");
1065         if (!test_opt(sb, XATTR_USER))
1066                 seq_puts(seq, ",nouser_xattr");
1067 #endif
1068 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1069         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1070                 seq_puts(seq, ",acl");
1071         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1072                 seq_puts(seq, ",noacl");
1073 #endif
1074         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1075                 seq_printf(seq, ",commit=%u",
1076                            (unsigned) (sbi->s_commit_interval / HZ));
1077         }
1078         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1079                 seq_printf(seq, ",min_batch_time=%u",
1080                            (unsigned) sbi->s_min_batch_time);
1081         }
1082         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1083                 seq_printf(seq, ",max_batch_time=%u",
1084                            (unsigned) sbi->s_min_batch_time);
1085         }
1086
1087         /*
1088          * We're changing the default of barrier mount option, so
1089          * let's always display its mount state so it's clear what its
1090          * status is.
1091          */
1092         seq_puts(seq, ",barrier=");
1093         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1094         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1095                 seq_puts(seq, ",journal_async_commit");
1096         else if (test_opt(sb, JOURNAL_CHECKSUM))
1097                 seq_puts(seq, ",journal_checksum");
1098         if (test_opt(sb, I_VERSION))
1099                 seq_puts(seq, ",i_version");
1100         if (!test_opt(sb, DELALLOC) &&
1101             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1102                 seq_puts(seq, ",nodelalloc");
1103
1104         if (!test_opt(sb, MBLK_IO_SUBMIT))
1105                 seq_puts(seq, ",nomblk_io_submit");
1106         if (sbi->s_stripe)
1107                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1108         /*
1109          * journal mode get enabled in different ways
1110          * So just print the value even if we didn't specify it
1111          */
1112         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1113                 seq_puts(seq, ",data=journal");
1114         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1115                 seq_puts(seq, ",data=ordered");
1116         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1117                 seq_puts(seq, ",data=writeback");
1118
1119         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1120                 seq_printf(seq, ",inode_readahead_blks=%u",
1121                            sbi->s_inode_readahead_blks);
1122
1123         if (test_opt(sb, DATA_ERR_ABORT))
1124                 seq_puts(seq, ",data_err=abort");
1125
1126         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1127                 seq_puts(seq, ",noauto_da_alloc");
1128
1129         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1130                 seq_puts(seq, ",discard");
1131
1132         if (test_opt(sb, NOLOAD))
1133                 seq_puts(seq, ",norecovery");
1134
1135         if (test_opt(sb, DIOREAD_NOLOCK))
1136                 seq_puts(seq, ",dioread_nolock");
1137
1138         if (test_opt(sb, BLOCK_VALIDITY) &&
1139             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1140                 seq_puts(seq, ",block_validity");
1141
1142         if (!test_opt(sb, INIT_INODE_TABLE))
1143                 seq_puts(seq, ",noinit_inode_table");
1144         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1145                 seq_printf(seq, ",init_inode_table=%u",
1146                            (unsigned) sbi->s_li_wait_mult);
1147
1148         ext4_show_quota_options(seq, sb);
1149
1150         return 0;
1151 }
1152
1153 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1154                                         u64 ino, u32 generation)
1155 {
1156         struct inode *inode;
1157
1158         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1159                 return ERR_PTR(-ESTALE);
1160         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1161                 return ERR_PTR(-ESTALE);
1162
1163         /* iget isn't really right if the inode is currently unallocated!!
1164          *
1165          * ext4_read_inode will return a bad_inode if the inode had been
1166          * deleted, so we should be safe.
1167          *
1168          * Currently we don't know the generation for parent directory, so
1169          * a generation of 0 means "accept any"
1170          */
1171         inode = ext4_iget(sb, ino);
1172         if (IS_ERR(inode))
1173                 return ERR_CAST(inode);
1174         if (generation && inode->i_generation != generation) {
1175                 iput(inode);
1176                 return ERR_PTR(-ESTALE);
1177         }
1178
1179         return inode;
1180 }
1181
1182 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1183                                         int fh_len, int fh_type)
1184 {
1185         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1186                                     ext4_nfs_get_inode);
1187 }
1188
1189 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1190                                         int fh_len, int fh_type)
1191 {
1192         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1193                                     ext4_nfs_get_inode);
1194 }
1195
1196 /*
1197  * Try to release metadata pages (indirect blocks, directories) which are
1198  * mapped via the block device.  Since these pages could have journal heads
1199  * which would prevent try_to_free_buffers() from freeing them, we must use
1200  * jbd2 layer's try_to_free_buffers() function to release them.
1201  */
1202 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1203                                  gfp_t wait)
1204 {
1205         journal_t *journal = EXT4_SB(sb)->s_journal;
1206
1207         WARN_ON(PageChecked(page));
1208         if (!page_has_buffers(page))
1209                 return 0;
1210         if (journal)
1211                 return jbd2_journal_try_to_free_buffers(journal, page,
1212                                                         wait & ~__GFP_WAIT);
1213         return try_to_free_buffers(page);
1214 }
1215
1216 #ifdef CONFIG_QUOTA
1217 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1218 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1219
1220 static int ext4_write_dquot(struct dquot *dquot);
1221 static int ext4_acquire_dquot(struct dquot *dquot);
1222 static int ext4_release_dquot(struct dquot *dquot);
1223 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1224 static int ext4_write_info(struct super_block *sb, int type);
1225 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1226                          struct path *path);
1227 static int ext4_quota_off(struct super_block *sb, int type);
1228 static int ext4_quota_on_mount(struct super_block *sb, int type);
1229 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1230                                size_t len, loff_t off);
1231 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1232                                 const char *data, size_t len, loff_t off);
1233
1234 static const struct dquot_operations ext4_quota_operations = {
1235         .get_reserved_space = ext4_get_reserved_space,
1236         .write_dquot    = ext4_write_dquot,
1237         .acquire_dquot  = ext4_acquire_dquot,
1238         .release_dquot  = ext4_release_dquot,
1239         .mark_dirty     = ext4_mark_dquot_dirty,
1240         .write_info     = ext4_write_info,
1241         .alloc_dquot    = dquot_alloc,
1242         .destroy_dquot  = dquot_destroy,
1243 };
1244
1245 static const struct quotactl_ops ext4_qctl_operations = {
1246         .quota_on       = ext4_quota_on,
1247         .quota_off      = ext4_quota_off,
1248         .quota_sync     = dquot_quota_sync,
1249         .get_info       = dquot_get_dqinfo,
1250         .set_info       = dquot_set_dqinfo,
1251         .get_dqblk      = dquot_get_dqblk,
1252         .set_dqblk      = dquot_set_dqblk
1253 };
1254 #endif
1255
1256 static const struct super_operations ext4_sops = {
1257         .alloc_inode    = ext4_alloc_inode,
1258         .destroy_inode  = ext4_destroy_inode,
1259         .write_inode    = ext4_write_inode,
1260         .dirty_inode    = ext4_dirty_inode,
1261         .drop_inode     = ext4_drop_inode,
1262         .evict_inode    = ext4_evict_inode,
1263         .put_super      = ext4_put_super,
1264         .sync_fs        = ext4_sync_fs,
1265         .freeze_fs      = ext4_freeze,
1266         .unfreeze_fs    = ext4_unfreeze,
1267         .statfs         = ext4_statfs,
1268         .remount_fs     = ext4_remount,
1269         .show_options   = ext4_show_options,
1270 #ifdef CONFIG_QUOTA
1271         .quota_read     = ext4_quota_read,
1272         .quota_write    = ext4_quota_write,
1273 #endif
1274         .bdev_try_to_free_page = bdev_try_to_free_page,
1275 };
1276
1277 static const struct super_operations ext4_nojournal_sops = {
1278         .alloc_inode    = ext4_alloc_inode,
1279         .destroy_inode  = ext4_destroy_inode,
1280         .write_inode    = ext4_write_inode,
1281         .dirty_inode    = ext4_dirty_inode,
1282         .drop_inode     = ext4_drop_inode,
1283         .evict_inode    = ext4_evict_inode,
1284         .write_super    = ext4_write_super,
1285         .put_super      = ext4_put_super,
1286         .statfs         = ext4_statfs,
1287         .remount_fs     = ext4_remount,
1288         .show_options   = ext4_show_options,
1289 #ifdef CONFIG_QUOTA
1290         .quota_read     = ext4_quota_read,
1291         .quota_write    = ext4_quota_write,
1292 #endif
1293         .bdev_try_to_free_page = bdev_try_to_free_page,
1294 };
1295
1296 static const struct export_operations ext4_export_ops = {
1297         .fh_to_dentry = ext4_fh_to_dentry,
1298         .fh_to_parent = ext4_fh_to_parent,
1299         .get_parent = ext4_get_parent,
1300 };
1301
1302 enum {
1303         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1304         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1305         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1306         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1307         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1308         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1309         Opt_journal_update, Opt_journal_dev,
1310         Opt_journal_checksum, Opt_journal_async_commit,
1311         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1312         Opt_data_err_abort, Opt_data_err_ignore,
1313         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1314         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1315         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1316         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1317         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1318         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1319         Opt_inode_readahead_blks, Opt_journal_ioprio,
1320         Opt_dioread_nolock, Opt_dioread_lock,
1321         Opt_discard, Opt_nodiscard,
1322         Opt_init_inode_table, Opt_noinit_inode_table,
1323 };
1324
1325 static const match_table_t tokens = {
1326         {Opt_bsd_df, "bsddf"},
1327         {Opt_minix_df, "minixdf"},
1328         {Opt_grpid, "grpid"},
1329         {Opt_grpid, "bsdgroups"},
1330         {Opt_nogrpid, "nogrpid"},
1331         {Opt_nogrpid, "sysvgroups"},
1332         {Opt_resgid, "resgid=%u"},
1333         {Opt_resuid, "resuid=%u"},
1334         {Opt_sb, "sb=%u"},
1335         {Opt_err_cont, "errors=continue"},
1336         {Opt_err_panic, "errors=panic"},
1337         {Opt_err_ro, "errors=remount-ro"},
1338         {Opt_nouid32, "nouid32"},
1339         {Opt_debug, "debug"},
1340         {Opt_oldalloc, "oldalloc"},
1341         {Opt_orlov, "orlov"},
1342         {Opt_user_xattr, "user_xattr"},
1343         {Opt_nouser_xattr, "nouser_xattr"},
1344         {Opt_acl, "acl"},
1345         {Opt_noacl, "noacl"},
1346         {Opt_noload, "noload"},
1347         {Opt_noload, "norecovery"},
1348         {Opt_nobh, "nobh"},
1349         {Opt_bh, "bh"},
1350         {Opt_commit, "commit=%u"},
1351         {Opt_min_batch_time, "min_batch_time=%u"},
1352         {Opt_max_batch_time, "max_batch_time=%u"},
1353         {Opt_journal_update, "journal=update"},
1354         {Opt_journal_dev, "journal_dev=%u"},
1355         {Opt_journal_checksum, "journal_checksum"},
1356         {Opt_journal_async_commit, "journal_async_commit"},
1357         {Opt_abort, "abort"},
1358         {Opt_data_journal, "data=journal"},
1359         {Opt_data_ordered, "data=ordered"},
1360         {Opt_data_writeback, "data=writeback"},
1361         {Opt_data_err_abort, "data_err=abort"},
1362         {Opt_data_err_ignore, "data_err=ignore"},
1363         {Opt_offusrjquota, "usrjquota="},
1364         {Opt_usrjquota, "usrjquota=%s"},
1365         {Opt_offgrpjquota, "grpjquota="},
1366         {Opt_grpjquota, "grpjquota=%s"},
1367         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1368         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1369         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1370         {Opt_grpquota, "grpquota"},
1371         {Opt_noquota, "noquota"},
1372         {Opt_quota, "quota"},
1373         {Opt_usrquota, "usrquota"},
1374         {Opt_barrier, "barrier=%u"},
1375         {Opt_barrier, "barrier"},
1376         {Opt_nobarrier, "nobarrier"},
1377         {Opt_i_version, "i_version"},
1378         {Opt_stripe, "stripe=%u"},
1379         {Opt_resize, "resize"},
1380         {Opt_delalloc, "delalloc"},
1381         {Opt_nodelalloc, "nodelalloc"},
1382         {Opt_mblk_io_submit, "mblk_io_submit"},
1383         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1384         {Opt_block_validity, "block_validity"},
1385         {Opt_noblock_validity, "noblock_validity"},
1386         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1387         {Opt_journal_ioprio, "journal_ioprio=%u"},
1388         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1389         {Opt_auto_da_alloc, "auto_da_alloc"},
1390         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1391         {Opt_dioread_nolock, "dioread_nolock"},
1392         {Opt_dioread_lock, "dioread_lock"},
1393         {Opt_discard, "discard"},
1394         {Opt_nodiscard, "nodiscard"},
1395         {Opt_init_inode_table, "init_itable=%u"},
1396         {Opt_init_inode_table, "init_itable"},
1397         {Opt_noinit_inode_table, "noinit_itable"},
1398         {Opt_err, NULL},
1399 };
1400
1401 static ext4_fsblk_t get_sb_block(void **data)
1402 {
1403         ext4_fsblk_t    sb_block;
1404         char            *options = (char *) *data;
1405
1406         if (!options || strncmp(options, "sb=", 3) != 0)
1407                 return 1;       /* Default location */
1408
1409         options += 3;
1410         /* TODO: use simple_strtoll with >32bit ext4 */
1411         sb_block = simple_strtoul(options, &options, 0);
1412         if (*options && *options != ',') {
1413                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1414                        (char *) *data);
1415                 return 1;
1416         }
1417         if (*options == ',')
1418                 options++;
1419         *data = (void *) options;
1420
1421         return sb_block;
1422 }
1423
1424 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1425 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1426         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1427
1428 #ifdef CONFIG_QUOTA
1429 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1430 {
1431         struct ext4_sb_info *sbi = EXT4_SB(sb);
1432         char *qname;
1433
1434         if (sb_any_quota_loaded(sb) &&
1435                 !sbi->s_qf_names[qtype]) {
1436                 ext4_msg(sb, KERN_ERR,
1437                         "Cannot change journaled "
1438                         "quota options when quota turned on");
1439                 return 0;
1440         }
1441         qname = match_strdup(args);
1442         if (!qname) {
1443                 ext4_msg(sb, KERN_ERR,
1444                         "Not enough memory for storing quotafile name");
1445                 return 0;
1446         }
1447         if (sbi->s_qf_names[qtype] &&
1448                 strcmp(sbi->s_qf_names[qtype], qname)) {
1449                 ext4_msg(sb, KERN_ERR,
1450                         "%s quota file already specified", QTYPE2NAME(qtype));
1451                 kfree(qname);
1452                 return 0;
1453         }
1454         sbi->s_qf_names[qtype] = qname;
1455         if (strchr(sbi->s_qf_names[qtype], '/')) {
1456                 ext4_msg(sb, KERN_ERR,
1457                         "quotafile must be on filesystem root");
1458                 kfree(sbi->s_qf_names[qtype]);
1459                 sbi->s_qf_names[qtype] = NULL;
1460                 return 0;
1461         }
1462         set_opt(sb, QUOTA);
1463         return 1;
1464 }
1465
1466 static int clear_qf_name(struct super_block *sb, int qtype)
1467 {
1468
1469         struct ext4_sb_info *sbi = EXT4_SB(sb);
1470
1471         if (sb_any_quota_loaded(sb) &&
1472                 sbi->s_qf_names[qtype]) {
1473                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1474                         " when quota turned on");
1475                 return 0;
1476         }
1477         /*
1478          * The space will be released later when all options are confirmed
1479          * to be correct
1480          */
1481         sbi->s_qf_names[qtype] = NULL;
1482         return 1;
1483 }
1484 #endif
1485
1486 static int parse_options(char *options, struct super_block *sb,
1487                          unsigned long *journal_devnum,
1488                          unsigned int *journal_ioprio,
1489                          ext4_fsblk_t *n_blocks_count, int is_remount)
1490 {
1491         struct ext4_sb_info *sbi = EXT4_SB(sb);
1492         char *p;
1493         substring_t args[MAX_OPT_ARGS];
1494         int data_opt = 0;
1495         int option;
1496 #ifdef CONFIG_QUOTA
1497         int qfmt;
1498 #endif
1499
1500         if (!options)
1501                 return 1;
1502
1503         while ((p = strsep(&options, ",")) != NULL) {
1504                 int token;
1505                 if (!*p)
1506                         continue;
1507
1508                 /*
1509                  * Initialize args struct so we know whether arg was
1510                  * found; some options take optional arguments.
1511                  */
1512                 args[0].to = args[0].from = NULL;
1513                 token = match_token(p, tokens, args);
1514                 switch (token) {
1515                 case Opt_bsd_df:
1516                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1517                         clear_opt(sb, MINIX_DF);
1518                         break;
1519                 case Opt_minix_df:
1520                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1521                         set_opt(sb, MINIX_DF);
1522
1523                         break;
1524                 case Opt_grpid:
1525                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1526                         set_opt(sb, GRPID);
1527
1528                         break;
1529                 case Opt_nogrpid:
1530                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1531                         clear_opt(sb, GRPID);
1532
1533                         break;
1534                 case Opt_resuid:
1535                         if (match_int(&args[0], &option))
1536                                 return 0;
1537                         sbi->s_resuid = option;
1538                         break;
1539                 case Opt_resgid:
1540                         if (match_int(&args[0], &option))
1541                                 return 0;
1542                         sbi->s_resgid = option;
1543                         break;
1544                 case Opt_sb:
1545                         /* handled by get_sb_block() instead of here */
1546                         /* *sb_block = match_int(&args[0]); */
1547                         break;
1548                 case Opt_err_panic:
1549                         clear_opt(sb, ERRORS_CONT);
1550                         clear_opt(sb, ERRORS_RO);
1551                         set_opt(sb, ERRORS_PANIC);
1552                         break;
1553                 case Opt_err_ro:
1554                         clear_opt(sb, ERRORS_CONT);
1555                         clear_opt(sb, ERRORS_PANIC);
1556                         set_opt(sb, ERRORS_RO);
1557                         break;
1558                 case Opt_err_cont:
1559                         clear_opt(sb, ERRORS_RO);
1560                         clear_opt(sb, ERRORS_PANIC);
1561                         set_opt(sb, ERRORS_CONT);
1562                         break;
1563                 case Opt_nouid32:
1564                         set_opt(sb, NO_UID32);
1565                         break;
1566                 case Opt_debug:
1567                         set_opt(sb, DEBUG);
1568                         break;
1569                 case Opt_oldalloc:
1570                         set_opt(sb, OLDALLOC);
1571                         break;
1572                 case Opt_orlov:
1573                         clear_opt(sb, OLDALLOC);
1574                         break;
1575 #ifdef CONFIG_EXT4_FS_XATTR
1576                 case Opt_user_xattr:
1577                         set_opt(sb, XATTR_USER);
1578                         break;
1579                 case Opt_nouser_xattr:
1580                         clear_opt(sb, XATTR_USER);
1581                         break;
1582 #else
1583                 case Opt_user_xattr:
1584                 case Opt_nouser_xattr:
1585                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1586                         break;
1587 #endif
1588 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1589                 case Opt_acl:
1590                         set_opt(sb, POSIX_ACL);
1591                         break;
1592                 case Opt_noacl:
1593                         clear_opt(sb, POSIX_ACL);
1594                         break;
1595 #else
1596                 case Opt_acl:
1597                 case Opt_noacl:
1598                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1599                         break;
1600 #endif
1601                 case Opt_journal_update:
1602                         /* @@@ FIXME */
1603                         /* Eventually we will want to be able to create
1604                            a journal file here.  For now, only allow the
1605                            user to specify an existing inode to be the
1606                            journal file. */
1607                         if (is_remount) {
1608                                 ext4_msg(sb, KERN_ERR,
1609                                          "Cannot specify journal on remount");
1610                                 return 0;
1611                         }
1612                         set_opt(sb, UPDATE_JOURNAL);
1613                         break;
1614                 case Opt_journal_dev:
1615                         if (is_remount) {
1616                                 ext4_msg(sb, KERN_ERR,
1617                                         "Cannot specify journal on remount");
1618                                 return 0;
1619                         }
1620                         if (match_int(&args[0], &option))
1621                                 return 0;
1622                         *journal_devnum = option;
1623                         break;
1624                 case Opt_journal_checksum:
1625                         set_opt(sb, JOURNAL_CHECKSUM);
1626                         break;
1627                 case Opt_journal_async_commit:
1628                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1629                         set_opt(sb, JOURNAL_CHECKSUM);
1630                         break;
1631                 case Opt_noload:
1632                         set_opt(sb, NOLOAD);
1633                         break;
1634                 case Opt_commit:
1635                         if (match_int(&args[0], &option))
1636                                 return 0;
1637                         if (option < 0)
1638                                 return 0;
1639                         if (option == 0)
1640                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1641                         sbi->s_commit_interval = HZ * option;
1642                         break;
1643                 case Opt_max_batch_time:
1644                         if (match_int(&args[0], &option))
1645                                 return 0;
1646                         if (option < 0)
1647                                 return 0;
1648                         if (option == 0)
1649                                 option = EXT4_DEF_MAX_BATCH_TIME;
1650                         sbi->s_max_batch_time = option;
1651                         break;
1652                 case Opt_min_batch_time:
1653                         if (match_int(&args[0], &option))
1654                                 return 0;
1655                         if (option < 0)
1656                                 return 0;
1657                         sbi->s_min_batch_time = option;
1658                         break;
1659                 case Opt_data_journal:
1660                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1661                         goto datacheck;
1662                 case Opt_data_ordered:
1663                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1664                         goto datacheck;
1665                 case Opt_data_writeback:
1666                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1667                 datacheck:
1668                         if (is_remount) {
1669                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1670                                         ext4_msg(sb, KERN_ERR,
1671                                                 "Cannot change data mode on remount");
1672                                         return 0;
1673                                 }
1674                         } else {
1675                                 clear_opt(sb, DATA_FLAGS);
1676                                 sbi->s_mount_opt |= data_opt;
1677                         }
1678                         break;
1679                 case Opt_data_err_abort:
1680                         set_opt(sb, DATA_ERR_ABORT);
1681                         break;
1682                 case Opt_data_err_ignore:
1683                         clear_opt(sb, DATA_ERR_ABORT);
1684                         break;
1685 #ifdef CONFIG_QUOTA
1686                 case Opt_usrjquota:
1687                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1688                                 return 0;
1689                         break;
1690                 case Opt_grpjquota:
1691                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1692                                 return 0;
1693                         break;
1694                 case Opt_offusrjquota:
1695                         if (!clear_qf_name(sb, USRQUOTA))
1696                                 return 0;
1697                         break;
1698                 case Opt_offgrpjquota:
1699                         if (!clear_qf_name(sb, GRPQUOTA))
1700                                 return 0;
1701                         break;
1702
1703                 case Opt_jqfmt_vfsold:
1704                         qfmt = QFMT_VFS_OLD;
1705                         goto set_qf_format;
1706                 case Opt_jqfmt_vfsv0:
1707                         qfmt = QFMT_VFS_V0;
1708                         goto set_qf_format;
1709                 case Opt_jqfmt_vfsv1:
1710                         qfmt = QFMT_VFS_V1;
1711 set_qf_format:
1712                         if (sb_any_quota_loaded(sb) &&
1713                             sbi->s_jquota_fmt != qfmt) {
1714                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1715                                         "journaled quota options when "
1716                                         "quota turned on");
1717                                 return 0;
1718                         }
1719                         sbi->s_jquota_fmt = qfmt;
1720                         break;
1721                 case Opt_quota:
1722                 case Opt_usrquota:
1723                         set_opt(sb, QUOTA);
1724                         set_opt(sb, USRQUOTA);
1725                         break;
1726                 case Opt_grpquota:
1727                         set_opt(sb, QUOTA);
1728                         set_opt(sb, GRPQUOTA);
1729                         break;
1730                 case Opt_noquota:
1731                         if (sb_any_quota_loaded(sb)) {
1732                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1733                                         "options when quota turned on");
1734                                 return 0;
1735                         }
1736                         clear_opt(sb, QUOTA);
1737                         clear_opt(sb, USRQUOTA);
1738                         clear_opt(sb, GRPQUOTA);
1739                         break;
1740 #else
1741                 case Opt_quota:
1742                 case Opt_usrquota:
1743                 case Opt_grpquota:
1744                         ext4_msg(sb, KERN_ERR,
1745                                 "quota options not supported");
1746                         break;
1747                 case Opt_usrjquota:
1748                 case Opt_grpjquota:
1749                 case Opt_offusrjquota:
1750                 case Opt_offgrpjquota:
1751                 case Opt_jqfmt_vfsold:
1752                 case Opt_jqfmt_vfsv0:
1753                 case Opt_jqfmt_vfsv1:
1754                         ext4_msg(sb, KERN_ERR,
1755                                 "journaled quota options not supported");
1756                         break;
1757                 case Opt_noquota:
1758                         break;
1759 #endif
1760                 case Opt_abort:
1761                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1762                         break;
1763                 case Opt_nobarrier:
1764                         clear_opt(sb, BARRIER);
1765                         break;
1766                 case Opt_barrier:
1767                         if (args[0].from) {
1768                                 if (match_int(&args[0], &option))
1769                                         return 0;
1770                         } else
1771                                 option = 1;     /* No argument, default to 1 */
1772                         if (option)
1773                                 set_opt(sb, BARRIER);
1774                         else
1775                                 clear_opt(sb, BARRIER);
1776                         break;
1777                 case Opt_ignore:
1778                         break;
1779                 case Opt_resize:
1780                         if (!is_remount) {
1781                                 ext4_msg(sb, KERN_ERR,
1782                                         "resize option only available "
1783                                         "for remount");
1784                                 return 0;
1785                         }
1786                         if (match_int(&args[0], &option) != 0)
1787                                 return 0;
1788                         *n_blocks_count = option;
1789                         break;
1790                 case Opt_nobh:
1791                         ext4_msg(sb, KERN_WARNING,
1792                                  "Ignoring deprecated nobh option");
1793                         break;
1794                 case Opt_bh:
1795                         ext4_msg(sb, KERN_WARNING,
1796                                  "Ignoring deprecated bh option");
1797                         break;
1798                 case Opt_i_version:
1799                         set_opt(sb, I_VERSION);
1800                         sb->s_flags |= MS_I_VERSION;
1801                         break;
1802                 case Opt_nodelalloc:
1803                         clear_opt(sb, DELALLOC);
1804                         break;
1805                 case Opt_mblk_io_submit:
1806                         set_opt(sb, MBLK_IO_SUBMIT);
1807                         break;
1808                 case Opt_nomblk_io_submit:
1809                         clear_opt(sb, MBLK_IO_SUBMIT);
1810                         break;
1811                 case Opt_stripe:
1812                         if (match_int(&args[0], &option))
1813                                 return 0;
1814                         if (option < 0)
1815                                 return 0;
1816                         sbi->s_stripe = option;
1817                         break;
1818                 case Opt_delalloc:
1819                         set_opt(sb, DELALLOC);
1820                         break;
1821                 case Opt_block_validity:
1822                         set_opt(sb, BLOCK_VALIDITY);
1823                         break;
1824                 case Opt_noblock_validity:
1825                         clear_opt(sb, BLOCK_VALIDITY);
1826                         break;
1827                 case Opt_inode_readahead_blks:
1828                         if (match_int(&args[0], &option))
1829                                 return 0;
1830                         if (option < 0 || option > (1 << 30))
1831                                 return 0;
1832                         if (option && !is_power_of_2(option)) {
1833                                 ext4_msg(sb, KERN_ERR,
1834                                          "EXT4-fs: inode_readahead_blks"
1835                                          " must be a power of 2");
1836                                 return 0;
1837                         }
1838                         sbi->s_inode_readahead_blks = option;
1839                         break;
1840                 case Opt_journal_ioprio:
1841                         if (match_int(&args[0], &option))
1842                                 return 0;
1843                         if (option < 0 || option > 7)
1844                                 break;
1845                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1846                                                             option);
1847                         break;
1848                 case Opt_noauto_da_alloc:
1849                         set_opt(sb, NO_AUTO_DA_ALLOC);
1850                         break;
1851                 case Opt_auto_da_alloc:
1852                         if (args[0].from) {
1853                                 if (match_int(&args[0], &option))
1854                                         return 0;
1855                         } else
1856                                 option = 1;     /* No argument, default to 1 */
1857                         if (option)
1858                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1859                         else
1860                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1861                         break;
1862                 case Opt_discard:
1863                         set_opt(sb, DISCARD);
1864                         break;
1865                 case Opt_nodiscard:
1866                         clear_opt(sb, DISCARD);
1867                         break;
1868                 case Opt_dioread_nolock:
1869                         set_opt(sb, DIOREAD_NOLOCK);
1870                         break;
1871                 case Opt_dioread_lock:
1872                         clear_opt(sb, DIOREAD_NOLOCK);
1873                         break;
1874                 case Opt_init_inode_table:
1875                         set_opt(sb, INIT_INODE_TABLE);
1876                         if (args[0].from) {
1877                                 if (match_int(&args[0], &option))
1878                                         return 0;
1879                         } else
1880                                 option = EXT4_DEF_LI_WAIT_MULT;
1881                         if (option < 0)
1882                                 return 0;
1883                         sbi->s_li_wait_mult = option;
1884                         break;
1885                 case Opt_noinit_inode_table:
1886                         clear_opt(sb, INIT_INODE_TABLE);
1887                         break;
1888                 default:
1889                         ext4_msg(sb, KERN_ERR,
1890                                "Unrecognized mount option \"%s\" "
1891                                "or missing value", p);
1892                         return 0;
1893                 }
1894         }
1895 #ifdef CONFIG_QUOTA
1896         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1897                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1898                         clear_opt(sb, USRQUOTA);
1899
1900                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1901                         clear_opt(sb, GRPQUOTA);
1902
1903                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1904                         ext4_msg(sb, KERN_ERR, "old and new quota "
1905                                         "format mixing");
1906                         return 0;
1907                 }
1908
1909                 if (!sbi->s_jquota_fmt) {
1910                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1911                                         "not specified");
1912                         return 0;
1913                 }
1914         } else {
1915                 if (sbi->s_jquota_fmt) {
1916                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1917                                         "specified with no journaling "
1918                                         "enabled");
1919                         return 0;
1920                 }
1921         }
1922 #endif
1923         return 1;
1924 }
1925
1926 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1927                             int read_only)
1928 {
1929         struct ext4_sb_info *sbi = EXT4_SB(sb);
1930         int res = 0;
1931
1932         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1933                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1934                          "forcing read-only mode");
1935                 res = MS_RDONLY;
1936         }
1937         if (read_only)
1938                 return res;
1939         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1940                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1941                          "running e2fsck is recommended");
1942         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1943                 ext4_msg(sb, KERN_WARNING,
1944                          "warning: mounting fs with errors, "
1945                          "running e2fsck is recommended");
1946         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1947                  le16_to_cpu(es->s_mnt_count) >=
1948                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1949                 ext4_msg(sb, KERN_WARNING,
1950                          "warning: maximal mount count reached, "
1951                          "running e2fsck is recommended");
1952         else if (le32_to_cpu(es->s_checkinterval) &&
1953                 (le32_to_cpu(es->s_lastcheck) +
1954                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1955                 ext4_msg(sb, KERN_WARNING,
1956                          "warning: checktime reached, "
1957                          "running e2fsck is recommended");
1958         if (!sbi->s_journal)
1959                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1960         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1961                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1962         le16_add_cpu(&es->s_mnt_count, 1);
1963         es->s_mtime = cpu_to_le32(get_seconds());
1964         ext4_update_dynamic_rev(sb);
1965         if (sbi->s_journal)
1966                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1967
1968         ext4_commit_super(sb, 1);
1969         if (test_opt(sb, DEBUG))
1970                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1971                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1972                         sb->s_blocksize,
1973                         sbi->s_groups_count,
1974                         EXT4_BLOCKS_PER_GROUP(sb),
1975                         EXT4_INODES_PER_GROUP(sb),
1976                         sbi->s_mount_opt, sbi->s_mount_opt2);
1977
1978         cleancache_init_fs(sb);
1979         return res;
1980 }
1981
1982 static int ext4_fill_flex_info(struct super_block *sb)
1983 {
1984         struct ext4_sb_info *sbi = EXT4_SB(sb);
1985         struct ext4_group_desc *gdp = NULL;
1986         ext4_group_t flex_group_count;
1987         ext4_group_t flex_group;
1988         int groups_per_flex = 0;
1989         size_t size;
1990         int i;
1991
1992         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1993         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1994
1995         if (groups_per_flex < 2) {
1996                 sbi->s_log_groups_per_flex = 0;
1997                 return 1;
1998         }
1999
2000         /* We allocate both existing and potentially added groups */
2001         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2002                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2003                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2004         size = flex_group_count * sizeof(struct flex_groups);
2005         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2006         if (sbi->s_flex_groups == NULL) {
2007                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2008                          flex_group_count);
2009                 goto failed;
2010         }
2011
2012         for (i = 0; i < sbi->s_groups_count; i++) {
2013                 gdp = ext4_get_group_desc(sb, i, NULL);
2014
2015                 flex_group = ext4_flex_group(sbi, i);
2016                 atomic_add(ext4_free_inodes_count(sb, gdp),
2017                            &sbi->s_flex_groups[flex_group].free_inodes);
2018                 atomic_add(ext4_free_blks_count(sb, gdp),
2019                            &sbi->s_flex_groups[flex_group].free_blocks);
2020                 atomic_add(ext4_used_dirs_count(sb, gdp),
2021                            &sbi->s_flex_groups[flex_group].used_dirs);
2022         }
2023
2024         return 1;
2025 failed:
2026         return 0;
2027 }
2028
2029 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2030                             struct ext4_group_desc *gdp)
2031 {
2032         __u16 crc = 0;
2033
2034         if (sbi->s_es->s_feature_ro_compat &
2035             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2036                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2037                 __le32 le_group = cpu_to_le32(block_group);
2038
2039                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2040                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2041                 crc = crc16(crc, (__u8 *)gdp, offset);
2042                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2043                 /* for checksum of struct ext4_group_desc do the rest...*/
2044                 if ((sbi->s_es->s_feature_incompat &
2045                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2046                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2047                         crc = crc16(crc, (__u8 *)gdp + offset,
2048                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2049                                         offset);
2050         }
2051
2052         return cpu_to_le16(crc);
2053 }
2054
2055 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2056                                 struct ext4_group_desc *gdp)
2057 {
2058         if ((sbi->s_es->s_feature_ro_compat &
2059              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2060             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2061                 return 0;
2062
2063         return 1;
2064 }
2065
2066 /* Called at mount-time, super-block is locked */
2067 static int ext4_check_descriptors(struct super_block *sb,
2068                                   ext4_group_t *first_not_zeroed)
2069 {
2070         struct ext4_sb_info *sbi = EXT4_SB(sb);
2071         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2072         ext4_fsblk_t last_block;
2073         ext4_fsblk_t block_bitmap;
2074         ext4_fsblk_t inode_bitmap;
2075         ext4_fsblk_t inode_table;
2076         int flexbg_flag = 0;
2077         ext4_group_t i, grp = sbi->s_groups_count;
2078
2079         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2080                 flexbg_flag = 1;
2081
2082         ext4_debug("Checking group descriptors");
2083
2084         for (i = 0; i < sbi->s_groups_count; i++) {
2085                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2086
2087                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2088                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2089                 else
2090                         last_block = first_block +
2091                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2092
2093                 if ((grp == sbi->s_groups_count) &&
2094                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2095                         grp = i;
2096
2097                 block_bitmap = ext4_block_bitmap(sb, gdp);
2098                 if (block_bitmap < first_block || block_bitmap > last_block) {
2099                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2100                                "Block bitmap for group %u not in group "
2101                                "(block %llu)!", i, block_bitmap);
2102                         return 0;
2103                 }
2104                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2105                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2106                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2107                                "Inode bitmap for group %u not in group "
2108                                "(block %llu)!", i, inode_bitmap);
2109                         return 0;
2110                 }
2111                 inode_table = ext4_inode_table(sb, gdp);
2112                 if (inode_table < first_block ||
2113                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2114                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2115                                "Inode table for group %u not in group "
2116                                "(block %llu)!", i, inode_table);
2117                         return 0;
2118                 }
2119                 ext4_lock_group(sb, i);
2120                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2121                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2122                                  "Checksum for group %u failed (%u!=%u)",
2123                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2124                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2125                         if (!(sb->s_flags & MS_RDONLY)) {
2126                                 ext4_unlock_group(sb, i);
2127                                 return 0;
2128                         }
2129                 }
2130                 ext4_unlock_group(sb, i);
2131                 if (!flexbg_flag)
2132                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2133         }
2134         if (NULL != first_not_zeroed)
2135                 *first_not_zeroed = grp;
2136
2137         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2138         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2139         return 1;
2140 }
2141
2142 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2143  * the superblock) which were deleted from all directories, but held open by
2144  * a process at the time of a crash.  We walk the list and try to delete these
2145  * inodes at recovery time (only with a read-write filesystem).
2146  *
2147  * In order to keep the orphan inode chain consistent during traversal (in
2148  * case of crash during recovery), we link each inode into the superblock
2149  * orphan list_head and handle it the same way as an inode deletion during
2150  * normal operation (which journals the operations for us).
2151  *
2152  * We only do an iget() and an iput() on each inode, which is very safe if we
2153  * accidentally point at an in-use or already deleted inode.  The worst that
2154  * can happen in this case is that we get a "bit already cleared" message from
2155  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2156  * e2fsck was run on this filesystem, and it must have already done the orphan
2157  * inode cleanup for us, so we can safely abort without any further action.
2158  */
2159 static void ext4_orphan_cleanup(struct super_block *sb,
2160                                 struct ext4_super_block *es)
2161 {
2162         unsigned int s_flags = sb->s_flags;
2163         int nr_orphans = 0, nr_truncates = 0;
2164 #ifdef CONFIG_QUOTA
2165         int i;
2166 #endif
2167         if (!es->s_last_orphan) {
2168                 jbd_debug(4, "no orphan inodes to clean up\n");
2169                 return;
2170         }
2171
2172         if (bdev_read_only(sb->s_bdev)) {
2173                 ext4_msg(sb, KERN_ERR, "write access "
2174                         "unavailable, skipping orphan cleanup");
2175                 return;
2176         }
2177
2178         /* Check if feature set would not allow a r/w mount */
2179         if (!ext4_feature_set_ok(sb, 0)) {
2180                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2181                          "unknown ROCOMPAT features");
2182                 return;
2183         }
2184
2185         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2186                 if (es->s_last_orphan)
2187                         jbd_debug(1, "Errors on filesystem, "
2188                                   "clearing orphan list.\n");
2189                 es->s_last_orphan = 0;
2190                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2191                 return;
2192         }
2193
2194         if (s_flags & MS_RDONLY) {
2195                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2196                 sb->s_flags &= ~MS_RDONLY;
2197         }
2198 #ifdef CONFIG_QUOTA
2199         /* Needed for iput() to work correctly and not trash data */
2200         sb->s_flags |= MS_ACTIVE;
2201         /* Turn on quotas so that they are updated correctly */
2202         for (i = 0; i < MAXQUOTAS; i++) {
2203                 if (EXT4_SB(sb)->s_qf_names[i]) {
2204                         int ret = ext4_quota_on_mount(sb, i);
2205                         if (ret < 0)
2206                                 ext4_msg(sb, KERN_ERR,
2207                                         "Cannot turn on journaled "
2208                                         "quota: error %d", ret);
2209                 }
2210         }
2211 #endif
2212
2213         while (es->s_last_orphan) {
2214                 struct inode *inode;
2215
2216                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2217                 if (IS_ERR(inode)) {
2218                         es->s_last_orphan = 0;
2219                         break;
2220                 }
2221
2222                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2223                 dquot_initialize(inode);
2224                 if (inode->i_nlink) {
2225                         ext4_msg(sb, KERN_DEBUG,
2226                                 "%s: truncating inode %lu to %lld bytes",
2227                                 __func__, inode->i_ino, inode->i_size);
2228                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2229                                   inode->i_ino, inode->i_size);
2230                         ext4_truncate(inode);
2231                         nr_truncates++;
2232                 } else {
2233                         ext4_msg(sb, KERN_DEBUG,
2234                                 "%s: deleting unreferenced inode %lu",
2235                                 __func__, inode->i_ino);
2236                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2237                                   inode->i_ino);
2238                         nr_orphans++;
2239                 }
2240                 iput(inode);  /* The delete magic happens here! */
2241         }
2242
2243 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2244
2245         if (nr_orphans)
2246                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2247                        PLURAL(nr_orphans));
2248         if (nr_truncates)
2249                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2250                        PLURAL(nr_truncates));
2251 #ifdef CONFIG_QUOTA
2252         /* Turn quotas off */
2253         for (i = 0; i < MAXQUOTAS; i++) {
2254                 if (sb_dqopt(sb)->files[i])
2255                         dquot_quota_off(sb, i);
2256         }
2257 #endif
2258         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2259 }
2260
2261 /*
2262  * Maximal extent format file size.
2263  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2264  * extent format containers, within a sector_t, and within i_blocks
2265  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2266  * so that won't be a limiting factor.
2267  *
2268  * However there is other limiting factor. We do store extents in the form
2269  * of starting block and length, hence the resulting length of the extent
2270  * covering maximum file size must fit into on-disk format containers as
2271  * well. Given that length is always by 1 unit bigger than max unit (because
2272  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2273  *
2274  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2275  */
2276 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2277 {
2278         loff_t res;
2279         loff_t upper_limit = MAX_LFS_FILESIZE;
2280
2281         /* small i_blocks in vfs inode? */
2282         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2283                 /*
2284                  * CONFIG_LBDAF is not enabled implies the inode
2285                  * i_block represent total blocks in 512 bytes
2286                  * 32 == size of vfs inode i_blocks * 8
2287                  */
2288                 upper_limit = (1LL << 32) - 1;
2289
2290                 /* total blocks in file system block size */
2291                 upper_limit >>= (blkbits - 9);
2292                 upper_limit <<= blkbits;
2293         }
2294
2295         /*
2296          * 32-bit extent-start container, ee_block. We lower the maxbytes
2297          * by one fs block, so ee_len can cover the extent of maximum file
2298          * size
2299          */
2300         res = (1LL << 32) - 1;
2301         res <<= blkbits;
2302
2303         /* Sanity check against vm- & vfs- imposed limits */
2304         if (res > upper_limit)
2305                 res = upper_limit;
2306
2307         return res;
2308 }
2309
2310 /*
2311  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2312  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2313  * We need to be 1 filesystem block less than the 2^48 sector limit.
2314  */
2315 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2316 {
2317         loff_t res = EXT4_NDIR_BLOCKS;
2318         int meta_blocks;
2319         loff_t upper_limit;
2320         /* This is calculated to be the largest file size for a dense, block
2321          * mapped file such that the file's total number of 512-byte sectors,
2322          * including data and all indirect blocks, does not exceed (2^48 - 1).
2323          *
2324          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2325          * number of 512-byte sectors of the file.
2326          */
2327
2328         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2329                 /*
2330                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2331                  * the inode i_block field represents total file blocks in
2332                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2333                  */
2334                 upper_limit = (1LL << 32) - 1;
2335
2336                 /* total blocks in file system block size */
2337                 upper_limit >>= (bits - 9);
2338
2339         } else {
2340                 /*
2341                  * We use 48 bit ext4_inode i_blocks
2342                  * With EXT4_HUGE_FILE_FL set the i_blocks
2343                  * represent total number of blocks in
2344                  * file system block size
2345                  */
2346                 upper_limit = (1LL << 48) - 1;
2347
2348         }
2349
2350         /* indirect blocks */
2351         meta_blocks = 1;
2352         /* double indirect blocks */
2353         meta_blocks += 1 + (1LL << (bits-2));
2354         /* tripple indirect blocks */
2355         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2356
2357         upper_limit -= meta_blocks;
2358         upper_limit <<= bits;
2359
2360         res += 1LL << (bits-2);
2361         res += 1LL << (2*(bits-2));
2362         res += 1LL << (3*(bits-2));
2363         res <<= bits;
2364         if (res > upper_limit)
2365                 res = upper_limit;
2366
2367         if (res > MAX_LFS_FILESIZE)
2368                 res = MAX_LFS_FILESIZE;
2369
2370         return res;
2371 }
2372
2373 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2374                                    ext4_fsblk_t logical_sb_block, int nr)
2375 {
2376         struct ext4_sb_info *sbi = EXT4_SB(sb);
2377         ext4_group_t bg, first_meta_bg;
2378         int has_super = 0;
2379
2380         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2381
2382         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2383             nr < first_meta_bg)
2384                 return logical_sb_block + nr + 1;
2385         bg = sbi->s_desc_per_block * nr;
2386         if (ext4_bg_has_super(sb, bg))
2387                 has_super = 1;
2388
2389         return (has_super + ext4_group_first_block_no(sb, bg));
2390 }
2391
2392 /**
2393  * ext4_get_stripe_size: Get the stripe size.
2394  * @sbi: In memory super block info
2395  *
2396  * If we have specified it via mount option, then
2397  * use the mount option value. If the value specified at mount time is
2398  * greater than the blocks per group use the super block value.
2399  * If the super block value is greater than blocks per group return 0.
2400  * Allocator needs it be less than blocks per group.
2401  *
2402  */
2403 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2404 {
2405         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2406         unsigned long stripe_width =
2407                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2408         int ret;
2409
2410         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2411                 ret = sbi->s_stripe;
2412         else if (stripe_width <= sbi->s_blocks_per_group)
2413                 ret = stripe_width;
2414         else if (stride <= sbi->s_blocks_per_group)
2415                 ret = stride;
2416         else
2417                 ret = 0;
2418
2419         /*
2420          * If the stripe width is 1, this makes no sense and
2421          * we set it to 0 to turn off stripe handling code.
2422          */
2423         if (ret <= 1)
2424                 ret = 0;
2425
2426         return ret;
2427 }
2428
2429 /* sysfs supprt */
2430
2431 struct ext4_attr {
2432         struct attribute attr;
2433         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2434         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2435                          const char *, size_t);
2436         int offset;
2437 };
2438
2439 static int parse_strtoul(const char *buf,
2440                 unsigned long max, unsigned long *value)
2441 {
2442         char *endp;
2443
2444         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2445         endp = skip_spaces(endp);
2446         if (*endp || *value > max)
2447                 return -EINVAL;
2448
2449         return 0;
2450 }
2451
2452 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2453                                               struct ext4_sb_info *sbi,
2454                                               char *buf)
2455 {
2456         return snprintf(buf, PAGE_SIZE, "%llu\n",
2457                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2458 }
2459
2460 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2461                                          struct ext4_sb_info *sbi, char *buf)
2462 {
2463         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2464
2465         if (!sb->s_bdev->bd_part)
2466                 return snprintf(buf, PAGE_SIZE, "0\n");
2467         return snprintf(buf, PAGE_SIZE, "%lu\n",
2468                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2469                          sbi->s_sectors_written_start) >> 1);
2470 }
2471
2472 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2473                                           struct ext4_sb_info *sbi, char *buf)
2474 {
2475         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2476
2477         if (!sb->s_bdev->bd_part)
2478                 return snprintf(buf, PAGE_SIZE, "0\n");
2479         return snprintf(buf, PAGE_SIZE, "%llu\n",
2480                         (unsigned long long)(sbi->s_kbytes_written +
2481                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2482                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2483 }
2484
2485 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2486                                       struct ext4_sb_info *sbi, char *buf)
2487 {
2488         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2489 }
2490
2491 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2492                                         struct ext4_sb_info *sbi, char *buf)
2493 {
2494         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2495 }
2496
2497 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2498                                           struct ext4_sb_info *sbi,
2499                                           const char *buf, size_t count)
2500 {
2501         unsigned long t;
2502
2503         if (parse_strtoul(buf, 0x40000000, &t))
2504                 return -EINVAL;
2505
2506         if (t && !is_power_of_2(t))
2507                 return -EINVAL;
2508
2509         sbi->s_inode_readahead_blks = t;
2510         return count;
2511 }
2512
2513 static ssize_t sbi_ui_show(struct ext4_attr *a,
2514                            struct ext4_sb_info *sbi, char *buf)
2515 {
2516         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2517
2518         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2519 }
2520
2521 static ssize_t sbi_ui_store(struct ext4_attr *a,
2522                             struct ext4_sb_info *sbi,
2523                             const char *buf, size_t count)
2524 {
2525         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2526         unsigned long t;
2527
2528         if (parse_strtoul(buf, 0xffffffff, &t))
2529                 return -EINVAL;
2530         *ui = t;
2531         return count;
2532 }
2533
2534 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2535 static struct ext4_attr ext4_attr_##_name = {                   \
2536         .attr = {.name = __stringify(_name), .mode = _mode },   \
2537         .show   = _show,                                        \
2538         .store  = _store,                                       \
2539         .offset = offsetof(struct ext4_sb_info, _elname),       \
2540 }
2541 #define EXT4_ATTR(name, mode, show, store) \
2542 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2543
2544 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2545 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2546 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2547 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2548         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2549 #define ATTR_LIST(name) &ext4_attr_##name.attr
2550
2551 EXT4_RO_ATTR(delayed_allocation_blocks);
2552 EXT4_RO_ATTR(session_write_kbytes);
2553 EXT4_RO_ATTR(lifetime_write_kbytes);
2554 EXT4_RO_ATTR(extent_cache_hits);
2555 EXT4_RO_ATTR(extent_cache_misses);
2556 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2557                  inode_readahead_blks_store, s_inode_readahead_blks);
2558 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2559 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2560 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2561 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2562 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2563 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2564 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2565 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2566
2567 static struct attribute *ext4_attrs[] = {
2568         ATTR_LIST(delayed_allocation_blocks),
2569         ATTR_LIST(session_write_kbytes),
2570         ATTR_LIST(lifetime_write_kbytes),
2571         ATTR_LIST(extent_cache_hits),
2572         ATTR_LIST(extent_cache_misses),
2573         ATTR_LIST(inode_readahead_blks),
2574         ATTR_LIST(inode_goal),
2575         ATTR_LIST(mb_stats),
2576         ATTR_LIST(mb_max_to_scan),
2577         ATTR_LIST(mb_min_to_scan),
2578         ATTR_LIST(mb_order2_req),
2579         ATTR_LIST(mb_stream_req),
2580         ATTR_LIST(mb_group_prealloc),
2581         ATTR_LIST(max_writeback_mb_bump),
2582         NULL,
2583 };
2584
2585 /* Features this copy of ext4 supports */
2586 EXT4_INFO_ATTR(lazy_itable_init);
2587 EXT4_INFO_ATTR(batched_discard);
2588
2589 static struct attribute *ext4_feat_attrs[] = {
2590         ATTR_LIST(lazy_itable_init),
2591         ATTR_LIST(batched_discard),
2592         NULL,
2593 };
2594
2595 static ssize_t ext4_attr_show(struct kobject *kobj,
2596                               struct attribute *attr, char *buf)
2597 {
2598         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2599                                                 s_kobj);
2600         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2601
2602         return a->show ? a->show(a, sbi, buf) : 0;
2603 }
2604
2605 static ssize_t ext4_attr_store(struct kobject *kobj,
2606                                struct attribute *attr,
2607                                const char *buf, size_t len)
2608 {
2609         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2610                                                 s_kobj);
2611         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2612
2613         return a->store ? a->store(a, sbi, buf, len) : 0;
2614 }
2615
2616 static void ext4_sb_release(struct kobject *kobj)
2617 {
2618         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2619                                                 s_kobj);
2620         complete(&sbi->s_kobj_unregister);
2621 }
2622
2623 static const struct sysfs_ops ext4_attr_ops = {
2624         .show   = ext4_attr_show,
2625         .store  = ext4_attr_store,
2626 };
2627
2628 static struct kobj_type ext4_ktype = {
2629         .default_attrs  = ext4_attrs,
2630         .sysfs_ops      = &ext4_attr_ops,
2631         .release        = ext4_sb_release,
2632 };
2633
2634 static void ext4_feat_release(struct kobject *kobj)
2635 {
2636         complete(&ext4_feat->f_kobj_unregister);
2637 }
2638
2639 static struct kobj_type ext4_feat_ktype = {
2640         .default_attrs  = ext4_feat_attrs,
2641         .sysfs_ops      = &ext4_attr_ops,
2642         .release        = ext4_feat_release,
2643 };
2644
2645 /*
2646  * Check whether this filesystem can be mounted based on
2647  * the features present and the RDONLY/RDWR mount requested.
2648  * Returns 1 if this filesystem can be mounted as requested,
2649  * 0 if it cannot be.
2650  */
2651 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2652 {
2653         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2654                 ext4_msg(sb, KERN_ERR,
2655                         "Couldn't mount because of "
2656                         "unsupported optional features (%x)",
2657                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2658                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2659                 return 0;
2660         }
2661
2662         if (readonly)
2663                 return 1;
2664
2665         /* Check that feature set is OK for a read-write mount */
2666         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2667                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2668                          "unsupported optional features (%x)",
2669                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2670                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2671                 return 0;
2672         }
2673         /*
2674          * Large file size enabled file system can only be mounted
2675          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2676          */
2677         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2678                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2679                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2680                                  "cannot be mounted RDWR without "
2681                                  "CONFIG_LBDAF");
2682                         return 0;
2683                 }
2684         }
2685         return 1;
2686 }
2687
2688 /*
2689  * This function is called once a day if we have errors logged
2690  * on the file system
2691  */
2692 static void print_daily_error_info(unsigned long arg)
2693 {
2694         struct super_block *sb = (struct super_block *) arg;
2695         struct ext4_sb_info *sbi;
2696         struct ext4_super_block *es;
2697
2698         sbi = EXT4_SB(sb);
2699         es = sbi->s_es;
2700
2701         if (es->s_error_count)
2702                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2703                          le32_to_cpu(es->s_error_count));
2704         if (es->s_first_error_time) {
2705                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2706                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2707                        (int) sizeof(es->s_first_error_func),
2708                        es->s_first_error_func,
2709                        le32_to_cpu(es->s_first_error_line));
2710                 if (es->s_first_error_ino)
2711                         printk(": inode %u",
2712                                le32_to_cpu(es->s_first_error_ino));
2713                 if (es->s_first_error_block)
2714                         printk(": block %llu", (unsigned long long)
2715                                le64_to_cpu(es->s_first_error_block));
2716                 printk("\n");
2717         }
2718         if (es->s_last_error_time) {
2719                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2720                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2721                        (int) sizeof(es->s_last_error_func),
2722                        es->s_last_error_func,
2723                        le32_to_cpu(es->s_last_error_line));
2724                 if (es->s_last_error_ino)
2725                         printk(": inode %u",
2726                                le32_to_cpu(es->s_last_error_ino));
2727                 if (es->s_last_error_block)
2728                         printk(": block %llu", (unsigned long long)
2729                                le64_to_cpu(es->s_last_error_block));
2730                 printk("\n");
2731         }
2732         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2733 }
2734
2735 /* Find next suitable group and run ext4_init_inode_table */
2736 static int ext4_run_li_request(struct ext4_li_request *elr)
2737 {
2738         struct ext4_group_desc *gdp = NULL;
2739         ext4_group_t group, ngroups;
2740         struct super_block *sb;
2741         unsigned long timeout = 0;
2742         int ret = 0;
2743
2744         sb = elr->lr_super;
2745         ngroups = EXT4_SB(sb)->s_groups_count;
2746
2747         for (group = elr->lr_next_group; group < ngroups; group++) {
2748                 gdp = ext4_get_group_desc(sb, group, NULL);
2749                 if (!gdp) {
2750                         ret = 1;
2751                         break;
2752                 }
2753
2754                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2755                         break;
2756         }
2757
2758         if (group == ngroups)
2759                 ret = 1;
2760
2761         if (!ret) {
2762                 timeout = jiffies;
2763                 ret = ext4_init_inode_table(sb, group,
2764                                             elr->lr_timeout ? 0 : 1);
2765                 if (elr->lr_timeout == 0) {
2766                         timeout = (jiffies - timeout) *
2767                                   elr->lr_sbi->s_li_wait_mult;
2768                         elr->lr_timeout = timeout;
2769                 }
2770                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2771                 elr->lr_next_group = group + 1;
2772         }
2773
2774         return ret;
2775 }
2776
2777 /*
2778  * Remove lr_request from the list_request and free the
2779  * request structure. Should be called with li_list_mtx held
2780  */
2781 static void ext4_remove_li_request(struct ext4_li_request *elr)
2782 {
2783         struct ext4_sb_info *sbi;
2784
2785         if (!elr)
2786                 return;
2787
2788         sbi = elr->lr_sbi;
2789
2790         list_del(&elr->lr_request);
2791         sbi->s_li_request = NULL;
2792         kfree(elr);
2793 }
2794
2795 static void ext4_unregister_li_request(struct super_block *sb)
2796 {
2797         mutex_lock(&ext4_li_mtx);
2798         if (!ext4_li_info) {
2799                 mutex_unlock(&ext4_li_mtx);
2800                 return;
2801         }
2802
2803         mutex_lock(&ext4_li_info->li_list_mtx);
2804         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2805         mutex_unlock(&ext4_li_info->li_list_mtx);
2806         mutex_unlock(&ext4_li_mtx);
2807 }
2808
2809 static struct task_struct *ext4_lazyinit_task;
2810
2811 /*
2812  * This is the function where ext4lazyinit thread lives. It walks
2813  * through the request list searching for next scheduled filesystem.
2814  * When such a fs is found, run the lazy initialization request
2815  * (ext4_rn_li_request) and keep track of the time spend in this
2816  * function. Based on that time we compute next schedule time of
2817  * the request. When walking through the list is complete, compute
2818  * next waking time and put itself into sleep.
2819  */
2820 static int ext4_lazyinit_thread(void *arg)
2821 {
2822         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2823         struct list_head *pos, *n;
2824         struct ext4_li_request *elr;
2825         unsigned long next_wakeup, cur;
2826
2827         BUG_ON(NULL == eli);
2828
2829 cont_thread:
2830         while (true) {
2831                 next_wakeup = MAX_JIFFY_OFFSET;
2832
2833                 mutex_lock(&eli->li_list_mtx);
2834                 if (list_empty(&eli->li_request_list)) {
2835                         mutex_unlock(&eli->li_list_mtx);
2836                         goto exit_thread;
2837                 }
2838
2839                 list_for_each_safe(pos, n, &eli->li_request_list) {
2840                         elr = list_entry(pos, struct ext4_li_request,
2841                                          lr_request);
2842
2843                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2844                                 if (ext4_run_li_request(elr) != 0) {
2845                                         /* error, remove the lazy_init job */
2846                                         ext4_remove_li_request(elr);
2847                                         continue;
2848                                 }
2849                         }
2850
2851                         if (time_before(elr->lr_next_sched, next_wakeup))
2852                                 next_wakeup = elr->lr_next_sched;
2853                 }
2854                 mutex_unlock(&eli->li_list_mtx);
2855
2856                 if (freezing(current))
2857                         refrigerator();
2858
2859                 cur = jiffies;
2860                 if ((time_after_eq(cur, next_wakeup)) ||
2861                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2862                         cond_resched();
2863                         continue;
2864                 }
2865
2866                 schedule_timeout_interruptible(next_wakeup - cur);
2867
2868                 if (kthread_should_stop()) {
2869                         ext4_clear_request_list();
2870                         goto exit_thread;
2871                 }
2872         }
2873
2874 exit_thread:
2875         /*
2876          * It looks like the request list is empty, but we need
2877          * to check it under the li_list_mtx lock, to prevent any
2878          * additions into it, and of course we should lock ext4_li_mtx
2879          * to atomically free the list and ext4_li_info, because at
2880          * this point another ext4 filesystem could be registering
2881          * new one.
2882          */
2883         mutex_lock(&ext4_li_mtx);
2884         mutex_lock(&eli->li_list_mtx);
2885         if (!list_empty(&eli->li_request_list)) {
2886                 mutex_unlock(&eli->li_list_mtx);
2887                 mutex_unlock(&ext4_li_mtx);
2888                 goto cont_thread;
2889         }
2890         mutex_unlock(&eli->li_list_mtx);
2891         kfree(ext4_li_info);
2892         ext4_li_info = NULL;
2893         mutex_unlock(&ext4_li_mtx);
2894
2895         return 0;
2896 }
2897
2898 static void ext4_clear_request_list(void)
2899 {
2900         struct list_head *pos, *n;
2901         struct ext4_li_request *elr;
2902
2903         mutex_lock(&ext4_li_info->li_list_mtx);
2904         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2905                 elr = list_entry(pos, struct ext4_li_request,
2906                                  lr_request);
2907                 ext4_remove_li_request(elr);
2908         }
2909         mutex_unlock(&ext4_li_info->li_list_mtx);
2910 }
2911
2912 static int ext4_run_lazyinit_thread(void)
2913 {
2914         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2915                                          ext4_li_info, "ext4lazyinit");
2916         if (IS_ERR(ext4_lazyinit_task)) {
2917                 int err = PTR_ERR(ext4_lazyinit_task);
2918                 ext4_clear_request_list();
2919                 kfree(ext4_li_info);
2920                 ext4_li_info = NULL;
2921                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2922                                  "initialization thread\n",
2923                                  err);
2924                 return err;
2925         }
2926         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2927         return 0;
2928 }
2929
2930 /*
2931  * Check whether it make sense to run itable init. thread or not.
2932  * If there is at least one uninitialized inode table, return
2933  * corresponding group number, else the loop goes through all
2934  * groups and return total number of groups.
2935  */
2936 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2937 {
2938         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2939         struct ext4_group_desc *gdp = NULL;
2940
2941         for (group = 0; group < ngroups; group++) {
2942                 gdp = ext4_get_group_desc(sb, group, NULL);
2943                 if (!gdp)
2944                         continue;
2945
2946                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2947                         break;
2948         }
2949
2950         return group;
2951 }
2952
2953 static int ext4_li_info_new(void)
2954 {
2955         struct ext4_lazy_init *eli = NULL;
2956
2957         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2958         if (!eli)
2959                 return -ENOMEM;
2960
2961         INIT_LIST_HEAD(&eli->li_request_list);
2962         mutex_init(&eli->li_list_mtx);
2963
2964         eli->li_state |= EXT4_LAZYINIT_QUIT;
2965
2966         ext4_li_info = eli;
2967
2968         return 0;
2969 }
2970
2971 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2972                                             ext4_group_t start)
2973 {
2974         struct ext4_sb_info *sbi = EXT4_SB(sb);
2975         struct ext4_li_request *elr;
2976         unsigned long rnd;
2977
2978         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2979         if (!elr)
2980                 return NULL;
2981
2982         elr->lr_super = sb;
2983         elr->lr_sbi = sbi;
2984         elr->lr_next_group = start;
2985
2986         /*
2987          * Randomize first schedule time of the request to
2988          * spread the inode table initialization requests
2989          * better.
2990          */
2991         get_random_bytes(&rnd, sizeof(rnd));
2992         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2993                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2994
2995         return elr;
2996 }
2997
2998 static int ext4_register_li_request(struct super_block *sb,
2999                                     ext4_group_t first_not_zeroed)
3000 {
3001         struct ext4_sb_info *sbi = EXT4_SB(sb);
3002         struct ext4_li_request *elr;
3003         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3004         int ret = 0;
3005
3006         if (sbi->s_li_request != NULL) {
3007                 /*
3008                  * Reset timeout so it can be computed again, because
3009                  * s_li_wait_mult might have changed.
3010                  */
3011                 sbi->s_li_request->lr_timeout = 0;
3012                 return 0;
3013         }
3014
3015         if (first_not_zeroed == ngroups ||
3016             (sb->s_flags & MS_RDONLY) ||
3017             !test_opt(sb, INIT_INODE_TABLE))
3018                 return 0;
3019
3020         elr = ext4_li_request_new(sb, first_not_zeroed);
3021         if (!elr)
3022                 return -ENOMEM;
3023
3024         mutex_lock(&ext4_li_mtx);
3025
3026         if (NULL == ext4_li_info) {
3027                 ret = ext4_li_info_new();
3028                 if (ret)
3029                         goto out;
3030         }
3031
3032         mutex_lock(&ext4_li_info->li_list_mtx);
3033         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3034         mutex_unlock(&ext4_li_info->li_list_mtx);
3035
3036         sbi->s_li_request = elr;
3037         /*
3038          * set elr to NULL here since it has been inserted to
3039          * the request_list and the removal and free of it is
3040          * handled by ext4_clear_request_list from now on.
3041          */
3042         elr = NULL;
3043
3044         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3045                 ret = ext4_run_lazyinit_thread();
3046                 if (ret)
3047                         goto out;
3048         }
3049 out:
3050         mutex_unlock(&ext4_li_mtx);
3051         if (ret)
3052                 kfree(elr);
3053         return ret;
3054 }
3055
3056 /*
3057  * We do not need to lock anything since this is called on
3058  * module unload.
3059  */
3060 static void ext4_destroy_lazyinit_thread(void)
3061 {
3062         /*
3063          * If thread exited earlier
3064          * there's nothing to be done.
3065          */
3066         if (!ext4_li_info || !ext4_lazyinit_task)
3067                 return;
3068
3069         kthread_stop(ext4_lazyinit_task);
3070 }
3071
3072 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3073                                 __releases(kernel_lock)
3074                                 __acquires(kernel_lock)
3075 {
3076         char *orig_data = kstrdup(data, GFP_KERNEL);
3077         struct buffer_head *bh;
3078         struct ext4_super_block *es = NULL;
3079         struct ext4_sb_info *sbi;
3080         ext4_fsblk_t block;
3081         ext4_fsblk_t sb_block = get_sb_block(&data);
3082         ext4_fsblk_t logical_sb_block;
3083         unsigned long offset = 0;
3084         unsigned long journal_devnum = 0;
3085         unsigned long def_mount_opts;
3086         struct inode *root;
3087         char *cp;
3088         const char *descr;
3089         int ret = -ENOMEM;
3090         int blocksize;
3091         unsigned int db_count;
3092         unsigned int i;
3093         int needs_recovery, has_huge_files;
3094         __u64 blocks_count;
3095         int err;
3096         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3097         ext4_group_t first_not_zeroed;
3098
3099         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3100         if (!sbi)
3101                 goto out_free_orig;
3102
3103         sbi->s_blockgroup_lock =
3104                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3105         if (!sbi->s_blockgroup_lock) {
3106                 kfree(sbi);
3107                 goto out_free_orig;
3108         }
3109         sb->s_fs_info = sbi;
3110         sbi->s_mount_opt = 0;
3111         sbi->s_resuid = EXT4_DEF_RESUID;
3112         sbi->s_resgid = EXT4_DEF_RESGID;
3113         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3114         sbi->s_sb_block = sb_block;
3115         if (sb->s_bdev->bd_part)
3116                 sbi->s_sectors_written_start =
3117                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3118
3119         /* Cleanup superblock name */
3120         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3121                 *cp = '!';
3122
3123         ret = -EINVAL;
3124         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3125         if (!blocksize) {
3126                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3127                 goto out_fail;
3128         }
3129
3130         /*
3131          * The ext4 superblock will not be buffer aligned for other than 1kB
3132          * block sizes.  We need to calculate the offset from buffer start.
3133          */
3134         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3135                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3136                 offset = do_div(logical_sb_block, blocksize);
3137         } else {
3138                 logical_sb_block = sb_block;
3139         }
3140
3141         if (!(bh = sb_bread(sb, logical_sb_block))) {
3142                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3143                 goto out_fail;
3144         }
3145         /*
3146          * Note: s_es must be initialized as soon as possible because
3147          *       some ext4 macro-instructions depend on its value
3148          */
3149         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3150         sbi->s_es = es;
3151         sb->s_magic = le16_to_cpu(es->s_magic);
3152         if (sb->s_magic != EXT4_SUPER_MAGIC)
3153                 goto cantfind_ext4;
3154         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3155
3156         /* Set defaults before we parse the mount options */
3157         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3158         set_opt(sb, INIT_INODE_TABLE);
3159         if (def_mount_opts & EXT4_DEFM_DEBUG)
3160                 set_opt(sb, DEBUG);
3161         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3162                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3163                         "2.6.38");
3164                 set_opt(sb, GRPID);
3165         }
3166         if (def_mount_opts & EXT4_DEFM_UID16)
3167                 set_opt(sb, NO_UID32);
3168         /* xattr user namespace & acls are now defaulted on */
3169 #ifdef CONFIG_EXT4_FS_XATTR
3170         set_opt(sb, XATTR_USER);
3171 #endif
3172 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3173         set_opt(sb, POSIX_ACL);
3174 #endif
3175         set_opt(sb, MBLK_IO_SUBMIT);
3176         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3177                 set_opt(sb, JOURNAL_DATA);
3178         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3179                 set_opt(sb, ORDERED_DATA);
3180         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3181                 set_opt(sb, WRITEBACK_DATA);
3182
3183         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3184                 set_opt(sb, ERRORS_PANIC);
3185         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3186                 set_opt(sb, ERRORS_CONT);
3187         else
3188                 set_opt(sb, ERRORS_RO);
3189         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3190                 set_opt(sb, BLOCK_VALIDITY);
3191         if (def_mount_opts & EXT4_DEFM_DISCARD)
3192                 set_opt(sb, DISCARD);
3193
3194         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3195         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3196         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3197         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3198         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3199
3200         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3201                 set_opt(sb, BARRIER);
3202
3203         /*
3204          * enable delayed allocation by default
3205          * Use -o nodelalloc to turn it off
3206          */
3207         if (!IS_EXT3_SB(sb) &&
3208             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3209                 set_opt(sb, DELALLOC);
3210
3211         /*
3212          * set default s_li_wait_mult for lazyinit, for the case there is
3213          * no mount option specified.
3214          */
3215         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3216
3217         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3218                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3219                 ext4_msg(sb, KERN_WARNING,
3220                          "failed to parse options in superblock: %s",
3221                          sbi->s_es->s_mount_opts);
3222         }
3223         if (!parse_options((char *) data, sb, &journal_devnum,
3224                            &journal_ioprio, NULL, 0))
3225                 goto failed_mount;
3226
3227         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3228                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3229
3230         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3231             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3232              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3233              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3234                 ext4_msg(sb, KERN_WARNING,
3235                        "feature flags set on rev 0 fs, "
3236                        "running e2fsck is recommended");
3237
3238         if (IS_EXT2_SB(sb)) {
3239                 if (ext2_feature_set_ok(sb))
3240                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3241                                  "using the ext4 subsystem");
3242                 else {
3243                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3244                                  "to feature incompatibilities");
3245                         goto failed_mount;
3246                 }
3247         }
3248
3249         if (IS_EXT3_SB(sb)) {
3250                 if (ext3_feature_set_ok(sb))
3251                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3252                                  "using the ext4 subsystem");
3253                 else {
3254                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3255                                  "to feature incompatibilities");
3256                         goto failed_mount;
3257                 }
3258         }
3259
3260         /*
3261          * Check feature flags regardless of the revision level, since we
3262          * previously didn't change the revision level when setting the flags,
3263          * so there is a chance incompat flags are set on a rev 0 filesystem.
3264          */
3265         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3266                 goto failed_mount;
3267
3268         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3269
3270         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3271             blocksize > EXT4_MAX_BLOCK_SIZE) {
3272                 ext4_msg(sb, KERN_ERR,
3273                        "Unsupported filesystem blocksize %d", blocksize);
3274                 goto failed_mount;
3275         }
3276
3277         if (sb->s_blocksize != blocksize) {
3278                 /* Validate the filesystem blocksize */
3279                 if (!sb_set_blocksize(sb, blocksize)) {
3280                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3281                                         blocksize);
3282                         goto failed_mount;
3283                 }
3284
3285                 brelse(bh);
3286                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3287                 offset = do_div(logical_sb_block, blocksize);
3288                 bh = sb_bread(sb, logical_sb_block);
3289                 if (!bh) {
3290                         ext4_msg(sb, KERN_ERR,
3291                                "Can't read superblock on 2nd try");
3292                         goto failed_mount;
3293                 }
3294                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3295                 sbi->s_es = es;
3296                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3297                         ext4_msg(sb, KERN_ERR,
3298                                "Magic mismatch, very weird!");
3299                         goto failed_mount;
3300                 }
3301         }
3302
3303         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3304                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3305         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3306                                                       has_huge_files);
3307         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3308
3309         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3310                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3311                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3312         } else {
3313                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3314                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3315                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3316                     (!is_power_of_2(sbi->s_inode_size)) ||
3317                     (sbi->s_inode_size > blocksize)) {
3318                         ext4_msg(sb, KERN_ERR,
3319                                "unsupported inode size: %d",
3320                                sbi->s_inode_size);
3321                         goto failed_mount;
3322                 }
3323                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3324                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3325         }
3326
3327         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3328         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3329                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3330                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3331                     !is_power_of_2(sbi->s_desc_size)) {
3332                         ext4_msg(sb, KERN_ERR,
3333                                "unsupported descriptor size %lu",
3334                                sbi->s_desc_size);
3335                         goto failed_mount;
3336                 }
3337         } else
3338                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3339
3340         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3341         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3342         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3343                 goto cantfind_ext4;
3344
3345         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3346         if (sbi->s_inodes_per_block == 0)
3347                 goto cantfind_ext4;
3348         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3349                                         sbi->s_inodes_per_block;
3350         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3351         sbi->s_sbh = bh;
3352         sbi->s_mount_state = le16_to_cpu(es->s_state);
3353         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3354         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3355
3356         for (i = 0; i < 4; i++)
3357                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3358         sbi->s_def_hash_version = es->s_def_hash_version;
3359         i = le32_to_cpu(es->s_flags);
3360         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3361                 sbi->s_hash_unsigned = 3;
3362         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3363 #ifdef __CHAR_UNSIGNED__
3364                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3365                 sbi->s_hash_unsigned = 3;
3366 #else
3367                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3368 #endif
3369                 sb->s_dirt = 1;
3370         }
3371
3372         if (sbi->s_blocks_per_group > blocksize * 8) {
3373                 ext4_msg(sb, KERN_ERR,
3374                        "#blocks per group too big: %lu",
3375                        sbi->s_blocks_per_group);
3376                 goto failed_mount;
3377         }
3378         if (sbi->s_inodes_per_group > blocksize * 8) {
3379                 ext4_msg(sb, KERN_ERR,
3380                        "#inodes per group too big: %lu",
3381                        sbi->s_inodes_per_group);
3382                 goto failed_mount;
3383         }
3384
3385         /*
3386          * Test whether we have more sectors than will fit in sector_t,
3387          * and whether the max offset is addressable by the page cache.
3388          */
3389         err = generic_check_addressable(sb->s_blocksize_bits,
3390                                         ext4_blocks_count(es));
3391         if (err) {
3392                 ext4_msg(sb, KERN_ERR, "filesystem"
3393                          " too large to mount safely on this system");
3394                 if (sizeof(sector_t) < 8)
3395                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3396                 ret = err;
3397                 goto failed_mount;
3398         }
3399
3400         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3401                 goto cantfind_ext4;
3402
3403         /* check blocks count against device size */
3404         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3405         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3406                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3407                        "exceeds size of device (%llu blocks)",
3408                        ext4_blocks_count(es), blocks_count);
3409                 goto failed_mount;
3410         }
3411
3412         /*
3413          * It makes no sense for the first data block to be beyond the end
3414          * of the filesystem.
3415          */
3416         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3417                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3418                          "block %u is beyond end of filesystem (%llu)",
3419                          le32_to_cpu(es->s_first_data_block),
3420                          ext4_blocks_count(es));
3421                 goto failed_mount;
3422         }
3423         blocks_count = (ext4_blocks_count(es) -
3424                         le32_to_cpu(es->s_first_data_block) +
3425                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3426         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3427         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3428                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3429                        "(block count %llu, first data block %u, "
3430                        "blocks per group %lu)", sbi->s_groups_count,
3431                        ext4_blocks_count(es),
3432                        le32_to_cpu(es->s_first_data_block),
3433                        EXT4_BLOCKS_PER_GROUP(sb));
3434                 goto failed_mount;
3435         }
3436         sbi->s_groups_count = blocks_count;
3437         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3438                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3439         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3440                    EXT4_DESC_PER_BLOCK(sb);
3441         sbi->s_group_desc = ext4_kvmalloc(db_count *
3442                                           sizeof(struct buffer_head *),
3443                                           GFP_KERNEL);
3444         if (sbi->s_group_desc == NULL) {
3445                 ext4_msg(sb, KERN_ERR, "not enough memory");
3446                 goto failed_mount;
3447         }
3448
3449 #ifdef CONFIG_PROC_FS
3450         if (ext4_proc_root)
3451                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3452 #endif
3453
3454         bgl_lock_init(sbi->s_blockgroup_lock);
3455
3456         for (i = 0; i < db_count; i++) {
3457                 block = descriptor_loc(sb, logical_sb_block, i);
3458                 sbi->s_group_desc[i] = sb_bread(sb, block);
3459                 if (!sbi->s_group_desc[i]) {
3460                         ext4_msg(sb, KERN_ERR,
3461                                "can't read group descriptor %d", i);
3462                         db_count = i;
3463                         goto failed_mount2;
3464                 }
3465         }
3466         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3467                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3468                 goto failed_mount2;
3469         }
3470         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3471                 if (!ext4_fill_flex_info(sb)) {
3472                         ext4_msg(sb, KERN_ERR,
3473                                "unable to initialize "
3474                                "flex_bg meta info!");
3475                         goto failed_mount2;
3476                 }
3477
3478         sbi->s_gdb_count = db_count;
3479         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3480         spin_lock_init(&sbi->s_next_gen_lock);
3481
3482         init_timer(&sbi->s_err_report);
3483         sbi->s_err_report.function = print_daily_error_info;
3484         sbi->s_err_report.data = (unsigned long) sb;
3485
3486         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3487                         ext4_count_free_blocks(sb));
3488         if (!err) {
3489                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3490                                 ext4_count_free_inodes(sb));
3491         }
3492         if (!err) {
3493                 err = percpu_counter_init(&sbi->s_dirs_counter,
3494                                 ext4_count_dirs(sb));
3495         }
3496         if (!err) {
3497                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3498         }
3499         if (err) {
3500                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3501                 goto failed_mount3;
3502         }
3503
3504         sbi->s_stripe = ext4_get_stripe_size(sbi);
3505         sbi->s_max_writeback_mb_bump = 128;
3506
3507         /*
3508          * set up enough so that it can read an inode
3509          */
3510         if (!test_opt(sb, NOLOAD) &&
3511             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3512                 sb->s_op = &ext4_sops;
3513         else
3514                 sb->s_op = &ext4_nojournal_sops;
3515         sb->s_export_op = &ext4_export_ops;
3516         sb->s_xattr = ext4_xattr_handlers;
3517 #ifdef CONFIG_QUOTA
3518         sb->s_qcop = &ext4_qctl_operations;
3519         sb->dq_op = &ext4_quota_operations;
3520 #endif
3521         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3522
3523         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3524         mutex_init(&sbi->s_orphan_lock);
3525         sbi->s_resize_flags = 0;
3526
3527         sb->s_root = NULL;
3528
3529         needs_recovery = (es->s_last_orphan != 0 ||
3530                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3531                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3532
3533         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3534             !(sb->s_flags & MS_RDONLY))
3535                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3536                         goto failed_mount3;
3537
3538         /*
3539          * The first inode we look at is the journal inode.  Don't try
3540          * root first: it may be modified in the journal!
3541          */
3542         if (!test_opt(sb, NOLOAD) &&
3543             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3544                 if (ext4_load_journal(sb, es, journal_devnum))
3545                         goto failed_mount3;
3546         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3547               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3548                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3549                        "suppressed and not mounted read-only");
3550                 goto failed_mount_wq;
3551         } else {
3552                 clear_opt(sb, DATA_FLAGS);
3553                 sbi->s_journal = NULL;
3554                 needs_recovery = 0;
3555                 goto no_journal;
3556         }
3557
3558         if (ext4_blocks_count(es) > 0xffffffffULL &&
3559             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3560                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3561                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3562                 goto failed_mount_wq;
3563         }
3564
3565         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3566                 jbd2_journal_set_features(sbi->s_journal,
3567                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3568                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3569         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3570                 jbd2_journal_set_features(sbi->s_journal,
3571                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3572                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3573                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3574         } else {
3575                 jbd2_journal_clear_features(sbi->s_journal,
3576                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3577                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3578         }
3579
3580         /* We have now updated the journal if required, so we can
3581          * validate the data journaling mode. */
3582         switch (test_opt(sb, DATA_FLAGS)) {
3583         case 0:
3584                 /* No mode set, assume a default based on the journal
3585                  * capabilities: ORDERED_DATA if the journal can
3586                  * cope, else JOURNAL_DATA
3587                  */
3588                 if (jbd2_journal_check_available_features
3589                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3590                         set_opt(sb, ORDERED_DATA);
3591                 else
3592                         set_opt(sb, JOURNAL_DATA);
3593                 break;
3594
3595         case EXT4_MOUNT_ORDERED_DATA:
3596         case EXT4_MOUNT_WRITEBACK_DATA:
3597                 if (!jbd2_journal_check_available_features
3598                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3599                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3600                                "requested data journaling mode");
3601                         goto failed_mount_wq;
3602                 }
3603         default:
3604                 break;
3605         }
3606         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3607
3608         /*
3609          * The journal may have updated the bg summary counts, so we
3610          * need to update the global counters.
3611          */
3612         percpu_counter_set(&sbi->s_freeblocks_counter,
3613                            ext4_count_free_blocks(sb));
3614         percpu_counter_set(&sbi->s_freeinodes_counter,
3615                            ext4_count_free_inodes(sb));
3616         percpu_counter_set(&sbi->s_dirs_counter,
3617                            ext4_count_dirs(sb));
3618         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3619
3620 no_journal:
3621         /*
3622          * The maximum number of concurrent works can be high and
3623          * concurrency isn't really necessary.  Limit it to 1.
3624          */
3625         EXT4_SB(sb)->dio_unwritten_wq =
3626                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3627         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3628                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3629                 goto failed_mount_wq;
3630         }
3631
3632         /*
3633          * The jbd2_journal_load will have done any necessary log recovery,
3634          * so we can safely mount the rest of the filesystem now.
3635          */
3636
3637         root = ext4_iget(sb, EXT4_ROOT_INO);
3638         if (IS_ERR(root)) {
3639                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3640                 ret = PTR_ERR(root);
3641                 root = NULL;
3642                 goto failed_mount4;
3643         }
3644         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3645                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3646                 goto failed_mount4;
3647         }
3648         sb->s_root = d_alloc_root(root);
3649         if (!sb->s_root) {
3650                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3651                 ret = -ENOMEM;
3652                 goto failed_mount4;
3653         }
3654
3655         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3656
3657         /* determine the minimum size of new large inodes, if present */
3658         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3659                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3660                                                      EXT4_GOOD_OLD_INODE_SIZE;
3661                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3662                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3663                         if (sbi->s_want_extra_isize <
3664                             le16_to_cpu(es->s_want_extra_isize))
3665                                 sbi->s_want_extra_isize =
3666                                         le16_to_cpu(es->s_want_extra_isize);
3667                         if (sbi->s_want_extra_isize <
3668                             le16_to_cpu(es->s_min_extra_isize))
3669                                 sbi->s_want_extra_isize =
3670                                         le16_to_cpu(es->s_min_extra_isize);
3671                 }
3672         }
3673         /* Check if enough inode space is available */
3674         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3675                                                         sbi->s_inode_size) {
3676                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3677                                                        EXT4_GOOD_OLD_INODE_SIZE;
3678                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3679                          "available");
3680         }
3681
3682         if (test_opt(sb, DELALLOC) &&
3683             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3684                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3685                          "requested data journaling mode");
3686                 clear_opt(sb, DELALLOC);
3687         }
3688         if (test_opt(sb, DIOREAD_NOLOCK)) {
3689                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3690                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3691                                 "option - requested data journaling mode");
3692                         clear_opt(sb, DIOREAD_NOLOCK);
3693                 }
3694                 if (sb->s_blocksize < PAGE_SIZE) {
3695                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3696                                 "option - block size is too small");
3697                         clear_opt(sb, DIOREAD_NOLOCK);
3698                 }
3699         }
3700
3701         err = ext4_setup_system_zone(sb);
3702         if (err) {
3703                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3704                          "zone (%d)", err);
3705                 goto failed_mount4;
3706         }
3707
3708         ext4_ext_init(sb);
3709         err = ext4_mb_init(sb, needs_recovery);
3710         if (err) {
3711                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3712                          err);
3713                 goto failed_mount4;
3714         }
3715
3716         err = ext4_register_li_request(sb, first_not_zeroed);
3717         if (err)
3718                 goto failed_mount4;
3719
3720         sbi->s_kobj.kset = ext4_kset;
3721         init_completion(&sbi->s_kobj_unregister);
3722         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3723                                    "%s", sb->s_id);
3724         if (err) {
3725                 ext4_mb_release(sb);
3726                 ext4_ext_release(sb);
3727                 goto failed_mount4;
3728         };
3729
3730         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3731         ext4_orphan_cleanup(sb, es);
3732         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3733         if (needs_recovery) {
3734                 ext4_msg(sb, KERN_INFO, "recovery complete");
3735                 ext4_mark_recovery_complete(sb, es);
3736         }
3737         if (EXT4_SB(sb)->s_journal) {
3738                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3739                         descr = " journalled data mode";
3740                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3741                         descr = " ordered data mode";
3742                 else
3743                         descr = " writeback data mode";
3744         } else
3745                 descr = "out journal";
3746
3747         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3748                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3749                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3750
3751         if (es->s_error_count)
3752                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3753
3754         kfree(orig_data);
3755         return 0;
3756
3757 cantfind_ext4:
3758         if (!silent)
3759                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3760         goto failed_mount;
3761
3762 failed_mount4:
3763         iput(root);
3764         sb->s_root = NULL;
3765         ext4_msg(sb, KERN_ERR, "mount failed");
3766         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3767 failed_mount_wq:
3768         ext4_release_system_zone(sb);
3769         if (sbi->s_journal) {
3770                 jbd2_journal_destroy(sbi->s_journal);
3771                 sbi->s_journal = NULL;
3772         }
3773 failed_mount3:
3774         del_timer(&sbi->s_err_report);
3775         if (sbi->s_flex_groups)
3776                 ext4_kvfree(sbi->s_flex_groups);
3777         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3778         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3779         percpu_counter_destroy(&sbi->s_dirs_counter);
3780         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3781         if (sbi->s_mmp_tsk)
3782                 kthread_stop(sbi->s_mmp_tsk);
3783 failed_mount2:
3784         for (i = 0; i < db_count; i++)
3785                 brelse(sbi->s_group_desc[i]);
3786         ext4_kvfree(sbi->s_group_desc);
3787 failed_mount:
3788         if (sbi->s_proc) {
3789                 remove_proc_entry(sb->s_id, ext4_proc_root);
3790         }
3791 #ifdef CONFIG_QUOTA
3792         for (i = 0; i < MAXQUOTAS; i++)
3793                 kfree(sbi->s_qf_names[i]);
3794 #endif
3795         ext4_blkdev_remove(sbi);
3796         brelse(bh);
3797 out_fail:
3798         sb->s_fs_info = NULL;
3799         kfree(sbi->s_blockgroup_lock);
3800         kfree(sbi);
3801 out_free_orig:
3802         kfree(orig_data);
3803         return ret;
3804 }
3805
3806 /*
3807  * Setup any per-fs journal parameters now.  We'll do this both on
3808  * initial mount, once the journal has been initialised but before we've
3809  * done any recovery; and again on any subsequent remount.
3810  */
3811 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3812 {
3813         struct ext4_sb_info *sbi = EXT4_SB(sb);
3814
3815         journal->j_commit_interval = sbi->s_commit_interval;
3816         journal->j_min_batch_time = sbi->s_min_batch_time;
3817         journal->j_max_batch_time = sbi->s_max_batch_time;
3818
3819         write_lock(&journal->j_state_lock);
3820         if (test_opt(sb, BARRIER))
3821                 journal->j_flags |= JBD2_BARRIER;
3822         else
3823                 journal->j_flags &= ~JBD2_BARRIER;
3824         if (test_opt(sb, DATA_ERR_ABORT))
3825                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3826         else
3827                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3828         write_unlock(&journal->j_state_lock);
3829 }
3830
3831 static journal_t *ext4_get_journal(struct super_block *sb,
3832                                    unsigned int journal_inum)
3833 {
3834         struct inode *journal_inode;
3835         journal_t *journal;
3836
3837         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3838
3839         /* First, test for the existence of a valid inode on disk.  Bad
3840          * things happen if we iget() an unused inode, as the subsequent
3841          * iput() will try to delete it. */
3842
3843         journal_inode = ext4_iget(sb, journal_inum);
3844         if (IS_ERR(journal_inode)) {
3845                 ext4_msg(sb, KERN_ERR, "no journal found");
3846                 return NULL;
3847         }
3848         if (!journal_inode->i_nlink) {
3849                 make_bad_inode(journal_inode);
3850                 iput(journal_inode);
3851                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3852                 return NULL;
3853         }
3854
3855         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3856                   journal_inode, journal_inode->i_size);
3857         if (!S_ISREG(journal_inode->i_mode)) {
3858                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3859                 iput(journal_inode);
3860                 return NULL;
3861         }
3862
3863         journal = jbd2_journal_init_inode(journal_inode);
3864         if (!journal) {
3865                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3866                 iput(journal_inode);
3867                 return NULL;
3868         }
3869         journal->j_private = sb;
3870         ext4_init_journal_params(sb, journal);
3871         return journal;
3872 }
3873
3874 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3875                                        dev_t j_dev)
3876 {
3877         struct buffer_head *bh;
3878         journal_t *journal;
3879         ext4_fsblk_t start;
3880         ext4_fsblk_t len;
3881         int hblock, blocksize;
3882         ext4_fsblk_t sb_block;
3883         unsigned long offset;
3884         struct ext4_super_block *es;
3885         struct block_device *bdev;
3886
3887         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3888
3889         bdev = ext4_blkdev_get(j_dev, sb);
3890         if (bdev == NULL)
3891                 return NULL;
3892
3893         blocksize = sb->s_blocksize;
3894         hblock = bdev_logical_block_size(bdev);
3895         if (blocksize < hblock) {
3896                 ext4_msg(sb, KERN_ERR,
3897                         "blocksize too small for journal device");
3898                 goto out_bdev;
3899         }
3900
3901         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3902         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3903         set_blocksize(bdev, blocksize);
3904         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3905                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3906                        "external journal");
3907                 goto out_bdev;
3908         }
3909
3910         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3911         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3912             !(le32_to_cpu(es->s_feature_incompat) &
3913               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3914                 ext4_msg(sb, KERN_ERR, "external journal has "
3915                                         "bad superblock");
3916                 brelse(bh);
3917                 goto out_bdev;
3918         }
3919
3920         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3921                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3922                 brelse(bh);
3923                 goto out_bdev;
3924         }
3925
3926         len = ext4_blocks_count(es);
3927         start = sb_block + 1;
3928         brelse(bh);     /* we're done with the superblock */
3929
3930         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3931                                         start, len, blocksize);
3932         if (!journal) {
3933                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3934                 goto out_bdev;
3935         }
3936         journal->j_private = sb;
3937         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3938         wait_on_buffer(journal->j_sb_buffer);
3939         if (!buffer_uptodate(journal->j_sb_buffer)) {
3940                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3941                 goto out_journal;
3942         }
3943         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3944                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3945                                         "user (unsupported) - %d",
3946                         be32_to_cpu(journal->j_superblock->s_nr_users));
3947                 goto out_journal;
3948         }
3949         EXT4_SB(sb)->journal_bdev = bdev;
3950         ext4_init_journal_params(sb, journal);
3951         return journal;
3952
3953 out_journal:
3954         jbd2_journal_destroy(journal);
3955 out_bdev:
3956         ext4_blkdev_put(bdev);
3957         return NULL;
3958 }
3959
3960 static int ext4_load_journal(struct super_block *sb,
3961                              struct ext4_super_block *es,
3962                              unsigned long journal_devnum)
3963 {
3964         journal_t *journal;
3965         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3966         dev_t journal_dev;
3967         int err = 0;
3968         int really_read_only;
3969
3970         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3971
3972         if (journal_devnum &&
3973             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3974                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3975                         "numbers have changed");
3976                 journal_dev = new_decode_dev(journal_devnum);
3977         } else
3978                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3979
3980         really_read_only = bdev_read_only(sb->s_bdev);
3981
3982         /*
3983          * Are we loading a blank journal or performing recovery after a
3984          * crash?  For recovery, we need to check in advance whether we
3985          * can get read-write access to the device.
3986          */
3987         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3988                 if (sb->s_flags & MS_RDONLY) {
3989                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3990                                         "required on readonly filesystem");
3991                         if (really_read_only) {
3992                                 ext4_msg(sb, KERN_ERR, "write access "
3993                                         "unavailable, cannot proceed");
3994                                 return -EROFS;
3995                         }
3996                         ext4_msg(sb, KERN_INFO, "write access will "
3997                                "be enabled during recovery");
3998                 }
3999         }
4000
4001         if (journal_inum && journal_dev) {
4002                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4003                        "and inode journals!");
4004                 return -EINVAL;
4005         }
4006
4007         if (journal_inum) {
4008                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4009                         return -EINVAL;
4010         } else {
4011                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4012                         return -EINVAL;
4013         }
4014
4015         if (!(journal->j_flags & JBD2_BARRIER))
4016                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4017
4018         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4019                 err = jbd2_journal_update_format(journal);
4020                 if (err)  {
4021                         ext4_msg(sb, KERN_ERR, "error updating journal");
4022                         jbd2_journal_destroy(journal);
4023                         return err;
4024                 }
4025         }
4026
4027         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4028                 err = jbd2_journal_wipe(journal, !really_read_only);
4029         if (!err) {
4030                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4031                 if (save)
4032                         memcpy(save, ((char *) es) +
4033                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4034                 err = jbd2_journal_load(journal);
4035                 if (save)
4036                         memcpy(((char *) es) + EXT4_S_ERR_START,
4037                                save, EXT4_S_ERR_LEN);
4038                 kfree(save);
4039         }
4040
4041         if (err) {
4042                 ext4_msg(sb, KERN_ERR, "error loading journal");
4043                 jbd2_journal_destroy(journal);
4044                 return err;
4045         }
4046
4047         EXT4_SB(sb)->s_journal = journal;
4048         ext4_clear_journal_err(sb, es);
4049
4050         if (!really_read_only && journal_devnum &&
4051             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4052                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4053
4054                 /* Make sure we flush the recovery flag to disk. */
4055                 ext4_commit_super(sb, 1);
4056         }
4057
4058         return 0;
4059 }
4060
4061 static int ext4_commit_super(struct super_block *sb, int sync)
4062 {
4063         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4064         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4065         int error = 0;
4066
4067         if (!sbh)
4068                 return error;
4069         if (buffer_write_io_error(sbh)) {
4070                 /*
4071                  * Oh, dear.  A previous attempt to write the
4072                  * superblock failed.  This could happen because the
4073                  * USB device was yanked out.  Or it could happen to
4074                  * be a transient write error and maybe the block will
4075                  * be remapped.  Nothing we can do but to retry the
4076                  * write and hope for the best.
4077                  */
4078                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4079                        "superblock detected");
4080                 clear_buffer_write_io_error(sbh);
4081                 set_buffer_uptodate(sbh);
4082         }
4083         /*
4084          * If the file system is mounted read-only, don't update the
4085          * superblock write time.  This avoids updating the superblock
4086          * write time when we are mounting the root file system
4087          * read/only but we need to replay the journal; at that point,
4088          * for people who are east of GMT and who make their clock
4089          * tick in localtime for Windows bug-for-bug compatibility,
4090          * the clock is set in the future, and this will cause e2fsck
4091          * to complain and force a full file system check.
4092          */
4093         if (!(sb->s_flags & MS_RDONLY))
4094                 es->s_wtime = cpu_to_le32(get_seconds());
4095         if (sb->s_bdev->bd_part)
4096                 es->s_kbytes_written =
4097                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4098                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4099                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4100         else
4101                 es->s_kbytes_written =
4102                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4103         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4104                                            &EXT4_SB(sb)->s_freeblocks_counter));
4105         es->s_free_inodes_count =
4106                 cpu_to_le32(percpu_counter_sum_positive(
4107                                 &EXT4_SB(sb)->s_freeinodes_counter));
4108         sb->s_dirt = 0;
4109         BUFFER_TRACE(sbh, "marking dirty");
4110         mark_buffer_dirty(sbh);
4111         if (sync) {
4112                 error = sync_dirty_buffer(sbh);
4113                 if (error)
4114                         return error;
4115
4116                 error = buffer_write_io_error(sbh);
4117                 if (error) {
4118                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4119                                "superblock");
4120                         clear_buffer_write_io_error(sbh);
4121                         set_buffer_uptodate(sbh);
4122                 }
4123         }
4124         return error;
4125 }
4126
4127 /*
4128  * Have we just finished recovery?  If so, and if we are mounting (or
4129  * remounting) the filesystem readonly, then we will end up with a
4130  * consistent fs on disk.  Record that fact.
4131  */
4132 static void ext4_mark_recovery_complete(struct super_block *sb,
4133                                         struct ext4_super_block *es)
4134 {
4135         journal_t *journal = EXT4_SB(sb)->s_journal;
4136
4137         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4138                 BUG_ON(journal != NULL);
4139                 return;
4140         }
4141         jbd2_journal_lock_updates(journal);
4142         if (jbd2_journal_flush(journal) < 0)
4143                 goto out;
4144
4145         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4146             sb->s_flags & MS_RDONLY) {
4147                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4148                 ext4_commit_super(sb, 1);
4149         }
4150
4151 out:
4152         jbd2_journal_unlock_updates(journal);
4153 }
4154
4155 /*
4156  * If we are mounting (or read-write remounting) a filesystem whose journal
4157  * has recorded an error from a previous lifetime, move that error to the
4158  * main filesystem now.
4159  */
4160 static void ext4_clear_journal_err(struct super_block *sb,
4161                                    struct ext4_super_block *es)
4162 {
4163         journal_t *journal;
4164         int j_errno;
4165         const char *errstr;
4166
4167         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4168
4169         journal = EXT4_SB(sb)->s_journal;
4170
4171         /*
4172          * Now check for any error status which may have been recorded in the
4173          * journal by a prior ext4_error() or ext4_abort()
4174          */
4175
4176         j_errno = jbd2_journal_errno(journal);
4177         if (j_errno) {
4178                 char nbuf[16];
4179
4180                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4181                 ext4_warning(sb, "Filesystem error recorded "
4182                              "from previous mount: %s", errstr);
4183                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4184
4185                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4186                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4187                 ext4_commit_super(sb, 1);
4188
4189                 jbd2_journal_clear_err(journal);
4190         }
4191 }
4192
4193 /*
4194  * Force the running and committing transactions to commit,
4195  * and wait on the commit.
4196  */
4197 int ext4_force_commit(struct super_block *sb)
4198 {
4199         journal_t *journal;
4200         int ret = 0;
4201
4202         if (sb->s_flags & MS_RDONLY)
4203                 return 0;
4204
4205         journal = EXT4_SB(sb)->s_journal;
4206         if (journal) {
4207                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4208                 ret = ext4_journal_force_commit(journal);
4209         }
4210
4211         return ret;
4212 }
4213
4214 static void ext4_write_super(struct super_block *sb)
4215 {
4216         lock_super(sb);
4217         ext4_commit_super(sb, 1);
4218         unlock_super(sb);
4219 }
4220
4221 static int ext4_sync_fs(struct super_block *sb, int wait)
4222 {
4223         int ret = 0;
4224         tid_t target;
4225         struct ext4_sb_info *sbi = EXT4_SB(sb);
4226
4227         trace_ext4_sync_fs(sb, wait);
4228         flush_workqueue(sbi->dio_unwritten_wq);
4229         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4230                 if (wait)
4231                         jbd2_log_wait_commit(sbi->s_journal, target);
4232         }
4233         return ret;
4234 }
4235
4236 /*
4237  * LVM calls this function before a (read-only) snapshot is created.  This
4238  * gives us a chance to flush the journal completely and mark the fs clean.
4239  *
4240  * Note that only this function cannot bring a filesystem to be in a clean
4241  * state independently, because ext4 prevents a new handle from being started
4242  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4243  * the upper layer.
4244  */
4245 static int ext4_freeze(struct super_block *sb)
4246 {
4247         int error = 0;
4248         journal_t *journal;
4249
4250         if (sb->s_flags & MS_RDONLY)
4251                 return 0;
4252
4253         journal = EXT4_SB(sb)->s_journal;
4254
4255         /* Now we set up the journal barrier. */
4256         jbd2_journal_lock_updates(journal);
4257
4258         /*
4259          * Don't clear the needs_recovery flag if we failed to flush
4260          * the journal.
4261          */
4262         error = jbd2_journal_flush(journal);
4263         if (error < 0)
4264                 goto out;
4265
4266         /* Journal blocked and flushed, clear needs_recovery flag. */
4267         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4268         error = ext4_commit_super(sb, 1);
4269 out:
4270         /* we rely on s_frozen to stop further updates */
4271         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4272         return error;
4273 }
4274
4275 /*
4276  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4277  * flag here, even though the filesystem is not technically dirty yet.
4278  */
4279 static int ext4_unfreeze(struct super_block *sb)
4280 {
4281         if (sb->s_flags & MS_RDONLY)
4282                 return 0;
4283
4284         lock_super(sb);
4285         /* Reset the needs_recovery flag before the fs is unlocked. */
4286         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4287         ext4_commit_super(sb, 1);
4288         unlock_super(sb);
4289         return 0;
4290 }
4291
4292 /*
4293  * Structure to save mount options for ext4_remount's benefit
4294  */
4295 struct ext4_mount_options {
4296         unsigned long s_mount_opt;
4297         unsigned long s_mount_opt2;
4298         uid_t s_resuid;
4299         gid_t s_resgid;
4300         unsigned long s_commit_interval;
4301         u32 s_min_batch_time, s_max_batch_time;
4302 #ifdef CONFIG_QUOTA
4303         int s_jquota_fmt;
4304         char *s_qf_names[MAXQUOTAS];
4305 #endif
4306 };
4307
4308 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4309 {
4310         struct ext4_super_block *es;
4311         struct ext4_sb_info *sbi = EXT4_SB(sb);
4312         ext4_fsblk_t n_blocks_count = 0;
4313         unsigned long old_sb_flags;
4314         struct ext4_mount_options old_opts;
4315         int enable_quota = 0;
4316         ext4_group_t g;
4317         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4318         int err = 0;
4319 #ifdef CONFIG_QUOTA
4320         int i;
4321 #endif
4322         char *orig_data = kstrdup(data, GFP_KERNEL);
4323
4324         /* Store the original options */
4325         lock_super(sb);
4326         old_sb_flags = sb->s_flags;
4327         old_opts.s_mount_opt = sbi->s_mount_opt;
4328         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4329         old_opts.s_resuid = sbi->s_resuid;
4330         old_opts.s_resgid = sbi->s_resgid;
4331         old_opts.s_commit_interval = sbi->s_commit_interval;
4332         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4333         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4334 #ifdef CONFIG_QUOTA
4335         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4336         for (i = 0; i < MAXQUOTAS; i++)
4337                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4338 #endif
4339         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4340                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4341
4342         /*
4343          * Allow the "check" option to be passed as a remount option.
4344          */
4345         if (!parse_options(data, sb, NULL, &journal_ioprio,
4346                            &n_blocks_count, 1)) {
4347                 err = -EINVAL;
4348                 goto restore_opts;
4349         }
4350
4351         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4352                 ext4_abort(sb, "Abort forced by user");
4353
4354         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4355                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4356
4357         es = sbi->s_es;
4358
4359         if (sbi->s_journal) {
4360                 ext4_init_journal_params(sb, sbi->s_journal);
4361                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4362         }
4363
4364         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4365                 n_blocks_count > ext4_blocks_count(es)) {
4366                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4367                         err = -EROFS;
4368                         goto restore_opts;
4369                 }
4370
4371                 if (*flags & MS_RDONLY) {
4372                         err = dquot_suspend(sb, -1);
4373                         if (err < 0)
4374                                 goto restore_opts;
4375
4376                         /*
4377                          * First of all, the unconditional stuff we have to do
4378                          * to disable replay of the journal when we next remount
4379                          */
4380                         sb->s_flags |= MS_RDONLY;
4381
4382                         /*
4383                          * OK, test if we are remounting a valid rw partition
4384                          * readonly, and if so set the rdonly flag and then
4385                          * mark the partition as valid again.
4386                          */
4387                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4388                             (sbi->s_mount_state & EXT4_VALID_FS))
4389                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4390
4391                         if (sbi->s_journal)
4392                                 ext4_mark_recovery_complete(sb, es);
4393                 } else {
4394                         /* Make sure we can mount this feature set readwrite */
4395                         if (!ext4_feature_set_ok(sb, 0)) {
4396                                 err = -EROFS;
4397                                 goto restore_opts;
4398                         }
4399                         /*
4400                          * Make sure the group descriptor checksums
4401                          * are sane.  If they aren't, refuse to remount r/w.
4402                          */
4403                         for (g = 0; g < sbi->s_groups_count; g++) {
4404                                 struct ext4_group_desc *gdp =
4405                                         ext4_get_group_desc(sb, g, NULL);
4406
4407                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4408                                         ext4_msg(sb, KERN_ERR,
4409                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4410                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4411                                                le16_to_cpu(gdp->bg_checksum));
4412                                         err = -EINVAL;
4413                                         goto restore_opts;
4414                                 }
4415                         }
4416
4417                         /*
4418                          * If we have an unprocessed orphan list hanging
4419                          * around from a previously readonly bdev mount,
4420                          * require a full umount/remount for now.
4421                          */
4422                         if (es->s_last_orphan) {
4423                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4424                                        "remount RDWR because of unprocessed "
4425                                        "orphan inode list.  Please "
4426                                        "umount/remount instead");
4427                                 err = -EINVAL;
4428                                 goto restore_opts;
4429                         }
4430
4431                         /*
4432                          * Mounting a RDONLY partition read-write, so reread
4433                          * and store the current valid flag.  (It may have
4434                          * been changed by e2fsck since we originally mounted
4435                          * the partition.)
4436                          */
4437                         if (sbi->s_journal)
4438                                 ext4_clear_journal_err(sb, es);
4439                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4440                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4441                                 goto restore_opts;
4442                         if (!ext4_setup_super(sb, es, 0))
4443                                 sb->s_flags &= ~MS_RDONLY;
4444                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4445                                                      EXT4_FEATURE_INCOMPAT_MMP))
4446                                 if (ext4_multi_mount_protect(sb,
4447                                                 le64_to_cpu(es->s_mmp_block))) {
4448                                         err = -EROFS;
4449                                         goto restore_opts;
4450                                 }
4451                         enable_quota = 1;
4452                 }
4453         }
4454
4455         /*
4456          * Reinitialize lazy itable initialization thread based on
4457          * current settings
4458          */
4459         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4460                 ext4_unregister_li_request(sb);
4461         else {
4462                 ext4_group_t first_not_zeroed;
4463                 first_not_zeroed = ext4_has_uninit_itable(sb);
4464                 ext4_register_li_request(sb, first_not_zeroed);
4465         }
4466
4467         ext4_setup_system_zone(sb);
4468         if (sbi->s_journal == NULL)
4469                 ext4_commit_super(sb, 1);
4470
4471 #ifdef CONFIG_QUOTA
4472         /* Release old quota file names */
4473         for (i = 0; i < MAXQUOTAS; i++)
4474                 if (old_opts.s_qf_names[i] &&
4475                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4476                         kfree(old_opts.s_qf_names[i]);
4477 #endif
4478         unlock_super(sb);
4479         if (enable_quota)
4480                 dquot_resume(sb, -1);
4481
4482         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4483         kfree(orig_data);
4484         return 0;
4485
4486 restore_opts:
4487         sb->s_flags = old_sb_flags;
4488         sbi->s_mount_opt = old_opts.s_mount_opt;
4489         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4490         sbi->s_resuid = old_opts.s_resuid;
4491         sbi->s_resgid = old_opts.s_resgid;
4492         sbi->s_commit_interval = old_opts.s_commit_interval;
4493         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4494         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4495 #ifdef CONFIG_QUOTA
4496         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4497         for (i = 0; i < MAXQUOTAS; i++) {
4498                 if (sbi->s_qf_names[i] &&
4499                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4500                         kfree(sbi->s_qf_names[i]);
4501                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4502         }
4503 #endif
4504         unlock_super(sb);
4505         kfree(orig_data);
4506         return err;
4507 }
4508
4509 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4510 {
4511         struct super_block *sb = dentry->d_sb;
4512         struct ext4_sb_info *sbi = EXT4_SB(sb);
4513         struct ext4_super_block *es = sbi->s_es;
4514         u64 fsid;
4515         s64 bfree;
4516
4517         if (test_opt(sb, MINIX_DF)) {
4518                 sbi->s_overhead_last = 0;
4519         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4520                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4521                 ext4_fsblk_t overhead = 0;
4522
4523                 /*
4524                  * Compute the overhead (FS structures).  This is constant
4525                  * for a given filesystem unless the number of block groups
4526                  * changes so we cache the previous value until it does.
4527                  */
4528
4529                 /*
4530                  * All of the blocks before first_data_block are
4531                  * overhead
4532                  */
4533                 overhead = le32_to_cpu(es->s_first_data_block);
4534
4535                 /*
4536                  * Add the overhead attributed to the superblock and
4537                  * block group descriptors.  If the sparse superblocks
4538                  * feature is turned on, then not all groups have this.
4539                  */
4540                 for (i = 0; i < ngroups; i++) {
4541                         overhead += ext4_bg_has_super(sb, i) +
4542                                 ext4_bg_num_gdb(sb, i);
4543                         cond_resched();
4544                 }
4545
4546                 /*
4547                  * Every block group has an inode bitmap, a block
4548                  * bitmap, and an inode table.
4549                  */
4550                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4551                 sbi->s_overhead_last = overhead;
4552                 smp_wmb();
4553                 sbi->s_blocks_last = ext4_blocks_count(es);
4554         }
4555
4556         buf->f_type = EXT4_SUPER_MAGIC;
4557         buf->f_bsize = sb->s_blocksize;
4558         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4559         bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4560                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4561         /* prevent underflow in case that few free space is available */
4562         buf->f_bfree = max_t(s64, bfree, 0);
4563         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4564         if (buf->f_bfree < ext4_r_blocks_count(es))
4565                 buf->f_bavail = 0;
4566         buf->f_files = le32_to_cpu(es->s_inodes_count);
4567         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4568         buf->f_namelen = EXT4_NAME_LEN;
4569         fsid = le64_to_cpup((void *)es->s_uuid) ^
4570                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4571         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4572         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4573
4574         return 0;
4575 }
4576
4577 /* Helper function for writing quotas on sync - we need to start transaction
4578  * before quota file is locked for write. Otherwise the are possible deadlocks:
4579  * Process 1                         Process 2
4580  * ext4_create()                     quota_sync()
4581  *   jbd2_journal_start()                  write_dquot()
4582  *   dquot_initialize()                         down(dqio_mutex)
4583  *     down(dqio_mutex)                    jbd2_journal_start()
4584  *
4585  */
4586
4587 #ifdef CONFIG_QUOTA
4588
4589 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4590 {
4591         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4592 }
4593
4594 static int ext4_write_dquot(struct dquot *dquot)
4595 {
4596         int ret, err;
4597         handle_t *handle;
4598         struct inode *inode;
4599
4600         inode = dquot_to_inode(dquot);
4601         handle = ext4_journal_start(inode,
4602                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4603         if (IS_ERR(handle))
4604                 return PTR_ERR(handle);
4605         ret = dquot_commit(dquot);
4606         err = ext4_journal_stop(handle);
4607         if (!ret)
4608                 ret = err;
4609         return ret;
4610 }
4611
4612 static int ext4_acquire_dquot(struct dquot *dquot)
4613 {
4614         int ret, err;
4615         handle_t *handle;
4616
4617         handle = ext4_journal_start(dquot_to_inode(dquot),
4618                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4619         if (IS_ERR(handle))
4620                 return PTR_ERR(handle);
4621         ret = dquot_acquire(dquot);
4622         err = ext4_journal_stop(handle);
4623         if (!ret)
4624                 ret = err;
4625         return ret;
4626 }
4627
4628 static int ext4_release_dquot(struct dquot *dquot)
4629 {
4630         int ret, err;
4631         handle_t *handle;
4632
4633         handle = ext4_journal_start(dquot_to_inode(dquot),
4634                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4635         if (IS_ERR(handle)) {
4636                 /* Release dquot anyway to avoid endless cycle in dqput() */
4637                 dquot_release(dquot);
4638                 return PTR_ERR(handle);
4639         }
4640         ret = dquot_release(dquot);
4641         err = ext4_journal_stop(handle);
4642         if (!ret)
4643                 ret = err;
4644         return ret;
4645 }
4646
4647 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4648 {
4649         /* Are we journaling quotas? */
4650         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4651             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4652                 dquot_mark_dquot_dirty(dquot);
4653                 return ext4_write_dquot(dquot);
4654         } else {
4655                 return dquot_mark_dquot_dirty(dquot);
4656         }
4657 }
4658
4659 static int ext4_write_info(struct super_block *sb, int type)
4660 {
4661         int ret, err;
4662         handle_t *handle;
4663
4664         /* Data block + inode block */
4665         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4666         if (IS_ERR(handle))
4667                 return PTR_ERR(handle);
4668         ret = dquot_commit_info(sb, type);
4669         err = ext4_journal_stop(handle);
4670         if (!ret)
4671                 ret = err;
4672         return ret;
4673 }
4674
4675 /*
4676  * Turn on quotas during mount time - we need to find
4677  * the quota file and such...
4678  */
4679 static int ext4_quota_on_mount(struct super_block *sb, int type)
4680 {
4681         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4682                                         EXT4_SB(sb)->s_jquota_fmt, type);
4683 }
4684
4685 /*
4686  * Standard function to be called on quota_on
4687  */
4688 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4689                          struct path *path)
4690 {
4691         int err;
4692
4693         if (!test_opt(sb, QUOTA))
4694                 return -EINVAL;
4695
4696         /* Quotafile not on the same filesystem? */
4697         if (path->mnt->mnt_sb != sb)
4698                 return -EXDEV;
4699         /* Journaling quota? */
4700         if (EXT4_SB(sb)->s_qf_names[type]) {
4701                 /* Quotafile not in fs root? */
4702                 if (path->dentry->d_parent != sb->s_root)
4703                         ext4_msg(sb, KERN_WARNING,
4704                                 "Quota file not on filesystem root. "
4705                                 "Journaled quota will not work");
4706         }
4707
4708         /*
4709          * When we journal data on quota file, we have to flush journal to see
4710          * all updates to the file when we bypass pagecache...
4711          */
4712         if (EXT4_SB(sb)->s_journal &&
4713             ext4_should_journal_data(path->dentry->d_inode)) {
4714                 /*
4715                  * We don't need to lock updates but journal_flush() could
4716                  * otherwise be livelocked...
4717                  */
4718                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4719                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4720                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4721                 if (err)
4722                         return err;
4723         }
4724
4725         return dquot_quota_on(sb, type, format_id, path);
4726 }
4727
4728 static int ext4_quota_off(struct super_block *sb, int type)
4729 {
4730         struct inode *inode = sb_dqopt(sb)->files[type];
4731         handle_t *handle;
4732
4733         /* Force all delayed allocation blocks to be allocated.
4734          * Caller already holds s_umount sem */
4735         if (test_opt(sb, DELALLOC))
4736                 sync_filesystem(sb);
4737
4738         if (!inode)
4739                 goto out;
4740
4741         /* Update modification times of quota files when userspace can
4742          * start looking at them */
4743         handle = ext4_journal_start(inode, 1);
4744         if (IS_ERR(handle))
4745                 goto out;
4746         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4747         ext4_mark_inode_dirty(handle, inode);
4748         ext4_journal_stop(handle);
4749
4750 out:
4751         return dquot_quota_off(sb, type);
4752 }
4753
4754 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4755  * acquiring the locks... As quota files are never truncated and quota code
4756  * itself serializes the operations (and no one else should touch the files)
4757  * we don't have to be afraid of races */
4758 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4759                                size_t len, loff_t off)
4760 {
4761         struct inode *inode = sb_dqopt(sb)->files[type];
4762         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4763         int err = 0;
4764         int offset = off & (sb->s_blocksize - 1);
4765         int tocopy;
4766         size_t toread;
4767         struct buffer_head *bh;
4768         loff_t i_size = i_size_read(inode);
4769
4770         if (off > i_size)
4771                 return 0;
4772         if (off+len > i_size)
4773                 len = i_size-off;
4774         toread = len;
4775         while (toread > 0) {
4776                 tocopy = sb->s_blocksize - offset < toread ?
4777                                 sb->s_blocksize - offset : toread;
4778                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4779                 if (err)
4780                         return err;
4781                 if (!bh)        /* A hole? */
4782                         memset(data, 0, tocopy);
4783                 else
4784                         memcpy(data, bh->b_data+offset, tocopy);
4785                 brelse(bh);
4786                 offset = 0;
4787                 toread -= tocopy;
4788                 data += tocopy;
4789                 blk++;
4790         }
4791         return len;
4792 }
4793
4794 /* Write to quotafile (we know the transaction is already started and has
4795  * enough credits) */
4796 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4797                                 const char *data, size_t len, loff_t off)
4798 {
4799         struct inode *inode = sb_dqopt(sb)->files[type];
4800         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4801         int err = 0;
4802         int offset = off & (sb->s_blocksize - 1);
4803         struct buffer_head *bh;
4804         handle_t *handle = journal_current_handle();
4805
4806         if (EXT4_SB(sb)->s_journal && !handle) {
4807                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4808                         " cancelled because transaction is not started",
4809                         (unsigned long long)off, (unsigned long long)len);
4810                 return -EIO;
4811         }
4812         /*
4813          * Since we account only one data block in transaction credits,
4814          * then it is impossible to cross a block boundary.
4815          */
4816         if (sb->s_blocksize - offset < len) {
4817                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4818                         " cancelled because not block aligned",
4819                         (unsigned long long)off, (unsigned long long)len);
4820                 return -EIO;
4821         }
4822
4823         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4824         bh = ext4_bread(handle, inode, blk, 1, &err);
4825         if (!bh)
4826                 goto out;
4827         err = ext4_journal_get_write_access(handle, bh);
4828         if (err) {
4829                 brelse(bh);
4830                 goto out;
4831         }
4832         lock_buffer(bh);
4833         memcpy(bh->b_data+offset, data, len);
4834         flush_dcache_page(bh->b_page);
4835         unlock_buffer(bh);
4836         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4837         brelse(bh);
4838 out:
4839         if (err) {
4840                 mutex_unlock(&inode->i_mutex);
4841                 return err;
4842         }
4843         if (inode->i_size < off + len) {
4844                 i_size_write(inode, off + len);
4845                 EXT4_I(inode)->i_disksize = inode->i_size;
4846                 ext4_mark_inode_dirty(handle, inode);
4847         }
4848         mutex_unlock(&inode->i_mutex);
4849         return len;
4850 }
4851
4852 #endif
4853
4854 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4855                        const char *dev_name, void *data)
4856 {
4857         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4858 }
4859
4860 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4861 static inline void register_as_ext2(void)
4862 {
4863         int err = register_filesystem(&ext2_fs_type);
4864         if (err)
4865                 printk(KERN_WARNING
4866                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4867 }
4868
4869 static inline void unregister_as_ext2(void)
4870 {
4871         unregister_filesystem(&ext2_fs_type);
4872 }
4873
4874 static inline int ext2_feature_set_ok(struct super_block *sb)
4875 {
4876         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4877                 return 0;
4878         if (sb->s_flags & MS_RDONLY)
4879                 return 1;
4880         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4881                 return 0;
4882         return 1;
4883 }
4884 MODULE_ALIAS("ext2");
4885 #else
4886 static inline void register_as_ext2(void) { }
4887 static inline void unregister_as_ext2(void) { }
4888 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4889 #endif
4890
4891 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4892 static inline void register_as_ext3(void)
4893 {
4894         int err = register_filesystem(&ext3_fs_type);
4895         if (err)
4896                 printk(KERN_WARNING
4897                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4898 }
4899
4900 static inline void unregister_as_ext3(void)
4901 {
4902         unregister_filesystem(&ext3_fs_type);
4903 }
4904
4905 static inline int ext3_feature_set_ok(struct super_block *sb)
4906 {
4907         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4908                 return 0;
4909         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4910                 return 0;
4911         if (sb->s_flags & MS_RDONLY)
4912                 return 1;
4913         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4914                 return 0;
4915         return 1;
4916 }
4917 MODULE_ALIAS("ext3");
4918 #else
4919 static inline void register_as_ext3(void) { }
4920 static inline void unregister_as_ext3(void) { }
4921 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4922 #endif
4923
4924 static struct file_system_type ext4_fs_type = {
4925         .owner          = THIS_MODULE,
4926         .name           = "ext4",
4927         .mount          = ext4_mount,
4928         .kill_sb        = kill_block_super,
4929         .fs_flags       = FS_REQUIRES_DEV,
4930 };
4931
4932 static int __init ext4_init_feat_adverts(void)
4933 {
4934         struct ext4_features *ef;
4935         int ret = -ENOMEM;
4936
4937         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4938         if (!ef)
4939                 goto out;
4940
4941         ef->f_kobj.kset = ext4_kset;
4942         init_completion(&ef->f_kobj_unregister);
4943         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4944                                    "features");
4945         if (ret) {
4946                 kfree(ef);
4947                 goto out;
4948         }
4949
4950         ext4_feat = ef;
4951         ret = 0;
4952 out:
4953         return ret;
4954 }
4955
4956 static void ext4_exit_feat_adverts(void)
4957 {
4958         kobject_put(&ext4_feat->f_kobj);
4959         wait_for_completion(&ext4_feat->f_kobj_unregister);
4960         kfree(ext4_feat);
4961 }
4962
4963 /* Shared across all ext4 file systems */
4964 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4965 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4966
4967 static int __init ext4_init_fs(void)
4968 {
4969         int i, err;
4970
4971         ext4_check_flag_values();
4972
4973         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4974                 mutex_init(&ext4__aio_mutex[i]);
4975                 init_waitqueue_head(&ext4__ioend_wq[i]);
4976         }
4977
4978         err = ext4_init_pageio();
4979         if (err)
4980                 return err;
4981         err = ext4_init_system_zone();
4982         if (err)
4983                 goto out7;
4984         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4985         if (!ext4_kset)
4986                 goto out6;
4987         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4988         if (!ext4_proc_root)
4989                 goto out5;
4990
4991         err = ext4_init_feat_adverts();
4992         if (err)
4993                 goto out4;
4994
4995         err = ext4_init_mballoc();
4996         if (err)
4997                 goto out3;
4998
4999         err = ext4_init_xattr();
5000         if (err)
5001                 goto out2;
5002         err = init_inodecache();
5003         if (err)
5004                 goto out1;
5005         register_as_ext3();
5006         register_as_ext2();
5007         err = register_filesystem(&ext4_fs_type);
5008         if (err)
5009                 goto out;
5010
5011         ext4_li_info = NULL;
5012         mutex_init(&ext4_li_mtx);
5013         return 0;
5014 out:
5015         unregister_as_ext2();
5016         unregister_as_ext3();
5017         destroy_inodecache();
5018 out1:
5019         ext4_exit_xattr();
5020 out2:
5021         ext4_exit_mballoc();
5022 out3:
5023         ext4_exit_feat_adverts();
5024 out4:
5025         remove_proc_entry("fs/ext4", NULL);
5026 out5:
5027         kset_unregister(ext4_kset);
5028 out6:
5029         ext4_exit_system_zone();
5030 out7:
5031         ext4_exit_pageio();
5032         return err;
5033 }
5034
5035 static void __exit ext4_exit_fs(void)
5036 {
5037         ext4_destroy_lazyinit_thread();
5038         unregister_as_ext2();
5039         unregister_as_ext3();
5040         unregister_filesystem(&ext4_fs_type);
5041         destroy_inodecache();
5042         ext4_exit_xattr();
5043         ext4_exit_mballoc();
5044         ext4_exit_feat_adverts();
5045         remove_proc_entry("fs/ext4", NULL);
5046         kset_unregister(ext4_kset);
5047         ext4_exit_system_zone();
5048         ext4_exit_pageio();
5049 }
5050
5051 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5052 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5053 MODULE_LICENSE("GPL");
5054 module_init(ext4_init_fs)
5055 module_exit(ext4_exit_fs)