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 <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_jbd2.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "mballoc.h"
  51
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/ext4.h>
  54
  55 static struct proc_dir_entry *ext4_proc_root;
  56 static struct kset *ext4_kset;
  57 struct ext4_lazy_init *ext4_li_info;
  58 struct mutex ext4_li_mtx;
  59 struct ext4_features *ext4_feat;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static const char *ext4_decode_error(struct super_block *sb, int errno,
  70                                      char nbuf[16]);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static void ext4_write_super(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static void ext4_destroy_lazyinit_thread(void);
  79 static void ext4_unregister_li_request(struct super_block *sb);
  80
  81 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  82 static struct file_system_type ext3_fs_type = {
  83         .owner          = THIS_MODULE,
  84         .name           = "ext3",
  85         .mount          = ext4_mount,
  86         .kill_sb        = kill_block_super,
  87         .fs_flags       = FS_REQUIRES_DEV,
  88 };
  89 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  90 #else
  91 #define IS_EXT3_SB(sb) (0)
  92 #endif
  93
  94 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  95                                struct ext4_group_desc *bg)
  96 {
  97         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  98                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  99                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 100 }
 101
 102 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 103                                struct ext4_group_desc *bg)
 104 {
 105         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 106                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 107                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 108 }
 109
 110 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 111                               struct ext4_group_desc *bg)
 112 {
 113         return le32_to_cpu(bg->bg_inode_table_lo) |
 114                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 115                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 116 }
 117
 118 __u32 ext4_free_blks_count(struct super_block *sb,
 119                               struct ext4_group_desc *bg)
 120 {
 121         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 122                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 123                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 124 }
 125
 126 __u32 ext4_free_inodes_count(struct super_block *sb,
 127                               struct ext4_group_desc *bg)
 128 {
 129         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 130                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 131                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 132 }
 133
 134 __u32 ext4_used_dirs_count(struct super_block *sb,
 135                               struct ext4_group_desc *bg)
 136 {
 137         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 138                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 139                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 140 }
 141
 142 __u32 ext4_itable_unused_count(struct super_block *sb,
 143                               struct ext4_group_desc *bg)
 144 {
 145         return le16_to_cpu(bg->bg_itable_unused_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 148 }
 149
 150 void ext4_block_bitmap_set(struct super_block *sb,
 151                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 152 {
 153         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 154         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 155                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 156 }
 157
 158 void ext4_inode_bitmap_set(struct super_block *sb,
 159                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 160 {
 161         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 162         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 163                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 164 }
 165
 166 void ext4_inode_table_set(struct super_block *sb,
 167                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 168 {
 169         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 170         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 171                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 172 }
 173
 174 void ext4_free_blks_set(struct super_block *sb,
 175                           struct ext4_group_desc *bg, __u32 count)
 176 {
 177         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 178         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 179                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 180 }
 181
 182 void ext4_free_inodes_set(struct super_block *sb,
 183                           struct ext4_group_desc *bg, __u32 count)
 184 {
 185         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 186         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 187                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 188 }
 189
 190 void ext4_used_dirs_set(struct super_block *sb,
 191                           struct ext4_group_desc *bg, __u32 count)
 192 {
 193         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 194         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 195                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 196 }
 197
 198 void ext4_itable_unused_set(struct super_block *sb,
 199                           struct ext4_group_desc *bg, __u32 count)
 200 {
 201         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 204 }
 205
 206
 207 /* Just increment the non-pointer handle value */
 208 static handle_t *ext4_get_nojournal(void)
 209 {
 210         handle_t *handle = current->journal_info;
 211         unsigned long ref_cnt = (unsigned long)handle;
 212
 213         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 214
 215         ref_cnt++;
 216         handle = (handle_t *)ref_cnt;
 217
 218         current->journal_info = handle;
 219         return handle;
 220 }
 221
 222
 223 /* Decrement the non-pointer handle value */
 224 static void ext4_put_nojournal(handle_t *handle)
 225 {
 226         unsigned long ref_cnt = (unsigned long)handle;
 227
 228         BUG_ON(ref_cnt == 0);
 229
 230         ref_cnt--;
 231         handle = (handle_t *)ref_cnt;
 232
 233         current->journal_info = handle;
 234 }
 235
 236 /*
 237  * Wrappers for jbd2_journal_start/end.
 238  *
 239  * The only special thing we need to do here is to make sure that all
 240  * journal_end calls result in the superblock being marked dirty, so
 241  * that sync() will call the filesystem's write_super callback if
 242  * appropriate.
 243  */
 244 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 245 {
 246         journal_t *journal;
 247
 248         if (sb->s_flags & MS_RDONLY)
 249                 return ERR_PTR(-EROFS);
 250
 251         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 252         /* Special case here: if the journal has aborted behind our
 253          * backs (eg. EIO in the commit thread), then we still need to
 254          * take the FS itself readonly cleanly. */
 255         journal = EXT4_SB(sb)->s_journal;
 256         if (journal) {
 257                 if (is_journal_aborted(journal)) {
 258                         ext4_abort(sb, "Detected aborted journal");
 259                         return ERR_PTR(-EROFS);
 260                 }
 261                 return jbd2_journal_start(journal, nblocks);
 262         }
 263         return ext4_get_nojournal();
 264 }
 265
 266 /*
 267  * The only special thing we need to do here is to make sure that all
 268  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 269  * that sync() will call the filesystem's write_super callback if
 270  * appropriate.
 271  */
 272 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 273 {
 274         struct super_block *sb;
 275         int err;
 276         int rc;
 277
 278         if (!ext4_handle_valid(handle)) {
 279                 ext4_put_nojournal(handle);
 280                 return 0;
 281         }
 282         sb = handle->h_transaction->t_journal->j_private;
 283         err = handle->h_err;
 284         rc = jbd2_journal_stop(handle);
 285
 286         if (!err)
 287                 err = rc;
 288         if (err)
 289                 __ext4_std_error(sb, where, line, err);
 290         return err;
 291 }
 292
 293 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 294                                const char *err_fn, struct buffer_head *bh,
 295                                handle_t *handle, int err)
 296 {
 297         char nbuf[16];
 298         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 299
 300         BUG_ON(!ext4_handle_valid(handle));
 301
 302         if (bh)
 303                 BUFFER_TRACE(bh, "abort");
 304
 305         if (!handle->h_err)
 306                 handle->h_err = err;
 307
 308         if (is_handle_aborted(handle))
 309                 return;
 310
 311         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 312                caller, line, errstr, err_fn);
 313
 314         jbd2_journal_abort_handle(handle);
 315 }
 316
 317 static void __save_error_info(struct super_block *sb, const char *func,
 318                             unsigned int line)
 319 {
 320         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 321
 322         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 324         es->s_last_error_time = cpu_to_le32(get_seconds());
 325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 326         es->s_last_error_line = cpu_to_le32(line);
 327         if (!es->s_first_error_time) {
 328                 es->s_first_error_time = es->s_last_error_time;
 329                 strncpy(es->s_first_error_func, func,
 330                         sizeof(es->s_first_error_func));
 331                 es->s_first_error_line = cpu_to_le32(line);
 332                 es->s_first_error_ino = es->s_last_error_ino;
 333                 es->s_first_error_block = es->s_last_error_block;
 334         }
 335         /*
 336          * Start the daily error reporting function if it hasn't been
 337          * started already
 338          */
 339         if (!es->s_error_count)
 340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 341         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 342 }
 343
 344 static void save_error_info(struct super_block *sb, const char *func,
 345                             unsigned int line)
 346 {
 347         __save_error_info(sb, func, line);
 348         ext4_commit_super(sb, 1);
 349 }
 350
 351
 352 /* Deal with the reporting of failure conditions on a filesystem such as
 353  * inconsistencies detected or read IO failures.
 354  *
 355  * On ext2, we can store the error state of the filesystem in the
 356  * superblock.  That is not possible on ext4, because we may have other
 357  * write ordering constraints on the superblock which prevent us from
 358  * writing it out straight away; and given that the journal is about to
 359  * be aborted, we can't rely on the current, or future, transactions to
 360  * write out the superblock safely.
 361  *
 362  * We'll just use the jbd2_journal_abort() error code to record an error in
 363  * the journal instead.  On recovery, the journal will complain about
 364  * that error until we've noted it down and cleared it.
 365  */
 366
 367 static void ext4_handle_error(struct super_block *sb)
 368 {
 369         if (sb->s_flags & MS_RDONLY)
 370                 return;
 371
 372         if (!test_opt(sb, ERRORS_CONT)) {
 373                 journal_t *journal = EXT4_SB(sb)->s_journal;
 374
 375                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 376                 if (journal)
 377                         jbd2_journal_abort(journal, -EIO);
 378         }
 379         if (test_opt(sb, ERRORS_RO)) {
 380                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 381                 sb->s_flags |= MS_RDONLY;
 382         }
 383         if (test_opt(sb, ERRORS_PANIC))
 384                 panic("EXT4-fs (device %s): panic forced after error\n",
 385                         sb->s_id);
 386 }
 387
 388 void __ext4_error(struct super_block *sb, const char *function,
 389                   unsigned int line, const char *fmt, ...)
 390 {
 391         va_list args;
 392
 393         va_start(args, fmt);
 394         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: ",
 395                sb->s_id, function, line, current->comm);
 396         vprintk(fmt, args);
 397         printk("\n");
 398         va_end(args);
 399
 400         ext4_handle_error(sb);
 401 }
 402
 403 void ext4_error_inode(struct inode *inode, const char *function,
 404                       unsigned int line, ext4_fsblk_t block,
 405                       const char *fmt, ...)
 406 {
 407         va_list args;
 408         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 409
 410         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 411         es->s_last_error_block = cpu_to_le64(block);
 412         save_error_info(inode->i_sb, function, line);
 413         va_start(args, fmt);
 414         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 415                inode->i_sb->s_id, function, line, inode->i_ino);
 416         if (block)
 417                 printk("block %llu: ", block);
 418         printk("comm %s: ", current->comm);
 419         vprintk(fmt, args);
 420         printk("\n");
 421         va_end(args);
 422
 423         ext4_handle_error(inode->i_sb);
 424 }
 425
 426 void ext4_error_file(struct file *file, const char *function,
 427                      unsigned int line, const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct ext4_super_block *es;
 431         struct inode *inode = file->f_dentry->d_inode;
 432         char pathname[80], *path;
 433
 434         es = EXT4_SB(inode->i_sb)->s_es;
 435         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 436         save_error_info(inode->i_sb, function, line);
 437         va_start(args, fmt);
 438         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 439         if (!path)
 440                 path = "(unknown)";
 441         printk(KERN_CRIT
 442                "EXT4-fs error (device %s): %s:%d: inode #%lu "
 443                "(comm %s path %s): ",
 444                inode->i_sb->s_id, function, line, inode->i_ino,
 445                current->comm, path);
 446         vprintk(fmt, args);
 447         printk("\n");
 448         va_end(args);
 449
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 static const char *ext4_decode_error(struct super_block *sb, int errno,
 454                                      char nbuf[16])
 455 {
 456         char *errstr = NULL;
 457
 458         switch (errno) {
 459         case -EIO:
 460                 errstr = "IO failure";
 461                 break;
 462         case -ENOMEM:
 463                 errstr = "Out of memory";
 464                 break;
 465         case -EROFS:
 466                 if (!sb || (EXT4_SB(sb)->s_journal &&
 467                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 468                         errstr = "Journal has aborted";
 469                 else
 470                         errstr = "Readonly filesystem";
 471                 break;
 472         default:
 473                 /* If the caller passed in an extra buffer for unknown
 474                  * errors, textualise them now.  Else we just return
 475                  * NULL. */
 476                 if (nbuf) {
 477                         /* Check for truncated error codes... */
 478                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 479                                 errstr = nbuf;
 480                 }
 481                 break;
 482         }
 483
 484         return errstr;
 485 }
 486
 487 /* __ext4_std_error decodes expected errors from journaling functions
 488  * automatically and invokes the appropriate error response.  */
 489
 490 void __ext4_std_error(struct super_block *sb, const char *function,
 491                       unsigned int line, int errno)
 492 {
 493         char nbuf[16];
 494         const char *errstr;
 495
 496         /* Special case: if the error is EROFS, and we're not already
 497          * inside a transaction, then there's really no point in logging
 498          * an error. */
 499         if (errno == -EROFS && journal_current_handle() == NULL &&
 500             (sb->s_flags & MS_RDONLY))
 501                 return;
 502
 503         errstr = ext4_decode_error(sb, errno, nbuf);
 504         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 505                sb->s_id, function, line, errstr);
 506         save_error_info(sb, function, line);
 507
 508         ext4_handle_error(sb);
 509 }
 510
 511 /*
 512  * ext4_abort is a much stronger failure handler than ext4_error.  The
 513  * abort function may be used to deal with unrecoverable failures such
 514  * as journal IO errors or ENOMEM at a critical moment in log management.
 515  *
 516  * We unconditionally force the filesystem into an ABORT|READONLY state,
 517  * unless the error response on the fs has been set to panic in which
 518  * case we take the easy way out and panic immediately.
 519  */
 520
 521 void __ext4_abort(struct super_block *sb, const char *function,
 522                 unsigned int line, const char *fmt, ...)
 523 {
 524         va_list args;
 525
 526         save_error_info(sb, function, line);
 527         va_start(args, fmt);
 528         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 529                function, line);
 530         vprintk(fmt, args);
 531         printk("\n");
 532         va_end(args);
 533
 534         if ((sb->s_flags & MS_RDONLY) == 0) {
 535                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 536                 sb->s_flags |= MS_RDONLY;
 537                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 538                 if (EXT4_SB(sb)->s_journal)
 539                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 540                 save_error_info(sb, function, line);
 541         }
 542         if (test_opt(sb, ERRORS_PANIC))
 543                 panic("EXT4-fs panic from previous error\n");
 544 }
 545
 546 void ext4_msg (struct super_block * sb, const char *prefix,
 547                    const char *fmt, ...)
 548 {
 549         va_list args;
 550
 551         va_start(args, fmt);
 552         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 553         vprintk(fmt, args);
 554         printk("\n");
 555         va_end(args);
 556 }
 557
 558 void __ext4_warning(struct super_block *sb, const char *function,
 559                     unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         va_start(args, fmt);
 564         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: ",
 565                sb->s_id, function, line);
 566         vprintk(fmt, args);
 567         printk("\n");
 568         va_end(args);
 569 }
 570
 571 void __ext4_grp_locked_error(const char *function, unsigned int line,
 572                              struct super_block *sb, ext4_group_t grp,
 573                              unsigned long ino, ext4_fsblk_t block,
 574                              const char *fmt, ...)
 575 __releases(bitlock)
 576 __acquires(bitlock)
 577 {
 578         va_list args;
 579         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 580
 581         es->s_last_error_ino = cpu_to_le32(ino);
 582         es->s_last_error_block = cpu_to_le64(block);
 583         __save_error_info(sb, function, line);
 584         va_start(args, fmt);
 585         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u",
 586                sb->s_id, function, line, grp);
 587         if (ino)
 588                 printk("inode %lu: ", ino);
 589         if (block)
 590                 printk("block %llu:", (unsigned long long) block);
 591         vprintk(fmt, args);
 592         printk("\n");
 593         va_end(args);
 594
 595         if (test_opt(sb, ERRORS_CONT)) {
 596                 ext4_commit_super(sb, 0);
 597                 return;
 598         }
 599
 600         ext4_unlock_group(sb, grp);
 601         ext4_handle_error(sb);
 602         /*
 603          * We only get here in the ERRORS_RO case; relocking the group
 604          * may be dangerous, but nothing bad will happen since the
 605          * filesystem will have already been marked read/only and the
 606          * journal has been aborted.  We return 1 as a hint to callers
 607          * who might what to use the return value from
 608          * ext4_grp_locked_error() to distinguish beween the
 609          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 610          * aggressively from the ext4 function in question, with a
 611          * more appropriate error code.
 612          */
 613         ext4_lock_group(sb, grp);
 614         return;
 615 }
 616
 617 void ext4_update_dynamic_rev(struct super_block *sb)
 618 {
 619         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 620
 621         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 622                 return;
 623
 624         ext4_warning(sb,
 625                      "updating to rev %d because of new feature flag, "
 626                      "running e2fsck is recommended",
 627                      EXT4_DYNAMIC_REV);
 628
 629         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 630         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 631         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 632         /* leave es->s_feature_*compat flags alone */
 633         /* es->s_uuid will be set by e2fsck if empty */
 634
 635         /*
 636          * The rest of the superblock fields should be zero, and if not it
 637          * means they are likely already in use, so leave them alone.  We
 638          * can leave it up to e2fsck to clean up any inconsistencies there.
 639          */
 640 }
 641
 642 /*
 643  * Open the external journal device
 644  */
 645 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 646 {
 647         struct block_device *bdev;
 648         char b[BDEVNAME_SIZE];
 649
 650         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 651         if (IS_ERR(bdev))
 652                 goto fail;
 653         return bdev;
 654
 655 fail:
 656         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 657                         __bdevname(dev, b), PTR_ERR(bdev));
 658         return NULL;
 659 }
 660
 661 /*
 662  * Release the journal device
 663  */
 664 static int ext4_blkdev_put(struct block_device *bdev)
 665 {
 666         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 667 }
 668
 669 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 670 {
 671         struct block_device *bdev;
 672         int ret = -ENODEV;
 673
 674         bdev = sbi->journal_bdev;
 675         if (bdev) {
 676                 ret = ext4_blkdev_put(bdev);
 677                 sbi->journal_bdev = NULL;
 678         }
 679         return ret;
 680 }
 681
 682 static inline struct inode *orphan_list_entry(struct list_head *l)
 683 {
 684         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 685 }
 686
 687 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 688 {
 689         struct list_head *l;
 690
 691         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 692                  le32_to_cpu(sbi->s_es->s_last_orphan));
 693
 694         printk(KERN_ERR "sb_info orphan list:\n");
 695         list_for_each(l, &sbi->s_orphan) {
 696                 struct inode *inode = orphan_list_entry(l);
 697                 printk(KERN_ERR "  "
 698                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 699                        inode->i_sb->s_id, inode->i_ino, inode,
 700                        inode->i_mode, inode->i_nlink,
 701                        NEXT_ORPHAN(inode));
 702         }
 703 }
 704
 705 static void ext4_put_super(struct super_block *sb)
 706 {
 707         struct ext4_sb_info *sbi = EXT4_SB(sb);
 708         struct ext4_super_block *es = sbi->s_es;
 709         int i, err;
 710
 711         ext4_unregister_li_request(sb);
 712         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 713
 714         flush_workqueue(sbi->dio_unwritten_wq);
 715         destroy_workqueue(sbi->dio_unwritten_wq);
 716
 717         lock_super(sb);
 718         if (sb->s_dirt)
 719                 ext4_commit_super(sb, 1);
 720
 721         if (sbi->s_journal) {
 722                 err = jbd2_journal_destroy(sbi->s_journal);
 723                 sbi->s_journal = NULL;
 724                 if (err < 0)
 725                         ext4_abort(sb, "Couldn't clean up the journal");
 726         }
 727
 728         del_timer(&sbi->s_err_report);
 729         ext4_release_system_zone(sb);
 730         ext4_mb_release(sb);
 731         ext4_ext_release(sb);
 732         ext4_xattr_put_super(sb);
 733
 734         if (!(sb->s_flags & MS_RDONLY)) {
 735                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 736                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 737                 ext4_commit_super(sb, 1);
 738         }
 739         if (sbi->s_proc) {
 740                 remove_proc_entry(sb->s_id, ext4_proc_root);
 741         }
 742         kobject_del(&sbi->s_kobj);
 743
 744         for (i = 0; i < sbi->s_gdb_count; i++)
 745                 brelse(sbi->s_group_desc[i]);
 746         kfree(sbi->s_group_desc);
 747         if (is_vmalloc_addr(sbi->s_flex_groups))
 748                 vfree(sbi->s_flex_groups);
 749         else
 750                 kfree(sbi->s_flex_groups);
 751         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 752         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 753         percpu_counter_destroy(&sbi->s_dirs_counter);
 754         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 755         brelse(sbi->s_sbh);
 756 #ifdef CONFIG_QUOTA
 757         for (i = 0; i < MAXQUOTAS; i++)
 758                 kfree(sbi->s_qf_names[i]);
 759 #endif
 760
 761         /* Debugging code just in case the in-memory inode orphan list
 762          * isn't empty.  The on-disk one can be non-empty if we've
 763          * detected an error and taken the fs readonly, but the
 764          * in-memory list had better be clean by this point. */
 765         if (!list_empty(&sbi->s_orphan))
 766                 dump_orphan_list(sb, sbi);
 767         J_ASSERT(list_empty(&sbi->s_orphan));
 768
 769         invalidate_bdev(sb->s_bdev);
 770         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 771                 /*
 772                  * Invalidate the journal device's buffers.  We don't want them
 773                  * floating about in memory - the physical journal device may
 774                  * hotswapped, and it breaks the `ro-after' testing code.
 775                  */
 776                 sync_blockdev(sbi->journal_bdev);
 777                 invalidate_bdev(sbi->journal_bdev);
 778                 ext4_blkdev_remove(sbi);
 779         }
 780         sb->s_fs_info = NULL;
 781         /*
 782          * Now that we are completely done shutting down the
 783          * superblock, we need to actually destroy the kobject.
 784          */
 785         unlock_super(sb);
 786         kobject_put(&sbi->s_kobj);
 787         wait_for_completion(&sbi->s_kobj_unregister);
 788         kfree(sbi->s_blockgroup_lock);
 789         kfree(sbi);
 790 }
 791
 792 static struct kmem_cache *ext4_inode_cachep;
 793
 794 /*
 795  * Called inside transaction, so use GFP_NOFS
 796  */
 797 static struct inode *ext4_alloc_inode(struct super_block *sb)
 798 {
 799         struct ext4_inode_info *ei;
 800
 801         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 802         if (!ei)
 803                 return NULL;
 804
 805         ei->vfs_inode.i_version = 1;
 806         ei->vfs_inode.i_data.writeback_index = 0;
 807         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 808         INIT_LIST_HEAD(&ei->i_prealloc_list);
 809         spin_lock_init(&ei->i_prealloc_lock);
 810         /*
 811          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 812          * therefore it can be null here.  Don't check it, just initialize
 813          * jinode.
 814          */
 815         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 816         ei->i_reserved_data_blocks = 0;
 817         ei->i_reserved_meta_blocks = 0;
 818         ei->i_allocated_meta_blocks = 0;
 819         ei->i_da_metadata_calc_len = 0;
 820         ei->i_delalloc_reserved_flag = 0;
 821         spin_lock_init(&(ei->i_block_reservation_lock));
 822 #ifdef CONFIG_QUOTA
 823         ei->i_reserved_quota = 0;
 824 #endif
 825         INIT_LIST_HEAD(&ei->i_completed_io_list);
 826         spin_lock_init(&ei->i_completed_io_lock);
 827         ei->cur_aio_dio = NULL;
 828         ei->i_sync_tid = 0;
 829         ei->i_datasync_tid = 0;
 830         atomic_set(&ei->i_ioend_count, 0);
 831
 832         return &ei->vfs_inode;
 833 }
 834
 835 static int ext4_drop_inode(struct inode *inode)
 836 {
 837         int drop = generic_drop_inode(inode);
 838
 839         trace_ext4_drop_inode(inode, drop);
 840         return drop;
 841 }
 842
 843 static void ext4_destroy_inode(struct inode *inode)
 844 {
 845         ext4_ioend_wait(inode);
 846         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 847                 ext4_msg(inode->i_sb, KERN_ERR,
 848                          "Inode %lu (%p): orphan list check failed!",
 849                          inode->i_ino, EXT4_I(inode));
 850                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 851                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 852                                 true);
 853                 dump_stack();
 854         }
 855         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 856 }
 857
 858 static void init_once(void *foo)
 859 {
 860         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 861
 862         INIT_LIST_HEAD(&ei->i_orphan);
 863 #ifdef CONFIG_EXT4_FS_XATTR
 864         init_rwsem(&ei->xattr_sem);
 865 #endif
 866         init_rwsem(&ei->i_data_sem);
 867         inode_init_once(&ei->vfs_inode);
 868 }
 869
 870 static int init_inodecache(void)
 871 {
 872         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 873                                              sizeof(struct ext4_inode_info),
 874                                              0, (SLAB_RECLAIM_ACCOUNT|
 875                                                 SLAB_MEM_SPREAD),
 876                                              init_once);
 877         if (ext4_inode_cachep == NULL)
 878                 return -ENOMEM;
 879         return 0;
 880 }
 881
 882 static void destroy_inodecache(void)
 883 {
 884         kmem_cache_destroy(ext4_inode_cachep);
 885 }
 886
 887 void ext4_clear_inode(struct inode *inode)
 888 {
 889         invalidate_inode_buffers(inode);
 890         end_writeback(inode);
 891         dquot_drop(inode);
 892         ext4_discard_preallocations(inode);
 893         if (EXT4_JOURNAL(inode))
 894                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 895                                        &EXT4_I(inode)->jinode);
 896 }
 897
 898 static inline void ext4_show_quota_options(struct seq_file *seq,
 899                                            struct super_block *sb)
 900 {
 901 #if defined(CONFIG_QUOTA)
 902         struct ext4_sb_info *sbi = EXT4_SB(sb);
 903
 904         if (sbi->s_jquota_fmt) {
 905                 char *fmtname = "";
 906
 907                 switch (sbi->s_jquota_fmt) {
 908                 case QFMT_VFS_OLD:
 909                         fmtname = "vfsold";
 910                         break;
 911                 case QFMT_VFS_V0:
 912                         fmtname = "vfsv0";
 913                         break;
 914                 case QFMT_VFS_V1:
 915                         fmtname = "vfsv1";
 916                         break;
 917                 }
 918                 seq_printf(seq, ",jqfmt=%s", fmtname);
 919         }
 920
 921         if (sbi->s_qf_names[USRQUOTA])
 922                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 923
 924         if (sbi->s_qf_names[GRPQUOTA])
 925                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 926
 927         if (test_opt(sb, USRQUOTA))
 928                 seq_puts(seq, ",usrquota");
 929
 930         if (test_opt(sb, GRPQUOTA))
 931                 seq_puts(seq, ",grpquota");
 932 #endif
 933 }
 934
 935 /*
 936  * Show an option if
 937  *  - it's set to a non-default value OR
 938  *  - if the per-sb default is different from the global default
 939  */
 940 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 941 {
 942         int def_errors;
 943         unsigned long def_mount_opts;
 944         struct super_block *sb = vfs->mnt_sb;
 945         struct ext4_sb_info *sbi = EXT4_SB(sb);
 946         struct ext4_super_block *es = sbi->s_es;
 947
 948         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 949         def_errors     = le16_to_cpu(es->s_errors);
 950
 951         if (sbi->s_sb_block != 1)
 952                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 953         if (test_opt(sb, MINIX_DF))
 954                 seq_puts(seq, ",minixdf");
 955         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 956                 seq_puts(seq, ",grpid");
 957         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 958                 seq_puts(seq, ",nogrpid");
 959         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 960             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 961                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 962         }
 963         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 964             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 965                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 966         }
 967         if (test_opt(sb, ERRORS_RO)) {
 968                 if (def_errors == EXT4_ERRORS_PANIC ||
 969                     def_errors == EXT4_ERRORS_CONTINUE) {
 970                         seq_puts(seq, ",errors=remount-ro");
 971                 }
 972         }
 973         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 974                 seq_puts(seq, ",errors=continue");
 975         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 976                 seq_puts(seq, ",errors=panic");
 977         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 978                 seq_puts(seq, ",nouid32");
 979         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 980                 seq_puts(seq, ",debug");
 981         if (test_opt(sb, OLDALLOC))
 982                 seq_puts(seq, ",oldalloc");
 983 #ifdef CONFIG_EXT4_FS_XATTR
 984         if (test_opt(sb, XATTR_USER) &&
 985                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 986                 seq_puts(seq, ",user_xattr");
 987         if (!test_opt(sb, XATTR_USER) &&
 988             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 989                 seq_puts(seq, ",nouser_xattr");
 990         }
 991 #endif
 992 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 993         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 994                 seq_puts(seq, ",acl");
 995         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 996                 seq_puts(seq, ",noacl");
 997 #endif
 998         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 999                 seq_printf(seq, ",commit=%u",
1000                            (unsigned) (sbi->s_commit_interval / HZ));
1001         }
1002         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1003                 seq_printf(seq, ",min_batch_time=%u",
1004                            (unsigned) sbi->s_min_batch_time);
1005         }
1006         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1007                 seq_printf(seq, ",max_batch_time=%u",
1008                            (unsigned) sbi->s_min_batch_time);
1009         }
1010
1011         /*
1012          * We're changing the default of barrier mount option, so
1013          * let's always display its mount state so it's clear what its
1014          * status is.
1015          */
1016         seq_puts(seq, ",barrier=");
1017         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1018         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1019                 seq_puts(seq, ",journal_async_commit");
1020         else if (test_opt(sb, JOURNAL_CHECKSUM))
1021                 seq_puts(seq, ",journal_checksum");
1022         if (test_opt(sb, I_VERSION))
1023                 seq_puts(seq, ",i_version");
1024         if (!test_opt(sb, DELALLOC) &&
1025             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1026                 seq_puts(seq, ",nodelalloc");
1027
1028         if (sbi->s_stripe)
1029                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1030         /*
1031          * journal mode get enabled in different ways
1032          * So just print the value even if we didn't specify it
1033          */
1034         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1035                 seq_puts(seq, ",data=journal");
1036         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1037                 seq_puts(seq, ",data=ordered");
1038         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1039                 seq_puts(seq, ",data=writeback");
1040
1041         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1042                 seq_printf(seq, ",inode_readahead_blks=%u",
1043                            sbi->s_inode_readahead_blks);
1044
1045         if (test_opt(sb, DATA_ERR_ABORT))
1046                 seq_puts(seq, ",data_err=abort");
1047
1048         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1049                 seq_puts(seq, ",noauto_da_alloc");
1050
1051         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1052                 seq_puts(seq, ",discard");
1053
1054         if (test_opt(sb, NOLOAD))
1055                 seq_puts(seq, ",norecovery");
1056
1057         if (test_opt(sb, DIOREAD_NOLOCK))
1058                 seq_puts(seq, ",dioread_nolock");
1059
1060         if (test_opt(sb, BLOCK_VALIDITY) &&
1061             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1062                 seq_puts(seq, ",block_validity");
1063
1064         if (!test_opt(sb, INIT_INODE_TABLE))
1065                 seq_puts(seq, ",noinit_inode_table");
1066         else if (sbi->s_li_wait_mult)
1067                 seq_printf(seq, ",init_inode_table=%u",
1068                            (unsigned) sbi->s_li_wait_mult);
1069
1070         ext4_show_quota_options(seq, sb);
1071
1072         return 0;
1073 }
1074
1075 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1076                                         u64 ino, u32 generation)
1077 {
1078         struct inode *inode;
1079
1080         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1081                 return ERR_PTR(-ESTALE);
1082         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1083                 return ERR_PTR(-ESTALE);
1084
1085         /* iget isn't really right if the inode is currently unallocated!!
1086          *
1087          * ext4_read_inode will return a bad_inode if the inode had been
1088          * deleted, so we should be safe.
1089          *
1090          * Currently we don't know the generation for parent directory, so
1091          * a generation of 0 means "accept any"
1092          */
1093         inode = ext4_iget(sb, ino);
1094         if (IS_ERR(inode))
1095                 return ERR_CAST(inode);
1096         if (generation && inode->i_generation != generation) {
1097                 iput(inode);
1098                 return ERR_PTR(-ESTALE);
1099         }
1100
1101         return inode;
1102 }
1103
1104 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1105                                         int fh_len, int fh_type)
1106 {
1107         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1108                                     ext4_nfs_get_inode);
1109 }
1110
1111 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1112                                         int fh_len, int fh_type)
1113 {
1114         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1115                                     ext4_nfs_get_inode);
1116 }
1117
1118 /*
1119  * Try to release metadata pages (indirect blocks, directories) which are
1120  * mapped via the block device.  Since these pages could have journal heads
1121  * which would prevent try_to_free_buffers() from freeing them, we must use
1122  * jbd2 layer's try_to_free_buffers() function to release them.
1123  */
1124 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1125                                  gfp_t wait)
1126 {
1127         journal_t *journal = EXT4_SB(sb)->s_journal;
1128
1129         WARN_ON(PageChecked(page));
1130         if (!page_has_buffers(page))
1131                 return 0;
1132         if (journal)
1133                 return jbd2_journal_try_to_free_buffers(journal, page,
1134                                                         wait & ~__GFP_WAIT);
1135         return try_to_free_buffers(page);
1136 }
1137
1138 #ifdef CONFIG_QUOTA
1139 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1140 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1141
1142 static int ext4_write_dquot(struct dquot *dquot);
1143 static int ext4_acquire_dquot(struct dquot *dquot);
1144 static int ext4_release_dquot(struct dquot *dquot);
1145 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1146 static int ext4_write_info(struct super_block *sb, int type);
1147 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1148                                 char *path);
1149 static int ext4_quota_off(struct super_block *sb, int type);
1150 static int ext4_quota_on_mount(struct super_block *sb, int type);
1151 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1152                                size_t len, loff_t off);
1153 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1154                                 const char *data, size_t len, loff_t off);
1155
1156 static const struct dquot_operations ext4_quota_operations = {
1157 #ifdef CONFIG_QUOTA
1158         .get_reserved_space = ext4_get_reserved_space,
1159 #endif
1160         .write_dquot    = ext4_write_dquot,
1161         .acquire_dquot  = ext4_acquire_dquot,
1162         .release_dquot  = ext4_release_dquot,
1163         .mark_dirty     = ext4_mark_dquot_dirty,
1164         .write_info     = ext4_write_info,
1165         .alloc_dquot    = dquot_alloc,
1166         .destroy_dquot  = dquot_destroy,
1167 };
1168
1169 static const struct quotactl_ops ext4_qctl_operations = {
1170         .quota_on       = ext4_quota_on,
1171         .quota_off      = ext4_quota_off,
1172         .quota_sync     = dquot_quota_sync,
1173         .get_info       = dquot_get_dqinfo,
1174         .set_info       = dquot_set_dqinfo,
1175         .get_dqblk      = dquot_get_dqblk,
1176         .set_dqblk      = dquot_set_dqblk
1177 };
1178 #endif
1179
1180 static const struct super_operations ext4_sops = {
1181         .alloc_inode    = ext4_alloc_inode,
1182         .destroy_inode  = ext4_destroy_inode,
1183         .write_inode    = ext4_write_inode,
1184         .dirty_inode    = ext4_dirty_inode,
1185         .drop_inode     = ext4_drop_inode,
1186         .evict_inode    = ext4_evict_inode,
1187         .put_super      = ext4_put_super,
1188         .sync_fs        = ext4_sync_fs,
1189         .freeze_fs      = ext4_freeze,
1190         .unfreeze_fs    = ext4_unfreeze,
1191         .statfs         = ext4_statfs,
1192         .remount_fs     = ext4_remount,
1193         .show_options   = ext4_show_options,
1194 #ifdef CONFIG_QUOTA
1195         .quota_read     = ext4_quota_read,
1196         .quota_write    = ext4_quota_write,
1197 #endif
1198         .bdev_try_to_free_page = bdev_try_to_free_page,
1199         .trim_fs        = ext4_trim_fs
1200 };
1201
1202 static const struct super_operations ext4_nojournal_sops = {
1203         .alloc_inode    = ext4_alloc_inode,
1204         .destroy_inode  = ext4_destroy_inode,
1205         .write_inode    = ext4_write_inode,
1206         .dirty_inode    = ext4_dirty_inode,
1207         .drop_inode     = ext4_drop_inode,
1208         .evict_inode    = ext4_evict_inode,
1209         .write_super    = ext4_write_super,
1210         .put_super      = ext4_put_super,
1211         .statfs         = ext4_statfs,
1212         .remount_fs     = ext4_remount,
1213         .show_options   = ext4_show_options,
1214 #ifdef CONFIG_QUOTA
1215         .quota_read     = ext4_quota_read,
1216         .quota_write    = ext4_quota_write,
1217 #endif
1218         .bdev_try_to_free_page = bdev_try_to_free_page,
1219 };
1220
1221 static const struct export_operations ext4_export_ops = {
1222         .fh_to_dentry = ext4_fh_to_dentry,
1223         .fh_to_parent = ext4_fh_to_parent,
1224         .get_parent = ext4_get_parent,
1225 };
1226
1227 enum {
1228         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1229         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1230         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1231         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1232         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1233         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1234         Opt_journal_update, Opt_journal_dev,
1235         Opt_journal_checksum, Opt_journal_async_commit,
1236         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1237         Opt_data_err_abort, Opt_data_err_ignore,
1238         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1239         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1240         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1241         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1242         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1243         Opt_block_validity, Opt_noblock_validity,
1244         Opt_inode_readahead_blks, Opt_journal_ioprio,
1245         Opt_dioread_nolock, Opt_dioread_lock,
1246         Opt_discard, Opt_nodiscard,
1247         Opt_init_inode_table, Opt_noinit_inode_table,
1248 };
1249
1250 static const match_table_t tokens = {
1251         {Opt_bsd_df, "bsddf"},
1252         {Opt_minix_df, "minixdf"},
1253         {Opt_grpid, "grpid"},
1254         {Opt_grpid, "bsdgroups"},
1255         {Opt_nogrpid, "nogrpid"},
1256         {Opt_nogrpid, "sysvgroups"},
1257         {Opt_resgid, "resgid=%u"},
1258         {Opt_resuid, "resuid=%u"},
1259         {Opt_sb, "sb=%u"},
1260         {Opt_err_cont, "errors=continue"},
1261         {Opt_err_panic, "errors=panic"},
1262         {Opt_err_ro, "errors=remount-ro"},
1263         {Opt_nouid32, "nouid32"},
1264         {Opt_debug, "debug"},
1265         {Opt_oldalloc, "oldalloc"},
1266         {Opt_orlov, "orlov"},
1267         {Opt_user_xattr, "user_xattr"},
1268         {Opt_nouser_xattr, "nouser_xattr"},
1269         {Opt_acl, "acl"},
1270         {Opt_noacl, "noacl"},
1271         {Opt_noload, "noload"},
1272         {Opt_noload, "norecovery"},
1273         {Opt_nobh, "nobh"},
1274         {Opt_bh, "bh"},
1275         {Opt_commit, "commit=%u"},
1276         {Opt_min_batch_time, "min_batch_time=%u"},
1277         {Opt_max_batch_time, "max_batch_time=%u"},
1278         {Opt_journal_update, "journal=update"},
1279         {Opt_journal_dev, "journal_dev=%u"},
1280         {Opt_journal_checksum, "journal_checksum"},
1281         {Opt_journal_async_commit, "journal_async_commit"},
1282         {Opt_abort, "abort"},
1283         {Opt_data_journal, "data=journal"},
1284         {Opt_data_ordered, "data=ordered"},
1285         {Opt_data_writeback, "data=writeback"},
1286         {Opt_data_err_abort, "data_err=abort"},
1287         {Opt_data_err_ignore, "data_err=ignore"},
1288         {Opt_offusrjquota, "usrjquota="},
1289         {Opt_usrjquota, "usrjquota=%s"},
1290         {Opt_offgrpjquota, "grpjquota="},
1291         {Opt_grpjquota, "grpjquota=%s"},
1292         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1293         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1294         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1295         {Opt_grpquota, "grpquota"},
1296         {Opt_noquota, "noquota"},
1297         {Opt_quota, "quota"},
1298         {Opt_usrquota, "usrquota"},
1299         {Opt_barrier, "barrier=%u"},
1300         {Opt_barrier, "barrier"},
1301         {Opt_nobarrier, "nobarrier"},
1302         {Opt_i_version, "i_version"},
1303         {Opt_stripe, "stripe=%u"},
1304         {Opt_resize, "resize"},
1305         {Opt_delalloc, "delalloc"},
1306         {Opt_nodelalloc, "nodelalloc"},
1307         {Opt_block_validity, "block_validity"},
1308         {Opt_noblock_validity, "noblock_validity"},
1309         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1310         {Opt_journal_ioprio, "journal_ioprio=%u"},
1311         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1312         {Opt_auto_da_alloc, "auto_da_alloc"},
1313         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1314         {Opt_dioread_nolock, "dioread_nolock"},
1315         {Opt_dioread_lock, "dioread_lock"},
1316         {Opt_discard, "discard"},
1317         {Opt_nodiscard, "nodiscard"},
1318         {Opt_init_inode_table, "init_itable=%u"},
1319         {Opt_init_inode_table, "init_itable"},
1320         {Opt_noinit_inode_table, "noinit_itable"},
1321         {Opt_err, NULL},
1322 };
1323
1324 static ext4_fsblk_t get_sb_block(void **data)
1325 {
1326         ext4_fsblk_t    sb_block;
1327         char            *options = (char *) *data;
1328
1329         if (!options || strncmp(options, "sb=", 3) != 0)
1330                 return 1;       /* Default location */
1331
1332         options += 3;
1333         /* TODO: use simple_strtoll with >32bit ext4 */
1334         sb_block = simple_strtoul(options, &options, 0);
1335         if (*options && *options != ',') {
1336                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1337                        (char *) *data);
1338                 return 1;
1339         }
1340         if (*options == ',')
1341                 options++;
1342         *data = (void *) options;
1343
1344         return sb_block;
1345 }
1346
1347 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1348 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1349         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1350
1351 #ifdef CONFIG_QUOTA
1352 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1353 {
1354         struct ext4_sb_info *sbi = EXT4_SB(sb);
1355         char *qname;
1356
1357         if (sb_any_quota_loaded(sb) &&
1358                 !sbi->s_qf_names[qtype]) {
1359                 ext4_msg(sb, KERN_ERR,
1360                         "Cannot change journaled "
1361                         "quota options when quota turned on");
1362                 return 0;
1363         }
1364         qname = match_strdup(args);
1365         if (!qname) {
1366                 ext4_msg(sb, KERN_ERR,
1367                         "Not enough memory for storing quotafile name");
1368                 return 0;
1369         }
1370         if (sbi->s_qf_names[qtype] &&
1371                 strcmp(sbi->s_qf_names[qtype], qname)) {
1372                 ext4_msg(sb, KERN_ERR,
1373                         "%s quota file already specified", QTYPE2NAME(qtype));
1374                 kfree(qname);
1375                 return 0;
1376         }
1377         sbi->s_qf_names[qtype] = qname;
1378         if (strchr(sbi->s_qf_names[qtype], '/')) {
1379                 ext4_msg(sb, KERN_ERR,
1380                         "quotafile must be on filesystem root");
1381                 kfree(sbi->s_qf_names[qtype]);
1382                 sbi->s_qf_names[qtype] = NULL;
1383                 return 0;
1384         }
1385         set_opt(sbi->s_mount_opt, QUOTA);
1386         return 1;
1387 }
1388
1389 static int clear_qf_name(struct super_block *sb, int qtype)
1390 {
1391
1392         struct ext4_sb_info *sbi = EXT4_SB(sb);
1393
1394         if (sb_any_quota_loaded(sb) &&
1395                 sbi->s_qf_names[qtype]) {
1396                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1397                         " when quota turned on");
1398                 return 0;
1399         }
1400         /*
1401          * The space will be released later when all options are confirmed
1402          * to be correct
1403          */
1404         sbi->s_qf_names[qtype] = NULL;
1405         return 1;
1406 }
1407 #endif
1408
1409 static int parse_options(char *options, struct super_block *sb,
1410                          unsigned long *journal_devnum,
1411                          unsigned int *journal_ioprio,
1412                          ext4_fsblk_t *n_blocks_count, int is_remount)
1413 {
1414         struct ext4_sb_info *sbi = EXT4_SB(sb);
1415         char *p;
1416         substring_t args[MAX_OPT_ARGS];
1417         int data_opt = 0;
1418         int option;
1419 #ifdef CONFIG_QUOTA
1420         int qfmt;
1421 #endif
1422
1423         if (!options)
1424                 return 1;
1425
1426         while ((p = strsep(&options, ",")) != NULL) {
1427                 int token;
1428                 if (!*p)
1429                         continue;
1430
1431                 /*
1432                  * Initialize args struct so we know whether arg was
1433                  * found; some options take optional arguments.
1434                  */
1435                 args[0].to = args[0].from = 0;
1436                 token = match_token(p, tokens, args);
1437                 switch (token) {
1438                 case Opt_bsd_df:
1439                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1440                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1441                         break;
1442                 case Opt_minix_df:
1443                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1444                         set_opt(sbi->s_mount_opt, MINIX_DF);
1445
1446                         break;
1447                 case Opt_grpid:
1448                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1449                         set_opt(sbi->s_mount_opt, GRPID);
1450
1451                         break;
1452                 case Opt_nogrpid:
1453                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1454                         clear_opt(sbi->s_mount_opt, GRPID);
1455
1456                         break;
1457                 case Opt_resuid:
1458                         if (match_int(&args[0], &option))
1459                                 return 0;
1460                         sbi->s_resuid = option;
1461                         break;
1462                 case Opt_resgid:
1463                         if (match_int(&args[0], &option))
1464                                 return 0;
1465                         sbi->s_resgid = option;
1466                         break;
1467                 case Opt_sb:
1468                         /* handled by get_sb_block() instead of here */
1469                         /* *sb_block = match_int(&args[0]); */
1470                         break;
1471                 case Opt_err_panic:
1472                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1473                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1474                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1475                         break;
1476                 case Opt_err_ro:
1477                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1478                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1479                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1480                         break;
1481                 case Opt_err_cont:
1482                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1483                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1484                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1485                         break;
1486                 case Opt_nouid32:
1487                         set_opt(sbi->s_mount_opt, NO_UID32);
1488                         break;
1489                 case Opt_debug:
1490                         set_opt(sbi->s_mount_opt, DEBUG);
1491                         break;
1492                 case Opt_oldalloc:
1493                         set_opt(sbi->s_mount_opt, OLDALLOC);
1494                         break;
1495                 case Opt_orlov:
1496                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1497                         break;
1498 #ifdef CONFIG_EXT4_FS_XATTR
1499                 case Opt_user_xattr:
1500                         set_opt(sbi->s_mount_opt, XATTR_USER);
1501                         break;
1502                 case Opt_nouser_xattr:
1503                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1504                         break;
1505 #else
1506                 case Opt_user_xattr:
1507                 case Opt_nouser_xattr:
1508                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1509                         break;
1510 #endif
1511 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1512                 case Opt_acl:
1513                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1514                         break;
1515                 case Opt_noacl:
1516                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1517                         break;
1518 #else
1519                 case Opt_acl:
1520                 case Opt_noacl:
1521                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1522                         break;
1523 #endif
1524                 case Opt_journal_update:
1525                         /* @@@ FIXME */
1526                         /* Eventually we will want to be able to create
1527                            a journal file here.  For now, only allow the
1528                            user to specify an existing inode to be the
1529                            journal file. */
1530                         if (is_remount) {
1531                                 ext4_msg(sb, KERN_ERR,
1532                                          "Cannot specify journal on remount");
1533                                 return 0;
1534                         }
1535                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1536                         break;
1537                 case Opt_journal_dev:
1538                         if (is_remount) {
1539                                 ext4_msg(sb, KERN_ERR,
1540                                         "Cannot specify journal on remount");
1541                                 return 0;
1542                         }
1543                         if (match_int(&args[0], &option))
1544                                 return 0;
1545                         *journal_devnum = option;
1546                         break;
1547                 case Opt_journal_checksum:
1548                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1549                         break;
1550                 case Opt_journal_async_commit:
1551                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1552                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1553                         break;
1554                 case Opt_noload:
1555                         set_opt(sbi->s_mount_opt, NOLOAD);
1556                         break;
1557                 case Opt_commit:
1558                         if (match_int(&args[0], &option))
1559                                 return 0;
1560                         if (option < 0)
1561                                 return 0;
1562                         if (option == 0)
1563                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1564                         sbi->s_commit_interval = HZ * option;
1565                         break;
1566                 case Opt_max_batch_time:
1567                         if (match_int(&args[0], &option))
1568                                 return 0;
1569                         if (option < 0)
1570                                 return 0;
1571                         if (option == 0)
1572                                 option = EXT4_DEF_MAX_BATCH_TIME;
1573                         sbi->s_max_batch_time = option;
1574                         break;
1575                 case Opt_min_batch_time:
1576                         if (match_int(&args[0], &option))
1577                                 return 0;
1578                         if (option < 0)
1579                                 return 0;
1580                         sbi->s_min_batch_time = option;
1581                         break;
1582                 case Opt_data_journal:
1583                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1584                         goto datacheck;
1585                 case Opt_data_ordered:
1586                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1587                         goto datacheck;
1588                 case Opt_data_writeback:
1589                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1590                 datacheck:
1591                         if (is_remount) {
1592                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1593                                         ext4_msg(sb, KERN_ERR,
1594                                                 "Cannot change data mode on remount");
1595                                         return 0;
1596                                 }
1597                         } else {
1598                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1599                                 sbi->s_mount_opt |= data_opt;
1600                         }
1601                         break;
1602                 case Opt_data_err_abort:
1603                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1604                         break;
1605                 case Opt_data_err_ignore:
1606                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1607                         break;
1608 #ifdef CONFIG_QUOTA
1609                 case Opt_usrjquota:
1610                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1611                                 return 0;
1612                         break;
1613                 case Opt_grpjquota:
1614                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1615                                 return 0;
1616                         break;
1617                 case Opt_offusrjquota:
1618                         if (!clear_qf_name(sb, USRQUOTA))
1619                                 return 0;
1620                         break;
1621                 case Opt_offgrpjquota:
1622                         if (!clear_qf_name(sb, GRPQUOTA))
1623                                 return 0;
1624                         break;
1625
1626                 case Opt_jqfmt_vfsold:
1627                         qfmt = QFMT_VFS_OLD;
1628                         goto set_qf_format;
1629                 case Opt_jqfmt_vfsv0:
1630                         qfmt = QFMT_VFS_V0;
1631                         goto set_qf_format;
1632                 case Opt_jqfmt_vfsv1:
1633                         qfmt = QFMT_VFS_V1;
1634 set_qf_format:
1635                         if (sb_any_quota_loaded(sb) &&
1636                             sbi->s_jquota_fmt != qfmt) {
1637                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1638                                         "journaled quota options when "
1639                                         "quota turned on");
1640                                 return 0;
1641                         }
1642                         sbi->s_jquota_fmt = qfmt;
1643                         break;
1644                 case Opt_quota:
1645                 case Opt_usrquota:
1646                         set_opt(sbi->s_mount_opt, QUOTA);
1647                         set_opt(sbi->s_mount_opt, USRQUOTA);
1648                         break;
1649                 case Opt_grpquota:
1650                         set_opt(sbi->s_mount_opt, QUOTA);
1651                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1652                         break;
1653                 case Opt_noquota:
1654                         if (sb_any_quota_loaded(sb)) {
1655                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1656                                         "options when quota turned on");
1657                                 return 0;
1658                         }
1659                         clear_opt(sbi->s_mount_opt, QUOTA);
1660                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1661                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1662                         break;
1663 #else
1664                 case Opt_quota:
1665                 case Opt_usrquota:
1666                 case Opt_grpquota:
1667                         ext4_msg(sb, KERN_ERR,
1668                                 "quota options not supported");
1669                         break;
1670                 case Opt_usrjquota:
1671                 case Opt_grpjquota:
1672                 case Opt_offusrjquota:
1673                 case Opt_offgrpjquota:
1674                 case Opt_jqfmt_vfsold:
1675                 case Opt_jqfmt_vfsv0:
1676                 case Opt_jqfmt_vfsv1:
1677                         ext4_msg(sb, KERN_ERR,
1678                                 "journaled quota options not supported");
1679                         break;
1680                 case Opt_noquota:
1681                         break;
1682 #endif
1683                 case Opt_abort:
1684                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1685                         break;
1686                 case Opt_nobarrier:
1687                         clear_opt(sbi->s_mount_opt, BARRIER);
1688                         break;
1689                 case Opt_barrier:
1690                         if (args[0].from) {
1691                                 if (match_int(&args[0], &option))
1692                                         return 0;
1693                         } else
1694                                 option = 1;     /* No argument, default to 1 */
1695                         if (option)
1696                                 set_opt(sbi->s_mount_opt, BARRIER);
1697                         else
1698                                 clear_opt(sbi->s_mount_opt, BARRIER);
1699                         break;
1700                 case Opt_ignore:
1701                         break;
1702                 case Opt_resize:
1703                         if (!is_remount) {
1704                                 ext4_msg(sb, KERN_ERR,
1705                                         "resize option only available "
1706                                         "for remount");
1707                                 return 0;
1708                         }
1709                         if (match_int(&args[0], &option) != 0)
1710                                 return 0;
1711                         *n_blocks_count = option;
1712                         break;
1713                 case Opt_nobh:
1714                         ext4_msg(sb, KERN_WARNING,
1715                                  "Ignoring deprecated nobh option");
1716                         break;
1717                 case Opt_bh:
1718                         ext4_msg(sb, KERN_WARNING,
1719                                  "Ignoring deprecated bh option");
1720                         break;
1721                 case Opt_i_version:
1722                         set_opt(sbi->s_mount_opt, I_VERSION);
1723                         sb->s_flags |= MS_I_VERSION;
1724                         break;
1725                 case Opt_nodelalloc:
1726                         clear_opt(sbi->s_mount_opt, DELALLOC);
1727                         break;
1728                 case Opt_stripe:
1729                         if (match_int(&args[0], &option))
1730                                 return 0;
1731                         if (option < 0)
1732                                 return 0;
1733                         sbi->s_stripe = option;
1734                         break;
1735                 case Opt_delalloc:
1736                         set_opt(sbi->s_mount_opt, DELALLOC);
1737                         break;
1738                 case Opt_block_validity:
1739                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1740                         break;
1741                 case Opt_noblock_validity:
1742                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1743                         break;
1744                 case Opt_inode_readahead_blks:
1745                         if (match_int(&args[0], &option))
1746                                 return 0;
1747                         if (option < 0 || option > (1 << 30))
1748                                 return 0;
1749                         if (!is_power_of_2(option)) {
1750                                 ext4_msg(sb, KERN_ERR,
1751                                          "EXT4-fs: inode_readahead_blks"
1752                                          " must be a power of 2");
1753                                 return 0;
1754                         }
1755                         sbi->s_inode_readahead_blks = option;
1756                         break;
1757                 case Opt_journal_ioprio:
1758                         if (match_int(&args[0], &option))
1759                                 return 0;
1760                         if (option < 0 || option > 7)
1761                                 break;
1762                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1763                                                             option);
1764                         break;
1765                 case Opt_noauto_da_alloc:
1766                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1767                         break;
1768                 case Opt_auto_da_alloc:
1769                         if (args[0].from) {
1770                                 if (match_int(&args[0], &option))
1771                                         return 0;
1772                         } else
1773                                 option = 1;     /* No argument, default to 1 */
1774                         if (option)
1775                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1776                         else
1777                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1778                         break;
1779                 case Opt_discard:
1780                         set_opt(sbi->s_mount_opt, DISCARD);
1781                         break;
1782                 case Opt_nodiscard:
1783                         clear_opt(sbi->s_mount_opt, DISCARD);
1784                         break;
1785                 case Opt_dioread_nolock:
1786                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1787                         break;
1788                 case Opt_dioread_lock:
1789                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1790                         break;
1791                 case Opt_init_inode_table:
1792                         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1793                         if (args[0].from) {
1794                                 if (match_int(&args[0], &option))
1795                                         return 0;
1796                         } else
1797                                 option = EXT4_DEF_LI_WAIT_MULT;
1798                         if (option < 0)
1799                                 return 0;
1800                         sbi->s_li_wait_mult = option;
1801                         break;
1802                 case Opt_noinit_inode_table:
1803                         clear_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1804                         break;
1805                 default:
1806                         ext4_msg(sb, KERN_ERR,
1807                                "Unrecognized mount option \"%s\" "
1808                                "or missing value", p);
1809                         return 0;
1810                 }
1811         }
1812 #ifdef CONFIG_QUOTA
1813         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1814                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1815                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1816
1817                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1818                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1819
1820                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1821                         ext4_msg(sb, KERN_ERR, "old and new quota "
1822                                         "format mixing");
1823                         return 0;
1824                 }
1825
1826                 if (!sbi->s_jquota_fmt) {
1827                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1828                                         "not specified");
1829                         return 0;
1830                 }
1831         } else {
1832                 if (sbi->s_jquota_fmt) {
1833                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1834                                         "specified with no journaling "
1835                                         "enabled");
1836                         return 0;
1837                 }
1838         }
1839 #endif
1840         return 1;
1841 }
1842
1843 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1844                             int read_only)
1845 {
1846         struct ext4_sb_info *sbi = EXT4_SB(sb);
1847         int res = 0;
1848
1849         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1850                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1851                          "forcing read-only mode");
1852                 res = MS_RDONLY;
1853         }
1854         if (read_only)
1855                 return res;
1856         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1857                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1858                          "running e2fsck is recommended");
1859         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1860                 ext4_msg(sb, KERN_WARNING,
1861                          "warning: mounting fs with errors, "
1862                          "running e2fsck is recommended");
1863         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1864                  le16_to_cpu(es->s_mnt_count) >=
1865                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1866                 ext4_msg(sb, KERN_WARNING,
1867                          "warning: maximal mount count reached, "
1868                          "running e2fsck is recommended");
1869         else if (le32_to_cpu(es->s_checkinterval) &&
1870                 (le32_to_cpu(es->s_lastcheck) +
1871                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1872                 ext4_msg(sb, KERN_WARNING,
1873                          "warning: checktime reached, "
1874                          "running e2fsck is recommended");
1875         if (!sbi->s_journal)
1876                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1877         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1878                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1879         le16_add_cpu(&es->s_mnt_count, 1);
1880         es->s_mtime = cpu_to_le32(get_seconds());
1881         ext4_update_dynamic_rev(sb);
1882         if (sbi->s_journal)
1883                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1884
1885         ext4_commit_super(sb, 1);
1886         if (test_opt(sb, DEBUG))
1887                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1888                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1889                         sb->s_blocksize,
1890                         sbi->s_groups_count,
1891                         EXT4_BLOCKS_PER_GROUP(sb),
1892                         EXT4_INODES_PER_GROUP(sb),
1893                         sbi->s_mount_opt);
1894
1895         return res;
1896 }
1897
1898 static int ext4_fill_flex_info(struct super_block *sb)
1899 {
1900         struct ext4_sb_info *sbi = EXT4_SB(sb);
1901         struct ext4_group_desc *gdp = NULL;
1902         ext4_group_t flex_group_count;
1903         ext4_group_t flex_group;
1904         int groups_per_flex = 0;
1905         size_t size;
1906         int i;
1907
1908         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1909         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1910
1911         if (groups_per_flex < 2) {
1912                 sbi->s_log_groups_per_flex = 0;
1913                 return 1;
1914         }
1915
1916         /* We allocate both existing and potentially added groups */
1917         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1918                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1919                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1920         size = flex_group_count * sizeof(struct flex_groups);
1921         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1922         if (sbi->s_flex_groups == NULL) {
1923                 sbi->s_flex_groups = vmalloc(size);
1924                 if (sbi->s_flex_groups)
1925                         memset(sbi->s_flex_groups, 0, size);
1926         }
1927         if (sbi->s_flex_groups == NULL) {
1928                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1929                                 "%u flex groups", flex_group_count);
1930                 goto failed;
1931         }
1932
1933         for (i = 0; i < sbi->s_groups_count; i++) {
1934                 gdp = ext4_get_group_desc(sb, i, NULL);
1935
1936                 flex_group = ext4_flex_group(sbi, i);
1937                 atomic_add(ext4_free_inodes_count(sb, gdp),
1938                            &sbi->s_flex_groups[flex_group].free_inodes);
1939                 atomic_add(ext4_free_blks_count(sb, gdp),
1940                            &sbi->s_flex_groups[flex_group].free_blocks);
1941                 atomic_add(ext4_used_dirs_count(sb, gdp),
1942                            &sbi->s_flex_groups[flex_group].used_dirs);
1943         }
1944
1945         return 1;
1946 failed:
1947         return 0;
1948 }
1949
1950 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1951                             struct ext4_group_desc *gdp)
1952 {
1953         __u16 crc = 0;
1954
1955         if (sbi->s_es->s_feature_ro_compat &
1956             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1957                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1958                 __le32 le_group = cpu_to_le32(block_group);
1959
1960                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1961                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1962                 crc = crc16(crc, (__u8 *)gdp, offset);
1963                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1964                 /* for checksum of struct ext4_group_desc do the rest...*/
1965                 if ((sbi->s_es->s_feature_incompat &
1966                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1967                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1968                         crc = crc16(crc, (__u8 *)gdp + offset,
1969                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1970                                         offset);
1971         }
1972
1973         return cpu_to_le16(crc);
1974 }
1975
1976 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1977                                 struct ext4_group_desc *gdp)
1978 {
1979         if ((sbi->s_es->s_feature_ro_compat &
1980              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1981             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1982                 return 0;
1983
1984         return 1;
1985 }
1986
1987 /* Called at mount-time, super-block is locked */
1988 static int ext4_check_descriptors(struct super_block *sb,
1989                                   ext4_group_t *first_not_zeroed)
1990 {
1991         struct ext4_sb_info *sbi = EXT4_SB(sb);
1992         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1993         ext4_fsblk_t last_block;
1994         ext4_fsblk_t block_bitmap;
1995         ext4_fsblk_t inode_bitmap;
1996         ext4_fsblk_t inode_table;
1997         int flexbg_flag = 0;
1998         ext4_group_t i, grp = sbi->s_groups_count;
1999
2000         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2001                 flexbg_flag = 1;
2002
2003         ext4_debug("Checking group descriptors");
2004
2005         for (i = 0; i < sbi->s_groups_count; i++) {
2006                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2007
2008                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2009                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2010                 else
2011                         last_block = first_block +
2012                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2013
2014                 if ((grp == sbi->s_groups_count) &&
2015                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2016                         grp = i;
2017
2018                 block_bitmap = ext4_block_bitmap(sb, gdp);
2019                 if (block_bitmap < first_block || block_bitmap > last_block) {
2020                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2021                                "Block bitmap for group %u not in group "
2022                                "(block %llu)!", i, block_bitmap);
2023                         return 0;
2024                 }
2025                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2026                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2027                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2028                                "Inode bitmap for group %u not in group "
2029                                "(block %llu)!", i, inode_bitmap);
2030                         return 0;
2031                 }
2032                 inode_table = ext4_inode_table(sb, gdp);
2033                 if (inode_table < first_block ||
2034                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2035                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2036                                "Inode table for group %u not in group "
2037                                "(block %llu)!", i, inode_table);
2038                         return 0;
2039                 }
2040                 ext4_lock_group(sb, i);
2041                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2042                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2043                                  "Checksum for group %u failed (%u!=%u)",
2044                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2045                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2046                         if (!(sb->s_flags & MS_RDONLY)) {
2047                                 ext4_unlock_group(sb, i);
2048                                 return 0;
2049                         }
2050                 }
2051                 ext4_unlock_group(sb, i);
2052                 if (!flexbg_flag)
2053                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2054         }
2055         if (NULL != first_not_zeroed)
2056                 *first_not_zeroed = grp;
2057
2058         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2059         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2060         return 1;
2061 }
2062
2063 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2064  * the superblock) which were deleted from all directories, but held open by
2065  * a process at the time of a crash.  We walk the list and try to delete these
2066  * inodes at recovery time (only with a read-write filesystem).
2067  *
2068  * In order to keep the orphan inode chain consistent during traversal (in
2069  * case of crash during recovery), we link each inode into the superblock
2070  * orphan list_head and handle it the same way as an inode deletion during
2071  * normal operation (which journals the operations for us).
2072  *
2073  * We only do an iget() and an iput() on each inode, which is very safe if we
2074  * accidentally point at an in-use or already deleted inode.  The worst that
2075  * can happen in this case is that we get a "bit already cleared" message from
2076  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2077  * e2fsck was run on this filesystem, and it must have already done the orphan
2078  * inode cleanup for us, so we can safely abort without any further action.
2079  */
2080 static void ext4_orphan_cleanup(struct super_block *sb,
2081                                 struct ext4_super_block *es)
2082 {
2083         unsigned int s_flags = sb->s_flags;
2084         int nr_orphans = 0, nr_truncates = 0;
2085 #ifdef CONFIG_QUOTA
2086         int i;
2087 #endif
2088         if (!es->s_last_orphan) {
2089                 jbd_debug(4, "no orphan inodes to clean up\n");
2090                 return;
2091         }
2092
2093         if (bdev_read_only(sb->s_bdev)) {
2094                 ext4_msg(sb, KERN_ERR, "write access "
2095                         "unavailable, skipping orphan cleanup");
2096                 return;
2097         }
2098
2099         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2100                 if (es->s_last_orphan)
2101                         jbd_debug(1, "Errors on filesystem, "
2102                                   "clearing orphan list.\n");
2103                 es->s_last_orphan = 0;
2104                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2105                 return;
2106         }
2107
2108         if (s_flags & MS_RDONLY) {
2109                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2110                 sb->s_flags &= ~MS_RDONLY;
2111         }
2112 #ifdef CONFIG_QUOTA
2113         /* Needed for iput() to work correctly and not trash data */
2114         sb->s_flags |= MS_ACTIVE;
2115         /* Turn on quotas so that they are updated correctly */
2116         for (i = 0; i < MAXQUOTAS; i++) {
2117                 if (EXT4_SB(sb)->s_qf_names[i]) {
2118                         int ret = ext4_quota_on_mount(sb, i);
2119                         if (ret < 0)
2120                                 ext4_msg(sb, KERN_ERR,
2121                                         "Cannot turn on journaled "
2122                                         "quota: error %d", ret);
2123                 }
2124         }
2125 #endif
2126
2127         while (es->s_last_orphan) {
2128                 struct inode *inode;
2129
2130                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2131                 if (IS_ERR(inode)) {
2132                         es->s_last_orphan = 0;
2133                         break;
2134                 }
2135
2136                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2137                 dquot_initialize(inode);
2138                 if (inode->i_nlink) {
2139                         ext4_msg(sb, KERN_DEBUG,
2140                                 "%s: truncating inode %lu to %lld bytes",
2141                                 __func__, inode->i_ino, inode->i_size);
2142                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2143                                   inode->i_ino, inode->i_size);
2144                         ext4_truncate(inode);
2145                         nr_truncates++;
2146                 } else {
2147                         ext4_msg(sb, KERN_DEBUG,
2148                                 "%s: deleting unreferenced inode %lu",
2149                                 __func__, inode->i_ino);
2150                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2151                                   inode->i_ino);
2152                         nr_orphans++;
2153                 }
2154                 iput(inode);  /* The delete magic happens here! */
2155         }
2156
2157 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2158
2159         if (nr_orphans)
2160                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2161                        PLURAL(nr_orphans));
2162         if (nr_truncates)
2163                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2164                        PLURAL(nr_truncates));
2165 #ifdef CONFIG_QUOTA
2166         /* Turn quotas off */
2167         for (i = 0; i < MAXQUOTAS; i++) {
2168                 if (sb_dqopt(sb)->files[i])
2169                         dquot_quota_off(sb, i);
2170         }
2171 #endif
2172         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2173 }
2174
2175 /*
2176  * Maximal extent format file size.
2177  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2178  * extent format containers, within a sector_t, and within i_blocks
2179  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2180  * so that won't be a limiting factor.
2181  *
2182  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2183  */
2184 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2185 {
2186         loff_t res;
2187         loff_t upper_limit = MAX_LFS_FILESIZE;
2188
2189         /* small i_blocks in vfs inode? */
2190         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2191                 /*
2192                  * CONFIG_LBDAF is not enabled implies the inode
2193                  * i_block represent total blocks in 512 bytes
2194                  * 32 == size of vfs inode i_blocks * 8
2195                  */
2196                 upper_limit = (1LL << 32) - 1;
2197
2198                 /* total blocks in file system block size */
2199                 upper_limit >>= (blkbits - 9);
2200                 upper_limit <<= blkbits;
2201         }
2202
2203         /* 32-bit extent-start container, ee_block */
2204         res = 1LL << 32;
2205         res <<= blkbits;
2206         res -= 1;
2207
2208         /* Sanity check against vm- & vfs- imposed limits */
2209         if (res > upper_limit)
2210                 res = upper_limit;
2211
2212         return res;
2213 }
2214
2215 /*
2216  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2217  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2218  * We need to be 1 filesystem block less than the 2^48 sector limit.
2219  */
2220 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2221 {
2222         loff_t res = EXT4_NDIR_BLOCKS;
2223         int meta_blocks;
2224         loff_t upper_limit;
2225         /* This is calculated to be the largest file size for a dense, block
2226          * mapped file such that the file's total number of 512-byte sectors,
2227          * including data and all indirect blocks, does not exceed (2^48 - 1).
2228          *
2229          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2230          * number of 512-byte sectors of the file.
2231          */
2232
2233         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2234                 /*
2235                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2236                  * the inode i_block field represents total file blocks in
2237                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2238                  */
2239                 upper_limit = (1LL << 32) - 1;
2240
2241                 /* total blocks in file system block size */
2242                 upper_limit >>= (bits - 9);
2243
2244         } else {
2245                 /*
2246                  * We use 48 bit ext4_inode i_blocks
2247                  * With EXT4_HUGE_FILE_FL set the i_blocks
2248                  * represent total number of blocks in
2249                  * file system block size
2250                  */
2251                 upper_limit = (1LL << 48) - 1;
2252
2253         }
2254
2255         /* indirect blocks */
2256         meta_blocks = 1;
2257         /* double indirect blocks */
2258         meta_blocks += 1 + (1LL << (bits-2));
2259         /* tripple indirect blocks */
2260         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2261
2262         upper_limit -= meta_blocks;
2263         upper_limit <<= bits;
2264
2265         res += 1LL << (bits-2);
2266         res += 1LL << (2*(bits-2));
2267         res += 1LL << (3*(bits-2));
2268         res <<= bits;
2269         if (res > upper_limit)
2270                 res = upper_limit;
2271
2272         if (res > MAX_LFS_FILESIZE)
2273                 res = MAX_LFS_FILESIZE;
2274
2275         return res;
2276 }
2277
2278 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2279                                    ext4_fsblk_t logical_sb_block, int nr)
2280 {
2281         struct ext4_sb_info *sbi = EXT4_SB(sb);
2282         ext4_group_t bg, first_meta_bg;
2283         int has_super = 0;
2284
2285         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2286
2287         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2288             nr < first_meta_bg)
2289                 return logical_sb_block + nr + 1;
2290         bg = sbi->s_desc_per_block * nr;
2291         if (ext4_bg_has_super(sb, bg))
2292                 has_super = 1;
2293
2294         return (has_super + ext4_group_first_block_no(sb, bg));
2295 }
2296
2297 /**
2298  * ext4_get_stripe_size: Get the stripe size.
2299  * @sbi: In memory super block info
2300  *
2301  * If we have specified it via mount option, then
2302  * use the mount option value. If the value specified at mount time is
2303  * greater than the blocks per group use the super block value.
2304  * If the super block value is greater than blocks per group return 0.
2305  * Allocator needs it be less than blocks per group.
2306  *
2307  */
2308 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2309 {
2310         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2311         unsigned long stripe_width =
2312                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2313
2314         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2315                 return sbi->s_stripe;
2316
2317         if (stripe_width <= sbi->s_blocks_per_group)
2318                 return stripe_width;
2319
2320         if (stride <= sbi->s_blocks_per_group)
2321                 return stride;
2322
2323         return 0;
2324 }
2325
2326 /* sysfs supprt */
2327
2328 struct ext4_attr {
2329         struct attribute attr;
2330         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2331         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2332                          const char *, size_t);
2333         int offset;
2334 };
2335
2336 static int parse_strtoul(const char *buf,
2337                 unsigned long max, unsigned long *value)
2338 {
2339         char *endp;
2340
2341         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2342         endp = skip_spaces(endp);
2343         if (*endp || *value > max)
2344                 return -EINVAL;
2345
2346         return 0;
2347 }
2348
2349 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2350                                               struct ext4_sb_info *sbi,
2351                                               char *buf)
2352 {
2353         return snprintf(buf, PAGE_SIZE, "%llu\n",
2354                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2355 }
2356
2357 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2358                                          struct ext4_sb_info *sbi, char *buf)
2359 {
2360         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2361
2362         if (!sb->s_bdev->bd_part)
2363                 return snprintf(buf, PAGE_SIZE, "0\n");
2364         return snprintf(buf, PAGE_SIZE, "%lu\n",
2365                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2366                          sbi->s_sectors_written_start) >> 1);
2367 }
2368
2369 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2370                                           struct ext4_sb_info *sbi, char *buf)
2371 {
2372         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2373
2374         if (!sb->s_bdev->bd_part)
2375                 return snprintf(buf, PAGE_SIZE, "0\n");
2376         return snprintf(buf, PAGE_SIZE, "%llu\n",
2377                         (unsigned long long)(sbi->s_kbytes_written +
2378                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2379                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2380 }
2381
2382 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2383                                           struct ext4_sb_info *sbi,
2384                                           const char *buf, size_t count)
2385 {
2386         unsigned long t;
2387
2388         if (parse_strtoul(buf, 0x40000000, &t))
2389                 return -EINVAL;
2390
2391         if (!is_power_of_2(t))
2392                 return -EINVAL;
2393
2394         sbi->s_inode_readahead_blks = t;
2395         return count;
2396 }
2397
2398 static ssize_t sbi_ui_show(struct ext4_attr *a,
2399                            struct ext4_sb_info *sbi, char *buf)
2400 {
2401         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2402
2403         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2404 }
2405
2406 static ssize_t sbi_ui_store(struct ext4_attr *a,
2407                             struct ext4_sb_info *sbi,
2408                             const char *buf, size_t count)
2409 {
2410         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2411         unsigned long t;
2412
2413         if (parse_strtoul(buf, 0xffffffff, &t))
2414                 return -EINVAL;
2415         *ui = t;
2416         return count;
2417 }
2418
2419 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2420 static struct ext4_attr ext4_attr_##_name = {                   \
2421         .attr = {.name = __stringify(_name), .mode = _mode },   \
2422         .show   = _show,                                        \
2423         .store  = _store,                                       \
2424         .offset = offsetof(struct ext4_sb_info, _elname),       \
2425 }
2426 #define EXT4_ATTR(name, mode, show, store) \
2427 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2428
2429 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2430 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2431 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2432 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2433         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2434 #define ATTR_LIST(name) &ext4_attr_##name.attr
2435
2436 EXT4_RO_ATTR(delayed_allocation_blocks);
2437 EXT4_RO_ATTR(session_write_kbytes);
2438 EXT4_RO_ATTR(lifetime_write_kbytes);
2439 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2440                  inode_readahead_blks_store, s_inode_readahead_blks);
2441 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2442 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2443 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2444 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2445 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2446 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2447 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2448 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2449
2450 static struct attribute *ext4_attrs[] = {
2451         ATTR_LIST(delayed_allocation_blocks),
2452         ATTR_LIST(session_write_kbytes),
2453         ATTR_LIST(lifetime_write_kbytes),
2454         ATTR_LIST(inode_readahead_blks),
2455         ATTR_LIST(inode_goal),
2456         ATTR_LIST(mb_stats),
2457         ATTR_LIST(mb_max_to_scan),
2458         ATTR_LIST(mb_min_to_scan),
2459         ATTR_LIST(mb_order2_req),
2460         ATTR_LIST(mb_stream_req),
2461         ATTR_LIST(mb_group_prealloc),
2462         ATTR_LIST(max_writeback_mb_bump),
2463         NULL,
2464 };
2465
2466 /* Features this copy of ext4 supports */
2467 EXT4_INFO_ATTR(lazy_itable_init);
2468 EXT4_INFO_ATTR(batched_discard);
2469
2470 static struct attribute *ext4_feat_attrs[] = {
2471         ATTR_LIST(lazy_itable_init),
2472         ATTR_LIST(batched_discard),
2473         NULL,
2474 };
2475
2476 static ssize_t ext4_attr_show(struct kobject *kobj,
2477                               struct attribute *attr, char *buf)
2478 {
2479         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2480                                                 s_kobj);
2481         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2482
2483         return a->show ? a->show(a, sbi, buf) : 0;
2484 }
2485
2486 static ssize_t ext4_attr_store(struct kobject *kobj,
2487                                struct attribute *attr,
2488                                const char *buf, size_t len)
2489 {
2490         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2491                                                 s_kobj);
2492         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2493
2494         return a->store ? a->store(a, sbi, buf, len) : 0;
2495 }
2496
2497 static void ext4_sb_release(struct kobject *kobj)
2498 {
2499         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2500                                                 s_kobj);
2501         complete(&sbi->s_kobj_unregister);
2502 }
2503
2504 static const struct sysfs_ops ext4_attr_ops = {
2505         .show   = ext4_attr_show,
2506         .store  = ext4_attr_store,
2507 };
2508
2509 static struct kobj_type ext4_ktype = {
2510         .default_attrs  = ext4_attrs,
2511         .sysfs_ops      = &ext4_attr_ops,
2512         .release        = ext4_sb_release,
2513 };
2514
2515 static void ext4_feat_release(struct kobject *kobj)
2516 {
2517         complete(&ext4_feat->f_kobj_unregister);
2518 }
2519
2520 static struct kobj_type ext4_feat_ktype = {
2521         .default_attrs  = ext4_feat_attrs,
2522         .sysfs_ops      = &ext4_attr_ops,
2523         .release        = ext4_feat_release,
2524 };
2525
2526 /*
2527  * Check whether this filesystem can be mounted based on
2528  * the features present and the RDONLY/RDWR mount requested.
2529  * Returns 1 if this filesystem can be mounted as requested,
2530  * 0 if it cannot be.
2531  */
2532 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2533 {
2534         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2535                 ext4_msg(sb, KERN_ERR,
2536                         "Couldn't mount because of "
2537                         "unsupported optional features (%x)",
2538                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2539                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2540                 return 0;
2541         }
2542
2543         if (readonly)
2544                 return 1;
2545
2546         /* Check that feature set is OK for a read-write mount */
2547         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2548                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2549                          "unsupported optional features (%x)",
2550                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2551                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2552                 return 0;
2553         }
2554         /*
2555          * Large file size enabled file system can only be mounted
2556          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2557          */
2558         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2559                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2560                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2561                                  "cannot be mounted RDWR without "
2562                                  "CONFIG_LBDAF");
2563                         return 0;
2564                 }
2565         }
2566         return 1;
2567 }
2568
2569 /*
2570  * This function is called once a day if we have errors logged
2571  * on the file system
2572  */
2573 static void print_daily_error_info(unsigned long arg)
2574 {
2575         struct super_block *sb = (struct super_block *) arg;
2576         struct ext4_sb_info *sbi;
2577         struct ext4_super_block *es;
2578
2579         sbi = EXT4_SB(sb);
2580         es = sbi->s_es;
2581
2582         if (es->s_error_count)
2583                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2584                          le32_to_cpu(es->s_error_count));
2585         if (es->s_first_error_time) {
2586                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2587                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2588                        (int) sizeof(es->s_first_error_func),
2589                        es->s_first_error_func,
2590                        le32_to_cpu(es->s_first_error_line));
2591                 if (es->s_first_error_ino)
2592                         printk(": inode %u",
2593                                le32_to_cpu(es->s_first_error_ino));
2594                 if (es->s_first_error_block)
2595                         printk(": block %llu", (unsigned long long)
2596                                le64_to_cpu(es->s_first_error_block));
2597                 printk("\n");
2598         }
2599         if (es->s_last_error_time) {
2600                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2601                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2602                        (int) sizeof(es->s_last_error_func),
2603                        es->s_last_error_func,
2604                        le32_to_cpu(es->s_last_error_line));
2605                 if (es->s_last_error_ino)
2606                         printk(": inode %u",
2607                                le32_to_cpu(es->s_last_error_ino));
2608                 if (es->s_last_error_block)
2609                         printk(": block %llu", (unsigned long long)
2610                                le64_to_cpu(es->s_last_error_block));
2611                 printk("\n");
2612         }
2613         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2614 }
2615
2616 static void ext4_lazyinode_timeout(unsigned long data)
2617 {
2618         struct task_struct *p = (struct task_struct *)data;
2619         wake_up_process(p);
2620 }
2621
2622 /* Find next suitable group and run ext4_init_inode_table */
2623 static int ext4_run_li_request(struct ext4_li_request *elr)
2624 {
2625         struct ext4_group_desc *gdp = NULL;
2626         ext4_group_t group, ngroups;
2627         struct super_block *sb;
2628         unsigned long timeout = 0;
2629         int ret = 0;
2630
2631         sb = elr->lr_super;
2632         ngroups = EXT4_SB(sb)->s_groups_count;
2633
2634         for (group = elr->lr_next_group; group < ngroups; group++) {
2635                 gdp = ext4_get_group_desc(sb, group, NULL);
2636                 if (!gdp) {
2637                         ret = 1;
2638                         break;
2639                 }
2640
2641                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2642                         break;
2643         }
2644
2645         if (group == ngroups)
2646                 ret = 1;
2647
2648         if (!ret) {
2649                 timeout = jiffies;
2650                 ret = ext4_init_inode_table(sb, group,
2651                                             elr->lr_timeout ? 0 : 1);
2652                 if (elr->lr_timeout == 0) {
2653                         timeout = jiffies - timeout;
2654                         if (elr->lr_sbi->s_li_wait_mult)
2655                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2656                         else
2657                                 timeout *= 20;
2658                         elr->lr_timeout = timeout;
2659                 }
2660                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2661                 elr->lr_next_group = group + 1;
2662         }
2663
2664         return ret;
2665 }
2666
2667 /*
2668  * Remove lr_request from the list_request and free the
2669  * request tructure. Should be called with li_list_mtx held
2670  */
2671 static void ext4_remove_li_request(struct ext4_li_request *elr)
2672 {
2673         struct ext4_sb_info *sbi;
2674
2675         if (!elr)
2676                 return;
2677
2678         sbi = elr->lr_sbi;
2679
2680         list_del(&elr->lr_request);
2681         sbi->s_li_request = NULL;
2682         kfree(elr);
2683 }
2684
2685 static void ext4_unregister_li_request(struct super_block *sb)
2686 {
2687         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2688
2689         if (!ext4_li_info)
2690                 return;
2691
2692         mutex_lock(&ext4_li_info->li_list_mtx);
2693         ext4_remove_li_request(elr);
2694         mutex_unlock(&ext4_li_info->li_list_mtx);
2695 }
2696
2697 /*
2698  * This is the function where ext4lazyinit thread lives. It walks
2699  * through the request list searching for next scheduled filesystem.
2700  * When such a fs is found, run the lazy initialization request
2701  * (ext4_rn_li_request) and keep track of the time spend in this
2702  * function. Based on that time we compute next schedule time of
2703  * the request. When walking through the list is complete, compute
2704  * next waking time and put itself into sleep.
2705  */
2706 static int ext4_lazyinit_thread(void *arg)
2707 {
2708         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2709         struct list_head *pos, *n;
2710         struct ext4_li_request *elr;
2711         unsigned long next_wakeup;
2712         DEFINE_WAIT(wait);
2713
2714         BUG_ON(NULL == eli);
2715
2716         eli->li_timer.data = (unsigned long)current;
2717         eli->li_timer.function = ext4_lazyinode_timeout;
2718
2719         eli->li_task = current;
2720         wake_up(&eli->li_wait_task);
2721
2722 cont_thread:
2723         while (true) {
2724                 next_wakeup = MAX_JIFFY_OFFSET;
2725
2726                 mutex_lock(&eli->li_list_mtx);
2727                 if (list_empty(&eli->li_request_list)) {
2728                         mutex_unlock(&eli->li_list_mtx);
2729                         goto exit_thread;
2730                 }
2731
2732                 list_for_each_safe(pos, n, &eli->li_request_list) {
2733                         elr = list_entry(pos, struct ext4_li_request,
2734                                          lr_request);
2735
2736                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2737                                 if (ext4_run_li_request(elr) != 0) {
2738                                         /* error, remove the lazy_init job */
2739                                         ext4_remove_li_request(elr);
2740                                         continue;
2741                                 }
2742                         }
2743
2744                         if (time_before(elr->lr_next_sched, next_wakeup))
2745                                 next_wakeup = elr->lr_next_sched;
2746                 }
2747                 mutex_unlock(&eli->li_list_mtx);
2748
2749                 if (freezing(current))
2750                         refrigerator();
2751
2752                 if ((time_after_eq(jiffies, next_wakeup)) ||
2753                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2754                         cond_resched();
2755                         continue;
2756                 }
2757
2758                 eli->li_timer.expires = next_wakeup;
2759                 add_timer(&eli->li_timer);
2760                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2761                                 TASK_INTERRUPTIBLE);
2762                 if (time_before(jiffies, next_wakeup))
2763                         schedule();
2764                 finish_wait(&eli->li_wait_daemon, &wait);
2765         }
2766
2767 exit_thread:
2768         /*
2769          * It looks like the request list is empty, but we need
2770          * to check it under the li_list_mtx lock, to prevent any
2771          * additions into it, and of course we should lock ext4_li_mtx
2772          * to atomically free the list and ext4_li_info, because at
2773          * this point another ext4 filesystem could be registering
2774          * new one.
2775          */
2776         mutex_lock(&ext4_li_mtx);
2777         mutex_lock(&eli->li_list_mtx);
2778         if (!list_empty(&eli->li_request_list)) {
2779                 mutex_unlock(&eli->li_list_mtx);
2780                 mutex_unlock(&ext4_li_mtx);
2781                 goto cont_thread;
2782         }
2783         mutex_unlock(&eli->li_list_mtx);
2784         del_timer_sync(&ext4_li_info->li_timer);
2785         eli->li_task = NULL;
2786         wake_up(&eli->li_wait_task);
2787
2788         kfree(ext4_li_info);
2789         ext4_li_info = NULL;
2790         mutex_unlock(&ext4_li_mtx);
2791
2792         return 0;
2793 }
2794
2795 static void ext4_clear_request_list(void)
2796 {
2797         struct list_head *pos, *n;
2798         struct ext4_li_request *elr;
2799
2800         mutex_lock(&ext4_li_info->li_list_mtx);
2801         if (list_empty(&ext4_li_info->li_request_list))
2802                 return;
2803
2804         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2805                 elr = list_entry(pos, struct ext4_li_request,
2806                                  lr_request);
2807                 ext4_remove_li_request(elr);
2808         }
2809         mutex_unlock(&ext4_li_info->li_list_mtx);
2810 }
2811
2812 static int ext4_run_lazyinit_thread(void)
2813 {
2814         struct task_struct *t;
2815
2816         t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2817         if (IS_ERR(t)) {
2818                 int err = PTR_ERR(t);
2819                 ext4_clear_request_list();
2820                 del_timer_sync(&ext4_li_info->li_timer);
2821                 kfree(ext4_li_info);
2822                 ext4_li_info = NULL;
2823                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2824                                  "initialization thread\n",
2825                                  err);
2826                 return err;
2827         }
2828         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2829
2830         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2831         return 0;
2832 }
2833
2834 /*
2835  * Check whether it make sense to run itable init. thread or not.
2836  * If there is at least one uninitialized inode table, return
2837  * corresponding group number, else the loop goes through all
2838  * groups and return total number of groups.
2839  */
2840 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2841 {
2842         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2843         struct ext4_group_desc *gdp = NULL;
2844
2845         for (group = 0; group < ngroups; group++) {
2846                 gdp = ext4_get_group_desc(sb, group, NULL);
2847                 if (!gdp)
2848                         continue;
2849
2850                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2851                         break;
2852         }
2853
2854         return group;
2855 }
2856
2857 static int ext4_li_info_new(void)
2858 {
2859         struct ext4_lazy_init *eli = NULL;
2860
2861         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2862         if (!eli)
2863                 return -ENOMEM;
2864
2865         eli->li_task = NULL;
2866         INIT_LIST_HEAD(&eli->li_request_list);
2867         mutex_init(&eli->li_list_mtx);
2868
2869         init_waitqueue_head(&eli->li_wait_daemon);
2870         init_waitqueue_head(&eli->li_wait_task);
2871         init_timer(&eli->li_timer);
2872         eli->li_state |= EXT4_LAZYINIT_QUIT;
2873
2874         ext4_li_info = eli;
2875
2876         return 0;
2877 }
2878
2879 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2880                                             ext4_group_t start)
2881 {
2882         struct ext4_sb_info *sbi = EXT4_SB(sb);
2883         struct ext4_li_request *elr;
2884         unsigned long rnd;
2885
2886         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2887         if (!elr)
2888                 return NULL;
2889
2890         elr->lr_super = sb;
2891         elr->lr_sbi = sbi;
2892         elr->lr_next_group = start;
2893
2894         /*
2895          * Randomize first schedule time of the request to
2896          * spread the inode table initialization requests
2897          * better.
2898          */
2899         get_random_bytes(&rnd, sizeof(rnd));
2900         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2901                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2902
2903         return elr;
2904 }
2905
2906 static int ext4_register_li_request(struct super_block *sb,
2907                                     ext4_group_t first_not_zeroed)
2908 {
2909         struct ext4_sb_info *sbi = EXT4_SB(sb);
2910         struct ext4_li_request *elr;
2911         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2912         int ret;
2913
2914         if (sbi->s_li_request != NULL)
2915                 return 0;
2916
2917         if (first_not_zeroed == ngroups ||
2918             (sb->s_flags & MS_RDONLY) ||
2919             !test_opt(sb, INIT_INODE_TABLE)) {
2920                 sbi->s_li_request = NULL;
2921                 return 0;
2922         }
2923
2924         if (first_not_zeroed == ngroups) {
2925                 sbi->s_li_request = NULL;
2926                 return 0;
2927         }
2928
2929         elr = ext4_li_request_new(sb, first_not_zeroed);
2930         if (!elr)
2931                 return -ENOMEM;
2932
2933         mutex_lock(&ext4_li_mtx);
2934
2935         if (NULL == ext4_li_info) {
2936                 ret = ext4_li_info_new();
2937                 if (ret)
2938                         goto out;
2939         }
2940
2941         mutex_lock(&ext4_li_info->li_list_mtx);
2942         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2943         mutex_unlock(&ext4_li_info->li_list_mtx);
2944
2945         sbi->s_li_request = elr;
2946
2947         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2948                 ret = ext4_run_lazyinit_thread();
2949                 if (ret)
2950                         goto out;
2951         }
2952 out:
2953         mutex_unlock(&ext4_li_mtx);
2954         if (ret)
2955                 kfree(elr);
2956         return ret;
2957 }
2958
2959 /*
2960  * We do not need to lock anything since this is called on
2961  * module unload.
2962  */
2963 static void ext4_destroy_lazyinit_thread(void)
2964 {
2965         /*
2966          * If thread exited earlier
2967          * there's nothing to be done.
2968          */
2969         if (!ext4_li_info)
2970                 return;
2971
2972         ext4_clear_request_list();
2973
2974         while (ext4_li_info->li_task) {
2975                 wake_up(&ext4_li_info->li_wait_daemon);
2976                 wait_event(ext4_li_info->li_wait_task,
2977                            ext4_li_info->li_task == NULL);
2978         }
2979 }
2980
2981 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2982                                 __releases(kernel_lock)
2983                                 __acquires(kernel_lock)
2984 {
2985         char *orig_data = kstrdup(data, GFP_KERNEL);
2986         struct buffer_head *bh;
2987         struct ext4_super_block *es = NULL;
2988         struct ext4_sb_info *sbi;
2989         ext4_fsblk_t block;
2990         ext4_fsblk_t sb_block = get_sb_block(&data);
2991         ext4_fsblk_t logical_sb_block;
2992         unsigned long offset = 0;
2993         unsigned long journal_devnum = 0;
2994         unsigned long def_mount_opts;
2995         struct inode *root;
2996         char *cp;
2997         const char *descr;
2998         int ret = -ENOMEM;
2999         int blocksize;
3000         unsigned int db_count;
3001         unsigned int i;
3002         int needs_recovery, has_huge_files;
3003         __u64 blocks_count;
3004         int err;
3005         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3006         ext4_group_t first_not_zeroed;
3007
3008         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3009         if (!sbi)
3010                 goto out_free_orig;
3011
3012         sbi->s_blockgroup_lock =
3013                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3014         if (!sbi->s_blockgroup_lock) {
3015                 kfree(sbi);
3016                 goto out_free_orig;
3017         }
3018         sb->s_fs_info = sbi;
3019         sbi->s_mount_opt = 0;
3020         sbi->s_resuid = EXT4_DEF_RESUID;
3021         sbi->s_resgid = EXT4_DEF_RESGID;
3022         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3023         sbi->s_sb_block = sb_block;
3024         if (sb->s_bdev->bd_part)
3025                 sbi->s_sectors_written_start =
3026                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3027
3028         /* Cleanup superblock name */
3029         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3030                 *cp = '!';
3031
3032         ret = -EINVAL;
3033         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3034         if (!blocksize) {
3035                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3036                 goto out_fail;
3037         }
3038
3039         /*
3040          * The ext4 superblock will not be buffer aligned for other than 1kB
3041          * block sizes.  We need to calculate the offset from buffer start.
3042          */
3043         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3044                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3045                 offset = do_div(logical_sb_block, blocksize);
3046         } else {
3047                 logical_sb_block = sb_block;
3048         }
3049
3050         if (!(bh = sb_bread(sb, logical_sb_block))) {
3051                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3052                 goto out_fail;
3053         }
3054         /*
3055          * Note: s_es must be initialized as soon as possible because
3056          *       some ext4 macro-instructions depend on its value
3057          */
3058         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3059         sbi->s_es = es;
3060         sb->s_magic = le16_to_cpu(es->s_magic);
3061         if (sb->s_magic != EXT4_SUPER_MAGIC)
3062                 goto cantfind_ext4;
3063         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3064
3065         /* Set defaults before we parse the mount options */
3066         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3067         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
3068         if (def_mount_opts & EXT4_DEFM_DEBUG)
3069                 set_opt(sbi->s_mount_opt, DEBUG);
3070         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3071                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3072                         "2.6.38");
3073                 set_opt(sbi->s_mount_opt, GRPID);
3074         }
3075         if (def_mount_opts & EXT4_DEFM_UID16)
3076                 set_opt(sbi->s_mount_opt, NO_UID32);
3077 #ifdef CONFIG_EXT4_FS_XATTR
3078         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3079                 set_opt(sbi->s_mount_opt, XATTR_USER);
3080 #endif
3081 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3082         if (def_mount_opts & EXT4_DEFM_ACL)
3083                 set_opt(sbi->s_mount_opt, POSIX_ACL);
3084 #endif
3085         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3086                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3087         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3088                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
3089         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3090                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3091
3092         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3093                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
3094         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3095                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
3096         else
3097                 set_opt(sbi->s_mount_opt, ERRORS_RO);
3098         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3099                 set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
3100         if (def_mount_opts & EXT4_DEFM_DISCARD)
3101                 set_opt(sbi->s_mount_opt, DISCARD);
3102
3103         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3104         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3105         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3106         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3107         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3108
3109         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3110                 set_opt(sbi->s_mount_opt, BARRIER);
3111
3112         /*
3113          * enable delayed allocation by default
3114          * Use -o nodelalloc to turn it off
3115          */
3116         if (!IS_EXT3_SB(sb) &&
3117             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3118                 set_opt(sbi->s_mount_opt, DELALLOC);
3119
3120         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3121                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3122                 ext4_msg(sb, KERN_WARNING,
3123                          "failed to parse options in superblock: %s",
3124                          sbi->s_es->s_mount_opts);
3125         }
3126         if (!parse_options((char *) data, sb, &journal_devnum,
3127                            &journal_ioprio, NULL, 0))
3128                 goto failed_mount;
3129
3130         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3131                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3132
3133         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3134             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3135              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3136              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3137                 ext4_msg(sb, KERN_WARNING,
3138                        "feature flags set on rev 0 fs, "
3139                        "running e2fsck is recommended");
3140
3141         /*
3142          * Check feature flags regardless of the revision level, since we
3143          * previously didn't change the revision level when setting the flags,
3144          * so there is a chance incompat flags are set on a rev 0 filesystem.
3145          */
3146         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3147                 goto failed_mount;
3148
3149         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3150
3151         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3152             blocksize > EXT4_MAX_BLOCK_SIZE) {
3153                 ext4_msg(sb, KERN_ERR,
3154                        "Unsupported filesystem blocksize %d", blocksize);
3155                 goto failed_mount;
3156         }
3157
3158         if (sb->s_blocksize != blocksize) {
3159                 /* Validate the filesystem blocksize */
3160                 if (!sb_set_blocksize(sb, blocksize)) {
3161                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3162                                         blocksize);
3163                         goto failed_mount;
3164                 }
3165
3166                 brelse(bh);
3167                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3168                 offset = do_div(logical_sb_block, blocksize);
3169                 bh = sb_bread(sb, logical_sb_block);
3170                 if (!bh) {
3171                         ext4_msg(sb, KERN_ERR,
3172                                "Can't read superblock on 2nd try");
3173                         goto failed_mount;
3174                 }
3175                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3176                 sbi->s_es = es;
3177                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3178                         ext4_msg(sb, KERN_ERR,
3179                                "Magic mismatch, very weird!");
3180                         goto failed_mount;
3181                 }
3182         }
3183
3184         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3185                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3186         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3187                                                       has_huge_files);
3188         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3189
3190         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3191                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3192                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3193         } else {
3194                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3195                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3196                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3197                     (!is_power_of_2(sbi->s_inode_size)) ||
3198                     (sbi->s_inode_size > blocksize)) {
3199                         ext4_msg(sb, KERN_ERR,
3200                                "unsupported inode size: %d",
3201                                sbi->s_inode_size);
3202                         goto failed_mount;
3203                 }
3204                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3205                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3206         }
3207
3208         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3209         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3210                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3211                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3212                     !is_power_of_2(sbi->s_desc_size)) {
3213                         ext4_msg(sb, KERN_ERR,
3214                                "unsupported descriptor size %lu",
3215                                sbi->s_desc_size);
3216                         goto failed_mount;
3217                 }
3218         } else
3219                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3220
3221         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3222         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3223         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3224                 goto cantfind_ext4;
3225
3226         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3227         if (sbi->s_inodes_per_block == 0)
3228                 goto cantfind_ext4;
3229         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3230                                         sbi->s_inodes_per_block;
3231         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3232         sbi->s_sbh = bh;
3233         sbi->s_mount_state = le16_to_cpu(es->s_state);
3234         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3235         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3236
3237         for (i = 0; i < 4; i++)
3238                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3239         sbi->s_def_hash_version = es->s_def_hash_version;
3240         i = le32_to_cpu(es->s_flags);
3241         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3242                 sbi->s_hash_unsigned = 3;
3243         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3244 #ifdef __CHAR_UNSIGNED__
3245                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3246                 sbi->s_hash_unsigned = 3;
3247 #else
3248                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3249 #endif
3250                 sb->s_dirt = 1;
3251         }
3252
3253         if (sbi->s_blocks_per_group > blocksize * 8) {
3254                 ext4_msg(sb, KERN_ERR,
3255                        "#blocks per group too big: %lu",
3256                        sbi->s_blocks_per_group);
3257                 goto failed_mount;
3258         }
3259         if (sbi->s_inodes_per_group > blocksize * 8) {
3260                 ext4_msg(sb, KERN_ERR,
3261                        "#inodes per group too big: %lu",
3262                        sbi->s_inodes_per_group);
3263                 goto failed_mount;
3264         }
3265
3266         /*
3267          * Test whether we have more sectors than will fit in sector_t,
3268          * and whether the max offset is addressable by the page cache.
3269          */
3270         ret = generic_check_addressable(sb->s_blocksize_bits,
3271                                         ext4_blocks_count(es));
3272         if (ret) {
3273                 ext4_msg(sb, KERN_ERR, "filesystem"
3274                          " too large to mount safely on this system");
3275                 if (sizeof(sector_t) < 8)
3276                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3277                 goto failed_mount;
3278         }
3279
3280         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3281                 goto cantfind_ext4;
3282
3283         /* check blocks count against device size */
3284         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3285         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3286                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3287                        "exceeds size of device (%llu blocks)",
3288                        ext4_blocks_count(es), blocks_count);
3289                 goto failed_mount;
3290         }
3291
3292         /*
3293          * It makes no sense for the first data block to be beyond the end
3294          * of the filesystem.
3295          */
3296         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3297                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3298                          "block %u is beyond end of filesystem (%llu)",
3299                          le32_to_cpu(es->s_first_data_block),
3300                          ext4_blocks_count(es));
3301                 goto failed_mount;
3302         }
3303         blocks_count = (ext4_blocks_count(es) -
3304                         le32_to_cpu(es->s_first_data_block) +
3305                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3306         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3307         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3308                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3309                        "(block count %llu, first data block %u, "
3310                        "blocks per group %lu)", sbi->s_groups_count,
3311                        ext4_blocks_count(es),
3312                        le32_to_cpu(es->s_first_data_block),
3313                        EXT4_BLOCKS_PER_GROUP(sb));
3314                 goto failed_mount;
3315         }
3316         sbi->s_groups_count = blocks_count;
3317         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3318                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3319         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3320                    EXT4_DESC_PER_BLOCK(sb);
3321         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3322                                     GFP_KERNEL);
3323         if (sbi->s_group_desc == NULL) {
3324                 ext4_msg(sb, KERN_ERR, "not enough memory");
3325                 goto failed_mount;
3326         }
3327
3328 #ifdef CONFIG_PROC_FS
3329         if (ext4_proc_root)
3330                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3331 #endif
3332
3333         bgl_lock_init(sbi->s_blockgroup_lock);
3334
3335         for (i = 0; i < db_count; i++) {
3336                 block = descriptor_loc(sb, logical_sb_block, i);
3337                 sbi->s_group_desc[i] = sb_bread(sb, block);
3338                 if (!sbi->s_group_desc[i]) {
3339                         ext4_msg(sb, KERN_ERR,
3340                                "can't read group descriptor %d", i);
3341                         db_count = i;
3342                         goto failed_mount2;
3343                 }
3344         }
3345         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3346                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3347                 goto failed_mount2;
3348         }
3349         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3350                 if (!ext4_fill_flex_info(sb)) {
3351                         ext4_msg(sb, KERN_ERR,
3352                                "unable to initialize "
3353                                "flex_bg meta info!");
3354                         goto failed_mount2;
3355                 }
3356
3357         sbi->s_gdb_count = db_count;
3358         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3359         spin_lock_init(&sbi->s_next_gen_lock);
3360
3361         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3362                         ext4_count_free_blocks(sb));
3363         if (!err) {
3364                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3365                                 ext4_count_free_inodes(sb));
3366         }
3367         if (!err) {
3368                 err = percpu_counter_init(&sbi->s_dirs_counter,
3369                                 ext4_count_dirs(sb));
3370         }
3371         if (!err) {
3372                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3373         }
3374         if (err) {
3375                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3376                 goto failed_mount3;
3377         }
3378
3379         sbi->s_stripe = ext4_get_stripe_size(sbi);
3380         sbi->s_max_writeback_mb_bump = 128;
3381
3382         /*
3383          * set up enough so that it can read an inode
3384          */
3385         if (!test_opt(sb, NOLOAD) &&
3386             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3387                 sb->s_op = &ext4_sops;
3388         else
3389                 sb->s_op = &ext4_nojournal_sops;
3390         sb->s_export_op = &ext4_export_ops;
3391         sb->s_xattr = ext4_xattr_handlers;
3392 #ifdef CONFIG_QUOTA
3393         sb->s_qcop = &ext4_qctl_operations;
3394         sb->dq_op = &ext4_quota_operations;
3395 #endif
3396         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3397         mutex_init(&sbi->s_orphan_lock);
3398         mutex_init(&sbi->s_resize_lock);
3399
3400         sb->s_root = NULL;
3401
3402         needs_recovery = (es->s_last_orphan != 0 ||
3403                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3404                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3405
3406         /*
3407          * The first inode we look at is the journal inode.  Don't try
3408          * root first: it may be modified in the journal!
3409          */
3410         if (!test_opt(sb, NOLOAD) &&
3411             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3412                 if (ext4_load_journal(sb, es, journal_devnum))
3413                         goto failed_mount3;
3414         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3415               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3416                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3417                        "suppressed and not mounted read-only");
3418                 goto failed_mount_wq;
3419         } else {
3420                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
3421                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3422                 sbi->s_journal = NULL;
3423                 needs_recovery = 0;
3424                 goto no_journal;
3425         }
3426
3427         if (ext4_blocks_count(es) > 0xffffffffULL &&
3428             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3429                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3430                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3431                 goto failed_mount_wq;
3432         }
3433
3434         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3435                 jbd2_journal_set_features(sbi->s_journal,
3436                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3437                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3438         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3439                 jbd2_journal_set_features(sbi->s_journal,
3440                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3441                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3442                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3443         } else {
3444                 jbd2_journal_clear_features(sbi->s_journal,
3445                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3446                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3447         }
3448
3449         /* We have now updated the journal if required, so we can
3450          * validate the data journaling mode. */
3451         switch (test_opt(sb, DATA_FLAGS)) {
3452         case 0:
3453                 /* No mode set, assume a default based on the journal
3454                  * capabilities: ORDERED_DATA if the journal can
3455                  * cope, else JOURNAL_DATA
3456                  */
3457                 if (jbd2_journal_check_available_features
3458                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3459                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
3460                 else
3461                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3462                 break;
3463
3464         case EXT4_MOUNT_ORDERED_DATA:
3465         case EXT4_MOUNT_WRITEBACK_DATA:
3466                 if (!jbd2_journal_check_available_features
3467                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3468                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3469                                "requested data journaling mode");
3470                         goto failed_mount_wq;
3471                 }
3472         default:
3473                 break;
3474         }
3475         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3476
3477         /*
3478          * The journal may have updated the bg summary counts, so we
3479          * need to update the global counters.
3480          */
3481         percpu_counter_set(&sbi->s_freeblocks_counter,
3482                            ext4_count_free_blocks(sb));
3483         percpu_counter_set(&sbi->s_freeinodes_counter,
3484                            ext4_count_free_inodes(sb));
3485         percpu_counter_set(&sbi->s_dirs_counter,
3486                            ext4_count_dirs(sb));
3487         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3488
3489 no_journal:
3490         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3491         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3492                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3493                 goto failed_mount_wq;
3494         }
3495
3496         /*
3497          * The jbd2_journal_load will have done any necessary log recovery,
3498          * so we can safely mount the rest of the filesystem now.
3499          */
3500
3501         root = ext4_iget(sb, EXT4_ROOT_INO);
3502         if (IS_ERR(root)) {
3503                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3504                 ret = PTR_ERR(root);
3505                 goto failed_mount4;
3506         }
3507         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3508                 iput(root);
3509                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3510                 goto failed_mount4;
3511         }
3512         sb->s_root = d_alloc_root(root);
3513         if (!sb->s_root) {
3514                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3515                 iput(root);
3516                 ret = -ENOMEM;
3517                 goto failed_mount4;
3518         }
3519
3520         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3521
3522         /* determine the minimum size of new large inodes, if present */
3523         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3524                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3525                                                      EXT4_GOOD_OLD_INODE_SIZE;
3526                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3527                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3528                         if (sbi->s_want_extra_isize <
3529                             le16_to_cpu(es->s_want_extra_isize))
3530                                 sbi->s_want_extra_isize =
3531                                         le16_to_cpu(es->s_want_extra_isize);
3532                         if (sbi->s_want_extra_isize <
3533                             le16_to_cpu(es->s_min_extra_isize))
3534                                 sbi->s_want_extra_isize =
3535                                         le16_to_cpu(es->s_min_extra_isize);
3536                 }
3537         }
3538         /* Check if enough inode space is available */
3539         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3540                                                         sbi->s_inode_size) {
3541                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3542                                                        EXT4_GOOD_OLD_INODE_SIZE;
3543                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3544                          "available");
3545         }
3546
3547         if (test_opt(sb, DELALLOC) &&
3548             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3549                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3550                          "requested data journaling mode");
3551                 clear_opt(sbi->s_mount_opt, DELALLOC);
3552         }
3553         if (test_opt(sb, DIOREAD_NOLOCK)) {
3554                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3555                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3556                                 "option - requested data journaling mode");
3557                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3558                 }
3559                 if (sb->s_blocksize < PAGE_SIZE) {
3560                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3561                                 "option - block size is too small");
3562                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3563                 }
3564         }
3565
3566         err = ext4_setup_system_zone(sb);
3567         if (err) {
3568                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3569                          "zone (%d)", err);
3570                 goto failed_mount4;
3571         }
3572
3573         ext4_ext_init(sb);
3574         err = ext4_mb_init(sb, needs_recovery);
3575         if (err) {
3576                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3577                          err);
3578                 goto failed_mount4;
3579         }
3580
3581         err = ext4_register_li_request(sb, first_not_zeroed);
3582         if (err)
3583                 goto failed_mount4;
3584
3585         sbi->s_kobj.kset = ext4_kset;
3586         init_completion(&sbi->s_kobj_unregister);
3587         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3588                                    "%s", sb->s_id);
3589         if (err) {
3590                 ext4_mb_release(sb);
3591                 ext4_ext_release(sb);
3592                 goto failed_mount4;
3593         };
3594
3595         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3596         ext4_orphan_cleanup(sb, es);
3597         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3598         if (needs_recovery) {
3599                 ext4_msg(sb, KERN_INFO, "recovery complete");
3600                 ext4_mark_recovery_complete(sb, es);
3601         }
3602         if (EXT4_SB(sb)->s_journal) {
3603                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3604                         descr = " journalled data mode";
3605                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3606                         descr = " ordered data mode";
3607                 else
3608                         descr = " writeback data mode";
3609         } else
3610                 descr = "out journal";
3611
3612         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3613                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3614                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3615
3616         init_timer(&sbi->s_err_report);
3617         sbi->s_err_report.function = print_daily_error_info;
3618         sbi->s_err_report.data = (unsigned long) sb;
3619         if (es->s_error_count)
3620                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3621
3622         kfree(orig_data);
3623         return 0;
3624
3625 cantfind_ext4:
3626         if (!silent)
3627                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3628         goto failed_mount;
3629
3630 failed_mount4:
3631         ext4_msg(sb, KERN_ERR, "mount failed");
3632         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3633 failed_mount_wq:
3634         ext4_release_system_zone(sb);
3635         if (sbi->s_journal) {
3636                 jbd2_journal_destroy(sbi->s_journal);
3637                 sbi->s_journal = NULL;
3638         }
3639 failed_mount3:
3640         if (sbi->s_flex_groups) {
3641                 if (is_vmalloc_addr(sbi->s_flex_groups))
3642                         vfree(sbi->s_flex_groups);
3643                 else
3644                         kfree(sbi->s_flex_groups);
3645         }
3646         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3647         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3648         percpu_counter_destroy(&sbi->s_dirs_counter);
3649         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3650 failed_mount2:
3651         for (i = 0; i < db_count; i++)
3652                 brelse(sbi->s_group_desc[i]);
3653         kfree(sbi->s_group_desc);
3654 failed_mount:
3655         if (sbi->s_proc) {
3656                 remove_proc_entry(sb->s_id, ext4_proc_root);
3657         }
3658 #ifdef CONFIG_QUOTA
3659         for (i = 0; i < MAXQUOTAS; i++)
3660                 kfree(sbi->s_qf_names[i]);
3661 #endif
3662         ext4_blkdev_remove(sbi);
3663         brelse(bh);
3664 out_fail:
3665         sb->s_fs_info = NULL;
3666         kfree(sbi->s_blockgroup_lock);
3667         kfree(sbi);
3668 out_free_orig:
3669         kfree(orig_data);
3670         return ret;
3671 }
3672
3673 /*
3674  * Setup any per-fs journal parameters now.  We'll do this both on
3675  * initial mount, once the journal has been initialised but before we've
3676  * done any recovery; and again on any subsequent remount.
3677  */
3678 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3679 {
3680         struct ext4_sb_info *sbi = EXT4_SB(sb);
3681
3682         journal->j_commit_interval = sbi->s_commit_interval;
3683         journal->j_min_batch_time = sbi->s_min_batch_time;
3684         journal->j_max_batch_time = sbi->s_max_batch_time;
3685
3686         write_lock(&journal->j_state_lock);
3687         if (test_opt(sb, BARRIER))
3688                 journal->j_flags |= JBD2_BARRIER;
3689         else
3690                 journal->j_flags &= ~JBD2_BARRIER;
3691         if (test_opt(sb, DATA_ERR_ABORT))
3692                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3693         else
3694                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3695         write_unlock(&journal->j_state_lock);
3696 }
3697
3698 static journal_t *ext4_get_journal(struct super_block *sb,
3699                                    unsigned int journal_inum)
3700 {
3701         struct inode *journal_inode;
3702         journal_t *journal;
3703
3704         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3705
3706         /* First, test for the existence of a valid inode on disk.  Bad
3707          * things happen if we iget() an unused inode, as the subsequent
3708          * iput() will try to delete it. */
3709
3710         journal_inode = ext4_iget(sb, journal_inum);
3711         if (IS_ERR(journal_inode)) {
3712                 ext4_msg(sb, KERN_ERR, "no journal found");
3713                 return NULL;
3714         }
3715         if (!journal_inode->i_nlink) {
3716                 make_bad_inode(journal_inode);
3717                 iput(journal_inode);
3718                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3719                 return NULL;
3720         }
3721
3722         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3723                   journal_inode, journal_inode->i_size);
3724         if (!S_ISREG(journal_inode->i_mode)) {
3725                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3726                 iput(journal_inode);
3727                 return NULL;
3728         }
3729
3730         journal = jbd2_journal_init_inode(journal_inode);
3731         if (!journal) {
3732                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3733                 iput(journal_inode);
3734                 return NULL;
3735         }
3736         journal->j_private = sb;
3737         ext4_init_journal_params(sb, journal);
3738         return journal;
3739 }
3740
3741 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3742                                        dev_t j_dev)
3743 {
3744         struct buffer_head *bh;
3745         journal_t *journal;
3746         ext4_fsblk_t start;
3747         ext4_fsblk_t len;
3748         int hblock, blocksize;
3749         ext4_fsblk_t sb_block;
3750         unsigned long offset;
3751         struct ext4_super_block *es;
3752         struct block_device *bdev;
3753
3754         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3755
3756         bdev = ext4_blkdev_get(j_dev, sb);
3757         if (bdev == NULL)
3758                 return NULL;
3759
3760         blocksize = sb->s_blocksize;
3761         hblock = bdev_logical_block_size(bdev);
3762         if (blocksize < hblock) {
3763                 ext4_msg(sb, KERN_ERR,
3764                         "blocksize too small for journal device");
3765                 goto out_bdev;
3766         }
3767
3768         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3769         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3770         set_blocksize(bdev, blocksize);
3771         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3772                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3773                        "external journal");
3774                 goto out_bdev;
3775         }
3776
3777         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3778         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3779             !(le32_to_cpu(es->s_feature_incompat) &
3780               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3781                 ext4_msg(sb, KERN_ERR, "external journal has "
3782                                         "bad superblock");
3783                 brelse(bh);
3784                 goto out_bdev;
3785         }
3786
3787         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3788                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3789                 brelse(bh);
3790                 goto out_bdev;
3791         }
3792
3793         len = ext4_blocks_count(es);
3794         start = sb_block + 1;
3795         brelse(bh);     /* we're done with the superblock */
3796
3797         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3798                                         start, len, blocksize);
3799         if (!journal) {
3800                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3801                 goto out_bdev;
3802         }
3803         journal->j_private = sb;
3804         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3805         wait_on_buffer(journal->j_sb_buffer);
3806         if (!buffer_uptodate(journal->j_sb_buffer)) {
3807                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3808                 goto out_journal;
3809         }
3810         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3811                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3812                                         "user (unsupported) - %d",
3813                         be32_to_cpu(journal->j_superblock->s_nr_users));
3814                 goto out_journal;
3815         }
3816         EXT4_SB(sb)->journal_bdev = bdev;
3817         ext4_init_journal_params(sb, journal);
3818         return journal;
3819
3820 out_journal:
3821         jbd2_journal_destroy(journal);
3822 out_bdev:
3823         ext4_blkdev_put(bdev);
3824         return NULL;
3825 }
3826
3827 static int ext4_load_journal(struct super_block *sb,
3828                              struct ext4_super_block *es,
3829                              unsigned long journal_devnum)
3830 {
3831         journal_t *journal;
3832         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3833         dev_t journal_dev;
3834         int err = 0;
3835         int really_read_only;
3836
3837         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3838
3839         if (journal_devnum &&
3840             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3841                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3842                         "numbers have changed");
3843                 journal_dev = new_decode_dev(journal_devnum);
3844         } else
3845                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3846
3847         really_read_only = bdev_read_only(sb->s_bdev);
3848
3849         /*
3850          * Are we loading a blank journal or performing recovery after a
3851          * crash?  For recovery, we need to check in advance whether we
3852          * can get read-write access to the device.
3853          */
3854         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3855                 if (sb->s_flags & MS_RDONLY) {
3856                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3857                                         "required on readonly filesystem");
3858                         if (really_read_only) {
3859                                 ext4_msg(sb, KERN_ERR, "write access "
3860                                         "unavailable, cannot proceed");
3861                                 return -EROFS;
3862                         }
3863                         ext4_msg(sb, KERN_INFO, "write access will "
3864                                "be enabled during recovery");
3865                 }
3866         }
3867
3868         if (journal_inum && journal_dev) {
3869                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3870                        "and inode journals!");
3871                 return -EINVAL;
3872         }
3873
3874         if (journal_inum) {
3875                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3876                         return -EINVAL;
3877         } else {
3878                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3879                         return -EINVAL;
3880         }
3881
3882         if (!(journal->j_flags & JBD2_BARRIER))
3883                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3884
3885         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3886                 err = jbd2_journal_update_format(journal);
3887                 if (err)  {
3888                         ext4_msg(sb, KERN_ERR, "error updating journal");
3889                         jbd2_journal_destroy(journal);
3890                         return err;
3891                 }
3892         }
3893
3894         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3895                 err = jbd2_journal_wipe(journal, !really_read_only);
3896         if (!err) {
3897                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3898                 if (save)
3899                         memcpy(save, ((char *) es) +
3900                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3901                 err = jbd2_journal_load(journal);
3902                 if (save)
3903                         memcpy(((char *) es) + EXT4_S_ERR_START,
3904                                save, EXT4_S_ERR_LEN);
3905                 kfree(save);
3906         }
3907
3908         if (err) {
3909                 ext4_msg(sb, KERN_ERR, "error loading journal");
3910                 jbd2_journal_destroy(journal);
3911                 return err;
3912         }
3913
3914         EXT4_SB(sb)->s_journal = journal;
3915         ext4_clear_journal_err(sb, es);
3916
3917         if (!really_read_only && journal_devnum &&
3918             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3919                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3920
3921                 /* Make sure we flush the recovery flag to disk. */
3922                 ext4_commit_super(sb, 1);
3923         }
3924
3925         return 0;
3926 }
3927
3928 static int ext4_commit_super(struct super_block *sb, int sync)
3929 {
3930         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3931         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3932         int error = 0;
3933
3934         if (!sbh)
3935                 return error;
3936         if (buffer_write_io_error(sbh)) {
3937                 /*
3938                  * Oh, dear.  A previous attempt to write the
3939                  * superblock failed.  This could happen because the
3940                  * USB device was yanked out.  Or it could happen to
3941                  * be a transient write error and maybe the block will
3942                  * be remapped.  Nothing we can do but to retry the
3943                  * write and hope for the best.
3944                  */
3945                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3946                        "superblock detected");
3947                 clear_buffer_write_io_error(sbh);
3948                 set_buffer_uptodate(sbh);
3949         }
3950         /*
3951          * If the file system is mounted read-only, don't update the
3952          * superblock write time.  This avoids updating the superblock
3953          * write time when we are mounting the root file system
3954          * read/only but we need to replay the journal; at that point,
3955          * for people who are east of GMT and who make their clock
3956          * tick in localtime for Windows bug-for-bug compatibility,
3957          * the clock is set in the future, and this will cause e2fsck
3958          * to complain and force a full file system check.
3959          */
3960         if (!(sb->s_flags & MS_RDONLY))
3961                 es->s_wtime = cpu_to_le32(get_seconds());
3962         if (sb->s_bdev->bd_part)
3963                 es->s_kbytes_written =
3964                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3965                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3966                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3967         else
3968                 es->s_kbytes_written =
3969                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3970         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3971                                            &EXT4_SB(sb)->s_freeblocks_counter));
3972         es->s_free_inodes_count =
3973                 cpu_to_le32(percpu_counter_sum_positive(
3974                                 &EXT4_SB(sb)->s_freeinodes_counter));
3975         sb->s_dirt = 0;
3976         BUFFER_TRACE(sbh, "marking dirty");
3977         mark_buffer_dirty(sbh);
3978         if (sync) {
3979                 error = sync_dirty_buffer(sbh);
3980                 if (error)
3981                         return error;
3982
3983                 error = buffer_write_io_error(sbh);
3984                 if (error) {
3985                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3986                                "superblock");
3987                         clear_buffer_write_io_error(sbh);
3988                         set_buffer_uptodate(sbh);
3989                 }
3990         }
3991         return error;
3992 }
3993
3994 /*
3995  * Have we just finished recovery?  If so, and if we are mounting (or
3996  * remounting) the filesystem readonly, then we will end up with a
3997  * consistent fs on disk.  Record that fact.
3998  */
3999 static void ext4_mark_recovery_complete(struct super_block *sb,
4000                                         struct ext4_super_block *es)
4001 {
4002         journal_t *journal = EXT4_SB(sb)->s_journal;
4003
4004         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4005                 BUG_ON(journal != NULL);
4006                 return;
4007         }
4008         jbd2_journal_lock_updates(journal);
4009         if (jbd2_journal_flush(journal) < 0)
4010                 goto out;
4011
4012         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4013             sb->s_flags & MS_RDONLY) {
4014                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4015                 ext4_commit_super(sb, 1);
4016         }
4017
4018 out:
4019         jbd2_journal_unlock_updates(journal);
4020 }
4021
4022 /*
4023  * If we are mounting (or read-write remounting) a filesystem whose journal
4024  * has recorded an error from a previous lifetime, move that error to the
4025  * main filesystem now.
4026  */
4027 static void ext4_clear_journal_err(struct super_block *sb,
4028                                    struct ext4_super_block *es)
4029 {
4030         journal_t *journal;
4031         int j_errno;
4032         const char *errstr;
4033
4034         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4035
4036         journal = EXT4_SB(sb)->s_journal;
4037
4038         /*
4039          * Now check for any error status which may have been recorded in the
4040          * journal by a prior ext4_error() or ext4_abort()
4041          */
4042
4043         j_errno = jbd2_journal_errno(journal);
4044         if (j_errno) {
4045                 char nbuf[16];
4046
4047                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4048                 ext4_warning(sb, "Filesystem error recorded "
4049                              "from previous mount: %s", errstr);
4050                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4051
4052                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4053                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4054                 ext4_commit_super(sb, 1);
4055
4056                 jbd2_journal_clear_err(journal);
4057         }
4058 }
4059
4060 /*
4061  * Force the running and committing transactions to commit,
4062  * and wait on the commit.
4063  */
4064 int ext4_force_commit(struct super_block *sb)
4065 {
4066         journal_t *journal;
4067         int ret = 0;
4068
4069         if (sb->s_flags & MS_RDONLY)
4070                 return 0;
4071
4072         journal = EXT4_SB(sb)->s_journal;
4073         if (journal) {
4074                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4075                 ret = ext4_journal_force_commit(journal);
4076         }
4077
4078         return ret;
4079 }
4080
4081 static void ext4_write_super(struct super_block *sb)
4082 {
4083         lock_super(sb);
4084         ext4_commit_super(sb, 1);
4085         unlock_super(sb);
4086 }
4087
4088 static int ext4_sync_fs(struct super_block *sb, int wait)
4089 {
4090         int ret = 0;
4091         tid_t target;
4092         struct ext4_sb_info *sbi = EXT4_SB(sb);
4093
4094         trace_ext4_sync_fs(sb, wait);
4095         flush_workqueue(sbi->dio_unwritten_wq);
4096         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4097                 if (wait)
4098                         jbd2_log_wait_commit(sbi->s_journal, target);
4099         }
4100         return ret;
4101 }
4102
4103 /*
4104  * LVM calls this function before a (read-only) snapshot is created.  This
4105  * gives us a chance to flush the journal completely and mark the fs clean.
4106  */
4107 static int ext4_freeze(struct super_block *sb)
4108 {
4109         int error = 0;
4110         journal_t *journal;
4111
4112         if (sb->s_flags & MS_RDONLY)
4113                 return 0;
4114
4115         journal = EXT4_SB(sb)->s_journal;
4116
4117         /* Now we set up the journal barrier. */
4118         jbd2_journal_lock_updates(journal);
4119
4120         /*
4121          * Don't clear the needs_recovery flag if we failed to flush
4122          * the journal.
4123          */
4124         error = jbd2_journal_flush(journal);
4125         if (error < 0)
4126                 goto out;
4127
4128         /* Journal blocked and flushed, clear needs_recovery flag. */
4129         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4130         error = ext4_commit_super(sb, 1);
4131 out:
4132         /* we rely on s_frozen to stop further updates */
4133         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4134         return error;
4135 }
4136
4137 /*
4138  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4139  * flag here, even though the filesystem is not technically dirty yet.
4140  */
4141 static int ext4_unfreeze(struct super_block *sb)
4142 {
4143         if (sb->s_flags & MS_RDONLY)
4144                 return 0;
4145
4146         lock_super(sb);
4147         /* Reset the needs_recovery flag before the fs is unlocked. */
4148         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4149         ext4_commit_super(sb, 1);
4150         unlock_super(sb);
4151         return 0;
4152 }
4153
4154 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4155 {
4156         struct ext4_super_block *es;
4157         struct ext4_sb_info *sbi = EXT4_SB(sb);
4158         ext4_fsblk_t n_blocks_count = 0;
4159         unsigned long old_sb_flags;
4160         struct ext4_mount_options old_opts;
4161         int enable_quota = 0;
4162         ext4_group_t g;
4163         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4164         int err;
4165 #ifdef CONFIG_QUOTA
4166         int i;
4167 #endif
4168         char *orig_data = kstrdup(data, GFP_KERNEL);
4169
4170         /* Store the original options */
4171         lock_super(sb);
4172         old_sb_flags = sb->s_flags;
4173         old_opts.s_mount_opt = sbi->s_mount_opt;
4174         old_opts.s_resuid = sbi->s_resuid;
4175         old_opts.s_resgid = sbi->s_resgid;
4176         old_opts.s_commit_interval = sbi->s_commit_interval;
4177         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4178         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4179 #ifdef CONFIG_QUOTA
4180         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4181         for (i = 0; i < MAXQUOTAS; i++)
4182                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4183 #endif
4184         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4185                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4186
4187         /*
4188          * Allow the "check" option to be passed as a remount option.
4189          */
4190         if (!parse_options(data, sb, NULL, &journal_ioprio,
4191                            &n_blocks_count, 1)) {
4192                 err = -EINVAL;
4193                 goto restore_opts;
4194         }
4195
4196         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4197                 ext4_abort(sb, "Abort forced by user");
4198
4199         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4200                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4201
4202         es = sbi->s_es;
4203
4204         if (sbi->s_journal) {
4205                 ext4_init_journal_params(sb, sbi->s_journal);
4206                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4207         }
4208
4209         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4210                 n_blocks_count > ext4_blocks_count(es)) {
4211                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4212                         err = -EROFS;
4213                         goto restore_opts;
4214                 }
4215
4216                 if (*flags & MS_RDONLY) {
4217                         err = dquot_suspend(sb, -1);
4218                         if (err < 0)
4219                                 goto restore_opts;
4220
4221                         /*
4222                          * First of all, the unconditional stuff we have to do
4223                          * to disable replay of the journal when we next remount
4224                          */
4225                         sb->s_flags |= MS_RDONLY;
4226
4227                         /*
4228                          * OK, test if we are remounting a valid rw partition
4229                          * readonly, and if so set the rdonly flag and then
4230                          * mark the partition as valid again.
4231                          */
4232                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4233                             (sbi->s_mount_state & EXT4_VALID_FS))
4234                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4235
4236                         if (sbi->s_journal)
4237                                 ext4_mark_recovery_complete(sb, es);
4238                 } else {
4239                         /* Make sure we can mount this feature set readwrite */
4240                         if (!ext4_feature_set_ok(sb, 0)) {
4241                                 err = -EROFS;
4242                                 goto restore_opts;
4243                         }
4244                         /*
4245                          * Make sure the group descriptor checksums
4246                          * are sane.  If they aren't, refuse to remount r/w.
4247                          */
4248                         for (g = 0; g < sbi->s_groups_count; g++) {
4249                                 struct ext4_group_desc *gdp =
4250                                         ext4_get_group_desc(sb, g, NULL);
4251
4252                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4253                                         ext4_msg(sb, KERN_ERR,
4254                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4255                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4256                                                le16_to_cpu(gdp->bg_checksum));
4257                                         err = -EINVAL;
4258                                         goto restore_opts;
4259                                 }
4260                         }
4261
4262                         /*
4263                          * If we have an unprocessed orphan list hanging
4264                          * around from a previously readonly bdev mount,
4265                          * require a full umount/remount for now.
4266                          */
4267                         if (es->s_last_orphan) {
4268                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4269                                        "remount RDWR because of unprocessed "
4270                                        "orphan inode list.  Please "
4271                                        "umount/remount instead");
4272                                 err = -EINVAL;
4273                                 goto restore_opts;
4274                         }
4275
4276                         /*
4277                          * Mounting a RDONLY partition read-write, so reread
4278                          * and store the current valid flag.  (It may have
4279                          * been changed by e2fsck since we originally mounted
4280                          * the partition.)
4281                          */
4282                         if (sbi->s_journal)
4283                                 ext4_clear_journal_err(sb, es);
4284                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4285                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4286                                 goto restore_opts;
4287                         if (!ext4_setup_super(sb, es, 0))
4288                                 sb->s_flags &= ~MS_RDONLY;
4289                         enable_quota = 1;
4290                 }
4291         }
4292
4293         /*
4294          * Reinitialize lazy itable initialization thread based on
4295          * current settings
4296          */
4297         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4298                 ext4_unregister_li_request(sb);
4299         else {
4300                 ext4_group_t first_not_zeroed;
4301                 first_not_zeroed = ext4_has_uninit_itable(sb);
4302                 ext4_register_li_request(sb, first_not_zeroed);
4303         }
4304
4305         ext4_setup_system_zone(sb);
4306         if (sbi->s_journal == NULL)
4307                 ext4_commit_super(sb, 1);
4308
4309 #ifdef CONFIG_QUOTA
4310         /* Release old quota file names */
4311         for (i = 0; i < MAXQUOTAS; i++)
4312                 if (old_opts.s_qf_names[i] &&
4313                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4314                         kfree(old_opts.s_qf_names[i]);
4315 #endif
4316         unlock_super(sb);
4317         if (enable_quota)
4318                 dquot_resume(sb, -1);
4319
4320         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4321         kfree(orig_data);
4322         return 0;
4323
4324 restore_opts:
4325         sb->s_flags = old_sb_flags;
4326         sbi->s_mount_opt = old_opts.s_mount_opt;
4327         sbi->s_resuid = old_opts.s_resuid;
4328         sbi->s_resgid = old_opts.s_resgid;
4329         sbi->s_commit_interval = old_opts.s_commit_interval;
4330         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4331         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4332 #ifdef CONFIG_QUOTA
4333         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4334         for (i = 0; i < MAXQUOTAS; i++) {
4335                 if (sbi->s_qf_names[i] &&
4336                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4337                         kfree(sbi->s_qf_names[i]);
4338                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4339         }
4340 #endif
4341         unlock_super(sb);
4342         kfree(orig_data);
4343         return err;
4344 }
4345
4346 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4347 {
4348         struct super_block *sb = dentry->d_sb;
4349         struct ext4_sb_info *sbi = EXT4_SB(sb);
4350         struct ext4_super_block *es = sbi->s_es;
4351         u64 fsid;
4352
4353         if (test_opt(sb, MINIX_DF)) {
4354                 sbi->s_overhead_last = 0;
4355         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4356                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4357                 ext4_fsblk_t overhead = 0;
4358
4359                 /*
4360                  * Compute the overhead (FS structures).  This is constant
4361                  * for a given filesystem unless the number of block groups
4362                  * changes so we cache the previous value until it does.
4363                  */
4364
4365                 /*
4366                  * All of the blocks before first_data_block are
4367                  * overhead
4368                  */
4369                 overhead = le32_to_cpu(es->s_first_data_block);
4370
4371                 /*
4372                  * Add the overhead attributed to the superblock and
4373                  * block group descriptors.  If the sparse superblocks
4374                  * feature is turned on, then not all groups have this.
4375                  */
4376                 for (i = 0; i < ngroups; i++) {
4377                         overhead += ext4_bg_has_super(sb, i) +
4378                                 ext4_bg_num_gdb(sb, i);
4379                         cond_resched();
4380                 }
4381
4382                 /*
4383                  * Every block group has an inode bitmap, a block
4384                  * bitmap, and an inode table.
4385                  */
4386                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4387                 sbi->s_overhead_last = overhead;
4388                 smp_wmb();
4389                 sbi->s_blocks_last = ext4_blocks_count(es);
4390         }
4391
4392         buf->f_type = EXT4_SUPER_MAGIC;
4393         buf->f_bsize = sb->s_blocksize;
4394         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4395         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4396                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4397         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4398         if (buf->f_bfree < ext4_r_blocks_count(es))
4399                 buf->f_bavail = 0;
4400         buf->f_files = le32_to_cpu(es->s_inodes_count);
4401         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4402         buf->f_namelen = EXT4_NAME_LEN;
4403         fsid = le64_to_cpup((void *)es->s_uuid) ^
4404                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4405         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4406         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4407
4408         return 0;
4409 }
4410
4411 /* Helper function for writing quotas on sync - we need to start transaction
4412  * before quota file is locked for write. Otherwise the are possible deadlocks:
4413  * Process 1                         Process 2
4414  * ext4_create()                     quota_sync()
4415  *   jbd2_journal_start()                  write_dquot()
4416  *   dquot_initialize()                         down(dqio_mutex)
4417  *     down(dqio_mutex)                    jbd2_journal_start()
4418  *
4419  */
4420
4421 #ifdef CONFIG_QUOTA
4422
4423 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4424 {
4425         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4426 }
4427
4428 static int ext4_write_dquot(struct dquot *dquot)
4429 {
4430         int ret, err;
4431         handle_t *handle;
4432         struct inode *inode;
4433
4434         inode = dquot_to_inode(dquot);
4435         handle = ext4_journal_start(inode,
4436                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4437         if (IS_ERR(handle))
4438                 return PTR_ERR(handle);
4439         ret = dquot_commit(dquot);
4440         err = ext4_journal_stop(handle);
4441         if (!ret)
4442                 ret = err;
4443         return ret;
4444 }
4445
4446 static int ext4_acquire_dquot(struct dquot *dquot)
4447 {
4448         int ret, err;
4449         handle_t *handle;
4450
4451         handle = ext4_journal_start(dquot_to_inode(dquot),
4452                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4453         if (IS_ERR(handle))
4454                 return PTR_ERR(handle);
4455         ret = dquot_acquire(dquot);
4456         err = ext4_journal_stop(handle);
4457         if (!ret)
4458                 ret = err;
4459         return ret;
4460 }
4461
4462 static int ext4_release_dquot(struct dquot *dquot)
4463 {
4464         int ret, err;
4465         handle_t *handle;
4466
4467         handle = ext4_journal_start(dquot_to_inode(dquot),
4468                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4469         if (IS_ERR(handle)) {
4470                 /* Release dquot anyway to avoid endless cycle in dqput() */
4471                 dquot_release(dquot);
4472                 return PTR_ERR(handle);
4473         }
4474         ret = dquot_release(dquot);
4475         err = ext4_journal_stop(handle);
4476         if (!ret)
4477                 ret = err;
4478         return ret;
4479 }
4480
4481 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4482 {
4483         /* Are we journaling quotas? */
4484         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4485             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4486                 dquot_mark_dquot_dirty(dquot);
4487                 return ext4_write_dquot(dquot);
4488         } else {
4489                 return dquot_mark_dquot_dirty(dquot);
4490         }
4491 }
4492
4493 static int ext4_write_info(struct super_block *sb, int type)
4494 {
4495         int ret, err;
4496         handle_t *handle;
4497
4498         /* Data block + inode block */
4499         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4500         if (IS_ERR(handle))
4501                 return PTR_ERR(handle);
4502         ret = dquot_commit_info(sb, type);
4503         err = ext4_journal_stop(handle);
4504         if (!ret)
4505                 ret = err;
4506         return ret;
4507 }
4508
4509 /*
4510  * Turn on quotas during mount time - we need to find
4511  * the quota file and such...
4512  */
4513 static int ext4_quota_on_mount(struct super_block *sb, int type)
4514 {
4515         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4516                                         EXT4_SB(sb)->s_jquota_fmt, type);
4517 }
4518
4519 /*
4520  * Standard function to be called on quota_on
4521  */
4522 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4523                          char *name)
4524 {
4525         int err;
4526         struct path path;
4527
4528         if (!test_opt(sb, QUOTA))
4529                 return -EINVAL;
4530
4531         err = kern_path(name, LOOKUP_FOLLOW, &path);
4532         if (err)
4533                 return err;
4534
4535         /* Quotafile not on the same filesystem? */
4536         if (path.mnt->mnt_sb != sb) {
4537                 path_put(&path);
4538                 return -EXDEV;
4539         }
4540         /* Journaling quota? */
4541         if (EXT4_SB(sb)->s_qf_names[type]) {
4542                 /* Quotafile not in fs root? */
4543                 if (path.dentry->d_parent != sb->s_root)
4544                         ext4_msg(sb, KERN_WARNING,
4545                                 "Quota file not on filesystem root. "
4546                                 "Journaled quota will not work");
4547         }
4548
4549         /*
4550          * When we journal data on quota file, we have to flush journal to see
4551          * all updates to the file when we bypass pagecache...
4552          */
4553         if (EXT4_SB(sb)->s_journal &&
4554             ext4_should_journal_data(path.dentry->d_inode)) {
4555                 /*
4556                  * We don't need to lock updates but journal_flush() could
4557                  * otherwise be livelocked...
4558                  */
4559                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4560                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4561                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4562                 if (err) {
4563                         path_put(&path);
4564                         return err;
4565                 }
4566         }
4567
4568         err = dquot_quota_on_path(sb, type, format_id, &path);
4569         path_put(&path);
4570         return err;
4571 }
4572
4573 static int ext4_quota_off(struct super_block *sb, int type)
4574 {
4575         /* Force all delayed allocation blocks to be allocated.
4576          * Caller already holds s_umount sem */
4577         if (test_opt(sb, DELALLOC))
4578                 sync_filesystem(sb);
4579
4580         return dquot_quota_off(sb, type);
4581 }
4582
4583 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4584  * acquiring the locks... As quota files are never truncated and quota code
4585  * itself serializes the operations (and noone else should touch the files)
4586  * we don't have to be afraid of races */
4587 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4588                                size_t len, loff_t off)
4589 {
4590         struct inode *inode = sb_dqopt(sb)->files[type];
4591         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4592         int err = 0;
4593         int offset = off & (sb->s_blocksize - 1);
4594         int tocopy;
4595         size_t toread;
4596         struct buffer_head *bh;
4597         loff_t i_size = i_size_read(inode);
4598
4599         if (off > i_size)
4600                 return 0;
4601         if (off+len > i_size)
4602                 len = i_size-off;
4603         toread = len;
4604         while (toread > 0) {
4605                 tocopy = sb->s_blocksize - offset < toread ?
4606                                 sb->s_blocksize - offset : toread;
4607                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4608                 if (err)
4609                         return err;
4610                 if (!bh)        /* A hole? */
4611                         memset(data, 0, tocopy);
4612                 else
4613                         memcpy(data, bh->b_data+offset, tocopy);
4614                 brelse(bh);
4615                 offset = 0;
4616                 toread -= tocopy;
4617                 data += tocopy;
4618                 blk++;
4619         }
4620         return len;
4621 }
4622
4623 /* Write to quotafile (we know the transaction is already started and has
4624  * enough credits) */
4625 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4626                                 const char *data, size_t len, loff_t off)
4627 {
4628         struct inode *inode = sb_dqopt(sb)->files[type];
4629         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4630         int err = 0;
4631         int offset = off & (sb->s_blocksize - 1);
4632         struct buffer_head *bh;
4633         handle_t *handle = journal_current_handle();
4634
4635         if (EXT4_SB(sb)->s_journal && !handle) {
4636                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4637                         " cancelled because transaction is not started",
4638                         (unsigned long long)off, (unsigned long long)len);
4639                 return -EIO;
4640         }
4641         /*
4642          * Since we account only one data block in transaction credits,
4643          * then it is impossible to cross a block boundary.
4644          */
4645         if (sb->s_blocksize - offset < len) {
4646                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4647                         " cancelled because not block aligned",
4648                         (unsigned long long)off, (unsigned long long)len);
4649                 return -EIO;
4650         }
4651
4652         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4653         bh = ext4_bread(handle, inode, blk, 1, &err);
4654         if (!bh)
4655                 goto out;
4656         err = ext4_journal_get_write_access(handle, bh);
4657         if (err) {
4658                 brelse(bh);
4659                 goto out;
4660         }
4661         lock_buffer(bh);
4662         memcpy(bh->b_data+offset, data, len);
4663         flush_dcache_page(bh->b_page);
4664         unlock_buffer(bh);
4665         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4666         brelse(bh);
4667 out:
4668         if (err) {
4669                 mutex_unlock(&inode->i_mutex);
4670                 return err;
4671         }
4672         if (inode->i_size < off + len) {
4673                 i_size_write(inode, off + len);
4674                 EXT4_I(inode)->i_disksize = inode->i_size;
4675         }
4676         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4677         ext4_mark_inode_dirty(handle, inode);
4678         mutex_unlock(&inode->i_mutex);
4679         return len;
4680 }
4681
4682 #endif
4683
4684 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4685                        const char *dev_name, void *data)
4686 {
4687         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4688 }
4689
4690 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4691 static struct file_system_type ext2_fs_type = {
4692         .owner          = THIS_MODULE,
4693         .name           = "ext2",
4694         .mount          = ext4_mount,
4695         .kill_sb        = kill_block_super,
4696         .fs_flags       = FS_REQUIRES_DEV,
4697 };
4698
4699 static inline void register_as_ext2(void)
4700 {
4701         int err = register_filesystem(&ext2_fs_type);
4702         if (err)
4703                 printk(KERN_WARNING
4704                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4705 }
4706
4707 static inline void unregister_as_ext2(void)
4708 {
4709         unregister_filesystem(&ext2_fs_type);
4710 }
4711 MODULE_ALIAS("ext2");
4712 #else
4713 static inline void register_as_ext2(void) { }
4714 static inline void unregister_as_ext2(void) { }
4715 #endif
4716
4717 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4718 static inline void register_as_ext3(void)
4719 {
4720         int err = register_filesystem(&ext3_fs_type);
4721         if (err)
4722                 printk(KERN_WARNING
4723                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4724 }
4725
4726 static inline void unregister_as_ext3(void)
4727 {
4728         unregister_filesystem(&ext3_fs_type);
4729 }
4730 MODULE_ALIAS("ext3");
4731 #else
4732 static inline void register_as_ext3(void) { }
4733 static inline void unregister_as_ext3(void) { }
4734 #endif
4735
4736 static struct file_system_type ext4_fs_type = {
4737         .owner          = THIS_MODULE,
4738         .name           = "ext4",
4739         .mount          = ext4_mount,
4740         .kill_sb        = kill_block_super,
4741         .fs_flags       = FS_REQUIRES_DEV,
4742 };
4743
4744 int __init ext4_init_feat_adverts(void)
4745 {
4746         struct ext4_features *ef;
4747         int ret = -ENOMEM;
4748
4749         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4750         if (!ef)
4751                 goto out;
4752
4753         ef->f_kobj.kset = ext4_kset;
4754         init_completion(&ef->f_kobj_unregister);
4755         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4756                                    "features");
4757         if (ret) {
4758                 kfree(ef);
4759                 goto out;
4760         }
4761
4762         ext4_feat = ef;
4763         ret = 0;
4764 out:
4765         return ret;
4766 }
4767
4768 static int __init ext4_init_fs(void)
4769 {
4770         int err;
4771
4772         ext4_check_flag_values();
4773         err = ext4_init_pageio();
4774         if (err)
4775                 return err;
4776         err = ext4_init_system_zone();
4777         if (err)
4778                 goto out5;
4779         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4780         if (!ext4_kset)
4781                 goto out4;
4782         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4783
4784         err = ext4_init_feat_adverts();
4785
4786         err = ext4_init_mballoc();
4787         if (err)
4788                 goto out3;
4789
4790         err = ext4_init_xattr();
4791         if (err)
4792                 goto out2;
4793         err = init_inodecache();
4794         if (err)
4795                 goto out1;
4796         register_as_ext2();
4797         register_as_ext3();
4798         err = register_filesystem(&ext4_fs_type);
4799         if (err)
4800                 goto out;
4801
4802         ext4_li_info = NULL;
4803         mutex_init(&ext4_li_mtx);
4804         return 0;
4805 out:
4806         unregister_as_ext2();
4807         unregister_as_ext3();
4808         destroy_inodecache();
4809 out1:
4810         ext4_exit_xattr();
4811 out2:
4812         ext4_exit_mballoc();
4813 out3:
4814         kfree(ext4_feat);
4815         remove_proc_entry("fs/ext4", NULL);
4816         kset_unregister(ext4_kset);
4817 out4:
4818         ext4_exit_system_zone();
4819 out5:
4820         ext4_exit_pageio();
4821         return err;
4822 }
4823
4824 static void __exit ext4_exit_fs(void)
4825 {
4826         ext4_destroy_lazyinit_thread();
4827         unregister_as_ext2();
4828         unregister_as_ext3();
4829         unregister_filesystem(&ext4_fs_type);
4830         destroy_inodecache();
4831         ext4_exit_xattr();
4832         ext4_exit_mballoc();
4833         remove_proc_entry("fs/ext4", NULL);
4834         kset_unregister(ext4_kset);
4835         ext4_exit_system_zone();
4836         ext4_exit_pageio();
4837 }
4838
4839 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4840 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4841 MODULE_LICENSE("GPL");
4842 module_init(ext4_init_fs)
4843 module_exit(ext4_exit_fs)