fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/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);
 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         bd_release(bdev);
 667         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 668 }
 669
 670 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 671 {
 672         struct block_device *bdev;
 673         int ret = -ENODEV;
 674
 675         bdev = sbi->journal_bdev;
 676         if (bdev) {
 677                 ret = ext4_blkdev_put(bdev);
 678                 sbi->journal_bdev = NULL;
 679         }
 680         return ret;
 681 }
 682
 683 static inline struct inode *orphan_list_entry(struct list_head *l)
 684 {
 685         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 686 }
 687
 688 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 689 {
 690         struct list_head *l;
 691
 692         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 693                  le32_to_cpu(sbi->s_es->s_last_orphan));
 694
 695         printk(KERN_ERR "sb_info orphan list:\n");
 696         list_for_each(l, &sbi->s_orphan) {
 697                 struct inode *inode = orphan_list_entry(l);
 698                 printk(KERN_ERR "  "
 699                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 700                        inode->i_sb->s_id, inode->i_ino, inode,
 701                        inode->i_mode, inode->i_nlink,
 702                        NEXT_ORPHAN(inode));
 703         }
 704 }
 705
 706 static void ext4_put_super(struct super_block *sb)
 707 {
 708         struct ext4_sb_info *sbi = EXT4_SB(sb);
 709         struct ext4_super_block *es = sbi->s_es;
 710         int i, err;
 711
 712         ext4_unregister_li_request(sb);
 713         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 714
 715         flush_workqueue(sbi->dio_unwritten_wq);
 716         destroy_workqueue(sbi->dio_unwritten_wq);
 717
 718         lock_super(sb);
 719         if (sb->s_dirt)
 720                 ext4_commit_super(sb, 1);
 721
 722         if (sbi->s_journal) {
 723                 err = jbd2_journal_destroy(sbi->s_journal);
 724                 sbi->s_journal = NULL;
 725                 if (err < 0)
 726                         ext4_abort(sb, "Couldn't clean up the journal");
 727         }
 728
 729         del_timer(&sbi->s_err_report);
 730         ext4_release_system_zone(sb);
 731         ext4_mb_release(sb);
 732         ext4_ext_release(sb);
 733         ext4_xattr_put_super(sb);
 734
 735         if (!(sb->s_flags & MS_RDONLY)) {
 736                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 737                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 738                 ext4_commit_super(sb, 1);
 739         }
 740         if (sbi->s_proc) {
 741                 remove_proc_entry(sb->s_id, ext4_proc_root);
 742         }
 743         kobject_del(&sbi->s_kobj);
 744
 745         for (i = 0; i < sbi->s_gdb_count; i++)
 746                 brelse(sbi->s_group_desc[i]);
 747         kfree(sbi->s_group_desc);
 748         if (is_vmalloc_addr(sbi->s_flex_groups))
 749                 vfree(sbi->s_flex_groups);
 750         else
 751                 kfree(sbi->s_flex_groups);
 752         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 753         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 754         percpu_counter_destroy(&sbi->s_dirs_counter);
 755         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 756         brelse(sbi->s_sbh);
 757 #ifdef CONFIG_QUOTA
 758         for (i = 0; i < MAXQUOTAS; i++)
 759                 kfree(sbi->s_qf_names[i]);
 760 #endif
 761
 762         /* Debugging code just in case the in-memory inode orphan list
 763          * isn't empty.  The on-disk one can be non-empty if we've
 764          * detected an error and taken the fs readonly, but the
 765          * in-memory list had better be clean by this point. */
 766         if (!list_empty(&sbi->s_orphan))
 767                 dump_orphan_list(sb, sbi);
 768         J_ASSERT(list_empty(&sbi->s_orphan));
 769
 770         invalidate_bdev(sb->s_bdev);
 771         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 772                 /*
 773                  * Invalidate the journal device's buffers.  We don't want them
 774                  * floating about in memory - the physical journal device may
 775                  * hotswapped, and it breaks the `ro-after' testing code.
 776                  */
 777                 sync_blockdev(sbi->journal_bdev);
 778                 invalidate_bdev(sbi->journal_bdev);
 779                 ext4_blkdev_remove(sbi);
 780         }
 781         sb->s_fs_info = NULL;
 782         /*
 783          * Now that we are completely done shutting down the
 784          * superblock, we need to actually destroy the kobject.
 785          */
 786         unlock_super(sb);
 787         kobject_put(&sbi->s_kobj);
 788         wait_for_completion(&sbi->s_kobj_unregister);
 789         kfree(sbi->s_blockgroup_lock);
 790         kfree(sbi);
 791 }
 792
 793 static struct kmem_cache *ext4_inode_cachep;
 794
 795 /*
 796  * Called inside transaction, so use GFP_NOFS
 797  */
 798 static struct inode *ext4_alloc_inode(struct super_block *sb)
 799 {
 800         struct ext4_inode_info *ei;
 801
 802         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 803         if (!ei)
 804                 return NULL;
 805
 806         ei->vfs_inode.i_version = 1;
 807         ei->vfs_inode.i_data.writeback_index = 0;
 808         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 809         INIT_LIST_HEAD(&ei->i_prealloc_list);
 810         spin_lock_init(&ei->i_prealloc_lock);
 811         /*
 812          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 813          * therefore it can be null here.  Don't check it, just initialize
 814          * jinode.
 815          */
 816         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 817         ei->i_reserved_data_blocks = 0;
 818         ei->i_reserved_meta_blocks = 0;
 819         ei->i_allocated_meta_blocks = 0;
 820         ei->i_da_metadata_calc_len = 0;
 821         ei->i_delalloc_reserved_flag = 0;
 822         spin_lock_init(&(ei->i_block_reservation_lock));
 823 #ifdef CONFIG_QUOTA
 824         ei->i_reserved_quota = 0;
 825 #endif
 826         INIT_LIST_HEAD(&ei->i_completed_io_list);
 827         spin_lock_init(&ei->i_completed_io_lock);
 828         ei->cur_aio_dio = NULL;
 829         ei->i_sync_tid = 0;
 830         ei->i_datasync_tid = 0;
 831         atomic_set(&ei->i_ioend_count, 0);
 832
 833         return &ei->vfs_inode;
 834 }
 835
 836 static int ext4_drop_inode(struct inode *inode)
 837 {
 838         int drop = generic_drop_inode(inode);
 839
 840         trace_ext4_drop_inode(inode, drop);
 841         return drop;
 842 }
 843
 844 static void ext4_i_callback(struct rcu_head *head)
 845 {
 846         struct inode *inode = container_of(head, struct inode, i_rcu);
 847         INIT_LIST_HEAD(&inode->i_dentry);
 848         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 849 }
 850
 851 static void ext4_destroy_inode(struct inode *inode)
 852 {
 853         ext4_ioend_wait(inode);
 854         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 855                 ext4_msg(inode->i_sb, KERN_ERR,
 856                          "Inode %lu (%p): orphan list check failed!",
 857                          inode->i_ino, EXT4_I(inode));
 858                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 859                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 860                                 true);
 861                 dump_stack();
 862         }
 863         call_rcu(&inode->i_rcu, ext4_i_callback);
 864 }
 865
 866 static void init_once(void *foo)
 867 {
 868         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 869
 870         INIT_LIST_HEAD(&ei->i_orphan);
 871 #ifdef CONFIG_EXT4_FS_XATTR
 872         init_rwsem(&ei->xattr_sem);
 873 #endif
 874         init_rwsem(&ei->i_data_sem);
 875         inode_init_once(&ei->vfs_inode);
 876 }
 877
 878 static int init_inodecache(void)
 879 {
 880         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 881                                              sizeof(struct ext4_inode_info),
 882                                              0, (SLAB_RECLAIM_ACCOUNT|
 883                                                 SLAB_MEM_SPREAD),
 884                                              init_once);
 885         if (ext4_inode_cachep == NULL)
 886                 return -ENOMEM;
 887         return 0;
 888 }
 889
 890 static void destroy_inodecache(void)
 891 {
 892         kmem_cache_destroy(ext4_inode_cachep);
 893 }
 894
 895 void ext4_clear_inode(struct inode *inode)
 896 {
 897         invalidate_inode_buffers(inode);
 898         end_writeback(inode);
 899         dquot_drop(inode);
 900         ext4_discard_preallocations(inode);
 901         if (EXT4_JOURNAL(inode))
 902                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 903                                        &EXT4_I(inode)->jinode);
 904 }
 905
 906 static inline void ext4_show_quota_options(struct seq_file *seq,
 907                                            struct super_block *sb)
 908 {
 909 #if defined(CONFIG_QUOTA)
 910         struct ext4_sb_info *sbi = EXT4_SB(sb);
 911
 912         if (sbi->s_jquota_fmt) {
 913                 char *fmtname = "";
 914
 915                 switch (sbi->s_jquota_fmt) {
 916                 case QFMT_VFS_OLD:
 917                         fmtname = "vfsold";
 918                         break;
 919                 case QFMT_VFS_V0:
 920                         fmtname = "vfsv0";
 921                         break;
 922                 case QFMT_VFS_V1:
 923                         fmtname = "vfsv1";
 924                         break;
 925                 }
 926                 seq_printf(seq, ",jqfmt=%s", fmtname);
 927         }
 928
 929         if (sbi->s_qf_names[USRQUOTA])
 930                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 931
 932         if (sbi->s_qf_names[GRPQUOTA])
 933                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 934
 935         if (test_opt(sb, USRQUOTA))
 936                 seq_puts(seq, ",usrquota");
 937
 938         if (test_opt(sb, GRPQUOTA))
 939                 seq_puts(seq, ",grpquota");
 940 #endif
 941 }
 942
 943 /*
 944  * Show an option if
 945  *  - it's set to a non-default value OR
 946  *  - if the per-sb default is different from the global default
 947  */
 948 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 949 {
 950         int def_errors;
 951         unsigned long def_mount_opts;
 952         struct super_block *sb = vfs->mnt_sb;
 953         struct ext4_sb_info *sbi = EXT4_SB(sb);
 954         struct ext4_super_block *es = sbi->s_es;
 955
 956         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 957         def_errors     = le16_to_cpu(es->s_errors);
 958
 959         if (sbi->s_sb_block != 1)
 960                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 961         if (test_opt(sb, MINIX_DF))
 962                 seq_puts(seq, ",minixdf");
 963         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 964                 seq_puts(seq, ",grpid");
 965         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 966                 seq_puts(seq, ",nogrpid");
 967         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 968             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 969                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 970         }
 971         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 972             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 973                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 974         }
 975         if (test_opt(sb, ERRORS_RO)) {
 976                 if (def_errors == EXT4_ERRORS_PANIC ||
 977                     def_errors == EXT4_ERRORS_CONTINUE) {
 978                         seq_puts(seq, ",errors=remount-ro");
 979                 }
 980         }
 981         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 982                 seq_puts(seq, ",errors=continue");
 983         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 984                 seq_puts(seq, ",errors=panic");
 985         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 986                 seq_puts(seq, ",nouid32");
 987         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 988                 seq_puts(seq, ",debug");
 989         if (test_opt(sb, OLDALLOC))
 990                 seq_puts(seq, ",oldalloc");
 991 #ifdef CONFIG_EXT4_FS_XATTR
 992         if (test_opt(sb, XATTR_USER) &&
 993                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 994                 seq_puts(seq, ",user_xattr");
 995         if (!test_opt(sb, XATTR_USER) &&
 996             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 997                 seq_puts(seq, ",nouser_xattr");
 998         }
 999 #endif
1000 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1001         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1002                 seq_puts(seq, ",acl");
1003         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1004                 seq_puts(seq, ",noacl");
1005 #endif
1006         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1007                 seq_printf(seq, ",commit=%u",
1008                            (unsigned) (sbi->s_commit_interval / HZ));
1009         }
1010         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1011                 seq_printf(seq, ",min_batch_time=%u",
1012                            (unsigned) sbi->s_min_batch_time);
1013         }
1014         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1015                 seq_printf(seq, ",max_batch_time=%u",
1016                            (unsigned) sbi->s_min_batch_time);
1017         }
1018
1019         /*
1020          * We're changing the default of barrier mount option, so
1021          * let's always display its mount state so it's clear what its
1022          * status is.
1023          */
1024         seq_puts(seq, ",barrier=");
1025         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1026         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1027                 seq_puts(seq, ",journal_async_commit");
1028         else if (test_opt(sb, JOURNAL_CHECKSUM))
1029                 seq_puts(seq, ",journal_checksum");
1030         if (test_opt(sb, I_VERSION))
1031                 seq_puts(seq, ",i_version");
1032         if (!test_opt(sb, DELALLOC) &&
1033             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1034                 seq_puts(seq, ",nodelalloc");
1035
1036         if (test_opt(sb, MBLK_IO_SUBMIT))
1037                 seq_puts(seq, ",mblk_io_submit");
1038         if (sbi->s_stripe)
1039                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1040         /*
1041          * journal mode get enabled in different ways
1042          * So just print the value even if we didn't specify it
1043          */
1044         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1045                 seq_puts(seq, ",data=journal");
1046         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1047                 seq_puts(seq, ",data=ordered");
1048         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1049                 seq_puts(seq, ",data=writeback");
1050
1051         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1052                 seq_printf(seq, ",inode_readahead_blks=%u",
1053                            sbi->s_inode_readahead_blks);
1054
1055         if (test_opt(sb, DATA_ERR_ABORT))
1056                 seq_puts(seq, ",data_err=abort");
1057
1058         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1059                 seq_puts(seq, ",noauto_da_alloc");
1060
1061         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1062                 seq_puts(seq, ",discard");
1063
1064         if (test_opt(sb, NOLOAD))
1065                 seq_puts(seq, ",norecovery");
1066
1067         if (test_opt(sb, DIOREAD_NOLOCK))
1068                 seq_puts(seq, ",dioread_nolock");
1069
1070         if (test_opt(sb, BLOCK_VALIDITY) &&
1071             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1072                 seq_puts(seq, ",block_validity");
1073
1074         if (!test_opt(sb, INIT_INODE_TABLE))
1075                 seq_puts(seq, ",noinit_inode_table");
1076         else if (sbi->s_li_wait_mult)
1077                 seq_printf(seq, ",init_inode_table=%u",
1078                            (unsigned) sbi->s_li_wait_mult);
1079
1080         ext4_show_quota_options(seq, sb);
1081
1082         return 0;
1083 }
1084
1085 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1086                                         u64 ino, u32 generation)
1087 {
1088         struct inode *inode;
1089
1090         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1091                 return ERR_PTR(-ESTALE);
1092         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1093                 return ERR_PTR(-ESTALE);
1094
1095         /* iget isn't really right if the inode is currently unallocated!!
1096          *
1097          * ext4_read_inode will return a bad_inode if the inode had been
1098          * deleted, so we should be safe.
1099          *
1100          * Currently we don't know the generation for parent directory, so
1101          * a generation of 0 means "accept any"
1102          */
1103         inode = ext4_iget(sb, ino);
1104         if (IS_ERR(inode))
1105                 return ERR_CAST(inode);
1106         if (generation && inode->i_generation != generation) {
1107                 iput(inode);
1108                 return ERR_PTR(-ESTALE);
1109         }
1110
1111         return inode;
1112 }
1113
1114 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1115                                         int fh_len, int fh_type)
1116 {
1117         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1118                                     ext4_nfs_get_inode);
1119 }
1120
1121 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1122                                         int fh_len, int fh_type)
1123 {
1124         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1125                                     ext4_nfs_get_inode);
1126 }
1127
1128 /*
1129  * Try to release metadata pages (indirect blocks, directories) which are
1130  * mapped via the block device.  Since these pages could have journal heads
1131  * which would prevent try_to_free_buffers() from freeing them, we must use
1132  * jbd2 layer's try_to_free_buffers() function to release them.
1133  */
1134 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1135                                  gfp_t wait)
1136 {
1137         journal_t *journal = EXT4_SB(sb)->s_journal;
1138
1139         WARN_ON(PageChecked(page));
1140         if (!page_has_buffers(page))
1141                 return 0;
1142         if (journal)
1143                 return jbd2_journal_try_to_free_buffers(journal, page,
1144                                                         wait & ~__GFP_WAIT);
1145         return try_to_free_buffers(page);
1146 }
1147
1148 #ifdef CONFIG_QUOTA
1149 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1150 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1151
1152 static int ext4_write_dquot(struct dquot *dquot);
1153 static int ext4_acquire_dquot(struct dquot *dquot);
1154 static int ext4_release_dquot(struct dquot *dquot);
1155 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1156 static int ext4_write_info(struct super_block *sb, int type);
1157 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1158                                 char *path);
1159 static int ext4_quota_off(struct super_block *sb, int type);
1160 static int ext4_quota_on_mount(struct super_block *sb, int type);
1161 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1162                                size_t len, loff_t off);
1163 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1164                                 const char *data, size_t len, loff_t off);
1165
1166 static const struct dquot_operations ext4_quota_operations = {
1167 #ifdef CONFIG_QUOTA
1168         .get_reserved_space = ext4_get_reserved_space,
1169 #endif
1170         .write_dquot    = ext4_write_dquot,
1171         .acquire_dquot  = ext4_acquire_dquot,
1172         .release_dquot  = ext4_release_dquot,
1173         .mark_dirty     = ext4_mark_dquot_dirty,
1174         .write_info     = ext4_write_info,
1175         .alloc_dquot    = dquot_alloc,
1176         .destroy_dquot  = dquot_destroy,
1177 };
1178
1179 static const struct quotactl_ops ext4_qctl_operations = {
1180         .quota_on       = ext4_quota_on,
1181         .quota_off      = ext4_quota_off,
1182         .quota_sync     = dquot_quota_sync,
1183         .get_info       = dquot_get_dqinfo,
1184         .set_info       = dquot_set_dqinfo,
1185         .get_dqblk      = dquot_get_dqblk,
1186         .set_dqblk      = dquot_set_dqblk
1187 };
1188 #endif
1189
1190 static const struct super_operations ext4_sops = {
1191         .alloc_inode    = ext4_alloc_inode,
1192         .destroy_inode  = ext4_destroy_inode,
1193         .write_inode    = ext4_write_inode,
1194         .dirty_inode    = ext4_dirty_inode,
1195         .drop_inode     = ext4_drop_inode,
1196         .evict_inode    = ext4_evict_inode,
1197         .put_super      = ext4_put_super,
1198         .sync_fs        = ext4_sync_fs,
1199         .freeze_fs      = ext4_freeze,
1200         .unfreeze_fs    = ext4_unfreeze,
1201         .statfs         = ext4_statfs,
1202         .remount_fs     = ext4_remount,
1203         .show_options   = ext4_show_options,
1204 #ifdef CONFIG_QUOTA
1205         .quota_read     = ext4_quota_read,
1206         .quota_write    = ext4_quota_write,
1207 #endif
1208         .bdev_try_to_free_page = bdev_try_to_free_page,
1209 };
1210
1211 static const struct super_operations ext4_nojournal_sops = {
1212         .alloc_inode    = ext4_alloc_inode,
1213         .destroy_inode  = ext4_destroy_inode,
1214         .write_inode    = ext4_write_inode,
1215         .dirty_inode    = ext4_dirty_inode,
1216         .drop_inode     = ext4_drop_inode,
1217         .evict_inode    = ext4_evict_inode,
1218         .write_super    = ext4_write_super,
1219         .put_super      = ext4_put_super,
1220         .statfs         = ext4_statfs,
1221         .remount_fs     = ext4_remount,
1222         .show_options   = ext4_show_options,
1223 #ifdef CONFIG_QUOTA
1224         .quota_read     = ext4_quota_read,
1225         .quota_write    = ext4_quota_write,
1226 #endif
1227         .bdev_try_to_free_page = bdev_try_to_free_page,
1228 };
1229
1230 static const struct export_operations ext4_export_ops = {
1231         .fh_to_dentry = ext4_fh_to_dentry,
1232         .fh_to_parent = ext4_fh_to_parent,
1233         .get_parent = ext4_get_parent,
1234 };
1235
1236 enum {
1237         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1238         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1239         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1240         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1241         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1242         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1243         Opt_journal_update, Opt_journal_dev,
1244         Opt_journal_checksum, Opt_journal_async_commit,
1245         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1246         Opt_data_err_abort, Opt_data_err_ignore,
1247         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1248         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1249         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1250         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1251         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1252         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1253         Opt_inode_readahead_blks, Opt_journal_ioprio,
1254         Opt_dioread_nolock, Opt_dioread_lock,
1255         Opt_discard, Opt_nodiscard,
1256         Opt_init_inode_table, Opt_noinit_inode_table,
1257 };
1258
1259 static const match_table_t tokens = {
1260         {Opt_bsd_df, "bsddf"},
1261         {Opt_minix_df, "minixdf"},
1262         {Opt_grpid, "grpid"},
1263         {Opt_grpid, "bsdgroups"},
1264         {Opt_nogrpid, "nogrpid"},
1265         {Opt_nogrpid, "sysvgroups"},
1266         {Opt_resgid, "resgid=%u"},
1267         {Opt_resuid, "resuid=%u"},
1268         {Opt_sb, "sb=%u"},
1269         {Opt_err_cont, "errors=continue"},
1270         {Opt_err_panic, "errors=panic"},
1271         {Opt_err_ro, "errors=remount-ro"},
1272         {Opt_nouid32, "nouid32"},
1273         {Opt_debug, "debug"},
1274         {Opt_oldalloc, "oldalloc"},
1275         {Opt_orlov, "orlov"},
1276         {Opt_user_xattr, "user_xattr"},
1277         {Opt_nouser_xattr, "nouser_xattr"},
1278         {Opt_acl, "acl"},
1279         {Opt_noacl, "noacl"},
1280         {Opt_noload, "noload"},
1281         {Opt_noload, "norecovery"},
1282         {Opt_nobh, "nobh"},
1283         {Opt_bh, "bh"},
1284         {Opt_commit, "commit=%u"},
1285         {Opt_min_batch_time, "min_batch_time=%u"},
1286         {Opt_max_batch_time, "max_batch_time=%u"},
1287         {Opt_journal_update, "journal=update"},
1288         {Opt_journal_dev, "journal_dev=%u"},
1289         {Opt_journal_checksum, "journal_checksum"},
1290         {Opt_journal_async_commit, "journal_async_commit"},
1291         {Opt_abort, "abort"},
1292         {Opt_data_journal, "data=journal"},
1293         {Opt_data_ordered, "data=ordered"},
1294         {Opt_data_writeback, "data=writeback"},
1295         {Opt_data_err_abort, "data_err=abort"},
1296         {Opt_data_err_ignore, "data_err=ignore"},
1297         {Opt_offusrjquota, "usrjquota="},
1298         {Opt_usrjquota, "usrjquota=%s"},
1299         {Opt_offgrpjquota, "grpjquota="},
1300         {Opt_grpjquota, "grpjquota=%s"},
1301         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1302         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1303         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1304         {Opt_grpquota, "grpquota"},
1305         {Opt_noquota, "noquota"},
1306         {Opt_quota, "quota"},
1307         {Opt_usrquota, "usrquota"},
1308         {Opt_barrier, "barrier=%u"},
1309         {Opt_barrier, "barrier"},
1310         {Opt_nobarrier, "nobarrier"},
1311         {Opt_i_version, "i_version"},
1312         {Opt_stripe, "stripe=%u"},
1313         {Opt_resize, "resize"},
1314         {Opt_delalloc, "delalloc"},
1315         {Opt_nodelalloc, "nodelalloc"},
1316         {Opt_mblk_io_submit, "mblk_io_submit"},
1317         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1318         {Opt_block_validity, "block_validity"},
1319         {Opt_noblock_validity, "noblock_validity"},
1320         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1321         {Opt_journal_ioprio, "journal_ioprio=%u"},
1322         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1323         {Opt_auto_da_alloc, "auto_da_alloc"},
1324         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1325         {Opt_dioread_nolock, "dioread_nolock"},
1326         {Opt_dioread_lock, "dioread_lock"},
1327         {Opt_discard, "discard"},
1328         {Opt_nodiscard, "nodiscard"},
1329         {Opt_init_inode_table, "init_itable=%u"},
1330         {Opt_init_inode_table, "init_itable"},
1331         {Opt_noinit_inode_table, "noinit_itable"},
1332         {Opt_err, NULL},
1333 };
1334
1335 static ext4_fsblk_t get_sb_block(void **data)
1336 {
1337         ext4_fsblk_t    sb_block;
1338         char            *options = (char *) *data;
1339
1340         if (!options || strncmp(options, "sb=", 3) != 0)
1341                 return 1;       /* Default location */
1342
1343         options += 3;
1344         /* TODO: use simple_strtoll with >32bit ext4 */
1345         sb_block = simple_strtoul(options, &options, 0);
1346         if (*options && *options != ',') {
1347                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1348                        (char *) *data);
1349                 return 1;
1350         }
1351         if (*options == ',')
1352                 options++;
1353         *data = (void *) options;
1354
1355         return sb_block;
1356 }
1357
1358 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1359 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1360         "Contact [email protected] if you think we should keep it.\n";
1361
1362 #ifdef CONFIG_QUOTA
1363 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1364 {
1365         struct ext4_sb_info *sbi = EXT4_SB(sb);
1366         char *qname;
1367
1368         if (sb_any_quota_loaded(sb) &&
1369                 !sbi->s_qf_names[qtype]) {
1370                 ext4_msg(sb, KERN_ERR,
1371                         "Cannot change journaled "
1372                         "quota options when quota turned on");
1373                 return 0;
1374         }
1375         qname = match_strdup(args);
1376         if (!qname) {
1377                 ext4_msg(sb, KERN_ERR,
1378                         "Not enough memory for storing quotafile name");
1379                 return 0;
1380         }
1381         if (sbi->s_qf_names[qtype] &&
1382                 strcmp(sbi->s_qf_names[qtype], qname)) {
1383                 ext4_msg(sb, KERN_ERR,
1384                         "%s quota file already specified", QTYPE2NAME(qtype));
1385                 kfree(qname);
1386                 return 0;
1387         }
1388         sbi->s_qf_names[qtype] = qname;
1389         if (strchr(sbi->s_qf_names[qtype], '/')) {
1390                 ext4_msg(sb, KERN_ERR,
1391                         "quotafile must be on filesystem root");
1392                 kfree(sbi->s_qf_names[qtype]);
1393                 sbi->s_qf_names[qtype] = NULL;
1394                 return 0;
1395         }
1396         set_opt(sbi->s_mount_opt, QUOTA);
1397         return 1;
1398 }
1399
1400 static int clear_qf_name(struct super_block *sb, int qtype)
1401 {
1402
1403         struct ext4_sb_info *sbi = EXT4_SB(sb);
1404
1405         if (sb_any_quota_loaded(sb) &&
1406                 sbi->s_qf_names[qtype]) {
1407                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1408                         " when quota turned on");
1409                 return 0;
1410         }
1411         /*
1412          * The space will be released later when all options are confirmed
1413          * to be correct
1414          */
1415         sbi->s_qf_names[qtype] = NULL;
1416         return 1;
1417 }
1418 #endif
1419
1420 static int parse_options(char *options, struct super_block *sb,
1421                          unsigned long *journal_devnum,
1422                          unsigned int *journal_ioprio,
1423                          ext4_fsblk_t *n_blocks_count, int is_remount)
1424 {
1425         struct ext4_sb_info *sbi = EXT4_SB(sb);
1426         char *p;
1427         substring_t args[MAX_OPT_ARGS];
1428         int data_opt = 0;
1429         int option;
1430 #ifdef CONFIG_QUOTA
1431         int qfmt;
1432 #endif
1433
1434         if (!options)
1435                 return 1;
1436
1437         while ((p = strsep(&options, ",")) != NULL) {
1438                 int token;
1439                 if (!*p)
1440                         continue;
1441
1442                 /*
1443                  * Initialize args struct so we know whether arg was
1444                  * found; some options take optional arguments.
1445                  */
1446                 args[0].to = args[0].from = 0;
1447                 token = match_token(p, tokens, args);
1448                 switch (token) {
1449                 case Opt_bsd_df:
1450                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1451                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1452                         break;
1453                 case Opt_minix_df:
1454                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1455                         set_opt(sbi->s_mount_opt, MINIX_DF);
1456
1457                         break;
1458                 case Opt_grpid:
1459                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1460                         set_opt(sbi->s_mount_opt, GRPID);
1461
1462                         break;
1463                 case Opt_nogrpid:
1464                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1465                         clear_opt(sbi->s_mount_opt, GRPID);
1466
1467                         break;
1468                 case Opt_resuid:
1469                         if (match_int(&args[0], &option))
1470                                 return 0;
1471                         sbi->s_resuid = option;
1472                         break;
1473                 case Opt_resgid:
1474                         if (match_int(&args[0], &option))
1475                                 return 0;
1476                         sbi->s_resgid = option;
1477                         break;
1478                 case Opt_sb:
1479                         /* handled by get_sb_block() instead of here */
1480                         /* *sb_block = match_int(&args[0]); */
1481                         break;
1482                 case Opt_err_panic:
1483                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1484                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1485                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1486                         break;
1487                 case Opt_err_ro:
1488                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1489                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1490                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1491                         break;
1492                 case Opt_err_cont:
1493                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1494                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1495                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1496                         break;
1497                 case Opt_nouid32:
1498                         set_opt(sbi->s_mount_opt, NO_UID32);
1499                         break;
1500                 case Opt_debug:
1501                         set_opt(sbi->s_mount_opt, DEBUG);
1502                         break;
1503                 case Opt_oldalloc:
1504                         set_opt(sbi->s_mount_opt, OLDALLOC);
1505                         break;
1506                 case Opt_orlov:
1507                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1508                         break;
1509 #ifdef CONFIG_EXT4_FS_XATTR
1510                 case Opt_user_xattr:
1511                         set_opt(sbi->s_mount_opt, XATTR_USER);
1512                         break;
1513                 case Opt_nouser_xattr:
1514                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1515                         break;
1516 #else
1517                 case Opt_user_xattr:
1518                 case Opt_nouser_xattr:
1519                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1520                         break;
1521 #endif
1522 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1523                 case Opt_acl:
1524                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1525                         break;
1526                 case Opt_noacl:
1527                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1528                         break;
1529 #else
1530                 case Opt_acl:
1531                 case Opt_noacl:
1532                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1533                         break;
1534 #endif
1535                 case Opt_journal_update:
1536                         /* @@@ FIXME */
1537                         /* Eventually we will want to be able to create
1538                            a journal file here.  For now, only allow the
1539                            user to specify an existing inode to be the
1540                            journal file. */
1541                         if (is_remount) {
1542                                 ext4_msg(sb, KERN_ERR,
1543                                          "Cannot specify journal on remount");
1544                                 return 0;
1545                         }
1546                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1547                         break;
1548                 case Opt_journal_dev:
1549                         if (is_remount) {
1550                                 ext4_msg(sb, KERN_ERR,
1551                                         "Cannot specify journal on remount");
1552                                 return 0;
1553                         }
1554                         if (match_int(&args[0], &option))
1555                                 return 0;
1556                         *journal_devnum = option;
1557                         break;
1558                 case Opt_journal_checksum:
1559                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1560                         break;
1561                 case Opt_journal_async_commit:
1562                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1563                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1564                         break;
1565                 case Opt_noload:
1566                         set_opt(sbi->s_mount_opt, NOLOAD);
1567                         break;
1568                 case Opt_commit:
1569                         if (match_int(&args[0], &option))
1570                                 return 0;
1571                         if (option < 0)
1572                                 return 0;
1573                         if (option == 0)
1574                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1575                         sbi->s_commit_interval = HZ * option;
1576                         break;
1577                 case Opt_max_batch_time:
1578                         if (match_int(&args[0], &option))
1579                                 return 0;
1580                         if (option < 0)
1581                                 return 0;
1582                         if (option == 0)
1583                                 option = EXT4_DEF_MAX_BATCH_TIME;
1584                         sbi->s_max_batch_time = option;
1585                         break;
1586                 case Opt_min_batch_time:
1587                         if (match_int(&args[0], &option))
1588                                 return 0;
1589                         if (option < 0)
1590                                 return 0;
1591                         sbi->s_min_batch_time = option;
1592                         break;
1593                 case Opt_data_journal:
1594                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1595                         goto datacheck;
1596                 case Opt_data_ordered:
1597                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1598                         goto datacheck;
1599                 case Opt_data_writeback:
1600                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1601                 datacheck:
1602                         if (is_remount) {
1603                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1604                                         ext4_msg(sb, KERN_ERR,
1605                                                 "Cannot change data mode on remount");
1606                                         return 0;
1607                                 }
1608                         } else {
1609                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1610                                 sbi->s_mount_opt |= data_opt;
1611                         }
1612                         break;
1613                 case Opt_data_err_abort:
1614                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1615                         break;
1616                 case Opt_data_err_ignore:
1617                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1618                         break;
1619 #ifdef CONFIG_QUOTA
1620                 case Opt_usrjquota:
1621                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1622                                 return 0;
1623                         break;
1624                 case Opt_grpjquota:
1625                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1626                                 return 0;
1627                         break;
1628                 case Opt_offusrjquota:
1629                         if (!clear_qf_name(sb, USRQUOTA))
1630                                 return 0;
1631                         break;
1632                 case Opt_offgrpjquota:
1633                         if (!clear_qf_name(sb, GRPQUOTA))
1634                                 return 0;
1635                         break;
1636
1637                 case Opt_jqfmt_vfsold:
1638                         qfmt = QFMT_VFS_OLD;
1639                         goto set_qf_format;
1640                 case Opt_jqfmt_vfsv0:
1641                         qfmt = QFMT_VFS_V0;
1642                         goto set_qf_format;
1643                 case Opt_jqfmt_vfsv1:
1644                         qfmt = QFMT_VFS_V1;
1645 set_qf_format:
1646                         if (sb_any_quota_loaded(sb) &&
1647                             sbi->s_jquota_fmt != qfmt) {
1648                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1649                                         "journaled quota options when "
1650                                         "quota turned on");
1651                                 return 0;
1652                         }
1653                         sbi->s_jquota_fmt = qfmt;
1654                         break;
1655                 case Opt_quota:
1656                 case Opt_usrquota:
1657                         set_opt(sbi->s_mount_opt, QUOTA);
1658                         set_opt(sbi->s_mount_opt, USRQUOTA);
1659                         break;
1660                 case Opt_grpquota:
1661                         set_opt(sbi->s_mount_opt, QUOTA);
1662                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1663                         break;
1664                 case Opt_noquota:
1665                         if (sb_any_quota_loaded(sb)) {
1666                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1667                                         "options when quota turned on");
1668                                 return 0;
1669                         }
1670                         clear_opt(sbi->s_mount_opt, QUOTA);
1671                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1672                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1673                         break;
1674 #else
1675                 case Opt_quota:
1676                 case Opt_usrquota:
1677                 case Opt_grpquota:
1678                         ext4_msg(sb, KERN_ERR,
1679                                 "quota options not supported");
1680                         break;
1681                 case Opt_usrjquota:
1682                 case Opt_grpjquota:
1683                 case Opt_offusrjquota:
1684                 case Opt_offgrpjquota:
1685                 case Opt_jqfmt_vfsold:
1686                 case Opt_jqfmt_vfsv0:
1687                 case Opt_jqfmt_vfsv1:
1688                         ext4_msg(sb, KERN_ERR,
1689                                 "journaled quota options not supported");
1690                         break;
1691                 case Opt_noquota:
1692                         break;
1693 #endif
1694                 case Opt_abort:
1695                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1696                         break;
1697                 case Opt_nobarrier:
1698                         clear_opt(sbi->s_mount_opt, BARRIER);
1699                         break;
1700                 case Opt_barrier:
1701                         if (args[0].from) {
1702                                 if (match_int(&args[0], &option))
1703                                         return 0;
1704                         } else
1705                                 option = 1;     /* No argument, default to 1 */
1706                         if (option)
1707                                 set_opt(sbi->s_mount_opt, BARRIER);
1708                         else
1709                                 clear_opt(sbi->s_mount_opt, BARRIER);
1710                         break;
1711                 case Opt_ignore:
1712                         break;
1713                 case Opt_resize:
1714                         if (!is_remount) {
1715                                 ext4_msg(sb, KERN_ERR,
1716                                         "resize option only available "
1717                                         "for remount");
1718                                 return 0;
1719                         }
1720                         if (match_int(&args[0], &option) != 0)
1721                                 return 0;
1722                         *n_blocks_count = option;
1723                         break;
1724                 case Opt_nobh:
1725                         ext4_msg(sb, KERN_WARNING,
1726                                  "Ignoring deprecated nobh option");
1727                         break;
1728                 case Opt_bh:
1729                         ext4_msg(sb, KERN_WARNING,
1730                                  "Ignoring deprecated bh option");
1731                         break;
1732                 case Opt_i_version:
1733                         set_opt(sbi->s_mount_opt, I_VERSION);
1734                         sb->s_flags |= MS_I_VERSION;
1735                         break;
1736                 case Opt_nodelalloc:
1737                         clear_opt(sbi->s_mount_opt, DELALLOC);
1738                         break;
1739                 case Opt_mblk_io_submit:
1740                         set_opt(sbi->s_mount_opt, MBLK_IO_SUBMIT);
1741                         break;
1742                 case Opt_nomblk_io_submit:
1743                         clear_opt(sbi->s_mount_opt, MBLK_IO_SUBMIT);
1744                         break;
1745                 case Opt_stripe:
1746                         if (match_int(&args[0], &option))
1747                                 return 0;
1748                         if (option < 0)
1749                                 return 0;
1750                         sbi->s_stripe = option;
1751                         break;
1752                 case Opt_delalloc:
1753                         set_opt(sbi->s_mount_opt, DELALLOC);
1754                         break;
1755                 case Opt_block_validity:
1756                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1757                         break;
1758                 case Opt_noblock_validity:
1759                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1760                         break;
1761                 case Opt_inode_readahead_blks:
1762                         if (match_int(&args[0], &option))
1763                                 return 0;
1764                         if (option < 0 || option > (1 << 30))
1765                                 return 0;
1766                         if (!is_power_of_2(option)) {
1767                                 ext4_msg(sb, KERN_ERR,
1768                                          "EXT4-fs: inode_readahead_blks"
1769                                          " must be a power of 2");
1770                                 return 0;
1771                         }
1772                         sbi->s_inode_readahead_blks = option;
1773                         break;
1774                 case Opt_journal_ioprio:
1775                         if (match_int(&args[0], &option))
1776                                 return 0;
1777                         if (option < 0 || option > 7)
1778                                 break;
1779                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1780                                                             option);
1781                         break;
1782                 case Opt_noauto_da_alloc:
1783                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1784                         break;
1785                 case Opt_auto_da_alloc:
1786                         if (args[0].from) {
1787                                 if (match_int(&args[0], &option))
1788                                         return 0;
1789                         } else
1790                                 option = 1;     /* No argument, default to 1 */
1791                         if (option)
1792                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1793                         else
1794                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1795                         break;
1796                 case Opt_discard:
1797                         set_opt(sbi->s_mount_opt, DISCARD);
1798                         break;
1799                 case Opt_nodiscard:
1800                         clear_opt(sbi->s_mount_opt, DISCARD);
1801                         break;
1802                 case Opt_dioread_nolock:
1803                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1804                         break;
1805                 case Opt_dioread_lock:
1806                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1807                         break;
1808                 case Opt_init_inode_table:
1809                         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1810                         if (args[0].from) {
1811                                 if (match_int(&args[0], &option))
1812                                         return 0;
1813                         } else
1814                                 option = EXT4_DEF_LI_WAIT_MULT;
1815                         if (option < 0)
1816                                 return 0;
1817                         sbi->s_li_wait_mult = option;
1818                         break;
1819                 case Opt_noinit_inode_table:
1820                         clear_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1821                         break;
1822                 default:
1823                         ext4_msg(sb, KERN_ERR,
1824                                "Unrecognized mount option \"%s\" "
1825                                "or missing value", p);
1826                         return 0;
1827                 }
1828         }
1829 #ifdef CONFIG_QUOTA
1830         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1831                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1832                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1833
1834                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1835                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1836
1837                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1838                         ext4_msg(sb, KERN_ERR, "old and new quota "
1839                                         "format mixing");
1840                         return 0;
1841                 }
1842
1843                 if (!sbi->s_jquota_fmt) {
1844                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1845                                         "not specified");
1846                         return 0;
1847                 }
1848         } else {
1849                 if (sbi->s_jquota_fmt) {
1850                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1851                                         "specified with no journaling "
1852                                         "enabled");
1853                         return 0;
1854                 }
1855         }
1856 #endif
1857         return 1;
1858 }
1859
1860 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1861                             int read_only)
1862 {
1863         struct ext4_sb_info *sbi = EXT4_SB(sb);
1864         int res = 0;
1865
1866         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1867                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1868                          "forcing read-only mode");
1869                 res = MS_RDONLY;
1870         }
1871         if (read_only)
1872                 return res;
1873         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1874                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1875                          "running e2fsck is recommended");
1876         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1877                 ext4_msg(sb, KERN_WARNING,
1878                          "warning: mounting fs with errors, "
1879                          "running e2fsck is recommended");
1880         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1881                  le16_to_cpu(es->s_mnt_count) >=
1882                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "warning: maximal mount count reached, "
1885                          "running e2fsck is recommended");
1886         else if (le32_to_cpu(es->s_checkinterval) &&
1887                 (le32_to_cpu(es->s_lastcheck) +
1888                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "warning: checktime reached, "
1891                          "running e2fsck is recommended");
1892         if (!sbi->s_journal)
1893                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1894         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1895                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1896         le16_add_cpu(&es->s_mnt_count, 1);
1897         es->s_mtime = cpu_to_le32(get_seconds());
1898         ext4_update_dynamic_rev(sb);
1899         if (sbi->s_journal)
1900                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1901
1902         ext4_commit_super(sb, 1);
1903         if (test_opt(sb, DEBUG))
1904                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1905                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1906                         sb->s_blocksize,
1907                         sbi->s_groups_count,
1908                         EXT4_BLOCKS_PER_GROUP(sb),
1909                         EXT4_INODES_PER_GROUP(sb),
1910                         sbi->s_mount_opt);
1911
1912         return res;
1913 }
1914
1915 static int ext4_fill_flex_info(struct super_block *sb)
1916 {
1917         struct ext4_sb_info *sbi = EXT4_SB(sb);
1918         struct ext4_group_desc *gdp = NULL;
1919         ext4_group_t flex_group_count;
1920         ext4_group_t flex_group;
1921         int groups_per_flex = 0;
1922         size_t size;
1923         int i;
1924
1925         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1926         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1927
1928         if (groups_per_flex < 2) {
1929                 sbi->s_log_groups_per_flex = 0;
1930                 return 1;
1931         }
1932
1933         /* We allocate both existing and potentially added groups */
1934         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1935                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1936                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1937         size = flex_group_count * sizeof(struct flex_groups);
1938         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1939         if (sbi->s_flex_groups == NULL) {
1940                 sbi->s_flex_groups = vmalloc(size);
1941                 if (sbi->s_flex_groups)
1942                         memset(sbi->s_flex_groups, 0, size);
1943         }
1944         if (sbi->s_flex_groups == NULL) {
1945                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1946                                 "%u flex groups", flex_group_count);
1947                 goto failed;
1948         }
1949
1950         for (i = 0; i < sbi->s_groups_count; i++) {
1951                 gdp = ext4_get_group_desc(sb, i, NULL);
1952
1953                 flex_group = ext4_flex_group(sbi, i);
1954                 atomic_add(ext4_free_inodes_count(sb, gdp),
1955                            &sbi->s_flex_groups[flex_group].free_inodes);
1956                 atomic_add(ext4_free_blks_count(sb, gdp),
1957                            &sbi->s_flex_groups[flex_group].free_blocks);
1958                 atomic_add(ext4_used_dirs_count(sb, gdp),
1959                            &sbi->s_flex_groups[flex_group].used_dirs);
1960         }
1961
1962         return 1;
1963 failed:
1964         return 0;
1965 }
1966
1967 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1968                             struct ext4_group_desc *gdp)
1969 {
1970         __u16 crc = 0;
1971
1972         if (sbi->s_es->s_feature_ro_compat &
1973             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1974                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1975                 __le32 le_group = cpu_to_le32(block_group);
1976
1977                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1978                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1979                 crc = crc16(crc, (__u8 *)gdp, offset);
1980                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1981                 /* for checksum of struct ext4_group_desc do the rest...*/
1982                 if ((sbi->s_es->s_feature_incompat &
1983                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1984                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1985                         crc = crc16(crc, (__u8 *)gdp + offset,
1986                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1987                                         offset);
1988         }
1989
1990         return cpu_to_le16(crc);
1991 }
1992
1993 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1994                                 struct ext4_group_desc *gdp)
1995 {
1996         if ((sbi->s_es->s_feature_ro_compat &
1997              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1998             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1999                 return 0;
2000
2001         return 1;
2002 }
2003
2004 /* Called at mount-time, super-block is locked */
2005 static int ext4_check_descriptors(struct super_block *sb,
2006                                   ext4_group_t *first_not_zeroed)
2007 {
2008         struct ext4_sb_info *sbi = EXT4_SB(sb);
2009         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2010         ext4_fsblk_t last_block;
2011         ext4_fsblk_t block_bitmap;
2012         ext4_fsblk_t inode_bitmap;
2013         ext4_fsblk_t inode_table;
2014         int flexbg_flag = 0;
2015         ext4_group_t i, grp = sbi->s_groups_count;
2016
2017         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2018                 flexbg_flag = 1;
2019
2020         ext4_debug("Checking group descriptors");
2021
2022         for (i = 0; i < sbi->s_groups_count; i++) {
2023                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2024
2025                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2026                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2027                 else
2028                         last_block = first_block +
2029                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2030
2031                 if ((grp == sbi->s_groups_count) &&
2032                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2033                         grp = i;
2034
2035                 block_bitmap = ext4_block_bitmap(sb, gdp);
2036                 if (block_bitmap < first_block || block_bitmap > last_block) {
2037                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2038                                "Block bitmap for group %u not in group "
2039                                "(block %llu)!", i, block_bitmap);
2040                         return 0;
2041                 }
2042                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2043                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2044                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2045                                "Inode bitmap for group %u not in group "
2046                                "(block %llu)!", i, inode_bitmap);
2047                         return 0;
2048                 }
2049                 inode_table = ext4_inode_table(sb, gdp);
2050                 if (inode_table < first_block ||
2051                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2052                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2053                                "Inode table for group %u not in group "
2054                                "(block %llu)!", i, inode_table);
2055                         return 0;
2056                 }
2057                 ext4_lock_group(sb, i);
2058                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2059                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2060                                  "Checksum for group %u failed (%u!=%u)",
2061                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2062                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2063                         if (!(sb->s_flags & MS_RDONLY)) {
2064                                 ext4_unlock_group(sb, i);
2065                                 return 0;
2066                         }
2067                 }
2068                 ext4_unlock_group(sb, i);
2069                 if (!flexbg_flag)
2070                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2071         }
2072         if (NULL != first_not_zeroed)
2073                 *first_not_zeroed = grp;
2074
2075         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2076         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2077         return 1;
2078 }
2079
2080 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2081  * the superblock) which were deleted from all directories, but held open by
2082  * a process at the time of a crash.  We walk the list and try to delete these
2083  * inodes at recovery time (only with a read-write filesystem).
2084  *
2085  * In order to keep the orphan inode chain consistent during traversal (in
2086  * case of crash during recovery), we link each inode into the superblock
2087  * orphan list_head and handle it the same way as an inode deletion during
2088  * normal operation (which journals the operations for us).
2089  *
2090  * We only do an iget() and an iput() on each inode, which is very safe if we
2091  * accidentally point at an in-use or already deleted inode.  The worst that
2092  * can happen in this case is that we get a "bit already cleared" message from
2093  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2094  * e2fsck was run on this filesystem, and it must have already done the orphan
2095  * inode cleanup for us, so we can safely abort without any further action.
2096  */
2097 static void ext4_orphan_cleanup(struct super_block *sb,
2098                                 struct ext4_super_block *es)
2099 {
2100         unsigned int s_flags = sb->s_flags;
2101         int nr_orphans = 0, nr_truncates = 0;
2102 #ifdef CONFIG_QUOTA
2103         int i;
2104 #endif
2105         if (!es->s_last_orphan) {
2106                 jbd_debug(4, "no orphan inodes to clean up\n");
2107                 return;
2108         }
2109
2110         if (bdev_read_only(sb->s_bdev)) {
2111                 ext4_msg(sb, KERN_ERR, "write access "
2112                         "unavailable, skipping orphan cleanup");
2113                 return;
2114         }
2115
2116         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2117                 if (es->s_last_orphan)
2118                         jbd_debug(1, "Errors on filesystem, "
2119                                   "clearing orphan list.\n");
2120                 es->s_last_orphan = 0;
2121                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2122                 return;
2123         }
2124
2125         if (s_flags & MS_RDONLY) {
2126                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2127                 sb->s_flags &= ~MS_RDONLY;
2128         }
2129 #ifdef CONFIG_QUOTA
2130         /* Needed for iput() to work correctly and not trash data */
2131         sb->s_flags |= MS_ACTIVE;
2132         /* Turn on quotas so that they are updated correctly */
2133         for (i = 0; i < MAXQUOTAS; i++) {
2134                 if (EXT4_SB(sb)->s_qf_names[i]) {
2135                         int ret = ext4_quota_on_mount(sb, i);
2136                         if (ret < 0)
2137                                 ext4_msg(sb, KERN_ERR,
2138                                         "Cannot turn on journaled "
2139                                         "quota: error %d", ret);
2140                 }
2141         }
2142 #endif
2143
2144         while (es->s_last_orphan) {
2145                 struct inode *inode;
2146
2147                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2148                 if (IS_ERR(inode)) {
2149                         es->s_last_orphan = 0;
2150                         break;
2151                 }
2152
2153                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2154                 dquot_initialize(inode);
2155                 if (inode->i_nlink) {
2156                         ext4_msg(sb, KERN_DEBUG,
2157                                 "%s: truncating inode %lu to %lld bytes",
2158                                 __func__, inode->i_ino, inode->i_size);
2159                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2160                                   inode->i_ino, inode->i_size);
2161                         ext4_truncate(inode);
2162                         nr_truncates++;
2163                 } else {
2164                         ext4_msg(sb, KERN_DEBUG,
2165                                 "%s: deleting unreferenced inode %lu",
2166                                 __func__, inode->i_ino);
2167                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2168                                   inode->i_ino);
2169                         nr_orphans++;
2170                 }
2171                 iput(inode);  /* The delete magic happens here! */
2172         }
2173
2174 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2175
2176         if (nr_orphans)
2177                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2178                        PLURAL(nr_orphans));
2179         if (nr_truncates)
2180                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2181                        PLURAL(nr_truncates));
2182 #ifdef CONFIG_QUOTA
2183         /* Turn quotas off */
2184         for (i = 0; i < MAXQUOTAS; i++) {
2185                 if (sb_dqopt(sb)->files[i])
2186                         dquot_quota_off(sb, i);
2187         }
2188 #endif
2189         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2190 }
2191
2192 /*
2193  * Maximal extent format file size.
2194  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2195  * extent format containers, within a sector_t, and within i_blocks
2196  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2197  * so that won't be a limiting factor.
2198  *
2199  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2200  */
2201 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2202 {
2203         loff_t res;
2204         loff_t upper_limit = MAX_LFS_FILESIZE;
2205
2206         /* small i_blocks in vfs inode? */
2207         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2208                 /*
2209                  * CONFIG_LBDAF is not enabled implies the inode
2210                  * i_block represent total blocks in 512 bytes
2211                  * 32 == size of vfs inode i_blocks * 8
2212                  */
2213                 upper_limit = (1LL << 32) - 1;
2214
2215                 /* total blocks in file system block size */
2216                 upper_limit >>= (blkbits - 9);
2217                 upper_limit <<= blkbits;
2218         }
2219
2220         /* 32-bit extent-start container, ee_block */
2221         res = 1LL << 32;
2222         res <<= blkbits;
2223         res -= 1;
2224
2225         /* Sanity check against vm- & vfs- imposed limits */
2226         if (res > upper_limit)
2227                 res = upper_limit;
2228
2229         return res;
2230 }
2231
2232 /*
2233  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2234  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2235  * We need to be 1 filesystem block less than the 2^48 sector limit.
2236  */
2237 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2238 {
2239         loff_t res = EXT4_NDIR_BLOCKS;
2240         int meta_blocks;
2241         loff_t upper_limit;
2242         /* This is calculated to be the largest file size for a dense, block
2243          * mapped file such that the file's total number of 512-byte sectors,
2244          * including data and all indirect blocks, does not exceed (2^48 - 1).
2245          *
2246          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2247          * number of 512-byte sectors of the file.
2248          */
2249
2250         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2251                 /*
2252                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2253                  * the inode i_block field represents total file blocks in
2254                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2255                  */
2256                 upper_limit = (1LL << 32) - 1;
2257
2258                 /* total blocks in file system block size */
2259                 upper_limit >>= (bits - 9);
2260
2261         } else {
2262                 /*
2263                  * We use 48 bit ext4_inode i_blocks
2264                  * With EXT4_HUGE_FILE_FL set the i_blocks
2265                  * represent total number of blocks in
2266                  * file system block size
2267                  */
2268                 upper_limit = (1LL << 48) - 1;
2269
2270         }
2271
2272         /* indirect blocks */
2273         meta_blocks = 1;
2274         /* double indirect blocks */
2275         meta_blocks += 1 + (1LL << (bits-2));
2276         /* tripple indirect blocks */
2277         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2278
2279         upper_limit -= meta_blocks;
2280         upper_limit <<= bits;
2281
2282         res += 1LL << (bits-2);
2283         res += 1LL << (2*(bits-2));
2284         res += 1LL << (3*(bits-2));
2285         res <<= bits;
2286         if (res > upper_limit)
2287                 res = upper_limit;
2288
2289         if (res > MAX_LFS_FILESIZE)
2290                 res = MAX_LFS_FILESIZE;
2291
2292         return res;
2293 }
2294
2295 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2296                                    ext4_fsblk_t logical_sb_block, int nr)
2297 {
2298         struct ext4_sb_info *sbi = EXT4_SB(sb);
2299         ext4_group_t bg, first_meta_bg;
2300         int has_super = 0;
2301
2302         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2303
2304         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2305             nr < first_meta_bg)
2306                 return logical_sb_block + nr + 1;
2307         bg = sbi->s_desc_per_block * nr;
2308         if (ext4_bg_has_super(sb, bg))
2309                 has_super = 1;
2310
2311         return (has_super + ext4_group_first_block_no(sb, bg));
2312 }
2313
2314 /**
2315  * ext4_get_stripe_size: Get the stripe size.
2316  * @sbi: In memory super block info
2317  *
2318  * If we have specified it via mount option, then
2319  * use the mount option value. If the value specified at mount time is
2320  * greater than the blocks per group use the super block value.
2321  * If the super block value is greater than blocks per group return 0.
2322  * Allocator needs it be less than blocks per group.
2323  *
2324  */
2325 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2326 {
2327         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2328         unsigned long stripe_width =
2329                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2330
2331         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2332                 return sbi->s_stripe;
2333
2334         if (stripe_width <= sbi->s_blocks_per_group)
2335                 return stripe_width;
2336
2337         if (stride <= sbi->s_blocks_per_group)
2338                 return stride;
2339
2340         return 0;
2341 }
2342
2343 /* sysfs supprt */
2344
2345 struct ext4_attr {
2346         struct attribute attr;
2347         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2348         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2349                          const char *, size_t);
2350         int offset;
2351 };
2352
2353 static int parse_strtoul(const char *buf,
2354                 unsigned long max, unsigned long *value)
2355 {
2356         char *endp;
2357
2358         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2359         endp = skip_spaces(endp);
2360         if (*endp || *value > max)
2361                 return -EINVAL;
2362
2363         return 0;
2364 }
2365
2366 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2367                                               struct ext4_sb_info *sbi,
2368                                               char *buf)
2369 {
2370         return snprintf(buf, PAGE_SIZE, "%llu\n",
2371                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2372 }
2373
2374 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2375                                          struct ext4_sb_info *sbi, char *buf)
2376 {
2377         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2378
2379         if (!sb->s_bdev->bd_part)
2380                 return snprintf(buf, PAGE_SIZE, "0\n");
2381         return snprintf(buf, PAGE_SIZE, "%lu\n",
2382                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2383                          sbi->s_sectors_written_start) >> 1);
2384 }
2385
2386 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2387                                           struct ext4_sb_info *sbi, char *buf)
2388 {
2389         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2390
2391         if (!sb->s_bdev->bd_part)
2392                 return snprintf(buf, PAGE_SIZE, "0\n");
2393         return snprintf(buf, PAGE_SIZE, "%llu\n",
2394                         (unsigned long long)(sbi->s_kbytes_written +
2395                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2396                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2397 }
2398
2399 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2400                                           struct ext4_sb_info *sbi,
2401                                           const char *buf, size_t count)
2402 {
2403         unsigned long t;
2404
2405         if (parse_strtoul(buf, 0x40000000, &t))
2406                 return -EINVAL;
2407
2408         if (!is_power_of_2(t))
2409                 return -EINVAL;
2410
2411         sbi->s_inode_readahead_blks = t;
2412         return count;
2413 }
2414
2415 static ssize_t sbi_ui_show(struct ext4_attr *a,
2416                            struct ext4_sb_info *sbi, char *buf)
2417 {
2418         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2419
2420         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2421 }
2422
2423 static ssize_t sbi_ui_store(struct ext4_attr *a,
2424                             struct ext4_sb_info *sbi,
2425                             const char *buf, size_t count)
2426 {
2427         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2428         unsigned long t;
2429
2430         if (parse_strtoul(buf, 0xffffffff, &t))
2431                 return -EINVAL;
2432         *ui = t;
2433         return count;
2434 }
2435
2436 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2437 static struct ext4_attr ext4_attr_##_name = {                   \
2438         .attr = {.name = __stringify(_name), .mode = _mode },   \
2439         .show   = _show,                                        \
2440         .store  = _store,                                       \
2441         .offset = offsetof(struct ext4_sb_info, _elname),       \
2442 }
2443 #define EXT4_ATTR(name, mode, show, store) \
2444 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2445
2446 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2447 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2448 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2449 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2450         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2451 #define ATTR_LIST(name) &ext4_attr_##name.attr
2452
2453 EXT4_RO_ATTR(delayed_allocation_blocks);
2454 EXT4_RO_ATTR(session_write_kbytes);
2455 EXT4_RO_ATTR(lifetime_write_kbytes);
2456 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2457                  inode_readahead_blks_store, s_inode_readahead_blks);
2458 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2459 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2460 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2461 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2462 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2463 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2464 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2465 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2466
2467 static struct attribute *ext4_attrs[] = {
2468         ATTR_LIST(delayed_allocation_blocks),
2469         ATTR_LIST(session_write_kbytes),
2470         ATTR_LIST(lifetime_write_kbytes),
2471         ATTR_LIST(inode_readahead_blks),
2472         ATTR_LIST(inode_goal),
2473         ATTR_LIST(mb_stats),
2474         ATTR_LIST(mb_max_to_scan),
2475         ATTR_LIST(mb_min_to_scan),
2476         ATTR_LIST(mb_order2_req),
2477         ATTR_LIST(mb_stream_req),
2478         ATTR_LIST(mb_group_prealloc),
2479         ATTR_LIST(max_writeback_mb_bump),
2480         NULL,
2481 };
2482
2483 /* Features this copy of ext4 supports */
2484 EXT4_INFO_ATTR(lazy_itable_init);
2485 EXT4_INFO_ATTR(batched_discard);
2486
2487 static struct attribute *ext4_feat_attrs[] = {
2488         ATTR_LIST(lazy_itable_init),
2489         ATTR_LIST(batched_discard),
2490         NULL,
2491 };
2492
2493 static ssize_t ext4_attr_show(struct kobject *kobj,
2494                               struct attribute *attr, char *buf)
2495 {
2496         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2497                                                 s_kobj);
2498         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2499
2500         return a->show ? a->show(a, sbi, buf) : 0;
2501 }
2502
2503 static ssize_t ext4_attr_store(struct kobject *kobj,
2504                                struct attribute *attr,
2505                                const char *buf, size_t len)
2506 {
2507         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2508                                                 s_kobj);
2509         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2510
2511         return a->store ? a->store(a, sbi, buf, len) : 0;
2512 }
2513
2514 static void ext4_sb_release(struct kobject *kobj)
2515 {
2516         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2517                                                 s_kobj);
2518         complete(&sbi->s_kobj_unregister);
2519 }
2520
2521 static const struct sysfs_ops ext4_attr_ops = {
2522         .show   = ext4_attr_show,
2523         .store  = ext4_attr_store,
2524 };
2525
2526 static struct kobj_type ext4_ktype = {
2527         .default_attrs  = ext4_attrs,
2528         .sysfs_ops      = &ext4_attr_ops,
2529         .release        = ext4_sb_release,
2530 };
2531
2532 static void ext4_feat_release(struct kobject *kobj)
2533 {
2534         complete(&ext4_feat->f_kobj_unregister);
2535 }
2536
2537 static struct kobj_type ext4_feat_ktype = {
2538         .default_attrs  = ext4_feat_attrs,
2539         .sysfs_ops      = &ext4_attr_ops,
2540         .release        = ext4_feat_release,
2541 };
2542
2543 /*
2544  * Check whether this filesystem can be mounted based on
2545  * the features present and the RDONLY/RDWR mount requested.
2546  * Returns 1 if this filesystem can be mounted as requested,
2547  * 0 if it cannot be.
2548  */
2549 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2550 {
2551         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2552                 ext4_msg(sb, KERN_ERR,
2553                         "Couldn't mount because of "
2554                         "unsupported optional features (%x)",
2555                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2556                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2557                 return 0;
2558         }
2559
2560         if (readonly)
2561                 return 1;
2562
2563         /* Check that feature set is OK for a read-write mount */
2564         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2565                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2566                          "unsupported optional features (%x)",
2567                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2568                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2569                 return 0;
2570         }
2571         /*
2572          * Large file size enabled file system can only be mounted
2573          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2574          */
2575         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2576                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2577                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2578                                  "cannot be mounted RDWR without "
2579                                  "CONFIG_LBDAF");
2580                         return 0;
2581                 }
2582         }
2583         return 1;
2584 }
2585
2586 /*
2587  * This function is called once a day if we have errors logged
2588  * on the file system
2589  */
2590 static void print_daily_error_info(unsigned long arg)
2591 {
2592         struct super_block *sb = (struct super_block *) arg;
2593         struct ext4_sb_info *sbi;
2594         struct ext4_super_block *es;
2595
2596         sbi = EXT4_SB(sb);
2597         es = sbi->s_es;
2598
2599         if (es->s_error_count)
2600                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2601                          le32_to_cpu(es->s_error_count));
2602         if (es->s_first_error_time) {
2603                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2604                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2605                        (int) sizeof(es->s_first_error_func),
2606                        es->s_first_error_func,
2607                        le32_to_cpu(es->s_first_error_line));
2608                 if (es->s_first_error_ino)
2609                         printk(": inode %u",
2610                                le32_to_cpu(es->s_first_error_ino));
2611                 if (es->s_first_error_block)
2612                         printk(": block %llu", (unsigned long long)
2613                                le64_to_cpu(es->s_first_error_block));
2614                 printk("\n");
2615         }
2616         if (es->s_last_error_time) {
2617                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2618                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2619                        (int) sizeof(es->s_last_error_func),
2620                        es->s_last_error_func,
2621                        le32_to_cpu(es->s_last_error_line));
2622                 if (es->s_last_error_ino)
2623                         printk(": inode %u",
2624                                le32_to_cpu(es->s_last_error_ino));
2625                 if (es->s_last_error_block)
2626                         printk(": block %llu", (unsigned long long)
2627                                le64_to_cpu(es->s_last_error_block));
2628                 printk("\n");
2629         }
2630         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2631 }
2632
2633 static void ext4_lazyinode_timeout(unsigned long data)
2634 {
2635         struct task_struct *p = (struct task_struct *)data;
2636         wake_up_process(p);
2637 }
2638
2639 /* Find next suitable group and run ext4_init_inode_table */
2640 static int ext4_run_li_request(struct ext4_li_request *elr)
2641 {
2642         struct ext4_group_desc *gdp = NULL;
2643         ext4_group_t group, ngroups;
2644         struct super_block *sb;
2645         unsigned long timeout = 0;
2646         int ret = 0;
2647
2648         sb = elr->lr_super;
2649         ngroups = EXT4_SB(sb)->s_groups_count;
2650
2651         for (group = elr->lr_next_group; group < ngroups; group++) {
2652                 gdp = ext4_get_group_desc(sb, group, NULL);
2653                 if (!gdp) {
2654                         ret = 1;
2655                         break;
2656                 }
2657
2658                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2659                         break;
2660         }
2661
2662         if (group == ngroups)
2663                 ret = 1;
2664
2665         if (!ret) {
2666                 timeout = jiffies;
2667                 ret = ext4_init_inode_table(sb, group,
2668                                             elr->lr_timeout ? 0 : 1);
2669                 if (elr->lr_timeout == 0) {
2670                         timeout = jiffies - timeout;
2671                         if (elr->lr_sbi->s_li_wait_mult)
2672                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2673                         else
2674                                 timeout *= 20;
2675                         elr->lr_timeout = timeout;
2676                 }
2677                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2678                 elr->lr_next_group = group + 1;
2679         }
2680
2681         return ret;
2682 }
2683
2684 /*
2685  * Remove lr_request from the list_request and free the
2686  * request tructure. Should be called with li_list_mtx held
2687  */
2688 static void ext4_remove_li_request(struct ext4_li_request *elr)
2689 {
2690         struct ext4_sb_info *sbi;
2691
2692         if (!elr)
2693                 return;
2694
2695         sbi = elr->lr_sbi;
2696
2697         list_del(&elr->lr_request);
2698         sbi->s_li_request = NULL;
2699         kfree(elr);
2700 }
2701
2702 static void ext4_unregister_li_request(struct super_block *sb)
2703 {
2704         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2705
2706         if (!ext4_li_info)
2707                 return;
2708
2709         mutex_lock(&ext4_li_info->li_list_mtx);
2710         ext4_remove_li_request(elr);
2711         mutex_unlock(&ext4_li_info->li_list_mtx);
2712 }
2713
2714 /*
2715  * This is the function where ext4lazyinit thread lives. It walks
2716  * through the request list searching for next scheduled filesystem.
2717  * When such a fs is found, run the lazy initialization request
2718  * (ext4_rn_li_request) and keep track of the time spend in this
2719  * function. Based on that time we compute next schedule time of
2720  * the request. When walking through the list is complete, compute
2721  * next waking time and put itself into sleep.
2722  */
2723 static int ext4_lazyinit_thread(void *arg)
2724 {
2725         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2726         struct list_head *pos, *n;
2727         struct ext4_li_request *elr;
2728         unsigned long next_wakeup;
2729         DEFINE_WAIT(wait);
2730
2731         BUG_ON(NULL == eli);
2732
2733         eli->li_timer.data = (unsigned long)current;
2734         eli->li_timer.function = ext4_lazyinode_timeout;
2735
2736         eli->li_task = current;
2737         wake_up(&eli->li_wait_task);
2738
2739 cont_thread:
2740         while (true) {
2741                 next_wakeup = MAX_JIFFY_OFFSET;
2742
2743                 mutex_lock(&eli->li_list_mtx);
2744                 if (list_empty(&eli->li_request_list)) {
2745                         mutex_unlock(&eli->li_list_mtx);
2746                         goto exit_thread;
2747                 }
2748
2749                 list_for_each_safe(pos, n, &eli->li_request_list) {
2750                         elr = list_entry(pos, struct ext4_li_request,
2751                                          lr_request);
2752
2753                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2754                                 if (ext4_run_li_request(elr) != 0) {
2755                                         /* error, remove the lazy_init job */
2756                                         ext4_remove_li_request(elr);
2757                                         continue;
2758                                 }
2759                         }
2760
2761                         if (time_before(elr->lr_next_sched, next_wakeup))
2762                                 next_wakeup = elr->lr_next_sched;
2763                 }
2764                 mutex_unlock(&eli->li_list_mtx);
2765
2766                 if (freezing(current))
2767                         refrigerator();
2768
2769                 if ((time_after_eq(jiffies, next_wakeup)) ||
2770                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2771                         cond_resched();
2772                         continue;
2773                 }
2774
2775                 eli->li_timer.expires = next_wakeup;
2776                 add_timer(&eli->li_timer);
2777                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2778                                 TASK_INTERRUPTIBLE);
2779                 if (time_before(jiffies, next_wakeup))
2780                         schedule();
2781                 finish_wait(&eli->li_wait_daemon, &wait);
2782         }
2783
2784 exit_thread:
2785         /*
2786          * It looks like the request list is empty, but we need
2787          * to check it under the li_list_mtx lock, to prevent any
2788          * additions into it, and of course we should lock ext4_li_mtx
2789          * to atomically free the list and ext4_li_info, because at
2790          * this point another ext4 filesystem could be registering
2791          * new one.
2792          */
2793         mutex_lock(&ext4_li_mtx);
2794         mutex_lock(&eli->li_list_mtx);
2795         if (!list_empty(&eli->li_request_list)) {
2796                 mutex_unlock(&eli->li_list_mtx);
2797                 mutex_unlock(&ext4_li_mtx);
2798                 goto cont_thread;
2799         }
2800         mutex_unlock(&eli->li_list_mtx);
2801         del_timer_sync(&ext4_li_info->li_timer);
2802         eli->li_task = NULL;
2803         wake_up(&eli->li_wait_task);
2804
2805         kfree(ext4_li_info);
2806         ext4_li_info = NULL;
2807         mutex_unlock(&ext4_li_mtx);
2808
2809         return 0;
2810 }
2811
2812 static void ext4_clear_request_list(void)
2813 {
2814         struct list_head *pos, *n;
2815         struct ext4_li_request *elr;
2816
2817         mutex_lock(&ext4_li_info->li_list_mtx);
2818         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2819                 elr = list_entry(pos, struct ext4_li_request,
2820                                  lr_request);
2821                 ext4_remove_li_request(elr);
2822         }
2823         mutex_unlock(&ext4_li_info->li_list_mtx);
2824 }
2825
2826 static int ext4_run_lazyinit_thread(void)
2827 {
2828         struct task_struct *t;
2829
2830         t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2831         if (IS_ERR(t)) {
2832                 int err = PTR_ERR(t);
2833                 ext4_clear_request_list();
2834                 del_timer_sync(&ext4_li_info->li_timer);
2835                 kfree(ext4_li_info);
2836                 ext4_li_info = NULL;
2837                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2838                                  "initialization thread\n",
2839                                  err);
2840                 return err;
2841         }
2842         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2843
2844         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2845         return 0;
2846 }
2847
2848 /*
2849  * Check whether it make sense to run itable init. thread or not.
2850  * If there is at least one uninitialized inode table, return
2851  * corresponding group number, else the loop goes through all
2852  * groups and return total number of groups.
2853  */
2854 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2855 {
2856         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2857         struct ext4_group_desc *gdp = NULL;
2858
2859         for (group = 0; group < ngroups; group++) {
2860                 gdp = ext4_get_group_desc(sb, group, NULL);
2861                 if (!gdp)
2862                         continue;
2863
2864                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2865                         break;
2866         }
2867
2868         return group;
2869 }
2870
2871 static int ext4_li_info_new(void)
2872 {
2873         struct ext4_lazy_init *eli = NULL;
2874
2875         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2876         if (!eli)
2877                 return -ENOMEM;
2878
2879         eli->li_task = NULL;
2880         INIT_LIST_HEAD(&eli->li_request_list);
2881         mutex_init(&eli->li_list_mtx);
2882
2883         init_waitqueue_head(&eli->li_wait_daemon);
2884         init_waitqueue_head(&eli->li_wait_task);
2885         init_timer(&eli->li_timer);
2886         eli->li_state |= EXT4_LAZYINIT_QUIT;
2887
2888         ext4_li_info = eli;
2889
2890         return 0;
2891 }
2892
2893 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2894                                             ext4_group_t start)
2895 {
2896         struct ext4_sb_info *sbi = EXT4_SB(sb);
2897         struct ext4_li_request *elr;
2898         unsigned long rnd;
2899
2900         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2901         if (!elr)
2902                 return NULL;
2903
2904         elr->lr_super = sb;
2905         elr->lr_sbi = sbi;
2906         elr->lr_next_group = start;
2907
2908         /*
2909          * Randomize first schedule time of the request to
2910          * spread the inode table initialization requests
2911          * better.
2912          */
2913         get_random_bytes(&rnd, sizeof(rnd));
2914         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2915                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2916
2917         return elr;
2918 }
2919
2920 static int ext4_register_li_request(struct super_block *sb,
2921                                     ext4_group_t first_not_zeroed)
2922 {
2923         struct ext4_sb_info *sbi = EXT4_SB(sb);
2924         struct ext4_li_request *elr;
2925         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2926         int ret;
2927
2928         if (sbi->s_li_request != NULL)
2929                 return 0;
2930
2931         if (first_not_zeroed == ngroups ||
2932             (sb->s_flags & MS_RDONLY) ||
2933             !test_opt(sb, INIT_INODE_TABLE)) {
2934                 sbi->s_li_request = NULL;
2935                 return 0;
2936         }
2937
2938         if (first_not_zeroed == ngroups) {
2939                 sbi->s_li_request = NULL;
2940                 return 0;
2941         }
2942
2943         elr = ext4_li_request_new(sb, first_not_zeroed);
2944         if (!elr)
2945                 return -ENOMEM;
2946
2947         mutex_lock(&ext4_li_mtx);
2948
2949         if (NULL == ext4_li_info) {
2950                 ret = ext4_li_info_new();
2951                 if (ret)
2952                         goto out;
2953         }
2954
2955         mutex_lock(&ext4_li_info->li_list_mtx);
2956         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2957         mutex_unlock(&ext4_li_info->li_list_mtx);
2958
2959         sbi->s_li_request = elr;
2960
2961         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2962                 ret = ext4_run_lazyinit_thread();
2963                 if (ret)
2964                         goto out;
2965         }
2966 out:
2967         mutex_unlock(&ext4_li_mtx);
2968         if (ret)
2969                 kfree(elr);
2970         return ret;
2971 }
2972
2973 /*
2974  * We do not need to lock anything since this is called on
2975  * module unload.
2976  */
2977 static void ext4_destroy_lazyinit_thread(void)
2978 {
2979         /*
2980          * If thread exited earlier
2981          * there's nothing to be done.
2982          */
2983         if (!ext4_li_info)
2984                 return;
2985
2986         ext4_clear_request_list();
2987
2988         while (ext4_li_info->li_task) {
2989                 wake_up(&ext4_li_info->li_wait_daemon);
2990                 wait_event(ext4_li_info->li_wait_task,
2991                            ext4_li_info->li_task == NULL);
2992         }
2993 }
2994
2995 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2996                                 __releases(kernel_lock)
2997                                 __acquires(kernel_lock)
2998 {
2999         char *orig_data = kstrdup(data, GFP_KERNEL);
3000         struct buffer_head *bh;
3001         struct ext4_super_block *es = NULL;
3002         struct ext4_sb_info *sbi;
3003         ext4_fsblk_t block;
3004         ext4_fsblk_t sb_block = get_sb_block(&data);
3005         ext4_fsblk_t logical_sb_block;
3006         unsigned long offset = 0;
3007         unsigned long journal_devnum = 0;
3008         unsigned long def_mount_opts;
3009         struct inode *root;
3010         char *cp;
3011         const char *descr;
3012         int ret = -ENOMEM;
3013         int blocksize;
3014         unsigned int db_count;
3015         unsigned int i;
3016         int needs_recovery, has_huge_files;
3017         __u64 blocks_count;
3018         int err;
3019         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3020         ext4_group_t first_not_zeroed;
3021
3022         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3023         if (!sbi)
3024                 goto out_free_orig;
3025
3026         sbi->s_blockgroup_lock =
3027                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3028         if (!sbi->s_blockgroup_lock) {
3029                 kfree(sbi);
3030                 goto out_free_orig;
3031         }
3032         sb->s_fs_info = sbi;
3033         sbi->s_mount_opt = 0;
3034         sbi->s_resuid = EXT4_DEF_RESUID;
3035         sbi->s_resgid = EXT4_DEF_RESGID;
3036         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3037         sbi->s_sb_block = sb_block;
3038         if (sb->s_bdev->bd_part)
3039                 sbi->s_sectors_written_start =
3040                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3041
3042         /* Cleanup superblock name */
3043         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3044                 *cp = '!';
3045
3046         ret = -EINVAL;
3047         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3048         if (!blocksize) {
3049                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3050                 goto out_fail;
3051         }
3052
3053         /*
3054          * The ext4 superblock will not be buffer aligned for other than 1kB
3055          * block sizes.  We need to calculate the offset from buffer start.
3056          */
3057         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3058                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3059                 offset = do_div(logical_sb_block, blocksize);
3060         } else {
3061                 logical_sb_block = sb_block;
3062         }
3063
3064         if (!(bh = sb_bread(sb, logical_sb_block))) {
3065                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3066                 goto out_fail;
3067         }
3068         /*
3069          * Note: s_es must be initialized as soon as possible because
3070          *       some ext4 macro-instructions depend on its value
3071          */
3072         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3073         sbi->s_es = es;
3074         sb->s_magic = le16_to_cpu(es->s_magic);
3075         if (sb->s_magic != EXT4_SUPER_MAGIC)
3076                 goto cantfind_ext4;
3077         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3078
3079         /* Set defaults before we parse the mount options */
3080         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3081         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
3082         if (def_mount_opts & EXT4_DEFM_DEBUG)
3083                 set_opt(sbi->s_mount_opt, DEBUG);
3084         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3085                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3086                         "2.6.38");
3087                 set_opt(sbi->s_mount_opt, GRPID);
3088         }
3089         if (def_mount_opts & EXT4_DEFM_UID16)
3090                 set_opt(sbi->s_mount_opt, NO_UID32);
3091 #ifdef CONFIG_EXT4_FS_XATTR
3092         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3093                 set_opt(sbi->s_mount_opt, XATTR_USER);
3094 #endif
3095 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3096         if (def_mount_opts & EXT4_DEFM_ACL)
3097                 set_opt(sbi->s_mount_opt, POSIX_ACL);
3098 #endif
3099         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3100                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3101         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3102                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
3103         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3104                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3105
3106         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3107                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
3108         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3109                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
3110         else
3111                 set_opt(sbi->s_mount_opt, ERRORS_RO);
3112         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3113                 set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
3114         if (def_mount_opts & EXT4_DEFM_DISCARD)
3115                 set_opt(sbi->s_mount_opt, DISCARD);
3116
3117         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3118         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3119         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3120         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3121         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3122
3123         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3124                 set_opt(sbi->s_mount_opt, BARRIER);
3125
3126         /*
3127          * enable delayed allocation by default
3128          * Use -o nodelalloc to turn it off
3129          */
3130         if (!IS_EXT3_SB(sb) &&
3131             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3132                 set_opt(sbi->s_mount_opt, DELALLOC);
3133
3134         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3135                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3136                 ext4_msg(sb, KERN_WARNING,
3137                          "failed to parse options in superblock: %s",
3138                          sbi->s_es->s_mount_opts);
3139         }
3140         if (!parse_options((char *) data, sb, &journal_devnum,
3141                            &journal_ioprio, NULL, 0))
3142                 goto failed_mount;
3143
3144         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3145                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3146
3147         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3148             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3149              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3150              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3151                 ext4_msg(sb, KERN_WARNING,
3152                        "feature flags set on rev 0 fs, "
3153                        "running e2fsck is recommended");
3154
3155         /*
3156          * Check feature flags regardless of the revision level, since we
3157          * previously didn't change the revision level when setting the flags,
3158          * so there is a chance incompat flags are set on a rev 0 filesystem.
3159          */
3160         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3161                 goto failed_mount;
3162
3163         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3164
3165         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3166             blocksize > EXT4_MAX_BLOCK_SIZE) {
3167                 ext4_msg(sb, KERN_ERR,
3168                        "Unsupported filesystem blocksize %d", blocksize);
3169                 goto failed_mount;
3170         }
3171
3172         if (sb->s_blocksize != blocksize) {
3173                 /* Validate the filesystem blocksize */
3174                 if (!sb_set_blocksize(sb, blocksize)) {
3175                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3176                                         blocksize);
3177                         goto failed_mount;
3178                 }
3179
3180                 brelse(bh);
3181                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3182                 offset = do_div(logical_sb_block, blocksize);
3183                 bh = sb_bread(sb, logical_sb_block);
3184                 if (!bh) {
3185                         ext4_msg(sb, KERN_ERR,
3186                                "Can't read superblock on 2nd try");
3187                         goto failed_mount;
3188                 }
3189                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3190                 sbi->s_es = es;
3191                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3192                         ext4_msg(sb, KERN_ERR,
3193                                "Magic mismatch, very weird!");
3194                         goto failed_mount;
3195                 }
3196         }
3197
3198         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3199                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3200         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3201                                                       has_huge_files);
3202         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3203
3204         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3205                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3206                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3207         } else {
3208                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3209                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3210                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3211                     (!is_power_of_2(sbi->s_inode_size)) ||
3212                     (sbi->s_inode_size > blocksize)) {
3213                         ext4_msg(sb, KERN_ERR,
3214                                "unsupported inode size: %d",
3215                                sbi->s_inode_size);
3216                         goto failed_mount;
3217                 }
3218                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3219                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3220         }
3221
3222         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3223         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3224                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3225                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3226                     !is_power_of_2(sbi->s_desc_size)) {
3227                         ext4_msg(sb, KERN_ERR,
3228                                "unsupported descriptor size %lu",
3229                                sbi->s_desc_size);
3230                         goto failed_mount;
3231                 }
3232         } else
3233                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3234
3235         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3236         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3237         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3238                 goto cantfind_ext4;
3239
3240         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3241         if (sbi->s_inodes_per_block == 0)
3242                 goto cantfind_ext4;
3243         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3244                                         sbi->s_inodes_per_block;
3245         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3246         sbi->s_sbh = bh;
3247         sbi->s_mount_state = le16_to_cpu(es->s_state);
3248         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3249         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3250
3251         for (i = 0; i < 4; i++)
3252                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3253         sbi->s_def_hash_version = es->s_def_hash_version;
3254         i = le32_to_cpu(es->s_flags);
3255         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3256                 sbi->s_hash_unsigned = 3;
3257         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3258 #ifdef __CHAR_UNSIGNED__
3259                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3260                 sbi->s_hash_unsigned = 3;
3261 #else
3262                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3263 #endif
3264                 sb->s_dirt = 1;
3265         }
3266
3267         if (sbi->s_blocks_per_group > blocksize * 8) {
3268                 ext4_msg(sb, KERN_ERR,
3269                        "#blocks per group too big: %lu",
3270                        sbi->s_blocks_per_group);
3271                 goto failed_mount;
3272         }
3273         if (sbi->s_inodes_per_group > blocksize * 8) {
3274                 ext4_msg(sb, KERN_ERR,
3275                        "#inodes per group too big: %lu",
3276                        sbi->s_inodes_per_group);
3277                 goto failed_mount;
3278         }
3279
3280         /*
3281          * Test whether we have more sectors than will fit in sector_t,
3282          * and whether the max offset is addressable by the page cache.
3283          */
3284         err = generic_check_addressable(sb->s_blocksize_bits,
3285                                         ext4_blocks_count(es));
3286         if (err) {
3287                 ext4_msg(sb, KERN_ERR, "filesystem"
3288                          " too large to mount safely on this system");
3289                 if (sizeof(sector_t) < 8)
3290                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3291                 ret = err;
3292                 goto failed_mount;
3293         }
3294
3295         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3296                 goto cantfind_ext4;
3297
3298         /* check blocks count against device size */
3299         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3300         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3301                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3302                        "exceeds size of device (%llu blocks)",
3303                        ext4_blocks_count(es), blocks_count);
3304                 goto failed_mount;
3305         }
3306
3307         /*
3308          * It makes no sense for the first data block to be beyond the end
3309          * of the filesystem.
3310          */
3311         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3312                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3313                          "block %u is beyond end of filesystem (%llu)",
3314                          le32_to_cpu(es->s_first_data_block),
3315                          ext4_blocks_count(es));
3316                 goto failed_mount;
3317         }
3318         blocks_count = (ext4_blocks_count(es) -
3319                         le32_to_cpu(es->s_first_data_block) +
3320                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3321         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3322         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3323                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3324                        "(block count %llu, first data block %u, "
3325                        "blocks per group %lu)", sbi->s_groups_count,
3326                        ext4_blocks_count(es),
3327                        le32_to_cpu(es->s_first_data_block),
3328                        EXT4_BLOCKS_PER_GROUP(sb));
3329                 goto failed_mount;
3330         }
3331         sbi->s_groups_count = blocks_count;
3332         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3333                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3334         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3335                    EXT4_DESC_PER_BLOCK(sb);
3336         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3337                                     GFP_KERNEL);
3338         if (sbi->s_group_desc == NULL) {
3339                 ext4_msg(sb, KERN_ERR, "not enough memory");
3340                 goto failed_mount;
3341         }
3342
3343 #ifdef CONFIG_PROC_FS
3344         if (ext4_proc_root)
3345                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3346 #endif
3347
3348         bgl_lock_init(sbi->s_blockgroup_lock);
3349
3350         for (i = 0; i < db_count; i++) {
3351                 block = descriptor_loc(sb, logical_sb_block, i);
3352                 sbi->s_group_desc[i] = sb_bread(sb, block);
3353                 if (!sbi->s_group_desc[i]) {
3354                         ext4_msg(sb, KERN_ERR,
3355                                "can't read group descriptor %d", i);
3356                         db_count = i;
3357                         goto failed_mount2;
3358                 }
3359         }
3360         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3361                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3362                 goto failed_mount2;
3363         }
3364         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3365                 if (!ext4_fill_flex_info(sb)) {
3366                         ext4_msg(sb, KERN_ERR,
3367                                "unable to initialize "
3368                                "flex_bg meta info!");
3369                         goto failed_mount2;
3370                 }
3371
3372         sbi->s_gdb_count = db_count;
3373         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3374         spin_lock_init(&sbi->s_next_gen_lock);
3375
3376         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3377                         ext4_count_free_blocks(sb));
3378         if (!err) {
3379                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3380                                 ext4_count_free_inodes(sb));
3381         }
3382         if (!err) {
3383                 err = percpu_counter_init(&sbi->s_dirs_counter,
3384                                 ext4_count_dirs(sb));
3385         }
3386         if (!err) {
3387                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3388         }
3389         if (err) {
3390                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3391                 goto failed_mount3;
3392         }
3393
3394         sbi->s_stripe = ext4_get_stripe_size(sbi);
3395         sbi->s_max_writeback_mb_bump = 128;
3396
3397         /*
3398          * set up enough so that it can read an inode
3399          */
3400         if (!test_opt(sb, NOLOAD) &&
3401             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3402                 sb->s_op = &ext4_sops;
3403         else
3404                 sb->s_op = &ext4_nojournal_sops;
3405         sb->s_export_op = &ext4_export_ops;
3406         sb->s_xattr = ext4_xattr_handlers;
3407 #ifdef CONFIG_QUOTA
3408         sb->s_qcop = &ext4_qctl_operations;
3409         sb->dq_op = &ext4_quota_operations;
3410 #endif
3411         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3412         mutex_init(&sbi->s_orphan_lock);
3413         mutex_init(&sbi->s_resize_lock);
3414
3415         sb->s_root = NULL;
3416
3417         needs_recovery = (es->s_last_orphan != 0 ||
3418                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3419                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3420
3421         /*
3422          * The first inode we look at is the journal inode.  Don't try
3423          * root first: it may be modified in the journal!
3424          */
3425         if (!test_opt(sb, NOLOAD) &&
3426             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3427                 if (ext4_load_journal(sb, es, journal_devnum))
3428                         goto failed_mount3;
3429         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3430               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3431                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3432                        "suppressed and not mounted read-only");
3433                 goto failed_mount_wq;
3434         } else {
3435                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
3436                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3437                 sbi->s_journal = NULL;
3438                 needs_recovery = 0;
3439                 goto no_journal;
3440         }
3441
3442         if (ext4_blocks_count(es) > 0xffffffffULL &&
3443             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3444                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3445                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3446                 goto failed_mount_wq;
3447         }
3448
3449         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3450                 jbd2_journal_set_features(sbi->s_journal,
3451                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3452                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3453         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3454                 jbd2_journal_set_features(sbi->s_journal,
3455                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3456                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3457                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3458         } else {
3459                 jbd2_journal_clear_features(sbi->s_journal,
3460                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3461                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3462         }
3463
3464         /* We have now updated the journal if required, so we can
3465          * validate the data journaling mode. */
3466         switch (test_opt(sb, DATA_FLAGS)) {
3467         case 0:
3468                 /* No mode set, assume a default based on the journal
3469                  * capabilities: ORDERED_DATA if the journal can
3470                  * cope, else JOURNAL_DATA
3471                  */
3472                 if (jbd2_journal_check_available_features
3473                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3474                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
3475                 else
3476                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3477                 break;
3478
3479         case EXT4_MOUNT_ORDERED_DATA:
3480         case EXT4_MOUNT_WRITEBACK_DATA:
3481                 if (!jbd2_journal_check_available_features
3482                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3483                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3484                                "requested data journaling mode");
3485                         goto failed_mount_wq;
3486                 }
3487         default:
3488                 break;
3489         }
3490         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3491
3492         /*
3493          * The journal may have updated the bg summary counts, so we
3494          * need to update the global counters.
3495          */
3496         percpu_counter_set(&sbi->s_freeblocks_counter,
3497                            ext4_count_free_blocks(sb));
3498         percpu_counter_set(&sbi->s_freeinodes_counter,
3499                            ext4_count_free_inodes(sb));
3500         percpu_counter_set(&sbi->s_dirs_counter,
3501                            ext4_count_dirs(sb));
3502         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3503
3504 no_journal:
3505         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3506         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3507                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3508                 goto failed_mount_wq;
3509         }
3510
3511         /*
3512          * The jbd2_journal_load will have done any necessary log recovery,
3513          * so we can safely mount the rest of the filesystem now.
3514          */
3515
3516         root = ext4_iget(sb, EXT4_ROOT_INO);
3517         if (IS_ERR(root)) {
3518                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3519                 ret = PTR_ERR(root);
3520                 goto failed_mount4;
3521         }
3522         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3523                 iput(root);
3524                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3525                 goto failed_mount4;
3526         }
3527         sb->s_root = d_alloc_root(root);
3528         if (!sb->s_root) {
3529                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3530                 iput(root);
3531                 ret = -ENOMEM;
3532                 goto failed_mount4;
3533         }
3534
3535         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3536
3537         /* determine the minimum size of new large inodes, if present */
3538         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3539                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3540                                                      EXT4_GOOD_OLD_INODE_SIZE;
3541                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3542                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3543                         if (sbi->s_want_extra_isize <
3544                             le16_to_cpu(es->s_want_extra_isize))
3545                                 sbi->s_want_extra_isize =
3546                                         le16_to_cpu(es->s_want_extra_isize);
3547                         if (sbi->s_want_extra_isize <
3548                             le16_to_cpu(es->s_min_extra_isize))
3549                                 sbi->s_want_extra_isize =
3550                                         le16_to_cpu(es->s_min_extra_isize);
3551                 }
3552         }
3553         /* Check if enough inode space is available */
3554         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3555                                                         sbi->s_inode_size) {
3556                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3557                                                        EXT4_GOOD_OLD_INODE_SIZE;
3558                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3559                          "available");
3560         }
3561
3562         if (test_opt(sb, DELALLOC) &&
3563             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3564                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3565                          "requested data journaling mode");
3566                 clear_opt(sbi->s_mount_opt, DELALLOC);
3567         }
3568         if (test_opt(sb, DIOREAD_NOLOCK)) {
3569                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3570                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3571                                 "option - requested data journaling mode");
3572                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3573                 }
3574                 if (sb->s_blocksize < PAGE_SIZE) {
3575                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3576                                 "option - block size is too small");
3577                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3578                 }
3579         }
3580
3581         err = ext4_setup_system_zone(sb);
3582         if (err) {
3583                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3584                          "zone (%d)", err);
3585                 goto failed_mount4;
3586         }
3587
3588         ext4_ext_init(sb);
3589         err = ext4_mb_init(sb, needs_recovery);
3590         if (err) {
3591                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3592                          err);
3593                 goto failed_mount4;
3594         }
3595
3596         err = ext4_register_li_request(sb, first_not_zeroed);
3597         if (err)
3598                 goto failed_mount4;
3599
3600         sbi->s_kobj.kset = ext4_kset;
3601         init_completion(&sbi->s_kobj_unregister);
3602         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3603                                    "%s", sb->s_id);
3604         if (err) {
3605                 ext4_mb_release(sb);
3606                 ext4_ext_release(sb);
3607                 goto failed_mount4;
3608         };
3609
3610         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3611         ext4_orphan_cleanup(sb, es);
3612         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3613         if (needs_recovery) {
3614                 ext4_msg(sb, KERN_INFO, "recovery complete");
3615                 ext4_mark_recovery_complete(sb, es);
3616         }
3617         if (EXT4_SB(sb)->s_journal) {
3618                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3619                         descr = " journalled data mode";
3620                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3621                         descr = " ordered data mode";
3622                 else
3623                         descr = " writeback data mode";
3624         } else
3625                 descr = "out journal";
3626
3627         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3628                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3629                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3630
3631         init_timer(&sbi->s_err_report);
3632         sbi->s_err_report.function = print_daily_error_info;
3633         sbi->s_err_report.data = (unsigned long) sb;
3634         if (es->s_error_count)
3635                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3636
3637         kfree(orig_data);
3638         return 0;
3639
3640 cantfind_ext4:
3641         if (!silent)
3642                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3643         goto failed_mount;
3644
3645 failed_mount4:
3646         ext4_msg(sb, KERN_ERR, "mount failed");
3647         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3648 failed_mount_wq:
3649         ext4_release_system_zone(sb);
3650         if (sbi->s_journal) {
3651                 jbd2_journal_destroy(sbi->s_journal);
3652                 sbi->s_journal = NULL;
3653         }
3654 failed_mount3:
3655         if (sbi->s_flex_groups) {
3656                 if (is_vmalloc_addr(sbi->s_flex_groups))
3657                         vfree(sbi->s_flex_groups);
3658                 else
3659                         kfree(sbi->s_flex_groups);
3660         }
3661         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3662         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3663         percpu_counter_destroy(&sbi->s_dirs_counter);
3664         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3665 failed_mount2:
3666         for (i = 0; i < db_count; i++)
3667                 brelse(sbi->s_group_desc[i]);
3668         kfree(sbi->s_group_desc);
3669 failed_mount:
3670         if (sbi->s_proc) {
3671                 remove_proc_entry(sb->s_id, ext4_proc_root);
3672         }
3673 #ifdef CONFIG_QUOTA
3674         for (i = 0; i < MAXQUOTAS; i++)
3675                 kfree(sbi->s_qf_names[i]);
3676 #endif
3677         ext4_blkdev_remove(sbi);
3678         brelse(bh);
3679 out_fail:
3680         sb->s_fs_info = NULL;
3681         kfree(sbi->s_blockgroup_lock);
3682         kfree(sbi);
3683 out_free_orig:
3684         kfree(orig_data);
3685         return ret;
3686 }
3687
3688 /*
3689  * Setup any per-fs journal parameters now.  We'll do this both on
3690  * initial mount, once the journal has been initialised but before we've
3691  * done any recovery; and again on any subsequent remount.
3692  */
3693 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3694 {
3695         struct ext4_sb_info *sbi = EXT4_SB(sb);
3696
3697         journal->j_commit_interval = sbi->s_commit_interval;
3698         journal->j_min_batch_time = sbi->s_min_batch_time;
3699         journal->j_max_batch_time = sbi->s_max_batch_time;
3700
3701         write_lock(&journal->j_state_lock);
3702         if (test_opt(sb, BARRIER))
3703                 journal->j_flags |= JBD2_BARRIER;
3704         else
3705                 journal->j_flags &= ~JBD2_BARRIER;
3706         if (test_opt(sb, DATA_ERR_ABORT))
3707                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3708         else
3709                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3710         write_unlock(&journal->j_state_lock);
3711 }
3712
3713 static journal_t *ext4_get_journal(struct super_block *sb,
3714                                    unsigned int journal_inum)
3715 {
3716         struct inode *journal_inode;
3717         journal_t *journal;
3718
3719         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3720
3721         /* First, test for the existence of a valid inode on disk.  Bad
3722          * things happen if we iget() an unused inode, as the subsequent
3723          * iput() will try to delete it. */
3724
3725         journal_inode = ext4_iget(sb, journal_inum);
3726         if (IS_ERR(journal_inode)) {
3727                 ext4_msg(sb, KERN_ERR, "no journal found");
3728                 return NULL;
3729         }
3730         if (!journal_inode->i_nlink) {
3731                 make_bad_inode(journal_inode);
3732                 iput(journal_inode);
3733                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3734                 return NULL;
3735         }
3736
3737         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3738                   journal_inode, journal_inode->i_size);
3739         if (!S_ISREG(journal_inode->i_mode)) {
3740                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3741                 iput(journal_inode);
3742                 return NULL;
3743         }
3744
3745         journal = jbd2_journal_init_inode(journal_inode);
3746         if (!journal) {
3747                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3748                 iput(journal_inode);
3749                 return NULL;
3750         }
3751         journal->j_private = sb;
3752         ext4_init_journal_params(sb, journal);
3753         return journal;
3754 }
3755
3756 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3757                                        dev_t j_dev)
3758 {
3759         struct buffer_head *bh;
3760         journal_t *journal;
3761         ext4_fsblk_t start;
3762         ext4_fsblk_t len;
3763         int hblock, blocksize;
3764         ext4_fsblk_t sb_block;
3765         unsigned long offset;
3766         struct ext4_super_block *es;
3767         struct block_device *bdev;
3768
3769         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3770
3771         bdev = ext4_blkdev_get(j_dev, sb);
3772         if (bdev == NULL)
3773                 return NULL;
3774
3775         if (bd_claim(bdev, sb)) {
3776                 ext4_msg(sb, KERN_ERR,
3777                         "failed to claim external journal device");
3778                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3779                 return NULL;
3780         }
3781
3782         blocksize = sb->s_blocksize;
3783         hblock = bdev_logical_block_size(bdev);
3784         if (blocksize < hblock) {
3785                 ext4_msg(sb, KERN_ERR,
3786                         "blocksize too small for journal device");
3787                 goto out_bdev;
3788         }
3789
3790         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3791         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3792         set_blocksize(bdev, blocksize);
3793         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3794                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3795                        "external journal");
3796                 goto out_bdev;
3797         }
3798
3799         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3800         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3801             !(le32_to_cpu(es->s_feature_incompat) &
3802               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3803                 ext4_msg(sb, KERN_ERR, "external journal has "
3804                                         "bad superblock");
3805                 brelse(bh);
3806                 goto out_bdev;
3807         }
3808
3809         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3810                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3811                 brelse(bh);
3812                 goto out_bdev;
3813         }
3814
3815         len = ext4_blocks_count(es);
3816         start = sb_block + 1;
3817         brelse(bh);     /* we're done with the superblock */
3818
3819         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3820                                         start, len, blocksize);
3821         if (!journal) {
3822                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3823                 goto out_bdev;
3824         }
3825         journal->j_private = sb;
3826         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3827         wait_on_buffer(journal->j_sb_buffer);
3828         if (!buffer_uptodate(journal->j_sb_buffer)) {
3829                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3830                 goto out_journal;
3831         }
3832         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3833                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3834                                         "user (unsupported) - %d",
3835                         be32_to_cpu(journal->j_superblock->s_nr_users));
3836                 goto out_journal;
3837         }
3838         EXT4_SB(sb)->journal_bdev = bdev;
3839         ext4_init_journal_params(sb, journal);
3840         return journal;
3841
3842 out_journal:
3843         jbd2_journal_destroy(journal);
3844 out_bdev:
3845         ext4_blkdev_put(bdev);
3846         return NULL;
3847 }
3848
3849 static int ext4_load_journal(struct super_block *sb,
3850                              struct ext4_super_block *es,
3851                              unsigned long journal_devnum)
3852 {
3853         journal_t *journal;
3854         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3855         dev_t journal_dev;
3856         int err = 0;
3857         int really_read_only;
3858
3859         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3860
3861         if (journal_devnum &&
3862             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3863                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3864                         "numbers have changed");
3865                 journal_dev = new_decode_dev(journal_devnum);
3866         } else
3867                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3868
3869         really_read_only = bdev_read_only(sb->s_bdev);
3870
3871         /*
3872          * Are we loading a blank journal or performing recovery after a
3873          * crash?  For recovery, we need to check in advance whether we
3874          * can get read-write access to the device.
3875          */
3876         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3877                 if (sb->s_flags & MS_RDONLY) {
3878                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3879                                         "required on readonly filesystem");
3880                         if (really_read_only) {
3881                                 ext4_msg(sb, KERN_ERR, "write access "
3882                                         "unavailable, cannot proceed");
3883                                 return -EROFS;
3884                         }
3885                         ext4_msg(sb, KERN_INFO, "write access will "
3886                                "be enabled during recovery");
3887                 }
3888         }
3889
3890         if (journal_inum && journal_dev) {
3891                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3892                        "and inode journals!");
3893                 return -EINVAL;
3894         }
3895
3896         if (journal_inum) {
3897                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3898                         return -EINVAL;
3899         } else {
3900                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3901                         return -EINVAL;
3902         }
3903
3904         if (!(journal->j_flags & JBD2_BARRIER))
3905                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3906
3907         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3908                 err = jbd2_journal_update_format(journal);
3909                 if (err)  {
3910                         ext4_msg(sb, KERN_ERR, "error updating journal");
3911                         jbd2_journal_destroy(journal);
3912                         return err;
3913                 }
3914         }
3915
3916         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3917                 err = jbd2_journal_wipe(journal, !really_read_only);
3918         if (!err) {
3919                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3920                 if (save)
3921                         memcpy(save, ((char *) es) +
3922                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3923                 err = jbd2_journal_load(journal);
3924                 if (save)
3925                         memcpy(((char *) es) + EXT4_S_ERR_START,
3926                                save, EXT4_S_ERR_LEN);
3927                 kfree(save);
3928         }
3929
3930         if (err) {
3931                 ext4_msg(sb, KERN_ERR, "error loading journal");
3932                 jbd2_journal_destroy(journal);
3933                 return err;
3934         }
3935
3936         EXT4_SB(sb)->s_journal = journal;
3937         ext4_clear_journal_err(sb, es);
3938
3939         if (!really_read_only && journal_devnum &&
3940             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3941                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3942
3943                 /* Make sure we flush the recovery flag to disk. */
3944                 ext4_commit_super(sb, 1);
3945         }
3946
3947         return 0;
3948 }
3949
3950 static int ext4_commit_super(struct super_block *sb, int sync)
3951 {
3952         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3953         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3954         int error = 0;
3955
3956         if (!sbh)
3957                 return error;
3958         if (buffer_write_io_error(sbh)) {
3959                 /*
3960                  * Oh, dear.  A previous attempt to write the
3961                  * superblock failed.  This could happen because the
3962                  * USB device was yanked out.  Or it could happen to
3963                  * be a transient write error and maybe the block will
3964                  * be remapped.  Nothing we can do but to retry the
3965                  * write and hope for the best.
3966                  */
3967                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3968                        "superblock detected");
3969                 clear_buffer_write_io_error(sbh);
3970                 set_buffer_uptodate(sbh);
3971         }
3972         /*
3973          * If the file system is mounted read-only, don't update the
3974          * superblock write time.  This avoids updating the superblock
3975          * write time when we are mounting the root file system
3976          * read/only but we need to replay the journal; at that point,
3977          * for people who are east of GMT and who make their clock
3978          * tick in localtime for Windows bug-for-bug compatibility,
3979          * the clock is set in the future, and this will cause e2fsck
3980          * to complain and force a full file system check.
3981          */
3982         if (!(sb->s_flags & MS_RDONLY))
3983                 es->s_wtime = cpu_to_le32(get_seconds());
3984         if (sb->s_bdev->bd_part)
3985                 es->s_kbytes_written =
3986                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3987                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3988                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3989         else
3990                 es->s_kbytes_written =
3991                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3992         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3993                                            &EXT4_SB(sb)->s_freeblocks_counter));
3994         es->s_free_inodes_count =
3995                 cpu_to_le32(percpu_counter_sum_positive(
3996                                 &EXT4_SB(sb)->s_freeinodes_counter));
3997         sb->s_dirt = 0;
3998         BUFFER_TRACE(sbh, "marking dirty");
3999         mark_buffer_dirty(sbh);
4000         if (sync) {
4001                 error = sync_dirty_buffer(sbh);
4002                 if (error)
4003                         return error;
4004
4005                 error = buffer_write_io_error(sbh);
4006                 if (error) {
4007                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4008                                "superblock");
4009                         clear_buffer_write_io_error(sbh);
4010                         set_buffer_uptodate(sbh);
4011                 }
4012         }
4013         return error;
4014 }
4015
4016 /*
4017  * Have we just finished recovery?  If so, and if we are mounting (or
4018  * remounting) the filesystem readonly, then we will end up with a
4019  * consistent fs on disk.  Record that fact.
4020  */
4021 static void ext4_mark_recovery_complete(struct super_block *sb,
4022                                         struct ext4_super_block *es)
4023 {
4024         journal_t *journal = EXT4_SB(sb)->s_journal;
4025
4026         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4027                 BUG_ON(journal != NULL);
4028                 return;
4029         }
4030         jbd2_journal_lock_updates(journal);
4031         if (jbd2_journal_flush(journal) < 0)
4032                 goto out;
4033
4034         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4035             sb->s_flags & MS_RDONLY) {
4036                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4037                 ext4_commit_super(sb, 1);
4038         }
4039
4040 out:
4041         jbd2_journal_unlock_updates(journal);
4042 }
4043
4044 /*
4045  * If we are mounting (or read-write remounting) a filesystem whose journal
4046  * has recorded an error from a previous lifetime, move that error to the
4047  * main filesystem now.
4048  */
4049 static void ext4_clear_journal_err(struct super_block *sb,
4050                                    struct ext4_super_block *es)
4051 {
4052         journal_t *journal;
4053         int j_errno;
4054         const char *errstr;
4055
4056         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4057
4058         journal = EXT4_SB(sb)->s_journal;
4059
4060         /*
4061          * Now check for any error status which may have been recorded in the
4062          * journal by a prior ext4_error() or ext4_abort()
4063          */
4064
4065         j_errno = jbd2_journal_errno(journal);
4066         if (j_errno) {
4067                 char nbuf[16];
4068
4069                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4070                 ext4_warning(sb, "Filesystem error recorded "
4071                              "from previous mount: %s", errstr);
4072                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4073
4074                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4075                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4076                 ext4_commit_super(sb, 1);
4077
4078                 jbd2_journal_clear_err(journal);
4079         }
4080 }
4081
4082 /*
4083  * Force the running and committing transactions to commit,
4084  * and wait on the commit.
4085  */
4086 int ext4_force_commit(struct super_block *sb)
4087 {
4088         journal_t *journal;
4089         int ret = 0;
4090
4091         if (sb->s_flags & MS_RDONLY)
4092                 return 0;
4093
4094         journal = EXT4_SB(sb)->s_journal;
4095         if (journal) {
4096                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4097                 ret = ext4_journal_force_commit(journal);
4098         }
4099
4100         return ret;
4101 }
4102
4103 static void ext4_write_super(struct super_block *sb)
4104 {
4105         lock_super(sb);
4106         ext4_commit_super(sb, 1);
4107         unlock_super(sb);
4108 }
4109
4110 static int ext4_sync_fs(struct super_block *sb, int wait)
4111 {
4112         int ret = 0;
4113         tid_t target;
4114         struct ext4_sb_info *sbi = EXT4_SB(sb);
4115
4116         trace_ext4_sync_fs(sb, wait);
4117         flush_workqueue(sbi->dio_unwritten_wq);
4118         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4119                 if (wait)
4120                         jbd2_log_wait_commit(sbi->s_journal, target);
4121         }
4122         return ret;
4123 }
4124
4125 /*
4126  * LVM calls this function before a (read-only) snapshot is created.  This
4127  * gives us a chance to flush the journal completely and mark the fs clean.
4128  */
4129 static int ext4_freeze(struct super_block *sb)
4130 {
4131         int error = 0;
4132         journal_t *journal;
4133
4134         if (sb->s_flags & MS_RDONLY)
4135                 return 0;
4136
4137         journal = EXT4_SB(sb)->s_journal;
4138
4139         /* Now we set up the journal barrier. */
4140         jbd2_journal_lock_updates(journal);
4141
4142         /*
4143          * Don't clear the needs_recovery flag if we failed to flush
4144          * the journal.
4145          */
4146         error = jbd2_journal_flush(journal);
4147         if (error < 0)
4148                 goto out;
4149
4150         /* Journal blocked and flushed, clear needs_recovery flag. */
4151         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4152         error = ext4_commit_super(sb, 1);
4153 out:
4154         /* we rely on s_frozen to stop further updates */
4155         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4156         return error;
4157 }
4158
4159 /*
4160  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4161  * flag here, even though the filesystem is not technically dirty yet.
4162  */
4163 static int ext4_unfreeze(struct super_block *sb)
4164 {
4165         if (sb->s_flags & MS_RDONLY)
4166                 return 0;
4167
4168         lock_super(sb);
4169         /* Reset the needs_recovery flag before the fs is unlocked. */
4170         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4171         ext4_commit_super(sb, 1);
4172         unlock_super(sb);
4173         return 0;
4174 }
4175
4176 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4177 {
4178         struct ext4_super_block *es;
4179         struct ext4_sb_info *sbi = EXT4_SB(sb);
4180         ext4_fsblk_t n_blocks_count = 0;
4181         unsigned long old_sb_flags;
4182         struct ext4_mount_options old_opts;
4183         int enable_quota = 0;
4184         ext4_group_t g;
4185         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4186         int err;
4187 #ifdef CONFIG_QUOTA
4188         int i;
4189 #endif
4190         char *orig_data = kstrdup(data, GFP_KERNEL);
4191
4192         /* Store the original options */
4193         lock_super(sb);
4194         old_sb_flags = sb->s_flags;
4195         old_opts.s_mount_opt = sbi->s_mount_opt;
4196         old_opts.s_resuid = sbi->s_resuid;
4197         old_opts.s_resgid = sbi->s_resgid;
4198         old_opts.s_commit_interval = sbi->s_commit_interval;
4199         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4200         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4201 #ifdef CONFIG_QUOTA
4202         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4203         for (i = 0; i < MAXQUOTAS; i++)
4204                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4205 #endif
4206         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4207                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4208
4209         /*
4210          * Allow the "check" option to be passed as a remount option.
4211          */
4212         if (!parse_options(data, sb, NULL, &journal_ioprio,
4213                            &n_blocks_count, 1)) {
4214                 err = -EINVAL;
4215                 goto restore_opts;
4216         }
4217
4218         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4219                 ext4_abort(sb, "Abort forced by user");
4220
4221         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4222                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4223
4224         es = sbi->s_es;
4225
4226         if (sbi->s_journal) {
4227                 ext4_init_journal_params(sb, sbi->s_journal);
4228                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4229         }
4230
4231         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4232                 n_blocks_count > ext4_blocks_count(es)) {
4233                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4234                         err = -EROFS;
4235                         goto restore_opts;
4236                 }
4237
4238                 if (*flags & MS_RDONLY) {
4239                         err = dquot_suspend(sb, -1);
4240                         if (err < 0)
4241                                 goto restore_opts;
4242
4243                         /*
4244                          * First of all, the unconditional stuff we have to do
4245                          * to disable replay of the journal when we next remount
4246                          */
4247                         sb->s_flags |= MS_RDONLY;
4248
4249                         /*
4250                          * OK, test if we are remounting a valid rw partition
4251                          * readonly, and if so set the rdonly flag and then
4252                          * mark the partition as valid again.
4253                          */
4254                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4255                             (sbi->s_mount_state & EXT4_VALID_FS))
4256                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4257
4258                         if (sbi->s_journal)
4259                                 ext4_mark_recovery_complete(sb, es);
4260                 } else {
4261                         /* Make sure we can mount this feature set readwrite */
4262                         if (!ext4_feature_set_ok(sb, 0)) {
4263                                 err = -EROFS;
4264                                 goto restore_opts;
4265                         }
4266                         /*
4267                          * Make sure the group descriptor checksums
4268                          * are sane.  If they aren't, refuse to remount r/w.
4269                          */
4270                         for (g = 0; g < sbi->s_groups_count; g++) {
4271                                 struct ext4_group_desc *gdp =
4272                                         ext4_get_group_desc(sb, g, NULL);
4273
4274                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4275                                         ext4_msg(sb, KERN_ERR,
4276                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4277                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4278                                                le16_to_cpu(gdp->bg_checksum));
4279                                         err = -EINVAL;
4280                                         goto restore_opts;
4281                                 }
4282                         }
4283
4284                         /*
4285                          * If we have an unprocessed orphan list hanging
4286                          * around from a previously readonly bdev mount,
4287                          * require a full umount/remount for now.
4288                          */
4289                         if (es->s_last_orphan) {
4290                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4291                                        "remount RDWR because of unprocessed "
4292                                        "orphan inode list.  Please "
4293                                        "umount/remount instead");
4294                                 err = -EINVAL;
4295                                 goto restore_opts;
4296                         }
4297
4298                         /*
4299                          * Mounting a RDONLY partition read-write, so reread
4300                          * and store the current valid flag.  (It may have
4301                          * been changed by e2fsck since we originally mounted
4302                          * the partition.)
4303                          */
4304                         if (sbi->s_journal)
4305                                 ext4_clear_journal_err(sb, es);
4306                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4307                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4308                                 goto restore_opts;
4309                         if (!ext4_setup_super(sb, es, 0))
4310                                 sb->s_flags &= ~MS_RDONLY;
4311                         enable_quota = 1;
4312                 }
4313         }
4314
4315         /*
4316          * Reinitialize lazy itable initialization thread based on
4317          * current settings
4318          */
4319         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4320                 ext4_unregister_li_request(sb);
4321         else {
4322                 ext4_group_t first_not_zeroed;
4323                 first_not_zeroed = ext4_has_uninit_itable(sb);
4324                 ext4_register_li_request(sb, first_not_zeroed);
4325         }
4326
4327         ext4_setup_system_zone(sb);
4328         if (sbi->s_journal == NULL)
4329                 ext4_commit_super(sb, 1);
4330
4331 #ifdef CONFIG_QUOTA
4332         /* Release old quota file names */
4333         for (i = 0; i < MAXQUOTAS; i++)
4334                 if (old_opts.s_qf_names[i] &&
4335                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4336                         kfree(old_opts.s_qf_names[i]);
4337 #endif
4338         unlock_super(sb);
4339         if (enable_quota)
4340                 dquot_resume(sb, -1);
4341
4342         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4343         kfree(orig_data);
4344         return 0;
4345
4346 restore_opts:
4347         sb->s_flags = old_sb_flags;
4348         sbi->s_mount_opt = old_opts.s_mount_opt;
4349         sbi->s_resuid = old_opts.s_resuid;
4350         sbi->s_resgid = old_opts.s_resgid;
4351         sbi->s_commit_interval = old_opts.s_commit_interval;
4352         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4353         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4354 #ifdef CONFIG_QUOTA
4355         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4356         for (i = 0; i < MAXQUOTAS; i++) {
4357                 if (sbi->s_qf_names[i] &&
4358                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4359                         kfree(sbi->s_qf_names[i]);
4360                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4361         }
4362 #endif
4363         unlock_super(sb);
4364         kfree(orig_data);
4365         return err;
4366 }
4367
4368 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4369 {
4370         struct super_block *sb = dentry->d_sb;
4371         struct ext4_sb_info *sbi = EXT4_SB(sb);
4372         struct ext4_super_block *es = sbi->s_es;
4373         u64 fsid;
4374
4375         if (test_opt(sb, MINIX_DF)) {
4376                 sbi->s_overhead_last = 0;
4377         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4378                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4379                 ext4_fsblk_t overhead = 0;
4380
4381                 /*
4382                  * Compute the overhead (FS structures).  This is constant
4383                  * for a given filesystem unless the number of block groups
4384                  * changes so we cache the previous value until it does.
4385                  */
4386
4387                 /*
4388                  * All of the blocks before first_data_block are
4389                  * overhead
4390                  */
4391                 overhead = le32_to_cpu(es->s_first_data_block);
4392
4393                 /*
4394                  * Add the overhead attributed to the superblock and
4395                  * block group descriptors.  If the sparse superblocks
4396                  * feature is turned on, then not all groups have this.
4397                  */
4398                 for (i = 0; i < ngroups; i++) {
4399                         overhead += ext4_bg_has_super(sb, i) +
4400                                 ext4_bg_num_gdb(sb, i);
4401                         cond_resched();
4402                 }
4403
4404                 /*
4405                  * Every block group has an inode bitmap, a block
4406                  * bitmap, and an inode table.
4407                  */
4408                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4409                 sbi->s_overhead_last = overhead;
4410                 smp_wmb();
4411                 sbi->s_blocks_last = ext4_blocks_count(es);
4412         }
4413
4414         buf->f_type = EXT4_SUPER_MAGIC;
4415         buf->f_bsize = sb->s_blocksize;
4416         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4417         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4418                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4419         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4420         if (buf->f_bfree < ext4_r_blocks_count(es))
4421                 buf->f_bavail = 0;
4422         buf->f_files = le32_to_cpu(es->s_inodes_count);
4423         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4424         buf->f_namelen = EXT4_NAME_LEN;
4425         fsid = le64_to_cpup((void *)es->s_uuid) ^
4426                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4427         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4428         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4429
4430         return 0;
4431 }
4432
4433 /* Helper function for writing quotas on sync - we need to start transaction
4434  * before quota file is locked for write. Otherwise the are possible deadlocks:
4435  * Process 1                         Process 2
4436  * ext4_create()                     quota_sync()
4437  *   jbd2_journal_start()                  write_dquot()
4438  *   dquot_initialize()                         down(dqio_mutex)
4439  *     down(dqio_mutex)                    jbd2_journal_start()
4440  *
4441  */
4442
4443 #ifdef CONFIG_QUOTA
4444
4445 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4446 {
4447         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4448 }
4449
4450 static int ext4_write_dquot(struct dquot *dquot)
4451 {
4452         int ret, err;
4453         handle_t *handle;
4454         struct inode *inode;
4455
4456         inode = dquot_to_inode(dquot);
4457         handle = ext4_journal_start(inode,
4458                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4459         if (IS_ERR(handle))
4460                 return PTR_ERR(handle);
4461         ret = dquot_commit(dquot);
4462         err = ext4_journal_stop(handle);
4463         if (!ret)
4464                 ret = err;
4465         return ret;
4466 }
4467
4468 static int ext4_acquire_dquot(struct dquot *dquot)
4469 {
4470         int ret, err;
4471         handle_t *handle;
4472
4473         handle = ext4_journal_start(dquot_to_inode(dquot),
4474                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4475         if (IS_ERR(handle))
4476                 return PTR_ERR(handle);
4477         ret = dquot_acquire(dquot);
4478         err = ext4_journal_stop(handle);
4479         if (!ret)
4480                 ret = err;
4481         return ret;
4482 }
4483
4484 static int ext4_release_dquot(struct dquot *dquot)
4485 {
4486         int ret, err;
4487         handle_t *handle;
4488
4489         handle = ext4_journal_start(dquot_to_inode(dquot),
4490                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4491         if (IS_ERR(handle)) {
4492                 /* Release dquot anyway to avoid endless cycle in dqput() */
4493                 dquot_release(dquot);
4494                 return PTR_ERR(handle);
4495         }
4496         ret = dquot_release(dquot);
4497         err = ext4_journal_stop(handle);
4498         if (!ret)
4499                 ret = err;
4500         return ret;
4501 }
4502
4503 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4504 {
4505         /* Are we journaling quotas? */
4506         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4507             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4508                 dquot_mark_dquot_dirty(dquot);
4509                 return ext4_write_dquot(dquot);
4510         } else {
4511                 return dquot_mark_dquot_dirty(dquot);
4512         }
4513 }
4514
4515 static int ext4_write_info(struct super_block *sb, int type)
4516 {
4517         int ret, err;
4518         handle_t *handle;
4519
4520         /* Data block + inode block */
4521         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4522         if (IS_ERR(handle))
4523                 return PTR_ERR(handle);
4524         ret = dquot_commit_info(sb, type);
4525         err = ext4_journal_stop(handle);
4526         if (!ret)
4527                 ret = err;
4528         return ret;
4529 }
4530
4531 /*
4532  * Turn on quotas during mount time - we need to find
4533  * the quota file and such...
4534  */
4535 static int ext4_quota_on_mount(struct super_block *sb, int type)
4536 {
4537         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4538                                         EXT4_SB(sb)->s_jquota_fmt, type);
4539 }
4540
4541 /*
4542  * Standard function to be called on quota_on
4543  */
4544 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4545                          char *name)
4546 {
4547         int err;
4548         struct path path;
4549
4550         if (!test_opt(sb, QUOTA))
4551                 return -EINVAL;
4552
4553         err = kern_path(name, LOOKUP_FOLLOW, &path);
4554         if (err)
4555                 return err;
4556
4557         /* Quotafile not on the same filesystem? */
4558         if (path.mnt->mnt_sb != sb) {
4559                 path_put(&path);
4560                 return -EXDEV;
4561         }
4562         /* Journaling quota? */
4563         if (EXT4_SB(sb)->s_qf_names[type]) {
4564                 /* Quotafile not in fs root? */
4565                 if (path.dentry->d_parent != sb->s_root)
4566                         ext4_msg(sb, KERN_WARNING,
4567                                 "Quota file not on filesystem root. "
4568                                 "Journaled quota will not work");
4569         }
4570
4571         /*
4572          * When we journal data on quota file, we have to flush journal to see
4573          * all updates to the file when we bypass pagecache...
4574          */
4575         if (EXT4_SB(sb)->s_journal &&
4576             ext4_should_journal_data(path.dentry->d_inode)) {
4577                 /*
4578                  * We don't need to lock updates but journal_flush() could
4579                  * otherwise be livelocked...
4580                  */
4581                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4582                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4583                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4584                 if (err) {
4585                         path_put(&path);
4586                         return err;
4587                 }
4588         }
4589
4590         err = dquot_quota_on_path(sb, type, format_id, &path);
4591         path_put(&path);
4592         return err;
4593 }
4594
4595 static int ext4_quota_off(struct super_block *sb, int type)
4596 {
4597         /* Force all delayed allocation blocks to be allocated.
4598          * Caller already holds s_umount sem */
4599         if (test_opt(sb, DELALLOC))
4600                 sync_filesystem(sb);
4601
4602         return dquot_quota_off(sb, type);
4603 }
4604
4605 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4606  * acquiring the locks... As quota files are never truncated and quota code
4607  * itself serializes the operations (and noone else should touch the files)
4608  * we don't have to be afraid of races */
4609 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4610                                size_t len, loff_t off)
4611 {
4612         struct inode *inode = sb_dqopt(sb)->files[type];
4613         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4614         int err = 0;
4615         int offset = off & (sb->s_blocksize - 1);
4616         int tocopy;
4617         size_t toread;
4618         struct buffer_head *bh;
4619         loff_t i_size = i_size_read(inode);
4620
4621         if (off > i_size)
4622                 return 0;
4623         if (off+len > i_size)
4624                 len = i_size-off;
4625         toread = len;
4626         while (toread > 0) {
4627                 tocopy = sb->s_blocksize - offset < toread ?
4628                                 sb->s_blocksize - offset : toread;
4629                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4630                 if (err)
4631                         return err;
4632                 if (!bh)        /* A hole? */
4633                         memset(data, 0, tocopy);
4634                 else
4635                         memcpy(data, bh->b_data+offset, tocopy);
4636                 brelse(bh);
4637                 offset = 0;
4638                 toread -= tocopy;
4639                 data += tocopy;
4640                 blk++;
4641         }
4642         return len;
4643 }
4644
4645 /* Write to quotafile (we know the transaction is already started and has
4646  * enough credits) */
4647 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4648                                 const char *data, size_t len, loff_t off)
4649 {
4650         struct inode *inode = sb_dqopt(sb)->files[type];
4651         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4652         int err = 0;
4653         int offset = off & (sb->s_blocksize - 1);
4654         struct buffer_head *bh;
4655         handle_t *handle = journal_current_handle();
4656
4657         if (EXT4_SB(sb)->s_journal && !handle) {
4658                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4659                         " cancelled because transaction is not started",
4660                         (unsigned long long)off, (unsigned long long)len);
4661                 return -EIO;
4662         }
4663         /*
4664          * Since we account only one data block in transaction credits,
4665          * then it is impossible to cross a block boundary.
4666          */
4667         if (sb->s_blocksize - offset < len) {
4668                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4669                         " cancelled because not block aligned",
4670                         (unsigned long long)off, (unsigned long long)len);
4671                 return -EIO;
4672         }
4673
4674         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4675         bh = ext4_bread(handle, inode, blk, 1, &err);
4676         if (!bh)
4677                 goto out;
4678         err = ext4_journal_get_write_access(handle, bh);
4679         if (err) {
4680                 brelse(bh);
4681                 goto out;
4682         }
4683         lock_buffer(bh);
4684         memcpy(bh->b_data+offset, data, len);
4685         flush_dcache_page(bh->b_page);
4686         unlock_buffer(bh);
4687         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4688         brelse(bh);
4689 out:
4690         if (err) {
4691                 mutex_unlock(&inode->i_mutex);
4692                 return err;
4693         }
4694         if (inode->i_size < off + len) {
4695                 i_size_write(inode, off + len);
4696                 EXT4_I(inode)->i_disksize = inode->i_size;
4697         }
4698         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4699         ext4_mark_inode_dirty(handle, inode);
4700         mutex_unlock(&inode->i_mutex);
4701         return len;
4702 }
4703
4704 #endif
4705
4706 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4707                        const char *dev_name, void *data)
4708 {
4709         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4710 }
4711
4712 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4713 static struct file_system_type ext2_fs_type = {
4714         .owner          = THIS_MODULE,
4715         .name           = "ext2",
4716         .mount          = ext4_mount,
4717         .kill_sb        = kill_block_super,
4718         .fs_flags       = FS_REQUIRES_DEV,
4719 };
4720
4721 static inline void register_as_ext2(void)
4722 {
4723         int err = register_filesystem(&ext2_fs_type);
4724         if (err)
4725                 printk(KERN_WARNING
4726                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4727 }
4728
4729 static inline void unregister_as_ext2(void)
4730 {
4731         unregister_filesystem(&ext2_fs_type);
4732 }
4733 MODULE_ALIAS("ext2");
4734 #else
4735 static inline void register_as_ext2(void) { }
4736 static inline void unregister_as_ext2(void) { }
4737 #endif
4738
4739 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4740 static inline void register_as_ext3(void)
4741 {
4742         int err = register_filesystem(&ext3_fs_type);
4743         if (err)
4744                 printk(KERN_WARNING
4745                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4746 }
4747
4748 static inline void unregister_as_ext3(void)
4749 {
4750         unregister_filesystem(&ext3_fs_type);
4751 }
4752 MODULE_ALIAS("ext3");
4753 #else
4754 static inline void register_as_ext3(void) { }
4755 static inline void unregister_as_ext3(void) { }
4756 #endif
4757
4758 static struct file_system_type ext4_fs_type = {
4759         .owner          = THIS_MODULE,
4760         .name           = "ext4",
4761         .mount          = ext4_mount,
4762         .kill_sb        = kill_block_super,
4763         .fs_flags       = FS_REQUIRES_DEV,
4764 };
4765
4766 int __init ext4_init_feat_adverts(void)
4767 {
4768         struct ext4_features *ef;
4769         int ret = -ENOMEM;
4770
4771         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4772         if (!ef)
4773                 goto out;
4774
4775         ef->f_kobj.kset = ext4_kset;
4776         init_completion(&ef->f_kobj_unregister);
4777         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4778                                    "features");
4779         if (ret) {
4780                 kfree(ef);
4781                 goto out;
4782         }
4783
4784         ext4_feat = ef;
4785         ret = 0;
4786 out:
4787         return ret;
4788 }
4789
4790 static int __init ext4_init_fs(void)
4791 {
4792         int err;
4793
4794         ext4_check_flag_values();
4795         err = ext4_init_pageio();
4796         if (err)
4797                 return err;
4798         err = ext4_init_system_zone();
4799         if (err)
4800                 goto out5;
4801         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4802         if (!ext4_kset)
4803                 goto out4;
4804         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4805
4806         err = ext4_init_feat_adverts();
4807
4808         err = ext4_init_mballoc();
4809         if (err)
4810                 goto out3;
4811
4812         err = ext4_init_xattr();
4813         if (err)
4814                 goto out2;
4815         err = init_inodecache();
4816         if (err)
4817                 goto out1;
4818         register_as_ext2();
4819         register_as_ext3();
4820         err = register_filesystem(&ext4_fs_type);
4821         if (err)
4822                 goto out;
4823
4824         ext4_li_info = NULL;
4825         mutex_init(&ext4_li_mtx);
4826         return 0;
4827 out:
4828         unregister_as_ext2();
4829         unregister_as_ext3();
4830         destroy_inodecache();
4831 out1:
4832         ext4_exit_xattr();
4833 out2:
4834         ext4_exit_mballoc();
4835 out3:
4836         kfree(ext4_feat);
4837         remove_proc_entry("fs/ext4", NULL);
4838         kset_unregister(ext4_kset);
4839 out4:
4840         ext4_exit_system_zone();
4841 out5:
4842         ext4_exit_pageio();
4843         return err;
4844 }
4845
4846 static void __exit ext4_exit_fs(void)
4847 {
4848         ext4_destroy_lazyinit_thread();
4849         unregister_as_ext2();
4850         unregister_as_ext3();
4851         unregister_filesystem(&ext4_fs_type);
4852         destroy_inodecache();
4853         ext4_exit_xattr();
4854         ext4_exit_mballoc();
4855         remove_proc_entry("fs/ext4", NULL);
4856         kset_unregister(ext4_kset);
4857         ext4_exit_system_zone();
4858         ext4_exit_pageio();
4859 }
4860
4861 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4862 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4863 MODULE_LICENSE("GPL");
4864 module_init(ext4_init_fs)
4865 module_exit(ext4_exit_fs)