fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_jbd2.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "mballoc.h"
  51
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/ext4.h>
  54
  55 static struct proc_dir_entry *ext4_proc_root;
  56 static struct kset *ext4_kset;
  57 struct ext4_lazy_init *ext4_li_info;
  58 struct mutex ext4_li_mtx;
  59 struct ext4_features *ext4_feat;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static const char *ext4_decode_error(struct super_block *sb, int errno,
  70                                      char nbuf[16]);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static void ext4_write_super(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static void ext4_destroy_lazyinit_thread(void);
  79 static void ext4_unregister_li_request(struct super_block *sb);
  80
  81 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  82 static struct file_system_type ext3_fs_type = {
  83         .owner          = THIS_MODULE,
  84         .name           = "ext3",
  85         .mount          = ext4_mount,
  86         .kill_sb        = kill_block_super,
  87         .fs_flags       = FS_REQUIRES_DEV,
  88 };
  89 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  90 #else
  91 #define IS_EXT3_SB(sb) (0)
  92 #endif
  93
  94 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  95                                struct ext4_group_desc *bg)
  96 {
  97         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  98                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  99                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 100 }
 101
 102 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 103                                struct ext4_group_desc *bg)
 104 {
 105         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 106                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 107                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 108 }
 109
 110 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 111                               struct ext4_group_desc *bg)
 112 {
 113         return le32_to_cpu(bg->bg_inode_table_lo) |
 114                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 115                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 116 }
 117
 118 __u32 ext4_free_blks_count(struct super_block *sb,
 119                               struct ext4_group_desc *bg)
 120 {
 121         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 122                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 123                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 124 }
 125
 126 __u32 ext4_free_inodes_count(struct super_block *sb,
 127                               struct ext4_group_desc *bg)
 128 {
 129         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 130                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 131                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 132 }
 133
 134 __u32 ext4_used_dirs_count(struct super_block *sb,
 135                               struct ext4_group_desc *bg)
 136 {
 137         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 138                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 139                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 140 }
 141
 142 __u32 ext4_itable_unused_count(struct super_block *sb,
 143                               struct ext4_group_desc *bg)
 144 {
 145         return le16_to_cpu(bg->bg_itable_unused_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 148 }
 149
 150 void ext4_block_bitmap_set(struct super_block *sb,
 151                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 152 {
 153         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 154         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 155                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 156 }
 157
 158 void ext4_inode_bitmap_set(struct super_block *sb,
 159                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 160 {
 161         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 162         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 163                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 164 }
 165
 166 void ext4_inode_table_set(struct super_block *sb,
 167                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 168 {
 169         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 170         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 171                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 172 }
 173
 174 void ext4_free_blks_set(struct super_block *sb,
 175                           struct ext4_group_desc *bg, __u32 count)
 176 {
 177         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 178         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 179                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 180 }
 181
 182 void ext4_free_inodes_set(struct super_block *sb,
 183                           struct ext4_group_desc *bg, __u32 count)
 184 {
 185         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 186         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 187                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 188 }
 189
 190 void ext4_used_dirs_set(struct super_block *sb,
 191                           struct ext4_group_desc *bg, __u32 count)
 192 {
 193         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 194         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 195                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 196 }
 197
 198 void ext4_itable_unused_set(struct super_block *sb,
 199                           struct ext4_group_desc *bg, __u32 count)
 200 {
 201         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 204 }
 205
 206
 207 /* Just increment the non-pointer handle value */
 208 static handle_t *ext4_get_nojournal(void)
 209 {
 210         handle_t *handle = current->journal_info;
 211         unsigned long ref_cnt = (unsigned long)handle;
 212
 213         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 214
 215         ref_cnt++;
 216         handle = (handle_t *)ref_cnt;
 217
 218         current->journal_info = handle;
 219         return handle;
 220 }
 221
 222
 223 /* Decrement the non-pointer handle value */
 224 static void ext4_put_nojournal(handle_t *handle)
 225 {
 226         unsigned long ref_cnt = (unsigned long)handle;
 227
 228         BUG_ON(ref_cnt == 0);
 229
 230         ref_cnt--;
 231         handle = (handle_t *)ref_cnt;
 232
 233         current->journal_info = handle;
 234 }
 235
 236 /*
 237  * Wrappers for jbd2_journal_start/end.
 238  *
 239  * The only special thing we need to do here is to make sure that all
 240  * journal_end calls result in the superblock being marked dirty, so
 241  * that sync() will call the filesystem's write_super callback if
 242  * appropriate.
 243  */
 244 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 245 {
 246         journal_t *journal;
 247
 248         if (sb->s_flags & MS_RDONLY)
 249                 return ERR_PTR(-EROFS);
 250
 251         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 252         /* Special case here: if the journal has aborted behind our
 253          * backs (eg. EIO in the commit thread), then we still need to
 254          * take the FS itself readonly cleanly. */
 255         journal = EXT4_SB(sb)->s_journal;
 256         if (journal) {
 257                 if (is_journal_aborted(journal)) {
 258                         ext4_abort(sb, "Detected aborted journal");
 259                         return ERR_PTR(-EROFS);
 260                 }
 261                 return jbd2_journal_start(journal, nblocks);
 262         }
 263         return ext4_get_nojournal();
 264 }
 265
 266 /*
 267  * The only special thing we need to do here is to make sure that all
 268  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 269  * that sync() will call the filesystem's write_super callback if
 270  * appropriate.
 271  */
 272 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 273 {
 274         struct super_block *sb;
 275         int err;
 276         int rc;
 277
 278         if (!ext4_handle_valid(handle)) {
 279                 ext4_put_nojournal(handle);
 280                 return 0;
 281         }
 282         sb = handle->h_transaction->t_journal->j_private;
 283         err = handle->h_err;
 284         rc = jbd2_journal_stop(handle);
 285
 286         if (!err)
 287                 err = rc;
 288         if (err)
 289                 __ext4_std_error(sb, where, line, err);
 290         return err;
 291 }
 292
 293 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 294                                const char *err_fn, struct buffer_head *bh,
 295                                handle_t *handle, int err)
 296 {
 297         char nbuf[16];
 298         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 299
 300         BUG_ON(!ext4_handle_valid(handle));
 301
 302         if (bh)
 303                 BUFFER_TRACE(bh, "abort");
 304
 305         if (!handle->h_err)
 306                 handle->h_err = err;
 307
 308         if (is_handle_aborted(handle))
 309                 return;
 310
 311         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 312                caller, line, errstr, err_fn);
 313
 314         jbd2_journal_abort_handle(handle);
 315 }
 316
 317 static void __save_error_info(struct super_block *sb, const char *func,
 318                             unsigned int line)
 319 {
 320         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 321
 322         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 324         es->s_last_error_time = cpu_to_le32(get_seconds());
 325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 326         es->s_last_error_line = cpu_to_le32(line);
 327         if (!es->s_first_error_time) {
 328                 es->s_first_error_time = es->s_last_error_time;
 329                 strncpy(es->s_first_error_func, func,
 330                         sizeof(es->s_first_error_func));
 331                 es->s_first_error_line = cpu_to_le32(line);
 332                 es->s_first_error_ino = es->s_last_error_ino;
 333                 es->s_first_error_block = es->s_last_error_block;
 334         }
 335         /*
 336          * Start the daily error reporting function if it hasn't been
 337          * started already
 338          */
 339         if (!es->s_error_count)
 340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 341         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 342 }
 343
 344 static void save_error_info(struct super_block *sb, const char *func,
 345                             unsigned int line)
 346 {
 347         __save_error_info(sb, func, line);
 348         ext4_commit_super(sb, 1);
 349 }
 350
 351
 352 /* Deal with the reporting of failure conditions on a filesystem such as
 353  * inconsistencies detected or read IO failures.
 354  *
 355  * On ext2, we can store the error state of the filesystem in the
 356  * superblock.  That is not possible on ext4, because we may have other
 357  * write ordering constraints on the superblock which prevent us from
 358  * writing it out straight away; and given that the journal is about to
 359  * be aborted, we can't rely on the current, or future, transactions to
 360  * write out the superblock safely.
 361  *
 362  * We'll just use the jbd2_journal_abort() error code to record an error in
 363  * the journal instead.  On recovery, the journal will complain about
 364  * that error until we've noted it down and cleared it.
 365  */
 366
 367 static void ext4_handle_error(struct super_block *sb)
 368 {
 369         if (sb->s_flags & MS_RDONLY)
 370                 return;
 371
 372         if (!test_opt(sb, ERRORS_CONT)) {
 373                 journal_t *journal = EXT4_SB(sb)->s_journal;
 374
 375                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 376                 if (journal)
 377                         jbd2_journal_abort(journal, -EIO);
 378         }
 379         if (test_opt(sb, ERRORS_RO)) {
 380                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 381                 sb->s_flags |= MS_RDONLY;
 382         }
 383         if (test_opt(sb, ERRORS_PANIC))
 384                 panic("EXT4-fs (device %s): panic forced after error\n",
 385                         sb->s_id);
 386 }
 387
 388 void __ext4_error(struct super_block *sb, const char *function,
 389                   unsigned int line, const char *fmt, ...)
 390 {
 391         va_list args;
 392
 393         va_start(args, fmt);
 394         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: ",
 395                sb->s_id, function, line, current->comm);
 396         vprintk(fmt, args);
 397         printk("\n");
 398         va_end(args);
 399
 400         ext4_handle_error(sb);
 401 }
 402
 403 void ext4_error_inode(struct inode *inode, const char *function,
 404                       unsigned int line, ext4_fsblk_t block,
 405                       const char *fmt, ...)
 406 {
 407         va_list args;
 408         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 409
 410         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 411         es->s_last_error_block = cpu_to_le64(block);
 412         save_error_info(inode->i_sb, function, line);
 413         va_start(args, fmt);
 414         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 415                inode->i_sb->s_id, function, line, inode->i_ino);
 416         if (block)
 417                 printk("block %llu: ", block);
 418         printk("comm %s: ", current->comm);
 419         vprintk(fmt, args);
 420         printk("\n");
 421         va_end(args);
 422
 423         ext4_handle_error(inode->i_sb);
 424 }
 425
 426 void ext4_error_file(struct file *file, const char *function,
 427                      unsigned int line, const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct ext4_super_block *es;
 431         struct inode *inode = file->f_dentry->d_inode;
 432         char pathname[80], *path;
 433
 434         es = EXT4_SB(inode->i_sb)->s_es;
 435         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 436         save_error_info(inode->i_sb, function, line);
 437         va_start(args, fmt);
 438         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 439         if (!path)
 440                 path = "(unknown)";
 441         printk(KERN_CRIT
 442                "EXT4-fs error (device %s): %s:%d: inode #%lu "
 443                "(comm %s path %s): ",
 444                inode->i_sb->s_id, function, line, inode->i_ino,
 445                current->comm, path);
 446         vprintk(fmt, args);
 447         printk("\n");
 448         va_end(args);
 449
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 static const char *ext4_decode_error(struct super_block *sb, int errno,
 454                                      char nbuf[16])
 455 {
 456         char *errstr = NULL;
 457
 458         switch (errno) {
 459         case -EIO:
 460                 errstr = "IO failure";
 461                 break;
 462         case -ENOMEM:
 463                 errstr = "Out of memory";
 464                 break;
 465         case -EROFS:
 466                 if (!sb || (EXT4_SB(sb)->s_journal &&
 467                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 468                         errstr = "Journal has aborted";
 469                 else
 470                         errstr = "Readonly filesystem";
 471                 break;
 472         default:
 473                 /* If the caller passed in an extra buffer for unknown
 474                  * errors, textualise them now.  Else we just return
 475                  * NULL. */
 476                 if (nbuf) {
 477                         /* Check for truncated error codes... */
 478                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 479                                 errstr = nbuf;
 480                 }
 481                 break;
 482         }
 483
 484         return errstr;
 485 }
 486
 487 /* __ext4_std_error decodes expected errors from journaling functions
 488  * automatically and invokes the appropriate error response.  */
 489
 490 void __ext4_std_error(struct super_block *sb, const char *function,
 491                       unsigned int line, int errno)
 492 {
 493         char nbuf[16];
 494         const char *errstr;
 495
 496         /* Special case: if the error is EROFS, and we're not already
 497          * inside a transaction, then there's really no point in logging
 498          * an error. */
 499         if (errno == -EROFS && journal_current_handle() == NULL &&
 500             (sb->s_flags & MS_RDONLY))
 501                 return;
 502
 503         errstr = ext4_decode_error(sb, errno, nbuf);
 504         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 505                sb->s_id, function, line, errstr);
 506         save_error_info(sb, function, line);
 507
 508         ext4_handle_error(sb);
 509 }
 510
 511 /*
 512  * ext4_abort is a much stronger failure handler than ext4_error.  The
 513  * abort function may be used to deal with unrecoverable failures such
 514  * as journal IO errors or ENOMEM at a critical moment in log management.
 515  *
 516  * We unconditionally force the filesystem into an ABORT|READONLY state,
 517  * unless the error response on the fs has been set to panic in which
 518  * case we take the easy way out and panic immediately.
 519  */
 520
 521 void __ext4_abort(struct super_block *sb, const char *function,
 522                 unsigned int line, const char *fmt, ...)
 523 {
 524         va_list args;
 525
 526         save_error_info(sb, function, line);
 527         va_start(args, fmt);
 528         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 529                function, line);
 530         vprintk(fmt, args);
 531         printk("\n");
 532         va_end(args);
 533
 534         if ((sb->s_flags & MS_RDONLY) == 0) {
 535                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 536                 sb->s_flags |= MS_RDONLY;
 537                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 538                 if (EXT4_SB(sb)->s_journal)
 539                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 540                 save_error_info(sb, function, line);
 541         }
 542         if (test_opt(sb, ERRORS_PANIC))
 543                 panic("EXT4-fs panic from previous error\n");
 544 }
 545
 546 void ext4_msg (struct super_block * sb, const char *prefix,
 547                    const char *fmt, ...)
 548 {
 549         va_list args;
 550
 551         va_start(args, fmt);
 552         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 553         vprintk(fmt, args);
 554         printk("\n");
 555         va_end(args);
 556 }
 557
 558 void __ext4_warning(struct super_block *sb, const char *function,
 559                     unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         va_start(args, fmt);
 564         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: ",
 565                sb->s_id, function, line);
 566         vprintk(fmt, args);
 567         printk("\n");
 568         va_end(args);
 569 }
 570
 571 void __ext4_grp_locked_error(const char *function, unsigned int line,
 572                              struct super_block *sb, ext4_group_t grp,
 573                              unsigned long ino, ext4_fsblk_t block,
 574                              const char *fmt, ...)
 575 __releases(bitlock)
 576 __acquires(bitlock)
 577 {
 578         va_list args;
 579         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 580
 581         es->s_last_error_ino = cpu_to_le32(ino);
 582         es->s_last_error_block = cpu_to_le64(block);
 583         __save_error_info(sb, function, line);
 584         va_start(args, fmt);
 585         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u",
 586                sb->s_id, function, line, grp);
 587         if (ino)
 588                 printk("inode %lu: ", ino);
 589         if (block)
 590                 printk("block %llu:", (unsigned long long) block);
 591         vprintk(fmt, args);
 592         printk("\n");
 593         va_end(args);
 594
 595         if (test_opt(sb, ERRORS_CONT)) {
 596                 ext4_commit_super(sb, 0);
 597                 return;
 598         }
 599
 600         ext4_unlock_group(sb, grp);
 601         ext4_handle_error(sb);
 602         /*
 603          * We only get here in the ERRORS_RO case; relocking the group
 604          * may be dangerous, but nothing bad will happen since the
 605          * filesystem will have already been marked read/only and the
 606          * journal has been aborted.  We return 1 as a hint to callers
 607          * who might what to use the return value from
 608          * ext4_grp_locked_error() to distinguish beween the
 609          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 610          * aggressively from the ext4 function in question, with a
 611          * more appropriate error code.
 612          */
 613         ext4_lock_group(sb, grp);
 614         return;
 615 }
 616
 617 void ext4_update_dynamic_rev(struct super_block *sb)
 618 {
 619         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 620
 621         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 622                 return;
 623
 624         ext4_warning(sb,
 625                      "updating to rev %d because of new feature flag, "
 626                      "running e2fsck is recommended",
 627                      EXT4_DYNAMIC_REV);
 628
 629         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 630         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 631         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 632         /* leave es->s_feature_*compat flags alone */
 633         /* es->s_uuid will be set by e2fsck if empty */
 634
 635         /*
 636          * The rest of the superblock fields should be zero, and if not it
 637          * means they are likely already in use, so leave them alone.  We
 638          * can leave it up to e2fsck to clean up any inconsistencies there.
 639          */
 640 }
 641
 642 /*
 643  * Open the external journal device
 644  */
 645 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 646 {
 647         struct block_device *bdev;
 648         char b[BDEVNAME_SIZE];
 649
 650         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 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
 832         return &ei->vfs_inode;
 833 }
 834
 835 static void ext4_destroy_inode(struct inode *inode)
 836 {
 837         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 838                 ext4_msg(inode->i_sb, KERN_ERR,
 839                          "Inode %lu (%p): orphan list check failed!",
 840                          inode->i_ino, EXT4_I(inode));
 841                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 842                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 843                                 true);
 844                 dump_stack();
 845         }
 846         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 847 }
 848
 849 static void init_once(void *foo)
 850 {
 851         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 852
 853         INIT_LIST_HEAD(&ei->i_orphan);
 854 #ifdef CONFIG_EXT4_FS_XATTR
 855         init_rwsem(&ei->xattr_sem);
 856 #endif
 857         init_rwsem(&ei->i_data_sem);
 858         inode_init_once(&ei->vfs_inode);
 859 }
 860
 861 static int init_inodecache(void)
 862 {
 863         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 864                                              sizeof(struct ext4_inode_info),
 865                                              0, (SLAB_RECLAIM_ACCOUNT|
 866                                                 SLAB_MEM_SPREAD),
 867                                              init_once);
 868         if (ext4_inode_cachep == NULL)
 869                 return -ENOMEM;
 870         return 0;
 871 }
 872
 873 static void destroy_inodecache(void)
 874 {
 875         kmem_cache_destroy(ext4_inode_cachep);
 876 }
 877
 878 void ext4_clear_inode(struct inode *inode)
 879 {
 880         invalidate_inode_buffers(inode);
 881         end_writeback(inode);
 882         dquot_drop(inode);
 883         ext4_discard_preallocations(inode);
 884         if (EXT4_JOURNAL(inode))
 885                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 886                                        &EXT4_I(inode)->jinode);
 887 }
 888
 889 static inline void ext4_show_quota_options(struct seq_file *seq,
 890                                            struct super_block *sb)
 891 {
 892 #if defined(CONFIG_QUOTA)
 893         struct ext4_sb_info *sbi = EXT4_SB(sb);
 894
 895         if (sbi->s_jquota_fmt) {
 896                 char *fmtname = "";
 897
 898                 switch (sbi->s_jquota_fmt) {
 899                 case QFMT_VFS_OLD:
 900                         fmtname = "vfsold";
 901                         break;
 902                 case QFMT_VFS_V0:
 903                         fmtname = "vfsv0";
 904                         break;
 905                 case QFMT_VFS_V1:
 906                         fmtname = "vfsv1";
 907                         break;
 908                 }
 909                 seq_printf(seq, ",jqfmt=%s", fmtname);
 910         }
 911
 912         if (sbi->s_qf_names[USRQUOTA])
 913                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 914
 915         if (sbi->s_qf_names[GRPQUOTA])
 916                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 917
 918         if (test_opt(sb, USRQUOTA))
 919                 seq_puts(seq, ",usrquota");
 920
 921         if (test_opt(sb, GRPQUOTA))
 922                 seq_puts(seq, ",grpquota");
 923 #endif
 924 }
 925
 926 /*
 927  * Show an option if
 928  *  - it's set to a non-default value OR
 929  *  - if the per-sb default is different from the global default
 930  */
 931 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 932 {
 933         int def_errors;
 934         unsigned long def_mount_opts;
 935         struct super_block *sb = vfs->mnt_sb;
 936         struct ext4_sb_info *sbi = EXT4_SB(sb);
 937         struct ext4_super_block *es = sbi->s_es;
 938
 939         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 940         def_errors     = le16_to_cpu(es->s_errors);
 941
 942         if (sbi->s_sb_block != 1)
 943                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 944         if (test_opt(sb, MINIX_DF))
 945                 seq_puts(seq, ",minixdf");
 946         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 947                 seq_puts(seq, ",grpid");
 948         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 949                 seq_puts(seq, ",nogrpid");
 950         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 951             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 952                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 953         }
 954         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 955             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 956                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 957         }
 958         if (test_opt(sb, ERRORS_RO)) {
 959                 if (def_errors == EXT4_ERRORS_PANIC ||
 960                     def_errors == EXT4_ERRORS_CONTINUE) {
 961                         seq_puts(seq, ",errors=remount-ro");
 962                 }
 963         }
 964         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 965                 seq_puts(seq, ",errors=continue");
 966         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 967                 seq_puts(seq, ",errors=panic");
 968         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 969                 seq_puts(seq, ",nouid32");
 970         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 971                 seq_puts(seq, ",debug");
 972         if (test_opt(sb, OLDALLOC))
 973                 seq_puts(seq, ",oldalloc");
 974 #ifdef CONFIG_EXT4_FS_XATTR
 975         if (test_opt(sb, XATTR_USER) &&
 976                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 977                 seq_puts(seq, ",user_xattr");
 978         if (!test_opt(sb, XATTR_USER) &&
 979             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 980                 seq_puts(seq, ",nouser_xattr");
 981         }
 982 #endif
 983 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 984         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 985                 seq_puts(seq, ",acl");
 986         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 987                 seq_puts(seq, ",noacl");
 988 #endif
 989         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 990                 seq_printf(seq, ",commit=%u",
 991                            (unsigned) (sbi->s_commit_interval / HZ));
 992         }
 993         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 994                 seq_printf(seq, ",min_batch_time=%u",
 995                            (unsigned) sbi->s_min_batch_time);
 996         }
 997         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 998                 seq_printf(seq, ",max_batch_time=%u",
 999                            (unsigned) sbi->s_min_batch_time);
1000         }
1001
1002         /*
1003          * We're changing the default of barrier mount option, so
1004          * let's always display its mount state so it's clear what its
1005          * status is.
1006          */
1007         seq_puts(seq, ",barrier=");
1008         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1009         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1010                 seq_puts(seq, ",journal_async_commit");
1011         else if (test_opt(sb, JOURNAL_CHECKSUM))
1012                 seq_puts(seq, ",journal_checksum");
1013         if (test_opt(sb, I_VERSION))
1014                 seq_puts(seq, ",i_version");
1015         if (!test_opt(sb, DELALLOC) &&
1016             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1017                 seq_puts(seq, ",nodelalloc");
1018
1019         if (sbi->s_stripe)
1020                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1021         /*
1022          * journal mode get enabled in different ways
1023          * So just print the value even if we didn't specify it
1024          */
1025         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1026                 seq_puts(seq, ",data=journal");
1027         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1028                 seq_puts(seq, ",data=ordered");
1029         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1030                 seq_puts(seq, ",data=writeback");
1031
1032         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1033                 seq_printf(seq, ",inode_readahead_blks=%u",
1034                            sbi->s_inode_readahead_blks);
1035
1036         if (test_opt(sb, DATA_ERR_ABORT))
1037                 seq_puts(seq, ",data_err=abort");
1038
1039         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1040                 seq_puts(seq, ",noauto_da_alloc");
1041
1042         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1043                 seq_puts(seq, ",discard");
1044
1045         if (test_opt(sb, NOLOAD))
1046                 seq_puts(seq, ",norecovery");
1047
1048         if (test_opt(sb, DIOREAD_NOLOCK))
1049                 seq_puts(seq, ",dioread_nolock");
1050
1051         if (test_opt(sb, BLOCK_VALIDITY) &&
1052             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1053                 seq_puts(seq, ",block_validity");
1054
1055         if (!test_opt(sb, INIT_INODE_TABLE))
1056                 seq_puts(seq, ",noinit_inode_table");
1057         else if (sbi->s_li_wait_mult)
1058                 seq_printf(seq, ",init_inode_table=%u",
1059                            (unsigned) sbi->s_li_wait_mult);
1060
1061         ext4_show_quota_options(seq, sb);
1062
1063         return 0;
1064 }
1065
1066 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1067                                         u64 ino, u32 generation)
1068 {
1069         struct inode *inode;
1070
1071         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1072                 return ERR_PTR(-ESTALE);
1073         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1074                 return ERR_PTR(-ESTALE);
1075
1076         /* iget isn't really right if the inode is currently unallocated!!
1077          *
1078          * ext4_read_inode will return a bad_inode if the inode had been
1079          * deleted, so we should be safe.
1080          *
1081          * Currently we don't know the generation for parent directory, so
1082          * a generation of 0 means "accept any"
1083          */
1084         inode = ext4_iget(sb, ino);
1085         if (IS_ERR(inode))
1086                 return ERR_CAST(inode);
1087         if (generation && inode->i_generation != generation) {
1088                 iput(inode);
1089                 return ERR_PTR(-ESTALE);
1090         }
1091
1092         return inode;
1093 }
1094
1095 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1096                                         int fh_len, int fh_type)
1097 {
1098         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1099                                     ext4_nfs_get_inode);
1100 }
1101
1102 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1103                                         int fh_len, int fh_type)
1104 {
1105         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1106                                     ext4_nfs_get_inode);
1107 }
1108
1109 /*
1110  * Try to release metadata pages (indirect blocks, directories) which are
1111  * mapped via the block device.  Since these pages could have journal heads
1112  * which would prevent try_to_free_buffers() from freeing them, we must use
1113  * jbd2 layer's try_to_free_buffers() function to release them.
1114  */
1115 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1116                                  gfp_t wait)
1117 {
1118         journal_t *journal = EXT4_SB(sb)->s_journal;
1119
1120         WARN_ON(PageChecked(page));
1121         if (!page_has_buffers(page))
1122                 return 0;
1123         if (journal)
1124                 return jbd2_journal_try_to_free_buffers(journal, page,
1125                                                         wait & ~__GFP_WAIT);
1126         return try_to_free_buffers(page);
1127 }
1128
1129 #ifdef CONFIG_QUOTA
1130 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1131 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1132
1133 static int ext4_write_dquot(struct dquot *dquot);
1134 static int ext4_acquire_dquot(struct dquot *dquot);
1135 static int ext4_release_dquot(struct dquot *dquot);
1136 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1137 static int ext4_write_info(struct super_block *sb, int type);
1138 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1139                                 char *path);
1140 static int ext4_quota_off(struct super_block *sb, int type);
1141 static int ext4_quota_on_mount(struct super_block *sb, int type);
1142 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1143                                size_t len, loff_t off);
1144 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1145                                 const char *data, size_t len, loff_t off);
1146
1147 static const struct dquot_operations ext4_quota_operations = {
1148 #ifdef CONFIG_QUOTA
1149         .get_reserved_space = ext4_get_reserved_space,
1150 #endif
1151         .write_dquot    = ext4_write_dquot,
1152         .acquire_dquot  = ext4_acquire_dquot,
1153         .release_dquot  = ext4_release_dquot,
1154         .mark_dirty     = ext4_mark_dquot_dirty,
1155         .write_info     = ext4_write_info,
1156         .alloc_dquot    = dquot_alloc,
1157         .destroy_dquot  = dquot_destroy,
1158 };
1159
1160 static const struct quotactl_ops ext4_qctl_operations = {
1161         .quota_on       = ext4_quota_on,
1162         .quota_off      = ext4_quota_off,
1163         .quota_sync     = dquot_quota_sync,
1164         .get_info       = dquot_get_dqinfo,
1165         .set_info       = dquot_set_dqinfo,
1166         .get_dqblk      = dquot_get_dqblk,
1167         .set_dqblk      = dquot_set_dqblk
1168 };
1169 #endif
1170
1171 static const struct super_operations ext4_sops = {
1172         .alloc_inode    = ext4_alloc_inode,
1173         .destroy_inode  = ext4_destroy_inode,
1174         .write_inode    = ext4_write_inode,
1175         .dirty_inode    = ext4_dirty_inode,
1176         .evict_inode    = ext4_evict_inode,
1177         .put_super      = ext4_put_super,
1178         .sync_fs        = ext4_sync_fs,
1179         .freeze_fs      = ext4_freeze,
1180         .unfreeze_fs    = ext4_unfreeze,
1181         .statfs         = ext4_statfs,
1182         .remount_fs     = ext4_remount,
1183         .show_options   = ext4_show_options,
1184 #ifdef CONFIG_QUOTA
1185         .quota_read     = ext4_quota_read,
1186         .quota_write    = ext4_quota_write,
1187 #endif
1188         .bdev_try_to_free_page = bdev_try_to_free_page,
1189         .trim_fs        = ext4_trim_fs
1190 };
1191
1192 static const struct super_operations ext4_nojournal_sops = {
1193         .alloc_inode    = ext4_alloc_inode,
1194         .destroy_inode  = ext4_destroy_inode,
1195         .write_inode    = ext4_write_inode,
1196         .dirty_inode    = ext4_dirty_inode,
1197         .evict_inode    = ext4_evict_inode,
1198         .write_super    = ext4_write_super,
1199         .put_super      = ext4_put_super,
1200         .statfs         = ext4_statfs,
1201         .remount_fs     = ext4_remount,
1202         .show_options   = ext4_show_options,
1203 #ifdef CONFIG_QUOTA
1204         .quota_read     = ext4_quota_read,
1205         .quota_write    = ext4_quota_write,
1206 #endif
1207         .bdev_try_to_free_page = bdev_try_to_free_page,
1208 };
1209
1210 static const struct export_operations ext4_export_ops = {
1211         .fh_to_dentry = ext4_fh_to_dentry,
1212         .fh_to_parent = ext4_fh_to_parent,
1213         .get_parent = ext4_get_parent,
1214 };
1215
1216 enum {
1217         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1218         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1219         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1220         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1221         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1222         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1223         Opt_journal_update, Opt_journal_dev,
1224         Opt_journal_checksum, Opt_journal_async_commit,
1225         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1226         Opt_data_err_abort, Opt_data_err_ignore,
1227         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1228         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1229         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1230         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1231         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1232         Opt_block_validity, Opt_noblock_validity,
1233         Opt_inode_readahead_blks, Opt_journal_ioprio,
1234         Opt_dioread_nolock, Opt_dioread_lock,
1235         Opt_discard, Opt_nodiscard,
1236         Opt_init_inode_table, Opt_noinit_inode_table,
1237 };
1238
1239 static const match_table_t tokens = {
1240         {Opt_bsd_df, "bsddf"},
1241         {Opt_minix_df, "minixdf"},
1242         {Opt_grpid, "grpid"},
1243         {Opt_grpid, "bsdgroups"},
1244         {Opt_nogrpid, "nogrpid"},
1245         {Opt_nogrpid, "sysvgroups"},
1246         {Opt_resgid, "resgid=%u"},
1247         {Opt_resuid, "resuid=%u"},
1248         {Opt_sb, "sb=%u"},
1249         {Opt_err_cont, "errors=continue"},
1250         {Opt_err_panic, "errors=panic"},
1251         {Opt_err_ro, "errors=remount-ro"},
1252         {Opt_nouid32, "nouid32"},
1253         {Opt_debug, "debug"},
1254         {Opt_oldalloc, "oldalloc"},
1255         {Opt_orlov, "orlov"},
1256         {Opt_user_xattr, "user_xattr"},
1257         {Opt_nouser_xattr, "nouser_xattr"},
1258         {Opt_acl, "acl"},
1259         {Opt_noacl, "noacl"},
1260         {Opt_noload, "noload"},
1261         {Opt_noload, "norecovery"},
1262         {Opt_nobh, "nobh"},
1263         {Opt_bh, "bh"},
1264         {Opt_commit, "commit=%u"},
1265         {Opt_min_batch_time, "min_batch_time=%u"},
1266         {Opt_max_batch_time, "max_batch_time=%u"},
1267         {Opt_journal_update, "journal=update"},
1268         {Opt_journal_dev, "journal_dev=%u"},
1269         {Opt_journal_checksum, "journal_checksum"},
1270         {Opt_journal_async_commit, "journal_async_commit"},
1271         {Opt_abort, "abort"},
1272         {Opt_data_journal, "data=journal"},
1273         {Opt_data_ordered, "data=ordered"},
1274         {Opt_data_writeback, "data=writeback"},
1275         {Opt_data_err_abort, "data_err=abort"},
1276         {Opt_data_err_ignore, "data_err=ignore"},
1277         {Opt_offusrjquota, "usrjquota="},
1278         {Opt_usrjquota, "usrjquota=%s"},
1279         {Opt_offgrpjquota, "grpjquota="},
1280         {Opt_grpjquota, "grpjquota=%s"},
1281         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1282         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1283         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1284         {Opt_grpquota, "grpquota"},
1285         {Opt_noquota, "noquota"},
1286         {Opt_quota, "quota"},
1287         {Opt_usrquota, "usrquota"},
1288         {Opt_barrier, "barrier=%u"},
1289         {Opt_barrier, "barrier"},
1290         {Opt_nobarrier, "nobarrier"},
1291         {Opt_i_version, "i_version"},
1292         {Opt_stripe, "stripe=%u"},
1293         {Opt_resize, "resize"},
1294         {Opt_delalloc, "delalloc"},
1295         {Opt_nodelalloc, "nodelalloc"},
1296         {Opt_block_validity, "block_validity"},
1297         {Opt_noblock_validity, "noblock_validity"},
1298         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1299         {Opt_journal_ioprio, "journal_ioprio=%u"},
1300         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1301         {Opt_auto_da_alloc, "auto_da_alloc"},
1302         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1303         {Opt_dioread_nolock, "dioread_nolock"},
1304         {Opt_dioread_lock, "dioread_lock"},
1305         {Opt_discard, "discard"},
1306         {Opt_nodiscard, "nodiscard"},
1307         {Opt_init_inode_table, "init_itable=%u"},
1308         {Opt_init_inode_table, "init_itable"},
1309         {Opt_noinit_inode_table, "noinit_itable"},
1310         {Opt_err, NULL},
1311 };
1312
1313 static ext4_fsblk_t get_sb_block(void **data)
1314 {
1315         ext4_fsblk_t    sb_block;
1316         char            *options = (char *) *data;
1317
1318         if (!options || strncmp(options, "sb=", 3) != 0)
1319                 return 1;       /* Default location */
1320
1321         options += 3;
1322         /* TODO: use simple_strtoll with >32bit ext4 */
1323         sb_block = simple_strtoul(options, &options, 0);
1324         if (*options && *options != ',') {
1325                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1326                        (char *) *data);
1327                 return 1;
1328         }
1329         if (*options == ',')
1330                 options++;
1331         *data = (void *) options;
1332
1333         return sb_block;
1334 }
1335
1336 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1337 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1338         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1339
1340 #ifdef CONFIG_QUOTA
1341 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1342 {
1343         struct ext4_sb_info *sbi = EXT4_SB(sb);
1344         char *qname;
1345
1346         if (sb_any_quota_loaded(sb) &&
1347                 !sbi->s_qf_names[qtype]) {
1348                 ext4_msg(sb, KERN_ERR,
1349                         "Cannot change journaled "
1350                         "quota options when quota turned on");
1351                 return 0;
1352         }
1353         qname = match_strdup(args);
1354         if (!qname) {
1355                 ext4_msg(sb, KERN_ERR,
1356                         "Not enough memory for storing quotafile name");
1357                 return 0;
1358         }
1359         if (sbi->s_qf_names[qtype] &&
1360                 strcmp(sbi->s_qf_names[qtype], qname)) {
1361                 ext4_msg(sb, KERN_ERR,
1362                         "%s quota file already specified", QTYPE2NAME(qtype));
1363                 kfree(qname);
1364                 return 0;
1365         }
1366         sbi->s_qf_names[qtype] = qname;
1367         if (strchr(sbi->s_qf_names[qtype], '/')) {
1368                 ext4_msg(sb, KERN_ERR,
1369                         "quotafile must be on filesystem root");
1370                 kfree(sbi->s_qf_names[qtype]);
1371                 sbi->s_qf_names[qtype] = NULL;
1372                 return 0;
1373         }
1374         set_opt(sbi->s_mount_opt, QUOTA);
1375         return 1;
1376 }
1377
1378 static int clear_qf_name(struct super_block *sb, int qtype)
1379 {
1380
1381         struct ext4_sb_info *sbi = EXT4_SB(sb);
1382
1383         if (sb_any_quota_loaded(sb) &&
1384                 sbi->s_qf_names[qtype]) {
1385                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1386                         " when quota turned on");
1387                 return 0;
1388         }
1389         /*
1390          * The space will be released later when all options are confirmed
1391          * to be correct
1392          */
1393         sbi->s_qf_names[qtype] = NULL;
1394         return 1;
1395 }
1396 #endif
1397
1398 static int parse_options(char *options, struct super_block *sb,
1399                          unsigned long *journal_devnum,
1400                          unsigned int *journal_ioprio,
1401                          ext4_fsblk_t *n_blocks_count, int is_remount)
1402 {
1403         struct ext4_sb_info *sbi = EXT4_SB(sb);
1404         char *p;
1405         substring_t args[MAX_OPT_ARGS];
1406         int data_opt = 0;
1407         int option;
1408 #ifdef CONFIG_QUOTA
1409         int qfmt;
1410 #endif
1411
1412         if (!options)
1413                 return 1;
1414
1415         while ((p = strsep(&options, ",")) != NULL) {
1416                 int token;
1417                 if (!*p)
1418                         continue;
1419
1420                 /*
1421                  * Initialize args struct so we know whether arg was
1422                  * found; some options take optional arguments.
1423                  */
1424                 args[0].to = args[0].from = 0;
1425                 token = match_token(p, tokens, args);
1426                 switch (token) {
1427                 case Opt_bsd_df:
1428                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1429                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1430                         break;
1431                 case Opt_minix_df:
1432                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1433                         set_opt(sbi->s_mount_opt, MINIX_DF);
1434
1435                         break;
1436                 case Opt_grpid:
1437                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1438                         set_opt(sbi->s_mount_opt, GRPID);
1439
1440                         break;
1441                 case Opt_nogrpid:
1442                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1443                         clear_opt(sbi->s_mount_opt, GRPID);
1444
1445                         break;
1446                 case Opt_resuid:
1447                         if (match_int(&args[0], &option))
1448                                 return 0;
1449                         sbi->s_resuid = option;
1450                         break;
1451                 case Opt_resgid:
1452                         if (match_int(&args[0], &option))
1453                                 return 0;
1454                         sbi->s_resgid = option;
1455                         break;
1456                 case Opt_sb:
1457                         /* handled by get_sb_block() instead of here */
1458                         /* *sb_block = match_int(&args[0]); */
1459                         break;
1460                 case Opt_err_panic:
1461                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1462                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1463                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1464                         break;
1465                 case Opt_err_ro:
1466                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1467                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1468                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1469                         break;
1470                 case Opt_err_cont:
1471                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1472                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1473                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1474                         break;
1475                 case Opt_nouid32:
1476                         set_opt(sbi->s_mount_opt, NO_UID32);
1477                         break;
1478                 case Opt_debug:
1479                         set_opt(sbi->s_mount_opt, DEBUG);
1480                         break;
1481                 case Opt_oldalloc:
1482                         set_opt(sbi->s_mount_opt, OLDALLOC);
1483                         break;
1484                 case Opt_orlov:
1485                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1486                         break;
1487 #ifdef CONFIG_EXT4_FS_XATTR
1488                 case Opt_user_xattr:
1489                         set_opt(sbi->s_mount_opt, XATTR_USER);
1490                         break;
1491                 case Opt_nouser_xattr:
1492                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1493                         break;
1494 #else
1495                 case Opt_user_xattr:
1496                 case Opt_nouser_xattr:
1497                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1498                         break;
1499 #endif
1500 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1501                 case Opt_acl:
1502                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1503                         break;
1504                 case Opt_noacl:
1505                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1506                         break;
1507 #else
1508                 case Opt_acl:
1509                 case Opt_noacl:
1510                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1511                         break;
1512 #endif
1513                 case Opt_journal_update:
1514                         /* @@@ FIXME */
1515                         /* Eventually we will want to be able to create
1516                            a journal file here.  For now, only allow the
1517                            user to specify an existing inode to be the
1518                            journal file. */
1519                         if (is_remount) {
1520                                 ext4_msg(sb, KERN_ERR,
1521                                          "Cannot specify journal on remount");
1522                                 return 0;
1523                         }
1524                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1525                         break;
1526                 case Opt_journal_dev:
1527                         if (is_remount) {
1528                                 ext4_msg(sb, KERN_ERR,
1529                                         "Cannot specify journal on remount");
1530                                 return 0;
1531                         }
1532                         if (match_int(&args[0], &option))
1533                                 return 0;
1534                         *journal_devnum = option;
1535                         break;
1536                 case Opt_journal_checksum:
1537                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1538                         break;
1539                 case Opt_journal_async_commit:
1540                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1541                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1542                         break;
1543                 case Opt_noload:
1544                         set_opt(sbi->s_mount_opt, NOLOAD);
1545                         break;
1546                 case Opt_commit:
1547                         if (match_int(&args[0], &option))
1548                                 return 0;
1549                         if (option < 0)
1550                                 return 0;
1551                         if (option == 0)
1552                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1553                         sbi->s_commit_interval = HZ * option;
1554                         break;
1555                 case Opt_max_batch_time:
1556                         if (match_int(&args[0], &option))
1557                                 return 0;
1558                         if (option < 0)
1559                                 return 0;
1560                         if (option == 0)
1561                                 option = EXT4_DEF_MAX_BATCH_TIME;
1562                         sbi->s_max_batch_time = option;
1563                         break;
1564                 case Opt_min_batch_time:
1565                         if (match_int(&args[0], &option))
1566                                 return 0;
1567                         if (option < 0)
1568                                 return 0;
1569                         sbi->s_min_batch_time = option;
1570                         break;
1571                 case Opt_data_journal:
1572                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1573                         goto datacheck;
1574                 case Opt_data_ordered:
1575                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1576                         goto datacheck;
1577                 case Opt_data_writeback:
1578                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1579                 datacheck:
1580                         if (is_remount) {
1581                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1582                                         ext4_msg(sb, KERN_ERR,
1583                                                 "Cannot change data mode on remount");
1584                                         return 0;
1585                                 }
1586                         } else {
1587                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1588                                 sbi->s_mount_opt |= data_opt;
1589                         }
1590                         break;
1591                 case Opt_data_err_abort:
1592                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1593                         break;
1594                 case Opt_data_err_ignore:
1595                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1596                         break;
1597 #ifdef CONFIG_QUOTA
1598                 case Opt_usrjquota:
1599                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1600                                 return 0;
1601                         break;
1602                 case Opt_grpjquota:
1603                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1604                                 return 0;
1605                         break;
1606                 case Opt_offusrjquota:
1607                         if (!clear_qf_name(sb, USRQUOTA))
1608                                 return 0;
1609                         break;
1610                 case Opt_offgrpjquota:
1611                         if (!clear_qf_name(sb, GRPQUOTA))
1612                                 return 0;
1613                         break;
1614
1615                 case Opt_jqfmt_vfsold:
1616                         qfmt = QFMT_VFS_OLD;
1617                         goto set_qf_format;
1618                 case Opt_jqfmt_vfsv0:
1619                         qfmt = QFMT_VFS_V0;
1620                         goto set_qf_format;
1621                 case Opt_jqfmt_vfsv1:
1622                         qfmt = QFMT_VFS_V1;
1623 set_qf_format:
1624                         if (sb_any_quota_loaded(sb) &&
1625                             sbi->s_jquota_fmt != qfmt) {
1626                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1627                                         "journaled quota options when "
1628                                         "quota turned on");
1629                                 return 0;
1630                         }
1631                         sbi->s_jquota_fmt = qfmt;
1632                         break;
1633                 case Opt_quota:
1634                 case Opt_usrquota:
1635                         set_opt(sbi->s_mount_opt, QUOTA);
1636                         set_opt(sbi->s_mount_opt, USRQUOTA);
1637                         break;
1638                 case Opt_grpquota:
1639                         set_opt(sbi->s_mount_opt, QUOTA);
1640                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1641                         break;
1642                 case Opt_noquota:
1643                         if (sb_any_quota_loaded(sb)) {
1644                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1645                                         "options when quota turned on");
1646                                 return 0;
1647                         }
1648                         clear_opt(sbi->s_mount_opt, QUOTA);
1649                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1650                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1651                         break;
1652 #else
1653                 case Opt_quota:
1654                 case Opt_usrquota:
1655                 case Opt_grpquota:
1656                         ext4_msg(sb, KERN_ERR,
1657                                 "quota options not supported");
1658                         break;
1659                 case Opt_usrjquota:
1660                 case Opt_grpjquota:
1661                 case Opt_offusrjquota:
1662                 case Opt_offgrpjquota:
1663                 case Opt_jqfmt_vfsold:
1664                 case Opt_jqfmt_vfsv0:
1665                 case Opt_jqfmt_vfsv1:
1666                         ext4_msg(sb, KERN_ERR,
1667                                 "journaled quota options not supported");
1668                         break;
1669                 case Opt_noquota:
1670                         break;
1671 #endif
1672                 case Opt_abort:
1673                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1674                         break;
1675                 case Opt_nobarrier:
1676                         clear_opt(sbi->s_mount_opt, BARRIER);
1677                         break;
1678                 case Opt_barrier:
1679                         if (args[0].from) {
1680                                 if (match_int(&args[0], &option))
1681                                         return 0;
1682                         } else
1683                                 option = 1;     /* No argument, default to 1 */
1684                         if (option)
1685                                 set_opt(sbi->s_mount_opt, BARRIER);
1686                         else
1687                                 clear_opt(sbi->s_mount_opt, BARRIER);
1688                         break;
1689                 case Opt_ignore:
1690                         break;
1691                 case Opt_resize:
1692                         if (!is_remount) {
1693                                 ext4_msg(sb, KERN_ERR,
1694                                         "resize option only available "
1695                                         "for remount");
1696                                 return 0;
1697                         }
1698                         if (match_int(&args[0], &option) != 0)
1699                                 return 0;
1700                         *n_blocks_count = option;
1701                         break;
1702                 case Opt_nobh:
1703                         ext4_msg(sb, KERN_WARNING,
1704                                  "Ignoring deprecated nobh option");
1705                         break;
1706                 case Opt_bh:
1707                         ext4_msg(sb, KERN_WARNING,
1708                                  "Ignoring deprecated bh option");
1709                         break;
1710                 case Opt_i_version:
1711                         set_opt(sbi->s_mount_opt, I_VERSION);
1712                         sb->s_flags |= MS_I_VERSION;
1713                         break;
1714                 case Opt_nodelalloc:
1715                         clear_opt(sbi->s_mount_opt, DELALLOC);
1716                         break;
1717                 case Opt_stripe:
1718                         if (match_int(&args[0], &option))
1719                                 return 0;
1720                         if (option < 0)
1721                                 return 0;
1722                         sbi->s_stripe = option;
1723                         break;
1724                 case Opt_delalloc:
1725                         set_opt(sbi->s_mount_opt, DELALLOC);
1726                         break;
1727                 case Opt_block_validity:
1728                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1729                         break;
1730                 case Opt_noblock_validity:
1731                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1732                         break;
1733                 case Opt_inode_readahead_blks:
1734                         if (match_int(&args[0], &option))
1735                                 return 0;
1736                         if (option < 0 || option > (1 << 30))
1737                                 return 0;
1738                         if (!is_power_of_2(option)) {
1739                                 ext4_msg(sb, KERN_ERR,
1740                                          "EXT4-fs: inode_readahead_blks"
1741                                          " must be a power of 2");
1742                                 return 0;
1743                         }
1744                         sbi->s_inode_readahead_blks = option;
1745                         break;
1746                 case Opt_journal_ioprio:
1747                         if (match_int(&args[0], &option))
1748                                 return 0;
1749                         if (option < 0 || option > 7)
1750                                 break;
1751                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1752                                                             option);
1753                         break;
1754                 case Opt_noauto_da_alloc:
1755                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1756                         break;
1757                 case Opt_auto_da_alloc:
1758                         if (args[0].from) {
1759                                 if (match_int(&args[0], &option))
1760                                         return 0;
1761                         } else
1762                                 option = 1;     /* No argument, default to 1 */
1763                         if (option)
1764                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1765                         else
1766                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1767                         break;
1768                 case Opt_discard:
1769                         set_opt(sbi->s_mount_opt, DISCARD);
1770                         break;
1771                 case Opt_nodiscard:
1772                         clear_opt(sbi->s_mount_opt, DISCARD);
1773                         break;
1774                 case Opt_dioread_nolock:
1775                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1776                         break;
1777                 case Opt_dioread_lock:
1778                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1779                         break;
1780                 case Opt_init_inode_table:
1781                         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1782                         if (args[0].from) {
1783                                 if (match_int(&args[0], &option))
1784                                         return 0;
1785                         } else
1786                                 option = EXT4_DEF_LI_WAIT_MULT;
1787                         if (option < 0)
1788                                 return 0;
1789                         sbi->s_li_wait_mult = option;
1790                         break;
1791                 case Opt_noinit_inode_table:
1792                         clear_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1793                         break;
1794                 default:
1795                         ext4_msg(sb, KERN_ERR,
1796                                "Unrecognized mount option \"%s\" "
1797                                "or missing value", p);
1798                         return 0;
1799                 }
1800         }
1801 #ifdef CONFIG_QUOTA
1802         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1803                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1804                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1805
1806                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1807                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1808
1809                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1810                         ext4_msg(sb, KERN_ERR, "old and new quota "
1811                                         "format mixing");
1812                         return 0;
1813                 }
1814
1815                 if (!sbi->s_jquota_fmt) {
1816                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1817                                         "not specified");
1818                         return 0;
1819                 }
1820         } else {
1821                 if (sbi->s_jquota_fmt) {
1822                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1823                                         "specified with no journaling "
1824                                         "enabled");
1825                         return 0;
1826                 }
1827         }
1828 #endif
1829         return 1;
1830 }
1831
1832 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1833                             int read_only)
1834 {
1835         struct ext4_sb_info *sbi = EXT4_SB(sb);
1836         int res = 0;
1837
1838         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1839                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1840                          "forcing read-only mode");
1841                 res = MS_RDONLY;
1842         }
1843         if (read_only)
1844                 return res;
1845         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1846                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1847                          "running e2fsck is recommended");
1848         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1849                 ext4_msg(sb, KERN_WARNING,
1850                          "warning: mounting fs with errors, "
1851                          "running e2fsck is recommended");
1852         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1853                  le16_to_cpu(es->s_mnt_count) >=
1854                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1855                 ext4_msg(sb, KERN_WARNING,
1856                          "warning: maximal mount count reached, "
1857                          "running e2fsck is recommended");
1858         else if (le32_to_cpu(es->s_checkinterval) &&
1859                 (le32_to_cpu(es->s_lastcheck) +
1860                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1861                 ext4_msg(sb, KERN_WARNING,
1862                          "warning: checktime reached, "
1863                          "running e2fsck is recommended");
1864         if (!sbi->s_journal)
1865                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1866         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1867                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1868         le16_add_cpu(&es->s_mnt_count, 1);
1869         es->s_mtime = cpu_to_le32(get_seconds());
1870         ext4_update_dynamic_rev(sb);
1871         if (sbi->s_journal)
1872                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1873
1874         ext4_commit_super(sb, 1);
1875         if (test_opt(sb, DEBUG))
1876                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1877                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1878                         sb->s_blocksize,
1879                         sbi->s_groups_count,
1880                         EXT4_BLOCKS_PER_GROUP(sb),
1881                         EXT4_INODES_PER_GROUP(sb),
1882                         sbi->s_mount_opt);
1883
1884         return res;
1885 }
1886
1887 static int ext4_fill_flex_info(struct super_block *sb)
1888 {
1889         struct ext4_sb_info *sbi = EXT4_SB(sb);
1890         struct ext4_group_desc *gdp = NULL;
1891         ext4_group_t flex_group_count;
1892         ext4_group_t flex_group;
1893         int groups_per_flex = 0;
1894         size_t size;
1895         int i;
1896
1897         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1898         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1899
1900         if (groups_per_flex < 2) {
1901                 sbi->s_log_groups_per_flex = 0;
1902                 return 1;
1903         }
1904
1905         /* We allocate both existing and potentially added groups */
1906         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1907                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1908                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1909         size = flex_group_count * sizeof(struct flex_groups);
1910         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1911         if (sbi->s_flex_groups == NULL) {
1912                 sbi->s_flex_groups = vmalloc(size);
1913                 if (sbi->s_flex_groups)
1914                         memset(sbi->s_flex_groups, 0, size);
1915         }
1916         if (sbi->s_flex_groups == NULL) {
1917                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1918                                 "%u flex groups", flex_group_count);
1919                 goto failed;
1920         }
1921
1922         for (i = 0; i < sbi->s_groups_count; i++) {
1923                 gdp = ext4_get_group_desc(sb, i, NULL);
1924
1925                 flex_group = ext4_flex_group(sbi, i);
1926                 atomic_add(ext4_free_inodes_count(sb, gdp),
1927                            &sbi->s_flex_groups[flex_group].free_inodes);
1928                 atomic_add(ext4_free_blks_count(sb, gdp),
1929                            &sbi->s_flex_groups[flex_group].free_blocks);
1930                 atomic_add(ext4_used_dirs_count(sb, gdp),
1931                            &sbi->s_flex_groups[flex_group].used_dirs);
1932         }
1933
1934         return 1;
1935 failed:
1936         return 0;
1937 }
1938
1939 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1940                             struct ext4_group_desc *gdp)
1941 {
1942         __u16 crc = 0;
1943
1944         if (sbi->s_es->s_feature_ro_compat &
1945             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1946                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1947                 __le32 le_group = cpu_to_le32(block_group);
1948
1949                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1950                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1951                 crc = crc16(crc, (__u8 *)gdp, offset);
1952                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1953                 /* for checksum of struct ext4_group_desc do the rest...*/
1954                 if ((sbi->s_es->s_feature_incompat &
1955                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1956                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1957                         crc = crc16(crc, (__u8 *)gdp + offset,
1958                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1959                                         offset);
1960         }
1961
1962         return cpu_to_le16(crc);
1963 }
1964
1965 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1966                                 struct ext4_group_desc *gdp)
1967 {
1968         if ((sbi->s_es->s_feature_ro_compat &
1969              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1970             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1971                 return 0;
1972
1973         return 1;
1974 }
1975
1976 /* Called at mount-time, super-block is locked */
1977 static int ext4_check_descriptors(struct super_block *sb,
1978                                   ext4_group_t *first_not_zeroed)
1979 {
1980         struct ext4_sb_info *sbi = EXT4_SB(sb);
1981         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1982         ext4_fsblk_t last_block;
1983         ext4_fsblk_t block_bitmap;
1984         ext4_fsblk_t inode_bitmap;
1985         ext4_fsblk_t inode_table;
1986         int flexbg_flag = 0;
1987         ext4_group_t i, grp = sbi->s_groups_count;
1988
1989         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1990                 flexbg_flag = 1;
1991
1992         ext4_debug("Checking group descriptors");
1993
1994         for (i = 0; i < sbi->s_groups_count; i++) {
1995                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1996
1997                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1998                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1999                 else
2000                         last_block = first_block +
2001                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2002
2003                 if ((grp == sbi->s_groups_count) &&
2004                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2005                         grp = i;
2006
2007                 block_bitmap = ext4_block_bitmap(sb, gdp);
2008                 if (block_bitmap < first_block || block_bitmap > last_block) {
2009                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2010                                "Block bitmap for group %u not in group "
2011                                "(block %llu)!", i, block_bitmap);
2012                         return 0;
2013                 }
2014                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2015                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2016                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2017                                "Inode bitmap for group %u not in group "
2018                                "(block %llu)!", i, inode_bitmap);
2019                         return 0;
2020                 }
2021                 inode_table = ext4_inode_table(sb, gdp);
2022                 if (inode_table < first_block ||
2023                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2024                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2025                                "Inode table for group %u not in group "
2026                                "(block %llu)!", i, inode_table);
2027                         return 0;
2028                 }
2029                 ext4_lock_group(sb, i);
2030                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2031                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2032                                  "Checksum for group %u failed (%u!=%u)",
2033                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2034                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2035                         if (!(sb->s_flags & MS_RDONLY)) {
2036                                 ext4_unlock_group(sb, i);
2037                                 return 0;
2038                         }
2039                 }
2040                 ext4_unlock_group(sb, i);
2041                 if (!flexbg_flag)
2042                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2043         }
2044         if (NULL != first_not_zeroed)
2045                 *first_not_zeroed = grp;
2046
2047         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2048         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2049         return 1;
2050 }
2051
2052 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2053  * the superblock) which were deleted from all directories, but held open by
2054  * a process at the time of a crash.  We walk the list and try to delete these
2055  * inodes at recovery time (only with a read-write filesystem).
2056  *
2057  * In order to keep the orphan inode chain consistent during traversal (in
2058  * case of crash during recovery), we link each inode into the superblock
2059  * orphan list_head and handle it the same way as an inode deletion during
2060  * normal operation (which journals the operations for us).
2061  *
2062  * We only do an iget() and an iput() on each inode, which is very safe if we
2063  * accidentally point at an in-use or already deleted inode.  The worst that
2064  * can happen in this case is that we get a "bit already cleared" message from
2065  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2066  * e2fsck was run on this filesystem, and it must have already done the orphan
2067  * inode cleanup for us, so we can safely abort without any further action.
2068  */
2069 static void ext4_orphan_cleanup(struct super_block *sb,
2070                                 struct ext4_super_block *es)
2071 {
2072         unsigned int s_flags = sb->s_flags;
2073         int nr_orphans = 0, nr_truncates = 0;
2074 #ifdef CONFIG_QUOTA
2075         int i;
2076 #endif
2077         if (!es->s_last_orphan) {
2078                 jbd_debug(4, "no orphan inodes to clean up\n");
2079                 return;
2080         }
2081
2082         if (bdev_read_only(sb->s_bdev)) {
2083                 ext4_msg(sb, KERN_ERR, "write access "
2084                         "unavailable, skipping orphan cleanup");
2085                 return;
2086         }
2087
2088         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2089                 if (es->s_last_orphan)
2090                         jbd_debug(1, "Errors on filesystem, "
2091                                   "clearing orphan list.\n");
2092                 es->s_last_orphan = 0;
2093                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2094                 return;
2095         }
2096
2097         if (s_flags & MS_RDONLY) {
2098                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2099                 sb->s_flags &= ~MS_RDONLY;
2100         }
2101 #ifdef CONFIG_QUOTA
2102         /* Needed for iput() to work correctly and not trash data */
2103         sb->s_flags |= MS_ACTIVE;
2104         /* Turn on quotas so that they are updated correctly */
2105         for (i = 0; i < MAXQUOTAS; i++) {
2106                 if (EXT4_SB(sb)->s_qf_names[i]) {
2107                         int ret = ext4_quota_on_mount(sb, i);
2108                         if (ret < 0)
2109                                 ext4_msg(sb, KERN_ERR,
2110                                         "Cannot turn on journaled "
2111                                         "quota: error %d", ret);
2112                 }
2113         }
2114 #endif
2115
2116         while (es->s_last_orphan) {
2117                 struct inode *inode;
2118
2119                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2120                 if (IS_ERR(inode)) {
2121                         es->s_last_orphan = 0;
2122                         break;
2123                 }
2124
2125                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2126                 dquot_initialize(inode);
2127                 if (inode->i_nlink) {
2128                         ext4_msg(sb, KERN_DEBUG,
2129                                 "%s: truncating inode %lu to %lld bytes",
2130                                 __func__, inode->i_ino, inode->i_size);
2131                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2132                                   inode->i_ino, inode->i_size);
2133                         ext4_truncate(inode);
2134                         nr_truncates++;
2135                 } else {
2136                         ext4_msg(sb, KERN_DEBUG,
2137                                 "%s: deleting unreferenced inode %lu",
2138                                 __func__, inode->i_ino);
2139                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2140                                   inode->i_ino);
2141                         nr_orphans++;
2142                 }
2143                 iput(inode);  /* The delete magic happens here! */
2144         }
2145
2146 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2147
2148         if (nr_orphans)
2149                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2150                        PLURAL(nr_orphans));
2151         if (nr_truncates)
2152                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2153                        PLURAL(nr_truncates));
2154 #ifdef CONFIG_QUOTA
2155         /* Turn quotas off */
2156         for (i = 0; i < MAXQUOTAS; i++) {
2157                 if (sb_dqopt(sb)->files[i])
2158                         dquot_quota_off(sb, i);
2159         }
2160 #endif
2161         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2162 }
2163
2164 /*
2165  * Maximal extent format file size.
2166  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2167  * extent format containers, within a sector_t, and within i_blocks
2168  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2169  * so that won't be a limiting factor.
2170  *
2171  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2172  */
2173 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2174 {
2175         loff_t res;
2176         loff_t upper_limit = MAX_LFS_FILESIZE;
2177
2178         /* small i_blocks in vfs inode? */
2179         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2180                 /*
2181                  * CONFIG_LBDAF is not enabled implies the inode
2182                  * i_block represent total blocks in 512 bytes
2183                  * 32 == size of vfs inode i_blocks * 8
2184                  */
2185                 upper_limit = (1LL << 32) - 1;
2186
2187                 /* total blocks in file system block size */
2188                 upper_limit >>= (blkbits - 9);
2189                 upper_limit <<= blkbits;
2190         }
2191
2192         /* 32-bit extent-start container, ee_block */
2193         res = 1LL << 32;
2194         res <<= blkbits;
2195         res -= 1;
2196
2197         /* Sanity check against vm- & vfs- imposed limits */
2198         if (res > upper_limit)
2199                 res = upper_limit;
2200
2201         return res;
2202 }
2203
2204 /*
2205  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2206  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2207  * We need to be 1 filesystem block less than the 2^48 sector limit.
2208  */
2209 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2210 {
2211         loff_t res = EXT4_NDIR_BLOCKS;
2212         int meta_blocks;
2213         loff_t upper_limit;
2214         /* This is calculated to be the largest file size for a dense, block
2215          * mapped file such that the file's total number of 512-byte sectors,
2216          * including data and all indirect blocks, does not exceed (2^48 - 1).
2217          *
2218          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2219          * number of 512-byte sectors of the file.
2220          */
2221
2222         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2223                 /*
2224                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2225                  * the inode i_block field represents total file blocks in
2226                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2227                  */
2228                 upper_limit = (1LL << 32) - 1;
2229
2230                 /* total blocks in file system block size */
2231                 upper_limit >>= (bits - 9);
2232
2233         } else {
2234                 /*
2235                  * We use 48 bit ext4_inode i_blocks
2236                  * With EXT4_HUGE_FILE_FL set the i_blocks
2237                  * represent total number of blocks in
2238                  * file system block size
2239                  */
2240                 upper_limit = (1LL << 48) - 1;
2241
2242         }
2243
2244         /* indirect blocks */
2245         meta_blocks = 1;
2246         /* double indirect blocks */
2247         meta_blocks += 1 + (1LL << (bits-2));
2248         /* tripple indirect blocks */
2249         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2250
2251         upper_limit -= meta_blocks;
2252         upper_limit <<= bits;
2253
2254         res += 1LL << (bits-2);
2255         res += 1LL << (2*(bits-2));
2256         res += 1LL << (3*(bits-2));
2257         res <<= bits;
2258         if (res > upper_limit)
2259                 res = upper_limit;
2260
2261         if (res > MAX_LFS_FILESIZE)
2262                 res = MAX_LFS_FILESIZE;
2263
2264         return res;
2265 }
2266
2267 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2268                                    ext4_fsblk_t logical_sb_block, int nr)
2269 {
2270         struct ext4_sb_info *sbi = EXT4_SB(sb);
2271         ext4_group_t bg, first_meta_bg;
2272         int has_super = 0;
2273
2274         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2275
2276         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2277             nr < first_meta_bg)
2278                 return logical_sb_block + nr + 1;
2279         bg = sbi->s_desc_per_block * nr;
2280         if (ext4_bg_has_super(sb, bg))
2281                 has_super = 1;
2282
2283         return (has_super + ext4_group_first_block_no(sb, bg));
2284 }
2285
2286 /**
2287  * ext4_get_stripe_size: Get the stripe size.
2288  * @sbi: In memory super block info
2289  *
2290  * If we have specified it via mount option, then
2291  * use the mount option value. If the value specified at mount time is
2292  * greater than the blocks per group use the super block value.
2293  * If the super block value is greater than blocks per group return 0.
2294  * Allocator needs it be less than blocks per group.
2295  *
2296  */
2297 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2298 {
2299         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2300         unsigned long stripe_width =
2301                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2302
2303         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2304                 return sbi->s_stripe;
2305
2306         if (stripe_width <= sbi->s_blocks_per_group)
2307                 return stripe_width;
2308
2309         if (stride <= sbi->s_blocks_per_group)
2310                 return stride;
2311
2312         return 0;
2313 }
2314
2315 /* sysfs supprt */
2316
2317 struct ext4_attr {
2318         struct attribute attr;
2319         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2320         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2321                          const char *, size_t);
2322         int offset;
2323 };
2324
2325 static int parse_strtoul(const char *buf,
2326                 unsigned long max, unsigned long *value)
2327 {
2328         char *endp;
2329
2330         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2331         endp = skip_spaces(endp);
2332         if (*endp || *value > max)
2333                 return -EINVAL;
2334
2335         return 0;
2336 }
2337
2338 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2339                                               struct ext4_sb_info *sbi,
2340                                               char *buf)
2341 {
2342         return snprintf(buf, PAGE_SIZE, "%llu\n",
2343                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2344 }
2345
2346 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2347                                          struct ext4_sb_info *sbi, char *buf)
2348 {
2349         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2350
2351         if (!sb->s_bdev->bd_part)
2352                 return snprintf(buf, PAGE_SIZE, "0\n");
2353         return snprintf(buf, PAGE_SIZE, "%lu\n",
2354                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2355                          sbi->s_sectors_written_start) >> 1);
2356 }
2357
2358 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2359                                           struct ext4_sb_info *sbi, char *buf)
2360 {
2361         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2362
2363         if (!sb->s_bdev->bd_part)
2364                 return snprintf(buf, PAGE_SIZE, "0\n");
2365         return snprintf(buf, PAGE_SIZE, "%llu\n",
2366                         (unsigned long long)(sbi->s_kbytes_written +
2367                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2368                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2369 }
2370
2371 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2372                                           struct ext4_sb_info *sbi,
2373                                           const char *buf, size_t count)
2374 {
2375         unsigned long t;
2376
2377         if (parse_strtoul(buf, 0x40000000, &t))
2378                 return -EINVAL;
2379
2380         if (!is_power_of_2(t))
2381                 return -EINVAL;
2382
2383         sbi->s_inode_readahead_blks = t;
2384         return count;
2385 }
2386
2387 static ssize_t sbi_ui_show(struct ext4_attr *a,
2388                            struct ext4_sb_info *sbi, char *buf)
2389 {
2390         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2391
2392         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2393 }
2394
2395 static ssize_t sbi_ui_store(struct ext4_attr *a,
2396                             struct ext4_sb_info *sbi,
2397                             const char *buf, size_t count)
2398 {
2399         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2400         unsigned long t;
2401
2402         if (parse_strtoul(buf, 0xffffffff, &t))
2403                 return -EINVAL;
2404         *ui = t;
2405         return count;
2406 }
2407
2408 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2409 static struct ext4_attr ext4_attr_##_name = {                   \
2410         .attr = {.name = __stringify(_name), .mode = _mode },   \
2411         .show   = _show,                                        \
2412         .store  = _store,                                       \
2413         .offset = offsetof(struct ext4_sb_info, _elname),       \
2414 }
2415 #define EXT4_ATTR(name, mode, show, store) \
2416 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2417
2418 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2419 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2420 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2421 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2422         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2423 #define ATTR_LIST(name) &ext4_attr_##name.attr
2424
2425 EXT4_RO_ATTR(delayed_allocation_blocks);
2426 EXT4_RO_ATTR(session_write_kbytes);
2427 EXT4_RO_ATTR(lifetime_write_kbytes);
2428 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2429                  inode_readahead_blks_store, s_inode_readahead_blks);
2430 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2431 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2432 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2433 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2434 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2435 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2436 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2437 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2438
2439 static struct attribute *ext4_attrs[] = {
2440         ATTR_LIST(delayed_allocation_blocks),
2441         ATTR_LIST(session_write_kbytes),
2442         ATTR_LIST(lifetime_write_kbytes),
2443         ATTR_LIST(inode_readahead_blks),
2444         ATTR_LIST(inode_goal),
2445         ATTR_LIST(mb_stats),
2446         ATTR_LIST(mb_max_to_scan),
2447         ATTR_LIST(mb_min_to_scan),
2448         ATTR_LIST(mb_order2_req),
2449         ATTR_LIST(mb_stream_req),
2450         ATTR_LIST(mb_group_prealloc),
2451         ATTR_LIST(max_writeback_mb_bump),
2452         NULL,
2453 };
2454
2455 /* Features this copy of ext4 supports */
2456 EXT4_INFO_ATTR(lazy_itable_init);
2457 EXT4_INFO_ATTR(batched_discard);
2458
2459 static struct attribute *ext4_feat_attrs[] = {
2460         ATTR_LIST(lazy_itable_init),
2461         ATTR_LIST(batched_discard),
2462         NULL,
2463 };
2464
2465 static ssize_t ext4_attr_show(struct kobject *kobj,
2466                               struct attribute *attr, char *buf)
2467 {
2468         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2469                                                 s_kobj);
2470         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2471
2472         return a->show ? a->show(a, sbi, buf) : 0;
2473 }
2474
2475 static ssize_t ext4_attr_store(struct kobject *kobj,
2476                                struct attribute *attr,
2477                                const char *buf, size_t len)
2478 {
2479         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2480                                                 s_kobj);
2481         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2482
2483         return a->store ? a->store(a, sbi, buf, len) : 0;
2484 }
2485
2486 static void ext4_sb_release(struct kobject *kobj)
2487 {
2488         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2489                                                 s_kobj);
2490         complete(&sbi->s_kobj_unregister);
2491 }
2492
2493 static const struct sysfs_ops ext4_attr_ops = {
2494         .show   = ext4_attr_show,
2495         .store  = ext4_attr_store,
2496 };
2497
2498 static struct kobj_type ext4_ktype = {
2499         .default_attrs  = ext4_attrs,
2500         .sysfs_ops      = &ext4_attr_ops,
2501         .release        = ext4_sb_release,
2502 };
2503
2504 static void ext4_feat_release(struct kobject *kobj)
2505 {
2506         complete(&ext4_feat->f_kobj_unregister);
2507 }
2508
2509 static struct kobj_type ext4_feat_ktype = {
2510         .default_attrs  = ext4_feat_attrs,
2511         .sysfs_ops      = &ext4_attr_ops,
2512         .release        = ext4_feat_release,
2513 };
2514
2515 /*
2516  * Check whether this filesystem can be mounted based on
2517  * the features present and the RDONLY/RDWR mount requested.
2518  * Returns 1 if this filesystem can be mounted as requested,
2519  * 0 if it cannot be.
2520  */
2521 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2522 {
2523         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2524                 ext4_msg(sb, KERN_ERR,
2525                         "Couldn't mount because of "
2526                         "unsupported optional features (%x)",
2527                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2528                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2529                 return 0;
2530         }
2531
2532         if (readonly)
2533                 return 1;
2534
2535         /* Check that feature set is OK for a read-write mount */
2536         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2537                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2538                          "unsupported optional features (%x)",
2539                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2540                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2541                 return 0;
2542         }
2543         /*
2544          * Large file size enabled file system can only be mounted
2545          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2546          */
2547         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2548                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2549                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2550                                  "cannot be mounted RDWR without "
2551                                  "CONFIG_LBDAF");
2552                         return 0;
2553                 }
2554         }
2555         return 1;
2556 }
2557
2558 /*
2559  * This function is called once a day if we have errors logged
2560  * on the file system
2561  */
2562 static void print_daily_error_info(unsigned long arg)
2563 {
2564         struct super_block *sb = (struct super_block *) arg;
2565         struct ext4_sb_info *sbi;
2566         struct ext4_super_block *es;
2567
2568         sbi = EXT4_SB(sb);
2569         es = sbi->s_es;
2570
2571         if (es->s_error_count)
2572                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2573                          le32_to_cpu(es->s_error_count));
2574         if (es->s_first_error_time) {
2575                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2576                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2577                        (int) sizeof(es->s_first_error_func),
2578                        es->s_first_error_func,
2579                        le32_to_cpu(es->s_first_error_line));
2580                 if (es->s_first_error_ino)
2581                         printk(": inode %u",
2582                                le32_to_cpu(es->s_first_error_ino));
2583                 if (es->s_first_error_block)
2584                         printk(": block %llu", (unsigned long long)
2585                                le64_to_cpu(es->s_first_error_block));
2586                 printk("\n");
2587         }
2588         if (es->s_last_error_time) {
2589                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2590                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2591                        (int) sizeof(es->s_last_error_func),
2592                        es->s_last_error_func,
2593                        le32_to_cpu(es->s_last_error_line));
2594                 if (es->s_last_error_ino)
2595                         printk(": inode %u",
2596                                le32_to_cpu(es->s_last_error_ino));
2597                 if (es->s_last_error_block)
2598                         printk(": block %llu", (unsigned long long)
2599                                le64_to_cpu(es->s_last_error_block));
2600                 printk("\n");
2601         }
2602         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2603 }
2604
2605 static void ext4_lazyinode_timeout(unsigned long data)
2606 {
2607         struct task_struct *p = (struct task_struct *)data;
2608         wake_up_process(p);
2609 }
2610
2611 /* Find next suitable group and run ext4_init_inode_table */
2612 static int ext4_run_li_request(struct ext4_li_request *elr)
2613 {
2614         struct ext4_group_desc *gdp = NULL;
2615         ext4_group_t group, ngroups;
2616         struct super_block *sb;
2617         unsigned long timeout = 0;
2618         int ret = 0;
2619
2620         sb = elr->lr_super;
2621         ngroups = EXT4_SB(sb)->s_groups_count;
2622
2623         for (group = elr->lr_next_group; group < ngroups; group++) {
2624                 gdp = ext4_get_group_desc(sb, group, NULL);
2625                 if (!gdp) {
2626                         ret = 1;
2627                         break;
2628                 }
2629
2630                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2631                         break;
2632         }
2633
2634         if (group == ngroups)
2635                 ret = 1;
2636
2637         if (!ret) {
2638                 timeout = jiffies;
2639                 ret = ext4_init_inode_table(sb, group,
2640                                             elr->lr_timeout ? 0 : 1);
2641                 if (elr->lr_timeout == 0) {
2642                         timeout = jiffies - timeout;
2643                         if (elr->lr_sbi->s_li_wait_mult)
2644                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2645                         else
2646                                 timeout *= 20;
2647                         elr->lr_timeout = timeout;
2648                 }
2649                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2650                 elr->lr_next_group = group + 1;
2651         }
2652
2653         return ret;
2654 }
2655
2656 /*
2657  * Remove lr_request from the list_request and free the
2658  * request tructure. Should be called with li_list_mtx held
2659  */
2660 static void ext4_remove_li_request(struct ext4_li_request *elr)
2661 {
2662         struct ext4_sb_info *sbi;
2663
2664         if (!elr)
2665                 return;
2666
2667         sbi = elr->lr_sbi;
2668
2669         list_del(&elr->lr_request);
2670         sbi->s_li_request = NULL;
2671         kfree(elr);
2672 }
2673
2674 static void ext4_unregister_li_request(struct super_block *sb)
2675 {
2676         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2677
2678         if (!ext4_li_info)
2679                 return;
2680
2681         mutex_lock(&ext4_li_info->li_list_mtx);
2682         ext4_remove_li_request(elr);
2683         mutex_unlock(&ext4_li_info->li_list_mtx);
2684 }
2685
2686 /*
2687  * This is the function where ext4lazyinit thread lives. It walks
2688  * through the request list searching for next scheduled filesystem.
2689  * When such a fs is found, run the lazy initialization request
2690  * (ext4_rn_li_request) and keep track of the time spend in this
2691  * function. Based on that time we compute next schedule time of
2692  * the request. When walking through the list is complete, compute
2693  * next waking time and put itself into sleep.
2694  */
2695 static int ext4_lazyinit_thread(void *arg)
2696 {
2697         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2698         struct list_head *pos, *n;
2699         struct ext4_li_request *elr;
2700         unsigned long next_wakeup;
2701         DEFINE_WAIT(wait);
2702         int ret;
2703
2704         BUG_ON(NULL == eli);
2705
2706         eli->li_timer.data = (unsigned long)current;
2707         eli->li_timer.function = ext4_lazyinode_timeout;
2708
2709         eli->li_task = current;
2710         wake_up(&eli->li_wait_task);
2711
2712 cont_thread:
2713         while (true) {
2714                 next_wakeup = MAX_JIFFY_OFFSET;
2715
2716                 mutex_lock(&eli->li_list_mtx);
2717                 if (list_empty(&eli->li_request_list)) {
2718                         mutex_unlock(&eli->li_list_mtx);
2719                         goto exit_thread;
2720                 }
2721
2722                 list_for_each_safe(pos, n, &eli->li_request_list) {
2723                         elr = list_entry(pos, struct ext4_li_request,
2724                                          lr_request);
2725
2726                         if (time_after_eq(jiffies, elr->lr_next_sched))
2727                                 ret = ext4_run_li_request(elr);
2728
2729                         if (ret) {
2730                                 ret = 0;
2731                                 ext4_remove_li_request(elr);
2732                                 continue;
2733                         }
2734
2735                         if (time_before(elr->lr_next_sched, next_wakeup))
2736                                 next_wakeup = elr->lr_next_sched;
2737                 }
2738                 mutex_unlock(&eli->li_list_mtx);
2739
2740                 if (freezing(current))
2741                         refrigerator();
2742
2743                 if (time_after_eq(jiffies, next_wakeup)) {
2744                         cond_resched();
2745                         continue;
2746                 }
2747
2748                 eli->li_timer.expires = next_wakeup;
2749                 add_timer(&eli->li_timer);
2750                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2751                                 TASK_INTERRUPTIBLE);
2752                 if (time_before(jiffies, next_wakeup))
2753                         schedule();
2754                 finish_wait(&eli->li_wait_daemon, &wait);
2755         }
2756
2757 exit_thread:
2758         /*
2759          * It looks like the request list is empty, but we need
2760          * to check it under the li_list_mtx lock, to prevent any
2761          * additions into it, and of course we should lock ext4_li_mtx
2762          * to atomically free the list and ext4_li_info, because at
2763          * this point another ext4 filesystem could be registering
2764          * new one.
2765          */
2766         mutex_lock(&ext4_li_mtx);
2767         mutex_lock(&eli->li_list_mtx);
2768         if (!list_empty(&eli->li_request_list)) {
2769                 mutex_unlock(&eli->li_list_mtx);
2770                 mutex_unlock(&ext4_li_mtx);
2771                 goto cont_thread;
2772         }
2773         mutex_unlock(&eli->li_list_mtx);
2774         del_timer_sync(&ext4_li_info->li_timer);
2775         eli->li_task = NULL;
2776         wake_up(&eli->li_wait_task);
2777
2778         kfree(ext4_li_info);
2779         ext4_li_info = NULL;
2780         mutex_unlock(&ext4_li_mtx);
2781
2782         return 0;
2783 }
2784
2785 static void ext4_clear_request_list(void)
2786 {
2787         struct list_head *pos, *n;
2788         struct ext4_li_request *elr;
2789
2790         mutex_lock(&ext4_li_info->li_list_mtx);
2791         if (list_empty(&ext4_li_info->li_request_list))
2792                 return;
2793
2794         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2795                 elr = list_entry(pos, struct ext4_li_request,
2796                                  lr_request);
2797                 ext4_remove_li_request(elr);
2798         }
2799         mutex_unlock(&ext4_li_info->li_list_mtx);
2800 }
2801
2802 static int ext4_run_lazyinit_thread(void)
2803 {
2804         struct task_struct *t;
2805
2806         t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2807         if (IS_ERR(t)) {
2808                 int err = PTR_ERR(t);
2809                 ext4_clear_request_list();
2810                 del_timer_sync(&ext4_li_info->li_timer);
2811                 kfree(ext4_li_info);
2812                 ext4_li_info = NULL;
2813                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2814                                  "initialization thread\n",
2815                                  err);
2816                 return err;
2817         }
2818         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2819
2820         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2821         return 0;
2822 }
2823
2824 /*
2825  * Check whether it make sense to run itable init. thread or not.
2826  * If there is at least one uninitialized inode table, return
2827  * corresponding group number, else the loop goes through all
2828  * groups and return total number of groups.
2829  */
2830 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2831 {
2832         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2833         struct ext4_group_desc *gdp = NULL;
2834
2835         for (group = 0; group < ngroups; group++) {
2836                 gdp = ext4_get_group_desc(sb, group, NULL);
2837                 if (!gdp)
2838                         continue;
2839
2840                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2841                         break;
2842         }
2843
2844         return group;
2845 }
2846
2847 static int ext4_li_info_new(void)
2848 {
2849         struct ext4_lazy_init *eli = NULL;
2850
2851         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2852         if (!eli)
2853                 return -ENOMEM;
2854
2855         eli->li_task = NULL;
2856         INIT_LIST_HEAD(&eli->li_request_list);
2857         mutex_init(&eli->li_list_mtx);
2858
2859         init_waitqueue_head(&eli->li_wait_daemon);
2860         init_waitqueue_head(&eli->li_wait_task);
2861         init_timer(&eli->li_timer);
2862         eli->li_state |= EXT4_LAZYINIT_QUIT;
2863
2864         ext4_li_info = eli;
2865
2866         return 0;
2867 }
2868
2869 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2870                                             ext4_group_t start)
2871 {
2872         struct ext4_sb_info *sbi = EXT4_SB(sb);
2873         struct ext4_li_request *elr;
2874         unsigned long rnd;
2875
2876         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2877         if (!elr)
2878                 return NULL;
2879
2880         elr->lr_super = sb;
2881         elr->lr_sbi = sbi;
2882         elr->lr_next_group = start;
2883
2884         /*
2885          * Randomize first schedule time of the request to
2886          * spread the inode table initialization requests
2887          * better.
2888          */
2889         get_random_bytes(&rnd, sizeof(rnd));
2890         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2891                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2892
2893         return elr;
2894 }
2895
2896 static int ext4_register_li_request(struct super_block *sb,
2897                                     ext4_group_t first_not_zeroed)
2898 {
2899         struct ext4_sb_info *sbi = EXT4_SB(sb);
2900         struct ext4_li_request *elr;
2901         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2902         int ret;
2903
2904         if (sbi->s_li_request != NULL)
2905                 return 0;
2906
2907         if (first_not_zeroed == ngroups ||
2908             (sb->s_flags & MS_RDONLY) ||
2909             !test_opt(sb, INIT_INODE_TABLE)) {
2910                 sbi->s_li_request = NULL;
2911                 return 0;
2912         }
2913
2914         if (first_not_zeroed == ngroups) {
2915                 sbi->s_li_request = NULL;
2916                 return 0;
2917         }
2918
2919         elr = ext4_li_request_new(sb, first_not_zeroed);
2920         if (!elr)
2921                 return -ENOMEM;
2922
2923         mutex_lock(&ext4_li_mtx);
2924
2925         if (NULL == ext4_li_info) {
2926                 ret = ext4_li_info_new();
2927                 if (ret)
2928                         goto out;
2929         }
2930
2931         mutex_lock(&ext4_li_info->li_list_mtx);
2932         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2933         mutex_unlock(&ext4_li_info->li_list_mtx);
2934
2935         sbi->s_li_request = elr;
2936
2937         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2938                 ret = ext4_run_lazyinit_thread();
2939                 if (ret)
2940                         goto out;
2941         }
2942 out:
2943         mutex_unlock(&ext4_li_mtx);
2944         if (ret)
2945                 kfree(elr);
2946         return ret;
2947 }
2948
2949 /*
2950  * We do not need to lock anything since this is called on
2951  * module unload.
2952  */
2953 static void ext4_destroy_lazyinit_thread(void)
2954 {
2955         /*
2956          * If thread exited earlier
2957          * there's nothing to be done.
2958          */
2959         if (!ext4_li_info)
2960                 return;
2961
2962         ext4_clear_request_list();
2963
2964         while (ext4_li_info->li_task) {
2965                 wake_up(&ext4_li_info->li_wait_daemon);
2966                 wait_event(ext4_li_info->li_wait_task,
2967                            ext4_li_info->li_task == NULL);
2968         }
2969 }
2970
2971 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2972                                 __releases(kernel_lock)
2973                                 __acquires(kernel_lock)
2974 {
2975         char *orig_data = kstrdup(data, GFP_KERNEL);
2976         struct buffer_head *bh;
2977         struct ext4_super_block *es = NULL;
2978         struct ext4_sb_info *sbi;
2979         ext4_fsblk_t block;
2980         ext4_fsblk_t sb_block = get_sb_block(&data);
2981         ext4_fsblk_t logical_sb_block;
2982         unsigned long offset = 0;
2983         unsigned long journal_devnum = 0;
2984         unsigned long def_mount_opts;
2985         struct inode *root;
2986         char *cp;
2987         const char *descr;
2988         int ret = -ENOMEM;
2989         int blocksize;
2990         unsigned int db_count;
2991         unsigned int i;
2992         int needs_recovery, has_huge_files;
2993         __u64 blocks_count;
2994         int err;
2995         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2996         ext4_group_t first_not_zeroed;
2997
2998         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2999         if (!sbi)
3000                 goto out_free_orig;
3001
3002         sbi->s_blockgroup_lock =
3003                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3004         if (!sbi->s_blockgroup_lock) {
3005                 kfree(sbi);
3006                 goto out_free_orig;
3007         }
3008         sb->s_fs_info = sbi;
3009         sbi->s_mount_opt = 0;
3010         sbi->s_resuid = EXT4_DEF_RESUID;
3011         sbi->s_resgid = EXT4_DEF_RESGID;
3012         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3013         sbi->s_sb_block = sb_block;
3014         if (sb->s_bdev->bd_part)
3015                 sbi->s_sectors_written_start =
3016                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3017
3018         /* Cleanup superblock name */
3019         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3020                 *cp = '!';
3021
3022         ret = -EINVAL;
3023         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3024         if (!blocksize) {
3025                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3026                 goto out_fail;
3027         }
3028
3029         /*
3030          * The ext4 superblock will not be buffer aligned for other than 1kB
3031          * block sizes.  We need to calculate the offset from buffer start.
3032          */
3033         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3034                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3035                 offset = do_div(logical_sb_block, blocksize);
3036         } else {
3037                 logical_sb_block = sb_block;
3038         }
3039
3040         if (!(bh = sb_bread(sb, logical_sb_block))) {
3041                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3042                 goto out_fail;
3043         }
3044         /*
3045          * Note: s_es must be initialized as soon as possible because
3046          *       some ext4 macro-instructions depend on its value
3047          */
3048         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3049         sbi->s_es = es;
3050         sb->s_magic = le16_to_cpu(es->s_magic);
3051         if (sb->s_magic != EXT4_SUPER_MAGIC)
3052                 goto cantfind_ext4;
3053         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3054
3055         /* Set defaults before we parse the mount options */
3056         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3057         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
3058         if (def_mount_opts & EXT4_DEFM_DEBUG)
3059                 set_opt(sbi->s_mount_opt, DEBUG);
3060         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3061                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3062                         "2.6.38");
3063                 set_opt(sbi->s_mount_opt, GRPID);
3064         }
3065         if (def_mount_opts & EXT4_DEFM_UID16)
3066                 set_opt(sbi->s_mount_opt, NO_UID32);
3067 #ifdef CONFIG_EXT4_FS_XATTR
3068         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3069                 set_opt(sbi->s_mount_opt, XATTR_USER);
3070 #endif
3071 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3072         if (def_mount_opts & EXT4_DEFM_ACL)
3073                 set_opt(sbi->s_mount_opt, POSIX_ACL);
3074 #endif
3075         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3076                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3077         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3078                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
3079         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3080                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3081
3082         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3083                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
3084         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3085                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
3086         else
3087                 set_opt(sbi->s_mount_opt, ERRORS_RO);
3088         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3089                 set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
3090         if (def_mount_opts & EXT4_DEFM_DISCARD)
3091                 set_opt(sbi->s_mount_opt, DISCARD);
3092
3093         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3094         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3095         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3096         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3097         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3098
3099         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3100                 set_opt(sbi->s_mount_opt, BARRIER);
3101
3102         /*
3103          * enable delayed allocation by default
3104          * Use -o nodelalloc to turn it off
3105          */
3106         if (!IS_EXT3_SB(sb) &&
3107             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3108                 set_opt(sbi->s_mount_opt, DELALLOC);
3109
3110         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3111                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3112                 ext4_msg(sb, KERN_WARNING,
3113                          "failed to parse options in superblock: %s",
3114                          sbi->s_es->s_mount_opts);
3115         }
3116         if (!parse_options((char *) data, sb, &journal_devnum,
3117                            &journal_ioprio, NULL, 0))
3118                 goto failed_mount;
3119
3120         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3121                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3122
3123         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3124             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3125              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3126              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3127                 ext4_msg(sb, KERN_WARNING,
3128                        "feature flags set on rev 0 fs, "
3129                        "running e2fsck is recommended");
3130
3131         /*
3132          * Check feature flags regardless of the revision level, since we
3133          * previously didn't change the revision level when setting the flags,
3134          * so there is a chance incompat flags are set on a rev 0 filesystem.
3135          */
3136         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3137                 goto failed_mount;
3138
3139         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3140
3141         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3142             blocksize > EXT4_MAX_BLOCK_SIZE) {
3143                 ext4_msg(sb, KERN_ERR,
3144                        "Unsupported filesystem blocksize %d", blocksize);
3145                 goto failed_mount;
3146         }
3147
3148         if (sb->s_blocksize != blocksize) {
3149                 /* Validate the filesystem blocksize */
3150                 if (!sb_set_blocksize(sb, blocksize)) {
3151                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3152                                         blocksize);
3153                         goto failed_mount;
3154                 }
3155
3156                 brelse(bh);
3157                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3158                 offset = do_div(logical_sb_block, blocksize);
3159                 bh = sb_bread(sb, logical_sb_block);
3160                 if (!bh) {
3161                         ext4_msg(sb, KERN_ERR,
3162                                "Can't read superblock on 2nd try");
3163                         goto failed_mount;
3164                 }
3165                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3166                 sbi->s_es = es;
3167                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3168                         ext4_msg(sb, KERN_ERR,
3169                                "Magic mismatch, very weird!");
3170                         goto failed_mount;
3171                 }
3172         }
3173
3174         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3175                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3176         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3177                                                       has_huge_files);
3178         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3179
3180         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3181                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3182                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3183         } else {
3184                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3185                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3186                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3187                     (!is_power_of_2(sbi->s_inode_size)) ||
3188                     (sbi->s_inode_size > blocksize)) {
3189                         ext4_msg(sb, KERN_ERR,
3190                                "unsupported inode size: %d",
3191                                sbi->s_inode_size);
3192                         goto failed_mount;
3193                 }
3194                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3195                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3196         }
3197
3198         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3199         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3200                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3201                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3202                     !is_power_of_2(sbi->s_desc_size)) {
3203                         ext4_msg(sb, KERN_ERR,
3204                                "unsupported descriptor size %lu",
3205                                sbi->s_desc_size);
3206                         goto failed_mount;
3207                 }
3208         } else
3209                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3210
3211         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3212         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3213         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3214                 goto cantfind_ext4;
3215
3216         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3217         if (sbi->s_inodes_per_block == 0)
3218                 goto cantfind_ext4;
3219         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3220                                         sbi->s_inodes_per_block;
3221         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3222         sbi->s_sbh = bh;
3223         sbi->s_mount_state = le16_to_cpu(es->s_state);
3224         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3225         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3226
3227         for (i = 0; i < 4; i++)
3228                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3229         sbi->s_def_hash_version = es->s_def_hash_version;
3230         i = le32_to_cpu(es->s_flags);
3231         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3232                 sbi->s_hash_unsigned = 3;
3233         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3234 #ifdef __CHAR_UNSIGNED__
3235                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3236                 sbi->s_hash_unsigned = 3;
3237 #else
3238                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3239 #endif
3240                 sb->s_dirt = 1;
3241         }
3242
3243         if (sbi->s_blocks_per_group > blocksize * 8) {
3244                 ext4_msg(sb, KERN_ERR,
3245                        "#blocks per group too big: %lu",
3246                        sbi->s_blocks_per_group);
3247                 goto failed_mount;
3248         }
3249         if (sbi->s_inodes_per_group > blocksize * 8) {
3250                 ext4_msg(sb, KERN_ERR,
3251                        "#inodes per group too big: %lu",
3252                        sbi->s_inodes_per_group);
3253                 goto failed_mount;
3254         }
3255
3256         /*
3257          * Test whether we have more sectors than will fit in sector_t,
3258          * and whether the max offset is addressable by the page cache.
3259          */
3260         ret = generic_check_addressable(sb->s_blocksize_bits,
3261                                         ext4_blocks_count(es));
3262         if (ret) {
3263                 ext4_msg(sb, KERN_ERR, "filesystem"
3264                          " too large to mount safely on this system");
3265                 if (sizeof(sector_t) < 8)
3266                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3267                 goto failed_mount;
3268         }
3269
3270         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3271                 goto cantfind_ext4;
3272
3273         /* check blocks count against device size */
3274         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3275         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3276                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3277                        "exceeds size of device (%llu blocks)",
3278                        ext4_blocks_count(es), blocks_count);
3279                 goto failed_mount;
3280         }
3281
3282         /*
3283          * It makes no sense for the first data block to be beyond the end
3284          * of the filesystem.
3285          */
3286         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3287                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3288                          "block %u is beyond end of filesystem (%llu)",
3289                          le32_to_cpu(es->s_first_data_block),
3290                          ext4_blocks_count(es));
3291                 goto failed_mount;
3292         }
3293         blocks_count = (ext4_blocks_count(es) -
3294                         le32_to_cpu(es->s_first_data_block) +
3295                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3296         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3297         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3298                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3299                        "(block count %llu, first data block %u, "
3300                        "blocks per group %lu)", sbi->s_groups_count,
3301                        ext4_blocks_count(es),
3302                        le32_to_cpu(es->s_first_data_block),
3303                        EXT4_BLOCKS_PER_GROUP(sb));
3304                 goto failed_mount;
3305         }
3306         sbi->s_groups_count = blocks_count;
3307         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3308                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3309         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3310                    EXT4_DESC_PER_BLOCK(sb);
3311         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3312                                     GFP_KERNEL);
3313         if (sbi->s_group_desc == NULL) {
3314                 ext4_msg(sb, KERN_ERR, "not enough memory");
3315                 goto failed_mount;
3316         }
3317
3318 #ifdef CONFIG_PROC_FS
3319         if (ext4_proc_root)
3320                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3321 #endif
3322
3323         bgl_lock_init(sbi->s_blockgroup_lock);
3324
3325         for (i = 0; i < db_count; i++) {
3326                 block = descriptor_loc(sb, logical_sb_block, i);
3327                 sbi->s_group_desc[i] = sb_bread(sb, block);
3328                 if (!sbi->s_group_desc[i]) {
3329                         ext4_msg(sb, KERN_ERR,
3330                                "can't read group descriptor %d", i);
3331                         db_count = i;
3332                         goto failed_mount2;
3333                 }
3334         }
3335         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3336                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3337                 goto failed_mount2;
3338         }
3339         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3340                 if (!ext4_fill_flex_info(sb)) {
3341                         ext4_msg(sb, KERN_ERR,
3342                                "unable to initialize "
3343                                "flex_bg meta info!");
3344                         goto failed_mount2;
3345                 }
3346
3347         sbi->s_gdb_count = db_count;
3348         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3349         spin_lock_init(&sbi->s_next_gen_lock);
3350
3351         sbi->s_stripe = ext4_get_stripe_size(sbi);
3352         sbi->s_max_writeback_mb_bump = 128;
3353
3354         /*
3355          * set up enough so that it can read an inode
3356          */
3357         if (!test_opt(sb, NOLOAD) &&
3358             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3359                 sb->s_op = &ext4_sops;
3360         else
3361                 sb->s_op = &ext4_nojournal_sops;
3362         sb->s_export_op = &ext4_export_ops;
3363         sb->s_xattr = ext4_xattr_handlers;
3364 #ifdef CONFIG_QUOTA
3365         sb->s_qcop = &ext4_qctl_operations;
3366         sb->dq_op = &ext4_quota_operations;
3367 #endif
3368         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3369         mutex_init(&sbi->s_orphan_lock);
3370         mutex_init(&sbi->s_resize_lock);
3371
3372         sb->s_root = NULL;
3373
3374         needs_recovery = (es->s_last_orphan != 0 ||
3375                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3376                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3377
3378         /*
3379          * The first inode we look at is the journal inode.  Don't try
3380          * root first: it may be modified in the journal!
3381          */
3382         if (!test_opt(sb, NOLOAD) &&
3383             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3384                 if (ext4_load_journal(sb, es, journal_devnum))
3385                         goto failed_mount3;
3386         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3387               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3388                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3389                        "suppressed and not mounted read-only");
3390                 goto failed_mount_wq;
3391         } else {
3392                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
3393                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3394                 sbi->s_journal = NULL;
3395                 needs_recovery = 0;
3396                 goto no_journal;
3397         }
3398
3399         if (ext4_blocks_count(es) > 0xffffffffULL &&
3400             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3401                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3402                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3403                 goto failed_mount_wq;
3404         }
3405
3406         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3407                 jbd2_journal_set_features(sbi->s_journal,
3408                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3409                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3410         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3411                 jbd2_journal_set_features(sbi->s_journal,
3412                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3413                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3414                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3415         } else {
3416                 jbd2_journal_clear_features(sbi->s_journal,
3417                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3418                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3419         }
3420
3421         /* We have now updated the journal if required, so we can
3422          * validate the data journaling mode. */
3423         switch (test_opt(sb, DATA_FLAGS)) {
3424         case 0:
3425                 /* No mode set, assume a default based on the journal
3426                  * capabilities: ORDERED_DATA if the journal can
3427                  * cope, else JOURNAL_DATA
3428                  */
3429                 if (jbd2_journal_check_available_features
3430                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3431                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
3432                 else
3433                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3434                 break;
3435
3436         case EXT4_MOUNT_ORDERED_DATA:
3437         case EXT4_MOUNT_WRITEBACK_DATA:
3438                 if (!jbd2_journal_check_available_features
3439                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3440                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3441                                "requested data journaling mode");
3442                         goto failed_mount_wq;
3443                 }
3444         default:
3445                 break;
3446         }
3447         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3448
3449 no_journal:
3450         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3451                                   ext4_count_free_blocks(sb));
3452         if (!err)
3453                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3454                                           ext4_count_free_inodes(sb));
3455         if (!err)
3456                 err = percpu_counter_init(&sbi->s_dirs_counter,
3457                                           ext4_count_dirs(sb));
3458         if (!err)
3459                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3460         if (err) {
3461                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3462                 goto failed_mount_wq;
3463         }
3464
3465         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3466         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3467                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3468                 goto failed_mount_wq;
3469         }
3470
3471         /*
3472          * The jbd2_journal_load will have done any necessary log recovery,
3473          * so we can safely mount the rest of the filesystem now.
3474          */
3475
3476         root = ext4_iget(sb, EXT4_ROOT_INO);
3477         if (IS_ERR(root)) {
3478                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3479                 ret = PTR_ERR(root);
3480                 goto failed_mount4;
3481         }
3482         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3483                 iput(root);
3484                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3485                 goto failed_mount4;
3486         }
3487         sb->s_root = d_alloc_root(root);
3488         if (!sb->s_root) {
3489                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3490                 iput(root);
3491                 ret = -ENOMEM;
3492                 goto failed_mount4;
3493         }
3494
3495         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3496
3497         /* determine the minimum size of new large inodes, if present */
3498         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3499                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3500                                                      EXT4_GOOD_OLD_INODE_SIZE;
3501                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3502                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3503                         if (sbi->s_want_extra_isize <
3504                             le16_to_cpu(es->s_want_extra_isize))
3505                                 sbi->s_want_extra_isize =
3506                                         le16_to_cpu(es->s_want_extra_isize);
3507                         if (sbi->s_want_extra_isize <
3508                             le16_to_cpu(es->s_min_extra_isize))
3509                                 sbi->s_want_extra_isize =
3510                                         le16_to_cpu(es->s_min_extra_isize);
3511                 }
3512         }
3513         /* Check if enough inode space is available */
3514         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3515                                                         sbi->s_inode_size) {
3516                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3517                                                        EXT4_GOOD_OLD_INODE_SIZE;
3518                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3519                          "available");
3520         }
3521
3522         if (test_opt(sb, DELALLOC) &&
3523             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3524                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3525                          "requested data journaling mode");
3526                 clear_opt(sbi->s_mount_opt, DELALLOC);
3527         }
3528         if (test_opt(sb, DIOREAD_NOLOCK)) {
3529                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3530                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3531                                 "option - requested data journaling mode");
3532                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3533                 }
3534                 if (sb->s_blocksize < PAGE_SIZE) {
3535                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3536                                 "option - block size is too small");
3537                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3538                 }
3539         }
3540
3541         err = ext4_setup_system_zone(sb);
3542         if (err) {
3543                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3544                          "zone (%d)", err);
3545                 goto failed_mount4;
3546         }
3547
3548         ext4_ext_init(sb);
3549         err = ext4_mb_init(sb, needs_recovery);
3550         if (err) {
3551                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3552                          err);
3553                 goto failed_mount4;
3554         }
3555
3556         err = ext4_register_li_request(sb, first_not_zeroed);
3557         if (err)
3558                 goto failed_mount4;
3559
3560         sbi->s_kobj.kset = ext4_kset;
3561         init_completion(&sbi->s_kobj_unregister);
3562         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3563                                    "%s", sb->s_id);
3564         if (err) {
3565                 ext4_mb_release(sb);
3566                 ext4_ext_release(sb);
3567                 goto failed_mount4;
3568         };
3569
3570         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3571         ext4_orphan_cleanup(sb, es);
3572         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3573         if (needs_recovery) {
3574                 ext4_msg(sb, KERN_INFO, "recovery complete");
3575                 ext4_mark_recovery_complete(sb, es);
3576         }
3577         if (EXT4_SB(sb)->s_journal) {
3578                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3579                         descr = " journalled data mode";
3580                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3581                         descr = " ordered data mode";
3582                 else
3583                         descr = " writeback data mode";
3584         } else
3585                 descr = "out journal";
3586
3587         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3588                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3589                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3590
3591         init_timer(&sbi->s_err_report);
3592         sbi->s_err_report.function = print_daily_error_info;
3593         sbi->s_err_report.data = (unsigned long) sb;
3594         if (es->s_error_count)
3595                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3596
3597         kfree(orig_data);
3598         return 0;
3599
3600 cantfind_ext4:
3601         if (!silent)
3602                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3603         goto failed_mount;
3604
3605 failed_mount4:
3606         ext4_msg(sb, KERN_ERR, "mount failed");
3607         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3608 failed_mount_wq:
3609         ext4_release_system_zone(sb);
3610         if (sbi->s_journal) {
3611                 jbd2_journal_destroy(sbi->s_journal);
3612                 sbi->s_journal = NULL;
3613         }
3614         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3615         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3616         percpu_counter_destroy(&sbi->s_dirs_counter);
3617         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3618 failed_mount3:
3619         if (sbi->s_flex_groups) {
3620                 if (is_vmalloc_addr(sbi->s_flex_groups))
3621                         vfree(sbi->s_flex_groups);
3622                 else
3623                         kfree(sbi->s_flex_groups);
3624         }
3625 failed_mount2:
3626         for (i = 0; i < db_count; i++)
3627                 brelse(sbi->s_group_desc[i]);
3628         kfree(sbi->s_group_desc);
3629 failed_mount:
3630         if (sbi->s_proc) {
3631                 remove_proc_entry(sb->s_id, ext4_proc_root);
3632         }
3633 #ifdef CONFIG_QUOTA
3634         for (i = 0; i < MAXQUOTAS; i++)
3635                 kfree(sbi->s_qf_names[i]);
3636 #endif
3637         ext4_blkdev_remove(sbi);
3638         brelse(bh);
3639 out_fail:
3640         sb->s_fs_info = NULL;
3641         kfree(sbi->s_blockgroup_lock);
3642         kfree(sbi);
3643 out_free_orig:
3644         kfree(orig_data);
3645         return ret;
3646 }
3647
3648 /*
3649  * Setup any per-fs journal parameters now.  We'll do this both on
3650  * initial mount, once the journal has been initialised but before we've
3651  * done any recovery; and again on any subsequent remount.
3652  */
3653 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3654 {
3655         struct ext4_sb_info *sbi = EXT4_SB(sb);
3656
3657         journal->j_commit_interval = sbi->s_commit_interval;
3658         journal->j_min_batch_time = sbi->s_min_batch_time;
3659         journal->j_max_batch_time = sbi->s_max_batch_time;
3660
3661         write_lock(&journal->j_state_lock);
3662         if (test_opt(sb, BARRIER))
3663                 journal->j_flags |= JBD2_BARRIER;
3664         else
3665                 journal->j_flags &= ~JBD2_BARRIER;
3666         if (test_opt(sb, DATA_ERR_ABORT))
3667                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3668         else
3669                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3670         write_unlock(&journal->j_state_lock);
3671 }
3672
3673 static journal_t *ext4_get_journal(struct super_block *sb,
3674                                    unsigned int journal_inum)
3675 {
3676         struct inode *journal_inode;
3677         journal_t *journal;
3678
3679         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3680
3681         /* First, test for the existence of a valid inode on disk.  Bad
3682          * things happen if we iget() an unused inode, as the subsequent
3683          * iput() will try to delete it. */
3684
3685         journal_inode = ext4_iget(sb, journal_inum);
3686         if (IS_ERR(journal_inode)) {
3687                 ext4_msg(sb, KERN_ERR, "no journal found");
3688                 return NULL;
3689         }
3690         if (!journal_inode->i_nlink) {
3691                 make_bad_inode(journal_inode);
3692                 iput(journal_inode);
3693                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3694                 return NULL;
3695         }
3696
3697         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3698                   journal_inode, journal_inode->i_size);
3699         if (!S_ISREG(journal_inode->i_mode)) {
3700                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3701                 iput(journal_inode);
3702                 return NULL;
3703         }
3704
3705         journal = jbd2_journal_init_inode(journal_inode);
3706         if (!journal) {
3707                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3708                 iput(journal_inode);
3709                 return NULL;
3710         }
3711         journal->j_private = sb;
3712         ext4_init_journal_params(sb, journal);
3713         return journal;
3714 }
3715
3716 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3717                                        dev_t j_dev)
3718 {
3719         struct buffer_head *bh;
3720         journal_t *journal;
3721         ext4_fsblk_t start;
3722         ext4_fsblk_t len;
3723         int hblock, blocksize;
3724         ext4_fsblk_t sb_block;
3725         unsigned long offset;
3726         struct ext4_super_block *es;
3727         struct block_device *bdev;
3728
3729         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3730
3731         bdev = ext4_blkdev_get(j_dev, sb);
3732         if (bdev == NULL)
3733                 return NULL;
3734
3735         if (bd_claim(bdev, sb)) {
3736                 ext4_msg(sb, KERN_ERR,
3737                         "failed to claim external journal device");
3738                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3739                 return NULL;
3740         }
3741
3742         blocksize = sb->s_blocksize;
3743         hblock = bdev_logical_block_size(bdev);
3744         if (blocksize < hblock) {
3745                 ext4_msg(sb, KERN_ERR,
3746                         "blocksize too small for journal device");
3747                 goto out_bdev;
3748         }
3749
3750         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3751         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3752         set_blocksize(bdev, blocksize);
3753         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3754                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3755                        "external journal");
3756                 goto out_bdev;
3757         }
3758
3759         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3760         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3761             !(le32_to_cpu(es->s_feature_incompat) &
3762               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3763                 ext4_msg(sb, KERN_ERR, "external journal has "
3764                                         "bad superblock");
3765                 brelse(bh);
3766                 goto out_bdev;
3767         }
3768
3769         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3770                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3771                 brelse(bh);
3772                 goto out_bdev;
3773         }
3774
3775         len = ext4_blocks_count(es);
3776         start = sb_block + 1;
3777         brelse(bh);     /* we're done with the superblock */
3778
3779         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3780                                         start, len, blocksize);
3781         if (!journal) {
3782                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3783                 goto out_bdev;
3784         }
3785         journal->j_private = sb;
3786         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3787         wait_on_buffer(journal->j_sb_buffer);
3788         if (!buffer_uptodate(journal->j_sb_buffer)) {
3789                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3790                 goto out_journal;
3791         }
3792         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3793                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3794                                         "user (unsupported) - %d",
3795                         be32_to_cpu(journal->j_superblock->s_nr_users));
3796                 goto out_journal;
3797         }
3798         EXT4_SB(sb)->journal_bdev = bdev;
3799         ext4_init_journal_params(sb, journal);
3800         return journal;
3801
3802 out_journal:
3803         jbd2_journal_destroy(journal);
3804 out_bdev:
3805         ext4_blkdev_put(bdev);
3806         return NULL;
3807 }
3808
3809 static int ext4_load_journal(struct super_block *sb,
3810                              struct ext4_super_block *es,
3811                              unsigned long journal_devnum)
3812 {
3813         journal_t *journal;
3814         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3815         dev_t journal_dev;
3816         int err = 0;
3817         int really_read_only;
3818
3819         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3820
3821         if (journal_devnum &&
3822             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3823                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3824                         "numbers have changed");
3825                 journal_dev = new_decode_dev(journal_devnum);
3826         } else
3827                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3828
3829         really_read_only = bdev_read_only(sb->s_bdev);
3830
3831         /*
3832          * Are we loading a blank journal or performing recovery after a
3833          * crash?  For recovery, we need to check in advance whether we
3834          * can get read-write access to the device.
3835          */
3836         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3837                 if (sb->s_flags & MS_RDONLY) {
3838                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3839                                         "required on readonly filesystem");
3840                         if (really_read_only) {
3841                                 ext4_msg(sb, KERN_ERR, "write access "
3842                                         "unavailable, cannot proceed");
3843                                 return -EROFS;
3844                         }
3845                         ext4_msg(sb, KERN_INFO, "write access will "
3846                                "be enabled during recovery");
3847                 }
3848         }
3849
3850         if (journal_inum && journal_dev) {
3851                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3852                        "and inode journals!");
3853                 return -EINVAL;
3854         }
3855
3856         if (journal_inum) {
3857                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3858                         return -EINVAL;
3859         } else {
3860                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3861                         return -EINVAL;
3862         }
3863
3864         if (!(journal->j_flags & JBD2_BARRIER))
3865                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3866
3867         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3868                 err = jbd2_journal_update_format(journal);
3869                 if (err)  {
3870                         ext4_msg(sb, KERN_ERR, "error updating journal");
3871                         jbd2_journal_destroy(journal);
3872                         return err;
3873                 }
3874         }
3875
3876         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3877                 err = jbd2_journal_wipe(journal, !really_read_only);
3878         if (!err) {
3879                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3880                 if (save)
3881                         memcpy(save, ((char *) es) +
3882                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3883                 err = jbd2_journal_load(journal);
3884                 if (save)
3885                         memcpy(((char *) es) + EXT4_S_ERR_START,
3886                                save, EXT4_S_ERR_LEN);
3887                 kfree(save);
3888         }
3889
3890         if (err) {
3891                 ext4_msg(sb, KERN_ERR, "error loading journal");
3892                 jbd2_journal_destroy(journal);
3893                 return err;
3894         }
3895
3896         EXT4_SB(sb)->s_journal = journal;
3897         ext4_clear_journal_err(sb, es);
3898
3899         if (!really_read_only && journal_devnum &&
3900             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3901                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3902
3903                 /* Make sure we flush the recovery flag to disk. */
3904                 ext4_commit_super(sb, 1);
3905         }
3906
3907         return 0;
3908 }
3909
3910 static int ext4_commit_super(struct super_block *sb, int sync)
3911 {
3912         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3913         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3914         int error = 0;
3915
3916         if (!sbh)
3917                 return error;
3918         if (buffer_write_io_error(sbh)) {
3919                 /*
3920                  * Oh, dear.  A previous attempt to write the
3921                  * superblock failed.  This could happen because the
3922                  * USB device was yanked out.  Or it could happen to
3923                  * be a transient write error and maybe the block will
3924                  * be remapped.  Nothing we can do but to retry the
3925                  * write and hope for the best.
3926                  */
3927                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3928                        "superblock detected");
3929                 clear_buffer_write_io_error(sbh);
3930                 set_buffer_uptodate(sbh);
3931         }
3932         /*
3933          * If the file system is mounted read-only, don't update the
3934          * superblock write time.  This avoids updating the superblock
3935          * write time when we are mounting the root file system
3936          * read/only but we need to replay the journal; at that point,
3937          * for people who are east of GMT and who make their clock
3938          * tick in localtime for Windows bug-for-bug compatibility,
3939          * the clock is set in the future, and this will cause e2fsck
3940          * to complain and force a full file system check.
3941          */
3942         if (!(sb->s_flags & MS_RDONLY))
3943                 es->s_wtime = cpu_to_le32(get_seconds());
3944         if (sb->s_bdev->bd_part)
3945                 es->s_kbytes_written =
3946                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3947                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3948                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3949         else
3950                 es->s_kbytes_written =
3951                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3952         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeblocks_counter))
3953                 ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3954                                         &EXT4_SB(sb)->s_freeblocks_counter));
3955         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
3956                 es->s_free_inodes_count =
3957                         cpu_to_le32(percpu_counter_sum_positive(
3958                                         &EXT4_SB(sb)->s_freeinodes_counter));
3959         sb->s_dirt = 0;
3960         BUFFER_TRACE(sbh, "marking dirty");
3961         mark_buffer_dirty(sbh);
3962         if (sync) {
3963                 error = sync_dirty_buffer(sbh);
3964                 if (error)
3965                         return error;
3966
3967                 error = buffer_write_io_error(sbh);
3968                 if (error) {
3969                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3970                                "superblock");
3971                         clear_buffer_write_io_error(sbh);
3972                         set_buffer_uptodate(sbh);
3973                 }
3974         }
3975         return error;
3976 }
3977
3978 /*
3979  * Have we just finished recovery?  If so, and if we are mounting (or
3980  * remounting) the filesystem readonly, then we will end up with a
3981  * consistent fs on disk.  Record that fact.
3982  */
3983 static void ext4_mark_recovery_complete(struct super_block *sb,
3984                                         struct ext4_super_block *es)
3985 {
3986         journal_t *journal = EXT4_SB(sb)->s_journal;
3987
3988         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3989                 BUG_ON(journal != NULL);
3990                 return;
3991         }
3992         jbd2_journal_lock_updates(journal);
3993         if (jbd2_journal_flush(journal) < 0)
3994                 goto out;
3995
3996         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3997             sb->s_flags & MS_RDONLY) {
3998                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3999                 ext4_commit_super(sb, 1);
4000         }
4001
4002 out:
4003         jbd2_journal_unlock_updates(journal);
4004 }
4005
4006 /*
4007  * If we are mounting (or read-write remounting) a filesystem whose journal
4008  * has recorded an error from a previous lifetime, move that error to the
4009  * main filesystem now.
4010  */
4011 static void ext4_clear_journal_err(struct super_block *sb,
4012                                    struct ext4_super_block *es)
4013 {
4014         journal_t *journal;
4015         int j_errno;
4016         const char *errstr;
4017
4018         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4019
4020         journal = EXT4_SB(sb)->s_journal;
4021
4022         /*
4023          * Now check for any error status which may have been recorded in the
4024          * journal by a prior ext4_error() or ext4_abort()
4025          */
4026
4027         j_errno = jbd2_journal_errno(journal);
4028         if (j_errno) {
4029                 char nbuf[16];
4030
4031                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4032                 ext4_warning(sb, "Filesystem error recorded "
4033                              "from previous mount: %s", errstr);
4034                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4035
4036                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4037                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4038                 ext4_commit_super(sb, 1);
4039
4040                 jbd2_journal_clear_err(journal);
4041         }
4042 }
4043
4044 /*
4045  * Force the running and committing transactions to commit,
4046  * and wait on the commit.
4047  */
4048 int ext4_force_commit(struct super_block *sb)
4049 {
4050         journal_t *journal;
4051         int ret = 0;
4052
4053         if (sb->s_flags & MS_RDONLY)
4054                 return 0;
4055
4056         journal = EXT4_SB(sb)->s_journal;
4057         if (journal) {
4058                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4059                 ret = ext4_journal_force_commit(journal);
4060         }
4061
4062         return ret;
4063 }
4064
4065 static void ext4_write_super(struct super_block *sb)
4066 {
4067         lock_super(sb);
4068         ext4_commit_super(sb, 1);
4069         unlock_super(sb);
4070 }
4071
4072 static int ext4_sync_fs(struct super_block *sb, int wait)
4073 {
4074         int ret = 0;
4075         tid_t target;
4076         struct ext4_sb_info *sbi = EXT4_SB(sb);
4077
4078         trace_ext4_sync_fs(sb, wait);
4079         flush_workqueue(sbi->dio_unwritten_wq);
4080         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4081                 if (wait)
4082                         jbd2_log_wait_commit(sbi->s_journal, target);
4083         }
4084         return ret;
4085 }
4086
4087 /*
4088  * LVM calls this function before a (read-only) snapshot is created.  This
4089  * gives us a chance to flush the journal completely and mark the fs clean.
4090  */
4091 static int ext4_freeze(struct super_block *sb)
4092 {
4093         int error = 0;
4094         journal_t *journal;
4095
4096         if (sb->s_flags & MS_RDONLY)
4097                 return 0;
4098
4099         journal = EXT4_SB(sb)->s_journal;
4100
4101         /* Now we set up the journal barrier. */
4102         jbd2_journal_lock_updates(journal);
4103
4104         /*
4105          * Don't clear the needs_recovery flag if we failed to flush
4106          * the journal.
4107          */
4108         error = jbd2_journal_flush(journal);
4109         if (error < 0)
4110                 goto out;
4111
4112         /* Journal blocked and flushed, clear needs_recovery flag. */
4113         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4114         error = ext4_commit_super(sb, 1);
4115 out:
4116         /* we rely on s_frozen to stop further updates */
4117         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4118         return error;
4119 }
4120
4121 /*
4122  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4123  * flag here, even though the filesystem is not technically dirty yet.
4124  */
4125 static int ext4_unfreeze(struct super_block *sb)
4126 {
4127         if (sb->s_flags & MS_RDONLY)
4128                 return 0;
4129
4130         lock_super(sb);
4131         /* Reset the needs_recovery flag before the fs is unlocked. */
4132         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4133         ext4_commit_super(sb, 1);
4134         unlock_super(sb);
4135         return 0;
4136 }
4137
4138 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4139 {
4140         struct ext4_super_block *es;
4141         struct ext4_sb_info *sbi = EXT4_SB(sb);
4142         ext4_fsblk_t n_blocks_count = 0;
4143         unsigned long old_sb_flags;
4144         struct ext4_mount_options old_opts;
4145         int enable_quota = 0;
4146         ext4_group_t g;
4147         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4148         int err;
4149 #ifdef CONFIG_QUOTA
4150         int i;
4151 #endif
4152         char *orig_data = kstrdup(data, GFP_KERNEL);
4153
4154         /* Store the original options */
4155         lock_super(sb);
4156         old_sb_flags = sb->s_flags;
4157         old_opts.s_mount_opt = sbi->s_mount_opt;
4158         old_opts.s_resuid = sbi->s_resuid;
4159         old_opts.s_resgid = sbi->s_resgid;
4160         old_opts.s_commit_interval = sbi->s_commit_interval;
4161         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4162         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4163 #ifdef CONFIG_QUOTA
4164         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4165         for (i = 0; i < MAXQUOTAS; i++)
4166                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4167 #endif
4168         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4169                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4170
4171         /*
4172          * Allow the "check" option to be passed as a remount option.
4173          */
4174         if (!parse_options(data, sb, NULL, &journal_ioprio,
4175                            &n_blocks_count, 1)) {
4176                 err = -EINVAL;
4177                 goto restore_opts;
4178         }
4179
4180         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4181                 ext4_abort(sb, "Abort forced by user");
4182
4183         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4184                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4185
4186         es = sbi->s_es;
4187
4188         if (sbi->s_journal) {
4189                 ext4_init_journal_params(sb, sbi->s_journal);
4190                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4191         }
4192
4193         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4194                 n_blocks_count > ext4_blocks_count(es)) {
4195                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4196                         err = -EROFS;
4197                         goto restore_opts;
4198                 }
4199
4200                 if (*flags & MS_RDONLY) {
4201                         err = dquot_suspend(sb, -1);
4202                         if (err < 0)
4203                                 goto restore_opts;
4204
4205                         /*
4206                          * First of all, the unconditional stuff we have to do
4207                          * to disable replay of the journal when we next remount
4208                          */
4209                         sb->s_flags |= MS_RDONLY;
4210
4211                         /*
4212                          * OK, test if we are remounting a valid rw partition
4213                          * readonly, and if so set the rdonly flag and then
4214                          * mark the partition as valid again.
4215                          */
4216                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4217                             (sbi->s_mount_state & EXT4_VALID_FS))
4218                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4219
4220                         if (sbi->s_journal)
4221                                 ext4_mark_recovery_complete(sb, es);
4222                 } else {
4223                         /* Make sure we can mount this feature set readwrite */
4224                         if (!ext4_feature_set_ok(sb, 0)) {
4225                                 err = -EROFS;
4226                                 goto restore_opts;
4227                         }
4228                         /*
4229                          * Make sure the group descriptor checksums
4230                          * are sane.  If they aren't, refuse to remount r/w.
4231                          */
4232                         for (g = 0; g < sbi->s_groups_count; g++) {
4233                                 struct ext4_group_desc *gdp =
4234                                         ext4_get_group_desc(sb, g, NULL);
4235
4236                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4237                                         ext4_msg(sb, KERN_ERR,
4238                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4239                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4240                                                le16_to_cpu(gdp->bg_checksum));
4241                                         err = -EINVAL;
4242                                         goto restore_opts;
4243                                 }
4244                         }
4245
4246                         /*
4247                          * If we have an unprocessed orphan list hanging
4248                          * around from a previously readonly bdev mount,
4249                          * require a full umount/remount for now.
4250                          */
4251                         if (es->s_last_orphan) {
4252                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4253                                        "remount RDWR because of unprocessed "
4254                                        "orphan inode list.  Please "
4255                                        "umount/remount instead");
4256                                 err = -EINVAL;
4257                                 goto restore_opts;
4258                         }
4259
4260                         /*
4261                          * Mounting a RDONLY partition read-write, so reread
4262                          * and store the current valid flag.  (It may have
4263                          * been changed by e2fsck since we originally mounted
4264                          * the partition.)
4265                          */
4266                         if (sbi->s_journal)
4267                                 ext4_clear_journal_err(sb, es);
4268                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4269                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4270                                 goto restore_opts;
4271                         if (!ext4_setup_super(sb, es, 0))
4272                                 sb->s_flags &= ~MS_RDONLY;
4273                         enable_quota = 1;
4274                 }
4275         }
4276
4277         /*
4278          * Reinitialize lazy itable initialization thread based on
4279          * current settings
4280          */
4281         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4282                 ext4_unregister_li_request(sb);
4283         else {
4284                 ext4_group_t first_not_zeroed;
4285                 first_not_zeroed = ext4_has_uninit_itable(sb);
4286                 ext4_register_li_request(sb, first_not_zeroed);
4287         }
4288
4289         ext4_setup_system_zone(sb);
4290         if (sbi->s_journal == NULL)
4291                 ext4_commit_super(sb, 1);
4292
4293 #ifdef CONFIG_QUOTA
4294         /* Release old quota file names */
4295         for (i = 0; i < MAXQUOTAS; i++)
4296                 if (old_opts.s_qf_names[i] &&
4297                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4298                         kfree(old_opts.s_qf_names[i]);
4299 #endif
4300         unlock_super(sb);
4301         if (enable_quota)
4302                 dquot_resume(sb, -1);
4303
4304         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4305         kfree(orig_data);
4306         return 0;
4307
4308 restore_opts:
4309         sb->s_flags = old_sb_flags;
4310         sbi->s_mount_opt = old_opts.s_mount_opt;
4311         sbi->s_resuid = old_opts.s_resuid;
4312         sbi->s_resgid = old_opts.s_resgid;
4313         sbi->s_commit_interval = old_opts.s_commit_interval;
4314         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4315         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4316 #ifdef CONFIG_QUOTA
4317         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4318         for (i = 0; i < MAXQUOTAS; i++) {
4319                 if (sbi->s_qf_names[i] &&
4320                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4321                         kfree(sbi->s_qf_names[i]);
4322                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4323         }
4324 #endif
4325         unlock_super(sb);
4326         kfree(orig_data);
4327         return err;
4328 }
4329
4330 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4331 {
4332         struct super_block *sb = dentry->d_sb;
4333         struct ext4_sb_info *sbi = EXT4_SB(sb);
4334         struct ext4_super_block *es = sbi->s_es;
4335         u64 fsid;
4336
4337         if (test_opt(sb, MINIX_DF)) {
4338                 sbi->s_overhead_last = 0;
4339         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4340                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4341                 ext4_fsblk_t overhead = 0;
4342
4343                 /*
4344                  * Compute the overhead (FS structures).  This is constant
4345                  * for a given filesystem unless the number of block groups
4346                  * changes so we cache the previous value until it does.
4347                  */
4348
4349                 /*
4350                  * All of the blocks before first_data_block are
4351                  * overhead
4352                  */
4353                 overhead = le32_to_cpu(es->s_first_data_block);
4354
4355                 /*
4356                  * Add the overhead attributed to the superblock and
4357                  * block group descriptors.  If the sparse superblocks
4358                  * feature is turned on, then not all groups have this.
4359                  */
4360                 for (i = 0; i < ngroups; i++) {
4361                         overhead += ext4_bg_has_super(sb, i) +
4362                                 ext4_bg_num_gdb(sb, i);
4363                         cond_resched();
4364                 }
4365
4366                 /*
4367                  * Every block group has an inode bitmap, a block
4368                  * bitmap, and an inode table.
4369                  */
4370                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4371                 sbi->s_overhead_last = overhead;
4372                 smp_wmb();
4373                 sbi->s_blocks_last = ext4_blocks_count(es);
4374         }
4375
4376         buf->f_type = EXT4_SUPER_MAGIC;
4377         buf->f_bsize = sb->s_blocksize;
4378         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4379         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4380                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4381         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4382         if (buf->f_bfree < ext4_r_blocks_count(es))
4383                 buf->f_bavail = 0;
4384         buf->f_files = le32_to_cpu(es->s_inodes_count);
4385         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4386         buf->f_namelen = EXT4_NAME_LEN;
4387         fsid = le64_to_cpup((void *)es->s_uuid) ^
4388                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4389         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4390         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4391
4392         return 0;
4393 }
4394
4395 /* Helper function for writing quotas on sync - we need to start transaction
4396  * before quota file is locked for write. Otherwise the are possible deadlocks:
4397  * Process 1                         Process 2
4398  * ext4_create()                     quota_sync()
4399  *   jbd2_journal_start()                  write_dquot()
4400  *   dquot_initialize()                         down(dqio_mutex)
4401  *     down(dqio_mutex)                    jbd2_journal_start()
4402  *
4403  */
4404
4405 #ifdef CONFIG_QUOTA
4406
4407 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4408 {
4409         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4410 }
4411
4412 static int ext4_write_dquot(struct dquot *dquot)
4413 {
4414         int ret, err;
4415         handle_t *handle;
4416         struct inode *inode;
4417
4418         inode = dquot_to_inode(dquot);
4419         handle = ext4_journal_start(inode,
4420                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4421         if (IS_ERR(handle))
4422                 return PTR_ERR(handle);
4423         ret = dquot_commit(dquot);
4424         err = ext4_journal_stop(handle);
4425         if (!ret)
4426                 ret = err;
4427         return ret;
4428 }
4429
4430 static int ext4_acquire_dquot(struct dquot *dquot)
4431 {
4432         int ret, err;
4433         handle_t *handle;
4434
4435         handle = ext4_journal_start(dquot_to_inode(dquot),
4436                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4437         if (IS_ERR(handle))
4438                 return PTR_ERR(handle);
4439         ret = dquot_acquire(dquot);
4440         err = ext4_journal_stop(handle);
4441         if (!ret)
4442                 ret = err;
4443         return ret;
4444 }
4445
4446 static int ext4_release_dquot(struct dquot *dquot)
4447 {
4448         int ret, err;
4449         handle_t *handle;
4450
4451         handle = ext4_journal_start(dquot_to_inode(dquot),
4452                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4453         if (IS_ERR(handle)) {
4454                 /* Release dquot anyway to avoid endless cycle in dqput() */
4455                 dquot_release(dquot);
4456                 return PTR_ERR(handle);
4457         }
4458         ret = dquot_release(dquot);
4459         err = ext4_journal_stop(handle);
4460         if (!ret)
4461                 ret = err;
4462         return ret;
4463 }
4464
4465 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4466 {
4467         /* Are we journaling quotas? */
4468         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4469             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4470                 dquot_mark_dquot_dirty(dquot);
4471                 return ext4_write_dquot(dquot);
4472         } else {
4473                 return dquot_mark_dquot_dirty(dquot);
4474         }
4475 }
4476
4477 static int ext4_write_info(struct super_block *sb, int type)
4478 {
4479         int ret, err;
4480         handle_t *handle;
4481
4482         /* Data block + inode block */
4483         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4484         if (IS_ERR(handle))
4485                 return PTR_ERR(handle);
4486         ret = dquot_commit_info(sb, type);
4487         err = ext4_journal_stop(handle);
4488         if (!ret)
4489                 ret = err;
4490         return ret;
4491 }
4492
4493 /*
4494  * Turn on quotas during mount time - we need to find
4495  * the quota file and such...
4496  */
4497 static int ext4_quota_on_mount(struct super_block *sb, int type)
4498 {
4499         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4500                                         EXT4_SB(sb)->s_jquota_fmt, type);
4501 }
4502
4503 /*
4504  * Standard function to be called on quota_on
4505  */
4506 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4507                          char *name)
4508 {
4509         int err;
4510         struct path path;
4511
4512         if (!test_opt(sb, QUOTA))
4513                 return -EINVAL;
4514
4515         err = kern_path(name, LOOKUP_FOLLOW, &path);
4516         if (err)
4517                 return err;
4518
4519         /* Quotafile not on the same filesystem? */
4520         if (path.mnt->mnt_sb != sb) {
4521                 path_put(&path);
4522                 return -EXDEV;
4523         }
4524         /* Journaling quota? */
4525         if (EXT4_SB(sb)->s_qf_names[type]) {
4526                 /* Quotafile not in fs root? */
4527                 if (path.dentry->d_parent != sb->s_root)
4528                         ext4_msg(sb, KERN_WARNING,
4529                                 "Quota file not on filesystem root. "
4530                                 "Journaled quota will not work");
4531         }
4532
4533         /*
4534          * When we journal data on quota file, we have to flush journal to see
4535          * all updates to the file when we bypass pagecache...
4536          */
4537         if (EXT4_SB(sb)->s_journal &&
4538             ext4_should_journal_data(path.dentry->d_inode)) {
4539                 /*
4540                  * We don't need to lock updates but journal_flush() could
4541                  * otherwise be livelocked...
4542                  */
4543                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4544                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4545                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4546                 if (err) {
4547                         path_put(&path);
4548                         return err;
4549                 }
4550         }
4551
4552         err = dquot_quota_on_path(sb, type, format_id, &path);
4553         path_put(&path);
4554         return err;
4555 }
4556
4557 static int ext4_quota_off(struct super_block *sb, int type)
4558 {
4559         /* Force all delayed allocation blocks to be allocated */
4560         if (test_opt(sb, DELALLOC)) {
4561                 down_read(&sb->s_umount);
4562                 sync_filesystem(sb);
4563                 up_read(&sb->s_umount);
4564         }
4565
4566         return dquot_quota_off(sb, type);
4567 }
4568
4569 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4570  * acquiring the locks... As quota files are never truncated and quota code
4571  * itself serializes the operations (and noone else should touch the files)
4572  * we don't have to be afraid of races */
4573 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4574                                size_t len, loff_t off)
4575 {
4576         struct inode *inode = sb_dqopt(sb)->files[type];
4577         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4578         int err = 0;
4579         int offset = off & (sb->s_blocksize - 1);
4580         int tocopy;
4581         size_t toread;
4582         struct buffer_head *bh;
4583         loff_t i_size = i_size_read(inode);
4584
4585         if (off > i_size)
4586                 return 0;
4587         if (off+len > i_size)
4588                 len = i_size-off;
4589         toread = len;
4590         while (toread > 0) {
4591                 tocopy = sb->s_blocksize - offset < toread ?
4592                                 sb->s_blocksize - offset : toread;
4593                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4594                 if (err)
4595                         return err;
4596                 if (!bh)        /* A hole? */
4597                         memset(data, 0, tocopy);
4598                 else
4599                         memcpy(data, bh->b_data+offset, tocopy);
4600                 brelse(bh);
4601                 offset = 0;
4602                 toread -= tocopy;
4603                 data += tocopy;
4604                 blk++;
4605         }
4606         return len;
4607 }
4608
4609 /* Write to quotafile (we know the transaction is already started and has
4610  * enough credits) */
4611 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4612                                 const char *data, size_t len, loff_t off)
4613 {
4614         struct inode *inode = sb_dqopt(sb)->files[type];
4615         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4616         int err = 0;
4617         int offset = off & (sb->s_blocksize - 1);
4618         struct buffer_head *bh;
4619         handle_t *handle = journal_current_handle();
4620
4621         if (EXT4_SB(sb)->s_journal && !handle) {
4622                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4623                         " cancelled because transaction is not started",
4624                         (unsigned long long)off, (unsigned long long)len);
4625                 return -EIO;
4626         }
4627         /*
4628          * Since we account only one data block in transaction credits,
4629          * then it is impossible to cross a block boundary.
4630          */
4631         if (sb->s_blocksize - offset < len) {
4632                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4633                         " cancelled because not block aligned",
4634                         (unsigned long long)off, (unsigned long long)len);
4635                 return -EIO;
4636         }
4637
4638         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4639         bh = ext4_bread(handle, inode, blk, 1, &err);
4640         if (!bh)
4641                 goto out;
4642         err = ext4_journal_get_write_access(handle, bh);
4643         if (err) {
4644                 brelse(bh);
4645                 goto out;
4646         }
4647         lock_buffer(bh);
4648         memcpy(bh->b_data+offset, data, len);
4649         flush_dcache_page(bh->b_page);
4650         unlock_buffer(bh);
4651         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4652         brelse(bh);
4653 out:
4654         if (err) {
4655                 mutex_unlock(&inode->i_mutex);
4656                 return err;
4657         }
4658         if (inode->i_size < off + len) {
4659                 i_size_write(inode, off + len);
4660                 EXT4_I(inode)->i_disksize = inode->i_size;
4661         }
4662         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4663         ext4_mark_inode_dirty(handle, inode);
4664         mutex_unlock(&inode->i_mutex);
4665         return len;
4666 }
4667
4668 #endif
4669
4670 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4671                        const char *dev_name, void *data)
4672 {
4673         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4674 }
4675
4676 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4677 static struct file_system_type ext2_fs_type = {
4678         .owner          = THIS_MODULE,
4679         .name           = "ext2",
4680         .mount          = ext4_mount,
4681         .kill_sb        = kill_block_super,
4682         .fs_flags       = FS_REQUIRES_DEV,
4683 };
4684
4685 static inline void register_as_ext2(void)
4686 {
4687         int err = register_filesystem(&ext2_fs_type);
4688         if (err)
4689                 printk(KERN_WARNING
4690                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4691 }
4692
4693 static inline void unregister_as_ext2(void)
4694 {
4695         unregister_filesystem(&ext2_fs_type);
4696 }
4697 MODULE_ALIAS("ext2");
4698 #else
4699 static inline void register_as_ext2(void) { }
4700 static inline void unregister_as_ext2(void) { }
4701 #endif
4702
4703 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4704 static inline void register_as_ext3(void)
4705 {
4706         int err = register_filesystem(&ext3_fs_type);
4707         if (err)
4708                 printk(KERN_WARNING
4709                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4710 }
4711
4712 static inline void unregister_as_ext3(void)
4713 {
4714         unregister_filesystem(&ext3_fs_type);
4715 }
4716 MODULE_ALIAS("ext3");
4717 #else
4718 static inline void register_as_ext3(void) { }
4719 static inline void unregister_as_ext3(void) { }
4720 #endif
4721
4722 static struct file_system_type ext4_fs_type = {
4723         .owner          = THIS_MODULE,
4724         .name           = "ext4",
4725         .mount          = ext4_mount,
4726         .kill_sb        = kill_block_super,
4727         .fs_flags       = FS_REQUIRES_DEV,
4728 };
4729
4730 int __init ext4_init_feat_adverts(void)
4731 {
4732         struct ext4_features *ef;
4733         int ret = -ENOMEM;
4734
4735         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4736         if (!ef)
4737                 goto out;
4738
4739         ef->f_kobj.kset = ext4_kset;
4740         init_completion(&ef->f_kobj_unregister);
4741         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4742                                    "features");
4743         if (ret) {
4744                 kfree(ef);
4745                 goto out;
4746         }
4747
4748         ext4_feat = ef;
4749         ret = 0;
4750 out:
4751         return ret;
4752 }
4753
4754 static int __init ext4_init_fs(void)
4755 {
4756         int err;
4757
4758         ext4_check_flag_values();
4759         err = ext4_init_pageio();
4760         if (err)
4761                 return err;
4762         err = ext4_init_system_zone();
4763         if (err)
4764                 goto out5;
4765         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4766         if (!ext4_kset)
4767                 goto out4;
4768         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4769
4770         err = ext4_init_feat_adverts();
4771
4772         err = ext4_init_mballoc();
4773         if (err)
4774                 goto out3;
4775
4776         err = ext4_init_xattr();
4777         if (err)
4778                 goto out2;
4779         err = init_inodecache();
4780         if (err)
4781                 goto out1;
4782         register_as_ext2();
4783         register_as_ext3();
4784         err = register_filesystem(&ext4_fs_type);
4785         if (err)
4786                 goto out;
4787
4788         ext4_li_info = NULL;
4789         mutex_init(&ext4_li_mtx);
4790         return 0;
4791 out:
4792         unregister_as_ext2();
4793         unregister_as_ext3();
4794         destroy_inodecache();
4795 out1:
4796         ext4_exit_xattr();
4797 out2:
4798         ext4_exit_mballoc();
4799 out3:
4800         kfree(ext4_feat);
4801         remove_proc_entry("fs/ext4", NULL);
4802         kset_unregister(ext4_kset);
4803 out4:
4804         ext4_exit_system_zone();
4805 out5:
4806         ext4_exit_pageio();
4807         return err;
4808 }
4809
4810 static void __exit ext4_exit_fs(void)
4811 {
4812         ext4_destroy_lazyinit_thread();
4813         unregister_as_ext2();
4814         unregister_as_ext3();
4815         unregister_filesystem(&ext4_fs_type);
4816         destroy_inodecache();
4817         ext4_exit_xattr();
4818         ext4_exit_mballoc();
4819         remove_proc_entry("fs/ext4", NULL);
4820         kset_unregister(ext4_kset);
4821         ext4_exit_system_zone();
4822         ext4_exit_pageio();
4823 }
4824
4825 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4826 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4827 MODULE_LICENSE("GPL");
4828 module_init(ext4_init_fs)
4829 module_exit(ext4_exit_fs)