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 static struct ext4_lazy_init *ext4_li_info;
  58 static struct mutex ext4_li_mtx;
  59 static 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 int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  84 static struct file_system_type ext3_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext3",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  92 #else
  93 #define IS_EXT3_SB(sb) (0)
  94 #endif
  95
  96 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  97                                struct ext4_group_desc *bg)
  98 {
  99         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 100                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 101                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 102 }
 103
 104 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 105                                struct ext4_group_desc *bg)
 106 {
 107         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 108                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 109                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 110 }
 111
 112 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 113                               struct ext4_group_desc *bg)
 114 {
 115         return le32_to_cpu(bg->bg_inode_table_lo) |
 116                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 117                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 118 }
 119
 120 __u32 ext4_free_blks_count(struct super_block *sb,
 121                               struct ext4_group_desc *bg)
 122 {
 123         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 124                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 125                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 126 }
 127
 128 __u32 ext4_free_inodes_count(struct super_block *sb,
 129                               struct ext4_group_desc *bg)
 130 {
 131         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 132                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 133                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 134 }
 135
 136 __u32 ext4_used_dirs_count(struct super_block *sb,
 137                               struct ext4_group_desc *bg)
 138 {
 139         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 140                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 141                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 142 }
 143
 144 __u32 ext4_itable_unused_count(struct super_block *sb,
 145                               struct ext4_group_desc *bg)
 146 {
 147         return le16_to_cpu(bg->bg_itable_unused_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 150 }
 151
 152 void ext4_block_bitmap_set(struct super_block *sb,
 153                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 154 {
 155         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 156         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 157                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 158 }
 159
 160 void ext4_inode_bitmap_set(struct super_block *sb,
 161                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 162 {
 163         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 164         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 165                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 166 }
 167
 168 void ext4_inode_table_set(struct super_block *sb,
 169                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 170 {
 171         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 172         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 173                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 174 }
 175
 176 void ext4_free_blks_set(struct super_block *sb,
 177                           struct ext4_group_desc *bg, __u32 count)
 178 {
 179         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 180         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 181                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 182 }
 183
 184 void ext4_free_inodes_set(struct super_block *sb,
 185                           struct ext4_group_desc *bg, __u32 count)
 186 {
 187         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 188         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 189                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 190 }
 191
 192 void ext4_used_dirs_set(struct super_block *sb,
 193                           struct ext4_group_desc *bg, __u32 count)
 194 {
 195         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 196         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 197                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 198 }
 199
 200 void ext4_itable_unused_set(struct super_block *sb,
 201                           struct ext4_group_desc *bg, __u32 count)
 202 {
 203         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 206 }
 207
 208
 209 /* Just increment the non-pointer handle value */
 210 static handle_t *ext4_get_nojournal(void)
 211 {
 212         handle_t *handle = current->journal_info;
 213         unsigned long ref_cnt = (unsigned long)handle;
 214
 215         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 216
 217         ref_cnt++;
 218         handle = (handle_t *)ref_cnt;
 219
 220         current->journal_info = handle;
 221         return handle;
 222 }
 223
 224
 225 /* Decrement the non-pointer handle value */
 226 static void ext4_put_nojournal(handle_t *handle)
 227 {
 228         unsigned long ref_cnt = (unsigned long)handle;
 229
 230         BUG_ON(ref_cnt == 0);
 231
 232         ref_cnt--;
 233         handle = (handle_t *)ref_cnt;
 234
 235         current->journal_info = handle;
 236 }
 237
 238 /*
 239  * Wrappers for jbd2_journal_start/end.
 240  *
 241  * The only special thing we need to do here is to make sure that all
 242  * journal_end calls result in the superblock being marked dirty, so
 243  * that sync() will call the filesystem's write_super callback if
 244  * appropriate.
 245  */
 246 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 247 {
 248         journal_t *journal;
 249
 250         if (sb->s_flags & MS_RDONLY)
 251                 return ERR_PTR(-EROFS);
 252
 253         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 254         /* Special case here: if the journal has aborted behind our
 255          * backs (eg. EIO in the commit thread), then we still need to
 256          * take the FS itself readonly cleanly. */
 257         journal = EXT4_SB(sb)->s_journal;
 258         if (journal) {
 259                 if (is_journal_aborted(journal)) {
 260                         ext4_abort(sb, "Detected aborted journal");
 261                         return ERR_PTR(-EROFS);
 262                 }
 263                 return jbd2_journal_start(journal, nblocks);
 264         }
 265         return ext4_get_nojournal();
 266 }
 267
 268 /*
 269  * The only special thing we need to do here is to make sure that all
 270  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 271  * that sync() will call the filesystem's write_super callback if
 272  * appropriate.
 273  */
 274 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 275 {
 276         struct super_block *sb;
 277         int err;
 278         int rc;
 279
 280         if (!ext4_handle_valid(handle)) {
 281                 ext4_put_nojournal(handle);
 282                 return 0;
 283         }
 284         sb = handle->h_transaction->t_journal->j_private;
 285         err = handle->h_err;
 286         rc = jbd2_journal_stop(handle);
 287
 288         if (!err)
 289                 err = rc;
 290         if (err)
 291                 __ext4_std_error(sb, where, line, err);
 292         return err;
 293 }
 294
 295 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 296                                const char *err_fn, struct buffer_head *bh,
 297                                handle_t *handle, int err)
 298 {
 299         char nbuf[16];
 300         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 301
 302         BUG_ON(!ext4_handle_valid(handle));
 303
 304         if (bh)
 305                 BUFFER_TRACE(bh, "abort");
 306
 307         if (!handle->h_err)
 308                 handle->h_err = err;
 309
 310         if (is_handle_aborted(handle))
 311                 return;
 312
 313         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 314                caller, line, errstr, err_fn);
 315
 316         jbd2_journal_abort_handle(handle);
 317 }
 318
 319 static void __save_error_info(struct super_block *sb, const char *func,
 320                             unsigned int line)
 321 {
 322         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 323
 324         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 325         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 326         es->s_last_error_time = cpu_to_le32(get_seconds());
 327         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 328         es->s_last_error_line = cpu_to_le32(line);
 329         if (!es->s_first_error_time) {
 330                 es->s_first_error_time = es->s_last_error_time;
 331                 strncpy(es->s_first_error_func, func,
 332                         sizeof(es->s_first_error_func));
 333                 es->s_first_error_line = cpu_to_le32(line);
 334                 es->s_first_error_ino = es->s_last_error_ino;
 335                 es->s_first_error_block = es->s_last_error_block;
 336         }
 337         /*
 338          * Start the daily error reporting function if it hasn't been
 339          * started already
 340          */
 341         if (!es->s_error_count)
 342                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 343         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 344 }
 345
 346 static void save_error_info(struct super_block *sb, const char *func,
 347                             unsigned int line)
 348 {
 349         __save_error_info(sb, func, line);
 350         ext4_commit_super(sb, 1);
 351 }
 352
 353
 354 /* Deal with the reporting of failure conditions on a filesystem such as
 355  * inconsistencies detected or read IO failures.
 356  *
 357  * On ext2, we can store the error state of the filesystem in the
 358  * superblock.  That is not possible on ext4, because we may have other
 359  * write ordering constraints on the superblock which prevent us from
 360  * writing it out straight away; and given that the journal is about to
 361  * be aborted, we can't rely on the current, or future, transactions to
 362  * write out the superblock safely.
 363  *
 364  * We'll just use the jbd2_journal_abort() error code to record an error in
 365  * the journal instead.  On recovery, the journal will complain about
 366  * that error until we've noted it down and cleared it.
 367  */
 368
 369 static void ext4_handle_error(struct super_block *sb)
 370 {
 371         if (sb->s_flags & MS_RDONLY)
 372                 return;
 373
 374         if (!test_opt(sb, ERRORS_CONT)) {
 375                 journal_t *journal = EXT4_SB(sb)->s_journal;
 376
 377                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 378                 if (journal)
 379                         jbd2_journal_abort(journal, -EIO);
 380         }
 381         if (test_opt(sb, ERRORS_RO)) {
 382                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 383                 sb->s_flags |= MS_RDONLY;
 384         }
 385         if (test_opt(sb, ERRORS_PANIC))
 386                 panic("EXT4-fs (device %s): panic forced after error\n",
 387                         sb->s_id);
 388 }
 389
 390 void __ext4_error(struct super_block *sb, const char *function,
 391                   unsigned int line, const char *fmt, ...)
 392 {
 393         struct va_format vaf;
 394         va_list args;
 395
 396         va_start(args, fmt);
 397         vaf.fmt = fmt;
 398         vaf.va = &args;
 399         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 400                sb->s_id, function, line, current->comm, &vaf);
 401         va_end(args);
 402
 403         ext4_handle_error(sb);
 404 }
 405
 406 void ext4_error_inode(struct inode *inode, const char *function,
 407                       unsigned int line, ext4_fsblk_t block,
 408                       const char *fmt, ...)
 409 {
 410         va_list args;
 411         struct va_format vaf;
 412         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 413
 414         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 415         es->s_last_error_block = cpu_to_le64(block);
 416         save_error_info(inode->i_sb, function, line);
 417         va_start(args, fmt);
 418         vaf.fmt = fmt;
 419         vaf.va = &args;
 420         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 421                inode->i_sb->s_id, function, line, inode->i_ino);
 422         if (block)
 423                 printk(KERN_CONT "block %llu: ", block);
 424         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 425         va_end(args);
 426
 427         ext4_handle_error(inode->i_sb);
 428 }
 429
 430 void ext4_error_file(struct file *file, const char *function,
 431                      unsigned int line, ext4_fsblk_t block,
 432                      const char *fmt, ...)
 433 {
 434         va_list args;
 435         struct va_format vaf;
 436         struct ext4_super_block *es;
 437         struct inode *inode = file->f_dentry->d_inode;
 438         char pathname[80], *path;
 439
 440         es = EXT4_SB(inode->i_sb)->s_es;
 441         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 442         save_error_info(inode->i_sb, function, line);
 443         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 444         if (IS_ERR(path))
 445                 path = "(unknown)";
 446         printk(KERN_CRIT
 447                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 448                inode->i_sb->s_id, function, line, inode->i_ino);
 449         if (block)
 450                 printk(KERN_CONT "block %llu: ", block);
 451         va_start(args, fmt);
 452         vaf.fmt = fmt;
 453         vaf.va = &args;
 454         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 455         va_end(args);
 456
 457         ext4_handle_error(inode->i_sb);
 458 }
 459
 460 static const char *ext4_decode_error(struct super_block *sb, int errno,
 461                                      char nbuf[16])
 462 {
 463         char *errstr = NULL;
 464
 465         switch (errno) {
 466         case -EIO:
 467                 errstr = "IO failure";
 468                 break;
 469         case -ENOMEM:
 470                 errstr = "Out of memory";
 471                 break;
 472         case -EROFS:
 473                 if (!sb || (EXT4_SB(sb)->s_journal &&
 474                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 475                         errstr = "Journal has aborted";
 476                 else
 477                         errstr = "Readonly filesystem";
 478                 break;
 479         default:
 480                 /* If the caller passed in an extra buffer for unknown
 481                  * errors, textualise them now.  Else we just return
 482                  * NULL. */
 483                 if (nbuf) {
 484                         /* Check for truncated error codes... */
 485                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 486                                 errstr = nbuf;
 487                 }
 488                 break;
 489         }
 490
 491         return errstr;
 492 }
 493
 494 /* __ext4_std_error decodes expected errors from journaling functions
 495  * automatically and invokes the appropriate error response.  */
 496
 497 void __ext4_std_error(struct super_block *sb, const char *function,
 498                       unsigned int line, int errno)
 499 {
 500         char nbuf[16];
 501         const char *errstr;
 502
 503         /* Special case: if the error is EROFS, and we're not already
 504          * inside a transaction, then there's really no point in logging
 505          * an error. */
 506         if (errno == -EROFS && journal_current_handle() == NULL &&
 507             (sb->s_flags & MS_RDONLY))
 508                 return;
 509
 510         errstr = ext4_decode_error(sb, errno, nbuf);
 511         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 512                sb->s_id, function, line, errstr);
 513         save_error_info(sb, function, line);
 514
 515         ext4_handle_error(sb);
 516 }
 517
 518 /*
 519  * ext4_abort is a much stronger failure handler than ext4_error.  The
 520  * abort function may be used to deal with unrecoverable failures such
 521  * as journal IO errors or ENOMEM at a critical moment in log management.
 522  *
 523  * We unconditionally force the filesystem into an ABORT|READONLY state,
 524  * unless the error response on the fs has been set to panic in which
 525  * case we take the easy way out and panic immediately.
 526  */
 527
 528 void __ext4_abort(struct super_block *sb, const char *function,
 529                 unsigned int line, const char *fmt, ...)
 530 {
 531         va_list args;
 532
 533         save_error_info(sb, function, line);
 534         va_start(args, fmt);
 535         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 536                function, line);
 537         vprintk(fmt, args);
 538         printk("\n");
 539         va_end(args);
 540
 541         if ((sb->s_flags & MS_RDONLY) == 0) {
 542                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 543                 sb->s_flags |= MS_RDONLY;
 544                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 545                 if (EXT4_SB(sb)->s_journal)
 546                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 547                 save_error_info(sb, function, line);
 548         }
 549         if (test_opt(sb, ERRORS_PANIC))
 550                 panic("EXT4-fs panic from previous error\n");
 551 }
 552
 553 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 554 {
 555         struct va_format vaf;
 556         va_list args;
 557
 558         va_start(args, fmt);
 559         vaf.fmt = fmt;
 560         vaf.va = &args;
 561         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 562         va_end(args);
 563 }
 564
 565 void __ext4_warning(struct super_block *sb, const char *function,
 566                     unsigned int line, const char *fmt, ...)
 567 {
 568         struct va_format vaf;
 569         va_list args;
 570
 571         va_start(args, fmt);
 572         vaf.fmt = fmt;
 573         vaf.va = &args;
 574         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 575                sb->s_id, function, line, &vaf);
 576         va_end(args);
 577 }
 578
 579 void __ext4_grp_locked_error(const char *function, unsigned int line,
 580                              struct super_block *sb, ext4_group_t grp,
 581                              unsigned long ino, ext4_fsblk_t block,
 582                              const char *fmt, ...)
 583 __releases(bitlock)
 584 __acquires(bitlock)
 585 {
 586         struct va_format vaf;
 587         va_list args;
 588         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 589
 590         es->s_last_error_ino = cpu_to_le32(ino);
 591         es->s_last_error_block = cpu_to_le64(block);
 592         __save_error_info(sb, function, line);
 593
 594         va_start(args, fmt);
 595
 596         vaf.fmt = fmt;
 597         vaf.va = &args;
 598         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 599                sb->s_id, function, line, grp);
 600         if (ino)
 601                 printk(KERN_CONT "inode %lu: ", ino);
 602         if (block)
 603                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 604         printk(KERN_CONT "%pV\n", &vaf);
 605         va_end(args);
 606
 607         if (test_opt(sb, ERRORS_CONT)) {
 608                 ext4_commit_super(sb, 0);
 609                 return;
 610         }
 611
 612         ext4_unlock_group(sb, grp);
 613         ext4_handle_error(sb);
 614         /*
 615          * We only get here in the ERRORS_RO case; relocking the group
 616          * may be dangerous, but nothing bad will happen since the
 617          * filesystem will have already been marked read/only and the
 618          * journal has been aborted.  We return 1 as a hint to callers
 619          * who might what to use the return value from
 620          * ext4_grp_locked_error() to distinguish beween the
 621          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 622          * aggressively from the ext4 function in question, with a
 623          * more appropriate error code.
 624          */
 625         ext4_lock_group(sb, grp);
 626         return;
 627 }
 628
 629 void ext4_update_dynamic_rev(struct super_block *sb)
 630 {
 631         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 632
 633         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 634                 return;
 635
 636         ext4_warning(sb,
 637                      "updating to rev %d because of new feature flag, "
 638                      "running e2fsck is recommended",
 639                      EXT4_DYNAMIC_REV);
 640
 641         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 642         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 643         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 644         /* leave es->s_feature_*compat flags alone */
 645         /* es->s_uuid will be set by e2fsck if empty */
 646
 647         /*
 648          * The rest of the superblock fields should be zero, and if not it
 649          * means they are likely already in use, so leave them alone.  We
 650          * can leave it up to e2fsck to clean up any inconsistencies there.
 651          */
 652 }
 653
 654 /*
 655  * Open the external journal device
 656  */
 657 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 658 {
 659         struct block_device *bdev;
 660         char b[BDEVNAME_SIZE];
 661
 662         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 663         if (IS_ERR(bdev))
 664                 goto fail;
 665         return bdev;
 666
 667 fail:
 668         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 669                         __bdevname(dev, b), PTR_ERR(bdev));
 670         return NULL;
 671 }
 672
 673 /*
 674  * Release the journal device
 675  */
 676 static int ext4_blkdev_put(struct block_device *bdev)
 677 {
 678         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 679 }
 680
 681 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 682 {
 683         struct block_device *bdev;
 684         int ret = -ENODEV;
 685
 686         bdev = sbi->journal_bdev;
 687         if (bdev) {
 688                 ret = ext4_blkdev_put(bdev);
 689                 sbi->journal_bdev = NULL;
 690         }
 691         return ret;
 692 }
 693
 694 static inline struct inode *orphan_list_entry(struct list_head *l)
 695 {
 696         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 697 }
 698
 699 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 700 {
 701         struct list_head *l;
 702
 703         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 704                  le32_to_cpu(sbi->s_es->s_last_orphan));
 705
 706         printk(KERN_ERR "sb_info orphan list:\n");
 707         list_for_each(l, &sbi->s_orphan) {
 708                 struct inode *inode = orphan_list_entry(l);
 709                 printk(KERN_ERR "  "
 710                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 711                        inode->i_sb->s_id, inode->i_ino, inode,
 712                        inode->i_mode, inode->i_nlink,
 713                        NEXT_ORPHAN(inode));
 714         }
 715 }
 716
 717 static void ext4_put_super(struct super_block *sb)
 718 {
 719         struct ext4_sb_info *sbi = EXT4_SB(sb);
 720         struct ext4_super_block *es = sbi->s_es;
 721         int i, err;
 722
 723         ext4_unregister_li_request(sb);
 724         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 725
 726         flush_workqueue(sbi->dio_unwritten_wq);
 727         destroy_workqueue(sbi->dio_unwritten_wq);
 728
 729         lock_super(sb);
 730         if (sb->s_dirt)
 731                 ext4_commit_super(sb, 1);
 732
 733         if (sbi->s_journal) {
 734                 err = jbd2_journal_destroy(sbi->s_journal);
 735                 sbi->s_journal = NULL;
 736                 if (err < 0)
 737                         ext4_abort(sb, "Couldn't clean up the journal");
 738         }
 739
 740         del_timer(&sbi->s_err_report);
 741         ext4_release_system_zone(sb);
 742         ext4_mb_release(sb);
 743         ext4_ext_release(sb);
 744         ext4_xattr_put_super(sb);
 745
 746         if (!(sb->s_flags & MS_RDONLY)) {
 747                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 748                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 749                 ext4_commit_super(sb, 1);
 750         }
 751         if (sbi->s_proc) {
 752                 remove_proc_entry(sb->s_id, ext4_proc_root);
 753         }
 754         kobject_del(&sbi->s_kobj);
 755
 756         for (i = 0; i < sbi->s_gdb_count; i++)
 757                 brelse(sbi->s_group_desc[i]);
 758         kfree(sbi->s_group_desc);
 759         if (is_vmalloc_addr(sbi->s_flex_groups))
 760                 vfree(sbi->s_flex_groups);
 761         else
 762                 kfree(sbi->s_flex_groups);
 763         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 764         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 765         percpu_counter_destroy(&sbi->s_dirs_counter);
 766         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 767         brelse(sbi->s_sbh);
 768 #ifdef CONFIG_QUOTA
 769         for (i = 0; i < MAXQUOTAS; i++)
 770                 kfree(sbi->s_qf_names[i]);
 771 #endif
 772
 773         /* Debugging code just in case the in-memory inode orphan list
 774          * isn't empty.  The on-disk one can be non-empty if we've
 775          * detected an error and taken the fs readonly, but the
 776          * in-memory list had better be clean by this point. */
 777         if (!list_empty(&sbi->s_orphan))
 778                 dump_orphan_list(sb, sbi);
 779         J_ASSERT(list_empty(&sbi->s_orphan));
 780
 781         invalidate_bdev(sb->s_bdev);
 782         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 783                 /*
 784                  * Invalidate the journal device's buffers.  We don't want them
 785                  * floating about in memory - the physical journal device may
 786                  * hotswapped, and it breaks the `ro-after' testing code.
 787                  */
 788                 sync_blockdev(sbi->journal_bdev);
 789                 invalidate_bdev(sbi->journal_bdev);
 790                 ext4_blkdev_remove(sbi);
 791         }
 792         sb->s_fs_info = NULL;
 793         /*
 794          * Now that we are completely done shutting down the
 795          * superblock, we need to actually destroy the kobject.
 796          */
 797         unlock_super(sb);
 798         kobject_put(&sbi->s_kobj);
 799         wait_for_completion(&sbi->s_kobj_unregister);
 800         kfree(sbi->s_blockgroup_lock);
 801         kfree(sbi);
 802 }
 803
 804 static struct kmem_cache *ext4_inode_cachep;
 805
 806 /*
 807  * Called inside transaction, so use GFP_NOFS
 808  */
 809 static struct inode *ext4_alloc_inode(struct super_block *sb)
 810 {
 811         struct ext4_inode_info *ei;
 812
 813         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 814         if (!ei)
 815                 return NULL;
 816
 817         ei->vfs_inode.i_version = 1;
 818         ei->vfs_inode.i_data.writeback_index = 0;
 819         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 820         INIT_LIST_HEAD(&ei->i_prealloc_list);
 821         spin_lock_init(&ei->i_prealloc_lock);
 822         ei->i_reserved_data_blocks = 0;
 823         ei->i_reserved_meta_blocks = 0;
 824         ei->i_allocated_meta_blocks = 0;
 825         ei->i_da_metadata_calc_len = 0;
 826         spin_lock_init(&(ei->i_block_reservation_lock));
 827 #ifdef CONFIG_QUOTA
 828         ei->i_reserved_quota = 0;
 829 #endif
 830         ei->jinode = NULL;
 831         INIT_LIST_HEAD(&ei->i_completed_io_list);
 832         spin_lock_init(&ei->i_completed_io_lock);
 833         ei->cur_aio_dio = NULL;
 834         ei->i_sync_tid = 0;
 835         ei->i_datasync_tid = 0;
 836         atomic_set(&ei->i_ioend_count, 0);
 837         atomic_set(&ei->i_aiodio_unwritten, 0);
 838
 839         return &ei->vfs_inode;
 840 }
 841
 842 static int ext4_drop_inode(struct inode *inode)
 843 {
 844         int drop = generic_drop_inode(inode);
 845
 846         trace_ext4_drop_inode(inode, drop);
 847         return drop;
 848 }
 849
 850 static void ext4_i_callback(struct rcu_head *head)
 851 {
 852         struct inode *inode = container_of(head, struct inode, i_rcu);
 853         INIT_LIST_HEAD(&inode->i_dentry);
 854         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 855 }
 856
 857 static void ext4_destroy_inode(struct inode *inode)
 858 {
 859         ext4_ioend_wait(inode);
 860         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 861                 ext4_msg(inode->i_sb, KERN_ERR,
 862                          "Inode %lu (%p): orphan list check failed!",
 863                          inode->i_ino, EXT4_I(inode));
 864                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 865                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 866                                 true);
 867                 dump_stack();
 868         }
 869         call_rcu(&inode->i_rcu, ext4_i_callback);
 870 }
 871
 872 static void init_once(void *foo)
 873 {
 874         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 875
 876         INIT_LIST_HEAD(&ei->i_orphan);
 877 #ifdef CONFIG_EXT4_FS_XATTR
 878         init_rwsem(&ei->xattr_sem);
 879 #endif
 880         init_rwsem(&ei->i_data_sem);
 881         inode_init_once(&ei->vfs_inode);
 882 }
 883
 884 static int init_inodecache(void)
 885 {
 886         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 887                                              sizeof(struct ext4_inode_info),
 888                                              0, (SLAB_RECLAIM_ACCOUNT|
 889                                                 SLAB_MEM_SPREAD),
 890                                              init_once);
 891         if (ext4_inode_cachep == NULL)
 892                 return -ENOMEM;
 893         return 0;
 894 }
 895
 896 static void destroy_inodecache(void)
 897 {
 898         kmem_cache_destroy(ext4_inode_cachep);
 899 }
 900
 901 void ext4_clear_inode(struct inode *inode)
 902 {
 903         invalidate_inode_buffers(inode);
 904         end_writeback(inode);
 905         dquot_drop(inode);
 906         ext4_discard_preallocations(inode);
 907         if (EXT4_I(inode)->jinode) {
 908                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 909                                                EXT4_I(inode)->jinode);
 910                 jbd2_free_inode(EXT4_I(inode)->jinode);
 911                 EXT4_I(inode)->jinode = NULL;
 912         }
 913 }
 914
 915 static inline void ext4_show_quota_options(struct seq_file *seq,
 916                                            struct super_block *sb)
 917 {
 918 #if defined(CONFIG_QUOTA)
 919         struct ext4_sb_info *sbi = EXT4_SB(sb);
 920
 921         if (sbi->s_jquota_fmt) {
 922                 char *fmtname = "";
 923
 924                 switch (sbi->s_jquota_fmt) {
 925                 case QFMT_VFS_OLD:
 926                         fmtname = "vfsold";
 927                         break;
 928                 case QFMT_VFS_V0:
 929                         fmtname = "vfsv0";
 930                         break;
 931                 case QFMT_VFS_V1:
 932                         fmtname = "vfsv1";
 933                         break;
 934                 }
 935                 seq_printf(seq, ",jqfmt=%s", fmtname);
 936         }
 937
 938         if (sbi->s_qf_names[USRQUOTA])
 939                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 940
 941         if (sbi->s_qf_names[GRPQUOTA])
 942                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 943
 944         if (test_opt(sb, USRQUOTA))
 945                 seq_puts(seq, ",usrquota");
 946
 947         if (test_opt(sb, GRPQUOTA))
 948                 seq_puts(seq, ",grpquota");
 949 #endif
 950 }
 951
 952 /*
 953  * Show an option if
 954  *  - it's set to a non-default value OR
 955  *  - if the per-sb default is different from the global default
 956  */
 957 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 958 {
 959         int def_errors;
 960         unsigned long def_mount_opts;
 961         struct super_block *sb = vfs->mnt_sb;
 962         struct ext4_sb_info *sbi = EXT4_SB(sb);
 963         struct ext4_super_block *es = sbi->s_es;
 964
 965         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 966         def_errors     = le16_to_cpu(es->s_errors);
 967
 968         if (sbi->s_sb_block != 1)
 969                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 970         if (test_opt(sb, MINIX_DF))
 971                 seq_puts(seq, ",minixdf");
 972         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 973                 seq_puts(seq, ",grpid");
 974         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 975                 seq_puts(seq, ",nogrpid");
 976         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 977             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 978                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 979         }
 980         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 981             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 982                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 983         }
 984         if (test_opt(sb, ERRORS_RO)) {
 985                 if (def_errors == EXT4_ERRORS_PANIC ||
 986                     def_errors == EXT4_ERRORS_CONTINUE) {
 987                         seq_puts(seq, ",errors=remount-ro");
 988                 }
 989         }
 990         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 991                 seq_puts(seq, ",errors=continue");
 992         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 993                 seq_puts(seq, ",errors=panic");
 994         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 995                 seq_puts(seq, ",nouid32");
 996         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 997                 seq_puts(seq, ",debug");
 998         if (test_opt(sb, OLDALLOC))
 999                 seq_puts(seq, ",oldalloc");
1000 #ifdef CONFIG_EXT4_FS_XATTR
1001         if (test_opt(sb, XATTR_USER))
1002                 seq_puts(seq, ",user_xattr");
1003         if (!test_opt(sb, XATTR_USER))
1004                 seq_puts(seq, ",nouser_xattr");
1005 #endif
1006 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1007         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1008                 seq_puts(seq, ",acl");
1009         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1010                 seq_puts(seq, ",noacl");
1011 #endif
1012         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1013                 seq_printf(seq, ",commit=%u",
1014                            (unsigned) (sbi->s_commit_interval / HZ));
1015         }
1016         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1017                 seq_printf(seq, ",min_batch_time=%u",
1018                            (unsigned) sbi->s_min_batch_time);
1019         }
1020         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1021                 seq_printf(seq, ",max_batch_time=%u",
1022                            (unsigned) sbi->s_min_batch_time);
1023         }
1024
1025         /*
1026          * We're changing the default of barrier mount option, so
1027          * let's always display its mount state so it's clear what its
1028          * status is.
1029          */
1030         seq_puts(seq, ",barrier=");
1031         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1032         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1033                 seq_puts(seq, ",journal_async_commit");
1034         else if (test_opt(sb, JOURNAL_CHECKSUM))
1035                 seq_puts(seq, ",journal_checksum");
1036         if (test_opt(sb, I_VERSION))
1037                 seq_puts(seq, ",i_version");
1038         if (!test_opt(sb, DELALLOC) &&
1039             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1040                 seq_puts(seq, ",nodelalloc");
1041
1042         if (!test_opt(sb, MBLK_IO_SUBMIT))
1043                 seq_puts(seq, ",nomblk_io_submit");
1044         if (sbi->s_stripe)
1045                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1046         /*
1047          * journal mode get enabled in different ways
1048          * So just print the value even if we didn't specify it
1049          */
1050         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1051                 seq_puts(seq, ",data=journal");
1052         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1053                 seq_puts(seq, ",data=ordered");
1054         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1055                 seq_puts(seq, ",data=writeback");
1056
1057         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1058                 seq_printf(seq, ",inode_readahead_blks=%u",
1059                            sbi->s_inode_readahead_blks);
1060
1061         if (test_opt(sb, DATA_ERR_ABORT))
1062                 seq_puts(seq, ",data_err=abort");
1063
1064         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1065                 seq_puts(seq, ",noauto_da_alloc");
1066
1067         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1068                 seq_puts(seq, ",discard");
1069
1070         if (test_opt(sb, NOLOAD))
1071                 seq_puts(seq, ",norecovery");
1072
1073         if (test_opt(sb, DIOREAD_NOLOCK))
1074                 seq_puts(seq, ",dioread_nolock");
1075
1076         if (test_opt(sb, BLOCK_VALIDITY) &&
1077             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1078                 seq_puts(seq, ",block_validity");
1079
1080         if (!test_opt(sb, INIT_INODE_TABLE))
1081                 seq_puts(seq, ",noinit_inode_table");
1082         else if (sbi->s_li_wait_mult)
1083                 seq_printf(seq, ",init_inode_table=%u",
1084                            (unsigned) sbi->s_li_wait_mult);
1085
1086         ext4_show_quota_options(seq, sb);
1087
1088         return 0;
1089 }
1090
1091 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1092                                         u64 ino, u32 generation)
1093 {
1094         struct inode *inode;
1095
1096         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1097                 return ERR_PTR(-ESTALE);
1098         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1099                 return ERR_PTR(-ESTALE);
1100
1101         /* iget isn't really right if the inode is currently unallocated!!
1102          *
1103          * ext4_read_inode will return a bad_inode if the inode had been
1104          * deleted, so we should be safe.
1105          *
1106          * Currently we don't know the generation for parent directory, so
1107          * a generation of 0 means "accept any"
1108          */
1109         inode = ext4_iget(sb, ino);
1110         if (IS_ERR(inode))
1111                 return ERR_CAST(inode);
1112         if (generation && inode->i_generation != generation) {
1113                 iput(inode);
1114                 return ERR_PTR(-ESTALE);
1115         }
1116
1117         return inode;
1118 }
1119
1120 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1121                                         int fh_len, int fh_type)
1122 {
1123         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1124                                     ext4_nfs_get_inode);
1125 }
1126
1127 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1128                                         int fh_len, int fh_type)
1129 {
1130         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1131                                     ext4_nfs_get_inode);
1132 }
1133
1134 /*
1135  * Try to release metadata pages (indirect blocks, directories) which are
1136  * mapped via the block device.  Since these pages could have journal heads
1137  * which would prevent try_to_free_buffers() from freeing them, we must use
1138  * jbd2 layer's try_to_free_buffers() function to release them.
1139  */
1140 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1141                                  gfp_t wait)
1142 {
1143         journal_t *journal = EXT4_SB(sb)->s_journal;
1144
1145         WARN_ON(PageChecked(page));
1146         if (!page_has_buffers(page))
1147                 return 0;
1148         if (journal)
1149                 return jbd2_journal_try_to_free_buffers(journal, page,
1150                                                         wait & ~__GFP_WAIT);
1151         return try_to_free_buffers(page);
1152 }
1153
1154 #ifdef CONFIG_QUOTA
1155 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1156 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1157
1158 static int ext4_write_dquot(struct dquot *dquot);
1159 static int ext4_acquire_dquot(struct dquot *dquot);
1160 static int ext4_release_dquot(struct dquot *dquot);
1161 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1162 static int ext4_write_info(struct super_block *sb, int type);
1163 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1164                          struct path *path);
1165 static int ext4_quota_off(struct super_block *sb, int type);
1166 static int ext4_quota_on_mount(struct super_block *sb, int type);
1167 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1168                                size_t len, loff_t off);
1169 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1170                                 const char *data, size_t len, loff_t off);
1171
1172 static const struct dquot_operations ext4_quota_operations = {
1173 #ifdef CONFIG_QUOTA
1174         .get_reserved_space = ext4_get_reserved_space,
1175 #endif
1176         .write_dquot    = ext4_write_dquot,
1177         .acquire_dquot  = ext4_acquire_dquot,
1178         .release_dquot  = ext4_release_dquot,
1179         .mark_dirty     = ext4_mark_dquot_dirty,
1180         .write_info     = ext4_write_info,
1181         .alloc_dquot    = dquot_alloc,
1182         .destroy_dquot  = dquot_destroy,
1183 };
1184
1185 static const struct quotactl_ops ext4_qctl_operations = {
1186         .quota_on       = ext4_quota_on,
1187         .quota_off      = ext4_quota_off,
1188         .quota_sync     = dquot_quota_sync,
1189         .get_info       = dquot_get_dqinfo,
1190         .set_info       = dquot_set_dqinfo,
1191         .get_dqblk      = dquot_get_dqblk,
1192         .set_dqblk      = dquot_set_dqblk
1193 };
1194 #endif
1195
1196 static const struct super_operations ext4_sops = {
1197         .alloc_inode    = ext4_alloc_inode,
1198         .destroy_inode  = ext4_destroy_inode,
1199         .write_inode    = ext4_write_inode,
1200         .dirty_inode    = ext4_dirty_inode,
1201         .drop_inode     = ext4_drop_inode,
1202         .evict_inode    = ext4_evict_inode,
1203         .put_super      = ext4_put_super,
1204         .sync_fs        = ext4_sync_fs,
1205         .freeze_fs      = ext4_freeze,
1206         .unfreeze_fs    = ext4_unfreeze,
1207         .statfs         = ext4_statfs,
1208         .remount_fs     = ext4_remount,
1209         .show_options   = ext4_show_options,
1210 #ifdef CONFIG_QUOTA
1211         .quota_read     = ext4_quota_read,
1212         .quota_write    = ext4_quota_write,
1213 #endif
1214         .bdev_try_to_free_page = bdev_try_to_free_page,
1215 };
1216
1217 static const struct super_operations ext4_nojournal_sops = {
1218         .alloc_inode    = ext4_alloc_inode,
1219         .destroy_inode  = ext4_destroy_inode,
1220         .write_inode    = ext4_write_inode,
1221         .dirty_inode    = ext4_dirty_inode,
1222         .drop_inode     = ext4_drop_inode,
1223         .evict_inode    = ext4_evict_inode,
1224         .write_super    = ext4_write_super,
1225         .put_super      = ext4_put_super,
1226         .statfs         = ext4_statfs,
1227         .remount_fs     = ext4_remount,
1228         .show_options   = ext4_show_options,
1229 #ifdef CONFIG_QUOTA
1230         .quota_read     = ext4_quota_read,
1231         .quota_write    = ext4_quota_write,
1232 #endif
1233         .bdev_try_to_free_page = bdev_try_to_free_page,
1234 };
1235
1236 static const struct export_operations ext4_export_ops = {
1237         .fh_to_dentry = ext4_fh_to_dentry,
1238         .fh_to_parent = ext4_fh_to_parent,
1239         .get_parent = ext4_get_parent,
1240 };
1241
1242 enum {
1243         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1244         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1245         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1246         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1247         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1248         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1249         Opt_journal_update, Opt_journal_dev,
1250         Opt_journal_checksum, Opt_journal_async_commit,
1251         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1252         Opt_data_err_abort, Opt_data_err_ignore,
1253         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1254         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1255         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1256         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1257         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1258         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1259         Opt_inode_readahead_blks, Opt_journal_ioprio,
1260         Opt_dioread_nolock, Opt_dioread_lock,
1261         Opt_discard, Opt_nodiscard,
1262         Opt_init_inode_table, Opt_noinit_inode_table,
1263 };
1264
1265 static const match_table_t tokens = {
1266         {Opt_bsd_df, "bsddf"},
1267         {Opt_minix_df, "minixdf"},
1268         {Opt_grpid, "grpid"},
1269         {Opt_grpid, "bsdgroups"},
1270         {Opt_nogrpid, "nogrpid"},
1271         {Opt_nogrpid, "sysvgroups"},
1272         {Opt_resgid, "resgid=%u"},
1273         {Opt_resuid, "resuid=%u"},
1274         {Opt_sb, "sb=%u"},
1275         {Opt_err_cont, "errors=continue"},
1276         {Opt_err_panic, "errors=panic"},
1277         {Opt_err_ro, "errors=remount-ro"},
1278         {Opt_nouid32, "nouid32"},
1279         {Opt_debug, "debug"},
1280         {Opt_oldalloc, "oldalloc"},
1281         {Opt_orlov, "orlov"},
1282         {Opt_user_xattr, "user_xattr"},
1283         {Opt_nouser_xattr, "nouser_xattr"},
1284         {Opt_acl, "acl"},
1285         {Opt_noacl, "noacl"},
1286         {Opt_noload, "noload"},
1287         {Opt_noload, "norecovery"},
1288         {Opt_nobh, "nobh"},
1289         {Opt_bh, "bh"},
1290         {Opt_commit, "commit=%u"},
1291         {Opt_min_batch_time, "min_batch_time=%u"},
1292         {Opt_max_batch_time, "max_batch_time=%u"},
1293         {Opt_journal_update, "journal=update"},
1294         {Opt_journal_dev, "journal_dev=%u"},
1295         {Opt_journal_checksum, "journal_checksum"},
1296         {Opt_journal_async_commit, "journal_async_commit"},
1297         {Opt_abort, "abort"},
1298         {Opt_data_journal, "data=journal"},
1299         {Opt_data_ordered, "data=ordered"},
1300         {Opt_data_writeback, "data=writeback"},
1301         {Opt_data_err_abort, "data_err=abort"},
1302         {Opt_data_err_ignore, "data_err=ignore"},
1303         {Opt_offusrjquota, "usrjquota="},
1304         {Opt_usrjquota, "usrjquota=%s"},
1305         {Opt_offgrpjquota, "grpjquota="},
1306         {Opt_grpjquota, "grpjquota=%s"},
1307         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1308         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1309         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1310         {Opt_grpquota, "grpquota"},
1311         {Opt_noquota, "noquota"},
1312         {Opt_quota, "quota"},
1313         {Opt_usrquota, "usrquota"},
1314         {Opt_barrier, "barrier=%u"},
1315         {Opt_barrier, "barrier"},
1316         {Opt_nobarrier, "nobarrier"},
1317         {Opt_i_version, "i_version"},
1318         {Opt_stripe, "stripe=%u"},
1319         {Opt_resize, "resize"},
1320         {Opt_delalloc, "delalloc"},
1321         {Opt_nodelalloc, "nodelalloc"},
1322         {Opt_mblk_io_submit, "mblk_io_submit"},
1323         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1324         {Opt_block_validity, "block_validity"},
1325         {Opt_noblock_validity, "noblock_validity"},
1326         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1327         {Opt_journal_ioprio, "journal_ioprio=%u"},
1328         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1329         {Opt_auto_da_alloc, "auto_da_alloc"},
1330         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1331         {Opt_dioread_nolock, "dioread_nolock"},
1332         {Opt_dioread_lock, "dioread_lock"},
1333         {Opt_discard, "discard"},
1334         {Opt_nodiscard, "nodiscard"},
1335         {Opt_init_inode_table, "init_itable=%u"},
1336         {Opt_init_inode_table, "init_itable"},
1337         {Opt_noinit_inode_table, "noinit_itable"},
1338         {Opt_err, NULL},
1339 };
1340
1341 static ext4_fsblk_t get_sb_block(void **data)
1342 {
1343         ext4_fsblk_t    sb_block;
1344         char            *options = (char *) *data;
1345
1346         if (!options || strncmp(options, "sb=", 3) != 0)
1347                 return 1;       /* Default location */
1348
1349         options += 3;
1350         /* TODO: use simple_strtoll with >32bit ext4 */
1351         sb_block = simple_strtoul(options, &options, 0);
1352         if (*options && *options != ',') {
1353                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1354                        (char *) *data);
1355                 return 1;
1356         }
1357         if (*options == ',')
1358                 options++;
1359         *data = (void *) options;
1360
1361         return sb_block;
1362 }
1363
1364 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1365 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1366         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1367
1368 #ifdef CONFIG_QUOTA
1369 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1370 {
1371         struct ext4_sb_info *sbi = EXT4_SB(sb);
1372         char *qname;
1373
1374         if (sb_any_quota_loaded(sb) &&
1375                 !sbi->s_qf_names[qtype]) {
1376                 ext4_msg(sb, KERN_ERR,
1377                         "Cannot change journaled "
1378                         "quota options when quota turned on");
1379                 return 0;
1380         }
1381         qname = match_strdup(args);
1382         if (!qname) {
1383                 ext4_msg(sb, KERN_ERR,
1384                         "Not enough memory for storing quotafile name");
1385                 return 0;
1386         }
1387         if (sbi->s_qf_names[qtype] &&
1388                 strcmp(sbi->s_qf_names[qtype], qname)) {
1389                 ext4_msg(sb, KERN_ERR,
1390                         "%s quota file already specified", QTYPE2NAME(qtype));
1391                 kfree(qname);
1392                 return 0;
1393         }
1394         sbi->s_qf_names[qtype] = qname;
1395         if (strchr(sbi->s_qf_names[qtype], '/')) {
1396                 ext4_msg(sb, KERN_ERR,
1397                         "quotafile must be on filesystem root");
1398                 kfree(sbi->s_qf_names[qtype]);
1399                 sbi->s_qf_names[qtype] = NULL;
1400                 return 0;
1401         }
1402         set_opt(sb, QUOTA);
1403         return 1;
1404 }
1405
1406 static int clear_qf_name(struct super_block *sb, int qtype)
1407 {
1408
1409         struct ext4_sb_info *sbi = EXT4_SB(sb);
1410
1411         if (sb_any_quota_loaded(sb) &&
1412                 sbi->s_qf_names[qtype]) {
1413                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1414                         " when quota turned on");
1415                 return 0;
1416         }
1417         /*
1418          * The space will be released later when all options are confirmed
1419          * to be correct
1420          */
1421         sbi->s_qf_names[qtype] = NULL;
1422         return 1;
1423 }
1424 #endif
1425
1426 static int parse_options(char *options, struct super_block *sb,
1427                          unsigned long *journal_devnum,
1428                          unsigned int *journal_ioprio,
1429                          ext4_fsblk_t *n_blocks_count, int is_remount)
1430 {
1431         struct ext4_sb_info *sbi = EXT4_SB(sb);
1432         char *p;
1433         substring_t args[MAX_OPT_ARGS];
1434         int data_opt = 0;
1435         int option;
1436 #ifdef CONFIG_QUOTA
1437         int qfmt;
1438 #endif
1439
1440         if (!options)
1441                 return 1;
1442
1443         while ((p = strsep(&options, ",")) != NULL) {
1444                 int token;
1445                 if (!*p)
1446                         continue;
1447
1448                 /*
1449                  * Initialize args struct so we know whether arg was
1450                  * found; some options take optional arguments.
1451                  */
1452                 args[0].to = args[0].from = NULL;
1453                 token = match_token(p, tokens, args);
1454                 switch (token) {
1455                 case Opt_bsd_df:
1456                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1457                         clear_opt(sb, MINIX_DF);
1458                         break;
1459                 case Opt_minix_df:
1460                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1461                         set_opt(sb, MINIX_DF);
1462
1463                         break;
1464                 case Opt_grpid:
1465                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1466                         set_opt(sb, GRPID);
1467
1468                         break;
1469                 case Opt_nogrpid:
1470                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1471                         clear_opt(sb, GRPID);
1472
1473                         break;
1474                 case Opt_resuid:
1475                         if (match_int(&args[0], &option))
1476                                 return 0;
1477                         sbi->s_resuid = option;
1478                         break;
1479                 case Opt_resgid:
1480                         if (match_int(&args[0], &option))
1481                                 return 0;
1482                         sbi->s_resgid = option;
1483                         break;
1484                 case Opt_sb:
1485                         /* handled by get_sb_block() instead of here */
1486                         /* *sb_block = match_int(&args[0]); */
1487                         break;
1488                 case Opt_err_panic:
1489                         clear_opt(sb, ERRORS_CONT);
1490                         clear_opt(sb, ERRORS_RO);
1491                         set_opt(sb, ERRORS_PANIC);
1492                         break;
1493                 case Opt_err_ro:
1494                         clear_opt(sb, ERRORS_CONT);
1495                         clear_opt(sb, ERRORS_PANIC);
1496                         set_opt(sb, ERRORS_RO);
1497                         break;
1498                 case Opt_err_cont:
1499                         clear_opt(sb, ERRORS_RO);
1500                         clear_opt(sb, ERRORS_PANIC);
1501                         set_opt(sb, ERRORS_CONT);
1502                         break;
1503                 case Opt_nouid32:
1504                         set_opt(sb, NO_UID32);
1505                         break;
1506                 case Opt_debug:
1507                         set_opt(sb, DEBUG);
1508                         break;
1509                 case Opt_oldalloc:
1510                         set_opt(sb, OLDALLOC);
1511                         break;
1512                 case Opt_orlov:
1513                         clear_opt(sb, OLDALLOC);
1514                         break;
1515 #ifdef CONFIG_EXT4_FS_XATTR
1516                 case Opt_user_xattr:
1517                         set_opt(sb, XATTR_USER);
1518                         break;
1519                 case Opt_nouser_xattr:
1520                         clear_opt(sb, XATTR_USER);
1521                         break;
1522 #else
1523                 case Opt_user_xattr:
1524                 case Opt_nouser_xattr:
1525                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1526                         break;
1527 #endif
1528 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1529                 case Opt_acl:
1530                         set_opt(sb, POSIX_ACL);
1531                         break;
1532                 case Opt_noacl:
1533                         clear_opt(sb, POSIX_ACL);
1534                         break;
1535 #else
1536                 case Opt_acl:
1537                 case Opt_noacl:
1538                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1539                         break;
1540 #endif
1541                 case Opt_journal_update:
1542                         /* @@@ FIXME */
1543                         /* Eventually we will want to be able to create
1544                            a journal file here.  For now, only allow the
1545                            user to specify an existing inode to be the
1546                            journal file. */
1547                         if (is_remount) {
1548                                 ext4_msg(sb, KERN_ERR,
1549                                          "Cannot specify journal on remount");
1550                                 return 0;
1551                         }
1552                         set_opt(sb, UPDATE_JOURNAL);
1553                         break;
1554                 case Opt_journal_dev:
1555                         if (is_remount) {
1556                                 ext4_msg(sb, KERN_ERR,
1557                                         "Cannot specify journal on remount");
1558                                 return 0;
1559                         }
1560                         if (match_int(&args[0], &option))
1561                                 return 0;
1562                         *journal_devnum = option;
1563                         break;
1564                 case Opt_journal_checksum:
1565                         set_opt(sb, JOURNAL_CHECKSUM);
1566                         break;
1567                 case Opt_journal_async_commit:
1568                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1569                         set_opt(sb, JOURNAL_CHECKSUM);
1570                         break;
1571                 case Opt_noload:
1572                         set_opt(sb, NOLOAD);
1573                         break;
1574                 case Opt_commit:
1575                         if (match_int(&args[0], &option))
1576                                 return 0;
1577                         if (option < 0)
1578                                 return 0;
1579                         if (option == 0)
1580                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1581                         sbi->s_commit_interval = HZ * option;
1582                         break;
1583                 case Opt_max_batch_time:
1584                         if (match_int(&args[0], &option))
1585                                 return 0;
1586                         if (option < 0)
1587                                 return 0;
1588                         if (option == 0)
1589                                 option = EXT4_DEF_MAX_BATCH_TIME;
1590                         sbi->s_max_batch_time = option;
1591                         break;
1592                 case Opt_min_batch_time:
1593                         if (match_int(&args[0], &option))
1594                                 return 0;
1595                         if (option < 0)
1596                                 return 0;
1597                         sbi->s_min_batch_time = option;
1598                         break;
1599                 case Opt_data_journal:
1600                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1601                         goto datacheck;
1602                 case Opt_data_ordered:
1603                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1604                         goto datacheck;
1605                 case Opt_data_writeback:
1606                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1607                 datacheck:
1608                         if (is_remount) {
1609                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1610                                         ext4_msg(sb, KERN_ERR,
1611                                                 "Cannot change data mode on remount");
1612                                         return 0;
1613                                 }
1614                         } else {
1615                                 clear_opt(sb, DATA_FLAGS);
1616                                 sbi->s_mount_opt |= data_opt;
1617                         }
1618                         break;
1619                 case Opt_data_err_abort:
1620                         set_opt(sb, DATA_ERR_ABORT);
1621                         break;
1622                 case Opt_data_err_ignore:
1623                         clear_opt(sb, DATA_ERR_ABORT);
1624                         break;
1625 #ifdef CONFIG_QUOTA
1626                 case Opt_usrjquota:
1627                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1628                                 return 0;
1629                         break;
1630                 case Opt_grpjquota:
1631                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1632                                 return 0;
1633                         break;
1634                 case Opt_offusrjquota:
1635                         if (!clear_qf_name(sb, USRQUOTA))
1636                                 return 0;
1637                         break;
1638                 case Opt_offgrpjquota:
1639                         if (!clear_qf_name(sb, GRPQUOTA))
1640                                 return 0;
1641                         break;
1642
1643                 case Opt_jqfmt_vfsold:
1644                         qfmt = QFMT_VFS_OLD;
1645                         goto set_qf_format;
1646                 case Opt_jqfmt_vfsv0:
1647                         qfmt = QFMT_VFS_V0;
1648                         goto set_qf_format;
1649                 case Opt_jqfmt_vfsv1:
1650                         qfmt = QFMT_VFS_V1;
1651 set_qf_format:
1652                         if (sb_any_quota_loaded(sb) &&
1653                             sbi->s_jquota_fmt != qfmt) {
1654                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1655                                         "journaled quota options when "
1656                                         "quota turned on");
1657                                 return 0;
1658                         }
1659                         sbi->s_jquota_fmt = qfmt;
1660                         break;
1661                 case Opt_quota:
1662                 case Opt_usrquota:
1663                         set_opt(sb, QUOTA);
1664                         set_opt(sb, USRQUOTA);
1665                         break;
1666                 case Opt_grpquota:
1667                         set_opt(sb, QUOTA);
1668                         set_opt(sb, GRPQUOTA);
1669                         break;
1670                 case Opt_noquota:
1671                         if (sb_any_quota_loaded(sb)) {
1672                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1673                                         "options when quota turned on");
1674                                 return 0;
1675                         }
1676                         clear_opt(sb, QUOTA);
1677                         clear_opt(sb, USRQUOTA);
1678                         clear_opt(sb, GRPQUOTA);
1679                         break;
1680 #else
1681                 case Opt_quota:
1682                 case Opt_usrquota:
1683                 case Opt_grpquota:
1684                         ext4_msg(sb, KERN_ERR,
1685                                 "quota options not supported");
1686                         break;
1687                 case Opt_usrjquota:
1688                 case Opt_grpjquota:
1689                 case Opt_offusrjquota:
1690                 case Opt_offgrpjquota:
1691                 case Opt_jqfmt_vfsold:
1692                 case Opt_jqfmt_vfsv0:
1693                 case Opt_jqfmt_vfsv1:
1694                         ext4_msg(sb, KERN_ERR,
1695                                 "journaled quota options not supported");
1696                         break;
1697                 case Opt_noquota:
1698                         break;
1699 #endif
1700                 case Opt_abort:
1701                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1702                         break;
1703                 case Opt_nobarrier:
1704                         clear_opt(sb, BARRIER);
1705                         break;
1706                 case Opt_barrier:
1707                         if (args[0].from) {
1708                                 if (match_int(&args[0], &option))
1709                                         return 0;
1710                         } else
1711                                 option = 1;     /* No argument, default to 1 */
1712                         if (option)
1713                                 set_opt(sb, BARRIER);
1714                         else
1715                                 clear_opt(sb, BARRIER);
1716                         break;
1717                 case Opt_ignore:
1718                         break;
1719                 case Opt_resize:
1720                         if (!is_remount) {
1721                                 ext4_msg(sb, KERN_ERR,
1722                                         "resize option only available "
1723                                         "for remount");
1724                                 return 0;
1725                         }
1726                         if (match_int(&args[0], &option) != 0)
1727                                 return 0;
1728                         *n_blocks_count = option;
1729                         break;
1730                 case Opt_nobh:
1731                         ext4_msg(sb, KERN_WARNING,
1732                                  "Ignoring deprecated nobh option");
1733                         break;
1734                 case Opt_bh:
1735                         ext4_msg(sb, KERN_WARNING,
1736                                  "Ignoring deprecated bh option");
1737                         break;
1738                 case Opt_i_version:
1739                         set_opt(sb, I_VERSION);
1740                         sb->s_flags |= MS_I_VERSION;
1741                         break;
1742                 case Opt_nodelalloc:
1743                         clear_opt(sb, DELALLOC);
1744                         break;
1745                 case Opt_mblk_io_submit:
1746                         set_opt(sb, MBLK_IO_SUBMIT);
1747                         break;
1748                 case Opt_nomblk_io_submit:
1749                         clear_opt(sb, MBLK_IO_SUBMIT);
1750                         break;
1751                 case Opt_stripe:
1752                         if (match_int(&args[0], &option))
1753                                 return 0;
1754                         if (option < 0)
1755                                 return 0;
1756                         sbi->s_stripe = option;
1757                         break;
1758                 case Opt_delalloc:
1759                         set_opt(sb, DELALLOC);
1760                         break;
1761                 case Opt_block_validity:
1762                         set_opt(sb, BLOCK_VALIDITY);
1763                         break;
1764                 case Opt_noblock_validity:
1765                         clear_opt(sb, BLOCK_VALIDITY);
1766                         break;
1767                 case Opt_inode_readahead_blks:
1768                         if (match_int(&args[0], &option))
1769                                 return 0;
1770                         if (option < 0 || option > (1 << 30))
1771                                 return 0;
1772                         if (option && !is_power_of_2(option)) {
1773                                 ext4_msg(sb, KERN_ERR,
1774                                          "EXT4-fs: inode_readahead_blks"
1775                                          " must be a power of 2");
1776                                 return 0;
1777                         }
1778                         sbi->s_inode_readahead_blks = option;
1779                         break;
1780                 case Opt_journal_ioprio:
1781                         if (match_int(&args[0], &option))
1782                                 return 0;
1783                         if (option < 0 || option > 7)
1784                                 break;
1785                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1786                                                             option);
1787                         break;
1788                 case Opt_noauto_da_alloc:
1789                         set_opt(sb, NO_AUTO_DA_ALLOC);
1790                         break;
1791                 case Opt_auto_da_alloc:
1792                         if (args[0].from) {
1793                                 if (match_int(&args[0], &option))
1794                                         return 0;
1795                         } else
1796                                 option = 1;     /* No argument, default to 1 */
1797                         if (option)
1798                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1799                         else
1800                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1801                         break;
1802                 case Opt_discard:
1803                         set_opt(sb, DISCARD);
1804                         break;
1805                 case Opt_nodiscard:
1806                         clear_opt(sb, DISCARD);
1807                         break;
1808                 case Opt_dioread_nolock:
1809                         set_opt(sb, DIOREAD_NOLOCK);
1810                         break;
1811                 case Opt_dioread_lock:
1812                         clear_opt(sb, DIOREAD_NOLOCK);
1813                         break;
1814                 case Opt_init_inode_table:
1815                         set_opt(sb, INIT_INODE_TABLE);
1816                         if (args[0].from) {
1817                                 if (match_int(&args[0], &option))
1818                                         return 0;
1819                         } else
1820                                 option = EXT4_DEF_LI_WAIT_MULT;
1821                         if (option < 0)
1822                                 return 0;
1823                         sbi->s_li_wait_mult = option;
1824                         break;
1825                 case Opt_noinit_inode_table:
1826                         clear_opt(sb, INIT_INODE_TABLE);
1827                         break;
1828                 default:
1829                         ext4_msg(sb, KERN_ERR,
1830                                "Unrecognized mount option \"%s\" "
1831                                "or missing value", p);
1832                         return 0;
1833                 }
1834         }
1835 #ifdef CONFIG_QUOTA
1836         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1837                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1838                         clear_opt(sb, USRQUOTA);
1839
1840                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1841                         clear_opt(sb, GRPQUOTA);
1842
1843                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1844                         ext4_msg(sb, KERN_ERR, "old and new quota "
1845                                         "format mixing");
1846                         return 0;
1847                 }
1848
1849                 if (!sbi->s_jquota_fmt) {
1850                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1851                                         "not specified");
1852                         return 0;
1853                 }
1854         } else {
1855                 if (sbi->s_jquota_fmt) {
1856                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1857                                         "specified with no journaling "
1858                                         "enabled");
1859                         return 0;
1860                 }
1861         }
1862 #endif
1863         return 1;
1864 }
1865
1866 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1867                             int read_only)
1868 {
1869         struct ext4_sb_info *sbi = EXT4_SB(sb);
1870         int res = 0;
1871
1872         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1873                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1874                          "forcing read-only mode");
1875                 res = MS_RDONLY;
1876         }
1877         if (read_only)
1878                 return res;
1879         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1880                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1881                          "running e2fsck is recommended");
1882         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "warning: mounting fs with errors, "
1885                          "running e2fsck is recommended");
1886         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1887                  le16_to_cpu(es->s_mnt_count) >=
1888                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "warning: maximal mount count reached, "
1891                          "running e2fsck is recommended");
1892         else if (le32_to_cpu(es->s_checkinterval) &&
1893                 (le32_to_cpu(es->s_lastcheck) +
1894                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1895                 ext4_msg(sb, KERN_WARNING,
1896                          "warning: checktime reached, "
1897                          "running e2fsck is recommended");
1898         if (!sbi->s_journal)
1899                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1900         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1901                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1902         le16_add_cpu(&es->s_mnt_count, 1);
1903         es->s_mtime = cpu_to_le32(get_seconds());
1904         ext4_update_dynamic_rev(sb);
1905         if (sbi->s_journal)
1906                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1907
1908         ext4_commit_super(sb, 1);
1909         if (test_opt(sb, DEBUG))
1910                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1911                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1912                         sb->s_blocksize,
1913                         sbi->s_groups_count,
1914                         EXT4_BLOCKS_PER_GROUP(sb),
1915                         EXT4_INODES_PER_GROUP(sb),
1916                         sbi->s_mount_opt, sbi->s_mount_opt2);
1917
1918         return res;
1919 }
1920
1921 static int ext4_fill_flex_info(struct super_block *sb)
1922 {
1923         struct ext4_sb_info *sbi = EXT4_SB(sb);
1924         struct ext4_group_desc *gdp = NULL;
1925         ext4_group_t flex_group_count;
1926         ext4_group_t flex_group;
1927         int groups_per_flex = 0;
1928         size_t size;
1929         int i;
1930
1931         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1932         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1933
1934         if (groups_per_flex < 2) {
1935                 sbi->s_log_groups_per_flex = 0;
1936                 return 1;
1937         }
1938
1939         /* We allocate both existing and potentially added groups */
1940         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1941                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1942                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1943         size = flex_group_count * sizeof(struct flex_groups);
1944         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1945         if (sbi->s_flex_groups == NULL) {
1946                 sbi->s_flex_groups = vzalloc(size);
1947                 if (sbi->s_flex_groups == NULL) {
1948                         ext4_msg(sb, KERN_ERR,
1949                                  "not enough memory for %u flex groups",
1950                                  flex_group_count);
1951                         goto failed;
1952                 }
1953         }
1954
1955         for (i = 0; i < sbi->s_groups_count; i++) {
1956                 gdp = ext4_get_group_desc(sb, i, NULL);
1957
1958                 flex_group = ext4_flex_group(sbi, i);
1959                 atomic_add(ext4_free_inodes_count(sb, gdp),
1960                            &sbi->s_flex_groups[flex_group].free_inodes);
1961                 atomic_add(ext4_free_blks_count(sb, gdp),
1962                            &sbi->s_flex_groups[flex_group].free_blocks);
1963                 atomic_add(ext4_used_dirs_count(sb, gdp),
1964                            &sbi->s_flex_groups[flex_group].used_dirs);
1965         }
1966
1967         return 1;
1968 failed:
1969         return 0;
1970 }
1971
1972 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1973                             struct ext4_group_desc *gdp)
1974 {
1975         __u16 crc = 0;
1976
1977         if (sbi->s_es->s_feature_ro_compat &
1978             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1979                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1980                 __le32 le_group = cpu_to_le32(block_group);
1981
1982                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1983                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1984                 crc = crc16(crc, (__u8 *)gdp, offset);
1985                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1986                 /* for checksum of struct ext4_group_desc do the rest...*/
1987                 if ((sbi->s_es->s_feature_incompat &
1988                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1989                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1990                         crc = crc16(crc, (__u8 *)gdp + offset,
1991                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1992                                         offset);
1993         }
1994
1995         return cpu_to_le16(crc);
1996 }
1997
1998 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1999                                 struct ext4_group_desc *gdp)
2000 {
2001         if ((sbi->s_es->s_feature_ro_compat &
2002              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2003             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2004                 return 0;
2005
2006         return 1;
2007 }
2008
2009 /* Called at mount-time, super-block is locked */
2010 static int ext4_check_descriptors(struct super_block *sb,
2011                                   ext4_group_t *first_not_zeroed)
2012 {
2013         struct ext4_sb_info *sbi = EXT4_SB(sb);
2014         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2015         ext4_fsblk_t last_block;
2016         ext4_fsblk_t block_bitmap;
2017         ext4_fsblk_t inode_bitmap;
2018         ext4_fsblk_t inode_table;
2019         int flexbg_flag = 0;
2020         ext4_group_t i, grp = sbi->s_groups_count;
2021
2022         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2023                 flexbg_flag = 1;
2024
2025         ext4_debug("Checking group descriptors");
2026
2027         for (i = 0; i < sbi->s_groups_count; i++) {
2028                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2029
2030                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2031                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2032                 else
2033                         last_block = first_block +
2034                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2035
2036                 if ((grp == sbi->s_groups_count) &&
2037                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2038                         grp = i;
2039
2040                 block_bitmap = ext4_block_bitmap(sb, gdp);
2041                 if (block_bitmap < first_block || block_bitmap > last_block) {
2042                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2043                                "Block bitmap for group %u not in group "
2044                                "(block %llu)!", i, block_bitmap);
2045                         return 0;
2046                 }
2047                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2048                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2049                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2050                                "Inode bitmap for group %u not in group "
2051                                "(block %llu)!", i, inode_bitmap);
2052                         return 0;
2053                 }
2054                 inode_table = ext4_inode_table(sb, gdp);
2055                 if (inode_table < first_block ||
2056                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2057                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2058                                "Inode table for group %u not in group "
2059                                "(block %llu)!", i, inode_table);
2060                         return 0;
2061                 }
2062                 ext4_lock_group(sb, i);
2063                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2064                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2065                                  "Checksum for group %u failed (%u!=%u)",
2066                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2067                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2068                         if (!(sb->s_flags & MS_RDONLY)) {
2069                                 ext4_unlock_group(sb, i);
2070                                 return 0;
2071                         }
2072                 }
2073                 ext4_unlock_group(sb, i);
2074                 if (!flexbg_flag)
2075                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2076         }
2077         if (NULL != first_not_zeroed)
2078                 *first_not_zeroed = grp;
2079
2080         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2081         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2082         return 1;
2083 }
2084
2085 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2086  * the superblock) which were deleted from all directories, but held open by
2087  * a process at the time of a crash.  We walk the list and try to delete these
2088  * inodes at recovery time (only with a read-write filesystem).
2089  *
2090  * In order to keep the orphan inode chain consistent during traversal (in
2091  * case of crash during recovery), we link each inode into the superblock
2092  * orphan list_head and handle it the same way as an inode deletion during
2093  * normal operation (which journals the operations for us).
2094  *
2095  * We only do an iget() and an iput() on each inode, which is very safe if we
2096  * accidentally point at an in-use or already deleted inode.  The worst that
2097  * can happen in this case is that we get a "bit already cleared" message from
2098  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2099  * e2fsck was run on this filesystem, and it must have already done the orphan
2100  * inode cleanup for us, so we can safely abort without any further action.
2101  */
2102 static void ext4_orphan_cleanup(struct super_block *sb,
2103                                 struct ext4_super_block *es)
2104 {
2105         unsigned int s_flags = sb->s_flags;
2106         int nr_orphans = 0, nr_truncates = 0;
2107 #ifdef CONFIG_QUOTA
2108         int i;
2109 #endif
2110         if (!es->s_last_orphan) {
2111                 jbd_debug(4, "no orphan inodes to clean up\n");
2112                 return;
2113         }
2114
2115         if (bdev_read_only(sb->s_bdev)) {
2116                 ext4_msg(sb, KERN_ERR, "write access "
2117                         "unavailable, skipping orphan cleanup");
2118                 return;
2119         }
2120
2121         /* Check if feature set would not allow a r/w mount */
2122         if (!ext4_feature_set_ok(sb, 0)) {
2123                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2124                          "unknown ROCOMPAT features");
2125                 return;
2126         }
2127
2128         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2129                 if (es->s_last_orphan)
2130                         jbd_debug(1, "Errors on filesystem, "
2131                                   "clearing orphan list.\n");
2132                 es->s_last_orphan = 0;
2133                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2134                 return;
2135         }
2136
2137         if (s_flags & MS_RDONLY) {
2138                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2139                 sb->s_flags &= ~MS_RDONLY;
2140         }
2141 #ifdef CONFIG_QUOTA
2142         /* Needed for iput() to work correctly and not trash data */
2143         sb->s_flags |= MS_ACTIVE;
2144         /* Turn on quotas so that they are updated correctly */
2145         for (i = 0; i < MAXQUOTAS; i++) {
2146                 if (EXT4_SB(sb)->s_qf_names[i]) {
2147                         int ret = ext4_quota_on_mount(sb, i);
2148                         if (ret < 0)
2149                                 ext4_msg(sb, KERN_ERR,
2150                                         "Cannot turn on journaled "
2151                                         "quota: error %d", ret);
2152                 }
2153         }
2154 #endif
2155
2156         while (es->s_last_orphan) {
2157                 struct inode *inode;
2158
2159                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2160                 if (IS_ERR(inode)) {
2161                         es->s_last_orphan = 0;
2162                         break;
2163                 }
2164
2165                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2166                 dquot_initialize(inode);
2167                 if (inode->i_nlink) {
2168                         ext4_msg(sb, KERN_DEBUG,
2169                                 "%s: truncating inode %lu to %lld bytes",
2170                                 __func__, inode->i_ino, inode->i_size);
2171                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2172                                   inode->i_ino, inode->i_size);
2173                         ext4_truncate(inode);
2174                         nr_truncates++;
2175                 } else {
2176                         ext4_msg(sb, KERN_DEBUG,
2177                                 "%s: deleting unreferenced inode %lu",
2178                                 __func__, inode->i_ino);
2179                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2180                                   inode->i_ino);
2181                         nr_orphans++;
2182                 }
2183                 iput(inode);  /* The delete magic happens here! */
2184         }
2185
2186 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2187
2188         if (nr_orphans)
2189                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2190                        PLURAL(nr_orphans));
2191         if (nr_truncates)
2192                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2193                        PLURAL(nr_truncates));
2194 #ifdef CONFIG_QUOTA
2195         /* Turn quotas off */
2196         for (i = 0; i < MAXQUOTAS; i++) {
2197                 if (sb_dqopt(sb)->files[i])
2198                         dquot_quota_off(sb, i);
2199         }
2200 #endif
2201         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2202 }
2203
2204 /*
2205  * Maximal extent format file size.
2206  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2207  * extent format containers, within a sector_t, and within i_blocks
2208  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2209  * so that won't be a limiting factor.
2210  *
2211  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2212  */
2213 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2214 {
2215         loff_t res;
2216         loff_t upper_limit = MAX_LFS_FILESIZE;
2217
2218         /* small i_blocks in vfs inode? */
2219         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2220                 /*
2221                  * CONFIG_LBDAF is not enabled implies the inode
2222                  * i_block represent total blocks in 512 bytes
2223                  * 32 == size of vfs inode i_blocks * 8
2224                  */
2225                 upper_limit = (1LL << 32) - 1;
2226
2227                 /* total blocks in file system block size */
2228                 upper_limit >>= (blkbits - 9);
2229                 upper_limit <<= blkbits;
2230         }
2231
2232         /* 32-bit extent-start container, ee_block */
2233         res = 1LL << 32;
2234         res <<= blkbits;
2235         res -= 1;
2236
2237         /* Sanity check against vm- & vfs- imposed limits */
2238         if (res > upper_limit)
2239                 res = upper_limit;
2240
2241         return res;
2242 }
2243
2244 /*
2245  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2246  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2247  * We need to be 1 filesystem block less than the 2^48 sector limit.
2248  */
2249 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2250 {
2251         loff_t res = EXT4_NDIR_BLOCKS;
2252         int meta_blocks;
2253         loff_t upper_limit;
2254         /* This is calculated to be the largest file size for a dense, block
2255          * mapped file such that the file's total number of 512-byte sectors,
2256          * including data and all indirect blocks, does not exceed (2^48 - 1).
2257          *
2258          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2259          * number of 512-byte sectors of the file.
2260          */
2261
2262         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2263                 /*
2264                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2265                  * the inode i_block field represents total file blocks in
2266                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2267                  */
2268                 upper_limit = (1LL << 32) - 1;
2269
2270                 /* total blocks in file system block size */
2271                 upper_limit >>= (bits - 9);
2272
2273         } else {
2274                 /*
2275                  * We use 48 bit ext4_inode i_blocks
2276                  * With EXT4_HUGE_FILE_FL set the i_blocks
2277                  * represent total number of blocks in
2278                  * file system block size
2279                  */
2280                 upper_limit = (1LL << 48) - 1;
2281
2282         }
2283
2284         /* indirect blocks */
2285         meta_blocks = 1;
2286         /* double indirect blocks */
2287         meta_blocks += 1 + (1LL << (bits-2));
2288         /* tripple indirect blocks */
2289         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2290
2291         upper_limit -= meta_blocks;
2292         upper_limit <<= bits;
2293
2294         res += 1LL << (bits-2);
2295         res += 1LL << (2*(bits-2));
2296         res += 1LL << (3*(bits-2));
2297         res <<= bits;
2298         if (res > upper_limit)
2299                 res = upper_limit;
2300
2301         if (res > MAX_LFS_FILESIZE)
2302                 res = MAX_LFS_FILESIZE;
2303
2304         return res;
2305 }
2306
2307 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2308                                    ext4_fsblk_t logical_sb_block, int nr)
2309 {
2310         struct ext4_sb_info *sbi = EXT4_SB(sb);
2311         ext4_group_t bg, first_meta_bg;
2312         int has_super = 0;
2313
2314         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2315
2316         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2317             nr < first_meta_bg)
2318                 return logical_sb_block + nr + 1;
2319         bg = sbi->s_desc_per_block * nr;
2320         if (ext4_bg_has_super(sb, bg))
2321                 has_super = 1;
2322
2323         return (has_super + ext4_group_first_block_no(sb, bg));
2324 }
2325
2326 /**
2327  * ext4_get_stripe_size: Get the stripe size.
2328  * @sbi: In memory super block info
2329  *
2330  * If we have specified it via mount option, then
2331  * use the mount option value. If the value specified at mount time is
2332  * greater than the blocks per group use the super block value.
2333  * If the super block value is greater than blocks per group return 0.
2334  * Allocator needs it be less than blocks per group.
2335  *
2336  */
2337 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2338 {
2339         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2340         unsigned long stripe_width =
2341                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2342
2343         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2344                 return sbi->s_stripe;
2345
2346         if (stripe_width <= sbi->s_blocks_per_group)
2347                 return stripe_width;
2348
2349         if (stride <= sbi->s_blocks_per_group)
2350                 return stride;
2351
2352         return 0;
2353 }
2354
2355 /* sysfs supprt */
2356
2357 struct ext4_attr {
2358         struct attribute attr;
2359         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2360         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2361                          const char *, size_t);
2362         int offset;
2363 };
2364
2365 static int parse_strtoul(const char *buf,
2366                 unsigned long max, unsigned long *value)
2367 {
2368         char *endp;
2369
2370         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2371         endp = skip_spaces(endp);
2372         if (*endp || *value > max)
2373                 return -EINVAL;
2374
2375         return 0;
2376 }
2377
2378 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2379                                               struct ext4_sb_info *sbi,
2380                                               char *buf)
2381 {
2382         return snprintf(buf, PAGE_SIZE, "%llu\n",
2383                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2384 }
2385
2386 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2387                                          struct ext4_sb_info *sbi, char *buf)
2388 {
2389         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2390
2391         if (!sb->s_bdev->bd_part)
2392                 return snprintf(buf, PAGE_SIZE, "0\n");
2393         return snprintf(buf, PAGE_SIZE, "%lu\n",
2394                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2395                          sbi->s_sectors_written_start) >> 1);
2396 }
2397
2398 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2399                                           struct ext4_sb_info *sbi, char *buf)
2400 {
2401         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2402
2403         if (!sb->s_bdev->bd_part)
2404                 return snprintf(buf, PAGE_SIZE, "0\n");
2405         return snprintf(buf, PAGE_SIZE, "%llu\n",
2406                         (unsigned long long)(sbi->s_kbytes_written +
2407                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2408                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2409 }
2410
2411 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2412                                           struct ext4_sb_info *sbi,
2413                                           const char *buf, size_t count)
2414 {
2415         unsigned long t;
2416
2417         if (parse_strtoul(buf, 0x40000000, &t))
2418                 return -EINVAL;
2419
2420         if (t && !is_power_of_2(t))
2421                 return -EINVAL;
2422
2423         sbi->s_inode_readahead_blks = t;
2424         return count;
2425 }
2426
2427 static ssize_t sbi_ui_show(struct ext4_attr *a,
2428                            struct ext4_sb_info *sbi, char *buf)
2429 {
2430         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2431
2432         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2433 }
2434
2435 static ssize_t sbi_ui_store(struct ext4_attr *a,
2436                             struct ext4_sb_info *sbi,
2437                             const char *buf, size_t count)
2438 {
2439         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2440         unsigned long t;
2441
2442         if (parse_strtoul(buf, 0xffffffff, &t))
2443                 return -EINVAL;
2444         *ui = t;
2445         return count;
2446 }
2447
2448 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2449 static struct ext4_attr ext4_attr_##_name = {                   \
2450         .attr = {.name = __stringify(_name), .mode = _mode },   \
2451         .show   = _show,                                        \
2452         .store  = _store,                                       \
2453         .offset = offsetof(struct ext4_sb_info, _elname),       \
2454 }
2455 #define EXT4_ATTR(name, mode, show, store) \
2456 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2457
2458 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2459 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2460 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2461 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2462         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2463 #define ATTR_LIST(name) &ext4_attr_##name.attr
2464
2465 EXT4_RO_ATTR(delayed_allocation_blocks);
2466 EXT4_RO_ATTR(session_write_kbytes);
2467 EXT4_RO_ATTR(lifetime_write_kbytes);
2468 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2469                  inode_readahead_blks_store, s_inode_readahead_blks);
2470 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2471 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2472 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2473 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2474 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2475 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2476 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2477 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2478
2479 static struct attribute *ext4_attrs[] = {
2480         ATTR_LIST(delayed_allocation_blocks),
2481         ATTR_LIST(session_write_kbytes),
2482         ATTR_LIST(lifetime_write_kbytes),
2483         ATTR_LIST(inode_readahead_blks),
2484         ATTR_LIST(inode_goal),
2485         ATTR_LIST(mb_stats),
2486         ATTR_LIST(mb_max_to_scan),
2487         ATTR_LIST(mb_min_to_scan),
2488         ATTR_LIST(mb_order2_req),
2489         ATTR_LIST(mb_stream_req),
2490         ATTR_LIST(mb_group_prealloc),
2491         ATTR_LIST(max_writeback_mb_bump),
2492         NULL,
2493 };
2494
2495 /* Features this copy of ext4 supports */
2496 EXT4_INFO_ATTR(lazy_itable_init);
2497 EXT4_INFO_ATTR(batched_discard);
2498
2499 static struct attribute *ext4_feat_attrs[] = {
2500         ATTR_LIST(lazy_itable_init),
2501         ATTR_LIST(batched_discard),
2502         NULL,
2503 };
2504
2505 static ssize_t ext4_attr_show(struct kobject *kobj,
2506                               struct attribute *attr, char *buf)
2507 {
2508         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2509                                                 s_kobj);
2510         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2511
2512         return a->show ? a->show(a, sbi, buf) : 0;
2513 }
2514
2515 static ssize_t ext4_attr_store(struct kobject *kobj,
2516                                struct attribute *attr,
2517                                const char *buf, size_t len)
2518 {
2519         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2520                                                 s_kobj);
2521         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2522
2523         return a->store ? a->store(a, sbi, buf, len) : 0;
2524 }
2525
2526 static void ext4_sb_release(struct kobject *kobj)
2527 {
2528         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2529                                                 s_kobj);
2530         complete(&sbi->s_kobj_unregister);
2531 }
2532
2533 static const struct sysfs_ops ext4_attr_ops = {
2534         .show   = ext4_attr_show,
2535         .store  = ext4_attr_store,
2536 };
2537
2538 static struct kobj_type ext4_ktype = {
2539         .default_attrs  = ext4_attrs,
2540         .sysfs_ops      = &ext4_attr_ops,
2541         .release        = ext4_sb_release,
2542 };
2543
2544 static void ext4_feat_release(struct kobject *kobj)
2545 {
2546         complete(&ext4_feat->f_kobj_unregister);
2547 }
2548
2549 static struct kobj_type ext4_feat_ktype = {
2550         .default_attrs  = ext4_feat_attrs,
2551         .sysfs_ops      = &ext4_attr_ops,
2552         .release        = ext4_feat_release,
2553 };
2554
2555 /*
2556  * Check whether this filesystem can be mounted based on
2557  * the features present and the RDONLY/RDWR mount requested.
2558  * Returns 1 if this filesystem can be mounted as requested,
2559  * 0 if it cannot be.
2560  */
2561 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2562 {
2563         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2564                 ext4_msg(sb, KERN_ERR,
2565                         "Couldn't mount because of "
2566                         "unsupported optional features (%x)",
2567                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2568                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2569                 return 0;
2570         }
2571
2572         if (readonly)
2573                 return 1;
2574
2575         /* Check that feature set is OK for a read-write mount */
2576         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2577                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2578                          "unsupported optional features (%x)",
2579                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2580                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2581                 return 0;
2582         }
2583         /*
2584          * Large file size enabled file system can only be mounted
2585          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2586          */
2587         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2588                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2589                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2590                                  "cannot be mounted RDWR without "
2591                                  "CONFIG_LBDAF");
2592                         return 0;
2593                 }
2594         }
2595         return 1;
2596 }
2597
2598 /*
2599  * This function is called once a day if we have errors logged
2600  * on the file system
2601  */
2602 static void print_daily_error_info(unsigned long arg)
2603 {
2604         struct super_block *sb = (struct super_block *) arg;
2605         struct ext4_sb_info *sbi;
2606         struct ext4_super_block *es;
2607
2608         sbi = EXT4_SB(sb);
2609         es = sbi->s_es;
2610
2611         if (es->s_error_count)
2612                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2613                          le32_to_cpu(es->s_error_count));
2614         if (es->s_first_error_time) {
2615                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2616                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2617                        (int) sizeof(es->s_first_error_func),
2618                        es->s_first_error_func,
2619                        le32_to_cpu(es->s_first_error_line));
2620                 if (es->s_first_error_ino)
2621                         printk(": inode %u",
2622                                le32_to_cpu(es->s_first_error_ino));
2623                 if (es->s_first_error_block)
2624                         printk(": block %llu", (unsigned long long)
2625                                le64_to_cpu(es->s_first_error_block));
2626                 printk("\n");
2627         }
2628         if (es->s_last_error_time) {
2629                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2630                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2631                        (int) sizeof(es->s_last_error_func),
2632                        es->s_last_error_func,
2633                        le32_to_cpu(es->s_last_error_line));
2634                 if (es->s_last_error_ino)
2635                         printk(": inode %u",
2636                                le32_to_cpu(es->s_last_error_ino));
2637                 if (es->s_last_error_block)
2638                         printk(": block %llu", (unsigned long long)
2639                                le64_to_cpu(es->s_last_error_block));
2640                 printk("\n");
2641         }
2642         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2643 }
2644
2645 static void ext4_lazyinode_timeout(unsigned long data)
2646 {
2647         struct task_struct *p = (struct task_struct *)data;
2648         wake_up_process(p);
2649 }
2650
2651 /* Find next suitable group and run ext4_init_inode_table */
2652 static int ext4_run_li_request(struct ext4_li_request *elr)
2653 {
2654         struct ext4_group_desc *gdp = NULL;
2655         ext4_group_t group, ngroups;
2656         struct super_block *sb;
2657         unsigned long timeout = 0;
2658         int ret = 0;
2659
2660         sb = elr->lr_super;
2661         ngroups = EXT4_SB(sb)->s_groups_count;
2662
2663         for (group = elr->lr_next_group; group < ngroups; group++) {
2664                 gdp = ext4_get_group_desc(sb, group, NULL);
2665                 if (!gdp) {
2666                         ret = 1;
2667                         break;
2668                 }
2669
2670                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2671                         break;
2672         }
2673
2674         if (group == ngroups)
2675                 ret = 1;
2676
2677         if (!ret) {
2678                 timeout = jiffies;
2679                 ret = ext4_init_inode_table(sb, group,
2680                                             elr->lr_timeout ? 0 : 1);
2681                 if (elr->lr_timeout == 0) {
2682                         timeout = jiffies - timeout;
2683                         if (elr->lr_sbi->s_li_wait_mult)
2684                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2685                         else
2686                                 timeout *= 20;
2687                         elr->lr_timeout = timeout;
2688                 }
2689                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2690                 elr->lr_next_group = group + 1;
2691         }
2692
2693         return ret;
2694 }
2695
2696 /*
2697  * Remove lr_request from the list_request and free the
2698  * request tructure. Should be called with li_list_mtx held
2699  */
2700 static void ext4_remove_li_request(struct ext4_li_request *elr)
2701 {
2702         struct ext4_sb_info *sbi;
2703
2704         if (!elr)
2705                 return;
2706
2707         sbi = elr->lr_sbi;
2708
2709         list_del(&elr->lr_request);
2710         sbi->s_li_request = NULL;
2711         kfree(elr);
2712 }
2713
2714 static void ext4_unregister_li_request(struct super_block *sb)
2715 {
2716         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2717
2718         if (!ext4_li_info)
2719                 return;
2720
2721         mutex_lock(&ext4_li_info->li_list_mtx);
2722         ext4_remove_li_request(elr);
2723         mutex_unlock(&ext4_li_info->li_list_mtx);
2724 }
2725
2726 static struct task_struct *ext4_lazyinit_task;
2727
2728 /*
2729  * This is the function where ext4lazyinit thread lives. It walks
2730  * through the request list searching for next scheduled filesystem.
2731  * When such a fs is found, run the lazy initialization request
2732  * (ext4_rn_li_request) and keep track of the time spend in this
2733  * function. Based on that time we compute next schedule time of
2734  * the request. When walking through the list is complete, compute
2735  * next waking time and put itself into sleep.
2736  */
2737 static int ext4_lazyinit_thread(void *arg)
2738 {
2739         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2740         struct list_head *pos, *n;
2741         struct ext4_li_request *elr;
2742         unsigned long next_wakeup;
2743         DEFINE_WAIT(wait);
2744
2745         BUG_ON(NULL == eli);
2746
2747         eli->li_timer.data = (unsigned long)current;
2748         eli->li_timer.function = ext4_lazyinode_timeout;
2749
2750         eli->li_task = current;
2751         wake_up(&eli->li_wait_task);
2752
2753 cont_thread:
2754         while (true) {
2755                 next_wakeup = MAX_JIFFY_OFFSET;
2756
2757                 mutex_lock(&eli->li_list_mtx);
2758                 if (list_empty(&eli->li_request_list)) {
2759                         mutex_unlock(&eli->li_list_mtx);
2760                         goto exit_thread;
2761                 }
2762
2763                 list_for_each_safe(pos, n, &eli->li_request_list) {
2764                         elr = list_entry(pos, struct ext4_li_request,
2765                                          lr_request);
2766
2767                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2768                                 if (ext4_run_li_request(elr) != 0) {
2769                                         /* error, remove the lazy_init job */
2770                                         ext4_remove_li_request(elr);
2771                                         continue;
2772                                 }
2773                         }
2774
2775                         if (time_before(elr->lr_next_sched, next_wakeup))
2776                                 next_wakeup = elr->lr_next_sched;
2777                 }
2778                 mutex_unlock(&eli->li_list_mtx);
2779
2780                 if (freezing(current))
2781                         refrigerator();
2782
2783                 if ((time_after_eq(jiffies, next_wakeup)) ||
2784                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2785                         cond_resched();
2786                         continue;
2787                 }
2788
2789                 eli->li_timer.expires = next_wakeup;
2790                 add_timer(&eli->li_timer);
2791                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2792                                 TASK_INTERRUPTIBLE);
2793                 if (time_before(jiffies, next_wakeup))
2794                         schedule();
2795                 finish_wait(&eli->li_wait_daemon, &wait);
2796                 if (kthread_should_stop()) {
2797                         ext4_clear_request_list();
2798                         goto exit_thread;
2799                 }
2800         }
2801
2802 exit_thread:
2803         /*
2804          * It looks like the request list is empty, but we need
2805          * to check it under the li_list_mtx lock, to prevent any
2806          * additions into it, and of course we should lock ext4_li_mtx
2807          * to atomically free the list and ext4_li_info, because at
2808          * this point another ext4 filesystem could be registering
2809          * new one.
2810          */
2811         mutex_lock(&ext4_li_mtx);
2812         mutex_lock(&eli->li_list_mtx);
2813         if (!list_empty(&eli->li_request_list)) {
2814                 mutex_unlock(&eli->li_list_mtx);
2815                 mutex_unlock(&ext4_li_mtx);
2816                 goto cont_thread;
2817         }
2818         mutex_unlock(&eli->li_list_mtx);
2819         del_timer_sync(&ext4_li_info->li_timer);
2820         eli->li_task = NULL;
2821         wake_up(&eli->li_wait_task);
2822
2823         kfree(ext4_li_info);
2824         ext4_lazyinit_task = NULL;
2825         ext4_li_info = NULL;
2826         mutex_unlock(&ext4_li_mtx);
2827
2828         return 0;
2829 }
2830
2831 static void ext4_clear_request_list(void)
2832 {
2833         struct list_head *pos, *n;
2834         struct ext4_li_request *elr;
2835
2836         mutex_lock(&ext4_li_info->li_list_mtx);
2837         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2838                 elr = list_entry(pos, struct ext4_li_request,
2839                                  lr_request);
2840                 ext4_remove_li_request(elr);
2841         }
2842         mutex_unlock(&ext4_li_info->li_list_mtx);
2843 }
2844
2845 static int ext4_run_lazyinit_thread(void)
2846 {
2847         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2848                                          ext4_li_info, "ext4lazyinit");
2849         if (IS_ERR(ext4_lazyinit_task)) {
2850                 int err = PTR_ERR(ext4_lazyinit_task);
2851                 ext4_clear_request_list();
2852                 del_timer_sync(&ext4_li_info->li_timer);
2853                 kfree(ext4_li_info);
2854                 ext4_li_info = NULL;
2855                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2856                                  "initialization thread\n",
2857                                  err);
2858                 return err;
2859         }
2860         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2861
2862         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2863         return 0;
2864 }
2865
2866 /*
2867  * Check whether it make sense to run itable init. thread or not.
2868  * If there is at least one uninitialized inode table, return
2869  * corresponding group number, else the loop goes through all
2870  * groups and return total number of groups.
2871  */
2872 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2873 {
2874         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2875         struct ext4_group_desc *gdp = NULL;
2876
2877         for (group = 0; group < ngroups; group++) {
2878                 gdp = ext4_get_group_desc(sb, group, NULL);
2879                 if (!gdp)
2880                         continue;
2881
2882                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2883                         break;
2884         }
2885
2886         return group;
2887 }
2888
2889 static int ext4_li_info_new(void)
2890 {
2891         struct ext4_lazy_init *eli = NULL;
2892
2893         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2894         if (!eli)
2895                 return -ENOMEM;
2896
2897         eli->li_task = NULL;
2898         INIT_LIST_HEAD(&eli->li_request_list);
2899         mutex_init(&eli->li_list_mtx);
2900
2901         init_waitqueue_head(&eli->li_wait_daemon);
2902         init_waitqueue_head(&eli->li_wait_task);
2903         init_timer(&eli->li_timer);
2904         eli->li_state |= EXT4_LAZYINIT_QUIT;
2905
2906         ext4_li_info = eli;
2907
2908         return 0;
2909 }
2910
2911 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2912                                             ext4_group_t start)
2913 {
2914         struct ext4_sb_info *sbi = EXT4_SB(sb);
2915         struct ext4_li_request *elr;
2916         unsigned long rnd;
2917
2918         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2919         if (!elr)
2920                 return NULL;
2921
2922         elr->lr_super = sb;
2923         elr->lr_sbi = sbi;
2924         elr->lr_next_group = start;
2925
2926         /*
2927          * Randomize first schedule time of the request to
2928          * spread the inode table initialization requests
2929          * better.
2930          */
2931         get_random_bytes(&rnd, sizeof(rnd));
2932         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2933                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2934
2935         return elr;
2936 }
2937
2938 static int ext4_register_li_request(struct super_block *sb,
2939                                     ext4_group_t first_not_zeroed)
2940 {
2941         struct ext4_sb_info *sbi = EXT4_SB(sb);
2942         struct ext4_li_request *elr;
2943         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2944         int ret = 0;
2945
2946         if (sbi->s_li_request != NULL)
2947                 return 0;
2948
2949         if (first_not_zeroed == ngroups ||
2950             (sb->s_flags & MS_RDONLY) ||
2951             !test_opt(sb, INIT_INODE_TABLE)) {
2952                 sbi->s_li_request = NULL;
2953                 return 0;
2954         }
2955
2956         if (first_not_zeroed == ngroups) {
2957                 sbi->s_li_request = NULL;
2958                 return 0;
2959         }
2960
2961         elr = ext4_li_request_new(sb, first_not_zeroed);
2962         if (!elr)
2963                 return -ENOMEM;
2964
2965         mutex_lock(&ext4_li_mtx);
2966
2967         if (NULL == ext4_li_info) {
2968                 ret = ext4_li_info_new();
2969                 if (ret)
2970                         goto out;
2971         }
2972
2973         mutex_lock(&ext4_li_info->li_list_mtx);
2974         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2975         mutex_unlock(&ext4_li_info->li_list_mtx);
2976
2977         sbi->s_li_request = elr;
2978         /*
2979          * set elr to NULL here since it has been inserted to
2980          * the request_list and the removal and free of it is
2981          * handled by ext4_clear_request_list from now on.
2982          */
2983         elr = NULL;
2984
2985         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2986                 ret = ext4_run_lazyinit_thread();
2987                 if (ret)
2988                         goto out;
2989         }
2990 out:
2991         mutex_unlock(&ext4_li_mtx);
2992         if (ret)
2993                 kfree(elr);
2994         return ret;
2995 }
2996
2997 /*
2998  * We do not need to lock anything since this is called on
2999  * module unload.
3000  */
3001 static void ext4_destroy_lazyinit_thread(void)
3002 {
3003         /*
3004          * If thread exited earlier
3005          * there's nothing to be done.
3006          */
3007         if (!ext4_li_info || !ext4_lazyinit_task)
3008                 return;
3009
3010         kthread_stop(ext4_lazyinit_task);
3011 }
3012
3013 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3014                                 __releases(kernel_lock)
3015                                 __acquires(kernel_lock)
3016 {
3017         char *orig_data = kstrdup(data, GFP_KERNEL);
3018         struct buffer_head *bh;
3019         struct ext4_super_block *es = NULL;
3020         struct ext4_sb_info *sbi;
3021         ext4_fsblk_t block;
3022         ext4_fsblk_t sb_block = get_sb_block(&data);
3023         ext4_fsblk_t logical_sb_block;
3024         unsigned long offset = 0;
3025         unsigned long journal_devnum = 0;
3026         unsigned long def_mount_opts;
3027         struct inode *root;
3028         char *cp;
3029         const char *descr;
3030         int ret = -ENOMEM;
3031         int blocksize;
3032         unsigned int db_count;
3033         unsigned int i;
3034         int needs_recovery, has_huge_files;
3035         __u64 blocks_count;
3036         int err;
3037         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3038         ext4_group_t first_not_zeroed;
3039
3040         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3041         if (!sbi)
3042                 goto out_free_orig;
3043
3044         sbi->s_blockgroup_lock =
3045                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3046         if (!sbi->s_blockgroup_lock) {
3047                 kfree(sbi);
3048                 goto out_free_orig;
3049         }
3050         sb->s_fs_info = sbi;
3051         sbi->s_mount_opt = 0;
3052         sbi->s_resuid = EXT4_DEF_RESUID;
3053         sbi->s_resgid = EXT4_DEF_RESGID;
3054         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3055         sbi->s_sb_block = sb_block;
3056         if (sb->s_bdev->bd_part)
3057                 sbi->s_sectors_written_start =
3058                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3059
3060         /* Cleanup superblock name */
3061         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3062                 *cp = '!';
3063
3064         ret = -EINVAL;
3065         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3066         if (!blocksize) {
3067                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3068                 goto out_fail;
3069         }
3070
3071         /*
3072          * The ext4 superblock will not be buffer aligned for other than 1kB
3073          * block sizes.  We need to calculate the offset from buffer start.
3074          */
3075         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3076                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3077                 offset = do_div(logical_sb_block, blocksize);
3078         } else {
3079                 logical_sb_block = sb_block;
3080         }
3081
3082         if (!(bh = sb_bread(sb, logical_sb_block))) {
3083                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3084                 goto out_fail;
3085         }
3086         /*
3087          * Note: s_es must be initialized as soon as possible because
3088          *       some ext4 macro-instructions depend on its value
3089          */
3090         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3091         sbi->s_es = es;
3092         sb->s_magic = le16_to_cpu(es->s_magic);
3093         if (sb->s_magic != EXT4_SUPER_MAGIC)
3094                 goto cantfind_ext4;
3095         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3096
3097         /* Set defaults before we parse the mount options */
3098         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3099         set_opt(sb, INIT_INODE_TABLE);
3100         if (def_mount_opts & EXT4_DEFM_DEBUG)
3101                 set_opt(sb, DEBUG);
3102         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3103                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3104                         "2.6.38");
3105                 set_opt(sb, GRPID);
3106         }
3107         if (def_mount_opts & EXT4_DEFM_UID16)
3108                 set_opt(sb, NO_UID32);
3109         /* xattr user namespace & acls are now defaulted on */
3110 #ifdef CONFIG_EXT4_FS_XATTR
3111         set_opt(sb, XATTR_USER);
3112 #endif
3113 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3114         set_opt(sb, POSIX_ACL);
3115 #endif
3116         set_opt(sb, MBLK_IO_SUBMIT);
3117         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3118                 set_opt(sb, JOURNAL_DATA);
3119         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3120                 set_opt(sb, ORDERED_DATA);
3121         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3122                 set_opt(sb, WRITEBACK_DATA);
3123
3124         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3125                 set_opt(sb, ERRORS_PANIC);
3126         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3127                 set_opt(sb, ERRORS_CONT);
3128         else
3129                 set_opt(sb, ERRORS_RO);
3130         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3131                 set_opt(sb, BLOCK_VALIDITY);
3132         if (def_mount_opts & EXT4_DEFM_DISCARD)
3133                 set_opt(sb, DISCARD);
3134
3135         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3136         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3137         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3138         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3139         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3140
3141         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3142                 set_opt(sb, BARRIER);
3143
3144         /*
3145          * enable delayed allocation by default
3146          * Use -o nodelalloc to turn it off
3147          */
3148         if (!IS_EXT3_SB(sb) &&
3149             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3150                 set_opt(sb, DELALLOC);
3151
3152         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3153                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3154                 ext4_msg(sb, KERN_WARNING,
3155                          "failed to parse options in superblock: %s",
3156                          sbi->s_es->s_mount_opts);
3157         }
3158         if (!parse_options((char *) data, sb, &journal_devnum,
3159                            &journal_ioprio, NULL, 0))
3160                 goto failed_mount;
3161
3162         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3163                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3164
3165         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3166             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3167              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3168              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3169                 ext4_msg(sb, KERN_WARNING,
3170                        "feature flags set on rev 0 fs, "
3171                        "running e2fsck is recommended");
3172
3173         /*
3174          * Check feature flags regardless of the revision level, since we
3175          * previously didn't change the revision level when setting the flags,
3176          * so there is a chance incompat flags are set on a rev 0 filesystem.
3177          */
3178         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3179                 goto failed_mount;
3180
3181         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3182
3183         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3184             blocksize > EXT4_MAX_BLOCK_SIZE) {
3185                 ext4_msg(sb, KERN_ERR,
3186                        "Unsupported filesystem blocksize %d", blocksize);
3187                 goto failed_mount;
3188         }
3189
3190         if (sb->s_blocksize != blocksize) {
3191                 /* Validate the filesystem blocksize */
3192                 if (!sb_set_blocksize(sb, blocksize)) {
3193                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3194                                         blocksize);
3195                         goto failed_mount;
3196                 }
3197
3198                 brelse(bh);
3199                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3200                 offset = do_div(logical_sb_block, blocksize);
3201                 bh = sb_bread(sb, logical_sb_block);
3202                 if (!bh) {
3203                         ext4_msg(sb, KERN_ERR,
3204                                "Can't read superblock on 2nd try");
3205                         goto failed_mount;
3206                 }
3207                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3208                 sbi->s_es = es;
3209                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3210                         ext4_msg(sb, KERN_ERR,
3211                                "Magic mismatch, very weird!");
3212                         goto failed_mount;
3213                 }
3214         }
3215
3216         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3217                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3218         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3219                                                       has_huge_files);
3220         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3221
3222         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3223                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3224                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3225         } else {
3226                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3227                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3228                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3229                     (!is_power_of_2(sbi->s_inode_size)) ||
3230                     (sbi->s_inode_size > blocksize)) {
3231                         ext4_msg(sb, KERN_ERR,
3232                                "unsupported inode size: %d",
3233                                sbi->s_inode_size);
3234                         goto failed_mount;
3235                 }
3236                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3237                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3238         }
3239
3240         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3241         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3242                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3243                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3244                     !is_power_of_2(sbi->s_desc_size)) {
3245                         ext4_msg(sb, KERN_ERR,
3246                                "unsupported descriptor size %lu",
3247                                sbi->s_desc_size);
3248                         goto failed_mount;
3249                 }
3250         } else
3251                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3252
3253         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3254         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3255         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3256                 goto cantfind_ext4;
3257
3258         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3259         if (sbi->s_inodes_per_block == 0)
3260                 goto cantfind_ext4;
3261         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3262                                         sbi->s_inodes_per_block;
3263         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3264         sbi->s_sbh = bh;
3265         sbi->s_mount_state = le16_to_cpu(es->s_state);
3266         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3267         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3268
3269         for (i = 0; i < 4; i++)
3270                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3271         sbi->s_def_hash_version = es->s_def_hash_version;
3272         i = le32_to_cpu(es->s_flags);
3273         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3274                 sbi->s_hash_unsigned = 3;
3275         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3276 #ifdef __CHAR_UNSIGNED__
3277                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3278                 sbi->s_hash_unsigned = 3;
3279 #else
3280                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3281 #endif
3282                 sb->s_dirt = 1;
3283         }
3284
3285         if (sbi->s_blocks_per_group > blocksize * 8) {
3286                 ext4_msg(sb, KERN_ERR,
3287                        "#blocks per group too big: %lu",
3288                        sbi->s_blocks_per_group);
3289                 goto failed_mount;
3290         }
3291         if (sbi->s_inodes_per_group > blocksize * 8) {
3292                 ext4_msg(sb, KERN_ERR,
3293                        "#inodes per group too big: %lu",
3294                        sbi->s_inodes_per_group);
3295                 goto failed_mount;
3296         }
3297
3298         /*
3299          * Test whether we have more sectors than will fit in sector_t,
3300          * and whether the max offset is addressable by the page cache.
3301          */
3302         err = generic_check_addressable(sb->s_blocksize_bits,
3303                                         ext4_blocks_count(es));
3304         if (err) {
3305                 ext4_msg(sb, KERN_ERR, "filesystem"
3306                          " too large to mount safely on this system");
3307                 if (sizeof(sector_t) < 8)
3308                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3309                 ret = err;
3310                 goto failed_mount;
3311         }
3312
3313         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3314                 goto cantfind_ext4;
3315
3316         /* check blocks count against device size */
3317         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3318         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3319                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3320                        "exceeds size of device (%llu blocks)",
3321                        ext4_blocks_count(es), blocks_count);
3322                 goto failed_mount;
3323         }
3324
3325         /*
3326          * It makes no sense for the first data block to be beyond the end
3327          * of the filesystem.
3328          */
3329         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3330                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3331                          "block %u is beyond end of filesystem (%llu)",
3332                          le32_to_cpu(es->s_first_data_block),
3333                          ext4_blocks_count(es));
3334                 goto failed_mount;
3335         }
3336         blocks_count = (ext4_blocks_count(es) -
3337                         le32_to_cpu(es->s_first_data_block) +
3338                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3339         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3340         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3341                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3342                        "(block count %llu, first data block %u, "
3343                        "blocks per group %lu)", sbi->s_groups_count,
3344                        ext4_blocks_count(es),
3345                        le32_to_cpu(es->s_first_data_block),
3346                        EXT4_BLOCKS_PER_GROUP(sb));
3347                 goto failed_mount;
3348         }
3349         sbi->s_groups_count = blocks_count;
3350         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3351                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3352         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3353                    EXT4_DESC_PER_BLOCK(sb);
3354         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3355                                     GFP_KERNEL);
3356         if (sbi->s_group_desc == NULL) {
3357                 ext4_msg(sb, KERN_ERR, "not enough memory");
3358                 goto failed_mount;
3359         }
3360
3361 #ifdef CONFIG_PROC_FS
3362         if (ext4_proc_root)
3363                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3364 #endif
3365
3366         bgl_lock_init(sbi->s_blockgroup_lock);
3367
3368         for (i = 0; i < db_count; i++) {
3369                 block = descriptor_loc(sb, logical_sb_block, i);
3370                 sbi->s_group_desc[i] = sb_bread(sb, block);
3371                 if (!sbi->s_group_desc[i]) {
3372                         ext4_msg(sb, KERN_ERR,
3373                                "can't read group descriptor %d", i);
3374                         db_count = i;
3375                         goto failed_mount2;
3376                 }
3377         }
3378         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3379                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3380                 goto failed_mount2;
3381         }
3382         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3383                 if (!ext4_fill_flex_info(sb)) {
3384                         ext4_msg(sb, KERN_ERR,
3385                                "unable to initialize "
3386                                "flex_bg meta info!");
3387                         goto failed_mount2;
3388                 }
3389
3390         sbi->s_gdb_count = db_count;
3391         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3392         spin_lock_init(&sbi->s_next_gen_lock);
3393
3394         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3395                         ext4_count_free_blocks(sb));
3396         if (!err) {
3397                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3398                                 ext4_count_free_inodes(sb));
3399         }
3400         if (!err) {
3401                 err = percpu_counter_init(&sbi->s_dirs_counter,
3402                                 ext4_count_dirs(sb));
3403         }
3404         if (!err) {
3405                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3406         }
3407         if (err) {
3408                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3409                 goto failed_mount3;
3410         }
3411
3412         sbi->s_stripe = ext4_get_stripe_size(sbi);
3413         sbi->s_max_writeback_mb_bump = 128;
3414
3415         /*
3416          * set up enough so that it can read an inode
3417          */
3418         if (!test_opt(sb, NOLOAD) &&
3419             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3420                 sb->s_op = &ext4_sops;
3421         else
3422                 sb->s_op = &ext4_nojournal_sops;
3423         sb->s_export_op = &ext4_export_ops;
3424         sb->s_xattr = ext4_xattr_handlers;
3425 #ifdef CONFIG_QUOTA
3426         sb->s_qcop = &ext4_qctl_operations;
3427         sb->dq_op = &ext4_quota_operations;
3428 #endif
3429         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3430
3431         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3432         mutex_init(&sbi->s_orphan_lock);
3433         mutex_init(&sbi->s_resize_lock);
3434
3435         sb->s_root = NULL;
3436
3437         needs_recovery = (es->s_last_orphan != 0 ||
3438                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3439                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3440
3441         /*
3442          * The first inode we look at is the journal inode.  Don't try
3443          * root first: it may be modified in the journal!
3444          */
3445         if (!test_opt(sb, NOLOAD) &&
3446             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3447                 if (ext4_load_journal(sb, es, journal_devnum))
3448                         goto failed_mount3;
3449         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3450               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3451                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3452                        "suppressed and not mounted read-only");
3453                 goto failed_mount_wq;
3454         } else {
3455                 clear_opt(sb, DATA_FLAGS);
3456                 set_opt(sb, WRITEBACK_DATA);
3457                 sbi->s_journal = NULL;
3458                 needs_recovery = 0;
3459                 goto no_journal;
3460         }
3461
3462         if (ext4_blocks_count(es) > 0xffffffffULL &&
3463             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3464                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3465                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3466                 goto failed_mount_wq;
3467         }
3468
3469         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3470                 jbd2_journal_set_features(sbi->s_journal,
3471                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3472                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3473         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3474                 jbd2_journal_set_features(sbi->s_journal,
3475                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3476                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3477                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3478         } else {
3479                 jbd2_journal_clear_features(sbi->s_journal,
3480                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3481                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3482         }
3483
3484         /* We have now updated the journal if required, so we can
3485          * validate the data journaling mode. */
3486         switch (test_opt(sb, DATA_FLAGS)) {
3487         case 0:
3488                 /* No mode set, assume a default based on the journal
3489                  * capabilities: ORDERED_DATA if the journal can
3490                  * cope, else JOURNAL_DATA
3491                  */
3492                 if (jbd2_journal_check_available_features
3493                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3494                         set_opt(sb, ORDERED_DATA);
3495                 else
3496                         set_opt(sb, JOURNAL_DATA);
3497                 break;
3498
3499         case EXT4_MOUNT_ORDERED_DATA:
3500         case EXT4_MOUNT_WRITEBACK_DATA:
3501                 if (!jbd2_journal_check_available_features
3502                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3503                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3504                                "requested data journaling mode");
3505                         goto failed_mount_wq;
3506                 }
3507         default:
3508                 break;
3509         }
3510         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3511
3512         /*
3513          * The journal may have updated the bg summary counts, so we
3514          * need to update the global counters.
3515          */
3516         percpu_counter_set(&sbi->s_freeblocks_counter,
3517                            ext4_count_free_blocks(sb));
3518         percpu_counter_set(&sbi->s_freeinodes_counter,
3519                            ext4_count_free_inodes(sb));
3520         percpu_counter_set(&sbi->s_dirs_counter,
3521                            ext4_count_dirs(sb));
3522         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3523
3524 no_journal:
3525         /*
3526          * The maximum number of concurrent works can be high and
3527          * concurrency isn't really necessary.  Limit it to 1.
3528          */
3529         EXT4_SB(sb)->dio_unwritten_wq =
3530                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3531         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3532                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3533                 goto failed_mount_wq;
3534         }
3535
3536         /*
3537          * The jbd2_journal_load will have done any necessary log recovery,
3538          * so we can safely mount the rest of the filesystem now.
3539          */
3540
3541         root = ext4_iget(sb, EXT4_ROOT_INO);
3542         if (IS_ERR(root)) {
3543                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3544                 ret = PTR_ERR(root);
3545                 root = NULL;
3546                 goto failed_mount4;
3547         }
3548         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3549                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3550                 goto failed_mount4;
3551         }
3552         sb->s_root = d_alloc_root(root);
3553         if (!sb->s_root) {
3554                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3555                 ret = -ENOMEM;
3556                 goto failed_mount4;
3557         }
3558
3559         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3560
3561         /* determine the minimum size of new large inodes, if present */
3562         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3563                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3564                                                      EXT4_GOOD_OLD_INODE_SIZE;
3565                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3566                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3567                         if (sbi->s_want_extra_isize <
3568                             le16_to_cpu(es->s_want_extra_isize))
3569                                 sbi->s_want_extra_isize =
3570                                         le16_to_cpu(es->s_want_extra_isize);
3571                         if (sbi->s_want_extra_isize <
3572                             le16_to_cpu(es->s_min_extra_isize))
3573                                 sbi->s_want_extra_isize =
3574                                         le16_to_cpu(es->s_min_extra_isize);
3575                 }
3576         }
3577         /* Check if enough inode space is available */
3578         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3579                                                         sbi->s_inode_size) {
3580                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3581                                                        EXT4_GOOD_OLD_INODE_SIZE;
3582                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3583                          "available");
3584         }
3585
3586         if (test_opt(sb, DELALLOC) &&
3587             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3588                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3589                          "requested data journaling mode");
3590                 clear_opt(sb, DELALLOC);
3591         }
3592         if (test_opt(sb, DIOREAD_NOLOCK)) {
3593                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3594                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3595                                 "option - requested data journaling mode");
3596                         clear_opt(sb, DIOREAD_NOLOCK);
3597                 }
3598                 if (sb->s_blocksize < PAGE_SIZE) {
3599                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3600                                 "option - block size is too small");
3601                         clear_opt(sb, DIOREAD_NOLOCK);
3602                 }
3603         }
3604
3605         err = ext4_setup_system_zone(sb);
3606         if (err) {
3607                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3608                          "zone (%d)", err);
3609                 goto failed_mount4;
3610         }
3611
3612         ext4_ext_init(sb);
3613         err = ext4_mb_init(sb, needs_recovery);
3614         if (err) {
3615                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3616                          err);
3617                 goto failed_mount4;
3618         }
3619
3620         err = ext4_register_li_request(sb, first_not_zeroed);
3621         if (err)
3622                 goto failed_mount4;
3623
3624         sbi->s_kobj.kset = ext4_kset;
3625         init_completion(&sbi->s_kobj_unregister);
3626         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3627                                    "%s", sb->s_id);
3628         if (err) {
3629                 ext4_mb_release(sb);
3630                 ext4_ext_release(sb);
3631                 goto failed_mount4;
3632         };
3633
3634         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3635         ext4_orphan_cleanup(sb, es);
3636         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3637         if (needs_recovery) {
3638                 ext4_msg(sb, KERN_INFO, "recovery complete");
3639                 ext4_mark_recovery_complete(sb, es);
3640         }
3641         if (EXT4_SB(sb)->s_journal) {
3642                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3643                         descr = " journalled data mode";
3644                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3645                         descr = " ordered data mode";
3646                 else
3647                         descr = " writeback data mode";
3648         } else
3649                 descr = "out journal";
3650
3651         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3652                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3653                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3654
3655         init_timer(&sbi->s_err_report);
3656         sbi->s_err_report.function = print_daily_error_info;
3657         sbi->s_err_report.data = (unsigned long) sb;
3658         if (es->s_error_count)
3659                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3660
3661         kfree(orig_data);
3662         return 0;
3663
3664 cantfind_ext4:
3665         if (!silent)
3666                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3667         goto failed_mount;
3668
3669 failed_mount4:
3670         iput(root);
3671         sb->s_root = NULL;
3672         ext4_msg(sb, KERN_ERR, "mount failed");
3673         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3674 failed_mount_wq:
3675         ext4_release_system_zone(sb);
3676         if (sbi->s_journal) {
3677                 jbd2_journal_destroy(sbi->s_journal);
3678                 sbi->s_journal = NULL;
3679         }
3680 failed_mount3:
3681         if (sbi->s_flex_groups) {
3682                 if (is_vmalloc_addr(sbi->s_flex_groups))
3683                         vfree(sbi->s_flex_groups);
3684                 else
3685                         kfree(sbi->s_flex_groups);
3686         }
3687         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3688         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3689         percpu_counter_destroy(&sbi->s_dirs_counter);
3690         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3691 failed_mount2:
3692         for (i = 0; i < db_count; i++)
3693                 brelse(sbi->s_group_desc[i]);
3694         kfree(sbi->s_group_desc);
3695 failed_mount:
3696         if (sbi->s_proc) {
3697                 remove_proc_entry(sb->s_id, ext4_proc_root);
3698         }
3699 #ifdef CONFIG_QUOTA
3700         for (i = 0; i < MAXQUOTAS; i++)
3701                 kfree(sbi->s_qf_names[i]);
3702 #endif
3703         ext4_blkdev_remove(sbi);
3704         brelse(bh);
3705 out_fail:
3706         sb->s_fs_info = NULL;
3707         kfree(sbi->s_blockgroup_lock);
3708         kfree(sbi);
3709 out_free_orig:
3710         kfree(orig_data);
3711         return ret;
3712 }
3713
3714 /*
3715  * Setup any per-fs journal parameters now.  We'll do this both on
3716  * initial mount, once the journal has been initialised but before we've
3717  * done any recovery; and again on any subsequent remount.
3718  */
3719 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3720 {
3721         struct ext4_sb_info *sbi = EXT4_SB(sb);
3722
3723         journal->j_commit_interval = sbi->s_commit_interval;
3724         journal->j_min_batch_time = sbi->s_min_batch_time;
3725         journal->j_max_batch_time = sbi->s_max_batch_time;
3726
3727         write_lock(&journal->j_state_lock);
3728         if (test_opt(sb, BARRIER))
3729                 journal->j_flags |= JBD2_BARRIER;
3730         else
3731                 journal->j_flags &= ~JBD2_BARRIER;
3732         if (test_opt(sb, DATA_ERR_ABORT))
3733                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3734         else
3735                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3736         write_unlock(&journal->j_state_lock);
3737 }
3738
3739 static journal_t *ext4_get_journal(struct super_block *sb,
3740                                    unsigned int journal_inum)
3741 {
3742         struct inode *journal_inode;
3743         journal_t *journal;
3744
3745         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3746
3747         /* First, test for the existence of a valid inode on disk.  Bad
3748          * things happen if we iget() an unused inode, as the subsequent
3749          * iput() will try to delete it. */
3750
3751         journal_inode = ext4_iget(sb, journal_inum);
3752         if (IS_ERR(journal_inode)) {
3753                 ext4_msg(sb, KERN_ERR, "no journal found");
3754                 return NULL;
3755         }
3756         if (!journal_inode->i_nlink) {
3757                 make_bad_inode(journal_inode);
3758                 iput(journal_inode);
3759                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3760                 return NULL;
3761         }
3762
3763         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3764                   journal_inode, journal_inode->i_size);
3765         if (!S_ISREG(journal_inode->i_mode)) {
3766                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3767                 iput(journal_inode);
3768                 return NULL;
3769         }
3770
3771         journal = jbd2_journal_init_inode(journal_inode);
3772         if (!journal) {
3773                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3774                 iput(journal_inode);
3775                 return NULL;
3776         }
3777         journal->j_private = sb;
3778         ext4_init_journal_params(sb, journal);
3779         return journal;
3780 }
3781
3782 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3783                                        dev_t j_dev)
3784 {
3785         struct buffer_head *bh;
3786         journal_t *journal;
3787         ext4_fsblk_t start;
3788         ext4_fsblk_t len;
3789         int hblock, blocksize;
3790         ext4_fsblk_t sb_block;
3791         unsigned long offset;
3792         struct ext4_super_block *es;
3793         struct block_device *bdev;
3794
3795         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3796
3797         bdev = ext4_blkdev_get(j_dev, sb);
3798         if (bdev == NULL)
3799                 return NULL;
3800
3801         blocksize = sb->s_blocksize;
3802         hblock = bdev_logical_block_size(bdev);
3803         if (blocksize < hblock) {
3804                 ext4_msg(sb, KERN_ERR,
3805                         "blocksize too small for journal device");
3806                 goto out_bdev;
3807         }
3808
3809         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3810         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3811         set_blocksize(bdev, blocksize);
3812         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3813                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3814                        "external journal");
3815                 goto out_bdev;
3816         }
3817
3818         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3819         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3820             !(le32_to_cpu(es->s_feature_incompat) &
3821               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3822                 ext4_msg(sb, KERN_ERR, "external journal has "
3823                                         "bad superblock");
3824                 brelse(bh);
3825                 goto out_bdev;
3826         }
3827
3828         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3829                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3830                 brelse(bh);
3831                 goto out_bdev;
3832         }
3833
3834         len = ext4_blocks_count(es);
3835         start = sb_block + 1;
3836         brelse(bh);     /* we're done with the superblock */
3837
3838         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3839                                         start, len, blocksize);
3840         if (!journal) {
3841                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3842                 goto out_bdev;
3843         }
3844         journal->j_private = sb;
3845         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3846         wait_on_buffer(journal->j_sb_buffer);
3847         if (!buffer_uptodate(journal->j_sb_buffer)) {
3848                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3849                 goto out_journal;
3850         }
3851         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3852                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3853                                         "user (unsupported) - %d",
3854                         be32_to_cpu(journal->j_superblock->s_nr_users));
3855                 goto out_journal;
3856         }
3857         EXT4_SB(sb)->journal_bdev = bdev;
3858         ext4_init_journal_params(sb, journal);
3859         return journal;
3860
3861 out_journal:
3862         jbd2_journal_destroy(journal);
3863 out_bdev:
3864         ext4_blkdev_put(bdev);
3865         return NULL;
3866 }
3867
3868 static int ext4_load_journal(struct super_block *sb,
3869                              struct ext4_super_block *es,
3870                              unsigned long journal_devnum)
3871 {
3872         journal_t *journal;
3873         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3874         dev_t journal_dev;
3875         int err = 0;
3876         int really_read_only;
3877
3878         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3879
3880         if (journal_devnum &&
3881             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3882                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3883                         "numbers have changed");
3884                 journal_dev = new_decode_dev(journal_devnum);
3885         } else
3886                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3887
3888         really_read_only = bdev_read_only(sb->s_bdev);
3889
3890         /*
3891          * Are we loading a blank journal or performing recovery after a
3892          * crash?  For recovery, we need to check in advance whether we
3893          * can get read-write access to the device.
3894          */
3895         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3896                 if (sb->s_flags & MS_RDONLY) {
3897                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3898                                         "required on readonly filesystem");
3899                         if (really_read_only) {
3900                                 ext4_msg(sb, KERN_ERR, "write access "
3901                                         "unavailable, cannot proceed");
3902                                 return -EROFS;
3903                         }
3904                         ext4_msg(sb, KERN_INFO, "write access will "
3905                                "be enabled during recovery");
3906                 }
3907         }
3908
3909         if (journal_inum && journal_dev) {
3910                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3911                        "and inode journals!");
3912                 return -EINVAL;
3913         }
3914
3915         if (journal_inum) {
3916                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3917                         return -EINVAL;
3918         } else {
3919                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3920                         return -EINVAL;
3921         }
3922
3923         if (!(journal->j_flags & JBD2_BARRIER))
3924                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3925
3926         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3927                 err = jbd2_journal_update_format(journal);
3928                 if (err)  {
3929                         ext4_msg(sb, KERN_ERR, "error updating journal");
3930                         jbd2_journal_destroy(journal);
3931                         return err;
3932                 }
3933         }
3934
3935         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3936                 err = jbd2_journal_wipe(journal, !really_read_only);
3937         if (!err) {
3938                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3939                 if (save)
3940                         memcpy(save, ((char *) es) +
3941                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3942                 err = jbd2_journal_load(journal);
3943                 if (save)
3944                         memcpy(((char *) es) + EXT4_S_ERR_START,
3945                                save, EXT4_S_ERR_LEN);
3946                 kfree(save);
3947         }
3948
3949         if (err) {
3950                 ext4_msg(sb, KERN_ERR, "error loading journal");
3951                 jbd2_journal_destroy(journal);
3952                 return err;
3953         }
3954
3955         EXT4_SB(sb)->s_journal = journal;
3956         ext4_clear_journal_err(sb, es);
3957
3958         if (!really_read_only && journal_devnum &&
3959             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3960                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3961
3962                 /* Make sure we flush the recovery flag to disk. */
3963                 ext4_commit_super(sb, 1);
3964         }
3965
3966         return 0;
3967 }
3968
3969 static int ext4_commit_super(struct super_block *sb, int sync)
3970 {
3971         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3972         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3973         int error = 0;
3974
3975         if (!sbh)
3976                 return error;
3977         if (buffer_write_io_error(sbh)) {
3978                 /*
3979                  * Oh, dear.  A previous attempt to write the
3980                  * superblock failed.  This could happen because the
3981                  * USB device was yanked out.  Or it could happen to
3982                  * be a transient write error and maybe the block will
3983                  * be remapped.  Nothing we can do but to retry the
3984                  * write and hope for the best.
3985                  */
3986                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3987                        "superblock detected");
3988                 clear_buffer_write_io_error(sbh);
3989                 set_buffer_uptodate(sbh);
3990         }
3991         /*
3992          * If the file system is mounted read-only, don't update the
3993          * superblock write time.  This avoids updating the superblock
3994          * write time when we are mounting the root file system
3995          * read/only but we need to replay the journal; at that point,
3996          * for people who are east of GMT and who make their clock
3997          * tick in localtime for Windows bug-for-bug compatibility,
3998          * the clock is set in the future, and this will cause e2fsck
3999          * to complain and force a full file system check.
4000          */
4001         if (!(sb->s_flags & MS_RDONLY))
4002                 es->s_wtime = cpu_to_le32(get_seconds());
4003         if (sb->s_bdev->bd_part)
4004                 es->s_kbytes_written =
4005                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4006                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4007                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4008         else
4009                 es->s_kbytes_written =
4010                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4011         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4012                                            &EXT4_SB(sb)->s_freeblocks_counter));
4013         es->s_free_inodes_count =
4014                 cpu_to_le32(percpu_counter_sum_positive(
4015                                 &EXT4_SB(sb)->s_freeinodes_counter));
4016         sb->s_dirt = 0;
4017         BUFFER_TRACE(sbh, "marking dirty");
4018         mark_buffer_dirty(sbh);
4019         if (sync) {
4020                 error = sync_dirty_buffer(sbh);
4021                 if (error)
4022                         return error;
4023
4024                 error = buffer_write_io_error(sbh);
4025                 if (error) {
4026                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4027                                "superblock");
4028                         clear_buffer_write_io_error(sbh);
4029                         set_buffer_uptodate(sbh);
4030                 }
4031         }
4032         return error;
4033 }
4034
4035 /*
4036  * Have we just finished recovery?  If so, and if we are mounting (or
4037  * remounting) the filesystem readonly, then we will end up with a
4038  * consistent fs on disk.  Record that fact.
4039  */
4040 static void ext4_mark_recovery_complete(struct super_block *sb,
4041                                         struct ext4_super_block *es)
4042 {
4043         journal_t *journal = EXT4_SB(sb)->s_journal;
4044
4045         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4046                 BUG_ON(journal != NULL);
4047                 return;
4048         }
4049         jbd2_journal_lock_updates(journal);
4050         if (jbd2_journal_flush(journal) < 0)
4051                 goto out;
4052
4053         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4054             sb->s_flags & MS_RDONLY) {
4055                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4056                 ext4_commit_super(sb, 1);
4057         }
4058
4059 out:
4060         jbd2_journal_unlock_updates(journal);
4061 }
4062
4063 /*
4064  * If we are mounting (or read-write remounting) a filesystem whose journal
4065  * has recorded an error from a previous lifetime, move that error to the
4066  * main filesystem now.
4067  */
4068 static void ext4_clear_journal_err(struct super_block *sb,
4069                                    struct ext4_super_block *es)
4070 {
4071         journal_t *journal;
4072         int j_errno;
4073         const char *errstr;
4074
4075         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4076
4077         journal = EXT4_SB(sb)->s_journal;
4078
4079         /*
4080          * Now check for any error status which may have been recorded in the
4081          * journal by a prior ext4_error() or ext4_abort()
4082          */
4083
4084         j_errno = jbd2_journal_errno(journal);
4085         if (j_errno) {
4086                 char nbuf[16];
4087
4088                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4089                 ext4_warning(sb, "Filesystem error recorded "
4090                              "from previous mount: %s", errstr);
4091                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4092
4093                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4094                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4095                 ext4_commit_super(sb, 1);
4096
4097                 jbd2_journal_clear_err(journal);
4098         }
4099 }
4100
4101 /*
4102  * Force the running and committing transactions to commit,
4103  * and wait on the commit.
4104  */
4105 int ext4_force_commit(struct super_block *sb)
4106 {
4107         journal_t *journal;
4108         int ret = 0;
4109
4110         if (sb->s_flags & MS_RDONLY)
4111                 return 0;
4112
4113         journal = EXT4_SB(sb)->s_journal;
4114         if (journal) {
4115                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4116                 ret = ext4_journal_force_commit(journal);
4117         }
4118
4119         return ret;
4120 }
4121
4122 static void ext4_write_super(struct super_block *sb)
4123 {
4124         lock_super(sb);
4125         ext4_commit_super(sb, 1);
4126         unlock_super(sb);
4127 }
4128
4129 static int ext4_sync_fs(struct super_block *sb, int wait)
4130 {
4131         int ret = 0;
4132         tid_t target;
4133         struct ext4_sb_info *sbi = EXT4_SB(sb);
4134
4135         trace_ext4_sync_fs(sb, wait);
4136         flush_workqueue(sbi->dio_unwritten_wq);
4137         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4138                 if (wait)
4139                         jbd2_log_wait_commit(sbi->s_journal, target);
4140         }
4141         return ret;
4142 }
4143
4144 /*
4145  * LVM calls this function before a (read-only) snapshot is created.  This
4146  * gives us a chance to flush the journal completely and mark the fs clean.
4147  */
4148 static int ext4_freeze(struct super_block *sb)
4149 {
4150         int error = 0;
4151         journal_t *journal;
4152
4153         if (sb->s_flags & MS_RDONLY)
4154                 return 0;
4155
4156         journal = EXT4_SB(sb)->s_journal;
4157
4158         /* Now we set up the journal barrier. */
4159         jbd2_journal_lock_updates(journal);
4160
4161         /*
4162          * Don't clear the needs_recovery flag if we failed to flush
4163          * the journal.
4164          */
4165         error = jbd2_journal_flush(journal);
4166         if (error < 0)
4167                 goto out;
4168
4169         /* Journal blocked and flushed, clear needs_recovery flag. */
4170         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4171         error = ext4_commit_super(sb, 1);
4172 out:
4173         /* we rely on s_frozen to stop further updates */
4174         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4175         return error;
4176 }
4177
4178 /*
4179  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4180  * flag here, even though the filesystem is not technically dirty yet.
4181  */
4182 static int ext4_unfreeze(struct super_block *sb)
4183 {
4184         if (sb->s_flags & MS_RDONLY)
4185                 return 0;
4186
4187         lock_super(sb);
4188         /* Reset the needs_recovery flag before the fs is unlocked. */
4189         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4190         ext4_commit_super(sb, 1);
4191         unlock_super(sb);
4192         return 0;
4193 }
4194
4195 /*
4196  * Structure to save mount options for ext4_remount's benefit
4197  */
4198 struct ext4_mount_options {
4199         unsigned long s_mount_opt;
4200         unsigned long s_mount_opt2;
4201         uid_t s_resuid;
4202         gid_t s_resgid;
4203         unsigned long s_commit_interval;
4204         u32 s_min_batch_time, s_max_batch_time;
4205 #ifdef CONFIG_QUOTA
4206         int s_jquota_fmt;
4207         char *s_qf_names[MAXQUOTAS];
4208 #endif
4209 };
4210
4211 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4212 {
4213         struct ext4_super_block *es;
4214         struct ext4_sb_info *sbi = EXT4_SB(sb);
4215         ext4_fsblk_t n_blocks_count = 0;
4216         unsigned long old_sb_flags;
4217         struct ext4_mount_options old_opts;
4218         int enable_quota = 0;
4219         ext4_group_t g;
4220         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4221         int err;
4222 #ifdef CONFIG_QUOTA
4223         int i;
4224 #endif
4225         char *orig_data = kstrdup(data, GFP_KERNEL);
4226
4227         /* Store the original options */
4228         lock_super(sb);
4229         old_sb_flags = sb->s_flags;
4230         old_opts.s_mount_opt = sbi->s_mount_opt;
4231         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4232         old_opts.s_resuid = sbi->s_resuid;
4233         old_opts.s_resgid = sbi->s_resgid;
4234         old_opts.s_commit_interval = sbi->s_commit_interval;
4235         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4236         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4237 #ifdef CONFIG_QUOTA
4238         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4239         for (i = 0; i < MAXQUOTAS; i++)
4240                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4241 #endif
4242         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4243                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4244
4245         /*
4246          * Allow the "check" option to be passed as a remount option.
4247          */
4248         if (!parse_options(data, sb, NULL, &journal_ioprio,
4249                            &n_blocks_count, 1)) {
4250                 err = -EINVAL;
4251                 goto restore_opts;
4252         }
4253
4254         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4255                 ext4_abort(sb, "Abort forced by user");
4256
4257         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4258                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4259
4260         es = sbi->s_es;
4261
4262         if (sbi->s_journal) {
4263                 ext4_init_journal_params(sb, sbi->s_journal);
4264                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4265         }
4266
4267         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4268                 n_blocks_count > ext4_blocks_count(es)) {
4269                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4270                         err = -EROFS;
4271                         goto restore_opts;
4272                 }
4273
4274                 if (*flags & MS_RDONLY) {
4275                         err = dquot_suspend(sb, -1);
4276                         if (err < 0)
4277                                 goto restore_opts;
4278
4279                         /*
4280                          * First of all, the unconditional stuff we have to do
4281                          * to disable replay of the journal when we next remount
4282                          */
4283                         sb->s_flags |= MS_RDONLY;
4284
4285                         /*
4286                          * OK, test if we are remounting a valid rw partition
4287                          * readonly, and if so set the rdonly flag and then
4288                          * mark the partition as valid again.
4289                          */
4290                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4291                             (sbi->s_mount_state & EXT4_VALID_FS))
4292                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4293
4294                         if (sbi->s_journal)
4295                                 ext4_mark_recovery_complete(sb, es);
4296                 } else {
4297                         /* Make sure we can mount this feature set readwrite */
4298                         if (!ext4_feature_set_ok(sb, 0)) {
4299                                 err = -EROFS;
4300                                 goto restore_opts;
4301                         }
4302                         /*
4303                          * Make sure the group descriptor checksums
4304                          * are sane.  If they aren't, refuse to remount r/w.
4305                          */
4306                         for (g = 0; g < sbi->s_groups_count; g++) {
4307                                 struct ext4_group_desc *gdp =
4308                                         ext4_get_group_desc(sb, g, NULL);
4309
4310                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4311                                         ext4_msg(sb, KERN_ERR,
4312                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4313                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4314                                                le16_to_cpu(gdp->bg_checksum));
4315                                         err = -EINVAL;
4316                                         goto restore_opts;
4317                                 }
4318                         }
4319
4320                         /*
4321                          * If we have an unprocessed orphan list hanging
4322                          * around from a previously readonly bdev mount,
4323                          * require a full umount/remount for now.
4324                          */
4325                         if (es->s_last_orphan) {
4326                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4327                                        "remount RDWR because of unprocessed "
4328                                        "orphan inode list.  Please "
4329                                        "umount/remount instead");
4330                                 err = -EINVAL;
4331                                 goto restore_opts;
4332                         }
4333
4334                         /*
4335                          * Mounting a RDONLY partition read-write, so reread
4336                          * and store the current valid flag.  (It may have
4337                          * been changed by e2fsck since we originally mounted
4338                          * the partition.)
4339                          */
4340                         if (sbi->s_journal)
4341                                 ext4_clear_journal_err(sb, es);
4342                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4343                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4344                                 goto restore_opts;
4345                         if (!ext4_setup_super(sb, es, 0))
4346                                 sb->s_flags &= ~MS_RDONLY;
4347                         enable_quota = 1;
4348                 }
4349         }
4350
4351         /*
4352          * Reinitialize lazy itable initialization thread based on
4353          * current settings
4354          */
4355         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4356                 ext4_unregister_li_request(sb);
4357         else {
4358                 ext4_group_t first_not_zeroed;
4359                 first_not_zeroed = ext4_has_uninit_itable(sb);
4360                 ext4_register_li_request(sb, first_not_zeroed);
4361         }
4362
4363         ext4_setup_system_zone(sb);
4364         if (sbi->s_journal == NULL)
4365                 ext4_commit_super(sb, 1);
4366
4367 #ifdef CONFIG_QUOTA
4368         /* Release old quota file names */
4369         for (i = 0; i < MAXQUOTAS; i++)
4370                 if (old_opts.s_qf_names[i] &&
4371                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4372                         kfree(old_opts.s_qf_names[i]);
4373 #endif
4374         unlock_super(sb);
4375         if (enable_quota)
4376                 dquot_resume(sb, -1);
4377
4378         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4379         kfree(orig_data);
4380         return 0;
4381
4382 restore_opts:
4383         sb->s_flags = old_sb_flags;
4384         sbi->s_mount_opt = old_opts.s_mount_opt;
4385         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4386         sbi->s_resuid = old_opts.s_resuid;
4387         sbi->s_resgid = old_opts.s_resgid;
4388         sbi->s_commit_interval = old_opts.s_commit_interval;
4389         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4390         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4391 #ifdef CONFIG_QUOTA
4392         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4393         for (i = 0; i < MAXQUOTAS; i++) {
4394                 if (sbi->s_qf_names[i] &&
4395                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4396                         kfree(sbi->s_qf_names[i]);
4397                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4398         }
4399 #endif
4400         unlock_super(sb);
4401         kfree(orig_data);
4402         return err;
4403 }
4404
4405 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4406 {
4407         struct super_block *sb = dentry->d_sb;
4408         struct ext4_sb_info *sbi = EXT4_SB(sb);
4409         struct ext4_super_block *es = sbi->s_es;
4410         u64 fsid;
4411
4412         if (test_opt(sb, MINIX_DF)) {
4413                 sbi->s_overhead_last = 0;
4414         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4415                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4416                 ext4_fsblk_t overhead = 0;
4417
4418                 /*
4419                  * Compute the overhead (FS structures).  This is constant
4420                  * for a given filesystem unless the number of block groups
4421                  * changes so we cache the previous value until it does.
4422                  */
4423
4424                 /*
4425                  * All of the blocks before first_data_block are
4426                  * overhead
4427                  */
4428                 overhead = le32_to_cpu(es->s_first_data_block);
4429
4430                 /*
4431                  * Add the overhead attributed to the superblock and
4432                  * block group descriptors.  If the sparse superblocks
4433                  * feature is turned on, then not all groups have this.
4434                  */
4435                 for (i = 0; i < ngroups; i++) {
4436                         overhead += ext4_bg_has_super(sb, i) +
4437                                 ext4_bg_num_gdb(sb, i);
4438                         cond_resched();
4439                 }
4440
4441                 /*
4442                  * Every block group has an inode bitmap, a block
4443                  * bitmap, and an inode table.
4444                  */
4445                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4446                 sbi->s_overhead_last = overhead;
4447                 smp_wmb();
4448                 sbi->s_blocks_last = ext4_blocks_count(es);
4449         }
4450
4451         buf->f_type = EXT4_SUPER_MAGIC;
4452         buf->f_bsize = sb->s_blocksize;
4453         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4454         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4455                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4456         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4457         if (buf->f_bfree < ext4_r_blocks_count(es))
4458                 buf->f_bavail = 0;
4459         buf->f_files = le32_to_cpu(es->s_inodes_count);
4460         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4461         buf->f_namelen = EXT4_NAME_LEN;
4462         fsid = le64_to_cpup((void *)es->s_uuid) ^
4463                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4464         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4465         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4466
4467         return 0;
4468 }
4469
4470 /* Helper function for writing quotas on sync - we need to start transaction
4471  * before quota file is locked for write. Otherwise the are possible deadlocks:
4472  * Process 1                         Process 2
4473  * ext4_create()                     quota_sync()
4474  *   jbd2_journal_start()                  write_dquot()
4475  *   dquot_initialize()                         down(dqio_mutex)
4476  *     down(dqio_mutex)                    jbd2_journal_start()
4477  *
4478  */
4479
4480 #ifdef CONFIG_QUOTA
4481
4482 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4483 {
4484         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4485 }
4486
4487 static int ext4_write_dquot(struct dquot *dquot)
4488 {
4489         int ret, err;
4490         handle_t *handle;
4491         struct inode *inode;
4492
4493         inode = dquot_to_inode(dquot);
4494         handle = ext4_journal_start(inode,
4495                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4496         if (IS_ERR(handle))
4497                 return PTR_ERR(handle);
4498         ret = dquot_commit(dquot);
4499         err = ext4_journal_stop(handle);
4500         if (!ret)
4501                 ret = err;
4502         return ret;
4503 }
4504
4505 static int ext4_acquire_dquot(struct dquot *dquot)
4506 {
4507         int ret, err;
4508         handle_t *handle;
4509
4510         handle = ext4_journal_start(dquot_to_inode(dquot),
4511                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4512         if (IS_ERR(handle))
4513                 return PTR_ERR(handle);
4514         ret = dquot_acquire(dquot);
4515         err = ext4_journal_stop(handle);
4516         if (!ret)
4517                 ret = err;
4518         return ret;
4519 }
4520
4521 static int ext4_release_dquot(struct dquot *dquot)
4522 {
4523         int ret, err;
4524         handle_t *handle;
4525
4526         handle = ext4_journal_start(dquot_to_inode(dquot),
4527                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4528         if (IS_ERR(handle)) {
4529                 /* Release dquot anyway to avoid endless cycle in dqput() */
4530                 dquot_release(dquot);
4531                 return PTR_ERR(handle);
4532         }
4533         ret = dquot_release(dquot);
4534         err = ext4_journal_stop(handle);
4535         if (!ret)
4536                 ret = err;
4537         return ret;
4538 }
4539
4540 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4541 {
4542         /* Are we journaling quotas? */
4543         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4544             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4545                 dquot_mark_dquot_dirty(dquot);
4546                 return ext4_write_dquot(dquot);
4547         } else {
4548                 return dquot_mark_dquot_dirty(dquot);
4549         }
4550 }
4551
4552 static int ext4_write_info(struct super_block *sb, int type)
4553 {
4554         int ret, err;
4555         handle_t *handle;
4556
4557         /* Data block + inode block */
4558         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4559         if (IS_ERR(handle))
4560                 return PTR_ERR(handle);
4561         ret = dquot_commit_info(sb, type);
4562         err = ext4_journal_stop(handle);
4563         if (!ret)
4564                 ret = err;
4565         return ret;
4566 }
4567
4568 /*
4569  * Turn on quotas during mount time - we need to find
4570  * the quota file and such...
4571  */
4572 static int ext4_quota_on_mount(struct super_block *sb, int type)
4573 {
4574         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4575                                         EXT4_SB(sb)->s_jquota_fmt, type);
4576 }
4577
4578 /*
4579  * Standard function to be called on quota_on
4580  */
4581 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4582                          struct path *path)
4583 {
4584         int err;
4585
4586         if (!test_opt(sb, QUOTA))
4587                 return -EINVAL;
4588
4589         /* Quotafile not on the same filesystem? */
4590         if (path->mnt->mnt_sb != sb)
4591                 return -EXDEV;
4592         /* Journaling quota? */
4593         if (EXT4_SB(sb)->s_qf_names[type]) {
4594                 /* Quotafile not in fs root? */
4595                 if (path->dentry->d_parent != sb->s_root)
4596                         ext4_msg(sb, KERN_WARNING,
4597                                 "Quota file not on filesystem root. "
4598                                 "Journaled quota will not work");
4599         }
4600
4601         /*
4602          * When we journal data on quota file, we have to flush journal to see
4603          * all updates to the file when we bypass pagecache...
4604          */
4605         if (EXT4_SB(sb)->s_journal &&
4606             ext4_should_journal_data(path->dentry->d_inode)) {
4607                 /*
4608                  * We don't need to lock updates but journal_flush() could
4609                  * otherwise be livelocked...
4610                  */
4611                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4612                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4613                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4614                 if (err)
4615                         return err;
4616         }
4617
4618         return dquot_quota_on(sb, type, format_id, path);
4619 }
4620
4621 static int ext4_quota_off(struct super_block *sb, int type)
4622 {
4623         struct inode *inode = sb_dqopt(sb)->files[type];
4624         handle_t *handle;
4625
4626         /* Force all delayed allocation blocks to be allocated.
4627          * Caller already holds s_umount sem */
4628         if (test_opt(sb, DELALLOC))
4629                 sync_filesystem(sb);
4630
4631         /* Update modification times of quota files when userspace can
4632          * start looking at them */
4633         handle = ext4_journal_start(inode, 1);
4634         if (IS_ERR(handle))
4635                 goto out;
4636         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4637         ext4_mark_inode_dirty(handle, inode);
4638         ext4_journal_stop(handle);
4639
4640 out:
4641         return dquot_quota_off(sb, type);
4642 }
4643
4644 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4645  * acquiring the locks... As quota files are never truncated and quota code
4646  * itself serializes the operations (and noone else should touch the files)
4647  * we don't have to be afraid of races */
4648 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4649                                size_t len, loff_t off)
4650 {
4651         struct inode *inode = sb_dqopt(sb)->files[type];
4652         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4653         int err = 0;
4654         int offset = off & (sb->s_blocksize - 1);
4655         int tocopy;
4656         size_t toread;
4657         struct buffer_head *bh;
4658         loff_t i_size = i_size_read(inode);
4659
4660         if (off > i_size)
4661                 return 0;
4662         if (off+len > i_size)
4663                 len = i_size-off;
4664         toread = len;
4665         while (toread > 0) {
4666                 tocopy = sb->s_blocksize - offset < toread ?
4667                                 sb->s_blocksize - offset : toread;
4668                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4669                 if (err)
4670                         return err;
4671                 if (!bh)        /* A hole? */
4672                         memset(data, 0, tocopy);
4673                 else
4674                         memcpy(data, bh->b_data+offset, tocopy);
4675                 brelse(bh);
4676                 offset = 0;
4677                 toread -= tocopy;
4678                 data += tocopy;
4679                 blk++;
4680         }
4681         return len;
4682 }
4683
4684 /* Write to quotafile (we know the transaction is already started and has
4685  * enough credits) */
4686 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4687                                 const char *data, size_t len, loff_t off)
4688 {
4689         struct inode *inode = sb_dqopt(sb)->files[type];
4690         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4691         int err = 0;
4692         int offset = off & (sb->s_blocksize - 1);
4693         struct buffer_head *bh;
4694         handle_t *handle = journal_current_handle();
4695
4696         if (EXT4_SB(sb)->s_journal && !handle) {
4697                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4698                         " cancelled because transaction is not started",
4699                         (unsigned long long)off, (unsigned long long)len);
4700                 return -EIO;
4701         }
4702         /*
4703          * Since we account only one data block in transaction credits,
4704          * then it is impossible to cross a block boundary.
4705          */
4706         if (sb->s_blocksize - offset < len) {
4707                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4708                         " cancelled because not block aligned",
4709                         (unsigned long long)off, (unsigned long long)len);
4710                 return -EIO;
4711         }
4712
4713         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4714         bh = ext4_bread(handle, inode, blk, 1, &err);
4715         if (!bh)
4716                 goto out;
4717         err = ext4_journal_get_write_access(handle, bh);
4718         if (err) {
4719                 brelse(bh);
4720                 goto out;
4721         }
4722         lock_buffer(bh);
4723         memcpy(bh->b_data+offset, data, len);
4724         flush_dcache_page(bh->b_page);
4725         unlock_buffer(bh);
4726         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4727         brelse(bh);
4728 out:
4729         if (err) {
4730                 mutex_unlock(&inode->i_mutex);
4731                 return err;
4732         }
4733         if (inode->i_size < off + len) {
4734                 i_size_write(inode, off + len);
4735                 EXT4_I(inode)->i_disksize = inode->i_size;
4736                 ext4_mark_inode_dirty(handle, inode);
4737         }
4738         mutex_unlock(&inode->i_mutex);
4739         return len;
4740 }
4741
4742 #endif
4743
4744 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4745                        const char *dev_name, void *data)
4746 {
4747         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4748 }
4749
4750 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4751 static struct file_system_type ext2_fs_type = {
4752         .owner          = THIS_MODULE,
4753         .name           = "ext2",
4754         .mount          = ext4_mount,
4755         .kill_sb        = kill_block_super,
4756         .fs_flags       = FS_REQUIRES_DEV,
4757 };
4758
4759 static inline void register_as_ext2(void)
4760 {
4761         int err = register_filesystem(&ext2_fs_type);
4762         if (err)
4763                 printk(KERN_WARNING
4764                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4765 }
4766
4767 static inline void unregister_as_ext2(void)
4768 {
4769         unregister_filesystem(&ext2_fs_type);
4770 }
4771 MODULE_ALIAS("ext2");
4772 #else
4773 static inline void register_as_ext2(void) { }
4774 static inline void unregister_as_ext2(void) { }
4775 #endif
4776
4777 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4778 static inline void register_as_ext3(void)
4779 {
4780         int err = register_filesystem(&ext3_fs_type);
4781         if (err)
4782                 printk(KERN_WARNING
4783                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4784 }
4785
4786 static inline void unregister_as_ext3(void)
4787 {
4788         unregister_filesystem(&ext3_fs_type);
4789 }
4790 MODULE_ALIAS("ext3");
4791 #else
4792 static inline void register_as_ext3(void) { }
4793 static inline void unregister_as_ext3(void) { }
4794 #endif
4795
4796 static struct file_system_type ext4_fs_type = {
4797         .owner          = THIS_MODULE,
4798         .name           = "ext4",
4799         .mount          = ext4_mount,
4800         .kill_sb        = kill_block_super,
4801         .fs_flags       = FS_REQUIRES_DEV,
4802 };
4803
4804 static int __init ext4_init_feat_adverts(void)
4805 {
4806         struct ext4_features *ef;
4807         int ret = -ENOMEM;
4808
4809         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4810         if (!ef)
4811                 goto out;
4812
4813         ef->f_kobj.kset = ext4_kset;
4814         init_completion(&ef->f_kobj_unregister);
4815         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4816                                    "features");
4817         if (ret) {
4818                 kfree(ef);
4819                 goto out;
4820         }
4821
4822         ext4_feat = ef;
4823         ret = 0;
4824 out:
4825         return ret;
4826 }
4827
4828 static void ext4_exit_feat_adverts(void)
4829 {
4830         kobject_put(&ext4_feat->f_kobj);
4831         wait_for_completion(&ext4_feat->f_kobj_unregister);
4832         kfree(ext4_feat);
4833 }
4834
4835 /* Shared across all ext4 file systems */
4836 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4837 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4838
4839 static int __init ext4_init_fs(void)
4840 {
4841         int i, err;
4842
4843         ext4_check_flag_values();
4844
4845         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4846                 mutex_init(&ext4__aio_mutex[i]);
4847                 init_waitqueue_head(&ext4__ioend_wq[i]);
4848         }
4849
4850         err = ext4_init_pageio();
4851         if (err)
4852                 return err;
4853         err = ext4_init_system_zone();
4854         if (err)
4855                 goto out7;
4856         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4857         if (!ext4_kset)
4858                 goto out6;
4859         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4860         if (!ext4_proc_root)
4861                 goto out5;
4862
4863         err = ext4_init_feat_adverts();
4864         if (err)
4865                 goto out4;
4866
4867         err = ext4_init_mballoc();
4868         if (err)
4869                 goto out3;
4870
4871         err = ext4_init_xattr();
4872         if (err)
4873                 goto out2;
4874         err = init_inodecache();
4875         if (err)
4876                 goto out1;
4877         register_as_ext2();
4878         register_as_ext3();
4879         err = register_filesystem(&ext4_fs_type);
4880         if (err)
4881                 goto out;
4882
4883         ext4_li_info = NULL;
4884         mutex_init(&ext4_li_mtx);
4885         return 0;
4886 out:
4887         unregister_as_ext2();
4888         unregister_as_ext3();
4889         destroy_inodecache();
4890 out1:
4891         ext4_exit_xattr();
4892 out2:
4893         ext4_exit_mballoc();
4894 out3:
4895         ext4_exit_feat_adverts();
4896 out4:
4897         remove_proc_entry("fs/ext4", NULL);
4898 out5:
4899         kset_unregister(ext4_kset);
4900 out6:
4901         ext4_exit_system_zone();
4902 out7:
4903         ext4_exit_pageio();
4904         return err;
4905 }
4906
4907 static void __exit ext4_exit_fs(void)
4908 {
4909         ext4_destroy_lazyinit_thread();
4910         unregister_as_ext2();
4911         unregister_as_ext3();
4912         unregister_filesystem(&ext4_fs_type);
4913         destroy_inodecache();
4914         ext4_exit_xattr();
4915         ext4_exit_mballoc();
4916         ext4_exit_feat_adverts();
4917         remove_proc_entry("fs/ext4", NULL);
4918         kset_unregister(ext4_kset);
4919         ext4_exit_system_zone();
4920         ext4_exit_pageio();
4921 }
4922
4923 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4924 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4925 MODULE_LICENSE("GPL");
4926 module_init(ext4_init_fs)
4927 module_exit(ext4_exit_fs)