fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct proc_dir_entry *ext4_proc_root;
  57 static struct kset *ext4_kset;
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ext4_features *ext4_feat;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static const char *ext4_decode_error(struct super_block *sb, int errno,
  71                                      char nbuf[16]);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static void ext4_write_super(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  95 #else
  96 #define IS_EXT2_SB(sb) (0)
  97 #endif
  98
  99
 100 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 101 static struct file_system_type ext3_fs_type = {
 102         .owner          = THIS_MODULE,
 103         .name           = "ext3",
 104         .mount          = ext4_mount,
 105         .kill_sb        = kill_block_super,
 106         .fs_flags       = FS_REQUIRES_DEV,
 107 };
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109 #else
 110 #define IS_EXT3_SB(sb) (0)
 111 #endif
 112
 113 void *ext4_kvmalloc(size_t size, gfp_t flags)
 114 {
 115         void *ret;
 116
 117         ret = kmalloc(size, flags);
 118         if (!ret)
 119                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 120         return ret;
 121 }
 122
 123 void *ext4_kvzalloc(size_t size, gfp_t flags)
 124 {
 125         void *ret;
 126
 127         ret = kzalloc(size, flags);
 128         if (!ret)
 129                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 130         return ret;
 131 }
 132
 133 void ext4_kvfree(void *ptr)
 134 {
 135         if (is_vmalloc_addr(ptr))
 136                 vfree(ptr);
 137         else
 138                 kfree(ptr);
 139
 140 }
 141
 142 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 143                                struct ext4_group_desc *bg)
 144 {
 145         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 148 }
 149
 150 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 151                                struct ext4_group_desc *bg)
 152 {
 153         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 154                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 155                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 156 }
 157
 158 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 159                               struct ext4_group_desc *bg)
 160 {
 161         return le32_to_cpu(bg->bg_inode_table_lo) |
 162                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 163                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 164 }
 165
 166 __u32 ext4_free_blks_count(struct super_block *sb,
 167                               struct ext4_group_desc *bg)
 168 {
 169         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 170                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 171                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 172 }
 173
 174 __u32 ext4_free_inodes_count(struct super_block *sb,
 175                               struct ext4_group_desc *bg)
 176 {
 177         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 178                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 179                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 180 }
 181
 182 __u32 ext4_used_dirs_count(struct super_block *sb,
 183                               struct ext4_group_desc *bg)
 184 {
 185         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 186                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 187                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 188 }
 189
 190 __u32 ext4_itable_unused_count(struct super_block *sb,
 191                               struct ext4_group_desc *bg)
 192 {
 193         return le16_to_cpu(bg->bg_itable_unused_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 196 }
 197
 198 void ext4_block_bitmap_set(struct super_block *sb,
 199                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 200 {
 201         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 204 }
 205
 206 void ext4_inode_bitmap_set(struct super_block *sb,
 207                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 208 {
 209         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 210         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 211                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 212 }
 213
 214 void ext4_inode_table_set(struct super_block *sb,
 215                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 216 {
 217         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 218         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 219                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 220 }
 221
 222 void ext4_free_blks_set(struct super_block *sb,
 223                           struct ext4_group_desc *bg, __u32 count)
 224 {
 225         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 226         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 227                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 228 }
 229
 230 void ext4_free_inodes_set(struct super_block *sb,
 231                           struct ext4_group_desc *bg, __u32 count)
 232 {
 233         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 234         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 235                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 236 }
 237
 238 void ext4_used_dirs_set(struct super_block *sb,
 239                           struct ext4_group_desc *bg, __u32 count)
 240 {
 241         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 242         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 243                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 244 }
 245
 246 void ext4_itable_unused_set(struct super_block *sb,
 247                           struct ext4_group_desc *bg, __u32 count)
 248 {
 249         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 252 }
 253
 254
 255 /* Just increment the non-pointer handle value */
 256 static handle_t *ext4_get_nojournal(void)
 257 {
 258         handle_t *handle = current->journal_info;
 259         unsigned long ref_cnt = (unsigned long)handle;
 260
 261         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 262
 263         ref_cnt++;
 264         handle = (handle_t *)ref_cnt;
 265
 266         current->journal_info = handle;
 267         return handle;
 268 }
 269
 270
 271 /* Decrement the non-pointer handle value */
 272 static void ext4_put_nojournal(handle_t *handle)
 273 {
 274         unsigned long ref_cnt = (unsigned long)handle;
 275
 276         BUG_ON(ref_cnt == 0);
 277
 278         ref_cnt--;
 279         handle = (handle_t *)ref_cnt;
 280
 281         current->journal_info = handle;
 282 }
 283
 284 /*
 285  * Wrappers for jbd2_journal_start/end.
 286  *
 287  * The only special thing we need to do here is to make sure that all
 288  * journal_end calls result in the superblock being marked dirty, so
 289  * that sync() will call the filesystem's write_super callback if
 290  * appropriate.
 291  *
 292  * To avoid j_barrier hold in userspace when a user calls freeze(),
 293  * ext4 prevents a new handle from being started by s_frozen, which
 294  * is in an upper layer.
 295  */
 296 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 297 {
 298         journal_t *journal;
 299         handle_t  *handle;
 300
 301         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 302         if (sb->s_flags & MS_RDONLY)
 303                 return ERR_PTR(-EROFS);
 304
 305         journal = EXT4_SB(sb)->s_journal;
 306         handle = ext4_journal_current_handle();
 307
 308         /*
 309          * If a handle has been started, it should be allowed to
 310          * finish, otherwise deadlock could happen between freeze
 311          * and others(e.g. truncate) due to the restart of the
 312          * journal handle if the filesystem is forzen and active
 313          * handles are not stopped.
 314          */
 315         if (!handle)
 316                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 317
 318         if (!journal)
 319                 return ext4_get_nojournal();
 320         /*
 321          * Special case here: if the journal has aborted behind our
 322          * backs (eg. EIO in the commit thread), then we still need to
 323          * take the FS itself readonly cleanly.
 324          */
 325         if (is_journal_aborted(journal)) {
 326                 ext4_abort(sb, "Detected aborted journal");
 327                 return ERR_PTR(-EROFS);
 328         }
 329         return jbd2_journal_start(journal, nblocks);
 330 }
 331
 332 /*
 333  * The only special thing we need to do here is to make sure that all
 334  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 335  * that sync() will call the filesystem's write_super callback if
 336  * appropriate.
 337  */
 338 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 339 {
 340         struct super_block *sb;
 341         int err;
 342         int rc;
 343
 344         if (!ext4_handle_valid(handle)) {
 345                 ext4_put_nojournal(handle);
 346                 return 0;
 347         }
 348         sb = handle->h_transaction->t_journal->j_private;
 349         err = handle->h_err;
 350         rc = jbd2_journal_stop(handle);
 351
 352         if (!err)
 353                 err = rc;
 354         if (err)
 355                 __ext4_std_error(sb, where, line, err);
 356         return err;
 357 }
 358
 359 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 360                                const char *err_fn, struct buffer_head *bh,
 361                                handle_t *handle, int err)
 362 {
 363         char nbuf[16];
 364         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 365
 366         BUG_ON(!ext4_handle_valid(handle));
 367
 368         if (bh)
 369                 BUFFER_TRACE(bh, "abort");
 370
 371         if (!handle->h_err)
 372                 handle->h_err = err;
 373
 374         if (is_handle_aborted(handle))
 375                 return;
 376
 377         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 378                caller, line, errstr, err_fn);
 379
 380         jbd2_journal_abort_handle(handle);
 381 }
 382
 383 static void __save_error_info(struct super_block *sb, const char *func,
 384                             unsigned int line)
 385 {
 386         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 387
 388         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 389         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 390         es->s_last_error_time = cpu_to_le32(get_seconds());
 391         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 392         es->s_last_error_line = cpu_to_le32(line);
 393         if (!es->s_first_error_time) {
 394                 es->s_first_error_time = es->s_last_error_time;
 395                 strncpy(es->s_first_error_func, func,
 396                         sizeof(es->s_first_error_func));
 397                 es->s_first_error_line = cpu_to_le32(line);
 398                 es->s_first_error_ino = es->s_last_error_ino;
 399                 es->s_first_error_block = es->s_last_error_block;
 400         }
 401         /*
 402          * Start the daily error reporting function if it hasn't been
 403          * started already
 404          */
 405         if (!es->s_error_count)
 406                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 407         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 408 }
 409
 410 static void save_error_info(struct super_block *sb, const char *func,
 411                             unsigned int line)
 412 {
 413         __save_error_info(sb, func, line);
 414         ext4_commit_super(sb, 1);
 415 }
 416
 417
 418 /* Deal with the reporting of failure conditions on a filesystem such as
 419  * inconsistencies detected or read IO failures.
 420  *
 421  * On ext2, we can store the error state of the filesystem in the
 422  * superblock.  That is not possible on ext4, because we may have other
 423  * write ordering constraints on the superblock which prevent us from
 424  * writing it out straight away; and given that the journal is about to
 425  * be aborted, we can't rely on the current, or future, transactions to
 426  * write out the superblock safely.
 427  *
 428  * We'll just use the jbd2_journal_abort() error code to record an error in
 429  * the journal instead.  On recovery, the journal will complain about
 430  * that error until we've noted it down and cleared it.
 431  */
 432
 433 static void ext4_handle_error(struct super_block *sb)
 434 {
 435         if (sb->s_flags & MS_RDONLY)
 436                 return;
 437
 438         if (!test_opt(sb, ERRORS_CONT)) {
 439                 journal_t *journal = EXT4_SB(sb)->s_journal;
 440
 441                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 442                 if (journal)
 443                         jbd2_journal_abort(journal, -EIO);
 444         }
 445         if (test_opt(sb, ERRORS_RO)) {
 446                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 447                 sb->s_flags |= MS_RDONLY;
 448         }
 449         if (test_opt(sb, ERRORS_PANIC))
 450                 panic("EXT4-fs (device %s): panic forced after error\n",
 451                         sb->s_id);
 452 }
 453
 454 void __ext4_error(struct super_block *sb, const char *function,
 455                   unsigned int line, const char *fmt, ...)
 456 {
 457         struct va_format vaf;
 458         va_list args;
 459
 460         va_start(args, fmt);
 461         vaf.fmt = fmt;
 462         vaf.va = &args;
 463         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 464                sb->s_id, function, line, current->comm, &vaf);
 465         va_end(args);
 466
 467         ext4_handle_error(sb);
 468 }
 469
 470 void ext4_error_inode(struct inode *inode, const char *function,
 471                       unsigned int line, ext4_fsblk_t block,
 472                       const char *fmt, ...)
 473 {
 474         va_list args;
 475         struct va_format vaf;
 476         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 477
 478         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 479         es->s_last_error_block = cpu_to_le64(block);
 480         save_error_info(inode->i_sb, function, line);
 481         va_start(args, fmt);
 482         vaf.fmt = fmt;
 483         vaf.va = &args;
 484         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 485                inode->i_sb->s_id, function, line, inode->i_ino);
 486         if (block)
 487                 printk(KERN_CONT "block %llu: ", block);
 488         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 489         va_end(args);
 490
 491         ext4_handle_error(inode->i_sb);
 492 }
 493
 494 void ext4_error_file(struct file *file, const char *function,
 495                      unsigned int line, ext4_fsblk_t block,
 496                      const char *fmt, ...)
 497 {
 498         va_list args;
 499         struct va_format vaf;
 500         struct ext4_super_block *es;
 501         struct inode *inode = file->f_dentry->d_inode;
 502         char pathname[80], *path;
 503
 504         es = EXT4_SB(inode->i_sb)->s_es;
 505         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 506         save_error_info(inode->i_sb, function, line);
 507         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 508         if (IS_ERR(path))
 509                 path = "(unknown)";
 510         printk(KERN_CRIT
 511                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 512                inode->i_sb->s_id, function, line, inode->i_ino);
 513         if (block)
 514                 printk(KERN_CONT "block %llu: ", block);
 515         va_start(args, fmt);
 516         vaf.fmt = fmt;
 517         vaf.va = &args;
 518         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 519         va_end(args);
 520
 521         ext4_handle_error(inode->i_sb);
 522 }
 523
 524 static const char *ext4_decode_error(struct super_block *sb, int errno,
 525                                      char nbuf[16])
 526 {
 527         char *errstr = NULL;
 528
 529         switch (errno) {
 530         case -EIO:
 531                 errstr = "IO failure";
 532                 break;
 533         case -ENOMEM:
 534                 errstr = "Out of memory";
 535                 break;
 536         case -EROFS:
 537                 if (!sb || (EXT4_SB(sb)->s_journal &&
 538                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 539                         errstr = "Journal has aborted";
 540                 else
 541                         errstr = "Readonly filesystem";
 542                 break;
 543         default:
 544                 /* If the caller passed in an extra buffer for unknown
 545                  * errors, textualise them now.  Else we just return
 546                  * NULL. */
 547                 if (nbuf) {
 548                         /* Check for truncated error codes... */
 549                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 550                                 errstr = nbuf;
 551                 }
 552                 break;
 553         }
 554
 555         return errstr;
 556 }
 557
 558 /* __ext4_std_error decodes expected errors from journaling functions
 559  * automatically and invokes the appropriate error response.  */
 560
 561 void __ext4_std_error(struct super_block *sb, const char *function,
 562                       unsigned int line, int errno)
 563 {
 564         char nbuf[16];
 565         const char *errstr;
 566
 567         /* Special case: if the error is EROFS, and we're not already
 568          * inside a transaction, then there's really no point in logging
 569          * an error. */
 570         if (errno == -EROFS && journal_current_handle() == NULL &&
 571             (sb->s_flags & MS_RDONLY))
 572                 return;
 573
 574         errstr = ext4_decode_error(sb, errno, nbuf);
 575         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 576                sb->s_id, function, line, errstr);
 577         save_error_info(sb, function, line);
 578
 579         ext4_handle_error(sb);
 580 }
 581
 582 /*
 583  * ext4_abort is a much stronger failure handler than ext4_error.  The
 584  * abort function may be used to deal with unrecoverable failures such
 585  * as journal IO errors or ENOMEM at a critical moment in log management.
 586  *
 587  * We unconditionally force the filesystem into an ABORT|READONLY state,
 588  * unless the error response on the fs has been set to panic in which
 589  * case we take the easy way out and panic immediately.
 590  */
 591
 592 void __ext4_abort(struct super_block *sb, const char *function,
 593                 unsigned int line, const char *fmt, ...)
 594 {
 595         va_list args;
 596
 597         save_error_info(sb, function, line);
 598         va_start(args, fmt);
 599         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 600                function, line);
 601         vprintk(fmt, args);
 602         printk("\n");
 603         va_end(args);
 604
 605         if ((sb->s_flags & MS_RDONLY) == 0) {
 606                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 607                 sb->s_flags |= MS_RDONLY;
 608                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 609                 if (EXT4_SB(sb)->s_journal)
 610                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 611                 save_error_info(sb, function, line);
 612         }
 613         if (test_opt(sb, ERRORS_PANIC))
 614                 panic("EXT4-fs panic from previous error\n");
 615 }
 616
 617 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 618 {
 619         struct va_format vaf;
 620         va_list args;
 621
 622         va_start(args, fmt);
 623         vaf.fmt = fmt;
 624         vaf.va = &args;
 625         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 626         va_end(args);
 627 }
 628
 629 void __ext4_warning(struct super_block *sb, const char *function,
 630                     unsigned int line, const char *fmt, ...)
 631 {
 632         struct va_format vaf;
 633         va_list args;
 634
 635         va_start(args, fmt);
 636         vaf.fmt = fmt;
 637         vaf.va = &args;
 638         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 639                sb->s_id, function, line, &vaf);
 640         va_end(args);
 641 }
 642
 643 void __ext4_grp_locked_error(const char *function, unsigned int line,
 644                              struct super_block *sb, ext4_group_t grp,
 645                              unsigned long ino, ext4_fsblk_t block,
 646                              const char *fmt, ...)
 647 __releases(bitlock)
 648 __acquires(bitlock)
 649 {
 650         struct va_format vaf;
 651         va_list args;
 652         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 653
 654         es->s_last_error_ino = cpu_to_le32(ino);
 655         es->s_last_error_block = cpu_to_le64(block);
 656         __save_error_info(sb, function, line);
 657
 658         va_start(args, fmt);
 659
 660         vaf.fmt = fmt;
 661         vaf.va = &args;
 662         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 663                sb->s_id, function, line, grp);
 664         if (ino)
 665                 printk(KERN_CONT "inode %lu: ", ino);
 666         if (block)
 667                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 668         printk(KERN_CONT "%pV\n", &vaf);
 669         va_end(args);
 670
 671         if (test_opt(sb, ERRORS_CONT)) {
 672                 ext4_commit_super(sb, 0);
 673                 return;
 674         }
 675
 676         ext4_unlock_group(sb, grp);
 677         ext4_handle_error(sb);
 678         /*
 679          * We only get here in the ERRORS_RO case; relocking the group
 680          * may be dangerous, but nothing bad will happen since the
 681          * filesystem will have already been marked read/only and the
 682          * journal has been aborted.  We return 1 as a hint to callers
 683          * who might what to use the return value from
 684          * ext4_grp_locked_error() to distinguish between the
 685          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 686          * aggressively from the ext4 function in question, with a
 687          * more appropriate error code.
 688          */
 689         ext4_lock_group(sb, grp);
 690         return;
 691 }
 692
 693 void ext4_update_dynamic_rev(struct super_block *sb)
 694 {
 695         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 696
 697         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 698                 return;
 699
 700         ext4_warning(sb,
 701                      "updating to rev %d because of new feature flag, "
 702                      "running e2fsck is recommended",
 703                      EXT4_DYNAMIC_REV);
 704
 705         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 706         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 707         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 708         /* leave es->s_feature_*compat flags alone */
 709         /* es->s_uuid will be set by e2fsck if empty */
 710
 711         /*
 712          * The rest of the superblock fields should be zero, and if not it
 713          * means they are likely already in use, so leave them alone.  We
 714          * can leave it up to e2fsck to clean up any inconsistencies there.
 715          */
 716 }
 717
 718 /*
 719  * Open the external journal device
 720  */
 721 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 722 {
 723         struct block_device *bdev;
 724         char b[BDEVNAME_SIZE];
 725
 726         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 727         if (IS_ERR(bdev))
 728                 goto fail;
 729         return bdev;
 730
 731 fail:
 732         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 733                         __bdevname(dev, b), PTR_ERR(bdev));
 734         return NULL;
 735 }
 736
 737 /*
 738  * Release the journal device
 739  */
 740 static int ext4_blkdev_put(struct block_device *bdev)
 741 {
 742         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 743 }
 744
 745 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 746 {
 747         struct block_device *bdev;
 748         int ret = -ENODEV;
 749
 750         bdev = sbi->journal_bdev;
 751         if (bdev) {
 752                 ret = ext4_blkdev_put(bdev);
 753                 sbi->journal_bdev = NULL;
 754         }
 755         return ret;
 756 }
 757
 758 static inline struct inode *orphan_list_entry(struct list_head *l)
 759 {
 760         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 761 }
 762
 763 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 764 {
 765         struct list_head *l;
 766
 767         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 768                  le32_to_cpu(sbi->s_es->s_last_orphan));
 769
 770         printk(KERN_ERR "sb_info orphan list:\n");
 771         list_for_each(l, &sbi->s_orphan) {
 772                 struct inode *inode = orphan_list_entry(l);
 773                 printk(KERN_ERR "  "
 774                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 775                        inode->i_sb->s_id, inode->i_ino, inode,
 776                        inode->i_mode, inode->i_nlink,
 777                        NEXT_ORPHAN(inode));
 778         }
 779 }
 780
 781 static void ext4_put_super(struct super_block *sb)
 782 {
 783         struct ext4_sb_info *sbi = EXT4_SB(sb);
 784         struct ext4_super_block *es = sbi->s_es;
 785         int i, err;
 786
 787         ext4_unregister_li_request(sb);
 788         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 789
 790         flush_workqueue(sbi->dio_unwritten_wq);
 791         destroy_workqueue(sbi->dio_unwritten_wq);
 792
 793         lock_super(sb);
 794         if (sb->s_dirt)
 795                 ext4_commit_super(sb, 1);
 796
 797         if (sbi->s_journal) {
 798                 err = jbd2_journal_destroy(sbi->s_journal);
 799                 sbi->s_journal = NULL;
 800                 if (err < 0)
 801                         ext4_abort(sb, "Couldn't clean up the journal");
 802         }
 803
 804         del_timer(&sbi->s_err_report);
 805         ext4_release_system_zone(sb);
 806         ext4_mb_release(sb);
 807         ext4_ext_release(sb);
 808         ext4_xattr_put_super(sb);
 809
 810         if (!(sb->s_flags & MS_RDONLY)) {
 811                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 812                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 813                 ext4_commit_super(sb, 1);
 814         }
 815         if (sbi->s_proc) {
 816                 remove_proc_entry(sb->s_id, ext4_proc_root);
 817         }
 818         kobject_del(&sbi->s_kobj);
 819
 820         for (i = 0; i < sbi->s_gdb_count; i++)
 821                 brelse(sbi->s_group_desc[i]);
 822         ext4_kvfree(sbi->s_group_desc);
 823         ext4_kvfree(sbi->s_flex_groups);
 824         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 825         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 826         percpu_counter_destroy(&sbi->s_dirs_counter);
 827         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 828         brelse(sbi->s_sbh);
 829 #ifdef CONFIG_QUOTA
 830         for (i = 0; i < MAXQUOTAS; i++)
 831                 kfree(sbi->s_qf_names[i]);
 832 #endif
 833
 834         /* Debugging code just in case the in-memory inode orphan list
 835          * isn't empty.  The on-disk one can be non-empty if we've
 836          * detected an error and taken the fs readonly, but the
 837          * in-memory list had better be clean by this point. */
 838         if (!list_empty(&sbi->s_orphan))
 839                 dump_orphan_list(sb, sbi);
 840         J_ASSERT(list_empty(&sbi->s_orphan));
 841
 842         invalidate_bdev(sb->s_bdev);
 843         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 844                 /*
 845                  * Invalidate the journal device's buffers.  We don't want them
 846                  * floating about in memory - the physical journal device may
 847                  * hotswapped, and it breaks the `ro-after' testing code.
 848                  */
 849                 sync_blockdev(sbi->journal_bdev);
 850                 invalidate_bdev(sbi->journal_bdev);
 851                 ext4_blkdev_remove(sbi);
 852         }
 853         if (sbi->s_mmp_tsk)
 854                 kthread_stop(sbi->s_mmp_tsk);
 855         sb->s_fs_info = NULL;
 856         /*
 857          * Now that we are completely done shutting down the
 858          * superblock, we need to actually destroy the kobject.
 859          */
 860         unlock_super(sb);
 861         kobject_put(&sbi->s_kobj);
 862         wait_for_completion(&sbi->s_kobj_unregister);
 863         kfree(sbi->s_blockgroup_lock);
 864         kfree(sbi);
 865 }
 866
 867 static struct kmem_cache *ext4_inode_cachep;
 868
 869 /*
 870  * Called inside transaction, so use GFP_NOFS
 871  */
 872 static struct inode *ext4_alloc_inode(struct super_block *sb)
 873 {
 874         struct ext4_inode_info *ei;
 875
 876         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 877         if (!ei)
 878                 return NULL;
 879
 880         ei->vfs_inode.i_version = 1;
 881         ei->vfs_inode.i_data.writeback_index = 0;
 882         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 883         INIT_LIST_HEAD(&ei->i_prealloc_list);
 884         spin_lock_init(&ei->i_prealloc_lock);
 885         ei->i_reserved_data_blocks = 0;
 886         ei->i_reserved_meta_blocks = 0;
 887         ei->i_allocated_meta_blocks = 0;
 888         ei->i_da_metadata_calc_len = 0;
 889         spin_lock_init(&(ei->i_block_reservation_lock));
 890 #ifdef CONFIG_QUOTA
 891         ei->i_reserved_quota = 0;
 892 #endif
 893         ei->jinode = NULL;
 894         INIT_LIST_HEAD(&ei->i_completed_io_list);
 895         spin_lock_init(&ei->i_completed_io_lock);
 896         ei->cur_aio_dio = NULL;
 897         ei->i_sync_tid = 0;
 898         ei->i_datasync_tid = 0;
 899         atomic_set(&ei->i_ioend_count, 0);
 900         atomic_set(&ei->i_aiodio_unwritten, 0);
 901
 902         return &ei->vfs_inode;
 903 }
 904
 905 static int ext4_drop_inode(struct inode *inode)
 906 {
 907         int drop = generic_drop_inode(inode);
 908
 909         trace_ext4_drop_inode(inode, drop);
 910         return drop;
 911 }
 912
 913 static void ext4_i_callback(struct rcu_head *head)
 914 {
 915         struct inode *inode = container_of(head, struct inode, i_rcu);
 916         INIT_LIST_HEAD(&inode->i_dentry);
 917         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 918 }
 919
 920 static void ext4_destroy_inode(struct inode *inode)
 921 {
 922         ext4_ioend_wait(inode);
 923         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 924                 ext4_msg(inode->i_sb, KERN_ERR,
 925                          "Inode %lu (%p): orphan list check failed!",
 926                          inode->i_ino, EXT4_I(inode));
 927                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 928                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 929                                 true);
 930                 dump_stack();
 931         }
 932         call_rcu(&inode->i_rcu, ext4_i_callback);
 933 }
 934
 935 static void init_once(void *foo)
 936 {
 937         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 938
 939         INIT_LIST_HEAD(&ei->i_orphan);
 940 #ifdef CONFIG_EXT4_FS_XATTR
 941         init_rwsem(&ei->xattr_sem);
 942 #endif
 943         init_rwsem(&ei->i_data_sem);
 944         inode_init_once(&ei->vfs_inode);
 945 }
 946
 947 static int init_inodecache(void)
 948 {
 949         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 950                                              sizeof(struct ext4_inode_info),
 951                                              0, (SLAB_RECLAIM_ACCOUNT|
 952                                                 SLAB_MEM_SPREAD),
 953                                              init_once);
 954         if (ext4_inode_cachep == NULL)
 955                 return -ENOMEM;
 956         return 0;
 957 }
 958
 959 static void destroy_inodecache(void)
 960 {
 961         kmem_cache_destroy(ext4_inode_cachep);
 962 }
 963
 964 void ext4_clear_inode(struct inode *inode)
 965 {
 966         invalidate_inode_buffers(inode);
 967         end_writeback(inode);
 968         dquot_drop(inode);
 969         ext4_discard_preallocations(inode);
 970         if (EXT4_I(inode)->jinode) {
 971                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 972                                                EXT4_I(inode)->jinode);
 973                 jbd2_free_inode(EXT4_I(inode)->jinode);
 974                 EXT4_I(inode)->jinode = NULL;
 975         }
 976 }
 977
 978 static inline void ext4_show_quota_options(struct seq_file *seq,
 979                                            struct super_block *sb)
 980 {
 981 #if defined(CONFIG_QUOTA)
 982         struct ext4_sb_info *sbi = EXT4_SB(sb);
 983
 984         if (sbi->s_jquota_fmt) {
 985                 char *fmtname = "";
 986
 987                 switch (sbi->s_jquota_fmt) {
 988                 case QFMT_VFS_OLD:
 989                         fmtname = "vfsold";
 990                         break;
 991                 case QFMT_VFS_V0:
 992                         fmtname = "vfsv0";
 993                         break;
 994                 case QFMT_VFS_V1:
 995                         fmtname = "vfsv1";
 996                         break;
 997                 }
 998                 seq_printf(seq, ",jqfmt=%s", fmtname);
 999         }
1000
1001         if (sbi->s_qf_names[USRQUOTA])
1002                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1003
1004         if (sbi->s_qf_names[GRPQUOTA])
1005                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1006
1007         if (test_opt(sb, USRQUOTA))
1008                 seq_puts(seq, ",usrquota");
1009
1010         if (test_opt(sb, GRPQUOTA))
1011                 seq_puts(seq, ",grpquota");
1012 #endif
1013 }
1014
1015 /*
1016  * Show an option if
1017  *  - it's set to a non-default value OR
1018  *  - if the per-sb default is different from the global default
1019  */
1020 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1021 {
1022         int def_errors;
1023         unsigned long def_mount_opts;
1024         struct super_block *sb = vfs->mnt_sb;
1025         struct ext4_sb_info *sbi = EXT4_SB(sb);
1026         struct ext4_super_block *es = sbi->s_es;
1027
1028         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1029         def_errors     = le16_to_cpu(es->s_errors);
1030
1031         if (sbi->s_sb_block != 1)
1032                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1033         if (test_opt(sb, MINIX_DF))
1034                 seq_puts(seq, ",minixdf");
1035         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1036                 seq_puts(seq, ",grpid");
1037         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1038                 seq_puts(seq, ",nogrpid");
1039         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1040             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1041                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1042         }
1043         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1044             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1045                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1046         }
1047         if (test_opt(sb, ERRORS_RO)) {
1048                 if (def_errors == EXT4_ERRORS_PANIC ||
1049                     def_errors == EXT4_ERRORS_CONTINUE) {
1050                         seq_puts(seq, ",errors=remount-ro");
1051                 }
1052         }
1053         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1054                 seq_puts(seq, ",errors=continue");
1055         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1056                 seq_puts(seq, ",errors=panic");
1057         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1058                 seq_puts(seq, ",nouid32");
1059         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1060                 seq_puts(seq, ",debug");
1061         if (test_opt(sb, OLDALLOC))
1062                 seq_puts(seq, ",oldalloc");
1063 #ifdef CONFIG_EXT4_FS_XATTR
1064         if (test_opt(sb, XATTR_USER))
1065                 seq_puts(seq, ",user_xattr");
1066         if (!test_opt(sb, XATTR_USER))
1067                 seq_puts(seq, ",nouser_xattr");
1068 #endif
1069 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1070         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1071                 seq_puts(seq, ",acl");
1072         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1073                 seq_puts(seq, ",noacl");
1074 #endif
1075         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1076                 seq_printf(seq, ",commit=%u",
1077                            (unsigned) (sbi->s_commit_interval / HZ));
1078         }
1079         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1080                 seq_printf(seq, ",min_batch_time=%u",
1081                            (unsigned) sbi->s_min_batch_time);
1082         }
1083         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1084                 seq_printf(seq, ",max_batch_time=%u",
1085                            (unsigned) sbi->s_min_batch_time);
1086         }
1087
1088         /*
1089          * We're changing the default of barrier mount option, so
1090          * let's always display its mount state so it's clear what its
1091          * status is.
1092          */
1093         seq_puts(seq, ",barrier=");
1094         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1095         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1096                 seq_puts(seq, ",journal_async_commit");
1097         else if (test_opt(sb, JOURNAL_CHECKSUM))
1098                 seq_puts(seq, ",journal_checksum");
1099         if (test_opt(sb, I_VERSION))
1100                 seq_puts(seq, ",i_version");
1101         if (!test_opt(sb, DELALLOC) &&
1102             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1103                 seq_puts(seq, ",nodelalloc");
1104
1105         if (!test_opt(sb, MBLK_IO_SUBMIT))
1106                 seq_puts(seq, ",nomblk_io_submit");
1107         if (sbi->s_stripe)
1108                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1109         /*
1110          * journal mode get enabled in different ways
1111          * So just print the value even if we didn't specify it
1112          */
1113         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1114                 seq_puts(seq, ",data=journal");
1115         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1116                 seq_puts(seq, ",data=ordered");
1117         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1118                 seq_puts(seq, ",data=writeback");
1119
1120         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1121                 seq_printf(seq, ",inode_readahead_blks=%u",
1122                            sbi->s_inode_readahead_blks);
1123
1124         if (test_opt(sb, DATA_ERR_ABORT))
1125                 seq_puts(seq, ",data_err=abort");
1126
1127         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1128                 seq_puts(seq, ",noauto_da_alloc");
1129
1130         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1131                 seq_puts(seq, ",discard");
1132
1133         if (test_opt(sb, NOLOAD))
1134                 seq_puts(seq, ",norecovery");
1135
1136         if (test_opt(sb, DIOREAD_NOLOCK))
1137                 seq_puts(seq, ",dioread_nolock");
1138
1139         if (test_opt(sb, BLOCK_VALIDITY) &&
1140             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1141                 seq_puts(seq, ",block_validity");
1142
1143         if (!test_opt(sb, INIT_INODE_TABLE))
1144                 seq_puts(seq, ",noinit_inode_table");
1145         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1146                 seq_printf(seq, ",init_inode_table=%u",
1147                            (unsigned) sbi->s_li_wait_mult);
1148
1149         ext4_show_quota_options(seq, sb);
1150
1151         return 0;
1152 }
1153
1154 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1155                                         u64 ino, u32 generation)
1156 {
1157         struct inode *inode;
1158
1159         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1160                 return ERR_PTR(-ESTALE);
1161         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1162                 return ERR_PTR(-ESTALE);
1163
1164         /* iget isn't really right if the inode is currently unallocated!!
1165          *
1166          * ext4_read_inode will return a bad_inode if the inode had been
1167          * deleted, so we should be safe.
1168          *
1169          * Currently we don't know the generation for parent directory, so
1170          * a generation of 0 means "accept any"
1171          */
1172         inode = ext4_iget(sb, ino);
1173         if (IS_ERR(inode))
1174                 return ERR_CAST(inode);
1175         if (generation && inode->i_generation != generation) {
1176                 iput(inode);
1177                 return ERR_PTR(-ESTALE);
1178         }
1179
1180         return inode;
1181 }
1182
1183 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1184                                         int fh_len, int fh_type)
1185 {
1186         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1187                                     ext4_nfs_get_inode);
1188 }
1189
1190 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1191                                         int fh_len, int fh_type)
1192 {
1193         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1194                                     ext4_nfs_get_inode);
1195 }
1196
1197 /*
1198  * Try to release metadata pages (indirect blocks, directories) which are
1199  * mapped via the block device.  Since these pages could have journal heads
1200  * which would prevent try_to_free_buffers() from freeing them, we must use
1201  * jbd2 layer's try_to_free_buffers() function to release them.
1202  */
1203 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1204                                  gfp_t wait)
1205 {
1206         journal_t *journal = EXT4_SB(sb)->s_journal;
1207
1208         WARN_ON(PageChecked(page));
1209         if (!page_has_buffers(page))
1210                 return 0;
1211         if (journal)
1212                 return jbd2_journal_try_to_free_buffers(journal, page,
1213                                                         wait & ~__GFP_WAIT);
1214         return try_to_free_buffers(page);
1215 }
1216
1217 #ifdef CONFIG_QUOTA
1218 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1219 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1220
1221 static int ext4_write_dquot(struct dquot *dquot);
1222 static int ext4_acquire_dquot(struct dquot *dquot);
1223 static int ext4_release_dquot(struct dquot *dquot);
1224 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1225 static int ext4_write_info(struct super_block *sb, int type);
1226 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1227                          struct path *path);
1228 static int ext4_quota_off(struct super_block *sb, int type);
1229 static int ext4_quota_on_mount(struct super_block *sb, int type);
1230 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1231                                size_t len, loff_t off);
1232 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1233                                 const char *data, size_t len, loff_t off);
1234
1235 static const struct dquot_operations ext4_quota_operations = {
1236         .get_reserved_space = ext4_get_reserved_space,
1237         .write_dquot    = ext4_write_dquot,
1238         .acquire_dquot  = ext4_acquire_dquot,
1239         .release_dquot  = ext4_release_dquot,
1240         .mark_dirty     = ext4_mark_dquot_dirty,
1241         .write_info     = ext4_write_info,
1242         .alloc_dquot    = dquot_alloc,
1243         .destroy_dquot  = dquot_destroy,
1244 };
1245
1246 static const struct quotactl_ops ext4_qctl_operations = {
1247         .quota_on       = ext4_quota_on,
1248         .quota_off      = ext4_quota_off,
1249         .quota_sync     = dquot_quota_sync,
1250         .get_info       = dquot_get_dqinfo,
1251         .set_info       = dquot_set_dqinfo,
1252         .get_dqblk      = dquot_get_dqblk,
1253         .set_dqblk      = dquot_set_dqblk
1254 };
1255 #endif
1256
1257 static const struct super_operations ext4_sops = {
1258         .alloc_inode    = ext4_alloc_inode,
1259         .destroy_inode  = ext4_destroy_inode,
1260         .write_inode    = ext4_write_inode,
1261         .dirty_inode    = ext4_dirty_inode,
1262         .drop_inode     = ext4_drop_inode,
1263         .evict_inode    = ext4_evict_inode,
1264         .put_super      = ext4_put_super,
1265         .sync_fs        = ext4_sync_fs,
1266         .freeze_fs      = ext4_freeze,
1267         .unfreeze_fs    = ext4_unfreeze,
1268         .statfs         = ext4_statfs,
1269         .remount_fs     = ext4_remount,
1270         .show_options   = ext4_show_options,
1271 #ifdef CONFIG_QUOTA
1272         .quota_read     = ext4_quota_read,
1273         .quota_write    = ext4_quota_write,
1274 #endif
1275         .bdev_try_to_free_page = bdev_try_to_free_page,
1276 };
1277
1278 static const struct super_operations ext4_nojournal_sops = {
1279         .alloc_inode    = ext4_alloc_inode,
1280         .destroy_inode  = ext4_destroy_inode,
1281         .write_inode    = ext4_write_inode,
1282         .dirty_inode    = ext4_dirty_inode,
1283         .drop_inode     = ext4_drop_inode,
1284         .evict_inode    = ext4_evict_inode,
1285         .write_super    = ext4_write_super,
1286         .put_super      = ext4_put_super,
1287         .statfs         = ext4_statfs,
1288         .remount_fs     = ext4_remount,
1289         .show_options   = ext4_show_options,
1290 #ifdef CONFIG_QUOTA
1291         .quota_read     = ext4_quota_read,
1292         .quota_write    = ext4_quota_write,
1293 #endif
1294         .bdev_try_to_free_page = bdev_try_to_free_page,
1295 };
1296
1297 static const struct export_operations ext4_export_ops = {
1298         .fh_to_dentry = ext4_fh_to_dentry,
1299         .fh_to_parent = ext4_fh_to_parent,
1300         .get_parent = ext4_get_parent,
1301 };
1302
1303 enum {
1304         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1305         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1306         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1307         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1308         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1309         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1310         Opt_journal_update, Opt_journal_dev,
1311         Opt_journal_checksum, Opt_journal_async_commit,
1312         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1313         Opt_data_err_abort, Opt_data_err_ignore,
1314         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1315         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1316         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1317         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1318         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1319         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1320         Opt_inode_readahead_blks, Opt_journal_ioprio,
1321         Opt_dioread_nolock, Opt_dioread_lock,
1322         Opt_discard, Opt_nodiscard,
1323         Opt_init_inode_table, Opt_noinit_inode_table,
1324 };
1325
1326 static const match_table_t tokens = {
1327         {Opt_bsd_df, "bsddf"},
1328         {Opt_minix_df, "minixdf"},
1329         {Opt_grpid, "grpid"},
1330         {Opt_grpid, "bsdgroups"},
1331         {Opt_nogrpid, "nogrpid"},
1332         {Opt_nogrpid, "sysvgroups"},
1333         {Opt_resgid, "resgid=%u"},
1334         {Opt_resuid, "resuid=%u"},
1335         {Opt_sb, "sb=%u"},
1336         {Opt_err_cont, "errors=continue"},
1337         {Opt_err_panic, "errors=panic"},
1338         {Opt_err_ro, "errors=remount-ro"},
1339         {Opt_nouid32, "nouid32"},
1340         {Opt_debug, "debug"},
1341         {Opt_oldalloc, "oldalloc"},
1342         {Opt_orlov, "orlov"},
1343         {Opt_user_xattr, "user_xattr"},
1344         {Opt_nouser_xattr, "nouser_xattr"},
1345         {Opt_acl, "acl"},
1346         {Opt_noacl, "noacl"},
1347         {Opt_noload, "noload"},
1348         {Opt_noload, "norecovery"},
1349         {Opt_nobh, "nobh"},
1350         {Opt_bh, "bh"},
1351         {Opt_commit, "commit=%u"},
1352         {Opt_min_batch_time, "min_batch_time=%u"},
1353         {Opt_max_batch_time, "max_batch_time=%u"},
1354         {Opt_journal_update, "journal=update"},
1355         {Opt_journal_dev, "journal_dev=%u"},
1356         {Opt_journal_checksum, "journal_checksum"},
1357         {Opt_journal_async_commit, "journal_async_commit"},
1358         {Opt_abort, "abort"},
1359         {Opt_data_journal, "data=journal"},
1360         {Opt_data_ordered, "data=ordered"},
1361         {Opt_data_writeback, "data=writeback"},
1362         {Opt_data_err_abort, "data_err=abort"},
1363         {Opt_data_err_ignore, "data_err=ignore"},
1364         {Opt_offusrjquota, "usrjquota="},
1365         {Opt_usrjquota, "usrjquota=%s"},
1366         {Opt_offgrpjquota, "grpjquota="},
1367         {Opt_grpjquota, "grpjquota=%s"},
1368         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1369         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1370         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1371         {Opt_grpquota, "grpquota"},
1372         {Opt_noquota, "noquota"},
1373         {Opt_quota, "quota"},
1374         {Opt_usrquota, "usrquota"},
1375         {Opt_barrier, "barrier=%u"},
1376         {Opt_barrier, "barrier"},
1377         {Opt_nobarrier, "nobarrier"},
1378         {Opt_i_version, "i_version"},
1379         {Opt_stripe, "stripe=%u"},
1380         {Opt_resize, "resize"},
1381         {Opt_delalloc, "delalloc"},
1382         {Opt_nodelalloc, "nodelalloc"},
1383         {Opt_mblk_io_submit, "mblk_io_submit"},
1384         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1385         {Opt_block_validity, "block_validity"},
1386         {Opt_noblock_validity, "noblock_validity"},
1387         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1388         {Opt_journal_ioprio, "journal_ioprio=%u"},
1389         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1390         {Opt_auto_da_alloc, "auto_da_alloc"},
1391         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1392         {Opt_dioread_nolock, "dioread_nolock"},
1393         {Opt_dioread_lock, "dioread_lock"},
1394         {Opt_discard, "discard"},
1395         {Opt_nodiscard, "nodiscard"},
1396         {Opt_init_inode_table, "init_itable=%u"},
1397         {Opt_init_inode_table, "init_itable"},
1398         {Opt_noinit_inode_table, "noinit_itable"},
1399         {Opt_err, NULL},
1400 };
1401
1402 static ext4_fsblk_t get_sb_block(void **data)
1403 {
1404         ext4_fsblk_t    sb_block;
1405         char            *options = (char *) *data;
1406
1407         if (!options || strncmp(options, "sb=", 3) != 0)
1408                 return 1;       /* Default location */
1409
1410         options += 3;
1411         /* TODO: use simple_strtoll with >32bit ext4 */
1412         sb_block = simple_strtoul(options, &options, 0);
1413         if (*options && *options != ',') {
1414                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1415                        (char *) *data);
1416                 return 1;
1417         }
1418         if (*options == ',')
1419                 options++;
1420         *data = (void *) options;
1421
1422         return sb_block;
1423 }
1424
1425 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1426 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1427         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1428
1429 #ifdef CONFIG_QUOTA
1430 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1431 {
1432         struct ext4_sb_info *sbi = EXT4_SB(sb);
1433         char *qname;
1434
1435         if (sb_any_quota_loaded(sb) &&
1436                 !sbi->s_qf_names[qtype]) {
1437                 ext4_msg(sb, KERN_ERR,
1438                         "Cannot change journaled "
1439                         "quota options when quota turned on");
1440                 return 0;
1441         }
1442         qname = match_strdup(args);
1443         if (!qname) {
1444                 ext4_msg(sb, KERN_ERR,
1445                         "Not enough memory for storing quotafile name");
1446                 return 0;
1447         }
1448         if (sbi->s_qf_names[qtype] &&
1449                 strcmp(sbi->s_qf_names[qtype], qname)) {
1450                 ext4_msg(sb, KERN_ERR,
1451                         "%s quota file already specified", QTYPE2NAME(qtype));
1452                 kfree(qname);
1453                 return 0;
1454         }
1455         sbi->s_qf_names[qtype] = qname;
1456         if (strchr(sbi->s_qf_names[qtype], '/')) {
1457                 ext4_msg(sb, KERN_ERR,
1458                         "quotafile must be on filesystem root");
1459                 kfree(sbi->s_qf_names[qtype]);
1460                 sbi->s_qf_names[qtype] = NULL;
1461                 return 0;
1462         }
1463         set_opt(sb, QUOTA);
1464         return 1;
1465 }
1466
1467 static int clear_qf_name(struct super_block *sb, int qtype)
1468 {
1469
1470         struct ext4_sb_info *sbi = EXT4_SB(sb);
1471
1472         if (sb_any_quota_loaded(sb) &&
1473                 sbi->s_qf_names[qtype]) {
1474                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1475                         " when quota turned on");
1476                 return 0;
1477         }
1478         /*
1479          * The space will be released later when all options are confirmed
1480          * to be correct
1481          */
1482         sbi->s_qf_names[qtype] = NULL;
1483         return 1;
1484 }
1485 #endif
1486
1487 static int parse_options(char *options, struct super_block *sb,
1488                          unsigned long *journal_devnum,
1489                          unsigned int *journal_ioprio,
1490                          ext4_fsblk_t *n_blocks_count, int is_remount)
1491 {
1492         struct ext4_sb_info *sbi = EXT4_SB(sb);
1493         char *p;
1494         substring_t args[MAX_OPT_ARGS];
1495         int data_opt = 0;
1496         int option;
1497 #ifdef CONFIG_QUOTA
1498         int qfmt;
1499 #endif
1500
1501         if (!options)
1502                 return 1;
1503
1504         while ((p = strsep(&options, ",")) != NULL) {
1505                 int token;
1506                 if (!*p)
1507                         continue;
1508
1509                 /*
1510                  * Initialize args struct so we know whether arg was
1511                  * found; some options take optional arguments.
1512                  */
1513                 args[0].to = args[0].from = NULL;
1514                 token = match_token(p, tokens, args);
1515                 switch (token) {
1516                 case Opt_bsd_df:
1517                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1518                         clear_opt(sb, MINIX_DF);
1519                         break;
1520                 case Opt_minix_df:
1521                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1522                         set_opt(sb, MINIX_DF);
1523
1524                         break;
1525                 case Opt_grpid:
1526                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1527                         set_opt(sb, GRPID);
1528
1529                         break;
1530                 case Opt_nogrpid:
1531                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1532                         clear_opt(sb, GRPID);
1533
1534                         break;
1535                 case Opt_resuid:
1536                         if (match_int(&args[0], &option))
1537                                 return 0;
1538                         sbi->s_resuid = option;
1539                         break;
1540                 case Opt_resgid:
1541                         if (match_int(&args[0], &option))
1542                                 return 0;
1543                         sbi->s_resgid = option;
1544                         break;
1545                 case Opt_sb:
1546                         /* handled by get_sb_block() instead of here */
1547                         /* *sb_block = match_int(&args[0]); */
1548                         break;
1549                 case Opt_err_panic:
1550                         clear_opt(sb, ERRORS_CONT);
1551                         clear_opt(sb, ERRORS_RO);
1552                         set_opt(sb, ERRORS_PANIC);
1553                         break;
1554                 case Opt_err_ro:
1555                         clear_opt(sb, ERRORS_CONT);
1556                         clear_opt(sb, ERRORS_PANIC);
1557                         set_opt(sb, ERRORS_RO);
1558                         break;
1559                 case Opt_err_cont:
1560                         clear_opt(sb, ERRORS_RO);
1561                         clear_opt(sb, ERRORS_PANIC);
1562                         set_opt(sb, ERRORS_CONT);
1563                         break;
1564                 case Opt_nouid32:
1565                         set_opt(sb, NO_UID32);
1566                         break;
1567                 case Opt_debug:
1568                         set_opt(sb, DEBUG);
1569                         break;
1570                 case Opt_oldalloc:
1571                         set_opt(sb, OLDALLOC);
1572                         break;
1573                 case Opt_orlov:
1574                         clear_opt(sb, OLDALLOC);
1575                         break;
1576 #ifdef CONFIG_EXT4_FS_XATTR
1577                 case Opt_user_xattr:
1578                         set_opt(sb, XATTR_USER);
1579                         break;
1580                 case Opt_nouser_xattr:
1581                         clear_opt(sb, XATTR_USER);
1582                         break;
1583 #else
1584                 case Opt_user_xattr:
1585                 case Opt_nouser_xattr:
1586                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1587                         break;
1588 #endif
1589 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1590                 case Opt_acl:
1591                         set_opt(sb, POSIX_ACL);
1592                         break;
1593                 case Opt_noacl:
1594                         clear_opt(sb, POSIX_ACL);
1595                         break;
1596 #else
1597                 case Opt_acl:
1598                 case Opt_noacl:
1599                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1600                         break;
1601 #endif
1602                 case Opt_journal_update:
1603                         /* @@@ FIXME */
1604                         /* Eventually we will want to be able to create
1605                            a journal file here.  For now, only allow the
1606                            user to specify an existing inode to be the
1607                            journal file. */
1608                         if (is_remount) {
1609                                 ext4_msg(sb, KERN_ERR,
1610                                          "Cannot specify journal on remount");
1611                                 return 0;
1612                         }
1613                         set_opt(sb, UPDATE_JOURNAL);
1614                         break;
1615                 case Opt_journal_dev:
1616                         if (is_remount) {
1617                                 ext4_msg(sb, KERN_ERR,
1618                                         "Cannot specify journal on remount");
1619                                 return 0;
1620                         }
1621                         if (match_int(&args[0], &option))
1622                                 return 0;
1623                         *journal_devnum = option;
1624                         break;
1625                 case Opt_journal_checksum:
1626                         set_opt(sb, JOURNAL_CHECKSUM);
1627                         break;
1628                 case Opt_journal_async_commit:
1629                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1630                         set_opt(sb, JOURNAL_CHECKSUM);
1631                         break;
1632                 case Opt_noload:
1633                         set_opt(sb, NOLOAD);
1634                         break;
1635                 case Opt_commit:
1636                         if (match_int(&args[0], &option))
1637                                 return 0;
1638                         if (option < 0)
1639                                 return 0;
1640                         if (option == 0)
1641                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1642                         sbi->s_commit_interval = HZ * option;
1643                         break;
1644                 case Opt_max_batch_time:
1645                         if (match_int(&args[0], &option))
1646                                 return 0;
1647                         if (option < 0)
1648                                 return 0;
1649                         if (option == 0)
1650                                 option = EXT4_DEF_MAX_BATCH_TIME;
1651                         sbi->s_max_batch_time = option;
1652                         break;
1653                 case Opt_min_batch_time:
1654                         if (match_int(&args[0], &option))
1655                                 return 0;
1656                         if (option < 0)
1657                                 return 0;
1658                         sbi->s_min_batch_time = option;
1659                         break;
1660                 case Opt_data_journal:
1661                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1662                         goto datacheck;
1663                 case Opt_data_ordered:
1664                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1665                         goto datacheck;
1666                 case Opt_data_writeback:
1667                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1668                 datacheck:
1669                         if (is_remount) {
1670                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1671                                         ext4_msg(sb, KERN_ERR,
1672                                                 "Cannot change data mode on remount");
1673                                         return 0;
1674                                 }
1675                         } else {
1676                                 clear_opt(sb, DATA_FLAGS);
1677                                 sbi->s_mount_opt |= data_opt;
1678                         }
1679                         break;
1680                 case Opt_data_err_abort:
1681                         set_opt(sb, DATA_ERR_ABORT);
1682                         break;
1683                 case Opt_data_err_ignore:
1684                         clear_opt(sb, DATA_ERR_ABORT);
1685                         break;
1686 #ifdef CONFIG_QUOTA
1687                 case Opt_usrjquota:
1688                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1689                                 return 0;
1690                         break;
1691                 case Opt_grpjquota:
1692                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1693                                 return 0;
1694                         break;
1695                 case Opt_offusrjquota:
1696                         if (!clear_qf_name(sb, USRQUOTA))
1697                                 return 0;
1698                         break;
1699                 case Opt_offgrpjquota:
1700                         if (!clear_qf_name(sb, GRPQUOTA))
1701                                 return 0;
1702                         break;
1703
1704                 case Opt_jqfmt_vfsold:
1705                         qfmt = QFMT_VFS_OLD;
1706                         goto set_qf_format;
1707                 case Opt_jqfmt_vfsv0:
1708                         qfmt = QFMT_VFS_V0;
1709                         goto set_qf_format;
1710                 case Opt_jqfmt_vfsv1:
1711                         qfmt = QFMT_VFS_V1;
1712 set_qf_format:
1713                         if (sb_any_quota_loaded(sb) &&
1714                             sbi->s_jquota_fmt != qfmt) {
1715                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1716                                         "journaled quota options when "
1717                                         "quota turned on");
1718                                 return 0;
1719                         }
1720                         sbi->s_jquota_fmt = qfmt;
1721                         break;
1722                 case Opt_quota:
1723                 case Opt_usrquota:
1724                         set_opt(sb, QUOTA);
1725                         set_opt(sb, USRQUOTA);
1726                         break;
1727                 case Opt_grpquota:
1728                         set_opt(sb, QUOTA);
1729                         set_opt(sb, GRPQUOTA);
1730                         break;
1731                 case Opt_noquota:
1732                         if (sb_any_quota_loaded(sb)) {
1733                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1734                                         "options when quota turned on");
1735                                 return 0;
1736                         }
1737                         clear_opt(sb, QUOTA);
1738                         clear_opt(sb, USRQUOTA);
1739                         clear_opt(sb, GRPQUOTA);
1740                         break;
1741 #else
1742                 case Opt_quota:
1743                 case Opt_usrquota:
1744                 case Opt_grpquota:
1745                         ext4_msg(sb, KERN_ERR,
1746                                 "quota options not supported");
1747                         break;
1748                 case Opt_usrjquota:
1749                 case Opt_grpjquota:
1750                 case Opt_offusrjquota:
1751                 case Opt_offgrpjquota:
1752                 case Opt_jqfmt_vfsold:
1753                 case Opt_jqfmt_vfsv0:
1754                 case Opt_jqfmt_vfsv1:
1755                         ext4_msg(sb, KERN_ERR,
1756                                 "journaled quota options not supported");
1757                         break;
1758                 case Opt_noquota:
1759                         break;
1760 #endif
1761                 case Opt_abort:
1762                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1763                         break;
1764                 case Opt_nobarrier:
1765                         clear_opt(sb, BARRIER);
1766                         break;
1767                 case Opt_barrier:
1768                         if (args[0].from) {
1769                                 if (match_int(&args[0], &option))
1770                                         return 0;
1771                         } else
1772                                 option = 1;     /* No argument, default to 1 */
1773                         if (option)
1774                                 set_opt(sb, BARRIER);
1775                         else
1776                                 clear_opt(sb, BARRIER);
1777                         break;
1778                 case Opt_ignore:
1779                         break;
1780                 case Opt_resize:
1781                         if (!is_remount) {
1782                                 ext4_msg(sb, KERN_ERR,
1783                                         "resize option only available "
1784                                         "for remount");
1785                                 return 0;
1786                         }
1787                         if (match_int(&args[0], &option) != 0)
1788                                 return 0;
1789                         *n_blocks_count = option;
1790                         break;
1791                 case Opt_nobh:
1792                         ext4_msg(sb, KERN_WARNING,
1793                                  "Ignoring deprecated nobh option");
1794                         break;
1795                 case Opt_bh:
1796                         ext4_msg(sb, KERN_WARNING,
1797                                  "Ignoring deprecated bh option");
1798                         break;
1799                 case Opt_i_version:
1800                         set_opt(sb, I_VERSION);
1801                         sb->s_flags |= MS_I_VERSION;
1802                         break;
1803                 case Opt_nodelalloc:
1804                         clear_opt(sb, DELALLOC);
1805                         break;
1806                 case Opt_mblk_io_submit:
1807                         set_opt(sb, MBLK_IO_SUBMIT);
1808                         break;
1809                 case Opt_nomblk_io_submit:
1810                         clear_opt(sb, MBLK_IO_SUBMIT);
1811                         break;
1812                 case Opt_stripe:
1813                         if (match_int(&args[0], &option))
1814                                 return 0;
1815                         if (option < 0)
1816                                 return 0;
1817                         sbi->s_stripe = option;
1818                         break;
1819                 case Opt_delalloc:
1820                         set_opt(sb, DELALLOC);
1821                         break;
1822                 case Opt_block_validity:
1823                         set_opt(sb, BLOCK_VALIDITY);
1824                         break;
1825                 case Opt_noblock_validity:
1826                         clear_opt(sb, BLOCK_VALIDITY);
1827                         break;
1828                 case Opt_inode_readahead_blks:
1829                         if (match_int(&args[0], &option))
1830                                 return 0;
1831                         if (option < 0 || option > (1 << 30))
1832                                 return 0;
1833                         if (option && !is_power_of_2(option)) {
1834                                 ext4_msg(sb, KERN_ERR,
1835                                          "EXT4-fs: inode_readahead_blks"
1836                                          " must be a power of 2");
1837                                 return 0;
1838                         }
1839                         sbi->s_inode_readahead_blks = option;
1840                         break;
1841                 case Opt_journal_ioprio:
1842                         if (match_int(&args[0], &option))
1843                                 return 0;
1844                         if (option < 0 || option > 7)
1845                                 break;
1846                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1847                                                             option);
1848                         break;
1849                 case Opt_noauto_da_alloc:
1850                         set_opt(sb, NO_AUTO_DA_ALLOC);
1851                         break;
1852                 case Opt_auto_da_alloc:
1853                         if (args[0].from) {
1854                                 if (match_int(&args[0], &option))
1855                                         return 0;
1856                         } else
1857                                 option = 1;     /* No argument, default to 1 */
1858                         if (option)
1859                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1860                         else
1861                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1862                         break;
1863                 case Opt_discard:
1864                         set_opt(sb, DISCARD);
1865                         break;
1866                 case Opt_nodiscard:
1867                         clear_opt(sb, DISCARD);
1868                         break;
1869                 case Opt_dioread_nolock:
1870                         set_opt(sb, DIOREAD_NOLOCK);
1871                         break;
1872                 case Opt_dioread_lock:
1873                         clear_opt(sb, DIOREAD_NOLOCK);
1874                         break;
1875                 case Opt_init_inode_table:
1876                         set_opt(sb, INIT_INODE_TABLE);
1877                         if (args[0].from) {
1878                                 if (match_int(&args[0], &option))
1879                                         return 0;
1880                         } else
1881                                 option = EXT4_DEF_LI_WAIT_MULT;
1882                         if (option < 0)
1883                                 return 0;
1884                         sbi->s_li_wait_mult = option;
1885                         break;
1886                 case Opt_noinit_inode_table:
1887                         clear_opt(sb, INIT_INODE_TABLE);
1888                         break;
1889                 default:
1890                         ext4_msg(sb, KERN_ERR,
1891                                "Unrecognized mount option \"%s\" "
1892                                "or missing value", p);
1893                         return 0;
1894                 }
1895         }
1896 #ifdef CONFIG_QUOTA
1897         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1898                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1899                         clear_opt(sb, USRQUOTA);
1900
1901                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1902                         clear_opt(sb, GRPQUOTA);
1903
1904                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1905                         ext4_msg(sb, KERN_ERR, "old and new quota "
1906                                         "format mixing");
1907                         return 0;
1908                 }
1909
1910                 if (!sbi->s_jquota_fmt) {
1911                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1912                                         "not specified");
1913                         return 0;
1914                 }
1915         } else {
1916                 if (sbi->s_jquota_fmt) {
1917                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1918                                         "specified with no journaling "
1919                                         "enabled");
1920                         return 0;
1921                 }
1922         }
1923 #endif
1924         return 1;
1925 }
1926
1927 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1928                             int read_only)
1929 {
1930         struct ext4_sb_info *sbi = EXT4_SB(sb);
1931         int res = 0;
1932
1933         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1934                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1935                          "forcing read-only mode");
1936                 res = MS_RDONLY;
1937         }
1938         if (read_only)
1939                 return res;
1940         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1941                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1942                          "running e2fsck is recommended");
1943         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1944                 ext4_msg(sb, KERN_WARNING,
1945                          "warning: mounting fs with errors, "
1946                          "running e2fsck is recommended");
1947         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1948                  le16_to_cpu(es->s_mnt_count) >=
1949                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1950                 ext4_msg(sb, KERN_WARNING,
1951                          "warning: maximal mount count reached, "
1952                          "running e2fsck is recommended");
1953         else if (le32_to_cpu(es->s_checkinterval) &&
1954                 (le32_to_cpu(es->s_lastcheck) +
1955                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1956                 ext4_msg(sb, KERN_WARNING,
1957                          "warning: checktime reached, "
1958                          "running e2fsck is recommended");
1959         if (!sbi->s_journal)
1960                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1961         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1962                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1963         le16_add_cpu(&es->s_mnt_count, 1);
1964         es->s_mtime = cpu_to_le32(get_seconds());
1965         ext4_update_dynamic_rev(sb);
1966         if (sbi->s_journal)
1967                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1968
1969         ext4_commit_super(sb, 1);
1970         if (test_opt(sb, DEBUG))
1971                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1972                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1973                         sb->s_blocksize,
1974                         sbi->s_groups_count,
1975                         EXT4_BLOCKS_PER_GROUP(sb),
1976                         EXT4_INODES_PER_GROUP(sb),
1977                         sbi->s_mount_opt, sbi->s_mount_opt2);
1978
1979         cleancache_init_fs(sb);
1980         return res;
1981 }
1982
1983 static int ext4_fill_flex_info(struct super_block *sb)
1984 {
1985         struct ext4_sb_info *sbi = EXT4_SB(sb);
1986         struct ext4_group_desc *gdp = NULL;
1987         ext4_group_t flex_group_count;
1988         ext4_group_t flex_group;
1989         int groups_per_flex = 0;
1990         size_t size;
1991         int i;
1992
1993         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1994         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1995
1996         if (groups_per_flex < 2) {
1997                 sbi->s_log_groups_per_flex = 0;
1998                 return 1;
1999         }
2000
2001         /* We allocate both existing and potentially added groups */
2002         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2003                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2004                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2005         size = flex_group_count * sizeof(struct flex_groups);
2006         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2007         if (sbi->s_flex_groups == NULL) {
2008                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2009                          flex_group_count);
2010                 goto failed;
2011         }
2012
2013         for (i = 0; i < sbi->s_groups_count; i++) {
2014                 gdp = ext4_get_group_desc(sb, i, NULL);
2015
2016                 flex_group = ext4_flex_group(sbi, i);
2017                 atomic_add(ext4_free_inodes_count(sb, gdp),
2018                            &sbi->s_flex_groups[flex_group].free_inodes);
2019                 atomic_add(ext4_free_blks_count(sb, gdp),
2020                            &sbi->s_flex_groups[flex_group].free_blocks);
2021                 atomic_add(ext4_used_dirs_count(sb, gdp),
2022                            &sbi->s_flex_groups[flex_group].used_dirs);
2023         }
2024
2025         return 1;
2026 failed:
2027         return 0;
2028 }
2029
2030 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2031                             struct ext4_group_desc *gdp)
2032 {
2033         __u16 crc = 0;
2034
2035         if (sbi->s_es->s_feature_ro_compat &
2036             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2037                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2038                 __le32 le_group = cpu_to_le32(block_group);
2039
2040                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2041                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2042                 crc = crc16(crc, (__u8 *)gdp, offset);
2043                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2044                 /* for checksum of struct ext4_group_desc do the rest...*/
2045                 if ((sbi->s_es->s_feature_incompat &
2046                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2047                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2048                         crc = crc16(crc, (__u8 *)gdp + offset,
2049                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2050                                         offset);
2051         }
2052
2053         return cpu_to_le16(crc);
2054 }
2055
2056 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2057                                 struct ext4_group_desc *gdp)
2058 {
2059         if ((sbi->s_es->s_feature_ro_compat &
2060              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2061             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2062                 return 0;
2063
2064         return 1;
2065 }
2066
2067 /* Called at mount-time, super-block is locked */
2068 static int ext4_check_descriptors(struct super_block *sb,
2069                                   ext4_group_t *first_not_zeroed)
2070 {
2071         struct ext4_sb_info *sbi = EXT4_SB(sb);
2072         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2073         ext4_fsblk_t last_block;
2074         ext4_fsblk_t block_bitmap;
2075         ext4_fsblk_t inode_bitmap;
2076         ext4_fsblk_t inode_table;
2077         int flexbg_flag = 0;
2078         ext4_group_t i, grp = sbi->s_groups_count;
2079
2080         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2081                 flexbg_flag = 1;
2082
2083         ext4_debug("Checking group descriptors");
2084
2085         for (i = 0; i < sbi->s_groups_count; i++) {
2086                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2087
2088                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2089                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2090                 else
2091                         last_block = first_block +
2092                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2093
2094                 if ((grp == sbi->s_groups_count) &&
2095                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2096                         grp = i;
2097
2098                 block_bitmap = ext4_block_bitmap(sb, gdp);
2099                 if (block_bitmap < first_block || block_bitmap > last_block) {
2100                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2101                                "Block bitmap for group %u not in group "
2102                                "(block %llu)!", i, block_bitmap);
2103                         return 0;
2104                 }
2105                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2106                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Inode bitmap for group %u not in group "
2109                                "(block %llu)!", i, inode_bitmap);
2110                         return 0;
2111                 }
2112                 inode_table = ext4_inode_table(sb, gdp);
2113                 if (inode_table < first_block ||
2114                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2115                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2116                                "Inode table for group %u not in group "
2117                                "(block %llu)!", i, inode_table);
2118                         return 0;
2119                 }
2120                 ext4_lock_group(sb, i);
2121                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2122                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2123                                  "Checksum for group %u failed (%u!=%u)",
2124                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2125                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2126                         if (!(sb->s_flags & MS_RDONLY)) {
2127                                 ext4_unlock_group(sb, i);
2128                                 return 0;
2129                         }
2130                 }
2131                 ext4_unlock_group(sb, i);
2132                 if (!flexbg_flag)
2133                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2134         }
2135         if (NULL != first_not_zeroed)
2136                 *first_not_zeroed = grp;
2137
2138         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2139         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2140         return 1;
2141 }
2142
2143 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2144  * the superblock) which were deleted from all directories, but held open by
2145  * a process at the time of a crash.  We walk the list and try to delete these
2146  * inodes at recovery time (only with a read-write filesystem).
2147  *
2148  * In order to keep the orphan inode chain consistent during traversal (in
2149  * case of crash during recovery), we link each inode into the superblock
2150  * orphan list_head and handle it the same way as an inode deletion during
2151  * normal operation (which journals the operations for us).
2152  *
2153  * We only do an iget() and an iput() on each inode, which is very safe if we
2154  * accidentally point at an in-use or already deleted inode.  The worst that
2155  * can happen in this case is that we get a "bit already cleared" message from
2156  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2157  * e2fsck was run on this filesystem, and it must have already done the orphan
2158  * inode cleanup for us, so we can safely abort without any further action.
2159  */
2160 static void ext4_orphan_cleanup(struct super_block *sb,
2161                                 struct ext4_super_block *es)
2162 {
2163         unsigned int s_flags = sb->s_flags;
2164         int nr_orphans = 0, nr_truncates = 0;
2165 #ifdef CONFIG_QUOTA
2166         int i;
2167 #endif
2168         if (!es->s_last_orphan) {
2169                 jbd_debug(4, "no orphan inodes to clean up\n");
2170                 return;
2171         }
2172
2173         if (bdev_read_only(sb->s_bdev)) {
2174                 ext4_msg(sb, KERN_ERR, "write access "
2175                         "unavailable, skipping orphan cleanup");
2176                 return;
2177         }
2178
2179         /* Check if feature set would not allow a r/w mount */
2180         if (!ext4_feature_set_ok(sb, 0)) {
2181                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2182                          "unknown ROCOMPAT features");
2183                 return;
2184         }
2185
2186         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2187                 if (es->s_last_orphan)
2188                         jbd_debug(1, "Errors on filesystem, "
2189                                   "clearing orphan list.\n");
2190                 es->s_last_orphan = 0;
2191                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2192                 return;
2193         }
2194
2195         if (s_flags & MS_RDONLY) {
2196                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2197                 sb->s_flags &= ~MS_RDONLY;
2198         }
2199 #ifdef CONFIG_QUOTA
2200         /* Needed for iput() to work correctly and not trash data */
2201         sb->s_flags |= MS_ACTIVE;
2202         /* Turn on quotas so that they are updated correctly */
2203         for (i = 0; i < MAXQUOTAS; i++) {
2204                 if (EXT4_SB(sb)->s_qf_names[i]) {
2205                         int ret = ext4_quota_on_mount(sb, i);
2206                         if (ret < 0)
2207                                 ext4_msg(sb, KERN_ERR,
2208                                         "Cannot turn on journaled "
2209                                         "quota: error %d", ret);
2210                 }
2211         }
2212 #endif
2213
2214         while (es->s_last_orphan) {
2215                 struct inode *inode;
2216
2217                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2218                 if (IS_ERR(inode)) {
2219                         es->s_last_orphan = 0;
2220                         break;
2221                 }
2222
2223                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2224                 dquot_initialize(inode);
2225                 if (inode->i_nlink) {
2226                         ext4_msg(sb, KERN_DEBUG,
2227                                 "%s: truncating inode %lu to %lld bytes",
2228                                 __func__, inode->i_ino, inode->i_size);
2229                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2230                                   inode->i_ino, inode->i_size);
2231                         ext4_truncate(inode);
2232                         nr_truncates++;
2233                 } else {
2234                         ext4_msg(sb, KERN_DEBUG,
2235                                 "%s: deleting unreferenced inode %lu",
2236                                 __func__, inode->i_ino);
2237                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2238                                   inode->i_ino);
2239                         nr_orphans++;
2240                 }
2241                 iput(inode);  /* The delete magic happens here! */
2242         }
2243
2244 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2245
2246         if (nr_orphans)
2247                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2248                        PLURAL(nr_orphans));
2249         if (nr_truncates)
2250                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2251                        PLURAL(nr_truncates));
2252 #ifdef CONFIG_QUOTA
2253         /* Turn quotas off */
2254         for (i = 0; i < MAXQUOTAS; i++) {
2255                 if (sb_dqopt(sb)->files[i])
2256                         dquot_quota_off(sb, i);
2257         }
2258 #endif
2259         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2260 }
2261
2262 /*
2263  * Maximal extent format file size.
2264  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2265  * extent format containers, within a sector_t, and within i_blocks
2266  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2267  * so that won't be a limiting factor.
2268  *
2269  * However there is other limiting factor. We do store extents in the form
2270  * of starting block and length, hence the resulting length of the extent
2271  * covering maximum file size must fit into on-disk format containers as
2272  * well. Given that length is always by 1 unit bigger than max unit (because
2273  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2274  *
2275  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2276  */
2277 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2278 {
2279         loff_t res;
2280         loff_t upper_limit = MAX_LFS_FILESIZE;
2281
2282         /* small i_blocks in vfs inode? */
2283         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2284                 /*
2285                  * CONFIG_LBDAF is not enabled implies the inode
2286                  * i_block represent total blocks in 512 bytes
2287                  * 32 == size of vfs inode i_blocks * 8
2288                  */
2289                 upper_limit = (1LL << 32) - 1;
2290
2291                 /* total blocks in file system block size */
2292                 upper_limit >>= (blkbits - 9);
2293                 upper_limit <<= blkbits;
2294         }
2295
2296         /*
2297          * 32-bit extent-start container, ee_block. We lower the maxbytes
2298          * by one fs block, so ee_len can cover the extent of maximum file
2299          * size
2300          */
2301         res = (1LL << 32) - 1;
2302         res <<= blkbits;
2303
2304         /* Sanity check against vm- & vfs- imposed limits */
2305         if (res > upper_limit)
2306                 res = upper_limit;
2307
2308         return res;
2309 }
2310
2311 /*
2312  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2313  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2314  * We need to be 1 filesystem block less than the 2^48 sector limit.
2315  */
2316 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2317 {
2318         loff_t res = EXT4_NDIR_BLOCKS;
2319         int meta_blocks;
2320         loff_t upper_limit;
2321         /* This is calculated to be the largest file size for a dense, block
2322          * mapped file such that the file's total number of 512-byte sectors,
2323          * including data and all indirect blocks, does not exceed (2^48 - 1).
2324          *
2325          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2326          * number of 512-byte sectors of the file.
2327          */
2328
2329         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2330                 /*
2331                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2332                  * the inode i_block field represents total file blocks in
2333                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2334                  */
2335                 upper_limit = (1LL << 32) - 1;
2336
2337                 /* total blocks in file system block size */
2338                 upper_limit >>= (bits - 9);
2339
2340         } else {
2341                 /*
2342                  * We use 48 bit ext4_inode i_blocks
2343                  * With EXT4_HUGE_FILE_FL set the i_blocks
2344                  * represent total number of blocks in
2345                  * file system block size
2346                  */
2347                 upper_limit = (1LL << 48) - 1;
2348
2349         }
2350
2351         /* indirect blocks */
2352         meta_blocks = 1;
2353         /* double indirect blocks */
2354         meta_blocks += 1 + (1LL << (bits-2));
2355         /* tripple indirect blocks */
2356         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2357
2358         upper_limit -= meta_blocks;
2359         upper_limit <<= bits;
2360
2361         res += 1LL << (bits-2);
2362         res += 1LL << (2*(bits-2));
2363         res += 1LL << (3*(bits-2));
2364         res <<= bits;
2365         if (res > upper_limit)
2366                 res = upper_limit;
2367
2368         if (res > MAX_LFS_FILESIZE)
2369                 res = MAX_LFS_FILESIZE;
2370
2371         return res;
2372 }
2373
2374 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2375                                    ext4_fsblk_t logical_sb_block, int nr)
2376 {
2377         struct ext4_sb_info *sbi = EXT4_SB(sb);
2378         ext4_group_t bg, first_meta_bg;
2379         int has_super = 0;
2380
2381         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2382
2383         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2384             nr < first_meta_bg)
2385                 return logical_sb_block + nr + 1;
2386         bg = sbi->s_desc_per_block * nr;
2387         if (ext4_bg_has_super(sb, bg))
2388                 has_super = 1;
2389
2390         return (has_super + ext4_group_first_block_no(sb, bg));
2391 }
2392
2393 /**
2394  * ext4_get_stripe_size: Get the stripe size.
2395  * @sbi: In memory super block info
2396  *
2397  * If we have specified it via mount option, then
2398  * use the mount option value. If the value specified at mount time is
2399  * greater than the blocks per group use the super block value.
2400  * If the super block value is greater than blocks per group return 0.
2401  * Allocator needs it be less than blocks per group.
2402  *
2403  */
2404 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2405 {
2406         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2407         unsigned long stripe_width =
2408                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2409         int ret;
2410
2411         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2412                 ret = sbi->s_stripe;
2413         else if (stripe_width <= sbi->s_blocks_per_group)
2414                 ret = stripe_width;
2415         else if (stride <= sbi->s_blocks_per_group)
2416                 ret = stride;
2417         else
2418                 ret = 0;
2419
2420         /*
2421          * If the stripe width is 1, this makes no sense and
2422          * we set it to 0 to turn off stripe handling code.
2423          */
2424         if (ret <= 1)
2425                 ret = 0;
2426
2427         return ret;
2428 }
2429
2430 /* sysfs supprt */
2431
2432 struct ext4_attr {
2433         struct attribute attr;
2434         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2435         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2436                          const char *, size_t);
2437         int offset;
2438 };
2439
2440 static int parse_strtoul(const char *buf,
2441                 unsigned long max, unsigned long *value)
2442 {
2443         char *endp;
2444
2445         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2446         endp = skip_spaces(endp);
2447         if (*endp || *value > max)
2448                 return -EINVAL;
2449
2450         return 0;
2451 }
2452
2453 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2454                                               struct ext4_sb_info *sbi,
2455                                               char *buf)
2456 {
2457         return snprintf(buf, PAGE_SIZE, "%llu\n",
2458                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2459 }
2460
2461 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2462                                          struct ext4_sb_info *sbi, char *buf)
2463 {
2464         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2465
2466         if (!sb->s_bdev->bd_part)
2467                 return snprintf(buf, PAGE_SIZE, "0\n");
2468         return snprintf(buf, PAGE_SIZE, "%lu\n",
2469                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2470                          sbi->s_sectors_written_start) >> 1);
2471 }
2472
2473 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2474                                           struct ext4_sb_info *sbi, char *buf)
2475 {
2476         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2477
2478         if (!sb->s_bdev->bd_part)
2479                 return snprintf(buf, PAGE_SIZE, "0\n");
2480         return snprintf(buf, PAGE_SIZE, "%llu\n",
2481                         (unsigned long long)(sbi->s_kbytes_written +
2482                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2483                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2484 }
2485
2486 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2487                                       struct ext4_sb_info *sbi, char *buf)
2488 {
2489         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2490 }
2491
2492 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2493                                         struct ext4_sb_info *sbi, char *buf)
2494 {
2495         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2496 }
2497
2498 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2499                                           struct ext4_sb_info *sbi,
2500                                           const char *buf, size_t count)
2501 {
2502         unsigned long t;
2503
2504         if (parse_strtoul(buf, 0x40000000, &t))
2505                 return -EINVAL;
2506
2507         if (t && !is_power_of_2(t))
2508                 return -EINVAL;
2509
2510         sbi->s_inode_readahead_blks = t;
2511         return count;
2512 }
2513
2514 static ssize_t sbi_ui_show(struct ext4_attr *a,
2515                            struct ext4_sb_info *sbi, char *buf)
2516 {
2517         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2518
2519         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2520 }
2521
2522 static ssize_t sbi_ui_store(struct ext4_attr *a,
2523                             struct ext4_sb_info *sbi,
2524                             const char *buf, size_t count)
2525 {
2526         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2527         unsigned long t;
2528
2529         if (parse_strtoul(buf, 0xffffffff, &t))
2530                 return -EINVAL;
2531         *ui = t;
2532         return count;
2533 }
2534
2535 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2536 static struct ext4_attr ext4_attr_##_name = {                   \
2537         .attr = {.name = __stringify(_name), .mode = _mode },   \
2538         .show   = _show,                                        \
2539         .store  = _store,                                       \
2540         .offset = offsetof(struct ext4_sb_info, _elname),       \
2541 }
2542 #define EXT4_ATTR(name, mode, show, store) \
2543 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2544
2545 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2546 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2547 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2548 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2549         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2550 #define ATTR_LIST(name) &ext4_attr_##name.attr
2551
2552 EXT4_RO_ATTR(delayed_allocation_blocks);
2553 EXT4_RO_ATTR(session_write_kbytes);
2554 EXT4_RO_ATTR(lifetime_write_kbytes);
2555 EXT4_RO_ATTR(extent_cache_hits);
2556 EXT4_RO_ATTR(extent_cache_misses);
2557 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2558                  inode_readahead_blks_store, s_inode_readahead_blks);
2559 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2560 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2561 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2562 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2563 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2564 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2565 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2566 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2567
2568 static struct attribute *ext4_attrs[] = {
2569         ATTR_LIST(delayed_allocation_blocks),
2570         ATTR_LIST(session_write_kbytes),
2571         ATTR_LIST(lifetime_write_kbytes),
2572         ATTR_LIST(extent_cache_hits),
2573         ATTR_LIST(extent_cache_misses),
2574         ATTR_LIST(inode_readahead_blks),
2575         ATTR_LIST(inode_goal),
2576         ATTR_LIST(mb_stats),
2577         ATTR_LIST(mb_max_to_scan),
2578         ATTR_LIST(mb_min_to_scan),
2579         ATTR_LIST(mb_order2_req),
2580         ATTR_LIST(mb_stream_req),
2581         ATTR_LIST(mb_group_prealloc),
2582         ATTR_LIST(max_writeback_mb_bump),
2583         NULL,
2584 };
2585
2586 /* Features this copy of ext4 supports */
2587 EXT4_INFO_ATTR(lazy_itable_init);
2588 EXT4_INFO_ATTR(batched_discard);
2589
2590 static struct attribute *ext4_feat_attrs[] = {
2591         ATTR_LIST(lazy_itable_init),
2592         ATTR_LIST(batched_discard),
2593         NULL,
2594 };
2595
2596 static ssize_t ext4_attr_show(struct kobject *kobj,
2597                               struct attribute *attr, char *buf)
2598 {
2599         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2600                                                 s_kobj);
2601         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2602
2603         return a->show ? a->show(a, sbi, buf) : 0;
2604 }
2605
2606 static ssize_t ext4_attr_store(struct kobject *kobj,
2607                                struct attribute *attr,
2608                                const char *buf, size_t len)
2609 {
2610         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2611                                                 s_kobj);
2612         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2613
2614         return a->store ? a->store(a, sbi, buf, len) : 0;
2615 }
2616
2617 static void ext4_sb_release(struct kobject *kobj)
2618 {
2619         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2620                                                 s_kobj);
2621         complete(&sbi->s_kobj_unregister);
2622 }
2623
2624 static const struct sysfs_ops ext4_attr_ops = {
2625         .show   = ext4_attr_show,
2626         .store  = ext4_attr_store,
2627 };
2628
2629 static struct kobj_type ext4_ktype = {
2630         .default_attrs  = ext4_attrs,
2631         .sysfs_ops      = &ext4_attr_ops,
2632         .release        = ext4_sb_release,
2633 };
2634
2635 static void ext4_feat_release(struct kobject *kobj)
2636 {
2637         complete(&ext4_feat->f_kobj_unregister);
2638 }
2639
2640 static struct kobj_type ext4_feat_ktype = {
2641         .default_attrs  = ext4_feat_attrs,
2642         .sysfs_ops      = &ext4_attr_ops,
2643         .release        = ext4_feat_release,
2644 };
2645
2646 /*
2647  * Check whether this filesystem can be mounted based on
2648  * the features present and the RDONLY/RDWR mount requested.
2649  * Returns 1 if this filesystem can be mounted as requested,
2650  * 0 if it cannot be.
2651  */
2652 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2653 {
2654         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2655                 ext4_msg(sb, KERN_ERR,
2656                         "Couldn't mount because of "
2657                         "unsupported optional features (%x)",
2658                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2659                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2660                 return 0;
2661         }
2662
2663         if (readonly)
2664                 return 1;
2665
2666         /* Check that feature set is OK for a read-write mount */
2667         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2668                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2669                          "unsupported optional features (%x)",
2670                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2671                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2672                 return 0;
2673         }
2674         /*
2675          * Large file size enabled file system can only be mounted
2676          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2677          */
2678         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2679                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2680                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2681                                  "cannot be mounted RDWR without "
2682                                  "CONFIG_LBDAF");
2683                         return 0;
2684                 }
2685         }
2686         return 1;
2687 }
2688
2689 /*
2690  * This function is called once a day if we have errors logged
2691  * on the file system
2692  */
2693 static void print_daily_error_info(unsigned long arg)
2694 {
2695         struct super_block *sb = (struct super_block *) arg;
2696         struct ext4_sb_info *sbi;
2697         struct ext4_super_block *es;
2698
2699         sbi = EXT4_SB(sb);
2700         es = sbi->s_es;
2701
2702         if (es->s_error_count)
2703                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2704                          le32_to_cpu(es->s_error_count));
2705         if (es->s_first_error_time) {
2706                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2707                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2708                        (int) sizeof(es->s_first_error_func),
2709                        es->s_first_error_func,
2710                        le32_to_cpu(es->s_first_error_line));
2711                 if (es->s_first_error_ino)
2712                         printk(": inode %u",
2713                                le32_to_cpu(es->s_first_error_ino));
2714                 if (es->s_first_error_block)
2715                         printk(": block %llu", (unsigned long long)
2716                                le64_to_cpu(es->s_first_error_block));
2717                 printk("\n");
2718         }
2719         if (es->s_last_error_time) {
2720                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2721                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2722                        (int) sizeof(es->s_last_error_func),
2723                        es->s_last_error_func,
2724                        le32_to_cpu(es->s_last_error_line));
2725                 if (es->s_last_error_ino)
2726                         printk(": inode %u",
2727                                le32_to_cpu(es->s_last_error_ino));
2728                 if (es->s_last_error_block)
2729                         printk(": block %llu", (unsigned long long)
2730                                le64_to_cpu(es->s_last_error_block));
2731                 printk("\n");
2732         }
2733         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2734 }
2735
2736 /* Find next suitable group and run ext4_init_inode_table */
2737 static int ext4_run_li_request(struct ext4_li_request *elr)
2738 {
2739         struct ext4_group_desc *gdp = NULL;
2740         ext4_group_t group, ngroups;
2741         struct super_block *sb;
2742         unsigned long timeout = 0;
2743         int ret = 0;
2744
2745         sb = elr->lr_super;
2746         ngroups = EXT4_SB(sb)->s_groups_count;
2747
2748         for (group = elr->lr_next_group; group < ngroups; group++) {
2749                 gdp = ext4_get_group_desc(sb, group, NULL);
2750                 if (!gdp) {
2751                         ret = 1;
2752                         break;
2753                 }
2754
2755                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2756                         break;
2757         }
2758
2759         if (group == ngroups)
2760                 ret = 1;
2761
2762         if (!ret) {
2763                 timeout = jiffies;
2764                 ret = ext4_init_inode_table(sb, group,
2765                                             elr->lr_timeout ? 0 : 1);
2766                 if (elr->lr_timeout == 0) {
2767                         timeout = (jiffies - timeout) *
2768                                   elr->lr_sbi->s_li_wait_mult;
2769                         elr->lr_timeout = timeout;
2770                 }
2771                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2772                 elr->lr_next_group = group + 1;
2773         }
2774
2775         return ret;
2776 }
2777
2778 /*
2779  * Remove lr_request from the list_request and free the
2780  * request structure. Should be called with li_list_mtx held
2781  */
2782 static void ext4_remove_li_request(struct ext4_li_request *elr)
2783 {
2784         struct ext4_sb_info *sbi;
2785
2786         if (!elr)
2787                 return;
2788
2789         sbi = elr->lr_sbi;
2790
2791         list_del(&elr->lr_request);
2792         sbi->s_li_request = NULL;
2793         kfree(elr);
2794 }
2795
2796 static void ext4_unregister_li_request(struct super_block *sb)
2797 {
2798         mutex_lock(&ext4_li_mtx);
2799         if (!ext4_li_info) {
2800                 mutex_unlock(&ext4_li_mtx);
2801                 return;
2802         }
2803
2804         mutex_lock(&ext4_li_info->li_list_mtx);
2805         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2806         mutex_unlock(&ext4_li_info->li_list_mtx);
2807         mutex_unlock(&ext4_li_mtx);
2808 }
2809
2810 static struct task_struct *ext4_lazyinit_task;
2811
2812 /*
2813  * This is the function where ext4lazyinit thread lives. It walks
2814  * through the request list searching for next scheduled filesystem.
2815  * When such a fs is found, run the lazy initialization request
2816  * (ext4_rn_li_request) and keep track of the time spend in this
2817  * function. Based on that time we compute next schedule time of
2818  * the request. When walking through the list is complete, compute
2819  * next waking time and put itself into sleep.
2820  */
2821 static int ext4_lazyinit_thread(void *arg)
2822 {
2823         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2824         struct list_head *pos, *n;
2825         struct ext4_li_request *elr;
2826         unsigned long next_wakeup, cur;
2827
2828         BUG_ON(NULL == eli);
2829
2830 cont_thread:
2831         while (true) {
2832                 next_wakeup = MAX_JIFFY_OFFSET;
2833
2834                 mutex_lock(&eli->li_list_mtx);
2835                 if (list_empty(&eli->li_request_list)) {
2836                         mutex_unlock(&eli->li_list_mtx);
2837                         goto exit_thread;
2838                 }
2839
2840                 list_for_each_safe(pos, n, &eli->li_request_list) {
2841                         elr = list_entry(pos, struct ext4_li_request,
2842                                          lr_request);
2843
2844                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2845                                 if (ext4_run_li_request(elr) != 0) {
2846                                         /* error, remove the lazy_init job */
2847                                         ext4_remove_li_request(elr);
2848                                         continue;
2849                                 }
2850                         }
2851
2852                         if (time_before(elr->lr_next_sched, next_wakeup))
2853                                 next_wakeup = elr->lr_next_sched;
2854                 }
2855                 mutex_unlock(&eli->li_list_mtx);
2856
2857                 if (freezing(current))
2858                         refrigerator();
2859
2860                 cur = jiffies;
2861                 if ((time_after_eq(cur, next_wakeup)) ||
2862                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2863                         cond_resched();
2864                         continue;
2865                 }
2866
2867                 schedule_timeout_interruptible(next_wakeup - cur);
2868
2869                 if (kthread_should_stop()) {
2870                         ext4_clear_request_list();
2871                         goto exit_thread;
2872                 }
2873         }
2874
2875 exit_thread:
2876         /*
2877          * It looks like the request list is empty, but we need
2878          * to check it under the li_list_mtx lock, to prevent any
2879          * additions into it, and of course we should lock ext4_li_mtx
2880          * to atomically free the list and ext4_li_info, because at
2881          * this point another ext4 filesystem could be registering
2882          * new one.
2883          */
2884         mutex_lock(&ext4_li_mtx);
2885         mutex_lock(&eli->li_list_mtx);
2886         if (!list_empty(&eli->li_request_list)) {
2887                 mutex_unlock(&eli->li_list_mtx);
2888                 mutex_unlock(&ext4_li_mtx);
2889                 goto cont_thread;
2890         }
2891         mutex_unlock(&eli->li_list_mtx);
2892         kfree(ext4_li_info);
2893         ext4_li_info = NULL;
2894         mutex_unlock(&ext4_li_mtx);
2895
2896         return 0;
2897 }
2898
2899 static void ext4_clear_request_list(void)
2900 {
2901         struct list_head *pos, *n;
2902         struct ext4_li_request *elr;
2903
2904         mutex_lock(&ext4_li_info->li_list_mtx);
2905         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2906                 elr = list_entry(pos, struct ext4_li_request,
2907                                  lr_request);
2908                 ext4_remove_li_request(elr);
2909         }
2910         mutex_unlock(&ext4_li_info->li_list_mtx);
2911 }
2912
2913 static int ext4_run_lazyinit_thread(void)
2914 {
2915         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2916                                          ext4_li_info, "ext4lazyinit");
2917         if (IS_ERR(ext4_lazyinit_task)) {
2918                 int err = PTR_ERR(ext4_lazyinit_task);
2919                 ext4_clear_request_list();
2920                 kfree(ext4_li_info);
2921                 ext4_li_info = NULL;
2922                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2923                                  "initialization thread\n",
2924                                  err);
2925                 return err;
2926         }
2927         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2928         return 0;
2929 }
2930
2931 /*
2932  * Check whether it make sense to run itable init. thread or not.
2933  * If there is at least one uninitialized inode table, return
2934  * corresponding group number, else the loop goes through all
2935  * groups and return total number of groups.
2936  */
2937 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2938 {
2939         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2940         struct ext4_group_desc *gdp = NULL;
2941
2942         for (group = 0; group < ngroups; group++) {
2943                 gdp = ext4_get_group_desc(sb, group, NULL);
2944                 if (!gdp)
2945                         continue;
2946
2947                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2948                         break;
2949         }
2950
2951         return group;
2952 }
2953
2954 static int ext4_li_info_new(void)
2955 {
2956         struct ext4_lazy_init *eli = NULL;
2957
2958         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2959         if (!eli)
2960                 return -ENOMEM;
2961
2962         INIT_LIST_HEAD(&eli->li_request_list);
2963         mutex_init(&eli->li_list_mtx);
2964
2965         eli->li_state |= EXT4_LAZYINIT_QUIT;
2966
2967         ext4_li_info = eli;
2968
2969         return 0;
2970 }
2971
2972 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2973                                             ext4_group_t start)
2974 {
2975         struct ext4_sb_info *sbi = EXT4_SB(sb);
2976         struct ext4_li_request *elr;
2977         unsigned long rnd;
2978
2979         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2980         if (!elr)
2981                 return NULL;
2982
2983         elr->lr_super = sb;
2984         elr->lr_sbi = sbi;
2985         elr->lr_next_group = start;
2986
2987         /*
2988          * Randomize first schedule time of the request to
2989          * spread the inode table initialization requests
2990          * better.
2991          */
2992         get_random_bytes(&rnd, sizeof(rnd));
2993         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2994                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2995
2996         return elr;
2997 }
2998
2999 static int ext4_register_li_request(struct super_block *sb,
3000                                     ext4_group_t first_not_zeroed)
3001 {
3002         struct ext4_sb_info *sbi = EXT4_SB(sb);
3003         struct ext4_li_request *elr;
3004         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3005         int ret = 0;
3006
3007         if (sbi->s_li_request != NULL) {
3008                 /*
3009                  * Reset timeout so it can be computed again, because
3010                  * s_li_wait_mult might have changed.
3011                  */
3012                 sbi->s_li_request->lr_timeout = 0;
3013                 return 0;
3014         }
3015
3016         if (first_not_zeroed == ngroups ||
3017             (sb->s_flags & MS_RDONLY) ||
3018             !test_opt(sb, INIT_INODE_TABLE))
3019                 return 0;
3020
3021         elr = ext4_li_request_new(sb, first_not_zeroed);
3022         if (!elr)
3023                 return -ENOMEM;
3024
3025         mutex_lock(&ext4_li_mtx);
3026
3027         if (NULL == ext4_li_info) {
3028                 ret = ext4_li_info_new();
3029                 if (ret)
3030                         goto out;
3031         }
3032
3033         mutex_lock(&ext4_li_info->li_list_mtx);
3034         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3035         mutex_unlock(&ext4_li_info->li_list_mtx);
3036
3037         sbi->s_li_request = elr;
3038         /*
3039          * set elr to NULL here since it has been inserted to
3040          * the request_list and the removal and free of it is
3041          * handled by ext4_clear_request_list from now on.
3042          */
3043         elr = NULL;
3044
3045         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3046                 ret = ext4_run_lazyinit_thread();
3047                 if (ret)
3048                         goto out;
3049         }
3050 out:
3051         mutex_unlock(&ext4_li_mtx);
3052         if (ret)
3053                 kfree(elr);
3054         return ret;
3055 }
3056
3057 /*
3058  * We do not need to lock anything since this is called on
3059  * module unload.
3060  */
3061 static void ext4_destroy_lazyinit_thread(void)
3062 {
3063         /*
3064          * If thread exited earlier
3065          * there's nothing to be done.
3066          */
3067         if (!ext4_li_info || !ext4_lazyinit_task)
3068                 return;
3069
3070         kthread_stop(ext4_lazyinit_task);
3071 }
3072
3073 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3074                                 __releases(kernel_lock)
3075                                 __acquires(kernel_lock)
3076 {
3077         char *orig_data = kstrdup(data, GFP_KERNEL);
3078         struct buffer_head *bh;
3079         struct ext4_super_block *es = NULL;
3080         struct ext4_sb_info *sbi;
3081         ext4_fsblk_t block;
3082         ext4_fsblk_t sb_block = get_sb_block(&data);
3083         ext4_fsblk_t logical_sb_block;
3084         unsigned long offset = 0;
3085         unsigned long journal_devnum = 0;
3086         unsigned long def_mount_opts;
3087         struct inode *root;
3088         char *cp;
3089         const char *descr;
3090         int ret = -ENOMEM;
3091         int blocksize;
3092         unsigned int db_count;
3093         unsigned int i;
3094         int needs_recovery, has_huge_files;
3095         __u64 blocks_count;
3096         int err;
3097         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3098         ext4_group_t first_not_zeroed;
3099
3100         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3101         if (!sbi)
3102                 goto out_free_orig;
3103
3104         sbi->s_blockgroup_lock =
3105                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3106         if (!sbi->s_blockgroup_lock) {
3107                 kfree(sbi);
3108                 goto out_free_orig;
3109         }
3110         sb->s_fs_info = sbi;
3111         sbi->s_mount_opt = 0;
3112         sbi->s_resuid = EXT4_DEF_RESUID;
3113         sbi->s_resgid = EXT4_DEF_RESGID;
3114         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3115         sbi->s_sb_block = sb_block;
3116         if (sb->s_bdev->bd_part)
3117                 sbi->s_sectors_written_start =
3118                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3119
3120         /* Cleanup superblock name */
3121         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3122                 *cp = '!';
3123
3124         ret = -EINVAL;
3125         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3126         if (!blocksize) {
3127                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3128                 goto out_fail;
3129         }
3130
3131         /*
3132          * The ext4 superblock will not be buffer aligned for other than 1kB
3133          * block sizes.  We need to calculate the offset from buffer start.
3134          */
3135         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3136                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3137                 offset = do_div(logical_sb_block, blocksize);
3138         } else {
3139                 logical_sb_block = sb_block;
3140         }
3141
3142         if (!(bh = sb_bread(sb, logical_sb_block))) {
3143                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3144                 goto out_fail;
3145         }
3146         /*
3147          * Note: s_es must be initialized as soon as possible because
3148          *       some ext4 macro-instructions depend on its value
3149          */
3150         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3151         sbi->s_es = es;
3152         sb->s_magic = le16_to_cpu(es->s_magic);
3153         if (sb->s_magic != EXT4_SUPER_MAGIC)
3154                 goto cantfind_ext4;
3155         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3156
3157         /* Set defaults before we parse the mount options */
3158         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3159         set_opt(sb, INIT_INODE_TABLE);
3160         if (def_mount_opts & EXT4_DEFM_DEBUG)
3161                 set_opt(sb, DEBUG);
3162         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3163                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3164                         "2.6.38");
3165                 set_opt(sb, GRPID);
3166         }
3167         if (def_mount_opts & EXT4_DEFM_UID16)
3168                 set_opt(sb, NO_UID32);
3169         /* xattr user namespace & acls are now defaulted on */
3170 #ifdef CONFIG_EXT4_FS_XATTR
3171         set_opt(sb, XATTR_USER);
3172 #endif
3173 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3174         set_opt(sb, POSIX_ACL);
3175 #endif
3176         set_opt(sb, MBLK_IO_SUBMIT);
3177         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3178                 set_opt(sb, JOURNAL_DATA);
3179         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3180                 set_opt(sb, ORDERED_DATA);
3181         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3182                 set_opt(sb, WRITEBACK_DATA);
3183
3184         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3185                 set_opt(sb, ERRORS_PANIC);
3186         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3187                 set_opt(sb, ERRORS_CONT);
3188         else
3189                 set_opt(sb, ERRORS_RO);
3190         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3191                 set_opt(sb, BLOCK_VALIDITY);
3192         if (def_mount_opts & EXT4_DEFM_DISCARD)
3193                 set_opt(sb, DISCARD);
3194
3195         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3196         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3197         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3198         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3199         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3200
3201         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3202                 set_opt(sb, BARRIER);
3203
3204         /*
3205          * enable delayed allocation by default
3206          * Use -o nodelalloc to turn it off
3207          */
3208         if (!IS_EXT3_SB(sb) &&
3209             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3210                 set_opt(sb, DELALLOC);
3211
3212         /*
3213          * set default s_li_wait_mult for lazyinit, for the case there is
3214          * no mount option specified.
3215          */
3216         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3217
3218         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3219                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3220                 ext4_msg(sb, KERN_WARNING,
3221                          "failed to parse options in superblock: %s",
3222                          sbi->s_es->s_mount_opts);
3223         }
3224         if (!parse_options((char *) data, sb, &journal_devnum,
3225                            &journal_ioprio, NULL, 0))
3226                 goto failed_mount;
3227
3228         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3229                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3230
3231         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3232             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3233              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3234              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3235                 ext4_msg(sb, KERN_WARNING,
3236                        "feature flags set on rev 0 fs, "
3237                        "running e2fsck is recommended");
3238
3239         if (IS_EXT2_SB(sb)) {
3240                 if (ext2_feature_set_ok(sb))
3241                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3242                                  "using the ext4 subsystem");
3243                 else {
3244                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3245                                  "to feature incompatibilities");
3246                         goto failed_mount;
3247                 }
3248         }
3249
3250         if (IS_EXT3_SB(sb)) {
3251                 if (ext3_feature_set_ok(sb))
3252                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3253                                  "using the ext4 subsystem");
3254                 else {
3255                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3256                                  "to feature incompatibilities");
3257                         goto failed_mount;
3258                 }
3259         }
3260
3261         /*
3262          * Check feature flags regardless of the revision level, since we
3263          * previously didn't change the revision level when setting the flags,
3264          * so there is a chance incompat flags are set on a rev 0 filesystem.
3265          */
3266         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3267                 goto failed_mount;
3268
3269         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3270
3271         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3272             blocksize > EXT4_MAX_BLOCK_SIZE) {
3273                 ext4_msg(sb, KERN_ERR,
3274                        "Unsupported filesystem blocksize %d", blocksize);
3275                 goto failed_mount;
3276         }
3277
3278         if (sb->s_blocksize != blocksize) {
3279                 /* Validate the filesystem blocksize */
3280                 if (!sb_set_blocksize(sb, blocksize)) {
3281                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3282                                         blocksize);
3283                         goto failed_mount;
3284                 }
3285
3286                 brelse(bh);
3287                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3288                 offset = do_div(logical_sb_block, blocksize);
3289                 bh = sb_bread(sb, logical_sb_block);
3290                 if (!bh) {
3291                         ext4_msg(sb, KERN_ERR,
3292                                "Can't read superblock on 2nd try");
3293                         goto failed_mount;
3294                 }
3295                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3296                 sbi->s_es = es;
3297                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3298                         ext4_msg(sb, KERN_ERR,
3299                                "Magic mismatch, very weird!");
3300                         goto failed_mount;
3301                 }
3302         }
3303
3304         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3305                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3306         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3307                                                       has_huge_files);
3308         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3309
3310         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3311                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3312                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3313         } else {
3314                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3315                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3316                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3317                     (!is_power_of_2(sbi->s_inode_size)) ||
3318                     (sbi->s_inode_size > blocksize)) {
3319                         ext4_msg(sb, KERN_ERR,
3320                                "unsupported inode size: %d",
3321                                sbi->s_inode_size);
3322                         goto failed_mount;
3323                 }
3324                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3325                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3326         }
3327
3328         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3329         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3330                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3331                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3332                     !is_power_of_2(sbi->s_desc_size)) {
3333                         ext4_msg(sb, KERN_ERR,
3334                                "unsupported descriptor size %lu",
3335                                sbi->s_desc_size);
3336                         goto failed_mount;
3337                 }
3338         } else
3339                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3340
3341         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3342         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3343         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3344                 goto cantfind_ext4;
3345
3346         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3347         if (sbi->s_inodes_per_block == 0)
3348                 goto cantfind_ext4;
3349         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3350                                         sbi->s_inodes_per_block;
3351         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3352         sbi->s_sbh = bh;
3353         sbi->s_mount_state = le16_to_cpu(es->s_state);
3354         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3355         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3356
3357         for (i = 0; i < 4; i++)
3358                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3359         sbi->s_def_hash_version = es->s_def_hash_version;
3360         i = le32_to_cpu(es->s_flags);
3361         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3362                 sbi->s_hash_unsigned = 3;
3363         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3364 #ifdef __CHAR_UNSIGNED__
3365                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3366                 sbi->s_hash_unsigned = 3;
3367 #else
3368                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3369 #endif
3370                 sb->s_dirt = 1;
3371         }
3372
3373         if (sbi->s_blocks_per_group > blocksize * 8) {
3374                 ext4_msg(sb, KERN_ERR,
3375                        "#blocks per group too big: %lu",
3376                        sbi->s_blocks_per_group);
3377                 goto failed_mount;
3378         }
3379         if (sbi->s_inodes_per_group > blocksize * 8) {
3380                 ext4_msg(sb, KERN_ERR,
3381                        "#inodes per group too big: %lu",
3382                        sbi->s_inodes_per_group);
3383                 goto failed_mount;
3384         }
3385
3386         /*
3387          * Test whether we have more sectors than will fit in sector_t,
3388          * and whether the max offset is addressable by the page cache.
3389          */
3390         err = generic_check_addressable(sb->s_blocksize_bits,
3391                                         ext4_blocks_count(es));
3392         if (err) {
3393                 ext4_msg(sb, KERN_ERR, "filesystem"
3394                          " too large to mount safely on this system");
3395                 if (sizeof(sector_t) < 8)
3396                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3397                 ret = err;
3398                 goto failed_mount;
3399         }
3400
3401         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3402                 goto cantfind_ext4;
3403
3404         /* check blocks count against device size */
3405         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3406         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3407                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3408                        "exceeds size of device (%llu blocks)",
3409                        ext4_blocks_count(es), blocks_count);
3410                 goto failed_mount;
3411         }
3412
3413         /*
3414          * It makes no sense for the first data block to be beyond the end
3415          * of the filesystem.
3416          */
3417         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3418                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3419                          "block %u is beyond end of filesystem (%llu)",
3420                          le32_to_cpu(es->s_first_data_block),
3421                          ext4_blocks_count(es));
3422                 goto failed_mount;
3423         }
3424         blocks_count = (ext4_blocks_count(es) -
3425                         le32_to_cpu(es->s_first_data_block) +
3426                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3427         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3428         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3429                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3430                        "(block count %llu, first data block %u, "
3431                        "blocks per group %lu)", sbi->s_groups_count,
3432                        ext4_blocks_count(es),
3433                        le32_to_cpu(es->s_first_data_block),
3434                        EXT4_BLOCKS_PER_GROUP(sb));
3435                 goto failed_mount;
3436         }
3437         sbi->s_groups_count = blocks_count;
3438         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3439                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3440         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3441                    EXT4_DESC_PER_BLOCK(sb);
3442         sbi->s_group_desc = ext4_kvmalloc(db_count *
3443                                           sizeof(struct buffer_head *),
3444                                           GFP_KERNEL);
3445         if (sbi->s_group_desc == NULL) {
3446                 ext4_msg(sb, KERN_ERR, "not enough memory");
3447                 goto failed_mount;
3448         }
3449
3450 #ifdef CONFIG_PROC_FS
3451         if (ext4_proc_root)
3452                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3453 #endif
3454
3455         bgl_lock_init(sbi->s_blockgroup_lock);
3456
3457         for (i = 0; i < db_count; i++) {
3458                 block = descriptor_loc(sb, logical_sb_block, i);
3459                 sbi->s_group_desc[i] = sb_bread(sb, block);
3460                 if (!sbi->s_group_desc[i]) {
3461                         ext4_msg(sb, KERN_ERR,
3462                                "can't read group descriptor %d", i);
3463                         db_count = i;
3464                         goto failed_mount2;
3465                 }
3466         }
3467         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3468                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3469                 goto failed_mount2;
3470         }
3471         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3472                 if (!ext4_fill_flex_info(sb)) {
3473                         ext4_msg(sb, KERN_ERR,
3474                                "unable to initialize "
3475                                "flex_bg meta info!");
3476                         goto failed_mount2;
3477                 }
3478
3479         sbi->s_gdb_count = db_count;
3480         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3481         spin_lock_init(&sbi->s_next_gen_lock);
3482
3483         init_timer(&sbi->s_err_report);
3484         sbi->s_err_report.function = print_daily_error_info;
3485         sbi->s_err_report.data = (unsigned long) sb;
3486
3487         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3488                         ext4_count_free_blocks(sb));
3489         if (!err) {
3490                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3491                                 ext4_count_free_inodes(sb));
3492         }
3493         if (!err) {
3494                 err = percpu_counter_init(&sbi->s_dirs_counter,
3495                                 ext4_count_dirs(sb));
3496         }
3497         if (!err) {
3498                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3499         }
3500         if (err) {
3501                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3502                 goto failed_mount3;
3503         }
3504
3505         sbi->s_stripe = ext4_get_stripe_size(sbi);
3506         sbi->s_max_writeback_mb_bump = 128;
3507
3508         /*
3509          * set up enough so that it can read an inode
3510          */
3511         if (!test_opt(sb, NOLOAD) &&
3512             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3513                 sb->s_op = &ext4_sops;
3514         else
3515                 sb->s_op = &ext4_nojournal_sops;
3516         sb->s_export_op = &ext4_export_ops;
3517         sb->s_xattr = ext4_xattr_handlers;
3518 #ifdef CONFIG_QUOTA
3519         sb->s_qcop = &ext4_qctl_operations;
3520         sb->dq_op = &ext4_quota_operations;
3521 #endif
3522         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3523
3524         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3525         mutex_init(&sbi->s_orphan_lock);
3526         sbi->s_resize_flags = 0;
3527
3528         sb->s_root = NULL;
3529
3530         needs_recovery = (es->s_last_orphan != 0 ||
3531                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3532                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3533
3534         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3535             !(sb->s_flags & MS_RDONLY))
3536                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3537                         goto failed_mount3;
3538
3539         /*
3540          * The first inode we look at is the journal inode.  Don't try
3541          * root first: it may be modified in the journal!
3542          */
3543         if (!test_opt(sb, NOLOAD) &&
3544             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3545                 if (ext4_load_journal(sb, es, journal_devnum))
3546                         goto failed_mount3;
3547         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3548               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3549                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3550                        "suppressed and not mounted read-only");
3551                 goto failed_mount_wq;
3552         } else {
3553                 clear_opt(sb, DATA_FLAGS);
3554                 sbi->s_journal = NULL;
3555                 needs_recovery = 0;
3556                 goto no_journal;
3557         }
3558
3559         if (ext4_blocks_count(es) > 0xffffffffULL &&
3560             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3561                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3562                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3563                 goto failed_mount_wq;
3564         }
3565
3566         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3567                 jbd2_journal_set_features(sbi->s_journal,
3568                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3569                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3570         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3571                 jbd2_journal_set_features(sbi->s_journal,
3572                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3573                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3574                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3575         } else {
3576                 jbd2_journal_clear_features(sbi->s_journal,
3577                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3578                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3579         }
3580
3581         /* We have now updated the journal if required, so we can
3582          * validate the data journaling mode. */
3583         switch (test_opt(sb, DATA_FLAGS)) {
3584         case 0:
3585                 /* No mode set, assume a default based on the journal
3586                  * capabilities: ORDERED_DATA if the journal can
3587                  * cope, else JOURNAL_DATA
3588                  */
3589                 if (jbd2_journal_check_available_features
3590                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3591                         set_opt(sb, ORDERED_DATA);
3592                 else
3593                         set_opt(sb, JOURNAL_DATA);
3594                 break;
3595
3596         case EXT4_MOUNT_ORDERED_DATA:
3597         case EXT4_MOUNT_WRITEBACK_DATA:
3598                 if (!jbd2_journal_check_available_features
3599                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3600                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3601                                "requested data journaling mode");
3602                         goto failed_mount_wq;
3603                 }
3604         default:
3605                 break;
3606         }
3607         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3608
3609         /*
3610          * The journal may have updated the bg summary counts, so we
3611          * need to update the global counters.
3612          */
3613         percpu_counter_set(&sbi->s_freeblocks_counter,
3614                            ext4_count_free_blocks(sb));
3615         percpu_counter_set(&sbi->s_freeinodes_counter,
3616                            ext4_count_free_inodes(sb));
3617         percpu_counter_set(&sbi->s_dirs_counter,
3618                            ext4_count_dirs(sb));
3619         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3620
3621 no_journal:
3622         /*
3623          * The maximum number of concurrent works can be high and
3624          * concurrency isn't really necessary.  Limit it to 1.
3625          */
3626         EXT4_SB(sb)->dio_unwritten_wq =
3627                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3628         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3629                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3630                 goto failed_mount_wq;
3631         }
3632
3633         /*
3634          * The jbd2_journal_load will have done any necessary log recovery,
3635          * so we can safely mount the rest of the filesystem now.
3636          */
3637
3638         root = ext4_iget(sb, EXT4_ROOT_INO);
3639         if (IS_ERR(root)) {
3640                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3641                 ret = PTR_ERR(root);
3642                 root = NULL;
3643                 goto failed_mount4;
3644         }
3645         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3646                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3647                 goto failed_mount4;
3648         }
3649         sb->s_root = d_alloc_root(root);
3650         if (!sb->s_root) {
3651                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3652                 ret = -ENOMEM;
3653                 goto failed_mount4;
3654         }
3655
3656         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3657
3658         /* determine the minimum size of new large inodes, if present */
3659         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3660                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3661                                                      EXT4_GOOD_OLD_INODE_SIZE;
3662                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3663                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3664                         if (sbi->s_want_extra_isize <
3665                             le16_to_cpu(es->s_want_extra_isize))
3666                                 sbi->s_want_extra_isize =
3667                                         le16_to_cpu(es->s_want_extra_isize);
3668                         if (sbi->s_want_extra_isize <
3669                             le16_to_cpu(es->s_min_extra_isize))
3670                                 sbi->s_want_extra_isize =
3671                                         le16_to_cpu(es->s_min_extra_isize);
3672                 }
3673         }
3674         /* Check if enough inode space is available */
3675         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3676                                                         sbi->s_inode_size) {
3677                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3678                                                        EXT4_GOOD_OLD_INODE_SIZE;
3679                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3680                          "available");
3681         }
3682
3683         if (test_opt(sb, DELALLOC) &&
3684             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3685                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3686                          "requested data journaling mode");
3687                 clear_opt(sb, DELALLOC);
3688         }
3689         if (test_opt(sb, DIOREAD_NOLOCK)) {
3690                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3691                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3692                                 "option - requested data journaling mode");
3693                         clear_opt(sb, DIOREAD_NOLOCK);
3694                 }
3695                 if (sb->s_blocksize < PAGE_SIZE) {
3696                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3697                                 "option - block size is too small");
3698                         clear_opt(sb, DIOREAD_NOLOCK);
3699                 }
3700         }
3701
3702         err = ext4_setup_system_zone(sb);
3703         if (err) {
3704                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3705                          "zone (%d)", err);
3706                 goto failed_mount4;
3707         }
3708
3709         ext4_ext_init(sb);
3710         err = ext4_mb_init(sb, needs_recovery);
3711         if (err) {
3712                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3713                          err);
3714                 goto failed_mount4;
3715         }
3716
3717         err = ext4_register_li_request(sb, first_not_zeroed);
3718         if (err)
3719                 goto failed_mount4;
3720
3721         sbi->s_kobj.kset = ext4_kset;
3722         init_completion(&sbi->s_kobj_unregister);
3723         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3724                                    "%s", sb->s_id);
3725         if (err) {
3726                 ext4_mb_release(sb);
3727                 ext4_ext_release(sb);
3728                 goto failed_mount4;
3729         };
3730
3731         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3732         ext4_orphan_cleanup(sb, es);
3733         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3734         if (needs_recovery) {
3735                 ext4_msg(sb, KERN_INFO, "recovery complete");
3736                 ext4_mark_recovery_complete(sb, es);
3737         }
3738         if (EXT4_SB(sb)->s_journal) {
3739                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3740                         descr = " journalled data mode";
3741                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3742                         descr = " ordered data mode";
3743                 else
3744                         descr = " writeback data mode";
3745         } else
3746                 descr = "out journal";
3747
3748         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3749                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3750                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3751
3752         if (es->s_error_count)
3753                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3754
3755         kfree(orig_data);
3756         return 0;
3757
3758 cantfind_ext4:
3759         if (!silent)
3760                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3761         goto failed_mount;
3762
3763 failed_mount4:
3764         iput(root);
3765         sb->s_root = NULL;
3766         ext4_msg(sb, KERN_ERR, "mount failed");
3767         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3768 failed_mount_wq:
3769         ext4_release_system_zone(sb);
3770         if (sbi->s_journal) {
3771                 jbd2_journal_destroy(sbi->s_journal);
3772                 sbi->s_journal = NULL;
3773         }
3774 failed_mount3:
3775         del_timer(&sbi->s_err_report);
3776         if (sbi->s_flex_groups)
3777                 ext4_kvfree(sbi->s_flex_groups);
3778         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3779         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3780         percpu_counter_destroy(&sbi->s_dirs_counter);
3781         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3782         if (sbi->s_mmp_tsk)
3783                 kthread_stop(sbi->s_mmp_tsk);
3784 failed_mount2:
3785         for (i = 0; i < db_count; i++)
3786                 brelse(sbi->s_group_desc[i]);
3787         ext4_kvfree(sbi->s_group_desc);
3788 failed_mount:
3789         if (sbi->s_proc) {
3790                 remove_proc_entry(sb->s_id, ext4_proc_root);
3791         }
3792 #ifdef CONFIG_QUOTA
3793         for (i = 0; i < MAXQUOTAS; i++)
3794                 kfree(sbi->s_qf_names[i]);
3795 #endif
3796         ext4_blkdev_remove(sbi);
3797         brelse(bh);
3798 out_fail:
3799         sb->s_fs_info = NULL;
3800         kfree(sbi->s_blockgroup_lock);
3801         kfree(sbi);
3802 out_free_orig:
3803         kfree(orig_data);
3804         return ret;
3805 }
3806
3807 /*
3808  * Setup any per-fs journal parameters now.  We'll do this both on
3809  * initial mount, once the journal has been initialised but before we've
3810  * done any recovery; and again on any subsequent remount.
3811  */
3812 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3813 {
3814         struct ext4_sb_info *sbi = EXT4_SB(sb);
3815
3816         journal->j_commit_interval = sbi->s_commit_interval;
3817         journal->j_min_batch_time = sbi->s_min_batch_time;
3818         journal->j_max_batch_time = sbi->s_max_batch_time;
3819
3820         write_lock(&journal->j_state_lock);
3821         if (test_opt(sb, BARRIER))
3822                 journal->j_flags |= JBD2_BARRIER;
3823         else
3824                 journal->j_flags &= ~JBD2_BARRIER;
3825         if (test_opt(sb, DATA_ERR_ABORT))
3826                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3827         else
3828                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3829         write_unlock(&journal->j_state_lock);
3830 }
3831
3832 static journal_t *ext4_get_journal(struct super_block *sb,
3833                                    unsigned int journal_inum)
3834 {
3835         struct inode *journal_inode;
3836         journal_t *journal;
3837
3838         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3839
3840         /* First, test for the existence of a valid inode on disk.  Bad
3841          * things happen if we iget() an unused inode, as the subsequent
3842          * iput() will try to delete it. */
3843
3844         journal_inode = ext4_iget(sb, journal_inum);
3845         if (IS_ERR(journal_inode)) {
3846                 ext4_msg(sb, KERN_ERR, "no journal found");
3847                 return NULL;
3848         }
3849         if (!journal_inode->i_nlink) {
3850                 make_bad_inode(journal_inode);
3851                 iput(journal_inode);
3852                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3853                 return NULL;
3854         }
3855
3856         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3857                   journal_inode, journal_inode->i_size);
3858         if (!S_ISREG(journal_inode->i_mode)) {
3859                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3860                 iput(journal_inode);
3861                 return NULL;
3862         }
3863
3864         journal = jbd2_journal_init_inode(journal_inode);
3865         if (!journal) {
3866                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3867                 iput(journal_inode);
3868                 return NULL;
3869         }
3870         journal->j_private = sb;
3871         ext4_init_journal_params(sb, journal);
3872         return journal;
3873 }
3874
3875 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3876                                        dev_t j_dev)
3877 {
3878         struct buffer_head *bh;
3879         journal_t *journal;
3880         ext4_fsblk_t start;
3881         ext4_fsblk_t len;
3882         int hblock, blocksize;
3883         ext4_fsblk_t sb_block;
3884         unsigned long offset;
3885         struct ext4_super_block *es;
3886         struct block_device *bdev;
3887
3888         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3889
3890         bdev = ext4_blkdev_get(j_dev, sb);
3891         if (bdev == NULL)
3892                 return NULL;
3893
3894         blocksize = sb->s_blocksize;
3895         hblock = bdev_logical_block_size(bdev);
3896         if (blocksize < hblock) {
3897                 ext4_msg(sb, KERN_ERR,
3898                         "blocksize too small for journal device");
3899                 goto out_bdev;
3900         }
3901
3902         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3903         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3904         set_blocksize(bdev, blocksize);
3905         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3906                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3907                        "external journal");
3908                 goto out_bdev;
3909         }
3910
3911         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3912         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3913             !(le32_to_cpu(es->s_feature_incompat) &
3914               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3915                 ext4_msg(sb, KERN_ERR, "external journal has "
3916                                         "bad superblock");
3917                 brelse(bh);
3918                 goto out_bdev;
3919         }
3920
3921         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3922                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3923                 brelse(bh);
3924                 goto out_bdev;
3925         }
3926
3927         len = ext4_blocks_count(es);
3928         start = sb_block + 1;
3929         brelse(bh);     /* we're done with the superblock */
3930
3931         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3932                                         start, len, blocksize);
3933         if (!journal) {
3934                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3935                 goto out_bdev;
3936         }
3937         journal->j_private = sb;
3938         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3939         wait_on_buffer(journal->j_sb_buffer);
3940         if (!buffer_uptodate(journal->j_sb_buffer)) {
3941                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3942                 goto out_journal;
3943         }
3944         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3945                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3946                                         "user (unsupported) - %d",
3947                         be32_to_cpu(journal->j_superblock->s_nr_users));
3948                 goto out_journal;
3949         }
3950         EXT4_SB(sb)->journal_bdev = bdev;
3951         ext4_init_journal_params(sb, journal);
3952         return journal;
3953
3954 out_journal:
3955         jbd2_journal_destroy(journal);
3956 out_bdev:
3957         ext4_blkdev_put(bdev);
3958         return NULL;
3959 }
3960
3961 static int ext4_load_journal(struct super_block *sb,
3962                              struct ext4_super_block *es,
3963                              unsigned long journal_devnum)
3964 {
3965         journal_t *journal;
3966         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3967         dev_t journal_dev;
3968         int err = 0;
3969         int really_read_only;
3970
3971         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3972
3973         if (journal_devnum &&
3974             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3975                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3976                         "numbers have changed");
3977                 journal_dev = new_decode_dev(journal_devnum);
3978         } else
3979                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3980
3981         really_read_only = bdev_read_only(sb->s_bdev);
3982
3983         /*
3984          * Are we loading a blank journal or performing recovery after a
3985          * crash?  For recovery, we need to check in advance whether we
3986          * can get read-write access to the device.
3987          */
3988         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3989                 if (sb->s_flags & MS_RDONLY) {
3990                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3991                                         "required on readonly filesystem");
3992                         if (really_read_only) {
3993                                 ext4_msg(sb, KERN_ERR, "write access "
3994                                         "unavailable, cannot proceed");
3995                                 return -EROFS;
3996                         }
3997                         ext4_msg(sb, KERN_INFO, "write access will "
3998                                "be enabled during recovery");
3999                 }
4000         }
4001
4002         if (journal_inum && journal_dev) {
4003                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4004                        "and inode journals!");
4005                 return -EINVAL;
4006         }
4007
4008         if (journal_inum) {
4009                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4010                         return -EINVAL;
4011         } else {
4012                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4013                         return -EINVAL;
4014         }
4015
4016         if (!(journal->j_flags & JBD2_BARRIER))
4017                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4018
4019         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4020                 err = jbd2_journal_update_format(journal);
4021                 if (err)  {
4022                         ext4_msg(sb, KERN_ERR, "error updating journal");
4023                         jbd2_journal_destroy(journal);
4024                         return err;
4025                 }
4026         }
4027
4028         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4029                 err = jbd2_journal_wipe(journal, !really_read_only);
4030         if (!err) {
4031                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4032                 if (save)
4033                         memcpy(save, ((char *) es) +
4034                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4035                 err = jbd2_journal_load(journal);
4036                 if (save)
4037                         memcpy(((char *) es) + EXT4_S_ERR_START,
4038                                save, EXT4_S_ERR_LEN);
4039                 kfree(save);
4040         }
4041
4042         if (err) {
4043                 ext4_msg(sb, KERN_ERR, "error loading journal");
4044                 jbd2_journal_destroy(journal);
4045                 return err;
4046         }
4047
4048         EXT4_SB(sb)->s_journal = journal;
4049         ext4_clear_journal_err(sb, es);
4050
4051         if (!really_read_only && journal_devnum &&
4052             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4053                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4054
4055                 /* Make sure we flush the recovery flag to disk. */
4056                 ext4_commit_super(sb, 1);
4057         }
4058
4059         return 0;
4060 }
4061
4062 static int ext4_commit_super(struct super_block *sb, int sync)
4063 {
4064         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4065         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4066         int error = 0;
4067
4068         if (!sbh)
4069                 return error;
4070         if (buffer_write_io_error(sbh)) {
4071                 /*
4072                  * Oh, dear.  A previous attempt to write the
4073                  * superblock failed.  This could happen because the
4074                  * USB device was yanked out.  Or it could happen to
4075                  * be a transient write error and maybe the block will
4076                  * be remapped.  Nothing we can do but to retry the
4077                  * write and hope for the best.
4078                  */
4079                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4080                        "superblock detected");
4081                 clear_buffer_write_io_error(sbh);
4082                 set_buffer_uptodate(sbh);
4083         }
4084         /*
4085          * If the file system is mounted read-only, don't update the
4086          * superblock write time.  This avoids updating the superblock
4087          * write time when we are mounting the root file system
4088          * read/only but we need to replay the journal; at that point,
4089          * for people who are east of GMT and who make their clock
4090          * tick in localtime for Windows bug-for-bug compatibility,
4091          * the clock is set in the future, and this will cause e2fsck
4092          * to complain and force a full file system check.
4093          */
4094         if (!(sb->s_flags & MS_RDONLY))
4095                 es->s_wtime = cpu_to_le32(get_seconds());
4096         if (sb->s_bdev->bd_part)
4097                 es->s_kbytes_written =
4098                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4099                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4100                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4101         else
4102                 es->s_kbytes_written =
4103                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4104         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4105                                            &EXT4_SB(sb)->s_freeblocks_counter));
4106         es->s_free_inodes_count =
4107                 cpu_to_le32(percpu_counter_sum_positive(
4108                                 &EXT4_SB(sb)->s_freeinodes_counter));
4109         sb->s_dirt = 0;
4110         BUFFER_TRACE(sbh, "marking dirty");
4111         mark_buffer_dirty(sbh);
4112         if (sync) {
4113                 error = sync_dirty_buffer(sbh);
4114                 if (error)
4115                         return error;
4116
4117                 error = buffer_write_io_error(sbh);
4118                 if (error) {
4119                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4120                                "superblock");
4121                         clear_buffer_write_io_error(sbh);
4122                         set_buffer_uptodate(sbh);
4123                 }
4124         }
4125         return error;
4126 }
4127
4128 /*
4129  * Have we just finished recovery?  If so, and if we are mounting (or
4130  * remounting) the filesystem readonly, then we will end up with a
4131  * consistent fs on disk.  Record that fact.
4132  */
4133 static void ext4_mark_recovery_complete(struct super_block *sb,
4134                                         struct ext4_super_block *es)
4135 {
4136         journal_t *journal = EXT4_SB(sb)->s_journal;
4137
4138         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4139                 BUG_ON(journal != NULL);
4140                 return;
4141         }
4142         jbd2_journal_lock_updates(journal);
4143         if (jbd2_journal_flush(journal) < 0)
4144                 goto out;
4145
4146         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4147             sb->s_flags & MS_RDONLY) {
4148                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4149                 ext4_commit_super(sb, 1);
4150         }
4151
4152 out:
4153         jbd2_journal_unlock_updates(journal);
4154 }
4155
4156 /*
4157  * If we are mounting (or read-write remounting) a filesystem whose journal
4158  * has recorded an error from a previous lifetime, move that error to the
4159  * main filesystem now.
4160  */
4161 static void ext4_clear_journal_err(struct super_block *sb,
4162                                    struct ext4_super_block *es)
4163 {
4164         journal_t *journal;
4165         int j_errno;
4166         const char *errstr;
4167
4168         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4169
4170         journal = EXT4_SB(sb)->s_journal;
4171
4172         /*
4173          * Now check for any error status which may have been recorded in the
4174          * journal by a prior ext4_error() or ext4_abort()
4175          */
4176
4177         j_errno = jbd2_journal_errno(journal);
4178         if (j_errno) {
4179                 char nbuf[16];
4180
4181                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4182                 ext4_warning(sb, "Filesystem error recorded "
4183                              "from previous mount: %s", errstr);
4184                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4185
4186                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4187                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4188                 ext4_commit_super(sb, 1);
4189
4190                 jbd2_journal_clear_err(journal);
4191         }
4192 }
4193
4194 /*
4195  * Force the running and committing transactions to commit,
4196  * and wait on the commit.
4197  */
4198 int ext4_force_commit(struct super_block *sb)
4199 {
4200         journal_t *journal;
4201         int ret = 0;
4202
4203         if (sb->s_flags & MS_RDONLY)
4204                 return 0;
4205
4206         journal = EXT4_SB(sb)->s_journal;
4207         if (journal) {
4208                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4209                 ret = ext4_journal_force_commit(journal);
4210         }
4211
4212         return ret;
4213 }
4214
4215 static void ext4_write_super(struct super_block *sb)
4216 {
4217         lock_super(sb);
4218         ext4_commit_super(sb, 1);
4219         unlock_super(sb);
4220 }
4221
4222 static int ext4_sync_fs(struct super_block *sb, int wait)
4223 {
4224         int ret = 0;
4225         tid_t target;
4226         struct ext4_sb_info *sbi = EXT4_SB(sb);
4227
4228         trace_ext4_sync_fs(sb, wait);
4229         flush_workqueue(sbi->dio_unwritten_wq);
4230         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4231                 if (wait)
4232                         jbd2_log_wait_commit(sbi->s_journal, target);
4233         }
4234         return ret;
4235 }
4236
4237 /*
4238  * LVM calls this function before a (read-only) snapshot is created.  This
4239  * gives us a chance to flush the journal completely and mark the fs clean.
4240  *
4241  * Note that only this function cannot bring a filesystem to be in a clean
4242  * state independently, because ext4 prevents a new handle from being started
4243  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4244  * the upper layer.
4245  */
4246 static int ext4_freeze(struct super_block *sb)
4247 {
4248         int error = 0;
4249         journal_t *journal;
4250
4251         if (sb->s_flags & MS_RDONLY)
4252                 return 0;
4253
4254         journal = EXT4_SB(sb)->s_journal;
4255
4256         /* Now we set up the journal barrier. */
4257         jbd2_journal_lock_updates(journal);
4258
4259         /*
4260          * Don't clear the needs_recovery flag if we failed to flush
4261          * the journal.
4262          */
4263         error = jbd2_journal_flush(journal);
4264         if (error < 0)
4265                 goto out;
4266
4267         /* Journal blocked and flushed, clear needs_recovery flag. */
4268         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4269         error = ext4_commit_super(sb, 1);
4270 out:
4271         /* we rely on s_frozen to stop further updates */
4272         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4273         return error;
4274 }
4275
4276 /*
4277  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4278  * flag here, even though the filesystem is not technically dirty yet.
4279  */
4280 static int ext4_unfreeze(struct super_block *sb)
4281 {
4282         if (sb->s_flags & MS_RDONLY)
4283                 return 0;
4284
4285         lock_super(sb);
4286         /* Reset the needs_recovery flag before the fs is unlocked. */
4287         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4288         ext4_commit_super(sb, 1);
4289         unlock_super(sb);
4290         return 0;
4291 }
4292
4293 /*
4294  * Structure to save mount options for ext4_remount's benefit
4295  */
4296 struct ext4_mount_options {
4297         unsigned long s_mount_opt;
4298         unsigned long s_mount_opt2;
4299         uid_t s_resuid;
4300         gid_t s_resgid;
4301         unsigned long s_commit_interval;
4302         u32 s_min_batch_time, s_max_batch_time;
4303 #ifdef CONFIG_QUOTA
4304         int s_jquota_fmt;
4305         char *s_qf_names[MAXQUOTAS];
4306 #endif
4307 };
4308
4309 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4310 {
4311         struct ext4_super_block *es;
4312         struct ext4_sb_info *sbi = EXT4_SB(sb);
4313         ext4_fsblk_t n_blocks_count = 0;
4314         unsigned long old_sb_flags;
4315         struct ext4_mount_options old_opts;
4316         int enable_quota = 0;
4317         ext4_group_t g;
4318         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4319         int err = 0;
4320 #ifdef CONFIG_QUOTA
4321         int i;
4322 #endif
4323         char *orig_data = kstrdup(data, GFP_KERNEL);
4324
4325         /* Store the original options */
4326         lock_super(sb);
4327         old_sb_flags = sb->s_flags;
4328         old_opts.s_mount_opt = sbi->s_mount_opt;
4329         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4330         old_opts.s_resuid = sbi->s_resuid;
4331         old_opts.s_resgid = sbi->s_resgid;
4332         old_opts.s_commit_interval = sbi->s_commit_interval;
4333         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4334         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4335 #ifdef CONFIG_QUOTA
4336         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4337         for (i = 0; i < MAXQUOTAS; i++)
4338                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4339 #endif
4340         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4341                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4342
4343         /*
4344          * Allow the "check" option to be passed as a remount option.
4345          */
4346         if (!parse_options(data, sb, NULL, &journal_ioprio,
4347                            &n_blocks_count, 1)) {
4348                 err = -EINVAL;
4349                 goto restore_opts;
4350         }
4351
4352         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4353                 ext4_abort(sb, "Abort forced by user");
4354
4355         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4356                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4357
4358         es = sbi->s_es;
4359
4360         if (sbi->s_journal) {
4361                 ext4_init_journal_params(sb, sbi->s_journal);
4362                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4363         }
4364
4365         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4366                 n_blocks_count > ext4_blocks_count(es)) {
4367                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4368                         err = -EROFS;
4369                         goto restore_opts;
4370                 }
4371
4372                 if (*flags & MS_RDONLY) {
4373                         err = dquot_suspend(sb, -1);
4374                         if (err < 0)
4375                                 goto restore_opts;
4376
4377                         /*
4378                          * First of all, the unconditional stuff we have to do
4379                          * to disable replay of the journal when we next remount
4380                          */
4381                         sb->s_flags |= MS_RDONLY;
4382
4383                         /*
4384                          * OK, test if we are remounting a valid rw partition
4385                          * readonly, and if so set the rdonly flag and then
4386                          * mark the partition as valid again.
4387                          */
4388                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4389                             (sbi->s_mount_state & EXT4_VALID_FS))
4390                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4391
4392                         if (sbi->s_journal)
4393                                 ext4_mark_recovery_complete(sb, es);
4394                 } else {
4395                         /* Make sure we can mount this feature set readwrite */
4396                         if (!ext4_feature_set_ok(sb, 0)) {
4397                                 err = -EROFS;
4398                                 goto restore_opts;
4399                         }
4400                         /*
4401                          * Make sure the group descriptor checksums
4402                          * are sane.  If they aren't, refuse to remount r/w.
4403                          */
4404                         for (g = 0; g < sbi->s_groups_count; g++) {
4405                                 struct ext4_group_desc *gdp =
4406                                         ext4_get_group_desc(sb, g, NULL);
4407
4408                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4409                                         ext4_msg(sb, KERN_ERR,
4410                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4411                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4412                                                le16_to_cpu(gdp->bg_checksum));
4413                                         err = -EINVAL;
4414                                         goto restore_opts;
4415                                 }
4416                         }
4417
4418                         /*
4419                          * If we have an unprocessed orphan list hanging
4420                          * around from a previously readonly bdev mount,
4421                          * require a full umount/remount for now.
4422                          */
4423                         if (es->s_last_orphan) {
4424                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4425                                        "remount RDWR because of unprocessed "
4426                                        "orphan inode list.  Please "
4427                                        "umount/remount instead");
4428                                 err = -EINVAL;
4429                                 goto restore_opts;
4430                         }
4431
4432                         /*
4433                          * Mounting a RDONLY partition read-write, so reread
4434                          * and store the current valid flag.  (It may have
4435                          * been changed by e2fsck since we originally mounted
4436                          * the partition.)
4437                          */
4438                         if (sbi->s_journal)
4439                                 ext4_clear_journal_err(sb, es);
4440                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4441                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4442                                 goto restore_opts;
4443                         if (!ext4_setup_super(sb, es, 0))
4444                                 sb->s_flags &= ~MS_RDONLY;
4445                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4446                                                      EXT4_FEATURE_INCOMPAT_MMP))
4447                                 if (ext4_multi_mount_protect(sb,
4448                                                 le64_to_cpu(es->s_mmp_block))) {
4449                                         err = -EROFS;
4450                                         goto restore_opts;
4451                                 }
4452                         enable_quota = 1;
4453                 }
4454         }
4455
4456         /*
4457          * Reinitialize lazy itable initialization thread based on
4458          * current settings
4459          */
4460         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4461                 ext4_unregister_li_request(sb);
4462         else {
4463                 ext4_group_t first_not_zeroed;
4464                 first_not_zeroed = ext4_has_uninit_itable(sb);
4465                 ext4_register_li_request(sb, first_not_zeroed);
4466         }
4467
4468         ext4_setup_system_zone(sb);
4469         if (sbi->s_journal == NULL)
4470                 ext4_commit_super(sb, 1);
4471
4472 #ifdef CONFIG_QUOTA
4473         /* Release old quota file names */
4474         for (i = 0; i < MAXQUOTAS; i++)
4475                 if (old_opts.s_qf_names[i] &&
4476                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4477                         kfree(old_opts.s_qf_names[i]);
4478 #endif
4479         unlock_super(sb);
4480         if (enable_quota)
4481                 dquot_resume(sb, -1);
4482
4483         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4484         kfree(orig_data);
4485         return 0;
4486
4487 restore_opts:
4488         sb->s_flags = old_sb_flags;
4489         sbi->s_mount_opt = old_opts.s_mount_opt;
4490         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4491         sbi->s_resuid = old_opts.s_resuid;
4492         sbi->s_resgid = old_opts.s_resgid;
4493         sbi->s_commit_interval = old_opts.s_commit_interval;
4494         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4495         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4496 #ifdef CONFIG_QUOTA
4497         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4498         for (i = 0; i < MAXQUOTAS; i++) {
4499                 if (sbi->s_qf_names[i] &&
4500                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4501                         kfree(sbi->s_qf_names[i]);
4502                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4503         }
4504 #endif
4505         unlock_super(sb);
4506         kfree(orig_data);
4507         return err;
4508 }
4509
4510 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4511 {
4512         struct super_block *sb = dentry->d_sb;
4513         struct ext4_sb_info *sbi = EXT4_SB(sb);
4514         struct ext4_super_block *es = sbi->s_es;
4515         u64 fsid;
4516         s64 bfree;
4517
4518         if (test_opt(sb, MINIX_DF)) {
4519                 sbi->s_overhead_last = 0;
4520         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4521                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4522                 ext4_fsblk_t overhead = 0;
4523
4524                 /*
4525                  * Compute the overhead (FS structures).  This is constant
4526                  * for a given filesystem unless the number of block groups
4527                  * changes so we cache the previous value until it does.
4528                  */
4529
4530                 /*
4531                  * All of the blocks before first_data_block are
4532                  * overhead
4533                  */
4534                 overhead = le32_to_cpu(es->s_first_data_block);
4535
4536                 /*
4537                  * Add the overhead attributed to the superblock and
4538                  * block group descriptors.  If the sparse superblocks
4539                  * feature is turned on, then not all groups have this.
4540                  */
4541                 for (i = 0; i < ngroups; i++) {
4542                         overhead += ext4_bg_has_super(sb, i) +
4543                                 ext4_bg_num_gdb(sb, i);
4544                         cond_resched();
4545                 }
4546
4547                 /*
4548                  * Every block group has an inode bitmap, a block
4549                  * bitmap, and an inode table.
4550                  */
4551                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4552                 sbi->s_overhead_last = overhead;
4553                 smp_wmb();
4554                 sbi->s_blocks_last = ext4_blocks_count(es);
4555         }
4556
4557         buf->f_type = EXT4_SUPER_MAGIC;
4558         buf->f_bsize = sb->s_blocksize;
4559         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4560         bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4561                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4562         /* prevent underflow in case that few free space is available */
4563         buf->f_bfree = max_t(s64, bfree, 0);
4564         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4565         if (buf->f_bfree < ext4_r_blocks_count(es))
4566                 buf->f_bavail = 0;
4567         buf->f_files = le32_to_cpu(es->s_inodes_count);
4568         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4569         buf->f_namelen = EXT4_NAME_LEN;
4570         fsid = le64_to_cpup((void *)es->s_uuid) ^
4571                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4572         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4573         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4574
4575         return 0;
4576 }
4577
4578 /* Helper function for writing quotas on sync - we need to start transaction
4579  * before quota file is locked for write. Otherwise the are possible deadlocks:
4580  * Process 1                         Process 2
4581  * ext4_create()                     quota_sync()
4582  *   jbd2_journal_start()                  write_dquot()
4583  *   dquot_initialize()                         down(dqio_mutex)
4584  *     down(dqio_mutex)                    jbd2_journal_start()
4585  *
4586  */
4587
4588 #ifdef CONFIG_QUOTA
4589
4590 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4591 {
4592         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4593 }
4594
4595 static int ext4_write_dquot(struct dquot *dquot)
4596 {
4597         int ret, err;
4598         handle_t *handle;
4599         struct inode *inode;
4600
4601         inode = dquot_to_inode(dquot);
4602         handle = ext4_journal_start(inode,
4603                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4604         if (IS_ERR(handle))
4605                 return PTR_ERR(handle);
4606         ret = dquot_commit(dquot);
4607         err = ext4_journal_stop(handle);
4608         if (!ret)
4609                 ret = err;
4610         return ret;
4611 }
4612
4613 static int ext4_acquire_dquot(struct dquot *dquot)
4614 {
4615         int ret, err;
4616         handle_t *handle;
4617
4618         handle = ext4_journal_start(dquot_to_inode(dquot),
4619                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4620         if (IS_ERR(handle))
4621                 return PTR_ERR(handle);
4622         ret = dquot_acquire(dquot);
4623         err = ext4_journal_stop(handle);
4624         if (!ret)
4625                 ret = err;
4626         return ret;
4627 }
4628
4629 static int ext4_release_dquot(struct dquot *dquot)
4630 {
4631         int ret, err;
4632         handle_t *handle;
4633
4634         handle = ext4_journal_start(dquot_to_inode(dquot),
4635                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4636         if (IS_ERR(handle)) {
4637                 /* Release dquot anyway to avoid endless cycle in dqput() */
4638                 dquot_release(dquot);
4639                 return PTR_ERR(handle);
4640         }
4641         ret = dquot_release(dquot);
4642         err = ext4_journal_stop(handle);
4643         if (!ret)
4644                 ret = err;
4645         return ret;
4646 }
4647
4648 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4649 {
4650         /* Are we journaling quotas? */
4651         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4652             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4653                 dquot_mark_dquot_dirty(dquot);
4654                 return ext4_write_dquot(dquot);
4655         } else {
4656                 return dquot_mark_dquot_dirty(dquot);
4657         }
4658 }
4659
4660 static int ext4_write_info(struct super_block *sb, int type)
4661 {
4662         int ret, err;
4663         handle_t *handle;
4664
4665         /* Data block + inode block */
4666         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4667         if (IS_ERR(handle))
4668                 return PTR_ERR(handle);
4669         ret = dquot_commit_info(sb, type);
4670         err = ext4_journal_stop(handle);
4671         if (!ret)
4672                 ret = err;
4673         return ret;
4674 }
4675
4676 /*
4677  * Turn on quotas during mount time - we need to find
4678  * the quota file and such...
4679  */
4680 static int ext4_quota_on_mount(struct super_block *sb, int type)
4681 {
4682         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4683                                         EXT4_SB(sb)->s_jquota_fmt, type);
4684 }
4685
4686 /*
4687  * Standard function to be called on quota_on
4688  */
4689 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4690                          struct path *path)
4691 {
4692         int err;
4693
4694         if (!test_opt(sb, QUOTA))
4695                 return -EINVAL;
4696
4697         /* Quotafile not on the same filesystem? */
4698         if (path->mnt->mnt_sb != sb)
4699                 return -EXDEV;
4700         /* Journaling quota? */
4701         if (EXT4_SB(sb)->s_qf_names[type]) {
4702                 /* Quotafile not in fs root? */
4703                 if (path->dentry->d_parent != sb->s_root)
4704                         ext4_msg(sb, KERN_WARNING,
4705                                 "Quota file not on filesystem root. "
4706                                 "Journaled quota will not work");
4707         }
4708
4709         /*
4710          * When we journal data on quota file, we have to flush journal to see
4711          * all updates to the file when we bypass pagecache...
4712          */
4713         if (EXT4_SB(sb)->s_journal &&
4714             ext4_should_journal_data(path->dentry->d_inode)) {
4715                 /*
4716                  * We don't need to lock updates but journal_flush() could
4717                  * otherwise be livelocked...
4718                  */
4719                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4720                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4721                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4722                 if (err)
4723                         return err;
4724         }
4725
4726         return dquot_quota_on(sb, type, format_id, path);
4727 }
4728
4729 static int ext4_quota_off(struct super_block *sb, int type)
4730 {
4731         struct inode *inode = sb_dqopt(sb)->files[type];
4732         handle_t *handle;
4733
4734         /* Force all delayed allocation blocks to be allocated.
4735          * Caller already holds s_umount sem */
4736         if (test_opt(sb, DELALLOC))
4737                 sync_filesystem(sb);
4738
4739         if (!inode)
4740                 goto out;
4741
4742         /* Update modification times of quota files when userspace can
4743          * start looking at them */
4744         handle = ext4_journal_start(inode, 1);
4745         if (IS_ERR(handle))
4746                 goto out;
4747         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4748         ext4_mark_inode_dirty(handle, inode);
4749         ext4_journal_stop(handle);
4750
4751 out:
4752         return dquot_quota_off(sb, type);
4753 }
4754
4755 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4756  * acquiring the locks... As quota files are never truncated and quota code
4757  * itself serializes the operations (and no one else should touch the files)
4758  * we don't have to be afraid of races */
4759 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4760                                size_t len, loff_t off)
4761 {
4762         struct inode *inode = sb_dqopt(sb)->files[type];
4763         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4764         int err = 0;
4765         int offset = off & (sb->s_blocksize - 1);
4766         int tocopy;
4767         size_t toread;
4768         struct buffer_head *bh;
4769         loff_t i_size = i_size_read(inode);
4770
4771         if (off > i_size)
4772                 return 0;
4773         if (off+len > i_size)
4774                 len = i_size-off;
4775         toread = len;
4776         while (toread > 0) {
4777                 tocopy = sb->s_blocksize - offset < toread ?
4778                                 sb->s_blocksize - offset : toread;
4779                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4780                 if (err)
4781                         return err;
4782                 if (!bh)        /* A hole? */
4783                         memset(data, 0, tocopy);
4784                 else
4785                         memcpy(data, bh->b_data+offset, tocopy);
4786                 brelse(bh);
4787                 offset = 0;
4788                 toread -= tocopy;
4789                 data += tocopy;
4790                 blk++;
4791         }
4792         return len;
4793 }
4794
4795 /* Write to quotafile (we know the transaction is already started and has
4796  * enough credits) */
4797 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4798                                 const char *data, size_t len, loff_t off)
4799 {
4800         struct inode *inode = sb_dqopt(sb)->files[type];
4801         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4802         int err = 0;
4803         int offset = off & (sb->s_blocksize - 1);
4804         struct buffer_head *bh;
4805         handle_t *handle = journal_current_handle();
4806
4807         if (EXT4_SB(sb)->s_journal && !handle) {
4808                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4809                         " cancelled because transaction is not started",
4810                         (unsigned long long)off, (unsigned long long)len);
4811                 return -EIO;
4812         }
4813         /*
4814          * Since we account only one data block in transaction credits,
4815          * then it is impossible to cross a block boundary.
4816          */
4817         if (sb->s_blocksize - offset < len) {
4818                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4819                         " cancelled because not block aligned",
4820                         (unsigned long long)off, (unsigned long long)len);
4821                 return -EIO;
4822         }
4823
4824         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4825         bh = ext4_bread(handle, inode, blk, 1, &err);
4826         if (!bh)
4827                 goto out;
4828         err = ext4_journal_get_write_access(handle, bh);
4829         if (err) {
4830                 brelse(bh);
4831                 goto out;
4832         }
4833         lock_buffer(bh);
4834         memcpy(bh->b_data+offset, data, len);
4835         flush_dcache_page(bh->b_page);
4836         unlock_buffer(bh);
4837         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4838         brelse(bh);
4839 out:
4840         if (err) {
4841                 mutex_unlock(&inode->i_mutex);
4842                 return err;
4843         }
4844         if (inode->i_size < off + len) {
4845                 i_size_write(inode, off + len);
4846                 EXT4_I(inode)->i_disksize = inode->i_size;
4847                 ext4_mark_inode_dirty(handle, inode);
4848         }
4849         mutex_unlock(&inode->i_mutex);
4850         return len;
4851 }
4852
4853 #endif
4854
4855 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4856                        const char *dev_name, void *data)
4857 {
4858         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4859 }
4860
4861 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4862 static inline void register_as_ext2(void)
4863 {
4864         int err = register_filesystem(&ext2_fs_type);
4865         if (err)
4866                 printk(KERN_WARNING
4867                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4868 }
4869
4870 static inline void unregister_as_ext2(void)
4871 {
4872         unregister_filesystem(&ext2_fs_type);
4873 }
4874
4875 static inline int ext2_feature_set_ok(struct super_block *sb)
4876 {
4877         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4878                 return 0;
4879         if (sb->s_flags & MS_RDONLY)
4880                 return 1;
4881         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4882                 return 0;
4883         return 1;
4884 }
4885 MODULE_ALIAS("ext2");
4886 #else
4887 static inline void register_as_ext2(void) { }
4888 static inline void unregister_as_ext2(void) { }
4889 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4890 #endif
4891
4892 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4893 static inline void register_as_ext3(void)
4894 {
4895         int err = register_filesystem(&ext3_fs_type);
4896         if (err)
4897                 printk(KERN_WARNING
4898                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4899 }
4900
4901 static inline void unregister_as_ext3(void)
4902 {
4903         unregister_filesystem(&ext3_fs_type);
4904 }
4905
4906 static inline int ext3_feature_set_ok(struct super_block *sb)
4907 {
4908         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4909                 return 0;
4910         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4911                 return 0;
4912         if (sb->s_flags & MS_RDONLY)
4913                 return 1;
4914         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4915                 return 0;
4916         return 1;
4917 }
4918 MODULE_ALIAS("ext3");
4919 #else
4920 static inline void register_as_ext3(void) { }
4921 static inline void unregister_as_ext3(void) { }
4922 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4923 #endif
4924
4925 static struct file_system_type ext4_fs_type = {
4926         .owner          = THIS_MODULE,
4927         .name           = "ext4",
4928         .mount          = ext4_mount,
4929         .kill_sb        = kill_block_super,
4930         .fs_flags       = FS_REQUIRES_DEV,
4931 };
4932
4933 static int __init ext4_init_feat_adverts(void)
4934 {
4935         struct ext4_features *ef;
4936         int ret = -ENOMEM;
4937
4938         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4939         if (!ef)
4940                 goto out;
4941
4942         ef->f_kobj.kset = ext4_kset;
4943         init_completion(&ef->f_kobj_unregister);
4944         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4945                                    "features");
4946         if (ret) {
4947                 kfree(ef);
4948                 goto out;
4949         }
4950
4951         ext4_feat = ef;
4952         ret = 0;
4953 out:
4954         return ret;
4955 }
4956
4957 static void ext4_exit_feat_adverts(void)
4958 {
4959         kobject_put(&ext4_feat->f_kobj);
4960         wait_for_completion(&ext4_feat->f_kobj_unregister);
4961         kfree(ext4_feat);
4962 }
4963
4964 /* Shared across all ext4 file systems */
4965 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4966 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4967
4968 static int __init ext4_init_fs(void)
4969 {
4970         int i, err;
4971
4972         ext4_check_flag_values();
4973
4974         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4975                 mutex_init(&ext4__aio_mutex[i]);
4976                 init_waitqueue_head(&ext4__ioend_wq[i]);
4977         }
4978
4979         err = ext4_init_pageio();
4980         if (err)
4981                 return err;
4982         err = ext4_init_system_zone();
4983         if (err)
4984                 goto out7;
4985         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4986         if (!ext4_kset)
4987                 goto out6;
4988         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4989         if (!ext4_proc_root)
4990                 goto out5;
4991
4992         err = ext4_init_feat_adverts();
4993         if (err)
4994                 goto out4;
4995
4996         err = ext4_init_mballoc();
4997         if (err)
4998                 goto out3;
4999
5000         err = ext4_init_xattr();
5001         if (err)
5002                 goto out2;
5003         err = init_inodecache();
5004         if (err)
5005                 goto out1;
5006         register_as_ext3();
5007         register_as_ext2();
5008         err = register_filesystem(&ext4_fs_type);
5009         if (err)
5010                 goto out;
5011
5012         ext4_li_info = NULL;
5013         mutex_init(&ext4_li_mtx);
5014         return 0;
5015 out:
5016         unregister_as_ext2();
5017         unregister_as_ext3();
5018         destroy_inodecache();
5019 out1:
5020         ext4_exit_xattr();
5021 out2:
5022         ext4_exit_mballoc();
5023 out3:
5024         ext4_exit_feat_adverts();
5025 out4:
5026         remove_proc_entry("fs/ext4", NULL);
5027 out5:
5028         kset_unregister(ext4_kset);
5029 out6:
5030         ext4_exit_system_zone();
5031 out7:
5032         ext4_exit_pageio();
5033         return err;
5034 }
5035
5036 static void __exit ext4_exit_fs(void)
5037 {
5038         ext4_destroy_lazyinit_thread();
5039         unregister_as_ext2();
5040         unregister_as_ext3();
5041         unregister_filesystem(&ext4_fs_type);
5042         destroy_inodecache();
5043         ext4_exit_xattr();
5044         ext4_exit_mballoc();
5045         ext4_exit_feat_adverts();
5046         remove_proc_entry("fs/ext4", NULL);
5047         kset_unregister(ext4_kset);
5048         ext4_exit_system_zone();
5049         ext4_exit_pageio();
5050 }
5051
5052 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5053 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5054 MODULE_LICENSE("GPL");
5055 module_init(ext4_init_fs)
5056 module_exit(ext4_exit_fs)