fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_jbd2.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "mballoc.h"
  51
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/ext4.h>
  54
  55 static struct proc_dir_entry *ext4_proc_root;
  56 static struct kset *ext4_kset;
  57 static struct ext4_lazy_init *ext4_li_info;
  58 static struct mutex ext4_li_mtx;
  59 static struct ext4_features *ext4_feat;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static const char *ext4_decode_error(struct super_block *sb, int errno,
  70                                      char nbuf[16]);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static void ext4_write_super(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  84 static struct file_system_type ext3_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext3",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  92 #else
  93 #define IS_EXT3_SB(sb) (0)
  94 #endif
  95
  96 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  97                                struct ext4_group_desc *bg)
  98 {
  99         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 100                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 101                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 102 }
 103
 104 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 105                                struct ext4_group_desc *bg)
 106 {
 107         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 108                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 109                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 110 }
 111
 112 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 113                               struct ext4_group_desc *bg)
 114 {
 115         return le32_to_cpu(bg->bg_inode_table_lo) |
 116                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 117                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 118 }
 119
 120 __u32 ext4_free_blks_count(struct super_block *sb,
 121                               struct ext4_group_desc *bg)
 122 {
 123         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 124                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 125                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 126 }
 127
 128 __u32 ext4_free_inodes_count(struct super_block *sb,
 129                               struct ext4_group_desc *bg)
 130 {
 131         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 132                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 133                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 134 }
 135
 136 __u32 ext4_used_dirs_count(struct super_block *sb,
 137                               struct ext4_group_desc *bg)
 138 {
 139         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 140                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 141                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 142 }
 143
 144 __u32 ext4_itable_unused_count(struct super_block *sb,
 145                               struct ext4_group_desc *bg)
 146 {
 147         return le16_to_cpu(bg->bg_itable_unused_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 150 }
 151
 152 void ext4_block_bitmap_set(struct super_block *sb,
 153                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 154 {
 155         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 156         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 157                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 158 }
 159
 160 void ext4_inode_bitmap_set(struct super_block *sb,
 161                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 162 {
 163         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 164         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 165                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 166 }
 167
 168 void ext4_inode_table_set(struct super_block *sb,
 169                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 170 {
 171         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 172         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 173                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 174 }
 175
 176 void ext4_free_blks_set(struct super_block *sb,
 177                           struct ext4_group_desc *bg, __u32 count)
 178 {
 179         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 180         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 181                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 182 }
 183
 184 void ext4_free_inodes_set(struct super_block *sb,
 185                           struct ext4_group_desc *bg, __u32 count)
 186 {
 187         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 188         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 189                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 190 }
 191
 192 void ext4_used_dirs_set(struct super_block *sb,
 193                           struct ext4_group_desc *bg, __u32 count)
 194 {
 195         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 196         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 197                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 198 }
 199
 200 void ext4_itable_unused_set(struct super_block *sb,
 201                           struct ext4_group_desc *bg, __u32 count)
 202 {
 203         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 206 }
 207
 208
 209 /* Just increment the non-pointer handle value */
 210 static handle_t *ext4_get_nojournal(void)
 211 {
 212         handle_t *handle = current->journal_info;
 213         unsigned long ref_cnt = (unsigned long)handle;
 214
 215         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 216
 217         ref_cnt++;
 218         handle = (handle_t *)ref_cnt;
 219
 220         current->journal_info = handle;
 221         return handle;
 222 }
 223
 224
 225 /* Decrement the non-pointer handle value */
 226 static void ext4_put_nojournal(handle_t *handle)
 227 {
 228         unsigned long ref_cnt = (unsigned long)handle;
 229
 230         BUG_ON(ref_cnt == 0);
 231
 232         ref_cnt--;
 233         handle = (handle_t *)ref_cnt;
 234
 235         current->journal_info = handle;
 236 }
 237
 238 /*
 239  * Wrappers for jbd2_journal_start/end.
 240  *
 241  * The only special thing we need to do here is to make sure that all
 242  * journal_end calls result in the superblock being marked dirty, so
 243  * that sync() will call the filesystem's write_super callback if
 244  * appropriate.
 245  *
 246  * To avoid j_barrier hold in userspace when a user calls freeze(),
 247  * ext4 prevents a new handle from being started by s_frozen, which
 248  * is in an upper layer.
 249  */
 250 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 251 {
 252         journal_t *journal;
 253         handle_t  *handle;
 254
 255         if (sb->s_flags & MS_RDONLY)
 256                 return ERR_PTR(-EROFS);
 257
 258         journal = EXT4_SB(sb)->s_journal;
 259         handle = ext4_journal_current_handle();
 260
 261         /*
 262          * If a handle has been started, it should be allowed to
 263          * finish, otherwise deadlock could happen between freeze
 264          * and others(e.g. truncate) due to the restart of the
 265          * journal handle if the filesystem is forzen and active
 266          * handles are not stopped.
 267          */
 268         if (!handle)
 269                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 270
 271         if (!journal)
 272                 return ext4_get_nojournal();
 273         /*
 274          * Special case here: if the journal has aborted behind our
 275          * backs (eg. EIO in the commit thread), then we still need to
 276          * take the FS itself readonly cleanly.
 277          */
 278         if (is_journal_aborted(journal)) {
 279                 ext4_abort(sb, "Detected aborted journal");
 280                 return ERR_PTR(-EROFS);
 281         }
 282         return jbd2_journal_start(journal, nblocks);
 283 }
 284
 285 /*
 286  * The only special thing we need to do here is to make sure that all
 287  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 288  * that sync() will call the filesystem's write_super callback if
 289  * appropriate.
 290  */
 291 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 292 {
 293         struct super_block *sb;
 294         int err;
 295         int rc;
 296
 297         if (!ext4_handle_valid(handle)) {
 298                 ext4_put_nojournal(handle);
 299                 return 0;
 300         }
 301         sb = handle->h_transaction->t_journal->j_private;
 302         err = handle->h_err;
 303         rc = jbd2_journal_stop(handle);
 304
 305         if (!err)
 306                 err = rc;
 307         if (err)
 308                 __ext4_std_error(sb, where, line, err);
 309         return err;
 310 }
 311
 312 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 313                                const char *err_fn, struct buffer_head *bh,
 314                                handle_t *handle, int err)
 315 {
 316         char nbuf[16];
 317         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 318
 319         BUG_ON(!ext4_handle_valid(handle));
 320
 321         if (bh)
 322                 BUFFER_TRACE(bh, "abort");
 323
 324         if (!handle->h_err)
 325                 handle->h_err = err;
 326
 327         if (is_handle_aborted(handle))
 328                 return;
 329
 330         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 331                caller, line, errstr, err_fn);
 332
 333         jbd2_journal_abort_handle(handle);
 334 }
 335
 336 static void __save_error_info(struct super_block *sb, const char *func,
 337                             unsigned int line)
 338 {
 339         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 340
 341         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 342         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 343         es->s_last_error_time = cpu_to_le32(get_seconds());
 344         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 345         es->s_last_error_line = cpu_to_le32(line);
 346         if (!es->s_first_error_time) {
 347                 es->s_first_error_time = es->s_last_error_time;
 348                 strncpy(es->s_first_error_func, func,
 349                         sizeof(es->s_first_error_func));
 350                 es->s_first_error_line = cpu_to_le32(line);
 351                 es->s_first_error_ino = es->s_last_error_ino;
 352                 es->s_first_error_block = es->s_last_error_block;
 353         }
 354         /*
 355          * Start the daily error reporting function if it hasn't been
 356          * started already
 357          */
 358         if (!es->s_error_count)
 359                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 360         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 361 }
 362
 363 static void save_error_info(struct super_block *sb, const char *func,
 364                             unsigned int line)
 365 {
 366         __save_error_info(sb, func, line);
 367         ext4_commit_super(sb, 1);
 368 }
 369
 370
 371 /* Deal with the reporting of failure conditions on a filesystem such as
 372  * inconsistencies detected or read IO failures.
 373  *
 374  * On ext2, we can store the error state of the filesystem in the
 375  * superblock.  That is not possible on ext4, because we may have other
 376  * write ordering constraints on the superblock which prevent us from
 377  * writing it out straight away; and given that the journal is about to
 378  * be aborted, we can't rely on the current, or future, transactions to
 379  * write out the superblock safely.
 380  *
 381  * We'll just use the jbd2_journal_abort() error code to record an error in
 382  * the journal instead.  On recovery, the journal will complain about
 383  * that error until we've noted it down and cleared it.
 384  */
 385
 386 static void ext4_handle_error(struct super_block *sb)
 387 {
 388         if (sb->s_flags & MS_RDONLY)
 389                 return;
 390
 391         if (!test_opt(sb, ERRORS_CONT)) {
 392                 journal_t *journal = EXT4_SB(sb)->s_journal;
 393
 394                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 395                 if (journal)
 396                         jbd2_journal_abort(journal, -EIO);
 397         }
 398         if (test_opt(sb, ERRORS_RO)) {
 399                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 400                 sb->s_flags |= MS_RDONLY;
 401         }
 402         if (test_opt(sb, ERRORS_PANIC))
 403                 panic("EXT4-fs (device %s): panic forced after error\n",
 404                         sb->s_id);
 405 }
 406
 407 void __ext4_error(struct super_block *sb, const char *function,
 408                   unsigned int line, const char *fmt, ...)
 409 {
 410         struct va_format vaf;
 411         va_list args;
 412
 413         va_start(args, fmt);
 414         vaf.fmt = fmt;
 415         vaf.va = &args;
 416         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 417                sb->s_id, function, line, current->comm, &vaf);
 418         va_end(args);
 419
 420         ext4_handle_error(sb);
 421 }
 422
 423 void ext4_error_inode(struct inode *inode, const char *function,
 424                       unsigned int line, ext4_fsblk_t block,
 425                       const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         save_error_info(inode->i_sb, function, line);
 434         va_start(args, fmt);
 435         vaf.fmt = fmt;
 436         vaf.va = &args;
 437         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 438                inode->i_sb->s_id, function, line, inode->i_ino);
 439         if (block)
 440                 printk(KERN_CONT "block %llu: ", block);
 441         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 442         va_end(args);
 443
 444         ext4_handle_error(inode->i_sb);
 445 }
 446
 447 void ext4_error_file(struct file *file, const char *function,
 448                      unsigned int line, ext4_fsblk_t block,
 449                      const char *fmt, ...)
 450 {
 451         va_list args;
 452         struct va_format vaf;
 453         struct ext4_super_block *es;
 454         struct inode *inode = file->f_dentry->d_inode;
 455         char pathname[80], *path;
 456
 457         es = EXT4_SB(inode->i_sb)->s_es;
 458         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 459         save_error_info(inode->i_sb, function, line);
 460         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 461         if (IS_ERR(path))
 462                 path = "(unknown)";
 463         printk(KERN_CRIT
 464                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 465                inode->i_sb->s_id, function, line, inode->i_ino);
 466         if (block)
 467                 printk(KERN_CONT "block %llu: ", block);
 468         va_start(args, fmt);
 469         vaf.fmt = fmt;
 470         vaf.va = &args;
 471         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 472         va_end(args);
 473
 474         ext4_handle_error(inode->i_sb);
 475 }
 476
 477 static const char *ext4_decode_error(struct super_block *sb, int errno,
 478                                      char nbuf[16])
 479 {
 480         char *errstr = NULL;
 481
 482         switch (errno) {
 483         case -EIO:
 484                 errstr = "IO failure";
 485                 break;
 486         case -ENOMEM:
 487                 errstr = "Out of memory";
 488                 break;
 489         case -EROFS:
 490                 if (!sb || (EXT4_SB(sb)->s_journal &&
 491                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 492                         errstr = "Journal has aborted";
 493                 else
 494                         errstr = "Readonly filesystem";
 495                 break;
 496         default:
 497                 /* If the caller passed in an extra buffer for unknown
 498                  * errors, textualise them now.  Else we just return
 499                  * NULL. */
 500                 if (nbuf) {
 501                         /* Check for truncated error codes... */
 502                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 503                                 errstr = nbuf;
 504                 }
 505                 break;
 506         }
 507
 508         return errstr;
 509 }
 510
 511 /* __ext4_std_error decodes expected errors from journaling functions
 512  * automatically and invokes the appropriate error response.  */
 513
 514 void __ext4_std_error(struct super_block *sb, const char *function,
 515                       unsigned int line, int errno)
 516 {
 517         char nbuf[16];
 518         const char *errstr;
 519
 520         /* Special case: if the error is EROFS, and we're not already
 521          * inside a transaction, then there's really no point in logging
 522          * an error. */
 523         if (errno == -EROFS && journal_current_handle() == NULL &&
 524             (sb->s_flags & MS_RDONLY))
 525                 return;
 526
 527         errstr = ext4_decode_error(sb, errno, nbuf);
 528         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 529                sb->s_id, function, line, errstr);
 530         save_error_info(sb, function, line);
 531
 532         ext4_handle_error(sb);
 533 }
 534
 535 /*
 536  * ext4_abort is a much stronger failure handler than ext4_error.  The
 537  * abort function may be used to deal with unrecoverable failures such
 538  * as journal IO errors or ENOMEM at a critical moment in log management.
 539  *
 540  * We unconditionally force the filesystem into an ABORT|READONLY state,
 541  * unless the error response on the fs has been set to panic in which
 542  * case we take the easy way out and panic immediately.
 543  */
 544
 545 void __ext4_abort(struct super_block *sb, const char *function,
 546                 unsigned int line, const char *fmt, ...)
 547 {
 548         va_list args;
 549
 550         save_error_info(sb, function, line);
 551         va_start(args, fmt);
 552         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 553                function, line);
 554         vprintk(fmt, args);
 555         printk("\n");
 556         va_end(args);
 557
 558         if ((sb->s_flags & MS_RDONLY) == 0) {
 559                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 560                 sb->s_flags |= MS_RDONLY;
 561                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 562                 if (EXT4_SB(sb)->s_journal)
 563                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 564                 save_error_info(sb, function, line);
 565         }
 566         if (test_opt(sb, ERRORS_PANIC))
 567                 panic("EXT4-fs panic from previous error\n");
 568 }
 569
 570 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 571 {
 572         struct va_format vaf;
 573         va_list args;
 574
 575         va_start(args, fmt);
 576         vaf.fmt = fmt;
 577         vaf.va = &args;
 578         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 579         va_end(args);
 580 }
 581
 582 void __ext4_warning(struct super_block *sb, const char *function,
 583                     unsigned int line, const char *fmt, ...)
 584 {
 585         struct va_format vaf;
 586         va_list args;
 587
 588         va_start(args, fmt);
 589         vaf.fmt = fmt;
 590         vaf.va = &args;
 591         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 592                sb->s_id, function, line, &vaf);
 593         va_end(args);
 594 }
 595
 596 void __ext4_grp_locked_error(const char *function, unsigned int line,
 597                              struct super_block *sb, ext4_group_t grp,
 598                              unsigned long ino, ext4_fsblk_t block,
 599                              const char *fmt, ...)
 600 __releases(bitlock)
 601 __acquires(bitlock)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 606
 607         es->s_last_error_ino = cpu_to_le32(ino);
 608         es->s_last_error_block = cpu_to_le64(block);
 609         __save_error_info(sb, function, line);
 610
 611         va_start(args, fmt);
 612
 613         vaf.fmt = fmt;
 614         vaf.va = &args;
 615         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 616                sb->s_id, function, line, grp);
 617         if (ino)
 618                 printk(KERN_CONT "inode %lu: ", ino);
 619         if (block)
 620                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 621         printk(KERN_CONT "%pV\n", &vaf);
 622         va_end(args);
 623
 624         if (test_opt(sb, ERRORS_CONT)) {
 625                 ext4_commit_super(sb, 0);
 626                 return;
 627         }
 628
 629         ext4_unlock_group(sb, grp);
 630         ext4_handle_error(sb);
 631         /*
 632          * We only get here in the ERRORS_RO case; relocking the group
 633          * may be dangerous, but nothing bad will happen since the
 634          * filesystem will have already been marked read/only and the
 635          * journal has been aborted.  We return 1 as a hint to callers
 636          * who might what to use the return value from
 637          * ext4_grp_locked_error() to distinguish between the
 638          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 639          * aggressively from the ext4 function in question, with a
 640          * more appropriate error code.
 641          */
 642         ext4_lock_group(sb, grp);
 643         return;
 644 }
 645
 646 void ext4_update_dynamic_rev(struct super_block *sb)
 647 {
 648         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 649
 650         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 651                 return;
 652
 653         ext4_warning(sb,
 654                      "updating to rev %d because of new feature flag, "
 655                      "running e2fsck is recommended",
 656                      EXT4_DYNAMIC_REV);
 657
 658         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 659         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 660         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 661         /* leave es->s_feature_*compat flags alone */
 662         /* es->s_uuid will be set by e2fsck if empty */
 663
 664         /*
 665          * The rest of the superblock fields should be zero, and if not it
 666          * means they are likely already in use, so leave them alone.  We
 667          * can leave it up to e2fsck to clean up any inconsistencies there.
 668          */
 669 }
 670
 671 /*
 672  * Open the external journal device
 673  */
 674 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 675 {
 676         struct block_device *bdev;
 677         char b[BDEVNAME_SIZE];
 678
 679         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 680         if (IS_ERR(bdev))
 681                 goto fail;
 682         return bdev;
 683
 684 fail:
 685         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 686                         __bdevname(dev, b), PTR_ERR(bdev));
 687         return NULL;
 688 }
 689
 690 /*
 691  * Release the journal device
 692  */
 693 static int ext4_blkdev_put(struct block_device *bdev)
 694 {
 695         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 696 }
 697
 698 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 699 {
 700         struct block_device *bdev;
 701         int ret = -ENODEV;
 702
 703         bdev = sbi->journal_bdev;
 704         if (bdev) {
 705                 ret = ext4_blkdev_put(bdev);
 706                 sbi->journal_bdev = NULL;
 707         }
 708         return ret;
 709 }
 710
 711 static inline struct inode *orphan_list_entry(struct list_head *l)
 712 {
 713         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 714 }
 715
 716 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 717 {
 718         struct list_head *l;
 719
 720         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 721                  le32_to_cpu(sbi->s_es->s_last_orphan));
 722
 723         printk(KERN_ERR "sb_info orphan list:\n");
 724         list_for_each(l, &sbi->s_orphan) {
 725                 struct inode *inode = orphan_list_entry(l);
 726                 printk(KERN_ERR "  "
 727                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 728                        inode->i_sb->s_id, inode->i_ino, inode,
 729                        inode->i_mode, inode->i_nlink,
 730                        NEXT_ORPHAN(inode));
 731         }
 732 }
 733
 734 static void ext4_put_super(struct super_block *sb)
 735 {
 736         struct ext4_sb_info *sbi = EXT4_SB(sb);
 737         struct ext4_super_block *es = sbi->s_es;
 738         int i, err;
 739
 740         ext4_unregister_li_request(sb);
 741         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 742
 743         flush_workqueue(sbi->dio_unwritten_wq);
 744         destroy_workqueue(sbi->dio_unwritten_wq);
 745
 746         lock_super(sb);
 747         if (sb->s_dirt)
 748                 ext4_commit_super(sb, 1);
 749
 750         if (sbi->s_journal) {
 751                 err = jbd2_journal_destroy(sbi->s_journal);
 752                 sbi->s_journal = NULL;
 753                 if (err < 0)
 754                         ext4_abort(sb, "Couldn't clean up the journal");
 755         }
 756
 757         del_timer(&sbi->s_err_report);
 758         ext4_release_system_zone(sb);
 759         ext4_mb_release(sb);
 760         ext4_ext_release(sb);
 761         ext4_xattr_put_super(sb);
 762
 763         if (!(sb->s_flags & MS_RDONLY)) {
 764                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 765                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 766                 ext4_commit_super(sb, 1);
 767         }
 768         if (sbi->s_proc) {
 769                 remove_proc_entry(sb->s_id, ext4_proc_root);
 770         }
 771         kobject_del(&sbi->s_kobj);
 772
 773         for (i = 0; i < sbi->s_gdb_count; i++)
 774                 brelse(sbi->s_group_desc[i]);
 775         kfree(sbi->s_group_desc);
 776         if (is_vmalloc_addr(sbi->s_flex_groups))
 777                 vfree(sbi->s_flex_groups);
 778         else
 779                 kfree(sbi->s_flex_groups);
 780         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 781         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 782         percpu_counter_destroy(&sbi->s_dirs_counter);
 783         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 784         brelse(sbi->s_sbh);
 785 #ifdef CONFIG_QUOTA
 786         for (i = 0; i < MAXQUOTAS; i++)
 787                 kfree(sbi->s_qf_names[i]);
 788 #endif
 789
 790         /* Debugging code just in case the in-memory inode orphan list
 791          * isn't empty.  The on-disk one can be non-empty if we've
 792          * detected an error and taken the fs readonly, but the
 793          * in-memory list had better be clean by this point. */
 794         if (!list_empty(&sbi->s_orphan))
 795                 dump_orphan_list(sb, sbi);
 796         J_ASSERT(list_empty(&sbi->s_orphan));
 797
 798         invalidate_bdev(sb->s_bdev);
 799         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 800                 /*
 801                  * Invalidate the journal device's buffers.  We don't want them
 802                  * floating about in memory - the physical journal device may
 803                  * hotswapped, and it breaks the `ro-after' testing code.
 804                  */
 805                 sync_blockdev(sbi->journal_bdev);
 806                 invalidate_bdev(sbi->journal_bdev);
 807                 ext4_blkdev_remove(sbi);
 808         }
 809         sb->s_fs_info = NULL;
 810         /*
 811          * Now that we are completely done shutting down the
 812          * superblock, we need to actually destroy the kobject.
 813          */
 814         unlock_super(sb);
 815         kobject_put(&sbi->s_kobj);
 816         wait_for_completion(&sbi->s_kobj_unregister);
 817         kfree(sbi->s_blockgroup_lock);
 818         kfree(sbi);
 819 }
 820
 821 static struct kmem_cache *ext4_inode_cachep;
 822
 823 /*
 824  * Called inside transaction, so use GFP_NOFS
 825  */
 826 static struct inode *ext4_alloc_inode(struct super_block *sb)
 827 {
 828         struct ext4_inode_info *ei;
 829
 830         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 831         if (!ei)
 832                 return NULL;
 833
 834         ei->vfs_inode.i_version = 1;
 835         ei->vfs_inode.i_data.writeback_index = 0;
 836         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 837         INIT_LIST_HEAD(&ei->i_prealloc_list);
 838         spin_lock_init(&ei->i_prealloc_lock);
 839         ei->i_reserved_data_blocks = 0;
 840         ei->i_reserved_meta_blocks = 0;
 841         ei->i_allocated_meta_blocks = 0;
 842         ei->i_da_metadata_calc_len = 0;
 843         spin_lock_init(&(ei->i_block_reservation_lock));
 844 #ifdef CONFIG_QUOTA
 845         ei->i_reserved_quota = 0;
 846 #endif
 847         ei->jinode = NULL;
 848         INIT_LIST_HEAD(&ei->i_completed_io_list);
 849         spin_lock_init(&ei->i_completed_io_lock);
 850         ei->cur_aio_dio = NULL;
 851         ei->i_sync_tid = 0;
 852         ei->i_datasync_tid = 0;
 853         atomic_set(&ei->i_ioend_count, 0);
 854         atomic_set(&ei->i_aiodio_unwritten, 0);
 855
 856         return &ei->vfs_inode;
 857 }
 858
 859 static int ext4_drop_inode(struct inode *inode)
 860 {
 861         int drop = generic_drop_inode(inode);
 862
 863         trace_ext4_drop_inode(inode, drop);
 864         return drop;
 865 }
 866
 867 static void ext4_i_callback(struct rcu_head *head)
 868 {
 869         struct inode *inode = container_of(head, struct inode, i_rcu);
 870         INIT_LIST_HEAD(&inode->i_dentry);
 871         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 872 }
 873
 874 static void ext4_destroy_inode(struct inode *inode)
 875 {
 876         ext4_ioend_wait(inode);
 877         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 878                 ext4_msg(inode->i_sb, KERN_ERR,
 879                          "Inode %lu (%p): orphan list check failed!",
 880                          inode->i_ino, EXT4_I(inode));
 881                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 882                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 883                                 true);
 884                 dump_stack();
 885         }
 886         call_rcu(&inode->i_rcu, ext4_i_callback);
 887 }
 888
 889 static void init_once(void *foo)
 890 {
 891         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 892
 893         INIT_LIST_HEAD(&ei->i_orphan);
 894 #ifdef CONFIG_EXT4_FS_XATTR
 895         init_rwsem(&ei->xattr_sem);
 896 #endif
 897         init_rwsem(&ei->i_data_sem);
 898         inode_init_once(&ei->vfs_inode);
 899 }
 900
 901 static int init_inodecache(void)
 902 {
 903         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 904                                              sizeof(struct ext4_inode_info),
 905                                              0, (SLAB_RECLAIM_ACCOUNT|
 906                                                 SLAB_MEM_SPREAD),
 907                                              init_once);
 908         if (ext4_inode_cachep == NULL)
 909                 return -ENOMEM;
 910         return 0;
 911 }
 912
 913 static void destroy_inodecache(void)
 914 {
 915         kmem_cache_destroy(ext4_inode_cachep);
 916 }
 917
 918 void ext4_clear_inode(struct inode *inode)
 919 {
 920         invalidate_inode_buffers(inode);
 921         end_writeback(inode);
 922         dquot_drop(inode);
 923         ext4_discard_preallocations(inode);
 924         if (EXT4_I(inode)->jinode) {
 925                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 926                                                EXT4_I(inode)->jinode);
 927                 jbd2_free_inode(EXT4_I(inode)->jinode);
 928                 EXT4_I(inode)->jinode = NULL;
 929         }
 930 }
 931
 932 static inline void ext4_show_quota_options(struct seq_file *seq,
 933                                            struct super_block *sb)
 934 {
 935 #if defined(CONFIG_QUOTA)
 936         struct ext4_sb_info *sbi = EXT4_SB(sb);
 937
 938         if (sbi->s_jquota_fmt) {
 939                 char *fmtname = "";
 940
 941                 switch (sbi->s_jquota_fmt) {
 942                 case QFMT_VFS_OLD:
 943                         fmtname = "vfsold";
 944                         break;
 945                 case QFMT_VFS_V0:
 946                         fmtname = "vfsv0";
 947                         break;
 948                 case QFMT_VFS_V1:
 949                         fmtname = "vfsv1";
 950                         break;
 951                 }
 952                 seq_printf(seq, ",jqfmt=%s", fmtname);
 953         }
 954
 955         if (sbi->s_qf_names[USRQUOTA])
 956                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 957
 958         if (sbi->s_qf_names[GRPQUOTA])
 959                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 960
 961         if (test_opt(sb, USRQUOTA))
 962                 seq_puts(seq, ",usrquota");
 963
 964         if (test_opt(sb, GRPQUOTA))
 965                 seq_puts(seq, ",grpquota");
 966 #endif
 967 }
 968
 969 /*
 970  * Show an option if
 971  *  - it's set to a non-default value OR
 972  *  - if the per-sb default is different from the global default
 973  */
 974 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 975 {
 976         int def_errors;
 977         unsigned long def_mount_opts;
 978         struct super_block *sb = vfs->mnt_sb;
 979         struct ext4_sb_info *sbi = EXT4_SB(sb);
 980         struct ext4_super_block *es = sbi->s_es;
 981
 982         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 983         def_errors     = le16_to_cpu(es->s_errors);
 984
 985         if (sbi->s_sb_block != 1)
 986                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 987         if (test_opt(sb, MINIX_DF))
 988                 seq_puts(seq, ",minixdf");
 989         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 990                 seq_puts(seq, ",grpid");
 991         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 992                 seq_puts(seq, ",nogrpid");
 993         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 994             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 995                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 996         }
 997         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 998             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 999                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1000         }
1001         if (test_opt(sb, ERRORS_RO)) {
1002                 if (def_errors == EXT4_ERRORS_PANIC ||
1003                     def_errors == EXT4_ERRORS_CONTINUE) {
1004                         seq_puts(seq, ",errors=remount-ro");
1005                 }
1006         }
1007         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1008                 seq_puts(seq, ",errors=continue");
1009         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1010                 seq_puts(seq, ",errors=panic");
1011         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1012                 seq_puts(seq, ",nouid32");
1013         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1014                 seq_puts(seq, ",debug");
1015         if (test_opt(sb, OLDALLOC))
1016                 seq_puts(seq, ",oldalloc");
1017 #ifdef CONFIG_EXT4_FS_XATTR
1018         if (test_opt(sb, XATTR_USER))
1019                 seq_puts(seq, ",user_xattr");
1020         if (!test_opt(sb, XATTR_USER))
1021                 seq_puts(seq, ",nouser_xattr");
1022 #endif
1023 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1024         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1025                 seq_puts(seq, ",acl");
1026         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1027                 seq_puts(seq, ",noacl");
1028 #endif
1029         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1030                 seq_printf(seq, ",commit=%u",
1031                            (unsigned) (sbi->s_commit_interval / HZ));
1032         }
1033         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1034                 seq_printf(seq, ",min_batch_time=%u",
1035                            (unsigned) sbi->s_min_batch_time);
1036         }
1037         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1038                 seq_printf(seq, ",max_batch_time=%u",
1039                            (unsigned) sbi->s_min_batch_time);
1040         }
1041
1042         /*
1043          * We're changing the default of barrier mount option, so
1044          * let's always display its mount state so it's clear what its
1045          * status is.
1046          */
1047         seq_puts(seq, ",barrier=");
1048         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1049         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1050                 seq_puts(seq, ",journal_async_commit");
1051         else if (test_opt(sb, JOURNAL_CHECKSUM))
1052                 seq_puts(seq, ",journal_checksum");
1053         if (test_opt(sb, I_VERSION))
1054                 seq_puts(seq, ",i_version");
1055         if (!test_opt(sb, DELALLOC) &&
1056             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1057                 seq_puts(seq, ",nodelalloc");
1058
1059         if (!test_opt(sb, MBLK_IO_SUBMIT))
1060                 seq_puts(seq, ",nomblk_io_submit");
1061         if (sbi->s_stripe)
1062                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1063         /*
1064          * journal mode get enabled in different ways
1065          * So just print the value even if we didn't specify it
1066          */
1067         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1068                 seq_puts(seq, ",data=journal");
1069         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1070                 seq_puts(seq, ",data=ordered");
1071         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1072                 seq_puts(seq, ",data=writeback");
1073
1074         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1075                 seq_printf(seq, ",inode_readahead_blks=%u",
1076                            sbi->s_inode_readahead_blks);
1077
1078         if (test_opt(sb, DATA_ERR_ABORT))
1079                 seq_puts(seq, ",data_err=abort");
1080
1081         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1082                 seq_puts(seq, ",noauto_da_alloc");
1083
1084         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1085                 seq_puts(seq, ",discard");
1086
1087         if (test_opt(sb, NOLOAD))
1088                 seq_puts(seq, ",norecovery");
1089
1090         if (test_opt(sb, DIOREAD_NOLOCK))
1091                 seq_puts(seq, ",dioread_nolock");
1092
1093         if (test_opt(sb, BLOCK_VALIDITY) &&
1094             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1095                 seq_puts(seq, ",block_validity");
1096
1097         if (!test_opt(sb, INIT_INODE_TABLE))
1098                 seq_puts(seq, ",noinit_inode_table");
1099         else if (sbi->s_li_wait_mult)
1100                 seq_printf(seq, ",init_inode_table=%u",
1101                            (unsigned) sbi->s_li_wait_mult);
1102
1103         ext4_show_quota_options(seq, sb);
1104
1105         return 0;
1106 }
1107
1108 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1109                                         u64 ino, u32 generation)
1110 {
1111         struct inode *inode;
1112
1113         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1114                 return ERR_PTR(-ESTALE);
1115         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1116                 return ERR_PTR(-ESTALE);
1117
1118         /* iget isn't really right if the inode is currently unallocated!!
1119          *
1120          * ext4_read_inode will return a bad_inode if the inode had been
1121          * deleted, so we should be safe.
1122          *
1123          * Currently we don't know the generation for parent directory, so
1124          * a generation of 0 means "accept any"
1125          */
1126         inode = ext4_iget(sb, ino);
1127         if (IS_ERR(inode))
1128                 return ERR_CAST(inode);
1129         if (generation && inode->i_generation != generation) {
1130                 iput(inode);
1131                 return ERR_PTR(-ESTALE);
1132         }
1133
1134         return inode;
1135 }
1136
1137 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1138                                         int fh_len, int fh_type)
1139 {
1140         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1141                                     ext4_nfs_get_inode);
1142 }
1143
1144 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1145                                         int fh_len, int fh_type)
1146 {
1147         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1148                                     ext4_nfs_get_inode);
1149 }
1150
1151 /*
1152  * Try to release metadata pages (indirect blocks, directories) which are
1153  * mapped via the block device.  Since these pages could have journal heads
1154  * which would prevent try_to_free_buffers() from freeing them, we must use
1155  * jbd2 layer's try_to_free_buffers() function to release them.
1156  */
1157 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1158                                  gfp_t wait)
1159 {
1160         journal_t *journal = EXT4_SB(sb)->s_journal;
1161
1162         WARN_ON(PageChecked(page));
1163         if (!page_has_buffers(page))
1164                 return 0;
1165         if (journal)
1166                 return jbd2_journal_try_to_free_buffers(journal, page,
1167                                                         wait & ~__GFP_WAIT);
1168         return try_to_free_buffers(page);
1169 }
1170
1171 #ifdef CONFIG_QUOTA
1172 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1173 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1174
1175 static int ext4_write_dquot(struct dquot *dquot);
1176 static int ext4_acquire_dquot(struct dquot *dquot);
1177 static int ext4_release_dquot(struct dquot *dquot);
1178 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1179 static int ext4_write_info(struct super_block *sb, int type);
1180 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1181                          struct path *path);
1182 static int ext4_quota_off(struct super_block *sb, int type);
1183 static int ext4_quota_on_mount(struct super_block *sb, int type);
1184 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1185                                size_t len, loff_t off);
1186 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1187                                 const char *data, size_t len, loff_t off);
1188
1189 static const struct dquot_operations ext4_quota_operations = {
1190 #ifdef CONFIG_QUOTA
1191         .get_reserved_space = ext4_get_reserved_space,
1192 #endif
1193         .write_dquot    = ext4_write_dquot,
1194         .acquire_dquot  = ext4_acquire_dquot,
1195         .release_dquot  = ext4_release_dquot,
1196         .mark_dirty     = ext4_mark_dquot_dirty,
1197         .write_info     = ext4_write_info,
1198         .alloc_dquot    = dquot_alloc,
1199         .destroy_dquot  = dquot_destroy,
1200 };
1201
1202 static const struct quotactl_ops ext4_qctl_operations = {
1203         .quota_on       = ext4_quota_on,
1204         .quota_off      = ext4_quota_off,
1205         .quota_sync     = dquot_quota_sync,
1206         .get_info       = dquot_get_dqinfo,
1207         .set_info       = dquot_set_dqinfo,
1208         .get_dqblk      = dquot_get_dqblk,
1209         .set_dqblk      = dquot_set_dqblk
1210 };
1211 #endif
1212
1213 static const struct super_operations ext4_sops = {
1214         .alloc_inode    = ext4_alloc_inode,
1215         .destroy_inode  = ext4_destroy_inode,
1216         .write_inode    = ext4_write_inode,
1217         .dirty_inode    = ext4_dirty_inode,
1218         .drop_inode     = ext4_drop_inode,
1219         .evict_inode    = ext4_evict_inode,
1220         .put_super      = ext4_put_super,
1221         .sync_fs        = ext4_sync_fs,
1222         .freeze_fs      = ext4_freeze,
1223         .unfreeze_fs    = ext4_unfreeze,
1224         .statfs         = ext4_statfs,
1225         .remount_fs     = ext4_remount,
1226         .show_options   = ext4_show_options,
1227 #ifdef CONFIG_QUOTA
1228         .quota_read     = ext4_quota_read,
1229         .quota_write    = ext4_quota_write,
1230 #endif
1231         .bdev_try_to_free_page = bdev_try_to_free_page,
1232 };
1233
1234 static const struct super_operations ext4_nojournal_sops = {
1235         .alloc_inode    = ext4_alloc_inode,
1236         .destroy_inode  = ext4_destroy_inode,
1237         .write_inode    = ext4_write_inode,
1238         .dirty_inode    = ext4_dirty_inode,
1239         .drop_inode     = ext4_drop_inode,
1240         .evict_inode    = ext4_evict_inode,
1241         .write_super    = ext4_write_super,
1242         .put_super      = ext4_put_super,
1243         .statfs         = ext4_statfs,
1244         .remount_fs     = ext4_remount,
1245         .show_options   = ext4_show_options,
1246 #ifdef CONFIG_QUOTA
1247         .quota_read     = ext4_quota_read,
1248         .quota_write    = ext4_quota_write,
1249 #endif
1250         .bdev_try_to_free_page = bdev_try_to_free_page,
1251 };
1252
1253 static const struct export_operations ext4_export_ops = {
1254         .fh_to_dentry = ext4_fh_to_dentry,
1255         .fh_to_parent = ext4_fh_to_parent,
1256         .get_parent = ext4_get_parent,
1257 };
1258
1259 enum {
1260         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1261         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1262         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1263         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1264         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1265         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1266         Opt_journal_update, Opt_journal_dev,
1267         Opt_journal_checksum, Opt_journal_async_commit,
1268         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1269         Opt_data_err_abort, Opt_data_err_ignore,
1270         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1271         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1272         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1273         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1274         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1275         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1276         Opt_inode_readahead_blks, Opt_journal_ioprio,
1277         Opt_dioread_nolock, Opt_dioread_lock,
1278         Opt_discard, Opt_nodiscard,
1279         Opt_init_inode_table, Opt_noinit_inode_table,
1280 };
1281
1282 static const match_table_t tokens = {
1283         {Opt_bsd_df, "bsddf"},
1284         {Opt_minix_df, "minixdf"},
1285         {Opt_grpid, "grpid"},
1286         {Opt_grpid, "bsdgroups"},
1287         {Opt_nogrpid, "nogrpid"},
1288         {Opt_nogrpid, "sysvgroups"},
1289         {Opt_resgid, "resgid=%u"},
1290         {Opt_resuid, "resuid=%u"},
1291         {Opt_sb, "sb=%u"},
1292         {Opt_err_cont, "errors=continue"},
1293         {Opt_err_panic, "errors=panic"},
1294         {Opt_err_ro, "errors=remount-ro"},
1295         {Opt_nouid32, "nouid32"},
1296         {Opt_debug, "debug"},
1297         {Opt_oldalloc, "oldalloc"},
1298         {Opt_orlov, "orlov"},
1299         {Opt_user_xattr, "user_xattr"},
1300         {Opt_nouser_xattr, "nouser_xattr"},
1301         {Opt_acl, "acl"},
1302         {Opt_noacl, "noacl"},
1303         {Opt_noload, "noload"},
1304         {Opt_noload, "norecovery"},
1305         {Opt_nobh, "nobh"},
1306         {Opt_bh, "bh"},
1307         {Opt_commit, "commit=%u"},
1308         {Opt_min_batch_time, "min_batch_time=%u"},
1309         {Opt_max_batch_time, "max_batch_time=%u"},
1310         {Opt_journal_update, "journal=update"},
1311         {Opt_journal_dev, "journal_dev=%u"},
1312         {Opt_journal_checksum, "journal_checksum"},
1313         {Opt_journal_async_commit, "journal_async_commit"},
1314         {Opt_abort, "abort"},
1315         {Opt_data_journal, "data=journal"},
1316         {Opt_data_ordered, "data=ordered"},
1317         {Opt_data_writeback, "data=writeback"},
1318         {Opt_data_err_abort, "data_err=abort"},
1319         {Opt_data_err_ignore, "data_err=ignore"},
1320         {Opt_offusrjquota, "usrjquota="},
1321         {Opt_usrjquota, "usrjquota=%s"},
1322         {Opt_offgrpjquota, "grpjquota="},
1323         {Opt_grpjquota, "grpjquota=%s"},
1324         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1325         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1326         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1327         {Opt_grpquota, "grpquota"},
1328         {Opt_noquota, "noquota"},
1329         {Opt_quota, "quota"},
1330         {Opt_usrquota, "usrquota"},
1331         {Opt_barrier, "barrier=%u"},
1332         {Opt_barrier, "barrier"},
1333         {Opt_nobarrier, "nobarrier"},
1334         {Opt_i_version, "i_version"},
1335         {Opt_stripe, "stripe=%u"},
1336         {Opt_resize, "resize"},
1337         {Opt_delalloc, "delalloc"},
1338         {Opt_nodelalloc, "nodelalloc"},
1339         {Opt_mblk_io_submit, "mblk_io_submit"},
1340         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1341         {Opt_block_validity, "block_validity"},
1342         {Opt_noblock_validity, "noblock_validity"},
1343         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1344         {Opt_journal_ioprio, "journal_ioprio=%u"},
1345         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1346         {Opt_auto_da_alloc, "auto_da_alloc"},
1347         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1348         {Opt_dioread_nolock, "dioread_nolock"},
1349         {Opt_dioread_lock, "dioread_lock"},
1350         {Opt_discard, "discard"},
1351         {Opt_nodiscard, "nodiscard"},
1352         {Opt_init_inode_table, "init_itable=%u"},
1353         {Opt_init_inode_table, "init_itable"},
1354         {Opt_noinit_inode_table, "noinit_itable"},
1355         {Opt_err, NULL},
1356 };
1357
1358 static ext4_fsblk_t get_sb_block(void **data)
1359 {
1360         ext4_fsblk_t    sb_block;
1361         char            *options = (char *) *data;
1362
1363         if (!options || strncmp(options, "sb=", 3) != 0)
1364                 return 1;       /* Default location */
1365
1366         options += 3;
1367         /* TODO: use simple_strtoll with >32bit ext4 */
1368         sb_block = simple_strtoul(options, &options, 0);
1369         if (*options && *options != ',') {
1370                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1371                        (char *) *data);
1372                 return 1;
1373         }
1374         if (*options == ',')
1375                 options++;
1376         *data = (void *) options;
1377
1378         return sb_block;
1379 }
1380
1381 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1382 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1383         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1384
1385 #ifdef CONFIG_QUOTA
1386 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1387 {
1388         struct ext4_sb_info *sbi = EXT4_SB(sb);
1389         char *qname;
1390
1391         if (sb_any_quota_loaded(sb) &&
1392                 !sbi->s_qf_names[qtype]) {
1393                 ext4_msg(sb, KERN_ERR,
1394                         "Cannot change journaled "
1395                         "quota options when quota turned on");
1396                 return 0;
1397         }
1398         qname = match_strdup(args);
1399         if (!qname) {
1400                 ext4_msg(sb, KERN_ERR,
1401                         "Not enough memory for storing quotafile name");
1402                 return 0;
1403         }
1404         if (sbi->s_qf_names[qtype] &&
1405                 strcmp(sbi->s_qf_names[qtype], qname)) {
1406                 ext4_msg(sb, KERN_ERR,
1407                         "%s quota file already specified", QTYPE2NAME(qtype));
1408                 kfree(qname);
1409                 return 0;
1410         }
1411         sbi->s_qf_names[qtype] = qname;
1412         if (strchr(sbi->s_qf_names[qtype], '/')) {
1413                 ext4_msg(sb, KERN_ERR,
1414                         "quotafile must be on filesystem root");
1415                 kfree(sbi->s_qf_names[qtype]);
1416                 sbi->s_qf_names[qtype] = NULL;
1417                 return 0;
1418         }
1419         set_opt(sb, QUOTA);
1420         return 1;
1421 }
1422
1423 static int clear_qf_name(struct super_block *sb, int qtype)
1424 {
1425
1426         struct ext4_sb_info *sbi = EXT4_SB(sb);
1427
1428         if (sb_any_quota_loaded(sb) &&
1429                 sbi->s_qf_names[qtype]) {
1430                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1431                         " when quota turned on");
1432                 return 0;
1433         }
1434         /*
1435          * The space will be released later when all options are confirmed
1436          * to be correct
1437          */
1438         sbi->s_qf_names[qtype] = NULL;
1439         return 1;
1440 }
1441 #endif
1442
1443 static int parse_options(char *options, struct super_block *sb,
1444                          unsigned long *journal_devnum,
1445                          unsigned int *journal_ioprio,
1446                          ext4_fsblk_t *n_blocks_count, int is_remount)
1447 {
1448         struct ext4_sb_info *sbi = EXT4_SB(sb);
1449         char *p;
1450         substring_t args[MAX_OPT_ARGS];
1451         int data_opt = 0;
1452         int option;
1453 #ifdef CONFIG_QUOTA
1454         int qfmt;
1455 #endif
1456
1457         if (!options)
1458                 return 1;
1459
1460         while ((p = strsep(&options, ",")) != NULL) {
1461                 int token;
1462                 if (!*p)
1463                         continue;
1464
1465                 /*
1466                  * Initialize args struct so we know whether arg was
1467                  * found; some options take optional arguments.
1468                  */
1469                 args[0].to = args[0].from = NULL;
1470                 token = match_token(p, tokens, args);
1471                 switch (token) {
1472                 case Opt_bsd_df:
1473                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1474                         clear_opt(sb, MINIX_DF);
1475                         break;
1476                 case Opt_minix_df:
1477                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1478                         set_opt(sb, MINIX_DF);
1479
1480                         break;
1481                 case Opt_grpid:
1482                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1483                         set_opt(sb, GRPID);
1484
1485                         break;
1486                 case Opt_nogrpid:
1487                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1488                         clear_opt(sb, GRPID);
1489
1490                         break;
1491                 case Opt_resuid:
1492                         if (match_int(&args[0], &option))
1493                                 return 0;
1494                         sbi->s_resuid = option;
1495                         break;
1496                 case Opt_resgid:
1497                         if (match_int(&args[0], &option))
1498                                 return 0;
1499                         sbi->s_resgid = option;
1500                         break;
1501                 case Opt_sb:
1502                         /* handled by get_sb_block() instead of here */
1503                         /* *sb_block = match_int(&args[0]); */
1504                         break;
1505                 case Opt_err_panic:
1506                         clear_opt(sb, ERRORS_CONT);
1507                         clear_opt(sb, ERRORS_RO);
1508                         set_opt(sb, ERRORS_PANIC);
1509                         break;
1510                 case Opt_err_ro:
1511                         clear_opt(sb, ERRORS_CONT);
1512                         clear_opt(sb, ERRORS_PANIC);
1513                         set_opt(sb, ERRORS_RO);
1514                         break;
1515                 case Opt_err_cont:
1516                         clear_opt(sb, ERRORS_RO);
1517                         clear_opt(sb, ERRORS_PANIC);
1518                         set_opt(sb, ERRORS_CONT);
1519                         break;
1520                 case Opt_nouid32:
1521                         set_opt(sb, NO_UID32);
1522                         break;
1523                 case Opt_debug:
1524                         set_opt(sb, DEBUG);
1525                         break;
1526                 case Opt_oldalloc:
1527                         set_opt(sb, OLDALLOC);
1528                         break;
1529                 case Opt_orlov:
1530                         clear_opt(sb, OLDALLOC);
1531                         break;
1532 #ifdef CONFIG_EXT4_FS_XATTR
1533                 case Opt_user_xattr:
1534                         set_opt(sb, XATTR_USER);
1535                         break;
1536                 case Opt_nouser_xattr:
1537                         clear_opt(sb, XATTR_USER);
1538                         break;
1539 #else
1540                 case Opt_user_xattr:
1541                 case Opt_nouser_xattr:
1542                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1543                         break;
1544 #endif
1545 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1546                 case Opt_acl:
1547                         set_opt(sb, POSIX_ACL);
1548                         break;
1549                 case Opt_noacl:
1550                         clear_opt(sb, POSIX_ACL);
1551                         break;
1552 #else
1553                 case Opt_acl:
1554                 case Opt_noacl:
1555                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1556                         break;
1557 #endif
1558                 case Opt_journal_update:
1559                         /* @@@ FIXME */
1560                         /* Eventually we will want to be able to create
1561                            a journal file here.  For now, only allow the
1562                            user to specify an existing inode to be the
1563                            journal file. */
1564                         if (is_remount) {
1565                                 ext4_msg(sb, KERN_ERR,
1566                                          "Cannot specify journal on remount");
1567                                 return 0;
1568                         }
1569                         set_opt(sb, UPDATE_JOURNAL);
1570                         break;
1571                 case Opt_journal_dev:
1572                         if (is_remount) {
1573                                 ext4_msg(sb, KERN_ERR,
1574                                         "Cannot specify journal on remount");
1575                                 return 0;
1576                         }
1577                         if (match_int(&args[0], &option))
1578                                 return 0;
1579                         *journal_devnum = option;
1580                         break;
1581                 case Opt_journal_checksum:
1582                         set_opt(sb, JOURNAL_CHECKSUM);
1583                         break;
1584                 case Opt_journal_async_commit:
1585                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1586                         set_opt(sb, JOURNAL_CHECKSUM);
1587                         break;
1588                 case Opt_noload:
1589                         set_opt(sb, NOLOAD);
1590                         break;
1591                 case Opt_commit:
1592                         if (match_int(&args[0], &option))
1593                                 return 0;
1594                         if (option < 0)
1595                                 return 0;
1596                         if (option == 0)
1597                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1598                         sbi->s_commit_interval = HZ * option;
1599                         break;
1600                 case Opt_max_batch_time:
1601                         if (match_int(&args[0], &option))
1602                                 return 0;
1603                         if (option < 0)
1604                                 return 0;
1605                         if (option == 0)
1606                                 option = EXT4_DEF_MAX_BATCH_TIME;
1607                         sbi->s_max_batch_time = option;
1608                         break;
1609                 case Opt_min_batch_time:
1610                         if (match_int(&args[0], &option))
1611                                 return 0;
1612                         if (option < 0)
1613                                 return 0;
1614                         sbi->s_min_batch_time = option;
1615                         break;
1616                 case Opt_data_journal:
1617                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1618                         goto datacheck;
1619                 case Opt_data_ordered:
1620                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1621                         goto datacheck;
1622                 case Opt_data_writeback:
1623                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1624                 datacheck:
1625                         if (is_remount) {
1626                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1627                                         ext4_msg(sb, KERN_ERR,
1628                                                 "Cannot change data mode on remount");
1629                                         return 0;
1630                                 }
1631                         } else {
1632                                 clear_opt(sb, DATA_FLAGS);
1633                                 sbi->s_mount_opt |= data_opt;
1634                         }
1635                         break;
1636                 case Opt_data_err_abort:
1637                         set_opt(sb, DATA_ERR_ABORT);
1638                         break;
1639                 case Opt_data_err_ignore:
1640                         clear_opt(sb, DATA_ERR_ABORT);
1641                         break;
1642 #ifdef CONFIG_QUOTA
1643                 case Opt_usrjquota:
1644                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1645                                 return 0;
1646                         break;
1647                 case Opt_grpjquota:
1648                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1649                                 return 0;
1650                         break;
1651                 case Opt_offusrjquota:
1652                         if (!clear_qf_name(sb, USRQUOTA))
1653                                 return 0;
1654                         break;
1655                 case Opt_offgrpjquota:
1656                         if (!clear_qf_name(sb, GRPQUOTA))
1657                                 return 0;
1658                         break;
1659
1660                 case Opt_jqfmt_vfsold:
1661                         qfmt = QFMT_VFS_OLD;
1662                         goto set_qf_format;
1663                 case Opt_jqfmt_vfsv0:
1664                         qfmt = QFMT_VFS_V0;
1665                         goto set_qf_format;
1666                 case Opt_jqfmt_vfsv1:
1667                         qfmt = QFMT_VFS_V1;
1668 set_qf_format:
1669                         if (sb_any_quota_loaded(sb) &&
1670                             sbi->s_jquota_fmt != qfmt) {
1671                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1672                                         "journaled quota options when "
1673                                         "quota turned on");
1674                                 return 0;
1675                         }
1676                         sbi->s_jquota_fmt = qfmt;
1677                         break;
1678                 case Opt_quota:
1679                 case Opt_usrquota:
1680                         set_opt(sb, QUOTA);
1681                         set_opt(sb, USRQUOTA);
1682                         break;
1683                 case Opt_grpquota:
1684                         set_opt(sb, QUOTA);
1685                         set_opt(sb, GRPQUOTA);
1686                         break;
1687                 case Opt_noquota:
1688                         if (sb_any_quota_loaded(sb)) {
1689                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1690                                         "options when quota turned on");
1691                                 return 0;
1692                         }
1693                         clear_opt(sb, QUOTA);
1694                         clear_opt(sb, USRQUOTA);
1695                         clear_opt(sb, GRPQUOTA);
1696                         break;
1697 #else
1698                 case Opt_quota:
1699                 case Opt_usrquota:
1700                 case Opt_grpquota:
1701                         ext4_msg(sb, KERN_ERR,
1702                                 "quota options not supported");
1703                         break;
1704                 case Opt_usrjquota:
1705                 case Opt_grpjquota:
1706                 case Opt_offusrjquota:
1707                 case Opt_offgrpjquota:
1708                 case Opt_jqfmt_vfsold:
1709                 case Opt_jqfmt_vfsv0:
1710                 case Opt_jqfmt_vfsv1:
1711                         ext4_msg(sb, KERN_ERR,
1712                                 "journaled quota options not supported");
1713                         break;
1714                 case Opt_noquota:
1715                         break;
1716 #endif
1717                 case Opt_abort:
1718                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1719                         break;
1720                 case Opt_nobarrier:
1721                         clear_opt(sb, BARRIER);
1722                         break;
1723                 case Opt_barrier:
1724                         if (args[0].from) {
1725                                 if (match_int(&args[0], &option))
1726                                         return 0;
1727                         } else
1728                                 option = 1;     /* No argument, default to 1 */
1729                         if (option)
1730                                 set_opt(sb, BARRIER);
1731                         else
1732                                 clear_opt(sb, BARRIER);
1733                         break;
1734                 case Opt_ignore:
1735                         break;
1736                 case Opt_resize:
1737                         if (!is_remount) {
1738                                 ext4_msg(sb, KERN_ERR,
1739                                         "resize option only available "
1740                                         "for remount");
1741                                 return 0;
1742                         }
1743                         if (match_int(&args[0], &option) != 0)
1744                                 return 0;
1745                         *n_blocks_count = option;
1746                         break;
1747                 case Opt_nobh:
1748                         ext4_msg(sb, KERN_WARNING,
1749                                  "Ignoring deprecated nobh option");
1750                         break;
1751                 case Opt_bh:
1752                         ext4_msg(sb, KERN_WARNING,
1753                                  "Ignoring deprecated bh option");
1754                         break;
1755                 case Opt_i_version:
1756                         set_opt(sb, I_VERSION);
1757                         sb->s_flags |= MS_I_VERSION;
1758                         break;
1759                 case Opt_nodelalloc:
1760                         clear_opt(sb, DELALLOC);
1761                         break;
1762                 case Opt_mblk_io_submit:
1763                         set_opt(sb, MBLK_IO_SUBMIT);
1764                         break;
1765                 case Opt_nomblk_io_submit:
1766                         clear_opt(sb, MBLK_IO_SUBMIT);
1767                         break;
1768                 case Opt_stripe:
1769                         if (match_int(&args[0], &option))
1770                                 return 0;
1771                         if (option < 0)
1772                                 return 0;
1773                         sbi->s_stripe = option;
1774                         break;
1775                 case Opt_delalloc:
1776                         set_opt(sb, DELALLOC);
1777                         break;
1778                 case Opt_block_validity:
1779                         set_opt(sb, BLOCK_VALIDITY);
1780                         break;
1781                 case Opt_noblock_validity:
1782                         clear_opt(sb, BLOCK_VALIDITY);
1783                         break;
1784                 case Opt_inode_readahead_blks:
1785                         if (match_int(&args[0], &option))
1786                                 return 0;
1787                         if (option < 0 || option > (1 << 30))
1788                                 return 0;
1789                         if (option && !is_power_of_2(option)) {
1790                                 ext4_msg(sb, KERN_ERR,
1791                                          "EXT4-fs: inode_readahead_blks"
1792                                          " must be a power of 2");
1793                                 return 0;
1794                         }
1795                         sbi->s_inode_readahead_blks = option;
1796                         break;
1797                 case Opt_journal_ioprio:
1798                         if (match_int(&args[0], &option))
1799                                 return 0;
1800                         if (option < 0 || option > 7)
1801                                 break;
1802                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1803                                                             option);
1804                         break;
1805                 case Opt_noauto_da_alloc:
1806                         set_opt(sb, NO_AUTO_DA_ALLOC);
1807                         break;
1808                 case Opt_auto_da_alloc:
1809                         if (args[0].from) {
1810                                 if (match_int(&args[0], &option))
1811                                         return 0;
1812                         } else
1813                                 option = 1;     /* No argument, default to 1 */
1814                         if (option)
1815                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1816                         else
1817                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1818                         break;
1819                 case Opt_discard:
1820                         set_opt(sb, DISCARD);
1821                         break;
1822                 case Opt_nodiscard:
1823                         clear_opt(sb, DISCARD);
1824                         break;
1825                 case Opt_dioread_nolock:
1826                         set_opt(sb, DIOREAD_NOLOCK);
1827                         break;
1828                 case Opt_dioread_lock:
1829                         clear_opt(sb, DIOREAD_NOLOCK);
1830                         break;
1831                 case Opt_init_inode_table:
1832                         set_opt(sb, INIT_INODE_TABLE);
1833                         if (args[0].from) {
1834                                 if (match_int(&args[0], &option))
1835                                         return 0;
1836                         } else
1837                                 option = EXT4_DEF_LI_WAIT_MULT;
1838                         if (option < 0)
1839                                 return 0;
1840                         sbi->s_li_wait_mult = option;
1841                         break;
1842                 case Opt_noinit_inode_table:
1843                         clear_opt(sb, INIT_INODE_TABLE);
1844                         break;
1845                 default:
1846                         ext4_msg(sb, KERN_ERR,
1847                                "Unrecognized mount option \"%s\" "
1848                                "or missing value", p);
1849                         return 0;
1850                 }
1851         }
1852 #ifdef CONFIG_QUOTA
1853         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1854                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1855                         clear_opt(sb, USRQUOTA);
1856
1857                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1858                         clear_opt(sb, GRPQUOTA);
1859
1860                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1861                         ext4_msg(sb, KERN_ERR, "old and new quota "
1862                                         "format mixing");
1863                         return 0;
1864                 }
1865
1866                 if (!sbi->s_jquota_fmt) {
1867                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1868                                         "not specified");
1869                         return 0;
1870                 }
1871         } else {
1872                 if (sbi->s_jquota_fmt) {
1873                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1874                                         "specified with no journaling "
1875                                         "enabled");
1876                         return 0;
1877                 }
1878         }
1879 #endif
1880         return 1;
1881 }
1882
1883 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1884                             int read_only)
1885 {
1886         struct ext4_sb_info *sbi = EXT4_SB(sb);
1887         int res = 0;
1888
1889         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1890                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1891                          "forcing read-only mode");
1892                 res = MS_RDONLY;
1893         }
1894         if (read_only)
1895                 return res;
1896         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1897                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1898                          "running e2fsck is recommended");
1899         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1900                 ext4_msg(sb, KERN_WARNING,
1901                          "warning: mounting fs with errors, "
1902                          "running e2fsck is recommended");
1903         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1904                  le16_to_cpu(es->s_mnt_count) >=
1905                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1906                 ext4_msg(sb, KERN_WARNING,
1907                          "warning: maximal mount count reached, "
1908                          "running e2fsck is recommended");
1909         else if (le32_to_cpu(es->s_checkinterval) &&
1910                 (le32_to_cpu(es->s_lastcheck) +
1911                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1912                 ext4_msg(sb, KERN_WARNING,
1913                          "warning: checktime reached, "
1914                          "running e2fsck is recommended");
1915         if (!sbi->s_journal)
1916                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1917         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1918                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1919         le16_add_cpu(&es->s_mnt_count, 1);
1920         es->s_mtime = cpu_to_le32(get_seconds());
1921         ext4_update_dynamic_rev(sb);
1922         if (sbi->s_journal)
1923                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1924
1925         ext4_commit_super(sb, 1);
1926         if (test_opt(sb, DEBUG))
1927                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1928                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1929                         sb->s_blocksize,
1930                         sbi->s_groups_count,
1931                         EXT4_BLOCKS_PER_GROUP(sb),
1932                         EXT4_INODES_PER_GROUP(sb),
1933                         sbi->s_mount_opt, sbi->s_mount_opt2);
1934
1935         return res;
1936 }
1937
1938 static int ext4_fill_flex_info(struct super_block *sb)
1939 {
1940         struct ext4_sb_info *sbi = EXT4_SB(sb);
1941         struct ext4_group_desc *gdp = NULL;
1942         ext4_group_t flex_group_count;
1943         ext4_group_t flex_group;
1944         int groups_per_flex = 0;
1945         size_t size;
1946         int i;
1947
1948         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1949         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1950
1951         if (groups_per_flex < 2) {
1952                 sbi->s_log_groups_per_flex = 0;
1953                 return 1;
1954         }
1955
1956         /* We allocate both existing and potentially added groups */
1957         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1958                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1959                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1960         size = flex_group_count * sizeof(struct flex_groups);
1961         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1962         if (sbi->s_flex_groups == NULL) {
1963                 sbi->s_flex_groups = vzalloc(size);
1964                 if (sbi->s_flex_groups == NULL) {
1965                         ext4_msg(sb, KERN_ERR,
1966                                  "not enough memory for %u flex groups",
1967                                  flex_group_count);
1968                         goto failed;
1969                 }
1970         }
1971
1972         for (i = 0; i < sbi->s_groups_count; i++) {
1973                 gdp = ext4_get_group_desc(sb, i, NULL);
1974
1975                 flex_group = ext4_flex_group(sbi, i);
1976                 atomic_add(ext4_free_inodes_count(sb, gdp),
1977                            &sbi->s_flex_groups[flex_group].free_inodes);
1978                 atomic_add(ext4_free_blks_count(sb, gdp),
1979                            &sbi->s_flex_groups[flex_group].free_blocks);
1980                 atomic_add(ext4_used_dirs_count(sb, gdp),
1981                            &sbi->s_flex_groups[flex_group].used_dirs);
1982         }
1983
1984         return 1;
1985 failed:
1986         return 0;
1987 }
1988
1989 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1990                             struct ext4_group_desc *gdp)
1991 {
1992         __u16 crc = 0;
1993
1994         if (sbi->s_es->s_feature_ro_compat &
1995             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1996                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1997                 __le32 le_group = cpu_to_le32(block_group);
1998
1999                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2000                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2001                 crc = crc16(crc, (__u8 *)gdp, offset);
2002                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2003                 /* for checksum of struct ext4_group_desc do the rest...*/
2004                 if ((sbi->s_es->s_feature_incompat &
2005                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2006                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2007                         crc = crc16(crc, (__u8 *)gdp + offset,
2008                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2009                                         offset);
2010         }
2011
2012         return cpu_to_le16(crc);
2013 }
2014
2015 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2016                                 struct ext4_group_desc *gdp)
2017 {
2018         if ((sbi->s_es->s_feature_ro_compat &
2019              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2020             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2021                 return 0;
2022
2023         return 1;
2024 }
2025
2026 /* Called at mount-time, super-block is locked */
2027 static int ext4_check_descriptors(struct super_block *sb,
2028                                   ext4_group_t *first_not_zeroed)
2029 {
2030         struct ext4_sb_info *sbi = EXT4_SB(sb);
2031         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2032         ext4_fsblk_t last_block;
2033         ext4_fsblk_t block_bitmap;
2034         ext4_fsblk_t inode_bitmap;
2035         ext4_fsblk_t inode_table;
2036         int flexbg_flag = 0;
2037         ext4_group_t i, grp = sbi->s_groups_count;
2038
2039         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2040                 flexbg_flag = 1;
2041
2042         ext4_debug("Checking group descriptors");
2043
2044         for (i = 0; i < sbi->s_groups_count; i++) {
2045                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2046
2047                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2048                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2049                 else
2050                         last_block = first_block +
2051                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2052
2053                 if ((grp == sbi->s_groups_count) &&
2054                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2055                         grp = i;
2056
2057                 block_bitmap = ext4_block_bitmap(sb, gdp);
2058                 if (block_bitmap < first_block || block_bitmap > last_block) {
2059                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2060                                "Block bitmap for group %u not in group "
2061                                "(block %llu)!", i, block_bitmap);
2062                         return 0;
2063                 }
2064                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2065                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2066                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2067                                "Inode bitmap for group %u not in group "
2068                                "(block %llu)!", i, inode_bitmap);
2069                         return 0;
2070                 }
2071                 inode_table = ext4_inode_table(sb, gdp);
2072                 if (inode_table < first_block ||
2073                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2074                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2075                                "Inode table for group %u not in group "
2076                                "(block %llu)!", i, inode_table);
2077                         return 0;
2078                 }
2079                 ext4_lock_group(sb, i);
2080                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2081                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2082                                  "Checksum for group %u failed (%u!=%u)",
2083                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2084                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2085                         if (!(sb->s_flags & MS_RDONLY)) {
2086                                 ext4_unlock_group(sb, i);
2087                                 return 0;
2088                         }
2089                 }
2090                 ext4_unlock_group(sb, i);
2091                 if (!flexbg_flag)
2092                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2093         }
2094         if (NULL != first_not_zeroed)
2095                 *first_not_zeroed = grp;
2096
2097         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2098         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2099         return 1;
2100 }
2101
2102 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2103  * the superblock) which were deleted from all directories, but held open by
2104  * a process at the time of a crash.  We walk the list and try to delete these
2105  * inodes at recovery time (only with a read-write filesystem).
2106  *
2107  * In order to keep the orphan inode chain consistent during traversal (in
2108  * case of crash during recovery), we link each inode into the superblock
2109  * orphan list_head and handle it the same way as an inode deletion during
2110  * normal operation (which journals the operations for us).
2111  *
2112  * We only do an iget() and an iput() on each inode, which is very safe if we
2113  * accidentally point at an in-use or already deleted inode.  The worst that
2114  * can happen in this case is that we get a "bit already cleared" message from
2115  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2116  * e2fsck was run on this filesystem, and it must have already done the orphan
2117  * inode cleanup for us, so we can safely abort without any further action.
2118  */
2119 static void ext4_orphan_cleanup(struct super_block *sb,
2120                                 struct ext4_super_block *es)
2121 {
2122         unsigned int s_flags = sb->s_flags;
2123         int nr_orphans = 0, nr_truncates = 0;
2124 #ifdef CONFIG_QUOTA
2125         int i;
2126 #endif
2127         if (!es->s_last_orphan) {
2128                 jbd_debug(4, "no orphan inodes to clean up\n");
2129                 return;
2130         }
2131
2132         if (bdev_read_only(sb->s_bdev)) {
2133                 ext4_msg(sb, KERN_ERR, "write access "
2134                         "unavailable, skipping orphan cleanup");
2135                 return;
2136         }
2137
2138         /* Check if feature set would not allow a r/w mount */
2139         if (!ext4_feature_set_ok(sb, 0)) {
2140                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2141                          "unknown ROCOMPAT features");
2142                 return;
2143         }
2144
2145         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2146                 if (es->s_last_orphan)
2147                         jbd_debug(1, "Errors on filesystem, "
2148                                   "clearing orphan list.\n");
2149                 es->s_last_orphan = 0;
2150                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2151                 return;
2152         }
2153
2154         if (s_flags & MS_RDONLY) {
2155                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2156                 sb->s_flags &= ~MS_RDONLY;
2157         }
2158 #ifdef CONFIG_QUOTA
2159         /* Needed for iput() to work correctly and not trash data */
2160         sb->s_flags |= MS_ACTIVE;
2161         /* Turn on quotas so that they are updated correctly */
2162         for (i = 0; i < MAXQUOTAS; i++) {
2163                 if (EXT4_SB(sb)->s_qf_names[i]) {
2164                         int ret = ext4_quota_on_mount(sb, i);
2165                         if (ret < 0)
2166                                 ext4_msg(sb, KERN_ERR,
2167                                         "Cannot turn on journaled "
2168                                         "quota: error %d", ret);
2169                 }
2170         }
2171 #endif
2172
2173         while (es->s_last_orphan) {
2174                 struct inode *inode;
2175
2176                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2177                 if (IS_ERR(inode)) {
2178                         es->s_last_orphan = 0;
2179                         break;
2180                 }
2181
2182                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2183                 dquot_initialize(inode);
2184                 if (inode->i_nlink) {
2185                         ext4_msg(sb, KERN_DEBUG,
2186                                 "%s: truncating inode %lu to %lld bytes",
2187                                 __func__, inode->i_ino, inode->i_size);
2188                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2189                                   inode->i_ino, inode->i_size);
2190                         ext4_truncate(inode);
2191                         nr_truncates++;
2192                 } else {
2193                         ext4_msg(sb, KERN_DEBUG,
2194                                 "%s: deleting unreferenced inode %lu",
2195                                 __func__, inode->i_ino);
2196                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2197                                   inode->i_ino);
2198                         nr_orphans++;
2199                 }
2200                 iput(inode);  /* The delete magic happens here! */
2201         }
2202
2203 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2204
2205         if (nr_orphans)
2206                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2207                        PLURAL(nr_orphans));
2208         if (nr_truncates)
2209                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2210                        PLURAL(nr_truncates));
2211 #ifdef CONFIG_QUOTA
2212         /* Turn quotas off */
2213         for (i = 0; i < MAXQUOTAS; i++) {
2214                 if (sb_dqopt(sb)->files[i])
2215                         dquot_quota_off(sb, i);
2216         }
2217 #endif
2218         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2219 }
2220
2221 /*
2222  * Maximal extent format file size.
2223  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2224  * extent format containers, within a sector_t, and within i_blocks
2225  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2226  * so that won't be a limiting factor.
2227  *
2228  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2229  */
2230 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2231 {
2232         loff_t res;
2233         loff_t upper_limit = MAX_LFS_FILESIZE;
2234
2235         /* small i_blocks in vfs inode? */
2236         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2237                 /*
2238                  * CONFIG_LBDAF is not enabled implies the inode
2239                  * i_block represent total blocks in 512 bytes
2240                  * 32 == size of vfs inode i_blocks * 8
2241                  */
2242                 upper_limit = (1LL << 32) - 1;
2243
2244                 /* total blocks in file system block size */
2245                 upper_limit >>= (blkbits - 9);
2246                 upper_limit <<= blkbits;
2247         }
2248
2249         /* 32-bit extent-start container, ee_block */
2250         res = 1LL << 32;
2251         res <<= blkbits;
2252         res -= 1;
2253
2254         /* Sanity check against vm- & vfs- imposed limits */
2255         if (res > upper_limit)
2256                 res = upper_limit;
2257
2258         return res;
2259 }
2260
2261 /*
2262  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2263  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2264  * We need to be 1 filesystem block less than the 2^48 sector limit.
2265  */
2266 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2267 {
2268         loff_t res = EXT4_NDIR_BLOCKS;
2269         int meta_blocks;
2270         loff_t upper_limit;
2271         /* This is calculated to be the largest file size for a dense, block
2272          * mapped file such that the file's total number of 512-byte sectors,
2273          * including data and all indirect blocks, does not exceed (2^48 - 1).
2274          *
2275          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2276          * number of 512-byte sectors of the file.
2277          */
2278
2279         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2280                 /*
2281                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2282                  * the inode i_block field represents total file blocks in
2283                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2284                  */
2285                 upper_limit = (1LL << 32) - 1;
2286
2287                 /* total blocks in file system block size */
2288                 upper_limit >>= (bits - 9);
2289
2290         } else {
2291                 /*
2292                  * We use 48 bit ext4_inode i_blocks
2293                  * With EXT4_HUGE_FILE_FL set the i_blocks
2294                  * represent total number of blocks in
2295                  * file system block size
2296                  */
2297                 upper_limit = (1LL << 48) - 1;
2298
2299         }
2300
2301         /* indirect blocks */
2302         meta_blocks = 1;
2303         /* double indirect blocks */
2304         meta_blocks += 1 + (1LL << (bits-2));
2305         /* tripple indirect blocks */
2306         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2307
2308         upper_limit -= meta_blocks;
2309         upper_limit <<= bits;
2310
2311         res += 1LL << (bits-2);
2312         res += 1LL << (2*(bits-2));
2313         res += 1LL << (3*(bits-2));
2314         res <<= bits;
2315         if (res > upper_limit)
2316                 res = upper_limit;
2317
2318         if (res > MAX_LFS_FILESIZE)
2319                 res = MAX_LFS_FILESIZE;
2320
2321         return res;
2322 }
2323
2324 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2325                                    ext4_fsblk_t logical_sb_block, int nr)
2326 {
2327         struct ext4_sb_info *sbi = EXT4_SB(sb);
2328         ext4_group_t bg, first_meta_bg;
2329         int has_super = 0;
2330
2331         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2332
2333         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2334             nr < first_meta_bg)
2335                 return logical_sb_block + nr + 1;
2336         bg = sbi->s_desc_per_block * nr;
2337         if (ext4_bg_has_super(sb, bg))
2338                 has_super = 1;
2339
2340         return (has_super + ext4_group_first_block_no(sb, bg));
2341 }
2342
2343 /**
2344  * ext4_get_stripe_size: Get the stripe size.
2345  * @sbi: In memory super block info
2346  *
2347  * If we have specified it via mount option, then
2348  * use the mount option value. If the value specified at mount time is
2349  * greater than the blocks per group use the super block value.
2350  * If the super block value is greater than blocks per group return 0.
2351  * Allocator needs it be less than blocks per group.
2352  *
2353  */
2354 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2355 {
2356         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2357         unsigned long stripe_width =
2358                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2359
2360         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2361                 return sbi->s_stripe;
2362
2363         if (stripe_width <= sbi->s_blocks_per_group)
2364                 return stripe_width;
2365
2366         if (stride <= sbi->s_blocks_per_group)
2367                 return stride;
2368
2369         return 0;
2370 }
2371
2372 /* sysfs supprt */
2373
2374 struct ext4_attr {
2375         struct attribute attr;
2376         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2377         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2378                          const char *, size_t);
2379         int offset;
2380 };
2381
2382 static int parse_strtoul(const char *buf,
2383                 unsigned long max, unsigned long *value)
2384 {
2385         char *endp;
2386
2387         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2388         endp = skip_spaces(endp);
2389         if (*endp || *value > max)
2390                 return -EINVAL;
2391
2392         return 0;
2393 }
2394
2395 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2396                                               struct ext4_sb_info *sbi,
2397                                               char *buf)
2398 {
2399         return snprintf(buf, PAGE_SIZE, "%llu\n",
2400                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2401 }
2402
2403 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2404                                          struct ext4_sb_info *sbi, char *buf)
2405 {
2406         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2407
2408         if (!sb->s_bdev->bd_part)
2409                 return snprintf(buf, PAGE_SIZE, "0\n");
2410         return snprintf(buf, PAGE_SIZE, "%lu\n",
2411                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2412                          sbi->s_sectors_written_start) >> 1);
2413 }
2414
2415 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2416                                           struct ext4_sb_info *sbi, char *buf)
2417 {
2418         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2419
2420         if (!sb->s_bdev->bd_part)
2421                 return snprintf(buf, PAGE_SIZE, "0\n");
2422         return snprintf(buf, PAGE_SIZE, "%llu\n",
2423                         (unsigned long long)(sbi->s_kbytes_written +
2424                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2425                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2426 }
2427
2428 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2429                                           struct ext4_sb_info *sbi,
2430                                           const char *buf, size_t count)
2431 {
2432         unsigned long t;
2433
2434         if (parse_strtoul(buf, 0x40000000, &t))
2435                 return -EINVAL;
2436
2437         if (t && !is_power_of_2(t))
2438                 return -EINVAL;
2439
2440         sbi->s_inode_readahead_blks = t;
2441         return count;
2442 }
2443
2444 static ssize_t sbi_ui_show(struct ext4_attr *a,
2445                            struct ext4_sb_info *sbi, char *buf)
2446 {
2447         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2448
2449         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2450 }
2451
2452 static ssize_t sbi_ui_store(struct ext4_attr *a,
2453                             struct ext4_sb_info *sbi,
2454                             const char *buf, size_t count)
2455 {
2456         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2457         unsigned long t;
2458
2459         if (parse_strtoul(buf, 0xffffffff, &t))
2460                 return -EINVAL;
2461         *ui = t;
2462         return count;
2463 }
2464
2465 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2466 static struct ext4_attr ext4_attr_##_name = {                   \
2467         .attr = {.name = __stringify(_name), .mode = _mode },   \
2468         .show   = _show,                                        \
2469         .store  = _store,                                       \
2470         .offset = offsetof(struct ext4_sb_info, _elname),       \
2471 }
2472 #define EXT4_ATTR(name, mode, show, store) \
2473 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2474
2475 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2476 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2477 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2478 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2479         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2480 #define ATTR_LIST(name) &ext4_attr_##name.attr
2481
2482 EXT4_RO_ATTR(delayed_allocation_blocks);
2483 EXT4_RO_ATTR(session_write_kbytes);
2484 EXT4_RO_ATTR(lifetime_write_kbytes);
2485 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2486                  inode_readahead_blks_store, s_inode_readahead_blks);
2487 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2488 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2489 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2490 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2491 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2492 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2493 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2494 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2495
2496 static struct attribute *ext4_attrs[] = {
2497         ATTR_LIST(delayed_allocation_blocks),
2498         ATTR_LIST(session_write_kbytes),
2499         ATTR_LIST(lifetime_write_kbytes),
2500         ATTR_LIST(inode_readahead_blks),
2501         ATTR_LIST(inode_goal),
2502         ATTR_LIST(mb_stats),
2503         ATTR_LIST(mb_max_to_scan),
2504         ATTR_LIST(mb_min_to_scan),
2505         ATTR_LIST(mb_order2_req),
2506         ATTR_LIST(mb_stream_req),
2507         ATTR_LIST(mb_group_prealloc),
2508         ATTR_LIST(max_writeback_mb_bump),
2509         NULL,
2510 };
2511
2512 /* Features this copy of ext4 supports */
2513 EXT4_INFO_ATTR(lazy_itable_init);
2514 EXT4_INFO_ATTR(batched_discard);
2515
2516 static struct attribute *ext4_feat_attrs[] = {
2517         ATTR_LIST(lazy_itable_init),
2518         ATTR_LIST(batched_discard),
2519         NULL,
2520 };
2521
2522 static ssize_t ext4_attr_show(struct kobject *kobj,
2523                               struct attribute *attr, char *buf)
2524 {
2525         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2526                                                 s_kobj);
2527         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2528
2529         return a->show ? a->show(a, sbi, buf) : 0;
2530 }
2531
2532 static ssize_t ext4_attr_store(struct kobject *kobj,
2533                                struct attribute *attr,
2534                                const char *buf, size_t len)
2535 {
2536         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2537                                                 s_kobj);
2538         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2539
2540         return a->store ? a->store(a, sbi, buf, len) : 0;
2541 }
2542
2543 static void ext4_sb_release(struct kobject *kobj)
2544 {
2545         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2546                                                 s_kobj);
2547         complete(&sbi->s_kobj_unregister);
2548 }
2549
2550 static const struct sysfs_ops ext4_attr_ops = {
2551         .show   = ext4_attr_show,
2552         .store  = ext4_attr_store,
2553 };
2554
2555 static struct kobj_type ext4_ktype = {
2556         .default_attrs  = ext4_attrs,
2557         .sysfs_ops      = &ext4_attr_ops,
2558         .release        = ext4_sb_release,
2559 };
2560
2561 static void ext4_feat_release(struct kobject *kobj)
2562 {
2563         complete(&ext4_feat->f_kobj_unregister);
2564 }
2565
2566 static struct kobj_type ext4_feat_ktype = {
2567         .default_attrs  = ext4_feat_attrs,
2568         .sysfs_ops      = &ext4_attr_ops,
2569         .release        = ext4_feat_release,
2570 };
2571
2572 /*
2573  * Check whether this filesystem can be mounted based on
2574  * the features present and the RDONLY/RDWR mount requested.
2575  * Returns 1 if this filesystem can be mounted as requested,
2576  * 0 if it cannot be.
2577  */
2578 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2579 {
2580         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2581                 ext4_msg(sb, KERN_ERR,
2582                         "Couldn't mount because of "
2583                         "unsupported optional features (%x)",
2584                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2585                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2586                 return 0;
2587         }
2588
2589         if (readonly)
2590                 return 1;
2591
2592         /* Check that feature set is OK for a read-write mount */
2593         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2594                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2595                          "unsupported optional features (%x)",
2596                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2597                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2598                 return 0;
2599         }
2600         /*
2601          * Large file size enabled file system can only be mounted
2602          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2603          */
2604         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2605                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2606                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2607                                  "cannot be mounted RDWR without "
2608                                  "CONFIG_LBDAF");
2609                         return 0;
2610                 }
2611         }
2612         return 1;
2613 }
2614
2615 /*
2616  * This function is called once a day if we have errors logged
2617  * on the file system
2618  */
2619 static void print_daily_error_info(unsigned long arg)
2620 {
2621         struct super_block *sb = (struct super_block *) arg;
2622         struct ext4_sb_info *sbi;
2623         struct ext4_super_block *es;
2624
2625         sbi = EXT4_SB(sb);
2626         es = sbi->s_es;
2627
2628         if (es->s_error_count)
2629                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2630                          le32_to_cpu(es->s_error_count));
2631         if (es->s_first_error_time) {
2632                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2633                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2634                        (int) sizeof(es->s_first_error_func),
2635                        es->s_first_error_func,
2636                        le32_to_cpu(es->s_first_error_line));
2637                 if (es->s_first_error_ino)
2638                         printk(": inode %u",
2639                                le32_to_cpu(es->s_first_error_ino));
2640                 if (es->s_first_error_block)
2641                         printk(": block %llu", (unsigned long long)
2642                                le64_to_cpu(es->s_first_error_block));
2643                 printk("\n");
2644         }
2645         if (es->s_last_error_time) {
2646                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2647                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2648                        (int) sizeof(es->s_last_error_func),
2649                        es->s_last_error_func,
2650                        le32_to_cpu(es->s_last_error_line));
2651                 if (es->s_last_error_ino)
2652                         printk(": inode %u",
2653                                le32_to_cpu(es->s_last_error_ino));
2654                 if (es->s_last_error_block)
2655                         printk(": block %llu", (unsigned long long)
2656                                le64_to_cpu(es->s_last_error_block));
2657                 printk("\n");
2658         }
2659         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2660 }
2661
2662 static void ext4_lazyinode_timeout(unsigned long data)
2663 {
2664         struct task_struct *p = (struct task_struct *)data;
2665         wake_up_process(p);
2666 }
2667
2668 /* Find next suitable group and run ext4_init_inode_table */
2669 static int ext4_run_li_request(struct ext4_li_request *elr)
2670 {
2671         struct ext4_group_desc *gdp = NULL;
2672         ext4_group_t group, ngroups;
2673         struct super_block *sb;
2674         unsigned long timeout = 0;
2675         int ret = 0;
2676
2677         sb = elr->lr_super;
2678         ngroups = EXT4_SB(sb)->s_groups_count;
2679
2680         for (group = elr->lr_next_group; group < ngroups; group++) {
2681                 gdp = ext4_get_group_desc(sb, group, NULL);
2682                 if (!gdp) {
2683                         ret = 1;
2684                         break;
2685                 }
2686
2687                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2688                         break;
2689         }
2690
2691         if (group == ngroups)
2692                 ret = 1;
2693
2694         if (!ret) {
2695                 timeout = jiffies;
2696                 ret = ext4_init_inode_table(sb, group,
2697                                             elr->lr_timeout ? 0 : 1);
2698                 if (elr->lr_timeout == 0) {
2699                         timeout = jiffies - timeout;
2700                         if (elr->lr_sbi->s_li_wait_mult)
2701                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2702                         else
2703                                 timeout *= 20;
2704                         elr->lr_timeout = timeout;
2705                 }
2706                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2707                 elr->lr_next_group = group + 1;
2708         }
2709
2710         return ret;
2711 }
2712
2713 /*
2714  * Remove lr_request from the list_request and free the
2715  * request tructure. Should be called with li_list_mtx held
2716  */
2717 static void ext4_remove_li_request(struct ext4_li_request *elr)
2718 {
2719         struct ext4_sb_info *sbi;
2720
2721         if (!elr)
2722                 return;
2723
2724         sbi = elr->lr_sbi;
2725
2726         list_del(&elr->lr_request);
2727         sbi->s_li_request = NULL;
2728         kfree(elr);
2729 }
2730
2731 static void ext4_unregister_li_request(struct super_block *sb)
2732 {
2733         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2734
2735         if (!ext4_li_info)
2736                 return;
2737
2738         mutex_lock(&ext4_li_info->li_list_mtx);
2739         ext4_remove_li_request(elr);
2740         mutex_unlock(&ext4_li_info->li_list_mtx);
2741 }
2742
2743 static struct task_struct *ext4_lazyinit_task;
2744
2745 /*
2746  * This is the function where ext4lazyinit thread lives. It walks
2747  * through the request list searching for next scheduled filesystem.
2748  * When such a fs is found, run the lazy initialization request
2749  * (ext4_rn_li_request) and keep track of the time spend in this
2750  * function. Based on that time we compute next schedule time of
2751  * the request. When walking through the list is complete, compute
2752  * next waking time and put itself into sleep.
2753  */
2754 static int ext4_lazyinit_thread(void *arg)
2755 {
2756         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2757         struct list_head *pos, *n;
2758         struct ext4_li_request *elr;
2759         unsigned long next_wakeup;
2760         DEFINE_WAIT(wait);
2761
2762         BUG_ON(NULL == eli);
2763
2764         eli->li_timer.data = (unsigned long)current;
2765         eli->li_timer.function = ext4_lazyinode_timeout;
2766
2767         eli->li_task = current;
2768         wake_up(&eli->li_wait_task);
2769
2770 cont_thread:
2771         while (true) {
2772                 next_wakeup = MAX_JIFFY_OFFSET;
2773
2774                 mutex_lock(&eli->li_list_mtx);
2775                 if (list_empty(&eli->li_request_list)) {
2776                         mutex_unlock(&eli->li_list_mtx);
2777                         goto exit_thread;
2778                 }
2779
2780                 list_for_each_safe(pos, n, &eli->li_request_list) {
2781                         elr = list_entry(pos, struct ext4_li_request,
2782                                          lr_request);
2783
2784                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2785                                 if (ext4_run_li_request(elr) != 0) {
2786                                         /* error, remove the lazy_init job */
2787                                         ext4_remove_li_request(elr);
2788                                         continue;
2789                                 }
2790                         }
2791
2792                         if (time_before(elr->lr_next_sched, next_wakeup))
2793                                 next_wakeup = elr->lr_next_sched;
2794                 }
2795                 mutex_unlock(&eli->li_list_mtx);
2796
2797                 if (freezing(current))
2798                         refrigerator();
2799
2800                 if ((time_after_eq(jiffies, next_wakeup)) ||
2801                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2802                         cond_resched();
2803                         continue;
2804                 }
2805
2806                 eli->li_timer.expires = next_wakeup;
2807                 add_timer(&eli->li_timer);
2808                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2809                                 TASK_INTERRUPTIBLE);
2810                 if (time_before(jiffies, next_wakeup))
2811                         schedule();
2812                 finish_wait(&eli->li_wait_daemon, &wait);
2813                 if (kthread_should_stop()) {
2814                         ext4_clear_request_list();
2815                         goto exit_thread;
2816                 }
2817         }
2818
2819 exit_thread:
2820         /*
2821          * It looks like the request list is empty, but we need
2822          * to check it under the li_list_mtx lock, to prevent any
2823          * additions into it, and of course we should lock ext4_li_mtx
2824          * to atomically free the list and ext4_li_info, because at
2825          * this point another ext4 filesystem could be registering
2826          * new one.
2827          */
2828         mutex_lock(&ext4_li_mtx);
2829         mutex_lock(&eli->li_list_mtx);
2830         if (!list_empty(&eli->li_request_list)) {
2831                 mutex_unlock(&eli->li_list_mtx);
2832                 mutex_unlock(&ext4_li_mtx);
2833                 goto cont_thread;
2834         }
2835         mutex_unlock(&eli->li_list_mtx);
2836         del_timer_sync(&ext4_li_info->li_timer);
2837         eli->li_task = NULL;
2838         wake_up(&eli->li_wait_task);
2839
2840         kfree(ext4_li_info);
2841         ext4_lazyinit_task = NULL;
2842         ext4_li_info = NULL;
2843         mutex_unlock(&ext4_li_mtx);
2844
2845         return 0;
2846 }
2847
2848 static void ext4_clear_request_list(void)
2849 {
2850         struct list_head *pos, *n;
2851         struct ext4_li_request *elr;
2852
2853         mutex_lock(&ext4_li_info->li_list_mtx);
2854         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2855                 elr = list_entry(pos, struct ext4_li_request,
2856                                  lr_request);
2857                 ext4_remove_li_request(elr);
2858         }
2859         mutex_unlock(&ext4_li_info->li_list_mtx);
2860 }
2861
2862 static int ext4_run_lazyinit_thread(void)
2863 {
2864         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2865                                          ext4_li_info, "ext4lazyinit");
2866         if (IS_ERR(ext4_lazyinit_task)) {
2867                 int err = PTR_ERR(ext4_lazyinit_task);
2868                 ext4_clear_request_list();
2869                 del_timer_sync(&ext4_li_info->li_timer);
2870                 kfree(ext4_li_info);
2871                 ext4_li_info = NULL;
2872                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2873                                  "initialization thread\n",
2874                                  err);
2875                 return err;
2876         }
2877         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2878
2879         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2880         return 0;
2881 }
2882
2883 /*
2884  * Check whether it make sense to run itable init. thread or not.
2885  * If there is at least one uninitialized inode table, return
2886  * corresponding group number, else the loop goes through all
2887  * groups and return total number of groups.
2888  */
2889 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2890 {
2891         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2892         struct ext4_group_desc *gdp = NULL;
2893
2894         for (group = 0; group < ngroups; group++) {
2895                 gdp = ext4_get_group_desc(sb, group, NULL);
2896                 if (!gdp)
2897                         continue;
2898
2899                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2900                         break;
2901         }
2902
2903         return group;
2904 }
2905
2906 static int ext4_li_info_new(void)
2907 {
2908         struct ext4_lazy_init *eli = NULL;
2909
2910         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2911         if (!eli)
2912                 return -ENOMEM;
2913
2914         eli->li_task = NULL;
2915         INIT_LIST_HEAD(&eli->li_request_list);
2916         mutex_init(&eli->li_list_mtx);
2917
2918         init_waitqueue_head(&eli->li_wait_daemon);
2919         init_waitqueue_head(&eli->li_wait_task);
2920         init_timer(&eli->li_timer);
2921         eli->li_state |= EXT4_LAZYINIT_QUIT;
2922
2923         ext4_li_info = eli;
2924
2925         return 0;
2926 }
2927
2928 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2929                                             ext4_group_t start)
2930 {
2931         struct ext4_sb_info *sbi = EXT4_SB(sb);
2932         struct ext4_li_request *elr;
2933         unsigned long rnd;
2934
2935         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2936         if (!elr)
2937                 return NULL;
2938
2939         elr->lr_super = sb;
2940         elr->lr_sbi = sbi;
2941         elr->lr_next_group = start;
2942
2943         /*
2944          * Randomize first schedule time of the request to
2945          * spread the inode table initialization requests
2946          * better.
2947          */
2948         get_random_bytes(&rnd, sizeof(rnd));
2949         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2950                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2951
2952         return elr;
2953 }
2954
2955 static int ext4_register_li_request(struct super_block *sb,
2956                                     ext4_group_t first_not_zeroed)
2957 {
2958         struct ext4_sb_info *sbi = EXT4_SB(sb);
2959         struct ext4_li_request *elr;
2960         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2961         int ret = 0;
2962
2963         if (sbi->s_li_request != NULL)
2964                 return 0;
2965
2966         if (first_not_zeroed == ngroups ||
2967             (sb->s_flags & MS_RDONLY) ||
2968             !test_opt(sb, INIT_INODE_TABLE)) {
2969                 sbi->s_li_request = NULL;
2970                 return 0;
2971         }
2972
2973         if (first_not_zeroed == ngroups) {
2974                 sbi->s_li_request = NULL;
2975                 return 0;
2976         }
2977
2978         elr = ext4_li_request_new(sb, first_not_zeroed);
2979         if (!elr)
2980                 return -ENOMEM;
2981
2982         mutex_lock(&ext4_li_mtx);
2983
2984         if (NULL == ext4_li_info) {
2985                 ret = ext4_li_info_new();
2986                 if (ret)
2987                         goto out;
2988         }
2989
2990         mutex_lock(&ext4_li_info->li_list_mtx);
2991         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2992         mutex_unlock(&ext4_li_info->li_list_mtx);
2993
2994         sbi->s_li_request = elr;
2995         /*
2996          * set elr to NULL here since it has been inserted to
2997          * the request_list and the removal and free of it is
2998          * handled by ext4_clear_request_list from now on.
2999          */
3000         elr = NULL;
3001
3002         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3003                 ret = ext4_run_lazyinit_thread();
3004                 if (ret)
3005                         goto out;
3006         }
3007 out:
3008         mutex_unlock(&ext4_li_mtx);
3009         if (ret)
3010                 kfree(elr);
3011         return ret;
3012 }
3013
3014 /*
3015  * We do not need to lock anything since this is called on
3016  * module unload.
3017  */
3018 static void ext4_destroy_lazyinit_thread(void)
3019 {
3020         /*
3021          * If thread exited earlier
3022          * there's nothing to be done.
3023          */
3024         if (!ext4_li_info || !ext4_lazyinit_task)
3025                 return;
3026
3027         kthread_stop(ext4_lazyinit_task);
3028 }
3029
3030 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3031                                 __releases(kernel_lock)
3032                                 __acquires(kernel_lock)
3033 {
3034         char *orig_data = kstrdup(data, GFP_KERNEL);
3035         struct buffer_head *bh;
3036         struct ext4_super_block *es = NULL;
3037         struct ext4_sb_info *sbi;
3038         ext4_fsblk_t block;
3039         ext4_fsblk_t sb_block = get_sb_block(&data);
3040         ext4_fsblk_t logical_sb_block;
3041         unsigned long offset = 0;
3042         unsigned long journal_devnum = 0;
3043         unsigned long def_mount_opts;
3044         struct inode *root;
3045         char *cp;
3046         const char *descr;
3047         int ret = -ENOMEM;
3048         int blocksize;
3049         unsigned int db_count;
3050         unsigned int i;
3051         int needs_recovery, has_huge_files;
3052         __u64 blocks_count;
3053         int err;
3054         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3055         ext4_group_t first_not_zeroed;
3056
3057         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3058         if (!sbi)
3059                 goto out_free_orig;
3060
3061         sbi->s_blockgroup_lock =
3062                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3063         if (!sbi->s_blockgroup_lock) {
3064                 kfree(sbi);
3065                 goto out_free_orig;
3066         }
3067         sb->s_fs_info = sbi;
3068         sbi->s_mount_opt = 0;
3069         sbi->s_resuid = EXT4_DEF_RESUID;
3070         sbi->s_resgid = EXT4_DEF_RESGID;
3071         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3072         sbi->s_sb_block = sb_block;
3073         if (sb->s_bdev->bd_part)
3074                 sbi->s_sectors_written_start =
3075                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3076
3077         /* Cleanup superblock name */
3078         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3079                 *cp = '!';
3080
3081         ret = -EINVAL;
3082         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3083         if (!blocksize) {
3084                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3085                 goto out_fail;
3086         }
3087
3088         /*
3089          * The ext4 superblock will not be buffer aligned for other than 1kB
3090          * block sizes.  We need to calculate the offset from buffer start.
3091          */
3092         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3093                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3094                 offset = do_div(logical_sb_block, blocksize);
3095         } else {
3096                 logical_sb_block = sb_block;
3097         }
3098
3099         if (!(bh = sb_bread(sb, logical_sb_block))) {
3100                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3101                 goto out_fail;
3102         }
3103         /*
3104          * Note: s_es must be initialized as soon as possible because
3105          *       some ext4 macro-instructions depend on its value
3106          */
3107         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3108         sbi->s_es = es;
3109         sb->s_magic = le16_to_cpu(es->s_magic);
3110         if (sb->s_magic != EXT4_SUPER_MAGIC)
3111                 goto cantfind_ext4;
3112         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3113
3114         /* Set defaults before we parse the mount options */
3115         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3116         set_opt(sb, INIT_INODE_TABLE);
3117         if (def_mount_opts & EXT4_DEFM_DEBUG)
3118                 set_opt(sb, DEBUG);
3119         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3120                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3121                         "2.6.38");
3122                 set_opt(sb, GRPID);
3123         }
3124         if (def_mount_opts & EXT4_DEFM_UID16)
3125                 set_opt(sb, NO_UID32);
3126         /* xattr user namespace & acls are now defaulted on */
3127 #ifdef CONFIG_EXT4_FS_XATTR
3128         set_opt(sb, XATTR_USER);
3129 #endif
3130 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3131         set_opt(sb, POSIX_ACL);
3132 #endif
3133         set_opt(sb, MBLK_IO_SUBMIT);
3134         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3135                 set_opt(sb, JOURNAL_DATA);
3136         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3137                 set_opt(sb, ORDERED_DATA);
3138         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3139                 set_opt(sb, WRITEBACK_DATA);
3140
3141         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3142                 set_opt(sb, ERRORS_PANIC);
3143         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3144                 set_opt(sb, ERRORS_CONT);
3145         else
3146                 set_opt(sb, ERRORS_RO);
3147         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3148                 set_opt(sb, BLOCK_VALIDITY);
3149         if (def_mount_opts & EXT4_DEFM_DISCARD)
3150                 set_opt(sb, DISCARD);
3151
3152         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3153         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3154         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3155         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3156         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3157
3158         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3159                 set_opt(sb, BARRIER);
3160
3161         /*
3162          * enable delayed allocation by default
3163          * Use -o nodelalloc to turn it off
3164          */
3165         if (!IS_EXT3_SB(sb) &&
3166             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3167                 set_opt(sb, DELALLOC);
3168
3169         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3170                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3171                 ext4_msg(sb, KERN_WARNING,
3172                          "failed to parse options in superblock: %s",
3173                          sbi->s_es->s_mount_opts);
3174         }
3175         if (!parse_options((char *) data, sb, &journal_devnum,
3176                            &journal_ioprio, NULL, 0))
3177                 goto failed_mount;
3178
3179         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3180                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3181
3182         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3183             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3184              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3185              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3186                 ext4_msg(sb, KERN_WARNING,
3187                        "feature flags set on rev 0 fs, "
3188                        "running e2fsck is recommended");
3189
3190         /*
3191          * Check feature flags regardless of the revision level, since we
3192          * previously didn't change the revision level when setting the flags,
3193          * so there is a chance incompat flags are set on a rev 0 filesystem.
3194          */
3195         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3196                 goto failed_mount;
3197
3198         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3199
3200         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3201             blocksize > EXT4_MAX_BLOCK_SIZE) {
3202                 ext4_msg(sb, KERN_ERR,
3203                        "Unsupported filesystem blocksize %d", blocksize);
3204                 goto failed_mount;
3205         }
3206
3207         if (sb->s_blocksize != blocksize) {
3208                 /* Validate the filesystem blocksize */
3209                 if (!sb_set_blocksize(sb, blocksize)) {
3210                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3211                                         blocksize);
3212                         goto failed_mount;
3213                 }
3214
3215                 brelse(bh);
3216                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3217                 offset = do_div(logical_sb_block, blocksize);
3218                 bh = sb_bread(sb, logical_sb_block);
3219                 if (!bh) {
3220                         ext4_msg(sb, KERN_ERR,
3221                                "Can't read superblock on 2nd try");
3222                         goto failed_mount;
3223                 }
3224                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3225                 sbi->s_es = es;
3226                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3227                         ext4_msg(sb, KERN_ERR,
3228                                "Magic mismatch, very weird!");
3229                         goto failed_mount;
3230                 }
3231         }
3232
3233         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3234                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3235         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3236                                                       has_huge_files);
3237         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3238
3239         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3240                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3241                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3242         } else {
3243                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3244                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3245                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3246                     (!is_power_of_2(sbi->s_inode_size)) ||
3247                     (sbi->s_inode_size > blocksize)) {
3248                         ext4_msg(sb, KERN_ERR,
3249                                "unsupported inode size: %d",
3250                                sbi->s_inode_size);
3251                         goto failed_mount;
3252                 }
3253                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3254                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3255         }
3256
3257         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3258         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3259                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3260                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3261                     !is_power_of_2(sbi->s_desc_size)) {
3262                         ext4_msg(sb, KERN_ERR,
3263                                "unsupported descriptor size %lu",
3264                                sbi->s_desc_size);
3265                         goto failed_mount;
3266                 }
3267         } else
3268                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3269
3270         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3271         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3272         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3273                 goto cantfind_ext4;
3274
3275         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3276         if (sbi->s_inodes_per_block == 0)
3277                 goto cantfind_ext4;
3278         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3279                                         sbi->s_inodes_per_block;
3280         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3281         sbi->s_sbh = bh;
3282         sbi->s_mount_state = le16_to_cpu(es->s_state);
3283         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3284         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3285
3286         for (i = 0; i < 4; i++)
3287                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3288         sbi->s_def_hash_version = es->s_def_hash_version;
3289         i = le32_to_cpu(es->s_flags);
3290         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3291                 sbi->s_hash_unsigned = 3;
3292         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3293 #ifdef __CHAR_UNSIGNED__
3294                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3295                 sbi->s_hash_unsigned = 3;
3296 #else
3297                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3298 #endif
3299                 sb->s_dirt = 1;
3300         }
3301
3302         if (sbi->s_blocks_per_group > blocksize * 8) {
3303                 ext4_msg(sb, KERN_ERR,
3304                        "#blocks per group too big: %lu",
3305                        sbi->s_blocks_per_group);
3306                 goto failed_mount;
3307         }
3308         if (sbi->s_inodes_per_group > blocksize * 8) {
3309                 ext4_msg(sb, KERN_ERR,
3310                        "#inodes per group too big: %lu",
3311                        sbi->s_inodes_per_group);
3312                 goto failed_mount;
3313         }
3314
3315         /*
3316          * Test whether we have more sectors than will fit in sector_t,
3317          * and whether the max offset is addressable by the page cache.
3318          */
3319         err = generic_check_addressable(sb->s_blocksize_bits,
3320                                         ext4_blocks_count(es));
3321         if (err) {
3322                 ext4_msg(sb, KERN_ERR, "filesystem"
3323                          " too large to mount safely on this system");
3324                 if (sizeof(sector_t) < 8)
3325                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3326                 ret = err;
3327                 goto failed_mount;
3328         }
3329
3330         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3331                 goto cantfind_ext4;
3332
3333         /* check blocks count against device size */
3334         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3335         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3336                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3337                        "exceeds size of device (%llu blocks)",
3338                        ext4_blocks_count(es), blocks_count);
3339                 goto failed_mount;
3340         }
3341
3342         /*
3343          * It makes no sense for the first data block to be beyond the end
3344          * of the filesystem.
3345          */
3346         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3347                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3348                          "block %u is beyond end of filesystem (%llu)",
3349                          le32_to_cpu(es->s_first_data_block),
3350                          ext4_blocks_count(es));
3351                 goto failed_mount;
3352         }
3353         blocks_count = (ext4_blocks_count(es) -
3354                         le32_to_cpu(es->s_first_data_block) +
3355                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3356         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3357         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3358                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3359                        "(block count %llu, first data block %u, "
3360                        "blocks per group %lu)", sbi->s_groups_count,
3361                        ext4_blocks_count(es),
3362                        le32_to_cpu(es->s_first_data_block),
3363                        EXT4_BLOCKS_PER_GROUP(sb));
3364                 goto failed_mount;
3365         }
3366         sbi->s_groups_count = blocks_count;
3367         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3368                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3369         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3370                    EXT4_DESC_PER_BLOCK(sb);
3371         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3372                                     GFP_KERNEL);
3373         if (sbi->s_group_desc == NULL) {
3374                 ext4_msg(sb, KERN_ERR, "not enough memory");
3375                 goto failed_mount;
3376         }
3377
3378 #ifdef CONFIG_PROC_FS
3379         if (ext4_proc_root)
3380                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3381 #endif
3382
3383         bgl_lock_init(sbi->s_blockgroup_lock);
3384
3385         for (i = 0; i < db_count; i++) {
3386                 block = descriptor_loc(sb, logical_sb_block, i);
3387                 sbi->s_group_desc[i] = sb_bread(sb, block);
3388                 if (!sbi->s_group_desc[i]) {
3389                         ext4_msg(sb, KERN_ERR,
3390                                "can't read group descriptor %d", i);
3391                         db_count = i;
3392                         goto failed_mount2;
3393                 }
3394         }
3395         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3396                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3397                 goto failed_mount2;
3398         }
3399         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3400                 if (!ext4_fill_flex_info(sb)) {
3401                         ext4_msg(sb, KERN_ERR,
3402                                "unable to initialize "
3403                                "flex_bg meta info!");
3404                         goto failed_mount2;
3405                 }
3406
3407         sbi->s_gdb_count = db_count;
3408         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3409         spin_lock_init(&sbi->s_next_gen_lock);
3410
3411         init_timer(&sbi->s_err_report);
3412         sbi->s_err_report.function = print_daily_error_info;
3413         sbi->s_err_report.data = (unsigned long) sb;
3414
3415         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3416                         ext4_count_free_blocks(sb));
3417         if (!err) {
3418                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3419                                 ext4_count_free_inodes(sb));
3420         }
3421         if (!err) {
3422                 err = percpu_counter_init(&sbi->s_dirs_counter,
3423                                 ext4_count_dirs(sb));
3424         }
3425         if (!err) {
3426                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3427         }
3428         if (err) {
3429                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3430                 goto failed_mount3;
3431         }
3432
3433         sbi->s_stripe = ext4_get_stripe_size(sbi);
3434         sbi->s_max_writeback_mb_bump = 128;
3435
3436         /*
3437          * set up enough so that it can read an inode
3438          */
3439         if (!test_opt(sb, NOLOAD) &&
3440             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3441                 sb->s_op = &ext4_sops;
3442         else
3443                 sb->s_op = &ext4_nojournal_sops;
3444         sb->s_export_op = &ext4_export_ops;
3445         sb->s_xattr = ext4_xattr_handlers;
3446 #ifdef CONFIG_QUOTA
3447         sb->s_qcop = &ext4_qctl_operations;
3448         sb->dq_op = &ext4_quota_operations;
3449 #endif
3450         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3451
3452         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3453         mutex_init(&sbi->s_orphan_lock);
3454         mutex_init(&sbi->s_resize_lock);
3455
3456         sb->s_root = NULL;
3457
3458         needs_recovery = (es->s_last_orphan != 0 ||
3459                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3460                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3461
3462         /*
3463          * The first inode we look at is the journal inode.  Don't try
3464          * root first: it may be modified in the journal!
3465          */
3466         if (!test_opt(sb, NOLOAD) &&
3467             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3468                 if (ext4_load_journal(sb, es, journal_devnum))
3469                         goto failed_mount3;
3470         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3471               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3472                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3473                        "suppressed and not mounted read-only");
3474                 goto failed_mount_wq;
3475         } else {
3476                 clear_opt(sb, DATA_FLAGS);
3477                 set_opt(sb, WRITEBACK_DATA);
3478                 sbi->s_journal = NULL;
3479                 needs_recovery = 0;
3480                 goto no_journal;
3481         }
3482
3483         if (ext4_blocks_count(es) > 0xffffffffULL &&
3484             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3485                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3486                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3487                 goto failed_mount_wq;
3488         }
3489
3490         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3491                 jbd2_journal_set_features(sbi->s_journal,
3492                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3493                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3494         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3495                 jbd2_journal_set_features(sbi->s_journal,
3496                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3497                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3498                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3499         } else {
3500                 jbd2_journal_clear_features(sbi->s_journal,
3501                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3502                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3503         }
3504
3505         /* We have now updated the journal if required, so we can
3506          * validate the data journaling mode. */
3507         switch (test_opt(sb, DATA_FLAGS)) {
3508         case 0:
3509                 /* No mode set, assume a default based on the journal
3510                  * capabilities: ORDERED_DATA if the journal can
3511                  * cope, else JOURNAL_DATA
3512                  */
3513                 if (jbd2_journal_check_available_features
3514                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3515                         set_opt(sb, ORDERED_DATA);
3516                 else
3517                         set_opt(sb, JOURNAL_DATA);
3518                 break;
3519
3520         case EXT4_MOUNT_ORDERED_DATA:
3521         case EXT4_MOUNT_WRITEBACK_DATA:
3522                 if (!jbd2_journal_check_available_features
3523                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3524                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3525                                "requested data journaling mode");
3526                         goto failed_mount_wq;
3527                 }
3528         default:
3529                 break;
3530         }
3531         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3532
3533         /*
3534          * The journal may have updated the bg summary counts, so we
3535          * need to update the global counters.
3536          */
3537         percpu_counter_set(&sbi->s_freeblocks_counter,
3538                            ext4_count_free_blocks(sb));
3539         percpu_counter_set(&sbi->s_freeinodes_counter,
3540                            ext4_count_free_inodes(sb));
3541         percpu_counter_set(&sbi->s_dirs_counter,
3542                            ext4_count_dirs(sb));
3543         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3544
3545 no_journal:
3546         /*
3547          * The maximum number of concurrent works can be high and
3548          * concurrency isn't really necessary.  Limit it to 1.
3549          */
3550         EXT4_SB(sb)->dio_unwritten_wq =
3551                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3552         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3553                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3554                 goto failed_mount_wq;
3555         }
3556
3557         /*
3558          * The jbd2_journal_load will have done any necessary log recovery,
3559          * so we can safely mount the rest of the filesystem now.
3560          */
3561
3562         root = ext4_iget(sb, EXT4_ROOT_INO);
3563         if (IS_ERR(root)) {
3564                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3565                 ret = PTR_ERR(root);
3566                 root = NULL;
3567                 goto failed_mount4;
3568         }
3569         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3570                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3571                 goto failed_mount4;
3572         }
3573         sb->s_root = d_alloc_root(root);
3574         if (!sb->s_root) {
3575                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3576                 ret = -ENOMEM;
3577                 goto failed_mount4;
3578         }
3579
3580         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3581
3582         /* determine the minimum size of new large inodes, if present */
3583         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3584                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3585                                                      EXT4_GOOD_OLD_INODE_SIZE;
3586                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3587                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3588                         if (sbi->s_want_extra_isize <
3589                             le16_to_cpu(es->s_want_extra_isize))
3590                                 sbi->s_want_extra_isize =
3591                                         le16_to_cpu(es->s_want_extra_isize);
3592                         if (sbi->s_want_extra_isize <
3593                             le16_to_cpu(es->s_min_extra_isize))
3594                                 sbi->s_want_extra_isize =
3595                                         le16_to_cpu(es->s_min_extra_isize);
3596                 }
3597         }
3598         /* Check if enough inode space is available */
3599         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3600                                                         sbi->s_inode_size) {
3601                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3602                                                        EXT4_GOOD_OLD_INODE_SIZE;
3603                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3604                          "available");
3605         }
3606
3607         if (test_opt(sb, DELALLOC) &&
3608             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3609                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3610                          "requested data journaling mode");
3611                 clear_opt(sb, DELALLOC);
3612         }
3613         if (test_opt(sb, DIOREAD_NOLOCK)) {
3614                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3615                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3616                                 "option - requested data journaling mode");
3617                         clear_opt(sb, DIOREAD_NOLOCK);
3618                 }
3619                 if (sb->s_blocksize < PAGE_SIZE) {
3620                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3621                                 "option - block size is too small");
3622                         clear_opt(sb, DIOREAD_NOLOCK);
3623                 }
3624         }
3625
3626         err = ext4_setup_system_zone(sb);
3627         if (err) {
3628                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3629                          "zone (%d)", err);
3630                 goto failed_mount4;
3631         }
3632
3633         ext4_ext_init(sb);
3634         err = ext4_mb_init(sb, needs_recovery);
3635         if (err) {
3636                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3637                          err);
3638                 goto failed_mount4;
3639         }
3640
3641         err = ext4_register_li_request(sb, first_not_zeroed);
3642         if (err)
3643                 goto failed_mount4;
3644
3645         sbi->s_kobj.kset = ext4_kset;
3646         init_completion(&sbi->s_kobj_unregister);
3647         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3648                                    "%s", sb->s_id);
3649         if (err) {
3650                 ext4_mb_release(sb);
3651                 ext4_ext_release(sb);
3652                 goto failed_mount4;
3653         };
3654
3655         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3656         ext4_orphan_cleanup(sb, es);
3657         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3658         if (needs_recovery) {
3659                 ext4_msg(sb, KERN_INFO, "recovery complete");
3660                 ext4_mark_recovery_complete(sb, es);
3661         }
3662         if (EXT4_SB(sb)->s_journal) {
3663                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3664                         descr = " journalled data mode";
3665                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3666                         descr = " ordered data mode";
3667                 else
3668                         descr = " writeback data mode";
3669         } else
3670                 descr = "out journal";
3671
3672         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3673                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3674                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3675
3676         if (es->s_error_count)
3677                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3678
3679         kfree(orig_data);
3680         return 0;
3681
3682 cantfind_ext4:
3683         if (!silent)
3684                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3685         goto failed_mount;
3686
3687 failed_mount4:
3688         iput(root);
3689         sb->s_root = NULL;
3690         ext4_msg(sb, KERN_ERR, "mount failed");
3691         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3692 failed_mount_wq:
3693         ext4_release_system_zone(sb);
3694         if (sbi->s_journal) {
3695                 jbd2_journal_destroy(sbi->s_journal);
3696                 sbi->s_journal = NULL;
3697         }
3698 failed_mount3:
3699         del_timer(&sbi->s_err_report);
3700         if (sbi->s_flex_groups) {
3701                 if (is_vmalloc_addr(sbi->s_flex_groups))
3702                         vfree(sbi->s_flex_groups);
3703                 else
3704                         kfree(sbi->s_flex_groups);
3705         }
3706         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3707         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3708         percpu_counter_destroy(&sbi->s_dirs_counter);
3709         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3710 failed_mount2:
3711         for (i = 0; i < db_count; i++)
3712                 brelse(sbi->s_group_desc[i]);
3713         kfree(sbi->s_group_desc);
3714 failed_mount:
3715         if (sbi->s_proc) {
3716                 remove_proc_entry(sb->s_id, ext4_proc_root);
3717         }
3718 #ifdef CONFIG_QUOTA
3719         for (i = 0; i < MAXQUOTAS; i++)
3720                 kfree(sbi->s_qf_names[i]);
3721 #endif
3722         ext4_blkdev_remove(sbi);
3723         brelse(bh);
3724 out_fail:
3725         sb->s_fs_info = NULL;
3726         kfree(sbi->s_blockgroup_lock);
3727         kfree(sbi);
3728 out_free_orig:
3729         kfree(orig_data);
3730         return ret;
3731 }
3732
3733 /*
3734  * Setup any per-fs journal parameters now.  We'll do this both on
3735  * initial mount, once the journal has been initialised but before we've
3736  * done any recovery; and again on any subsequent remount.
3737  */
3738 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3739 {
3740         struct ext4_sb_info *sbi = EXT4_SB(sb);
3741
3742         journal->j_commit_interval = sbi->s_commit_interval;
3743         journal->j_min_batch_time = sbi->s_min_batch_time;
3744         journal->j_max_batch_time = sbi->s_max_batch_time;
3745
3746         write_lock(&journal->j_state_lock);
3747         if (test_opt(sb, BARRIER))
3748                 journal->j_flags |= JBD2_BARRIER;
3749         else
3750                 journal->j_flags &= ~JBD2_BARRIER;
3751         if (test_opt(sb, DATA_ERR_ABORT))
3752                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3753         else
3754                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3755         write_unlock(&journal->j_state_lock);
3756 }
3757
3758 static journal_t *ext4_get_journal(struct super_block *sb,
3759                                    unsigned int journal_inum)
3760 {
3761         struct inode *journal_inode;
3762         journal_t *journal;
3763
3764         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3765
3766         /* First, test for the existence of a valid inode on disk.  Bad
3767          * things happen if we iget() an unused inode, as the subsequent
3768          * iput() will try to delete it. */
3769
3770         journal_inode = ext4_iget(sb, journal_inum);
3771         if (IS_ERR(journal_inode)) {
3772                 ext4_msg(sb, KERN_ERR, "no journal found");
3773                 return NULL;
3774         }
3775         if (!journal_inode->i_nlink) {
3776                 make_bad_inode(journal_inode);
3777                 iput(journal_inode);
3778                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3779                 return NULL;
3780         }
3781
3782         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3783                   journal_inode, journal_inode->i_size);
3784         if (!S_ISREG(journal_inode->i_mode)) {
3785                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3786                 iput(journal_inode);
3787                 return NULL;
3788         }
3789
3790         journal = jbd2_journal_init_inode(journal_inode);
3791         if (!journal) {
3792                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3793                 iput(journal_inode);
3794                 return NULL;
3795         }
3796         journal->j_private = sb;
3797         ext4_init_journal_params(sb, journal);
3798         return journal;
3799 }
3800
3801 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3802                                        dev_t j_dev)
3803 {
3804         struct buffer_head *bh;
3805         journal_t *journal;
3806         ext4_fsblk_t start;
3807         ext4_fsblk_t len;
3808         int hblock, blocksize;
3809         ext4_fsblk_t sb_block;
3810         unsigned long offset;
3811         struct ext4_super_block *es;
3812         struct block_device *bdev;
3813
3814         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3815
3816         bdev = ext4_blkdev_get(j_dev, sb);
3817         if (bdev == NULL)
3818                 return NULL;
3819
3820         blocksize = sb->s_blocksize;
3821         hblock = bdev_logical_block_size(bdev);
3822         if (blocksize < hblock) {
3823                 ext4_msg(sb, KERN_ERR,
3824                         "blocksize too small for journal device");
3825                 goto out_bdev;
3826         }
3827
3828         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3829         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3830         set_blocksize(bdev, blocksize);
3831         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3832                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3833                        "external journal");
3834                 goto out_bdev;
3835         }
3836
3837         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3838         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3839             !(le32_to_cpu(es->s_feature_incompat) &
3840               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3841                 ext4_msg(sb, KERN_ERR, "external journal has "
3842                                         "bad superblock");
3843                 brelse(bh);
3844                 goto out_bdev;
3845         }
3846
3847         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3848                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3849                 brelse(bh);
3850                 goto out_bdev;
3851         }
3852
3853         len = ext4_blocks_count(es);
3854         start = sb_block + 1;
3855         brelse(bh);     /* we're done with the superblock */
3856
3857         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3858                                         start, len, blocksize);
3859         if (!journal) {
3860                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3861                 goto out_bdev;
3862         }
3863         journal->j_private = sb;
3864         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3865         wait_on_buffer(journal->j_sb_buffer);
3866         if (!buffer_uptodate(journal->j_sb_buffer)) {
3867                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3868                 goto out_journal;
3869         }
3870         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3871                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3872                                         "user (unsupported) - %d",
3873                         be32_to_cpu(journal->j_superblock->s_nr_users));
3874                 goto out_journal;
3875         }
3876         EXT4_SB(sb)->journal_bdev = bdev;
3877         ext4_init_journal_params(sb, journal);
3878         return journal;
3879
3880 out_journal:
3881         jbd2_journal_destroy(journal);
3882 out_bdev:
3883         ext4_blkdev_put(bdev);
3884         return NULL;
3885 }
3886
3887 static int ext4_load_journal(struct super_block *sb,
3888                              struct ext4_super_block *es,
3889                              unsigned long journal_devnum)
3890 {
3891         journal_t *journal;
3892         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3893         dev_t journal_dev;
3894         int err = 0;
3895         int really_read_only;
3896
3897         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3898
3899         if (journal_devnum &&
3900             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3901                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3902                         "numbers have changed");
3903                 journal_dev = new_decode_dev(journal_devnum);
3904         } else
3905                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3906
3907         really_read_only = bdev_read_only(sb->s_bdev);
3908
3909         /*
3910          * Are we loading a blank journal or performing recovery after a
3911          * crash?  For recovery, we need to check in advance whether we
3912          * can get read-write access to the device.
3913          */
3914         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3915                 if (sb->s_flags & MS_RDONLY) {
3916                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3917                                         "required on readonly filesystem");
3918                         if (really_read_only) {
3919                                 ext4_msg(sb, KERN_ERR, "write access "
3920                                         "unavailable, cannot proceed");
3921                                 return -EROFS;
3922                         }
3923                         ext4_msg(sb, KERN_INFO, "write access will "
3924                                "be enabled during recovery");
3925                 }
3926         }
3927
3928         if (journal_inum && journal_dev) {
3929                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3930                        "and inode journals!");
3931                 return -EINVAL;
3932         }
3933
3934         if (journal_inum) {
3935                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3936                         return -EINVAL;
3937         } else {
3938                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3939                         return -EINVAL;
3940         }
3941
3942         if (!(journal->j_flags & JBD2_BARRIER))
3943                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3944
3945         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3946                 err = jbd2_journal_update_format(journal);
3947                 if (err)  {
3948                         ext4_msg(sb, KERN_ERR, "error updating journal");
3949                         jbd2_journal_destroy(journal);
3950                         return err;
3951                 }
3952         }
3953
3954         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3955                 err = jbd2_journal_wipe(journal, !really_read_only);
3956         if (!err) {
3957                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3958                 if (save)
3959                         memcpy(save, ((char *) es) +
3960                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3961                 err = jbd2_journal_load(journal);
3962                 if (save)
3963                         memcpy(((char *) es) + EXT4_S_ERR_START,
3964                                save, EXT4_S_ERR_LEN);
3965                 kfree(save);
3966         }
3967
3968         if (err) {
3969                 ext4_msg(sb, KERN_ERR, "error loading journal");
3970                 jbd2_journal_destroy(journal);
3971                 return err;
3972         }
3973
3974         EXT4_SB(sb)->s_journal = journal;
3975         ext4_clear_journal_err(sb, es);
3976
3977         if (!really_read_only && journal_devnum &&
3978             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3979                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3980
3981                 /* Make sure we flush the recovery flag to disk. */
3982                 ext4_commit_super(sb, 1);
3983         }
3984
3985         return 0;
3986 }
3987
3988 static int ext4_commit_super(struct super_block *sb, int sync)
3989 {
3990         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3991         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3992         int error = 0;
3993
3994         if (!sbh)
3995                 return error;
3996         if (buffer_write_io_error(sbh)) {
3997                 /*
3998                  * Oh, dear.  A previous attempt to write the
3999                  * superblock failed.  This could happen because the
4000                  * USB device was yanked out.  Or it could happen to
4001                  * be a transient write error and maybe the block will
4002                  * be remapped.  Nothing we can do but to retry the
4003                  * write and hope for the best.
4004                  */
4005                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4006                        "superblock detected");
4007                 clear_buffer_write_io_error(sbh);
4008                 set_buffer_uptodate(sbh);
4009         }
4010         /*
4011          * If the file system is mounted read-only, don't update the
4012          * superblock write time.  This avoids updating the superblock
4013          * write time when we are mounting the root file system
4014          * read/only but we need to replay the journal; at that point,
4015          * for people who are east of GMT and who make their clock
4016          * tick in localtime for Windows bug-for-bug compatibility,
4017          * the clock is set in the future, and this will cause e2fsck
4018          * to complain and force a full file system check.
4019          */
4020         if (!(sb->s_flags & MS_RDONLY))
4021                 es->s_wtime = cpu_to_le32(get_seconds());
4022         if (sb->s_bdev->bd_part)
4023                 es->s_kbytes_written =
4024                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4025                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4026                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4027         else
4028                 es->s_kbytes_written =
4029                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4030         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4031                                            &EXT4_SB(sb)->s_freeblocks_counter));
4032         es->s_free_inodes_count =
4033                 cpu_to_le32(percpu_counter_sum_positive(
4034                                 &EXT4_SB(sb)->s_freeinodes_counter));
4035         sb->s_dirt = 0;
4036         BUFFER_TRACE(sbh, "marking dirty");
4037         mark_buffer_dirty(sbh);
4038         if (sync) {
4039                 error = sync_dirty_buffer(sbh);
4040                 if (error)
4041                         return error;
4042
4043                 error = buffer_write_io_error(sbh);
4044                 if (error) {
4045                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4046                                "superblock");
4047                         clear_buffer_write_io_error(sbh);
4048                         set_buffer_uptodate(sbh);
4049                 }
4050         }
4051         return error;
4052 }
4053
4054 /*
4055  * Have we just finished recovery?  If so, and if we are mounting (or
4056  * remounting) the filesystem readonly, then we will end up with a
4057  * consistent fs on disk.  Record that fact.
4058  */
4059 static void ext4_mark_recovery_complete(struct super_block *sb,
4060                                         struct ext4_super_block *es)
4061 {
4062         journal_t *journal = EXT4_SB(sb)->s_journal;
4063
4064         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4065                 BUG_ON(journal != NULL);
4066                 return;
4067         }
4068         jbd2_journal_lock_updates(journal);
4069         if (jbd2_journal_flush(journal) < 0)
4070                 goto out;
4071
4072         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4073             sb->s_flags & MS_RDONLY) {
4074                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4075                 ext4_commit_super(sb, 1);
4076         }
4077
4078 out:
4079         jbd2_journal_unlock_updates(journal);
4080 }
4081
4082 /*
4083  * If we are mounting (or read-write remounting) a filesystem whose journal
4084  * has recorded an error from a previous lifetime, move that error to the
4085  * main filesystem now.
4086  */
4087 static void ext4_clear_journal_err(struct super_block *sb,
4088                                    struct ext4_super_block *es)
4089 {
4090         journal_t *journal;
4091         int j_errno;
4092         const char *errstr;
4093
4094         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4095
4096         journal = EXT4_SB(sb)->s_journal;
4097
4098         /*
4099          * Now check for any error status which may have been recorded in the
4100          * journal by a prior ext4_error() or ext4_abort()
4101          */
4102
4103         j_errno = jbd2_journal_errno(journal);
4104         if (j_errno) {
4105                 char nbuf[16];
4106
4107                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4108                 ext4_warning(sb, "Filesystem error recorded "
4109                              "from previous mount: %s", errstr);
4110                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4111
4112                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4113                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4114                 ext4_commit_super(sb, 1);
4115
4116                 jbd2_journal_clear_err(journal);
4117         }
4118 }
4119
4120 /*
4121  * Force the running and committing transactions to commit,
4122  * and wait on the commit.
4123  */
4124 int ext4_force_commit(struct super_block *sb)
4125 {
4126         journal_t *journal;
4127         int ret = 0;
4128
4129         if (sb->s_flags & MS_RDONLY)
4130                 return 0;
4131
4132         journal = EXT4_SB(sb)->s_journal;
4133         if (journal) {
4134                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4135                 ret = ext4_journal_force_commit(journal);
4136         }
4137
4138         return ret;
4139 }
4140
4141 static void ext4_write_super(struct super_block *sb)
4142 {
4143         lock_super(sb);
4144         ext4_commit_super(sb, 1);
4145         unlock_super(sb);
4146 }
4147
4148 static int ext4_sync_fs(struct super_block *sb, int wait)
4149 {
4150         int ret = 0;
4151         tid_t target;
4152         struct ext4_sb_info *sbi = EXT4_SB(sb);
4153
4154         trace_ext4_sync_fs(sb, wait);
4155         flush_workqueue(sbi->dio_unwritten_wq);
4156         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4157                 if (wait)
4158                         jbd2_log_wait_commit(sbi->s_journal, target);
4159         }
4160         return ret;
4161 }
4162
4163 /*
4164  * LVM calls this function before a (read-only) snapshot is created.  This
4165  * gives us a chance to flush the journal completely and mark the fs clean.
4166  *
4167  * Note that only this function cannot bring a filesystem to be in a clean
4168  * state independently, because ext4 prevents a new handle from being started
4169  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4170  * the upper layer.
4171  */
4172 static int ext4_freeze(struct super_block *sb)
4173 {
4174         int error = 0;
4175         journal_t *journal;
4176
4177         if (sb->s_flags & MS_RDONLY)
4178                 return 0;
4179
4180         journal = EXT4_SB(sb)->s_journal;
4181
4182         /* Now we set up the journal barrier. */
4183         jbd2_journal_lock_updates(journal);
4184
4185         /*
4186          * Don't clear the needs_recovery flag if we failed to flush
4187          * the journal.
4188          */
4189         error = jbd2_journal_flush(journal);
4190         if (error < 0)
4191                 goto out;
4192
4193         /* Journal blocked and flushed, clear needs_recovery flag. */
4194         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4195         error = ext4_commit_super(sb, 1);
4196 out:
4197         /* we rely on s_frozen to stop further updates */
4198         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4199         return error;
4200 }
4201
4202 /*
4203  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4204  * flag here, even though the filesystem is not technically dirty yet.
4205  */
4206 static int ext4_unfreeze(struct super_block *sb)
4207 {
4208         if (sb->s_flags & MS_RDONLY)
4209                 return 0;
4210
4211         lock_super(sb);
4212         /* Reset the needs_recovery flag before the fs is unlocked. */
4213         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4214         ext4_commit_super(sb, 1);
4215         unlock_super(sb);
4216         return 0;
4217 }
4218
4219 /*
4220  * Structure to save mount options for ext4_remount's benefit
4221  */
4222 struct ext4_mount_options {
4223         unsigned long s_mount_opt;
4224         unsigned long s_mount_opt2;
4225         uid_t s_resuid;
4226         gid_t s_resgid;
4227         unsigned long s_commit_interval;
4228         u32 s_min_batch_time, s_max_batch_time;
4229 #ifdef CONFIG_QUOTA
4230         int s_jquota_fmt;
4231         char *s_qf_names[MAXQUOTAS];
4232 #endif
4233 };
4234
4235 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4236 {
4237         struct ext4_super_block *es;
4238         struct ext4_sb_info *sbi = EXT4_SB(sb);
4239         ext4_fsblk_t n_blocks_count = 0;
4240         unsigned long old_sb_flags;
4241         struct ext4_mount_options old_opts;
4242         int enable_quota = 0;
4243         ext4_group_t g;
4244         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4245         int err;
4246 #ifdef CONFIG_QUOTA
4247         int i;
4248 #endif
4249         char *orig_data = kstrdup(data, GFP_KERNEL);
4250
4251         /* Store the original options */
4252         lock_super(sb);
4253         old_sb_flags = sb->s_flags;
4254         old_opts.s_mount_opt = sbi->s_mount_opt;
4255         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4256         old_opts.s_resuid = sbi->s_resuid;
4257         old_opts.s_resgid = sbi->s_resgid;
4258         old_opts.s_commit_interval = sbi->s_commit_interval;
4259         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4260         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4261 #ifdef CONFIG_QUOTA
4262         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4263         for (i = 0; i < MAXQUOTAS; i++)
4264                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4265 #endif
4266         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4267                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4268
4269         /*
4270          * Allow the "check" option to be passed as a remount option.
4271          */
4272         if (!parse_options(data, sb, NULL, &journal_ioprio,
4273                            &n_blocks_count, 1)) {
4274                 err = -EINVAL;
4275                 goto restore_opts;
4276         }
4277
4278         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4279                 ext4_abort(sb, "Abort forced by user");
4280
4281         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4282                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4283
4284         es = sbi->s_es;
4285
4286         if (sbi->s_journal) {
4287                 ext4_init_journal_params(sb, sbi->s_journal);
4288                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4289         }
4290
4291         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4292                 n_blocks_count > ext4_blocks_count(es)) {
4293                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4294                         err = -EROFS;
4295                         goto restore_opts;
4296                 }
4297
4298                 if (*flags & MS_RDONLY) {
4299                         err = dquot_suspend(sb, -1);
4300                         if (err < 0)
4301                                 goto restore_opts;
4302
4303                         /*
4304                          * First of all, the unconditional stuff we have to do
4305                          * to disable replay of the journal when we next remount
4306                          */
4307                         sb->s_flags |= MS_RDONLY;
4308
4309                         /*
4310                          * OK, test if we are remounting a valid rw partition
4311                          * readonly, and if so set the rdonly flag and then
4312                          * mark the partition as valid again.
4313                          */
4314                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4315                             (sbi->s_mount_state & EXT4_VALID_FS))
4316                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4317
4318                         if (sbi->s_journal)
4319                                 ext4_mark_recovery_complete(sb, es);
4320                 } else {
4321                         /* Make sure we can mount this feature set readwrite */
4322                         if (!ext4_feature_set_ok(sb, 0)) {
4323                                 err = -EROFS;
4324                                 goto restore_opts;
4325                         }
4326                         /*
4327                          * Make sure the group descriptor checksums
4328                          * are sane.  If they aren't, refuse to remount r/w.
4329                          */
4330                         for (g = 0; g < sbi->s_groups_count; g++) {
4331                                 struct ext4_group_desc *gdp =
4332                                         ext4_get_group_desc(sb, g, NULL);
4333
4334                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4335                                         ext4_msg(sb, KERN_ERR,
4336                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4337                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4338                                                le16_to_cpu(gdp->bg_checksum));
4339                                         err = -EINVAL;
4340                                         goto restore_opts;
4341                                 }
4342                         }
4343
4344                         /*
4345                          * If we have an unprocessed orphan list hanging
4346                          * around from a previously readonly bdev mount,
4347                          * require a full umount/remount for now.
4348                          */
4349                         if (es->s_last_orphan) {
4350                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4351                                        "remount RDWR because of unprocessed "
4352                                        "orphan inode list.  Please "
4353                                        "umount/remount instead");
4354                                 err = -EINVAL;
4355                                 goto restore_opts;
4356                         }
4357
4358                         /*
4359                          * Mounting a RDONLY partition read-write, so reread
4360                          * and store the current valid flag.  (It may have
4361                          * been changed by e2fsck since we originally mounted
4362                          * the partition.)
4363                          */
4364                         if (sbi->s_journal)
4365                                 ext4_clear_journal_err(sb, es);
4366                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4367                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4368                                 goto restore_opts;
4369                         if (!ext4_setup_super(sb, es, 0))
4370                                 sb->s_flags &= ~MS_RDONLY;
4371                         enable_quota = 1;
4372                 }
4373         }
4374
4375         /*
4376          * Reinitialize lazy itable initialization thread based on
4377          * current settings
4378          */
4379         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4380                 ext4_unregister_li_request(sb);
4381         else {
4382                 ext4_group_t first_not_zeroed;
4383                 first_not_zeroed = ext4_has_uninit_itable(sb);
4384                 ext4_register_li_request(sb, first_not_zeroed);
4385         }
4386
4387         ext4_setup_system_zone(sb);
4388         if (sbi->s_journal == NULL)
4389                 ext4_commit_super(sb, 1);
4390
4391 #ifdef CONFIG_QUOTA
4392         /* Release old quota file names */
4393         for (i = 0; i < MAXQUOTAS; i++)
4394                 if (old_opts.s_qf_names[i] &&
4395                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4396                         kfree(old_opts.s_qf_names[i]);
4397 #endif
4398         unlock_super(sb);
4399         if (enable_quota)
4400                 dquot_resume(sb, -1);
4401
4402         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4403         kfree(orig_data);
4404         return 0;
4405
4406 restore_opts:
4407         sb->s_flags = old_sb_flags;
4408         sbi->s_mount_opt = old_opts.s_mount_opt;
4409         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4410         sbi->s_resuid = old_opts.s_resuid;
4411         sbi->s_resgid = old_opts.s_resgid;
4412         sbi->s_commit_interval = old_opts.s_commit_interval;
4413         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4414         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4415 #ifdef CONFIG_QUOTA
4416         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4417         for (i = 0; i < MAXQUOTAS; i++) {
4418                 if (sbi->s_qf_names[i] &&
4419                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4420                         kfree(sbi->s_qf_names[i]);
4421                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4422         }
4423 #endif
4424         unlock_super(sb);
4425         kfree(orig_data);
4426         return err;
4427 }
4428
4429 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4430 {
4431         struct super_block *sb = dentry->d_sb;
4432         struct ext4_sb_info *sbi = EXT4_SB(sb);
4433         struct ext4_super_block *es = sbi->s_es;
4434         u64 fsid;
4435
4436         if (test_opt(sb, MINIX_DF)) {
4437                 sbi->s_overhead_last = 0;
4438         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4439                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4440                 ext4_fsblk_t overhead = 0;
4441
4442                 /*
4443                  * Compute the overhead (FS structures).  This is constant
4444                  * for a given filesystem unless the number of block groups
4445                  * changes so we cache the previous value until it does.
4446                  */
4447
4448                 /*
4449                  * All of the blocks before first_data_block are
4450                  * overhead
4451                  */
4452                 overhead = le32_to_cpu(es->s_first_data_block);
4453
4454                 /*
4455                  * Add the overhead attributed to the superblock and
4456                  * block group descriptors.  If the sparse superblocks
4457                  * feature is turned on, then not all groups have this.
4458                  */
4459                 for (i = 0; i < ngroups; i++) {
4460                         overhead += ext4_bg_has_super(sb, i) +
4461                                 ext4_bg_num_gdb(sb, i);
4462                         cond_resched();
4463                 }
4464
4465                 /*
4466                  * Every block group has an inode bitmap, a block
4467                  * bitmap, and an inode table.
4468                  */
4469                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4470                 sbi->s_overhead_last = overhead;
4471                 smp_wmb();
4472                 sbi->s_blocks_last = ext4_blocks_count(es);
4473         }
4474
4475         buf->f_type = EXT4_SUPER_MAGIC;
4476         buf->f_bsize = sb->s_blocksize;
4477         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4478         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4479                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4480         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4481         if (buf->f_bfree < ext4_r_blocks_count(es))
4482                 buf->f_bavail = 0;
4483         buf->f_files = le32_to_cpu(es->s_inodes_count);
4484         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4485         buf->f_namelen = EXT4_NAME_LEN;
4486         fsid = le64_to_cpup((void *)es->s_uuid) ^
4487                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4488         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4489         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4490
4491         return 0;
4492 }
4493
4494 /* Helper function for writing quotas on sync - we need to start transaction
4495  * before quota file is locked for write. Otherwise the are possible deadlocks:
4496  * Process 1                         Process 2
4497  * ext4_create()                     quota_sync()
4498  *   jbd2_journal_start()                  write_dquot()
4499  *   dquot_initialize()                         down(dqio_mutex)
4500  *     down(dqio_mutex)                    jbd2_journal_start()
4501  *
4502  */
4503
4504 #ifdef CONFIG_QUOTA
4505
4506 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4507 {
4508         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4509 }
4510
4511 static int ext4_write_dquot(struct dquot *dquot)
4512 {
4513         int ret, err;
4514         handle_t *handle;
4515         struct inode *inode;
4516
4517         inode = dquot_to_inode(dquot);
4518         handle = ext4_journal_start(inode,
4519                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4520         if (IS_ERR(handle))
4521                 return PTR_ERR(handle);
4522         ret = dquot_commit(dquot);
4523         err = ext4_journal_stop(handle);
4524         if (!ret)
4525                 ret = err;
4526         return ret;
4527 }
4528
4529 static int ext4_acquire_dquot(struct dquot *dquot)
4530 {
4531         int ret, err;
4532         handle_t *handle;
4533
4534         handle = ext4_journal_start(dquot_to_inode(dquot),
4535                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4536         if (IS_ERR(handle))
4537                 return PTR_ERR(handle);
4538         ret = dquot_acquire(dquot);
4539         err = ext4_journal_stop(handle);
4540         if (!ret)
4541                 ret = err;
4542         return ret;
4543 }
4544
4545 static int ext4_release_dquot(struct dquot *dquot)
4546 {
4547         int ret, err;
4548         handle_t *handle;
4549
4550         handle = ext4_journal_start(dquot_to_inode(dquot),
4551                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4552         if (IS_ERR(handle)) {
4553                 /* Release dquot anyway to avoid endless cycle in dqput() */
4554                 dquot_release(dquot);
4555                 return PTR_ERR(handle);
4556         }
4557         ret = dquot_release(dquot);
4558         err = ext4_journal_stop(handle);
4559         if (!ret)
4560                 ret = err;
4561         return ret;
4562 }
4563
4564 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4565 {
4566         /* Are we journaling quotas? */
4567         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4568             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4569                 dquot_mark_dquot_dirty(dquot);
4570                 return ext4_write_dquot(dquot);
4571         } else {
4572                 return dquot_mark_dquot_dirty(dquot);
4573         }
4574 }
4575
4576 static int ext4_write_info(struct super_block *sb, int type)
4577 {
4578         int ret, err;
4579         handle_t *handle;
4580
4581         /* Data block + inode block */
4582         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4583         if (IS_ERR(handle))
4584                 return PTR_ERR(handle);
4585         ret = dquot_commit_info(sb, type);
4586         err = ext4_journal_stop(handle);
4587         if (!ret)
4588                 ret = err;
4589         return ret;
4590 }
4591
4592 /*
4593  * Turn on quotas during mount time - we need to find
4594  * the quota file and such...
4595  */
4596 static int ext4_quota_on_mount(struct super_block *sb, int type)
4597 {
4598         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4599                                         EXT4_SB(sb)->s_jquota_fmt, type);
4600 }
4601
4602 /*
4603  * Standard function to be called on quota_on
4604  */
4605 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4606                          struct path *path)
4607 {
4608         int err;
4609
4610         if (!test_opt(sb, QUOTA))
4611                 return -EINVAL;
4612
4613         /* Quotafile not on the same filesystem? */
4614         if (path->mnt->mnt_sb != sb)
4615                 return -EXDEV;
4616         /* Journaling quota? */
4617         if (EXT4_SB(sb)->s_qf_names[type]) {
4618                 /* Quotafile not in fs root? */
4619                 if (path->dentry->d_parent != sb->s_root)
4620                         ext4_msg(sb, KERN_WARNING,
4621                                 "Quota file not on filesystem root. "
4622                                 "Journaled quota will not work");
4623         }
4624
4625         /*
4626          * When we journal data on quota file, we have to flush journal to see
4627          * all updates to the file when we bypass pagecache...
4628          */
4629         if (EXT4_SB(sb)->s_journal &&
4630             ext4_should_journal_data(path->dentry->d_inode)) {
4631                 /*
4632                  * We don't need to lock updates but journal_flush() could
4633                  * otherwise be livelocked...
4634                  */
4635                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4636                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4637                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4638                 if (err)
4639                         return err;
4640         }
4641
4642         return dquot_quota_on(sb, type, format_id, path);
4643 }
4644
4645 static int ext4_quota_off(struct super_block *sb, int type)
4646 {
4647         struct inode *inode = sb_dqopt(sb)->files[type];
4648         handle_t *handle;
4649
4650         /* Force all delayed allocation blocks to be allocated.
4651          * Caller already holds s_umount sem */
4652         if (test_opt(sb, DELALLOC))
4653                 sync_filesystem(sb);
4654
4655         /* Update modification times of quota files when userspace can
4656          * start looking at them */
4657         handle = ext4_journal_start(inode, 1);
4658         if (IS_ERR(handle))
4659                 goto out;
4660         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4661         ext4_mark_inode_dirty(handle, inode);
4662         ext4_journal_stop(handle);
4663
4664 out:
4665         return dquot_quota_off(sb, type);
4666 }
4667
4668 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4669  * acquiring the locks... As quota files are never truncated and quota code
4670  * itself serializes the operations (and no one else should touch the files)
4671  * we don't have to be afraid of races */
4672 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4673                                size_t len, loff_t off)
4674 {
4675         struct inode *inode = sb_dqopt(sb)->files[type];
4676         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4677         int err = 0;
4678         int offset = off & (sb->s_blocksize - 1);
4679         int tocopy;
4680         size_t toread;
4681         struct buffer_head *bh;
4682         loff_t i_size = i_size_read(inode);
4683
4684         if (off > i_size)
4685                 return 0;
4686         if (off+len > i_size)
4687                 len = i_size-off;
4688         toread = len;
4689         while (toread > 0) {
4690                 tocopy = sb->s_blocksize - offset < toread ?
4691                                 sb->s_blocksize - offset : toread;
4692                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4693                 if (err)
4694                         return err;
4695                 if (!bh)        /* A hole? */
4696                         memset(data, 0, tocopy);
4697                 else
4698                         memcpy(data, bh->b_data+offset, tocopy);
4699                 brelse(bh);
4700                 offset = 0;
4701                 toread -= tocopy;
4702                 data += tocopy;
4703                 blk++;
4704         }
4705         return len;
4706 }
4707
4708 /* Write to quotafile (we know the transaction is already started and has
4709  * enough credits) */
4710 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4711                                 const char *data, size_t len, loff_t off)
4712 {
4713         struct inode *inode = sb_dqopt(sb)->files[type];
4714         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4715         int err = 0;
4716         int offset = off & (sb->s_blocksize - 1);
4717         struct buffer_head *bh;
4718         handle_t *handle = journal_current_handle();
4719
4720         if (EXT4_SB(sb)->s_journal && !handle) {
4721                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4722                         " cancelled because transaction is not started",
4723                         (unsigned long long)off, (unsigned long long)len);
4724                 return -EIO;
4725         }
4726         /*
4727          * Since we account only one data block in transaction credits,
4728          * then it is impossible to cross a block boundary.
4729          */
4730         if (sb->s_blocksize - offset < len) {
4731                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4732                         " cancelled because not block aligned",
4733                         (unsigned long long)off, (unsigned long long)len);
4734                 return -EIO;
4735         }
4736
4737         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4738         bh = ext4_bread(handle, inode, blk, 1, &err);
4739         if (!bh)
4740                 goto out;
4741         err = ext4_journal_get_write_access(handle, bh);
4742         if (err) {
4743                 brelse(bh);
4744                 goto out;
4745         }
4746         lock_buffer(bh);
4747         memcpy(bh->b_data+offset, data, len);
4748         flush_dcache_page(bh->b_page);
4749         unlock_buffer(bh);
4750         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4751         brelse(bh);
4752 out:
4753         if (err) {
4754                 mutex_unlock(&inode->i_mutex);
4755                 return err;
4756         }
4757         if (inode->i_size < off + len) {
4758                 i_size_write(inode, off + len);
4759                 EXT4_I(inode)->i_disksize = inode->i_size;
4760                 ext4_mark_inode_dirty(handle, inode);
4761         }
4762         mutex_unlock(&inode->i_mutex);
4763         return len;
4764 }
4765
4766 #endif
4767
4768 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4769                        const char *dev_name, void *data)
4770 {
4771         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4772 }
4773
4774 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4775 static struct file_system_type ext2_fs_type = {
4776         .owner          = THIS_MODULE,
4777         .name           = "ext2",
4778         .mount          = ext4_mount,
4779         .kill_sb        = kill_block_super,
4780         .fs_flags       = FS_REQUIRES_DEV,
4781 };
4782
4783 static inline void register_as_ext2(void)
4784 {
4785         int err = register_filesystem(&ext2_fs_type);
4786         if (err)
4787                 printk(KERN_WARNING
4788                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4789 }
4790
4791 static inline void unregister_as_ext2(void)
4792 {
4793         unregister_filesystem(&ext2_fs_type);
4794 }
4795 MODULE_ALIAS("ext2");
4796 #else
4797 static inline void register_as_ext2(void) { }
4798 static inline void unregister_as_ext2(void) { }
4799 #endif
4800
4801 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4802 static inline void register_as_ext3(void)
4803 {
4804         int err = register_filesystem(&ext3_fs_type);
4805         if (err)
4806                 printk(KERN_WARNING
4807                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4808 }
4809
4810 static inline void unregister_as_ext3(void)
4811 {
4812         unregister_filesystem(&ext3_fs_type);
4813 }
4814 MODULE_ALIAS("ext3");
4815 #else
4816 static inline void register_as_ext3(void) { }
4817 static inline void unregister_as_ext3(void) { }
4818 #endif
4819
4820 static struct file_system_type ext4_fs_type = {
4821         .owner          = THIS_MODULE,
4822         .name           = "ext4",
4823         .mount          = ext4_mount,
4824         .kill_sb        = kill_block_super,
4825         .fs_flags       = FS_REQUIRES_DEV,
4826 };
4827
4828 static int __init ext4_init_feat_adverts(void)
4829 {
4830         struct ext4_features *ef;
4831         int ret = -ENOMEM;
4832
4833         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4834         if (!ef)
4835                 goto out;
4836
4837         ef->f_kobj.kset = ext4_kset;
4838         init_completion(&ef->f_kobj_unregister);
4839         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4840                                    "features");
4841         if (ret) {
4842                 kfree(ef);
4843                 goto out;
4844         }
4845
4846         ext4_feat = ef;
4847         ret = 0;
4848 out:
4849         return ret;
4850 }
4851
4852 static void ext4_exit_feat_adverts(void)
4853 {
4854         kobject_put(&ext4_feat->f_kobj);
4855         wait_for_completion(&ext4_feat->f_kobj_unregister);
4856         kfree(ext4_feat);
4857 }
4858
4859 /* Shared across all ext4 file systems */
4860 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4861 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4862
4863 static int __init ext4_init_fs(void)
4864 {
4865         int i, err;
4866
4867         ext4_check_flag_values();
4868
4869         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4870                 mutex_init(&ext4__aio_mutex[i]);
4871                 init_waitqueue_head(&ext4__ioend_wq[i]);
4872         }
4873
4874         err = ext4_init_pageio();
4875         if (err)
4876                 return err;
4877         err = ext4_init_system_zone();
4878         if (err)
4879                 goto out7;
4880         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4881         if (!ext4_kset)
4882                 goto out6;
4883         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4884         if (!ext4_proc_root)
4885                 goto out5;
4886
4887         err = ext4_init_feat_adverts();
4888         if (err)
4889                 goto out4;
4890
4891         err = ext4_init_mballoc();
4892         if (err)
4893                 goto out3;
4894
4895         err = ext4_init_xattr();
4896         if (err)
4897                 goto out2;
4898         err = init_inodecache();
4899         if (err)
4900                 goto out1;
4901         register_as_ext2();
4902         register_as_ext3();
4903         err = register_filesystem(&ext4_fs_type);
4904         if (err)
4905                 goto out;
4906
4907         ext4_li_info = NULL;
4908         mutex_init(&ext4_li_mtx);
4909         return 0;
4910 out:
4911         unregister_as_ext2();
4912         unregister_as_ext3();
4913         destroy_inodecache();
4914 out1:
4915         ext4_exit_xattr();
4916 out2:
4917         ext4_exit_mballoc();
4918 out3:
4919         ext4_exit_feat_adverts();
4920 out4:
4921         remove_proc_entry("fs/ext4", NULL);
4922 out5:
4923         kset_unregister(ext4_kset);
4924 out6:
4925         ext4_exit_system_zone();
4926 out7:
4927         ext4_exit_pageio();
4928         return err;
4929 }
4930
4931 static void __exit ext4_exit_fs(void)
4932 {
4933         ext4_destroy_lazyinit_thread();
4934         unregister_as_ext2();
4935         unregister_as_ext3();
4936         unregister_filesystem(&ext4_fs_type);
4937         destroy_inodecache();
4938         ext4_exit_xattr();
4939         ext4_exit_mballoc();
4940         ext4_exit_feat_adverts();
4941         remove_proc_entry("fs/ext4", NULL);
4942         kset_unregister(ext4_kset);
4943         ext4_exit_system_zone();
4944         ext4_exit_pageio();
4945 }
4946
4947 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4948 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4949 MODULE_LICENSE("GPL");
4950 module_init(ext4_init_fs)
4951 module_exit(ext4_exit_fs)