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