fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434
 435 /* Deal with the reporting of failure conditions on a filesystem such as
 436  * inconsistencies detected or read IO failures.
 437  *
 438  * On ext2, we can store the error state of the filesystem in the
 439  * superblock.  That is not possible on ext4, because we may have other
 440  * write ordering constraints on the superblock which prevent us from
 441  * writing it out straight away; and given that the journal is about to
 442  * be aborted, we can't rely on the current, or future, transactions to
 443  * write out the superblock safely.
 444  *
 445  * We'll just use the jbd2_journal_abort() error code to record an error in
 446  * the journal instead.  On recovery, the journal will complain about
 447  * that error until we've noted it down and cleared it.
 448  */
 449
 450 static void ext4_handle_error(struct super_block *sb)
 451 {
 452         if (sb->s_flags & MS_RDONLY)
 453                 return;
 454
 455         if (!test_opt(sb, ERRORS_CONT)) {
 456                 journal_t *journal = EXT4_SB(sb)->s_journal;
 457
 458                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 459                 if (journal)
 460                         jbd2_journal_abort(journal, -EIO);
 461         }
 462         if (test_opt(sb, ERRORS_RO)) {
 463                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 464                 sb->s_flags |= MS_RDONLY;
 465         }
 466         if (test_opt(sb, ERRORS_PANIC))
 467                 panic("EXT4-fs (device %s): panic forced after error\n",
 468                         sb->s_id);
 469 }
 470
 471 void __ext4_error(struct super_block *sb, const char *function,
 472                   unsigned int line, const char *fmt, ...)
 473 {
 474         struct va_format vaf;
 475         va_list args;
 476
 477         va_start(args, fmt);
 478         vaf.fmt = fmt;
 479         vaf.va = &args;
 480         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 481                sb->s_id, function, line, current->comm, &vaf);
 482         va_end(args);
 483         save_error_info(sb, function, line);
 484
 485         ext4_handle_error(sb);
 486 }
 487
 488 void ext4_error_inode(struct inode *inode, const char *function,
 489                       unsigned int line, ext4_fsblk_t block,
 490                       const char *fmt, ...)
 491 {
 492         va_list args;
 493         struct va_format vaf;
 494         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 495
 496         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 497         es->s_last_error_block = cpu_to_le64(block);
 498         save_error_info(inode->i_sb, function, line);
 499         va_start(args, fmt);
 500         vaf.fmt = fmt;
 501         vaf.va = &args;
 502         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 503                inode->i_sb->s_id, function, line, inode->i_ino);
 504         if (block)
 505                 printk(KERN_CONT "block %llu: ", block);
 506         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 507         va_end(args);
 508
 509         ext4_handle_error(inode->i_sb);
 510 }
 511
 512 void ext4_error_file(struct file *file, const char *function,
 513                      unsigned int line, ext4_fsblk_t block,
 514                      const char *fmt, ...)
 515 {
 516         va_list args;
 517         struct va_format vaf;
 518         struct ext4_super_block *es;
 519         struct inode *inode = file->f_dentry->d_inode;
 520         char pathname[80], *path;
 521
 522         es = EXT4_SB(inode->i_sb)->s_es;
 523         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 524         save_error_info(inode->i_sb, function, line);
 525         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 526         if (IS_ERR(path))
 527                 path = "(unknown)";
 528         printk(KERN_CRIT
 529                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 530                inode->i_sb->s_id, function, line, inode->i_ino);
 531         if (block)
 532                 printk(KERN_CONT "block %llu: ", block);
 533         va_start(args, fmt);
 534         vaf.fmt = fmt;
 535         vaf.va = &args;
 536         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 537         va_end(args);
 538
 539         ext4_handle_error(inode->i_sb);
 540 }
 541
 542 static const char *ext4_decode_error(struct super_block *sb, int errno,
 543                                      char nbuf[16])
 544 {
 545         char *errstr = NULL;
 546
 547         switch (errno) {
 548         case -EIO:
 549                 errstr = "IO failure";
 550                 break;
 551         case -ENOMEM:
 552                 errstr = "Out of memory";
 553                 break;
 554         case -EROFS:
 555                 if (!sb || (EXT4_SB(sb)->s_journal &&
 556                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 557                         errstr = "Journal has aborted";
 558                 else
 559                         errstr = "Readonly filesystem";
 560                 break;
 561         default:
 562                 /* If the caller passed in an extra buffer for unknown
 563                  * errors, textualise them now.  Else we just return
 564                  * NULL. */
 565                 if (nbuf) {
 566                         /* Check for truncated error codes... */
 567                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 568                                 errstr = nbuf;
 569                 }
 570                 break;
 571         }
 572
 573         return errstr;
 574 }
 575
 576 /* __ext4_std_error decodes expected errors from journaling functions
 577  * automatically and invokes the appropriate error response.  */
 578
 579 void __ext4_std_error(struct super_block *sb, const char *function,
 580                       unsigned int line, int errno)
 581 {
 582         char nbuf[16];
 583         const char *errstr;
 584
 585         /* Special case: if the error is EROFS, and we're not already
 586          * inside a transaction, then there's really no point in logging
 587          * an error. */
 588         if (errno == -EROFS && journal_current_handle() == NULL &&
 589             (sb->s_flags & MS_RDONLY))
 590                 return;
 591
 592         errstr = ext4_decode_error(sb, errno, nbuf);
 593         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 594                sb->s_id, function, line, errstr);
 595         save_error_info(sb, function, line);
 596
 597         ext4_handle_error(sb);
 598 }
 599
 600 /*
 601  * ext4_abort is a much stronger failure handler than ext4_error.  The
 602  * abort function may be used to deal with unrecoverable failures such
 603  * as journal IO errors or ENOMEM at a critical moment in log management.
 604  *
 605  * We unconditionally force the filesystem into an ABORT|READONLY state,
 606  * unless the error response on the fs has been set to panic in which
 607  * case we take the easy way out and panic immediately.
 608  */
 609
 610 void __ext4_abort(struct super_block *sb, const char *function,
 611                 unsigned int line, const char *fmt, ...)
 612 {
 613         va_list args;
 614
 615         save_error_info(sb, function, line);
 616         va_start(args, fmt);
 617         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 618                function, line);
 619         vprintk(fmt, args);
 620         printk("\n");
 621         va_end(args);
 622
 623         if ((sb->s_flags & MS_RDONLY) == 0) {
 624                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 625                 sb->s_flags |= MS_RDONLY;
 626                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 627                 if (EXT4_SB(sb)->s_journal)
 628                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 629                 save_error_info(sb, function, line);
 630         }
 631         if (test_opt(sb, ERRORS_PANIC))
 632                 panic("EXT4-fs panic from previous error\n");
 633 }
 634
 635 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 636 {
 637         struct va_format vaf;
 638         va_list args;
 639
 640         va_start(args, fmt);
 641         vaf.fmt = fmt;
 642         vaf.va = &args;
 643         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 644         va_end(args);
 645 }
 646
 647 void __ext4_warning(struct super_block *sb, const char *function,
 648                     unsigned int line, const char *fmt, ...)
 649 {
 650         struct va_format vaf;
 651         va_list args;
 652
 653         va_start(args, fmt);
 654         vaf.fmt = fmt;
 655         vaf.va = &args;
 656         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 657                sb->s_id, function, line, &vaf);
 658         va_end(args);
 659 }
 660
 661 void __ext4_grp_locked_error(const char *function, unsigned int line,
 662                              struct super_block *sb, ext4_group_t grp,
 663                              unsigned long ino, ext4_fsblk_t block,
 664                              const char *fmt, ...)
 665 __releases(bitlock)
 666 __acquires(bitlock)
 667 {
 668         struct va_format vaf;
 669         va_list args;
 670         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 671
 672         es->s_last_error_ino = cpu_to_le32(ino);
 673         es->s_last_error_block = cpu_to_le64(block);
 674         __save_error_info(sb, function, line);
 675
 676         va_start(args, fmt);
 677
 678         vaf.fmt = fmt;
 679         vaf.va = &args;
 680         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 681                sb->s_id, function, line, grp);
 682         if (ino)
 683                 printk(KERN_CONT "inode %lu: ", ino);
 684         if (block)
 685                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 686         printk(KERN_CONT "%pV\n", &vaf);
 687         va_end(args);
 688
 689         if (test_opt(sb, ERRORS_CONT)) {
 690                 ext4_commit_super(sb, 0);
 691                 return;
 692         }
 693
 694         ext4_unlock_group(sb, grp);
 695         ext4_handle_error(sb);
 696         /*
 697          * We only get here in the ERRORS_RO case; relocking the group
 698          * may be dangerous, but nothing bad will happen since the
 699          * filesystem will have already been marked read/only and the
 700          * journal has been aborted.  We return 1 as a hint to callers
 701          * who might what to use the return value from
 702          * ext4_grp_locked_error() to distinguish between the
 703          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 704          * aggressively from the ext4 function in question, with a
 705          * more appropriate error code.
 706          */
 707         ext4_lock_group(sb, grp);
 708         return;
 709 }
 710
 711 void ext4_update_dynamic_rev(struct super_block *sb)
 712 {
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 716                 return;
 717
 718         ext4_warning(sb,
 719                      "updating to rev %d because of new feature flag, "
 720                      "running e2fsck is recommended",
 721                      EXT4_DYNAMIC_REV);
 722
 723         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 724         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 725         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 726         /* leave es->s_feature_*compat flags alone */
 727         /* es->s_uuid will be set by e2fsck if empty */
 728
 729         /*
 730          * The rest of the superblock fields should be zero, and if not it
 731          * means they are likely already in use, so leave them alone.  We
 732          * can leave it up to e2fsck to clean up any inconsistencies there.
 733          */
 734 }
 735
 736 /*
 737  * Open the external journal device
 738  */
 739 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 740 {
 741         struct block_device *bdev;
 742         char b[BDEVNAME_SIZE];
 743
 744         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 745         if (IS_ERR(bdev))
 746                 goto fail;
 747         return bdev;
 748
 749 fail:
 750         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 751                         __bdevname(dev, b), PTR_ERR(bdev));
 752         return NULL;
 753 }
 754
 755 /*
 756  * Release the journal device
 757  */
 758 static int ext4_blkdev_put(struct block_device *bdev)
 759 {
 760         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 761 }
 762
 763 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 764 {
 765         struct block_device *bdev;
 766         int ret = -ENODEV;
 767
 768         bdev = sbi->journal_bdev;
 769         if (bdev) {
 770                 ret = ext4_blkdev_put(bdev);
 771                 sbi->journal_bdev = NULL;
 772         }
 773         return ret;
 774 }
 775
 776 static inline struct inode *orphan_list_entry(struct list_head *l)
 777 {
 778         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 779 }
 780
 781 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 782 {
 783         struct list_head *l;
 784
 785         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 786                  le32_to_cpu(sbi->s_es->s_last_orphan));
 787
 788         printk(KERN_ERR "sb_info orphan list:\n");
 789         list_for_each(l, &sbi->s_orphan) {
 790                 struct inode *inode = orphan_list_entry(l);
 791                 printk(KERN_ERR "  "
 792                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 793                        inode->i_sb->s_id, inode->i_ino, inode,
 794                        inode->i_mode, inode->i_nlink,
 795                        NEXT_ORPHAN(inode));
 796         }
 797 }
 798
 799 static void ext4_put_super(struct super_block *sb)
 800 {
 801         struct ext4_sb_info *sbi = EXT4_SB(sb);
 802         struct ext4_super_block *es = sbi->s_es;
 803         int i, err;
 804
 805         ext4_unregister_li_request(sb);
 806         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 807
 808         flush_workqueue(sbi->dio_unwritten_wq);
 809         destroy_workqueue(sbi->dio_unwritten_wq);
 810
 811         lock_super(sb);
 812         if (sb->s_dirt)
 813                 ext4_commit_super(sb, 1);
 814
 815         if (sbi->s_journal) {
 816                 err = jbd2_journal_destroy(sbi->s_journal);
 817                 sbi->s_journal = NULL;
 818                 if (err < 0)
 819                         ext4_abort(sb, "Couldn't clean up the journal");
 820         }
 821
 822         del_timer_sync(&sbi->s_err_report);
 823         ext4_release_system_zone(sb);
 824         ext4_mb_release(sb);
 825         ext4_ext_release(sb);
 826         ext4_xattr_put_super(sb);
 827
 828         if (!(sb->s_flags & MS_RDONLY)) {
 829                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 830                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 831                 ext4_commit_super(sb, 1);
 832         }
 833         if (sbi->s_proc) {
 834                 remove_proc_entry(sb->s_id, ext4_proc_root);
 835         }
 836         kobject_del(&sbi->s_kobj);
 837
 838         for (i = 0; i < sbi->s_gdb_count; i++)
 839                 brelse(sbi->s_group_desc[i]);
 840         ext4_kvfree(sbi->s_group_desc);
 841         ext4_kvfree(sbi->s_flex_groups);
 842         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 843         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 844         percpu_counter_destroy(&sbi->s_dirs_counter);
 845         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 846         brelse(sbi->s_sbh);
 847 #ifdef CONFIG_QUOTA
 848         for (i = 0; i < MAXQUOTAS; i++)
 849                 kfree(sbi->s_qf_names[i]);
 850 #endif
 851
 852         /* Debugging code just in case the in-memory inode orphan list
 853          * isn't empty.  The on-disk one can be non-empty if we've
 854          * detected an error and taken the fs readonly, but the
 855          * in-memory list had better be clean by this point. */
 856         if (!list_empty(&sbi->s_orphan))
 857                 dump_orphan_list(sb, sbi);
 858         J_ASSERT(list_empty(&sbi->s_orphan));
 859
 860         sync_blockdev(sb->s_bdev);
 861         invalidate_bdev(sb->s_bdev);
 862         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 863                 /*
 864                  * Invalidate the journal device's buffers.  We don't want them
 865                  * floating about in memory - the physical journal device may
 866                  * hotswapped, and it breaks the `ro-after' testing code.
 867                  */
 868                 sync_blockdev(sbi->journal_bdev);
 869                 invalidate_bdev(sbi->journal_bdev);
 870                 ext4_blkdev_remove(sbi);
 871         }
 872         if (sbi->s_mmp_tsk)
 873                 kthread_stop(sbi->s_mmp_tsk);
 874         sb->s_fs_info = NULL;
 875         /*
 876          * Now that we are completely done shutting down the
 877          * superblock, we need to actually destroy the kobject.
 878          */
 879         unlock_super(sb);
 880         kobject_put(&sbi->s_kobj);
 881         wait_for_completion(&sbi->s_kobj_unregister);
 882         kfree(sbi->s_blockgroup_lock);
 883         kfree(sbi);
 884 }
 885
 886 static struct kmem_cache *ext4_inode_cachep;
 887
 888 /*
 889  * Called inside transaction, so use GFP_NOFS
 890  */
 891 static struct inode *ext4_alloc_inode(struct super_block *sb)
 892 {
 893         struct ext4_inode_info *ei;
 894
 895         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 896         if (!ei)
 897                 return NULL;
 898
 899         ei->vfs_inode.i_version = 1;
 900         ei->vfs_inode.i_data.writeback_index = 0;
 901         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 902         INIT_LIST_HEAD(&ei->i_prealloc_list);
 903         spin_lock_init(&ei->i_prealloc_lock);
 904         ei->i_reserved_data_blocks = 0;
 905         ei->i_reserved_meta_blocks = 0;
 906         ei->i_allocated_meta_blocks = 0;
 907         ei->i_da_metadata_calc_len = 0;
 908         ei->i_da_metadata_calc_last_lblock = 0;
 909         spin_lock_init(&(ei->i_block_reservation_lock));
 910 #ifdef CONFIG_QUOTA
 911         ei->i_reserved_quota = 0;
 912 #endif
 913         ei->jinode = NULL;
 914         INIT_LIST_HEAD(&ei->i_completed_io_list);
 915         spin_lock_init(&ei->i_completed_io_lock);
 916         ei->cur_aio_dio = NULL;
 917         ei->i_sync_tid = 0;
 918         ei->i_datasync_tid = 0;
 919         atomic_set(&ei->i_ioend_count, 0);
 920         atomic_set(&ei->i_aiodio_unwritten, 0);
 921
 922         return &ei->vfs_inode;
 923 }
 924
 925 static int ext4_drop_inode(struct inode *inode)
 926 {
 927         int drop = generic_drop_inode(inode);
 928
 929         trace_ext4_drop_inode(inode, drop);
 930         return drop;
 931 }
 932
 933 static void ext4_i_callback(struct rcu_head *head)
 934 {
 935         struct inode *inode = container_of(head, struct inode, i_rcu);
 936         INIT_LIST_HEAD(&inode->i_dentry);
 937         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 938 }
 939
 940 static void ext4_destroy_inode(struct inode *inode)
 941 {
 942         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 943                 ext4_msg(inode->i_sb, KERN_ERR,
 944                          "Inode %lu (%p): orphan list check failed!",
 945                          inode->i_ino, EXT4_I(inode));
 946                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 947                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 948                                 true);
 949                 dump_stack();
 950         }
 951         call_rcu(&inode->i_rcu, ext4_i_callback);
 952 }
 953
 954 static void init_once(void *foo)
 955 {
 956         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 957
 958         INIT_LIST_HEAD(&ei->i_orphan);
 959 #ifdef CONFIG_EXT4_FS_XATTR
 960         init_rwsem(&ei->xattr_sem);
 961 #endif
 962         init_rwsem(&ei->i_data_sem);
 963         inode_init_once(&ei->vfs_inode);
 964 }
 965
 966 static int init_inodecache(void)
 967 {
 968         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 969                                              sizeof(struct ext4_inode_info),
 970                                              0, (SLAB_RECLAIM_ACCOUNT|
 971                                                 SLAB_MEM_SPREAD),
 972                                              init_once);
 973         if (ext4_inode_cachep == NULL)
 974                 return -ENOMEM;
 975         return 0;
 976 }
 977
 978 static void destroy_inodecache(void)
 979 {
 980         kmem_cache_destroy(ext4_inode_cachep);
 981 }
 982
 983 void ext4_clear_inode(struct inode *inode)
 984 {
 985         invalidate_inode_buffers(inode);
 986         end_writeback(inode);
 987         dquot_drop(inode);
 988         ext4_discard_preallocations(inode);
 989         if (EXT4_I(inode)->jinode) {
 990                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 991                                                EXT4_I(inode)->jinode);
 992                 jbd2_free_inode(EXT4_I(inode)->jinode);
 993                 EXT4_I(inode)->jinode = NULL;
 994         }
 995 }
 996
 997 static inline void ext4_show_quota_options(struct seq_file *seq,
 998                                            struct super_block *sb)
 999 {
1000 #if defined(CONFIG_QUOTA)
1001         struct ext4_sb_info *sbi = EXT4_SB(sb);
1002
1003         if (sbi->s_jquota_fmt) {
1004                 char *fmtname = "";
1005
1006                 switch (sbi->s_jquota_fmt) {
1007                 case QFMT_VFS_OLD:
1008                         fmtname = "vfsold";
1009                         break;
1010                 case QFMT_VFS_V0:
1011                         fmtname = "vfsv0";
1012                         break;
1013                 case QFMT_VFS_V1:
1014                         fmtname = "vfsv1";
1015                         break;
1016                 }
1017                 seq_printf(seq, ",jqfmt=%s", fmtname);
1018         }
1019
1020         if (sbi->s_qf_names[USRQUOTA])
1021                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1022
1023         if (sbi->s_qf_names[GRPQUOTA])
1024                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1025
1026         if (test_opt(sb, USRQUOTA))
1027                 seq_puts(seq, ",usrquota");
1028
1029         if (test_opt(sb, GRPQUOTA))
1030                 seq_puts(seq, ",grpquota");
1031 #endif
1032 }
1033
1034 /*
1035  * Show an option if
1036  *  - it's set to a non-default value OR
1037  *  - if the per-sb default is different from the global default
1038  */
1039 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1040 {
1041         int def_errors;
1042         unsigned long def_mount_opts;
1043         struct super_block *sb = vfs->mnt_sb;
1044         struct ext4_sb_info *sbi = EXT4_SB(sb);
1045         struct ext4_super_block *es = sbi->s_es;
1046
1047         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1048         def_errors     = le16_to_cpu(es->s_errors);
1049
1050         if (sbi->s_sb_block != 1)
1051                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1052         if (test_opt(sb, MINIX_DF))
1053                 seq_puts(seq, ",minixdf");
1054         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1055                 seq_puts(seq, ",grpid");
1056         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1057                 seq_puts(seq, ",nogrpid");
1058         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1059             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1060                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1061         }
1062         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1063             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1064                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1065         }
1066         if (test_opt(sb, ERRORS_RO)) {
1067                 if (def_errors == EXT4_ERRORS_PANIC ||
1068                     def_errors == EXT4_ERRORS_CONTINUE) {
1069                         seq_puts(seq, ",errors=remount-ro");
1070                 }
1071         }
1072         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1073                 seq_puts(seq, ",errors=continue");
1074         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1075                 seq_puts(seq, ",errors=panic");
1076         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1077                 seq_puts(seq, ",nouid32");
1078         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1079                 seq_puts(seq, ",debug");
1080 #ifdef CONFIG_EXT4_FS_XATTR
1081         if (test_opt(sb, XATTR_USER))
1082                 seq_puts(seq, ",user_xattr");
1083         if (!test_opt(sb, XATTR_USER))
1084                 seq_puts(seq, ",nouser_xattr");
1085 #endif
1086 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1087         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1088                 seq_puts(seq, ",acl");
1089         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1090                 seq_puts(seq, ",noacl");
1091 #endif
1092         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1093                 seq_printf(seq, ",commit=%u",
1094                            (unsigned) (sbi->s_commit_interval / HZ));
1095         }
1096         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1097                 seq_printf(seq, ",min_batch_time=%u",
1098                            (unsigned) sbi->s_min_batch_time);
1099         }
1100         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1101                 seq_printf(seq, ",max_batch_time=%u",
1102                            (unsigned) sbi->s_max_batch_time);
1103         }
1104
1105         /*
1106          * We're changing the default of barrier mount option, so
1107          * let's always display its mount state so it's clear what its
1108          * status is.
1109          */
1110         seq_puts(seq, ",barrier=");
1111         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1112         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1113                 seq_puts(seq, ",journal_async_commit");
1114         else if (test_opt(sb, JOURNAL_CHECKSUM))
1115                 seq_puts(seq, ",journal_checksum");
1116         if (test_opt(sb, I_VERSION))
1117                 seq_puts(seq, ",i_version");
1118         if (!test_opt(sb, DELALLOC) &&
1119             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1120                 seq_puts(seq, ",nodelalloc");
1121
1122         if (!test_opt(sb, MBLK_IO_SUBMIT))
1123                 seq_puts(seq, ",nomblk_io_submit");
1124         if (sbi->s_stripe)
1125                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1126         /*
1127          * journal mode get enabled in different ways
1128          * So just print the value even if we didn't specify it
1129          */
1130         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1131                 seq_puts(seq, ",data=journal");
1132         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1133                 seq_puts(seq, ",data=ordered");
1134         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1135                 seq_puts(seq, ",data=writeback");
1136
1137         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1138                 seq_printf(seq, ",inode_readahead_blks=%u",
1139                            sbi->s_inode_readahead_blks);
1140
1141         if (test_opt(sb, DATA_ERR_ABORT))
1142                 seq_puts(seq, ",data_err=abort");
1143
1144         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1145                 seq_puts(seq, ",noauto_da_alloc");
1146
1147         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1148                 seq_puts(seq, ",discard");
1149
1150         if (test_opt(sb, NOLOAD))
1151                 seq_puts(seq, ",norecovery");
1152
1153         if (test_opt(sb, DIOREAD_NOLOCK))
1154                 seq_puts(seq, ",dioread_nolock");
1155
1156         if (test_opt(sb, BLOCK_VALIDITY) &&
1157             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1158                 seq_puts(seq, ",block_validity");
1159
1160         if (!test_opt(sb, INIT_INODE_TABLE))
1161                 seq_puts(seq, ",noinit_itable");
1162         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1163                 seq_printf(seq, ",init_itable=%u",
1164                            (unsigned) sbi->s_li_wait_mult);
1165
1166         ext4_show_quota_options(seq, sb);
1167
1168         return 0;
1169 }
1170
1171 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1172                                         u64 ino, u32 generation)
1173 {
1174         struct inode *inode;
1175
1176         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1177                 return ERR_PTR(-ESTALE);
1178         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1179                 return ERR_PTR(-ESTALE);
1180
1181         /* iget isn't really right if the inode is currently unallocated!!
1182          *
1183          * ext4_read_inode will return a bad_inode if the inode had been
1184          * deleted, so we should be safe.
1185          *
1186          * Currently we don't know the generation for parent directory, so
1187          * a generation of 0 means "accept any"
1188          */
1189         inode = ext4_iget_normal(sb, ino);
1190         if (IS_ERR(inode))
1191                 return ERR_CAST(inode);
1192         if (generation && inode->i_generation != generation) {
1193                 iput(inode);
1194                 return ERR_PTR(-ESTALE);
1195         }
1196
1197         return inode;
1198 }
1199
1200 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1201                                         int fh_len, int fh_type)
1202 {
1203         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1204                                     ext4_nfs_get_inode);
1205 }
1206
1207 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1208                                         int fh_len, int fh_type)
1209 {
1210         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1211                                     ext4_nfs_get_inode);
1212 }
1213
1214 /*
1215  * Try to release metadata pages (indirect blocks, directories) which are
1216  * mapped via the block device.  Since these pages could have journal heads
1217  * which would prevent try_to_free_buffers() from freeing them, we must use
1218  * jbd2 layer's try_to_free_buffers() function to release them.
1219  */
1220 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1221                                  gfp_t wait)
1222 {
1223         journal_t *journal = EXT4_SB(sb)->s_journal;
1224
1225         WARN_ON(PageChecked(page));
1226         if (!page_has_buffers(page))
1227                 return 0;
1228         if (journal)
1229                 return jbd2_journal_try_to_free_buffers(journal, page,
1230                                                         wait & ~__GFP_WAIT);
1231         return try_to_free_buffers(page);
1232 }
1233
1234 #ifdef CONFIG_QUOTA
1235 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1236 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1237
1238 static int ext4_write_dquot(struct dquot *dquot);
1239 static int ext4_acquire_dquot(struct dquot *dquot);
1240 static int ext4_release_dquot(struct dquot *dquot);
1241 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1242 static int ext4_write_info(struct super_block *sb, int type);
1243 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1244                          struct path *path);
1245 static int ext4_quota_off(struct super_block *sb, int type);
1246 static int ext4_quota_on_mount(struct super_block *sb, int type);
1247 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1248                                size_t len, loff_t off);
1249 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1250                                 const char *data, size_t len, loff_t off);
1251
1252 static const struct dquot_operations ext4_quota_operations = {
1253         .get_reserved_space = ext4_get_reserved_space,
1254         .write_dquot    = ext4_write_dquot,
1255         .acquire_dquot  = ext4_acquire_dquot,
1256         .release_dquot  = ext4_release_dquot,
1257         .mark_dirty     = ext4_mark_dquot_dirty,
1258         .write_info     = ext4_write_info,
1259         .alloc_dquot    = dquot_alloc,
1260         .destroy_dquot  = dquot_destroy,
1261 };
1262
1263 static const struct quotactl_ops ext4_qctl_operations = {
1264         .quota_on       = ext4_quota_on,
1265         .quota_off      = ext4_quota_off,
1266         .quota_sync     = dquot_quota_sync,
1267         .get_info       = dquot_get_dqinfo,
1268         .set_info       = dquot_set_dqinfo,
1269         .get_dqblk      = dquot_get_dqblk,
1270         .set_dqblk      = dquot_set_dqblk
1271 };
1272 #endif
1273
1274 static const struct super_operations ext4_sops = {
1275         .alloc_inode    = ext4_alloc_inode,
1276         .destroy_inode  = ext4_destroy_inode,
1277         .write_inode    = ext4_write_inode,
1278         .dirty_inode    = ext4_dirty_inode,
1279         .drop_inode     = ext4_drop_inode,
1280         .evict_inode    = ext4_evict_inode,
1281         .put_super      = ext4_put_super,
1282         .sync_fs        = ext4_sync_fs,
1283         .freeze_fs      = ext4_freeze,
1284         .unfreeze_fs    = ext4_unfreeze,
1285         .statfs         = ext4_statfs,
1286         .remount_fs     = ext4_remount,
1287         .show_options   = ext4_show_options,
1288 #ifdef CONFIG_QUOTA
1289         .quota_read     = ext4_quota_read,
1290         .quota_write    = ext4_quota_write,
1291 #endif
1292         .bdev_try_to_free_page = bdev_try_to_free_page,
1293 };
1294
1295 static const struct super_operations ext4_nojournal_sops = {
1296         .alloc_inode    = ext4_alloc_inode,
1297         .destroy_inode  = ext4_destroy_inode,
1298         .write_inode    = ext4_write_inode,
1299         .dirty_inode    = ext4_dirty_inode,
1300         .drop_inode     = ext4_drop_inode,
1301         .evict_inode    = ext4_evict_inode,
1302         .write_super    = ext4_write_super,
1303         .put_super      = ext4_put_super,
1304         .statfs         = ext4_statfs,
1305         .remount_fs     = ext4_remount,
1306         .show_options   = ext4_show_options,
1307 #ifdef CONFIG_QUOTA
1308         .quota_read     = ext4_quota_read,
1309         .quota_write    = ext4_quota_write,
1310 #endif
1311         .bdev_try_to_free_page = bdev_try_to_free_page,
1312 };
1313
1314 static const struct export_operations ext4_export_ops = {
1315         .fh_to_dentry = ext4_fh_to_dentry,
1316         .fh_to_parent = ext4_fh_to_parent,
1317         .get_parent = ext4_get_parent,
1318 };
1319
1320 enum {
1321         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1322         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1323         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1324         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1325         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1326         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1327         Opt_journal_update, Opt_journal_dev,
1328         Opt_journal_checksum, Opt_journal_async_commit,
1329         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1330         Opt_data_err_abort, Opt_data_err_ignore,
1331         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1332         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1333         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1334         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1335         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1336         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1337         Opt_inode_readahead_blks, Opt_journal_ioprio,
1338         Opt_dioread_nolock, Opt_dioread_lock,
1339         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1340 };
1341
1342 static const match_table_t tokens = {
1343         {Opt_bsd_df, "bsddf"},
1344         {Opt_minix_df, "minixdf"},
1345         {Opt_grpid, "grpid"},
1346         {Opt_grpid, "bsdgroups"},
1347         {Opt_nogrpid, "nogrpid"},
1348         {Opt_nogrpid, "sysvgroups"},
1349         {Opt_resgid, "resgid=%u"},
1350         {Opt_resuid, "resuid=%u"},
1351         {Opt_sb, "sb=%u"},
1352         {Opt_err_cont, "errors=continue"},
1353         {Opt_err_panic, "errors=panic"},
1354         {Opt_err_ro, "errors=remount-ro"},
1355         {Opt_nouid32, "nouid32"},
1356         {Opt_debug, "debug"},
1357         {Opt_oldalloc, "oldalloc"},
1358         {Opt_orlov, "orlov"},
1359         {Opt_user_xattr, "user_xattr"},
1360         {Opt_nouser_xattr, "nouser_xattr"},
1361         {Opt_acl, "acl"},
1362         {Opt_noacl, "noacl"},
1363         {Opt_noload, "noload"},
1364         {Opt_noload, "norecovery"},
1365         {Opt_nobh, "nobh"},
1366         {Opt_bh, "bh"},
1367         {Opt_commit, "commit=%u"},
1368         {Opt_min_batch_time, "min_batch_time=%u"},
1369         {Opt_max_batch_time, "max_batch_time=%u"},
1370         {Opt_journal_update, "journal=update"},
1371         {Opt_journal_dev, "journal_dev=%u"},
1372         {Opt_journal_checksum, "journal_checksum"},
1373         {Opt_journal_async_commit, "journal_async_commit"},
1374         {Opt_abort, "abort"},
1375         {Opt_data_journal, "data=journal"},
1376         {Opt_data_ordered, "data=ordered"},
1377         {Opt_data_writeback, "data=writeback"},
1378         {Opt_data_err_abort, "data_err=abort"},
1379         {Opt_data_err_ignore, "data_err=ignore"},
1380         {Opt_offusrjquota, "usrjquota="},
1381         {Opt_usrjquota, "usrjquota=%s"},
1382         {Opt_offgrpjquota, "grpjquota="},
1383         {Opt_grpjquota, "grpjquota=%s"},
1384         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1385         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1386         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1387         {Opt_grpquota, "grpquota"},
1388         {Opt_noquota, "noquota"},
1389         {Opt_quota, "quota"},
1390         {Opt_usrquota, "usrquota"},
1391         {Opt_barrier, "barrier=%u"},
1392         {Opt_barrier, "barrier"},
1393         {Opt_nobarrier, "nobarrier"},
1394         {Opt_i_version, "i_version"},
1395         {Opt_stripe, "stripe=%u"},
1396         {Opt_resize, "resize"},
1397         {Opt_delalloc, "delalloc"},
1398         {Opt_nodelalloc, "nodelalloc"},
1399         {Opt_mblk_io_submit, "mblk_io_submit"},
1400         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1401         {Opt_block_validity, "block_validity"},
1402         {Opt_noblock_validity, "noblock_validity"},
1403         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1404         {Opt_journal_ioprio, "journal_ioprio=%u"},
1405         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1406         {Opt_auto_da_alloc, "auto_da_alloc"},
1407         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1408         {Opt_dioread_nolock, "dioread_nolock"},
1409         {Opt_dioread_lock, "dioread_lock"},
1410         {Opt_discard, "discard"},
1411         {Opt_nodiscard, "nodiscard"},
1412         {Opt_init_itable, "init_itable=%u"},
1413         {Opt_init_itable, "init_itable"},
1414         {Opt_noinit_itable, "noinit_itable"},
1415         {Opt_err, NULL},
1416 };
1417
1418 static ext4_fsblk_t get_sb_block(void **data)
1419 {
1420         ext4_fsblk_t    sb_block;
1421         char            *options = (char *) *data;
1422
1423         if (!options || strncmp(options, "sb=", 3) != 0)
1424                 return 1;       /* Default location */
1425
1426         options += 3;
1427         /* TODO: use simple_strtoll with >32bit ext4 */
1428         sb_block = simple_strtoul(options, &options, 0);
1429         if (*options && *options != ',') {
1430                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1431                        (char *) *data);
1432                 return 1;
1433         }
1434         if (*options == ',')
1435                 options++;
1436         *data = (void *) options;
1437
1438         return sb_block;
1439 }
1440
1441 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1442 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1443         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1444
1445 #ifdef CONFIG_QUOTA
1446 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1447 {
1448         struct ext4_sb_info *sbi = EXT4_SB(sb);
1449         char *qname;
1450
1451         if (sb_any_quota_loaded(sb) &&
1452                 !sbi->s_qf_names[qtype]) {
1453                 ext4_msg(sb, KERN_ERR,
1454                         "Cannot change journaled "
1455                         "quota options when quota turned on");
1456                 return 0;
1457         }
1458         qname = match_strdup(args);
1459         if (!qname) {
1460                 ext4_msg(sb, KERN_ERR,
1461                         "Not enough memory for storing quotafile name");
1462                 return 0;
1463         }
1464         if (sbi->s_qf_names[qtype] &&
1465                 strcmp(sbi->s_qf_names[qtype], qname)) {
1466                 ext4_msg(sb, KERN_ERR,
1467                         "%s quota file already specified", QTYPE2NAME(qtype));
1468                 kfree(qname);
1469                 return 0;
1470         }
1471         sbi->s_qf_names[qtype] = qname;
1472         if (strchr(sbi->s_qf_names[qtype], '/')) {
1473                 ext4_msg(sb, KERN_ERR,
1474                         "quotafile must be on filesystem root");
1475                 kfree(sbi->s_qf_names[qtype]);
1476                 sbi->s_qf_names[qtype] = NULL;
1477                 return 0;
1478         }
1479         set_opt(sb, QUOTA);
1480         return 1;
1481 }
1482
1483 static int clear_qf_name(struct super_block *sb, int qtype)
1484 {
1485
1486         struct ext4_sb_info *sbi = EXT4_SB(sb);
1487
1488         if (sb_any_quota_loaded(sb) &&
1489                 sbi->s_qf_names[qtype]) {
1490                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1491                         " when quota turned on");
1492                 return 0;
1493         }
1494         /*
1495          * The space will be released later when all options are confirmed
1496          * to be correct
1497          */
1498         sbi->s_qf_names[qtype] = NULL;
1499         return 1;
1500 }
1501 #endif
1502
1503 static int parse_options(char *options, struct super_block *sb,
1504                          unsigned long *journal_devnum,
1505                          unsigned int *journal_ioprio,
1506                          ext4_fsblk_t *n_blocks_count, int is_remount)
1507 {
1508         struct ext4_sb_info *sbi = EXT4_SB(sb);
1509         char *p;
1510         substring_t args[MAX_OPT_ARGS];
1511         int data_opt = 0;
1512         int option;
1513 #ifdef CONFIG_QUOTA
1514         int qfmt;
1515 #endif
1516
1517         if (!options)
1518                 return 1;
1519
1520         while ((p = strsep(&options, ",")) != NULL) {
1521                 int token;
1522                 if (!*p)
1523                         continue;
1524
1525                 /*
1526                  * Initialize args struct so we know whether arg was
1527                  * found; some options take optional arguments.
1528                  */
1529                 args[0].to = args[0].from = NULL;
1530                 token = match_token(p, tokens, args);
1531                 switch (token) {
1532                 case Opt_bsd_df:
1533                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1534                         clear_opt(sb, MINIX_DF);
1535                         break;
1536                 case Opt_minix_df:
1537                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1538                         set_opt(sb, MINIX_DF);
1539
1540                         break;
1541                 case Opt_grpid:
1542                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1543                         set_opt(sb, GRPID);
1544
1545                         break;
1546                 case Opt_nogrpid:
1547                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1548                         clear_opt(sb, GRPID);
1549
1550                         break;
1551                 case Opt_resuid:
1552                         if (match_int(&args[0], &option))
1553                                 return 0;
1554                         sbi->s_resuid = option;
1555                         break;
1556                 case Opt_resgid:
1557                         if (match_int(&args[0], &option))
1558                                 return 0;
1559                         sbi->s_resgid = option;
1560                         break;
1561                 case Opt_sb:
1562                         /* handled by get_sb_block() instead of here */
1563                         /* *sb_block = match_int(&args[0]); */
1564                         break;
1565                 case Opt_err_panic:
1566                         clear_opt(sb, ERRORS_CONT);
1567                         clear_opt(sb, ERRORS_RO);
1568                         set_opt(sb, ERRORS_PANIC);
1569                         break;
1570                 case Opt_err_ro:
1571                         clear_opt(sb, ERRORS_CONT);
1572                         clear_opt(sb, ERRORS_PANIC);
1573                         set_opt(sb, ERRORS_RO);
1574                         break;
1575                 case Opt_err_cont:
1576                         clear_opt(sb, ERRORS_RO);
1577                         clear_opt(sb, ERRORS_PANIC);
1578                         set_opt(sb, ERRORS_CONT);
1579                         break;
1580                 case Opt_nouid32:
1581                         set_opt(sb, NO_UID32);
1582                         break;
1583                 case Opt_debug:
1584                         set_opt(sb, DEBUG);
1585                         break;
1586                 case Opt_oldalloc:
1587                         ext4_msg(sb, KERN_WARNING,
1588                                  "Ignoring deprecated oldalloc option");
1589                         break;
1590                 case Opt_orlov:
1591                         ext4_msg(sb, KERN_WARNING,
1592                                  "Ignoring deprecated orlov option");
1593                         break;
1594 #ifdef CONFIG_EXT4_FS_XATTR
1595                 case Opt_user_xattr:
1596                         set_opt(sb, XATTR_USER);
1597                         break;
1598                 case Opt_nouser_xattr:
1599                         clear_opt(sb, XATTR_USER);
1600                         break;
1601 #else
1602                 case Opt_user_xattr:
1603                 case Opt_nouser_xattr:
1604                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1605                         break;
1606 #endif
1607 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1608                 case Opt_acl:
1609                         set_opt(sb, POSIX_ACL);
1610                         break;
1611                 case Opt_noacl:
1612                         clear_opt(sb, POSIX_ACL);
1613                         break;
1614 #else
1615                 case Opt_acl:
1616                 case Opt_noacl:
1617                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1618                         break;
1619 #endif
1620                 case Opt_journal_update:
1621                         /* @@@ FIXME */
1622                         /* Eventually we will want to be able to create
1623                            a journal file here.  For now, only allow the
1624                            user to specify an existing inode to be the
1625                            journal file. */
1626                         if (is_remount) {
1627                                 ext4_msg(sb, KERN_ERR,
1628                                          "Cannot specify journal on remount");
1629                                 return 0;
1630                         }
1631                         set_opt(sb, UPDATE_JOURNAL);
1632                         break;
1633                 case Opt_journal_dev:
1634                         if (is_remount) {
1635                                 ext4_msg(sb, KERN_ERR,
1636                                         "Cannot specify journal on remount");
1637                                 return 0;
1638                         }
1639                         if (match_int(&args[0], &option))
1640                                 return 0;
1641                         *journal_devnum = option;
1642                         break;
1643                 case Opt_journal_checksum:
1644                         set_opt(sb, JOURNAL_CHECKSUM);
1645                         break;
1646                 case Opt_journal_async_commit:
1647                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1648                         set_opt(sb, JOURNAL_CHECKSUM);
1649                         break;
1650                 case Opt_noload:
1651                         set_opt(sb, NOLOAD);
1652                         break;
1653                 case Opt_commit:
1654                         if (match_int(&args[0], &option))
1655                                 return 0;
1656                         if (option < 0)
1657                                 return 0;
1658                         if (option == 0)
1659                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1660                         sbi->s_commit_interval = HZ * option;
1661                         break;
1662                 case Opt_max_batch_time:
1663                         if (match_int(&args[0], &option))
1664                                 return 0;
1665                         if (option < 0)
1666                                 return 0;
1667                         sbi->s_max_batch_time = option;
1668                         break;
1669                 case Opt_min_batch_time:
1670                         if (match_int(&args[0], &option))
1671                                 return 0;
1672                         if (option < 0)
1673                                 return 0;
1674                         sbi->s_min_batch_time = option;
1675                         break;
1676                 case Opt_data_journal:
1677                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1678                         goto datacheck;
1679                 case Opt_data_ordered:
1680                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1681                         goto datacheck;
1682                 case Opt_data_writeback:
1683                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1684                 datacheck:
1685                         if (is_remount) {
1686                                 if (!sbi->s_journal)
1687                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1688                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1689                                         ext4_msg(sb, KERN_ERR,
1690                                                 "Cannot change data mode on remount");
1691                                         return 0;
1692                                 }
1693                         } else {
1694                                 clear_opt(sb, DATA_FLAGS);
1695                                 sbi->s_mount_opt |= data_opt;
1696                         }
1697                         break;
1698                 case Opt_data_err_abort:
1699                         set_opt(sb, DATA_ERR_ABORT);
1700                         break;
1701                 case Opt_data_err_ignore:
1702                         clear_opt(sb, DATA_ERR_ABORT);
1703                         break;
1704 #ifdef CONFIG_QUOTA
1705                 case Opt_usrjquota:
1706                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1707                                 return 0;
1708                         break;
1709                 case Opt_grpjquota:
1710                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1711                                 return 0;
1712                         break;
1713                 case Opt_offusrjquota:
1714                         if (!clear_qf_name(sb, USRQUOTA))
1715                                 return 0;
1716                         break;
1717                 case Opt_offgrpjquota:
1718                         if (!clear_qf_name(sb, GRPQUOTA))
1719                                 return 0;
1720                         break;
1721
1722                 case Opt_jqfmt_vfsold:
1723                         qfmt = QFMT_VFS_OLD;
1724                         goto set_qf_format;
1725                 case Opt_jqfmt_vfsv0:
1726                         qfmt = QFMT_VFS_V0;
1727                         goto set_qf_format;
1728                 case Opt_jqfmt_vfsv1:
1729                         qfmt = QFMT_VFS_V1;
1730 set_qf_format:
1731                         if (sb_any_quota_loaded(sb) &&
1732                             sbi->s_jquota_fmt != qfmt) {
1733                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1734                                         "journaled quota options when "
1735                                         "quota turned on");
1736                                 return 0;
1737                         }
1738                         sbi->s_jquota_fmt = qfmt;
1739                         break;
1740                 case Opt_quota:
1741                 case Opt_usrquota:
1742                         set_opt(sb, QUOTA);
1743                         set_opt(sb, USRQUOTA);
1744                         break;
1745                 case Opt_grpquota:
1746                         set_opt(sb, QUOTA);
1747                         set_opt(sb, GRPQUOTA);
1748                         break;
1749                 case Opt_noquota:
1750                         if (sb_any_quota_loaded(sb)) {
1751                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1752                                         "options when quota turned on");
1753                                 return 0;
1754                         }
1755                         clear_opt(sb, QUOTA);
1756                         clear_opt(sb, USRQUOTA);
1757                         clear_opt(sb, GRPQUOTA);
1758                         break;
1759 #else
1760                 case Opt_quota:
1761                 case Opt_usrquota:
1762                 case Opt_grpquota:
1763                         ext4_msg(sb, KERN_ERR,
1764                                 "quota options not supported");
1765                         break;
1766                 case Opt_usrjquota:
1767                 case Opt_grpjquota:
1768                 case Opt_offusrjquota:
1769                 case Opt_offgrpjquota:
1770                 case Opt_jqfmt_vfsold:
1771                 case Opt_jqfmt_vfsv0:
1772                 case Opt_jqfmt_vfsv1:
1773                         ext4_msg(sb, KERN_ERR,
1774                                 "journaled quota options not supported");
1775                         break;
1776                 case Opt_noquota:
1777                         break;
1778 #endif
1779                 case Opt_abort:
1780                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1781                         break;
1782                 case Opt_nobarrier:
1783                         clear_opt(sb, BARRIER);
1784                         break;
1785                 case Opt_barrier:
1786                         if (args[0].from) {
1787                                 if (match_int(&args[0], &option))
1788                                         return 0;
1789                         } else
1790                                 option = 1;     /* No argument, default to 1 */
1791                         if (option)
1792                                 set_opt(sb, BARRIER);
1793                         else
1794                                 clear_opt(sb, BARRIER);
1795                         break;
1796                 case Opt_ignore:
1797                         break;
1798                 case Opt_resize:
1799                         if (!is_remount) {
1800                                 ext4_msg(sb, KERN_ERR,
1801                                         "resize option only available "
1802                                         "for remount");
1803                                 return 0;
1804                         }
1805                         if (match_int(&args[0], &option) != 0)
1806                                 return 0;
1807                         *n_blocks_count = option;
1808                         break;
1809                 case Opt_nobh:
1810                         ext4_msg(sb, KERN_WARNING,
1811                                  "Ignoring deprecated nobh option");
1812                         break;
1813                 case Opt_bh:
1814                         ext4_msg(sb, KERN_WARNING,
1815                                  "Ignoring deprecated bh option");
1816                         break;
1817                 case Opt_i_version:
1818                         set_opt(sb, I_VERSION);
1819                         sb->s_flags |= MS_I_VERSION;
1820                         break;
1821                 case Opt_nodelalloc:
1822                         clear_opt(sb, DELALLOC);
1823                         clear_opt2(sb, EXPLICIT_DELALLOC);
1824                         break;
1825                 case Opt_mblk_io_submit:
1826                         set_opt(sb, MBLK_IO_SUBMIT);
1827                         break;
1828                 case Opt_nomblk_io_submit:
1829                         clear_opt(sb, MBLK_IO_SUBMIT);
1830                         break;
1831                 case Opt_stripe:
1832                         if (match_int(&args[0], &option))
1833                                 return 0;
1834                         if (option < 0)
1835                                 return 0;
1836                         sbi->s_stripe = option;
1837                         break;
1838                 case Opt_delalloc:
1839                         set_opt(sb, DELALLOC);
1840                         set_opt2(sb, EXPLICIT_DELALLOC);
1841                         break;
1842                 case Opt_block_validity:
1843                         set_opt(sb, BLOCK_VALIDITY);
1844                         break;
1845                 case Opt_noblock_validity:
1846                         clear_opt(sb, BLOCK_VALIDITY);
1847                         break;
1848                 case Opt_inode_readahead_blks:
1849                         if (match_int(&args[0], &option))
1850                                 return 0;
1851                         if (option < 0 || option > (1 << 30))
1852                                 return 0;
1853                         if (option && !is_power_of_2(option)) {
1854                                 ext4_msg(sb, KERN_ERR,
1855                                          "EXT4-fs: inode_readahead_blks"
1856                                          " must be a power of 2");
1857                                 return 0;
1858                         }
1859                         sbi->s_inode_readahead_blks = option;
1860                         break;
1861                 case Opt_journal_ioprio:
1862                         if (match_int(&args[0], &option))
1863                                 return 0;
1864                         if (option < 0 || option > 7)
1865                                 break;
1866                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1867                                                             option);
1868                         break;
1869                 case Opt_noauto_da_alloc:
1870                         set_opt(sb, NO_AUTO_DA_ALLOC);
1871                         break;
1872                 case Opt_auto_da_alloc:
1873                         if (args[0].from) {
1874                                 if (match_int(&args[0], &option))
1875                                         return 0;
1876                         } else
1877                                 option = 1;     /* No argument, default to 1 */
1878                         if (option)
1879                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1880                         else
1881                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1882                         break;
1883                 case Opt_discard:
1884                         set_opt(sb, DISCARD);
1885                         break;
1886                 case Opt_nodiscard:
1887                         clear_opt(sb, DISCARD);
1888                         break;
1889                 case Opt_dioread_nolock:
1890                         set_opt(sb, DIOREAD_NOLOCK);
1891                         break;
1892                 case Opt_dioread_lock:
1893                         clear_opt(sb, DIOREAD_NOLOCK);
1894                         break;
1895                 case Opt_init_itable:
1896                         set_opt(sb, INIT_INODE_TABLE);
1897                         if (args[0].from) {
1898                                 if (match_int(&args[0], &option))
1899                                         return 0;
1900                         } else
1901                                 option = EXT4_DEF_LI_WAIT_MULT;
1902                         if (option < 0)
1903                                 return 0;
1904                         sbi->s_li_wait_mult = option;
1905                         break;
1906                 case Opt_noinit_itable:
1907                         clear_opt(sb, INIT_INODE_TABLE);
1908                         break;
1909                 default:
1910                         ext4_msg(sb, KERN_ERR,
1911                                "Unrecognized mount option \"%s\" "
1912                                "or missing value", p);
1913                         return 0;
1914                 }
1915         }
1916 #ifdef CONFIG_QUOTA
1917         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1918                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1919                         clear_opt(sb, USRQUOTA);
1920
1921                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1922                         clear_opt(sb, GRPQUOTA);
1923
1924                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1925                         ext4_msg(sb, KERN_ERR, "old and new quota "
1926                                         "format mixing");
1927                         return 0;
1928                 }
1929
1930                 if (!sbi->s_jquota_fmt) {
1931                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1932                                         "not specified");
1933                         return 0;
1934                 }
1935         }
1936 #endif
1937         if (test_opt(sb, DIOREAD_NOLOCK)) {
1938                 int blocksize =
1939                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1940
1941                 if (blocksize < PAGE_CACHE_SIZE) {
1942                         ext4_msg(sb, KERN_ERR, "can't mount with "
1943                                  "dioread_nolock if block size != PAGE_SIZE");
1944                         return 0;
1945                 }
1946         }
1947         return 1;
1948 }
1949
1950 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1951                             int read_only)
1952 {
1953         struct ext4_sb_info *sbi = EXT4_SB(sb);
1954         int res = 0;
1955
1956         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1957                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1958                          "forcing read-only mode");
1959                 res = MS_RDONLY;
1960         }
1961         if (read_only)
1962                 goto done;
1963         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1964                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1965                          "running e2fsck is recommended");
1966         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1967                 ext4_msg(sb, KERN_WARNING,
1968                          "warning: mounting fs with errors, "
1969                          "running e2fsck is recommended");
1970         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1971                  le16_to_cpu(es->s_mnt_count) >=
1972                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1973                 ext4_msg(sb, KERN_WARNING,
1974                          "warning: maximal mount count reached, "
1975                          "running e2fsck is recommended");
1976         else if (le32_to_cpu(es->s_checkinterval) &&
1977                 (le32_to_cpu(es->s_lastcheck) +
1978                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1979                 ext4_msg(sb, KERN_WARNING,
1980                          "warning: checktime reached, "
1981                          "running e2fsck is recommended");
1982         if (!sbi->s_journal)
1983                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1984         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1985                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1986         le16_add_cpu(&es->s_mnt_count, 1);
1987         es->s_mtime = cpu_to_le32(get_seconds());
1988         ext4_update_dynamic_rev(sb);
1989         if (sbi->s_journal)
1990                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1991
1992         ext4_commit_super(sb, 1);
1993 done:
1994         if (test_opt(sb, DEBUG))
1995                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1996                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1997                         sb->s_blocksize,
1998                         sbi->s_groups_count,
1999                         EXT4_BLOCKS_PER_GROUP(sb),
2000                         EXT4_INODES_PER_GROUP(sb),
2001                         sbi->s_mount_opt, sbi->s_mount_opt2);
2002
2003         cleancache_init_fs(sb);
2004         return res;
2005 }
2006
2007 static int ext4_fill_flex_info(struct super_block *sb)
2008 {
2009         struct ext4_sb_info *sbi = EXT4_SB(sb);
2010         struct ext4_group_desc *gdp = NULL;
2011         ext4_group_t flex_group_count;
2012         ext4_group_t flex_group;
2013         unsigned int groups_per_flex = 0;
2014         size_t size;
2015         int i;
2016
2017         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2018         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2019                 sbi->s_log_groups_per_flex = 0;
2020                 return 1;
2021         }
2022         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2023
2024         /* We allocate both existing and potentially added groups */
2025         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2026                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2027                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2028         size = flex_group_count * sizeof(struct flex_groups);
2029         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2030         if (sbi->s_flex_groups == NULL) {
2031                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2032                          flex_group_count);
2033                 goto failed;
2034         }
2035
2036         for (i = 0; i < sbi->s_groups_count; i++) {
2037                 gdp = ext4_get_group_desc(sb, i, NULL);
2038
2039                 flex_group = ext4_flex_group(sbi, i);
2040                 atomic_add(ext4_free_inodes_count(sb, gdp),
2041                            &sbi->s_flex_groups[flex_group].free_inodes);
2042                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2043                              &sbi->s_flex_groups[flex_group].free_clusters);
2044                 atomic_add(ext4_used_dirs_count(sb, gdp),
2045                            &sbi->s_flex_groups[flex_group].used_dirs);
2046         }
2047
2048         return 1;
2049 failed:
2050         return 0;
2051 }
2052
2053 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2054                             struct ext4_group_desc *gdp)
2055 {
2056         __u16 crc = 0;
2057
2058         if (sbi->s_es->s_feature_ro_compat &
2059             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2060                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2061                 __le32 le_group = cpu_to_le32(block_group);
2062
2063                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2064                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2065                 crc = crc16(crc, (__u8 *)gdp, offset);
2066                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2067                 /* for checksum of struct ext4_group_desc do the rest...*/
2068                 if ((sbi->s_es->s_feature_incompat &
2069                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2070                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2071                         crc = crc16(crc, (__u8 *)gdp + offset,
2072                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2073                                         offset);
2074         }
2075
2076         return cpu_to_le16(crc);
2077 }
2078
2079 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2080                                 struct ext4_group_desc *gdp)
2081 {
2082         if ((sbi->s_es->s_feature_ro_compat &
2083              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2084             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2085                 return 0;
2086
2087         return 1;
2088 }
2089
2090 /* Called at mount-time, super-block is locked */
2091 static int ext4_check_descriptors(struct super_block *sb,
2092                                   ext4_group_t *first_not_zeroed)
2093 {
2094         struct ext4_sb_info *sbi = EXT4_SB(sb);
2095         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2096         ext4_fsblk_t last_block;
2097         ext4_fsblk_t block_bitmap;
2098         ext4_fsblk_t inode_bitmap;
2099         ext4_fsblk_t inode_table;
2100         int flexbg_flag = 0;
2101         ext4_group_t i, grp = sbi->s_groups_count;
2102
2103         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2104                 flexbg_flag = 1;
2105
2106         ext4_debug("Checking group descriptors");
2107
2108         for (i = 0; i < sbi->s_groups_count; i++) {
2109                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2110
2111                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2112                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2113                 else
2114                         last_block = first_block +
2115                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2116
2117                 if ((grp == sbi->s_groups_count) &&
2118                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2119                         grp = i;
2120
2121                 block_bitmap = ext4_block_bitmap(sb, gdp);
2122                 if (block_bitmap < first_block || block_bitmap > last_block) {
2123                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2124                                "Block bitmap for group %u not in group "
2125                                "(block %llu)!", i, block_bitmap);
2126                         return 0;
2127                 }
2128                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2129                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2130                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2131                                "Inode bitmap for group %u not in group "
2132                                "(block %llu)!", i, inode_bitmap);
2133                         return 0;
2134                 }
2135                 inode_table = ext4_inode_table(sb, gdp);
2136                 if (inode_table < first_block ||
2137                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2138                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2139                                "Inode table for group %u not in group "
2140                                "(block %llu)!", i, inode_table);
2141                         return 0;
2142                 }
2143                 ext4_lock_group(sb, i);
2144                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2145                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2146                                  "Checksum for group %u failed (%u!=%u)",
2147                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2148                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2149                         if (!(sb->s_flags & MS_RDONLY)) {
2150                                 ext4_unlock_group(sb, i);
2151                                 return 0;
2152                         }
2153                 }
2154                 ext4_unlock_group(sb, i);
2155                 if (!flexbg_flag)
2156                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2157         }
2158         if (NULL != first_not_zeroed)
2159                 *first_not_zeroed = grp;
2160
2161         ext4_free_blocks_count_set(sbi->s_es,
2162                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2163         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2164         return 1;
2165 }
2166
2167 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2168  * the superblock) which were deleted from all directories, but held open by
2169  * a process at the time of a crash.  We walk the list and try to delete these
2170  * inodes at recovery time (only with a read-write filesystem).
2171  *
2172  * In order to keep the orphan inode chain consistent during traversal (in
2173  * case of crash during recovery), we link each inode into the superblock
2174  * orphan list_head and handle it the same way as an inode deletion during
2175  * normal operation (which journals the operations for us).
2176  *
2177  * We only do an iget() and an iput() on each inode, which is very safe if we
2178  * accidentally point at an in-use or already deleted inode.  The worst that
2179  * can happen in this case is that we get a "bit already cleared" message from
2180  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2181  * e2fsck was run on this filesystem, and it must have already done the orphan
2182  * inode cleanup for us, so we can safely abort without any further action.
2183  */
2184 static void ext4_orphan_cleanup(struct super_block *sb,
2185                                 struct ext4_super_block *es)
2186 {
2187         unsigned int s_flags = sb->s_flags;
2188         int nr_orphans = 0, nr_truncates = 0;
2189 #ifdef CONFIG_QUOTA
2190         int i;
2191 #endif
2192         if (!es->s_last_orphan) {
2193                 jbd_debug(4, "no orphan inodes to clean up\n");
2194                 return;
2195         }
2196
2197         if (bdev_read_only(sb->s_bdev)) {
2198                 ext4_msg(sb, KERN_ERR, "write access "
2199                         "unavailable, skipping orphan cleanup");
2200                 return;
2201         }
2202
2203         /* Check if feature set would not allow a r/w mount */
2204         if (!ext4_feature_set_ok(sb, 0)) {
2205                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2206                          "unknown ROCOMPAT features");
2207                 return;
2208         }
2209
2210         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2211                 if (es->s_last_orphan)
2212                         jbd_debug(1, "Errors on filesystem, "
2213                                   "clearing orphan list.\n");
2214                 es->s_last_orphan = 0;
2215                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2216                 return;
2217         }
2218
2219         if (s_flags & MS_RDONLY) {
2220                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2221                 sb->s_flags &= ~MS_RDONLY;
2222         }
2223 #ifdef CONFIG_QUOTA
2224         /* Needed for iput() to work correctly and not trash data */
2225         sb->s_flags |= MS_ACTIVE;
2226         /* Turn on quotas so that they are updated correctly */
2227         for (i = 0; i < MAXQUOTAS; i++) {
2228                 if (EXT4_SB(sb)->s_qf_names[i]) {
2229                         int ret = ext4_quota_on_mount(sb, i);
2230                         if (ret < 0)
2231                                 ext4_msg(sb, KERN_ERR,
2232                                         "Cannot turn on journaled "
2233                                         "quota: error %d", ret);
2234                 }
2235         }
2236 #endif
2237
2238         while (es->s_last_orphan) {
2239                 struct inode *inode;
2240
2241                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2242                 if (IS_ERR(inode)) {
2243                         es->s_last_orphan = 0;
2244                         break;
2245                 }
2246
2247                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2248                 dquot_initialize(inode);
2249                 if (inode->i_nlink) {
2250                         ext4_msg(sb, KERN_DEBUG,
2251                                 "%s: truncating inode %lu to %lld bytes",
2252                                 __func__, inode->i_ino, inode->i_size);
2253                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2254                                   inode->i_ino, inode->i_size);
2255                         mutex_lock(&inode->i_mutex);
2256                         ext4_truncate(inode);
2257                         mutex_unlock(&inode->i_mutex);
2258                         nr_truncates++;
2259                 } else {
2260                         ext4_msg(sb, KERN_DEBUG,
2261                                 "%s: deleting unreferenced inode %lu",
2262                                 __func__, inode->i_ino);
2263                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2264                                   inode->i_ino);
2265                         nr_orphans++;
2266                 }
2267                 iput(inode);  /* The delete magic happens here! */
2268         }
2269
2270 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2271
2272         if (nr_orphans)
2273                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2274                        PLURAL(nr_orphans));
2275         if (nr_truncates)
2276                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2277                        PLURAL(nr_truncates));
2278 #ifdef CONFIG_QUOTA
2279         /* Turn quotas off */
2280         for (i = 0; i < MAXQUOTAS; i++) {
2281                 if (sb_dqopt(sb)->files[i])
2282                         dquot_quota_off(sb, i);
2283         }
2284 #endif
2285         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2286 }
2287
2288 /*
2289  * Maximal extent format file size.
2290  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2291  * extent format containers, within a sector_t, and within i_blocks
2292  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2293  * so that won't be a limiting factor.
2294  *
2295  * However there is other limiting factor. We do store extents in the form
2296  * of starting block and length, hence the resulting length of the extent
2297  * covering maximum file size must fit into on-disk format containers as
2298  * well. Given that length is always by 1 unit bigger than max unit (because
2299  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2300  *
2301  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2302  */
2303 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2304 {
2305         loff_t res;
2306         loff_t upper_limit = MAX_LFS_FILESIZE;
2307
2308         /* small i_blocks in vfs inode? */
2309         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2310                 /*
2311                  * CONFIG_LBDAF is not enabled implies the inode
2312                  * i_block represent total blocks in 512 bytes
2313                  * 32 == size of vfs inode i_blocks * 8
2314                  */
2315                 upper_limit = (1LL << 32) - 1;
2316
2317                 /* total blocks in file system block size */
2318                 upper_limit >>= (blkbits - 9);
2319                 upper_limit <<= blkbits;
2320         }
2321
2322         /*
2323          * 32-bit extent-start container, ee_block. We lower the maxbytes
2324          * by one fs block, so ee_len can cover the extent of maximum file
2325          * size
2326          */
2327         res = (1LL << 32) - 1;
2328         res <<= blkbits;
2329
2330         /* Sanity check against vm- & vfs- imposed limits */
2331         if (res > upper_limit)
2332                 res = upper_limit;
2333
2334         return res;
2335 }
2336
2337 /*
2338  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2339  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2340  * We need to be 1 filesystem block less than the 2^48 sector limit.
2341  */
2342 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2343 {
2344         loff_t res = EXT4_NDIR_BLOCKS;
2345         int meta_blocks;
2346         loff_t upper_limit;
2347         /* This is calculated to be the largest file size for a dense, block
2348          * mapped file such that the file's total number of 512-byte sectors,
2349          * including data and all indirect blocks, does not exceed (2^48 - 1).
2350          *
2351          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2352          * number of 512-byte sectors of the file.
2353          */
2354
2355         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2356                 /*
2357                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2358                  * the inode i_block field represents total file blocks in
2359                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2360                  */
2361                 upper_limit = (1LL << 32) - 1;
2362
2363                 /* total blocks in file system block size */
2364                 upper_limit >>= (bits - 9);
2365
2366         } else {
2367                 /*
2368                  * We use 48 bit ext4_inode i_blocks
2369                  * With EXT4_HUGE_FILE_FL set the i_blocks
2370                  * represent total number of blocks in
2371                  * file system block size
2372                  */
2373                 upper_limit = (1LL << 48) - 1;
2374
2375         }
2376
2377         /* indirect blocks */
2378         meta_blocks = 1;
2379         /* double indirect blocks */
2380         meta_blocks += 1 + (1LL << (bits-2));
2381         /* tripple indirect blocks */
2382         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2383
2384         upper_limit -= meta_blocks;
2385         upper_limit <<= bits;
2386
2387         res += 1LL << (bits-2);
2388         res += 1LL << (2*(bits-2));
2389         res += 1LL << (3*(bits-2));
2390         res <<= bits;
2391         if (res > upper_limit)
2392                 res = upper_limit;
2393
2394         if (res > MAX_LFS_FILESIZE)
2395                 res = MAX_LFS_FILESIZE;
2396
2397         return res;
2398 }
2399
2400 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2401                                    ext4_fsblk_t logical_sb_block, int nr)
2402 {
2403         struct ext4_sb_info *sbi = EXT4_SB(sb);
2404         ext4_group_t bg, first_meta_bg;
2405         int has_super = 0;
2406
2407         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2408
2409         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2410             nr < first_meta_bg)
2411                 return logical_sb_block + nr + 1;
2412         bg = sbi->s_desc_per_block * nr;
2413         if (ext4_bg_has_super(sb, bg))
2414                 has_super = 1;
2415
2416         return (has_super + ext4_group_first_block_no(sb, bg));
2417 }
2418
2419 /**
2420  * ext4_get_stripe_size: Get the stripe size.
2421  * @sbi: In memory super block info
2422  *
2423  * If we have specified it via mount option, then
2424  * use the mount option value. If the value specified at mount time is
2425  * greater than the blocks per group use the super block value.
2426  * If the super block value is greater than blocks per group return 0.
2427  * Allocator needs it be less than blocks per group.
2428  *
2429  */
2430 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2431 {
2432         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2433         unsigned long stripe_width =
2434                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2435         int ret;
2436
2437         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2438                 ret = sbi->s_stripe;
2439         else if (stripe_width <= sbi->s_blocks_per_group)
2440                 ret = stripe_width;
2441         else if (stride <= sbi->s_blocks_per_group)
2442                 ret = stride;
2443         else
2444                 ret = 0;
2445
2446         /*
2447          * If the stripe width is 1, this makes no sense and
2448          * we set it to 0 to turn off stripe handling code.
2449          */
2450         if (ret <= 1)
2451                 ret = 0;
2452
2453         return ret;
2454 }
2455
2456 /* sysfs supprt */
2457
2458 struct ext4_attr {
2459         struct attribute attr;
2460         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2461         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2462                          const char *, size_t);
2463         int offset;
2464 };
2465
2466 static int parse_strtoul(const char *buf,
2467                 unsigned long max, unsigned long *value)
2468 {
2469         char *endp;
2470
2471         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2472         endp = skip_spaces(endp);
2473         if (*endp || *value > max)
2474                 return -EINVAL;
2475
2476         return 0;
2477 }
2478
2479 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2480                                               struct ext4_sb_info *sbi,
2481                                               char *buf)
2482 {
2483         return snprintf(buf, PAGE_SIZE, "%llu\n",
2484                 (s64) EXT4_C2B(sbi,
2485                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2486 }
2487
2488 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2489                                          struct ext4_sb_info *sbi, char *buf)
2490 {
2491         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2492
2493         if (!sb->s_bdev->bd_part)
2494                 return snprintf(buf, PAGE_SIZE, "0\n");
2495         return snprintf(buf, PAGE_SIZE, "%lu\n",
2496                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2497                          sbi->s_sectors_written_start) >> 1);
2498 }
2499
2500 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2501                                           struct ext4_sb_info *sbi, char *buf)
2502 {
2503         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2504
2505         if (!sb->s_bdev->bd_part)
2506                 return snprintf(buf, PAGE_SIZE, "0\n");
2507         return snprintf(buf, PAGE_SIZE, "%llu\n",
2508                         (unsigned long long)(sbi->s_kbytes_written +
2509                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2510                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2511 }
2512
2513 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2514                                           struct ext4_sb_info *sbi,
2515                                           const char *buf, size_t count)
2516 {
2517         unsigned long t;
2518
2519         if (parse_strtoul(buf, 0x40000000, &t))
2520                 return -EINVAL;
2521
2522         if (t && !is_power_of_2(t))
2523                 return -EINVAL;
2524
2525         sbi->s_inode_readahead_blks = t;
2526         return count;
2527 }
2528
2529 static ssize_t sbi_ui_show(struct ext4_attr *a,
2530                            struct ext4_sb_info *sbi, char *buf)
2531 {
2532         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2533
2534         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2535 }
2536
2537 static ssize_t sbi_ui_store(struct ext4_attr *a,
2538                             struct ext4_sb_info *sbi,
2539                             const char *buf, size_t count)
2540 {
2541         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2542         unsigned long t;
2543
2544         if (parse_strtoul(buf, 0xffffffff, &t))
2545                 return -EINVAL;
2546         *ui = t;
2547         return count;
2548 }
2549
2550 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2551 static struct ext4_attr ext4_attr_##_name = {                   \
2552         .attr = {.name = __stringify(_name), .mode = _mode },   \
2553         .show   = _show,                                        \
2554         .store  = _store,                                       \
2555         .offset = offsetof(struct ext4_sb_info, _elname),       \
2556 }
2557 #define EXT4_ATTR(name, mode, show, store) \
2558 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2559
2560 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2561 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2562 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2563 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2564         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2565 #define ATTR_LIST(name) &ext4_attr_##name.attr
2566
2567 EXT4_RO_ATTR(delayed_allocation_blocks);
2568 EXT4_RO_ATTR(session_write_kbytes);
2569 EXT4_RO_ATTR(lifetime_write_kbytes);
2570 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2571                  inode_readahead_blks_store, s_inode_readahead_blks);
2572 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2573 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2574 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2575 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2576 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2577 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2578 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2579 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2580
2581 static struct attribute *ext4_attrs[] = {
2582         ATTR_LIST(delayed_allocation_blocks),
2583         ATTR_LIST(session_write_kbytes),
2584         ATTR_LIST(lifetime_write_kbytes),
2585         ATTR_LIST(inode_readahead_blks),
2586         ATTR_LIST(inode_goal),
2587         ATTR_LIST(mb_stats),
2588         ATTR_LIST(mb_max_to_scan),
2589         ATTR_LIST(mb_min_to_scan),
2590         ATTR_LIST(mb_order2_req),
2591         ATTR_LIST(mb_stream_req),
2592         ATTR_LIST(mb_group_prealloc),
2593         ATTR_LIST(max_writeback_mb_bump),
2594         NULL,
2595 };
2596
2597 /* Features this copy of ext4 supports */
2598 EXT4_INFO_ATTR(lazy_itable_init);
2599 EXT4_INFO_ATTR(batched_discard);
2600
2601 static struct attribute *ext4_feat_attrs[] = {
2602         ATTR_LIST(lazy_itable_init),
2603         ATTR_LIST(batched_discard),
2604         NULL,
2605 };
2606
2607 static ssize_t ext4_attr_show(struct kobject *kobj,
2608                               struct attribute *attr, char *buf)
2609 {
2610         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2611                                                 s_kobj);
2612         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2613
2614         return a->show ? a->show(a, sbi, buf) : 0;
2615 }
2616
2617 static ssize_t ext4_attr_store(struct kobject *kobj,
2618                                struct attribute *attr,
2619                                const char *buf, size_t len)
2620 {
2621         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2622                                                 s_kobj);
2623         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2624
2625         return a->store ? a->store(a, sbi, buf, len) : 0;
2626 }
2627
2628 static void ext4_sb_release(struct kobject *kobj)
2629 {
2630         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2631                                                 s_kobj);
2632         complete(&sbi->s_kobj_unregister);
2633 }
2634
2635 static const struct sysfs_ops ext4_attr_ops = {
2636         .show   = ext4_attr_show,
2637         .store  = ext4_attr_store,
2638 };
2639
2640 static struct kobj_type ext4_ktype = {
2641         .default_attrs  = ext4_attrs,
2642         .sysfs_ops      = &ext4_attr_ops,
2643         .release        = ext4_sb_release,
2644 };
2645
2646 static void ext4_feat_release(struct kobject *kobj)
2647 {
2648         complete(&ext4_feat->f_kobj_unregister);
2649 }
2650
2651 static struct kobj_type ext4_feat_ktype = {
2652         .default_attrs  = ext4_feat_attrs,
2653         .sysfs_ops      = &ext4_attr_ops,
2654         .release        = ext4_feat_release,
2655 };
2656
2657 /*
2658  * Check whether this filesystem can be mounted based on
2659  * the features present and the RDONLY/RDWR mount requested.
2660  * Returns 1 if this filesystem can be mounted as requested,
2661  * 0 if it cannot be.
2662  */
2663 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2664 {
2665         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2666                 ext4_msg(sb, KERN_ERR,
2667                         "Couldn't mount because of "
2668                         "unsupported optional features (%x)",
2669                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2670                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2671                 return 0;
2672         }
2673
2674         if (readonly)
2675                 return 1;
2676
2677         /* Check that feature set is OK for a read-write mount */
2678         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2679                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2680                          "unsupported optional features (%x)",
2681                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2682                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2683                 return 0;
2684         }
2685         /*
2686          * Large file size enabled file system can only be mounted
2687          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2688          */
2689         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2690                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2691                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2692                                  "cannot be mounted RDWR without "
2693                                  "CONFIG_LBDAF");
2694                         return 0;
2695                 }
2696         }
2697         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2698             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2699                 ext4_msg(sb, KERN_ERR,
2700                          "Can't support bigalloc feature without "
2701                          "extents feature\n");
2702                 return 0;
2703         }
2704         return 1;
2705 }
2706
2707 /*
2708  * This function is called once a day if we have errors logged
2709  * on the file system
2710  */
2711 static void print_daily_error_info(unsigned long arg)
2712 {
2713         struct super_block *sb = (struct super_block *) arg;
2714         struct ext4_sb_info *sbi;
2715         struct ext4_super_block *es;
2716
2717         sbi = EXT4_SB(sb);
2718         es = sbi->s_es;
2719
2720         if (es->s_error_count)
2721                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2722                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2723                          le32_to_cpu(es->s_error_count));
2724         if (es->s_first_error_time) {
2725                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2726                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2727                        (int) sizeof(es->s_first_error_func),
2728                        es->s_first_error_func,
2729                        le32_to_cpu(es->s_first_error_line));
2730                 if (es->s_first_error_ino)
2731                         printk(": inode %u",
2732                                le32_to_cpu(es->s_first_error_ino));
2733                 if (es->s_first_error_block)
2734                         printk(": block %llu", (unsigned long long)
2735                                le64_to_cpu(es->s_first_error_block));
2736                 printk("\n");
2737         }
2738         if (es->s_last_error_time) {
2739                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2740                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2741                        (int) sizeof(es->s_last_error_func),
2742                        es->s_last_error_func,
2743                        le32_to_cpu(es->s_last_error_line));
2744                 if (es->s_last_error_ino)
2745                         printk(": inode %u",
2746                                le32_to_cpu(es->s_last_error_ino));
2747                 if (es->s_last_error_block)
2748                         printk(": block %llu", (unsigned long long)
2749                                le64_to_cpu(es->s_last_error_block));
2750                 printk("\n");
2751         }
2752         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2753 }
2754
2755 /* Find next suitable group and run ext4_init_inode_table */
2756 static int ext4_run_li_request(struct ext4_li_request *elr)
2757 {
2758         struct ext4_group_desc *gdp = NULL;
2759         ext4_group_t group, ngroups;
2760         struct super_block *sb;
2761         unsigned long timeout = 0;
2762         int ret = 0;
2763
2764         sb = elr->lr_super;
2765         ngroups = EXT4_SB(sb)->s_groups_count;
2766
2767         for (group = elr->lr_next_group; group < ngroups; group++) {
2768                 gdp = ext4_get_group_desc(sb, group, NULL);
2769                 if (!gdp) {
2770                         ret = 1;
2771                         break;
2772                 }
2773
2774                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2775                         break;
2776         }
2777
2778         if (group == ngroups)
2779                 ret = 1;
2780
2781         if (!ret) {
2782                 timeout = jiffies;
2783                 ret = ext4_init_inode_table(sb, group,
2784                                             elr->lr_timeout ? 0 : 1);
2785                 if (elr->lr_timeout == 0) {
2786                         timeout = (jiffies - timeout) *
2787                                   elr->lr_sbi->s_li_wait_mult;
2788                         elr->lr_timeout = timeout;
2789                 }
2790                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2791                 elr->lr_next_group = group + 1;
2792         }
2793
2794         return ret;
2795 }
2796
2797 /*
2798  * Remove lr_request from the list_request and free the
2799  * request structure. Should be called with li_list_mtx held
2800  */
2801 static void ext4_remove_li_request(struct ext4_li_request *elr)
2802 {
2803         struct ext4_sb_info *sbi;
2804
2805         if (!elr)
2806                 return;
2807
2808         sbi = elr->lr_sbi;
2809
2810         list_del(&elr->lr_request);
2811         sbi->s_li_request = NULL;
2812         kfree(elr);
2813 }
2814
2815 static void ext4_unregister_li_request(struct super_block *sb)
2816 {
2817         mutex_lock(&ext4_li_mtx);
2818         if (!ext4_li_info) {
2819                 mutex_unlock(&ext4_li_mtx);
2820                 return;
2821         }
2822
2823         mutex_lock(&ext4_li_info->li_list_mtx);
2824         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2825         mutex_unlock(&ext4_li_info->li_list_mtx);
2826         mutex_unlock(&ext4_li_mtx);
2827 }
2828
2829 static struct task_struct *ext4_lazyinit_task;
2830
2831 /*
2832  * This is the function where ext4lazyinit thread lives. It walks
2833  * through the request list searching for next scheduled filesystem.
2834  * When such a fs is found, run the lazy initialization request
2835  * (ext4_rn_li_request) and keep track of the time spend in this
2836  * function. Based on that time we compute next schedule time of
2837  * the request. When walking through the list is complete, compute
2838  * next waking time and put itself into sleep.
2839  */
2840 static int ext4_lazyinit_thread(void *arg)
2841 {
2842         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2843         struct list_head *pos, *n;
2844         struct ext4_li_request *elr;
2845         unsigned long next_wakeup, cur;
2846
2847         BUG_ON(NULL == eli);
2848
2849 cont_thread:
2850         while (true) {
2851                 next_wakeup = MAX_JIFFY_OFFSET;
2852
2853                 mutex_lock(&eli->li_list_mtx);
2854                 if (list_empty(&eli->li_request_list)) {
2855                         mutex_unlock(&eli->li_list_mtx);
2856                         goto exit_thread;
2857                 }
2858
2859                 list_for_each_safe(pos, n, &eli->li_request_list) {
2860                         elr = list_entry(pos, struct ext4_li_request,
2861                                          lr_request);
2862
2863                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2864                                 if (ext4_run_li_request(elr) != 0) {
2865                                         /* error, remove the lazy_init job */
2866                                         ext4_remove_li_request(elr);
2867                                         continue;
2868                                 }
2869                         }
2870
2871                         if (time_before(elr->lr_next_sched, next_wakeup))
2872                                 next_wakeup = elr->lr_next_sched;
2873                 }
2874                 mutex_unlock(&eli->li_list_mtx);
2875
2876                 if (freezing(current))
2877                         refrigerator();
2878
2879                 cur = jiffies;
2880                 if ((time_after_eq(cur, next_wakeup)) ||
2881                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2882                         cond_resched();
2883                         continue;
2884                 }
2885
2886                 schedule_timeout_interruptible(next_wakeup - cur);
2887
2888                 if (kthread_should_stop()) {
2889                         ext4_clear_request_list();
2890                         goto exit_thread;
2891                 }
2892         }
2893
2894 exit_thread:
2895         /*
2896          * It looks like the request list is empty, but we need
2897          * to check it under the li_list_mtx lock, to prevent any
2898          * additions into it, and of course we should lock ext4_li_mtx
2899          * to atomically free the list and ext4_li_info, because at
2900          * this point another ext4 filesystem could be registering
2901          * new one.
2902          */
2903         mutex_lock(&ext4_li_mtx);
2904         mutex_lock(&eli->li_list_mtx);
2905         if (!list_empty(&eli->li_request_list)) {
2906                 mutex_unlock(&eli->li_list_mtx);
2907                 mutex_unlock(&ext4_li_mtx);
2908                 goto cont_thread;
2909         }
2910         mutex_unlock(&eli->li_list_mtx);
2911         kfree(ext4_li_info);
2912         ext4_li_info = NULL;
2913         mutex_unlock(&ext4_li_mtx);
2914
2915         return 0;
2916 }
2917
2918 static void ext4_clear_request_list(void)
2919 {
2920         struct list_head *pos, *n;
2921         struct ext4_li_request *elr;
2922
2923         mutex_lock(&ext4_li_info->li_list_mtx);
2924         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2925                 elr = list_entry(pos, struct ext4_li_request,
2926                                  lr_request);
2927                 ext4_remove_li_request(elr);
2928         }
2929         mutex_unlock(&ext4_li_info->li_list_mtx);
2930 }
2931
2932 static int ext4_run_lazyinit_thread(void)
2933 {
2934         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2935                                          ext4_li_info, "ext4lazyinit");
2936         if (IS_ERR(ext4_lazyinit_task)) {
2937                 int err = PTR_ERR(ext4_lazyinit_task);
2938                 ext4_clear_request_list();
2939                 kfree(ext4_li_info);
2940                 ext4_li_info = NULL;
2941                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2942                                  "initialization thread\n",
2943                                  err);
2944                 return err;
2945         }
2946         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2947         return 0;
2948 }
2949
2950 /*
2951  * Check whether it make sense to run itable init. thread or not.
2952  * If there is at least one uninitialized inode table, return
2953  * corresponding group number, else the loop goes through all
2954  * groups and return total number of groups.
2955  */
2956 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2957 {
2958         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2959         struct ext4_group_desc *gdp = NULL;
2960
2961         for (group = 0; group < ngroups; group++) {
2962                 gdp = ext4_get_group_desc(sb, group, NULL);
2963                 if (!gdp)
2964                         continue;
2965
2966                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2967                         break;
2968         }
2969
2970         return group;
2971 }
2972
2973 static int ext4_li_info_new(void)
2974 {
2975         struct ext4_lazy_init *eli = NULL;
2976
2977         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2978         if (!eli)
2979                 return -ENOMEM;
2980
2981         INIT_LIST_HEAD(&eli->li_request_list);
2982         mutex_init(&eli->li_list_mtx);
2983
2984         eli->li_state |= EXT4_LAZYINIT_QUIT;
2985
2986         ext4_li_info = eli;
2987
2988         return 0;
2989 }
2990
2991 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2992                                             ext4_group_t start)
2993 {
2994         struct ext4_sb_info *sbi = EXT4_SB(sb);
2995         struct ext4_li_request *elr;
2996         unsigned long rnd;
2997
2998         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2999         if (!elr)
3000                 return NULL;
3001
3002         elr->lr_super = sb;
3003         elr->lr_sbi = sbi;
3004         elr->lr_next_group = start;
3005
3006         /*
3007          * Randomize first schedule time of the request to
3008          * spread the inode table initialization requests
3009          * better.
3010          */
3011         get_random_bytes(&rnd, sizeof(rnd));
3012         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3013                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3014
3015         return elr;
3016 }
3017
3018 static int ext4_register_li_request(struct super_block *sb,
3019                                     ext4_group_t first_not_zeroed)
3020 {
3021         struct ext4_sb_info *sbi = EXT4_SB(sb);
3022         struct ext4_li_request *elr;
3023         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3024         int ret = 0;
3025
3026         if (sbi->s_li_request != NULL) {
3027                 /*
3028                  * Reset timeout so it can be computed again, because
3029                  * s_li_wait_mult might have changed.
3030                  */
3031                 sbi->s_li_request->lr_timeout = 0;
3032                 return 0;
3033         }
3034
3035         if (first_not_zeroed == ngroups ||
3036             (sb->s_flags & MS_RDONLY) ||
3037             !test_opt(sb, INIT_INODE_TABLE))
3038                 return 0;
3039
3040         elr = ext4_li_request_new(sb, first_not_zeroed);
3041         if (!elr)
3042                 return -ENOMEM;
3043
3044         mutex_lock(&ext4_li_mtx);
3045
3046         if (NULL == ext4_li_info) {
3047                 ret = ext4_li_info_new();
3048                 if (ret)
3049                         goto out;
3050         }
3051
3052         mutex_lock(&ext4_li_info->li_list_mtx);
3053         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3054         mutex_unlock(&ext4_li_info->li_list_mtx);
3055
3056         sbi->s_li_request = elr;
3057         /*
3058          * set elr to NULL here since it has been inserted to
3059          * the request_list and the removal and free of it is
3060          * handled by ext4_clear_request_list from now on.
3061          */
3062         elr = NULL;
3063
3064         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3065                 ret = ext4_run_lazyinit_thread();
3066                 if (ret)
3067                         goto out;
3068         }
3069 out:
3070         mutex_unlock(&ext4_li_mtx);
3071         if (ret)
3072                 kfree(elr);
3073         return ret;
3074 }
3075
3076 /*
3077  * We do not need to lock anything since this is called on
3078  * module unload.
3079  */
3080 static void ext4_destroy_lazyinit_thread(void)
3081 {
3082         /*
3083          * If thread exited earlier
3084          * there's nothing to be done.
3085          */
3086         if (!ext4_li_info || !ext4_lazyinit_task)
3087                 return;
3088
3089         kthread_stop(ext4_lazyinit_task);
3090 }
3091
3092 /*
3093  * Note: calculating the overhead so we can be compatible with
3094  * historical BSD practice is quite difficult in the face of
3095  * clusters/bigalloc.  This is because multiple metadata blocks from
3096  * different block group can end up in the same allocation cluster.
3097  * Calculating the exact overhead in the face of clustered allocation
3098  * requires either O(all block bitmaps) in memory or O(number of block
3099  * groups**2) in time.  We will still calculate the superblock for
3100  * older file systems --- and if we come across with a bigalloc file
3101  * system with zero in s_overhead_clusters the estimate will be close to
3102  * correct especially for very large cluster sizes --- but for newer
3103  * file systems, it's better to calculate this figure once at mkfs
3104  * time, and store it in the superblock.  If the superblock value is
3105  * present (even for non-bigalloc file systems), we will use it.
3106  */
3107 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3108                           char *buf)
3109 {
3110         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3111         struct ext4_group_desc  *gdp;
3112         ext4_fsblk_t            first_block, last_block, b;
3113         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3114         int                     s, j, count = 0;
3115
3116         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3117                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3118                         sbi->s_itb_per_group + 2);
3119
3120         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3121                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3122         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3123         for (i = 0; i < ngroups; i++) {
3124                 gdp = ext4_get_group_desc(sb, i, NULL);
3125                 b = ext4_block_bitmap(sb, gdp);
3126                 if (b >= first_block && b <= last_block) {
3127                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3128                         count++;
3129                 }
3130                 b = ext4_inode_bitmap(sb, gdp);
3131                 if (b >= first_block && b <= last_block) {
3132                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3133                         count++;
3134                 }
3135                 b = ext4_inode_table(sb, gdp);
3136                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3137                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3138                                 int c = EXT4_B2C(sbi, b - first_block);
3139                                 ext4_set_bit(c, buf);
3140                                 count++;
3141                         }
3142                 if (i != grp)
3143                         continue;
3144                 s = 0;
3145                 if (ext4_bg_has_super(sb, grp)) {
3146                         ext4_set_bit(s++, buf);
3147                         count++;
3148                 }
3149                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3150                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3151                         count++;
3152                 }
3153         }
3154         if (!count)
3155                 return 0;
3156         return EXT4_CLUSTERS_PER_GROUP(sb) -
3157                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3158 }
3159
3160 /*
3161  * Compute the overhead and stash it in sbi->s_overhead
3162  */
3163 int ext4_calculate_overhead(struct super_block *sb)
3164 {
3165         struct ext4_sb_info *sbi = EXT4_SB(sb);
3166         struct ext4_super_block *es = sbi->s_es;
3167         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3168         ext4_fsblk_t overhead = 0;
3169         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3170
3171         memset(buf, 0, PAGE_SIZE);
3172         if (!buf)
3173                 return -ENOMEM;
3174
3175         /*
3176          * Compute the overhead (FS structures).  This is constant
3177          * for a given filesystem unless the number of block groups
3178          * changes so we cache the previous value until it does.
3179          */
3180
3181         /*
3182          * All of the blocks before first_data_block are overhead
3183          */
3184         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3185
3186         /*
3187          * Add the overhead found in each block group
3188          */
3189         for (i = 0; i < ngroups; i++) {
3190                 int blks;
3191
3192                 blks = count_overhead(sb, i, buf);
3193                 overhead += blks;
3194                 if (blks)
3195                         memset(buf, 0, PAGE_SIZE);
3196                 cond_resched();
3197         }
3198         sbi->s_overhead = overhead;
3199         smp_wmb();
3200         free_page((unsigned long) buf);
3201         return 0;
3202 }
3203
3204 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3205 {
3206         char *orig_data = kstrdup(data, GFP_KERNEL);
3207         struct buffer_head *bh;
3208         struct ext4_super_block *es = NULL;
3209         struct ext4_sb_info *sbi;
3210         ext4_fsblk_t block;
3211         ext4_fsblk_t sb_block = get_sb_block(&data);
3212         ext4_fsblk_t logical_sb_block;
3213         unsigned long offset = 0;
3214         unsigned long journal_devnum = 0;
3215         unsigned long def_mount_opts;
3216         struct inode *root;
3217         char *cp;
3218         const char *descr;
3219         int ret = -ENOMEM;
3220         int blocksize, clustersize;
3221         unsigned int db_count;
3222         unsigned int i;
3223         int needs_recovery, has_huge_files, has_bigalloc;
3224         __u64 blocks_count;
3225         int err;
3226         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3227         ext4_group_t first_not_zeroed;
3228
3229         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3230         if (!sbi)
3231                 goto out_free_orig;
3232
3233         sbi->s_blockgroup_lock =
3234                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3235         if (!sbi->s_blockgroup_lock) {
3236                 kfree(sbi);
3237                 goto out_free_orig;
3238         }
3239         sb->s_fs_info = sbi;
3240         sbi->s_mount_opt = 0;
3241         sbi->s_resuid = EXT4_DEF_RESUID;
3242         sbi->s_resgid = EXT4_DEF_RESGID;
3243         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3244         sbi->s_sb_block = sb_block;
3245         if (sb->s_bdev->bd_part)
3246                 sbi->s_sectors_written_start =
3247                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3248
3249         /* Cleanup superblock name */
3250         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3251                 *cp = '!';
3252
3253         ret = -EINVAL;
3254         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3255         if (!blocksize) {
3256                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3257                 goto out_fail;
3258         }
3259
3260         /*
3261          * The ext4 superblock will not be buffer aligned for other than 1kB
3262          * block sizes.  We need to calculate the offset from buffer start.
3263          */
3264         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3265                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3266                 offset = do_div(logical_sb_block, blocksize);
3267         } else {
3268                 logical_sb_block = sb_block;
3269         }
3270
3271         if (!(bh = sb_bread(sb, logical_sb_block))) {
3272                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3273                 goto out_fail;
3274         }
3275         /*
3276          * Note: s_es must be initialized as soon as possible because
3277          *       some ext4 macro-instructions depend on its value
3278          */
3279         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3280         sbi->s_es = es;
3281         sb->s_magic = le16_to_cpu(es->s_magic);
3282         if (sb->s_magic != EXT4_SUPER_MAGIC)
3283                 goto cantfind_ext4;
3284         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3285
3286         /* Set defaults before we parse the mount options */
3287         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3288         set_opt(sb, INIT_INODE_TABLE);
3289         if (def_mount_opts & EXT4_DEFM_DEBUG)
3290                 set_opt(sb, DEBUG);
3291         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3292                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3293                         "2.6.38");
3294                 set_opt(sb, GRPID);
3295         }
3296         if (def_mount_opts & EXT4_DEFM_UID16)
3297                 set_opt(sb, NO_UID32);
3298         /* xattr user namespace & acls are now defaulted on */
3299 #ifdef CONFIG_EXT4_FS_XATTR
3300         set_opt(sb, XATTR_USER);
3301 #endif
3302 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3303         set_opt(sb, POSIX_ACL);
3304 #endif
3305         set_opt(sb, MBLK_IO_SUBMIT);
3306         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3307                 set_opt(sb, JOURNAL_DATA);
3308         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3309                 set_opt(sb, ORDERED_DATA);
3310         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3311                 set_opt(sb, WRITEBACK_DATA);
3312
3313         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3314                 set_opt(sb, ERRORS_PANIC);
3315         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3316                 set_opt(sb, ERRORS_CONT);
3317         else
3318                 set_opt(sb, ERRORS_RO);
3319         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3320                 set_opt(sb, BLOCK_VALIDITY);
3321         if (def_mount_opts & EXT4_DEFM_DISCARD)
3322                 set_opt(sb, DISCARD);
3323
3324         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3325         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3326         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3327         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3328         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3329
3330         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3331                 set_opt(sb, BARRIER);
3332
3333         /*
3334          * enable delayed allocation by default
3335          * Use -o nodelalloc to turn it off
3336          */
3337         if (!IS_EXT3_SB(sb) &&
3338             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3339                 set_opt(sb, DELALLOC);
3340
3341         /*
3342          * set default s_li_wait_mult for lazyinit, for the case there is
3343          * no mount option specified.
3344          */
3345         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3346
3347         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3348                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3349                 ext4_msg(sb, KERN_WARNING,
3350                          "failed to parse options in superblock: %s",
3351                          sbi->s_es->s_mount_opts);
3352         }
3353         if (!parse_options((char *) data, sb, &journal_devnum,
3354                            &journal_ioprio, NULL, 0))
3355                 goto failed_mount;
3356
3357         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3358                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3359                             "with data=journal disables delayed "
3360                             "allocation and O_DIRECT support!\n");
3361                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3362                         ext4_msg(sb, KERN_ERR, "can't mount with "
3363                                  "both data=journal and delalloc");
3364                         goto failed_mount;
3365                 }
3366                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3367                         ext4_msg(sb, KERN_ERR, "can't mount with "
3368                                  "both data=journal and dioread_nolock");
3369                         goto failed_mount;
3370                 }
3371                 if (test_opt(sb, DELALLOC))
3372                         clear_opt(sb, DELALLOC);
3373         }
3374
3375         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3376                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3377
3378         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3379             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3380              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3381              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3382                 ext4_msg(sb, KERN_WARNING,
3383                        "feature flags set on rev 0 fs, "
3384                        "running e2fsck is recommended");
3385
3386         if (IS_EXT2_SB(sb)) {
3387                 if (ext2_feature_set_ok(sb))
3388                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3389                                  "using the ext4 subsystem");
3390                 else {
3391                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3392                                  "to feature incompatibilities");
3393                         goto failed_mount;
3394                 }
3395         }
3396
3397         if (IS_EXT3_SB(sb)) {
3398                 if (ext3_feature_set_ok(sb))
3399                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3400                                  "using the ext4 subsystem");
3401                 else {
3402                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3403                                  "to feature incompatibilities");
3404                         goto failed_mount;
3405                 }
3406         }
3407
3408         /*
3409          * Check feature flags regardless of the revision level, since we
3410          * previously didn't change the revision level when setting the flags,
3411          * so there is a chance incompat flags are set on a rev 0 filesystem.
3412          */
3413         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3414                 goto failed_mount;
3415
3416         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3417         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3418             blocksize > EXT4_MAX_BLOCK_SIZE) {
3419                 ext4_msg(sb, KERN_ERR,
3420                        "Unsupported filesystem blocksize %d", blocksize);
3421                 goto failed_mount;
3422         }
3423
3424         if (sb->s_blocksize != blocksize) {
3425                 /* Validate the filesystem blocksize */
3426                 if (!sb_set_blocksize(sb, blocksize)) {
3427                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3428                                         blocksize);
3429                         goto failed_mount;
3430                 }
3431
3432                 brelse(bh);
3433                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3434                 offset = do_div(logical_sb_block, blocksize);
3435                 bh = sb_bread(sb, logical_sb_block);
3436                 if (!bh) {
3437                         ext4_msg(sb, KERN_ERR,
3438                                "Can't read superblock on 2nd try");
3439                         goto failed_mount;
3440                 }
3441                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3442                 sbi->s_es = es;
3443                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3444                         ext4_msg(sb, KERN_ERR,
3445                                "Magic mismatch, very weird!");
3446                         goto failed_mount;
3447                 }
3448         }
3449
3450         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3451                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3452         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3453                                                       has_huge_files);
3454         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3455
3456         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3457                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3458                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3459         } else {
3460                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3461                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3462                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3463                     (!is_power_of_2(sbi->s_inode_size)) ||
3464                     (sbi->s_inode_size > blocksize)) {
3465                         ext4_msg(sb, KERN_ERR,
3466                                "unsupported inode size: %d",
3467                                sbi->s_inode_size);
3468                         goto failed_mount;
3469                 }
3470                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3471                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3472         }
3473
3474         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3475         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3476                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3477                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3478                     !is_power_of_2(sbi->s_desc_size)) {
3479                         ext4_msg(sb, KERN_ERR,
3480                                "unsupported descriptor size %lu",
3481                                sbi->s_desc_size);
3482                         goto failed_mount;
3483                 }
3484         } else
3485                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3486
3487         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3488         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3489         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3490                 goto cantfind_ext4;
3491
3492         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3493         if (sbi->s_inodes_per_block == 0)
3494                 goto cantfind_ext4;
3495         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3496                                         sbi->s_inodes_per_block;
3497         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3498         sbi->s_sbh = bh;
3499         sbi->s_mount_state = le16_to_cpu(es->s_state);
3500         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3501         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3502
3503         for (i = 0; i < 4; i++)
3504                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3505         sbi->s_def_hash_version = es->s_def_hash_version;
3506         i = le32_to_cpu(es->s_flags);
3507         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3508                 sbi->s_hash_unsigned = 3;
3509         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3510 #ifdef __CHAR_UNSIGNED__
3511                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3512                 sbi->s_hash_unsigned = 3;
3513 #else
3514                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3515 #endif
3516                 sb->s_dirt = 1;
3517         }
3518
3519         /* Handle clustersize */
3520         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3521         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3522                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3523         if (has_bigalloc) {
3524                 if (clustersize < blocksize) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                  "cluster size (%d) smaller than "
3527                                  "block size (%d)", clustersize, blocksize);
3528                         goto failed_mount;
3529                 }
3530                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3531                         le32_to_cpu(es->s_log_block_size);
3532                 sbi->s_clusters_per_group =
3533                         le32_to_cpu(es->s_clusters_per_group);
3534                 if (sbi->s_clusters_per_group > blocksize * 8) {
3535                         ext4_msg(sb, KERN_ERR,
3536                                  "#clusters per group too big: %lu",
3537                                  sbi->s_clusters_per_group);
3538                         goto failed_mount;
3539                 }
3540                 if (sbi->s_blocks_per_group !=
3541                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3542                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3543                                  "clusters per group (%lu) inconsistent",
3544                                  sbi->s_blocks_per_group,
3545                                  sbi->s_clusters_per_group);
3546                         goto failed_mount;
3547                 }
3548         } else {
3549                 if (clustersize != blocksize) {
3550                         ext4_warning(sb, "fragment/cluster size (%d) != "
3551                                      "block size (%d)", clustersize,
3552                                      blocksize);
3553                         clustersize = blocksize;
3554                 }
3555                 if (sbi->s_blocks_per_group > blocksize * 8) {
3556                         ext4_msg(sb, KERN_ERR,
3557                                  "#blocks per group too big: %lu",
3558                                  sbi->s_blocks_per_group);
3559                         goto failed_mount;
3560                 }
3561                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3562                 sbi->s_cluster_bits = 0;
3563         }
3564         sbi->s_cluster_ratio = clustersize / blocksize;
3565
3566         if (sbi->s_inodes_per_group > blocksize * 8) {
3567                 ext4_msg(sb, KERN_ERR,
3568                        "#inodes per group too big: %lu",
3569                        sbi->s_inodes_per_group);
3570                 goto failed_mount;
3571         }
3572
3573         /*
3574          * Test whether we have more sectors than will fit in sector_t,
3575          * and whether the max offset is addressable by the page cache.
3576          */
3577         err = generic_check_addressable(sb->s_blocksize_bits,
3578                                         ext4_blocks_count(es));
3579         if (err) {
3580                 ext4_msg(sb, KERN_ERR, "filesystem"
3581                          " too large to mount safely on this system");
3582                 if (sizeof(sector_t) < 8)
3583                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3584                 ret = err;
3585                 goto failed_mount;
3586         }
3587
3588         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3589                 goto cantfind_ext4;
3590
3591         /* check blocks count against device size */
3592         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3593         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3594                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3595                        "exceeds size of device (%llu blocks)",
3596                        ext4_blocks_count(es), blocks_count);
3597                 goto failed_mount;
3598         }
3599
3600         /*
3601          * It makes no sense for the first data block to be beyond the end
3602          * of the filesystem.
3603          */
3604         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3605                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3606                          "block %u is beyond end of filesystem (%llu)",
3607                          le32_to_cpu(es->s_first_data_block),
3608                          ext4_blocks_count(es));
3609                 goto failed_mount;
3610         }
3611         blocks_count = (ext4_blocks_count(es) -
3612                         le32_to_cpu(es->s_first_data_block) +
3613                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3614         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3615         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3616                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3617                        "(block count %llu, first data block %u, "
3618                        "blocks per group %lu)", sbi->s_groups_count,
3619                        ext4_blocks_count(es),
3620                        le32_to_cpu(es->s_first_data_block),
3621                        EXT4_BLOCKS_PER_GROUP(sb));
3622                 goto failed_mount;
3623         }
3624         sbi->s_groups_count = blocks_count;
3625         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3626                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3627         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3628                    EXT4_DESC_PER_BLOCK(sb);
3629         sbi->s_group_desc = ext4_kvmalloc(db_count *
3630                                           sizeof(struct buffer_head *),
3631                                           GFP_KERNEL);
3632         if (sbi->s_group_desc == NULL) {
3633                 ext4_msg(sb, KERN_ERR, "not enough memory");
3634                 goto failed_mount;
3635         }
3636
3637         if (ext4_proc_root)
3638                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3639
3640         bgl_lock_init(sbi->s_blockgroup_lock);
3641
3642         for (i = 0; i < db_count; i++) {
3643                 block = descriptor_loc(sb, logical_sb_block, i);
3644                 sbi->s_group_desc[i] = sb_bread(sb, block);
3645                 if (!sbi->s_group_desc[i]) {
3646                         ext4_msg(sb, KERN_ERR,
3647                                "can't read group descriptor %d", i);
3648                         db_count = i;
3649                         goto failed_mount2;
3650                 }
3651         }
3652         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3653                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3654                 goto failed_mount2;
3655         }
3656         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3657                 if (!ext4_fill_flex_info(sb)) {
3658                         ext4_msg(sb, KERN_ERR,
3659                                "unable to initialize "
3660                                "flex_bg meta info!");
3661                         goto failed_mount2;
3662                 }
3663
3664         sbi->s_gdb_count = db_count;
3665         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3666         spin_lock_init(&sbi->s_next_gen_lock);
3667
3668         init_timer(&sbi->s_err_report);
3669         sbi->s_err_report.function = print_daily_error_info;
3670         sbi->s_err_report.data = (unsigned long) sb;
3671
3672         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3673                         ext4_count_free_clusters(sb));
3674         if (!err) {
3675                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3676                                 ext4_count_free_inodes(sb));
3677         }
3678         if (!err) {
3679                 err = percpu_counter_init(&sbi->s_dirs_counter,
3680                                 ext4_count_dirs(sb));
3681         }
3682         if (!err) {
3683                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3684         }
3685         if (err) {
3686                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3687                 goto failed_mount3;
3688         }
3689
3690         sbi->s_stripe = ext4_get_stripe_size(sbi);
3691         sbi->s_max_writeback_mb_bump = 128;
3692
3693         /*
3694          * set up enough so that it can read an inode
3695          */
3696         if (!test_opt(sb, NOLOAD) &&
3697             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3698                 sb->s_op = &ext4_sops;
3699         else
3700                 sb->s_op = &ext4_nojournal_sops;
3701         sb->s_export_op = &ext4_export_ops;
3702         sb->s_xattr = ext4_xattr_handlers;
3703 #ifdef CONFIG_QUOTA
3704         sb->s_qcop = &ext4_qctl_operations;
3705         sb->dq_op = &ext4_quota_operations;
3706 #endif
3707         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3708
3709         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3710         mutex_init(&sbi->s_orphan_lock);
3711         sbi->s_resize_flags = 0;
3712
3713         sb->s_root = NULL;
3714
3715         needs_recovery = (es->s_last_orphan != 0 ||
3716                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3717                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3718
3719         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3720             !(sb->s_flags & MS_RDONLY))
3721                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3722                         goto failed_mount3;
3723
3724         /*
3725          * The first inode we look at is the journal inode.  Don't try
3726          * root first: it may be modified in the journal!
3727          */
3728         if (!test_opt(sb, NOLOAD) &&
3729             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3730                 if (ext4_load_journal(sb, es, journal_devnum))
3731                         goto failed_mount3;
3732         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3733               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3734                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3735                        "suppressed and not mounted read-only");
3736                 goto failed_mount_wq;
3737         } else {
3738                 clear_opt(sb, DATA_FLAGS);
3739                 sbi->s_journal = NULL;
3740                 needs_recovery = 0;
3741                 goto no_journal;
3742         }
3743
3744         if (ext4_blocks_count(es) > 0xffffffffULL &&
3745             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3746                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3747                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3748                 goto failed_mount_wq;
3749         }
3750
3751         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3752                 jbd2_journal_set_features(sbi->s_journal,
3753                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3754                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3755         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3756                 jbd2_journal_set_features(sbi->s_journal,
3757                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3758                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3759                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3760         } else {
3761                 jbd2_journal_clear_features(sbi->s_journal,
3762                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3763                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3764         }
3765
3766         /* We have now updated the journal if required, so we can
3767          * validate the data journaling mode. */
3768         switch (test_opt(sb, DATA_FLAGS)) {
3769         case 0:
3770                 /* No mode set, assume a default based on the journal
3771                  * capabilities: ORDERED_DATA if the journal can
3772                  * cope, else JOURNAL_DATA
3773                  */
3774                 if (jbd2_journal_check_available_features
3775                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3776                         set_opt(sb, ORDERED_DATA);
3777                 else
3778                         set_opt(sb, JOURNAL_DATA);
3779                 break;
3780
3781         case EXT4_MOUNT_ORDERED_DATA:
3782         case EXT4_MOUNT_WRITEBACK_DATA:
3783                 if (!jbd2_journal_check_available_features
3784                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3785                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3786                                "requested data journaling mode");
3787                         goto failed_mount_wq;
3788                 }
3789         default:
3790                 break;
3791         }
3792         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3793
3794         /*
3795          * The journal may have updated the bg summary counts, so we
3796          * need to update the global counters.
3797          */
3798         percpu_counter_set(&sbi->s_freeclusters_counter,
3799                            ext4_count_free_clusters(sb));
3800         percpu_counter_set(&sbi->s_freeinodes_counter,
3801                            ext4_count_free_inodes(sb));
3802         percpu_counter_set(&sbi->s_dirs_counter,
3803                            ext4_count_dirs(sb));
3804         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3805
3806 no_journal:
3807         /*
3808          * Get the # of file system overhead blocks from the
3809          * superblock if present.
3810          */
3811         if (es->s_overhead_clusters)
3812                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3813         else {
3814                 ret = ext4_calculate_overhead(sb);
3815                 if (ret)
3816                         goto failed_mount_wq;
3817         }
3818
3819         /*
3820          * The maximum number of concurrent works can be high and
3821          * concurrency isn't really necessary.  Limit it to 1.
3822          */
3823         EXT4_SB(sb)->dio_unwritten_wq =
3824                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3825         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3826                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3827                 goto failed_mount_wq;
3828         }
3829
3830         /*
3831          * The jbd2_journal_load will have done any necessary log recovery,
3832          * so we can safely mount the rest of the filesystem now.
3833          */
3834
3835         root = ext4_iget(sb, EXT4_ROOT_INO);
3836         if (IS_ERR(root)) {
3837                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3838                 ret = PTR_ERR(root);
3839                 root = NULL;
3840                 goto failed_mount4;
3841         }
3842         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3843                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3844                 goto failed_mount4;
3845         }
3846         sb->s_root = d_alloc_root(root);
3847         if (!sb->s_root) {
3848                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3849                 ret = -ENOMEM;
3850                 goto failed_mount4;
3851         }
3852
3853         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3854                 sb->s_flags |= MS_RDONLY;
3855
3856         /* determine the minimum size of new large inodes, if present */
3857         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3858                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3859                                                      EXT4_GOOD_OLD_INODE_SIZE;
3860                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3861                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3862                         if (sbi->s_want_extra_isize <
3863                             le16_to_cpu(es->s_want_extra_isize))
3864                                 sbi->s_want_extra_isize =
3865                                         le16_to_cpu(es->s_want_extra_isize);
3866                         if (sbi->s_want_extra_isize <
3867                             le16_to_cpu(es->s_min_extra_isize))
3868                                 sbi->s_want_extra_isize =
3869                                         le16_to_cpu(es->s_min_extra_isize);
3870                 }
3871         }
3872         /* Check if enough inode space is available */
3873         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3874                                                         sbi->s_inode_size) {
3875                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3876                                                        EXT4_GOOD_OLD_INODE_SIZE;
3877                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3878                          "available");
3879         }
3880
3881         err = ext4_setup_system_zone(sb);
3882         if (err) {
3883                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3884                          "zone (%d)", err);
3885                 goto failed_mount4;
3886         }
3887
3888         ext4_ext_init(sb);
3889         err = ext4_mb_init(sb, needs_recovery);
3890         if (err) {
3891                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3892                          err);
3893                 goto failed_mount5;
3894         }
3895
3896         err = ext4_register_li_request(sb, first_not_zeroed);
3897         if (err)
3898                 goto failed_mount6;
3899
3900         sbi->s_kobj.kset = ext4_kset;
3901         init_completion(&sbi->s_kobj_unregister);
3902         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3903                                    "%s", sb->s_id);
3904         if (err)
3905                 goto failed_mount7;
3906
3907         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3908         ext4_orphan_cleanup(sb, es);
3909         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3910         if (needs_recovery) {
3911                 ext4_msg(sb, KERN_INFO, "recovery complete");
3912                 ext4_mark_recovery_complete(sb, es);
3913         }
3914         if (EXT4_SB(sb)->s_journal) {
3915                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3916                         descr = " journalled data mode";
3917                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3918                         descr = " ordered data mode";
3919                 else
3920                         descr = " writeback data mode";
3921         } else
3922                 descr = "out journal";
3923
3924         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3925                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3926                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3927
3928         if (es->s_error_count)
3929                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3930
3931         kfree(orig_data);
3932         return 0;
3933
3934 cantfind_ext4:
3935         if (!silent)
3936                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3937         goto failed_mount;
3938
3939 failed_mount7:
3940         ext4_unregister_li_request(sb);
3941 failed_mount6:
3942         ext4_ext_release(sb);
3943 failed_mount5:
3944         ext4_mb_release(sb);
3945         ext4_release_system_zone(sb);
3946 failed_mount4:
3947         iput(root);
3948         sb->s_root = NULL;
3949         ext4_msg(sb, KERN_ERR, "mount failed");
3950         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3951 failed_mount_wq:
3952         if (sbi->s_journal) {
3953                 jbd2_journal_destroy(sbi->s_journal);
3954                 sbi->s_journal = NULL;
3955         }
3956 failed_mount3:
3957         del_timer_sync(&sbi->s_err_report);
3958         if (sbi->s_flex_groups)
3959                 ext4_kvfree(sbi->s_flex_groups);
3960         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3961         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3962         percpu_counter_destroy(&sbi->s_dirs_counter);
3963         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3964         if (sbi->s_mmp_tsk)
3965                 kthread_stop(sbi->s_mmp_tsk);
3966 failed_mount2:
3967         for (i = 0; i < db_count; i++)
3968                 brelse(sbi->s_group_desc[i]);
3969         ext4_kvfree(sbi->s_group_desc);
3970 failed_mount:
3971         if (sbi->s_proc) {
3972                 remove_proc_entry(sb->s_id, ext4_proc_root);
3973         }
3974 #ifdef CONFIG_QUOTA
3975         for (i = 0; i < MAXQUOTAS; i++)
3976                 kfree(sbi->s_qf_names[i]);
3977 #endif
3978         ext4_blkdev_remove(sbi);
3979         brelse(bh);
3980 out_fail:
3981         sb->s_fs_info = NULL;
3982         kfree(sbi->s_blockgroup_lock);
3983         kfree(sbi);
3984 out_free_orig:
3985         kfree(orig_data);
3986         return ret;
3987 }
3988
3989 /*
3990  * Setup any per-fs journal parameters now.  We'll do this both on
3991  * initial mount, once the journal has been initialised but before we've
3992  * done any recovery; and again on any subsequent remount.
3993  */
3994 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3995 {
3996         struct ext4_sb_info *sbi = EXT4_SB(sb);
3997
3998         journal->j_commit_interval = sbi->s_commit_interval;
3999         journal->j_min_batch_time = sbi->s_min_batch_time;
4000         journal->j_max_batch_time = sbi->s_max_batch_time;
4001
4002         write_lock(&journal->j_state_lock);
4003         if (test_opt(sb, BARRIER))
4004                 journal->j_flags |= JBD2_BARRIER;
4005         else
4006                 journal->j_flags &= ~JBD2_BARRIER;
4007         if (test_opt(sb, DATA_ERR_ABORT))
4008                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4009         else
4010                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4011         write_unlock(&journal->j_state_lock);
4012 }
4013
4014 static journal_t *ext4_get_journal(struct super_block *sb,
4015                                    unsigned int journal_inum)
4016 {
4017         struct inode *journal_inode;
4018         journal_t *journal;
4019
4020         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4021
4022         /* First, test for the existence of a valid inode on disk.  Bad
4023          * things happen if we iget() an unused inode, as the subsequent
4024          * iput() will try to delete it. */
4025
4026         journal_inode = ext4_iget(sb, journal_inum);
4027         if (IS_ERR(journal_inode)) {
4028                 ext4_msg(sb, KERN_ERR, "no journal found");
4029                 return NULL;
4030         }
4031         if (!journal_inode->i_nlink) {
4032                 make_bad_inode(journal_inode);
4033                 iput(journal_inode);
4034                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4035                 return NULL;
4036         }
4037
4038         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4039                   journal_inode, journal_inode->i_size);
4040         if (!S_ISREG(journal_inode->i_mode)) {
4041                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4042                 iput(journal_inode);
4043                 return NULL;
4044         }
4045
4046         journal = jbd2_journal_init_inode(journal_inode);
4047         if (!journal) {
4048                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4049                 iput(journal_inode);
4050                 return NULL;
4051         }
4052         journal->j_private = sb;
4053         ext4_init_journal_params(sb, journal);
4054         return journal;
4055 }
4056
4057 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4058                                        dev_t j_dev)
4059 {
4060         struct buffer_head *bh;
4061         journal_t *journal;
4062         ext4_fsblk_t start;
4063         ext4_fsblk_t len;
4064         int hblock, blocksize;
4065         ext4_fsblk_t sb_block;
4066         unsigned long offset;
4067         struct ext4_super_block *es;
4068         struct block_device *bdev;
4069
4070         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4071
4072         bdev = ext4_blkdev_get(j_dev, sb);
4073         if (bdev == NULL)
4074                 return NULL;
4075
4076         blocksize = sb->s_blocksize;
4077         hblock = bdev_logical_block_size(bdev);
4078         if (blocksize < hblock) {
4079                 ext4_msg(sb, KERN_ERR,
4080                         "blocksize too small for journal device");
4081                 goto out_bdev;
4082         }
4083
4084         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4085         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4086         set_blocksize(bdev, blocksize);
4087         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4088                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4089                        "external journal");
4090                 goto out_bdev;
4091         }
4092
4093         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
4094         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4095             !(le32_to_cpu(es->s_feature_incompat) &
4096               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4097                 ext4_msg(sb, KERN_ERR, "external journal has "
4098                                         "bad superblock");
4099                 brelse(bh);
4100                 goto out_bdev;
4101         }
4102
4103         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4104                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4105                 brelse(bh);
4106                 goto out_bdev;
4107         }
4108
4109         len = ext4_blocks_count(es);
4110         start = sb_block + 1;
4111         brelse(bh);     /* we're done with the superblock */
4112
4113         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4114                                         start, len, blocksize);
4115         if (!journal) {
4116                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4117                 goto out_bdev;
4118         }
4119         journal->j_private = sb;
4120         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4121         wait_on_buffer(journal->j_sb_buffer);
4122         if (!buffer_uptodate(journal->j_sb_buffer)) {
4123                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4124                 goto out_journal;
4125         }
4126         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4127                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4128                                         "user (unsupported) - %d",
4129                         be32_to_cpu(journal->j_superblock->s_nr_users));
4130                 goto out_journal;
4131         }
4132         EXT4_SB(sb)->journal_bdev = bdev;
4133         ext4_init_journal_params(sb, journal);
4134         return journal;
4135
4136 out_journal:
4137         jbd2_journal_destroy(journal);
4138 out_bdev:
4139         ext4_blkdev_put(bdev);
4140         return NULL;
4141 }
4142
4143 static int ext4_load_journal(struct super_block *sb,
4144                              struct ext4_super_block *es,
4145                              unsigned long journal_devnum)
4146 {
4147         journal_t *journal;
4148         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4149         dev_t journal_dev;
4150         int err = 0;
4151         int really_read_only;
4152
4153         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4154
4155         if (journal_devnum &&
4156             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4157                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4158                         "numbers have changed");
4159                 journal_dev = new_decode_dev(journal_devnum);
4160         } else
4161                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4162
4163         really_read_only = bdev_read_only(sb->s_bdev);
4164
4165         /*
4166          * Are we loading a blank journal or performing recovery after a
4167          * crash?  For recovery, we need to check in advance whether we
4168          * can get read-write access to the device.
4169          */
4170         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4171                 if (sb->s_flags & MS_RDONLY) {
4172                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4173                                         "required on readonly filesystem");
4174                         if (really_read_only) {
4175                                 ext4_msg(sb, KERN_ERR, "write access "
4176                                         "unavailable, cannot proceed");
4177                                 return -EROFS;
4178                         }
4179                         ext4_msg(sb, KERN_INFO, "write access will "
4180                                "be enabled during recovery");
4181                 }
4182         }
4183
4184         if (journal_inum && journal_dev) {
4185                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4186                        "and inode journals!");
4187                 return -EINVAL;
4188         }
4189
4190         if (journal_inum) {
4191                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4192                         return -EINVAL;
4193         } else {
4194                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4195                         return -EINVAL;
4196         }
4197
4198         if (!(journal->j_flags & JBD2_BARRIER))
4199                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4200
4201         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4202                 err = jbd2_journal_update_format(journal);
4203                 if (err)  {
4204                         ext4_msg(sb, KERN_ERR, "error updating journal");
4205                         jbd2_journal_destroy(journal);
4206                         return err;
4207                 }
4208         }
4209
4210         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4211                 err = jbd2_journal_wipe(journal, !really_read_only);
4212         if (!err) {
4213                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4214                 if (save)
4215                         memcpy(save, ((char *) es) +
4216                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4217                 err = jbd2_journal_load(journal);
4218                 if (save)
4219                         memcpy(((char *) es) + EXT4_S_ERR_START,
4220                                save, EXT4_S_ERR_LEN);
4221                 kfree(save);
4222         }
4223
4224         if (err) {
4225                 ext4_msg(sb, KERN_ERR, "error loading journal");
4226                 jbd2_journal_destroy(journal);
4227                 return err;
4228         }
4229
4230         EXT4_SB(sb)->s_journal = journal;
4231         ext4_clear_journal_err(sb, es);
4232
4233         if (!really_read_only && journal_devnum &&
4234             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4235                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4236
4237                 /* Make sure we flush the recovery flag to disk. */
4238                 ext4_commit_super(sb, 1);
4239         }
4240
4241         return 0;
4242 }
4243
4244 static int ext4_commit_super(struct super_block *sb, int sync)
4245 {
4246         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4247         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4248         int error = 0;
4249
4250         if (!sbh || block_device_ejected(sb))
4251                 return error;
4252         if (buffer_write_io_error(sbh)) {
4253                 /*
4254                  * Oh, dear.  A previous attempt to write the
4255                  * superblock failed.  This could happen because the
4256                  * USB device was yanked out.  Or it could happen to
4257                  * be a transient write error and maybe the block will
4258                  * be remapped.  Nothing we can do but to retry the
4259                  * write and hope for the best.
4260                  */
4261                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4262                        "superblock detected");
4263                 clear_buffer_write_io_error(sbh);
4264                 set_buffer_uptodate(sbh);
4265         }
4266         /*
4267          * If the file system is mounted read-only, don't update the
4268          * superblock write time.  This avoids updating the superblock
4269          * write time when we are mounting the root file system
4270          * read/only but we need to replay the journal; at that point,
4271          * for people who are east of GMT and who make their clock
4272          * tick in localtime for Windows bug-for-bug compatibility,
4273          * the clock is set in the future, and this will cause e2fsck
4274          * to complain and force a full file system check.
4275          */
4276         if (!(sb->s_flags & MS_RDONLY))
4277                 es->s_wtime = cpu_to_le32(get_seconds());
4278         if (sb->s_bdev->bd_part)
4279                 es->s_kbytes_written =
4280                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4281                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4282                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4283         else
4284                 es->s_kbytes_written =
4285                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4286         ext4_free_blocks_count_set(es,
4287                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4288                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4289         es->s_free_inodes_count =
4290                 cpu_to_le32(percpu_counter_sum_positive(
4291                                 &EXT4_SB(sb)->s_freeinodes_counter));
4292         sb->s_dirt = 0;
4293         BUFFER_TRACE(sbh, "marking dirty");
4294         mark_buffer_dirty(sbh);
4295         if (sync) {
4296                 error = sync_dirty_buffer(sbh);
4297                 if (error)
4298                         return error;
4299
4300                 error = buffer_write_io_error(sbh);
4301                 if (error) {
4302                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4303                                "superblock");
4304                         clear_buffer_write_io_error(sbh);
4305                         set_buffer_uptodate(sbh);
4306                 }
4307         }
4308         return error;
4309 }
4310
4311 /*
4312  * Have we just finished recovery?  If so, and if we are mounting (or
4313  * remounting) the filesystem readonly, then we will end up with a
4314  * consistent fs on disk.  Record that fact.
4315  */
4316 static void ext4_mark_recovery_complete(struct super_block *sb,
4317                                         struct ext4_super_block *es)
4318 {
4319         journal_t *journal = EXT4_SB(sb)->s_journal;
4320
4321         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4322                 BUG_ON(journal != NULL);
4323                 return;
4324         }
4325         jbd2_journal_lock_updates(journal);
4326         if (jbd2_journal_flush(journal) < 0)
4327                 goto out;
4328
4329         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4330             sb->s_flags & MS_RDONLY) {
4331                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4332                 ext4_commit_super(sb, 1);
4333         }
4334
4335 out:
4336         jbd2_journal_unlock_updates(journal);
4337 }
4338
4339 /*
4340  * If we are mounting (or read-write remounting) a filesystem whose journal
4341  * has recorded an error from a previous lifetime, move that error to the
4342  * main filesystem now.
4343  */
4344 static void ext4_clear_journal_err(struct super_block *sb,
4345                                    struct ext4_super_block *es)
4346 {
4347         journal_t *journal;
4348         int j_errno;
4349         const char *errstr;
4350
4351         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4352
4353         journal = EXT4_SB(sb)->s_journal;
4354
4355         /*
4356          * Now check for any error status which may have been recorded in the
4357          * journal by a prior ext4_error() or ext4_abort()
4358          */
4359
4360         j_errno = jbd2_journal_errno(journal);
4361         if (j_errno) {
4362                 char nbuf[16];
4363
4364                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4365                 ext4_warning(sb, "Filesystem error recorded "
4366                              "from previous mount: %s", errstr);
4367                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4368
4369                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4370                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4371                 ext4_commit_super(sb, 1);
4372
4373                 jbd2_journal_clear_err(journal);
4374         }
4375 }
4376
4377 /*
4378  * Force the running and committing transactions to commit,
4379  * and wait on the commit.
4380  */
4381 int ext4_force_commit(struct super_block *sb)
4382 {
4383         journal_t *journal;
4384         int ret = 0;
4385
4386         if (sb->s_flags & MS_RDONLY)
4387                 return 0;
4388
4389         journal = EXT4_SB(sb)->s_journal;
4390         if (journal) {
4391                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4392                 ret = ext4_journal_force_commit(journal);
4393         }
4394
4395         return ret;
4396 }
4397
4398 static void ext4_write_super(struct super_block *sb)
4399 {
4400         lock_super(sb);
4401         ext4_commit_super(sb, 1);
4402         unlock_super(sb);
4403 }
4404
4405 static int ext4_sync_fs(struct super_block *sb, int wait)
4406 {
4407         int ret = 0;
4408         tid_t target;
4409         struct ext4_sb_info *sbi = EXT4_SB(sb);
4410
4411         trace_ext4_sync_fs(sb, wait);
4412         flush_workqueue(sbi->dio_unwritten_wq);
4413         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4414                 if (wait)
4415                         jbd2_log_wait_commit(sbi->s_journal, target);
4416         }
4417         return ret;
4418 }
4419
4420 /*
4421  * LVM calls this function before a (read-only) snapshot is created.  This
4422  * gives us a chance to flush the journal completely and mark the fs clean.
4423  *
4424  * Note that only this function cannot bring a filesystem to be in a clean
4425  * state independently, because ext4 prevents a new handle from being started
4426  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4427  * the upper layer.
4428  */
4429 static int ext4_freeze(struct super_block *sb)
4430 {
4431         int error = 0;
4432         journal_t *journal;
4433
4434         if (sb->s_flags & MS_RDONLY)
4435                 return 0;
4436
4437         journal = EXT4_SB(sb)->s_journal;
4438
4439         /* Now we set up the journal barrier. */
4440         jbd2_journal_lock_updates(journal);
4441
4442         /*
4443          * Don't clear the needs_recovery flag if we failed to flush
4444          * the journal.
4445          */
4446         error = jbd2_journal_flush(journal);
4447         if (error < 0)
4448                 goto out;
4449
4450         /* Journal blocked and flushed, clear needs_recovery flag. */
4451         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4452         error = ext4_commit_super(sb, 1);
4453 out:
4454         /* we rely on s_frozen to stop further updates */
4455         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4456         return error;
4457 }
4458
4459 /*
4460  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4461  * flag here, even though the filesystem is not technically dirty yet.
4462  */
4463 static int ext4_unfreeze(struct super_block *sb)
4464 {
4465         if (sb->s_flags & MS_RDONLY)
4466                 return 0;
4467
4468         lock_super(sb);
4469         /* Reset the needs_recovery flag before the fs is unlocked. */
4470         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4471         ext4_commit_super(sb, 1);
4472         unlock_super(sb);
4473         return 0;
4474 }
4475
4476 /*
4477  * Structure to save mount options for ext4_remount's benefit
4478  */
4479 struct ext4_mount_options {
4480         unsigned long s_mount_opt;
4481         unsigned long s_mount_opt2;
4482         uid_t s_resuid;
4483         gid_t s_resgid;
4484         unsigned long s_commit_interval;
4485         u32 s_min_batch_time, s_max_batch_time;
4486 #ifdef CONFIG_QUOTA
4487         int s_jquota_fmt;
4488         char *s_qf_names[MAXQUOTAS];
4489 #endif
4490 };
4491
4492 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4493 {
4494         struct ext4_super_block *es;
4495         struct ext4_sb_info *sbi = EXT4_SB(sb);
4496         ext4_fsblk_t n_blocks_count = 0;
4497         unsigned long old_sb_flags;
4498         struct ext4_mount_options old_opts;
4499         int enable_quota = 0;
4500         ext4_group_t g;
4501         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4502         int err = 0;
4503 #ifdef CONFIG_QUOTA
4504         int i;
4505 #endif
4506         char *orig_data = kstrdup(data, GFP_KERNEL);
4507
4508         /* Store the original options */
4509         lock_super(sb);
4510         old_sb_flags = sb->s_flags;
4511         old_opts.s_mount_opt = sbi->s_mount_opt;
4512         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4513         old_opts.s_resuid = sbi->s_resuid;
4514         old_opts.s_resgid = sbi->s_resgid;
4515         old_opts.s_commit_interval = sbi->s_commit_interval;
4516         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4517         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4518 #ifdef CONFIG_QUOTA
4519         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4520         for (i = 0; i < MAXQUOTAS; i++)
4521                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4522 #endif
4523         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4524                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4525
4526         /*
4527          * Allow the "check" option to be passed as a remount option.
4528          */
4529         if (!parse_options(data, sb, NULL, &journal_ioprio,
4530                            &n_blocks_count, 1)) {
4531                 err = -EINVAL;
4532                 goto restore_opts;
4533         }
4534
4535         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4536                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4537                         ext4_msg(sb, KERN_ERR, "can't mount with "
4538                                  "both data=journal and delalloc");
4539                         err = -EINVAL;
4540                         goto restore_opts;
4541                 }
4542                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4543                         ext4_msg(sb, KERN_ERR, "can't mount with "
4544                                  "both data=journal and dioread_nolock");
4545                         err = -EINVAL;
4546                         goto restore_opts;
4547                 }
4548         }
4549
4550         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4551                 ext4_abort(sb, "Abort forced by user");
4552
4553         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4554                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4555
4556         es = sbi->s_es;
4557
4558         if (sbi->s_journal) {
4559                 ext4_init_journal_params(sb, sbi->s_journal);
4560                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4561         }
4562
4563         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4564                 n_blocks_count > ext4_blocks_count(es)) {
4565                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4566                         err = -EROFS;
4567                         goto restore_opts;
4568                 }
4569
4570                 if (*flags & MS_RDONLY) {
4571                         err = dquot_suspend(sb, -1);
4572                         if (err < 0)
4573                                 goto restore_opts;
4574
4575                         /*
4576                          * First of all, the unconditional stuff we have to do
4577                          * to disable replay of the journal when we next remount
4578                          */
4579                         sb->s_flags |= MS_RDONLY;
4580
4581                         /*
4582                          * OK, test if we are remounting a valid rw partition
4583                          * readonly, and if so set the rdonly flag and then
4584                          * mark the partition as valid again.
4585                          */
4586                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4587                             (sbi->s_mount_state & EXT4_VALID_FS))
4588                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4589
4590                         if (sbi->s_journal)
4591                                 ext4_mark_recovery_complete(sb, es);
4592                 } else {
4593                         /* Make sure we can mount this feature set readwrite */
4594                         if (!ext4_feature_set_ok(sb, 0)) {
4595                                 err = -EROFS;
4596                                 goto restore_opts;
4597                         }
4598                         /*
4599                          * Make sure the group descriptor checksums
4600                          * are sane.  If they aren't, refuse to remount r/w.
4601                          */
4602                         for (g = 0; g < sbi->s_groups_count; g++) {
4603                                 struct ext4_group_desc *gdp =
4604                                         ext4_get_group_desc(sb, g, NULL);
4605
4606                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4607                                         ext4_msg(sb, KERN_ERR,
4608                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4609                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4610                                                le16_to_cpu(gdp->bg_checksum));
4611                                         err = -EINVAL;
4612                                         goto restore_opts;
4613                                 }
4614                         }
4615
4616                         /*
4617                          * If we have an unprocessed orphan list hanging
4618                          * around from a previously readonly bdev mount,
4619                          * require a full umount/remount for now.
4620                          */
4621                         if (es->s_last_orphan) {
4622                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4623                                        "remount RDWR because of unprocessed "
4624                                        "orphan inode list.  Please "
4625                                        "umount/remount instead");
4626                                 err = -EINVAL;
4627                                 goto restore_opts;
4628                         }
4629
4630                         /*
4631                          * Mounting a RDONLY partition read-write, so reread
4632                          * and store the current valid flag.  (It may have
4633                          * been changed by e2fsck since we originally mounted
4634                          * the partition.)
4635                          */
4636                         if (sbi->s_journal)
4637                                 ext4_clear_journal_err(sb, es);
4638                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4639                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4640                                 goto restore_opts;
4641                         if (!ext4_setup_super(sb, es, 0))
4642                                 sb->s_flags &= ~MS_RDONLY;
4643                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4644                                                      EXT4_FEATURE_INCOMPAT_MMP))
4645                                 if (ext4_multi_mount_protect(sb,
4646                                                 le64_to_cpu(es->s_mmp_block))) {
4647                                         err = -EROFS;
4648                                         goto restore_opts;
4649                                 }
4650                         enable_quota = 1;
4651                 }
4652         }
4653
4654         /*
4655          * Reinitialize lazy itable initialization thread based on
4656          * current settings
4657          */
4658         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4659                 ext4_unregister_li_request(sb);
4660         else {
4661                 ext4_group_t first_not_zeroed;
4662                 first_not_zeroed = ext4_has_uninit_itable(sb);
4663                 ext4_register_li_request(sb, first_not_zeroed);
4664         }
4665
4666         ext4_setup_system_zone(sb);
4667         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4668                 ext4_commit_super(sb, 1);
4669
4670 #ifdef CONFIG_QUOTA
4671         /* Release old quota file names */
4672         for (i = 0; i < MAXQUOTAS; i++)
4673                 if (old_opts.s_qf_names[i] &&
4674                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4675                         kfree(old_opts.s_qf_names[i]);
4676 #endif
4677         unlock_super(sb);
4678         if (enable_quota)
4679                 dquot_resume(sb, -1);
4680
4681         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4682         kfree(orig_data);
4683         return 0;
4684
4685 restore_opts:
4686         sb->s_flags = old_sb_flags;
4687         sbi->s_mount_opt = old_opts.s_mount_opt;
4688         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4689         sbi->s_resuid = old_opts.s_resuid;
4690         sbi->s_resgid = old_opts.s_resgid;
4691         sbi->s_commit_interval = old_opts.s_commit_interval;
4692         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4693         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4694 #ifdef CONFIG_QUOTA
4695         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4696         for (i = 0; i < MAXQUOTAS; i++) {
4697                 if (sbi->s_qf_names[i] &&
4698                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4699                         kfree(sbi->s_qf_names[i]);
4700                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4701         }
4702 #endif
4703         unlock_super(sb);
4704         kfree(orig_data);
4705         return err;
4706 }
4707
4708 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4709 {
4710         struct super_block *sb = dentry->d_sb;
4711         struct ext4_sb_info *sbi = EXT4_SB(sb);
4712         struct ext4_super_block *es = sbi->s_es;
4713         ext4_fsblk_t overhead = 0;
4714         u64 fsid;
4715         s64 bfree;
4716
4717         if (!test_opt(sb, MINIX_DF))
4718                 overhead = sbi->s_overhead;
4719
4720         buf->f_type = EXT4_SUPER_MAGIC;
4721         buf->f_bsize = sb->s_blocksize;
4722         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4723         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4724                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4725         /* prevent underflow in case that few free space is available */
4726         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4727         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4728         if (buf->f_bfree < ext4_r_blocks_count(es))
4729                 buf->f_bavail = 0;
4730         buf->f_files = le32_to_cpu(es->s_inodes_count);
4731         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4732         buf->f_namelen = EXT4_NAME_LEN;
4733         fsid = le64_to_cpup((void *)es->s_uuid) ^
4734                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4735         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4736         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4737
4738         return 0;
4739 }
4740
4741 /* Helper function for writing quotas on sync - we need to start transaction
4742  * before quota file is locked for write. Otherwise the are possible deadlocks:
4743  * Process 1                         Process 2
4744  * ext4_create()                     quota_sync()
4745  *   jbd2_journal_start()                  write_dquot()
4746  *   dquot_initialize()                         down(dqio_mutex)
4747  *     down(dqio_mutex)                    jbd2_journal_start()
4748  *
4749  */
4750
4751 #ifdef CONFIG_QUOTA
4752
4753 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4754 {
4755         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4756 }
4757
4758 static int ext4_write_dquot(struct dquot *dquot)
4759 {
4760         int ret, err;
4761         handle_t *handle;
4762         struct inode *inode;
4763
4764         inode = dquot_to_inode(dquot);
4765         handle = ext4_journal_start(inode,
4766                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4767         if (IS_ERR(handle))
4768                 return PTR_ERR(handle);
4769         ret = dquot_commit(dquot);
4770         err = ext4_journal_stop(handle);
4771         if (!ret)
4772                 ret = err;
4773         return ret;
4774 }
4775
4776 static int ext4_acquire_dquot(struct dquot *dquot)
4777 {
4778         int ret, err;
4779         handle_t *handle;
4780
4781         handle = ext4_journal_start(dquot_to_inode(dquot),
4782                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4783         if (IS_ERR(handle))
4784                 return PTR_ERR(handle);
4785         ret = dquot_acquire(dquot);
4786         err = ext4_journal_stop(handle);
4787         if (!ret)
4788                 ret = err;
4789         return ret;
4790 }
4791
4792 static int ext4_release_dquot(struct dquot *dquot)
4793 {
4794         int ret, err;
4795         handle_t *handle;
4796
4797         handle = ext4_journal_start(dquot_to_inode(dquot),
4798                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4799         if (IS_ERR(handle)) {
4800                 /* Release dquot anyway to avoid endless cycle in dqput() */
4801                 dquot_release(dquot);
4802                 return PTR_ERR(handle);
4803         }
4804         ret = dquot_release(dquot);
4805         err = ext4_journal_stop(handle);
4806         if (!ret)
4807                 ret = err;
4808         return ret;
4809 }
4810
4811 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4812 {
4813         /* Are we journaling quotas? */
4814         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4815             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4816                 dquot_mark_dquot_dirty(dquot);
4817                 return ext4_write_dquot(dquot);
4818         } else {
4819                 return dquot_mark_dquot_dirty(dquot);
4820         }
4821 }
4822
4823 static int ext4_write_info(struct super_block *sb, int type)
4824 {
4825         int ret, err;
4826         handle_t *handle;
4827
4828         /* Data block + inode block */
4829         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4830         if (IS_ERR(handle))
4831                 return PTR_ERR(handle);
4832         ret = dquot_commit_info(sb, type);
4833         err = ext4_journal_stop(handle);
4834         if (!ret)
4835                 ret = err;
4836         return ret;
4837 }
4838
4839 /*
4840  * Turn on quotas during mount time - we need to find
4841  * the quota file and such...
4842  */
4843 static int ext4_quota_on_mount(struct super_block *sb, int type)
4844 {
4845         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4846                                         EXT4_SB(sb)->s_jquota_fmt, type);
4847 }
4848
4849 /*
4850  * Standard function to be called on quota_on
4851  */
4852 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4853                          struct path *path)
4854 {
4855         int err;
4856
4857         if (!test_opt(sb, QUOTA))
4858                 return -EINVAL;
4859
4860         /* Quotafile not on the same filesystem? */
4861         if (path->mnt->mnt_sb != sb)
4862                 return -EXDEV;
4863         /* Journaling quota? */
4864         if (EXT4_SB(sb)->s_qf_names[type]) {
4865                 /* Quotafile not in fs root? */
4866                 if (path->dentry->d_parent != sb->s_root)
4867                         ext4_msg(sb, KERN_WARNING,
4868                                 "Quota file not on filesystem root. "
4869                                 "Journaled quota will not work");
4870         }
4871
4872         /*
4873          * When we journal data on quota file, we have to flush journal to see
4874          * all updates to the file when we bypass pagecache...
4875          */
4876         if (EXT4_SB(sb)->s_journal &&
4877             ext4_should_journal_data(path->dentry->d_inode)) {
4878                 /*
4879                  * We don't need to lock updates but journal_flush() could
4880                  * otherwise be livelocked...
4881                  */
4882                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4883                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4884                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4885                 if (err)
4886                         return err;
4887         }
4888
4889         return dquot_quota_on(sb, type, format_id, path);
4890 }
4891
4892 static int ext4_quota_off(struct super_block *sb, int type)
4893 {
4894         struct inode *inode = sb_dqopt(sb)->files[type];
4895         handle_t *handle;
4896
4897         /* Force all delayed allocation blocks to be allocated.
4898          * Caller already holds s_umount sem */
4899         if (test_opt(sb, DELALLOC))
4900                 sync_filesystem(sb);
4901
4902         if (!inode)
4903                 goto out;
4904
4905         /* Update modification times of quota files when userspace can
4906          * start looking at them */
4907         handle = ext4_journal_start(inode, 1);
4908         if (IS_ERR(handle))
4909                 goto out;
4910         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4911         ext4_mark_inode_dirty(handle, inode);
4912         ext4_journal_stop(handle);
4913
4914 out:
4915         return dquot_quota_off(sb, type);
4916 }
4917
4918 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4919  * acquiring the locks... As quota files are never truncated and quota code
4920  * itself serializes the operations (and no one else should touch the files)
4921  * we don't have to be afraid of races */
4922 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4923                                size_t len, loff_t off)
4924 {
4925         struct inode *inode = sb_dqopt(sb)->files[type];
4926         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4927         int err = 0;
4928         int offset = off & (sb->s_blocksize - 1);
4929         int tocopy;
4930         size_t toread;
4931         struct buffer_head *bh;
4932         loff_t i_size = i_size_read(inode);
4933
4934         if (off > i_size)
4935                 return 0;
4936         if (off+len > i_size)
4937                 len = i_size-off;
4938         toread = len;
4939         while (toread > 0) {
4940                 tocopy = sb->s_blocksize - offset < toread ?
4941                                 sb->s_blocksize - offset : toread;
4942                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4943                 if (err)
4944                         return err;
4945                 if (!bh)        /* A hole? */
4946                         memset(data, 0, tocopy);
4947                 else
4948                         memcpy(data, bh->b_data+offset, tocopy);
4949                 brelse(bh);
4950                 offset = 0;
4951                 toread -= tocopy;
4952                 data += tocopy;
4953                 blk++;
4954         }
4955         return len;
4956 }
4957
4958 /* Write to quotafile (we know the transaction is already started and has
4959  * enough credits) */
4960 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4961                                 const char *data, size_t len, loff_t off)
4962 {
4963         struct inode *inode = sb_dqopt(sb)->files[type];
4964         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4965         int err = 0;
4966         int offset = off & (sb->s_blocksize - 1);
4967         struct buffer_head *bh;
4968         handle_t *handle = journal_current_handle();
4969
4970         if (EXT4_SB(sb)->s_journal && !handle) {
4971                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4972                         " cancelled because transaction is not started",
4973                         (unsigned long long)off, (unsigned long long)len);
4974                 return -EIO;
4975         }
4976         /*
4977          * Since we account only one data block in transaction credits,
4978          * then it is impossible to cross a block boundary.
4979          */
4980         if (sb->s_blocksize - offset < len) {
4981                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4982                         " cancelled because not block aligned",
4983                         (unsigned long long)off, (unsigned long long)len);
4984                 return -EIO;
4985         }
4986
4987         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4988         bh = ext4_bread(handle, inode, blk, 1, &err);
4989         if (!bh)
4990                 goto out;
4991         err = ext4_journal_get_write_access(handle, bh);
4992         if (err) {
4993                 brelse(bh);
4994                 goto out;
4995         }
4996         lock_buffer(bh);
4997         memcpy(bh->b_data+offset, data, len);
4998         flush_dcache_page(bh->b_page);
4999         unlock_buffer(bh);
5000         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5001         brelse(bh);
5002 out:
5003         if (err) {
5004                 mutex_unlock(&inode->i_mutex);
5005                 return err;
5006         }
5007         if (inode->i_size < off + len) {
5008                 i_size_write(inode, off + len);
5009                 EXT4_I(inode)->i_disksize = inode->i_size;
5010                 ext4_mark_inode_dirty(handle, inode);
5011         }
5012         mutex_unlock(&inode->i_mutex);
5013         return len;
5014 }
5015
5016 #endif
5017
5018 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5019                        const char *dev_name, void *data)
5020 {
5021         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5022 }
5023
5024 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5025 static inline void register_as_ext2(void)
5026 {
5027         int err = register_filesystem(&ext2_fs_type);
5028         if (err)
5029                 printk(KERN_WARNING
5030                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5031 }
5032
5033 static inline void unregister_as_ext2(void)
5034 {
5035         unregister_filesystem(&ext2_fs_type);
5036 }
5037
5038 static inline int ext2_feature_set_ok(struct super_block *sb)
5039 {
5040         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5041                 return 0;
5042         if (sb->s_flags & MS_RDONLY)
5043                 return 1;
5044         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5045                 return 0;
5046         return 1;
5047 }
5048 MODULE_ALIAS("ext2");
5049 #else
5050 static inline void register_as_ext2(void) { }
5051 static inline void unregister_as_ext2(void) { }
5052 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5053 #endif
5054
5055 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5056 static inline void register_as_ext3(void)
5057 {
5058         int err = register_filesystem(&ext3_fs_type);
5059         if (err)
5060                 printk(KERN_WARNING
5061                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5062 }
5063
5064 static inline void unregister_as_ext3(void)
5065 {
5066         unregister_filesystem(&ext3_fs_type);
5067 }
5068
5069 static inline int ext3_feature_set_ok(struct super_block *sb)
5070 {
5071         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5072                 return 0;
5073         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5074                 return 0;
5075         if (sb->s_flags & MS_RDONLY)
5076                 return 1;
5077         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5078                 return 0;
5079         return 1;
5080 }
5081 MODULE_ALIAS("ext3");
5082 #else
5083 static inline void register_as_ext3(void) { }
5084 static inline void unregister_as_ext3(void) { }
5085 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5086 #endif
5087
5088 static struct file_system_type ext4_fs_type = {
5089         .owner          = THIS_MODULE,
5090         .name           = "ext4",
5091         .mount          = ext4_mount,
5092         .kill_sb        = kill_block_super,
5093         .fs_flags       = FS_REQUIRES_DEV,
5094 };
5095
5096 static int __init ext4_init_feat_adverts(void)
5097 {
5098         struct ext4_features *ef;
5099         int ret = -ENOMEM;
5100
5101         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5102         if (!ef)
5103                 goto out;
5104
5105         ef->f_kobj.kset = ext4_kset;
5106         init_completion(&ef->f_kobj_unregister);
5107         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5108                                    "features");
5109         if (ret) {
5110                 kfree(ef);
5111                 goto out;
5112         }
5113
5114         ext4_feat = ef;
5115         ret = 0;
5116 out:
5117         return ret;
5118 }
5119
5120 static void ext4_exit_feat_adverts(void)
5121 {
5122         kobject_put(&ext4_feat->f_kobj);
5123         wait_for_completion(&ext4_feat->f_kobj_unregister);
5124         kfree(ext4_feat);
5125 }
5126
5127 /* Shared across all ext4 file systems */
5128 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5129 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5130
5131 static int __init ext4_init_fs(void)
5132 {
5133         int i, err;
5134
5135         ext4_check_flag_values();
5136
5137         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5138                 mutex_init(&ext4__aio_mutex[i]);
5139                 init_waitqueue_head(&ext4__ioend_wq[i]);
5140         }
5141
5142         err = ext4_init_pageio();
5143         if (err)
5144                 return err;
5145         err = ext4_init_system_zone();
5146         if (err)
5147                 goto out6;
5148         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5149         if (!ext4_kset)
5150                 goto out5;
5151         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5152
5153         err = ext4_init_feat_adverts();
5154         if (err)
5155                 goto out4;
5156
5157         err = ext4_init_mballoc();
5158         if (err)
5159                 goto out3;
5160
5161         err = ext4_init_xattr();
5162         if (err)
5163                 goto out2;
5164         err = init_inodecache();
5165         if (err)
5166                 goto out1;
5167         register_as_ext3();
5168         register_as_ext2();
5169         err = register_filesystem(&ext4_fs_type);
5170         if (err)
5171                 goto out;
5172
5173         ext4_li_info = NULL;
5174         mutex_init(&ext4_li_mtx);
5175         return 0;
5176 out:
5177         unregister_as_ext2();
5178         unregister_as_ext3();
5179         destroy_inodecache();
5180 out1:
5181         ext4_exit_xattr();
5182 out2:
5183         ext4_exit_mballoc();
5184 out3:
5185         ext4_exit_feat_adverts();
5186 out4:
5187         if (ext4_proc_root)
5188                 remove_proc_entry("fs/ext4", NULL);
5189         kset_unregister(ext4_kset);
5190 out5:
5191         ext4_exit_system_zone();
5192 out6:
5193         ext4_exit_pageio();
5194         return err;
5195 }
5196
5197 static void __exit ext4_exit_fs(void)
5198 {
5199         ext4_destroy_lazyinit_thread();
5200         unregister_as_ext2();
5201         unregister_as_ext3();
5202         unregister_filesystem(&ext4_fs_type);
5203         destroy_inodecache();
5204         ext4_exit_xattr();
5205         ext4_exit_mballoc();
5206         ext4_exit_feat_adverts();
5207         remove_proc_entry("fs/ext4", NULL);
5208         kset_unregister(ext4_kset);
5209         ext4_exit_system_zone();
5210         ext4_exit_pageio();
5211 }
5212
5213 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5214 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5215 MODULE_LICENSE("GPL");
5216 module_init(ext4_init_fs)
5217 module_exit(ext4_exit_fs)