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         invalidate_bdev(sb->s_bdev);
 861         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 862                 /*
 863                  * Invalidate the journal device's buffers.  We don't want them
 864                  * floating about in memory - the physical journal device may
 865                  * hotswapped, and it breaks the `ro-after' testing code.
 866                  */
 867                 sync_blockdev(sbi->journal_bdev);
 868                 invalidate_bdev(sbi->journal_bdev);
 869                 ext4_blkdev_remove(sbi);
 870         }
 871         if (sbi->s_mmp_tsk)
 872                 kthread_stop(sbi->s_mmp_tsk);
 873         sb->s_fs_info = NULL;
 874         /*
 875          * Now that we are completely done shutting down the
 876          * superblock, we need to actually destroy the kobject.
 877          */
 878         unlock_super(sb);
 879         kobject_put(&sbi->s_kobj);
 880         wait_for_completion(&sbi->s_kobj_unregister);
 881         kfree(sbi->s_blockgroup_lock);
 882         kfree(sbi);
 883 }
 884
 885 static struct kmem_cache *ext4_inode_cachep;
 886
 887 /*
 888  * Called inside transaction, so use GFP_NOFS
 889  */
 890 static struct inode *ext4_alloc_inode(struct super_block *sb)
 891 {
 892         struct ext4_inode_info *ei;
 893
 894         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 895         if (!ei)
 896                 return NULL;
 897
 898         ei->vfs_inode.i_version = 1;
 899         ei->vfs_inode.i_data.writeback_index = 0;
 900         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 901         INIT_LIST_HEAD(&ei->i_prealloc_list);
 902         spin_lock_init(&ei->i_prealloc_lock);
 903         ei->i_reserved_data_blocks = 0;
 904         ei->i_reserved_meta_blocks = 0;
 905         ei->i_allocated_meta_blocks = 0;
 906         ei->i_da_metadata_calc_len = 0;
 907         ei->i_da_metadata_calc_last_lblock = 0;
 908         spin_lock_init(&(ei->i_block_reservation_lock));
 909 #ifdef CONFIG_QUOTA
 910         ei->i_reserved_quota = 0;
 911 #endif
 912         ei->jinode = NULL;
 913         INIT_LIST_HEAD(&ei->i_completed_io_list);
 914         spin_lock_init(&ei->i_completed_io_lock);
 915         ei->cur_aio_dio = NULL;
 916         ei->i_sync_tid = 0;
 917         ei->i_datasync_tid = 0;
 918         atomic_set(&ei->i_ioend_count, 0);
 919         atomic_set(&ei->i_aiodio_unwritten, 0);
 920
 921         return &ei->vfs_inode;
 922 }
 923
 924 static int ext4_drop_inode(struct inode *inode)
 925 {
 926         int drop = generic_drop_inode(inode);
 927
 928         trace_ext4_drop_inode(inode, drop);
 929         return drop;
 930 }
 931
 932 static void ext4_i_callback(struct rcu_head *head)
 933 {
 934         struct inode *inode = container_of(head, struct inode, i_rcu);
 935         INIT_LIST_HEAD(&inode->i_dentry);
 936         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 937 }
 938
 939 static void ext4_destroy_inode(struct inode *inode)
 940 {
 941         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 942                 ext4_msg(inode->i_sb, KERN_ERR,
 943                          "Inode %lu (%p): orphan list check failed!",
 944                          inode->i_ino, EXT4_I(inode));
 945                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 946                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 947                                 true);
 948                 dump_stack();
 949         }
 950         call_rcu(&inode->i_rcu, ext4_i_callback);
 951 }
 952
 953 static void init_once(void *foo)
 954 {
 955         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 956
 957         INIT_LIST_HEAD(&ei->i_orphan);
 958 #ifdef CONFIG_EXT4_FS_XATTR
 959         init_rwsem(&ei->xattr_sem);
 960 #endif
 961         init_rwsem(&ei->i_data_sem);
 962         inode_init_once(&ei->vfs_inode);
 963 }
 964
 965 static int init_inodecache(void)
 966 {
 967         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 968                                              sizeof(struct ext4_inode_info),
 969                                              0, (SLAB_RECLAIM_ACCOUNT|
 970                                                 SLAB_MEM_SPREAD),
 971                                              init_once);
 972         if (ext4_inode_cachep == NULL)
 973                 return -ENOMEM;
 974         return 0;
 975 }
 976
 977 static void destroy_inodecache(void)
 978 {
 979         kmem_cache_destroy(ext4_inode_cachep);
 980 }
 981
 982 void ext4_clear_inode(struct inode *inode)
 983 {
 984         invalidate_inode_buffers(inode);
 985         end_writeback(inode);
 986         dquot_drop(inode);
 987         ext4_discard_preallocations(inode);
 988         if (EXT4_I(inode)->jinode) {
 989                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 990                                                EXT4_I(inode)->jinode);
 991                 jbd2_free_inode(EXT4_I(inode)->jinode);
 992                 EXT4_I(inode)->jinode = NULL;
 993         }
 994 }
 995
 996 static inline void ext4_show_quota_options(struct seq_file *seq,
 997                                            struct super_block *sb)
 998 {
 999 #if defined(CONFIG_QUOTA)
1000         struct ext4_sb_info *sbi = EXT4_SB(sb);
1001
1002         if (sbi->s_jquota_fmt) {
1003                 char *fmtname = "";
1004
1005                 switch (sbi->s_jquota_fmt) {
1006                 case QFMT_VFS_OLD:
1007                         fmtname = "vfsold";
1008                         break;
1009                 case QFMT_VFS_V0:
1010                         fmtname = "vfsv0";
1011                         break;
1012                 case QFMT_VFS_V1:
1013                         fmtname = "vfsv1";
1014                         break;
1015                 }
1016                 seq_printf(seq, ",jqfmt=%s", fmtname);
1017         }
1018
1019         if (sbi->s_qf_names[USRQUOTA])
1020                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1021
1022         if (sbi->s_qf_names[GRPQUOTA])
1023                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1024
1025         if (test_opt(sb, USRQUOTA))
1026                 seq_puts(seq, ",usrquota");
1027
1028         if (test_opt(sb, GRPQUOTA))
1029                 seq_puts(seq, ",grpquota");
1030 #endif
1031 }
1032
1033 /*
1034  * Show an option if
1035  *  - it's set to a non-default value OR
1036  *  - if the per-sb default is different from the global default
1037  */
1038 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1039 {
1040         int def_errors;
1041         unsigned long def_mount_opts;
1042         struct super_block *sb = vfs->mnt_sb;
1043         struct ext4_sb_info *sbi = EXT4_SB(sb);
1044         struct ext4_super_block *es = sbi->s_es;
1045
1046         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1047         def_errors     = le16_to_cpu(es->s_errors);
1048
1049         if (sbi->s_sb_block != 1)
1050                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1051         if (test_opt(sb, MINIX_DF))
1052                 seq_puts(seq, ",minixdf");
1053         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1054                 seq_puts(seq, ",grpid");
1055         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1056                 seq_puts(seq, ",nogrpid");
1057         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1058             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1059                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1060         }
1061         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1062             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1063                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1064         }
1065         if (test_opt(sb, ERRORS_RO)) {
1066                 if (def_errors == EXT4_ERRORS_PANIC ||
1067                     def_errors == EXT4_ERRORS_CONTINUE) {
1068                         seq_puts(seq, ",errors=remount-ro");
1069                 }
1070         }
1071         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1072                 seq_puts(seq, ",errors=continue");
1073         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1074                 seq_puts(seq, ",errors=panic");
1075         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1076                 seq_puts(seq, ",nouid32");
1077         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1078                 seq_puts(seq, ",debug");
1079 #ifdef CONFIG_EXT4_FS_XATTR
1080         if (test_opt(sb, XATTR_USER))
1081                 seq_puts(seq, ",user_xattr");
1082         if (!test_opt(sb, XATTR_USER))
1083                 seq_puts(seq, ",nouser_xattr");
1084 #endif
1085 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1086         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1087                 seq_puts(seq, ",acl");
1088         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1089                 seq_puts(seq, ",noacl");
1090 #endif
1091         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1092                 seq_printf(seq, ",commit=%u",
1093                            (unsigned) (sbi->s_commit_interval / HZ));
1094         }
1095         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1096                 seq_printf(seq, ",min_batch_time=%u",
1097                            (unsigned) sbi->s_min_batch_time);
1098         }
1099         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1100                 seq_printf(seq, ",max_batch_time=%u",
1101                            (unsigned) sbi->s_max_batch_time);
1102         }
1103
1104         /*
1105          * We're changing the default of barrier mount option, so
1106          * let's always display its mount state so it's clear what its
1107          * status is.
1108          */
1109         seq_puts(seq, ",barrier=");
1110         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1111         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1112                 seq_puts(seq, ",journal_async_commit");
1113         else if (test_opt(sb, JOURNAL_CHECKSUM))
1114                 seq_puts(seq, ",journal_checksum");
1115         if (test_opt(sb, I_VERSION))
1116                 seq_puts(seq, ",i_version");
1117         if (!test_opt(sb, DELALLOC) &&
1118             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1119                 seq_puts(seq, ",nodelalloc");
1120
1121         if (!test_opt(sb, MBLK_IO_SUBMIT))
1122                 seq_puts(seq, ",nomblk_io_submit");
1123         if (sbi->s_stripe)
1124                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1125         /*
1126          * journal mode get enabled in different ways
1127          * So just print the value even if we didn't specify it
1128          */
1129         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1130                 seq_puts(seq, ",data=journal");
1131         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1132                 seq_puts(seq, ",data=ordered");
1133         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1134                 seq_puts(seq, ",data=writeback");
1135
1136         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1137                 seq_printf(seq, ",inode_readahead_blks=%u",
1138                            sbi->s_inode_readahead_blks);
1139
1140         if (test_opt(sb, DATA_ERR_ABORT))
1141                 seq_puts(seq, ",data_err=abort");
1142
1143         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1144                 seq_puts(seq, ",noauto_da_alloc");
1145
1146         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1147                 seq_puts(seq, ",discard");
1148
1149         if (test_opt(sb, NOLOAD))
1150                 seq_puts(seq, ",norecovery");
1151
1152         if (test_opt(sb, DIOREAD_NOLOCK))
1153                 seq_puts(seq, ",dioread_nolock");
1154
1155         if (test_opt(sb, BLOCK_VALIDITY) &&
1156             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1157                 seq_puts(seq, ",block_validity");
1158
1159         if (!test_opt(sb, INIT_INODE_TABLE))
1160                 seq_puts(seq, ",noinit_itable");
1161         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1162                 seq_printf(seq, ",init_itable=%u",
1163                            (unsigned) sbi->s_li_wait_mult);
1164
1165         ext4_show_quota_options(seq, sb);
1166
1167         return 0;
1168 }
1169
1170 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1171                                         u64 ino, u32 generation)
1172 {
1173         struct inode *inode;
1174
1175         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1176                 return ERR_PTR(-ESTALE);
1177         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1178                 return ERR_PTR(-ESTALE);
1179
1180         /* iget isn't really right if the inode is currently unallocated!!
1181          *
1182          * ext4_read_inode will return a bad_inode if the inode had been
1183          * deleted, so we should be safe.
1184          *
1185          * Currently we don't know the generation for parent directory, so
1186          * a generation of 0 means "accept any"
1187          */
1188         inode = ext4_iget(sb, ino);
1189         if (IS_ERR(inode))
1190                 return ERR_CAST(inode);
1191         if (generation && inode->i_generation != generation) {
1192                 iput(inode);
1193                 return ERR_PTR(-ESTALE);
1194         }
1195
1196         return inode;
1197 }
1198
1199 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1200                                         int fh_len, int fh_type)
1201 {
1202         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1203                                     ext4_nfs_get_inode);
1204 }
1205
1206 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1207                                         int fh_len, int fh_type)
1208 {
1209         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1210                                     ext4_nfs_get_inode);
1211 }
1212
1213 /*
1214  * Try to release metadata pages (indirect blocks, directories) which are
1215  * mapped via the block device.  Since these pages could have journal heads
1216  * which would prevent try_to_free_buffers() from freeing them, we must use
1217  * jbd2 layer's try_to_free_buffers() function to release them.
1218  */
1219 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1220                                  gfp_t wait)
1221 {
1222         journal_t *journal = EXT4_SB(sb)->s_journal;
1223
1224         WARN_ON(PageChecked(page));
1225         if (!page_has_buffers(page))
1226                 return 0;
1227         if (journal)
1228                 return jbd2_journal_try_to_free_buffers(journal, page,
1229                                                         wait & ~__GFP_WAIT);
1230         return try_to_free_buffers(page);
1231 }
1232
1233 #ifdef CONFIG_QUOTA
1234 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1235 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1236
1237 static int ext4_write_dquot(struct dquot *dquot);
1238 static int ext4_acquire_dquot(struct dquot *dquot);
1239 static int ext4_release_dquot(struct dquot *dquot);
1240 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1241 static int ext4_write_info(struct super_block *sb, int type);
1242 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1243                          struct path *path);
1244 static int ext4_quota_off(struct super_block *sb, int type);
1245 static int ext4_quota_on_mount(struct super_block *sb, int type);
1246 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1247                                size_t len, loff_t off);
1248 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1249                                 const char *data, size_t len, loff_t off);
1250
1251 static const struct dquot_operations ext4_quota_operations = {
1252         .get_reserved_space = ext4_get_reserved_space,
1253         .write_dquot    = ext4_write_dquot,
1254         .acquire_dquot  = ext4_acquire_dquot,
1255         .release_dquot  = ext4_release_dquot,
1256         .mark_dirty     = ext4_mark_dquot_dirty,
1257         .write_info     = ext4_write_info,
1258         .alloc_dquot    = dquot_alloc,
1259         .destroy_dquot  = dquot_destroy,
1260 };
1261
1262 static const struct quotactl_ops ext4_qctl_operations = {
1263         .quota_on       = ext4_quota_on,
1264         .quota_off      = ext4_quota_off,
1265         .quota_sync     = dquot_quota_sync,
1266         .get_info       = dquot_get_dqinfo,
1267         .set_info       = dquot_set_dqinfo,
1268         .get_dqblk      = dquot_get_dqblk,
1269         .set_dqblk      = dquot_set_dqblk
1270 };
1271 #endif
1272
1273 static const struct super_operations ext4_sops = {
1274         .alloc_inode    = ext4_alloc_inode,
1275         .destroy_inode  = ext4_destroy_inode,
1276         .write_inode    = ext4_write_inode,
1277         .dirty_inode    = ext4_dirty_inode,
1278         .drop_inode     = ext4_drop_inode,
1279         .evict_inode    = ext4_evict_inode,
1280         .put_super      = ext4_put_super,
1281         .sync_fs        = ext4_sync_fs,
1282         .freeze_fs      = ext4_freeze,
1283         .unfreeze_fs    = ext4_unfreeze,
1284         .statfs         = ext4_statfs,
1285         .remount_fs     = ext4_remount,
1286         .show_options   = ext4_show_options,
1287 #ifdef CONFIG_QUOTA
1288         .quota_read     = ext4_quota_read,
1289         .quota_write    = ext4_quota_write,
1290 #endif
1291         .bdev_try_to_free_page = bdev_try_to_free_page,
1292 };
1293
1294 static const struct super_operations ext4_nojournal_sops = {
1295         .alloc_inode    = ext4_alloc_inode,
1296         .destroy_inode  = ext4_destroy_inode,
1297         .write_inode    = ext4_write_inode,
1298         .dirty_inode    = ext4_dirty_inode,
1299         .drop_inode     = ext4_drop_inode,
1300         .evict_inode    = ext4_evict_inode,
1301         .write_super    = ext4_write_super,
1302         .put_super      = ext4_put_super,
1303         .statfs         = ext4_statfs,
1304         .remount_fs     = ext4_remount,
1305         .show_options   = ext4_show_options,
1306 #ifdef CONFIG_QUOTA
1307         .quota_read     = ext4_quota_read,
1308         .quota_write    = ext4_quota_write,
1309 #endif
1310         .bdev_try_to_free_page = bdev_try_to_free_page,
1311 };
1312
1313 static const struct export_operations ext4_export_ops = {
1314         .fh_to_dentry = ext4_fh_to_dentry,
1315         .fh_to_parent = ext4_fh_to_parent,
1316         .get_parent = ext4_get_parent,
1317 };
1318
1319 enum {
1320         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1321         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1322         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1323         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1324         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1325         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1326         Opt_journal_update, Opt_journal_dev,
1327         Opt_journal_checksum, Opt_journal_async_commit,
1328         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1329         Opt_data_err_abort, Opt_data_err_ignore,
1330         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1331         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1332         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1333         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1334         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1335         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1336         Opt_inode_readahead_blks, Opt_journal_ioprio,
1337         Opt_dioread_nolock, Opt_dioread_lock,
1338         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1339 };
1340
1341 static const match_table_t tokens = {
1342         {Opt_bsd_df, "bsddf"},
1343         {Opt_minix_df, "minixdf"},
1344         {Opt_grpid, "grpid"},
1345         {Opt_grpid, "bsdgroups"},
1346         {Opt_nogrpid, "nogrpid"},
1347         {Opt_nogrpid, "sysvgroups"},
1348         {Opt_resgid, "resgid=%u"},
1349         {Opt_resuid, "resuid=%u"},
1350         {Opt_sb, "sb=%u"},
1351         {Opt_err_cont, "errors=continue"},
1352         {Opt_err_panic, "errors=panic"},
1353         {Opt_err_ro, "errors=remount-ro"},
1354         {Opt_nouid32, "nouid32"},
1355         {Opt_debug, "debug"},
1356         {Opt_oldalloc, "oldalloc"},
1357         {Opt_orlov, "orlov"},
1358         {Opt_user_xattr, "user_xattr"},
1359         {Opt_nouser_xattr, "nouser_xattr"},
1360         {Opt_acl, "acl"},
1361         {Opt_noacl, "noacl"},
1362         {Opt_noload, "noload"},
1363         {Opt_noload, "norecovery"},
1364         {Opt_nobh, "nobh"},
1365         {Opt_bh, "bh"},
1366         {Opt_commit, "commit=%u"},
1367         {Opt_min_batch_time, "min_batch_time=%u"},
1368         {Opt_max_batch_time, "max_batch_time=%u"},
1369         {Opt_journal_update, "journal=update"},
1370         {Opt_journal_dev, "journal_dev=%u"},
1371         {Opt_journal_checksum, "journal_checksum"},
1372         {Opt_journal_async_commit, "journal_async_commit"},
1373         {Opt_abort, "abort"},
1374         {Opt_data_journal, "data=journal"},
1375         {Opt_data_ordered, "data=ordered"},
1376         {Opt_data_writeback, "data=writeback"},
1377         {Opt_data_err_abort, "data_err=abort"},
1378         {Opt_data_err_ignore, "data_err=ignore"},
1379         {Opt_offusrjquota, "usrjquota="},
1380         {Opt_usrjquota, "usrjquota=%s"},
1381         {Opt_offgrpjquota, "grpjquota="},
1382         {Opt_grpjquota, "grpjquota=%s"},
1383         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1384         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1385         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1386         {Opt_grpquota, "grpquota"},
1387         {Opt_noquota, "noquota"},
1388         {Opt_quota, "quota"},
1389         {Opt_usrquota, "usrquota"},
1390         {Opt_barrier, "barrier=%u"},
1391         {Opt_barrier, "barrier"},
1392         {Opt_nobarrier, "nobarrier"},
1393         {Opt_i_version, "i_version"},
1394         {Opt_stripe, "stripe=%u"},
1395         {Opt_resize, "resize"},
1396         {Opt_delalloc, "delalloc"},
1397         {Opt_nodelalloc, "nodelalloc"},
1398         {Opt_mblk_io_submit, "mblk_io_submit"},
1399         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1400         {Opt_block_validity, "block_validity"},
1401         {Opt_noblock_validity, "noblock_validity"},
1402         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1403         {Opt_journal_ioprio, "journal_ioprio=%u"},
1404         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1405         {Opt_auto_da_alloc, "auto_da_alloc"},
1406         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1407         {Opt_dioread_nolock, "dioread_nolock"},
1408         {Opt_dioread_lock, "dioread_lock"},
1409         {Opt_discard, "discard"},
1410         {Opt_nodiscard, "nodiscard"},
1411         {Opt_init_itable, "init_itable=%u"},
1412         {Opt_init_itable, "init_itable"},
1413         {Opt_noinit_itable, "noinit_itable"},
1414         {Opt_err, NULL},
1415 };
1416
1417 static ext4_fsblk_t get_sb_block(void **data)
1418 {
1419         ext4_fsblk_t    sb_block;
1420         char            *options = (char *) *data;
1421
1422         if (!options || strncmp(options, "sb=", 3) != 0)
1423                 return 1;       /* Default location */
1424
1425         options += 3;
1426         /* TODO: use simple_strtoll with >32bit ext4 */
1427         sb_block = simple_strtoul(options, &options, 0);
1428         if (*options && *options != ',') {
1429                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1430                        (char *) *data);
1431                 return 1;
1432         }
1433         if (*options == ',')
1434                 options++;
1435         *data = (void *) options;
1436
1437         return sb_block;
1438 }
1439
1440 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1441 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1442         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1443
1444 #ifdef CONFIG_QUOTA
1445 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1446 {
1447         struct ext4_sb_info *sbi = EXT4_SB(sb);
1448         char *qname;
1449
1450         if (sb_any_quota_loaded(sb) &&
1451                 !sbi->s_qf_names[qtype]) {
1452                 ext4_msg(sb, KERN_ERR,
1453                         "Cannot change journaled "
1454                         "quota options when quota turned on");
1455                 return 0;
1456         }
1457         qname = match_strdup(args);
1458         if (!qname) {
1459                 ext4_msg(sb, KERN_ERR,
1460                         "Not enough memory for storing quotafile name");
1461                 return 0;
1462         }
1463         if (sbi->s_qf_names[qtype] &&
1464                 strcmp(sbi->s_qf_names[qtype], qname)) {
1465                 ext4_msg(sb, KERN_ERR,
1466                         "%s quota file already specified", QTYPE2NAME(qtype));
1467                 kfree(qname);
1468                 return 0;
1469         }
1470         sbi->s_qf_names[qtype] = qname;
1471         if (strchr(sbi->s_qf_names[qtype], '/')) {
1472                 ext4_msg(sb, KERN_ERR,
1473                         "quotafile must be on filesystem root");
1474                 kfree(sbi->s_qf_names[qtype]);
1475                 sbi->s_qf_names[qtype] = NULL;
1476                 return 0;
1477         }
1478         set_opt(sb, QUOTA);
1479         return 1;
1480 }
1481
1482 static int clear_qf_name(struct super_block *sb, int qtype)
1483 {
1484
1485         struct ext4_sb_info *sbi = EXT4_SB(sb);
1486
1487         if (sb_any_quota_loaded(sb) &&
1488                 sbi->s_qf_names[qtype]) {
1489                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1490                         " when quota turned on");
1491                 return 0;
1492         }
1493         /*
1494          * The space will be released later when all options are confirmed
1495          * to be correct
1496          */
1497         sbi->s_qf_names[qtype] = NULL;
1498         return 1;
1499 }
1500 #endif
1501
1502 static int parse_options(char *options, struct super_block *sb,
1503                          unsigned long *journal_devnum,
1504                          unsigned int *journal_ioprio,
1505                          ext4_fsblk_t *n_blocks_count, int is_remount)
1506 {
1507         struct ext4_sb_info *sbi = EXT4_SB(sb);
1508         char *p;
1509         substring_t args[MAX_OPT_ARGS];
1510         int data_opt = 0;
1511         int option;
1512 #ifdef CONFIG_QUOTA
1513         int qfmt;
1514 #endif
1515
1516         if (!options)
1517                 return 1;
1518
1519         while ((p = strsep(&options, ",")) != NULL) {
1520                 int token;
1521                 if (!*p)
1522                         continue;
1523
1524                 /*
1525                  * Initialize args struct so we know whether arg was
1526                  * found; some options take optional arguments.
1527                  */
1528                 args[0].to = args[0].from = NULL;
1529                 token = match_token(p, tokens, args);
1530                 switch (token) {
1531                 case Opt_bsd_df:
1532                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1533                         clear_opt(sb, MINIX_DF);
1534                         break;
1535                 case Opt_minix_df:
1536                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1537                         set_opt(sb, MINIX_DF);
1538
1539                         break;
1540                 case Opt_grpid:
1541                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1542                         set_opt(sb, GRPID);
1543
1544                         break;
1545                 case Opt_nogrpid:
1546                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1547                         clear_opt(sb, GRPID);
1548
1549                         break;
1550                 case Opt_resuid:
1551                         if (match_int(&args[0], &option))
1552                                 return 0;
1553                         sbi->s_resuid = option;
1554                         break;
1555                 case Opt_resgid:
1556                         if (match_int(&args[0], &option))
1557                                 return 0;
1558                         sbi->s_resgid = option;
1559                         break;
1560                 case Opt_sb:
1561                         /* handled by get_sb_block() instead of here */
1562                         /* *sb_block = match_int(&args[0]); */
1563                         break;
1564                 case Opt_err_panic:
1565                         clear_opt(sb, ERRORS_CONT);
1566                         clear_opt(sb, ERRORS_RO);
1567                         set_opt(sb, ERRORS_PANIC);
1568                         break;
1569                 case Opt_err_ro:
1570                         clear_opt(sb, ERRORS_CONT);
1571                         clear_opt(sb, ERRORS_PANIC);
1572                         set_opt(sb, ERRORS_RO);
1573                         break;
1574                 case Opt_err_cont:
1575                         clear_opt(sb, ERRORS_RO);
1576                         clear_opt(sb, ERRORS_PANIC);
1577                         set_opt(sb, ERRORS_CONT);
1578                         break;
1579                 case Opt_nouid32:
1580                         set_opt(sb, NO_UID32);
1581                         break;
1582                 case Opt_debug:
1583                         set_opt(sb, DEBUG);
1584                         break;
1585                 case Opt_oldalloc:
1586                         ext4_msg(sb, KERN_WARNING,
1587                                  "Ignoring deprecated oldalloc option");
1588                         break;
1589                 case Opt_orlov:
1590                         ext4_msg(sb, KERN_WARNING,
1591                                  "Ignoring deprecated orlov option");
1592                         break;
1593 #ifdef CONFIG_EXT4_FS_XATTR
1594                 case Opt_user_xattr:
1595                         set_opt(sb, XATTR_USER);
1596                         break;
1597                 case Opt_nouser_xattr:
1598                         clear_opt(sb, XATTR_USER);
1599                         break;
1600 #else
1601                 case Opt_user_xattr:
1602                 case Opt_nouser_xattr:
1603                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1604                         break;
1605 #endif
1606 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1607                 case Opt_acl:
1608                         set_opt(sb, POSIX_ACL);
1609                         break;
1610                 case Opt_noacl:
1611                         clear_opt(sb, POSIX_ACL);
1612                         break;
1613 #else
1614                 case Opt_acl:
1615                 case Opt_noacl:
1616                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1617                         break;
1618 #endif
1619                 case Opt_journal_update:
1620                         /* @@@ FIXME */
1621                         /* Eventually we will want to be able to create
1622                            a journal file here.  For now, only allow the
1623                            user to specify an existing inode to be the
1624                            journal file. */
1625                         if (is_remount) {
1626                                 ext4_msg(sb, KERN_ERR,
1627                                          "Cannot specify journal on remount");
1628                                 return 0;
1629                         }
1630                         set_opt(sb, UPDATE_JOURNAL);
1631                         break;
1632                 case Opt_journal_dev:
1633                         if (is_remount) {
1634                                 ext4_msg(sb, KERN_ERR,
1635                                         "Cannot specify journal on remount");
1636                                 return 0;
1637                         }
1638                         if (match_int(&args[0], &option))
1639                                 return 0;
1640                         *journal_devnum = option;
1641                         break;
1642                 case Opt_journal_checksum:
1643                         set_opt(sb, JOURNAL_CHECKSUM);
1644                         break;
1645                 case Opt_journal_async_commit:
1646                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1647                         set_opt(sb, JOURNAL_CHECKSUM);
1648                         break;
1649                 case Opt_noload:
1650                         set_opt(sb, NOLOAD);
1651                         break;
1652                 case Opt_commit:
1653                         if (match_int(&args[0], &option))
1654                                 return 0;
1655                         if (option < 0)
1656                                 return 0;
1657                         if (option == 0)
1658                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1659                         sbi->s_commit_interval = HZ * option;
1660                         break;
1661                 case Opt_max_batch_time:
1662                         if (match_int(&args[0], &option))
1663                                 return 0;
1664                         if (option < 0)
1665                                 return 0;
1666                         if (option == 0)
1667                                 option = EXT4_DEF_MAX_BATCH_TIME;
1668                         sbi->s_max_batch_time = option;
1669                         break;
1670                 case Opt_min_batch_time:
1671                         if (match_int(&args[0], &option))
1672                                 return 0;
1673                         if (option < 0)
1674                                 return 0;
1675                         sbi->s_min_batch_time = option;
1676                         break;
1677                 case Opt_data_journal:
1678                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1679                         goto datacheck;
1680                 case Opt_data_ordered:
1681                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1682                         goto datacheck;
1683                 case Opt_data_writeback:
1684                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1685                 datacheck:
1686                         if (is_remount) {
1687                                 if (!sbi->s_journal)
1688                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1689                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1690                                         ext4_msg(sb, KERN_ERR,
1691                                                 "Cannot change data mode on remount");
1692                                         return 0;
1693                                 }
1694                         } else {
1695                                 clear_opt(sb, DATA_FLAGS);
1696                                 sbi->s_mount_opt |= data_opt;
1697                         }
1698                         break;
1699                 case Opt_data_err_abort:
1700                         set_opt(sb, DATA_ERR_ABORT);
1701                         break;
1702                 case Opt_data_err_ignore:
1703                         clear_opt(sb, DATA_ERR_ABORT);
1704                         break;
1705 #ifdef CONFIG_QUOTA
1706                 case Opt_usrjquota:
1707                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1708                                 return 0;
1709                         break;
1710                 case Opt_grpjquota:
1711                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1712                                 return 0;
1713                         break;
1714                 case Opt_offusrjquota:
1715                         if (!clear_qf_name(sb, USRQUOTA))
1716                                 return 0;
1717                         break;
1718                 case Opt_offgrpjquota:
1719                         if (!clear_qf_name(sb, GRPQUOTA))
1720                                 return 0;
1721                         break;
1722
1723                 case Opt_jqfmt_vfsold:
1724                         qfmt = QFMT_VFS_OLD;
1725                         goto set_qf_format;
1726                 case Opt_jqfmt_vfsv0:
1727                         qfmt = QFMT_VFS_V0;
1728                         goto set_qf_format;
1729                 case Opt_jqfmt_vfsv1:
1730                         qfmt = QFMT_VFS_V1;
1731 set_qf_format:
1732                         if (sb_any_quota_loaded(sb) &&
1733                             sbi->s_jquota_fmt != qfmt) {
1734                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1735                                         "journaled quota options when "
1736                                         "quota turned on");
1737                                 return 0;
1738                         }
1739                         sbi->s_jquota_fmt = qfmt;
1740                         break;
1741                 case Opt_quota:
1742                 case Opt_usrquota:
1743                         set_opt(sb, QUOTA);
1744                         set_opt(sb, USRQUOTA);
1745                         break;
1746                 case Opt_grpquota:
1747                         set_opt(sb, QUOTA);
1748                         set_opt(sb, GRPQUOTA);
1749                         break;
1750                 case Opt_noquota:
1751                         if (sb_any_quota_loaded(sb)) {
1752                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1753                                         "options when quota turned on");
1754                                 return 0;
1755                         }
1756                         clear_opt(sb, QUOTA);
1757                         clear_opt(sb, USRQUOTA);
1758                         clear_opt(sb, GRPQUOTA);
1759                         break;
1760 #else
1761                 case Opt_quota:
1762                 case Opt_usrquota:
1763                 case Opt_grpquota:
1764                         ext4_msg(sb, KERN_ERR,
1765                                 "quota options not supported");
1766                         break;
1767                 case Opt_usrjquota:
1768                 case Opt_grpjquota:
1769                 case Opt_offusrjquota:
1770                 case Opt_offgrpjquota:
1771                 case Opt_jqfmt_vfsold:
1772                 case Opt_jqfmt_vfsv0:
1773                 case Opt_jqfmt_vfsv1:
1774                         ext4_msg(sb, KERN_ERR,
1775                                 "journaled quota options not supported");
1776                         break;
1777                 case Opt_noquota:
1778                         break;
1779 #endif
1780                 case Opt_abort:
1781                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1782                         break;
1783                 case Opt_nobarrier:
1784                         clear_opt(sb, BARRIER);
1785                         break;
1786                 case Opt_barrier:
1787                         if (args[0].from) {
1788                                 if (match_int(&args[0], &option))
1789                                         return 0;
1790                         } else
1791                                 option = 1;     /* No argument, default to 1 */
1792                         if (option)
1793                                 set_opt(sb, BARRIER);
1794                         else
1795                                 clear_opt(sb, BARRIER);
1796                         break;
1797                 case Opt_ignore:
1798                         break;
1799                 case Opt_resize:
1800                         if (!is_remount) {
1801                                 ext4_msg(sb, KERN_ERR,
1802                                         "resize option only available "
1803                                         "for remount");
1804                                 return 0;
1805                         }
1806                         if (match_int(&args[0], &option) != 0)
1807                                 return 0;
1808                         *n_blocks_count = option;
1809                         break;
1810                 case Opt_nobh:
1811                         ext4_msg(sb, KERN_WARNING,
1812                                  "Ignoring deprecated nobh option");
1813                         break;
1814                 case Opt_bh:
1815                         ext4_msg(sb, KERN_WARNING,
1816                                  "Ignoring deprecated bh option");
1817                         break;
1818                 case Opt_i_version:
1819                         set_opt(sb, I_VERSION);
1820                         sb->s_flags |= MS_I_VERSION;
1821                         break;
1822                 case Opt_nodelalloc:
1823                         clear_opt(sb, DELALLOC);
1824                         clear_opt2(sb, EXPLICIT_DELALLOC);
1825                         break;
1826                 case Opt_mblk_io_submit:
1827                         set_opt(sb, MBLK_IO_SUBMIT);
1828                         break;
1829                 case Opt_nomblk_io_submit:
1830                         clear_opt(sb, MBLK_IO_SUBMIT);
1831                         break;
1832                 case Opt_stripe:
1833                         if (match_int(&args[0], &option))
1834                                 return 0;
1835                         if (option < 0)
1836                                 return 0;
1837                         sbi->s_stripe = option;
1838                         break;
1839                 case Opt_delalloc:
1840                         set_opt(sb, DELALLOC);
1841                         set_opt2(sb, EXPLICIT_DELALLOC);
1842                         break;
1843                 case Opt_block_validity:
1844                         set_opt(sb, BLOCK_VALIDITY);
1845                         break;
1846                 case Opt_noblock_validity:
1847                         clear_opt(sb, BLOCK_VALIDITY);
1848                         break;
1849                 case Opt_inode_readahead_blks:
1850                         if (match_int(&args[0], &option))
1851                                 return 0;
1852                         if (option < 0 || option > (1 << 30))
1853                                 return 0;
1854                         if (option && !is_power_of_2(option)) {
1855                                 ext4_msg(sb, KERN_ERR,
1856                                          "EXT4-fs: inode_readahead_blks"
1857                                          " must be a power of 2");
1858                                 return 0;
1859                         }
1860                         sbi->s_inode_readahead_blks = option;
1861                         break;
1862                 case Opt_journal_ioprio:
1863                         if (match_int(&args[0], &option))
1864                                 return 0;
1865                         if (option < 0 || option > 7)
1866                                 break;
1867                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1868                                                             option);
1869                         break;
1870                 case Opt_noauto_da_alloc:
1871                         set_opt(sb, NO_AUTO_DA_ALLOC);
1872                         break;
1873                 case Opt_auto_da_alloc:
1874                         if (args[0].from) {
1875                                 if (match_int(&args[0], &option))
1876                                         return 0;
1877                         } else
1878                                 option = 1;     /* No argument, default to 1 */
1879                         if (option)
1880                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1881                         else
1882                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1883                         break;
1884                 case Opt_discard:
1885                         set_opt(sb, DISCARD);
1886                         break;
1887                 case Opt_nodiscard:
1888                         clear_opt(sb, DISCARD);
1889                         break;
1890                 case Opt_dioread_nolock:
1891                         set_opt(sb, DIOREAD_NOLOCK);
1892                         break;
1893                 case Opt_dioread_lock:
1894                         clear_opt(sb, DIOREAD_NOLOCK);
1895                         break;
1896                 case Opt_init_itable:
1897                         set_opt(sb, INIT_INODE_TABLE);
1898                         if (args[0].from) {
1899                                 if (match_int(&args[0], &option))
1900                                         return 0;
1901                         } else
1902                                 option = EXT4_DEF_LI_WAIT_MULT;
1903                         if (option < 0)
1904                                 return 0;
1905                         sbi->s_li_wait_mult = option;
1906                         break;
1907                 case Opt_noinit_itable:
1908                         clear_opt(sb, INIT_INODE_TABLE);
1909                         break;
1910                 default:
1911                         ext4_msg(sb, KERN_ERR,
1912                                "Unrecognized mount option \"%s\" "
1913                                "or missing value", p);
1914                         return 0;
1915                 }
1916         }
1917 #ifdef CONFIG_QUOTA
1918         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1919                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1920                         clear_opt(sb, USRQUOTA);
1921
1922                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1923                         clear_opt(sb, GRPQUOTA);
1924
1925                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1926                         ext4_msg(sb, KERN_ERR, "old and new quota "
1927                                         "format mixing");
1928                         return 0;
1929                 }
1930
1931                 if (!sbi->s_jquota_fmt) {
1932                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1933                                         "not specified");
1934                         return 0;
1935                 }
1936         } else {
1937                 if (sbi->s_jquota_fmt) {
1938                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1939                                         "specified with no journaling "
1940                                         "enabled");
1941                         return 0;
1942                 }
1943         }
1944 #endif
1945         if (test_opt(sb, DIOREAD_NOLOCK)) {
1946                 int blocksize =
1947                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1948
1949                 if (blocksize < PAGE_CACHE_SIZE) {
1950                         ext4_msg(sb, KERN_ERR, "can't mount with "
1951                                  "dioread_nolock if block size != PAGE_SIZE");
1952                         return 0;
1953                 }
1954         }
1955         return 1;
1956 }
1957
1958 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1959                             int read_only)
1960 {
1961         struct ext4_sb_info *sbi = EXT4_SB(sb);
1962         int res = 0;
1963
1964         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1965                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1966                          "forcing read-only mode");
1967                 res = MS_RDONLY;
1968         }
1969         if (read_only)
1970                 goto done;
1971         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1972                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1973                          "running e2fsck is recommended");
1974         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1975                 ext4_msg(sb, KERN_WARNING,
1976                          "warning: mounting fs with errors, "
1977                          "running e2fsck is recommended");
1978         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1979                  le16_to_cpu(es->s_mnt_count) >=
1980                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1981                 ext4_msg(sb, KERN_WARNING,
1982                          "warning: maximal mount count reached, "
1983                          "running e2fsck is recommended");
1984         else if (le32_to_cpu(es->s_checkinterval) &&
1985                 (le32_to_cpu(es->s_lastcheck) +
1986                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1987                 ext4_msg(sb, KERN_WARNING,
1988                          "warning: checktime reached, "
1989                          "running e2fsck is recommended");
1990         if (!sbi->s_journal)
1991                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1992         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1993                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1994         le16_add_cpu(&es->s_mnt_count, 1);
1995         es->s_mtime = cpu_to_le32(get_seconds());
1996         ext4_update_dynamic_rev(sb);
1997         if (sbi->s_journal)
1998                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1999
2000         ext4_commit_super(sb, 1);
2001 done:
2002         if (test_opt(sb, DEBUG))
2003                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2004                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2005                         sb->s_blocksize,
2006                         sbi->s_groups_count,
2007                         EXT4_BLOCKS_PER_GROUP(sb),
2008                         EXT4_INODES_PER_GROUP(sb),
2009                         sbi->s_mount_opt, sbi->s_mount_opt2);
2010
2011         cleancache_init_fs(sb);
2012         return res;
2013 }
2014
2015 static int ext4_fill_flex_info(struct super_block *sb)
2016 {
2017         struct ext4_sb_info *sbi = EXT4_SB(sb);
2018         struct ext4_group_desc *gdp = NULL;
2019         ext4_group_t flex_group_count;
2020         ext4_group_t flex_group;
2021         unsigned int groups_per_flex = 0;
2022         size_t size;
2023         int i;
2024
2025         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2026         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2027                 sbi->s_log_groups_per_flex = 0;
2028                 return 1;
2029         }
2030         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2031
2032         /* We allocate both existing and potentially added groups */
2033         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2034                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2035                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2036         size = flex_group_count * sizeof(struct flex_groups);
2037         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2038         if (sbi->s_flex_groups == NULL) {
2039                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2040                          flex_group_count);
2041                 goto failed;
2042         }
2043
2044         for (i = 0; i < sbi->s_groups_count; i++) {
2045                 gdp = ext4_get_group_desc(sb, i, NULL);
2046
2047                 flex_group = ext4_flex_group(sbi, i);
2048                 atomic_add(ext4_free_inodes_count(sb, gdp),
2049                            &sbi->s_flex_groups[flex_group].free_inodes);
2050                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2051                              &sbi->s_flex_groups[flex_group].free_clusters);
2052                 atomic_add(ext4_used_dirs_count(sb, gdp),
2053                            &sbi->s_flex_groups[flex_group].used_dirs);
2054         }
2055
2056         return 1;
2057 failed:
2058         return 0;
2059 }
2060
2061 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2062                             struct ext4_group_desc *gdp)
2063 {
2064         __u16 crc = 0;
2065
2066         if (sbi->s_es->s_feature_ro_compat &
2067             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2068                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2069                 __le32 le_group = cpu_to_le32(block_group);
2070
2071                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2072                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2073                 crc = crc16(crc, (__u8 *)gdp, offset);
2074                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2075                 /* for checksum of struct ext4_group_desc do the rest...*/
2076                 if ((sbi->s_es->s_feature_incompat &
2077                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2078                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2079                         crc = crc16(crc, (__u8 *)gdp + offset,
2080                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2081                                         offset);
2082         }
2083
2084         return cpu_to_le16(crc);
2085 }
2086
2087 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2088                                 struct ext4_group_desc *gdp)
2089 {
2090         if ((sbi->s_es->s_feature_ro_compat &
2091              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2092             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2093                 return 0;
2094
2095         return 1;
2096 }
2097
2098 /* Called at mount-time, super-block is locked */
2099 static int ext4_check_descriptors(struct super_block *sb,
2100                                   ext4_group_t *first_not_zeroed)
2101 {
2102         struct ext4_sb_info *sbi = EXT4_SB(sb);
2103         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2104         ext4_fsblk_t last_block;
2105         ext4_fsblk_t block_bitmap;
2106         ext4_fsblk_t inode_bitmap;
2107         ext4_fsblk_t inode_table;
2108         int flexbg_flag = 0;
2109         ext4_group_t i, grp = sbi->s_groups_count;
2110
2111         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2112                 flexbg_flag = 1;
2113
2114         ext4_debug("Checking group descriptors");
2115
2116         for (i = 0; i < sbi->s_groups_count; i++) {
2117                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2118
2119                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2120                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2121                 else
2122                         last_block = first_block +
2123                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2124
2125                 if ((grp == sbi->s_groups_count) &&
2126                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2127                         grp = i;
2128
2129                 block_bitmap = ext4_block_bitmap(sb, gdp);
2130                 if (block_bitmap < first_block || block_bitmap > last_block) {
2131                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2132                                "Block bitmap for group %u not in group "
2133                                "(block %llu)!", i, block_bitmap);
2134                         return 0;
2135                 }
2136                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2137                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2138                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2139                                "Inode bitmap for group %u not in group "
2140                                "(block %llu)!", i, inode_bitmap);
2141                         return 0;
2142                 }
2143                 inode_table = ext4_inode_table(sb, gdp);
2144                 if (inode_table < first_block ||
2145                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                "Inode table for group %u not in group "
2148                                "(block %llu)!", i, inode_table);
2149                         return 0;
2150                 }
2151                 ext4_lock_group(sb, i);
2152                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2153                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2154                                  "Checksum for group %u failed (%u!=%u)",
2155                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2156                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2157                         if (!(sb->s_flags & MS_RDONLY)) {
2158                                 ext4_unlock_group(sb, i);
2159                                 return 0;
2160                         }
2161                 }
2162                 ext4_unlock_group(sb, i);
2163                 if (!flexbg_flag)
2164                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2165         }
2166         if (NULL != first_not_zeroed)
2167                 *first_not_zeroed = grp;
2168
2169         ext4_free_blocks_count_set(sbi->s_es,
2170                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2171         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2172         return 1;
2173 }
2174
2175 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2176  * the superblock) which were deleted from all directories, but held open by
2177  * a process at the time of a crash.  We walk the list and try to delete these
2178  * inodes at recovery time (only with a read-write filesystem).
2179  *
2180  * In order to keep the orphan inode chain consistent during traversal (in
2181  * case of crash during recovery), we link each inode into the superblock
2182  * orphan list_head and handle it the same way as an inode deletion during
2183  * normal operation (which journals the operations for us).
2184  *
2185  * We only do an iget() and an iput() on each inode, which is very safe if we
2186  * accidentally point at an in-use or already deleted inode.  The worst that
2187  * can happen in this case is that we get a "bit already cleared" message from
2188  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2189  * e2fsck was run on this filesystem, and it must have already done the orphan
2190  * inode cleanup for us, so we can safely abort without any further action.
2191  */
2192 static void ext4_orphan_cleanup(struct super_block *sb,
2193                                 struct ext4_super_block *es)
2194 {
2195         unsigned int s_flags = sb->s_flags;
2196         int nr_orphans = 0, nr_truncates = 0;
2197 #ifdef CONFIG_QUOTA
2198         int i;
2199 #endif
2200         if (!es->s_last_orphan) {
2201                 jbd_debug(4, "no orphan inodes to clean up\n");
2202                 return;
2203         }
2204
2205         if (bdev_read_only(sb->s_bdev)) {
2206                 ext4_msg(sb, KERN_ERR, "write access "
2207                         "unavailable, skipping orphan cleanup");
2208                 return;
2209         }
2210
2211         /* Check if feature set would not allow a r/w mount */
2212         if (!ext4_feature_set_ok(sb, 0)) {
2213                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2214                          "unknown ROCOMPAT features");
2215                 return;
2216         }
2217
2218         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2219                 if (es->s_last_orphan)
2220                         jbd_debug(1, "Errors on filesystem, "
2221                                   "clearing orphan list.\n");
2222                 es->s_last_orphan = 0;
2223                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2224                 return;
2225         }
2226
2227         if (s_flags & MS_RDONLY) {
2228                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2229                 sb->s_flags &= ~MS_RDONLY;
2230         }
2231 #ifdef CONFIG_QUOTA
2232         /* Needed for iput() to work correctly and not trash data */
2233         sb->s_flags |= MS_ACTIVE;
2234         /* Turn on quotas so that they are updated correctly */
2235         for (i = 0; i < MAXQUOTAS; i++) {
2236                 if (EXT4_SB(sb)->s_qf_names[i]) {
2237                         int ret = ext4_quota_on_mount(sb, i);
2238                         if (ret < 0)
2239                                 ext4_msg(sb, KERN_ERR,
2240                                         "Cannot turn on journaled "
2241                                         "quota: error %d", ret);
2242                 }
2243         }
2244 #endif
2245
2246         while (es->s_last_orphan) {
2247                 struct inode *inode;
2248
2249                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2250                 if (IS_ERR(inode)) {
2251                         es->s_last_orphan = 0;
2252                         break;
2253                 }
2254
2255                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2256                 dquot_initialize(inode);
2257                 if (inode->i_nlink) {
2258                         ext4_msg(sb, KERN_DEBUG,
2259                                 "%s: truncating inode %lu to %lld bytes",
2260                                 __func__, inode->i_ino, inode->i_size);
2261                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2262                                   inode->i_ino, inode->i_size);
2263                         mutex_lock(&inode->i_mutex);
2264                         ext4_truncate(inode);
2265                         mutex_unlock(&inode->i_mutex);
2266                         nr_truncates++;
2267                 } else {
2268                         ext4_msg(sb, KERN_DEBUG,
2269                                 "%s: deleting unreferenced inode %lu",
2270                                 __func__, inode->i_ino);
2271                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2272                                   inode->i_ino);
2273                         nr_orphans++;
2274                 }
2275                 iput(inode);  /* The delete magic happens here! */
2276         }
2277
2278 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2279
2280         if (nr_orphans)
2281                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2282                        PLURAL(nr_orphans));
2283         if (nr_truncates)
2284                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2285                        PLURAL(nr_truncates));
2286 #ifdef CONFIG_QUOTA
2287         /* Turn quotas off */
2288         for (i = 0; i < MAXQUOTAS; i++) {
2289                 if (sb_dqopt(sb)->files[i])
2290                         dquot_quota_off(sb, i);
2291         }
2292 #endif
2293         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2294 }
2295
2296 /*
2297  * Maximal extent format file size.
2298  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2299  * extent format containers, within a sector_t, and within i_blocks
2300  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2301  * so that won't be a limiting factor.
2302  *
2303  * However there is other limiting factor. We do store extents in the form
2304  * of starting block and length, hence the resulting length of the extent
2305  * covering maximum file size must fit into on-disk format containers as
2306  * well. Given that length is always by 1 unit bigger than max unit (because
2307  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2308  *
2309  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2310  */
2311 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2312 {
2313         loff_t res;
2314         loff_t upper_limit = MAX_LFS_FILESIZE;
2315
2316         /* small i_blocks in vfs inode? */
2317         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2318                 /*
2319                  * CONFIG_LBDAF is not enabled implies the inode
2320                  * i_block represent total blocks in 512 bytes
2321                  * 32 == size of vfs inode i_blocks * 8
2322                  */
2323                 upper_limit = (1LL << 32) - 1;
2324
2325                 /* total blocks in file system block size */
2326                 upper_limit >>= (blkbits - 9);
2327                 upper_limit <<= blkbits;
2328         }
2329
2330         /*
2331          * 32-bit extent-start container, ee_block. We lower the maxbytes
2332          * by one fs block, so ee_len can cover the extent of maximum file
2333          * size
2334          */
2335         res = (1LL << 32) - 1;
2336         res <<= blkbits;
2337
2338         /* Sanity check against vm- & vfs- imposed limits */
2339         if (res > upper_limit)
2340                 res = upper_limit;
2341
2342         return res;
2343 }
2344
2345 /*
2346  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2347  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2348  * We need to be 1 filesystem block less than the 2^48 sector limit.
2349  */
2350 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2351 {
2352         loff_t res = EXT4_NDIR_BLOCKS;
2353         int meta_blocks;
2354         loff_t upper_limit;
2355         /* This is calculated to be the largest file size for a dense, block
2356          * mapped file such that the file's total number of 512-byte sectors,
2357          * including data and all indirect blocks, does not exceed (2^48 - 1).
2358          *
2359          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2360          * number of 512-byte sectors of the file.
2361          */
2362
2363         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2364                 /*
2365                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2366                  * the inode i_block field represents total file blocks in
2367                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2368                  */
2369                 upper_limit = (1LL << 32) - 1;
2370
2371                 /* total blocks in file system block size */
2372                 upper_limit >>= (bits - 9);
2373
2374         } else {
2375                 /*
2376                  * We use 48 bit ext4_inode i_blocks
2377                  * With EXT4_HUGE_FILE_FL set the i_blocks
2378                  * represent total number of blocks in
2379                  * file system block size
2380                  */
2381                 upper_limit = (1LL << 48) - 1;
2382
2383         }
2384
2385         /* indirect blocks */
2386         meta_blocks = 1;
2387         /* double indirect blocks */
2388         meta_blocks += 1 + (1LL << (bits-2));
2389         /* tripple indirect blocks */
2390         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2391
2392         upper_limit -= meta_blocks;
2393         upper_limit <<= bits;
2394
2395         res += 1LL << (bits-2);
2396         res += 1LL << (2*(bits-2));
2397         res += 1LL << (3*(bits-2));
2398         res <<= bits;
2399         if (res > upper_limit)
2400                 res = upper_limit;
2401
2402         if (res > MAX_LFS_FILESIZE)
2403                 res = MAX_LFS_FILESIZE;
2404
2405         return res;
2406 }
2407
2408 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2409                                    ext4_fsblk_t logical_sb_block, int nr)
2410 {
2411         struct ext4_sb_info *sbi = EXT4_SB(sb);
2412         ext4_group_t bg, first_meta_bg;
2413         int has_super = 0;
2414
2415         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2416
2417         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2418             nr < first_meta_bg)
2419                 return logical_sb_block + nr + 1;
2420         bg = sbi->s_desc_per_block * nr;
2421         if (ext4_bg_has_super(sb, bg))
2422                 has_super = 1;
2423
2424         return (has_super + ext4_group_first_block_no(sb, bg));
2425 }
2426
2427 /**
2428  * ext4_get_stripe_size: Get the stripe size.
2429  * @sbi: In memory super block info
2430  *
2431  * If we have specified it via mount option, then
2432  * use the mount option value. If the value specified at mount time is
2433  * greater than the blocks per group use the super block value.
2434  * If the super block value is greater than blocks per group return 0.
2435  * Allocator needs it be less than blocks per group.
2436  *
2437  */
2438 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2439 {
2440         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2441         unsigned long stripe_width =
2442                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2443         int ret;
2444
2445         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2446                 ret = sbi->s_stripe;
2447         else if (stripe_width <= sbi->s_blocks_per_group)
2448                 ret = stripe_width;
2449         else if (stride <= sbi->s_blocks_per_group)
2450                 ret = stride;
2451         else
2452                 ret = 0;
2453
2454         /*
2455          * If the stripe width is 1, this makes no sense and
2456          * we set it to 0 to turn off stripe handling code.
2457          */
2458         if (ret <= 1)
2459                 ret = 0;
2460
2461         return ret;
2462 }
2463
2464 /* sysfs supprt */
2465
2466 struct ext4_attr {
2467         struct attribute attr;
2468         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2469         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2470                          const char *, size_t);
2471         int offset;
2472 };
2473
2474 static int parse_strtoul(const char *buf,
2475                 unsigned long max, unsigned long *value)
2476 {
2477         char *endp;
2478
2479         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2480         endp = skip_spaces(endp);
2481         if (*endp || *value > max)
2482                 return -EINVAL;
2483
2484         return 0;
2485 }
2486
2487 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2488                                               struct ext4_sb_info *sbi,
2489                                               char *buf)
2490 {
2491         return snprintf(buf, PAGE_SIZE, "%llu\n",
2492                 (s64) EXT4_C2B(sbi,
2493                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2494 }
2495
2496 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2497                                          struct ext4_sb_info *sbi, char *buf)
2498 {
2499         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2500
2501         if (!sb->s_bdev->bd_part)
2502                 return snprintf(buf, PAGE_SIZE, "0\n");
2503         return snprintf(buf, PAGE_SIZE, "%lu\n",
2504                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2505                          sbi->s_sectors_written_start) >> 1);
2506 }
2507
2508 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2509                                           struct ext4_sb_info *sbi, char *buf)
2510 {
2511         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2512
2513         if (!sb->s_bdev->bd_part)
2514                 return snprintf(buf, PAGE_SIZE, "0\n");
2515         return snprintf(buf, PAGE_SIZE, "%llu\n",
2516                         (unsigned long long)(sbi->s_kbytes_written +
2517                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2518                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2519 }
2520
2521 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2522                                           struct ext4_sb_info *sbi,
2523                                           const char *buf, size_t count)
2524 {
2525         unsigned long t;
2526
2527         if (parse_strtoul(buf, 0x40000000, &t))
2528                 return -EINVAL;
2529
2530         if (t && !is_power_of_2(t))
2531                 return -EINVAL;
2532
2533         sbi->s_inode_readahead_blks = t;
2534         return count;
2535 }
2536
2537 static ssize_t sbi_ui_show(struct ext4_attr *a,
2538                            struct ext4_sb_info *sbi, char *buf)
2539 {
2540         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2541
2542         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2543 }
2544
2545 static ssize_t sbi_ui_store(struct ext4_attr *a,
2546                             struct ext4_sb_info *sbi,
2547                             const char *buf, size_t count)
2548 {
2549         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2550         unsigned long t;
2551
2552         if (parse_strtoul(buf, 0xffffffff, &t))
2553                 return -EINVAL;
2554         *ui = t;
2555         return count;
2556 }
2557
2558 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2559 static struct ext4_attr ext4_attr_##_name = {                   \
2560         .attr = {.name = __stringify(_name), .mode = _mode },   \
2561         .show   = _show,                                        \
2562         .store  = _store,                                       \
2563         .offset = offsetof(struct ext4_sb_info, _elname),       \
2564 }
2565 #define EXT4_ATTR(name, mode, show, store) \
2566 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2567
2568 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2569 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2570 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2571 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2572         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2573 #define ATTR_LIST(name) &ext4_attr_##name.attr
2574
2575 EXT4_RO_ATTR(delayed_allocation_blocks);
2576 EXT4_RO_ATTR(session_write_kbytes);
2577 EXT4_RO_ATTR(lifetime_write_kbytes);
2578 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2579                  inode_readahead_blks_store, s_inode_readahead_blks);
2580 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2581 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2582 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2583 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2584 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2585 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2586 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2587 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2588
2589 static struct attribute *ext4_attrs[] = {
2590         ATTR_LIST(delayed_allocation_blocks),
2591         ATTR_LIST(session_write_kbytes),
2592         ATTR_LIST(lifetime_write_kbytes),
2593         ATTR_LIST(inode_readahead_blks),
2594         ATTR_LIST(inode_goal),
2595         ATTR_LIST(mb_stats),
2596         ATTR_LIST(mb_max_to_scan),
2597         ATTR_LIST(mb_min_to_scan),
2598         ATTR_LIST(mb_order2_req),
2599         ATTR_LIST(mb_stream_req),
2600         ATTR_LIST(mb_group_prealloc),
2601         ATTR_LIST(max_writeback_mb_bump),
2602         NULL,
2603 };
2604
2605 /* Features this copy of ext4 supports */
2606 EXT4_INFO_ATTR(lazy_itable_init);
2607 EXT4_INFO_ATTR(batched_discard);
2608
2609 static struct attribute *ext4_feat_attrs[] = {
2610         ATTR_LIST(lazy_itable_init),
2611         ATTR_LIST(batched_discard),
2612         NULL,
2613 };
2614
2615 static ssize_t ext4_attr_show(struct kobject *kobj,
2616                               struct attribute *attr, char *buf)
2617 {
2618         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2619                                                 s_kobj);
2620         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2621
2622         return a->show ? a->show(a, sbi, buf) : 0;
2623 }
2624
2625 static ssize_t ext4_attr_store(struct kobject *kobj,
2626                                struct attribute *attr,
2627                                const char *buf, size_t len)
2628 {
2629         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2630                                                 s_kobj);
2631         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2632
2633         return a->store ? a->store(a, sbi, buf, len) : 0;
2634 }
2635
2636 static void ext4_sb_release(struct kobject *kobj)
2637 {
2638         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2639                                                 s_kobj);
2640         complete(&sbi->s_kobj_unregister);
2641 }
2642
2643 static const struct sysfs_ops ext4_attr_ops = {
2644         .show   = ext4_attr_show,
2645         .store  = ext4_attr_store,
2646 };
2647
2648 static struct kobj_type ext4_ktype = {
2649         .default_attrs  = ext4_attrs,
2650         .sysfs_ops      = &ext4_attr_ops,
2651         .release        = ext4_sb_release,
2652 };
2653
2654 static void ext4_feat_release(struct kobject *kobj)
2655 {
2656         complete(&ext4_feat->f_kobj_unregister);
2657 }
2658
2659 static struct kobj_type ext4_feat_ktype = {
2660         .default_attrs  = ext4_feat_attrs,
2661         .sysfs_ops      = &ext4_attr_ops,
2662         .release        = ext4_feat_release,
2663 };
2664
2665 /*
2666  * Check whether this filesystem can be mounted based on
2667  * the features present and the RDONLY/RDWR mount requested.
2668  * Returns 1 if this filesystem can be mounted as requested,
2669  * 0 if it cannot be.
2670  */
2671 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2672 {
2673         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2674                 ext4_msg(sb, KERN_ERR,
2675                         "Couldn't mount because of "
2676                         "unsupported optional features (%x)",
2677                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2678                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2679                 return 0;
2680         }
2681
2682         if (readonly)
2683                 return 1;
2684
2685         /* Check that feature set is OK for a read-write mount */
2686         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2687                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2688                          "unsupported optional features (%x)",
2689                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2690                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2691                 return 0;
2692         }
2693         /*
2694          * Large file size enabled file system can only be mounted
2695          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2696          */
2697         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2698                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2699                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2700                                  "cannot be mounted RDWR without "
2701                                  "CONFIG_LBDAF");
2702                         return 0;
2703                 }
2704         }
2705         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2706             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2707                 ext4_msg(sb, KERN_ERR,
2708                          "Can't support bigalloc feature without "
2709                          "extents feature\n");
2710                 return 0;
2711         }
2712         return 1;
2713 }
2714
2715 /*
2716  * This function is called once a day if we have errors logged
2717  * on the file system
2718  */
2719 static void print_daily_error_info(unsigned long arg)
2720 {
2721         struct super_block *sb = (struct super_block *) arg;
2722         struct ext4_sb_info *sbi;
2723         struct ext4_super_block *es;
2724
2725         sbi = EXT4_SB(sb);
2726         es = sbi->s_es;
2727
2728         if (es->s_error_count)
2729                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2730                          le32_to_cpu(es->s_error_count));
2731         if (es->s_first_error_time) {
2732                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2733                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2734                        (int) sizeof(es->s_first_error_func),
2735                        es->s_first_error_func,
2736                        le32_to_cpu(es->s_first_error_line));
2737                 if (es->s_first_error_ino)
2738                         printk(": inode %u",
2739                                le32_to_cpu(es->s_first_error_ino));
2740                 if (es->s_first_error_block)
2741                         printk(": block %llu", (unsigned long long)
2742                                le64_to_cpu(es->s_first_error_block));
2743                 printk("\n");
2744         }
2745         if (es->s_last_error_time) {
2746                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2747                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2748                        (int) sizeof(es->s_last_error_func),
2749                        es->s_last_error_func,
2750                        le32_to_cpu(es->s_last_error_line));
2751                 if (es->s_last_error_ino)
2752                         printk(": inode %u",
2753                                le32_to_cpu(es->s_last_error_ino));
2754                 if (es->s_last_error_block)
2755                         printk(": block %llu", (unsigned long long)
2756                                le64_to_cpu(es->s_last_error_block));
2757                 printk("\n");
2758         }
2759         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2760 }
2761
2762 /* Find next suitable group and run ext4_init_inode_table */
2763 static int ext4_run_li_request(struct ext4_li_request *elr)
2764 {
2765         struct ext4_group_desc *gdp = NULL;
2766         ext4_group_t group, ngroups;
2767         struct super_block *sb;
2768         unsigned long timeout = 0;
2769         int ret = 0;
2770
2771         sb = elr->lr_super;
2772         ngroups = EXT4_SB(sb)->s_groups_count;
2773
2774         for (group = elr->lr_next_group; group < ngroups; group++) {
2775                 gdp = ext4_get_group_desc(sb, group, NULL);
2776                 if (!gdp) {
2777                         ret = 1;
2778                         break;
2779                 }
2780
2781                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2782                         break;
2783         }
2784
2785         if (group == ngroups)
2786                 ret = 1;
2787
2788         if (!ret) {
2789                 timeout = jiffies;
2790                 ret = ext4_init_inode_table(sb, group,
2791                                             elr->lr_timeout ? 0 : 1);
2792                 if (elr->lr_timeout == 0) {
2793                         timeout = (jiffies - timeout) *
2794                                   elr->lr_sbi->s_li_wait_mult;
2795                         elr->lr_timeout = timeout;
2796                 }
2797                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2798                 elr->lr_next_group = group + 1;
2799         }
2800
2801         return ret;
2802 }
2803
2804 /*
2805  * Remove lr_request from the list_request and free the
2806  * request structure. Should be called with li_list_mtx held
2807  */
2808 static void ext4_remove_li_request(struct ext4_li_request *elr)
2809 {
2810         struct ext4_sb_info *sbi;
2811
2812         if (!elr)
2813                 return;
2814
2815         sbi = elr->lr_sbi;
2816
2817         list_del(&elr->lr_request);
2818         sbi->s_li_request = NULL;
2819         kfree(elr);
2820 }
2821
2822 static void ext4_unregister_li_request(struct super_block *sb)
2823 {
2824         mutex_lock(&ext4_li_mtx);
2825         if (!ext4_li_info) {
2826                 mutex_unlock(&ext4_li_mtx);
2827                 return;
2828         }
2829
2830         mutex_lock(&ext4_li_info->li_list_mtx);
2831         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2832         mutex_unlock(&ext4_li_info->li_list_mtx);
2833         mutex_unlock(&ext4_li_mtx);
2834 }
2835
2836 static struct task_struct *ext4_lazyinit_task;
2837
2838 /*
2839  * This is the function where ext4lazyinit thread lives. It walks
2840  * through the request list searching for next scheduled filesystem.
2841  * When such a fs is found, run the lazy initialization request
2842  * (ext4_rn_li_request) and keep track of the time spend in this
2843  * function. Based on that time we compute next schedule time of
2844  * the request. When walking through the list is complete, compute
2845  * next waking time and put itself into sleep.
2846  */
2847 static int ext4_lazyinit_thread(void *arg)
2848 {
2849         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2850         struct list_head *pos, *n;
2851         struct ext4_li_request *elr;
2852         unsigned long next_wakeup, cur;
2853
2854         BUG_ON(NULL == eli);
2855
2856 cont_thread:
2857         while (true) {
2858                 next_wakeup = MAX_JIFFY_OFFSET;
2859
2860                 mutex_lock(&eli->li_list_mtx);
2861                 if (list_empty(&eli->li_request_list)) {
2862                         mutex_unlock(&eli->li_list_mtx);
2863                         goto exit_thread;
2864                 }
2865
2866                 list_for_each_safe(pos, n, &eli->li_request_list) {
2867                         elr = list_entry(pos, struct ext4_li_request,
2868                                          lr_request);
2869
2870                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2871                                 if (ext4_run_li_request(elr) != 0) {
2872                                         /* error, remove the lazy_init job */
2873                                         ext4_remove_li_request(elr);
2874                                         continue;
2875                                 }
2876                         }
2877
2878                         if (time_before(elr->lr_next_sched, next_wakeup))
2879                                 next_wakeup = elr->lr_next_sched;
2880                 }
2881                 mutex_unlock(&eli->li_list_mtx);
2882
2883                 if (freezing(current))
2884                         refrigerator();
2885
2886                 cur = jiffies;
2887                 if ((time_after_eq(cur, next_wakeup)) ||
2888                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2889                         cond_resched();
2890                         continue;
2891                 }
2892
2893                 schedule_timeout_interruptible(next_wakeup - cur);
2894
2895                 if (kthread_should_stop()) {
2896                         ext4_clear_request_list();
2897                         goto exit_thread;
2898                 }
2899         }
2900
2901 exit_thread:
2902         /*
2903          * It looks like the request list is empty, but we need
2904          * to check it under the li_list_mtx lock, to prevent any
2905          * additions into it, and of course we should lock ext4_li_mtx
2906          * to atomically free the list and ext4_li_info, because at
2907          * this point another ext4 filesystem could be registering
2908          * new one.
2909          */
2910         mutex_lock(&ext4_li_mtx);
2911         mutex_lock(&eli->li_list_mtx);
2912         if (!list_empty(&eli->li_request_list)) {
2913                 mutex_unlock(&eli->li_list_mtx);
2914                 mutex_unlock(&ext4_li_mtx);
2915                 goto cont_thread;
2916         }
2917         mutex_unlock(&eli->li_list_mtx);
2918         kfree(ext4_li_info);
2919         ext4_li_info = NULL;
2920         mutex_unlock(&ext4_li_mtx);
2921
2922         return 0;
2923 }
2924
2925 static void ext4_clear_request_list(void)
2926 {
2927         struct list_head *pos, *n;
2928         struct ext4_li_request *elr;
2929
2930         mutex_lock(&ext4_li_info->li_list_mtx);
2931         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2932                 elr = list_entry(pos, struct ext4_li_request,
2933                                  lr_request);
2934                 ext4_remove_li_request(elr);
2935         }
2936         mutex_unlock(&ext4_li_info->li_list_mtx);
2937 }
2938
2939 static int ext4_run_lazyinit_thread(void)
2940 {
2941         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2942                                          ext4_li_info, "ext4lazyinit");
2943         if (IS_ERR(ext4_lazyinit_task)) {
2944                 int err = PTR_ERR(ext4_lazyinit_task);
2945                 ext4_clear_request_list();
2946                 kfree(ext4_li_info);
2947                 ext4_li_info = NULL;
2948                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2949                                  "initialization thread\n",
2950                                  err);
2951                 return err;
2952         }
2953         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2954         return 0;
2955 }
2956
2957 /*
2958  * Check whether it make sense to run itable init. thread or not.
2959  * If there is at least one uninitialized inode table, return
2960  * corresponding group number, else the loop goes through all
2961  * groups and return total number of groups.
2962  */
2963 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2964 {
2965         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2966         struct ext4_group_desc *gdp = NULL;
2967
2968         for (group = 0; group < ngroups; group++) {
2969                 gdp = ext4_get_group_desc(sb, group, NULL);
2970                 if (!gdp)
2971                         continue;
2972
2973                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2974                         break;
2975         }
2976
2977         return group;
2978 }
2979
2980 static int ext4_li_info_new(void)
2981 {
2982         struct ext4_lazy_init *eli = NULL;
2983
2984         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2985         if (!eli)
2986                 return -ENOMEM;
2987
2988         INIT_LIST_HEAD(&eli->li_request_list);
2989         mutex_init(&eli->li_list_mtx);
2990
2991         eli->li_state |= EXT4_LAZYINIT_QUIT;
2992
2993         ext4_li_info = eli;
2994
2995         return 0;
2996 }
2997
2998 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2999                                             ext4_group_t start)
3000 {
3001         struct ext4_sb_info *sbi = EXT4_SB(sb);
3002         struct ext4_li_request *elr;
3003         unsigned long rnd;
3004
3005         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3006         if (!elr)
3007                 return NULL;
3008
3009         elr->lr_super = sb;
3010         elr->lr_sbi = sbi;
3011         elr->lr_next_group = start;
3012
3013         /*
3014          * Randomize first schedule time of the request to
3015          * spread the inode table initialization requests
3016          * better.
3017          */
3018         get_random_bytes(&rnd, sizeof(rnd));
3019         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3020                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3021
3022         return elr;
3023 }
3024
3025 static int ext4_register_li_request(struct super_block *sb,
3026                                     ext4_group_t first_not_zeroed)
3027 {
3028         struct ext4_sb_info *sbi = EXT4_SB(sb);
3029         struct ext4_li_request *elr;
3030         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3031         int ret = 0;
3032
3033         if (sbi->s_li_request != NULL) {
3034                 /*
3035                  * Reset timeout so it can be computed again, because
3036                  * s_li_wait_mult might have changed.
3037                  */
3038                 sbi->s_li_request->lr_timeout = 0;
3039                 return 0;
3040         }
3041
3042         if (first_not_zeroed == ngroups ||
3043             (sb->s_flags & MS_RDONLY) ||
3044             !test_opt(sb, INIT_INODE_TABLE))
3045                 return 0;
3046
3047         elr = ext4_li_request_new(sb, first_not_zeroed);
3048         if (!elr)
3049                 return -ENOMEM;
3050
3051         mutex_lock(&ext4_li_mtx);
3052
3053         if (NULL == ext4_li_info) {
3054                 ret = ext4_li_info_new();
3055                 if (ret)
3056                         goto out;
3057         }
3058
3059         mutex_lock(&ext4_li_info->li_list_mtx);
3060         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3061         mutex_unlock(&ext4_li_info->li_list_mtx);
3062
3063         sbi->s_li_request = elr;
3064         /*
3065          * set elr to NULL here since it has been inserted to
3066          * the request_list and the removal and free of it is
3067          * handled by ext4_clear_request_list from now on.
3068          */
3069         elr = NULL;
3070
3071         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3072                 ret = ext4_run_lazyinit_thread();
3073                 if (ret)
3074                         goto out;
3075         }
3076 out:
3077         mutex_unlock(&ext4_li_mtx);
3078         if (ret)
3079                 kfree(elr);
3080         return ret;
3081 }
3082
3083 /*
3084  * We do not need to lock anything since this is called on
3085  * module unload.
3086  */
3087 static void ext4_destroy_lazyinit_thread(void)
3088 {
3089         /*
3090          * If thread exited earlier
3091          * there's nothing to be done.
3092          */
3093         if (!ext4_li_info || !ext4_lazyinit_task)
3094                 return;
3095
3096         kthread_stop(ext4_lazyinit_task);
3097 }
3098
3099 /*
3100  * Note: calculating the overhead so we can be compatible with
3101  * historical BSD practice is quite difficult in the face of
3102  * clusters/bigalloc.  This is because multiple metadata blocks from
3103  * different block group can end up in the same allocation cluster.
3104  * Calculating the exact overhead in the face of clustered allocation
3105  * requires either O(all block bitmaps) in memory or O(number of block
3106  * groups**2) in time.  We will still calculate the superblock for
3107  * older file systems --- and if we come across with a bigalloc file
3108  * system with zero in s_overhead_clusters the estimate will be close to
3109  * correct especially for very large cluster sizes --- but for newer
3110  * file systems, it's better to calculate this figure once at mkfs
3111  * time, and store it in the superblock.  If the superblock value is
3112  * present (even for non-bigalloc file systems), we will use it.
3113  */
3114 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3115                           char *buf)
3116 {
3117         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3118         struct ext4_group_desc  *gdp;
3119         ext4_fsblk_t            first_block, last_block, b;
3120         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3121         int                     s, j, count = 0;
3122
3123         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3124                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3125                         sbi->s_itb_per_group + 2);
3126
3127         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3128                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3129         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3130         for (i = 0; i < ngroups; i++) {
3131                 gdp = ext4_get_group_desc(sb, i, NULL);
3132                 b = ext4_block_bitmap(sb, gdp);
3133                 if (b >= first_block && b <= last_block) {
3134                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3135                         count++;
3136                 }
3137                 b = ext4_inode_bitmap(sb, gdp);
3138                 if (b >= first_block && b <= last_block) {
3139                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3140                         count++;
3141                 }
3142                 b = ext4_inode_table(sb, gdp);
3143                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3144                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3145                                 int c = EXT4_B2C(sbi, b - first_block);
3146                                 ext4_set_bit(c, buf);
3147                                 count++;
3148                         }
3149                 if (i != grp)
3150                         continue;
3151                 s = 0;
3152                 if (ext4_bg_has_super(sb, grp)) {
3153                         ext4_set_bit(s++, buf);
3154                         count++;
3155                 }
3156                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3157                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3158                         count++;
3159                 }
3160         }
3161         if (!count)
3162                 return 0;
3163         return EXT4_CLUSTERS_PER_GROUP(sb) -
3164                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3165 }
3166
3167 /*
3168  * Compute the overhead and stash it in sbi->s_overhead
3169  */
3170 int ext4_calculate_overhead(struct super_block *sb)
3171 {
3172         struct ext4_sb_info *sbi = EXT4_SB(sb);
3173         struct ext4_super_block *es = sbi->s_es;
3174         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3175         ext4_fsblk_t overhead = 0;
3176         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3177
3178         memset(buf, 0, PAGE_SIZE);
3179         if (!buf)
3180                 return -ENOMEM;
3181
3182         /*
3183          * Compute the overhead (FS structures).  This is constant
3184          * for a given filesystem unless the number of block groups
3185          * changes so we cache the previous value until it does.
3186          */
3187
3188         /*
3189          * All of the blocks before first_data_block are overhead
3190          */
3191         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3192
3193         /*
3194          * Add the overhead found in each block group
3195          */
3196         for (i = 0; i < ngroups; i++) {
3197                 int blks;
3198
3199                 blks = count_overhead(sb, i, buf);
3200                 overhead += blks;
3201                 if (blks)
3202                         memset(buf, 0, PAGE_SIZE);
3203                 cond_resched();
3204         }
3205         sbi->s_overhead = overhead;
3206         smp_wmb();
3207         free_page((unsigned long) buf);
3208         return 0;
3209 }
3210
3211 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3212 {
3213         char *orig_data = kstrdup(data, GFP_KERNEL);
3214         struct buffer_head *bh;
3215         struct ext4_super_block *es = NULL;
3216         struct ext4_sb_info *sbi;
3217         ext4_fsblk_t block;
3218         ext4_fsblk_t sb_block = get_sb_block(&data);
3219         ext4_fsblk_t logical_sb_block;
3220         unsigned long offset = 0;
3221         unsigned long journal_devnum = 0;
3222         unsigned long def_mount_opts;
3223         struct inode *root;
3224         char *cp;
3225         const char *descr;
3226         int ret = -ENOMEM;
3227         int blocksize, clustersize;
3228         unsigned int db_count;
3229         unsigned int i;
3230         int needs_recovery, has_huge_files, has_bigalloc;
3231         __u64 blocks_count;
3232         int err;
3233         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3234         ext4_group_t first_not_zeroed;
3235
3236         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3237         if (!sbi)
3238                 goto out_free_orig;
3239
3240         sbi->s_blockgroup_lock =
3241                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3242         if (!sbi->s_blockgroup_lock) {
3243                 kfree(sbi);
3244                 goto out_free_orig;
3245         }
3246         sb->s_fs_info = sbi;
3247         sbi->s_mount_opt = 0;
3248         sbi->s_resuid = EXT4_DEF_RESUID;
3249         sbi->s_resgid = EXT4_DEF_RESGID;
3250         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3251         sbi->s_sb_block = sb_block;
3252         if (sb->s_bdev->bd_part)
3253                 sbi->s_sectors_written_start =
3254                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3255
3256         /* Cleanup superblock name */
3257         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3258                 *cp = '!';
3259
3260         ret = -EINVAL;
3261         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3262         if (!blocksize) {
3263                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3264                 goto out_fail;
3265         }
3266
3267         /*
3268          * The ext4 superblock will not be buffer aligned for other than 1kB
3269          * block sizes.  We need to calculate the offset from buffer start.
3270          */
3271         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3272                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3273                 offset = do_div(logical_sb_block, blocksize);
3274         } else {
3275                 logical_sb_block = sb_block;
3276         }
3277
3278         if (!(bh = sb_bread(sb, logical_sb_block))) {
3279                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3280                 goto out_fail;
3281         }
3282         /*
3283          * Note: s_es must be initialized as soon as possible because
3284          *       some ext4 macro-instructions depend on its value
3285          */
3286         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3287         sbi->s_es = es;
3288         sb->s_magic = le16_to_cpu(es->s_magic);
3289         if (sb->s_magic != EXT4_SUPER_MAGIC)
3290                 goto cantfind_ext4;
3291         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3292
3293         /* Set defaults before we parse the mount options */
3294         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3295         set_opt(sb, INIT_INODE_TABLE);
3296         if (def_mount_opts & EXT4_DEFM_DEBUG)
3297                 set_opt(sb, DEBUG);
3298         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3299                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3300                         "2.6.38");
3301                 set_opt(sb, GRPID);
3302         }
3303         if (def_mount_opts & EXT4_DEFM_UID16)
3304                 set_opt(sb, NO_UID32);
3305         /* xattr user namespace & acls are now defaulted on */
3306 #ifdef CONFIG_EXT4_FS_XATTR
3307         set_opt(sb, XATTR_USER);
3308 #endif
3309 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3310         set_opt(sb, POSIX_ACL);
3311 #endif
3312         set_opt(sb, MBLK_IO_SUBMIT);
3313         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3314                 set_opt(sb, JOURNAL_DATA);
3315         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3316                 set_opt(sb, ORDERED_DATA);
3317         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3318                 set_opt(sb, WRITEBACK_DATA);
3319
3320         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3321                 set_opt(sb, ERRORS_PANIC);
3322         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3323                 set_opt(sb, ERRORS_CONT);
3324         else
3325                 set_opt(sb, ERRORS_RO);
3326         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3327                 set_opt(sb, BLOCK_VALIDITY);
3328         if (def_mount_opts & EXT4_DEFM_DISCARD)
3329                 set_opt(sb, DISCARD);
3330
3331         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3332         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3333         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3334         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3335         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3336
3337         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3338                 set_opt(sb, BARRIER);
3339
3340         /*
3341          * enable delayed allocation by default
3342          * Use -o nodelalloc to turn it off
3343          */
3344         if (!IS_EXT3_SB(sb) &&
3345             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3346                 set_opt(sb, DELALLOC);
3347
3348         /*
3349          * set default s_li_wait_mult for lazyinit, for the case there is
3350          * no mount option specified.
3351          */
3352         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3353
3354         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3355                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3356                 ext4_msg(sb, KERN_WARNING,
3357                          "failed to parse options in superblock: %s",
3358                          sbi->s_es->s_mount_opts);
3359         }
3360         if (!parse_options((char *) data, sb, &journal_devnum,
3361                            &journal_ioprio, NULL, 0))
3362                 goto failed_mount;
3363
3364         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3365                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3366                             "with data=journal disables delayed "
3367                             "allocation and O_DIRECT support!\n");
3368                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3369                         ext4_msg(sb, KERN_ERR, "can't mount with "
3370                                  "both data=journal and delalloc");
3371                         goto failed_mount;
3372                 }
3373                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3374                         ext4_msg(sb, KERN_ERR, "can't mount with "
3375                                  "both data=journal and dioread_nolock");
3376                         goto failed_mount;
3377                 }
3378                 if (test_opt(sb, DELALLOC))
3379                         clear_opt(sb, DELALLOC);
3380         }
3381
3382         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3383                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3384
3385         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3386             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3387              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3388              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3389                 ext4_msg(sb, KERN_WARNING,
3390                        "feature flags set on rev 0 fs, "
3391                        "running e2fsck is recommended");
3392
3393         if (IS_EXT2_SB(sb)) {
3394                 if (ext2_feature_set_ok(sb))
3395                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3396                                  "using the ext4 subsystem");
3397                 else {
3398                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3399                                  "to feature incompatibilities");
3400                         goto failed_mount;
3401                 }
3402         }
3403
3404         if (IS_EXT3_SB(sb)) {
3405                 if (ext3_feature_set_ok(sb))
3406                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3407                                  "using the ext4 subsystem");
3408                 else {
3409                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3410                                  "to feature incompatibilities");
3411                         goto failed_mount;
3412                 }
3413         }
3414
3415         /*
3416          * Check feature flags regardless of the revision level, since we
3417          * previously didn't change the revision level when setting the flags,
3418          * so there is a chance incompat flags are set on a rev 0 filesystem.
3419          */
3420         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3421                 goto failed_mount;
3422
3423         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3424         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3425             blocksize > EXT4_MAX_BLOCK_SIZE) {
3426                 ext4_msg(sb, KERN_ERR,
3427                        "Unsupported filesystem blocksize %d", blocksize);
3428                 goto failed_mount;
3429         }
3430
3431         if (sb->s_blocksize != blocksize) {
3432                 /* Validate the filesystem blocksize */
3433                 if (!sb_set_blocksize(sb, blocksize)) {
3434                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3435                                         blocksize);
3436                         goto failed_mount;
3437                 }
3438
3439                 brelse(bh);
3440                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3441                 offset = do_div(logical_sb_block, blocksize);
3442                 bh = sb_bread(sb, logical_sb_block);
3443                 if (!bh) {
3444                         ext4_msg(sb, KERN_ERR,
3445                                "Can't read superblock on 2nd try");
3446                         goto failed_mount;
3447                 }
3448                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3449                 sbi->s_es = es;
3450                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3451                         ext4_msg(sb, KERN_ERR,
3452                                "Magic mismatch, very weird!");
3453                         goto failed_mount;
3454                 }
3455         }
3456
3457         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3458                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3459         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3460                                                       has_huge_files);
3461         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3462
3463         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3464                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3465                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3466         } else {
3467                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3468                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3469                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3470                     (!is_power_of_2(sbi->s_inode_size)) ||
3471                     (sbi->s_inode_size > blocksize)) {
3472                         ext4_msg(sb, KERN_ERR,
3473                                "unsupported inode size: %d",
3474                                sbi->s_inode_size);
3475                         goto failed_mount;
3476                 }
3477                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3478                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3479         }
3480
3481         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3482         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3483                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3484                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3485                     !is_power_of_2(sbi->s_desc_size)) {
3486                         ext4_msg(sb, KERN_ERR,
3487                                "unsupported descriptor size %lu",
3488                                sbi->s_desc_size);
3489                         goto failed_mount;
3490                 }
3491         } else
3492                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3493
3494         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3495         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3496         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3497                 goto cantfind_ext4;
3498
3499         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3500         if (sbi->s_inodes_per_block == 0)
3501                 goto cantfind_ext4;
3502         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3503                                         sbi->s_inodes_per_block;
3504         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3505         sbi->s_sbh = bh;
3506         sbi->s_mount_state = le16_to_cpu(es->s_state);
3507         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3508         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3509
3510         for (i = 0; i < 4; i++)
3511                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3512         sbi->s_def_hash_version = es->s_def_hash_version;
3513         i = le32_to_cpu(es->s_flags);
3514         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3515                 sbi->s_hash_unsigned = 3;
3516         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3517 #ifdef __CHAR_UNSIGNED__
3518                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3519                 sbi->s_hash_unsigned = 3;
3520 #else
3521                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3522 #endif
3523                 sb->s_dirt = 1;
3524         }
3525
3526         /* Handle clustersize */
3527         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3528         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3529                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3530         if (has_bigalloc) {
3531                 if (clustersize < blocksize) {
3532                         ext4_msg(sb, KERN_ERR,
3533                                  "cluster size (%d) smaller than "
3534                                  "block size (%d)", clustersize, blocksize);
3535                         goto failed_mount;
3536                 }
3537                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3538                         le32_to_cpu(es->s_log_block_size);
3539                 sbi->s_clusters_per_group =
3540                         le32_to_cpu(es->s_clusters_per_group);
3541                 if (sbi->s_clusters_per_group > blocksize * 8) {
3542                         ext4_msg(sb, KERN_ERR,
3543                                  "#clusters per group too big: %lu",
3544                                  sbi->s_clusters_per_group);
3545                         goto failed_mount;
3546                 }
3547                 if (sbi->s_blocks_per_group !=
3548                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3549                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3550                                  "clusters per group (%lu) inconsistent",
3551                                  sbi->s_blocks_per_group,
3552                                  sbi->s_clusters_per_group);
3553                         goto failed_mount;
3554                 }
3555         } else {
3556                 if (clustersize != blocksize) {
3557                         ext4_warning(sb, "fragment/cluster size (%d) != "
3558                                      "block size (%d)", clustersize,
3559                                      blocksize);
3560                         clustersize = blocksize;
3561                 }
3562                 if (sbi->s_blocks_per_group > blocksize * 8) {
3563                         ext4_msg(sb, KERN_ERR,
3564                                  "#blocks per group too big: %lu",
3565                                  sbi->s_blocks_per_group);
3566                         goto failed_mount;
3567                 }
3568                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3569                 sbi->s_cluster_bits = 0;
3570         }
3571         sbi->s_cluster_ratio = clustersize / blocksize;
3572
3573         if (sbi->s_inodes_per_group > blocksize * 8) {
3574                 ext4_msg(sb, KERN_ERR,
3575                        "#inodes per group too big: %lu",
3576                        sbi->s_inodes_per_group);
3577                 goto failed_mount;
3578         }
3579
3580         /*
3581          * Test whether we have more sectors than will fit in sector_t,
3582          * and whether the max offset is addressable by the page cache.
3583          */
3584         err = generic_check_addressable(sb->s_blocksize_bits,
3585                                         ext4_blocks_count(es));
3586         if (err) {
3587                 ext4_msg(sb, KERN_ERR, "filesystem"
3588                          " too large to mount safely on this system");
3589                 if (sizeof(sector_t) < 8)
3590                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3591                 ret = err;
3592                 goto failed_mount;
3593         }
3594
3595         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3596                 goto cantfind_ext4;
3597
3598         /* check blocks count against device size */
3599         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3600         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3601                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3602                        "exceeds size of device (%llu blocks)",
3603                        ext4_blocks_count(es), blocks_count);
3604                 goto failed_mount;
3605         }
3606
3607         /*
3608          * It makes no sense for the first data block to be beyond the end
3609          * of the filesystem.
3610          */
3611         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3612                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3613                          "block %u is beyond end of filesystem (%llu)",
3614                          le32_to_cpu(es->s_first_data_block),
3615                          ext4_blocks_count(es));
3616                 goto failed_mount;
3617         }
3618         blocks_count = (ext4_blocks_count(es) -
3619                         le32_to_cpu(es->s_first_data_block) +
3620                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3621         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3622         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3623                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3624                        "(block count %llu, first data block %u, "
3625                        "blocks per group %lu)", sbi->s_groups_count,
3626                        ext4_blocks_count(es),
3627                        le32_to_cpu(es->s_first_data_block),
3628                        EXT4_BLOCKS_PER_GROUP(sb));
3629                 goto failed_mount;
3630         }
3631         sbi->s_groups_count = blocks_count;
3632         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3633                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3634         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3635                    EXT4_DESC_PER_BLOCK(sb);
3636         sbi->s_group_desc = ext4_kvmalloc(db_count *
3637                                           sizeof(struct buffer_head *),
3638                                           GFP_KERNEL);
3639         if (sbi->s_group_desc == NULL) {
3640                 ext4_msg(sb, KERN_ERR, "not enough memory");
3641                 goto failed_mount;
3642         }
3643
3644         if (ext4_proc_root)
3645                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3646
3647         bgl_lock_init(sbi->s_blockgroup_lock);
3648
3649         for (i = 0; i < db_count; i++) {
3650                 block = descriptor_loc(sb, logical_sb_block, i);
3651                 sbi->s_group_desc[i] = sb_bread(sb, block);
3652                 if (!sbi->s_group_desc[i]) {
3653                         ext4_msg(sb, KERN_ERR,
3654                                "can't read group descriptor %d", i);
3655                         db_count = i;
3656                         goto failed_mount2;
3657                 }
3658         }
3659         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3660                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3661                 goto failed_mount2;
3662         }
3663         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3664                 if (!ext4_fill_flex_info(sb)) {
3665                         ext4_msg(sb, KERN_ERR,
3666                                "unable to initialize "
3667                                "flex_bg meta info!");
3668                         goto failed_mount2;
3669                 }
3670
3671         sbi->s_gdb_count = db_count;
3672         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3673         spin_lock_init(&sbi->s_next_gen_lock);
3674
3675         init_timer(&sbi->s_err_report);
3676         sbi->s_err_report.function = print_daily_error_info;
3677         sbi->s_err_report.data = (unsigned long) sb;
3678
3679         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3680                         ext4_count_free_clusters(sb));
3681         if (!err) {
3682                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3683                                 ext4_count_free_inodes(sb));
3684         }
3685         if (!err) {
3686                 err = percpu_counter_init(&sbi->s_dirs_counter,
3687                                 ext4_count_dirs(sb));
3688         }
3689         if (!err) {
3690                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3691         }
3692         if (err) {
3693                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3694                 goto failed_mount3;
3695         }
3696
3697         sbi->s_stripe = ext4_get_stripe_size(sbi);
3698         sbi->s_max_writeback_mb_bump = 128;
3699
3700         /*
3701          * set up enough so that it can read an inode
3702          */
3703         if (!test_opt(sb, NOLOAD) &&
3704             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3705                 sb->s_op = &ext4_sops;
3706         else
3707                 sb->s_op = &ext4_nojournal_sops;
3708         sb->s_export_op = &ext4_export_ops;
3709         sb->s_xattr = ext4_xattr_handlers;
3710 #ifdef CONFIG_QUOTA
3711         sb->s_qcop = &ext4_qctl_operations;
3712         sb->dq_op = &ext4_quota_operations;
3713 #endif
3714         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3715
3716         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3717         mutex_init(&sbi->s_orphan_lock);
3718         sbi->s_resize_flags = 0;
3719
3720         sb->s_root = NULL;
3721
3722         needs_recovery = (es->s_last_orphan != 0 ||
3723                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3724                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3725
3726         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3727             !(sb->s_flags & MS_RDONLY))
3728                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3729                         goto failed_mount3;
3730
3731         /*
3732          * The first inode we look at is the journal inode.  Don't try
3733          * root first: it may be modified in the journal!
3734          */
3735         if (!test_opt(sb, NOLOAD) &&
3736             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3737                 if (ext4_load_journal(sb, es, journal_devnum))
3738                         goto failed_mount3;
3739         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3740               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3741                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3742                        "suppressed and not mounted read-only");
3743                 goto failed_mount_wq;
3744         } else {
3745                 clear_opt(sb, DATA_FLAGS);
3746                 sbi->s_journal = NULL;
3747                 needs_recovery = 0;
3748                 goto no_journal;
3749         }
3750
3751         if (ext4_blocks_count(es) > 0xffffffffULL &&
3752             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3753                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3754                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3755                 goto failed_mount_wq;
3756         }
3757
3758         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3759                 jbd2_journal_set_features(sbi->s_journal,
3760                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3761                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3762         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3763                 jbd2_journal_set_features(sbi->s_journal,
3764                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3765                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3766                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3767         } else {
3768                 jbd2_journal_clear_features(sbi->s_journal,
3769                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3770                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3771         }
3772
3773         /* We have now updated the journal if required, so we can
3774          * validate the data journaling mode. */
3775         switch (test_opt(sb, DATA_FLAGS)) {
3776         case 0:
3777                 /* No mode set, assume a default based on the journal
3778                  * capabilities: ORDERED_DATA if the journal can
3779                  * cope, else JOURNAL_DATA
3780                  */
3781                 if (jbd2_journal_check_available_features
3782                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3783                         set_opt(sb, ORDERED_DATA);
3784                 else
3785                         set_opt(sb, JOURNAL_DATA);
3786                 break;
3787
3788         case EXT4_MOUNT_ORDERED_DATA:
3789         case EXT4_MOUNT_WRITEBACK_DATA:
3790                 if (!jbd2_journal_check_available_features
3791                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3792                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3793                                "requested data journaling mode");
3794                         goto failed_mount_wq;
3795                 }
3796         default:
3797                 break;
3798         }
3799         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3800
3801         /*
3802          * The journal may have updated the bg summary counts, so we
3803          * need to update the global counters.
3804          */
3805         percpu_counter_set(&sbi->s_freeclusters_counter,
3806                            ext4_count_free_clusters(sb));
3807         percpu_counter_set(&sbi->s_freeinodes_counter,
3808                            ext4_count_free_inodes(sb));
3809         percpu_counter_set(&sbi->s_dirs_counter,
3810                            ext4_count_dirs(sb));
3811         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3812
3813 no_journal:
3814         /*
3815          * Get the # of file system overhead blocks from the
3816          * superblock if present.
3817          */
3818         if (es->s_overhead_clusters)
3819                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3820         else {
3821                 ret = ext4_calculate_overhead(sb);
3822                 if (ret)
3823                         goto failed_mount_wq;
3824         }
3825
3826         /*
3827          * The maximum number of concurrent works can be high and
3828          * concurrency isn't really necessary.  Limit it to 1.
3829          */
3830         EXT4_SB(sb)->dio_unwritten_wq =
3831                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3832         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3833                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3834                 goto failed_mount_wq;
3835         }
3836
3837         /*
3838          * The jbd2_journal_load will have done any necessary log recovery,
3839          * so we can safely mount the rest of the filesystem now.
3840          */
3841
3842         root = ext4_iget(sb, EXT4_ROOT_INO);
3843         if (IS_ERR(root)) {
3844                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3845                 ret = PTR_ERR(root);
3846                 root = NULL;
3847                 goto failed_mount4;
3848         }
3849         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3850                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3851                 goto failed_mount4;
3852         }
3853         sb->s_root = d_alloc_root(root);
3854         if (!sb->s_root) {
3855                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3856                 ret = -ENOMEM;
3857                 goto failed_mount4;
3858         }
3859
3860         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3861                 sb->s_flags |= MS_RDONLY;
3862
3863         /* determine the minimum size of new large inodes, if present */
3864         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3865                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3866                                                      EXT4_GOOD_OLD_INODE_SIZE;
3867                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3868                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3869                         if (sbi->s_want_extra_isize <
3870                             le16_to_cpu(es->s_want_extra_isize))
3871                                 sbi->s_want_extra_isize =
3872                                         le16_to_cpu(es->s_want_extra_isize);
3873                         if (sbi->s_want_extra_isize <
3874                             le16_to_cpu(es->s_min_extra_isize))
3875                                 sbi->s_want_extra_isize =
3876                                         le16_to_cpu(es->s_min_extra_isize);
3877                 }
3878         }
3879         /* Check if enough inode space is available */
3880         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3881                                                         sbi->s_inode_size) {
3882                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3883                                                        EXT4_GOOD_OLD_INODE_SIZE;
3884                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3885                          "available");
3886         }
3887
3888         err = ext4_setup_system_zone(sb);
3889         if (err) {
3890                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3891                          "zone (%d)", err);
3892                 goto failed_mount4;
3893         }
3894
3895         ext4_ext_init(sb);
3896         err = ext4_mb_init(sb, needs_recovery);
3897         if (err) {
3898                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3899                          err);
3900                 goto failed_mount5;
3901         }
3902
3903         err = ext4_register_li_request(sb, first_not_zeroed);
3904         if (err)
3905                 goto failed_mount6;
3906
3907         sbi->s_kobj.kset = ext4_kset;
3908         init_completion(&sbi->s_kobj_unregister);
3909         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3910                                    "%s", sb->s_id);
3911         if (err)
3912                 goto failed_mount7;
3913
3914         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3915         ext4_orphan_cleanup(sb, es);
3916         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3917         if (needs_recovery) {
3918                 ext4_msg(sb, KERN_INFO, "recovery complete");
3919                 ext4_mark_recovery_complete(sb, es);
3920         }
3921         if (EXT4_SB(sb)->s_journal) {
3922                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3923                         descr = " journalled data mode";
3924                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3925                         descr = " ordered data mode";
3926                 else
3927                         descr = " writeback data mode";
3928         } else
3929                 descr = "out journal";
3930
3931         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3932                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3933                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3934
3935         if (es->s_error_count)
3936                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3937
3938         kfree(orig_data);
3939         return 0;
3940
3941 cantfind_ext4:
3942         if (!silent)
3943                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3944         goto failed_mount;
3945
3946 failed_mount7:
3947         ext4_unregister_li_request(sb);
3948 failed_mount6:
3949         ext4_ext_release(sb);
3950 failed_mount5:
3951         ext4_mb_release(sb);
3952         ext4_release_system_zone(sb);
3953 failed_mount4:
3954         iput(root);
3955         sb->s_root = NULL;
3956         ext4_msg(sb, KERN_ERR, "mount failed");
3957         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3958 failed_mount_wq:
3959         if (sbi->s_journal) {
3960                 jbd2_journal_destroy(sbi->s_journal);
3961                 sbi->s_journal = NULL;
3962         }
3963 failed_mount3:
3964         del_timer_sync(&sbi->s_err_report);
3965         if (sbi->s_flex_groups)
3966                 ext4_kvfree(sbi->s_flex_groups);
3967         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3968         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3969         percpu_counter_destroy(&sbi->s_dirs_counter);
3970         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3971         if (sbi->s_mmp_tsk)
3972                 kthread_stop(sbi->s_mmp_tsk);
3973 failed_mount2:
3974         for (i = 0; i < db_count; i++)
3975                 brelse(sbi->s_group_desc[i]);
3976         ext4_kvfree(sbi->s_group_desc);
3977 failed_mount:
3978         if (sbi->s_proc) {
3979                 remove_proc_entry(sb->s_id, ext4_proc_root);
3980         }
3981 #ifdef CONFIG_QUOTA
3982         for (i = 0; i < MAXQUOTAS; i++)
3983                 kfree(sbi->s_qf_names[i]);
3984 #endif
3985         ext4_blkdev_remove(sbi);
3986         brelse(bh);
3987 out_fail:
3988         sb->s_fs_info = NULL;
3989         kfree(sbi->s_blockgroup_lock);
3990         kfree(sbi);
3991 out_free_orig:
3992         kfree(orig_data);
3993         return ret;
3994 }
3995
3996 /*
3997  * Setup any per-fs journal parameters now.  We'll do this both on
3998  * initial mount, once the journal has been initialised but before we've
3999  * done any recovery; and again on any subsequent remount.
4000  */
4001 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4002 {
4003         struct ext4_sb_info *sbi = EXT4_SB(sb);
4004
4005         journal->j_commit_interval = sbi->s_commit_interval;
4006         journal->j_min_batch_time = sbi->s_min_batch_time;
4007         journal->j_max_batch_time = sbi->s_max_batch_time;
4008
4009         write_lock(&journal->j_state_lock);
4010         if (test_opt(sb, BARRIER))
4011                 journal->j_flags |= JBD2_BARRIER;
4012         else
4013                 journal->j_flags &= ~JBD2_BARRIER;
4014         if (test_opt(sb, DATA_ERR_ABORT))
4015                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4016         else
4017                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4018         write_unlock(&journal->j_state_lock);
4019 }
4020
4021 static journal_t *ext4_get_journal(struct super_block *sb,
4022                                    unsigned int journal_inum)
4023 {
4024         struct inode *journal_inode;
4025         journal_t *journal;
4026
4027         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4028
4029         /* First, test for the existence of a valid inode on disk.  Bad
4030          * things happen if we iget() an unused inode, as the subsequent
4031          * iput() will try to delete it. */
4032
4033         journal_inode = ext4_iget(sb, journal_inum);
4034         if (IS_ERR(journal_inode)) {
4035                 ext4_msg(sb, KERN_ERR, "no journal found");
4036                 return NULL;
4037         }
4038         if (!journal_inode->i_nlink) {
4039                 make_bad_inode(journal_inode);
4040                 iput(journal_inode);
4041                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4042                 return NULL;
4043         }
4044
4045         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4046                   journal_inode, journal_inode->i_size);
4047         if (!S_ISREG(journal_inode->i_mode)) {
4048                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4049                 iput(journal_inode);
4050                 return NULL;
4051         }
4052
4053         journal = jbd2_journal_init_inode(journal_inode);
4054         if (!journal) {
4055                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4056                 iput(journal_inode);
4057                 return NULL;
4058         }
4059         journal->j_private = sb;
4060         ext4_init_journal_params(sb, journal);
4061         return journal;
4062 }
4063
4064 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4065                                        dev_t j_dev)
4066 {
4067         struct buffer_head *bh;
4068         journal_t *journal;
4069         ext4_fsblk_t start;
4070         ext4_fsblk_t len;
4071         int hblock, blocksize;
4072         ext4_fsblk_t sb_block;
4073         unsigned long offset;
4074         struct ext4_super_block *es;
4075         struct block_device *bdev;
4076
4077         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4078
4079         bdev = ext4_blkdev_get(j_dev, sb);
4080         if (bdev == NULL)
4081                 return NULL;
4082
4083         blocksize = sb->s_blocksize;
4084         hblock = bdev_logical_block_size(bdev);
4085         if (blocksize < hblock) {
4086                 ext4_msg(sb, KERN_ERR,
4087                         "blocksize too small for journal device");
4088                 goto out_bdev;
4089         }
4090
4091         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4092         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4093         set_blocksize(bdev, blocksize);
4094         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4095                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4096                        "external journal");
4097                 goto out_bdev;
4098         }
4099
4100         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
4101         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4102             !(le32_to_cpu(es->s_feature_incompat) &
4103               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4104                 ext4_msg(sb, KERN_ERR, "external journal has "
4105                                         "bad superblock");
4106                 brelse(bh);
4107                 goto out_bdev;
4108         }
4109
4110         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4111                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4112                 brelse(bh);
4113                 goto out_bdev;
4114         }
4115
4116         len = ext4_blocks_count(es);
4117         start = sb_block + 1;
4118         brelse(bh);     /* we're done with the superblock */
4119
4120         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4121                                         start, len, blocksize);
4122         if (!journal) {
4123                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4124                 goto out_bdev;
4125         }
4126         journal->j_private = sb;
4127         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4128         wait_on_buffer(journal->j_sb_buffer);
4129         if (!buffer_uptodate(journal->j_sb_buffer)) {
4130                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4131                 goto out_journal;
4132         }
4133         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4134                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4135                                         "user (unsupported) - %d",
4136                         be32_to_cpu(journal->j_superblock->s_nr_users));
4137                 goto out_journal;
4138         }
4139         EXT4_SB(sb)->journal_bdev = bdev;
4140         ext4_init_journal_params(sb, journal);
4141         return journal;
4142
4143 out_journal:
4144         jbd2_journal_destroy(journal);
4145 out_bdev:
4146         ext4_blkdev_put(bdev);
4147         return NULL;
4148 }
4149
4150 static int ext4_load_journal(struct super_block *sb,
4151                              struct ext4_super_block *es,
4152                              unsigned long journal_devnum)
4153 {
4154         journal_t *journal;
4155         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4156         dev_t journal_dev;
4157         int err = 0;
4158         int really_read_only;
4159
4160         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4161
4162         if (journal_devnum &&
4163             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4164                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4165                         "numbers have changed");
4166                 journal_dev = new_decode_dev(journal_devnum);
4167         } else
4168                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4169
4170         really_read_only = bdev_read_only(sb->s_bdev);
4171
4172         /*
4173          * Are we loading a blank journal or performing recovery after a
4174          * crash?  For recovery, we need to check in advance whether we
4175          * can get read-write access to the device.
4176          */
4177         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4178                 if (sb->s_flags & MS_RDONLY) {
4179                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4180                                         "required on readonly filesystem");
4181                         if (really_read_only) {
4182                                 ext4_msg(sb, KERN_ERR, "write access "
4183                                         "unavailable, cannot proceed");
4184                                 return -EROFS;
4185                         }
4186                         ext4_msg(sb, KERN_INFO, "write access will "
4187                                "be enabled during recovery");
4188                 }
4189         }
4190
4191         if (journal_inum && journal_dev) {
4192                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4193                        "and inode journals!");
4194                 return -EINVAL;
4195         }
4196
4197         if (journal_inum) {
4198                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4199                         return -EINVAL;
4200         } else {
4201                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4202                         return -EINVAL;
4203         }
4204
4205         if (!(journal->j_flags & JBD2_BARRIER))
4206                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4207
4208         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4209                 err = jbd2_journal_update_format(journal);
4210                 if (err)  {
4211                         ext4_msg(sb, KERN_ERR, "error updating journal");
4212                         jbd2_journal_destroy(journal);
4213                         return err;
4214                 }
4215         }
4216
4217         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4218                 err = jbd2_journal_wipe(journal, !really_read_only);
4219         if (!err) {
4220                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4221                 if (save)
4222                         memcpy(save, ((char *) es) +
4223                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4224                 err = jbd2_journal_load(journal);
4225                 if (save)
4226                         memcpy(((char *) es) + EXT4_S_ERR_START,
4227                                save, EXT4_S_ERR_LEN);
4228                 kfree(save);
4229         }
4230
4231         if (err) {
4232                 ext4_msg(sb, KERN_ERR, "error loading journal");
4233                 jbd2_journal_destroy(journal);
4234                 return err;
4235         }
4236
4237         EXT4_SB(sb)->s_journal = journal;
4238         ext4_clear_journal_err(sb, es);
4239
4240         if (!really_read_only && journal_devnum &&
4241             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4242                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4243
4244                 /* Make sure we flush the recovery flag to disk. */
4245                 ext4_commit_super(sb, 1);
4246         }
4247
4248         return 0;
4249 }
4250
4251 static int ext4_commit_super(struct super_block *sb, int sync)
4252 {
4253         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4254         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4255         int error = 0;
4256
4257         if (!sbh || block_device_ejected(sb))
4258                 return error;
4259         if (buffer_write_io_error(sbh)) {
4260                 /*
4261                  * Oh, dear.  A previous attempt to write the
4262                  * superblock failed.  This could happen because the
4263                  * USB device was yanked out.  Or it could happen to
4264                  * be a transient write error and maybe the block will
4265                  * be remapped.  Nothing we can do but to retry the
4266                  * write and hope for the best.
4267                  */
4268                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4269                        "superblock detected");
4270                 clear_buffer_write_io_error(sbh);
4271                 set_buffer_uptodate(sbh);
4272         }
4273         /*
4274          * If the file system is mounted read-only, don't update the
4275          * superblock write time.  This avoids updating the superblock
4276          * write time when we are mounting the root file system
4277          * read/only but we need to replay the journal; at that point,
4278          * for people who are east of GMT and who make their clock
4279          * tick in localtime for Windows bug-for-bug compatibility,
4280          * the clock is set in the future, and this will cause e2fsck
4281          * to complain and force a full file system check.
4282          */
4283         if (!(sb->s_flags & MS_RDONLY))
4284                 es->s_wtime = cpu_to_le32(get_seconds());
4285         if (sb->s_bdev->bd_part)
4286                 es->s_kbytes_written =
4287                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4288                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4289                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4290         else
4291                 es->s_kbytes_written =
4292                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4293         ext4_free_blocks_count_set(es,
4294                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4295                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4296         es->s_free_inodes_count =
4297                 cpu_to_le32(percpu_counter_sum_positive(
4298                                 &EXT4_SB(sb)->s_freeinodes_counter));
4299         sb->s_dirt = 0;
4300         BUFFER_TRACE(sbh, "marking dirty");
4301         mark_buffer_dirty(sbh);
4302         if (sync) {
4303                 error = sync_dirty_buffer(sbh);
4304                 if (error)
4305                         return error;
4306
4307                 error = buffer_write_io_error(sbh);
4308                 if (error) {
4309                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4310                                "superblock");
4311                         clear_buffer_write_io_error(sbh);
4312                         set_buffer_uptodate(sbh);
4313                 }
4314         }
4315         return error;
4316 }
4317
4318 /*
4319  * Have we just finished recovery?  If so, and if we are mounting (or
4320  * remounting) the filesystem readonly, then we will end up with a
4321  * consistent fs on disk.  Record that fact.
4322  */
4323 static void ext4_mark_recovery_complete(struct super_block *sb,
4324                                         struct ext4_super_block *es)
4325 {
4326         journal_t *journal = EXT4_SB(sb)->s_journal;
4327
4328         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4329                 BUG_ON(journal != NULL);
4330                 return;
4331         }
4332         jbd2_journal_lock_updates(journal);
4333         if (jbd2_journal_flush(journal) < 0)
4334                 goto out;
4335
4336         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4337             sb->s_flags & MS_RDONLY) {
4338                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4339                 ext4_commit_super(sb, 1);
4340         }
4341
4342 out:
4343         jbd2_journal_unlock_updates(journal);
4344 }
4345
4346 /*
4347  * If we are mounting (or read-write remounting) a filesystem whose journal
4348  * has recorded an error from a previous lifetime, move that error to the
4349  * main filesystem now.
4350  */
4351 static void ext4_clear_journal_err(struct super_block *sb,
4352                                    struct ext4_super_block *es)
4353 {
4354         journal_t *journal;
4355         int j_errno;
4356         const char *errstr;
4357
4358         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4359
4360         journal = EXT4_SB(sb)->s_journal;
4361
4362         /*
4363          * Now check for any error status which may have been recorded in the
4364          * journal by a prior ext4_error() or ext4_abort()
4365          */
4366
4367         j_errno = jbd2_journal_errno(journal);
4368         if (j_errno) {
4369                 char nbuf[16];
4370
4371                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4372                 ext4_warning(sb, "Filesystem error recorded "
4373                              "from previous mount: %s", errstr);
4374                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4375
4376                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4377                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4378                 ext4_commit_super(sb, 1);
4379
4380                 jbd2_journal_clear_err(journal);
4381         }
4382 }
4383
4384 /*
4385  * Force the running and committing transactions to commit,
4386  * and wait on the commit.
4387  */
4388 int ext4_force_commit(struct super_block *sb)
4389 {
4390         journal_t *journal;
4391         int ret = 0;
4392
4393         if (sb->s_flags & MS_RDONLY)
4394                 return 0;
4395
4396         journal = EXT4_SB(sb)->s_journal;
4397         if (journal) {
4398                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4399                 ret = ext4_journal_force_commit(journal);
4400         }
4401
4402         return ret;
4403 }
4404
4405 static void ext4_write_super(struct super_block *sb)
4406 {
4407         lock_super(sb);
4408         ext4_commit_super(sb, 1);
4409         unlock_super(sb);
4410 }
4411
4412 static int ext4_sync_fs(struct super_block *sb, int wait)
4413 {
4414         int ret = 0;
4415         tid_t target;
4416         struct ext4_sb_info *sbi = EXT4_SB(sb);
4417
4418         trace_ext4_sync_fs(sb, wait);
4419         flush_workqueue(sbi->dio_unwritten_wq);
4420         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4421                 if (wait)
4422                         jbd2_log_wait_commit(sbi->s_journal, target);
4423         }
4424         return ret;
4425 }
4426
4427 /*
4428  * LVM calls this function before a (read-only) snapshot is created.  This
4429  * gives us a chance to flush the journal completely and mark the fs clean.
4430  *
4431  * Note that only this function cannot bring a filesystem to be in a clean
4432  * state independently, because ext4 prevents a new handle from being started
4433  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4434  * the upper layer.
4435  */
4436 static int ext4_freeze(struct super_block *sb)
4437 {
4438         int error = 0;
4439         journal_t *journal;
4440
4441         if (sb->s_flags & MS_RDONLY)
4442                 return 0;
4443
4444         journal = EXT4_SB(sb)->s_journal;
4445
4446         /* Now we set up the journal barrier. */
4447         jbd2_journal_lock_updates(journal);
4448
4449         /*
4450          * Don't clear the needs_recovery flag if we failed to flush
4451          * the journal.
4452          */
4453         error = jbd2_journal_flush(journal);
4454         if (error < 0)
4455                 goto out;
4456
4457         /* Journal blocked and flushed, clear needs_recovery flag. */
4458         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4459         error = ext4_commit_super(sb, 1);
4460 out:
4461         /* we rely on s_frozen to stop further updates */
4462         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4463         return error;
4464 }
4465
4466 /*
4467  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4468  * flag here, even though the filesystem is not technically dirty yet.
4469  */
4470 static int ext4_unfreeze(struct super_block *sb)
4471 {
4472         if (sb->s_flags & MS_RDONLY)
4473                 return 0;
4474
4475         lock_super(sb);
4476         /* Reset the needs_recovery flag before the fs is unlocked. */
4477         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4478         ext4_commit_super(sb, 1);
4479         unlock_super(sb);
4480         return 0;
4481 }
4482
4483 /*
4484  * Structure to save mount options for ext4_remount's benefit
4485  */
4486 struct ext4_mount_options {
4487         unsigned long s_mount_opt;
4488         unsigned long s_mount_opt2;
4489         uid_t s_resuid;
4490         gid_t s_resgid;
4491         unsigned long s_commit_interval;
4492         u32 s_min_batch_time, s_max_batch_time;
4493 #ifdef CONFIG_QUOTA
4494         int s_jquota_fmt;
4495         char *s_qf_names[MAXQUOTAS];
4496 #endif
4497 };
4498
4499 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4500 {
4501         struct ext4_super_block *es;
4502         struct ext4_sb_info *sbi = EXT4_SB(sb);
4503         ext4_fsblk_t n_blocks_count = 0;
4504         unsigned long old_sb_flags;
4505         struct ext4_mount_options old_opts;
4506         int enable_quota = 0;
4507         ext4_group_t g;
4508         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4509         int err = 0;
4510 #ifdef CONFIG_QUOTA
4511         int i;
4512 #endif
4513         char *orig_data = kstrdup(data, GFP_KERNEL);
4514
4515         /* Store the original options */
4516         lock_super(sb);
4517         old_sb_flags = sb->s_flags;
4518         old_opts.s_mount_opt = sbi->s_mount_opt;
4519         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4520         old_opts.s_resuid = sbi->s_resuid;
4521         old_opts.s_resgid = sbi->s_resgid;
4522         old_opts.s_commit_interval = sbi->s_commit_interval;
4523         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4524         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4525 #ifdef CONFIG_QUOTA
4526         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4527         for (i = 0; i < MAXQUOTAS; i++)
4528                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4529 #endif
4530         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4531                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4532
4533         /*
4534          * Allow the "check" option to be passed as a remount option.
4535          */
4536         if (!parse_options(data, sb, NULL, &journal_ioprio,
4537                            &n_blocks_count, 1)) {
4538                 err = -EINVAL;
4539                 goto restore_opts;
4540         }
4541
4542         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4543                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4544                         ext4_msg(sb, KERN_ERR, "can't mount with "
4545                                  "both data=journal and delalloc");
4546                         err = -EINVAL;
4547                         goto restore_opts;
4548                 }
4549                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4550                         ext4_msg(sb, KERN_ERR, "can't mount with "
4551                                  "both data=journal and dioread_nolock");
4552                         err = -EINVAL;
4553                         goto restore_opts;
4554                 }
4555         }
4556
4557         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4558                 ext4_abort(sb, "Abort forced by user");
4559
4560         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4561                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4562
4563         es = sbi->s_es;
4564
4565         if (sbi->s_journal) {
4566                 ext4_init_journal_params(sb, sbi->s_journal);
4567                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4568         }
4569
4570         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4571                 n_blocks_count > ext4_blocks_count(es)) {
4572                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4573                         err = -EROFS;
4574                         goto restore_opts;
4575                 }
4576
4577                 if (*flags & MS_RDONLY) {
4578                         err = dquot_suspend(sb, -1);
4579                         if (err < 0)
4580                                 goto restore_opts;
4581
4582                         /*
4583                          * First of all, the unconditional stuff we have to do
4584                          * to disable replay of the journal when we next remount
4585                          */
4586                         sb->s_flags |= MS_RDONLY;
4587
4588                         /*
4589                          * OK, test if we are remounting a valid rw partition
4590                          * readonly, and if so set the rdonly flag and then
4591                          * mark the partition as valid again.
4592                          */
4593                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4594                             (sbi->s_mount_state & EXT4_VALID_FS))
4595                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4596
4597                         if (sbi->s_journal)
4598                                 ext4_mark_recovery_complete(sb, es);
4599                 } else {
4600                         /* Make sure we can mount this feature set readwrite */
4601                         if (!ext4_feature_set_ok(sb, 0)) {
4602                                 err = -EROFS;
4603                                 goto restore_opts;
4604                         }
4605                         /*
4606                          * Make sure the group descriptor checksums
4607                          * are sane.  If they aren't, refuse to remount r/w.
4608                          */
4609                         for (g = 0; g < sbi->s_groups_count; g++) {
4610                                 struct ext4_group_desc *gdp =
4611                                         ext4_get_group_desc(sb, g, NULL);
4612
4613                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4614                                         ext4_msg(sb, KERN_ERR,
4615                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4616                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4617                                                le16_to_cpu(gdp->bg_checksum));
4618                                         err = -EINVAL;
4619                                         goto restore_opts;
4620                                 }
4621                         }
4622
4623                         /*
4624                          * If we have an unprocessed orphan list hanging
4625                          * around from a previously readonly bdev mount,
4626                          * require a full umount/remount for now.
4627                          */
4628                         if (es->s_last_orphan) {
4629                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4630                                        "remount RDWR because of unprocessed "
4631                                        "orphan inode list.  Please "
4632                                        "umount/remount instead");
4633                                 err = -EINVAL;
4634                                 goto restore_opts;
4635                         }
4636
4637                         /*
4638                          * Mounting a RDONLY partition read-write, so reread
4639                          * and store the current valid flag.  (It may have
4640                          * been changed by e2fsck since we originally mounted
4641                          * the partition.)
4642                          */
4643                         if (sbi->s_journal)
4644                                 ext4_clear_journal_err(sb, es);
4645                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4646                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4647                                 goto restore_opts;
4648                         if (!ext4_setup_super(sb, es, 0))
4649                                 sb->s_flags &= ~MS_RDONLY;
4650                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4651                                                      EXT4_FEATURE_INCOMPAT_MMP))
4652                                 if (ext4_multi_mount_protect(sb,
4653                                                 le64_to_cpu(es->s_mmp_block))) {
4654                                         err = -EROFS;
4655                                         goto restore_opts;
4656                                 }
4657                         enable_quota = 1;
4658                 }
4659         }
4660
4661         /*
4662          * Reinitialize lazy itable initialization thread based on
4663          * current settings
4664          */
4665         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4666                 ext4_unregister_li_request(sb);
4667         else {
4668                 ext4_group_t first_not_zeroed;
4669                 first_not_zeroed = ext4_has_uninit_itable(sb);
4670                 ext4_register_li_request(sb, first_not_zeroed);
4671         }
4672
4673         ext4_setup_system_zone(sb);
4674         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4675                 ext4_commit_super(sb, 1);
4676
4677 #ifdef CONFIG_QUOTA
4678         /* Release old quota file names */
4679         for (i = 0; i < MAXQUOTAS; i++)
4680                 if (old_opts.s_qf_names[i] &&
4681                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4682                         kfree(old_opts.s_qf_names[i]);
4683 #endif
4684         unlock_super(sb);
4685         if (enable_quota)
4686                 dquot_resume(sb, -1);
4687
4688         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4689         kfree(orig_data);
4690         return 0;
4691
4692 restore_opts:
4693         sb->s_flags = old_sb_flags;
4694         sbi->s_mount_opt = old_opts.s_mount_opt;
4695         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4696         sbi->s_resuid = old_opts.s_resuid;
4697         sbi->s_resgid = old_opts.s_resgid;
4698         sbi->s_commit_interval = old_opts.s_commit_interval;
4699         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4700         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4701 #ifdef CONFIG_QUOTA
4702         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4703         for (i = 0; i < MAXQUOTAS; i++) {
4704                 if (sbi->s_qf_names[i] &&
4705                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4706                         kfree(sbi->s_qf_names[i]);
4707                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4708         }
4709 #endif
4710         unlock_super(sb);
4711         kfree(orig_data);
4712         return err;
4713 }
4714
4715 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4716 {
4717         struct super_block *sb = dentry->d_sb;
4718         struct ext4_sb_info *sbi = EXT4_SB(sb);
4719         struct ext4_super_block *es = sbi->s_es;
4720         ext4_fsblk_t overhead = 0;
4721         u64 fsid;
4722         s64 bfree;
4723
4724         if (!test_opt(sb, MINIX_DF))
4725                 overhead = sbi->s_overhead;
4726
4727         buf->f_type = EXT4_SUPER_MAGIC;
4728         buf->f_bsize = sb->s_blocksize;
4729         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4730         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4731                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4732         /* prevent underflow in case that few free space is available */
4733         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4734         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4735         if (buf->f_bfree < ext4_r_blocks_count(es))
4736                 buf->f_bavail = 0;
4737         buf->f_files = le32_to_cpu(es->s_inodes_count);
4738         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4739         buf->f_namelen = EXT4_NAME_LEN;
4740         fsid = le64_to_cpup((void *)es->s_uuid) ^
4741                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4742         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4743         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4744
4745         return 0;
4746 }
4747
4748 /* Helper function for writing quotas on sync - we need to start transaction
4749  * before quota file is locked for write. Otherwise the are possible deadlocks:
4750  * Process 1                         Process 2
4751  * ext4_create()                     quota_sync()
4752  *   jbd2_journal_start()                  write_dquot()
4753  *   dquot_initialize()                         down(dqio_mutex)
4754  *     down(dqio_mutex)                    jbd2_journal_start()
4755  *
4756  */
4757
4758 #ifdef CONFIG_QUOTA
4759
4760 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4761 {
4762         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4763 }
4764
4765 static int ext4_write_dquot(struct dquot *dquot)
4766 {
4767         int ret, err;
4768         handle_t *handle;
4769         struct inode *inode;
4770
4771         inode = dquot_to_inode(dquot);
4772         handle = ext4_journal_start(inode,
4773                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4774         if (IS_ERR(handle))
4775                 return PTR_ERR(handle);
4776         ret = dquot_commit(dquot);
4777         err = ext4_journal_stop(handle);
4778         if (!ret)
4779                 ret = err;
4780         return ret;
4781 }
4782
4783 static int ext4_acquire_dquot(struct dquot *dquot)
4784 {
4785         int ret, err;
4786         handle_t *handle;
4787
4788         handle = ext4_journal_start(dquot_to_inode(dquot),
4789                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4790         if (IS_ERR(handle))
4791                 return PTR_ERR(handle);
4792         ret = dquot_acquire(dquot);
4793         err = ext4_journal_stop(handle);
4794         if (!ret)
4795                 ret = err;
4796         return ret;
4797 }
4798
4799 static int ext4_release_dquot(struct dquot *dquot)
4800 {
4801         int ret, err;
4802         handle_t *handle;
4803
4804         handle = ext4_journal_start(dquot_to_inode(dquot),
4805                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4806         if (IS_ERR(handle)) {
4807                 /* Release dquot anyway to avoid endless cycle in dqput() */
4808                 dquot_release(dquot);
4809                 return PTR_ERR(handle);
4810         }
4811         ret = dquot_release(dquot);
4812         err = ext4_journal_stop(handle);
4813         if (!ret)
4814                 ret = err;
4815         return ret;
4816 }
4817
4818 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4819 {
4820         /* Are we journaling quotas? */
4821         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4822             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4823                 dquot_mark_dquot_dirty(dquot);
4824                 return ext4_write_dquot(dquot);
4825         } else {
4826                 return dquot_mark_dquot_dirty(dquot);
4827         }
4828 }
4829
4830 static int ext4_write_info(struct super_block *sb, int type)
4831 {
4832         int ret, err;
4833         handle_t *handle;
4834
4835         /* Data block + inode block */
4836         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4837         if (IS_ERR(handle))
4838                 return PTR_ERR(handle);
4839         ret = dquot_commit_info(sb, type);
4840         err = ext4_journal_stop(handle);
4841         if (!ret)
4842                 ret = err;
4843         return ret;
4844 }
4845
4846 /*
4847  * Turn on quotas during mount time - we need to find
4848  * the quota file and such...
4849  */
4850 static int ext4_quota_on_mount(struct super_block *sb, int type)
4851 {
4852         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4853                                         EXT4_SB(sb)->s_jquota_fmt, type);
4854 }
4855
4856 /*
4857  * Standard function to be called on quota_on
4858  */
4859 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4860                          struct path *path)
4861 {
4862         int err;
4863
4864         if (!test_opt(sb, QUOTA))
4865                 return -EINVAL;
4866
4867         /* Quotafile not on the same filesystem? */
4868         if (path->mnt->mnt_sb != sb)
4869                 return -EXDEV;
4870         /* Journaling quota? */
4871         if (EXT4_SB(sb)->s_qf_names[type]) {
4872                 /* Quotafile not in fs root? */
4873                 if (path->dentry->d_parent != sb->s_root)
4874                         ext4_msg(sb, KERN_WARNING,
4875                                 "Quota file not on filesystem root. "
4876                                 "Journaled quota will not work");
4877         }
4878
4879         /*
4880          * When we journal data on quota file, we have to flush journal to see
4881          * all updates to the file when we bypass pagecache...
4882          */
4883         if (EXT4_SB(sb)->s_journal &&
4884             ext4_should_journal_data(path->dentry->d_inode)) {
4885                 /*
4886                  * We don't need to lock updates but journal_flush() could
4887                  * otherwise be livelocked...
4888                  */
4889                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4890                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4891                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4892                 if (err)
4893                         return err;
4894         }
4895
4896         return dquot_quota_on(sb, type, format_id, path);
4897 }
4898
4899 static int ext4_quota_off(struct super_block *sb, int type)
4900 {
4901         struct inode *inode = sb_dqopt(sb)->files[type];
4902         handle_t *handle;
4903
4904         /* Force all delayed allocation blocks to be allocated.
4905          * Caller already holds s_umount sem */
4906         if (test_opt(sb, DELALLOC))
4907                 sync_filesystem(sb);
4908
4909         if (!inode)
4910                 goto out;
4911
4912         /* Update modification times of quota files when userspace can
4913          * start looking at them */
4914         handle = ext4_journal_start(inode, 1);
4915         if (IS_ERR(handle))
4916                 goto out;
4917         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4918         ext4_mark_inode_dirty(handle, inode);
4919         ext4_journal_stop(handle);
4920
4921 out:
4922         return dquot_quota_off(sb, type);
4923 }
4924
4925 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4926  * acquiring the locks... As quota files are never truncated and quota code
4927  * itself serializes the operations (and no one else should touch the files)
4928  * we don't have to be afraid of races */
4929 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4930                                size_t len, loff_t off)
4931 {
4932         struct inode *inode = sb_dqopt(sb)->files[type];
4933         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4934         int err = 0;
4935         int offset = off & (sb->s_blocksize - 1);
4936         int tocopy;
4937         size_t toread;
4938         struct buffer_head *bh;
4939         loff_t i_size = i_size_read(inode);
4940
4941         if (off > i_size)
4942                 return 0;
4943         if (off+len > i_size)
4944                 len = i_size-off;
4945         toread = len;
4946         while (toread > 0) {
4947                 tocopy = sb->s_blocksize - offset < toread ?
4948                                 sb->s_blocksize - offset : toread;
4949                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4950                 if (err)
4951                         return err;
4952                 if (!bh)        /* A hole? */
4953                         memset(data, 0, tocopy);
4954                 else
4955                         memcpy(data, bh->b_data+offset, tocopy);
4956                 brelse(bh);
4957                 offset = 0;
4958                 toread -= tocopy;
4959                 data += tocopy;
4960                 blk++;
4961         }
4962         return len;
4963 }
4964
4965 /* Write to quotafile (we know the transaction is already started and has
4966  * enough credits) */
4967 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4968                                 const char *data, size_t len, loff_t off)
4969 {
4970         struct inode *inode = sb_dqopt(sb)->files[type];
4971         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4972         int err = 0;
4973         int offset = off & (sb->s_blocksize - 1);
4974         struct buffer_head *bh;
4975         handle_t *handle = journal_current_handle();
4976
4977         if (EXT4_SB(sb)->s_journal && !handle) {
4978                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4979                         " cancelled because transaction is not started",
4980                         (unsigned long long)off, (unsigned long long)len);
4981                 return -EIO;
4982         }
4983         /*
4984          * Since we account only one data block in transaction credits,
4985          * then it is impossible to cross a block boundary.
4986          */
4987         if (sb->s_blocksize - offset < len) {
4988                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4989                         " cancelled because not block aligned",
4990                         (unsigned long long)off, (unsigned long long)len);
4991                 return -EIO;
4992         }
4993
4994         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4995         bh = ext4_bread(handle, inode, blk, 1, &err);
4996         if (!bh)
4997                 goto out;
4998         err = ext4_journal_get_write_access(handle, bh);
4999         if (err) {
5000                 brelse(bh);
5001                 goto out;
5002         }
5003         lock_buffer(bh);
5004         memcpy(bh->b_data+offset, data, len);
5005         flush_dcache_page(bh->b_page);
5006         unlock_buffer(bh);
5007         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5008         brelse(bh);
5009 out:
5010         if (err) {
5011                 mutex_unlock(&inode->i_mutex);
5012                 return err;
5013         }
5014         if (inode->i_size < off + len) {
5015                 i_size_write(inode, off + len);
5016                 EXT4_I(inode)->i_disksize = inode->i_size;
5017                 ext4_mark_inode_dirty(handle, inode);
5018         }
5019         mutex_unlock(&inode->i_mutex);
5020         return len;
5021 }
5022
5023 #endif
5024
5025 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5026                        const char *dev_name, void *data)
5027 {
5028         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5029 }
5030
5031 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5032 static inline void register_as_ext2(void)
5033 {
5034         int err = register_filesystem(&ext2_fs_type);
5035         if (err)
5036                 printk(KERN_WARNING
5037                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5038 }
5039
5040 static inline void unregister_as_ext2(void)
5041 {
5042         unregister_filesystem(&ext2_fs_type);
5043 }
5044
5045 static inline int ext2_feature_set_ok(struct super_block *sb)
5046 {
5047         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5048                 return 0;
5049         if (sb->s_flags & MS_RDONLY)
5050                 return 1;
5051         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5052                 return 0;
5053         return 1;
5054 }
5055 MODULE_ALIAS("ext2");
5056 #else
5057 static inline void register_as_ext2(void) { }
5058 static inline void unregister_as_ext2(void) { }
5059 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5060 #endif
5061
5062 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5063 static inline void register_as_ext3(void)
5064 {
5065         int err = register_filesystem(&ext3_fs_type);
5066         if (err)
5067                 printk(KERN_WARNING
5068                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5069 }
5070
5071 static inline void unregister_as_ext3(void)
5072 {
5073         unregister_filesystem(&ext3_fs_type);
5074 }
5075
5076 static inline int ext3_feature_set_ok(struct super_block *sb)
5077 {
5078         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5079                 return 0;
5080         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5081                 return 0;
5082         if (sb->s_flags & MS_RDONLY)
5083                 return 1;
5084         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5085                 return 0;
5086         return 1;
5087 }
5088 MODULE_ALIAS("ext3");
5089 #else
5090 static inline void register_as_ext3(void) { }
5091 static inline void unregister_as_ext3(void) { }
5092 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5093 #endif
5094
5095 static struct file_system_type ext4_fs_type = {
5096         .owner          = THIS_MODULE,
5097         .name           = "ext4",
5098         .mount          = ext4_mount,
5099         .kill_sb        = kill_block_super,
5100         .fs_flags       = FS_REQUIRES_DEV,
5101 };
5102
5103 static int __init ext4_init_feat_adverts(void)
5104 {
5105         struct ext4_features *ef;
5106         int ret = -ENOMEM;
5107
5108         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5109         if (!ef)
5110                 goto out;
5111
5112         ef->f_kobj.kset = ext4_kset;
5113         init_completion(&ef->f_kobj_unregister);
5114         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5115                                    "features");
5116         if (ret) {
5117                 kfree(ef);
5118                 goto out;
5119         }
5120
5121         ext4_feat = ef;
5122         ret = 0;
5123 out:
5124         return ret;
5125 }
5126
5127 static void ext4_exit_feat_adverts(void)
5128 {
5129         kobject_put(&ext4_feat->f_kobj);
5130         wait_for_completion(&ext4_feat->f_kobj_unregister);
5131         kfree(ext4_feat);
5132 }
5133
5134 /* Shared across all ext4 file systems */
5135 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5136 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5137
5138 static int __init ext4_init_fs(void)
5139 {
5140         int i, err;
5141
5142         ext4_check_flag_values();
5143
5144         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5145                 mutex_init(&ext4__aio_mutex[i]);
5146                 init_waitqueue_head(&ext4__ioend_wq[i]);
5147         }
5148
5149         err = ext4_init_pageio();
5150         if (err)
5151                 return err;
5152         err = ext4_init_system_zone();
5153         if (err)
5154                 goto out6;
5155         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5156         if (!ext4_kset)
5157                 goto out5;
5158         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5159
5160         err = ext4_init_feat_adverts();
5161         if (err)
5162                 goto out4;
5163
5164         err = ext4_init_mballoc();
5165         if (err)
5166                 goto out3;
5167
5168         err = ext4_init_xattr();
5169         if (err)
5170                 goto out2;
5171         err = init_inodecache();
5172         if (err)
5173                 goto out1;
5174         register_as_ext3();
5175         register_as_ext2();
5176         err = register_filesystem(&ext4_fs_type);
5177         if (err)
5178                 goto out;
5179
5180         ext4_li_info = NULL;
5181         mutex_init(&ext4_li_mtx);
5182         return 0;
5183 out:
5184         unregister_as_ext2();
5185         unregister_as_ext3();
5186         destroy_inodecache();
5187 out1:
5188         ext4_exit_xattr();
5189 out2:
5190         ext4_exit_mballoc();
5191 out3:
5192         ext4_exit_feat_adverts();
5193 out4:
5194         if (ext4_proc_root)
5195                 remove_proc_entry("fs/ext4", NULL);
5196         kset_unregister(ext4_kset);
5197 out5:
5198         ext4_exit_system_zone();
5199 out6:
5200         ext4_exit_pageio();
5201         return err;
5202 }
5203
5204 static void __exit ext4_exit_fs(void)
5205 {
5206         ext4_destroy_lazyinit_thread();
5207         unregister_as_ext2();
5208         unregister_as_ext3();
5209         unregister_filesystem(&ext4_fs_type);
5210         destroy_inodecache();
5211         ext4_exit_xattr();
5212         ext4_exit_mballoc();
5213         ext4_exit_feat_adverts();
5214         remove_proc_entry("fs/ext4", NULL);
5215         kset_unregister(ext4_kset);
5216         ext4_exit_system_zone();
5217         ext4_exit_pageio();
5218 }
5219
5220 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5221 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5222 MODULE_LICENSE("GPL");
5223 module_init(ext4_init_fs)
5224 module_exit(ext4_exit_fs)