fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct proc_dir_entry *ext4_proc_root;
  57 static struct kset *ext4_kset;
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ext4_features *ext4_feat;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static const char *ext4_decode_error(struct super_block *sb, int errno,
  71                                      char nbuf[16]);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static void ext4_write_super(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  95 #else
  96 #define IS_EXT2_SB(sb) (0)
  97 #endif
  98
  99
 100 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 101 static struct file_system_type ext3_fs_type = {
 102         .owner          = THIS_MODULE,
 103         .name           = "ext3",
 104         .mount          = ext4_mount,
 105         .kill_sb        = kill_block_super,
 106         .fs_flags       = FS_REQUIRES_DEV,
 107 };
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109 #else
 110 #define IS_EXT3_SB(sb) (0)
 111 #endif
 112
 113 void *ext4_kvmalloc(size_t size, gfp_t flags)
 114 {
 115         void *ret;
 116
 117         ret = kmalloc(size, flags);
 118         if (!ret)
 119                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 120         return ret;
 121 }
 122
 123 void *ext4_kvzalloc(size_t size, gfp_t flags)
 124 {
 125         void *ret;
 126
 127         ret = kzalloc(size, flags);
 128         if (!ret)
 129                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 130         return ret;
 131 }
 132
 133 void ext4_kvfree(void *ptr)
 134 {
 135         if (is_vmalloc_addr(ptr))
 136                 vfree(ptr);
 137         else
 138                 kfree(ptr);
 139
 140 }
 141
 142 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 143                                struct ext4_group_desc *bg)
 144 {
 145         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 148 }
 149
 150 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 151                                struct ext4_group_desc *bg)
 152 {
 153         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 154                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 155                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 156 }
 157
 158 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 159                               struct ext4_group_desc *bg)
 160 {
 161         return le32_to_cpu(bg->bg_inode_table_lo) |
 162                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 163                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 164 }
 165
 166 __u32 ext4_free_blks_count(struct super_block *sb,
 167                               struct ext4_group_desc *bg)
 168 {
 169         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 170                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 171                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 172 }
 173
 174 __u32 ext4_free_inodes_count(struct super_block *sb,
 175                               struct ext4_group_desc *bg)
 176 {
 177         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 178                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 179                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 180 }
 181
 182 __u32 ext4_used_dirs_count(struct super_block *sb,
 183                               struct ext4_group_desc *bg)
 184 {
 185         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 186                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 187                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 188 }
 189
 190 __u32 ext4_itable_unused_count(struct super_block *sb,
 191                               struct ext4_group_desc *bg)
 192 {
 193         return le16_to_cpu(bg->bg_itable_unused_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 196 }
 197
 198 void ext4_block_bitmap_set(struct super_block *sb,
 199                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 200 {
 201         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 204 }
 205
 206 void ext4_inode_bitmap_set(struct super_block *sb,
 207                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 208 {
 209         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 210         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 211                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 212 }
 213
 214 void ext4_inode_table_set(struct super_block *sb,
 215                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 216 {
 217         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 218         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 219                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 220 }
 221
 222 void ext4_free_blks_set(struct super_block *sb,
 223                           struct ext4_group_desc *bg, __u32 count)
 224 {
 225         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 226         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 227                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 228 }
 229
 230 void ext4_free_inodes_set(struct super_block *sb,
 231                           struct ext4_group_desc *bg, __u32 count)
 232 {
 233         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 234         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 235                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 236 }
 237
 238 void ext4_used_dirs_set(struct super_block *sb,
 239                           struct ext4_group_desc *bg, __u32 count)
 240 {
 241         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 242         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 243                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 244 }
 245
 246 void ext4_itable_unused_set(struct super_block *sb,
 247                           struct ext4_group_desc *bg, __u32 count)
 248 {
 249         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 252 }
 253
 254
 255 /* Just increment the non-pointer handle value */
 256 static handle_t *ext4_get_nojournal(void)
 257 {
 258         handle_t *handle = current->journal_info;
 259         unsigned long ref_cnt = (unsigned long)handle;
 260
 261         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 262
 263         ref_cnt++;
 264         handle = (handle_t *)ref_cnt;
 265
 266         current->journal_info = handle;
 267         return handle;
 268 }
 269
 270
 271 /* Decrement the non-pointer handle value */
 272 static void ext4_put_nojournal(handle_t *handle)
 273 {
 274         unsigned long ref_cnt = (unsigned long)handle;
 275
 276         BUG_ON(ref_cnt == 0);
 277
 278         ref_cnt--;
 279         handle = (handle_t *)ref_cnt;
 280
 281         current->journal_info = handle;
 282 }
 283
 284 /*
 285  * Wrappers for jbd2_journal_start/end.
 286  *
 287  * The only special thing we need to do here is to make sure that all
 288  * journal_end calls result in the superblock being marked dirty, so
 289  * that sync() will call the filesystem's write_super callback if
 290  * appropriate.
 291  *
 292  * To avoid j_barrier hold in userspace when a user calls freeze(),
 293  * ext4 prevents a new handle from being started by s_frozen, which
 294  * is in an upper layer.
 295  */
 296 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 297 {
 298         journal_t *journal;
 299         handle_t  *handle;
 300
 301         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 302         if (sb->s_flags & MS_RDONLY)
 303                 return ERR_PTR(-EROFS);
 304
 305         journal = EXT4_SB(sb)->s_journal;
 306         handle = ext4_journal_current_handle();
 307
 308         /*
 309          * If a handle has been started, it should be allowed to
 310          * finish, otherwise deadlock could happen between freeze
 311          * and others(e.g. truncate) due to the restart of the
 312          * journal handle if the filesystem is forzen and active
 313          * handles are not stopped.
 314          */
 315         if (!handle)
 316                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 317
 318         if (!journal)
 319                 return ext4_get_nojournal();
 320         /*
 321          * Special case here: if the journal has aborted behind our
 322          * backs (eg. EIO in the commit thread), then we still need to
 323          * take the FS itself readonly cleanly.
 324          */
 325         if (is_journal_aborted(journal)) {
 326                 ext4_abort(sb, "Detected aborted journal");
 327                 return ERR_PTR(-EROFS);
 328         }
 329         return jbd2_journal_start(journal, nblocks);
 330 }
 331
 332 /*
 333  * The only special thing we need to do here is to make sure that all
 334  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 335  * that sync() will call the filesystem's write_super callback if
 336  * appropriate.
 337  */
 338 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 339 {
 340         struct super_block *sb;
 341         int err;
 342         int rc;
 343
 344         if (!ext4_handle_valid(handle)) {
 345                 ext4_put_nojournal(handle);
 346                 return 0;
 347         }
 348         sb = handle->h_transaction->t_journal->j_private;
 349         err = handle->h_err;
 350         rc = jbd2_journal_stop(handle);
 351
 352         if (!err)
 353                 err = rc;
 354         if (err)
 355                 __ext4_std_error(sb, where, line, err);
 356         return err;
 357 }
 358
 359 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 360                                const char *err_fn, struct buffer_head *bh,
 361                                handle_t *handle, int err)
 362 {
 363         char nbuf[16];
 364         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 365
 366         BUG_ON(!ext4_handle_valid(handle));
 367
 368         if (bh)
 369                 BUFFER_TRACE(bh, "abort");
 370
 371         if (!handle->h_err)
 372                 handle->h_err = err;
 373
 374         if (is_handle_aborted(handle))
 375                 return;
 376
 377         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 378                caller, line, errstr, err_fn);
 379
 380         jbd2_journal_abort_handle(handle);
 381 }
 382
 383 static void __save_error_info(struct super_block *sb, const char *func,
 384                             unsigned int line)
 385 {
 386         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 387
 388         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 389         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 390         es->s_last_error_time = cpu_to_le32(get_seconds());
 391         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 392         es->s_last_error_line = cpu_to_le32(line);
 393         if (!es->s_first_error_time) {
 394                 es->s_first_error_time = es->s_last_error_time;
 395                 strncpy(es->s_first_error_func, func,
 396                         sizeof(es->s_first_error_func));
 397                 es->s_first_error_line = cpu_to_le32(line);
 398                 es->s_first_error_ino = es->s_last_error_ino;
 399                 es->s_first_error_block = es->s_last_error_block;
 400         }
 401         /*
 402          * Start the daily error reporting function if it hasn't been
 403          * started already
 404          */
 405         if (!es->s_error_count)
 406                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 407         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 408 }
 409
 410 static void save_error_info(struct super_block *sb, const char *func,
 411                             unsigned int line)
 412 {
 413         __save_error_info(sb, func, line);
 414         ext4_commit_super(sb, 1);
 415 }
 416
 417 /*
 418  * The del_gendisk() function uninitializes the disk-specific data
 419  * structures, including the bdi structure, without telling anyone
 420  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 421  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 422  * This is a kludge to prevent these oops until we can put in a proper
 423  * hook in del_gendisk() to inform the VFS and file system layers.
 424  */
 425 static int block_device_ejected(struct super_block *sb)
 426 {
 427         struct inode *bd_inode = sb->s_bdev->bd_inode;
 428         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 429
 430         return bdi->dev == NULL;
 431 }
 432
 433
 434 /* Deal with the reporting of failure conditions on a filesystem such as
 435  * inconsistencies detected or read IO failures.
 436  *
 437  * On ext2, we can store the error state of the filesystem in the
 438  * superblock.  That is not possible on ext4, because we may have other
 439  * write ordering constraints on the superblock which prevent us from
 440  * writing it out straight away; and given that the journal is about to
 441  * be aborted, we can't rely on the current, or future, transactions to
 442  * write out the superblock safely.
 443  *
 444  * We'll just use the jbd2_journal_abort() error code to record an error in
 445  * the journal instead.  On recovery, the journal will complain about
 446  * that error until we've noted it down and cleared it.
 447  */
 448
 449 static void ext4_handle_error(struct super_block *sb)
 450 {
 451         if (sb->s_flags & MS_RDONLY)
 452                 return;
 453
 454         if (!test_opt(sb, ERRORS_CONT)) {
 455                 journal_t *journal = EXT4_SB(sb)->s_journal;
 456
 457                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 458                 if (journal)
 459                         jbd2_journal_abort(journal, -EIO);
 460         }
 461         if (test_opt(sb, ERRORS_RO)) {
 462                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 463                 sb->s_flags |= MS_RDONLY;
 464         }
 465         if (test_opt(sb, ERRORS_PANIC))
 466                 panic("EXT4-fs (device %s): panic forced after error\n",
 467                         sb->s_id);
 468 }
 469
 470 void __ext4_error(struct super_block *sb, const char *function,
 471                   unsigned int line, const char *fmt, ...)
 472 {
 473         struct va_format vaf;
 474         va_list args;
 475
 476         va_start(args, fmt);
 477         vaf.fmt = fmt;
 478         vaf.va = &args;
 479         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 480                sb->s_id, function, line, current->comm, &vaf);
 481         va_end(args);
 482
 483         ext4_handle_error(sb);
 484 }
 485
 486 void ext4_error_inode(struct inode *inode, const char *function,
 487                       unsigned int line, ext4_fsblk_t block,
 488                       const char *fmt, ...)
 489 {
 490         va_list args;
 491         struct va_format vaf;
 492         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 493
 494         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 495         es->s_last_error_block = cpu_to_le64(block);
 496         save_error_info(inode->i_sb, function, line);
 497         va_start(args, fmt);
 498         vaf.fmt = fmt;
 499         vaf.va = &args;
 500         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 501                inode->i_sb->s_id, function, line, inode->i_ino);
 502         if (block)
 503                 printk(KERN_CONT "block %llu: ", block);
 504         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 505         va_end(args);
 506
 507         ext4_handle_error(inode->i_sb);
 508 }
 509
 510 void ext4_error_file(struct file *file, const char *function,
 511                      unsigned int line, ext4_fsblk_t block,
 512                      const char *fmt, ...)
 513 {
 514         va_list args;
 515         struct va_format vaf;
 516         struct ext4_super_block *es;
 517         struct inode *inode = file->f_dentry->d_inode;
 518         char pathname[80], *path;
 519
 520         es = EXT4_SB(inode->i_sb)->s_es;
 521         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 522         save_error_info(inode->i_sb, function, line);
 523         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 524         if (IS_ERR(path))
 525                 path = "(unknown)";
 526         printk(KERN_CRIT
 527                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 528                inode->i_sb->s_id, function, line, inode->i_ino);
 529         if (block)
 530                 printk(KERN_CONT "block %llu: ", block);
 531         va_start(args, fmt);
 532         vaf.fmt = fmt;
 533         vaf.va = &args;
 534         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 535         va_end(args);
 536
 537         ext4_handle_error(inode->i_sb);
 538 }
 539
 540 static const char *ext4_decode_error(struct super_block *sb, int errno,
 541                                      char nbuf[16])
 542 {
 543         char *errstr = NULL;
 544
 545         switch (errno) {
 546         case -EIO:
 547                 errstr = "IO failure";
 548                 break;
 549         case -ENOMEM:
 550                 errstr = "Out of memory";
 551                 break;
 552         case -EROFS:
 553                 if (!sb || (EXT4_SB(sb)->s_journal &&
 554                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 555                         errstr = "Journal has aborted";
 556                 else
 557                         errstr = "Readonly filesystem";
 558                 break;
 559         default:
 560                 /* If the caller passed in an extra buffer for unknown
 561                  * errors, textualise them now.  Else we just return
 562                  * NULL. */
 563                 if (nbuf) {
 564                         /* Check for truncated error codes... */
 565                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 566                                 errstr = nbuf;
 567                 }
 568                 break;
 569         }
 570
 571         return errstr;
 572 }
 573
 574 /* __ext4_std_error decodes expected errors from journaling functions
 575  * automatically and invokes the appropriate error response.  */
 576
 577 void __ext4_std_error(struct super_block *sb, const char *function,
 578                       unsigned int line, int errno)
 579 {
 580         char nbuf[16];
 581         const char *errstr;
 582
 583         /* Special case: if the error is EROFS, and we're not already
 584          * inside a transaction, then there's really no point in logging
 585          * an error. */
 586         if (errno == -EROFS && journal_current_handle() == NULL &&
 587             (sb->s_flags & MS_RDONLY))
 588                 return;
 589
 590         errstr = ext4_decode_error(sb, errno, nbuf);
 591         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 592                sb->s_id, function, line, errstr);
 593         save_error_info(sb, function, line);
 594
 595         ext4_handle_error(sb);
 596 }
 597
 598 /*
 599  * ext4_abort is a much stronger failure handler than ext4_error.  The
 600  * abort function may be used to deal with unrecoverable failures such
 601  * as journal IO errors or ENOMEM at a critical moment in log management.
 602  *
 603  * We unconditionally force the filesystem into an ABORT|READONLY state,
 604  * unless the error response on the fs has been set to panic in which
 605  * case we take the easy way out and panic immediately.
 606  */
 607
 608 void __ext4_abort(struct super_block *sb, const char *function,
 609                 unsigned int line, const char *fmt, ...)
 610 {
 611         va_list args;
 612
 613         save_error_info(sb, function, line);
 614         va_start(args, fmt);
 615         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 616                function, line);
 617         vprintk(fmt, args);
 618         printk("\n");
 619         va_end(args);
 620
 621         if ((sb->s_flags & MS_RDONLY) == 0) {
 622                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 623                 sb->s_flags |= MS_RDONLY;
 624                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 625                 if (EXT4_SB(sb)->s_journal)
 626                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 627                 save_error_info(sb, function, line);
 628         }
 629         if (test_opt(sb, ERRORS_PANIC))
 630                 panic("EXT4-fs panic from previous error\n");
 631 }
 632
 633 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 634 {
 635         struct va_format vaf;
 636         va_list args;
 637
 638         va_start(args, fmt);
 639         vaf.fmt = fmt;
 640         vaf.va = &args;
 641         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 642         va_end(args);
 643 }
 644
 645 void __ext4_warning(struct super_block *sb, const char *function,
 646                     unsigned int line, const char *fmt, ...)
 647 {
 648         struct va_format vaf;
 649         va_list args;
 650
 651         va_start(args, fmt);
 652         vaf.fmt = fmt;
 653         vaf.va = &args;
 654         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 655                sb->s_id, function, line, &vaf);
 656         va_end(args);
 657 }
 658
 659 void __ext4_grp_locked_error(const char *function, unsigned int line,
 660                              struct super_block *sb, ext4_group_t grp,
 661                              unsigned long ino, ext4_fsblk_t block,
 662                              const char *fmt, ...)
 663 __releases(bitlock)
 664 __acquires(bitlock)
 665 {
 666         struct va_format vaf;
 667         va_list args;
 668         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 669
 670         es->s_last_error_ino = cpu_to_le32(ino);
 671         es->s_last_error_block = cpu_to_le64(block);
 672         __save_error_info(sb, function, line);
 673
 674         va_start(args, fmt);
 675
 676         vaf.fmt = fmt;
 677         vaf.va = &args;
 678         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 679                sb->s_id, function, line, grp);
 680         if (ino)
 681                 printk(KERN_CONT "inode %lu: ", ino);
 682         if (block)
 683                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 684         printk(KERN_CONT "%pV\n", &vaf);
 685         va_end(args);
 686
 687         if (test_opt(sb, ERRORS_CONT)) {
 688                 ext4_commit_super(sb, 0);
 689                 return;
 690         }
 691
 692         ext4_unlock_group(sb, grp);
 693         ext4_handle_error(sb);
 694         /*
 695          * We only get here in the ERRORS_RO case; relocking the group
 696          * may be dangerous, but nothing bad will happen since the
 697          * filesystem will have already been marked read/only and the
 698          * journal has been aborted.  We return 1 as a hint to callers
 699          * who might what to use the return value from
 700          * ext4_grp_locked_error() to distinguish between the
 701          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 702          * aggressively from the ext4 function in question, with a
 703          * more appropriate error code.
 704          */
 705         ext4_lock_group(sb, grp);
 706         return;
 707 }
 708
 709 void ext4_update_dynamic_rev(struct super_block *sb)
 710 {
 711         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 712
 713         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 714                 return;
 715
 716         ext4_warning(sb,
 717                      "updating to rev %d because of new feature flag, "
 718                      "running e2fsck is recommended",
 719                      EXT4_DYNAMIC_REV);
 720
 721         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 722         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 723         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 724         /* leave es->s_feature_*compat flags alone */
 725         /* es->s_uuid will be set by e2fsck if empty */
 726
 727         /*
 728          * The rest of the superblock fields should be zero, and if not it
 729          * means they are likely already in use, so leave them alone.  We
 730          * can leave it up to e2fsck to clean up any inconsistencies there.
 731          */
 732 }
 733
 734 /*
 735  * Open the external journal device
 736  */
 737 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 738 {
 739         struct block_device *bdev;
 740         char b[BDEVNAME_SIZE];
 741
 742         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 743         if (IS_ERR(bdev))
 744                 goto fail;
 745         return bdev;
 746
 747 fail:
 748         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 749                         __bdevname(dev, b), PTR_ERR(bdev));
 750         return NULL;
 751 }
 752
 753 /*
 754  * Release the journal device
 755  */
 756 static int ext4_blkdev_put(struct block_device *bdev)
 757 {
 758         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 759 }
 760
 761 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 762 {
 763         struct block_device *bdev;
 764         int ret = -ENODEV;
 765
 766         bdev = sbi->journal_bdev;
 767         if (bdev) {
 768                 ret = ext4_blkdev_put(bdev);
 769                 sbi->journal_bdev = NULL;
 770         }
 771         return ret;
 772 }
 773
 774 static inline struct inode *orphan_list_entry(struct list_head *l)
 775 {
 776         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 777 }
 778
 779 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 780 {
 781         struct list_head *l;
 782
 783         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 784                  le32_to_cpu(sbi->s_es->s_last_orphan));
 785
 786         printk(KERN_ERR "sb_info orphan list:\n");
 787         list_for_each(l, &sbi->s_orphan) {
 788                 struct inode *inode = orphan_list_entry(l);
 789                 printk(KERN_ERR "  "
 790                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 791                        inode->i_sb->s_id, inode->i_ino, inode,
 792                        inode->i_mode, inode->i_nlink,
 793                        NEXT_ORPHAN(inode));
 794         }
 795 }
 796
 797 static void ext4_put_super(struct super_block *sb)
 798 {
 799         struct ext4_sb_info *sbi = EXT4_SB(sb);
 800         struct ext4_super_block *es = sbi->s_es;
 801         int i, err;
 802
 803         ext4_unregister_li_request(sb);
 804         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 805
 806         flush_workqueue(sbi->dio_unwritten_wq);
 807         destroy_workqueue(sbi->dio_unwritten_wq);
 808
 809         lock_super(sb);
 810         if (sb->s_dirt)
 811                 ext4_commit_super(sb, 1);
 812
 813         if (sbi->s_journal) {
 814                 err = jbd2_journal_destroy(sbi->s_journal);
 815                 sbi->s_journal = NULL;
 816                 if (err < 0)
 817                         ext4_abort(sb, "Couldn't clean up the journal");
 818         }
 819
 820         del_timer(&sbi->s_err_report);
 821         ext4_release_system_zone(sb);
 822         ext4_mb_release(sb);
 823         ext4_ext_release(sb);
 824         ext4_xattr_put_super(sb);
 825
 826         if (!(sb->s_flags & MS_RDONLY)) {
 827                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 828                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 829                 ext4_commit_super(sb, 1);
 830         }
 831         if (sbi->s_proc) {
 832                 remove_proc_entry(sb->s_id, ext4_proc_root);
 833         }
 834         kobject_del(&sbi->s_kobj);
 835
 836         for (i = 0; i < sbi->s_gdb_count; i++)
 837                 brelse(sbi->s_group_desc[i]);
 838         ext4_kvfree(sbi->s_group_desc);
 839         ext4_kvfree(sbi->s_flex_groups);
 840         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 841         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 842         percpu_counter_destroy(&sbi->s_dirs_counter);
 843         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 844         brelse(sbi->s_sbh);
 845 #ifdef CONFIG_QUOTA
 846         for (i = 0; i < MAXQUOTAS; i++)
 847                 kfree(sbi->s_qf_names[i]);
 848 #endif
 849
 850         /* Debugging code just in case the in-memory inode orphan list
 851          * isn't empty.  The on-disk one can be non-empty if we've
 852          * detected an error and taken the fs readonly, but the
 853          * in-memory list had better be clean by this point. */
 854         if (!list_empty(&sbi->s_orphan))
 855                 dump_orphan_list(sb, sbi);
 856         J_ASSERT(list_empty(&sbi->s_orphan));
 857
 858         invalidate_bdev(sb->s_bdev);
 859         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 860                 /*
 861                  * Invalidate the journal device's buffers.  We don't want them
 862                  * floating about in memory - the physical journal device may
 863                  * hotswapped, and it breaks the `ro-after' testing code.
 864                  */
 865                 sync_blockdev(sbi->journal_bdev);
 866                 invalidate_bdev(sbi->journal_bdev);
 867                 ext4_blkdev_remove(sbi);
 868         }
 869         if (sbi->s_mmp_tsk)
 870                 kthread_stop(sbi->s_mmp_tsk);
 871         sb->s_fs_info = NULL;
 872         /*
 873          * Now that we are completely done shutting down the
 874          * superblock, we need to actually destroy the kobject.
 875          */
 876         unlock_super(sb);
 877         kobject_put(&sbi->s_kobj);
 878         wait_for_completion(&sbi->s_kobj_unregister);
 879         kfree(sbi->s_blockgroup_lock);
 880         kfree(sbi);
 881 }
 882
 883 static struct kmem_cache *ext4_inode_cachep;
 884
 885 /*
 886  * Called inside transaction, so use GFP_NOFS
 887  */
 888 static struct inode *ext4_alloc_inode(struct super_block *sb)
 889 {
 890         struct ext4_inode_info *ei;
 891
 892         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 893         if (!ei)
 894                 return NULL;
 895
 896         ei->vfs_inode.i_version = 1;
 897         ei->vfs_inode.i_data.writeback_index = 0;
 898         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 899         INIT_LIST_HEAD(&ei->i_prealloc_list);
 900         spin_lock_init(&ei->i_prealloc_lock);
 901         ei->i_reserved_data_blocks = 0;
 902         ei->i_reserved_meta_blocks = 0;
 903         ei->i_allocated_meta_blocks = 0;
 904         ei->i_da_metadata_calc_len = 0;
 905         spin_lock_init(&(ei->i_block_reservation_lock));
 906 #ifdef CONFIG_QUOTA
 907         ei->i_reserved_quota = 0;
 908 #endif
 909         ei->jinode = NULL;
 910         INIT_LIST_HEAD(&ei->i_completed_io_list);
 911         spin_lock_init(&ei->i_completed_io_lock);
 912         ei->cur_aio_dio = NULL;
 913         ei->i_sync_tid = 0;
 914         ei->i_datasync_tid = 0;
 915         atomic_set(&ei->i_ioend_count, 0);
 916         atomic_set(&ei->i_aiodio_unwritten, 0);
 917
 918         return &ei->vfs_inode;
 919 }
 920
 921 static int ext4_drop_inode(struct inode *inode)
 922 {
 923         int drop = generic_drop_inode(inode);
 924
 925         trace_ext4_drop_inode(inode, drop);
 926         return drop;
 927 }
 928
 929 static void ext4_i_callback(struct rcu_head *head)
 930 {
 931         struct inode *inode = container_of(head, struct inode, i_rcu);
 932         INIT_LIST_HEAD(&inode->i_dentry);
 933         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 934 }
 935
 936 static void ext4_destroy_inode(struct inode *inode)
 937 {
 938         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 939                 ext4_msg(inode->i_sb, KERN_ERR,
 940                          "Inode %lu (%p): orphan list check failed!",
 941                          inode->i_ino, EXT4_I(inode));
 942                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 943                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 944                                 true);
 945                 dump_stack();
 946         }
 947         call_rcu(&inode->i_rcu, ext4_i_callback);
 948 }
 949
 950 static void init_once(void *foo)
 951 {
 952         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 953
 954         INIT_LIST_HEAD(&ei->i_orphan);
 955 #ifdef CONFIG_EXT4_FS_XATTR
 956         init_rwsem(&ei->xattr_sem);
 957 #endif
 958         init_rwsem(&ei->i_data_sem);
 959         inode_init_once(&ei->vfs_inode);
 960 }
 961
 962 static int init_inodecache(void)
 963 {
 964         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 965                                              sizeof(struct ext4_inode_info),
 966                                              0, (SLAB_RECLAIM_ACCOUNT|
 967                                                 SLAB_MEM_SPREAD),
 968                                              init_once);
 969         if (ext4_inode_cachep == NULL)
 970                 return -ENOMEM;
 971         return 0;
 972 }
 973
 974 static void destroy_inodecache(void)
 975 {
 976         kmem_cache_destroy(ext4_inode_cachep);
 977 }
 978
 979 void ext4_clear_inode(struct inode *inode)
 980 {
 981         invalidate_inode_buffers(inode);
 982         end_writeback(inode);
 983         dquot_drop(inode);
 984         ext4_discard_preallocations(inode);
 985         if (EXT4_I(inode)->jinode) {
 986                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 987                                                EXT4_I(inode)->jinode);
 988                 jbd2_free_inode(EXT4_I(inode)->jinode);
 989                 EXT4_I(inode)->jinode = NULL;
 990         }
 991 }
 992
 993 static inline void ext4_show_quota_options(struct seq_file *seq,
 994                                            struct super_block *sb)
 995 {
 996 #if defined(CONFIG_QUOTA)
 997         struct ext4_sb_info *sbi = EXT4_SB(sb);
 998
 999         if (sbi->s_jquota_fmt) {
1000                 char *fmtname = "";
1001
1002                 switch (sbi->s_jquota_fmt) {
1003                 case QFMT_VFS_OLD:
1004                         fmtname = "vfsold";
1005                         break;
1006                 case QFMT_VFS_V0:
1007                         fmtname = "vfsv0";
1008                         break;
1009                 case QFMT_VFS_V1:
1010                         fmtname = "vfsv1";
1011                         break;
1012                 }
1013                 seq_printf(seq, ",jqfmt=%s", fmtname);
1014         }
1015
1016         if (sbi->s_qf_names[USRQUOTA])
1017                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1018
1019         if (sbi->s_qf_names[GRPQUOTA])
1020                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1021
1022         if (test_opt(sb, USRQUOTA))
1023                 seq_puts(seq, ",usrquota");
1024
1025         if (test_opt(sb, GRPQUOTA))
1026                 seq_puts(seq, ",grpquota");
1027 #endif
1028 }
1029
1030 /*
1031  * Show an option if
1032  *  - it's set to a non-default value OR
1033  *  - if the per-sb default is different from the global default
1034  */
1035 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1036 {
1037         int def_errors;
1038         unsigned long def_mount_opts;
1039         struct super_block *sb = vfs->mnt_sb;
1040         struct ext4_sb_info *sbi = EXT4_SB(sb);
1041         struct ext4_super_block *es = sbi->s_es;
1042
1043         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1044         def_errors     = le16_to_cpu(es->s_errors);
1045
1046         if (sbi->s_sb_block != 1)
1047                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1048         if (test_opt(sb, MINIX_DF))
1049                 seq_puts(seq, ",minixdf");
1050         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1051                 seq_puts(seq, ",grpid");
1052         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1053                 seq_puts(seq, ",nogrpid");
1054         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1055             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1056                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1057         }
1058         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1059             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1060                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1061         }
1062         if (test_opt(sb, ERRORS_RO)) {
1063                 if (def_errors == EXT4_ERRORS_PANIC ||
1064                     def_errors == EXT4_ERRORS_CONTINUE) {
1065                         seq_puts(seq, ",errors=remount-ro");
1066                 }
1067         }
1068         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1069                 seq_puts(seq, ",errors=continue");
1070         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1071                 seq_puts(seq, ",errors=panic");
1072         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1073                 seq_puts(seq, ",nouid32");
1074         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1075                 seq_puts(seq, ",debug");
1076         if (test_opt(sb, OLDALLOC))
1077                 seq_puts(seq, ",oldalloc");
1078 #ifdef CONFIG_EXT4_FS_XATTR
1079         if (test_opt(sb, XATTR_USER))
1080                 seq_puts(seq, ",user_xattr");
1081         if (!test_opt(sb, XATTR_USER))
1082                 seq_puts(seq, ",nouser_xattr");
1083 #endif
1084 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1085         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1086                 seq_puts(seq, ",acl");
1087         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1088                 seq_puts(seq, ",noacl");
1089 #endif
1090         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1091                 seq_printf(seq, ",commit=%u",
1092                            (unsigned) (sbi->s_commit_interval / HZ));
1093         }
1094         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1095                 seq_printf(seq, ",min_batch_time=%u",
1096                            (unsigned) sbi->s_min_batch_time);
1097         }
1098         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1099                 seq_printf(seq, ",max_batch_time=%u",
1100                            (unsigned) sbi->s_min_batch_time);
1101         }
1102
1103         /*
1104          * We're changing the default of barrier mount option, so
1105          * let's always display its mount state so it's clear what its
1106          * status is.
1107          */
1108         seq_puts(seq, ",barrier=");
1109         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1110         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1111                 seq_puts(seq, ",journal_async_commit");
1112         else if (test_opt(sb, JOURNAL_CHECKSUM))
1113                 seq_puts(seq, ",journal_checksum");
1114         if (test_opt(sb, I_VERSION))
1115                 seq_puts(seq, ",i_version");
1116         if (!test_opt(sb, DELALLOC) &&
1117             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1118                 seq_puts(seq, ",nodelalloc");
1119
1120         if (!test_opt(sb, MBLK_IO_SUBMIT))
1121                 seq_puts(seq, ",nomblk_io_submit");
1122         if (sbi->s_stripe)
1123                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1124         /*
1125          * journal mode get enabled in different ways
1126          * So just print the value even if we didn't specify it
1127          */
1128         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1129                 seq_puts(seq, ",data=journal");
1130         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1131                 seq_puts(seq, ",data=ordered");
1132         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1133                 seq_puts(seq, ",data=writeback");
1134
1135         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1136                 seq_printf(seq, ",inode_readahead_blks=%u",
1137                            sbi->s_inode_readahead_blks);
1138
1139         if (test_opt(sb, DATA_ERR_ABORT))
1140                 seq_puts(seq, ",data_err=abort");
1141
1142         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1143                 seq_puts(seq, ",noauto_da_alloc");
1144
1145         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1146                 seq_puts(seq, ",discard");
1147
1148         if (test_opt(sb, NOLOAD))
1149                 seq_puts(seq, ",norecovery");
1150
1151         if (test_opt(sb, DIOREAD_NOLOCK))
1152                 seq_puts(seq, ",dioread_nolock");
1153
1154         if (test_opt(sb, BLOCK_VALIDITY) &&
1155             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1156                 seq_puts(seq, ",block_validity");
1157
1158         if (!test_opt(sb, INIT_INODE_TABLE))
1159                 seq_puts(seq, ",noinit_inode_table");
1160         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1161                 seq_printf(seq, ",init_inode_table=%u",
1162                            (unsigned) sbi->s_li_wait_mult);
1163
1164         ext4_show_quota_options(seq, sb);
1165
1166         return 0;
1167 }
1168
1169 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1170                                         u64 ino, u32 generation)
1171 {
1172         struct inode *inode;
1173
1174         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1175                 return ERR_PTR(-ESTALE);
1176         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1177                 return ERR_PTR(-ESTALE);
1178
1179         /* iget isn't really right if the inode is currently unallocated!!
1180          *
1181          * ext4_read_inode will return a bad_inode if the inode had been
1182          * deleted, so we should be safe.
1183          *
1184          * Currently we don't know the generation for parent directory, so
1185          * a generation of 0 means "accept any"
1186          */
1187         inode = ext4_iget(sb, ino);
1188         if (IS_ERR(inode))
1189                 return ERR_CAST(inode);
1190         if (generation && inode->i_generation != generation) {
1191                 iput(inode);
1192                 return ERR_PTR(-ESTALE);
1193         }
1194
1195         return inode;
1196 }
1197
1198 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1199                                         int fh_len, int fh_type)
1200 {
1201         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1202                                     ext4_nfs_get_inode);
1203 }
1204
1205 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1206                                         int fh_len, int fh_type)
1207 {
1208         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1209                                     ext4_nfs_get_inode);
1210 }
1211
1212 /*
1213  * Try to release metadata pages (indirect blocks, directories) which are
1214  * mapped via the block device.  Since these pages could have journal heads
1215  * which would prevent try_to_free_buffers() from freeing them, we must use
1216  * jbd2 layer's try_to_free_buffers() function to release them.
1217  */
1218 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1219                                  gfp_t wait)
1220 {
1221         journal_t *journal = EXT4_SB(sb)->s_journal;
1222
1223         WARN_ON(PageChecked(page));
1224         if (!page_has_buffers(page))
1225                 return 0;
1226         if (journal)
1227                 return jbd2_journal_try_to_free_buffers(journal, page,
1228                                                         wait & ~__GFP_WAIT);
1229         return try_to_free_buffers(page);
1230 }
1231
1232 #ifdef CONFIG_QUOTA
1233 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1234 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1235
1236 static int ext4_write_dquot(struct dquot *dquot);
1237 static int ext4_acquire_dquot(struct dquot *dquot);
1238 static int ext4_release_dquot(struct dquot *dquot);
1239 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1240 static int ext4_write_info(struct super_block *sb, int type);
1241 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1242                          struct path *path);
1243 static int ext4_quota_off(struct super_block *sb, int type);
1244 static int ext4_quota_on_mount(struct super_block *sb, int type);
1245 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1246                                size_t len, loff_t off);
1247 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1248                                 const char *data, size_t len, loff_t off);
1249
1250 static const struct dquot_operations ext4_quota_operations = {
1251         .get_reserved_space = ext4_get_reserved_space,
1252         .write_dquot    = ext4_write_dquot,
1253         .acquire_dquot  = ext4_acquire_dquot,
1254         .release_dquot  = ext4_release_dquot,
1255         .mark_dirty     = ext4_mark_dquot_dirty,
1256         .write_info     = ext4_write_info,
1257         .alloc_dquot    = dquot_alloc,
1258         .destroy_dquot  = dquot_destroy,
1259 };
1260
1261 static const struct quotactl_ops ext4_qctl_operations = {
1262         .quota_on       = ext4_quota_on,
1263         .quota_off      = ext4_quota_off,
1264         .quota_sync     = dquot_quota_sync,
1265         .get_info       = dquot_get_dqinfo,
1266         .set_info       = dquot_set_dqinfo,
1267         .get_dqblk      = dquot_get_dqblk,
1268         .set_dqblk      = dquot_set_dqblk
1269 };
1270 #endif
1271
1272 static const struct super_operations ext4_sops = {
1273         .alloc_inode    = ext4_alloc_inode,
1274         .destroy_inode  = ext4_destroy_inode,
1275         .write_inode    = ext4_write_inode,
1276         .dirty_inode    = ext4_dirty_inode,
1277         .drop_inode     = ext4_drop_inode,
1278         .evict_inode    = ext4_evict_inode,
1279         .put_super      = ext4_put_super,
1280         .sync_fs        = ext4_sync_fs,
1281         .freeze_fs      = ext4_freeze,
1282         .unfreeze_fs    = ext4_unfreeze,
1283         .statfs         = ext4_statfs,
1284         .remount_fs     = ext4_remount,
1285         .show_options   = ext4_show_options,
1286 #ifdef CONFIG_QUOTA
1287         .quota_read     = ext4_quota_read,
1288         .quota_write    = ext4_quota_write,
1289 #endif
1290         .bdev_try_to_free_page = bdev_try_to_free_page,
1291 };
1292
1293 static const struct super_operations ext4_nojournal_sops = {
1294         .alloc_inode    = ext4_alloc_inode,
1295         .destroy_inode  = ext4_destroy_inode,
1296         .write_inode    = ext4_write_inode,
1297         .dirty_inode    = ext4_dirty_inode,
1298         .drop_inode     = ext4_drop_inode,
1299         .evict_inode    = ext4_evict_inode,
1300         .write_super    = ext4_write_super,
1301         .put_super      = ext4_put_super,
1302         .statfs         = ext4_statfs,
1303         .remount_fs     = ext4_remount,
1304         .show_options   = ext4_show_options,
1305 #ifdef CONFIG_QUOTA
1306         .quota_read     = ext4_quota_read,
1307         .quota_write    = ext4_quota_write,
1308 #endif
1309         .bdev_try_to_free_page = bdev_try_to_free_page,
1310 };
1311
1312 static const struct export_operations ext4_export_ops = {
1313         .fh_to_dentry = ext4_fh_to_dentry,
1314         .fh_to_parent = ext4_fh_to_parent,
1315         .get_parent = ext4_get_parent,
1316 };
1317
1318 enum {
1319         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1320         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1321         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1322         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1323         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1324         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1325         Opt_journal_update, Opt_journal_dev,
1326         Opt_journal_checksum, Opt_journal_async_commit,
1327         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1328         Opt_data_err_abort, Opt_data_err_ignore,
1329         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1330         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1331         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1332         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1333         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1334         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1335         Opt_inode_readahead_blks, Opt_journal_ioprio,
1336         Opt_dioread_nolock, Opt_dioread_lock,
1337         Opt_discard, Opt_nodiscard,
1338         Opt_init_inode_table, Opt_noinit_inode_table,
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_inode_table, "init_itable=%u"},
1412         {Opt_init_inode_table, "init_itable"},
1413         {Opt_noinit_inode_table, "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                         set_opt(sb, OLDALLOC);
1587                         break;
1588                 case Opt_orlov:
1589                         clear_opt(sb, OLDALLOC);
1590                         break;
1591 #ifdef CONFIG_EXT4_FS_XATTR
1592                 case Opt_user_xattr:
1593                         set_opt(sb, XATTR_USER);
1594                         break;
1595                 case Opt_nouser_xattr:
1596                         clear_opt(sb, XATTR_USER);
1597                         break;
1598 #else
1599                 case Opt_user_xattr:
1600                 case Opt_nouser_xattr:
1601                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1602                         break;
1603 #endif
1604 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1605                 case Opt_acl:
1606                         set_opt(sb, POSIX_ACL);
1607                         break;
1608                 case Opt_noacl:
1609                         clear_opt(sb, POSIX_ACL);
1610                         break;
1611 #else
1612                 case Opt_acl:
1613                 case Opt_noacl:
1614                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1615                         break;
1616 #endif
1617                 case Opt_journal_update:
1618                         /* @@@ FIXME */
1619                         /* Eventually we will want to be able to create
1620                            a journal file here.  For now, only allow the
1621                            user to specify an existing inode to be the
1622                            journal file. */
1623                         if (is_remount) {
1624                                 ext4_msg(sb, KERN_ERR,
1625                                          "Cannot specify journal on remount");
1626                                 return 0;
1627                         }
1628                         set_opt(sb, UPDATE_JOURNAL);
1629                         break;
1630                 case Opt_journal_dev:
1631                         if (is_remount) {
1632                                 ext4_msg(sb, KERN_ERR,
1633                                         "Cannot specify journal on remount");
1634                                 return 0;
1635                         }
1636                         if (match_int(&args[0], &option))
1637                                 return 0;
1638                         *journal_devnum = option;
1639                         break;
1640                 case Opt_journal_checksum:
1641                         set_opt(sb, JOURNAL_CHECKSUM);
1642                         break;
1643                 case Opt_journal_async_commit:
1644                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1645                         set_opt(sb, JOURNAL_CHECKSUM);
1646                         break;
1647                 case Opt_noload:
1648                         set_opt(sb, NOLOAD);
1649                         break;
1650                 case Opt_commit:
1651                         if (match_int(&args[0], &option))
1652                                 return 0;
1653                         if (option < 0)
1654                                 return 0;
1655                         if (option == 0)
1656                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1657                         sbi->s_commit_interval = HZ * option;
1658                         break;
1659                 case Opt_max_batch_time:
1660                         if (match_int(&args[0], &option))
1661                                 return 0;
1662                         if (option < 0)
1663                                 return 0;
1664                         if (option == 0)
1665                                 option = EXT4_DEF_MAX_BATCH_TIME;
1666                         sbi->s_max_batch_time = option;
1667                         break;
1668                 case Opt_min_batch_time:
1669                         if (match_int(&args[0], &option))
1670                                 return 0;
1671                         if (option < 0)
1672                                 return 0;
1673                         sbi->s_min_batch_time = option;
1674                         break;
1675                 case Opt_data_journal:
1676                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1677                         goto datacheck;
1678                 case Opt_data_ordered:
1679                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1680                         goto datacheck;
1681                 case Opt_data_writeback:
1682                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1683                 datacheck:
1684                         if (is_remount) {
1685                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1686                                         ext4_msg(sb, KERN_ERR,
1687                                                 "Cannot change data mode on remount");
1688                                         return 0;
1689                                 }
1690                         } else {
1691                                 clear_opt(sb, DATA_FLAGS);
1692                                 sbi->s_mount_opt |= data_opt;
1693                         }
1694                         break;
1695                 case Opt_data_err_abort:
1696                         set_opt(sb, DATA_ERR_ABORT);
1697                         break;
1698                 case Opt_data_err_ignore:
1699                         clear_opt(sb, DATA_ERR_ABORT);
1700                         break;
1701 #ifdef CONFIG_QUOTA
1702                 case Opt_usrjquota:
1703                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1704                                 return 0;
1705                         break;
1706                 case Opt_grpjquota:
1707                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1708                                 return 0;
1709                         break;
1710                 case Opt_offusrjquota:
1711                         if (!clear_qf_name(sb, USRQUOTA))
1712                                 return 0;
1713                         break;
1714                 case Opt_offgrpjquota:
1715                         if (!clear_qf_name(sb, GRPQUOTA))
1716                                 return 0;
1717                         break;
1718
1719                 case Opt_jqfmt_vfsold:
1720                         qfmt = QFMT_VFS_OLD;
1721                         goto set_qf_format;
1722                 case Opt_jqfmt_vfsv0:
1723                         qfmt = QFMT_VFS_V0;
1724                         goto set_qf_format;
1725                 case Opt_jqfmt_vfsv1:
1726                         qfmt = QFMT_VFS_V1;
1727 set_qf_format:
1728                         if (sb_any_quota_loaded(sb) &&
1729                             sbi->s_jquota_fmt != qfmt) {
1730                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1731                                         "journaled quota options when "
1732                                         "quota turned on");
1733                                 return 0;
1734                         }
1735                         sbi->s_jquota_fmt = qfmt;
1736                         break;
1737                 case Opt_quota:
1738                 case Opt_usrquota:
1739                         set_opt(sb, QUOTA);
1740                         set_opt(sb, USRQUOTA);
1741                         break;
1742                 case Opt_grpquota:
1743                         set_opt(sb, QUOTA);
1744                         set_opt(sb, GRPQUOTA);
1745                         break;
1746                 case Opt_noquota:
1747                         if (sb_any_quota_loaded(sb)) {
1748                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1749                                         "options when quota turned on");
1750                                 return 0;
1751                         }
1752                         clear_opt(sb, QUOTA);
1753                         clear_opt(sb, USRQUOTA);
1754                         clear_opt(sb, GRPQUOTA);
1755                         break;
1756 #else
1757                 case Opt_quota:
1758                 case Opt_usrquota:
1759                 case Opt_grpquota:
1760                         ext4_msg(sb, KERN_ERR,
1761                                 "quota options not supported");
1762                         break;
1763                 case Opt_usrjquota:
1764                 case Opt_grpjquota:
1765                 case Opt_offusrjquota:
1766                 case Opt_offgrpjquota:
1767                 case Opt_jqfmt_vfsold:
1768                 case Opt_jqfmt_vfsv0:
1769                 case Opt_jqfmt_vfsv1:
1770                         ext4_msg(sb, KERN_ERR,
1771                                 "journaled quota options not supported");
1772                         break;
1773                 case Opt_noquota:
1774                         break;
1775 #endif
1776                 case Opt_abort:
1777                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1778                         break;
1779                 case Opt_nobarrier:
1780                         clear_opt(sb, BARRIER);
1781                         break;
1782                 case Opt_barrier:
1783                         if (args[0].from) {
1784                                 if (match_int(&args[0], &option))
1785                                         return 0;
1786                         } else
1787                                 option = 1;     /* No argument, default to 1 */
1788                         if (option)
1789                                 set_opt(sb, BARRIER);
1790                         else
1791                                 clear_opt(sb, BARRIER);
1792                         break;
1793                 case Opt_ignore:
1794                         break;
1795                 case Opt_resize:
1796                         if (!is_remount) {
1797                                 ext4_msg(sb, KERN_ERR,
1798                                         "resize option only available "
1799                                         "for remount");
1800                                 return 0;
1801                         }
1802                         if (match_int(&args[0], &option) != 0)
1803                                 return 0;
1804                         *n_blocks_count = option;
1805                         break;
1806                 case Opt_nobh:
1807                         ext4_msg(sb, KERN_WARNING,
1808                                  "Ignoring deprecated nobh option");
1809                         break;
1810                 case Opt_bh:
1811                         ext4_msg(sb, KERN_WARNING,
1812                                  "Ignoring deprecated bh option");
1813                         break;
1814                 case Opt_i_version:
1815                         set_opt(sb, I_VERSION);
1816                         sb->s_flags |= MS_I_VERSION;
1817                         break;
1818                 case Opt_nodelalloc:
1819                         clear_opt(sb, DELALLOC);
1820                         clear_opt2(sb, EXPLICIT_DELALLOC);
1821                         break;
1822                 case Opt_mblk_io_submit:
1823                         set_opt(sb, MBLK_IO_SUBMIT);
1824                         break;
1825                 case Opt_nomblk_io_submit:
1826                         clear_opt(sb, MBLK_IO_SUBMIT);
1827                         break;
1828                 case Opt_stripe:
1829                         if (match_int(&args[0], &option))
1830                                 return 0;
1831                         if (option < 0)
1832                                 return 0;
1833                         sbi->s_stripe = option;
1834                         break;
1835                 case Opt_delalloc:
1836                         set_opt(sb, DELALLOC);
1837                         set_opt2(sb, EXPLICIT_DELALLOC);
1838                         break;
1839                 case Opt_block_validity:
1840                         set_opt(sb, BLOCK_VALIDITY);
1841                         break;
1842                 case Opt_noblock_validity:
1843                         clear_opt(sb, BLOCK_VALIDITY);
1844                         break;
1845                 case Opt_inode_readahead_blks:
1846                         if (match_int(&args[0], &option))
1847                                 return 0;
1848                         if (option < 0 || option > (1 << 30))
1849                                 return 0;
1850                         if (option && !is_power_of_2(option)) {
1851                                 ext4_msg(sb, KERN_ERR,
1852                                          "EXT4-fs: inode_readahead_blks"
1853                                          " must be a power of 2");
1854                                 return 0;
1855                         }
1856                         sbi->s_inode_readahead_blks = option;
1857                         break;
1858                 case Opt_journal_ioprio:
1859                         if (match_int(&args[0], &option))
1860                                 return 0;
1861                         if (option < 0 || option > 7)
1862                                 break;
1863                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1864                                                             option);
1865                         break;
1866                 case Opt_noauto_da_alloc:
1867                         set_opt(sb, NO_AUTO_DA_ALLOC);
1868                         break;
1869                 case Opt_auto_da_alloc:
1870                         if (args[0].from) {
1871                                 if (match_int(&args[0], &option))
1872                                         return 0;
1873                         } else
1874                                 option = 1;     /* No argument, default to 1 */
1875                         if (option)
1876                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1877                         else
1878                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1879                         break;
1880                 case Opt_discard:
1881                         set_opt(sb, DISCARD);
1882                         break;
1883                 case Opt_nodiscard:
1884                         clear_opt(sb, DISCARD);
1885                         break;
1886                 case Opt_dioread_nolock:
1887                         set_opt(sb, DIOREAD_NOLOCK);
1888                         break;
1889                 case Opt_dioread_lock:
1890                         clear_opt(sb, DIOREAD_NOLOCK);
1891                         break;
1892                 case Opt_init_inode_table:
1893                         set_opt(sb, INIT_INODE_TABLE);
1894                         if (args[0].from) {
1895                                 if (match_int(&args[0], &option))
1896                                         return 0;
1897                         } else
1898                                 option = EXT4_DEF_LI_WAIT_MULT;
1899                         if (option < 0)
1900                                 return 0;
1901                         sbi->s_li_wait_mult = option;
1902                         break;
1903                 case Opt_noinit_inode_table:
1904                         clear_opt(sb, INIT_INODE_TABLE);
1905                         break;
1906                 default:
1907                         ext4_msg(sb, KERN_ERR,
1908                                "Unrecognized mount option \"%s\" "
1909                                "or missing value", p);
1910                         return 0;
1911                 }
1912         }
1913 #ifdef CONFIG_QUOTA
1914         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1915                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1916                         clear_opt(sb, USRQUOTA);
1917
1918                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1919                         clear_opt(sb, GRPQUOTA);
1920
1921                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1922                         ext4_msg(sb, KERN_ERR, "old and new quota "
1923                                         "format mixing");
1924                         return 0;
1925                 }
1926
1927                 if (!sbi->s_jquota_fmt) {
1928                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1929                                         "not specified");
1930                         return 0;
1931                 }
1932         } else {
1933                 if (sbi->s_jquota_fmt) {
1934                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1935                                         "specified with no journaling "
1936                                         "enabled");
1937                         return 0;
1938                 }
1939         }
1940 #endif
1941         return 1;
1942 }
1943
1944 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1945                             int read_only)
1946 {
1947         struct ext4_sb_info *sbi = EXT4_SB(sb);
1948         int res = 0;
1949
1950         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1951                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1952                          "forcing read-only mode");
1953                 res = MS_RDONLY;
1954         }
1955         if (read_only)
1956                 goto done;
1957         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1958                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1959                          "running e2fsck is recommended");
1960         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1961                 ext4_msg(sb, KERN_WARNING,
1962                          "warning: mounting fs with errors, "
1963                          "running e2fsck is recommended");
1964         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1965                  le16_to_cpu(es->s_mnt_count) >=
1966                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1967                 ext4_msg(sb, KERN_WARNING,
1968                          "warning: maximal mount count reached, "
1969                          "running e2fsck is recommended");
1970         else if (le32_to_cpu(es->s_checkinterval) &&
1971                 (le32_to_cpu(es->s_lastcheck) +
1972                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1973                 ext4_msg(sb, KERN_WARNING,
1974                          "warning: checktime reached, "
1975                          "running e2fsck is recommended");
1976         if (!sbi->s_journal)
1977                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1978         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1979                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1980         le16_add_cpu(&es->s_mnt_count, 1);
1981         es->s_mtime = cpu_to_le32(get_seconds());
1982         ext4_update_dynamic_rev(sb);
1983         if (sbi->s_journal)
1984                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1985
1986         ext4_commit_super(sb, 1);
1987 done:
1988         if (test_opt(sb, DEBUG))
1989                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1990                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1991                         sb->s_blocksize,
1992                         sbi->s_groups_count,
1993                         EXT4_BLOCKS_PER_GROUP(sb),
1994                         EXT4_INODES_PER_GROUP(sb),
1995                         sbi->s_mount_opt, sbi->s_mount_opt2);
1996
1997         cleancache_init_fs(sb);
1998         return res;
1999 }
2000
2001 static int ext4_fill_flex_info(struct super_block *sb)
2002 {
2003         struct ext4_sb_info *sbi = EXT4_SB(sb);
2004         struct ext4_group_desc *gdp = NULL;
2005         ext4_group_t flex_group_count;
2006         ext4_group_t flex_group;
2007         int groups_per_flex = 0;
2008         size_t size;
2009         int i;
2010
2011         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2012         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2013
2014         if (groups_per_flex < 2) {
2015                 sbi->s_log_groups_per_flex = 0;
2016                 return 1;
2017         }
2018
2019         /* We allocate both existing and potentially added groups */
2020         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2021                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2022                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2023         size = flex_group_count * sizeof(struct flex_groups);
2024         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2025         if (sbi->s_flex_groups == NULL) {
2026                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2027                          flex_group_count);
2028                 goto failed;
2029         }
2030
2031         for (i = 0; i < sbi->s_groups_count; i++) {
2032                 gdp = ext4_get_group_desc(sb, i, NULL);
2033
2034                 flex_group = ext4_flex_group(sbi, i);
2035                 atomic_add(ext4_free_inodes_count(sb, gdp),
2036                            &sbi->s_flex_groups[flex_group].free_inodes);
2037                 atomic_add(ext4_free_blks_count(sb, gdp),
2038                            &sbi->s_flex_groups[flex_group].free_clusters);
2039                 atomic_add(ext4_used_dirs_count(sb, gdp),
2040                            &sbi->s_flex_groups[flex_group].used_dirs);
2041         }
2042
2043         return 1;
2044 failed:
2045         return 0;
2046 }
2047
2048 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2049                             struct ext4_group_desc *gdp)
2050 {
2051         __u16 crc = 0;
2052
2053         if (sbi->s_es->s_feature_ro_compat &
2054             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2055                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2056                 __le32 le_group = cpu_to_le32(block_group);
2057
2058                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2059                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2060                 crc = crc16(crc, (__u8 *)gdp, offset);
2061                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2062                 /* for checksum of struct ext4_group_desc do the rest...*/
2063                 if ((sbi->s_es->s_feature_incompat &
2064                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2065                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2066                         crc = crc16(crc, (__u8 *)gdp + offset,
2067                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2068                                         offset);
2069         }
2070
2071         return cpu_to_le16(crc);
2072 }
2073
2074 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2075                                 struct ext4_group_desc *gdp)
2076 {
2077         if ((sbi->s_es->s_feature_ro_compat &
2078              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2079             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2080                 return 0;
2081
2082         return 1;
2083 }
2084
2085 /* Called at mount-time, super-block is locked */
2086 static int ext4_check_descriptors(struct super_block *sb,
2087                                   ext4_group_t *first_not_zeroed)
2088 {
2089         struct ext4_sb_info *sbi = EXT4_SB(sb);
2090         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2091         ext4_fsblk_t last_block;
2092         ext4_fsblk_t block_bitmap;
2093         ext4_fsblk_t inode_bitmap;
2094         ext4_fsblk_t inode_table;
2095         int flexbg_flag = 0;
2096         ext4_group_t i, grp = sbi->s_groups_count;
2097
2098         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2099                 flexbg_flag = 1;
2100
2101         ext4_debug("Checking group descriptors");
2102
2103         for (i = 0; i < sbi->s_groups_count; i++) {
2104                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2105
2106                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2107                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2108                 else
2109                         last_block = first_block +
2110                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2111
2112                 if ((grp == sbi->s_groups_count) &&
2113                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2114                         grp = i;
2115
2116                 block_bitmap = ext4_block_bitmap(sb, gdp);
2117                 if (block_bitmap < first_block || block_bitmap > last_block) {
2118                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2119                                "Block bitmap for group %u not in group "
2120                                "(block %llu)!", i, block_bitmap);
2121                         return 0;
2122                 }
2123                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2124                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2125                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2126                                "Inode bitmap for group %u not in group "
2127                                "(block %llu)!", i, inode_bitmap);
2128                         return 0;
2129                 }
2130                 inode_table = ext4_inode_table(sb, gdp);
2131                 if (inode_table < first_block ||
2132                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2133                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2134                                "Inode table for group %u not in group "
2135                                "(block %llu)!", i, inode_table);
2136                         return 0;
2137                 }
2138                 ext4_lock_group(sb, i);
2139                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2140                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2141                                  "Checksum for group %u failed (%u!=%u)",
2142                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2143                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2144                         if (!(sb->s_flags & MS_RDONLY)) {
2145                                 ext4_unlock_group(sb, i);
2146                                 return 0;
2147                         }
2148                 }
2149                 ext4_unlock_group(sb, i);
2150                 if (!flexbg_flag)
2151                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2152         }
2153         if (NULL != first_not_zeroed)
2154                 *first_not_zeroed = grp;
2155
2156         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2157         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2158         return 1;
2159 }
2160
2161 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2162  * the superblock) which were deleted from all directories, but held open by
2163  * a process at the time of a crash.  We walk the list and try to delete these
2164  * inodes at recovery time (only with a read-write filesystem).
2165  *
2166  * In order to keep the orphan inode chain consistent during traversal (in
2167  * case of crash during recovery), we link each inode into the superblock
2168  * orphan list_head and handle it the same way as an inode deletion during
2169  * normal operation (which journals the operations for us).
2170  *
2171  * We only do an iget() and an iput() on each inode, which is very safe if we
2172  * accidentally point at an in-use or already deleted inode.  The worst that
2173  * can happen in this case is that we get a "bit already cleared" message from
2174  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2175  * e2fsck was run on this filesystem, and it must have already done the orphan
2176  * inode cleanup for us, so we can safely abort without any further action.
2177  */
2178 static void ext4_orphan_cleanup(struct super_block *sb,
2179                                 struct ext4_super_block *es)
2180 {
2181         unsigned int s_flags = sb->s_flags;
2182         int nr_orphans = 0, nr_truncates = 0;
2183 #ifdef CONFIG_QUOTA
2184         int i;
2185 #endif
2186         if (!es->s_last_orphan) {
2187                 jbd_debug(4, "no orphan inodes to clean up\n");
2188                 return;
2189         }
2190
2191         if (bdev_read_only(sb->s_bdev)) {
2192                 ext4_msg(sb, KERN_ERR, "write access "
2193                         "unavailable, skipping orphan cleanup");
2194                 return;
2195         }
2196
2197         /* Check if feature set would not allow a r/w mount */
2198         if (!ext4_feature_set_ok(sb, 0)) {
2199                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2200                          "unknown ROCOMPAT features");
2201                 return;
2202         }
2203
2204         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2205                 if (es->s_last_orphan)
2206                         jbd_debug(1, "Errors on filesystem, "
2207                                   "clearing orphan list.\n");
2208                 es->s_last_orphan = 0;
2209                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2210                 return;
2211         }
2212
2213         if (s_flags & MS_RDONLY) {
2214                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2215                 sb->s_flags &= ~MS_RDONLY;
2216         }
2217 #ifdef CONFIG_QUOTA
2218         /* Needed for iput() to work correctly and not trash data */
2219         sb->s_flags |= MS_ACTIVE;
2220         /* Turn on quotas so that they are updated correctly */
2221         for (i = 0; i < MAXQUOTAS; i++) {
2222                 if (EXT4_SB(sb)->s_qf_names[i]) {
2223                         int ret = ext4_quota_on_mount(sb, i);
2224                         if (ret < 0)
2225                                 ext4_msg(sb, KERN_ERR,
2226                                         "Cannot turn on journaled "
2227                                         "quota: error %d", ret);
2228                 }
2229         }
2230 #endif
2231
2232         while (es->s_last_orphan) {
2233                 struct inode *inode;
2234
2235                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2236                 if (IS_ERR(inode)) {
2237                         es->s_last_orphan = 0;
2238                         break;
2239                 }
2240
2241                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2242                 dquot_initialize(inode);
2243                 if (inode->i_nlink) {
2244                         ext4_msg(sb, KERN_DEBUG,
2245                                 "%s: truncating inode %lu to %lld bytes",
2246                                 __func__, inode->i_ino, inode->i_size);
2247                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2248                                   inode->i_ino, inode->i_size);
2249                         ext4_truncate(inode);
2250                         nr_truncates++;
2251                 } else {
2252                         ext4_msg(sb, KERN_DEBUG,
2253                                 "%s: deleting unreferenced inode %lu",
2254                                 __func__, inode->i_ino);
2255                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2256                                   inode->i_ino);
2257                         nr_orphans++;
2258                 }
2259                 iput(inode);  /* The delete magic happens here! */
2260         }
2261
2262 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2263
2264         if (nr_orphans)
2265                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2266                        PLURAL(nr_orphans));
2267         if (nr_truncates)
2268                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2269                        PLURAL(nr_truncates));
2270 #ifdef CONFIG_QUOTA
2271         /* Turn quotas off */
2272         for (i = 0; i < MAXQUOTAS; i++) {
2273                 if (sb_dqopt(sb)->files[i])
2274                         dquot_quota_off(sb, i);
2275         }
2276 #endif
2277         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2278 }
2279
2280 /*
2281  * Maximal extent format file size.
2282  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2283  * extent format containers, within a sector_t, and within i_blocks
2284  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2285  * so that won't be a limiting factor.
2286  *
2287  * However there is other limiting factor. We do store extents in the form
2288  * of starting block and length, hence the resulting length of the extent
2289  * covering maximum file size must fit into on-disk format containers as
2290  * well. Given that length is always by 1 unit bigger than max unit (because
2291  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2292  *
2293  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2294  */
2295 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2296 {
2297         loff_t res;
2298         loff_t upper_limit = MAX_LFS_FILESIZE;
2299
2300         /* small i_blocks in vfs inode? */
2301         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2302                 /*
2303                  * CONFIG_LBDAF is not enabled implies the inode
2304                  * i_block represent total blocks in 512 bytes
2305                  * 32 == size of vfs inode i_blocks * 8
2306                  */
2307                 upper_limit = (1LL << 32) - 1;
2308
2309                 /* total blocks in file system block size */
2310                 upper_limit >>= (blkbits - 9);
2311                 upper_limit <<= blkbits;
2312         }
2313
2314         /*
2315          * 32-bit extent-start container, ee_block. We lower the maxbytes
2316          * by one fs block, so ee_len can cover the extent of maximum file
2317          * size
2318          */
2319         res = (1LL << 32) - 1;
2320         res <<= blkbits;
2321
2322         /* Sanity check against vm- & vfs- imposed limits */
2323         if (res > upper_limit)
2324                 res = upper_limit;
2325
2326         return res;
2327 }
2328
2329 /*
2330  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2331  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2332  * We need to be 1 filesystem block less than the 2^48 sector limit.
2333  */
2334 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2335 {
2336         loff_t res = EXT4_NDIR_BLOCKS;
2337         int meta_blocks;
2338         loff_t upper_limit;
2339         /* This is calculated to be the largest file size for a dense, block
2340          * mapped file such that the file's total number of 512-byte sectors,
2341          * including data and all indirect blocks, does not exceed (2^48 - 1).
2342          *
2343          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2344          * number of 512-byte sectors of the file.
2345          */
2346
2347         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2348                 /*
2349                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2350                  * the inode i_block field represents total file blocks in
2351                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2352                  */
2353                 upper_limit = (1LL << 32) - 1;
2354
2355                 /* total blocks in file system block size */
2356                 upper_limit >>= (bits - 9);
2357
2358         } else {
2359                 /*
2360                  * We use 48 bit ext4_inode i_blocks
2361                  * With EXT4_HUGE_FILE_FL set the i_blocks
2362                  * represent total number of blocks in
2363                  * file system block size
2364                  */
2365                 upper_limit = (1LL << 48) - 1;
2366
2367         }
2368
2369         /* indirect blocks */
2370         meta_blocks = 1;
2371         /* double indirect blocks */
2372         meta_blocks += 1 + (1LL << (bits-2));
2373         /* tripple indirect blocks */
2374         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2375
2376         upper_limit -= meta_blocks;
2377         upper_limit <<= bits;
2378
2379         res += 1LL << (bits-2);
2380         res += 1LL << (2*(bits-2));
2381         res += 1LL << (3*(bits-2));
2382         res <<= bits;
2383         if (res > upper_limit)
2384                 res = upper_limit;
2385
2386         if (res > MAX_LFS_FILESIZE)
2387                 res = MAX_LFS_FILESIZE;
2388
2389         return res;
2390 }
2391
2392 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2393                                    ext4_fsblk_t logical_sb_block, int nr)
2394 {
2395         struct ext4_sb_info *sbi = EXT4_SB(sb);
2396         ext4_group_t bg, first_meta_bg;
2397         int has_super = 0;
2398
2399         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2400
2401         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2402             nr < first_meta_bg)
2403                 return logical_sb_block + nr + 1;
2404         bg = sbi->s_desc_per_block * nr;
2405         if (ext4_bg_has_super(sb, bg))
2406                 has_super = 1;
2407
2408         return (has_super + ext4_group_first_block_no(sb, bg));
2409 }
2410
2411 /**
2412  * ext4_get_stripe_size: Get the stripe size.
2413  * @sbi: In memory super block info
2414  *
2415  * If we have specified it via mount option, then
2416  * use the mount option value. If the value specified at mount time is
2417  * greater than the blocks per group use the super block value.
2418  * If the super block value is greater than blocks per group return 0.
2419  * Allocator needs it be less than blocks per group.
2420  *
2421  */
2422 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2423 {
2424         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2425         unsigned long stripe_width =
2426                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2427         int ret;
2428
2429         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2430                 ret = sbi->s_stripe;
2431         else if (stripe_width <= sbi->s_blocks_per_group)
2432                 ret = stripe_width;
2433         else if (stride <= sbi->s_blocks_per_group)
2434                 ret = stride;
2435         else
2436                 ret = 0;
2437
2438         /*
2439          * If the stripe width is 1, this makes no sense and
2440          * we set it to 0 to turn off stripe handling code.
2441          */
2442         if (ret <= 1)
2443                 ret = 0;
2444
2445         return ret;
2446 }
2447
2448 /* sysfs supprt */
2449
2450 struct ext4_attr {
2451         struct attribute attr;
2452         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2453         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2454                          const char *, size_t);
2455         int offset;
2456 };
2457
2458 static int parse_strtoul(const char *buf,
2459                 unsigned long max, unsigned long *value)
2460 {
2461         char *endp;
2462
2463         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2464         endp = skip_spaces(endp);
2465         if (*endp || *value > max)
2466                 return -EINVAL;
2467
2468         return 0;
2469 }
2470
2471 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2472                                               struct ext4_sb_info *sbi,
2473                                               char *buf)
2474 {
2475         return snprintf(buf, PAGE_SIZE, "%llu\n",
2476                 (s64) EXT4_C2B(sbi,
2477                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2478 }
2479
2480 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2481                                          struct ext4_sb_info *sbi, char *buf)
2482 {
2483         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2484
2485         if (!sb->s_bdev->bd_part)
2486                 return snprintf(buf, PAGE_SIZE, "0\n");
2487         return snprintf(buf, PAGE_SIZE, "%lu\n",
2488                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2489                          sbi->s_sectors_written_start) >> 1);
2490 }
2491
2492 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2493                                           struct ext4_sb_info *sbi, char *buf)
2494 {
2495         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2496
2497         if (!sb->s_bdev->bd_part)
2498                 return snprintf(buf, PAGE_SIZE, "0\n");
2499         return snprintf(buf, PAGE_SIZE, "%llu\n",
2500                         (unsigned long long)(sbi->s_kbytes_written +
2501                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2502                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2503 }
2504
2505 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2506                                       struct ext4_sb_info *sbi, char *buf)
2507 {
2508         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2509 }
2510
2511 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2512                                         struct ext4_sb_info *sbi, char *buf)
2513 {
2514         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2515 }
2516
2517 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2518                                           struct ext4_sb_info *sbi,
2519                                           const char *buf, size_t count)
2520 {
2521         unsigned long t;
2522
2523         if (parse_strtoul(buf, 0x40000000, &t))
2524                 return -EINVAL;
2525
2526         if (t && !is_power_of_2(t))
2527                 return -EINVAL;
2528
2529         sbi->s_inode_readahead_blks = t;
2530         return count;
2531 }
2532
2533 static ssize_t sbi_ui_show(struct ext4_attr *a,
2534                            struct ext4_sb_info *sbi, char *buf)
2535 {
2536         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2537
2538         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2539 }
2540
2541 static ssize_t sbi_ui_store(struct ext4_attr *a,
2542                             struct ext4_sb_info *sbi,
2543                             const char *buf, size_t count)
2544 {
2545         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2546         unsigned long t;
2547
2548         if (parse_strtoul(buf, 0xffffffff, &t))
2549                 return -EINVAL;
2550         *ui = t;
2551         return count;
2552 }
2553
2554 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2555 static struct ext4_attr ext4_attr_##_name = {                   \
2556         .attr = {.name = __stringify(_name), .mode = _mode },   \
2557         .show   = _show,                                        \
2558         .store  = _store,                                       \
2559         .offset = offsetof(struct ext4_sb_info, _elname),       \
2560 }
2561 #define EXT4_ATTR(name, mode, show, store) \
2562 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2563
2564 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2565 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2566 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2567 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2568         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2569 #define ATTR_LIST(name) &ext4_attr_##name.attr
2570
2571 EXT4_RO_ATTR(delayed_allocation_blocks);
2572 EXT4_RO_ATTR(session_write_kbytes);
2573 EXT4_RO_ATTR(lifetime_write_kbytes);
2574 EXT4_RO_ATTR(extent_cache_hits);
2575 EXT4_RO_ATTR(extent_cache_misses);
2576 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2577                  inode_readahead_blks_store, s_inode_readahead_blks);
2578 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2579 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2580 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2581 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2582 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2583 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2584 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2585 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2586
2587 static struct attribute *ext4_attrs[] = {
2588         ATTR_LIST(delayed_allocation_blocks),
2589         ATTR_LIST(session_write_kbytes),
2590         ATTR_LIST(lifetime_write_kbytes),
2591         ATTR_LIST(extent_cache_hits),
2592         ATTR_LIST(extent_cache_misses),
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 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3100                                 __releases(kernel_lock)
3101                                 __acquires(kernel_lock)
3102 {
3103         char *orig_data = kstrdup(data, GFP_KERNEL);
3104         struct buffer_head *bh;
3105         struct ext4_super_block *es = NULL;
3106         struct ext4_sb_info *sbi;
3107         ext4_fsblk_t block;
3108         ext4_fsblk_t sb_block = get_sb_block(&data);
3109         ext4_fsblk_t logical_sb_block;
3110         unsigned long offset = 0;
3111         unsigned long journal_devnum = 0;
3112         unsigned long def_mount_opts;
3113         struct inode *root;
3114         char *cp;
3115         const char *descr;
3116         int ret = -ENOMEM;
3117         int blocksize, clustersize;
3118         unsigned int db_count;
3119         unsigned int i;
3120         int needs_recovery, has_huge_files, has_bigalloc;
3121         __u64 blocks_count;
3122         int err;
3123         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3124         ext4_group_t first_not_zeroed;
3125
3126         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3127         if (!sbi)
3128                 goto out_free_orig;
3129
3130         sbi->s_blockgroup_lock =
3131                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3132         if (!sbi->s_blockgroup_lock) {
3133                 kfree(sbi);
3134                 goto out_free_orig;
3135         }
3136         sb->s_fs_info = sbi;
3137         sbi->s_mount_opt = 0;
3138         sbi->s_resuid = EXT4_DEF_RESUID;
3139         sbi->s_resgid = EXT4_DEF_RESGID;
3140         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3141         sbi->s_sb_block = sb_block;
3142         if (sb->s_bdev->bd_part)
3143                 sbi->s_sectors_written_start =
3144                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3145
3146         /* Cleanup superblock name */
3147         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3148                 *cp = '!';
3149
3150         ret = -EINVAL;
3151         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3152         if (!blocksize) {
3153                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3154                 goto out_fail;
3155         }
3156
3157         /*
3158          * The ext4 superblock will not be buffer aligned for other than 1kB
3159          * block sizes.  We need to calculate the offset from buffer start.
3160          */
3161         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3162                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3163                 offset = do_div(logical_sb_block, blocksize);
3164         } else {
3165                 logical_sb_block = sb_block;
3166         }
3167
3168         if (!(bh = sb_bread(sb, logical_sb_block))) {
3169                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3170                 goto out_fail;
3171         }
3172         /*
3173          * Note: s_es must be initialized as soon as possible because
3174          *       some ext4 macro-instructions depend on its value
3175          */
3176         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3177         sbi->s_es = es;
3178         sb->s_magic = le16_to_cpu(es->s_magic);
3179         if (sb->s_magic != EXT4_SUPER_MAGIC)
3180                 goto cantfind_ext4;
3181         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3182
3183         /* Set defaults before we parse the mount options */
3184         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3185         set_opt(sb, INIT_INODE_TABLE);
3186         if (def_mount_opts & EXT4_DEFM_DEBUG)
3187                 set_opt(sb, DEBUG);
3188         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3189                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3190                         "2.6.38");
3191                 set_opt(sb, GRPID);
3192         }
3193         if (def_mount_opts & EXT4_DEFM_UID16)
3194                 set_opt(sb, NO_UID32);
3195         /* xattr user namespace & acls are now defaulted on */
3196 #ifdef CONFIG_EXT4_FS_XATTR
3197         set_opt(sb, XATTR_USER);
3198 #endif
3199 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3200         set_opt(sb, POSIX_ACL);
3201 #endif
3202         set_opt(sb, MBLK_IO_SUBMIT);
3203         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3204                 set_opt(sb, JOURNAL_DATA);
3205         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3206                 set_opt(sb, ORDERED_DATA);
3207         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3208                 set_opt(sb, WRITEBACK_DATA);
3209
3210         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3211                 set_opt(sb, ERRORS_PANIC);
3212         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3213                 set_opt(sb, ERRORS_CONT);
3214         else
3215                 set_opt(sb, ERRORS_RO);
3216         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3217                 set_opt(sb, BLOCK_VALIDITY);
3218         if (def_mount_opts & EXT4_DEFM_DISCARD)
3219                 set_opt(sb, DISCARD);
3220
3221         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3222         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3223         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3224         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3225         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3226
3227         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3228                 set_opt(sb, BARRIER);
3229
3230         /*
3231          * enable delayed allocation by default
3232          * Use -o nodelalloc to turn it off
3233          */
3234         if (!IS_EXT3_SB(sb) &&
3235             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3236                 set_opt(sb, DELALLOC);
3237
3238         /*
3239          * set default s_li_wait_mult for lazyinit, for the case there is
3240          * no mount option specified.
3241          */
3242         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3243
3244         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3245                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3246                 ext4_msg(sb, KERN_WARNING,
3247                          "failed to parse options in superblock: %s",
3248                          sbi->s_es->s_mount_opts);
3249         }
3250         if (!parse_options((char *) data, sb, &journal_devnum,
3251                            &journal_ioprio, NULL, 0))
3252                 goto failed_mount;
3253
3254         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3255                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3256                             "with data=journal disables delayed "
3257                             "allocation and O_DIRECT support!\n");
3258                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3259                         ext4_msg(sb, KERN_ERR, "can't mount with "
3260                                  "both data=journal and delalloc");
3261                         goto failed_mount;
3262                 }
3263                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3264                         ext4_msg(sb, KERN_ERR, "can't mount with "
3265                                  "both data=journal and delalloc");
3266                         goto failed_mount;
3267                 }
3268                 if (test_opt(sb, DELALLOC))
3269                         clear_opt(sb, DELALLOC);
3270         }
3271
3272         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3273         if (test_opt(sb, DIOREAD_NOLOCK)) {
3274                 if (blocksize < PAGE_SIZE) {
3275                         ext4_msg(sb, KERN_ERR, "can't mount with "
3276                                  "dioread_nolock if block size != PAGE_SIZE");
3277                         goto failed_mount;
3278                 }
3279         }
3280
3281         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3282                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3283
3284         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3285             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3286              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3287              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3288                 ext4_msg(sb, KERN_WARNING,
3289                        "feature flags set on rev 0 fs, "
3290                        "running e2fsck is recommended");
3291
3292         if (IS_EXT2_SB(sb)) {
3293                 if (ext2_feature_set_ok(sb))
3294                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3295                                  "using the ext4 subsystem");
3296                 else {
3297                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3298                                  "to feature incompatibilities");
3299                         goto failed_mount;
3300                 }
3301         }
3302
3303         if (IS_EXT3_SB(sb)) {
3304                 if (ext3_feature_set_ok(sb))
3305                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3306                                  "using the ext4 subsystem");
3307                 else {
3308                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3309                                  "to feature incompatibilities");
3310                         goto failed_mount;
3311                 }
3312         }
3313
3314         /*
3315          * Check feature flags regardless of the revision level, since we
3316          * previously didn't change the revision level when setting the flags,
3317          * so there is a chance incompat flags are set on a rev 0 filesystem.
3318          */
3319         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3320                 goto failed_mount;
3321
3322         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3323             blocksize > EXT4_MAX_BLOCK_SIZE) {
3324                 ext4_msg(sb, KERN_ERR,
3325                        "Unsupported filesystem blocksize %d", blocksize);
3326                 goto failed_mount;
3327         }
3328
3329         if (sb->s_blocksize != blocksize) {
3330                 /* Validate the filesystem blocksize */
3331                 if (!sb_set_blocksize(sb, blocksize)) {
3332                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3333                                         blocksize);
3334                         goto failed_mount;
3335                 }
3336
3337                 brelse(bh);
3338                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3339                 offset = do_div(logical_sb_block, blocksize);
3340                 bh = sb_bread(sb, logical_sb_block);
3341                 if (!bh) {
3342                         ext4_msg(sb, KERN_ERR,
3343                                "Can't read superblock on 2nd try");
3344                         goto failed_mount;
3345                 }
3346                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3347                 sbi->s_es = es;
3348                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3349                         ext4_msg(sb, KERN_ERR,
3350                                "Magic mismatch, very weird!");
3351                         goto failed_mount;
3352                 }
3353         }
3354
3355         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3356                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3357         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3358                                                       has_huge_files);
3359         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3360
3361         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3362                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3363                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3364         } else {
3365                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3366                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3367                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3368                     (!is_power_of_2(sbi->s_inode_size)) ||
3369                     (sbi->s_inode_size > blocksize)) {
3370                         ext4_msg(sb, KERN_ERR,
3371                                "unsupported inode size: %d",
3372                                sbi->s_inode_size);
3373                         goto failed_mount;
3374                 }
3375                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3376                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3377         }
3378
3379         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3380         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3381                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3382                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3383                     !is_power_of_2(sbi->s_desc_size)) {
3384                         ext4_msg(sb, KERN_ERR,
3385                                "unsupported descriptor size %lu",
3386                                sbi->s_desc_size);
3387                         goto failed_mount;
3388                 }
3389         } else
3390                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3391
3392         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3393         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3394         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3395                 goto cantfind_ext4;
3396
3397         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3398         if (sbi->s_inodes_per_block == 0)
3399                 goto cantfind_ext4;
3400         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3401                                         sbi->s_inodes_per_block;
3402         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3403         sbi->s_sbh = bh;
3404         sbi->s_mount_state = le16_to_cpu(es->s_state);
3405         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3406         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3407
3408         for (i = 0; i < 4; i++)
3409                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3410         sbi->s_def_hash_version = es->s_def_hash_version;
3411         i = le32_to_cpu(es->s_flags);
3412         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3413                 sbi->s_hash_unsigned = 3;
3414         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3415 #ifdef __CHAR_UNSIGNED__
3416                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3417                 sbi->s_hash_unsigned = 3;
3418 #else
3419                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3420 #endif
3421                 sb->s_dirt = 1;
3422         }
3423
3424         /* Handle clustersize */
3425         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3426         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3427                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3428         if (has_bigalloc) {
3429                 if (clustersize < blocksize) {
3430                         ext4_msg(sb, KERN_ERR,
3431                                  "cluster size (%d) smaller than "
3432                                  "block size (%d)", clustersize, blocksize);
3433                         goto failed_mount;
3434                 }
3435                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3436                         le32_to_cpu(es->s_log_block_size);
3437                 sbi->s_clusters_per_group =
3438                         le32_to_cpu(es->s_clusters_per_group);
3439                 if (sbi->s_clusters_per_group > blocksize * 8) {
3440                         ext4_msg(sb, KERN_ERR,
3441                                  "#clusters per group too big: %lu",
3442                                  sbi->s_clusters_per_group);
3443                         goto failed_mount;
3444                 }
3445                 if (sbi->s_blocks_per_group !=
3446                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3447                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3448                                  "clusters per group (%lu) inconsistent",
3449                                  sbi->s_blocks_per_group,
3450                                  sbi->s_clusters_per_group);
3451                         goto failed_mount;
3452                 }
3453         } else {
3454                 if (clustersize != blocksize) {
3455                         ext4_warning(sb, "fragment/cluster size (%d) != "
3456                                      "block size (%d)", clustersize,
3457                                      blocksize);
3458                         clustersize = blocksize;
3459                 }
3460                 if (sbi->s_blocks_per_group > blocksize * 8) {
3461                         ext4_msg(sb, KERN_ERR,
3462                                  "#blocks per group too big: %lu",
3463                                  sbi->s_blocks_per_group);
3464                         goto failed_mount;
3465                 }
3466                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3467                 sbi->s_cluster_bits = 0;
3468         }
3469         sbi->s_cluster_ratio = clustersize / blocksize;
3470
3471         if (sbi->s_inodes_per_group > blocksize * 8) {
3472                 ext4_msg(sb, KERN_ERR,
3473                        "#inodes per group too big: %lu",
3474                        sbi->s_inodes_per_group);
3475                 goto failed_mount;
3476         }
3477
3478         /*
3479          * Test whether we have more sectors than will fit in sector_t,
3480          * and whether the max offset is addressable by the page cache.
3481          */
3482         err = generic_check_addressable(sb->s_blocksize_bits,
3483                                         ext4_blocks_count(es));
3484         if (err) {
3485                 ext4_msg(sb, KERN_ERR, "filesystem"
3486                          " too large to mount safely on this system");
3487                 if (sizeof(sector_t) < 8)
3488                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3489                 ret = err;
3490                 goto failed_mount;
3491         }
3492
3493         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3494                 goto cantfind_ext4;
3495
3496         /* check blocks count against device size */
3497         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3498         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3499                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3500                        "exceeds size of device (%llu blocks)",
3501                        ext4_blocks_count(es), blocks_count);
3502                 goto failed_mount;
3503         }
3504
3505         /*
3506          * It makes no sense for the first data block to be beyond the end
3507          * of the filesystem.
3508          */
3509         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3510                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3511                          "block %u is beyond end of filesystem (%llu)",
3512                          le32_to_cpu(es->s_first_data_block),
3513                          ext4_blocks_count(es));
3514                 goto failed_mount;
3515         }
3516         blocks_count = (ext4_blocks_count(es) -
3517                         le32_to_cpu(es->s_first_data_block) +
3518                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3519         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3520         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3521                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3522                        "(block count %llu, first data block %u, "
3523                        "blocks per group %lu)", sbi->s_groups_count,
3524                        ext4_blocks_count(es),
3525                        le32_to_cpu(es->s_first_data_block),
3526                        EXT4_BLOCKS_PER_GROUP(sb));
3527                 goto failed_mount;
3528         }
3529         sbi->s_groups_count = blocks_count;
3530         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3531                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3532         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3533                    EXT4_DESC_PER_BLOCK(sb);
3534         sbi->s_group_desc = ext4_kvmalloc(db_count *
3535                                           sizeof(struct buffer_head *),
3536                                           GFP_KERNEL);
3537         if (sbi->s_group_desc == NULL) {
3538                 ext4_msg(sb, KERN_ERR, "not enough memory");
3539                 goto failed_mount;
3540         }
3541
3542 #ifdef CONFIG_PROC_FS
3543         if (ext4_proc_root)
3544                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3545 #endif
3546
3547         bgl_lock_init(sbi->s_blockgroup_lock);
3548
3549         for (i = 0; i < db_count; i++) {
3550                 block = descriptor_loc(sb, logical_sb_block, i);
3551                 sbi->s_group_desc[i] = sb_bread(sb, block);
3552                 if (!sbi->s_group_desc[i]) {
3553                         ext4_msg(sb, KERN_ERR,
3554                                "can't read group descriptor %d", i);
3555                         db_count = i;
3556                         goto failed_mount2;
3557                 }
3558         }
3559         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3560                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3561                 goto failed_mount2;
3562         }
3563         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3564                 if (!ext4_fill_flex_info(sb)) {
3565                         ext4_msg(sb, KERN_ERR,
3566                                "unable to initialize "
3567                                "flex_bg meta info!");
3568                         goto failed_mount2;
3569                 }
3570
3571         sbi->s_gdb_count = db_count;
3572         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3573         spin_lock_init(&sbi->s_next_gen_lock);
3574
3575         init_timer(&sbi->s_err_report);
3576         sbi->s_err_report.function = print_daily_error_info;
3577         sbi->s_err_report.data = (unsigned long) sb;
3578
3579         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3580                         ext4_count_free_blocks(sb));
3581         if (!err) {
3582                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3583                                 ext4_count_free_inodes(sb));
3584         }
3585         if (!err) {
3586                 err = percpu_counter_init(&sbi->s_dirs_counter,
3587                                 ext4_count_dirs(sb));
3588         }
3589         if (!err) {
3590                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3591         }
3592         if (err) {
3593                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3594                 goto failed_mount3;
3595         }
3596
3597         sbi->s_stripe = ext4_get_stripe_size(sbi);
3598         sbi->s_max_writeback_mb_bump = 128;
3599
3600         /*
3601          * set up enough so that it can read an inode
3602          */
3603         if (!test_opt(sb, NOLOAD) &&
3604             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3605                 sb->s_op = &ext4_sops;
3606         else
3607                 sb->s_op = &ext4_nojournal_sops;
3608         sb->s_export_op = &ext4_export_ops;
3609         sb->s_xattr = ext4_xattr_handlers;
3610 #ifdef CONFIG_QUOTA
3611         sb->s_qcop = &ext4_qctl_operations;
3612         sb->dq_op = &ext4_quota_operations;
3613 #endif
3614         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3615
3616         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3617         mutex_init(&sbi->s_orphan_lock);
3618         sbi->s_resize_flags = 0;
3619
3620         sb->s_root = NULL;
3621
3622         needs_recovery = (es->s_last_orphan != 0 ||
3623                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3624                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3625
3626         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3627             !(sb->s_flags & MS_RDONLY))
3628                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3629                         goto failed_mount3;
3630
3631         /*
3632          * The first inode we look at is the journal inode.  Don't try
3633          * root first: it may be modified in the journal!
3634          */
3635         if (!test_opt(sb, NOLOAD) &&
3636             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3637                 if (ext4_load_journal(sb, es, journal_devnum))
3638                         goto failed_mount3;
3639         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3640               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3641                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3642                        "suppressed and not mounted read-only");
3643                 goto failed_mount_wq;
3644         } else {
3645                 clear_opt(sb, DATA_FLAGS);
3646                 sbi->s_journal = NULL;
3647                 needs_recovery = 0;
3648                 goto no_journal;
3649         }
3650
3651         if (ext4_blocks_count(es) > 0xffffffffULL &&
3652             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3653                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3654                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3655                 goto failed_mount_wq;
3656         }
3657
3658         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3659                 jbd2_journal_set_features(sbi->s_journal,
3660                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3661                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3662         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3663                 jbd2_journal_set_features(sbi->s_journal,
3664                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3665                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3666                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3667         } else {
3668                 jbd2_journal_clear_features(sbi->s_journal,
3669                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3670                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3671         }
3672
3673         /* We have now updated the journal if required, so we can
3674          * validate the data journaling mode. */
3675         switch (test_opt(sb, DATA_FLAGS)) {
3676         case 0:
3677                 /* No mode set, assume a default based on the journal
3678                  * capabilities: ORDERED_DATA if the journal can
3679                  * cope, else JOURNAL_DATA
3680                  */
3681                 if (jbd2_journal_check_available_features
3682                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3683                         set_opt(sb, ORDERED_DATA);
3684                 else
3685                         set_opt(sb, JOURNAL_DATA);
3686                 break;
3687
3688         case EXT4_MOUNT_ORDERED_DATA:
3689         case EXT4_MOUNT_WRITEBACK_DATA:
3690                 if (!jbd2_journal_check_available_features
3691                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3692                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3693                                "requested data journaling mode");
3694                         goto failed_mount_wq;
3695                 }
3696         default:
3697                 break;
3698         }
3699         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3700
3701         /*
3702          * The journal may have updated the bg summary counts, so we
3703          * need to update the global counters.
3704          */
3705         percpu_counter_set(&sbi->s_freeclusters_counter,
3706                            ext4_count_free_blocks(sb));
3707         percpu_counter_set(&sbi->s_freeinodes_counter,
3708                            ext4_count_free_inodes(sb));
3709         percpu_counter_set(&sbi->s_dirs_counter,
3710                            ext4_count_dirs(sb));
3711         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3712
3713 no_journal:
3714         /*
3715          * The maximum number of concurrent works can be high and
3716          * concurrency isn't really necessary.  Limit it to 1.
3717          */
3718         EXT4_SB(sb)->dio_unwritten_wq =
3719                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3720         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3721                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3722                 goto failed_mount_wq;
3723         }
3724
3725         /*
3726          * The jbd2_journal_load will have done any necessary log recovery,
3727          * so we can safely mount the rest of the filesystem now.
3728          */
3729
3730         root = ext4_iget(sb, EXT4_ROOT_INO);
3731         if (IS_ERR(root)) {
3732                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3733                 ret = PTR_ERR(root);
3734                 root = NULL;
3735                 goto failed_mount4;
3736         }
3737         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3738                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3739                 goto failed_mount4;
3740         }
3741         sb->s_root = d_alloc_root(root);
3742         if (!sb->s_root) {
3743                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3744                 ret = -ENOMEM;
3745                 goto failed_mount4;
3746         }
3747
3748         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3749
3750         /* determine the minimum size of new large inodes, if present */
3751         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3752                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3753                                                      EXT4_GOOD_OLD_INODE_SIZE;
3754                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3755                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3756                         if (sbi->s_want_extra_isize <
3757                             le16_to_cpu(es->s_want_extra_isize))
3758                                 sbi->s_want_extra_isize =
3759                                         le16_to_cpu(es->s_want_extra_isize);
3760                         if (sbi->s_want_extra_isize <
3761                             le16_to_cpu(es->s_min_extra_isize))
3762                                 sbi->s_want_extra_isize =
3763                                         le16_to_cpu(es->s_min_extra_isize);
3764                 }
3765         }
3766         /* Check if enough inode space is available */
3767         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3768                                                         sbi->s_inode_size) {
3769                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3770                                                        EXT4_GOOD_OLD_INODE_SIZE;
3771                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3772                          "available");
3773         }
3774
3775         err = ext4_setup_system_zone(sb);
3776         if (err) {
3777                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3778                          "zone (%d)", err);
3779                 goto failed_mount4;
3780         }
3781
3782         ext4_ext_init(sb);
3783         err = ext4_mb_init(sb, needs_recovery);
3784         if (err) {
3785                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3786                          err);
3787                 goto failed_mount4;
3788         }
3789
3790         err = ext4_register_li_request(sb, first_not_zeroed);
3791         if (err)
3792                 goto failed_mount4;
3793
3794         sbi->s_kobj.kset = ext4_kset;
3795         init_completion(&sbi->s_kobj_unregister);
3796         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3797                                    "%s", sb->s_id);
3798         if (err) {
3799                 ext4_mb_release(sb);
3800                 ext4_ext_release(sb);
3801                 goto failed_mount4;
3802         };
3803
3804         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3805         ext4_orphan_cleanup(sb, es);
3806         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3807         if (needs_recovery) {
3808                 ext4_msg(sb, KERN_INFO, "recovery complete");
3809                 ext4_mark_recovery_complete(sb, es);
3810         }
3811         if (EXT4_SB(sb)->s_journal) {
3812                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3813                         descr = " journalled data mode";
3814                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3815                         descr = " ordered data mode";
3816                 else
3817                         descr = " writeback data mode";
3818         } else
3819                 descr = "out journal";
3820
3821         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3822                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3823                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3824
3825         if (es->s_error_count)
3826                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3827
3828         kfree(orig_data);
3829         return 0;
3830
3831 cantfind_ext4:
3832         if (!silent)
3833                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3834         goto failed_mount;
3835
3836 failed_mount4:
3837         iput(root);
3838         sb->s_root = NULL;
3839         ext4_msg(sb, KERN_ERR, "mount failed");
3840         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3841 failed_mount_wq:
3842         ext4_release_system_zone(sb);
3843         if (sbi->s_journal) {
3844                 jbd2_journal_destroy(sbi->s_journal);
3845                 sbi->s_journal = NULL;
3846         }
3847 failed_mount3:
3848         del_timer(&sbi->s_err_report);
3849         if (sbi->s_flex_groups)
3850                 ext4_kvfree(sbi->s_flex_groups);
3851         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3852         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3853         percpu_counter_destroy(&sbi->s_dirs_counter);
3854         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3855         if (sbi->s_mmp_tsk)
3856                 kthread_stop(sbi->s_mmp_tsk);
3857 failed_mount2:
3858         for (i = 0; i < db_count; i++)
3859                 brelse(sbi->s_group_desc[i]);
3860         ext4_kvfree(sbi->s_group_desc);
3861 failed_mount:
3862         if (sbi->s_proc) {
3863                 remove_proc_entry(sb->s_id, ext4_proc_root);
3864         }
3865 #ifdef CONFIG_QUOTA
3866         for (i = 0; i < MAXQUOTAS; i++)
3867                 kfree(sbi->s_qf_names[i]);
3868 #endif
3869         ext4_blkdev_remove(sbi);
3870         brelse(bh);
3871 out_fail:
3872         sb->s_fs_info = NULL;
3873         kfree(sbi->s_blockgroup_lock);
3874         kfree(sbi);
3875 out_free_orig:
3876         kfree(orig_data);
3877         return ret;
3878 }
3879
3880 /*
3881  * Setup any per-fs journal parameters now.  We'll do this both on
3882  * initial mount, once the journal has been initialised but before we've
3883  * done any recovery; and again on any subsequent remount.
3884  */
3885 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3886 {
3887         struct ext4_sb_info *sbi = EXT4_SB(sb);
3888
3889         journal->j_commit_interval = sbi->s_commit_interval;
3890         journal->j_min_batch_time = sbi->s_min_batch_time;
3891         journal->j_max_batch_time = sbi->s_max_batch_time;
3892
3893         write_lock(&journal->j_state_lock);
3894         if (test_opt(sb, BARRIER))
3895                 journal->j_flags |= JBD2_BARRIER;
3896         else
3897                 journal->j_flags &= ~JBD2_BARRIER;
3898         if (test_opt(sb, DATA_ERR_ABORT))
3899                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3900         else
3901                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3902         write_unlock(&journal->j_state_lock);
3903 }
3904
3905 static journal_t *ext4_get_journal(struct super_block *sb,
3906                                    unsigned int journal_inum)
3907 {
3908         struct inode *journal_inode;
3909         journal_t *journal;
3910
3911         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3912
3913         /* First, test for the existence of a valid inode on disk.  Bad
3914          * things happen if we iget() an unused inode, as the subsequent
3915          * iput() will try to delete it. */
3916
3917         journal_inode = ext4_iget(sb, journal_inum);
3918         if (IS_ERR(journal_inode)) {
3919                 ext4_msg(sb, KERN_ERR, "no journal found");
3920                 return NULL;
3921         }
3922         if (!journal_inode->i_nlink) {
3923                 make_bad_inode(journal_inode);
3924                 iput(journal_inode);
3925                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3926                 return NULL;
3927         }
3928
3929         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3930                   journal_inode, journal_inode->i_size);
3931         if (!S_ISREG(journal_inode->i_mode)) {
3932                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3933                 iput(journal_inode);
3934                 return NULL;
3935         }
3936
3937         journal = jbd2_journal_init_inode(journal_inode);
3938         if (!journal) {
3939                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3940                 iput(journal_inode);
3941                 return NULL;
3942         }
3943         journal->j_private = sb;
3944         ext4_init_journal_params(sb, journal);
3945         return journal;
3946 }
3947
3948 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3949                                        dev_t j_dev)
3950 {
3951         struct buffer_head *bh;
3952         journal_t *journal;
3953         ext4_fsblk_t start;
3954         ext4_fsblk_t len;
3955         int hblock, blocksize;
3956         ext4_fsblk_t sb_block;
3957         unsigned long offset;
3958         struct ext4_super_block *es;
3959         struct block_device *bdev;
3960
3961         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3962
3963         bdev = ext4_blkdev_get(j_dev, sb);
3964         if (bdev == NULL)
3965                 return NULL;
3966
3967         blocksize = sb->s_blocksize;
3968         hblock = bdev_logical_block_size(bdev);
3969         if (blocksize < hblock) {
3970                 ext4_msg(sb, KERN_ERR,
3971                         "blocksize too small for journal device");
3972                 goto out_bdev;
3973         }
3974
3975         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3976         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3977         set_blocksize(bdev, blocksize);
3978         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3979                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3980                        "external journal");
3981                 goto out_bdev;
3982         }
3983
3984         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3985         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3986             !(le32_to_cpu(es->s_feature_incompat) &
3987               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3988                 ext4_msg(sb, KERN_ERR, "external journal has "
3989                                         "bad superblock");
3990                 brelse(bh);
3991                 goto out_bdev;
3992         }
3993
3994         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3995                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3996                 brelse(bh);
3997                 goto out_bdev;
3998         }
3999
4000         len = ext4_blocks_count(es);
4001         start = sb_block + 1;
4002         brelse(bh);     /* we're done with the superblock */
4003
4004         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4005                                         start, len, blocksize);
4006         if (!journal) {
4007                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4008                 goto out_bdev;
4009         }
4010         journal->j_private = sb;
4011         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4012         wait_on_buffer(journal->j_sb_buffer);
4013         if (!buffer_uptodate(journal->j_sb_buffer)) {
4014                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4015                 goto out_journal;
4016         }
4017         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4018                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4019                                         "user (unsupported) - %d",
4020                         be32_to_cpu(journal->j_superblock->s_nr_users));
4021                 goto out_journal;
4022         }
4023         EXT4_SB(sb)->journal_bdev = bdev;
4024         ext4_init_journal_params(sb, journal);
4025         return journal;
4026
4027 out_journal:
4028         jbd2_journal_destroy(journal);
4029 out_bdev:
4030         ext4_blkdev_put(bdev);
4031         return NULL;
4032 }
4033
4034 static int ext4_load_journal(struct super_block *sb,
4035                              struct ext4_super_block *es,
4036                              unsigned long journal_devnum)
4037 {
4038         journal_t *journal;
4039         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4040         dev_t journal_dev;
4041         int err = 0;
4042         int really_read_only;
4043
4044         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4045
4046         if (journal_devnum &&
4047             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4048                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4049                         "numbers have changed");
4050                 journal_dev = new_decode_dev(journal_devnum);
4051         } else
4052                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4053
4054         really_read_only = bdev_read_only(sb->s_bdev);
4055
4056         /*
4057          * Are we loading a blank journal or performing recovery after a
4058          * crash?  For recovery, we need to check in advance whether we
4059          * can get read-write access to the device.
4060          */
4061         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4062                 if (sb->s_flags & MS_RDONLY) {
4063                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4064                                         "required on readonly filesystem");
4065                         if (really_read_only) {
4066                                 ext4_msg(sb, KERN_ERR, "write access "
4067                                         "unavailable, cannot proceed");
4068                                 return -EROFS;
4069                         }
4070                         ext4_msg(sb, KERN_INFO, "write access will "
4071                                "be enabled during recovery");
4072                 }
4073         }
4074
4075         if (journal_inum && journal_dev) {
4076                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4077                        "and inode journals!");
4078                 return -EINVAL;
4079         }
4080
4081         if (journal_inum) {
4082                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4083                         return -EINVAL;
4084         } else {
4085                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4086                         return -EINVAL;
4087         }
4088
4089         if (!(journal->j_flags & JBD2_BARRIER))
4090                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4091
4092         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4093                 err = jbd2_journal_update_format(journal);
4094                 if (err)  {
4095                         ext4_msg(sb, KERN_ERR, "error updating journal");
4096                         jbd2_journal_destroy(journal);
4097                         return err;
4098                 }
4099         }
4100
4101         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4102                 err = jbd2_journal_wipe(journal, !really_read_only);
4103         if (!err) {
4104                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4105                 if (save)
4106                         memcpy(save, ((char *) es) +
4107                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4108                 err = jbd2_journal_load(journal);
4109                 if (save)
4110                         memcpy(((char *) es) + EXT4_S_ERR_START,
4111                                save, EXT4_S_ERR_LEN);
4112                 kfree(save);
4113         }
4114
4115         if (err) {
4116                 ext4_msg(sb, KERN_ERR, "error loading journal");
4117                 jbd2_journal_destroy(journal);
4118                 return err;
4119         }
4120
4121         EXT4_SB(sb)->s_journal = journal;
4122         ext4_clear_journal_err(sb, es);
4123
4124         if (!really_read_only && journal_devnum &&
4125             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4126                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4127
4128                 /* Make sure we flush the recovery flag to disk. */
4129                 ext4_commit_super(sb, 1);
4130         }
4131
4132         return 0;
4133 }
4134
4135 static int ext4_commit_super(struct super_block *sb, int sync)
4136 {
4137         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4138         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4139         int error = 0;
4140
4141         if (!sbh || block_device_ejected(sb))
4142                 return error;
4143         if (buffer_write_io_error(sbh)) {
4144                 /*
4145                  * Oh, dear.  A previous attempt to write the
4146                  * superblock failed.  This could happen because the
4147                  * USB device was yanked out.  Or it could happen to
4148                  * be a transient write error and maybe the block will
4149                  * be remapped.  Nothing we can do but to retry the
4150                  * write and hope for the best.
4151                  */
4152                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4153                        "superblock detected");
4154                 clear_buffer_write_io_error(sbh);
4155                 set_buffer_uptodate(sbh);
4156         }
4157         /*
4158          * If the file system is mounted read-only, don't update the
4159          * superblock write time.  This avoids updating the superblock
4160          * write time when we are mounting the root file system
4161          * read/only but we need to replay the journal; at that point,
4162          * for people who are east of GMT and who make their clock
4163          * tick in localtime for Windows bug-for-bug compatibility,
4164          * the clock is set in the future, and this will cause e2fsck
4165          * to complain and force a full file system check.
4166          */
4167         if (!(sb->s_flags & MS_RDONLY))
4168                 es->s_wtime = cpu_to_le32(get_seconds());
4169         if (sb->s_bdev->bd_part)
4170                 es->s_kbytes_written =
4171                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4172                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4173                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4174         else
4175                 es->s_kbytes_written =
4176                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4177         ext4_free_blocks_count_set(es,
4178                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4179                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4180         es->s_free_inodes_count =
4181                 cpu_to_le32(percpu_counter_sum_positive(
4182                                 &EXT4_SB(sb)->s_freeinodes_counter));
4183         sb->s_dirt = 0;
4184         BUFFER_TRACE(sbh, "marking dirty");
4185         mark_buffer_dirty(sbh);
4186         if (sync) {
4187                 error = sync_dirty_buffer(sbh);
4188                 if (error)
4189                         return error;
4190
4191                 error = buffer_write_io_error(sbh);
4192                 if (error) {
4193                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4194                                "superblock");
4195                         clear_buffer_write_io_error(sbh);
4196                         set_buffer_uptodate(sbh);
4197                 }
4198         }
4199         return error;
4200 }
4201
4202 /*
4203  * Have we just finished recovery?  If so, and if we are mounting (or
4204  * remounting) the filesystem readonly, then we will end up with a
4205  * consistent fs on disk.  Record that fact.
4206  */
4207 static void ext4_mark_recovery_complete(struct super_block *sb,
4208                                         struct ext4_super_block *es)
4209 {
4210         journal_t *journal = EXT4_SB(sb)->s_journal;
4211
4212         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4213                 BUG_ON(journal != NULL);
4214                 return;
4215         }
4216         jbd2_journal_lock_updates(journal);
4217         if (jbd2_journal_flush(journal) < 0)
4218                 goto out;
4219
4220         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4221             sb->s_flags & MS_RDONLY) {
4222                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4223                 ext4_commit_super(sb, 1);
4224         }
4225
4226 out:
4227         jbd2_journal_unlock_updates(journal);
4228 }
4229
4230 /*
4231  * If we are mounting (or read-write remounting) a filesystem whose journal
4232  * has recorded an error from a previous lifetime, move that error to the
4233  * main filesystem now.
4234  */
4235 static void ext4_clear_journal_err(struct super_block *sb,
4236                                    struct ext4_super_block *es)
4237 {
4238         journal_t *journal;
4239         int j_errno;
4240         const char *errstr;
4241
4242         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4243
4244         journal = EXT4_SB(sb)->s_journal;
4245
4246         /*
4247          * Now check for any error status which may have been recorded in the
4248          * journal by a prior ext4_error() or ext4_abort()
4249          */
4250
4251         j_errno = jbd2_journal_errno(journal);
4252         if (j_errno) {
4253                 char nbuf[16];
4254
4255                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4256                 ext4_warning(sb, "Filesystem error recorded "
4257                              "from previous mount: %s", errstr);
4258                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4259
4260                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4261                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4262                 ext4_commit_super(sb, 1);
4263
4264                 jbd2_journal_clear_err(journal);
4265         }
4266 }
4267
4268 /*
4269  * Force the running and committing transactions to commit,
4270  * and wait on the commit.
4271  */
4272 int ext4_force_commit(struct super_block *sb)
4273 {
4274         journal_t *journal;
4275         int ret = 0;
4276
4277         if (sb->s_flags & MS_RDONLY)
4278                 return 0;
4279
4280         journal = EXT4_SB(sb)->s_journal;
4281         if (journal) {
4282                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4283                 ret = ext4_journal_force_commit(journal);
4284         }
4285
4286         return ret;
4287 }
4288
4289 static void ext4_write_super(struct super_block *sb)
4290 {
4291         lock_super(sb);
4292         ext4_commit_super(sb, 1);
4293         unlock_super(sb);
4294 }
4295
4296 static int ext4_sync_fs(struct super_block *sb, int wait)
4297 {
4298         int ret = 0;
4299         tid_t target;
4300         struct ext4_sb_info *sbi = EXT4_SB(sb);
4301
4302         trace_ext4_sync_fs(sb, wait);
4303         flush_workqueue(sbi->dio_unwritten_wq);
4304         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4305                 if (wait)
4306                         jbd2_log_wait_commit(sbi->s_journal, target);
4307         }
4308         return ret;
4309 }
4310
4311 /*
4312  * LVM calls this function before a (read-only) snapshot is created.  This
4313  * gives us a chance to flush the journal completely and mark the fs clean.
4314  *
4315  * Note that only this function cannot bring a filesystem to be in a clean
4316  * state independently, because ext4 prevents a new handle from being started
4317  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4318  * the upper layer.
4319  */
4320 static int ext4_freeze(struct super_block *sb)
4321 {
4322         int error = 0;
4323         journal_t *journal;
4324
4325         if (sb->s_flags & MS_RDONLY)
4326                 return 0;
4327
4328         journal = EXT4_SB(sb)->s_journal;
4329
4330         /* Now we set up the journal barrier. */
4331         jbd2_journal_lock_updates(journal);
4332
4333         /*
4334          * Don't clear the needs_recovery flag if we failed to flush
4335          * the journal.
4336          */
4337         error = jbd2_journal_flush(journal);
4338         if (error < 0)
4339                 goto out;
4340
4341         /* Journal blocked and flushed, clear needs_recovery flag. */
4342         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4343         error = ext4_commit_super(sb, 1);
4344 out:
4345         /* we rely on s_frozen to stop further updates */
4346         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4347         return error;
4348 }
4349
4350 /*
4351  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4352  * flag here, even though the filesystem is not technically dirty yet.
4353  */
4354 static int ext4_unfreeze(struct super_block *sb)
4355 {
4356         if (sb->s_flags & MS_RDONLY)
4357                 return 0;
4358
4359         lock_super(sb);
4360         /* Reset the needs_recovery flag before the fs is unlocked. */
4361         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4362         ext4_commit_super(sb, 1);
4363         unlock_super(sb);
4364         return 0;
4365 }
4366
4367 /*
4368  * Structure to save mount options for ext4_remount's benefit
4369  */
4370 struct ext4_mount_options {
4371         unsigned long s_mount_opt;
4372         unsigned long s_mount_opt2;
4373         uid_t s_resuid;
4374         gid_t s_resgid;
4375         unsigned long s_commit_interval;
4376         u32 s_min_batch_time, s_max_batch_time;
4377 #ifdef CONFIG_QUOTA
4378         int s_jquota_fmt;
4379         char *s_qf_names[MAXQUOTAS];
4380 #endif
4381 };
4382
4383 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4384 {
4385         struct ext4_super_block *es;
4386         struct ext4_sb_info *sbi = EXT4_SB(sb);
4387         ext4_fsblk_t n_blocks_count = 0;
4388         unsigned long old_sb_flags;
4389         struct ext4_mount_options old_opts;
4390         int enable_quota = 0;
4391         ext4_group_t g;
4392         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4393         int err = 0;
4394 #ifdef CONFIG_QUOTA
4395         int i;
4396 #endif
4397         char *orig_data = kstrdup(data, GFP_KERNEL);
4398
4399         /* Store the original options */
4400         lock_super(sb);
4401         old_sb_flags = sb->s_flags;
4402         old_opts.s_mount_opt = sbi->s_mount_opt;
4403         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4404         old_opts.s_resuid = sbi->s_resuid;
4405         old_opts.s_resgid = sbi->s_resgid;
4406         old_opts.s_commit_interval = sbi->s_commit_interval;
4407         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4408         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4409 #ifdef CONFIG_QUOTA
4410         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4411         for (i = 0; i < MAXQUOTAS; i++)
4412                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4413 #endif
4414         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4415                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4416
4417         /*
4418          * Allow the "check" option to be passed as a remount option.
4419          */
4420         if (!parse_options(data, sb, NULL, &journal_ioprio,
4421                            &n_blocks_count, 1)) {
4422                 err = -EINVAL;
4423                 goto restore_opts;
4424         }
4425
4426         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4427                 ext4_abort(sb, "Abort forced by user");
4428
4429         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4430                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4431
4432         es = sbi->s_es;
4433
4434         if (sbi->s_journal) {
4435                 ext4_init_journal_params(sb, sbi->s_journal);
4436                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4437         }
4438
4439         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4440                 n_blocks_count > ext4_blocks_count(es)) {
4441                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4442                         err = -EROFS;
4443                         goto restore_opts;
4444                 }
4445
4446                 if (*flags & MS_RDONLY) {
4447                         err = dquot_suspend(sb, -1);
4448                         if (err < 0)
4449                                 goto restore_opts;
4450
4451                         /*
4452                          * First of all, the unconditional stuff we have to do
4453                          * to disable replay of the journal when we next remount
4454                          */
4455                         sb->s_flags |= MS_RDONLY;
4456
4457                         /*
4458                          * OK, test if we are remounting a valid rw partition
4459                          * readonly, and if so set the rdonly flag and then
4460                          * mark the partition as valid again.
4461                          */
4462                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4463                             (sbi->s_mount_state & EXT4_VALID_FS))
4464                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4465
4466                         if (sbi->s_journal)
4467                                 ext4_mark_recovery_complete(sb, es);
4468                 } else {
4469                         /* Make sure we can mount this feature set readwrite */
4470                         if (!ext4_feature_set_ok(sb, 0)) {
4471                                 err = -EROFS;
4472                                 goto restore_opts;
4473                         }
4474                         /*
4475                          * Make sure the group descriptor checksums
4476                          * are sane.  If they aren't, refuse to remount r/w.
4477                          */
4478                         for (g = 0; g < sbi->s_groups_count; g++) {
4479                                 struct ext4_group_desc *gdp =
4480                                         ext4_get_group_desc(sb, g, NULL);
4481
4482                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4483                                         ext4_msg(sb, KERN_ERR,
4484                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4485                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4486                                                le16_to_cpu(gdp->bg_checksum));
4487                                         err = -EINVAL;
4488                                         goto restore_opts;
4489                                 }
4490                         }
4491
4492                         /*
4493                          * If we have an unprocessed orphan list hanging
4494                          * around from a previously readonly bdev mount,
4495                          * require a full umount/remount for now.
4496                          */
4497                         if (es->s_last_orphan) {
4498                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4499                                        "remount RDWR because of unprocessed "
4500                                        "orphan inode list.  Please "
4501                                        "umount/remount instead");
4502                                 err = -EINVAL;
4503                                 goto restore_opts;
4504                         }
4505
4506                         /*
4507                          * Mounting a RDONLY partition read-write, so reread
4508                          * and store the current valid flag.  (It may have
4509                          * been changed by e2fsck since we originally mounted
4510                          * the partition.)
4511                          */
4512                         if (sbi->s_journal)
4513                                 ext4_clear_journal_err(sb, es);
4514                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4515                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4516                                 goto restore_opts;
4517                         if (!ext4_setup_super(sb, es, 0))
4518                                 sb->s_flags &= ~MS_RDONLY;
4519                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4520                                                      EXT4_FEATURE_INCOMPAT_MMP))
4521                                 if (ext4_multi_mount_protect(sb,
4522                                                 le64_to_cpu(es->s_mmp_block))) {
4523                                         err = -EROFS;
4524                                         goto restore_opts;
4525                                 }
4526                         enable_quota = 1;
4527                 }
4528         }
4529
4530         /*
4531          * Reinitialize lazy itable initialization thread based on
4532          * current settings
4533          */
4534         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4535                 ext4_unregister_li_request(sb);
4536         else {
4537                 ext4_group_t first_not_zeroed;
4538                 first_not_zeroed = ext4_has_uninit_itable(sb);
4539                 ext4_register_li_request(sb, first_not_zeroed);
4540         }
4541
4542         ext4_setup_system_zone(sb);
4543         if (sbi->s_journal == NULL)
4544                 ext4_commit_super(sb, 1);
4545
4546 #ifdef CONFIG_QUOTA
4547         /* Release old quota file names */
4548         for (i = 0; i < MAXQUOTAS; i++)
4549                 if (old_opts.s_qf_names[i] &&
4550                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4551                         kfree(old_opts.s_qf_names[i]);
4552 #endif
4553         unlock_super(sb);
4554         if (enable_quota)
4555                 dquot_resume(sb, -1);
4556
4557         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4558         kfree(orig_data);
4559         return 0;
4560
4561 restore_opts:
4562         sb->s_flags = old_sb_flags;
4563         sbi->s_mount_opt = old_opts.s_mount_opt;
4564         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4565         sbi->s_resuid = old_opts.s_resuid;
4566         sbi->s_resgid = old_opts.s_resgid;
4567         sbi->s_commit_interval = old_opts.s_commit_interval;
4568         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4569         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4570 #ifdef CONFIG_QUOTA
4571         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4572         for (i = 0; i < MAXQUOTAS; i++) {
4573                 if (sbi->s_qf_names[i] &&
4574                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4575                         kfree(sbi->s_qf_names[i]);
4576                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4577         }
4578 #endif
4579         unlock_super(sb);
4580         kfree(orig_data);
4581         return err;
4582 }
4583
4584 /*
4585  * Note: calculating the overhead so we can be compatible with
4586  * historical BSD practice is quite difficult in the face of
4587  * clusters/bigalloc.  This is because multiple metadata blocks from
4588  * different block group can end up in the same allocation cluster.
4589  * Calculating the exact overhead in the face of clustered allocation
4590  * requires either O(all block bitmaps) in memory or O(number of block
4591  * groups**2) in time.  We will still calculate the superblock for
4592  * older file systems --- and if we come across with a bigalloc file
4593  * system with zero in s_overhead_clusters the estimate will be close to
4594  * correct especially for very large cluster sizes --- but for newer
4595  * file systems, it's better to calculate this figure once at mkfs
4596  * time, and store it in the superblock.  If the superblock value is
4597  * present (even for non-bigalloc file systems), we will use it.
4598  */
4599 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4600 {
4601         struct super_block *sb = dentry->d_sb;
4602         struct ext4_sb_info *sbi = EXT4_SB(sb);
4603         struct ext4_super_block *es = sbi->s_es;
4604         struct ext4_group_desc *gdp;
4605         u64 fsid;
4606         s64 bfree;
4607
4608         if (test_opt(sb, MINIX_DF)) {
4609                 sbi->s_overhead_last = 0;
4610         } else if (es->s_overhead_clusters) {
4611                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4612         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4613                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4614                 ext4_fsblk_t overhead = 0;
4615
4616                 /*
4617                  * Compute the overhead (FS structures).  This is constant
4618                  * for a given filesystem unless the number of block groups
4619                  * changes so we cache the previous value until it does.
4620                  */
4621
4622                 /*
4623                  * All of the blocks before first_data_block are
4624                  * overhead
4625                  */
4626                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4627
4628                 /*
4629                  * Add the overhead found in each block group
4630                  */
4631                 for (i = 0; i < ngroups; i++) {
4632                         gdp = ext4_get_group_desc(sb, i, NULL);
4633                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4634                         cond_resched();
4635                 }
4636                 sbi->s_overhead_last = overhead;
4637                 smp_wmb();
4638                 sbi->s_blocks_last = ext4_blocks_count(es);
4639         }
4640
4641         buf->f_type = EXT4_SUPER_MAGIC;
4642         buf->f_bsize = sb->s_blocksize;
4643         buf->f_blocks = (ext4_blocks_count(es) -
4644                          EXT4_C2B(sbi, sbi->s_overhead_last));
4645         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4646                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4647         /* prevent underflow in case that few free space is available */
4648         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4649         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4650         if (buf->f_bfree < ext4_r_blocks_count(es))
4651                 buf->f_bavail = 0;
4652         buf->f_files = le32_to_cpu(es->s_inodes_count);
4653         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4654         buf->f_namelen = EXT4_NAME_LEN;
4655         fsid = le64_to_cpup((void *)es->s_uuid) ^
4656                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4657         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4658         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4659
4660         return 0;
4661 }
4662
4663 /* Helper function for writing quotas on sync - we need to start transaction
4664  * before quota file is locked for write. Otherwise the are possible deadlocks:
4665  * Process 1                         Process 2
4666  * ext4_create()                     quota_sync()
4667  *   jbd2_journal_start()                  write_dquot()
4668  *   dquot_initialize()                         down(dqio_mutex)
4669  *     down(dqio_mutex)                    jbd2_journal_start()
4670  *
4671  */
4672
4673 #ifdef CONFIG_QUOTA
4674
4675 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4676 {
4677         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4678 }
4679
4680 static int ext4_write_dquot(struct dquot *dquot)
4681 {
4682         int ret, err;
4683         handle_t *handle;
4684         struct inode *inode;
4685
4686         inode = dquot_to_inode(dquot);
4687         handle = ext4_journal_start(inode,
4688                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4689         if (IS_ERR(handle))
4690                 return PTR_ERR(handle);
4691         ret = dquot_commit(dquot);
4692         err = ext4_journal_stop(handle);
4693         if (!ret)
4694                 ret = err;
4695         return ret;
4696 }
4697
4698 static int ext4_acquire_dquot(struct dquot *dquot)
4699 {
4700         int ret, err;
4701         handle_t *handle;
4702
4703         handle = ext4_journal_start(dquot_to_inode(dquot),
4704                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4705         if (IS_ERR(handle))
4706                 return PTR_ERR(handle);
4707         ret = dquot_acquire(dquot);
4708         err = ext4_journal_stop(handle);
4709         if (!ret)
4710                 ret = err;
4711         return ret;
4712 }
4713
4714 static int ext4_release_dquot(struct dquot *dquot)
4715 {
4716         int ret, err;
4717         handle_t *handle;
4718
4719         handle = ext4_journal_start(dquot_to_inode(dquot),
4720                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4721         if (IS_ERR(handle)) {
4722                 /* Release dquot anyway to avoid endless cycle in dqput() */
4723                 dquot_release(dquot);
4724                 return PTR_ERR(handle);
4725         }
4726         ret = dquot_release(dquot);
4727         err = ext4_journal_stop(handle);
4728         if (!ret)
4729                 ret = err;
4730         return ret;
4731 }
4732
4733 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4734 {
4735         /* Are we journaling quotas? */
4736         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4737             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4738                 dquot_mark_dquot_dirty(dquot);
4739                 return ext4_write_dquot(dquot);
4740         } else {
4741                 return dquot_mark_dquot_dirty(dquot);
4742         }
4743 }
4744
4745 static int ext4_write_info(struct super_block *sb, int type)
4746 {
4747         int ret, err;
4748         handle_t *handle;
4749
4750         /* Data block + inode block */
4751         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4752         if (IS_ERR(handle))
4753                 return PTR_ERR(handle);
4754         ret = dquot_commit_info(sb, type);
4755         err = ext4_journal_stop(handle);
4756         if (!ret)
4757                 ret = err;
4758         return ret;
4759 }
4760
4761 /*
4762  * Turn on quotas during mount time - we need to find
4763  * the quota file and such...
4764  */
4765 static int ext4_quota_on_mount(struct super_block *sb, int type)
4766 {
4767         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4768                                         EXT4_SB(sb)->s_jquota_fmt, type);
4769 }
4770
4771 /*
4772  * Standard function to be called on quota_on
4773  */
4774 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4775                          struct path *path)
4776 {
4777         int err;
4778
4779         if (!test_opt(sb, QUOTA))
4780                 return -EINVAL;
4781
4782         /* Quotafile not on the same filesystem? */
4783         if (path->mnt->mnt_sb != sb)
4784                 return -EXDEV;
4785         /* Journaling quota? */
4786         if (EXT4_SB(sb)->s_qf_names[type]) {
4787                 /* Quotafile not in fs root? */
4788                 if (path->dentry->d_parent != sb->s_root)
4789                         ext4_msg(sb, KERN_WARNING,
4790                                 "Quota file not on filesystem root. "
4791                                 "Journaled quota will not work");
4792         }
4793
4794         /*
4795          * When we journal data on quota file, we have to flush journal to see
4796          * all updates to the file when we bypass pagecache...
4797          */
4798         if (EXT4_SB(sb)->s_journal &&
4799             ext4_should_journal_data(path->dentry->d_inode)) {
4800                 /*
4801                  * We don't need to lock updates but journal_flush() could
4802                  * otherwise be livelocked...
4803                  */
4804                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4805                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4806                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4807                 if (err)
4808                         return err;
4809         }
4810
4811         return dquot_quota_on(sb, type, format_id, path);
4812 }
4813
4814 static int ext4_quota_off(struct super_block *sb, int type)
4815 {
4816         struct inode *inode = sb_dqopt(sb)->files[type];
4817         handle_t *handle;
4818
4819         /* Force all delayed allocation blocks to be allocated.
4820          * Caller already holds s_umount sem */
4821         if (test_opt(sb, DELALLOC))
4822                 sync_filesystem(sb);
4823
4824         if (!inode)
4825                 goto out;
4826
4827         /* Update modification times of quota files when userspace can
4828          * start looking at them */
4829         handle = ext4_journal_start(inode, 1);
4830         if (IS_ERR(handle))
4831                 goto out;
4832         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4833         ext4_mark_inode_dirty(handle, inode);
4834         ext4_journal_stop(handle);
4835
4836 out:
4837         return dquot_quota_off(sb, type);
4838 }
4839
4840 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4841  * acquiring the locks... As quota files are never truncated and quota code
4842  * itself serializes the operations (and no one else should touch the files)
4843  * we don't have to be afraid of races */
4844 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4845                                size_t len, loff_t off)
4846 {
4847         struct inode *inode = sb_dqopt(sb)->files[type];
4848         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4849         int err = 0;
4850         int offset = off & (sb->s_blocksize - 1);
4851         int tocopy;
4852         size_t toread;
4853         struct buffer_head *bh;
4854         loff_t i_size = i_size_read(inode);
4855
4856         if (off > i_size)
4857                 return 0;
4858         if (off+len > i_size)
4859                 len = i_size-off;
4860         toread = len;
4861         while (toread > 0) {
4862                 tocopy = sb->s_blocksize - offset < toread ?
4863                                 sb->s_blocksize - offset : toread;
4864                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4865                 if (err)
4866                         return err;
4867                 if (!bh)        /* A hole? */
4868                         memset(data, 0, tocopy);
4869                 else
4870                         memcpy(data, bh->b_data+offset, tocopy);
4871                 brelse(bh);
4872                 offset = 0;
4873                 toread -= tocopy;
4874                 data += tocopy;
4875                 blk++;
4876         }
4877         return len;
4878 }
4879
4880 /* Write to quotafile (we know the transaction is already started and has
4881  * enough credits) */
4882 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4883                                 const char *data, size_t len, loff_t off)
4884 {
4885         struct inode *inode = sb_dqopt(sb)->files[type];
4886         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4887         int err = 0;
4888         int offset = off & (sb->s_blocksize - 1);
4889         struct buffer_head *bh;
4890         handle_t *handle = journal_current_handle();
4891
4892         if (EXT4_SB(sb)->s_journal && !handle) {
4893                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4894                         " cancelled because transaction is not started",
4895                         (unsigned long long)off, (unsigned long long)len);
4896                 return -EIO;
4897         }
4898         /*
4899          * Since we account only one data block in transaction credits,
4900          * then it is impossible to cross a block boundary.
4901          */
4902         if (sb->s_blocksize - offset < len) {
4903                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4904                         " cancelled because not block aligned",
4905                         (unsigned long long)off, (unsigned long long)len);
4906                 return -EIO;
4907         }
4908
4909         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4910         bh = ext4_bread(handle, inode, blk, 1, &err);
4911         if (!bh)
4912                 goto out;
4913         err = ext4_journal_get_write_access(handle, bh);
4914         if (err) {
4915                 brelse(bh);
4916                 goto out;
4917         }
4918         lock_buffer(bh);
4919         memcpy(bh->b_data+offset, data, len);
4920         flush_dcache_page(bh->b_page);
4921         unlock_buffer(bh);
4922         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4923         brelse(bh);
4924 out:
4925         if (err) {
4926                 mutex_unlock(&inode->i_mutex);
4927                 return err;
4928         }
4929         if (inode->i_size < off + len) {
4930                 i_size_write(inode, off + len);
4931                 EXT4_I(inode)->i_disksize = inode->i_size;
4932                 ext4_mark_inode_dirty(handle, inode);
4933         }
4934         mutex_unlock(&inode->i_mutex);
4935         return len;
4936 }
4937
4938 #endif
4939
4940 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4941                        const char *dev_name, void *data)
4942 {
4943         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4944 }
4945
4946 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4947 static inline void register_as_ext2(void)
4948 {
4949         int err = register_filesystem(&ext2_fs_type);
4950         if (err)
4951                 printk(KERN_WARNING
4952                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4953 }
4954
4955 static inline void unregister_as_ext2(void)
4956 {
4957         unregister_filesystem(&ext2_fs_type);
4958 }
4959
4960 static inline int ext2_feature_set_ok(struct super_block *sb)
4961 {
4962         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4963                 return 0;
4964         if (sb->s_flags & MS_RDONLY)
4965                 return 1;
4966         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4967                 return 0;
4968         return 1;
4969 }
4970 MODULE_ALIAS("ext2");
4971 #else
4972 static inline void register_as_ext2(void) { }
4973 static inline void unregister_as_ext2(void) { }
4974 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4975 #endif
4976
4977 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4978 static inline void register_as_ext3(void)
4979 {
4980         int err = register_filesystem(&ext3_fs_type);
4981         if (err)
4982                 printk(KERN_WARNING
4983                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4984 }
4985
4986 static inline void unregister_as_ext3(void)
4987 {
4988         unregister_filesystem(&ext3_fs_type);
4989 }
4990
4991 static inline int ext3_feature_set_ok(struct super_block *sb)
4992 {
4993         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4994                 return 0;
4995         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4996                 return 0;
4997         if (sb->s_flags & MS_RDONLY)
4998                 return 1;
4999         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5000                 return 0;
5001         return 1;
5002 }
5003 MODULE_ALIAS("ext3");
5004 #else
5005 static inline void register_as_ext3(void) { }
5006 static inline void unregister_as_ext3(void) { }
5007 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5008 #endif
5009
5010 static struct file_system_type ext4_fs_type = {
5011         .owner          = THIS_MODULE,
5012         .name           = "ext4",
5013         .mount          = ext4_mount,
5014         .kill_sb        = kill_block_super,
5015         .fs_flags       = FS_REQUIRES_DEV,
5016 };
5017
5018 static int __init ext4_init_feat_adverts(void)
5019 {
5020         struct ext4_features *ef;
5021         int ret = -ENOMEM;
5022
5023         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5024         if (!ef)
5025                 goto out;
5026
5027         ef->f_kobj.kset = ext4_kset;
5028         init_completion(&ef->f_kobj_unregister);
5029         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5030                                    "features");
5031         if (ret) {
5032                 kfree(ef);
5033                 goto out;
5034         }
5035
5036         ext4_feat = ef;
5037         ret = 0;
5038 out:
5039         return ret;
5040 }
5041
5042 static void ext4_exit_feat_adverts(void)
5043 {
5044         kobject_put(&ext4_feat->f_kobj);
5045         wait_for_completion(&ext4_feat->f_kobj_unregister);
5046         kfree(ext4_feat);
5047 }
5048
5049 /* Shared across all ext4 file systems */
5050 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5051 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5052
5053 static int __init ext4_init_fs(void)
5054 {
5055         int i, err;
5056
5057         ext4_check_flag_values();
5058
5059         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5060                 mutex_init(&ext4__aio_mutex[i]);
5061                 init_waitqueue_head(&ext4__ioend_wq[i]);
5062         }
5063
5064         err = ext4_init_pageio();
5065         if (err)
5066                 return err;
5067         err = ext4_init_system_zone();
5068         if (err)
5069                 goto out7;
5070         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5071         if (!ext4_kset)
5072                 goto out6;
5073         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5074         if (!ext4_proc_root)
5075                 goto out5;
5076
5077         err = ext4_init_feat_adverts();
5078         if (err)
5079                 goto out4;
5080
5081         err = ext4_init_mballoc();
5082         if (err)
5083                 goto out3;
5084
5085         err = ext4_init_xattr();
5086         if (err)
5087                 goto out2;
5088         err = init_inodecache();
5089         if (err)
5090                 goto out1;
5091         register_as_ext3();
5092         register_as_ext2();
5093         err = register_filesystem(&ext4_fs_type);
5094         if (err)
5095                 goto out;
5096
5097         ext4_li_info = NULL;
5098         mutex_init(&ext4_li_mtx);
5099         return 0;
5100 out:
5101         unregister_as_ext2();
5102         unregister_as_ext3();
5103         destroy_inodecache();
5104 out1:
5105         ext4_exit_xattr();
5106 out2:
5107         ext4_exit_mballoc();
5108 out3:
5109         ext4_exit_feat_adverts();
5110 out4:
5111         remove_proc_entry("fs/ext4", NULL);
5112 out5:
5113         kset_unregister(ext4_kset);
5114 out6:
5115         ext4_exit_system_zone();
5116 out7:
5117         ext4_exit_pageio();
5118         return err;
5119 }
5120
5121 static void __exit ext4_exit_fs(void)
5122 {
5123         ext4_destroy_lazyinit_thread();
5124         unregister_as_ext2();
5125         unregister_as_ext3();
5126         unregister_filesystem(&ext4_fs_type);
5127         destroy_inodecache();
5128         ext4_exit_xattr();
5129         ext4_exit_mballoc();
5130         ext4_exit_feat_adverts();
5131         remove_proc_entry("fs/ext4", NULL);
5132         kset_unregister(ext4_kset);
5133         ext4_exit_system_zone();
5134         ext4_exit_pageio();
5135 }
5136
5137 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5138 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5139 MODULE_LICENSE("GPL");
5140 module_init(ext4_init_fs)
5141 module_exit(ext4_exit_fs)