fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "richacl.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct proc_dir_entry *ext4_proc_root;
  59 static struct kset *ext4_kset;
  60 static struct ext4_lazy_init *ext4_li_info;
  61 static struct mutex ext4_li_mtx;
  62 static struct ext4_features *ext4_feat;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static const char *ext4_decode_error(struct super_block *sb, int errno,
  74                                      char nbuf[16]);
  75 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  76 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  77 static int ext4_unfreeze(struct super_block *sb);
  78 static void ext4_write_super(struct super_block *sb);
  79 static int ext4_freeze(struct super_block *sb);
  80 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  81                        const char *dev_name, void *data);
  82 static inline int ext2_feature_set_ok(struct super_block *sb);
  83 static inline int ext3_feature_set_ok(struct super_block *sb);
  84 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  85 static void ext4_destroy_lazyinit_thread(void);
  86 static void ext4_unregister_li_request(struct super_block *sb);
  87 static void ext4_clear_request_list(void);
  88
  89 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  90 static struct file_system_type ext2_fs_type = {
  91         .owner          = THIS_MODULE,
  92         .name           = "ext2",
  93         .mount          = ext4_mount,
  94         .kill_sb        = kill_block_super,
  95         .fs_flags       = FS_REQUIRES_DEV,
  96 };
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 112 #else
 113 #define IS_EXT3_SB(sb) (0)
 114 #endif
 115
 116 void *ext4_kvmalloc(size_t size, gfp_t flags)
 117 {
 118         void *ret;
 119
 120         ret = kmalloc(size, flags);
 121         if (!ret)
 122                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 123         return ret;
 124 }
 125
 126 void *ext4_kvzalloc(size_t size, gfp_t flags)
 127 {
 128         void *ret;
 129
 130         ret = kzalloc(size, flags);
 131         if (!ret)
 132                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 133         return ret;
 134 }
 135
 136 void ext4_kvfree(void *ptr)
 137 {
 138         if (is_vmalloc_addr(ptr))
 139                 vfree(ptr);
 140         else
 141                 kfree(ptr);
 142
 143 }
 144
 145 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 146                                struct ext4_group_desc *bg)
 147 {
 148         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 149                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 150                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 151 }
 152
 153 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 154                                struct ext4_group_desc *bg)
 155 {
 156         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 157                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 158                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 159 }
 160
 161 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 162                               struct ext4_group_desc *bg)
 163 {
 164         return le32_to_cpu(bg->bg_inode_table_lo) |
 165                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 166                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 167 }
 168
 169 __u32 ext4_free_group_clusters(struct super_block *sb,
 170                                struct ext4_group_desc *bg)
 171 {
 172         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 173                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 174                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 175 }
 176
 177 __u32 ext4_free_inodes_count(struct super_block *sb,
 178                               struct ext4_group_desc *bg)
 179 {
 180         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 181                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 182                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 183 }
 184
 185 __u32 ext4_used_dirs_count(struct super_block *sb,
 186                               struct ext4_group_desc *bg)
 187 {
 188         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 189                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 190                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 191 }
 192
 193 __u32 ext4_itable_unused_count(struct super_block *sb,
 194                               struct ext4_group_desc *bg)
 195 {
 196         return le16_to_cpu(bg->bg_itable_unused_lo) |
 197                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 198                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 199 }
 200
 201 void ext4_block_bitmap_set(struct super_block *sb,
 202                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 203 {
 204         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 205         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 206                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 207 }
 208
 209 void ext4_inode_bitmap_set(struct super_block *sb,
 210                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 211 {
 212         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 213         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 214                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 215 }
 216
 217 void ext4_inode_table_set(struct super_block *sb,
 218                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 219 {
 220         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 221         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 222                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 223 }
 224
 225 void ext4_free_group_clusters_set(struct super_block *sb,
 226                                   struct ext4_group_desc *bg, __u32 count)
 227 {
 228         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 229         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 230                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 231 }
 232
 233 void ext4_free_inodes_set(struct super_block *sb,
 234                           struct ext4_group_desc *bg, __u32 count)
 235 {
 236         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 237         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 238                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 239 }
 240
 241 void ext4_used_dirs_set(struct super_block *sb,
 242                           struct ext4_group_desc *bg, __u32 count)
 243 {
 244         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 245         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 246                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 247 }
 248
 249 void ext4_itable_unused_set(struct super_block *sb,
 250                           struct ext4_group_desc *bg, __u32 count)
 251 {
 252         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 253         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 254                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 255 }
 256
 257
 258 /* Just increment the non-pointer handle value */
 259 static handle_t *ext4_get_nojournal(void)
 260 {
 261         handle_t *handle = current->journal_info;
 262         unsigned long ref_cnt = (unsigned long)handle;
 263
 264         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 265
 266         ref_cnt++;
 267         handle = (handle_t *)ref_cnt;
 268
 269         current->journal_info = handle;
 270         return handle;
 271 }
 272
 273
 274 /* Decrement the non-pointer handle value */
 275 static void ext4_put_nojournal(handle_t *handle)
 276 {
 277         unsigned long ref_cnt = (unsigned long)handle;
 278
 279         BUG_ON(ref_cnt == 0);
 280
 281         ref_cnt--;
 282         handle = (handle_t *)ref_cnt;
 283
 284         current->journal_info = handle;
 285 }
 286
 287 /*
 288  * Wrappers for jbd2_journal_start/end.
 289  *
 290  * The only special thing we need to do here is to make sure that all
 291  * journal_end calls result in the superblock being marked dirty, so
 292  * that sync() will call the filesystem's write_super callback if
 293  * appropriate.
 294  *
 295  * To avoid j_barrier hold in userspace when a user calls freeze(),
 296  * ext4 prevents a new handle from being started by s_frozen, which
 297  * is in an upper layer.
 298  */
 299 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 300 {
 301         journal_t *journal;
 302         handle_t  *handle;
 303
 304         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 305         if (sb->s_flags & MS_RDONLY)
 306                 return ERR_PTR(-EROFS);
 307
 308         journal = EXT4_SB(sb)->s_journal;
 309         handle = ext4_journal_current_handle();
 310
 311         /*
 312          * If a handle has been started, it should be allowed to
 313          * finish, otherwise deadlock could happen between freeze
 314          * and others(e.g. truncate) due to the restart of the
 315          * journal handle if the filesystem is forzen and active
 316          * handles are not stopped.
 317          */
 318         if (!handle)
 319                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 320
 321         if (!journal)
 322                 return ext4_get_nojournal();
 323         /*
 324          * Special case here: if the journal has aborted behind our
 325          * backs (eg. EIO in the commit thread), then we still need to
 326          * take the FS itself readonly cleanly.
 327          */
 328         if (is_journal_aborted(journal)) {
 329                 ext4_abort(sb, "Detected aborted journal");
 330                 return ERR_PTR(-EROFS);
 331         }
 332         return jbd2_journal_start(journal, nblocks);
 333 }
 334
 335 /*
 336  * The only special thing we need to do here is to make sure that all
 337  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 338  * that sync() will call the filesystem's write_super callback if
 339  * appropriate.
 340  */
 341 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 342 {
 343         struct super_block *sb;
 344         int err;
 345         int rc;
 346
 347         if (!ext4_handle_valid(handle)) {
 348                 ext4_put_nojournal(handle);
 349                 return 0;
 350         }
 351         sb = handle->h_transaction->t_journal->j_private;
 352         err = handle->h_err;
 353         rc = jbd2_journal_stop(handle);
 354
 355         if (!err)
 356                 err = rc;
 357         if (err)
 358                 __ext4_std_error(sb, where, line, err);
 359         return err;
 360 }
 361
 362 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 363                                const char *err_fn, struct buffer_head *bh,
 364                                handle_t *handle, int err)
 365 {
 366         char nbuf[16];
 367         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 368
 369         BUG_ON(!ext4_handle_valid(handle));
 370
 371         if (bh)
 372                 BUFFER_TRACE(bh, "abort");
 373
 374         if (!handle->h_err)
 375                 handle->h_err = err;
 376
 377         if (is_handle_aborted(handle))
 378                 return;
 379
 380         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 381                caller, line, errstr, err_fn);
 382
 383         jbd2_journal_abort_handle(handle);
 384 }
 385
 386 static void __save_error_info(struct super_block *sb, const char *func,
 387                             unsigned int line)
 388 {
 389         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 390
 391         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 392         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 393         es->s_last_error_time = cpu_to_le32(get_seconds());
 394         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 395         es->s_last_error_line = cpu_to_le32(line);
 396         if (!es->s_first_error_time) {
 397                 es->s_first_error_time = es->s_last_error_time;
 398                 strncpy(es->s_first_error_func, func,
 399                         sizeof(es->s_first_error_func));
 400                 es->s_first_error_line = cpu_to_le32(line);
 401                 es->s_first_error_ino = es->s_last_error_ino;
 402                 es->s_first_error_block = es->s_last_error_block;
 403         }
 404         /*
 405          * Start the daily error reporting function if it hasn't been
 406          * started already
 407          */
 408         if (!es->s_error_count)
 409                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 410         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 411 }
 412
 413 static void save_error_info(struct super_block *sb, const char *func,
 414                             unsigned int line)
 415 {
 416         __save_error_info(sb, func, line);
 417         ext4_commit_super(sb, 1);
 418 }
 419
 420 /*
 421  * The del_gendisk() function uninitializes the disk-specific data
 422  * structures, including the bdi structure, without telling anyone
 423  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 424  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 425  * This is a kludge to prevent these oops until we can put in a proper
 426  * hook in del_gendisk() to inform the VFS and file system layers.
 427  */
 428 static int block_device_ejected(struct super_block *sb)
 429 {
 430         struct inode *bd_inode = sb->s_bdev->bd_inode;
 431         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 432
 433         return bdi->dev == NULL;
 434 }
 435
 436 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 437 {
 438         struct super_block              *sb = journal->j_private;
 439         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 440         int                             error = is_journal_aborted(journal);
 441         struct ext4_journal_cb_entry    *jce, *tmp;
 442
 443         spin_lock(&sbi->s_md_lock);
 444         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 445                 list_del_init(&jce->jce_list);
 446                 spin_unlock(&sbi->s_md_lock);
 447                 jce->jce_func(sb, jce, error);
 448                 spin_lock(&sbi->s_md_lock);
 449         }
 450         spin_unlock(&sbi->s_md_lock);
 451 }
 452
 453 /* Deal with the reporting of failure conditions on a filesystem such as
 454  * inconsistencies detected or read IO failures.
 455  *
 456  * On ext2, we can store the error state of the filesystem in the
 457  * superblock.  That is not possible on ext4, because we may have other
 458  * write ordering constraints on the superblock which prevent us from
 459  * writing it out straight away; and given that the journal is about to
 460  * be aborted, we can't rely on the current, or future, transactions to
 461  * write out the superblock safely.
 462  *
 463  * We'll just use the jbd2_journal_abort() error code to record an error in
 464  * the journal instead.  On recovery, the journal will complain about
 465  * that error until we've noted it down and cleared it.
 466  */
 467
 468 static void ext4_handle_error(struct super_block *sb)
 469 {
 470         if (sb->s_flags & MS_RDONLY)
 471                 return;
 472
 473         if (!test_opt(sb, ERRORS_CONT)) {
 474                 journal_t *journal = EXT4_SB(sb)->s_journal;
 475
 476                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 477                 if (journal)
 478                         jbd2_journal_abort(journal, -EIO);
 479         }
 480         if (test_opt(sb, ERRORS_RO)) {
 481                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 482                 sb->s_flags |= MS_RDONLY;
 483         }
 484         if (test_opt(sb, ERRORS_PANIC))
 485                 panic("EXT4-fs (device %s): panic forced after error\n",
 486                         sb->s_id);
 487 }
 488
 489 void __ext4_error(struct super_block *sb, const char *function,
 490                   unsigned int line, const char *fmt, ...)
 491 {
 492         struct va_format vaf;
 493         va_list args;
 494
 495         va_start(args, fmt);
 496         vaf.fmt = fmt;
 497         vaf.va = &args;
 498         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 499                sb->s_id, function, line, current->comm, &vaf);
 500         va_end(args);
 501
 502         ext4_handle_error(sb);
 503 }
 504
 505 void ext4_error_inode(struct inode *inode, const char *function,
 506                       unsigned int line, ext4_fsblk_t block,
 507                       const char *fmt, ...)
 508 {
 509         va_list args;
 510         struct va_format vaf;
 511         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 512
 513         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 514         es->s_last_error_block = cpu_to_le64(block);
 515         save_error_info(inode->i_sb, function, line);
 516         va_start(args, fmt);
 517         vaf.fmt = fmt;
 518         vaf.va = &args;
 519         if (block)
 520                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 521                        "inode #%lu: block %llu: comm %s: %pV\n",
 522                        inode->i_sb->s_id, function, line, inode->i_ino,
 523                        block, current->comm, &vaf);
 524         else
 525                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 526                        "inode #%lu: comm %s: %pV\n",
 527                        inode->i_sb->s_id, function, line, inode->i_ino,
 528                        current->comm, &vaf);
 529         va_end(args);
 530
 531         ext4_handle_error(inode->i_sb);
 532 }
 533
 534 void ext4_error_file(struct file *file, const char *function,
 535                      unsigned int line, ext4_fsblk_t block,
 536                      const char *fmt, ...)
 537 {
 538         va_list args;
 539         struct va_format vaf;
 540         struct ext4_super_block *es;
 541         struct inode *inode = file->f_dentry->d_inode;
 542         char pathname[80], *path;
 543
 544         es = EXT4_SB(inode->i_sb)->s_es;
 545         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 546         save_error_info(inode->i_sb, function, line);
 547         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 548         if (IS_ERR(path))
 549                 path = "(unknown)";
 550         va_start(args, fmt);
 551         vaf.fmt = fmt;
 552         vaf.va = &args;
 553         if (block)
 554                 printk(KERN_CRIT
 555                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 556                        "block %llu: comm %s: path %s: %pV\n",
 557                        inode->i_sb->s_id, function, line, inode->i_ino,
 558                        block, current->comm, path, &vaf);
 559         else
 560                 printk(KERN_CRIT
 561                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 562                        "comm %s: path %s: %pV\n",
 563                        inode->i_sb->s_id, function, line, inode->i_ino,
 564                        current->comm, path, &vaf);
 565         va_end(args);
 566
 567         ext4_handle_error(inode->i_sb);
 568 }
 569
 570 static const char *ext4_decode_error(struct super_block *sb, int errno,
 571                                      char nbuf[16])
 572 {
 573         char *errstr = NULL;
 574
 575         switch (errno) {
 576         case -EIO:
 577                 errstr = "IO failure";
 578                 break;
 579         case -ENOMEM:
 580                 errstr = "Out of memory";
 581                 break;
 582         case -EROFS:
 583                 if (!sb || (EXT4_SB(sb)->s_journal &&
 584                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 585                         errstr = "Journal has aborted";
 586                 else
 587                         errstr = "Readonly filesystem";
 588                 break;
 589         default:
 590                 /* If the caller passed in an extra buffer for unknown
 591                  * errors, textualise them now.  Else we just return
 592                  * NULL. */
 593                 if (nbuf) {
 594                         /* Check for truncated error codes... */
 595                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 596                                 errstr = nbuf;
 597                 }
 598                 break;
 599         }
 600
 601         return errstr;
 602 }
 603
 604 /* __ext4_std_error decodes expected errors from journaling functions
 605  * automatically and invokes the appropriate error response.  */
 606
 607 void __ext4_std_error(struct super_block *sb, const char *function,
 608                       unsigned int line, int errno)
 609 {
 610         char nbuf[16];
 611         const char *errstr;
 612
 613         /* Special case: if the error is EROFS, and we're not already
 614          * inside a transaction, then there's really no point in logging
 615          * an error. */
 616         if (errno == -EROFS && journal_current_handle() == NULL &&
 617             (sb->s_flags & MS_RDONLY))
 618                 return;
 619
 620         errstr = ext4_decode_error(sb, errno, nbuf);
 621         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 622                sb->s_id, function, line, errstr);
 623         save_error_info(sb, function, line);
 624
 625         ext4_handle_error(sb);
 626 }
 627
 628 /*
 629  * ext4_abort is a much stronger failure handler than ext4_error.  The
 630  * abort function may be used to deal with unrecoverable failures such
 631  * as journal IO errors or ENOMEM at a critical moment in log management.
 632  *
 633  * We unconditionally force the filesystem into an ABORT|READONLY state,
 634  * unless the error response on the fs has been set to panic in which
 635  * case we take the easy way out and panic immediately.
 636  */
 637
 638 void __ext4_abort(struct super_block *sb, const char *function,
 639                 unsigned int line, const char *fmt, ...)
 640 {
 641         va_list args;
 642
 643         save_error_info(sb, function, line);
 644         va_start(args, fmt);
 645         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 646                function, line);
 647         vprintk(fmt, args);
 648         printk("\n");
 649         va_end(args);
 650
 651         if ((sb->s_flags & MS_RDONLY) == 0) {
 652                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 653                 sb->s_flags |= MS_RDONLY;
 654                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 655                 if (EXT4_SB(sb)->s_journal)
 656                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 657                 save_error_info(sb, function, line);
 658         }
 659         if (test_opt(sb, ERRORS_PANIC))
 660                 panic("EXT4-fs panic from previous error\n");
 661 }
 662
 663 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 664 {
 665         struct va_format vaf;
 666         va_list args;
 667
 668         va_start(args, fmt);
 669         vaf.fmt = fmt;
 670         vaf.va = &args;
 671         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 672         va_end(args);
 673 }
 674
 675 void __ext4_warning(struct super_block *sb, const char *function,
 676                     unsigned int line, const char *fmt, ...)
 677 {
 678         struct va_format vaf;
 679         va_list args;
 680
 681         va_start(args, fmt);
 682         vaf.fmt = fmt;
 683         vaf.va = &args;
 684         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 685                sb->s_id, function, line, &vaf);
 686         va_end(args);
 687 }
 688
 689 void __ext4_grp_locked_error(const char *function, unsigned int line,
 690                              struct super_block *sb, ext4_group_t grp,
 691                              unsigned long ino, ext4_fsblk_t block,
 692                              const char *fmt, ...)
 693 __releases(bitlock)
 694 __acquires(bitlock)
 695 {
 696         struct va_format vaf;
 697         va_list args;
 698         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 699
 700         es->s_last_error_ino = cpu_to_le32(ino);
 701         es->s_last_error_block = cpu_to_le64(block);
 702         __save_error_info(sb, function, line);
 703
 704         va_start(args, fmt);
 705
 706         vaf.fmt = fmt;
 707         vaf.va = &args;
 708         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 709                sb->s_id, function, line, grp);
 710         if (ino)
 711                 printk(KERN_CONT "inode %lu: ", ino);
 712         if (block)
 713                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 714         printk(KERN_CONT "%pV\n", &vaf);
 715         va_end(args);
 716
 717         if (test_opt(sb, ERRORS_CONT)) {
 718                 ext4_commit_super(sb, 0);
 719                 return;
 720         }
 721
 722         ext4_unlock_group(sb, grp);
 723         ext4_handle_error(sb);
 724         /*
 725          * We only get here in the ERRORS_RO case; relocking the group
 726          * may be dangerous, but nothing bad will happen since the
 727          * filesystem will have already been marked read/only and the
 728          * journal has been aborted.  We return 1 as a hint to callers
 729          * who might what to use the return value from
 730          * ext4_grp_locked_error() to distinguish between the
 731          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 732          * aggressively from the ext4 function in question, with a
 733          * more appropriate error code.
 734          */
 735         ext4_lock_group(sb, grp);
 736         return;
 737 }
 738
 739 void ext4_update_dynamic_rev(struct super_block *sb)
 740 {
 741         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 742
 743         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 744                 return;
 745
 746         ext4_warning(sb,
 747                      "updating to rev %d because of new feature flag, "
 748                      "running e2fsck is recommended",
 749                      EXT4_DYNAMIC_REV);
 750
 751         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 752         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 753         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 754         /* leave es->s_feature_*compat flags alone */
 755         /* es->s_uuid will be set by e2fsck if empty */
 756
 757         /*
 758          * The rest of the superblock fields should be zero, and if not it
 759          * means they are likely already in use, so leave them alone.  We
 760          * can leave it up to e2fsck to clean up any inconsistencies there.
 761          */
 762 }
 763
 764 /*
 765  * Open the external journal device
 766  */
 767 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 768 {
 769         struct block_device *bdev;
 770         char b[BDEVNAME_SIZE];
 771
 772         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 773         if (IS_ERR(bdev))
 774                 goto fail;
 775         return bdev;
 776
 777 fail:
 778         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 779                         __bdevname(dev, b), PTR_ERR(bdev));
 780         return NULL;
 781 }
 782
 783 /*
 784  * Release the journal device
 785  */
 786 static int ext4_blkdev_put(struct block_device *bdev)
 787 {
 788         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 789 }
 790
 791 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 792 {
 793         struct block_device *bdev;
 794         int ret = -ENODEV;
 795
 796         bdev = sbi->journal_bdev;
 797         if (bdev) {
 798                 ret = ext4_blkdev_put(bdev);
 799                 sbi->journal_bdev = NULL;
 800         }
 801         return ret;
 802 }
 803
 804 static inline struct inode *orphan_list_entry(struct list_head *l)
 805 {
 806         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 807 }
 808
 809 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 810 {
 811         struct list_head *l;
 812
 813         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 814                  le32_to_cpu(sbi->s_es->s_last_orphan));
 815
 816         printk(KERN_ERR "sb_info orphan list:\n");
 817         list_for_each(l, &sbi->s_orphan) {
 818                 struct inode *inode = orphan_list_entry(l);
 819                 printk(KERN_ERR "  "
 820                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 821                        inode->i_sb->s_id, inode->i_ino, inode,
 822                        inode->i_mode, inode->i_nlink,
 823                        NEXT_ORPHAN(inode));
 824         }
 825 }
 826
 827 static void ext4_put_super(struct super_block *sb)
 828 {
 829         struct ext4_sb_info *sbi = EXT4_SB(sb);
 830         struct ext4_super_block *es = sbi->s_es;
 831         int i, err;
 832
 833         ext4_unregister_li_request(sb);
 834         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 835
 836         flush_workqueue(sbi->dio_unwritten_wq);
 837         destroy_workqueue(sbi->dio_unwritten_wq);
 838
 839         lock_super(sb);
 840         if (sbi->s_journal) {
 841                 err = jbd2_journal_destroy(sbi->s_journal);
 842                 sbi->s_journal = NULL;
 843                 if (err < 0)
 844                         ext4_abort(sb, "Couldn't clean up the journal");
 845         }
 846
 847         del_timer(&sbi->s_err_report);
 848         ext4_release_system_zone(sb);
 849         ext4_mb_release(sb);
 850         ext4_ext_release(sb);
 851         ext4_xattr_put_super(sb);
 852
 853         if (!(sb->s_flags & MS_RDONLY)) {
 854                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 855                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 856         }
 857         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
 858                 ext4_commit_super(sb, 1);
 859
 860         if (sbi->s_proc) {
 861                 remove_proc_entry("options", sbi->s_proc);
 862                 remove_proc_entry(sb->s_id, ext4_proc_root);
 863         }
 864         kobject_del(&sbi->s_kobj);
 865
 866         for (i = 0; i < sbi->s_gdb_count; i++)
 867                 brelse(sbi->s_group_desc[i]);
 868         ext4_kvfree(sbi->s_group_desc);
 869         ext4_kvfree(sbi->s_flex_groups);
 870         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 871         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 872         percpu_counter_destroy(&sbi->s_dirs_counter);
 873         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 874         brelse(sbi->s_sbh);
 875 #ifdef CONFIG_QUOTA
 876         for (i = 0; i < MAXQUOTAS; i++)
 877                 kfree(sbi->s_qf_names[i]);
 878 #endif
 879
 880         /* Debugging code just in case the in-memory inode orphan list
 881          * isn't empty.  The on-disk one can be non-empty if we've
 882          * detected an error and taken the fs readonly, but the
 883          * in-memory list had better be clean by this point. */
 884         if (!list_empty(&sbi->s_orphan))
 885                 dump_orphan_list(sb, sbi);
 886         J_ASSERT(list_empty(&sbi->s_orphan));
 887
 888         invalidate_bdev(sb->s_bdev);
 889         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 890                 /*
 891                  * Invalidate the journal device's buffers.  We don't want them
 892                  * floating about in memory - the physical journal device may
 893                  * hotswapped, and it breaks the `ro-after' testing code.
 894                  */
 895                 sync_blockdev(sbi->journal_bdev);
 896                 invalidate_bdev(sbi->journal_bdev);
 897                 ext4_blkdev_remove(sbi);
 898         }
 899         if (sbi->s_mmp_tsk)
 900                 kthread_stop(sbi->s_mmp_tsk);
 901         sb->s_fs_info = NULL;
 902         /*
 903          * Now that we are completely done shutting down the
 904          * superblock, we need to actually destroy the kobject.
 905          */
 906         unlock_super(sb);
 907         kobject_put(&sbi->s_kobj);
 908         wait_for_completion(&sbi->s_kobj_unregister);
 909         kfree(sbi->s_blockgroup_lock);
 910         kfree(sbi);
 911 }
 912
 913 static struct kmem_cache *ext4_inode_cachep;
 914
 915 /*
 916  * Called inside transaction, so use GFP_NOFS
 917  */
 918 static struct inode *ext4_alloc_inode(struct super_block *sb)
 919 {
 920         struct ext4_inode_info *ei;
 921
 922         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 923         if (!ei)
 924                 return NULL;
 925 #ifdef CONFIG_EXT4_FS_RICHACL
 926         ei->i_richacl = EXT4_RICHACL_NOT_CACHED;
 927 #endif
 928         ei->vfs_inode.i_version = 1;
 929         ei->vfs_inode.i_data.writeback_index = 0;
 930         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 931         INIT_LIST_HEAD(&ei->i_prealloc_list);
 932         spin_lock_init(&ei->i_prealloc_lock);
 933         ei->i_reserved_data_blocks = 0;
 934         ei->i_reserved_meta_blocks = 0;
 935         ei->i_allocated_meta_blocks = 0;
 936         ei->i_da_metadata_calc_len = 0;
 937         spin_lock_init(&(ei->i_block_reservation_lock));
 938 #ifdef CONFIG_QUOTA
 939         ei->i_reserved_quota = 0;
 940 #endif
 941         ei->jinode = NULL;
 942         INIT_LIST_HEAD(&ei->i_completed_io_list);
 943         spin_lock_init(&ei->i_completed_io_lock);
 944         ei->cur_aio_dio = NULL;
 945         ei->i_sync_tid = 0;
 946         ei->i_datasync_tid = 0;
 947         atomic_set(&ei->i_ioend_count, 0);
 948         atomic_set(&ei->i_aiodio_unwritten, 0);
 949
 950         return &ei->vfs_inode;
 951 }
 952
 953 static int ext4_drop_inode(struct inode *inode)
 954 {
 955         int drop = generic_drop_inode(inode);
 956
 957         trace_ext4_drop_inode(inode, drop);
 958         return drop;
 959 }
 960
 961 static void ext4_i_callback(struct rcu_head *head)
 962 {
 963         struct inode *inode = container_of(head, struct inode, i_rcu);
 964         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 965 }
 966
 967 static void ext4_destroy_inode(struct inode *inode)
 968 {
 969         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 970                 ext4_msg(inode->i_sb, KERN_ERR,
 971                          "Inode %lu (%p): orphan list check failed!",
 972                          inode->i_ino, EXT4_I(inode));
 973                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 974                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 975                                 true);
 976                 dump_stack();
 977         }
 978         call_rcu(&inode->i_rcu, ext4_i_callback);
 979 }
 980
 981 static void init_once(void *foo)
 982 {
 983         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 984
 985         INIT_LIST_HEAD(&ei->i_orphan);
 986 #ifdef CONFIG_EXT4_FS_XATTR
 987         init_rwsem(&ei->xattr_sem);
 988 #endif
 989         init_rwsem(&ei->i_data_sem);
 990         inode_init_once(&ei->vfs_inode);
 991 }
 992
 993 static int init_inodecache(void)
 994 {
 995         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 996                                              sizeof(struct ext4_inode_info),
 997                                              0, (SLAB_RECLAIM_ACCOUNT|
 998                                                 SLAB_MEM_SPREAD),
 999                                              init_once);
1000         if (ext4_inode_cachep == NULL)
1001                 return -ENOMEM;
1002         return 0;
1003 }
1004
1005 static void destroy_inodecache(void)
1006 {
1007         kmem_cache_destroy(ext4_inode_cachep);
1008 }
1009
1010 void ext4_clear_inode(struct inode *inode)
1011 {
1012         invalidate_inode_buffers(inode);
1013         end_writeback(inode);
1014         dquot_drop(inode);
1015 #ifdef CONFIG_EXT4_FS_RICHACL
1016         if (EXT4_I(inode)->i_richacl &&
1017                 EXT4_I(inode)->i_richacl != EXT4_RICHACL_NOT_CACHED) {
1018                 richacl_put(EXT4_I(inode)->i_richacl);
1019                 EXT4_I(inode)->i_richacl = EXT4_RICHACL_NOT_CACHED;
1020         }
1021 #endif
1022         ext4_discard_preallocations(inode);
1023         if (EXT4_I(inode)->jinode) {
1024                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1025                                                EXT4_I(inode)->jinode);
1026                 jbd2_free_inode(EXT4_I(inode)->jinode);
1027                 EXT4_I(inode)->jinode = NULL;
1028         }
1029 }
1030
1031 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1032                                         u64 ino, u32 generation)
1033 {
1034         struct inode *inode;
1035
1036         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1037                 return ERR_PTR(-ESTALE);
1038         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1039                 return ERR_PTR(-ESTALE);
1040
1041         /* iget isn't really right if the inode is currently unallocated!!
1042          *
1043          * ext4_read_inode will return a bad_inode if the inode had been
1044          * deleted, so we should be safe.
1045          *
1046          * Currently we don't know the generation for parent directory, so
1047          * a generation of 0 means "accept any"
1048          */
1049         inode = ext4_iget(sb, ino);
1050         if (IS_ERR(inode))
1051                 return ERR_CAST(inode);
1052         if (generation && inode->i_generation != generation) {
1053                 iput(inode);
1054                 return ERR_PTR(-ESTALE);
1055         }
1056
1057         return inode;
1058 }
1059
1060 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1061                                         int fh_len, int fh_type)
1062 {
1063         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1064                                     ext4_nfs_get_inode);
1065 }
1066
1067 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1068                                         int fh_len, int fh_type)
1069 {
1070         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1071                                     ext4_nfs_get_inode);
1072 }
1073
1074 /*
1075  * Try to release metadata pages (indirect blocks, directories) which are
1076  * mapped via the block device.  Since these pages could have journal heads
1077  * which would prevent try_to_free_buffers() from freeing them, we must use
1078  * jbd2 layer's try_to_free_buffers() function to release them.
1079  */
1080 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1081                                  gfp_t wait)
1082 {
1083         journal_t *journal = EXT4_SB(sb)->s_journal;
1084
1085         WARN_ON(PageChecked(page));
1086         if (!page_has_buffers(page))
1087                 return 0;
1088         if (journal)
1089                 return jbd2_journal_try_to_free_buffers(journal, page,
1090                                                         wait & ~__GFP_WAIT);
1091         return try_to_free_buffers(page);
1092 }
1093
1094 #ifdef CONFIG_QUOTA
1095 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1096 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1097
1098 static int ext4_write_dquot(struct dquot *dquot);
1099 static int ext4_acquire_dquot(struct dquot *dquot);
1100 static int ext4_release_dquot(struct dquot *dquot);
1101 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1102 static int ext4_write_info(struct super_block *sb, int type);
1103 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1104                          struct path *path);
1105 static int ext4_quota_off(struct super_block *sb, int type);
1106 static int ext4_quota_on_mount(struct super_block *sb, int type);
1107 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1108                                size_t len, loff_t off);
1109 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1110                                 const char *data, size_t len, loff_t off);
1111
1112 static const struct dquot_operations ext4_quota_operations = {
1113         .get_reserved_space = ext4_get_reserved_space,
1114         .write_dquot    = ext4_write_dquot,
1115         .acquire_dquot  = ext4_acquire_dquot,
1116         .release_dquot  = ext4_release_dquot,
1117         .mark_dirty     = ext4_mark_dquot_dirty,
1118         .write_info     = ext4_write_info,
1119         .alloc_dquot    = dquot_alloc,
1120         .destroy_dquot  = dquot_destroy,
1121 };
1122
1123 static const struct quotactl_ops ext4_qctl_operations = {
1124         .quota_on       = ext4_quota_on,
1125         .quota_off      = ext4_quota_off,
1126         .quota_sync     = dquot_quota_sync,
1127         .get_info       = dquot_get_dqinfo,
1128         .set_info       = dquot_set_dqinfo,
1129         .get_dqblk      = dquot_get_dqblk,
1130         .set_dqblk      = dquot_set_dqblk
1131 };
1132 #endif
1133
1134 static const struct super_operations ext4_sops = {
1135         .alloc_inode    = ext4_alloc_inode,
1136         .destroy_inode  = ext4_destroy_inode,
1137         .write_inode    = ext4_write_inode,
1138         .dirty_inode    = ext4_dirty_inode,
1139         .drop_inode     = ext4_drop_inode,
1140         .evict_inode    = ext4_evict_inode,
1141         .put_super      = ext4_put_super,
1142         .sync_fs        = ext4_sync_fs,
1143         .freeze_fs      = ext4_freeze,
1144         .unfreeze_fs    = ext4_unfreeze,
1145         .statfs         = ext4_statfs,
1146         .remount_fs     = ext4_remount,
1147         .show_options   = ext4_show_options,
1148 #ifdef CONFIG_QUOTA
1149         .quota_read     = ext4_quota_read,
1150         .quota_write    = ext4_quota_write,
1151 #endif
1152         .bdev_try_to_free_page = bdev_try_to_free_page,
1153 };
1154
1155 static const struct super_operations ext4_nojournal_sops = {
1156         .alloc_inode    = ext4_alloc_inode,
1157         .destroy_inode  = ext4_destroy_inode,
1158         .write_inode    = ext4_write_inode,
1159         .dirty_inode    = ext4_dirty_inode,
1160         .drop_inode     = ext4_drop_inode,
1161         .evict_inode    = ext4_evict_inode,
1162         .write_super    = ext4_write_super,
1163         .put_super      = ext4_put_super,
1164         .statfs         = ext4_statfs,
1165         .remount_fs     = ext4_remount,
1166         .show_options   = ext4_show_options,
1167 #ifdef CONFIG_QUOTA
1168         .quota_read     = ext4_quota_read,
1169         .quota_write    = ext4_quota_write,
1170 #endif
1171         .bdev_try_to_free_page = bdev_try_to_free_page,
1172 };
1173
1174 static const struct export_operations ext4_export_ops = {
1175         .fh_to_dentry = ext4_fh_to_dentry,
1176         .fh_to_parent = ext4_fh_to_parent,
1177         .get_parent = ext4_get_parent,
1178 };
1179
1180 enum {
1181         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1182         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1183         Opt_nouid32, Opt_debug, Opt_removed,
1184         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_richacl, Opt_noacl,
1185         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1186         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1187         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1188         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1189         Opt_data_err_abort, Opt_data_err_ignore,
1190         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1191         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1192         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1193         Opt_usrquota, Opt_grpquota, Opt_i_version,
1194         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1195         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1196         Opt_inode_readahead_blks, Opt_journal_ioprio,
1197         Opt_dioread_nolock, Opt_dioread_lock,
1198         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1199 };
1200
1201 static const match_table_t tokens = {
1202         {Opt_bsd_df, "bsddf"},
1203         {Opt_minix_df, "minixdf"},
1204         {Opt_grpid, "grpid"},
1205         {Opt_grpid, "bsdgroups"},
1206         {Opt_nogrpid, "nogrpid"},
1207         {Opt_nogrpid, "sysvgroups"},
1208         {Opt_resgid, "resgid=%u"},
1209         {Opt_resuid, "resuid=%u"},
1210         {Opt_sb, "sb=%u"},
1211         {Opt_err_cont, "errors=continue"},
1212         {Opt_err_panic, "errors=panic"},
1213         {Opt_err_ro, "errors=remount-ro"},
1214         {Opt_nouid32, "nouid32"},
1215         {Opt_debug, "debug"},
1216         {Opt_removed, "oldalloc"},
1217         {Opt_removed, "orlov"},
1218         {Opt_user_xattr, "user_xattr"},
1219         {Opt_nouser_xattr, "nouser_xattr"},
1220         {Opt_acl, "acl"},
1221         {Opt_richacl, "richacl"},
1222         {Opt_noacl, "noacl"},
1223         {Opt_noload, "norecovery"},
1224         {Opt_noload, "noload"},
1225         {Opt_removed, "nobh"},
1226         {Opt_removed, "bh"},
1227         {Opt_commit, "commit=%u"},
1228         {Opt_min_batch_time, "min_batch_time=%u"},
1229         {Opt_max_batch_time, "max_batch_time=%u"},
1230         {Opt_journal_dev, "journal_dev=%u"},
1231         {Opt_journal_checksum, "journal_checksum"},
1232         {Opt_journal_async_commit, "journal_async_commit"},
1233         {Opt_abort, "abort"},
1234         {Opt_data_journal, "data=journal"},
1235         {Opt_data_ordered, "data=ordered"},
1236         {Opt_data_writeback, "data=writeback"},
1237         {Opt_data_err_abort, "data_err=abort"},
1238         {Opt_data_err_ignore, "data_err=ignore"},
1239         {Opt_offusrjquota, "usrjquota="},
1240         {Opt_usrjquota, "usrjquota=%s"},
1241         {Opt_offgrpjquota, "grpjquota="},
1242         {Opt_grpjquota, "grpjquota=%s"},
1243         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1244         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1245         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1246         {Opt_grpquota, "grpquota"},
1247         {Opt_noquota, "noquota"},
1248         {Opt_quota, "quota"},
1249         {Opt_usrquota, "usrquota"},
1250         {Opt_barrier, "barrier=%u"},
1251         {Opt_barrier, "barrier"},
1252         {Opt_nobarrier, "nobarrier"},
1253         {Opt_i_version, "i_version"},
1254         {Opt_stripe, "stripe=%u"},
1255         {Opt_delalloc, "delalloc"},
1256         {Opt_nodelalloc, "nodelalloc"},
1257         {Opt_mblk_io_submit, "mblk_io_submit"},
1258         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1259         {Opt_block_validity, "block_validity"},
1260         {Opt_noblock_validity, "noblock_validity"},
1261         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1262         {Opt_journal_ioprio, "journal_ioprio=%u"},
1263         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1264         {Opt_auto_da_alloc, "auto_da_alloc"},
1265         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1266         {Opt_dioread_nolock, "dioread_nolock"},
1267         {Opt_dioread_lock, "dioread_lock"},
1268         {Opt_discard, "discard"},
1269         {Opt_nodiscard, "nodiscard"},
1270         {Opt_init_itable, "init_itable=%u"},
1271         {Opt_init_itable, "init_itable"},
1272         {Opt_noinit_itable, "noinit_itable"},
1273         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1274         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1275         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1276         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1277         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1278         {Opt_err, NULL},
1279 };
1280
1281 static ext4_fsblk_t get_sb_block(void **data)
1282 {
1283         ext4_fsblk_t    sb_block;
1284         char            *options = (char *) *data;
1285
1286         if (!options || strncmp(options, "sb=", 3) != 0)
1287                 return 1;       /* Default location */
1288
1289         options += 3;
1290         /* TODO: use simple_strtoll with >32bit ext4 */
1291         sb_block = simple_strtoul(options, &options, 0);
1292         if (*options && *options != ',') {
1293                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1294                        (char *) *data);
1295                 return 1;
1296         }
1297         if (*options == ',')
1298                 options++;
1299         *data = (void *) options;
1300
1301         return sb_block;
1302 }
1303
1304 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1305 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1306         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1307
1308 #ifdef CONFIG_QUOTA
1309 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1310 {
1311         struct ext4_sb_info *sbi = EXT4_SB(sb);
1312         char *qname;
1313
1314         if (sb_any_quota_loaded(sb) &&
1315                 !sbi->s_qf_names[qtype]) {
1316                 ext4_msg(sb, KERN_ERR,
1317                         "Cannot change journaled "
1318                         "quota options when quota turned on");
1319                 return -1;
1320         }
1321         qname = match_strdup(args);
1322         if (!qname) {
1323                 ext4_msg(sb, KERN_ERR,
1324                         "Not enough memory for storing quotafile name");
1325                 return -1;
1326         }
1327         if (sbi->s_qf_names[qtype] &&
1328                 strcmp(sbi->s_qf_names[qtype], qname)) {
1329                 ext4_msg(sb, KERN_ERR,
1330                         "%s quota file already specified", QTYPE2NAME(qtype));
1331                 kfree(qname);
1332                 return -1;
1333         }
1334         sbi->s_qf_names[qtype] = qname;
1335         if (strchr(sbi->s_qf_names[qtype], '/')) {
1336                 ext4_msg(sb, KERN_ERR,
1337                         "quotafile must be on filesystem root");
1338                 kfree(sbi->s_qf_names[qtype]);
1339                 sbi->s_qf_names[qtype] = NULL;
1340                 return -1;
1341         }
1342         set_opt(sb, QUOTA);
1343         return 1;
1344 }
1345
1346 static int clear_qf_name(struct super_block *sb, int qtype)
1347 {
1348
1349         struct ext4_sb_info *sbi = EXT4_SB(sb);
1350
1351         if (sb_any_quota_loaded(sb) &&
1352                 sbi->s_qf_names[qtype]) {
1353                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1354                         " when quota turned on");
1355                 return -1;
1356         }
1357         /*
1358          * The space will be released later when all options are confirmed
1359          * to be correct
1360          */
1361         sbi->s_qf_names[qtype] = NULL;
1362         return 1;
1363 }
1364 #endif
1365
1366 #define MOPT_SET        0x0001
1367 #define MOPT_CLEAR      0x0002
1368 #define MOPT_NOSUPPORT  0x0004
1369 #define MOPT_EXPLICIT   0x0008
1370 #define MOPT_CLEAR_ERR  0x0010
1371 #define MOPT_GTE0       0x0020
1372 #ifdef CONFIG_QUOTA
1373 #define MOPT_Q          0
1374 #define MOPT_QFMT       0x0040
1375 #else
1376 #define MOPT_Q          MOPT_NOSUPPORT
1377 #define MOPT_QFMT       MOPT_NOSUPPORT
1378 #endif
1379 #define MOPT_DATAJ      0x0080
1380
1381 static const struct mount_opts {
1382         int     token;
1383         int     mount_opt;
1384         int     flags;
1385 } ext4_mount_opts[] = {
1386         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1387         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1388         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1389         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1390         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1391         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1392         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1393         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1394         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1395         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1396         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1397         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1398         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1399         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1400         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1401         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1402                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1403         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1404         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1405         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1406         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1407         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1408         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1409         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1410         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1411         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1412         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1413         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1414         {Opt_commit, 0, MOPT_GTE0},
1415         {Opt_max_batch_time, 0, MOPT_GTE0},
1416         {Opt_min_batch_time, 0, MOPT_GTE0},
1417         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1418         {Opt_init_itable, 0, MOPT_GTE0},
1419         {Opt_stripe, 0, MOPT_GTE0},
1420         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1421         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1422         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1423 #ifdef CONFIG_EXT4_FS_XATTR
1424         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1425         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1426 #else
1427         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1428         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1429 #endif
1430 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1431         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1432         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1433 #else
1434         {Opt_acl, 0, MOPT_NOSUPPORT},
1435         {Opt_noacl, 0, MOPT_NOSUPPORT},
1436 #endif
1437         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1438         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1439         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1440         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1441                                                         MOPT_SET | MOPT_Q},
1442         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1443                                                         MOPT_SET | MOPT_Q},
1444         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1445                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1446         {Opt_usrjquota, 0, MOPT_Q},
1447         {Opt_grpjquota, 0, MOPT_Q},
1448         {Opt_offusrjquota, 0, MOPT_Q},
1449         {Opt_offgrpjquota, 0, MOPT_Q},
1450         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1451         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1452         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1453         {Opt_err, 0, 0}
1454 };
1455
1456 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1457                             substring_t *args, unsigned long *journal_devnum,
1458                             unsigned int *journal_ioprio, int is_remount)
1459 {
1460         struct ext4_sb_info *sbi = EXT4_SB(sb);
1461         const struct mount_opts *m;
1462         int arg = 0;
1463
1464 #ifdef CONFIG_QUOTA
1465         if (token == Opt_usrjquota)
1466                 return set_qf_name(sb, USRQUOTA, &args[0]);
1467         else if (token == Opt_grpjquota)
1468                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1469         else if (token == Opt_offusrjquota)
1470                 return clear_qf_name(sb, USRQUOTA);
1471         else if (token == Opt_offgrpjquota)
1472                 return clear_qf_name(sb, GRPQUOTA);
1473 #endif
1474         if (args->from && match_int(args, &arg))
1475                 return -1;
1476         switch (token) {
1477         case Opt_noacl:
1478         case Opt_nouser_xattr:
1479                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1480                 break;
1481         case Opt_richacl:
1482                 sb->s_flags |= MS_RICHACL;
1483                 return 1;
1484         case Opt_sb:
1485                 return 1;       /* handled by get_sb_block() */
1486         case Opt_removed:
1487                 ext4_msg(sb, KERN_WARNING,
1488                          "Ignoring removed %s option", opt);
1489                 return 1;
1490         case Opt_resuid:
1491                 sbi->s_resuid = arg;
1492                 return 1;
1493         case Opt_resgid:
1494                 sbi->s_resgid = arg;
1495                 return 1;
1496         case Opt_abort:
1497                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1498                 return 1;
1499         case Opt_i_version:
1500                 sb->s_flags |= MS_I_VERSION;
1501                 return 1;
1502         case Opt_journal_dev:
1503                 if (is_remount) {
1504                         ext4_msg(sb, KERN_ERR,
1505                                  "Cannot specify journal on remount");
1506                         return -1;
1507                 }
1508                 *journal_devnum = arg;
1509                 return 1;
1510         case Opt_journal_ioprio:
1511                 if (arg < 0 || arg > 7)
1512                         return -1;
1513                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1514                 return 1;
1515         }
1516
1517         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1518                 if (token != m->token)
1519                         continue;
1520                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1521                         return -1;
1522                 if (m->flags & MOPT_EXPLICIT)
1523                         set_opt2(sb, EXPLICIT_DELALLOC);
1524                 if (m->flags & MOPT_CLEAR_ERR)
1525                         clear_opt(sb, ERRORS_MASK);
1526                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1527                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1528                                  "options when quota turned on");
1529                         return -1;
1530                 }
1531
1532                 if (m->flags & MOPT_NOSUPPORT) {
1533                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1534                 } else if (token == Opt_commit) {
1535                         if (arg == 0)
1536                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1537                         sbi->s_commit_interval = HZ * arg;
1538                 } else if (token == Opt_max_batch_time) {
1539                         if (arg == 0)
1540                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1541                         sbi->s_max_batch_time = arg;
1542                 } else if (token == Opt_min_batch_time) {
1543                         sbi->s_min_batch_time = arg;
1544                 } else if (token == Opt_inode_readahead_blks) {
1545                         if (arg > (1 << 30))
1546                                 return -1;
1547                         if (arg && !is_power_of_2(arg)) {
1548                                 ext4_msg(sb, KERN_ERR,
1549                                          "EXT4-fs: inode_readahead_blks"
1550                                          " must be a power of 2");
1551                                 return -1;
1552                         }
1553                         sbi->s_inode_readahead_blks = arg;
1554                 } else if (token == Opt_init_itable) {
1555                         set_opt(sb, INIT_INODE_TABLE);
1556                         if (!args->from)
1557                                 arg = EXT4_DEF_LI_WAIT_MULT;
1558                         sbi->s_li_wait_mult = arg;
1559                 } else if (token == Opt_stripe) {
1560                         sbi->s_stripe = arg;
1561                 } else if (m->flags & MOPT_DATAJ) {
1562                         if (is_remount) {
1563                                 if (!sbi->s_journal)
1564                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1565                                 else if (test_opt(sb, DATA_FLAGS) !=
1566                                          m->mount_opt) {
1567                                         ext4_msg(sb, KERN_ERR,
1568                                          "Cannot change data mode on remount");
1569                                         return -1;
1570                                 }
1571                         } else {
1572                                 clear_opt(sb, DATA_FLAGS);
1573                                 sbi->s_mount_opt |= m->mount_opt;
1574                         }
1575 #ifdef CONFIG_QUOTA
1576                 } else if (m->flags & MOPT_QFMT) {
1577                         if (sb_any_quota_loaded(sb) &&
1578                             sbi->s_jquota_fmt != m->mount_opt) {
1579                                 ext4_msg(sb, KERN_ERR, "Cannot "
1580                                          "change journaled quota options "
1581                                          "when quota turned on");
1582                                 return -1;
1583                         }
1584                         sbi->s_jquota_fmt = m->mount_opt;
1585 #endif
1586                 } else {
1587                         if (!args->from)
1588                                 arg = 1;
1589                         if (m->flags & MOPT_CLEAR)
1590                                 arg = !arg;
1591                         else if (unlikely(!(m->flags & MOPT_SET))) {
1592                                 ext4_msg(sb, KERN_WARNING,
1593                                          "buggy handling of option %s", opt);
1594                                 WARN_ON(1);
1595                                 return -1;
1596                         }
1597                         if (arg != 0)
1598                                 sbi->s_mount_opt |= m->mount_opt;
1599                         else
1600                                 sbi->s_mount_opt &= ~m->mount_opt;
1601                 }
1602                 return 1;
1603         }
1604         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1605                  "or missing value", opt);
1606         return -1;
1607 }
1608
1609 static int parse_options(char *options, struct super_block *sb,
1610                          unsigned long *journal_devnum,
1611                          unsigned int *journal_ioprio,
1612                          int is_remount)
1613 {
1614         struct ext4_sb_info *sbi = EXT4_SB(sb);
1615         char *p;
1616         substring_t args[MAX_OPT_ARGS];
1617         int token;
1618
1619         if (!options)
1620                 return 1;
1621
1622         while ((p = strsep(&options, ",")) != NULL) {
1623                 if (!*p)
1624                         continue;
1625                 /*
1626                  * Initialize args struct so we know whether arg was
1627                  * found; some options take optional arguments.
1628                  */
1629                 args[0].to = args[0].from = 0;
1630                 token = match_token(p, tokens, args);
1631                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1632                                      journal_ioprio, is_remount) < 0)
1633                         return 0;
1634         }
1635 #ifdef CONFIG_QUOTA
1636         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1637                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1638                         clear_opt(sb, USRQUOTA);
1639
1640                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1641                         clear_opt(sb, GRPQUOTA);
1642
1643                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1644                         ext4_msg(sb, KERN_ERR, "old and new quota "
1645                                         "format mixing");
1646                         return 0;
1647                 }
1648
1649                 if (!sbi->s_jquota_fmt) {
1650                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1651                                         "not specified");
1652                         return 0;
1653                 }
1654         } else {
1655                 if (sbi->s_jquota_fmt) {
1656                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1657                                         "specified with no journaling "
1658                                         "enabled");
1659                         return 0;
1660                 }
1661         }
1662 #endif
1663 #if defined(CONFIG_EXT4_FS_RICHACL) && defined(CONFIG_EXT4_FS_POSIX_ACL)
1664         if (test_opt(sb, POSIX_ACL) && (sb->s_flags & MS_RICHACL))
1665                 clear_opt(sb, POSIX_ACL);
1666 #endif
1667         return 1;
1668 }
1669
1670 static inline void ext4_show_quota_options(struct seq_file *seq,
1671                                            struct super_block *sb)
1672 {
1673 #if defined(CONFIG_QUOTA)
1674         struct ext4_sb_info *sbi = EXT4_SB(sb);
1675
1676         if (sbi->s_jquota_fmt) {
1677                 char *fmtname = "";
1678
1679                 switch (sbi->s_jquota_fmt) {
1680                 case QFMT_VFS_OLD:
1681                         fmtname = "vfsold";
1682                         break;
1683                 case QFMT_VFS_V0:
1684                         fmtname = "vfsv0";
1685                         break;
1686                 case QFMT_VFS_V1:
1687                         fmtname = "vfsv1";
1688                         break;
1689                 }
1690                 seq_printf(seq, ",jqfmt=%s", fmtname);
1691         }
1692
1693         if (sbi->s_qf_names[USRQUOTA])
1694                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1695
1696         if (sbi->s_qf_names[GRPQUOTA])
1697                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1698
1699         if (test_opt(sb, USRQUOTA))
1700                 seq_puts(seq, ",usrquota");
1701
1702         if (test_opt(sb, GRPQUOTA))
1703                 seq_puts(seq, ",grpquota");
1704 #endif
1705 }
1706
1707 static const char *token2str(int token)
1708 {
1709         static const struct match_token *t;
1710
1711         for (t = tokens; t->token != Opt_err; t++)
1712                 if (t->token == token && !strchr(t->pattern, '='))
1713                         break;
1714         return t->pattern;
1715 }
1716
1717 /*
1718  * Show an option if
1719  *  - it's set to a non-default value OR
1720  *  - if the per-sb default is different from the global default
1721  */
1722 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1723                               int nodefs)
1724 {
1725         struct ext4_sb_info *sbi = EXT4_SB(sb);
1726         struct ext4_super_block *es = sbi->s_es;
1727         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1728         const struct mount_opts *m;
1729         char sep = nodefs ? '\n' : ',';
1730
1731 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1732 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1733
1734         if (sbi->s_sb_block != 1)
1735                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1736
1737         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1738                 int want_set = m->flags & MOPT_SET;
1739                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1740                     (m->flags & MOPT_CLEAR_ERR))
1741                         continue;
1742                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1743                         continue; /* skip if same as the default */
1744                 if ((want_set &&
1745                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1746                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1747                         continue; /* select Opt_noFoo vs Opt_Foo */
1748                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1749         }
1750
1751         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1752             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1753                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1754         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1755             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1756                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1757         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1758         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1759                 SEQ_OPTS_PUTS("errors=remount-ro");
1760         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1761                 SEQ_OPTS_PUTS("errors=continue");
1762         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1763                 SEQ_OPTS_PUTS("errors=panic");
1764         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1765                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1766         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1767                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1768         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1769                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1770         if (sb->s_flags & MS_I_VERSION)
1771                 SEQ_OPTS_PUTS("i_version");
1772         if (nodefs || sbi->s_stripe)
1773                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1774         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1775                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1776                         SEQ_OPTS_PUTS("data=journal");
1777                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1778                         SEQ_OPTS_PUTS("data=ordered");
1779                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1780                         SEQ_OPTS_PUTS("data=writeback");
1781         }
1782         if (nodefs ||
1783             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1784                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1785                                sbi->s_inode_readahead_blks);
1786
1787         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1788                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1789                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1790
1791         if (sb->s_flags & MS_RICHACL)
1792                 SEQ_OPTS_PUTS("richacl");
1793
1794         ext4_show_quota_options(seq, sb);
1795         return 0;
1796 }
1797
1798 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1799 {
1800         return _ext4_show_options(seq, root->d_sb, 0);
1801 }
1802
1803 static int options_seq_show(struct seq_file *seq, void *offset)
1804 {
1805         struct super_block *sb = seq->private;
1806         int rc;
1807
1808         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1809         rc = _ext4_show_options(seq, sb, 1);
1810         seq_puts(seq, "\n");
1811         return rc;
1812 }
1813
1814 static int options_open_fs(struct inode *inode, struct file *file)
1815 {
1816         return single_open(file, options_seq_show, PDE(inode)->data);
1817 }
1818
1819 static const struct file_operations ext4_seq_options_fops = {
1820         .owner = THIS_MODULE,
1821         .open = options_open_fs,
1822         .read = seq_read,
1823         .llseek = seq_lseek,
1824         .release = single_release,
1825 };
1826
1827 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1828                             int read_only)
1829 {
1830         struct ext4_sb_info *sbi = EXT4_SB(sb);
1831         int res = 0;
1832
1833         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1834                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1835                          "forcing read-only mode");
1836                 res = MS_RDONLY;
1837         }
1838         if (read_only)
1839                 goto done;
1840         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1841                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1842                          "running e2fsck is recommended");
1843         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1844                 ext4_msg(sb, KERN_WARNING,
1845                          "warning: mounting fs with errors, "
1846                          "running e2fsck is recommended");
1847         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1848                  le16_to_cpu(es->s_mnt_count) >=
1849                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1850                 ext4_msg(sb, KERN_WARNING,
1851                          "warning: maximal mount count reached, "
1852                          "running e2fsck is recommended");
1853         else if (le32_to_cpu(es->s_checkinterval) &&
1854                 (le32_to_cpu(es->s_lastcheck) +
1855                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1856                 ext4_msg(sb, KERN_WARNING,
1857                          "warning: checktime reached, "
1858                          "running e2fsck is recommended");
1859         if (!sbi->s_journal)
1860                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1861         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1862                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1863         le16_add_cpu(&es->s_mnt_count, 1);
1864         es->s_mtime = cpu_to_le32(get_seconds());
1865         ext4_update_dynamic_rev(sb);
1866         if (sbi->s_journal)
1867                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1868
1869         ext4_commit_super(sb, 1);
1870 done:
1871         if (test_opt(sb, DEBUG))
1872                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1873                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1874                         sb->s_blocksize,
1875                         sbi->s_groups_count,
1876                         EXT4_BLOCKS_PER_GROUP(sb),
1877                         EXT4_INODES_PER_GROUP(sb),
1878                         sbi->s_mount_opt, sbi->s_mount_opt2);
1879
1880         cleancache_init_fs(sb);
1881         return res;
1882 }
1883
1884 static int ext4_fill_flex_info(struct super_block *sb)
1885 {
1886         struct ext4_sb_info *sbi = EXT4_SB(sb);
1887         struct ext4_group_desc *gdp = NULL;
1888         ext4_group_t flex_group_count;
1889         ext4_group_t flex_group;
1890         unsigned int groups_per_flex = 0;
1891         size_t size;
1892         int i;
1893
1894         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1895         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1896                 sbi->s_log_groups_per_flex = 0;
1897                 return 1;
1898         }
1899         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1900
1901         /* We allocate both existing and potentially added groups */
1902         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1903                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1904                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1905         size = flex_group_count * sizeof(struct flex_groups);
1906         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1907         if (sbi->s_flex_groups == NULL) {
1908                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1909                          flex_group_count);
1910                 goto failed;
1911         }
1912
1913         for (i = 0; i < sbi->s_groups_count; i++) {
1914                 gdp = ext4_get_group_desc(sb, i, NULL);
1915
1916                 flex_group = ext4_flex_group(sbi, i);
1917                 atomic_add(ext4_free_inodes_count(sb, gdp),
1918                            &sbi->s_flex_groups[flex_group].free_inodes);
1919                 atomic_add(ext4_free_group_clusters(sb, gdp),
1920                            &sbi->s_flex_groups[flex_group].free_clusters);
1921                 atomic_add(ext4_used_dirs_count(sb, gdp),
1922                            &sbi->s_flex_groups[flex_group].used_dirs);
1923         }
1924
1925         return 1;
1926 failed:
1927         return 0;
1928 }
1929
1930 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1931                             struct ext4_group_desc *gdp)
1932 {
1933         __u16 crc = 0;
1934
1935         if (sbi->s_es->s_feature_ro_compat &
1936             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1937                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1938                 __le32 le_group = cpu_to_le32(block_group);
1939
1940                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1941                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1942                 crc = crc16(crc, (__u8 *)gdp, offset);
1943                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1944                 /* for checksum of struct ext4_group_desc do the rest...*/
1945                 if ((sbi->s_es->s_feature_incompat &
1946                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1947                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1948                         crc = crc16(crc, (__u8 *)gdp + offset,
1949                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1950                                         offset);
1951         }
1952
1953         return cpu_to_le16(crc);
1954 }
1955
1956 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1957                                 struct ext4_group_desc *gdp)
1958 {
1959         if ((sbi->s_es->s_feature_ro_compat &
1960              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1961             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1962                 return 0;
1963
1964         return 1;
1965 }
1966
1967 /* Called at mount-time, super-block is locked */
1968 static int ext4_check_descriptors(struct super_block *sb,
1969                                   ext4_group_t *first_not_zeroed)
1970 {
1971         struct ext4_sb_info *sbi = EXT4_SB(sb);
1972         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1973         ext4_fsblk_t last_block;
1974         ext4_fsblk_t block_bitmap;
1975         ext4_fsblk_t inode_bitmap;
1976         ext4_fsblk_t inode_table;
1977         int flexbg_flag = 0;
1978         ext4_group_t i, grp = sbi->s_groups_count;
1979
1980         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1981                 flexbg_flag = 1;
1982
1983         ext4_debug("Checking group descriptors");
1984
1985         for (i = 0; i < sbi->s_groups_count; i++) {
1986                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1987
1988                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1989                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1990                 else
1991                         last_block = first_block +
1992                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1993
1994                 if ((grp == sbi->s_groups_count) &&
1995                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1996                         grp = i;
1997
1998                 block_bitmap = ext4_block_bitmap(sb, gdp);
1999                 if (block_bitmap < first_block || block_bitmap > last_block) {
2000                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2001                                "Block bitmap for group %u not in group "
2002                                "(block %llu)!", i, block_bitmap);
2003                         return 0;
2004                 }
2005                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2006                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2007                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2008                                "Inode bitmap for group %u not in group "
2009                                "(block %llu)!", i, inode_bitmap);
2010                         return 0;
2011                 }
2012                 inode_table = ext4_inode_table(sb, gdp);
2013                 if (inode_table < first_block ||
2014                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2015                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2016                                "Inode table for group %u not in group "
2017                                "(block %llu)!", i, inode_table);
2018                         return 0;
2019                 }
2020                 ext4_lock_group(sb, i);
2021                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2022                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2023                                  "Checksum for group %u failed (%u!=%u)",
2024                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2025                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2026                         if (!(sb->s_flags & MS_RDONLY)) {
2027                                 ext4_unlock_group(sb, i);
2028                                 return 0;
2029                         }
2030                 }
2031                 ext4_unlock_group(sb, i);
2032                 if (!flexbg_flag)
2033                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2034         }
2035         if (NULL != first_not_zeroed)
2036                 *first_not_zeroed = grp;
2037
2038         ext4_free_blocks_count_set(sbi->s_es,
2039                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2040         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2041         return 1;
2042 }
2043
2044 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2045  * the superblock) which were deleted from all directories, but held open by
2046  * a process at the time of a crash.  We walk the list and try to delete these
2047  * inodes at recovery time (only with a read-write filesystem).
2048  *
2049  * In order to keep the orphan inode chain consistent during traversal (in
2050  * case of crash during recovery), we link each inode into the superblock
2051  * orphan list_head and handle it the same way as an inode deletion during
2052  * normal operation (which journals the operations for us).
2053  *
2054  * We only do an iget() and an iput() on each inode, which is very safe if we
2055  * accidentally point at an in-use or already deleted inode.  The worst that
2056  * can happen in this case is that we get a "bit already cleared" message from
2057  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2058  * e2fsck was run on this filesystem, and it must have already done the orphan
2059  * inode cleanup for us, so we can safely abort without any further action.
2060  */
2061 static void ext4_orphan_cleanup(struct super_block *sb,
2062                                 struct ext4_super_block *es)
2063 {
2064         unsigned int s_flags = sb->s_flags;
2065         int nr_orphans = 0, nr_truncates = 0;
2066 #ifdef CONFIG_QUOTA
2067         int i;
2068 #endif
2069         if (!es->s_last_orphan) {
2070                 jbd_debug(4, "no orphan inodes to clean up\n");
2071                 return;
2072         }
2073
2074         if (bdev_read_only(sb->s_bdev)) {
2075                 ext4_msg(sb, KERN_ERR, "write access "
2076                         "unavailable, skipping orphan cleanup");
2077                 return;
2078         }
2079
2080         /* Check if feature set would not allow a r/w mount */
2081         if (!ext4_feature_set_ok(sb, 0)) {
2082                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2083                          "unknown ROCOMPAT features");
2084                 return;
2085         }
2086
2087         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2088                 if (es->s_last_orphan)
2089                         jbd_debug(1, "Errors on filesystem, "
2090                                   "clearing orphan list.\n");
2091                 es->s_last_orphan = 0;
2092                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2093                 return;
2094         }
2095
2096         if (s_flags & MS_RDONLY) {
2097                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2098                 sb->s_flags &= ~MS_RDONLY;
2099         }
2100 #ifdef CONFIG_QUOTA
2101         /* Needed for iput() to work correctly and not trash data */
2102         sb->s_flags |= MS_ACTIVE;
2103         /* Turn on quotas so that they are updated correctly */
2104         for (i = 0; i < MAXQUOTAS; i++) {
2105                 if (EXT4_SB(sb)->s_qf_names[i]) {
2106                         int ret = ext4_quota_on_mount(sb, i);
2107                         if (ret < 0)
2108                                 ext4_msg(sb, KERN_ERR,
2109                                         "Cannot turn on journaled "
2110                                         "quota: error %d", ret);
2111                 }
2112         }
2113 #endif
2114
2115         while (es->s_last_orphan) {
2116                 struct inode *inode;
2117
2118                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2119                 if (IS_ERR(inode)) {
2120                         es->s_last_orphan = 0;
2121                         break;
2122                 }
2123
2124                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2125                 dquot_initialize(inode);
2126                 if (inode->i_nlink) {
2127                         ext4_msg(sb, KERN_DEBUG,
2128                                 "%s: truncating inode %lu to %lld bytes",
2129                                 __func__, inode->i_ino, inode->i_size);
2130                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2131                                   inode->i_ino, inode->i_size);
2132                         ext4_truncate(inode);
2133                         nr_truncates++;
2134                 } else {
2135                         ext4_msg(sb, KERN_DEBUG,
2136                                 "%s: deleting unreferenced inode %lu",
2137                                 __func__, inode->i_ino);
2138                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2139                                   inode->i_ino);
2140                         nr_orphans++;
2141                 }
2142                 iput(inode);  /* The delete magic happens here! */
2143         }
2144
2145 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2146
2147         if (nr_orphans)
2148                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2149                        PLURAL(nr_orphans));
2150         if (nr_truncates)
2151                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2152                        PLURAL(nr_truncates));
2153 #ifdef CONFIG_QUOTA
2154         /* Turn quotas off */
2155         for (i = 0; i < MAXQUOTAS; i++) {
2156                 if (sb_dqopt(sb)->files[i])
2157                         dquot_quota_off(sb, i);
2158         }
2159 #endif
2160         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2161 }
2162
2163 /*
2164  * Maximal extent format file size.
2165  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2166  * extent format containers, within a sector_t, and within i_blocks
2167  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2168  * so that won't be a limiting factor.
2169  *
2170  * However there is other limiting factor. We do store extents in the form
2171  * of starting block and length, hence the resulting length of the extent
2172  * covering maximum file size must fit into on-disk format containers as
2173  * well. Given that length is always by 1 unit bigger than max unit (because
2174  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2175  *
2176  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2177  */
2178 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2179 {
2180         loff_t res;
2181         loff_t upper_limit = MAX_LFS_FILESIZE;
2182
2183         /* small i_blocks in vfs inode? */
2184         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2185                 /*
2186                  * CONFIG_LBDAF is not enabled implies the inode
2187                  * i_block represent total blocks in 512 bytes
2188                  * 32 == size of vfs inode i_blocks * 8
2189                  */
2190                 upper_limit = (1LL << 32) - 1;
2191
2192                 /* total blocks in file system block size */
2193                 upper_limit >>= (blkbits - 9);
2194                 upper_limit <<= blkbits;
2195         }
2196
2197         /*
2198          * 32-bit extent-start container, ee_block. We lower the maxbytes
2199          * by one fs block, so ee_len can cover the extent of maximum file
2200          * size
2201          */
2202         res = (1LL << 32) - 1;
2203         res <<= blkbits;
2204
2205         /* Sanity check against vm- & vfs- imposed limits */
2206         if (res > upper_limit)
2207                 res = upper_limit;
2208
2209         return res;
2210 }
2211
2212 /*
2213  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2214  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2215  * We need to be 1 filesystem block less than the 2^48 sector limit.
2216  */
2217 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2218 {
2219         loff_t res = EXT4_NDIR_BLOCKS;
2220         int meta_blocks;
2221         loff_t upper_limit;
2222         /* This is calculated to be the largest file size for a dense, block
2223          * mapped file such that the file's total number of 512-byte sectors,
2224          * including data and all indirect blocks, does not exceed (2^48 - 1).
2225          *
2226          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2227          * number of 512-byte sectors of the file.
2228          */
2229
2230         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2231                 /*
2232                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2233                  * the inode i_block field represents total file blocks in
2234                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2235                  */
2236                 upper_limit = (1LL << 32) - 1;
2237
2238                 /* total blocks in file system block size */
2239                 upper_limit >>= (bits - 9);
2240
2241         } else {
2242                 /*
2243                  * We use 48 bit ext4_inode i_blocks
2244                  * With EXT4_HUGE_FILE_FL set the i_blocks
2245                  * represent total number of blocks in
2246                  * file system block size
2247                  */
2248                 upper_limit = (1LL << 48) - 1;
2249
2250         }
2251
2252         /* indirect blocks */
2253         meta_blocks = 1;
2254         /* double indirect blocks */
2255         meta_blocks += 1 + (1LL << (bits-2));
2256         /* tripple indirect blocks */
2257         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2258
2259         upper_limit -= meta_blocks;
2260         upper_limit <<= bits;
2261
2262         res += 1LL << (bits-2);
2263         res += 1LL << (2*(bits-2));
2264         res += 1LL << (3*(bits-2));
2265         res <<= bits;
2266         if (res > upper_limit)
2267                 res = upper_limit;
2268
2269         if (res > MAX_LFS_FILESIZE)
2270                 res = MAX_LFS_FILESIZE;
2271
2272         return res;
2273 }
2274
2275 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2276                                    ext4_fsblk_t logical_sb_block, int nr)
2277 {
2278         struct ext4_sb_info *sbi = EXT4_SB(sb);
2279         ext4_group_t bg, first_meta_bg;
2280         int has_super = 0;
2281
2282         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2283
2284         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2285             nr < first_meta_bg)
2286                 return logical_sb_block + nr + 1;
2287         bg = sbi->s_desc_per_block * nr;
2288         if (ext4_bg_has_super(sb, bg))
2289                 has_super = 1;
2290
2291         return (has_super + ext4_group_first_block_no(sb, bg));
2292 }
2293
2294 /**
2295  * ext4_get_stripe_size: Get the stripe size.
2296  * @sbi: In memory super block info
2297  *
2298  * If we have specified it via mount option, then
2299  * use the mount option value. If the value specified at mount time is
2300  * greater than the blocks per group use the super block value.
2301  * If the super block value is greater than blocks per group return 0.
2302  * Allocator needs it be less than blocks per group.
2303  *
2304  */
2305 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2306 {
2307         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2308         unsigned long stripe_width =
2309                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2310         int ret;
2311
2312         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2313                 ret = sbi->s_stripe;
2314         else if (stripe_width <= sbi->s_blocks_per_group)
2315                 ret = stripe_width;
2316         else if (stride <= sbi->s_blocks_per_group)
2317                 ret = stride;
2318         else
2319                 ret = 0;
2320
2321         /*
2322          * If the stripe width is 1, this makes no sense and
2323          * we set it to 0 to turn off stripe handling code.
2324          */
2325         if (ret <= 1)
2326                 ret = 0;
2327
2328         return ret;
2329 }
2330
2331 /* sysfs supprt */
2332
2333 struct ext4_attr {
2334         struct attribute attr;
2335         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2336         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2337                          const char *, size_t);
2338         int offset;
2339 };
2340
2341 static int parse_strtoul(const char *buf,
2342                 unsigned long max, unsigned long *value)
2343 {
2344         char *endp;
2345
2346         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2347         endp = skip_spaces(endp);
2348         if (*endp || *value > max)
2349                 return -EINVAL;
2350
2351         return 0;
2352 }
2353
2354 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2355                                               struct ext4_sb_info *sbi,
2356                                               char *buf)
2357 {
2358         return snprintf(buf, PAGE_SIZE, "%llu\n",
2359                 (s64) EXT4_C2B(sbi,
2360                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2361 }
2362
2363 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2364                                          struct ext4_sb_info *sbi, char *buf)
2365 {
2366         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2367
2368         if (!sb->s_bdev->bd_part)
2369                 return snprintf(buf, PAGE_SIZE, "0\n");
2370         return snprintf(buf, PAGE_SIZE, "%lu\n",
2371                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2372                          sbi->s_sectors_written_start) >> 1);
2373 }
2374
2375 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2376                                           struct ext4_sb_info *sbi, char *buf)
2377 {
2378         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2379
2380         if (!sb->s_bdev->bd_part)
2381                 return snprintf(buf, PAGE_SIZE, "0\n");
2382         return snprintf(buf, PAGE_SIZE, "%llu\n",
2383                         (unsigned long long)(sbi->s_kbytes_written +
2384                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2385                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2386 }
2387
2388 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2389                                           struct ext4_sb_info *sbi,
2390                                           const char *buf, size_t count)
2391 {
2392         unsigned long t;
2393
2394         if (parse_strtoul(buf, 0x40000000, &t))
2395                 return -EINVAL;
2396
2397         if (t && !is_power_of_2(t))
2398                 return -EINVAL;
2399
2400         sbi->s_inode_readahead_blks = t;
2401         return count;
2402 }
2403
2404 static ssize_t sbi_ui_show(struct ext4_attr *a,
2405                            struct ext4_sb_info *sbi, char *buf)
2406 {
2407         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2408
2409         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2410 }
2411
2412 static ssize_t sbi_ui_store(struct ext4_attr *a,
2413                             struct ext4_sb_info *sbi,
2414                             const char *buf, size_t count)
2415 {
2416         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2417         unsigned long t;
2418
2419         if (parse_strtoul(buf, 0xffffffff, &t))
2420                 return -EINVAL;
2421         *ui = t;
2422         return count;
2423 }
2424
2425 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2426 static struct ext4_attr ext4_attr_##_name = {                   \
2427         .attr = {.name = __stringify(_name), .mode = _mode },   \
2428         .show   = _show,                                        \
2429         .store  = _store,                                       \
2430         .offset = offsetof(struct ext4_sb_info, _elname),       \
2431 }
2432 #define EXT4_ATTR(name, mode, show, store) \
2433 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2434
2435 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2436 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2437 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2438 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2439         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2440 #define ATTR_LIST(name) &ext4_attr_##name.attr
2441
2442 EXT4_RO_ATTR(delayed_allocation_blocks);
2443 EXT4_RO_ATTR(session_write_kbytes);
2444 EXT4_RO_ATTR(lifetime_write_kbytes);
2445 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2446                  inode_readahead_blks_store, s_inode_readahead_blks);
2447 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2448 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2449 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2450 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2451 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2452 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2453 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2454 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2455
2456 static struct attribute *ext4_attrs[] = {
2457         ATTR_LIST(delayed_allocation_blocks),
2458         ATTR_LIST(session_write_kbytes),
2459         ATTR_LIST(lifetime_write_kbytes),
2460         ATTR_LIST(inode_readahead_blks),
2461         ATTR_LIST(inode_goal),
2462         ATTR_LIST(mb_stats),
2463         ATTR_LIST(mb_max_to_scan),
2464         ATTR_LIST(mb_min_to_scan),
2465         ATTR_LIST(mb_order2_req),
2466         ATTR_LIST(mb_stream_req),
2467         ATTR_LIST(mb_group_prealloc),
2468         ATTR_LIST(max_writeback_mb_bump),
2469         NULL,
2470 };
2471
2472 /* Features this copy of ext4 supports */
2473 EXT4_INFO_ATTR(lazy_itable_init);
2474 EXT4_INFO_ATTR(batched_discard);
2475
2476 static struct attribute *ext4_feat_attrs[] = {
2477         ATTR_LIST(lazy_itable_init),
2478         ATTR_LIST(batched_discard),
2479         NULL,
2480 };
2481
2482 static ssize_t ext4_attr_show(struct kobject *kobj,
2483                               struct attribute *attr, char *buf)
2484 {
2485         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2486                                                 s_kobj);
2487         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2488
2489         return a->show ? a->show(a, sbi, buf) : 0;
2490 }
2491
2492 static ssize_t ext4_attr_store(struct kobject *kobj,
2493                                struct attribute *attr,
2494                                const char *buf, size_t len)
2495 {
2496         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2497                                                 s_kobj);
2498         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2499
2500         return a->store ? a->store(a, sbi, buf, len) : 0;
2501 }
2502
2503 static void ext4_sb_release(struct kobject *kobj)
2504 {
2505         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2506                                                 s_kobj);
2507         complete(&sbi->s_kobj_unregister);
2508 }
2509
2510 static const struct sysfs_ops ext4_attr_ops = {
2511         .show   = ext4_attr_show,
2512         .store  = ext4_attr_store,
2513 };
2514
2515 static struct kobj_type ext4_ktype = {
2516         .default_attrs  = ext4_attrs,
2517         .sysfs_ops      = &ext4_attr_ops,
2518         .release        = ext4_sb_release,
2519 };
2520
2521 static void ext4_feat_release(struct kobject *kobj)
2522 {
2523         complete(&ext4_feat->f_kobj_unregister);
2524 }
2525
2526 static struct kobj_type ext4_feat_ktype = {
2527         .default_attrs  = ext4_feat_attrs,
2528         .sysfs_ops      = &ext4_attr_ops,
2529         .release        = ext4_feat_release,
2530 };
2531
2532 /*
2533  * Check whether this filesystem can be mounted based on
2534  * the features present and the RDONLY/RDWR mount requested.
2535  * Returns 1 if this filesystem can be mounted as requested,
2536  * 0 if it cannot be.
2537  */
2538 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2539 {
2540         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2541                 ext4_msg(sb, KERN_ERR,
2542                         "Couldn't mount because of "
2543                         "unsupported optional features (%x)",
2544                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2545                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2546                 return 0;
2547         }
2548
2549         if (readonly)
2550                 return 1;
2551
2552         /* Check that feature set is OK for a read-write mount */
2553         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2554                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2555                          "unsupported optional features (%x)",
2556                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2557                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2558                 return 0;
2559         }
2560         /*
2561          * Large file size enabled file system can only be mounted
2562          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2563          */
2564         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2565                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2566                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2567                                  "cannot be mounted RDWR without "
2568                                  "CONFIG_LBDAF");
2569                         return 0;
2570                 }
2571         }
2572         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2573             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2574                 ext4_msg(sb, KERN_ERR,
2575                          "Can't support bigalloc feature without "
2576                          "extents feature\n");
2577                 return 0;
2578         }
2579         return 1;
2580 }
2581
2582 /*
2583  * This function is called once a day if we have errors logged
2584  * on the file system
2585  */
2586 static void print_daily_error_info(unsigned long arg)
2587 {
2588         struct super_block *sb = (struct super_block *) arg;
2589         struct ext4_sb_info *sbi;
2590         struct ext4_super_block *es;
2591
2592         sbi = EXT4_SB(sb);
2593         es = sbi->s_es;
2594
2595         if (es->s_error_count)
2596                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2597                          le32_to_cpu(es->s_error_count));
2598         if (es->s_first_error_time) {
2599                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2600                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2601                        (int) sizeof(es->s_first_error_func),
2602                        es->s_first_error_func,
2603                        le32_to_cpu(es->s_first_error_line));
2604                 if (es->s_first_error_ino)
2605                         printk(": inode %u",
2606                                le32_to_cpu(es->s_first_error_ino));
2607                 if (es->s_first_error_block)
2608                         printk(": block %llu", (unsigned long long)
2609                                le64_to_cpu(es->s_first_error_block));
2610                 printk("\n");
2611         }
2612         if (es->s_last_error_time) {
2613                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2614                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2615                        (int) sizeof(es->s_last_error_func),
2616                        es->s_last_error_func,
2617                        le32_to_cpu(es->s_last_error_line));
2618                 if (es->s_last_error_ino)
2619                         printk(": inode %u",
2620                                le32_to_cpu(es->s_last_error_ino));
2621                 if (es->s_last_error_block)
2622                         printk(": block %llu", (unsigned long long)
2623                                le64_to_cpu(es->s_last_error_block));
2624                 printk("\n");
2625         }
2626         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2627 }
2628
2629 /* Find next suitable group and run ext4_init_inode_table */
2630 static int ext4_run_li_request(struct ext4_li_request *elr)
2631 {
2632         struct ext4_group_desc *gdp = NULL;
2633         ext4_group_t group, ngroups;
2634         struct super_block *sb;
2635         unsigned long timeout = 0;
2636         int ret = 0;
2637
2638         sb = elr->lr_super;
2639         ngroups = EXT4_SB(sb)->s_groups_count;
2640
2641         for (group = elr->lr_next_group; group < ngroups; group++) {
2642                 gdp = ext4_get_group_desc(sb, group, NULL);
2643                 if (!gdp) {
2644                         ret = 1;
2645                         break;
2646                 }
2647
2648                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2649                         break;
2650         }
2651
2652         if (group == ngroups)
2653                 ret = 1;
2654
2655         if (!ret) {
2656                 timeout = jiffies;
2657                 ret = ext4_init_inode_table(sb, group,
2658                                             elr->lr_timeout ? 0 : 1);
2659                 if (elr->lr_timeout == 0) {
2660                         timeout = (jiffies - timeout) *
2661                                   elr->lr_sbi->s_li_wait_mult;
2662                         elr->lr_timeout = timeout;
2663                 }
2664                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2665                 elr->lr_next_group = group + 1;
2666         }
2667
2668         return ret;
2669 }
2670
2671 /*
2672  * Remove lr_request from the list_request and free the
2673  * request structure. Should be called with li_list_mtx held
2674  */
2675 static void ext4_remove_li_request(struct ext4_li_request *elr)
2676 {
2677         struct ext4_sb_info *sbi;
2678
2679         if (!elr)
2680                 return;
2681
2682         sbi = elr->lr_sbi;
2683
2684         list_del(&elr->lr_request);
2685         sbi->s_li_request = NULL;
2686         kfree(elr);
2687 }
2688
2689 static void ext4_unregister_li_request(struct super_block *sb)
2690 {
2691         mutex_lock(&ext4_li_mtx);
2692         if (!ext4_li_info) {
2693                 mutex_unlock(&ext4_li_mtx);
2694                 return;
2695         }
2696
2697         mutex_lock(&ext4_li_info->li_list_mtx);
2698         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2699         mutex_unlock(&ext4_li_info->li_list_mtx);
2700         mutex_unlock(&ext4_li_mtx);
2701 }
2702
2703 static struct task_struct *ext4_lazyinit_task;
2704
2705 /*
2706  * This is the function where ext4lazyinit thread lives. It walks
2707  * through the request list searching for next scheduled filesystem.
2708  * When such a fs is found, run the lazy initialization request
2709  * (ext4_rn_li_request) and keep track of the time spend in this
2710  * function. Based on that time we compute next schedule time of
2711  * the request. When walking through the list is complete, compute
2712  * next waking time and put itself into sleep.
2713  */
2714 static int ext4_lazyinit_thread(void *arg)
2715 {
2716         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2717         struct list_head *pos, *n;
2718         struct ext4_li_request *elr;
2719         unsigned long next_wakeup, cur;
2720
2721         BUG_ON(NULL == eli);
2722
2723 cont_thread:
2724         while (true) {
2725                 next_wakeup = MAX_JIFFY_OFFSET;
2726
2727                 mutex_lock(&eli->li_list_mtx);
2728                 if (list_empty(&eli->li_request_list)) {
2729                         mutex_unlock(&eli->li_list_mtx);
2730                         goto exit_thread;
2731                 }
2732
2733                 list_for_each_safe(pos, n, &eli->li_request_list) {
2734                         elr = list_entry(pos, struct ext4_li_request,
2735                                          lr_request);
2736
2737                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2738                                 if (ext4_run_li_request(elr) != 0) {
2739                                         /* error, remove the lazy_init job */
2740                                         ext4_remove_li_request(elr);
2741                                         continue;
2742                                 }
2743                         }
2744
2745                         if (time_before(elr->lr_next_sched, next_wakeup))
2746                                 next_wakeup = elr->lr_next_sched;
2747                 }
2748                 mutex_unlock(&eli->li_list_mtx);
2749
2750                 try_to_freeze();
2751
2752                 cur = jiffies;
2753                 if ((time_after_eq(cur, next_wakeup)) ||
2754                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2755                         cond_resched();
2756                         continue;
2757                 }
2758
2759                 schedule_timeout_interruptible(next_wakeup - cur);
2760
2761                 if (kthread_should_stop()) {
2762                         ext4_clear_request_list();
2763                         goto exit_thread;
2764                 }
2765         }
2766
2767 exit_thread:
2768         /*
2769          * It looks like the request list is empty, but we need
2770          * to check it under the li_list_mtx lock, to prevent any
2771          * additions into it, and of course we should lock ext4_li_mtx
2772          * to atomically free the list and ext4_li_info, because at
2773          * this point another ext4 filesystem could be registering
2774          * new one.
2775          */
2776         mutex_lock(&ext4_li_mtx);
2777         mutex_lock(&eli->li_list_mtx);
2778         if (!list_empty(&eli->li_request_list)) {
2779                 mutex_unlock(&eli->li_list_mtx);
2780                 mutex_unlock(&ext4_li_mtx);
2781                 goto cont_thread;
2782         }
2783         mutex_unlock(&eli->li_list_mtx);
2784         kfree(ext4_li_info);
2785         ext4_li_info = NULL;
2786         mutex_unlock(&ext4_li_mtx);
2787
2788         return 0;
2789 }
2790
2791 static void ext4_clear_request_list(void)
2792 {
2793         struct list_head *pos, *n;
2794         struct ext4_li_request *elr;
2795
2796         mutex_lock(&ext4_li_info->li_list_mtx);
2797         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2798                 elr = list_entry(pos, struct ext4_li_request,
2799                                  lr_request);
2800                 ext4_remove_li_request(elr);
2801         }
2802         mutex_unlock(&ext4_li_info->li_list_mtx);
2803 }
2804
2805 static int ext4_run_lazyinit_thread(void)
2806 {
2807         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2808                                          ext4_li_info, "ext4lazyinit");
2809         if (IS_ERR(ext4_lazyinit_task)) {
2810                 int err = PTR_ERR(ext4_lazyinit_task);
2811                 ext4_clear_request_list();
2812                 kfree(ext4_li_info);
2813                 ext4_li_info = NULL;
2814                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2815                                  "initialization thread\n",
2816                                  err);
2817                 return err;
2818         }
2819         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2820         return 0;
2821 }
2822
2823 /*
2824  * Check whether it make sense to run itable init. thread or not.
2825  * If there is at least one uninitialized inode table, return
2826  * corresponding group number, else the loop goes through all
2827  * groups and return total number of groups.
2828  */
2829 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2830 {
2831         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2832         struct ext4_group_desc *gdp = NULL;
2833
2834         for (group = 0; group < ngroups; group++) {
2835                 gdp = ext4_get_group_desc(sb, group, NULL);
2836                 if (!gdp)
2837                         continue;
2838
2839                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2840                         break;
2841         }
2842
2843         return group;
2844 }
2845
2846 static int ext4_li_info_new(void)
2847 {
2848         struct ext4_lazy_init *eli = NULL;
2849
2850         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2851         if (!eli)
2852                 return -ENOMEM;
2853
2854         INIT_LIST_HEAD(&eli->li_request_list);
2855         mutex_init(&eli->li_list_mtx);
2856
2857         eli->li_state |= EXT4_LAZYINIT_QUIT;
2858
2859         ext4_li_info = eli;
2860
2861         return 0;
2862 }
2863
2864 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2865                                             ext4_group_t start)
2866 {
2867         struct ext4_sb_info *sbi = EXT4_SB(sb);
2868         struct ext4_li_request *elr;
2869         unsigned long rnd;
2870
2871         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2872         if (!elr)
2873                 return NULL;
2874
2875         elr->lr_super = sb;
2876         elr->lr_sbi = sbi;
2877         elr->lr_next_group = start;
2878
2879         /*
2880          * Randomize first schedule time of the request to
2881          * spread the inode table initialization requests
2882          * better.
2883          */
2884         get_random_bytes(&rnd, sizeof(rnd));
2885         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2886                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2887
2888         return elr;
2889 }
2890
2891 static int ext4_register_li_request(struct super_block *sb,
2892                                     ext4_group_t first_not_zeroed)
2893 {
2894         struct ext4_sb_info *sbi = EXT4_SB(sb);
2895         struct ext4_li_request *elr;
2896         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2897         int ret = 0;
2898
2899         if (sbi->s_li_request != NULL) {
2900                 /*
2901                  * Reset timeout so it can be computed again, because
2902                  * s_li_wait_mult might have changed.
2903                  */
2904                 sbi->s_li_request->lr_timeout = 0;
2905                 return 0;
2906         }
2907
2908         if (first_not_zeroed == ngroups ||
2909             (sb->s_flags & MS_RDONLY) ||
2910             !test_opt(sb, INIT_INODE_TABLE))
2911                 return 0;
2912
2913         elr = ext4_li_request_new(sb, first_not_zeroed);
2914         if (!elr)
2915                 return -ENOMEM;
2916
2917         mutex_lock(&ext4_li_mtx);
2918
2919         if (NULL == ext4_li_info) {
2920                 ret = ext4_li_info_new();
2921                 if (ret)
2922                         goto out;
2923         }
2924
2925         mutex_lock(&ext4_li_info->li_list_mtx);
2926         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2927         mutex_unlock(&ext4_li_info->li_list_mtx);
2928
2929         sbi->s_li_request = elr;
2930         /*
2931          * set elr to NULL here since it has been inserted to
2932          * the request_list and the removal and free of it is
2933          * handled by ext4_clear_request_list from now on.
2934          */
2935         elr = NULL;
2936
2937         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2938                 ret = ext4_run_lazyinit_thread();
2939                 if (ret)
2940                         goto out;
2941         }
2942 out:
2943         mutex_unlock(&ext4_li_mtx);
2944         if (ret)
2945                 kfree(elr);
2946         return ret;
2947 }
2948
2949 /*
2950  * We do not need to lock anything since this is called on
2951  * module unload.
2952  */
2953 static void ext4_destroy_lazyinit_thread(void)
2954 {
2955         /*
2956          * If thread exited earlier
2957          * there's nothing to be done.
2958          */
2959         if (!ext4_li_info || !ext4_lazyinit_task)
2960                 return;
2961
2962         kthread_stop(ext4_lazyinit_task);
2963 }
2964
2965 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2966 {
2967         char *orig_data = kstrdup(data, GFP_KERNEL);
2968         struct buffer_head *bh;
2969         struct ext4_super_block *es = NULL;
2970         struct ext4_sb_info *sbi;
2971         ext4_fsblk_t block;
2972         ext4_fsblk_t sb_block = get_sb_block(&data);
2973         ext4_fsblk_t logical_sb_block;
2974         unsigned long offset = 0;
2975         unsigned long journal_devnum = 0;
2976         unsigned long def_mount_opts;
2977         struct inode *root;
2978         char *cp;
2979         const char *descr;
2980         int ret = -ENOMEM;
2981         int blocksize, clustersize;
2982         unsigned int db_count;
2983         unsigned int i;
2984         int needs_recovery, has_huge_files, has_bigalloc;
2985         __u64 blocks_count;
2986         int err;
2987         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2988         ext4_group_t first_not_zeroed;
2989         unsigned long acl_flags = 0;
2990
2991         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2992         if (!sbi)
2993                 goto out_free_orig;
2994
2995         sbi->s_blockgroup_lock =
2996                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2997         if (!sbi->s_blockgroup_lock) {
2998                 kfree(sbi);
2999                 goto out_free_orig;
3000         }
3001         sb->s_fs_info = sbi;
3002         sbi->s_mount_opt = 0;
3003         sbi->s_resuid = EXT4_DEF_RESUID;
3004         sbi->s_resgid = EXT4_DEF_RESGID;
3005         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3006         sbi->s_sb_block = sb_block;
3007         if (sb->s_bdev->bd_part)
3008                 sbi->s_sectors_written_start =
3009                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3010
3011         /* Cleanup superblock name */
3012         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3013                 *cp = '!';
3014
3015         ret = -EINVAL;
3016         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3017         if (!blocksize) {
3018                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3019                 goto out_fail;
3020         }
3021
3022         /*
3023          * The ext4 superblock will not be buffer aligned for other than 1kB
3024          * block sizes.  We need to calculate the offset from buffer start.
3025          */
3026         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3027                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3028                 offset = do_div(logical_sb_block, blocksize);
3029         } else {
3030                 logical_sb_block = sb_block;
3031         }
3032
3033         if (!(bh = sb_bread(sb, logical_sb_block))) {
3034                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3035                 goto out_fail;
3036         }
3037         /*
3038          * Note: s_es must be initialized as soon as possible because
3039          *       some ext4 macro-instructions depend on its value
3040          */
3041         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3042         sbi->s_es = es;
3043         sb->s_magic = le16_to_cpu(es->s_magic);
3044         if (sb->s_magic != EXT4_SUPER_MAGIC)
3045                 goto cantfind_ext4;
3046         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3047
3048         /* Set defaults before we parse the mount options */
3049         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3050         set_opt(sb, INIT_INODE_TABLE);
3051         if (def_mount_opts & EXT4_DEFM_DEBUG)
3052                 set_opt(sb, DEBUG);
3053         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3054                 set_opt(sb, GRPID);
3055         if (def_mount_opts & EXT4_DEFM_UID16)
3056                 set_opt(sb, NO_UID32);
3057         /* xattr user namespace & acls are now defaulted on */
3058 #ifdef CONFIG_EXT4_FS_XATTR
3059         set_opt(sb, XATTR_USER);
3060 #endif
3061 #if defined(CONFIG_EXT4_FS_POSIX_ACL)
3062         set_opt(sb, POSIX_ACL);
3063 #endif
3064         set_opt(sb, MBLK_IO_SUBMIT);
3065         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3066                 set_opt(sb, JOURNAL_DATA);
3067         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3068                 set_opt(sb, ORDERED_DATA);
3069         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3070                 set_opt(sb, WRITEBACK_DATA);
3071
3072         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3073                 set_opt(sb, ERRORS_PANIC);
3074         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3075                 set_opt(sb, ERRORS_CONT);
3076         else
3077                 set_opt(sb, ERRORS_RO);
3078         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3079                 set_opt(sb, BLOCK_VALIDITY);
3080         if (def_mount_opts & EXT4_DEFM_DISCARD)
3081                 set_opt(sb, DISCARD);
3082
3083         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3084         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3085         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3086         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3087         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3088
3089         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3090                 set_opt(sb, BARRIER);
3091
3092         /*
3093          * enable delayed allocation by default
3094          * Use -o nodelalloc to turn it off
3095          */
3096         if (!IS_EXT3_SB(sb) &&
3097             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3098                 set_opt(sb, DELALLOC);
3099
3100         /*
3101          * set default s_li_wait_mult for lazyinit, for the case there is
3102          * no mount option specified.
3103          */
3104         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3105
3106         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3107                            &journal_devnum, &journal_ioprio, 0)) {
3108                 ext4_msg(sb, KERN_WARNING,
3109                          "failed to parse options in superblock: %s",
3110                          sbi->s_es->s_mount_opts);
3111         }
3112         sbi->s_def_mount_opt = sbi->s_mount_opt;
3113         if (!parse_options((char *) data, sb, &journal_devnum,
3114                            &journal_ioprio, 0))
3115                 goto failed_mount;
3116
3117         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3118                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3119                             "with data=journal disables delayed "
3120                             "allocation and O_DIRECT support!\n");
3121                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3122                         ext4_msg(sb, KERN_ERR, "can't mount with "
3123                                  "both data=journal and delalloc");
3124                         goto failed_mount;
3125                 }
3126                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3127                         ext4_msg(sb, KERN_ERR, "can't mount with "
3128                                  "both data=journal and delalloc");
3129                         goto failed_mount;
3130                 }
3131                 if (test_opt(sb, DELALLOC))
3132                         clear_opt(sb, DELALLOC);
3133         }
3134
3135         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3136         if (test_opt(sb, DIOREAD_NOLOCK)) {
3137                 if (blocksize < PAGE_SIZE) {
3138                         ext4_msg(sb, KERN_ERR, "can't mount with "
3139                                  "dioread_nolock if block size != PAGE_SIZE");
3140                         goto failed_mount;
3141                 }
3142         }
3143
3144         if (sb->s_flags & MS_RICHACL)
3145                 acl_flags = MS_RICHACL;
3146         else if (test_opt(sb, POSIX_ACL))
3147                 acl_flags = MS_POSIXACL;
3148
3149         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_RICHACL)) | acl_flags;
3150
3151         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3152             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3153              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3154              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3155                 ext4_msg(sb, KERN_WARNING,
3156                        "feature flags set on rev 0 fs, "
3157                        "running e2fsck is recommended");
3158
3159         if (IS_EXT2_SB(sb)) {
3160                 if (ext2_feature_set_ok(sb))
3161                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3162                                  "using the ext4 subsystem");
3163                 else {
3164                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3165                                  "to feature incompatibilities");
3166                         goto failed_mount;
3167                 }
3168         }
3169
3170         if (IS_EXT3_SB(sb)) {
3171                 if (ext3_feature_set_ok(sb))
3172                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3173                                  "using the ext4 subsystem");
3174                 else {
3175                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3176                                  "to feature incompatibilities");
3177                         goto failed_mount;
3178                 }
3179         }
3180
3181         /*
3182          * Check feature flags regardless of the revision level, since we
3183          * previously didn't change the revision level when setting the flags,
3184          * so there is a chance incompat flags are set on a rev 0 filesystem.
3185          */
3186         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3187                 goto failed_mount;
3188
3189         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3190             blocksize > EXT4_MAX_BLOCK_SIZE) {
3191                 ext4_msg(sb, KERN_ERR,
3192                        "Unsupported filesystem blocksize %d", blocksize);
3193                 goto failed_mount;
3194         }
3195
3196         if (sb->s_blocksize != blocksize) {
3197                 /* Validate the filesystem blocksize */
3198                 if (!sb_set_blocksize(sb, blocksize)) {
3199                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3200                                         blocksize);
3201                         goto failed_mount;
3202                 }
3203
3204                 brelse(bh);
3205                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3206                 offset = do_div(logical_sb_block, blocksize);
3207                 bh = sb_bread(sb, logical_sb_block);
3208                 if (!bh) {
3209                         ext4_msg(sb, KERN_ERR,
3210                                "Can't read superblock on 2nd try");
3211                         goto failed_mount;
3212                 }
3213                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3214                 sbi->s_es = es;
3215                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3216                         ext4_msg(sb, KERN_ERR,
3217                                "Magic mismatch, very weird!");
3218                         goto failed_mount;
3219                 }
3220         }
3221
3222         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3223                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3224         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3225                                                       has_huge_files);
3226         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3227
3228         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3229                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3230                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3231         } else {
3232                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3233                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3234                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3235                     (!is_power_of_2(sbi->s_inode_size)) ||
3236                     (sbi->s_inode_size > blocksize)) {
3237                         ext4_msg(sb, KERN_ERR,
3238                                "unsupported inode size: %d",
3239                                sbi->s_inode_size);
3240                         goto failed_mount;
3241                 }
3242                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3243                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3244         }
3245
3246         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3247         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3248                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3249                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3250                     !is_power_of_2(sbi->s_desc_size)) {
3251                         ext4_msg(sb, KERN_ERR,
3252                                "unsupported descriptor size %lu",
3253                                sbi->s_desc_size);
3254                         goto failed_mount;
3255                 }
3256         } else
3257                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3258
3259         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3260         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3261         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3262                 goto cantfind_ext4;
3263
3264         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3265         if (sbi->s_inodes_per_block == 0)
3266                 goto cantfind_ext4;
3267         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3268                                         sbi->s_inodes_per_block;
3269         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3270         sbi->s_sbh = bh;
3271         sbi->s_mount_state = le16_to_cpu(es->s_state);
3272         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3273         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3274
3275         for (i = 0; i < 4; i++)
3276                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3277         sbi->s_def_hash_version = es->s_def_hash_version;
3278         i = le32_to_cpu(es->s_flags);
3279         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3280                 sbi->s_hash_unsigned = 3;
3281         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3282 #ifdef __CHAR_UNSIGNED__
3283                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3284                 sbi->s_hash_unsigned = 3;
3285 #else
3286                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3287 #endif
3288         }
3289
3290         /* Handle clustersize */
3291         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3292         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3293                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3294         if (has_bigalloc) {
3295                 if (clustersize < blocksize) {
3296                         ext4_msg(sb, KERN_ERR,
3297                                  "cluster size (%d) smaller than "
3298                                  "block size (%d)", clustersize, blocksize);
3299                         goto failed_mount;
3300                 }
3301                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3302                         le32_to_cpu(es->s_log_block_size);
3303                 sbi->s_clusters_per_group =
3304                         le32_to_cpu(es->s_clusters_per_group);
3305                 if (sbi->s_clusters_per_group > blocksize * 8) {
3306                         ext4_msg(sb, KERN_ERR,
3307                                  "#clusters per group too big: %lu",
3308                                  sbi->s_clusters_per_group);
3309                         goto failed_mount;
3310                 }
3311                 if (sbi->s_blocks_per_group !=
3312                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3313                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3314                                  "clusters per group (%lu) inconsistent",
3315                                  sbi->s_blocks_per_group,
3316                                  sbi->s_clusters_per_group);
3317                         goto failed_mount;
3318                 }
3319         } else {
3320                 if (clustersize != blocksize) {
3321                         ext4_warning(sb, "fragment/cluster size (%d) != "
3322                                      "block size (%d)", clustersize,
3323                                      blocksize);
3324                         clustersize = blocksize;
3325                 }
3326                 if (sbi->s_blocks_per_group > blocksize * 8) {
3327                         ext4_msg(sb, KERN_ERR,
3328                                  "#blocks per group too big: %lu",
3329                                  sbi->s_blocks_per_group);
3330                         goto failed_mount;
3331                 }
3332                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3333                 sbi->s_cluster_bits = 0;
3334         }
3335         sbi->s_cluster_ratio = clustersize / blocksize;
3336
3337         if (sbi->s_inodes_per_group > blocksize * 8) {
3338                 ext4_msg(sb, KERN_ERR,
3339                        "#inodes per group too big: %lu",
3340                        sbi->s_inodes_per_group);
3341                 goto failed_mount;
3342         }
3343
3344         /*
3345          * Test whether we have more sectors than will fit in sector_t,
3346          * and whether the max offset is addressable by the page cache.
3347          */
3348         err = generic_check_addressable(sb->s_blocksize_bits,
3349                                         ext4_blocks_count(es));
3350         if (err) {
3351                 ext4_msg(sb, KERN_ERR, "filesystem"
3352                          " too large to mount safely on this system");
3353                 if (sizeof(sector_t) < 8)
3354                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3355                 ret = err;
3356                 goto failed_mount;
3357         }
3358
3359         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3360                 goto cantfind_ext4;
3361
3362         /* check blocks count against device size */
3363         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3364         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3365                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3366                        "exceeds size of device (%llu blocks)",
3367                        ext4_blocks_count(es), blocks_count);
3368                 goto failed_mount;
3369         }
3370
3371         /*
3372          * It makes no sense for the first data block to be beyond the end
3373          * of the filesystem.
3374          */
3375         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3376                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3377                          "block %u is beyond end of filesystem (%llu)",
3378                          le32_to_cpu(es->s_first_data_block),
3379                          ext4_blocks_count(es));
3380                 goto failed_mount;
3381         }
3382         blocks_count = (ext4_blocks_count(es) -
3383                         le32_to_cpu(es->s_first_data_block) +
3384                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3385         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3386         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3387                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3388                        "(block count %llu, first data block %u, "
3389                        "blocks per group %lu)", sbi->s_groups_count,
3390                        ext4_blocks_count(es),
3391                        le32_to_cpu(es->s_first_data_block),
3392                        EXT4_BLOCKS_PER_GROUP(sb));
3393                 goto failed_mount;
3394         }
3395         sbi->s_groups_count = blocks_count;
3396         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3397                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3398         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3399                    EXT4_DESC_PER_BLOCK(sb);
3400         sbi->s_group_desc = ext4_kvmalloc(db_count *
3401                                           sizeof(struct buffer_head *),
3402                                           GFP_KERNEL);
3403         if (sbi->s_group_desc == NULL) {
3404                 ext4_msg(sb, KERN_ERR, "not enough memory");
3405                 goto failed_mount;
3406         }
3407
3408         if (ext4_proc_root)
3409                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3410
3411         if (sbi->s_proc)
3412                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3413                                  &ext4_seq_options_fops, sb);
3414
3415         bgl_lock_init(sbi->s_blockgroup_lock);
3416
3417         for (i = 0; i < db_count; i++) {
3418                 block = descriptor_loc(sb, logical_sb_block, i);
3419                 sbi->s_group_desc[i] = sb_bread(sb, block);
3420                 if (!sbi->s_group_desc[i]) {
3421                         ext4_msg(sb, KERN_ERR,
3422                                "can't read group descriptor %d", i);
3423                         db_count = i;
3424                         goto failed_mount2;
3425                 }
3426         }
3427         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3428                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3429                 goto failed_mount2;
3430         }
3431         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3432                 if (!ext4_fill_flex_info(sb)) {
3433                         ext4_msg(sb, KERN_ERR,
3434                                "unable to initialize "
3435                                "flex_bg meta info!");
3436                         goto failed_mount2;
3437                 }
3438
3439         sbi->s_gdb_count = db_count;
3440         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3441         spin_lock_init(&sbi->s_next_gen_lock);
3442
3443         init_timer(&sbi->s_err_report);
3444         sbi->s_err_report.function = print_daily_error_info;
3445         sbi->s_err_report.data = (unsigned long) sb;
3446
3447         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3448                         ext4_count_free_clusters(sb));
3449         if (!err) {
3450                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3451                                 ext4_count_free_inodes(sb));
3452         }
3453         if (!err) {
3454                 err = percpu_counter_init(&sbi->s_dirs_counter,
3455                                 ext4_count_dirs(sb));
3456         }
3457         if (!err) {
3458                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3459         }
3460         if (err) {
3461                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3462                 goto failed_mount3;
3463         }
3464
3465         sbi->s_stripe = ext4_get_stripe_size(sbi);
3466         sbi->s_max_writeback_mb_bump = 128;
3467
3468         /*
3469          * set up enough so that it can read an inode
3470          */
3471         if (!test_opt(sb, NOLOAD) &&
3472             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3473                 sb->s_op = &ext4_sops;
3474         else
3475                 sb->s_op = &ext4_nojournal_sops;
3476         sb->s_export_op = &ext4_export_ops;
3477         sb->s_xattr = ext4_xattr_handlers;
3478 #ifdef CONFIG_QUOTA
3479         sb->s_qcop = &ext4_qctl_operations;
3480         sb->dq_op = &ext4_quota_operations;
3481 #endif
3482         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3483
3484         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3485         mutex_init(&sbi->s_orphan_lock);
3486         sbi->s_resize_flags = 0;
3487
3488         sb->s_root = NULL;
3489
3490         needs_recovery = (es->s_last_orphan != 0 ||
3491                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3492                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3493
3494         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3495             !(sb->s_flags & MS_RDONLY))
3496                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3497                         goto failed_mount3;
3498
3499         /*
3500          * The first inode we look at is the journal inode.  Don't try
3501          * root first: it may be modified in the journal!
3502          */
3503         if (!test_opt(sb, NOLOAD) &&
3504             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3505                 if (ext4_load_journal(sb, es, journal_devnum))
3506                         goto failed_mount3;
3507         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3508               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3509                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3510                        "suppressed and not mounted read-only");
3511                 goto failed_mount_wq;
3512         } else {
3513                 clear_opt(sb, DATA_FLAGS);
3514                 sbi->s_journal = NULL;
3515                 needs_recovery = 0;
3516                 goto no_journal;
3517         }
3518
3519         if (ext4_blocks_count(es) > 0xffffffffULL &&
3520             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3521                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3522                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3523                 goto failed_mount_wq;
3524         }
3525
3526         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3527                 jbd2_journal_set_features(sbi->s_journal,
3528                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3529                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3530         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3531                 jbd2_journal_set_features(sbi->s_journal,
3532                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3533                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3534                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3535         } else {
3536                 jbd2_journal_clear_features(sbi->s_journal,
3537                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3538                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3539         }
3540
3541         /* We have now updated the journal if required, so we can
3542          * validate the data journaling mode. */
3543         switch (test_opt(sb, DATA_FLAGS)) {
3544         case 0:
3545                 /* No mode set, assume a default based on the journal
3546                  * capabilities: ORDERED_DATA if the journal can
3547                  * cope, else JOURNAL_DATA
3548                  */
3549                 if (jbd2_journal_check_available_features
3550                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3551                         set_opt(sb, ORDERED_DATA);
3552                 else
3553                         set_opt(sb, JOURNAL_DATA);
3554                 break;
3555
3556         case EXT4_MOUNT_ORDERED_DATA:
3557         case EXT4_MOUNT_WRITEBACK_DATA:
3558                 if (!jbd2_journal_check_available_features
3559                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3560                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3561                                "requested data journaling mode");
3562                         goto failed_mount_wq;
3563                 }
3564         default:
3565                 break;
3566         }
3567         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3568
3569         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3570
3571         /*
3572          * The journal may have updated the bg summary counts, so we
3573          * need to update the global counters.
3574          */
3575         percpu_counter_set(&sbi->s_freeclusters_counter,
3576                            ext4_count_free_clusters(sb));
3577         percpu_counter_set(&sbi->s_freeinodes_counter,
3578                            ext4_count_free_inodes(sb));
3579         percpu_counter_set(&sbi->s_dirs_counter,
3580                            ext4_count_dirs(sb));
3581         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3582
3583 no_journal:
3584         /*
3585          * The maximum number of concurrent works can be high and
3586          * concurrency isn't really necessary.  Limit it to 1.
3587          */
3588         EXT4_SB(sb)->dio_unwritten_wq =
3589                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3590         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3591                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3592                 goto failed_mount_wq;
3593         }
3594
3595         /*
3596          * The jbd2_journal_load will have done any necessary log recovery,
3597          * so we can safely mount the rest of the filesystem now.
3598          */
3599
3600         root = ext4_iget(sb, EXT4_ROOT_INO);
3601         if (IS_ERR(root)) {
3602                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3603                 ret = PTR_ERR(root);
3604                 root = NULL;
3605                 goto failed_mount4;
3606         }
3607         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3608                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3609                 iput(root);
3610                 goto failed_mount4;
3611         }
3612         sb->s_root = d_make_root(root);
3613         if (!sb->s_root) {
3614                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3615                 ret = -ENOMEM;
3616                 goto failed_mount4;
3617         }
3618
3619         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3620
3621         /* determine the minimum size of new large inodes, if present */
3622         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3623                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3624                                                      EXT4_GOOD_OLD_INODE_SIZE;
3625                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3626                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3627                         if (sbi->s_want_extra_isize <
3628                             le16_to_cpu(es->s_want_extra_isize))
3629                                 sbi->s_want_extra_isize =
3630                                         le16_to_cpu(es->s_want_extra_isize);
3631                         if (sbi->s_want_extra_isize <
3632                             le16_to_cpu(es->s_min_extra_isize))
3633                                 sbi->s_want_extra_isize =
3634                                         le16_to_cpu(es->s_min_extra_isize);
3635                 }
3636         }
3637         /* Check if enough inode space is available */
3638         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3639                                                         sbi->s_inode_size) {
3640                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3641                                                        EXT4_GOOD_OLD_INODE_SIZE;
3642                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3643                          "available");
3644         }
3645
3646         err = ext4_setup_system_zone(sb);
3647         if (err) {
3648                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3649                          "zone (%d)", err);
3650                 goto failed_mount4a;
3651         }
3652
3653         ext4_ext_init(sb);
3654         err = ext4_mb_init(sb, needs_recovery);
3655         if (err) {
3656                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3657                          err);
3658                 goto failed_mount5;
3659         }
3660
3661         err = ext4_register_li_request(sb, first_not_zeroed);
3662         if (err)
3663                 goto failed_mount6;
3664
3665         sbi->s_kobj.kset = ext4_kset;
3666         init_completion(&sbi->s_kobj_unregister);
3667         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3668                                    "%s", sb->s_id);
3669         if (err)
3670                 goto failed_mount7;
3671
3672         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3673         ext4_orphan_cleanup(sb, es);
3674         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3675         if (needs_recovery) {
3676                 ext4_msg(sb, KERN_INFO, "recovery complete");
3677                 ext4_mark_recovery_complete(sb, es);
3678         }
3679         if (EXT4_SB(sb)->s_journal) {
3680                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3681                         descr = " journalled data mode";
3682                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3683                         descr = " ordered data mode";
3684                 else
3685                         descr = " writeback data mode";
3686         } else
3687                 descr = "out journal";
3688
3689         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3690                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3691                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3692
3693         if (es->s_error_count)
3694                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3695
3696         kfree(orig_data);
3697         return 0;
3698
3699 cantfind_ext4:
3700         if (!silent)
3701                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3702         goto failed_mount;
3703
3704 failed_mount7:
3705         ext4_unregister_li_request(sb);
3706 failed_mount6:
3707         ext4_mb_release(sb);
3708 failed_mount5:
3709         ext4_ext_release(sb);
3710         ext4_release_system_zone(sb);
3711 failed_mount4a:
3712         dput(sb->s_root);
3713         sb->s_root = NULL;
3714 failed_mount4:
3715         ext4_msg(sb, KERN_ERR, "mount failed");
3716         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3717 failed_mount_wq:
3718         if (sbi->s_journal) {
3719                 jbd2_journal_destroy(sbi->s_journal);
3720                 sbi->s_journal = NULL;
3721         }
3722 failed_mount3:
3723         del_timer(&sbi->s_err_report);
3724         if (sbi->s_flex_groups)
3725                 ext4_kvfree(sbi->s_flex_groups);
3726         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3727         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3728         percpu_counter_destroy(&sbi->s_dirs_counter);
3729         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3730         if (sbi->s_mmp_tsk)
3731                 kthread_stop(sbi->s_mmp_tsk);
3732 failed_mount2:
3733         for (i = 0; i < db_count; i++)
3734                 brelse(sbi->s_group_desc[i]);
3735         ext4_kvfree(sbi->s_group_desc);
3736 failed_mount:
3737         if (sbi->s_proc) {
3738                 remove_proc_entry("options", sbi->s_proc);
3739                 remove_proc_entry(sb->s_id, ext4_proc_root);
3740         }
3741 #ifdef CONFIG_QUOTA
3742         for (i = 0; i < MAXQUOTAS; i++)
3743                 kfree(sbi->s_qf_names[i]);
3744 #endif
3745         ext4_blkdev_remove(sbi);
3746         brelse(bh);
3747 out_fail:
3748         sb->s_fs_info = NULL;
3749         kfree(sbi->s_blockgroup_lock);
3750         kfree(sbi);
3751 out_free_orig:
3752         kfree(orig_data);
3753         return ret;
3754 }
3755
3756 /*
3757  * Setup any per-fs journal parameters now.  We'll do this both on
3758  * initial mount, once the journal has been initialised but before we've
3759  * done any recovery; and again on any subsequent remount.
3760  */
3761 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3762 {
3763         struct ext4_sb_info *sbi = EXT4_SB(sb);
3764
3765         journal->j_commit_interval = sbi->s_commit_interval;
3766         journal->j_min_batch_time = sbi->s_min_batch_time;
3767         journal->j_max_batch_time = sbi->s_max_batch_time;
3768
3769         write_lock(&journal->j_state_lock);
3770         if (test_opt(sb, BARRIER))
3771                 journal->j_flags |= JBD2_BARRIER;
3772         else
3773                 journal->j_flags &= ~JBD2_BARRIER;
3774         if (test_opt(sb, DATA_ERR_ABORT))
3775                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3776         else
3777                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3778         write_unlock(&journal->j_state_lock);
3779 }
3780
3781 static journal_t *ext4_get_journal(struct super_block *sb,
3782                                    unsigned int journal_inum)
3783 {
3784         struct inode *journal_inode;
3785         journal_t *journal;
3786
3787         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3788
3789         /* First, test for the existence of a valid inode on disk.  Bad
3790          * things happen if we iget() an unused inode, as the subsequent
3791          * iput() will try to delete it. */
3792
3793         journal_inode = ext4_iget(sb, journal_inum);
3794         if (IS_ERR(journal_inode)) {
3795                 ext4_msg(sb, KERN_ERR, "no journal found");
3796                 return NULL;
3797         }
3798         if (!journal_inode->i_nlink) {
3799                 make_bad_inode(journal_inode);
3800                 iput(journal_inode);
3801                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3802                 return NULL;
3803         }
3804
3805         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3806                   journal_inode, journal_inode->i_size);
3807         if (!S_ISREG(journal_inode->i_mode)) {
3808                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3809                 iput(journal_inode);
3810                 return NULL;
3811         }
3812
3813         journal = jbd2_journal_init_inode(journal_inode);
3814         if (!journal) {
3815                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3816                 iput(journal_inode);
3817                 return NULL;
3818         }
3819         journal->j_private = sb;
3820         ext4_init_journal_params(sb, journal);
3821         return journal;
3822 }
3823
3824 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3825                                        dev_t j_dev)
3826 {
3827         struct buffer_head *bh;
3828         journal_t *journal;
3829         ext4_fsblk_t start;
3830         ext4_fsblk_t len;
3831         int hblock, blocksize;
3832         ext4_fsblk_t sb_block;
3833         unsigned long offset;
3834         struct ext4_super_block *es;
3835         struct block_device *bdev;
3836
3837         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3838
3839         bdev = ext4_blkdev_get(j_dev, sb);
3840         if (bdev == NULL)
3841                 return NULL;
3842
3843         blocksize = sb->s_blocksize;
3844         hblock = bdev_logical_block_size(bdev);
3845         if (blocksize < hblock) {
3846                 ext4_msg(sb, KERN_ERR,
3847                         "blocksize too small for journal device");
3848                 goto out_bdev;
3849         }
3850
3851         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3852         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3853         set_blocksize(bdev, blocksize);
3854         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3855                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3856                        "external journal");
3857                 goto out_bdev;
3858         }
3859
3860         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3861         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3862             !(le32_to_cpu(es->s_feature_incompat) &
3863               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3864                 ext4_msg(sb, KERN_ERR, "external journal has "
3865                                         "bad superblock");
3866                 brelse(bh);
3867                 goto out_bdev;
3868         }
3869
3870         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3871                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3872                 brelse(bh);
3873                 goto out_bdev;
3874         }
3875
3876         len = ext4_blocks_count(es);
3877         start = sb_block + 1;
3878         brelse(bh);     /* we're done with the superblock */
3879
3880         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3881                                         start, len, blocksize);
3882         if (!journal) {
3883                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3884                 goto out_bdev;
3885         }
3886         journal->j_private = sb;
3887         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3888         wait_on_buffer(journal->j_sb_buffer);
3889         if (!buffer_uptodate(journal->j_sb_buffer)) {
3890                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3891                 goto out_journal;
3892         }
3893         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3894                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3895                                         "user (unsupported) - %d",
3896                         be32_to_cpu(journal->j_superblock->s_nr_users));
3897                 goto out_journal;
3898         }
3899         EXT4_SB(sb)->journal_bdev = bdev;
3900         ext4_init_journal_params(sb, journal);
3901         return journal;
3902
3903 out_journal:
3904         jbd2_journal_destroy(journal);
3905 out_bdev:
3906         ext4_blkdev_put(bdev);
3907         return NULL;
3908 }
3909
3910 static int ext4_load_journal(struct super_block *sb,
3911                              struct ext4_super_block *es,
3912                              unsigned long journal_devnum)
3913 {
3914         journal_t *journal;
3915         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3916         dev_t journal_dev;
3917         int err = 0;
3918         int really_read_only;
3919
3920         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3921
3922         if (journal_devnum &&
3923             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3924                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3925                         "numbers have changed");
3926                 journal_dev = new_decode_dev(journal_devnum);
3927         } else
3928                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3929
3930         really_read_only = bdev_read_only(sb->s_bdev);
3931
3932         /*
3933          * Are we loading a blank journal or performing recovery after a
3934          * crash?  For recovery, we need to check in advance whether we
3935          * can get read-write access to the device.
3936          */
3937         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3938                 if (sb->s_flags & MS_RDONLY) {
3939                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3940                                         "required on readonly filesystem");
3941                         if (really_read_only) {
3942                                 ext4_msg(sb, KERN_ERR, "write access "
3943                                         "unavailable, cannot proceed");
3944                                 return -EROFS;
3945                         }
3946                         ext4_msg(sb, KERN_INFO, "write access will "
3947                                "be enabled during recovery");
3948                 }
3949         }
3950
3951         if (journal_inum && journal_dev) {
3952                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3953                        "and inode journals!");
3954                 return -EINVAL;
3955         }
3956
3957         if (journal_inum) {
3958                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3959                         return -EINVAL;
3960         } else {
3961                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3962                         return -EINVAL;
3963         }
3964
3965         if (!(journal->j_flags & JBD2_BARRIER))
3966                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3967
3968         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3969                 err = jbd2_journal_wipe(journal, !really_read_only);
3970         if (!err) {
3971                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3972                 if (save)
3973                         memcpy(save, ((char *) es) +
3974                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3975                 err = jbd2_journal_load(journal);
3976                 if (save)
3977                         memcpy(((char *) es) + EXT4_S_ERR_START,
3978                                save, EXT4_S_ERR_LEN);
3979                 kfree(save);
3980         }
3981
3982         if (err) {
3983                 ext4_msg(sb, KERN_ERR, "error loading journal");
3984                 jbd2_journal_destroy(journal);
3985                 return err;
3986         }
3987
3988         EXT4_SB(sb)->s_journal = journal;
3989         ext4_clear_journal_err(sb, es);
3990
3991         if (!really_read_only && journal_devnum &&
3992             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3993                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3994
3995                 /* Make sure we flush the recovery flag to disk. */
3996                 ext4_commit_super(sb, 1);
3997         }
3998
3999         return 0;
4000 }
4001
4002 static int ext4_commit_super(struct super_block *sb, int sync)
4003 {
4004         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4005         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4006         int error = 0;
4007
4008         if (!sbh || block_device_ejected(sb))
4009                 return error;
4010         if (buffer_write_io_error(sbh)) {
4011                 /*
4012                  * Oh, dear.  A previous attempt to write the
4013                  * superblock failed.  This could happen because the
4014                  * USB device was yanked out.  Or it could happen to
4015                  * be a transient write error and maybe the block will
4016                  * be remapped.  Nothing we can do but to retry the
4017                  * write and hope for the best.
4018                  */
4019                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4020                        "superblock detected");
4021                 clear_buffer_write_io_error(sbh);
4022                 set_buffer_uptodate(sbh);
4023         }
4024         /*
4025          * If the file system is mounted read-only, don't update the
4026          * superblock write time.  This avoids updating the superblock
4027          * write time when we are mounting the root file system
4028          * read/only but we need to replay the journal; at that point,
4029          * for people who are east of GMT and who make their clock
4030          * tick in localtime for Windows bug-for-bug compatibility,
4031          * the clock is set in the future, and this will cause e2fsck
4032          * to complain and force a full file system check.
4033          */
4034         if (!(sb->s_flags & MS_RDONLY))
4035                 es->s_wtime = cpu_to_le32(get_seconds());
4036         if (sb->s_bdev->bd_part)
4037                 es->s_kbytes_written =
4038                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4039                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4040                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4041         else
4042                 es->s_kbytes_written =
4043                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4044         ext4_free_blocks_count_set(es,
4045                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4046                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4047         es->s_free_inodes_count =
4048                 cpu_to_le32(percpu_counter_sum_positive(
4049                                 &EXT4_SB(sb)->s_freeinodes_counter));
4050         sb->s_dirt = 0;
4051         BUFFER_TRACE(sbh, "marking dirty");
4052         mark_buffer_dirty(sbh);
4053         if (sync) {
4054                 error = sync_dirty_buffer(sbh);
4055                 if (error)
4056                         return error;
4057
4058                 error = buffer_write_io_error(sbh);
4059                 if (error) {
4060                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4061                                "superblock");
4062                         clear_buffer_write_io_error(sbh);
4063                         set_buffer_uptodate(sbh);
4064                 }
4065         }
4066         return error;
4067 }
4068
4069 /*
4070  * Have we just finished recovery?  If so, and if we are mounting (or
4071  * remounting) the filesystem readonly, then we will end up with a
4072  * consistent fs on disk.  Record that fact.
4073  */
4074 static void ext4_mark_recovery_complete(struct super_block *sb,
4075                                         struct ext4_super_block *es)
4076 {
4077         journal_t *journal = EXT4_SB(sb)->s_journal;
4078
4079         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4080                 BUG_ON(journal != NULL);
4081                 return;
4082         }
4083         jbd2_journal_lock_updates(journal);
4084         if (jbd2_journal_flush(journal) < 0)
4085                 goto out;
4086
4087         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4088             sb->s_flags & MS_RDONLY) {
4089                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4090                 ext4_commit_super(sb, 1);
4091         }
4092
4093 out:
4094         jbd2_journal_unlock_updates(journal);
4095 }
4096
4097 /*
4098  * If we are mounting (or read-write remounting) a filesystem whose journal
4099  * has recorded an error from a previous lifetime, move that error to the
4100  * main filesystem now.
4101  */
4102 static void ext4_clear_journal_err(struct super_block *sb,
4103                                    struct ext4_super_block *es)
4104 {
4105         journal_t *journal;
4106         int j_errno;
4107         const char *errstr;
4108
4109         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4110
4111         journal = EXT4_SB(sb)->s_journal;
4112
4113         /*
4114          * Now check for any error status which may have been recorded in the
4115          * journal by a prior ext4_error() or ext4_abort()
4116          */
4117
4118         j_errno = jbd2_journal_errno(journal);
4119         if (j_errno) {
4120                 char nbuf[16];
4121
4122                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4123                 ext4_warning(sb, "Filesystem error recorded "
4124                              "from previous mount: %s", errstr);
4125                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4126
4127                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4128                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4129                 ext4_commit_super(sb, 1);
4130
4131                 jbd2_journal_clear_err(journal);
4132         }
4133 }
4134
4135 /*
4136  * Force the running and committing transactions to commit,
4137  * and wait on the commit.
4138  */
4139 int ext4_force_commit(struct super_block *sb)
4140 {
4141         journal_t *journal;
4142         int ret = 0;
4143
4144         if (sb->s_flags & MS_RDONLY)
4145                 return 0;
4146
4147         journal = EXT4_SB(sb)->s_journal;
4148         if (journal) {
4149                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4150                 ret = ext4_journal_force_commit(journal);
4151         }
4152
4153         return ret;
4154 }
4155
4156 static void ext4_write_super(struct super_block *sb)
4157 {
4158         lock_super(sb);
4159         ext4_commit_super(sb, 1);
4160         unlock_super(sb);
4161 }
4162
4163 static int ext4_sync_fs(struct super_block *sb, int wait)
4164 {
4165         int ret = 0;
4166         tid_t target;
4167         struct ext4_sb_info *sbi = EXT4_SB(sb);
4168
4169         trace_ext4_sync_fs(sb, wait);
4170         flush_workqueue(sbi->dio_unwritten_wq);
4171         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4172                 if (wait)
4173                         jbd2_log_wait_commit(sbi->s_journal, target);
4174         }
4175         return ret;
4176 }
4177
4178 /*
4179  * LVM calls this function before a (read-only) snapshot is created.  This
4180  * gives us a chance to flush the journal completely and mark the fs clean.
4181  *
4182  * Note that only this function cannot bring a filesystem to be in a clean
4183  * state independently, because ext4 prevents a new handle from being started
4184  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4185  * the upper layer.
4186  */
4187 static int ext4_freeze(struct super_block *sb)
4188 {
4189         int error = 0;
4190         journal_t *journal;
4191
4192         if (sb->s_flags & MS_RDONLY)
4193                 return 0;
4194
4195         journal = EXT4_SB(sb)->s_journal;
4196
4197         /* Now we set up the journal barrier. */
4198         jbd2_journal_lock_updates(journal);
4199
4200         /*
4201          * Don't clear the needs_recovery flag if we failed to flush
4202          * the journal.
4203          */
4204         error = jbd2_journal_flush(journal);
4205         if (error < 0)
4206                 goto out;
4207
4208         /* Journal blocked and flushed, clear needs_recovery flag. */
4209         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4210         error = ext4_commit_super(sb, 1);
4211 out:
4212         /* we rely on s_frozen to stop further updates */
4213         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4214         return error;
4215 }
4216
4217 /*
4218  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4219  * flag here, even though the filesystem is not technically dirty yet.
4220  */
4221 static int ext4_unfreeze(struct super_block *sb)
4222 {
4223         if (sb->s_flags & MS_RDONLY)
4224                 return 0;
4225
4226         lock_super(sb);
4227         /* Reset the needs_recovery flag before the fs is unlocked. */
4228         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4229         ext4_commit_super(sb, 1);
4230         unlock_super(sb);
4231         return 0;
4232 }
4233
4234 /*
4235  * Structure to save mount options for ext4_remount's benefit
4236  */
4237 struct ext4_mount_options {
4238         unsigned long s_mount_opt;
4239         unsigned long s_mount_opt2;
4240         uid_t s_resuid;
4241         gid_t s_resgid;
4242         unsigned long s_commit_interval;
4243         u32 s_min_batch_time, s_max_batch_time;
4244 #ifdef CONFIG_QUOTA
4245         int s_jquota_fmt;
4246         char *s_qf_names[MAXQUOTAS];
4247 #endif
4248 };
4249
4250 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4251 {
4252         struct ext4_super_block *es;
4253         struct ext4_sb_info *sbi = EXT4_SB(sb);
4254         unsigned long old_sb_flags;
4255         struct ext4_mount_options old_opts;
4256         int enable_quota = 0;
4257         ext4_group_t g;
4258         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4259         int err = 0;
4260         unsigned long acl_flags = 0;
4261 #ifdef CONFIG_QUOTA
4262         int i;
4263 #endif
4264         char *orig_data = kstrdup(data, GFP_KERNEL);
4265
4266         /* Store the original options */
4267         lock_super(sb);
4268         old_sb_flags = sb->s_flags;
4269         old_opts.s_mount_opt = sbi->s_mount_opt;
4270         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4271         old_opts.s_resuid = sbi->s_resuid;
4272         old_opts.s_resgid = sbi->s_resgid;
4273         old_opts.s_commit_interval = sbi->s_commit_interval;
4274         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4275         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4276 #ifdef CONFIG_QUOTA
4277         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4278         for (i = 0; i < MAXQUOTAS; i++)
4279                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4280 #endif
4281         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4282                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4283
4284         /*
4285          * Allow the "check" option to be passed as a remount option.
4286          */
4287         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4288                 err = -EINVAL;
4289                 goto restore_opts;
4290         }
4291
4292         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4293                 ext4_abort(sb, "Abort forced by user");
4294
4295         if (sb->s_flags & MS_RICHACL)
4296                 acl_flags = MS_RICHACL;
4297         else if (test_opt(sb, POSIX_ACL))
4298                 acl_flags = MS_POSIXACL;
4299
4300         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_RICHACL)) | acl_flags;
4301
4302         es = sbi->s_es;
4303
4304         if (sbi->s_journal) {
4305                 ext4_init_journal_params(sb, sbi->s_journal);
4306                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4307         }
4308
4309         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4310                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4311                         err = -EROFS;
4312                         goto restore_opts;
4313                 }
4314
4315                 if (*flags & MS_RDONLY) {
4316                         err = dquot_suspend(sb, -1);
4317                         if (err < 0)
4318                                 goto restore_opts;
4319
4320                         /*
4321                          * First of all, the unconditional stuff we have to do
4322                          * to disable replay of the journal when we next remount
4323                          */
4324                         sb->s_flags |= MS_RDONLY;
4325
4326                         /*
4327                          * OK, test if we are remounting a valid rw partition
4328                          * readonly, and if so set the rdonly flag and then
4329                          * mark the partition as valid again.
4330                          */
4331                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4332                             (sbi->s_mount_state & EXT4_VALID_FS))
4333                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4334
4335                         if (sbi->s_journal)
4336                                 ext4_mark_recovery_complete(sb, es);
4337                 } else {
4338                         /* Make sure we can mount this feature set readwrite */
4339                         if (!ext4_feature_set_ok(sb, 0)) {
4340                                 err = -EROFS;
4341                                 goto restore_opts;
4342                         }
4343                         /*
4344                          * Make sure the group descriptor checksums
4345                          * are sane.  If they aren't, refuse to remount r/w.
4346                          */
4347                         for (g = 0; g < sbi->s_groups_count; g++) {
4348                                 struct ext4_group_desc *gdp =
4349                                         ext4_get_group_desc(sb, g, NULL);
4350
4351                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4352                                         ext4_msg(sb, KERN_ERR,
4353                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4354                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4355                                                le16_to_cpu(gdp->bg_checksum));
4356                                         err = -EINVAL;
4357                                         goto restore_opts;
4358                                 }
4359                         }
4360
4361                         /*
4362                          * If we have an unprocessed orphan list hanging
4363                          * around from a previously readonly bdev mount,
4364                          * require a full umount/remount for now.
4365                          */
4366                         if (es->s_last_orphan) {
4367                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4368                                        "remount RDWR because of unprocessed "
4369                                        "orphan inode list.  Please "
4370                                        "umount/remount instead");
4371                                 err = -EINVAL;
4372                                 goto restore_opts;
4373                         }
4374
4375                         /*
4376                          * Mounting a RDONLY partition read-write, so reread
4377                          * and store the current valid flag.  (It may have
4378                          * been changed by e2fsck since we originally mounted
4379                          * the partition.)
4380                          */
4381                         if (sbi->s_journal)
4382                                 ext4_clear_journal_err(sb, es);
4383                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4384                         if (!ext4_setup_super(sb, es, 0))
4385                                 sb->s_flags &= ~MS_RDONLY;
4386                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4387                                                      EXT4_FEATURE_INCOMPAT_MMP))
4388                                 if (ext4_multi_mount_protect(sb,
4389                                                 le64_to_cpu(es->s_mmp_block))) {
4390                                         err = -EROFS;
4391                                         goto restore_opts;
4392                                 }
4393                         enable_quota = 1;
4394                 }
4395         }
4396
4397         /*
4398          * Reinitialize lazy itable initialization thread based on
4399          * current settings
4400          */
4401         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4402                 ext4_unregister_li_request(sb);
4403         else {
4404                 ext4_group_t first_not_zeroed;
4405                 first_not_zeroed = ext4_has_uninit_itable(sb);
4406                 ext4_register_li_request(sb, first_not_zeroed);
4407         }
4408
4409         ext4_setup_system_zone(sb);
4410         if (sbi->s_journal == NULL)
4411                 ext4_commit_super(sb, 1);
4412
4413 #ifdef CONFIG_QUOTA
4414         /* Release old quota file names */
4415         for (i = 0; i < MAXQUOTAS; i++)
4416                 if (old_opts.s_qf_names[i] &&
4417                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4418                         kfree(old_opts.s_qf_names[i]);
4419 #endif
4420         unlock_super(sb);
4421         if (enable_quota)
4422                 dquot_resume(sb, -1);
4423
4424         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4425         kfree(orig_data);
4426         return 0;
4427
4428 restore_opts:
4429         sb->s_flags = old_sb_flags;
4430         sbi->s_mount_opt = old_opts.s_mount_opt;
4431         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4432         sbi->s_resuid = old_opts.s_resuid;
4433         sbi->s_resgid = old_opts.s_resgid;
4434         sbi->s_commit_interval = old_opts.s_commit_interval;
4435         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4436         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4437 #ifdef CONFIG_QUOTA
4438         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4439         for (i = 0; i < MAXQUOTAS; i++) {
4440                 if (sbi->s_qf_names[i] &&
4441                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4442                         kfree(sbi->s_qf_names[i]);
4443                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4444         }
4445 #endif
4446         unlock_super(sb);
4447         kfree(orig_data);
4448         return err;
4449 }
4450
4451 /*
4452  * Note: calculating the overhead so we can be compatible with
4453  * historical BSD practice is quite difficult in the face of
4454  * clusters/bigalloc.  This is because multiple metadata blocks from
4455  * different block group can end up in the same allocation cluster.
4456  * Calculating the exact overhead in the face of clustered allocation
4457  * requires either O(all block bitmaps) in memory or O(number of block
4458  * groups**2) in time.  We will still calculate the superblock for
4459  * older file systems --- and if we come across with a bigalloc file
4460  * system with zero in s_overhead_clusters the estimate will be close to
4461  * correct especially for very large cluster sizes --- but for newer
4462  * file systems, it's better to calculate this figure once at mkfs
4463  * time, and store it in the superblock.  If the superblock value is
4464  * present (even for non-bigalloc file systems), we will use it.
4465  */
4466 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4467 {
4468         struct super_block *sb = dentry->d_sb;
4469         struct ext4_sb_info *sbi = EXT4_SB(sb);
4470         struct ext4_super_block *es = sbi->s_es;
4471         struct ext4_group_desc *gdp;
4472         u64 fsid;
4473         s64 bfree;
4474
4475         if (test_opt(sb, MINIX_DF)) {
4476                 sbi->s_overhead_last = 0;
4477         } else if (es->s_overhead_clusters) {
4478                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4479         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4480                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4481                 ext4_fsblk_t overhead = 0;
4482
4483                 /*
4484                  * Compute the overhead (FS structures).  This is constant
4485                  * for a given filesystem unless the number of block groups
4486                  * changes so we cache the previous value until it does.
4487                  */
4488
4489                 /*
4490                  * All of the blocks before first_data_block are
4491                  * overhead
4492                  */
4493                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4494
4495                 /*
4496                  * Add the overhead found in each block group
4497                  */
4498                 for (i = 0; i < ngroups; i++) {
4499                         gdp = ext4_get_group_desc(sb, i, NULL);
4500                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4501                         cond_resched();
4502                 }
4503                 sbi->s_overhead_last = overhead;
4504                 smp_wmb();
4505                 sbi->s_blocks_last = ext4_blocks_count(es);
4506         }
4507
4508         buf->f_type = EXT4_SUPER_MAGIC;
4509         buf->f_bsize = sb->s_blocksize;
4510         buf->f_blocks = (ext4_blocks_count(es) -
4511                          EXT4_C2B(sbi, sbi->s_overhead_last));
4512         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4513                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4514         /* prevent underflow in case that few free space is available */
4515         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4516         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4517         if (buf->f_bfree < ext4_r_blocks_count(es))
4518                 buf->f_bavail = 0;
4519         buf->f_files = le32_to_cpu(es->s_inodes_count);
4520         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4521         buf->f_namelen = EXT4_NAME_LEN;
4522         fsid = le64_to_cpup((void *)es->s_uuid) ^
4523                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4524         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4525         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4526
4527         return 0;
4528 }
4529
4530 /* Helper function for writing quotas on sync - we need to start transaction
4531  * before quota file is locked for write. Otherwise the are possible deadlocks:
4532  * Process 1                         Process 2
4533  * ext4_create()                     quota_sync()
4534  *   jbd2_journal_start()                  write_dquot()
4535  *   dquot_initialize()                         down(dqio_mutex)
4536  *     down(dqio_mutex)                    jbd2_journal_start()
4537  *
4538  */
4539
4540 #ifdef CONFIG_QUOTA
4541
4542 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4543 {
4544         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4545 }
4546
4547 static int ext4_write_dquot(struct dquot *dquot)
4548 {
4549         int ret, err;
4550         handle_t *handle;
4551         struct inode *inode;
4552
4553         inode = dquot_to_inode(dquot);
4554         handle = ext4_journal_start(inode,
4555                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4556         if (IS_ERR(handle))
4557                 return PTR_ERR(handle);
4558         ret = dquot_commit(dquot);
4559         err = ext4_journal_stop(handle);
4560         if (!ret)
4561                 ret = err;
4562         return ret;
4563 }
4564
4565 static int ext4_acquire_dquot(struct dquot *dquot)
4566 {
4567         int ret, err;
4568         handle_t *handle;
4569
4570         handle = ext4_journal_start(dquot_to_inode(dquot),
4571                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4572         if (IS_ERR(handle))
4573                 return PTR_ERR(handle);
4574         ret = dquot_acquire(dquot);
4575         err = ext4_journal_stop(handle);
4576         if (!ret)
4577                 ret = err;
4578         return ret;
4579 }
4580
4581 static int ext4_release_dquot(struct dquot *dquot)
4582 {
4583         int ret, err;
4584         handle_t *handle;
4585
4586         handle = ext4_journal_start(dquot_to_inode(dquot),
4587                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4588         if (IS_ERR(handle)) {
4589                 /* Release dquot anyway to avoid endless cycle in dqput() */
4590                 dquot_release(dquot);
4591                 return PTR_ERR(handle);
4592         }
4593         ret = dquot_release(dquot);
4594         err = ext4_journal_stop(handle);
4595         if (!ret)
4596                 ret = err;
4597         return ret;
4598 }
4599
4600 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4601 {
4602         /* Are we journaling quotas? */
4603         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4604             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4605                 dquot_mark_dquot_dirty(dquot);
4606                 return ext4_write_dquot(dquot);
4607         } else {
4608                 return dquot_mark_dquot_dirty(dquot);
4609         }
4610 }
4611
4612 static int ext4_write_info(struct super_block *sb, int type)
4613 {
4614         int ret, err;
4615         handle_t *handle;
4616
4617         /* Data block + inode block */
4618         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4619         if (IS_ERR(handle))
4620                 return PTR_ERR(handle);
4621         ret = dquot_commit_info(sb, type);
4622         err = ext4_journal_stop(handle);
4623         if (!ret)
4624                 ret = err;
4625         return ret;
4626 }
4627
4628 /*
4629  * Turn on quotas during mount time - we need to find
4630  * the quota file and such...
4631  */
4632 static int ext4_quota_on_mount(struct super_block *sb, int type)
4633 {
4634         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4635                                         EXT4_SB(sb)->s_jquota_fmt, type);
4636 }
4637
4638 /*
4639  * Standard function to be called on quota_on
4640  */
4641 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4642                          struct path *path)
4643 {
4644         int err;
4645
4646         if (!test_opt(sb, QUOTA))
4647                 return -EINVAL;
4648
4649         /* Quotafile not on the same filesystem? */
4650         if (path->dentry->d_sb != sb)
4651                 return -EXDEV;
4652         /* Journaling quota? */
4653         if (EXT4_SB(sb)->s_qf_names[type]) {
4654                 /* Quotafile not in fs root? */
4655                 if (path->dentry->d_parent != sb->s_root)
4656                         ext4_msg(sb, KERN_WARNING,
4657                                 "Quota file not on filesystem root. "
4658                                 "Journaled quota will not work");
4659         }
4660
4661         /*
4662          * When we journal data on quota file, we have to flush journal to see
4663          * all updates to the file when we bypass pagecache...
4664          */
4665         if (EXT4_SB(sb)->s_journal &&
4666             ext4_should_journal_data(path->dentry->d_inode)) {
4667                 /*
4668                  * We don't need to lock updates but journal_flush() could
4669                  * otherwise be livelocked...
4670                  */
4671                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4672                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4673                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4674                 if (err)
4675                         return err;
4676         }
4677
4678         return dquot_quota_on(sb, type, format_id, path);
4679 }
4680
4681 static int ext4_quota_off(struct super_block *sb, int type)
4682 {
4683         struct inode *inode = sb_dqopt(sb)->files[type];
4684         handle_t *handle;
4685
4686         /* Force all delayed allocation blocks to be allocated.
4687          * Caller already holds s_umount sem */
4688         if (test_opt(sb, DELALLOC))
4689                 sync_filesystem(sb);
4690
4691         if (!inode)
4692                 goto out;
4693
4694         /* Update modification times of quota files when userspace can
4695          * start looking at them */
4696         handle = ext4_journal_start(inode, 1);
4697         if (IS_ERR(handle))
4698                 goto out;
4699         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4700         ext4_mark_inode_dirty(handle, inode);
4701         ext4_journal_stop(handle);
4702
4703 out:
4704         return dquot_quota_off(sb, type);
4705 }
4706
4707 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4708  * acquiring the locks... As quota files are never truncated and quota code
4709  * itself serializes the operations (and no one else should touch the files)
4710  * we don't have to be afraid of races */
4711 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4712                                size_t len, loff_t off)
4713 {
4714         struct inode *inode = sb_dqopt(sb)->files[type];
4715         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4716         int err = 0;
4717         int offset = off & (sb->s_blocksize - 1);
4718         int tocopy;
4719         size_t toread;
4720         struct buffer_head *bh;
4721         loff_t i_size = i_size_read(inode);
4722
4723         if (off > i_size)
4724                 return 0;
4725         if (off+len > i_size)
4726                 len = i_size-off;
4727         toread = len;
4728         while (toread > 0) {
4729                 tocopy = sb->s_blocksize - offset < toread ?
4730                                 sb->s_blocksize - offset : toread;
4731                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4732                 if (err)
4733                         return err;
4734                 if (!bh)        /* A hole? */
4735                         memset(data, 0, tocopy);
4736                 else
4737                         memcpy(data, bh->b_data+offset, tocopy);
4738                 brelse(bh);
4739                 offset = 0;
4740                 toread -= tocopy;
4741                 data += tocopy;
4742                 blk++;
4743         }
4744         return len;
4745 }
4746
4747 /* Write to quotafile (we know the transaction is already started and has
4748  * enough credits) */
4749 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4750                                 const char *data, size_t len, loff_t off)
4751 {
4752         struct inode *inode = sb_dqopt(sb)->files[type];
4753         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4754         int err = 0;
4755         int offset = off & (sb->s_blocksize - 1);
4756         struct buffer_head *bh;
4757         handle_t *handle = journal_current_handle();
4758
4759         if (EXT4_SB(sb)->s_journal && !handle) {
4760                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4761                         " cancelled because transaction is not started",
4762                         (unsigned long long)off, (unsigned long long)len);
4763                 return -EIO;
4764         }
4765         /*
4766          * Since we account only one data block in transaction credits,
4767          * then it is impossible to cross a block boundary.
4768          */
4769         if (sb->s_blocksize - offset < len) {
4770                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4771                         " cancelled because not block aligned",
4772                         (unsigned long long)off, (unsigned long long)len);
4773                 return -EIO;
4774         }
4775
4776         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4777         bh = ext4_bread(handle, inode, blk, 1, &err);
4778         if (!bh)
4779                 goto out;
4780         err = ext4_journal_get_write_access(handle, bh);
4781         if (err) {
4782                 brelse(bh);
4783                 goto out;
4784         }
4785         lock_buffer(bh);
4786         memcpy(bh->b_data+offset, data, len);
4787         flush_dcache_page(bh->b_page);
4788         unlock_buffer(bh);
4789         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4790         brelse(bh);
4791 out:
4792         if (err) {
4793                 mutex_unlock(&inode->i_mutex);
4794                 return err;
4795         }
4796         if (inode->i_size < off + len) {
4797                 i_size_write(inode, off + len);
4798                 EXT4_I(inode)->i_disksize = inode->i_size;
4799                 ext4_mark_inode_dirty(handle, inode);
4800         }
4801         mutex_unlock(&inode->i_mutex);
4802         return len;
4803 }
4804
4805 #endif
4806
4807 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4808                        const char *dev_name, void *data)
4809 {
4810         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4811 }
4812
4813 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4814 static inline void register_as_ext2(void)
4815 {
4816         int err = register_filesystem(&ext2_fs_type);
4817         if (err)
4818                 printk(KERN_WARNING
4819                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4820 }
4821
4822 static inline void unregister_as_ext2(void)
4823 {
4824         unregister_filesystem(&ext2_fs_type);
4825 }
4826
4827 static inline int ext2_feature_set_ok(struct super_block *sb)
4828 {
4829         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4830                 return 0;
4831         if (sb->s_flags & MS_RDONLY)
4832                 return 1;
4833         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4834                 return 0;
4835         return 1;
4836 }
4837 MODULE_ALIAS("ext2");
4838 #else
4839 static inline void register_as_ext2(void) { }
4840 static inline void unregister_as_ext2(void) { }
4841 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4842 #endif
4843
4844 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4845 static inline void register_as_ext3(void)
4846 {
4847         int err = register_filesystem(&ext3_fs_type);
4848         if (err)
4849                 printk(KERN_WARNING
4850                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4851 }
4852
4853 static inline void unregister_as_ext3(void)
4854 {
4855         unregister_filesystem(&ext3_fs_type);
4856 }
4857
4858 static inline int ext3_feature_set_ok(struct super_block *sb)
4859 {
4860         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4861                 return 0;
4862         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4863                 return 0;
4864         if (sb->s_flags & MS_RDONLY)
4865                 return 1;
4866         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4867                 return 0;
4868         return 1;
4869 }
4870 MODULE_ALIAS("ext3");
4871 #else
4872 static inline void register_as_ext3(void) { }
4873 static inline void unregister_as_ext3(void) { }
4874 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4875 #endif
4876
4877 static struct file_system_type ext4_fs_type = {
4878         .owner          = THIS_MODULE,
4879         .name           = "ext4",
4880         .mount          = ext4_mount,
4881         .kill_sb        = kill_block_super,
4882         .fs_flags       = FS_REQUIRES_DEV,
4883 };
4884
4885 static int __init ext4_init_feat_adverts(void)
4886 {
4887         struct ext4_features *ef;
4888         int ret = -ENOMEM;
4889
4890         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4891         if (!ef)
4892                 goto out;
4893
4894         ef->f_kobj.kset = ext4_kset;
4895         init_completion(&ef->f_kobj_unregister);
4896         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4897                                    "features");
4898         if (ret) {
4899                 kfree(ef);
4900                 goto out;
4901         }
4902
4903         ext4_feat = ef;
4904         ret = 0;
4905 out:
4906         return ret;
4907 }
4908
4909 static void ext4_exit_feat_adverts(void)
4910 {
4911         kobject_put(&ext4_feat->f_kobj);
4912         wait_for_completion(&ext4_feat->f_kobj_unregister);
4913         kfree(ext4_feat);
4914 }
4915
4916 /* Shared across all ext4 file systems */
4917 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4918 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4919
4920 static int __init ext4_init_fs(void)
4921 {
4922         int i, err;
4923
4924         ext4_li_info = NULL;
4925         mutex_init(&ext4_li_mtx);
4926
4927         ext4_check_flag_values();
4928
4929         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4930                 mutex_init(&ext4__aio_mutex[i]);
4931                 init_waitqueue_head(&ext4__ioend_wq[i]);
4932         }
4933
4934         err = ext4_init_pageio();
4935         if (err)
4936                 return err;
4937         err = ext4_init_system_zone();
4938         if (err)
4939                 goto out6;
4940         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4941         if (!ext4_kset)
4942                 goto out5;
4943         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4944
4945         err = ext4_init_feat_adverts();
4946         if (err)
4947                 goto out4;
4948
4949         err = ext4_init_mballoc();
4950         if (err)
4951                 goto out3;
4952
4953         err = ext4_init_xattr();
4954         if (err)
4955                 goto out2;
4956         err = init_inodecache();
4957         if (err)
4958                 goto out1;
4959         register_as_ext3();
4960         register_as_ext2();
4961         err = register_filesystem(&ext4_fs_type);
4962         if (err)
4963                 goto out;
4964
4965         return 0;
4966 out:
4967         unregister_as_ext2();
4968         unregister_as_ext3();
4969         destroy_inodecache();
4970 out1:
4971         ext4_exit_xattr();
4972 out2:
4973         ext4_exit_mballoc();
4974 out3:
4975         ext4_exit_feat_adverts();
4976 out4:
4977         if (ext4_proc_root)
4978                 remove_proc_entry("fs/ext4", NULL);
4979         kset_unregister(ext4_kset);
4980 out5:
4981         ext4_exit_system_zone();
4982 out6:
4983         ext4_exit_pageio();
4984         return err;
4985 }
4986
4987 static void __exit ext4_exit_fs(void)
4988 {
4989         ext4_destroy_lazyinit_thread();
4990         unregister_as_ext2();
4991         unregister_as_ext3();
4992         unregister_filesystem(&ext4_fs_type);
4993         destroy_inodecache();
4994         ext4_exit_xattr();
4995         ext4_exit_mballoc();
4996         ext4_exit_feat_adverts();
4997         remove_proc_entry("fs/ext4", NULL);
4998         kset_unregister(ext4_kset);
4999         ext4_exit_system_zone();
5000         ext4_exit_pageio();
5001 }
5002
5003 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5004 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5005 MODULE_LICENSE("GPL");
5006 module_init(ext4_init_fs)
5007 module_exit(ext4_exit_fs)