fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb,
 147                               struct ext4_super_block *es)
 148 {
 149         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 150                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 151                 return;
 152
 153         es->s_checksum = ext4_superblock_csum(sb, es);
 154 }
 155
 156 void *ext4_kvmalloc(size_t size, gfp_t flags)
 157 {
 158         void *ret;
 159
 160         ret = kmalloc(size, flags);
 161         if (!ret)
 162                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 163         return ret;
 164 }
 165
 166 void *ext4_kvzalloc(size_t size, gfp_t flags)
 167 {
 168         void *ret;
 169
 170         ret = kzalloc(size, flags);
 171         if (!ret)
 172                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 173         return ret;
 174 }
 175
 176 void ext4_kvfree(void *ptr)
 177 {
 178         if (is_vmalloc_addr(ptr))
 179                 vfree(ptr);
 180         else
 181                 kfree(ptr);
 182
 183 }
 184
 185 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 186                                struct ext4_group_desc *bg)
 187 {
 188         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 189                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 190                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 191 }
 192
 193 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 194                                struct ext4_group_desc *bg)
 195 {
 196         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 197                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 198                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 199 }
 200
 201 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 202                               struct ext4_group_desc *bg)
 203 {
 204         return le32_to_cpu(bg->bg_inode_table_lo) |
 205                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 206                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 207 }
 208
 209 __u32 ext4_free_group_clusters(struct super_block *sb,
 210                                struct ext4_group_desc *bg)
 211 {
 212         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 213                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 214                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 215 }
 216
 217 __u32 ext4_free_inodes_count(struct super_block *sb,
 218                               struct ext4_group_desc *bg)
 219 {
 220         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 221                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 222                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 223 }
 224
 225 __u32 ext4_used_dirs_count(struct super_block *sb,
 226                               struct ext4_group_desc *bg)
 227 {
 228         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 229                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 230                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 231 }
 232
 233 __u32 ext4_itable_unused_count(struct super_block *sb,
 234                               struct ext4_group_desc *bg)
 235 {
 236         return le16_to_cpu(bg->bg_itable_unused_lo) |
 237                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 238                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 239 }
 240
 241 void ext4_block_bitmap_set(struct super_block *sb,
 242                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 243 {
 244         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 245         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 246                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 247 }
 248
 249 void ext4_inode_bitmap_set(struct super_block *sb,
 250                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 251 {
 252         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 253         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 254                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 255 }
 256
 257 void ext4_inode_table_set(struct super_block *sb,
 258                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 259 {
 260         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 261         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 262                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 263 }
 264
 265 void ext4_free_group_clusters_set(struct super_block *sb,
 266                                   struct ext4_group_desc *bg, __u32 count)
 267 {
 268         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 269         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 270                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 271 }
 272
 273 void ext4_free_inodes_set(struct super_block *sb,
 274                           struct ext4_group_desc *bg, __u32 count)
 275 {
 276         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 277         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 278                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 279 }
 280
 281 void ext4_used_dirs_set(struct super_block *sb,
 282                           struct ext4_group_desc *bg, __u32 count)
 283 {
 284         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 285         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 286                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 287 }
 288
 289 void ext4_itable_unused_set(struct super_block *sb,
 290                           struct ext4_group_desc *bg, __u32 count)
 291 {
 292         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 293         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 294                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 295 }
 296
 297
 298 /* Just increment the non-pointer handle value */
 299 static handle_t *ext4_get_nojournal(void)
 300 {
 301         handle_t *handle = current->journal_info;
 302         unsigned long ref_cnt = (unsigned long)handle;
 303
 304         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 305
 306         ref_cnt++;
 307         handle = (handle_t *)ref_cnt;
 308
 309         current->journal_info = handle;
 310         return handle;
 311 }
 312
 313
 314 /* Decrement the non-pointer handle value */
 315 static void ext4_put_nojournal(handle_t *handle)
 316 {
 317         unsigned long ref_cnt = (unsigned long)handle;
 318
 319         BUG_ON(ref_cnt == 0);
 320
 321         ref_cnt--;
 322         handle = (handle_t *)ref_cnt;
 323
 324         current->journal_info = handle;
 325 }
 326
 327 /*
 328  * Wrappers for jbd2_journal_start/end.
 329  *
 330  * The only special thing we need to do here is to make sure that all
 331  * journal_end calls result in the superblock being marked dirty, so
 332  * that sync() will call the filesystem's write_super callback if
 333  * appropriate.
 334  */
 335 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 336 {
 337         journal_t *journal;
 338
 339         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 340         if (sb->s_flags & MS_RDONLY)
 341                 return ERR_PTR(-EROFS);
 342
 343         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 344         journal = EXT4_SB(sb)->s_journal;
 345         if (!journal)
 346                 return ext4_get_nojournal();
 347         /*
 348          * Special case here: if the journal has aborted behind our
 349          * backs (eg. EIO in the commit thread), then we still need to
 350          * take the FS itself readonly cleanly.
 351          */
 352         if (is_journal_aborted(journal)) {
 353                 ext4_abort(sb, "Detected aborted journal");
 354                 return ERR_PTR(-EROFS);
 355         }
 356         return jbd2_journal_start(journal, nblocks);
 357 }
 358
 359 /*
 360  * The only special thing we need to do here is to make sure that all
 361  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 362  * that sync() will call the filesystem's write_super callback if
 363  * appropriate.
 364  */
 365 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 366 {
 367         struct super_block *sb;
 368         int err;
 369         int rc;
 370
 371         if (!ext4_handle_valid(handle)) {
 372                 ext4_put_nojournal(handle);
 373                 return 0;
 374         }
 375         sb = handle->h_transaction->t_journal->j_private;
 376         err = handle->h_err;
 377         rc = jbd2_journal_stop(handle);
 378
 379         if (!err)
 380                 err = rc;
 381         if (err)
 382                 __ext4_std_error(sb, where, line, err);
 383         return err;
 384 }
 385
 386 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 387                                const char *err_fn, struct buffer_head *bh,
 388                                handle_t *handle, int err)
 389 {
 390         char nbuf[16];
 391         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 392
 393         BUG_ON(!ext4_handle_valid(handle));
 394
 395         if (bh)
 396                 BUFFER_TRACE(bh, "abort");
 397
 398         if (!handle->h_err)
 399                 handle->h_err = err;
 400
 401         if (is_handle_aborted(handle))
 402                 return;
 403
 404         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 405                caller, line, errstr, err_fn);
 406
 407         jbd2_journal_abort_handle(handle);
 408 }
 409
 410 static void __save_error_info(struct super_block *sb, const char *func,
 411                             unsigned int line)
 412 {
 413         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 414
 415         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 416         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 417         es->s_last_error_time = cpu_to_le32(get_seconds());
 418         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 419         es->s_last_error_line = cpu_to_le32(line);
 420         if (!es->s_first_error_time) {
 421                 es->s_first_error_time = es->s_last_error_time;
 422                 strncpy(es->s_first_error_func, func,
 423                         sizeof(es->s_first_error_func));
 424                 es->s_first_error_line = cpu_to_le32(line);
 425                 es->s_first_error_ino = es->s_last_error_ino;
 426                 es->s_first_error_block = es->s_last_error_block;
 427         }
 428         /*
 429          * Start the daily error reporting function if it hasn't been
 430          * started already
 431          */
 432         if (!es->s_error_count)
 433                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 434         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 435 }
 436
 437 static void save_error_info(struct super_block *sb, const char *func,
 438                             unsigned int line)
 439 {
 440         __save_error_info(sb, func, line);
 441         ext4_commit_super(sb, 1);
 442 }
 443
 444 /*
 445  * The del_gendisk() function uninitializes the disk-specific data
 446  * structures, including the bdi structure, without telling anyone
 447  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 448  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 449  * This is a kludge to prevent these oops until we can put in a proper
 450  * hook in del_gendisk() to inform the VFS and file system layers.
 451  */
 452 static int block_device_ejected(struct super_block *sb)
 453 {
 454         struct inode *bd_inode = sb->s_bdev->bd_inode;
 455         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 456
 457         return bdi->dev == NULL;
 458 }
 459
 460 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 461 {
 462         struct super_block              *sb = journal->j_private;
 463         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 464         int                             error = is_journal_aborted(journal);
 465         struct ext4_journal_cb_entry    *jce, *tmp;
 466
 467         spin_lock(&sbi->s_md_lock);
 468         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 469                 list_del_init(&jce->jce_list);
 470                 spin_unlock(&sbi->s_md_lock);
 471                 jce->jce_func(sb, jce, error);
 472                 spin_lock(&sbi->s_md_lock);
 473         }
 474         spin_unlock(&sbi->s_md_lock);
 475 }
 476
 477 /* Deal with the reporting of failure conditions on a filesystem such as
 478  * inconsistencies detected or read IO failures.
 479  *
 480  * On ext2, we can store the error state of the filesystem in the
 481  * superblock.  That is not possible on ext4, because we may have other
 482  * write ordering constraints on the superblock which prevent us from
 483  * writing it out straight away; and given that the journal is about to
 484  * be aborted, we can't rely on the current, or future, transactions to
 485  * write out the superblock safely.
 486  *
 487  * We'll just use the jbd2_journal_abort() error code to record an error in
 488  * the journal instead.  On recovery, the journal will complain about
 489  * that error until we've noted it down and cleared it.
 490  */
 491
 492 static void ext4_handle_error(struct super_block *sb)
 493 {
 494         if (sb->s_flags & MS_RDONLY)
 495                 return;
 496
 497         if (!test_opt(sb, ERRORS_CONT)) {
 498                 journal_t *journal = EXT4_SB(sb)->s_journal;
 499
 500                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 501                 if (journal)
 502                         jbd2_journal_abort(journal, -EIO);
 503         }
 504         if (test_opt(sb, ERRORS_RO)) {
 505                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 506                 sb->s_flags |= MS_RDONLY;
 507         }
 508         if (test_opt(sb, ERRORS_PANIC))
 509                 panic("EXT4-fs (device %s): panic forced after error\n",
 510                         sb->s_id);
 511 }
 512
 513 void __ext4_error(struct super_block *sb, const char *function,
 514                   unsigned int line, const char *fmt, ...)
 515 {
 516         struct va_format vaf;
 517         va_list args;
 518
 519         va_start(args, fmt);
 520         vaf.fmt = fmt;
 521         vaf.va = &args;
 522         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 523                sb->s_id, function, line, current->comm, &vaf);
 524         va_end(args);
 525         save_error_info(sb, function, line);
 526
 527         ext4_handle_error(sb);
 528 }
 529
 530 void ext4_error_inode(struct inode *inode, const char *function,
 531                       unsigned int line, ext4_fsblk_t block,
 532                       const char *fmt, ...)
 533 {
 534         va_list args;
 535         struct va_format vaf;
 536         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 537
 538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 539         es->s_last_error_block = cpu_to_le64(block);
 540         save_error_info(inode->i_sb, function, line);
 541         va_start(args, fmt);
 542         vaf.fmt = fmt;
 543         vaf.va = &args;
 544         if (block)
 545                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 546                        "inode #%lu: block %llu: comm %s: %pV\n",
 547                        inode->i_sb->s_id, function, line, inode->i_ino,
 548                        block, current->comm, &vaf);
 549         else
 550                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 551                        "inode #%lu: comm %s: %pV\n",
 552                        inode->i_sb->s_id, function, line, inode->i_ino,
 553                        current->comm, &vaf);
 554         va_end(args);
 555
 556         ext4_handle_error(inode->i_sb);
 557 }
 558
 559 void ext4_error_file(struct file *file, const char *function,
 560                      unsigned int line, ext4_fsblk_t block,
 561                      const char *fmt, ...)
 562 {
 563         va_list args;
 564         struct va_format vaf;
 565         struct ext4_super_block *es;
 566         struct inode *inode = file->f_dentry->d_inode;
 567         char pathname[80], *path;
 568
 569         es = EXT4_SB(inode->i_sb)->s_es;
 570         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 571         save_error_info(inode->i_sb, function, line);
 572         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 573         if (IS_ERR(path))
 574                 path = "(unknown)";
 575         va_start(args, fmt);
 576         vaf.fmt = fmt;
 577         vaf.va = &args;
 578         if (block)
 579                 printk(KERN_CRIT
 580                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 581                        "block %llu: comm %s: path %s: %pV\n",
 582                        inode->i_sb->s_id, function, line, inode->i_ino,
 583                        block, current->comm, path, &vaf);
 584         else
 585                 printk(KERN_CRIT
 586                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 587                        "comm %s: path %s: %pV\n",
 588                        inode->i_sb->s_id, function, line, inode->i_ino,
 589                        current->comm, path, &vaf);
 590         va_end(args);
 591
 592         ext4_handle_error(inode->i_sb);
 593 }
 594
 595 static const char *ext4_decode_error(struct super_block *sb, int errno,
 596                                      char nbuf[16])
 597 {
 598         char *errstr = NULL;
 599
 600         switch (errno) {
 601         case -EIO:
 602                 errstr = "IO failure";
 603                 break;
 604         case -ENOMEM:
 605                 errstr = "Out of memory";
 606                 break;
 607         case -EROFS:
 608                 if (!sb || (EXT4_SB(sb)->s_journal &&
 609                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 610                         errstr = "Journal has aborted";
 611                 else
 612                         errstr = "Readonly filesystem";
 613                 break;
 614         default:
 615                 /* If the caller passed in an extra buffer for unknown
 616                  * errors, textualise them now.  Else we just return
 617                  * NULL. */
 618                 if (nbuf) {
 619                         /* Check for truncated error codes... */
 620                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 621                                 errstr = nbuf;
 622                 }
 623                 break;
 624         }
 625
 626         return errstr;
 627 }
 628
 629 /* __ext4_std_error decodes expected errors from journaling functions
 630  * automatically and invokes the appropriate error response.  */
 631
 632 void __ext4_std_error(struct super_block *sb, const char *function,
 633                       unsigned int line, int errno)
 634 {
 635         char nbuf[16];
 636         const char *errstr;
 637
 638         /* Special case: if the error is EROFS, and we're not already
 639          * inside a transaction, then there's really no point in logging
 640          * an error. */
 641         if (errno == -EROFS && journal_current_handle() == NULL &&
 642             (sb->s_flags & MS_RDONLY))
 643                 return;
 644
 645         errstr = ext4_decode_error(sb, errno, nbuf);
 646         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 647                sb->s_id, function, line, errstr);
 648         save_error_info(sb, function, line);
 649
 650         ext4_handle_error(sb);
 651 }
 652
 653 /*
 654  * ext4_abort is a much stronger failure handler than ext4_error.  The
 655  * abort function may be used to deal with unrecoverable failures such
 656  * as journal IO errors or ENOMEM at a critical moment in log management.
 657  *
 658  * We unconditionally force the filesystem into an ABORT|READONLY state,
 659  * unless the error response on the fs has been set to panic in which
 660  * case we take the easy way out and panic immediately.
 661  */
 662
 663 void __ext4_abort(struct super_block *sb, const char *function,
 664                 unsigned int line, const char *fmt, ...)
 665 {
 666         va_list args;
 667
 668         save_error_info(sb, function, line);
 669         va_start(args, fmt);
 670         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 671                function, line);
 672         vprintk(fmt, args);
 673         printk("\n");
 674         va_end(args);
 675
 676         if ((sb->s_flags & MS_RDONLY) == 0) {
 677                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 678                 sb->s_flags |= MS_RDONLY;
 679                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 680                 if (EXT4_SB(sb)->s_journal)
 681                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 682                 save_error_info(sb, function, line);
 683         }
 684         if (test_opt(sb, ERRORS_PANIC))
 685                 panic("EXT4-fs panic from previous error\n");
 686 }
 687
 688 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 689 {
 690         struct va_format vaf;
 691         va_list args;
 692
 693         va_start(args, fmt);
 694         vaf.fmt = fmt;
 695         vaf.va = &args;
 696         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 697         va_end(args);
 698 }
 699
 700 void __ext4_warning(struct super_block *sb, const char *function,
 701                     unsigned int line, const char *fmt, ...)
 702 {
 703         struct va_format vaf;
 704         va_list args;
 705
 706         va_start(args, fmt);
 707         vaf.fmt = fmt;
 708         vaf.va = &args;
 709         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 710                sb->s_id, function, line, &vaf);
 711         va_end(args);
 712 }
 713
 714 void __ext4_grp_locked_error(const char *function, unsigned int line,
 715                              struct super_block *sb, ext4_group_t grp,
 716                              unsigned long ino, ext4_fsblk_t block,
 717                              const char *fmt, ...)
 718 __releases(bitlock)
 719 __acquires(bitlock)
 720 {
 721         struct va_format vaf;
 722         va_list args;
 723         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 724
 725         es->s_last_error_ino = cpu_to_le32(ino);
 726         es->s_last_error_block = cpu_to_le64(block);
 727         __save_error_info(sb, function, line);
 728
 729         va_start(args, fmt);
 730
 731         vaf.fmt = fmt;
 732         vaf.va = &args;
 733         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 734                sb->s_id, function, line, grp);
 735         if (ino)
 736                 printk(KERN_CONT "inode %lu: ", ino);
 737         if (block)
 738                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 739         printk(KERN_CONT "%pV\n", &vaf);
 740         va_end(args);
 741
 742         if (test_opt(sb, ERRORS_CONT)) {
 743                 ext4_commit_super(sb, 0);
 744                 return;
 745         }
 746
 747         ext4_unlock_group(sb, grp);
 748         ext4_handle_error(sb);
 749         /*
 750          * We only get here in the ERRORS_RO case; relocking the group
 751          * may be dangerous, but nothing bad will happen since the
 752          * filesystem will have already been marked read/only and the
 753          * journal has been aborted.  We return 1 as a hint to callers
 754          * who might what to use the return value from
 755          * ext4_grp_locked_error() to distinguish between the
 756          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 757          * aggressively from the ext4 function in question, with a
 758          * more appropriate error code.
 759          */
 760         ext4_lock_group(sb, grp);
 761         return;
 762 }
 763
 764 void ext4_update_dynamic_rev(struct super_block *sb)
 765 {
 766         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 767
 768         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 769                 return;
 770
 771         ext4_warning(sb,
 772                      "updating to rev %d because of new feature flag, "
 773                      "running e2fsck is recommended",
 774                      EXT4_DYNAMIC_REV);
 775
 776         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 777         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 778         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 779         /* leave es->s_feature_*compat flags alone */
 780         /* es->s_uuid will be set by e2fsck if empty */
 781
 782         /*
 783          * The rest of the superblock fields should be zero, and if not it
 784          * means they are likely already in use, so leave them alone.  We
 785          * can leave it up to e2fsck to clean up any inconsistencies there.
 786          */
 787 }
 788
 789 /*
 790  * Open the external journal device
 791  */
 792 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 793 {
 794         struct block_device *bdev;
 795         char b[BDEVNAME_SIZE];
 796
 797         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 798         if (IS_ERR(bdev))
 799                 goto fail;
 800         return bdev;
 801
 802 fail:
 803         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 804                         __bdevname(dev, b), PTR_ERR(bdev));
 805         return NULL;
 806 }
 807
 808 /*
 809  * Release the journal device
 810  */
 811 static int ext4_blkdev_put(struct block_device *bdev)
 812 {
 813         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 814 }
 815
 816 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 817 {
 818         struct block_device *bdev;
 819         int ret = -ENODEV;
 820
 821         bdev = sbi->journal_bdev;
 822         if (bdev) {
 823                 ret = ext4_blkdev_put(bdev);
 824                 sbi->journal_bdev = NULL;
 825         }
 826         return ret;
 827 }
 828
 829 static inline struct inode *orphan_list_entry(struct list_head *l)
 830 {
 831         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 832 }
 833
 834 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 835 {
 836         struct list_head *l;
 837
 838         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 839                  le32_to_cpu(sbi->s_es->s_last_orphan));
 840
 841         printk(KERN_ERR "sb_info orphan list:\n");
 842         list_for_each(l, &sbi->s_orphan) {
 843                 struct inode *inode = orphan_list_entry(l);
 844                 printk(KERN_ERR "  "
 845                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 846                        inode->i_sb->s_id, inode->i_ino, inode,
 847                        inode->i_mode, inode->i_nlink,
 848                        NEXT_ORPHAN(inode));
 849         }
 850 }
 851
 852 static void ext4_put_super(struct super_block *sb)
 853 {
 854         struct ext4_sb_info *sbi = EXT4_SB(sb);
 855         struct ext4_super_block *es = sbi->s_es;
 856         int i, err;
 857
 858         ext4_unregister_li_request(sb);
 859         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 860
 861         flush_workqueue(sbi->dio_unwritten_wq);
 862         destroy_workqueue(sbi->dio_unwritten_wq);
 863
 864         lock_super(sb);
 865         if (sbi->s_journal) {
 866                 err = jbd2_journal_destroy(sbi->s_journal);
 867                 sbi->s_journal = NULL;
 868                 if (err < 0)
 869                         ext4_abort(sb, "Couldn't clean up the journal");
 870         }
 871
 872         del_timer(&sbi->s_err_report);
 873         ext4_release_system_zone(sb);
 874         ext4_mb_release(sb);
 875         ext4_ext_release(sb);
 876         ext4_xattr_put_super(sb);
 877
 878         if (!(sb->s_flags & MS_RDONLY)) {
 879                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 880                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 881         }
 882         if (!(sb->s_flags & MS_RDONLY))
 883                 ext4_commit_super(sb, 1);
 884
 885         if (sbi->s_proc) {
 886                 remove_proc_entry("options", sbi->s_proc);
 887                 remove_proc_entry(sb->s_id, ext4_proc_root);
 888         }
 889         kobject_del(&sbi->s_kobj);
 890
 891         for (i = 0; i < sbi->s_gdb_count; i++)
 892                 brelse(sbi->s_group_desc[i]);
 893         ext4_kvfree(sbi->s_group_desc);
 894         ext4_kvfree(sbi->s_flex_groups);
 895         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 896         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 897         percpu_counter_destroy(&sbi->s_dirs_counter);
 898         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 899         brelse(sbi->s_sbh);
 900 #ifdef CONFIG_QUOTA
 901         for (i = 0; i < MAXQUOTAS; i++)
 902                 kfree(sbi->s_qf_names[i]);
 903 #endif
 904
 905         /* Debugging code just in case the in-memory inode orphan list
 906          * isn't empty.  The on-disk one can be non-empty if we've
 907          * detected an error and taken the fs readonly, but the
 908          * in-memory list had better be clean by this point. */
 909         if (!list_empty(&sbi->s_orphan))
 910                 dump_orphan_list(sb, sbi);
 911         J_ASSERT(list_empty(&sbi->s_orphan));
 912
 913         invalidate_bdev(sb->s_bdev);
 914         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 915                 /*
 916                  * Invalidate the journal device's buffers.  We don't want them
 917                  * floating about in memory - the physical journal device may
 918                  * hotswapped, and it breaks the `ro-after' testing code.
 919                  */
 920                 sync_blockdev(sbi->journal_bdev);
 921                 invalidate_bdev(sbi->journal_bdev);
 922                 ext4_blkdev_remove(sbi);
 923         }
 924         if (sbi->s_mmp_tsk)
 925                 kthread_stop(sbi->s_mmp_tsk);
 926         sb->s_fs_info = NULL;
 927         /*
 928          * Now that we are completely done shutting down the
 929          * superblock, we need to actually destroy the kobject.
 930          */
 931         unlock_super(sb);
 932         kobject_put(&sbi->s_kobj);
 933         wait_for_completion(&sbi->s_kobj_unregister);
 934         if (sbi->s_chksum_driver)
 935                 crypto_free_shash(sbi->s_chksum_driver);
 936         kfree(sbi->s_blockgroup_lock);
 937         kfree(sbi);
 938 }
 939
 940 static struct kmem_cache *ext4_inode_cachep;
 941
 942 /*
 943  * Called inside transaction, so use GFP_NOFS
 944  */
 945 static struct inode *ext4_alloc_inode(struct super_block *sb)
 946 {
 947         struct ext4_inode_info *ei;
 948
 949         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 950         if (!ei)
 951                 return NULL;
 952
 953         ei->vfs_inode.i_version = 1;
 954         ei->vfs_inode.i_data.writeback_index = 0;
 955         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 956         INIT_LIST_HEAD(&ei->i_prealloc_list);
 957         spin_lock_init(&ei->i_prealloc_lock);
 958         ei->i_reserved_data_blocks = 0;
 959         ei->i_reserved_meta_blocks = 0;
 960         ei->i_allocated_meta_blocks = 0;
 961         ei->i_da_metadata_calc_len = 0;
 962         ei->i_da_metadata_calc_last_lblock = 0;
 963         spin_lock_init(&(ei->i_block_reservation_lock));
 964 #ifdef CONFIG_QUOTA
 965         ei->i_reserved_quota = 0;
 966 #endif
 967         ei->jinode = NULL;
 968         INIT_LIST_HEAD(&ei->i_completed_io_list);
 969         spin_lock_init(&ei->i_completed_io_lock);
 970         ei->cur_aio_dio = NULL;
 971         ei->i_sync_tid = 0;
 972         ei->i_datasync_tid = 0;
 973         atomic_set(&ei->i_ioend_count, 0);
 974         atomic_set(&ei->i_aiodio_unwritten, 0);
 975
 976         return &ei->vfs_inode;
 977 }
 978
 979 static int ext4_drop_inode(struct inode *inode)
 980 {
 981         int drop = generic_drop_inode(inode);
 982
 983         trace_ext4_drop_inode(inode, drop);
 984         return drop;
 985 }
 986
 987 static void ext4_i_callback(struct rcu_head *head)
 988 {
 989         struct inode *inode = container_of(head, struct inode, i_rcu);
 990         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 991 }
 992
 993 static void ext4_destroy_inode(struct inode *inode)
 994 {
 995         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 996                 ext4_msg(inode->i_sb, KERN_ERR,
 997                          "Inode %lu (%p): orphan list check failed!",
 998                          inode->i_ino, EXT4_I(inode));
 999                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1000                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1001                                 true);
1002                 dump_stack();
1003         }
1004         call_rcu(&inode->i_rcu, ext4_i_callback);
1005 }
1006
1007 static void init_once(void *foo)
1008 {
1009         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1010
1011         INIT_LIST_HEAD(&ei->i_orphan);
1012 #ifdef CONFIG_EXT4_FS_XATTR
1013         init_rwsem(&ei->xattr_sem);
1014 #endif
1015         init_rwsem(&ei->i_data_sem);
1016         inode_init_once(&ei->vfs_inode);
1017 }
1018
1019 static int init_inodecache(void)
1020 {
1021         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1022                                              sizeof(struct ext4_inode_info),
1023                                              0, (SLAB_RECLAIM_ACCOUNT|
1024                                                 SLAB_MEM_SPREAD),
1025                                              init_once);
1026         if (ext4_inode_cachep == NULL)
1027                 return -ENOMEM;
1028         return 0;
1029 }
1030
1031 static void destroy_inodecache(void)
1032 {
1033         kmem_cache_destroy(ext4_inode_cachep);
1034 }
1035
1036 void ext4_clear_inode(struct inode *inode)
1037 {
1038         invalidate_inode_buffers(inode);
1039         clear_inode(inode);
1040         dquot_drop(inode);
1041         ext4_discard_preallocations(inode);
1042         if (EXT4_I(inode)->jinode) {
1043                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1044                                                EXT4_I(inode)->jinode);
1045                 jbd2_free_inode(EXT4_I(inode)->jinode);
1046                 EXT4_I(inode)->jinode = NULL;
1047         }
1048 }
1049
1050 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1051                                         u64 ino, u32 generation)
1052 {
1053         struct inode *inode;
1054
1055         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1056                 return ERR_PTR(-ESTALE);
1057         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1058                 return ERR_PTR(-ESTALE);
1059
1060         /* iget isn't really right if the inode is currently unallocated!!
1061          *
1062          * ext4_read_inode will return a bad_inode if the inode had been
1063          * deleted, so we should be safe.
1064          *
1065          * Currently we don't know the generation for parent directory, so
1066          * a generation of 0 means "accept any"
1067          */
1068         inode = ext4_iget(sb, ino);
1069         if (IS_ERR(inode))
1070                 return ERR_CAST(inode);
1071         if (generation && inode->i_generation != generation) {
1072                 iput(inode);
1073                 return ERR_PTR(-ESTALE);
1074         }
1075
1076         return inode;
1077 }
1078
1079 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1080                                         int fh_len, int fh_type)
1081 {
1082         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1083                                     ext4_nfs_get_inode);
1084 }
1085
1086 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1087                                         int fh_len, int fh_type)
1088 {
1089         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1090                                     ext4_nfs_get_inode);
1091 }
1092
1093 /*
1094  * Try to release metadata pages (indirect blocks, directories) which are
1095  * mapped via the block device.  Since these pages could have journal heads
1096  * which would prevent try_to_free_buffers() from freeing them, we must use
1097  * jbd2 layer's try_to_free_buffers() function to release them.
1098  */
1099 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1100                                  gfp_t wait)
1101 {
1102         journal_t *journal = EXT4_SB(sb)->s_journal;
1103
1104         WARN_ON(PageChecked(page));
1105         if (!page_has_buffers(page))
1106                 return 0;
1107         if (journal)
1108                 return jbd2_journal_try_to_free_buffers(journal, page,
1109                                                         wait & ~__GFP_WAIT);
1110         return try_to_free_buffers(page);
1111 }
1112
1113 #ifdef CONFIG_QUOTA
1114 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1115 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1116
1117 static int ext4_write_dquot(struct dquot *dquot);
1118 static int ext4_acquire_dquot(struct dquot *dquot);
1119 static int ext4_release_dquot(struct dquot *dquot);
1120 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1121 static int ext4_write_info(struct super_block *sb, int type);
1122 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1123                          struct path *path);
1124 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1125                                  int format_id);
1126 static int ext4_quota_off(struct super_block *sb, int type);
1127 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1128 static int ext4_quota_on_mount(struct super_block *sb, int type);
1129 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1130                                size_t len, loff_t off);
1131 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1132                                 const char *data, size_t len, loff_t off);
1133 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1134                              unsigned int flags);
1135 static int ext4_enable_quotas(struct super_block *sb);
1136
1137 static const struct dquot_operations ext4_quota_operations = {
1138         .get_reserved_space = ext4_get_reserved_space,
1139         .write_dquot    = ext4_write_dquot,
1140         .acquire_dquot  = ext4_acquire_dquot,
1141         .release_dquot  = ext4_release_dquot,
1142         .mark_dirty     = ext4_mark_dquot_dirty,
1143         .write_info     = ext4_write_info,
1144         .alloc_dquot    = dquot_alloc,
1145         .destroy_dquot  = dquot_destroy,
1146 };
1147
1148 static const struct quotactl_ops ext4_qctl_operations = {
1149         .quota_on       = ext4_quota_on,
1150         .quota_off      = ext4_quota_off,
1151         .quota_sync     = dquot_quota_sync,
1152         .get_info       = dquot_get_dqinfo,
1153         .set_info       = dquot_set_dqinfo,
1154         .get_dqblk      = dquot_get_dqblk,
1155         .set_dqblk      = dquot_set_dqblk
1156 };
1157
1158 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1159         .quota_on_meta  = ext4_quota_on_sysfile,
1160         .quota_off      = ext4_quota_off_sysfile,
1161         .quota_sync     = dquot_quota_sync,
1162         .get_info       = dquot_get_dqinfo,
1163         .set_info       = dquot_set_dqinfo,
1164         .get_dqblk      = dquot_get_dqblk,
1165         .set_dqblk      = dquot_set_dqblk
1166 };
1167 #endif
1168
1169 static const struct super_operations ext4_sops = {
1170         .alloc_inode    = ext4_alloc_inode,
1171         .destroy_inode  = ext4_destroy_inode,
1172         .write_inode    = ext4_write_inode,
1173         .dirty_inode    = ext4_dirty_inode,
1174         .drop_inode     = ext4_drop_inode,
1175         .evict_inode    = ext4_evict_inode,
1176         .put_super      = ext4_put_super,
1177         .sync_fs        = ext4_sync_fs,
1178         .freeze_fs      = ext4_freeze,
1179         .unfreeze_fs    = ext4_unfreeze,
1180         .statfs         = ext4_statfs,
1181         .remount_fs     = ext4_remount,
1182         .show_options   = ext4_show_options,
1183 #ifdef CONFIG_QUOTA
1184         .quota_read     = ext4_quota_read,
1185         .quota_write    = ext4_quota_write,
1186 #endif
1187         .bdev_try_to_free_page = bdev_try_to_free_page,
1188 };
1189
1190 static const struct super_operations ext4_nojournal_sops = {
1191         .alloc_inode    = ext4_alloc_inode,
1192         .destroy_inode  = ext4_destroy_inode,
1193         .write_inode    = ext4_write_inode,
1194         .dirty_inode    = ext4_dirty_inode,
1195         .drop_inode     = ext4_drop_inode,
1196         .evict_inode    = ext4_evict_inode,
1197         .put_super      = ext4_put_super,
1198         .statfs         = ext4_statfs,
1199         .remount_fs     = ext4_remount,
1200         .show_options   = ext4_show_options,
1201 #ifdef CONFIG_QUOTA
1202         .quota_read     = ext4_quota_read,
1203         .quota_write    = ext4_quota_write,
1204 #endif
1205         .bdev_try_to_free_page = bdev_try_to_free_page,
1206 };
1207
1208 static const struct export_operations ext4_export_ops = {
1209         .fh_to_dentry = ext4_fh_to_dentry,
1210         .fh_to_parent = ext4_fh_to_parent,
1211         .get_parent = ext4_get_parent,
1212 };
1213
1214 enum {
1215         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1216         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1217         Opt_nouid32, Opt_debug, Opt_removed,
1218         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1219         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1220         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1221         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1222         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1223         Opt_data_err_abort, Opt_data_err_ignore,
1224         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1225         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1226         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1227         Opt_usrquota, Opt_grpquota, Opt_i_version,
1228         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1229         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1230         Opt_inode_readahead_blks, Opt_journal_ioprio,
1231         Opt_dioread_nolock, Opt_dioread_lock,
1232         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1233         Opt_max_dir_size_kb,
1234 };
1235
1236 static const match_table_t tokens = {
1237         {Opt_bsd_df, "bsddf"},
1238         {Opt_minix_df, "minixdf"},
1239         {Opt_grpid, "grpid"},
1240         {Opt_grpid, "bsdgroups"},
1241         {Opt_nogrpid, "nogrpid"},
1242         {Opt_nogrpid, "sysvgroups"},
1243         {Opt_resgid, "resgid=%u"},
1244         {Opt_resuid, "resuid=%u"},
1245         {Opt_sb, "sb=%u"},
1246         {Opt_err_cont, "errors=continue"},
1247         {Opt_err_panic, "errors=panic"},
1248         {Opt_err_ro, "errors=remount-ro"},
1249         {Opt_nouid32, "nouid32"},
1250         {Opt_debug, "debug"},
1251         {Opt_removed, "oldalloc"},
1252         {Opt_removed, "orlov"},
1253         {Opt_user_xattr, "user_xattr"},
1254         {Opt_nouser_xattr, "nouser_xattr"},
1255         {Opt_acl, "acl"},
1256         {Opt_noacl, "noacl"},
1257         {Opt_noload, "norecovery"},
1258         {Opt_noload, "noload"},
1259         {Opt_removed, "nobh"},
1260         {Opt_removed, "bh"},
1261         {Opt_commit, "commit=%u"},
1262         {Opt_min_batch_time, "min_batch_time=%u"},
1263         {Opt_max_batch_time, "max_batch_time=%u"},
1264         {Opt_journal_dev, "journal_dev=%u"},
1265         {Opt_journal_checksum, "journal_checksum"},
1266         {Opt_journal_async_commit, "journal_async_commit"},
1267         {Opt_abort, "abort"},
1268         {Opt_data_journal, "data=journal"},
1269         {Opt_data_ordered, "data=ordered"},
1270         {Opt_data_writeback, "data=writeback"},
1271         {Opt_data_err_abort, "data_err=abort"},
1272         {Opt_data_err_ignore, "data_err=ignore"},
1273         {Opt_offusrjquota, "usrjquota="},
1274         {Opt_usrjquota, "usrjquota=%s"},
1275         {Opt_offgrpjquota, "grpjquota="},
1276         {Opt_grpjquota, "grpjquota=%s"},
1277         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1278         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1279         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1280         {Opt_grpquota, "grpquota"},
1281         {Opt_noquota, "noquota"},
1282         {Opt_quota, "quota"},
1283         {Opt_usrquota, "usrquota"},
1284         {Opt_barrier, "barrier=%u"},
1285         {Opt_barrier, "barrier"},
1286         {Opt_nobarrier, "nobarrier"},
1287         {Opt_i_version, "i_version"},
1288         {Opt_stripe, "stripe=%u"},
1289         {Opt_delalloc, "delalloc"},
1290         {Opt_nodelalloc, "nodelalloc"},
1291         {Opt_mblk_io_submit, "mblk_io_submit"},
1292         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1293         {Opt_block_validity, "block_validity"},
1294         {Opt_noblock_validity, "noblock_validity"},
1295         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1296         {Opt_journal_ioprio, "journal_ioprio=%u"},
1297         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1298         {Opt_auto_da_alloc, "auto_da_alloc"},
1299         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1300         {Opt_dioread_nolock, "dioread_nolock"},
1301         {Opt_dioread_lock, "dioread_lock"},
1302         {Opt_discard, "discard"},
1303         {Opt_nodiscard, "nodiscard"},
1304         {Opt_init_itable, "init_itable=%u"},
1305         {Opt_init_itable, "init_itable"},
1306         {Opt_noinit_itable, "noinit_itable"},
1307         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1308         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1309         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1310         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1311         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1312         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1313         {Opt_err, NULL},
1314 };
1315
1316 static ext4_fsblk_t get_sb_block(void **data)
1317 {
1318         ext4_fsblk_t    sb_block;
1319         char            *options = (char *) *data;
1320
1321         if (!options || strncmp(options, "sb=", 3) != 0)
1322                 return 1;       /* Default location */
1323
1324         options += 3;
1325         /* TODO: use simple_strtoll with >32bit ext4 */
1326         sb_block = simple_strtoul(options, &options, 0);
1327         if (*options && *options != ',') {
1328                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1329                        (char *) *data);
1330                 return 1;
1331         }
1332         if (*options == ',')
1333                 options++;
1334         *data = (void *) options;
1335
1336         return sb_block;
1337 }
1338
1339 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1340 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1341         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1342
1343 #ifdef CONFIG_QUOTA
1344 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1345 {
1346         struct ext4_sb_info *sbi = EXT4_SB(sb);
1347         char *qname;
1348
1349         if (sb_any_quota_loaded(sb) &&
1350                 !sbi->s_qf_names[qtype]) {
1351                 ext4_msg(sb, KERN_ERR,
1352                         "Cannot change journaled "
1353                         "quota options when quota turned on");
1354                 return -1;
1355         }
1356         qname = match_strdup(args);
1357         if (!qname) {
1358                 ext4_msg(sb, KERN_ERR,
1359                         "Not enough memory for storing quotafile name");
1360                 return -1;
1361         }
1362         if (sbi->s_qf_names[qtype] &&
1363                 strcmp(sbi->s_qf_names[qtype], qname)) {
1364                 ext4_msg(sb, KERN_ERR,
1365                         "%s quota file already specified", QTYPE2NAME(qtype));
1366                 kfree(qname);
1367                 return -1;
1368         }
1369         sbi->s_qf_names[qtype] = qname;
1370         if (strchr(sbi->s_qf_names[qtype], '/')) {
1371                 ext4_msg(sb, KERN_ERR,
1372                         "quotafile must be on filesystem root");
1373                 kfree(sbi->s_qf_names[qtype]);
1374                 sbi->s_qf_names[qtype] = NULL;
1375                 return -1;
1376         }
1377         set_opt(sb, QUOTA);
1378         return 1;
1379 }
1380
1381 static int clear_qf_name(struct super_block *sb, int qtype)
1382 {
1383
1384         struct ext4_sb_info *sbi = EXT4_SB(sb);
1385
1386         if (sb_any_quota_loaded(sb) &&
1387                 sbi->s_qf_names[qtype]) {
1388                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1389                         " when quota turned on");
1390                 return -1;
1391         }
1392         /*
1393          * The space will be released later when all options are confirmed
1394          * to be correct
1395          */
1396         sbi->s_qf_names[qtype] = NULL;
1397         return 1;
1398 }
1399 #endif
1400
1401 #define MOPT_SET        0x0001
1402 #define MOPT_CLEAR      0x0002
1403 #define MOPT_NOSUPPORT  0x0004
1404 #define MOPT_EXPLICIT   0x0008
1405 #define MOPT_CLEAR_ERR  0x0010
1406 #define MOPT_GTE0       0x0020
1407 #ifdef CONFIG_QUOTA
1408 #define MOPT_Q          0
1409 #define MOPT_QFMT       0x0040
1410 #else
1411 #define MOPT_Q          MOPT_NOSUPPORT
1412 #define MOPT_QFMT       MOPT_NOSUPPORT
1413 #endif
1414 #define MOPT_DATAJ      0x0080
1415
1416 static const struct mount_opts {
1417         int     token;
1418         int     mount_opt;
1419         int     flags;
1420 } ext4_mount_opts[] = {
1421         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1422         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1423         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1424         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1425         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1426         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1427         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1428         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1429         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1430         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1431         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1432         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1433         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1434         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1435         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1436         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1437                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1438         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1439         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1440         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1441         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1442         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1443         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1444         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1445         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1446         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1447         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1448         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1449         {Opt_commit, 0, MOPT_GTE0},
1450         {Opt_max_batch_time, 0, MOPT_GTE0},
1451         {Opt_min_batch_time, 0, MOPT_GTE0},
1452         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1453         {Opt_init_itable, 0, MOPT_GTE0},
1454         {Opt_stripe, 0, MOPT_GTE0},
1455         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1456         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1457         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1458 #ifdef CONFIG_EXT4_FS_XATTR
1459         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1460         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1461 #else
1462         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1463         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1464 #endif
1465 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1466         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1467         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1468 #else
1469         {Opt_acl, 0, MOPT_NOSUPPORT},
1470         {Opt_noacl, 0, MOPT_NOSUPPORT},
1471 #endif
1472         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1473         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1474         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1475         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1476                                                         MOPT_SET | MOPT_Q},
1477         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1478                                                         MOPT_SET | MOPT_Q},
1479         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1480                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1481         {Opt_usrjquota, 0, MOPT_Q},
1482         {Opt_grpjquota, 0, MOPT_Q},
1483         {Opt_offusrjquota, 0, MOPT_Q},
1484         {Opt_offgrpjquota, 0, MOPT_Q},
1485         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1486         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1487         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1488         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1489         {Opt_err, 0, 0}
1490 };
1491
1492 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1493                             substring_t *args, unsigned long *journal_devnum,
1494                             unsigned int *journal_ioprio, int is_remount)
1495 {
1496         struct ext4_sb_info *sbi = EXT4_SB(sb);
1497         const struct mount_opts *m;
1498         kuid_t uid;
1499         kgid_t gid;
1500         int arg = 0;
1501
1502 #ifdef CONFIG_QUOTA
1503         if (token == Opt_usrjquota)
1504                 return set_qf_name(sb, USRQUOTA, &args[0]);
1505         else if (token == Opt_grpjquota)
1506                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1507         else if (token == Opt_offusrjquota)
1508                 return clear_qf_name(sb, USRQUOTA);
1509         else if (token == Opt_offgrpjquota)
1510                 return clear_qf_name(sb, GRPQUOTA);
1511 #endif
1512         if (args->from && match_int(args, &arg))
1513                 return -1;
1514         switch (token) {
1515         case Opt_noacl:
1516         case Opt_nouser_xattr:
1517                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1518                 break;
1519         case Opt_sb:
1520                 return 1;       /* handled by get_sb_block() */
1521         case Opt_removed:
1522                 ext4_msg(sb, KERN_WARNING,
1523                          "Ignoring removed %s option", opt);
1524                 return 1;
1525         case Opt_resuid:
1526                 uid = make_kuid(current_user_ns(), arg);
1527                 if (!uid_valid(uid)) {
1528                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1529                         return -1;
1530                 }
1531                 sbi->s_resuid = uid;
1532                 return 1;
1533         case Opt_resgid:
1534                 gid = make_kgid(current_user_ns(), arg);
1535                 if (!gid_valid(gid)) {
1536                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1537                         return -1;
1538                 }
1539                 sbi->s_resgid = gid;
1540                 return 1;
1541         case Opt_abort:
1542                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1543                 return 1;
1544         case Opt_i_version:
1545                 sb->s_flags |= MS_I_VERSION;
1546                 return 1;
1547         case Opt_journal_dev:
1548                 if (is_remount) {
1549                         ext4_msg(sb, KERN_ERR,
1550                                  "Cannot specify journal on remount");
1551                         return -1;
1552                 }
1553                 *journal_devnum = arg;
1554                 return 1;
1555         case Opt_journal_ioprio:
1556                 if (arg < 0 || arg > 7)
1557                         return -1;
1558                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1559                 return 1;
1560         }
1561
1562         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1563                 if (token != m->token)
1564                         continue;
1565                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1566                         return -1;
1567                 if (m->flags & MOPT_EXPLICIT)
1568                         set_opt2(sb, EXPLICIT_DELALLOC);
1569                 if (m->flags & MOPT_CLEAR_ERR)
1570                         clear_opt(sb, ERRORS_MASK);
1571                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1572                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1573                                  "options when quota turned on");
1574                         return -1;
1575                 }
1576
1577                 if (m->flags & MOPT_NOSUPPORT) {
1578                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1579                 } else if (token == Opt_commit) {
1580                         if (arg == 0)
1581                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1582                         sbi->s_commit_interval = HZ * arg;
1583                 } else if (token == Opt_max_batch_time) {
1584                         if (arg == 0)
1585                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1586                         sbi->s_max_batch_time = arg;
1587                 } else if (token == Opt_min_batch_time) {
1588                         sbi->s_min_batch_time = arg;
1589                 } else if (token == Opt_inode_readahead_blks) {
1590                         if (arg > (1 << 30))
1591                                 return -1;
1592                         if (arg && !is_power_of_2(arg)) {
1593                                 ext4_msg(sb, KERN_ERR,
1594                                          "EXT4-fs: inode_readahead_blks"
1595                                          " must be a power of 2");
1596                                 return -1;
1597                         }
1598                         sbi->s_inode_readahead_blks = arg;
1599                 } else if (token == Opt_init_itable) {
1600                         set_opt(sb, INIT_INODE_TABLE);
1601                         if (!args->from)
1602                                 arg = EXT4_DEF_LI_WAIT_MULT;
1603                         sbi->s_li_wait_mult = arg;
1604                 } else if (token == Opt_max_dir_size_kb) {
1605                         sbi->s_max_dir_size_kb = arg;
1606                 } else if (token == Opt_stripe) {
1607                         sbi->s_stripe = arg;
1608                 } else if (m->flags & MOPT_DATAJ) {
1609                         if (is_remount) {
1610                                 if (!sbi->s_journal)
1611                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1612                                 else if (test_opt(sb, DATA_FLAGS) !=
1613                                          m->mount_opt) {
1614                                         ext4_msg(sb, KERN_ERR,
1615                                          "Cannot change data mode on remount");
1616                                         return -1;
1617                                 }
1618                         } else {
1619                                 clear_opt(sb, DATA_FLAGS);
1620                                 sbi->s_mount_opt |= m->mount_opt;
1621                         }
1622 #ifdef CONFIG_QUOTA
1623                 } else if (m->flags & MOPT_QFMT) {
1624                         if (sb_any_quota_loaded(sb) &&
1625                             sbi->s_jquota_fmt != m->mount_opt) {
1626                                 ext4_msg(sb, KERN_ERR, "Cannot "
1627                                          "change journaled quota options "
1628                                          "when quota turned on");
1629                                 return -1;
1630                         }
1631                         sbi->s_jquota_fmt = m->mount_opt;
1632 #endif
1633                 } else {
1634                         if (!args->from)
1635                                 arg = 1;
1636                         if (m->flags & MOPT_CLEAR)
1637                                 arg = !arg;
1638                         else if (unlikely(!(m->flags & MOPT_SET))) {
1639                                 ext4_msg(sb, KERN_WARNING,
1640                                          "buggy handling of option %s", opt);
1641                                 WARN_ON(1);
1642                                 return -1;
1643                         }
1644                         if (arg != 0)
1645                                 sbi->s_mount_opt |= m->mount_opt;
1646                         else
1647                                 sbi->s_mount_opt &= ~m->mount_opt;
1648                 }
1649                 return 1;
1650         }
1651         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1652                  "or missing value", opt);
1653         return -1;
1654 }
1655
1656 static int parse_options(char *options, struct super_block *sb,
1657                          unsigned long *journal_devnum,
1658                          unsigned int *journal_ioprio,
1659                          int is_remount)
1660 {
1661 #ifdef CONFIG_QUOTA
1662         struct ext4_sb_info *sbi = EXT4_SB(sb);
1663 #endif
1664         char *p;
1665         substring_t args[MAX_OPT_ARGS];
1666         int token;
1667
1668         if (!options)
1669                 return 1;
1670
1671         while ((p = strsep(&options, ",")) != NULL) {
1672                 if (!*p)
1673                         continue;
1674                 /*
1675                  * Initialize args struct so we know whether arg was
1676                  * found; some options take optional arguments.
1677                  */
1678                 args[0].to = args[0].from = 0;
1679                 token = match_token(p, tokens, args);
1680                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1681                                      journal_ioprio, is_remount) < 0)
1682                         return 0;
1683         }
1684 #ifdef CONFIG_QUOTA
1685         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1686                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1687                         clear_opt(sb, USRQUOTA);
1688
1689                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1690                         clear_opt(sb, GRPQUOTA);
1691
1692                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1693                         ext4_msg(sb, KERN_ERR, "old and new quota "
1694                                         "format mixing");
1695                         return 0;
1696                 }
1697
1698                 if (!sbi->s_jquota_fmt) {
1699                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1700                                         "not specified");
1701                         return 0;
1702                 }
1703         } else {
1704                 if (sbi->s_jquota_fmt) {
1705                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1706                                         "specified with no journaling "
1707                                         "enabled");
1708                         return 0;
1709                 }
1710         }
1711 #endif
1712         return 1;
1713 }
1714
1715 static inline void ext4_show_quota_options(struct seq_file *seq,
1716                                            struct super_block *sb)
1717 {
1718 #if defined(CONFIG_QUOTA)
1719         struct ext4_sb_info *sbi = EXT4_SB(sb);
1720
1721         if (sbi->s_jquota_fmt) {
1722                 char *fmtname = "";
1723
1724                 switch (sbi->s_jquota_fmt) {
1725                 case QFMT_VFS_OLD:
1726                         fmtname = "vfsold";
1727                         break;
1728                 case QFMT_VFS_V0:
1729                         fmtname = "vfsv0";
1730                         break;
1731                 case QFMT_VFS_V1:
1732                         fmtname = "vfsv1";
1733                         break;
1734                 }
1735                 seq_printf(seq, ",jqfmt=%s", fmtname);
1736         }
1737
1738         if (sbi->s_qf_names[USRQUOTA])
1739                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1740
1741         if (sbi->s_qf_names[GRPQUOTA])
1742                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1743
1744         if (test_opt(sb, USRQUOTA))
1745                 seq_puts(seq, ",usrquota");
1746
1747         if (test_opt(sb, GRPQUOTA))
1748                 seq_puts(seq, ",grpquota");
1749 #endif
1750 }
1751
1752 static const char *token2str(int token)
1753 {
1754         static const struct match_token *t;
1755
1756         for (t = tokens; t->token != Opt_err; t++)
1757                 if (t->token == token && !strchr(t->pattern, '='))
1758                         break;
1759         return t->pattern;
1760 }
1761
1762 /*
1763  * Show an option if
1764  *  - it's set to a non-default value OR
1765  *  - if the per-sb default is different from the global default
1766  */
1767 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1768                               int nodefs)
1769 {
1770         struct ext4_sb_info *sbi = EXT4_SB(sb);
1771         struct ext4_super_block *es = sbi->s_es;
1772         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1773         const struct mount_opts *m;
1774         char sep = nodefs ? '\n' : ',';
1775
1776 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1777 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1778
1779         if (sbi->s_sb_block != 1)
1780                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1781
1782         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1783                 int want_set = m->flags & MOPT_SET;
1784                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1785                     (m->flags & MOPT_CLEAR_ERR))
1786                         continue;
1787                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1788                         continue; /* skip if same as the default */
1789                 if ((want_set &&
1790                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1791                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1792                         continue; /* select Opt_noFoo vs Opt_Foo */
1793                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1794         }
1795
1796         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1797             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1798                 SEQ_OPTS_PRINT("resuid=%u",
1799                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1800         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1801             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1802                 SEQ_OPTS_PRINT("resgid=%u",
1803                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1804         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1805         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1806                 SEQ_OPTS_PUTS("errors=remount-ro");
1807         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1808                 SEQ_OPTS_PUTS("errors=continue");
1809         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1810                 SEQ_OPTS_PUTS("errors=panic");
1811         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1812                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1813         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1814                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1815         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1816                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1817         if (sb->s_flags & MS_I_VERSION)
1818                 SEQ_OPTS_PUTS("i_version");
1819         if (nodefs || sbi->s_stripe)
1820                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1821         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1822                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1823                         SEQ_OPTS_PUTS("data=journal");
1824                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1825                         SEQ_OPTS_PUTS("data=ordered");
1826                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1827                         SEQ_OPTS_PUTS("data=writeback");
1828         }
1829         if (nodefs ||
1830             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1831                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1832                                sbi->s_inode_readahead_blks);
1833
1834         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1835                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1836                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1837         if (nodefs || sbi->s_max_dir_size_kb)
1838                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1839
1840         ext4_show_quota_options(seq, sb);
1841         return 0;
1842 }
1843
1844 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1845 {
1846         return _ext4_show_options(seq, root->d_sb, 0);
1847 }
1848
1849 static int options_seq_show(struct seq_file *seq, void *offset)
1850 {
1851         struct super_block *sb = seq->private;
1852         int rc;
1853
1854         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1855         rc = _ext4_show_options(seq, sb, 1);
1856         seq_puts(seq, "\n");
1857         return rc;
1858 }
1859
1860 static int options_open_fs(struct inode *inode, struct file *file)
1861 {
1862         return single_open(file, options_seq_show, PDE(inode)->data);
1863 }
1864
1865 static const struct file_operations ext4_seq_options_fops = {
1866         .owner = THIS_MODULE,
1867         .open = options_open_fs,
1868         .read = seq_read,
1869         .llseek = seq_lseek,
1870         .release = single_release,
1871 };
1872
1873 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1874                             int read_only)
1875 {
1876         struct ext4_sb_info *sbi = EXT4_SB(sb);
1877         int res = 0;
1878
1879         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1880                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1881                          "forcing read-only mode");
1882                 res = MS_RDONLY;
1883         }
1884         if (read_only)
1885                 goto done;
1886         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1887                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1888                          "running e2fsck is recommended");
1889         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1890                 ext4_msg(sb, KERN_WARNING,
1891                          "warning: mounting fs with errors, "
1892                          "running e2fsck is recommended");
1893         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1894                  le16_to_cpu(es->s_mnt_count) >=
1895                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1896                 ext4_msg(sb, KERN_WARNING,
1897                          "warning: maximal mount count reached, "
1898                          "running e2fsck is recommended");
1899         else if (le32_to_cpu(es->s_checkinterval) &&
1900                 (le32_to_cpu(es->s_lastcheck) +
1901                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1902                 ext4_msg(sb, KERN_WARNING,
1903                          "warning: checktime reached, "
1904                          "running e2fsck is recommended");
1905         if (!sbi->s_journal)
1906                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1907         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1908                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1909         le16_add_cpu(&es->s_mnt_count, 1);
1910         es->s_mtime = cpu_to_le32(get_seconds());
1911         ext4_update_dynamic_rev(sb);
1912         if (sbi->s_journal)
1913                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1914
1915         ext4_commit_super(sb, 1);
1916 done:
1917         if (test_opt(sb, DEBUG))
1918                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1919                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1920                         sb->s_blocksize,
1921                         sbi->s_groups_count,
1922                         EXT4_BLOCKS_PER_GROUP(sb),
1923                         EXT4_INODES_PER_GROUP(sb),
1924                         sbi->s_mount_opt, sbi->s_mount_opt2);
1925
1926         cleancache_init_fs(sb);
1927         return res;
1928 }
1929
1930 static int ext4_fill_flex_info(struct super_block *sb)
1931 {
1932         struct ext4_sb_info *sbi = EXT4_SB(sb);
1933         struct ext4_group_desc *gdp = NULL;
1934         ext4_group_t flex_group_count;
1935         ext4_group_t flex_group;
1936         unsigned int groups_per_flex = 0;
1937         size_t size;
1938         int i;
1939
1940         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1941         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1942                 sbi->s_log_groups_per_flex = 0;
1943                 return 1;
1944         }
1945         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1946
1947         /* We allocate both existing and potentially added groups */
1948         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1949                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1950                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1951         size = flex_group_count * sizeof(struct flex_groups);
1952         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1953         if (sbi->s_flex_groups == NULL) {
1954                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1955                          flex_group_count);
1956                 goto failed;
1957         }
1958
1959         for (i = 0; i < sbi->s_groups_count; i++) {
1960                 gdp = ext4_get_group_desc(sb, i, NULL);
1961
1962                 flex_group = ext4_flex_group(sbi, i);
1963                 atomic_add(ext4_free_inodes_count(sb, gdp),
1964                            &sbi->s_flex_groups[flex_group].free_inodes);
1965                 atomic_add(ext4_free_group_clusters(sb, gdp),
1966                            &sbi->s_flex_groups[flex_group].free_clusters);
1967                 atomic_add(ext4_used_dirs_count(sb, gdp),
1968                            &sbi->s_flex_groups[flex_group].used_dirs);
1969         }
1970
1971         return 1;
1972 failed:
1973         return 0;
1974 }
1975
1976 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1977                                    struct ext4_group_desc *gdp)
1978 {
1979         int offset;
1980         __u16 crc = 0;
1981         __le32 le_group = cpu_to_le32(block_group);
1982
1983         if ((sbi->s_es->s_feature_ro_compat &
1984              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1985                 /* Use new metadata_csum algorithm */
1986                 __u16 old_csum;
1987                 __u32 csum32;
1988
1989                 old_csum = gdp->bg_checksum;
1990                 gdp->bg_checksum = 0;
1991                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1992                                      sizeof(le_group));
1993                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1994                                      sbi->s_desc_size);
1995                 gdp->bg_checksum = old_csum;
1996
1997                 crc = csum32 & 0xFFFF;
1998                 goto out;
1999         }
2000
2001         /* old crc16 code */
2002         offset = offsetof(struct ext4_group_desc, bg_checksum);
2003
2004         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2005         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2006         crc = crc16(crc, (__u8 *)gdp, offset);
2007         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2008         /* for checksum of struct ext4_group_desc do the rest...*/
2009         if ((sbi->s_es->s_feature_incompat &
2010              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2011             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2012                 crc = crc16(crc, (__u8 *)gdp + offset,
2013                             le16_to_cpu(sbi->s_es->s_desc_size) -
2014                                 offset);
2015
2016 out:
2017         return cpu_to_le16(crc);
2018 }
2019
2020 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2021                                 struct ext4_group_desc *gdp)
2022 {
2023         if (ext4_has_group_desc_csum(sb) &&
2024             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2025                                                       block_group, gdp)))
2026                 return 0;
2027
2028         return 1;
2029 }
2030
2031 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2032                               struct ext4_group_desc *gdp)
2033 {
2034         if (!ext4_has_group_desc_csum(sb))
2035                 return;
2036         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2037 }
2038
2039 /* Called at mount-time, super-block is locked */
2040 static int ext4_check_descriptors(struct super_block *sb,
2041                                   ext4_group_t *first_not_zeroed)
2042 {
2043         struct ext4_sb_info *sbi = EXT4_SB(sb);
2044         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2045         ext4_fsblk_t last_block;
2046         ext4_fsblk_t block_bitmap;
2047         ext4_fsblk_t inode_bitmap;
2048         ext4_fsblk_t inode_table;
2049         int flexbg_flag = 0;
2050         ext4_group_t i, grp = sbi->s_groups_count;
2051
2052         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2053                 flexbg_flag = 1;
2054
2055         ext4_debug("Checking group descriptors");
2056
2057         for (i = 0; i < sbi->s_groups_count; i++) {
2058                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2059
2060                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2061                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2062                 else
2063                         last_block = first_block +
2064                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2065
2066                 if ((grp == sbi->s_groups_count) &&
2067                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2068                         grp = i;
2069
2070                 block_bitmap = ext4_block_bitmap(sb, gdp);
2071                 if (block_bitmap < first_block || block_bitmap > last_block) {
2072                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2073                                "Block bitmap for group %u not in group "
2074                                "(block %llu)!", i, block_bitmap);
2075                         return 0;
2076                 }
2077                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2078                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2079                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2080                                "Inode bitmap for group %u not in group "
2081                                "(block %llu)!", i, inode_bitmap);
2082                         return 0;
2083                 }
2084                 inode_table = ext4_inode_table(sb, gdp);
2085                 if (inode_table < first_block ||
2086                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2087                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2088                                "Inode table for group %u not in group "
2089                                "(block %llu)!", i, inode_table);
2090                         return 0;
2091                 }
2092                 ext4_lock_group(sb, i);
2093                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2094                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2095                                  "Checksum for group %u failed (%u!=%u)",
2096                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2097                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2098                         if (!(sb->s_flags & MS_RDONLY)) {
2099                                 ext4_unlock_group(sb, i);
2100                                 return 0;
2101                         }
2102                 }
2103                 ext4_unlock_group(sb, i);
2104                 if (!flexbg_flag)
2105                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2106         }
2107         if (NULL != first_not_zeroed)
2108                 *first_not_zeroed = grp;
2109
2110         ext4_free_blocks_count_set(sbi->s_es,
2111                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2112         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2113         return 1;
2114 }
2115
2116 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2117  * the superblock) which were deleted from all directories, but held open by
2118  * a process at the time of a crash.  We walk the list and try to delete these
2119  * inodes at recovery time (only with a read-write filesystem).
2120  *
2121  * In order to keep the orphan inode chain consistent during traversal (in
2122  * case of crash during recovery), we link each inode into the superblock
2123  * orphan list_head and handle it the same way as an inode deletion during
2124  * normal operation (which journals the operations for us).
2125  *
2126  * We only do an iget() and an iput() on each inode, which is very safe if we
2127  * accidentally point at an in-use or already deleted inode.  The worst that
2128  * can happen in this case is that we get a "bit already cleared" message from
2129  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2130  * e2fsck was run on this filesystem, and it must have already done the orphan
2131  * inode cleanup for us, so we can safely abort without any further action.
2132  */
2133 static void ext4_orphan_cleanup(struct super_block *sb,
2134                                 struct ext4_super_block *es)
2135 {
2136         unsigned int s_flags = sb->s_flags;
2137         int nr_orphans = 0, nr_truncates = 0;
2138 #ifdef CONFIG_QUOTA
2139         int i;
2140 #endif
2141         if (!es->s_last_orphan) {
2142                 jbd_debug(4, "no orphan inodes to clean up\n");
2143                 return;
2144         }
2145
2146         if (bdev_read_only(sb->s_bdev)) {
2147                 ext4_msg(sb, KERN_ERR, "write access "
2148                         "unavailable, skipping orphan cleanup");
2149                 return;
2150         }
2151
2152         /* Check if feature set would not allow a r/w mount */
2153         if (!ext4_feature_set_ok(sb, 0)) {
2154                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2155                          "unknown ROCOMPAT features");
2156                 return;
2157         }
2158
2159         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2160                 if (es->s_last_orphan)
2161                         jbd_debug(1, "Errors on filesystem, "
2162                                   "clearing orphan list.\n");
2163                 es->s_last_orphan = 0;
2164                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2165                 return;
2166         }
2167
2168         if (s_flags & MS_RDONLY) {
2169                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2170                 sb->s_flags &= ~MS_RDONLY;
2171         }
2172 #ifdef CONFIG_QUOTA
2173         /* Needed for iput() to work correctly and not trash data */
2174         sb->s_flags |= MS_ACTIVE;
2175         /* Turn on quotas so that they are updated correctly */
2176         for (i = 0; i < MAXQUOTAS; i++) {
2177                 if (EXT4_SB(sb)->s_qf_names[i]) {
2178                         int ret = ext4_quota_on_mount(sb, i);
2179                         if (ret < 0)
2180                                 ext4_msg(sb, KERN_ERR,
2181                                         "Cannot turn on journaled "
2182                                         "quota: error %d", ret);
2183                 }
2184         }
2185 #endif
2186
2187         while (es->s_last_orphan) {
2188                 struct inode *inode;
2189
2190                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2191                 if (IS_ERR(inode)) {
2192                         es->s_last_orphan = 0;
2193                         break;
2194                 }
2195
2196                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2197                 dquot_initialize(inode);
2198                 if (inode->i_nlink) {
2199                         ext4_msg(sb, KERN_DEBUG,
2200                                 "%s: truncating inode %lu to %lld bytes",
2201                                 __func__, inode->i_ino, inode->i_size);
2202                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2203                                   inode->i_ino, inode->i_size);
2204                         ext4_truncate(inode);
2205                         nr_truncates++;
2206                 } else {
2207                         ext4_msg(sb, KERN_DEBUG,
2208                                 "%s: deleting unreferenced inode %lu",
2209                                 __func__, inode->i_ino);
2210                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2211                                   inode->i_ino);
2212                         nr_orphans++;
2213                 }
2214                 iput(inode);  /* The delete magic happens here! */
2215         }
2216
2217 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2218
2219         if (nr_orphans)
2220                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2221                        PLURAL(nr_orphans));
2222         if (nr_truncates)
2223                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2224                        PLURAL(nr_truncates));
2225 #ifdef CONFIG_QUOTA
2226         /* Turn quotas off */
2227         for (i = 0; i < MAXQUOTAS; i++) {
2228                 if (sb_dqopt(sb)->files[i])
2229                         dquot_quota_off(sb, i);
2230         }
2231 #endif
2232         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2233 }
2234
2235 /*
2236  * Maximal extent format file size.
2237  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2238  * extent format containers, within a sector_t, and within i_blocks
2239  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2240  * so that won't be a limiting factor.
2241  *
2242  * However there is other limiting factor. We do store extents in the form
2243  * of starting block and length, hence the resulting length of the extent
2244  * covering maximum file size must fit into on-disk format containers as
2245  * well. Given that length is always by 1 unit bigger than max unit (because
2246  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2247  *
2248  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2249  */
2250 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2251 {
2252         loff_t res;
2253         loff_t upper_limit = MAX_LFS_FILESIZE;
2254
2255         /* small i_blocks in vfs inode? */
2256         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2257                 /*
2258                  * CONFIG_LBDAF is not enabled implies the inode
2259                  * i_block represent total blocks in 512 bytes
2260                  * 32 == size of vfs inode i_blocks * 8
2261                  */
2262                 upper_limit = (1LL << 32) - 1;
2263
2264                 /* total blocks in file system block size */
2265                 upper_limit >>= (blkbits - 9);
2266                 upper_limit <<= blkbits;
2267         }
2268
2269         /*
2270          * 32-bit extent-start container, ee_block. We lower the maxbytes
2271          * by one fs block, so ee_len can cover the extent of maximum file
2272          * size
2273          */
2274         res = (1LL << 32) - 1;
2275         res <<= blkbits;
2276
2277         /* Sanity check against vm- & vfs- imposed limits */
2278         if (res > upper_limit)
2279                 res = upper_limit;
2280
2281         return res;
2282 }
2283
2284 /*
2285  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2286  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2287  * We need to be 1 filesystem block less than the 2^48 sector limit.
2288  */
2289 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2290 {
2291         loff_t res = EXT4_NDIR_BLOCKS;
2292         int meta_blocks;
2293         loff_t upper_limit;
2294         /* This is calculated to be the largest file size for a dense, block
2295          * mapped file such that the file's total number of 512-byte sectors,
2296          * including data and all indirect blocks, does not exceed (2^48 - 1).
2297          *
2298          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2299          * number of 512-byte sectors of the file.
2300          */
2301
2302         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2303                 /*
2304                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2305                  * the inode i_block field represents total file blocks in
2306                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2307                  */
2308                 upper_limit = (1LL << 32) - 1;
2309
2310                 /* total blocks in file system block size */
2311                 upper_limit >>= (bits - 9);
2312
2313         } else {
2314                 /*
2315                  * We use 48 bit ext4_inode i_blocks
2316                  * With EXT4_HUGE_FILE_FL set the i_blocks
2317                  * represent total number of blocks in
2318                  * file system block size
2319                  */
2320                 upper_limit = (1LL << 48) - 1;
2321
2322         }
2323
2324         /* indirect blocks */
2325         meta_blocks = 1;
2326         /* double indirect blocks */
2327         meta_blocks += 1 + (1LL << (bits-2));
2328         /* tripple indirect blocks */
2329         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2330
2331         upper_limit -= meta_blocks;
2332         upper_limit <<= bits;
2333
2334         res += 1LL << (bits-2);
2335         res += 1LL << (2*(bits-2));
2336         res += 1LL << (3*(bits-2));
2337         res <<= bits;
2338         if (res > upper_limit)
2339                 res = upper_limit;
2340
2341         if (res > MAX_LFS_FILESIZE)
2342                 res = MAX_LFS_FILESIZE;
2343
2344         return res;
2345 }
2346
2347 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2348                                    ext4_fsblk_t logical_sb_block, int nr)
2349 {
2350         struct ext4_sb_info *sbi = EXT4_SB(sb);
2351         ext4_group_t bg, first_meta_bg;
2352         int has_super = 0;
2353
2354         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2355
2356         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2357             nr < first_meta_bg)
2358                 return logical_sb_block + nr + 1;
2359         bg = sbi->s_desc_per_block * nr;
2360         if (ext4_bg_has_super(sb, bg))
2361                 has_super = 1;
2362
2363         return (has_super + ext4_group_first_block_no(sb, bg));
2364 }
2365
2366 /**
2367  * ext4_get_stripe_size: Get the stripe size.
2368  * @sbi: In memory super block info
2369  *
2370  * If we have specified it via mount option, then
2371  * use the mount option value. If the value specified at mount time is
2372  * greater than the blocks per group use the super block value.
2373  * If the super block value is greater than blocks per group return 0.
2374  * Allocator needs it be less than blocks per group.
2375  *
2376  */
2377 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2378 {
2379         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2380         unsigned long stripe_width =
2381                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2382         int ret;
2383
2384         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2385                 ret = sbi->s_stripe;
2386         else if (stripe_width <= sbi->s_blocks_per_group)
2387                 ret = stripe_width;
2388         else if (stride <= sbi->s_blocks_per_group)
2389                 ret = stride;
2390         else
2391                 ret = 0;
2392
2393         /*
2394          * If the stripe width is 1, this makes no sense and
2395          * we set it to 0 to turn off stripe handling code.
2396          */
2397         if (ret <= 1)
2398                 ret = 0;
2399
2400         return ret;
2401 }
2402
2403 /* sysfs supprt */
2404
2405 struct ext4_attr {
2406         struct attribute attr;
2407         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2408         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2409                          const char *, size_t);
2410         int offset;
2411 };
2412
2413 static int parse_strtoul(const char *buf,
2414                 unsigned long max, unsigned long *value)
2415 {
2416         char *endp;
2417
2418         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2419         endp = skip_spaces(endp);
2420         if (*endp || *value > max)
2421                 return -EINVAL;
2422
2423         return 0;
2424 }
2425
2426 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2427                                               struct ext4_sb_info *sbi,
2428                                               char *buf)
2429 {
2430         return snprintf(buf, PAGE_SIZE, "%llu\n",
2431                 (s64) EXT4_C2B(sbi,
2432                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2433 }
2434
2435 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2436                                          struct ext4_sb_info *sbi, char *buf)
2437 {
2438         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2439
2440         if (!sb->s_bdev->bd_part)
2441                 return snprintf(buf, PAGE_SIZE, "0\n");
2442         return snprintf(buf, PAGE_SIZE, "%lu\n",
2443                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2444                          sbi->s_sectors_written_start) >> 1);
2445 }
2446
2447 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2448                                           struct ext4_sb_info *sbi, char *buf)
2449 {
2450         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2451
2452         if (!sb->s_bdev->bd_part)
2453                 return snprintf(buf, PAGE_SIZE, "0\n");
2454         return snprintf(buf, PAGE_SIZE, "%llu\n",
2455                         (unsigned long long)(sbi->s_kbytes_written +
2456                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2457                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2458 }
2459
2460 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2461                                           struct ext4_sb_info *sbi,
2462                                           const char *buf, size_t count)
2463 {
2464         unsigned long t;
2465
2466         if (parse_strtoul(buf, 0x40000000, &t))
2467                 return -EINVAL;
2468
2469         if (t && !is_power_of_2(t))
2470                 return -EINVAL;
2471
2472         sbi->s_inode_readahead_blks = t;
2473         return count;
2474 }
2475
2476 static ssize_t sbi_ui_show(struct ext4_attr *a,
2477                            struct ext4_sb_info *sbi, char *buf)
2478 {
2479         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2480
2481         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2482 }
2483
2484 static ssize_t sbi_ui_store(struct ext4_attr *a,
2485                             struct ext4_sb_info *sbi,
2486                             const char *buf, size_t count)
2487 {
2488         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2489         unsigned long t;
2490
2491         if (parse_strtoul(buf, 0xffffffff, &t))
2492                 return -EINVAL;
2493         *ui = t;
2494         return count;
2495 }
2496
2497 static ssize_t trigger_test_error(struct ext4_attr *a,
2498                                   struct ext4_sb_info *sbi,
2499                                   const char *buf, size_t count)
2500 {
2501         int len = count;
2502
2503         if (!capable(CAP_SYS_ADMIN))
2504                 return -EPERM;
2505
2506         if (len && buf[len-1] == '\n')
2507                 len--;
2508
2509         if (len)
2510                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2511         return count;
2512 }
2513
2514 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2515 static struct ext4_attr ext4_attr_##_name = {                   \
2516         .attr = {.name = __stringify(_name), .mode = _mode },   \
2517         .show   = _show,                                        \
2518         .store  = _store,                                       \
2519         .offset = offsetof(struct ext4_sb_info, _elname),       \
2520 }
2521 #define EXT4_ATTR(name, mode, show, store) \
2522 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2523
2524 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2525 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2526 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2527 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2528         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2529 #define ATTR_LIST(name) &ext4_attr_##name.attr
2530
2531 EXT4_RO_ATTR(delayed_allocation_blocks);
2532 EXT4_RO_ATTR(session_write_kbytes);
2533 EXT4_RO_ATTR(lifetime_write_kbytes);
2534 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2535                  inode_readahead_blks_store, s_inode_readahead_blks);
2536 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2537 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2538 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2539 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2540 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2541 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2542 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2543 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2544 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2545
2546 static struct attribute *ext4_attrs[] = {
2547         ATTR_LIST(delayed_allocation_blocks),
2548         ATTR_LIST(session_write_kbytes),
2549         ATTR_LIST(lifetime_write_kbytes),
2550         ATTR_LIST(inode_readahead_blks),
2551         ATTR_LIST(inode_goal),
2552         ATTR_LIST(mb_stats),
2553         ATTR_LIST(mb_max_to_scan),
2554         ATTR_LIST(mb_min_to_scan),
2555         ATTR_LIST(mb_order2_req),
2556         ATTR_LIST(mb_stream_req),
2557         ATTR_LIST(mb_group_prealloc),
2558         ATTR_LIST(max_writeback_mb_bump),
2559         ATTR_LIST(trigger_fs_error),
2560         NULL,
2561 };
2562
2563 /* Features this copy of ext4 supports */
2564 EXT4_INFO_ATTR(lazy_itable_init);
2565 EXT4_INFO_ATTR(batched_discard);
2566
2567 static struct attribute *ext4_feat_attrs[] = {
2568         ATTR_LIST(lazy_itable_init),
2569         ATTR_LIST(batched_discard),
2570         NULL,
2571 };
2572
2573 static ssize_t ext4_attr_show(struct kobject *kobj,
2574                               struct attribute *attr, char *buf)
2575 {
2576         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2577                                                 s_kobj);
2578         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2579
2580         return a->show ? a->show(a, sbi, buf) : 0;
2581 }
2582
2583 static ssize_t ext4_attr_store(struct kobject *kobj,
2584                                struct attribute *attr,
2585                                const char *buf, size_t len)
2586 {
2587         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2588                                                 s_kobj);
2589         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2590
2591         return a->store ? a->store(a, sbi, buf, len) : 0;
2592 }
2593
2594 static void ext4_sb_release(struct kobject *kobj)
2595 {
2596         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2597                                                 s_kobj);
2598         complete(&sbi->s_kobj_unregister);
2599 }
2600
2601 static const struct sysfs_ops ext4_attr_ops = {
2602         .show   = ext4_attr_show,
2603         .store  = ext4_attr_store,
2604 };
2605
2606 static struct kobj_type ext4_ktype = {
2607         .default_attrs  = ext4_attrs,
2608         .sysfs_ops      = &ext4_attr_ops,
2609         .release        = ext4_sb_release,
2610 };
2611
2612 static void ext4_feat_release(struct kobject *kobj)
2613 {
2614         complete(&ext4_feat->f_kobj_unregister);
2615 }
2616
2617 static struct kobj_type ext4_feat_ktype = {
2618         .default_attrs  = ext4_feat_attrs,
2619         .sysfs_ops      = &ext4_attr_ops,
2620         .release        = ext4_feat_release,
2621 };
2622
2623 /*
2624  * Check whether this filesystem can be mounted based on
2625  * the features present and the RDONLY/RDWR mount requested.
2626  * Returns 1 if this filesystem can be mounted as requested,
2627  * 0 if it cannot be.
2628  */
2629 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2630 {
2631         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2632                 ext4_msg(sb, KERN_ERR,
2633                         "Couldn't mount because of "
2634                         "unsupported optional features (%x)",
2635                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2636                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2637                 return 0;
2638         }
2639
2640         if (readonly)
2641                 return 1;
2642
2643         /* Check that feature set is OK for a read-write mount */
2644         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2645                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2646                          "unsupported optional features (%x)",
2647                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2648                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2649                 return 0;
2650         }
2651         /*
2652          * Large file size enabled file system can only be mounted
2653          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2654          */
2655         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2656                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2657                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2658                                  "cannot be mounted RDWR without "
2659                                  "CONFIG_LBDAF");
2660                         return 0;
2661                 }
2662         }
2663         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2664             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2665                 ext4_msg(sb, KERN_ERR,
2666                          "Can't support bigalloc feature without "
2667                          "extents feature\n");
2668                 return 0;
2669         }
2670
2671 #ifndef CONFIG_QUOTA
2672         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2673             !readonly) {
2674                 ext4_msg(sb, KERN_ERR,
2675                          "Filesystem with quota feature cannot be mounted RDWR "
2676                          "without CONFIG_QUOTA");
2677                 return 0;
2678         }
2679 #endif  /* CONFIG_QUOTA */
2680         return 1;
2681 }
2682
2683 /*
2684  * This function is called once a day if we have errors logged
2685  * on the file system
2686  */
2687 static void print_daily_error_info(unsigned long arg)
2688 {
2689         struct super_block *sb = (struct super_block *) arg;
2690         struct ext4_sb_info *sbi;
2691         struct ext4_super_block *es;
2692
2693         sbi = EXT4_SB(sb);
2694         es = sbi->s_es;
2695
2696         if (es->s_error_count)
2697                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2698                          le32_to_cpu(es->s_error_count));
2699         if (es->s_first_error_time) {
2700                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2701                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2702                        (int) sizeof(es->s_first_error_func),
2703                        es->s_first_error_func,
2704                        le32_to_cpu(es->s_first_error_line));
2705                 if (es->s_first_error_ino)
2706                         printk(": inode %u",
2707                                le32_to_cpu(es->s_first_error_ino));
2708                 if (es->s_first_error_block)
2709                         printk(": block %llu", (unsigned long long)
2710                                le64_to_cpu(es->s_first_error_block));
2711                 printk("\n");
2712         }
2713         if (es->s_last_error_time) {
2714                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2715                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2716                        (int) sizeof(es->s_last_error_func),
2717                        es->s_last_error_func,
2718                        le32_to_cpu(es->s_last_error_line));
2719                 if (es->s_last_error_ino)
2720                         printk(": inode %u",
2721                                le32_to_cpu(es->s_last_error_ino));
2722                 if (es->s_last_error_block)
2723                         printk(": block %llu", (unsigned long long)
2724                                le64_to_cpu(es->s_last_error_block));
2725                 printk("\n");
2726         }
2727         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2728 }
2729
2730 /* Find next suitable group and run ext4_init_inode_table */
2731 static int ext4_run_li_request(struct ext4_li_request *elr)
2732 {
2733         struct ext4_group_desc *gdp = NULL;
2734         ext4_group_t group, ngroups;
2735         struct super_block *sb;
2736         unsigned long timeout = 0;
2737         int ret = 0;
2738
2739         sb = elr->lr_super;
2740         ngroups = EXT4_SB(sb)->s_groups_count;
2741
2742         sb_start_write(sb);
2743         for (group = elr->lr_next_group; group < ngroups; group++) {
2744                 gdp = ext4_get_group_desc(sb, group, NULL);
2745                 if (!gdp) {
2746                         ret = 1;
2747                         break;
2748                 }
2749
2750                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2751                         break;
2752         }
2753
2754         if (group == ngroups)
2755                 ret = 1;
2756
2757         if (!ret) {
2758                 timeout = jiffies;
2759                 ret = ext4_init_inode_table(sb, group,
2760                                             elr->lr_timeout ? 0 : 1);
2761                 if (elr->lr_timeout == 0) {
2762                         timeout = (jiffies - timeout) *
2763                                   elr->lr_sbi->s_li_wait_mult;
2764                         elr->lr_timeout = timeout;
2765                 }
2766                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2767                 elr->lr_next_group = group + 1;
2768         }
2769         sb_end_write(sb);
2770
2771         return ret;
2772 }
2773
2774 /*
2775  * Remove lr_request from the list_request and free the
2776  * request structure. Should be called with li_list_mtx held
2777  */
2778 static void ext4_remove_li_request(struct ext4_li_request *elr)
2779 {
2780         struct ext4_sb_info *sbi;
2781
2782         if (!elr)
2783                 return;
2784
2785         sbi = elr->lr_sbi;
2786
2787         list_del(&elr->lr_request);
2788         sbi->s_li_request = NULL;
2789         kfree(elr);
2790 }
2791
2792 static void ext4_unregister_li_request(struct super_block *sb)
2793 {
2794         mutex_lock(&ext4_li_mtx);
2795         if (!ext4_li_info) {
2796                 mutex_unlock(&ext4_li_mtx);
2797                 return;
2798         }
2799
2800         mutex_lock(&ext4_li_info->li_list_mtx);
2801         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2802         mutex_unlock(&ext4_li_info->li_list_mtx);
2803         mutex_unlock(&ext4_li_mtx);
2804 }
2805
2806 static struct task_struct *ext4_lazyinit_task;
2807
2808 /*
2809  * This is the function where ext4lazyinit thread lives. It walks
2810  * through the request list searching for next scheduled filesystem.
2811  * When such a fs is found, run the lazy initialization request
2812  * (ext4_rn_li_request) and keep track of the time spend in this
2813  * function. Based on that time we compute next schedule time of
2814  * the request. When walking through the list is complete, compute
2815  * next waking time and put itself into sleep.
2816  */
2817 static int ext4_lazyinit_thread(void *arg)
2818 {
2819         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2820         struct list_head *pos, *n;
2821         struct ext4_li_request *elr;
2822         unsigned long next_wakeup, cur;
2823
2824         BUG_ON(NULL == eli);
2825
2826 cont_thread:
2827         while (true) {
2828                 next_wakeup = MAX_JIFFY_OFFSET;
2829
2830                 mutex_lock(&eli->li_list_mtx);
2831                 if (list_empty(&eli->li_request_list)) {
2832                         mutex_unlock(&eli->li_list_mtx);
2833                         goto exit_thread;
2834                 }
2835
2836                 list_for_each_safe(pos, n, &eli->li_request_list) {
2837                         elr = list_entry(pos, struct ext4_li_request,
2838                                          lr_request);
2839
2840                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2841                                 if (ext4_run_li_request(elr) != 0) {
2842                                         /* error, remove the lazy_init job */
2843                                         ext4_remove_li_request(elr);
2844                                         continue;
2845                                 }
2846                         }
2847
2848                         if (time_before(elr->lr_next_sched, next_wakeup))
2849                                 next_wakeup = elr->lr_next_sched;
2850                 }
2851                 mutex_unlock(&eli->li_list_mtx);
2852
2853                 try_to_freeze();
2854
2855                 cur = jiffies;
2856                 if ((time_after_eq(cur, next_wakeup)) ||
2857                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2858                         cond_resched();
2859                         continue;
2860                 }
2861
2862                 schedule_timeout_interruptible(next_wakeup - cur);
2863
2864                 if (kthread_should_stop()) {
2865                         ext4_clear_request_list();
2866                         goto exit_thread;
2867                 }
2868         }
2869
2870 exit_thread:
2871         /*
2872          * It looks like the request list is empty, but we need
2873          * to check it under the li_list_mtx lock, to prevent any
2874          * additions into it, and of course we should lock ext4_li_mtx
2875          * to atomically free the list and ext4_li_info, because at
2876          * this point another ext4 filesystem could be registering
2877          * new one.
2878          */
2879         mutex_lock(&ext4_li_mtx);
2880         mutex_lock(&eli->li_list_mtx);
2881         if (!list_empty(&eli->li_request_list)) {
2882                 mutex_unlock(&eli->li_list_mtx);
2883                 mutex_unlock(&ext4_li_mtx);
2884                 goto cont_thread;
2885         }
2886         mutex_unlock(&eli->li_list_mtx);
2887         kfree(ext4_li_info);
2888         ext4_li_info = NULL;
2889         mutex_unlock(&ext4_li_mtx);
2890
2891         return 0;
2892 }
2893
2894 static void ext4_clear_request_list(void)
2895 {
2896         struct list_head *pos, *n;
2897         struct ext4_li_request *elr;
2898
2899         mutex_lock(&ext4_li_info->li_list_mtx);
2900         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2901                 elr = list_entry(pos, struct ext4_li_request,
2902                                  lr_request);
2903                 ext4_remove_li_request(elr);
2904         }
2905         mutex_unlock(&ext4_li_info->li_list_mtx);
2906 }
2907
2908 static int ext4_run_lazyinit_thread(void)
2909 {
2910         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2911                                          ext4_li_info, "ext4lazyinit");
2912         if (IS_ERR(ext4_lazyinit_task)) {
2913                 int err = PTR_ERR(ext4_lazyinit_task);
2914                 ext4_clear_request_list();
2915                 kfree(ext4_li_info);
2916                 ext4_li_info = NULL;
2917                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2918                                  "initialization thread\n",
2919                                  err);
2920                 return err;
2921         }
2922         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2923         return 0;
2924 }
2925
2926 /*
2927  * Check whether it make sense to run itable init. thread or not.
2928  * If there is at least one uninitialized inode table, return
2929  * corresponding group number, else the loop goes through all
2930  * groups and return total number of groups.
2931  */
2932 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2933 {
2934         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2935         struct ext4_group_desc *gdp = NULL;
2936
2937         for (group = 0; group < ngroups; group++) {
2938                 gdp = ext4_get_group_desc(sb, group, NULL);
2939                 if (!gdp)
2940                         continue;
2941
2942                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2943                         break;
2944         }
2945
2946         return group;
2947 }
2948
2949 static int ext4_li_info_new(void)
2950 {
2951         struct ext4_lazy_init *eli = NULL;
2952
2953         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2954         if (!eli)
2955                 return -ENOMEM;
2956
2957         INIT_LIST_HEAD(&eli->li_request_list);
2958         mutex_init(&eli->li_list_mtx);
2959
2960         eli->li_state |= EXT4_LAZYINIT_QUIT;
2961
2962         ext4_li_info = eli;
2963
2964         return 0;
2965 }
2966
2967 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2968                                             ext4_group_t start)
2969 {
2970         struct ext4_sb_info *sbi = EXT4_SB(sb);
2971         struct ext4_li_request *elr;
2972         unsigned long rnd;
2973
2974         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2975         if (!elr)
2976                 return NULL;
2977
2978         elr->lr_super = sb;
2979         elr->lr_sbi = sbi;
2980         elr->lr_next_group = start;
2981
2982         /*
2983          * Randomize first schedule time of the request to
2984          * spread the inode table initialization requests
2985          * better.
2986          */
2987         get_random_bytes(&rnd, sizeof(rnd));
2988         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2989                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2990
2991         return elr;
2992 }
2993
2994 static int ext4_register_li_request(struct super_block *sb,
2995                                     ext4_group_t first_not_zeroed)
2996 {
2997         struct ext4_sb_info *sbi = EXT4_SB(sb);
2998         struct ext4_li_request *elr;
2999         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3000         int ret = 0;
3001
3002         if (sbi->s_li_request != NULL) {
3003                 /*
3004                  * Reset timeout so it can be computed again, because
3005                  * s_li_wait_mult might have changed.
3006                  */
3007                 sbi->s_li_request->lr_timeout = 0;
3008                 return 0;
3009         }
3010
3011         if (first_not_zeroed == ngroups ||
3012             (sb->s_flags & MS_RDONLY) ||
3013             !test_opt(sb, INIT_INODE_TABLE))
3014                 return 0;
3015
3016         elr = ext4_li_request_new(sb, first_not_zeroed);
3017         if (!elr)
3018                 return -ENOMEM;
3019
3020         mutex_lock(&ext4_li_mtx);
3021
3022         if (NULL == ext4_li_info) {
3023                 ret = ext4_li_info_new();
3024                 if (ret)
3025                         goto out;
3026         }
3027
3028         mutex_lock(&ext4_li_info->li_list_mtx);
3029         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3030         mutex_unlock(&ext4_li_info->li_list_mtx);
3031
3032         sbi->s_li_request = elr;
3033         /*
3034          * set elr to NULL here since it has been inserted to
3035          * the request_list and the removal and free of it is
3036          * handled by ext4_clear_request_list from now on.
3037          */
3038         elr = NULL;
3039
3040         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3041                 ret = ext4_run_lazyinit_thread();
3042                 if (ret)
3043                         goto out;
3044         }
3045 out:
3046         mutex_unlock(&ext4_li_mtx);
3047         if (ret)
3048                 kfree(elr);
3049         return ret;
3050 }
3051
3052 /*
3053  * We do not need to lock anything since this is called on
3054  * module unload.
3055  */
3056 static void ext4_destroy_lazyinit_thread(void)
3057 {
3058         /*
3059          * If thread exited earlier
3060          * there's nothing to be done.
3061          */
3062         if (!ext4_li_info || !ext4_lazyinit_task)
3063                 return;
3064
3065         kthread_stop(ext4_lazyinit_task);
3066 }
3067
3068 static int set_journal_csum_feature_set(struct super_block *sb)
3069 {
3070         int ret = 1;
3071         int compat, incompat;
3072         struct ext4_sb_info *sbi = EXT4_SB(sb);
3073
3074         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3075                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3076                 /* journal checksum v2 */
3077                 compat = 0;
3078                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3079         } else {
3080                 /* journal checksum v1 */
3081                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3082                 incompat = 0;
3083         }
3084
3085         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3086                 ret = jbd2_journal_set_features(sbi->s_journal,
3087                                 compat, 0,
3088                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3089                                 incompat);
3090         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3091                 ret = jbd2_journal_set_features(sbi->s_journal,
3092                                 compat, 0,
3093                                 incompat);
3094                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3095                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3096         } else {
3097                 jbd2_journal_clear_features(sbi->s_journal,
3098                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3099                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3100                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3101         }
3102
3103         return ret;
3104 }
3105
3106 /*
3107  * Note: calculating the overhead so we can be compatible with
3108  * historical BSD practice is quite difficult in the face of
3109  * clusters/bigalloc.  This is because multiple metadata blocks from
3110  * different block group can end up in the same allocation cluster.
3111  * Calculating the exact overhead in the face of clustered allocation
3112  * requires either O(all block bitmaps) in memory or O(number of block
3113  * groups**2) in time.  We will still calculate the superblock for
3114  * older file systems --- and if we come across with a bigalloc file
3115  * system with zero in s_overhead_clusters the estimate will be close to
3116  * correct especially for very large cluster sizes --- but for newer
3117  * file systems, it's better to calculate this figure once at mkfs
3118  * time, and store it in the superblock.  If the superblock value is
3119  * present (even for non-bigalloc file systems), we will use it.
3120  */
3121 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3122                           char *buf)
3123 {
3124         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3125         struct ext4_group_desc  *gdp;
3126         ext4_fsblk_t            first_block, last_block, b;
3127         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3128         int                     s, j, count = 0;
3129
3130         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3131                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3132                         sbi->s_itb_per_group + 2);
3133
3134         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3135                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3136         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3137         for (i = 0; i < ngroups; i++) {
3138                 gdp = ext4_get_group_desc(sb, i, NULL);
3139                 b = ext4_block_bitmap(sb, gdp);
3140                 if (b >= first_block && b <= last_block) {
3141                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3142                         count++;
3143                 }
3144                 b = ext4_inode_bitmap(sb, gdp);
3145                 if (b >= first_block && b <= last_block) {
3146                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3147                         count++;
3148                 }
3149                 b = ext4_inode_table(sb, gdp);
3150                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3151                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3152                                 int c = EXT4_B2C(sbi, b - first_block);
3153                                 ext4_set_bit(c, buf);
3154                                 count++;
3155                         }
3156                 if (i != grp)
3157                         continue;
3158                 s = 0;
3159                 if (ext4_bg_has_super(sb, grp)) {
3160                         ext4_set_bit(s++, buf);
3161                         count++;
3162                 }
3163                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3164                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3165                         count++;
3166                 }
3167         }
3168         if (!count)
3169                 return 0;
3170         return EXT4_CLUSTERS_PER_GROUP(sb) -
3171                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3172 }
3173
3174 /*
3175  * Compute the overhead and stash it in sbi->s_overhead
3176  */
3177 int ext4_calculate_overhead(struct super_block *sb)
3178 {
3179         struct ext4_sb_info *sbi = EXT4_SB(sb);
3180         struct ext4_super_block *es = sbi->s_es;
3181         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3182         ext4_fsblk_t overhead = 0;
3183         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3184
3185         memset(buf, 0, PAGE_SIZE);
3186         if (!buf)
3187                 return -ENOMEM;
3188
3189         /*
3190          * Compute the overhead (FS structures).  This is constant
3191          * for a given filesystem unless the number of block groups
3192          * changes so we cache the previous value until it does.
3193          */
3194
3195         /*
3196          * All of the blocks before first_data_block are overhead
3197          */
3198         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3199
3200         /*
3201          * Add the overhead found in each block group
3202          */
3203         for (i = 0; i < ngroups; i++) {
3204                 int blks;
3205
3206                 blks = count_overhead(sb, i, buf);
3207                 overhead += blks;
3208                 if (blks)
3209                         memset(buf, 0, PAGE_SIZE);
3210                 cond_resched();
3211         }
3212         sbi->s_overhead = overhead;
3213         smp_wmb();
3214         free_page((unsigned long) buf);
3215         return 0;
3216 }
3217
3218 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3219 {
3220         char *orig_data = kstrdup(data, GFP_KERNEL);
3221         struct buffer_head *bh;
3222         struct ext4_super_block *es = NULL;
3223         struct ext4_sb_info *sbi;
3224         ext4_fsblk_t block;
3225         ext4_fsblk_t sb_block = get_sb_block(&data);
3226         ext4_fsblk_t logical_sb_block;
3227         unsigned long offset = 0;
3228         unsigned long journal_devnum = 0;
3229         unsigned long def_mount_opts;
3230         struct inode *root;
3231         char *cp;
3232         const char *descr;
3233         int ret = -ENOMEM;
3234         int blocksize, clustersize;
3235         unsigned int db_count;
3236         unsigned int i;
3237         int needs_recovery, has_huge_files, has_bigalloc;
3238         __u64 blocks_count;
3239         int err;
3240         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3241         ext4_group_t first_not_zeroed;
3242
3243         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3244         if (!sbi)
3245                 goto out_free_orig;
3246
3247         sbi->s_blockgroup_lock =
3248                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3249         if (!sbi->s_blockgroup_lock) {
3250                 kfree(sbi);
3251                 goto out_free_orig;
3252         }
3253         sb->s_fs_info = sbi;
3254         sbi->s_sb = sb;
3255         sbi->s_mount_opt = 0;
3256         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3257         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3258         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3259         sbi->s_sb_block = sb_block;
3260         if (sb->s_bdev->bd_part)
3261                 sbi->s_sectors_written_start =
3262                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3263
3264         /* Cleanup superblock name */
3265         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3266                 *cp = '!';
3267
3268         ret = -EINVAL;
3269         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3270         if (!blocksize) {
3271                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3272                 goto out_fail;
3273         }
3274
3275         /*
3276          * The ext4 superblock will not be buffer aligned for other than 1kB
3277          * block sizes.  We need to calculate the offset from buffer start.
3278          */
3279         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3280                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3281                 offset = do_div(logical_sb_block, blocksize);
3282         } else {
3283                 logical_sb_block = sb_block;
3284         }
3285
3286         if (!(bh = sb_bread(sb, logical_sb_block))) {
3287                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3288                 goto out_fail;
3289         }
3290         /*
3291          * Note: s_es must be initialized as soon as possible because
3292          *       some ext4 macro-instructions depend on its value
3293          */
3294         es = (struct ext4_super_block *) (bh->b_data + offset);
3295         sbi->s_es = es;
3296         sb->s_magic = le16_to_cpu(es->s_magic);
3297         if (sb->s_magic != EXT4_SUPER_MAGIC)
3298                 goto cantfind_ext4;
3299         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3300
3301         /* Warn if metadata_csum and gdt_csum are both set. */
3302         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3303                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3304             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3305                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3306                              "redundant flags; please run fsck.");
3307
3308         /* Check for a known checksum algorithm */
3309         if (!ext4_verify_csum_type(sb, es)) {
3310                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3311                          "unknown checksum algorithm.");
3312                 silent = 1;
3313                 goto cantfind_ext4;
3314         }
3315
3316         /* Load the checksum driver */
3317         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3318                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3319                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3320                 if (IS_ERR(sbi->s_chksum_driver)) {
3321                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3322                         ret = PTR_ERR(sbi->s_chksum_driver);
3323                         sbi->s_chksum_driver = NULL;
3324                         goto failed_mount;
3325                 }
3326         }
3327
3328         /* Check superblock checksum */
3329         if (!ext4_superblock_csum_verify(sb, es)) {
3330                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3331                          "invalid superblock checksum.  Run e2fsck?");
3332                 silent = 1;
3333                 goto cantfind_ext4;
3334         }
3335
3336         /* Precompute checksum seed for all metadata */
3337         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3338                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3339                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3340                                                sizeof(es->s_uuid));
3341
3342         /* Set defaults before we parse the mount options */
3343         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3344         set_opt(sb, INIT_INODE_TABLE);
3345         if (def_mount_opts & EXT4_DEFM_DEBUG)
3346                 set_opt(sb, DEBUG);
3347         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3348                 set_opt(sb, GRPID);
3349         if (def_mount_opts & EXT4_DEFM_UID16)
3350                 set_opt(sb, NO_UID32);
3351         /* xattr user namespace & acls are now defaulted on */
3352 #ifdef CONFIG_EXT4_FS_XATTR
3353         set_opt(sb, XATTR_USER);
3354 #endif
3355 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3356         set_opt(sb, POSIX_ACL);
3357 #endif
3358         set_opt(sb, MBLK_IO_SUBMIT);
3359         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3360                 set_opt(sb, JOURNAL_DATA);
3361         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3362                 set_opt(sb, ORDERED_DATA);
3363         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3364                 set_opt(sb, WRITEBACK_DATA);
3365
3366         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3367                 set_opt(sb, ERRORS_PANIC);
3368         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3369                 set_opt(sb, ERRORS_CONT);
3370         else
3371                 set_opt(sb, ERRORS_RO);
3372         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3373                 set_opt(sb, BLOCK_VALIDITY);
3374         if (def_mount_opts & EXT4_DEFM_DISCARD)
3375                 set_opt(sb, DISCARD);
3376
3377         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3378         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3379         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3380         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3381         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3382
3383         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3384                 set_opt(sb, BARRIER);
3385
3386         /*
3387          * enable delayed allocation by default
3388          * Use -o nodelalloc to turn it off
3389          */
3390         if (!IS_EXT3_SB(sb) &&
3391             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3392                 set_opt(sb, DELALLOC);
3393
3394         /*
3395          * set default s_li_wait_mult for lazyinit, for the case there is
3396          * no mount option specified.
3397          */
3398         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3399
3400         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3401                            &journal_devnum, &journal_ioprio, 0)) {
3402                 ext4_msg(sb, KERN_WARNING,
3403                          "failed to parse options in superblock: %s",
3404                          sbi->s_es->s_mount_opts);
3405         }
3406         sbi->s_def_mount_opt = sbi->s_mount_opt;
3407         if (!parse_options((char *) data, sb, &journal_devnum,
3408                            &journal_ioprio, 0))
3409                 goto failed_mount;
3410
3411         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3412                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3413                             "with data=journal disables delayed "
3414                             "allocation and O_DIRECT support!\n");
3415                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3416                         ext4_msg(sb, KERN_ERR, "can't mount with "
3417                                  "both data=journal and delalloc");
3418                         goto failed_mount;
3419                 }
3420                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3421                         ext4_msg(sb, KERN_ERR, "can't mount with "
3422                                  "both data=journal and delalloc");
3423                         goto failed_mount;
3424                 }
3425                 if (test_opt(sb, DELALLOC))
3426                         clear_opt(sb, DELALLOC);
3427         }
3428
3429         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3430         if (test_opt(sb, DIOREAD_NOLOCK)) {
3431                 if (blocksize < PAGE_SIZE) {
3432                         ext4_msg(sb, KERN_ERR, "can't mount with "
3433                                  "dioread_nolock if block size != PAGE_SIZE");
3434                         goto failed_mount;
3435                 }
3436         }
3437
3438         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3439                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3440
3441         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3442             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3443              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3444              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3445                 ext4_msg(sb, KERN_WARNING,
3446                        "feature flags set on rev 0 fs, "
3447                        "running e2fsck is recommended");
3448
3449         if (IS_EXT2_SB(sb)) {
3450                 if (ext2_feature_set_ok(sb))
3451                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3452                                  "using the ext4 subsystem");
3453                 else {
3454                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3455                                  "to feature incompatibilities");
3456                         goto failed_mount;
3457                 }
3458         }
3459
3460         if (IS_EXT3_SB(sb)) {
3461                 if (ext3_feature_set_ok(sb))
3462                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3463                                  "using the ext4 subsystem");
3464                 else {
3465                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3466                                  "to feature incompatibilities");
3467                         goto failed_mount;
3468                 }
3469         }
3470
3471         /*
3472          * Check feature flags regardless of the revision level, since we
3473          * previously didn't change the revision level when setting the flags,
3474          * so there is a chance incompat flags are set on a rev 0 filesystem.
3475          */
3476         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3477                 goto failed_mount;
3478
3479         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3480             blocksize > EXT4_MAX_BLOCK_SIZE) {
3481                 ext4_msg(sb, KERN_ERR,
3482                        "Unsupported filesystem blocksize %d", blocksize);
3483                 goto failed_mount;
3484         }
3485
3486         if (sb->s_blocksize != blocksize) {
3487                 /* Validate the filesystem blocksize */
3488                 if (!sb_set_blocksize(sb, blocksize)) {
3489                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3490                                         blocksize);
3491                         goto failed_mount;
3492                 }
3493
3494                 brelse(bh);
3495                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3496                 offset = do_div(logical_sb_block, blocksize);
3497                 bh = sb_bread(sb, logical_sb_block);
3498                 if (!bh) {
3499                         ext4_msg(sb, KERN_ERR,
3500                                "Can't read superblock on 2nd try");
3501                         goto failed_mount;
3502                 }
3503                 es = (struct ext4_super_block *)(bh->b_data + offset);
3504                 sbi->s_es = es;
3505                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3506                         ext4_msg(sb, KERN_ERR,
3507                                "Magic mismatch, very weird!");
3508                         goto failed_mount;
3509                 }
3510         }
3511
3512         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3513                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3514         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3515                                                       has_huge_files);
3516         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3517
3518         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3519                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3520                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3521         } else {
3522                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3523                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3524                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3525                     (!is_power_of_2(sbi->s_inode_size)) ||
3526                     (sbi->s_inode_size > blocksize)) {
3527                         ext4_msg(sb, KERN_ERR,
3528                                "unsupported inode size: %d",
3529                                sbi->s_inode_size);
3530                         goto failed_mount;
3531                 }
3532                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3533                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3534         }
3535
3536         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3538                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3539                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3540                     !is_power_of_2(sbi->s_desc_size)) {
3541                         ext4_msg(sb, KERN_ERR,
3542                                "unsupported descriptor size %lu",
3543                                sbi->s_desc_size);
3544                         goto failed_mount;
3545                 }
3546         } else
3547                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3548
3549         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3550         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3551         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3552                 goto cantfind_ext4;
3553
3554         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3555         if (sbi->s_inodes_per_block == 0)
3556                 goto cantfind_ext4;
3557         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3558                                         sbi->s_inodes_per_block;
3559         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3560         sbi->s_sbh = bh;
3561         sbi->s_mount_state = le16_to_cpu(es->s_state);
3562         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3563         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3564
3565         for (i = 0; i < 4; i++)
3566                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3567         sbi->s_def_hash_version = es->s_def_hash_version;
3568         i = le32_to_cpu(es->s_flags);
3569         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3570                 sbi->s_hash_unsigned = 3;
3571         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3572 #ifdef __CHAR_UNSIGNED__
3573                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3574                 sbi->s_hash_unsigned = 3;
3575 #else
3576                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3577 #endif
3578         }
3579
3580         /* Handle clustersize */
3581         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3582         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3583                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3584         if (has_bigalloc) {
3585                 if (clustersize < blocksize) {
3586                         ext4_msg(sb, KERN_ERR,
3587                                  "cluster size (%d) smaller than "
3588                                  "block size (%d)", clustersize, blocksize);
3589                         goto failed_mount;
3590                 }
3591                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3592                         le32_to_cpu(es->s_log_block_size);
3593                 sbi->s_clusters_per_group =
3594                         le32_to_cpu(es->s_clusters_per_group);
3595                 if (sbi->s_clusters_per_group > blocksize * 8) {
3596                         ext4_msg(sb, KERN_ERR,
3597                                  "#clusters per group too big: %lu",
3598                                  sbi->s_clusters_per_group);
3599                         goto failed_mount;
3600                 }
3601                 if (sbi->s_blocks_per_group !=
3602                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3603                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3604                                  "clusters per group (%lu) inconsistent",
3605                                  sbi->s_blocks_per_group,
3606                                  sbi->s_clusters_per_group);
3607                         goto failed_mount;
3608                 }
3609         } else {
3610                 if (clustersize != blocksize) {
3611                         ext4_warning(sb, "fragment/cluster size (%d) != "
3612                                      "block size (%d)", clustersize,
3613                                      blocksize);
3614                         clustersize = blocksize;
3615                 }
3616                 if (sbi->s_blocks_per_group > blocksize * 8) {
3617                         ext4_msg(sb, KERN_ERR,
3618                                  "#blocks per group too big: %lu",
3619                                  sbi->s_blocks_per_group);
3620                         goto failed_mount;
3621                 }
3622                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3623                 sbi->s_cluster_bits = 0;
3624         }
3625         sbi->s_cluster_ratio = clustersize / blocksize;
3626
3627         if (sbi->s_inodes_per_group > blocksize * 8) {
3628                 ext4_msg(sb, KERN_ERR,
3629                        "#inodes per group too big: %lu",
3630                        sbi->s_inodes_per_group);
3631                 goto failed_mount;
3632         }
3633
3634         /*
3635          * Test whether we have more sectors than will fit in sector_t,
3636          * and whether the max offset is addressable by the page cache.
3637          */
3638         err = generic_check_addressable(sb->s_blocksize_bits,
3639                                         ext4_blocks_count(es));
3640         if (err) {
3641                 ext4_msg(sb, KERN_ERR, "filesystem"
3642                          " too large to mount safely on this system");
3643                 if (sizeof(sector_t) < 8)
3644                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3645                 ret = err;
3646                 goto failed_mount;
3647         }
3648
3649         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3650                 goto cantfind_ext4;
3651
3652         /* check blocks count against device size */
3653         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3654         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3655                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3656                        "exceeds size of device (%llu blocks)",
3657                        ext4_blocks_count(es), blocks_count);
3658                 goto failed_mount;
3659         }
3660
3661         /*
3662          * It makes no sense for the first data block to be beyond the end
3663          * of the filesystem.
3664          */
3665         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3666                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3667                          "block %u is beyond end of filesystem (%llu)",
3668                          le32_to_cpu(es->s_first_data_block),
3669                          ext4_blocks_count(es));
3670                 goto failed_mount;
3671         }
3672         blocks_count = (ext4_blocks_count(es) -
3673                         le32_to_cpu(es->s_first_data_block) +
3674                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3675         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3676         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3677                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3678                        "(block count %llu, first data block %u, "
3679                        "blocks per group %lu)", sbi->s_groups_count,
3680                        ext4_blocks_count(es),
3681                        le32_to_cpu(es->s_first_data_block),
3682                        EXT4_BLOCKS_PER_GROUP(sb));
3683                 goto failed_mount;
3684         }
3685         sbi->s_groups_count = blocks_count;
3686         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3687                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3688         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3689                    EXT4_DESC_PER_BLOCK(sb);
3690         sbi->s_group_desc = ext4_kvmalloc(db_count *
3691                                           sizeof(struct buffer_head *),
3692                                           GFP_KERNEL);
3693         if (sbi->s_group_desc == NULL) {
3694                 ext4_msg(sb, KERN_ERR, "not enough memory");
3695                 ret = -ENOMEM;
3696                 goto failed_mount;
3697         }
3698
3699         if (ext4_proc_root)
3700                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3701
3702         if (sbi->s_proc)
3703                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3704                                  &ext4_seq_options_fops, sb);
3705
3706         bgl_lock_init(sbi->s_blockgroup_lock);
3707
3708         for (i = 0; i < db_count; i++) {
3709                 block = descriptor_loc(sb, logical_sb_block, i);
3710                 sbi->s_group_desc[i] = sb_bread(sb, block);
3711                 if (!sbi->s_group_desc[i]) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                "can't read group descriptor %d", i);
3714                         db_count = i;
3715                         goto failed_mount2;
3716                 }
3717         }
3718         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3719                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3720                 goto failed_mount2;
3721         }
3722         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3723                 if (!ext4_fill_flex_info(sb)) {
3724                         ext4_msg(sb, KERN_ERR,
3725                                "unable to initialize "
3726                                "flex_bg meta info!");
3727                         goto failed_mount2;
3728                 }
3729
3730         sbi->s_gdb_count = db_count;
3731         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3732         spin_lock_init(&sbi->s_next_gen_lock);
3733
3734         init_timer(&sbi->s_err_report);
3735         sbi->s_err_report.function = print_daily_error_info;
3736         sbi->s_err_report.data = (unsigned long) sb;
3737
3738         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3739                         ext4_count_free_clusters(sb));
3740         if (!err) {
3741                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3742                                 ext4_count_free_inodes(sb));
3743         }
3744         if (!err) {
3745                 err = percpu_counter_init(&sbi->s_dirs_counter,
3746                                 ext4_count_dirs(sb));
3747         }
3748         if (!err) {
3749                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3750         }
3751         if (err) {
3752                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3753                 ret = err;
3754                 goto failed_mount3;
3755         }
3756
3757         sbi->s_stripe = ext4_get_stripe_size(sbi);
3758         sbi->s_max_writeback_mb_bump = 128;
3759
3760         /*
3761          * set up enough so that it can read an inode
3762          */
3763         if (!test_opt(sb, NOLOAD) &&
3764             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3765                 sb->s_op = &ext4_sops;
3766         else
3767                 sb->s_op = &ext4_nojournal_sops;
3768         sb->s_export_op = &ext4_export_ops;
3769         sb->s_xattr = ext4_xattr_handlers;
3770 #ifdef CONFIG_QUOTA
3771         sb->s_qcop = &ext4_qctl_operations;
3772         sb->dq_op = &ext4_quota_operations;
3773
3774         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3775                 /* Use qctl operations for hidden quota files. */
3776                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3777         }
3778 #endif
3779         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3780
3781         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3782         mutex_init(&sbi->s_orphan_lock);
3783         sbi->s_resize_flags = 0;
3784
3785         sb->s_root = NULL;
3786
3787         needs_recovery = (es->s_last_orphan != 0 ||
3788                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3789                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3790
3791         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3792             !(sb->s_flags & MS_RDONLY))
3793                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3794                         goto failed_mount3;
3795
3796         /*
3797          * The first inode we look at is the journal inode.  Don't try
3798          * root first: it may be modified in the journal!
3799          */
3800         if (!test_opt(sb, NOLOAD) &&
3801             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3802                 if (ext4_load_journal(sb, es, journal_devnum))
3803                         goto failed_mount3;
3804         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3805               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3806                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3807                        "suppressed and not mounted read-only");
3808                 goto failed_mount_wq;
3809         } else {
3810                 clear_opt(sb, DATA_FLAGS);
3811                 sbi->s_journal = NULL;
3812                 needs_recovery = 0;
3813                 goto no_journal;
3814         }
3815
3816         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3817             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3818                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3819                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3820                 goto failed_mount_wq;
3821         }
3822
3823         if (!set_journal_csum_feature_set(sb)) {
3824                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3825                          "feature set");
3826                 goto failed_mount_wq;
3827         }
3828
3829         /* We have now updated the journal if required, so we can
3830          * validate the data journaling mode. */
3831         switch (test_opt(sb, DATA_FLAGS)) {
3832         case 0:
3833                 /* No mode set, assume a default based on the journal
3834                  * capabilities: ORDERED_DATA if the journal can
3835                  * cope, else JOURNAL_DATA
3836                  */
3837                 if (jbd2_journal_check_available_features
3838                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3839                         set_opt(sb, ORDERED_DATA);
3840                 else
3841                         set_opt(sb, JOURNAL_DATA);
3842                 break;
3843
3844         case EXT4_MOUNT_ORDERED_DATA:
3845         case EXT4_MOUNT_WRITEBACK_DATA:
3846                 if (!jbd2_journal_check_available_features
3847                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3848                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3849                                "requested data journaling mode");
3850                         goto failed_mount_wq;
3851                 }
3852         default:
3853                 break;
3854         }
3855         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3856
3857         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3858
3859         /*
3860          * The journal may have updated the bg summary counts, so we
3861          * need to update the global counters.
3862          */
3863         percpu_counter_set(&sbi->s_freeclusters_counter,
3864                            ext4_count_free_clusters(sb));
3865         percpu_counter_set(&sbi->s_freeinodes_counter,
3866                            ext4_count_free_inodes(sb));
3867         percpu_counter_set(&sbi->s_dirs_counter,
3868                            ext4_count_dirs(sb));
3869         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3870
3871 no_journal:
3872         /*
3873          * Get the # of file system overhead blocks from the
3874          * superblock if present.
3875          */
3876         if (es->s_overhead_clusters)
3877                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3878         else {
3879                 ret = ext4_calculate_overhead(sb);
3880                 if (ret)
3881                         goto failed_mount_wq;
3882         }
3883
3884         /*
3885          * The maximum number of concurrent works can be high and
3886          * concurrency isn't really necessary.  Limit it to 1.
3887          */
3888         EXT4_SB(sb)->dio_unwritten_wq =
3889                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3890         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3891                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3892                 goto failed_mount_wq;
3893         }
3894
3895         /*
3896          * The jbd2_journal_load will have done any necessary log recovery,
3897          * so we can safely mount the rest of the filesystem now.
3898          */
3899
3900         root = ext4_iget(sb, EXT4_ROOT_INO);
3901         if (IS_ERR(root)) {
3902                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3903                 ret = PTR_ERR(root);
3904                 root = NULL;
3905                 goto failed_mount4;
3906         }
3907         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3908                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3909                 iput(root);
3910                 goto failed_mount4;
3911         }
3912         sb->s_root = d_make_root(root);
3913         if (!sb->s_root) {
3914                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3915                 ret = -ENOMEM;
3916                 goto failed_mount4;
3917         }
3918
3919         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3920                 sb->s_flags |= MS_RDONLY;
3921
3922         /* determine the minimum size of new large inodes, if present */
3923         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3924                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3925                                                      EXT4_GOOD_OLD_INODE_SIZE;
3926                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3927                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3928                         if (sbi->s_want_extra_isize <
3929                             le16_to_cpu(es->s_want_extra_isize))
3930                                 sbi->s_want_extra_isize =
3931                                         le16_to_cpu(es->s_want_extra_isize);
3932                         if (sbi->s_want_extra_isize <
3933                             le16_to_cpu(es->s_min_extra_isize))
3934                                 sbi->s_want_extra_isize =
3935                                         le16_to_cpu(es->s_min_extra_isize);
3936                 }
3937         }
3938         /* Check if enough inode space is available */
3939         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3940                                                         sbi->s_inode_size) {
3941                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3942                                                        EXT4_GOOD_OLD_INODE_SIZE;
3943                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3944                          "available");
3945         }
3946
3947         err = ext4_setup_system_zone(sb);
3948         if (err) {
3949                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3950                          "zone (%d)", err);
3951                 goto failed_mount4a;
3952         }
3953
3954         ext4_ext_init(sb);
3955         err = ext4_mb_init(sb);
3956         if (err) {
3957                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3958                          err);
3959                 goto failed_mount5;
3960         }
3961
3962         err = ext4_register_li_request(sb, first_not_zeroed);
3963         if (err)
3964                 goto failed_mount6;
3965
3966         sbi->s_kobj.kset = ext4_kset;
3967         init_completion(&sbi->s_kobj_unregister);
3968         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3969                                    "%s", sb->s_id);
3970         if (err)
3971                 goto failed_mount7;
3972
3973         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3974         ext4_orphan_cleanup(sb, es);
3975         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3976         if (needs_recovery) {
3977                 ext4_msg(sb, KERN_INFO, "recovery complete");
3978                 ext4_mark_recovery_complete(sb, es);
3979         }
3980         if (EXT4_SB(sb)->s_journal) {
3981                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3982                         descr = " journalled data mode";
3983                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3984                         descr = " ordered data mode";
3985                 else
3986                         descr = " writeback data mode";
3987         } else
3988                 descr = "out journal";
3989
3990 #ifdef CONFIG_QUOTA
3991         /* Enable quota usage during mount. */
3992         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
3993             !(sb->s_flags & MS_RDONLY)) {
3994                 ret = ext4_enable_quotas(sb);
3995                 if (ret)
3996                         goto failed_mount7;
3997         }
3998 #endif  /* CONFIG_QUOTA */
3999
4000         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4001                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4002                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4003
4004         if (es->s_error_count)
4005                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4006
4007         kfree(orig_data);
4008         return 0;
4009
4010 cantfind_ext4:
4011         if (!silent)
4012                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4013         goto failed_mount;
4014
4015 failed_mount7:
4016         ext4_unregister_li_request(sb);
4017 failed_mount6:
4018         ext4_mb_release(sb);
4019 failed_mount5:
4020         ext4_ext_release(sb);
4021         ext4_release_system_zone(sb);
4022 failed_mount4a:
4023         dput(sb->s_root);
4024         sb->s_root = NULL;
4025 failed_mount4:
4026         ext4_msg(sb, KERN_ERR, "mount failed");
4027         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4028 failed_mount_wq:
4029         if (sbi->s_journal) {
4030                 jbd2_journal_destroy(sbi->s_journal);
4031                 sbi->s_journal = NULL;
4032         }
4033 failed_mount3:
4034         del_timer(&sbi->s_err_report);
4035         if (sbi->s_flex_groups)
4036                 ext4_kvfree(sbi->s_flex_groups);
4037         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4038         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4039         percpu_counter_destroy(&sbi->s_dirs_counter);
4040         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4041         if (sbi->s_mmp_tsk)
4042                 kthread_stop(sbi->s_mmp_tsk);
4043 failed_mount2:
4044         for (i = 0; i < db_count; i++)
4045                 brelse(sbi->s_group_desc[i]);
4046         ext4_kvfree(sbi->s_group_desc);
4047 failed_mount:
4048         if (sbi->s_chksum_driver)
4049                 crypto_free_shash(sbi->s_chksum_driver);
4050         if (sbi->s_proc) {
4051                 remove_proc_entry("options", sbi->s_proc);
4052                 remove_proc_entry(sb->s_id, ext4_proc_root);
4053         }
4054 #ifdef CONFIG_QUOTA
4055         for (i = 0; i < MAXQUOTAS; i++)
4056                 kfree(sbi->s_qf_names[i]);
4057 #endif
4058         ext4_blkdev_remove(sbi);
4059         brelse(bh);
4060 out_fail:
4061         sb->s_fs_info = NULL;
4062         kfree(sbi->s_blockgroup_lock);
4063         kfree(sbi);
4064 out_free_orig:
4065         kfree(orig_data);
4066         return ret;
4067 }
4068
4069 /*
4070  * Setup any per-fs journal parameters now.  We'll do this both on
4071  * initial mount, once the journal has been initialised but before we've
4072  * done any recovery; and again on any subsequent remount.
4073  */
4074 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4075 {
4076         struct ext4_sb_info *sbi = EXT4_SB(sb);
4077
4078         journal->j_commit_interval = sbi->s_commit_interval;
4079         journal->j_min_batch_time = sbi->s_min_batch_time;
4080         journal->j_max_batch_time = sbi->s_max_batch_time;
4081
4082         write_lock(&journal->j_state_lock);
4083         if (test_opt(sb, BARRIER))
4084                 journal->j_flags |= JBD2_BARRIER;
4085         else
4086                 journal->j_flags &= ~JBD2_BARRIER;
4087         if (test_opt(sb, DATA_ERR_ABORT))
4088                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4089         else
4090                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4091         write_unlock(&journal->j_state_lock);
4092 }
4093
4094 static journal_t *ext4_get_journal(struct super_block *sb,
4095                                    unsigned int journal_inum)
4096 {
4097         struct inode *journal_inode;
4098         journal_t *journal;
4099
4100         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4101
4102         /* First, test for the existence of a valid inode on disk.  Bad
4103          * things happen if we iget() an unused inode, as the subsequent
4104          * iput() will try to delete it. */
4105
4106         journal_inode = ext4_iget(sb, journal_inum);
4107         if (IS_ERR(journal_inode)) {
4108                 ext4_msg(sb, KERN_ERR, "no journal found");
4109                 return NULL;
4110         }
4111         if (!journal_inode->i_nlink) {
4112                 make_bad_inode(journal_inode);
4113                 iput(journal_inode);
4114                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4115                 return NULL;
4116         }
4117
4118         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4119                   journal_inode, journal_inode->i_size);
4120         if (!S_ISREG(journal_inode->i_mode)) {
4121                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4122                 iput(journal_inode);
4123                 return NULL;
4124         }
4125
4126         journal = jbd2_journal_init_inode(journal_inode);
4127         if (!journal) {
4128                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4129                 iput(journal_inode);
4130                 return NULL;
4131         }
4132         journal->j_private = sb;
4133         ext4_init_journal_params(sb, journal);
4134         return journal;
4135 }
4136
4137 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4138                                        dev_t j_dev)
4139 {
4140         struct buffer_head *bh;
4141         journal_t *journal;
4142         ext4_fsblk_t start;
4143         ext4_fsblk_t len;
4144         int hblock, blocksize;
4145         ext4_fsblk_t sb_block;
4146         unsigned long offset;
4147         struct ext4_super_block *es;
4148         struct block_device *bdev;
4149
4150         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4151
4152         bdev = ext4_blkdev_get(j_dev, sb);
4153         if (bdev == NULL)
4154                 return NULL;
4155
4156         blocksize = sb->s_blocksize;
4157         hblock = bdev_logical_block_size(bdev);
4158         if (blocksize < hblock) {
4159                 ext4_msg(sb, KERN_ERR,
4160                         "blocksize too small for journal device");
4161                 goto out_bdev;
4162         }
4163
4164         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4165         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4166         set_blocksize(bdev, blocksize);
4167         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4168                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4169                        "external journal");
4170                 goto out_bdev;
4171         }
4172
4173         es = (struct ext4_super_block *) (bh->b_data + offset);
4174         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4175             !(le32_to_cpu(es->s_feature_incompat) &
4176               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4177                 ext4_msg(sb, KERN_ERR, "external journal has "
4178                                         "bad superblock");
4179                 brelse(bh);
4180                 goto out_bdev;
4181         }
4182
4183         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4184                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4185                 brelse(bh);
4186                 goto out_bdev;
4187         }
4188
4189         len = ext4_blocks_count(es);
4190         start = sb_block + 1;
4191         brelse(bh);     /* we're done with the superblock */
4192
4193         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4194                                         start, len, blocksize);
4195         if (!journal) {
4196                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4197                 goto out_bdev;
4198         }
4199         journal->j_private = sb;
4200         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4201         wait_on_buffer(journal->j_sb_buffer);
4202         if (!buffer_uptodate(journal->j_sb_buffer)) {
4203                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4204                 goto out_journal;
4205         }
4206         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4207                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4208                                         "user (unsupported) - %d",
4209                         be32_to_cpu(journal->j_superblock->s_nr_users));
4210                 goto out_journal;
4211         }
4212         EXT4_SB(sb)->journal_bdev = bdev;
4213         ext4_init_journal_params(sb, journal);
4214         return journal;
4215
4216 out_journal:
4217         jbd2_journal_destroy(journal);
4218 out_bdev:
4219         ext4_blkdev_put(bdev);
4220         return NULL;
4221 }
4222
4223 static int ext4_load_journal(struct super_block *sb,
4224                              struct ext4_super_block *es,
4225                              unsigned long journal_devnum)
4226 {
4227         journal_t *journal;
4228         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4229         dev_t journal_dev;
4230         int err = 0;
4231         int really_read_only;
4232
4233         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4234
4235         if (journal_devnum &&
4236             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4237                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4238                         "numbers have changed");
4239                 journal_dev = new_decode_dev(journal_devnum);
4240         } else
4241                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4242
4243         really_read_only = bdev_read_only(sb->s_bdev);
4244
4245         /*
4246          * Are we loading a blank journal or performing recovery after a
4247          * crash?  For recovery, we need to check in advance whether we
4248          * can get read-write access to the device.
4249          */
4250         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4251                 if (sb->s_flags & MS_RDONLY) {
4252                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4253                                         "required on readonly filesystem");
4254                         if (really_read_only) {
4255                                 ext4_msg(sb, KERN_ERR, "write access "
4256                                         "unavailable, cannot proceed");
4257                                 return -EROFS;
4258                         }
4259                         ext4_msg(sb, KERN_INFO, "write access will "
4260                                "be enabled during recovery");
4261                 }
4262         }
4263
4264         if (journal_inum && journal_dev) {
4265                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4266                        "and inode journals!");
4267                 return -EINVAL;
4268         }
4269
4270         if (journal_inum) {
4271                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4272                         return -EINVAL;
4273         } else {
4274                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4275                         return -EINVAL;
4276         }
4277
4278         if (!(journal->j_flags & JBD2_BARRIER))
4279                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4280
4281         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4282                 err = jbd2_journal_wipe(journal, !really_read_only);
4283         if (!err) {
4284                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4285                 if (save)
4286                         memcpy(save, ((char *) es) +
4287                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4288                 err = jbd2_journal_load(journal);
4289                 if (save)
4290                         memcpy(((char *) es) + EXT4_S_ERR_START,
4291                                save, EXT4_S_ERR_LEN);
4292                 kfree(save);
4293         }
4294
4295         if (err) {
4296                 ext4_msg(sb, KERN_ERR, "error loading journal");
4297                 jbd2_journal_destroy(journal);
4298                 return err;
4299         }
4300
4301         EXT4_SB(sb)->s_journal = journal;
4302         ext4_clear_journal_err(sb, es);
4303
4304         if (!really_read_only && journal_devnum &&
4305             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4306                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4307
4308                 /* Make sure we flush the recovery flag to disk. */
4309                 ext4_commit_super(sb, 1);
4310         }
4311
4312         return 0;
4313 }
4314
4315 static int ext4_commit_super(struct super_block *sb, int sync)
4316 {
4317         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4318         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4319         int error = 0;
4320
4321         if (!sbh || block_device_ejected(sb))
4322                 return error;
4323         if (buffer_write_io_error(sbh)) {
4324                 /*
4325                  * Oh, dear.  A previous attempt to write the
4326                  * superblock failed.  This could happen because the
4327                  * USB device was yanked out.  Or it could happen to
4328                  * be a transient write error and maybe the block will
4329                  * be remapped.  Nothing we can do but to retry the
4330                  * write and hope for the best.
4331                  */
4332                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4333                        "superblock detected");
4334                 clear_buffer_write_io_error(sbh);
4335                 set_buffer_uptodate(sbh);
4336         }
4337         /*
4338          * If the file system is mounted read-only, don't update the
4339          * superblock write time.  This avoids updating the superblock
4340          * write time when we are mounting the root file system
4341          * read/only but we need to replay the journal; at that point,
4342          * for people who are east of GMT and who make their clock
4343          * tick in localtime for Windows bug-for-bug compatibility,
4344          * the clock is set in the future, and this will cause e2fsck
4345          * to complain and force a full file system check.
4346          */
4347         if (!(sb->s_flags & MS_RDONLY))
4348                 es->s_wtime = cpu_to_le32(get_seconds());
4349         if (sb->s_bdev->bd_part)
4350                 es->s_kbytes_written =
4351                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4352                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4353                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4354         else
4355                 es->s_kbytes_written =
4356                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4357         ext4_free_blocks_count_set(es,
4358                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4359                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4360         es->s_free_inodes_count =
4361                 cpu_to_le32(percpu_counter_sum_positive(
4362                                 &EXT4_SB(sb)->s_freeinodes_counter));
4363         BUFFER_TRACE(sbh, "marking dirty");
4364         ext4_superblock_csum_set(sb, es);
4365         mark_buffer_dirty(sbh);
4366         if (sync) {
4367                 error = sync_dirty_buffer(sbh);
4368                 if (error)
4369                         return error;
4370
4371                 error = buffer_write_io_error(sbh);
4372                 if (error) {
4373                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4374                                "superblock");
4375                         clear_buffer_write_io_error(sbh);
4376                         set_buffer_uptodate(sbh);
4377                 }
4378         }
4379         return error;
4380 }
4381
4382 /*
4383  * Have we just finished recovery?  If so, and if we are mounting (or
4384  * remounting) the filesystem readonly, then we will end up with a
4385  * consistent fs on disk.  Record that fact.
4386  */
4387 static void ext4_mark_recovery_complete(struct super_block *sb,
4388                                         struct ext4_super_block *es)
4389 {
4390         journal_t *journal = EXT4_SB(sb)->s_journal;
4391
4392         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4393                 BUG_ON(journal != NULL);
4394                 return;
4395         }
4396         jbd2_journal_lock_updates(journal);
4397         if (jbd2_journal_flush(journal) < 0)
4398                 goto out;
4399
4400         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4401             sb->s_flags & MS_RDONLY) {
4402                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4403                 ext4_commit_super(sb, 1);
4404         }
4405
4406 out:
4407         jbd2_journal_unlock_updates(journal);
4408 }
4409
4410 /*
4411  * If we are mounting (or read-write remounting) a filesystem whose journal
4412  * has recorded an error from a previous lifetime, move that error to the
4413  * main filesystem now.
4414  */
4415 static void ext4_clear_journal_err(struct super_block *sb,
4416                                    struct ext4_super_block *es)
4417 {
4418         journal_t *journal;
4419         int j_errno;
4420         const char *errstr;
4421
4422         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4423
4424         journal = EXT4_SB(sb)->s_journal;
4425
4426         /*
4427          * Now check for any error status which may have been recorded in the
4428          * journal by a prior ext4_error() or ext4_abort()
4429          */
4430
4431         j_errno = jbd2_journal_errno(journal);
4432         if (j_errno) {
4433                 char nbuf[16];
4434
4435                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4436                 ext4_warning(sb, "Filesystem error recorded "
4437                              "from previous mount: %s", errstr);
4438                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4439
4440                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4441                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4442                 ext4_commit_super(sb, 1);
4443
4444                 jbd2_journal_clear_err(journal);
4445                 jbd2_journal_update_sb_errno(journal);
4446         }
4447 }
4448
4449 /*
4450  * Force the running and committing transactions to commit,
4451  * and wait on the commit.
4452  */
4453 int ext4_force_commit(struct super_block *sb)
4454 {
4455         journal_t *journal;
4456         int ret = 0;
4457
4458         if (sb->s_flags & MS_RDONLY)
4459                 return 0;
4460
4461         journal = EXT4_SB(sb)->s_journal;
4462         if (journal)
4463                 ret = ext4_journal_force_commit(journal);
4464
4465         return ret;
4466 }
4467
4468 static int ext4_sync_fs(struct super_block *sb, int wait)
4469 {
4470         int ret = 0;
4471         tid_t target;
4472         struct ext4_sb_info *sbi = EXT4_SB(sb);
4473
4474         trace_ext4_sync_fs(sb, wait);
4475         flush_workqueue(sbi->dio_unwritten_wq);
4476         /*
4477          * Writeback quota in non-journalled quota case - journalled quota has
4478          * no dirty dquots
4479          */
4480         dquot_writeback_dquots(sb, -1);
4481         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4482                 if (wait)
4483                         jbd2_log_wait_commit(sbi->s_journal, target);
4484         }
4485         return ret;
4486 }
4487
4488 /*
4489  * LVM calls this function before a (read-only) snapshot is created.  This
4490  * gives us a chance to flush the journal completely and mark the fs clean.
4491  *
4492  * Note that only this function cannot bring a filesystem to be in a clean
4493  * state independently. It relies on upper layer to stop all data & metadata
4494  * modifications.
4495  */
4496 static int ext4_freeze(struct super_block *sb)
4497 {
4498         int error = 0;
4499         journal_t *journal;
4500
4501         if (sb->s_flags & MS_RDONLY)
4502                 return 0;
4503
4504         journal = EXT4_SB(sb)->s_journal;
4505
4506         /* Now we set up the journal barrier. */
4507         jbd2_journal_lock_updates(journal);
4508
4509         /*
4510          * Don't clear the needs_recovery flag if we failed to flush
4511          * the journal.
4512          */
4513         error = jbd2_journal_flush(journal);
4514         if (error < 0)
4515                 goto out;
4516
4517         /* Journal blocked and flushed, clear needs_recovery flag. */
4518         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4519         error = ext4_commit_super(sb, 1);
4520 out:
4521         /* we rely on upper layer to stop further updates */
4522         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4523         return error;
4524 }
4525
4526 /*
4527  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4528  * flag here, even though the filesystem is not technically dirty yet.
4529  */
4530 static int ext4_unfreeze(struct super_block *sb)
4531 {
4532         if (sb->s_flags & MS_RDONLY)
4533                 return 0;
4534
4535         lock_super(sb);
4536         /* Reset the needs_recovery flag before the fs is unlocked. */
4537         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4538         ext4_commit_super(sb, 1);
4539         unlock_super(sb);
4540         return 0;
4541 }
4542
4543 /*
4544  * Structure to save mount options for ext4_remount's benefit
4545  */
4546 struct ext4_mount_options {
4547         unsigned long s_mount_opt;
4548         unsigned long s_mount_opt2;
4549         kuid_t s_resuid;
4550         kgid_t s_resgid;
4551         unsigned long s_commit_interval;
4552         u32 s_min_batch_time, s_max_batch_time;
4553 #ifdef CONFIG_QUOTA
4554         int s_jquota_fmt;
4555         char *s_qf_names[MAXQUOTAS];
4556 #endif
4557 };
4558
4559 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4560 {
4561         struct ext4_super_block *es;
4562         struct ext4_sb_info *sbi = EXT4_SB(sb);
4563         unsigned long old_sb_flags;
4564         struct ext4_mount_options old_opts;
4565         int enable_quota = 0;
4566         ext4_group_t g;
4567         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4568         int err = 0;
4569 #ifdef CONFIG_QUOTA
4570         int i;
4571 #endif
4572         char *orig_data = kstrdup(data, GFP_KERNEL);
4573
4574         /* Store the original options */
4575         lock_super(sb);
4576         old_sb_flags = sb->s_flags;
4577         old_opts.s_mount_opt = sbi->s_mount_opt;
4578         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4579         old_opts.s_resuid = sbi->s_resuid;
4580         old_opts.s_resgid = sbi->s_resgid;
4581         old_opts.s_commit_interval = sbi->s_commit_interval;
4582         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4583         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4584 #ifdef CONFIG_QUOTA
4585         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4586         for (i = 0; i < MAXQUOTAS; i++)
4587                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4588 #endif
4589         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4590                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4591
4592         /*
4593          * Allow the "check" option to be passed as a remount option.
4594          */
4595         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4596                 err = -EINVAL;
4597                 goto restore_opts;
4598         }
4599
4600         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4601                 ext4_abort(sb, "Abort forced by user");
4602
4603         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4604                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4605
4606         es = sbi->s_es;
4607
4608         if (sbi->s_journal) {
4609                 ext4_init_journal_params(sb, sbi->s_journal);
4610                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4611         }
4612
4613         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4614                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4615                         err = -EROFS;
4616                         goto restore_opts;
4617                 }
4618
4619                 if (*flags & MS_RDONLY) {
4620                         err = dquot_suspend(sb, -1);
4621                         if (err < 0)
4622                                 goto restore_opts;
4623
4624                         /*
4625                          * First of all, the unconditional stuff we have to do
4626                          * to disable replay of the journal when we next remount
4627                          */
4628                         sb->s_flags |= MS_RDONLY;
4629
4630                         /*
4631                          * OK, test if we are remounting a valid rw partition
4632                          * readonly, and if so set the rdonly flag and then
4633                          * mark the partition as valid again.
4634                          */
4635                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4636                             (sbi->s_mount_state & EXT4_VALID_FS))
4637                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4638
4639                         if (sbi->s_journal)
4640                                 ext4_mark_recovery_complete(sb, es);
4641                 } else {
4642                         /* Make sure we can mount this feature set readwrite */
4643                         if (!ext4_feature_set_ok(sb, 0)) {
4644                                 err = -EROFS;
4645                                 goto restore_opts;
4646                         }
4647                         /*
4648                          * Make sure the group descriptor checksums
4649                          * are sane.  If they aren't, refuse to remount r/w.
4650                          */
4651                         for (g = 0; g < sbi->s_groups_count; g++) {
4652                                 struct ext4_group_desc *gdp =
4653                                         ext4_get_group_desc(sb, g, NULL);
4654
4655                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4656                                         ext4_msg(sb, KERN_ERR,
4657                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4658                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4659                                                le16_to_cpu(gdp->bg_checksum));
4660                                         err = -EINVAL;
4661                                         goto restore_opts;
4662                                 }
4663                         }
4664
4665                         /*
4666                          * If we have an unprocessed orphan list hanging
4667                          * around from a previously readonly bdev mount,
4668                          * require a full umount/remount for now.
4669                          */
4670                         if (es->s_last_orphan) {
4671                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4672                                        "remount RDWR because of unprocessed "
4673                                        "orphan inode list.  Please "
4674                                        "umount/remount instead");
4675                                 err = -EINVAL;
4676                                 goto restore_opts;
4677                         }
4678
4679                         /*
4680                          * Mounting a RDONLY partition read-write, so reread
4681                          * and store the current valid flag.  (It may have
4682                          * been changed by e2fsck since we originally mounted
4683                          * the partition.)
4684                          */
4685                         if (sbi->s_journal)
4686                                 ext4_clear_journal_err(sb, es);
4687                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4688                         if (!ext4_setup_super(sb, es, 0))
4689                                 sb->s_flags &= ~MS_RDONLY;
4690                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4691                                                      EXT4_FEATURE_INCOMPAT_MMP))
4692                                 if (ext4_multi_mount_protect(sb,
4693                                                 le64_to_cpu(es->s_mmp_block))) {
4694                                         err = -EROFS;
4695                                         goto restore_opts;
4696                                 }
4697                         enable_quota = 1;
4698                 }
4699         }
4700
4701         /*
4702          * Reinitialize lazy itable initialization thread based on
4703          * current settings
4704          */
4705         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4706                 ext4_unregister_li_request(sb);
4707         else {
4708                 ext4_group_t first_not_zeroed;
4709                 first_not_zeroed = ext4_has_uninit_itable(sb);
4710                 ext4_register_li_request(sb, first_not_zeroed);
4711         }
4712
4713         ext4_setup_system_zone(sb);
4714         if (sbi->s_journal == NULL)
4715                 ext4_commit_super(sb, 1);
4716
4717         unlock_super(sb);
4718 #ifdef CONFIG_QUOTA
4719         /* Release old quota file names */
4720         for (i = 0; i < MAXQUOTAS; i++)
4721                 if (old_opts.s_qf_names[i] &&
4722                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4723                         kfree(old_opts.s_qf_names[i]);
4724         if (enable_quota) {
4725                 if (sb_any_quota_suspended(sb))
4726                         dquot_resume(sb, -1);
4727                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4728                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4729                         err = ext4_enable_quotas(sb);
4730                         if (err) {
4731                                 lock_super(sb);
4732                                 goto restore_opts;
4733                         }
4734                 }
4735         }
4736 #endif
4737
4738         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4739         kfree(orig_data);
4740         return 0;
4741
4742 restore_opts:
4743         sb->s_flags = old_sb_flags;
4744         sbi->s_mount_opt = old_opts.s_mount_opt;
4745         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4746         sbi->s_resuid = old_opts.s_resuid;
4747         sbi->s_resgid = old_opts.s_resgid;
4748         sbi->s_commit_interval = old_opts.s_commit_interval;
4749         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4750         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4751 #ifdef CONFIG_QUOTA
4752         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4753         for (i = 0; i < MAXQUOTAS; i++) {
4754                 if (sbi->s_qf_names[i] &&
4755                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4756                         kfree(sbi->s_qf_names[i]);
4757                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4758         }
4759 #endif
4760         unlock_super(sb);
4761         kfree(orig_data);
4762         return err;
4763 }
4764
4765 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4766 {
4767         struct super_block *sb = dentry->d_sb;
4768         struct ext4_sb_info *sbi = EXT4_SB(sb);
4769         struct ext4_super_block *es = sbi->s_es;
4770         ext4_fsblk_t overhead = 0;
4771         u64 fsid;
4772         s64 bfree;
4773
4774         if (!test_opt(sb, MINIX_DF))
4775                 overhead = sbi->s_overhead;
4776
4777         buf->f_type = EXT4_SUPER_MAGIC;
4778         buf->f_bsize = sb->s_blocksize;
4779         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4780         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4781                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4782         /* prevent underflow in case that few free space is available */
4783         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4784         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4785         if (buf->f_bfree < ext4_r_blocks_count(es))
4786                 buf->f_bavail = 0;
4787         buf->f_files = le32_to_cpu(es->s_inodes_count);
4788         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4789         buf->f_namelen = EXT4_NAME_LEN;
4790         fsid = le64_to_cpup((void *)es->s_uuid) ^
4791                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4792         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4793         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4794
4795         return 0;
4796 }
4797
4798 /* Helper function for writing quotas on sync - we need to start transaction
4799  * before quota file is locked for write. Otherwise the are possible deadlocks:
4800  * Process 1                         Process 2
4801  * ext4_create()                     quota_sync()
4802  *   jbd2_journal_start()                  write_dquot()
4803  *   dquot_initialize()                         down(dqio_mutex)
4804  *     down(dqio_mutex)                    jbd2_journal_start()
4805  *
4806  */
4807
4808 #ifdef CONFIG_QUOTA
4809
4810 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4811 {
4812         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4813 }
4814
4815 static int ext4_write_dquot(struct dquot *dquot)
4816 {
4817         int ret, err;
4818         handle_t *handle;
4819         struct inode *inode;
4820
4821         inode = dquot_to_inode(dquot);
4822         handle = ext4_journal_start(inode,
4823                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4824         if (IS_ERR(handle))
4825                 return PTR_ERR(handle);
4826         ret = dquot_commit(dquot);
4827         err = ext4_journal_stop(handle);
4828         if (!ret)
4829                 ret = err;
4830         return ret;
4831 }
4832
4833 static int ext4_acquire_dquot(struct dquot *dquot)
4834 {
4835         int ret, err;
4836         handle_t *handle;
4837
4838         handle = ext4_journal_start(dquot_to_inode(dquot),
4839                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4840         if (IS_ERR(handle))
4841                 return PTR_ERR(handle);
4842         ret = dquot_acquire(dquot);
4843         err = ext4_journal_stop(handle);
4844         if (!ret)
4845                 ret = err;
4846         return ret;
4847 }
4848
4849 static int ext4_release_dquot(struct dquot *dquot)
4850 {
4851         int ret, err;
4852         handle_t *handle;
4853
4854         handle = ext4_journal_start(dquot_to_inode(dquot),
4855                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4856         if (IS_ERR(handle)) {
4857                 /* Release dquot anyway to avoid endless cycle in dqput() */
4858                 dquot_release(dquot);
4859                 return PTR_ERR(handle);
4860         }
4861         ret = dquot_release(dquot);
4862         err = ext4_journal_stop(handle);
4863         if (!ret)
4864                 ret = err;
4865         return ret;
4866 }
4867
4868 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4869 {
4870         /* Are we journaling quotas? */
4871         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4872             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4873                 dquot_mark_dquot_dirty(dquot);
4874                 return ext4_write_dquot(dquot);
4875         } else {
4876                 return dquot_mark_dquot_dirty(dquot);
4877         }
4878 }
4879
4880 static int ext4_write_info(struct super_block *sb, int type)
4881 {
4882         int ret, err;
4883         handle_t *handle;
4884
4885         /* Data block + inode block */
4886         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4887         if (IS_ERR(handle))
4888                 return PTR_ERR(handle);
4889         ret = dquot_commit_info(sb, type);
4890         err = ext4_journal_stop(handle);
4891         if (!ret)
4892                 ret = err;
4893         return ret;
4894 }
4895
4896 /*
4897  * Turn on quotas during mount time - we need to find
4898  * the quota file and such...
4899  */
4900 static int ext4_quota_on_mount(struct super_block *sb, int type)
4901 {
4902         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4903                                         EXT4_SB(sb)->s_jquota_fmt, type);
4904 }
4905
4906 /*
4907  * Standard function to be called on quota_on
4908  */
4909 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4910                          struct path *path)
4911 {
4912         int err;
4913
4914         if (!test_opt(sb, QUOTA))
4915                 return -EINVAL;
4916
4917         /* Quotafile not on the same filesystem? */
4918         if (path->dentry->d_sb != sb)
4919                 return -EXDEV;
4920         /* Journaling quota? */
4921         if (EXT4_SB(sb)->s_qf_names[type]) {
4922                 /* Quotafile not in fs root? */
4923                 if (path->dentry->d_parent != sb->s_root)
4924                         ext4_msg(sb, KERN_WARNING,
4925                                 "Quota file not on filesystem root. "
4926                                 "Journaled quota will not work");
4927         }
4928
4929         /*
4930          * When we journal data on quota file, we have to flush journal to see
4931          * all updates to the file when we bypass pagecache...
4932          */
4933         if (EXT4_SB(sb)->s_journal &&
4934             ext4_should_journal_data(path->dentry->d_inode)) {
4935                 /*
4936                  * We don't need to lock updates but journal_flush() could
4937                  * otherwise be livelocked...
4938                  */
4939                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4940                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4941                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4942                 if (err)
4943                         return err;
4944         }
4945
4946         return dquot_quota_on(sb, type, format_id, path);
4947 }
4948
4949 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4950                              unsigned int flags)
4951 {
4952         int err;
4953         struct inode *qf_inode;
4954         unsigned long qf_inums[MAXQUOTAS] = {
4955                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4956                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4957         };
4958
4959         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4960
4961         if (!qf_inums[type])
4962                 return -EPERM;
4963
4964         qf_inode = ext4_iget(sb, qf_inums[type]);
4965         if (IS_ERR(qf_inode)) {
4966                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4967                 return PTR_ERR(qf_inode);
4968         }
4969
4970         err = dquot_enable(qf_inode, type, format_id, flags);
4971         iput(qf_inode);
4972
4973         return err;
4974 }
4975
4976 /* Enable usage tracking for all quota types. */
4977 static int ext4_enable_quotas(struct super_block *sb)
4978 {
4979         int type, err = 0;
4980         unsigned long qf_inums[MAXQUOTAS] = {
4981                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4982                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4983         };
4984
4985         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
4986         for (type = 0; type < MAXQUOTAS; type++) {
4987                 if (qf_inums[type]) {
4988                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
4989                                                 DQUOT_USAGE_ENABLED);
4990                         if (err) {
4991                                 ext4_warning(sb,
4992                                         "Failed to enable quota (type=%d) "
4993                                         "tracking. Please run e2fsck to fix.",
4994                                         type);
4995                                 return err;
4996                         }
4997                 }
4998         }
4999         return 0;
5000 }
5001
5002 /*
5003  * quota_on function that is used when QUOTA feature is set.
5004  */
5005 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5006                                  int format_id)
5007 {
5008         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5009                 return -EINVAL;
5010
5011         /*
5012          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5013          */
5014         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5015 }
5016
5017 static int ext4_quota_off(struct super_block *sb, int type)
5018 {
5019         struct inode *inode = sb_dqopt(sb)->files[type];
5020         handle_t *handle;
5021
5022         /* Force all delayed allocation blocks to be allocated.
5023          * Caller already holds s_umount sem */
5024         if (test_opt(sb, DELALLOC))
5025                 sync_filesystem(sb);
5026
5027         if (!inode)
5028                 goto out;
5029
5030         /* Update modification times of quota files when userspace can
5031          * start looking at them */
5032         handle = ext4_journal_start(inode, 1);
5033         if (IS_ERR(handle))
5034                 goto out;
5035         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5036         ext4_mark_inode_dirty(handle, inode);
5037         ext4_journal_stop(handle);
5038
5039 out:
5040         return dquot_quota_off(sb, type);
5041 }
5042
5043 /*
5044  * quota_off function that is used when QUOTA feature is set.
5045  */
5046 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5047 {
5048         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5049                 return -EINVAL;
5050
5051         /* Disable only the limits. */
5052         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5053 }
5054
5055 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5056  * acquiring the locks... As quota files are never truncated and quota code
5057  * itself serializes the operations (and no one else should touch the files)
5058  * we don't have to be afraid of races */
5059 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5060                                size_t len, loff_t off)
5061 {
5062         struct inode *inode = sb_dqopt(sb)->files[type];
5063         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5064         int err = 0;
5065         int offset = off & (sb->s_blocksize - 1);
5066         int tocopy;
5067         size_t toread;
5068         struct buffer_head *bh;
5069         loff_t i_size = i_size_read(inode);
5070
5071         if (off > i_size)
5072                 return 0;
5073         if (off+len > i_size)
5074                 len = i_size-off;
5075         toread = len;
5076         while (toread > 0) {
5077                 tocopy = sb->s_blocksize - offset < toread ?
5078                                 sb->s_blocksize - offset : toread;
5079                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5080                 if (err)
5081                         return err;
5082                 if (!bh)        /* A hole? */
5083                         memset(data, 0, tocopy);
5084                 else
5085                         memcpy(data, bh->b_data+offset, tocopy);
5086                 brelse(bh);
5087                 offset = 0;
5088                 toread -= tocopy;
5089                 data += tocopy;
5090                 blk++;
5091         }
5092         return len;
5093 }
5094
5095 /* Write to quotafile (we know the transaction is already started and has
5096  * enough credits) */
5097 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5098                                 const char *data, size_t len, loff_t off)
5099 {
5100         struct inode *inode = sb_dqopt(sb)->files[type];
5101         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5102         int err = 0;
5103         int offset = off & (sb->s_blocksize - 1);
5104         struct buffer_head *bh;
5105         handle_t *handle = journal_current_handle();
5106
5107         if (EXT4_SB(sb)->s_journal && !handle) {
5108                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5109                         " cancelled because transaction is not started",
5110                         (unsigned long long)off, (unsigned long long)len);
5111                 return -EIO;
5112         }
5113         /*
5114          * Since we account only one data block in transaction credits,
5115          * then it is impossible to cross a block boundary.
5116          */
5117         if (sb->s_blocksize - offset < len) {
5118                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5119                         " cancelled because not block aligned",
5120                         (unsigned long long)off, (unsigned long long)len);
5121                 return -EIO;
5122         }
5123
5124         bh = ext4_bread(handle, inode, blk, 1, &err);
5125         if (!bh)
5126                 goto out;
5127         err = ext4_journal_get_write_access(handle, bh);
5128         if (err) {
5129                 brelse(bh);
5130                 goto out;
5131         }
5132         lock_buffer(bh);
5133         memcpy(bh->b_data+offset, data, len);
5134         flush_dcache_page(bh->b_page);
5135         unlock_buffer(bh);
5136         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5137         brelse(bh);
5138 out:
5139         if (err)
5140                 return err;
5141         if (inode->i_size < off + len) {
5142                 i_size_write(inode, off + len);
5143                 EXT4_I(inode)->i_disksize = inode->i_size;
5144                 ext4_mark_inode_dirty(handle, inode);
5145         }
5146         return len;
5147 }
5148
5149 #endif
5150
5151 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5152                        const char *dev_name, void *data)
5153 {
5154         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5155 }
5156
5157 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5158 static inline void register_as_ext2(void)
5159 {
5160         int err = register_filesystem(&ext2_fs_type);
5161         if (err)
5162                 printk(KERN_WARNING
5163                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5164 }
5165
5166 static inline void unregister_as_ext2(void)
5167 {
5168         unregister_filesystem(&ext2_fs_type);
5169 }
5170
5171 static inline int ext2_feature_set_ok(struct super_block *sb)
5172 {
5173         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5174                 return 0;
5175         if (sb->s_flags & MS_RDONLY)
5176                 return 1;
5177         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5178                 return 0;
5179         return 1;
5180 }
5181 MODULE_ALIAS("ext2");
5182 #else
5183 static inline void register_as_ext2(void) { }
5184 static inline void unregister_as_ext2(void) { }
5185 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5186 #endif
5187
5188 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5189 static inline void register_as_ext3(void)
5190 {
5191         int err = register_filesystem(&ext3_fs_type);
5192         if (err)
5193                 printk(KERN_WARNING
5194                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5195 }
5196
5197 static inline void unregister_as_ext3(void)
5198 {
5199         unregister_filesystem(&ext3_fs_type);
5200 }
5201
5202 static inline int ext3_feature_set_ok(struct super_block *sb)
5203 {
5204         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5205                 return 0;
5206         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5207                 return 0;
5208         if (sb->s_flags & MS_RDONLY)
5209                 return 1;
5210         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5211                 return 0;
5212         return 1;
5213 }
5214 MODULE_ALIAS("ext3");
5215 #else
5216 static inline void register_as_ext3(void) { }
5217 static inline void unregister_as_ext3(void) { }
5218 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5219 #endif
5220
5221 static struct file_system_type ext4_fs_type = {
5222         .owner          = THIS_MODULE,
5223         .name           = "ext4",
5224         .mount          = ext4_mount,
5225         .kill_sb        = kill_block_super,
5226         .fs_flags       = FS_REQUIRES_DEV,
5227 };
5228
5229 static int __init ext4_init_feat_adverts(void)
5230 {
5231         struct ext4_features *ef;
5232         int ret = -ENOMEM;
5233
5234         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5235         if (!ef)
5236                 goto out;
5237
5238         ef->f_kobj.kset = ext4_kset;
5239         init_completion(&ef->f_kobj_unregister);
5240         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5241                                    "features");
5242         if (ret) {
5243                 kfree(ef);
5244                 goto out;
5245         }
5246
5247         ext4_feat = ef;
5248         ret = 0;
5249 out:
5250         return ret;
5251 }
5252
5253 static void ext4_exit_feat_adverts(void)
5254 {
5255         kobject_put(&ext4_feat->f_kobj);
5256         wait_for_completion(&ext4_feat->f_kobj_unregister);
5257         kfree(ext4_feat);
5258 }
5259
5260 /* Shared across all ext4 file systems */
5261 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5262 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5263
5264 static int __init ext4_init_fs(void)
5265 {
5266         int i, err;
5267
5268         ext4_li_info = NULL;
5269         mutex_init(&ext4_li_mtx);
5270
5271         ext4_check_flag_values();
5272
5273         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5274                 mutex_init(&ext4__aio_mutex[i]);
5275                 init_waitqueue_head(&ext4__ioend_wq[i]);
5276         }
5277
5278         err = ext4_init_pageio();
5279         if (err)
5280                 return err;
5281         err = ext4_init_system_zone();
5282         if (err)
5283                 goto out6;
5284         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5285         if (!ext4_kset)
5286                 goto out5;
5287         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5288
5289         err = ext4_init_feat_adverts();
5290         if (err)
5291                 goto out4;
5292
5293         err = ext4_init_mballoc();
5294         if (err)
5295                 goto out3;
5296
5297         err = ext4_init_xattr();
5298         if (err)
5299                 goto out2;
5300         err = init_inodecache();
5301         if (err)
5302                 goto out1;
5303         register_as_ext3();
5304         register_as_ext2();
5305         err = register_filesystem(&ext4_fs_type);
5306         if (err)
5307                 goto out;
5308
5309         return 0;
5310 out:
5311         unregister_as_ext2();
5312         unregister_as_ext3();
5313         destroy_inodecache();
5314 out1:
5315         ext4_exit_xattr();
5316 out2:
5317         ext4_exit_mballoc();
5318 out3:
5319         ext4_exit_feat_adverts();
5320 out4:
5321         if (ext4_proc_root)
5322                 remove_proc_entry("fs/ext4", NULL);
5323         kset_unregister(ext4_kset);
5324 out5:
5325         ext4_exit_system_zone();
5326 out6:
5327         ext4_exit_pageio();
5328         return err;
5329 }
5330
5331 static void __exit ext4_exit_fs(void)
5332 {
5333         ext4_destroy_lazyinit_thread();
5334         unregister_as_ext2();
5335         unregister_as_ext3();
5336         unregister_filesystem(&ext4_fs_type);
5337         destroy_inodecache();
5338         ext4_exit_xattr();
5339         ext4_exit_mballoc();
5340         ext4_exit_feat_adverts();
5341         remove_proc_entry("fs/ext4", NULL);
5342         kset_unregister(ext4_kset);
5343         ext4_exit_system_zone();
5344         ext4_exit_pageio();
5345 }
5346
5347 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5348 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5349 MODULE_LICENSE("GPL");
5350 module_init(ext4_init_fs)
5351 module_exit(ext4_exit_fs)