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"       /* Needed for trace points definition */
  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 {
 148         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 149
 150         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 151                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 152                 return;
 153
 154         es->s_checksum = ext4_superblock_csum(sb, es);
 155 }
 156
 157 void *ext4_kvmalloc(size_t size, gfp_t flags)
 158 {
 159         void *ret;
 160
 161         ret = kmalloc(size, flags);
 162         if (!ret)
 163                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 164         return ret;
 165 }
 166
 167 void *ext4_kvzalloc(size_t size, gfp_t flags)
 168 {
 169         void *ret;
 170
 171         ret = kzalloc(size, flags);
 172         if (!ret)
 173                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 174         return ret;
 175 }
 176
 177 void ext4_kvfree(void *ptr)
 178 {
 179         if (is_vmalloc_addr(ptr))
 180                 vfree(ptr);
 181         else
 182                 kfree(ptr);
 183
 184 }
 185
 186 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 187                                struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 192 }
 193
 194 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 200 }
 201
 202 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le32_to_cpu(bg->bg_inode_table_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 208 }
 209
 210 __u32 ext4_free_group_clusters(struct super_block *sb,
 211                                struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_free_inodes_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 224 }
 225
 226 __u32 ext4_used_dirs_count(struct super_block *sb,
 227                               struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 232 }
 233
 234 __u32 ext4_itable_unused_count(struct super_block *sb,
 235                               struct ext4_group_desc *bg)
 236 {
 237         return le16_to_cpu(bg->bg_itable_unused_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 240 }
 241
 242 void ext4_block_bitmap_set(struct super_block *sb,
 243                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_inode_bitmap_set(struct super_block *sb,
 251                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 252 {
 253         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 256 }
 257
 258 void ext4_inode_table_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 260 {
 261         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 264 }
 265
 266 void ext4_free_group_clusters_set(struct super_block *sb,
 267                                   struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_free_inodes_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282 void ext4_used_dirs_set(struct super_block *sb,
 283                           struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290 void ext4_itable_unused_set(struct super_block *sb,
 291                           struct ext4_group_desc *bg, __u32 count)
 292 {
 293         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 296 }
 297
 298
 299 /* Just increment the non-pointer handle value */
 300 static handle_t *ext4_get_nojournal(void)
 301 {
 302         handle_t *handle = current->journal_info;
 303         unsigned long ref_cnt = (unsigned long)handle;
 304
 305         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 306
 307         ref_cnt++;
 308         handle = (handle_t *)ref_cnt;
 309
 310         current->journal_info = handle;
 311         return handle;
 312 }
 313
 314
 315 /* Decrement the non-pointer handle value */
 316 static void ext4_put_nojournal(handle_t *handle)
 317 {
 318         unsigned long ref_cnt = (unsigned long)handle;
 319
 320         BUG_ON(ref_cnt == 0);
 321
 322         ref_cnt--;
 323         handle = (handle_t *)ref_cnt;
 324
 325         current->journal_info = handle;
 326 }
 327
 328 /*
 329  * Wrappers for jbd2_journal_start/end.
 330  */
 331 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 332 {
 333         journal_t *journal;
 334
 335         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 336         if (sb->s_flags & MS_RDONLY)
 337                 return ERR_PTR(-EROFS);
 338
 339         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 340         journal = EXT4_SB(sb)->s_journal;
 341         if (!journal)
 342                 return ext4_get_nojournal();
 343         /*
 344          * Special case here: if the journal has aborted behind our
 345          * backs (eg. EIO in the commit thread), then we still need to
 346          * take the FS itself readonly cleanly.
 347          */
 348         if (is_journal_aborted(journal)) {
 349                 ext4_abort(sb, "Detected aborted journal");
 350                 return ERR_PTR(-EROFS);
 351         }
 352         return jbd2_journal_start(journal, nblocks);
 353 }
 354
 355 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 356 {
 357         struct super_block *sb;
 358         int err;
 359         int rc;
 360
 361         if (!ext4_handle_valid(handle)) {
 362                 ext4_put_nojournal(handle);
 363                 return 0;
 364         }
 365         sb = handle->h_transaction->t_journal->j_private;
 366         err = handle->h_err;
 367         rc = jbd2_journal_stop(handle);
 368
 369         if (!err)
 370                 err = rc;
 371         if (err)
 372                 __ext4_std_error(sb, where, line, err);
 373         return err;
 374 }
 375
 376 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 377                                const char *err_fn, struct buffer_head *bh,
 378                                handle_t *handle, int err)
 379 {
 380         char nbuf[16];
 381         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 382
 383         BUG_ON(!ext4_handle_valid(handle));
 384
 385         if (bh)
 386                 BUFFER_TRACE(bh, "abort");
 387
 388         if (!handle->h_err)
 389                 handle->h_err = err;
 390
 391         if (is_handle_aborted(handle))
 392                 return;
 393
 394         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 395                caller, line, errstr, err_fn);
 396
 397         jbd2_journal_abort_handle(handle);
 398 }
 399
 400 static void __save_error_info(struct super_block *sb, const char *func,
 401                             unsigned int line)
 402 {
 403         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 404
 405         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 406         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 407         es->s_last_error_time = cpu_to_le32(get_seconds());
 408         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 409         es->s_last_error_line = cpu_to_le32(line);
 410         if (!es->s_first_error_time) {
 411                 es->s_first_error_time = es->s_last_error_time;
 412                 strncpy(es->s_first_error_func, func,
 413                         sizeof(es->s_first_error_func));
 414                 es->s_first_error_line = cpu_to_le32(line);
 415                 es->s_first_error_ino = es->s_last_error_ino;
 416                 es->s_first_error_block = es->s_last_error_block;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, const char *func,
 428                             unsigned int line)
 429 {
 430         __save_error_info(sb, func, line);
 431         ext4_commit_super(sb, 1);
 432 }
 433
 434 /*
 435  * The del_gendisk() function uninitializes the disk-specific data
 436  * structures, including the bdi structure, without telling anyone
 437  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 438  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 439  * This is a kludge to prevent these oops until we can put in a proper
 440  * hook in del_gendisk() to inform the VFS and file system layers.
 441  */
 442 static int block_device_ejected(struct super_block *sb)
 443 {
 444         struct inode *bd_inode = sb->s_bdev->bd_inode;
 445         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 446
 447         return bdi->dev == NULL;
 448 }
 449
 450 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 451 {
 452         struct super_block              *sb = journal->j_private;
 453         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 454         int                             error = is_journal_aborted(journal);
 455         struct ext4_journal_cb_entry    *jce, *tmp;
 456
 457         spin_lock(&sbi->s_md_lock);
 458         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 459                 list_del_init(&jce->jce_list);
 460                 spin_unlock(&sbi->s_md_lock);
 461                 jce->jce_func(sb, jce, error);
 462                 spin_lock(&sbi->s_md_lock);
 463         }
 464         spin_unlock(&sbi->s_md_lock);
 465 }
 466
 467 /* Deal with the reporting of failure conditions on a filesystem such as
 468  * inconsistencies detected or read IO failures.
 469  *
 470  * On ext2, we can store the error state of the filesystem in the
 471  * superblock.  That is not possible on ext4, because we may have other
 472  * write ordering constraints on the superblock which prevent us from
 473  * writing it out straight away; and given that the journal is about to
 474  * be aborted, we can't rely on the current, or future, transactions to
 475  * write out the superblock safely.
 476  *
 477  * We'll just use the jbd2_journal_abort() error code to record an error in
 478  * the journal instead.  On recovery, the journal will complain about
 479  * that error until we've noted it down and cleared it.
 480  */
 481
 482 static void ext4_handle_error(struct super_block *sb)
 483 {
 484         if (sb->s_flags & MS_RDONLY)
 485                 return;
 486
 487         if (!test_opt(sb, ERRORS_CONT)) {
 488                 journal_t *journal = EXT4_SB(sb)->s_journal;
 489
 490                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 491                 if (journal)
 492                         jbd2_journal_abort(journal, -EIO);
 493         }
 494         if (test_opt(sb, ERRORS_RO)) {
 495                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 496                 sb->s_flags |= MS_RDONLY;
 497         }
 498         if (test_opt(sb, ERRORS_PANIC))
 499                 panic("EXT4-fs (device %s): panic forced after error\n",
 500                         sb->s_id);
 501 }
 502
 503 void __ext4_error(struct super_block *sb, const char *function,
 504                   unsigned int line, const char *fmt, ...)
 505 {
 506         struct va_format vaf;
 507         va_list args;
 508
 509         va_start(args, fmt);
 510         vaf.fmt = fmt;
 511         vaf.va = &args;
 512         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 513                sb->s_id, function, line, current->comm, &vaf);
 514         va_end(args);
 515         save_error_info(sb, function, line);
 516
 517         ext4_handle_error(sb);
 518 }
 519
 520 void ext4_error_inode(struct inode *inode, const char *function,
 521                       unsigned int line, ext4_fsblk_t block,
 522                       const char *fmt, ...)
 523 {
 524         va_list args;
 525         struct va_format vaf;
 526         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 527
 528         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 529         es->s_last_error_block = cpu_to_le64(block);
 530         save_error_info(inode->i_sb, function, line);
 531         va_start(args, fmt);
 532         vaf.fmt = fmt;
 533         vaf.va = &args;
 534         if (block)
 535                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 536                        "inode #%lu: block %llu: comm %s: %pV\n",
 537                        inode->i_sb->s_id, function, line, inode->i_ino,
 538                        block, current->comm, &vaf);
 539         else
 540                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 541                        "inode #%lu: comm %s: %pV\n",
 542                        inode->i_sb->s_id, function, line, inode->i_ino,
 543                        current->comm, &vaf);
 544         va_end(args);
 545
 546         ext4_handle_error(inode->i_sb);
 547 }
 548
 549 void ext4_error_file(struct file *file, const char *function,
 550                      unsigned int line, ext4_fsblk_t block,
 551                      const char *fmt, ...)
 552 {
 553         va_list args;
 554         struct va_format vaf;
 555         struct ext4_super_block *es;
 556         struct inode *inode = file->f_dentry->d_inode;
 557         char pathname[80], *path;
 558
 559         es = EXT4_SB(inode->i_sb)->s_es;
 560         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 561         save_error_info(inode->i_sb, function, line);
 562         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 563         if (IS_ERR(path))
 564                 path = "(unknown)";
 565         va_start(args, fmt);
 566         vaf.fmt = fmt;
 567         vaf.va = &args;
 568         if (block)
 569                 printk(KERN_CRIT
 570                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 571                        "block %llu: comm %s: path %s: %pV\n",
 572                        inode->i_sb->s_id, function, line, inode->i_ino,
 573                        block, current->comm, path, &vaf);
 574         else
 575                 printk(KERN_CRIT
 576                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 577                        "comm %s: path %s: %pV\n",
 578                        inode->i_sb->s_id, function, line, inode->i_ino,
 579                        current->comm, path, &vaf);
 580         va_end(args);
 581
 582         ext4_handle_error(inode->i_sb);
 583 }
 584
 585 static const char *ext4_decode_error(struct super_block *sb, int errno,
 586                                      char nbuf[16])
 587 {
 588         char *errstr = NULL;
 589
 590         switch (errno) {
 591         case -EIO:
 592                 errstr = "IO failure";
 593                 break;
 594         case -ENOMEM:
 595                 errstr = "Out of memory";
 596                 break;
 597         case -EROFS:
 598                 if (!sb || (EXT4_SB(sb)->s_journal &&
 599                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 600                         errstr = "Journal has aborted";
 601                 else
 602                         errstr = "Readonly filesystem";
 603                 break;
 604         default:
 605                 /* If the caller passed in an extra buffer for unknown
 606                  * errors, textualise them now.  Else we just return
 607                  * NULL. */
 608                 if (nbuf) {
 609                         /* Check for truncated error codes... */
 610                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 611                                 errstr = nbuf;
 612                 }
 613                 break;
 614         }
 615
 616         return errstr;
 617 }
 618
 619 /* __ext4_std_error decodes expected errors from journaling functions
 620  * automatically and invokes the appropriate error response.  */
 621
 622 void __ext4_std_error(struct super_block *sb, const char *function,
 623                       unsigned int line, int errno)
 624 {
 625         char nbuf[16];
 626         const char *errstr;
 627
 628         /* Special case: if the error is EROFS, and we're not already
 629          * inside a transaction, then there's really no point in logging
 630          * an error. */
 631         if (errno == -EROFS && journal_current_handle() == NULL &&
 632             (sb->s_flags & MS_RDONLY))
 633                 return;
 634
 635         errstr = ext4_decode_error(sb, errno, nbuf);
 636         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 637                sb->s_id, function, line, errstr);
 638         save_error_info(sb, function, line);
 639
 640         ext4_handle_error(sb);
 641 }
 642
 643 /*
 644  * ext4_abort is a much stronger failure handler than ext4_error.  The
 645  * abort function may be used to deal with unrecoverable failures such
 646  * as journal IO errors or ENOMEM at a critical moment in log management.
 647  *
 648  * We unconditionally force the filesystem into an ABORT|READONLY state,
 649  * unless the error response on the fs has been set to panic in which
 650  * case we take the easy way out and panic immediately.
 651  */
 652
 653 void __ext4_abort(struct super_block *sb, const char *function,
 654                 unsigned int line, const char *fmt, ...)
 655 {
 656         va_list args;
 657
 658         save_error_info(sb, function, line);
 659         va_start(args, fmt);
 660         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 661                function, line);
 662         vprintk(fmt, args);
 663         printk("\n");
 664         va_end(args);
 665
 666         if ((sb->s_flags & MS_RDONLY) == 0) {
 667                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 668                 sb->s_flags |= MS_RDONLY;
 669                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 670                 if (EXT4_SB(sb)->s_journal)
 671                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 672                 save_error_info(sb, function, line);
 673         }
 674         if (test_opt(sb, ERRORS_PANIC))
 675                 panic("EXT4-fs panic from previous error\n");
 676 }
 677
 678 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 679 {
 680         struct va_format vaf;
 681         va_list args;
 682
 683         va_start(args, fmt);
 684         vaf.fmt = fmt;
 685         vaf.va = &args;
 686         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 687         va_end(args);
 688 }
 689
 690 void __ext4_warning(struct super_block *sb, const char *function,
 691                     unsigned int line, const char *fmt, ...)
 692 {
 693         struct va_format vaf;
 694         va_list args;
 695
 696         va_start(args, fmt);
 697         vaf.fmt = fmt;
 698         vaf.va = &args;
 699         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 700                sb->s_id, function, line, &vaf);
 701         va_end(args);
 702 }
 703
 704 void __ext4_grp_locked_error(const char *function, unsigned int line,
 705                              struct super_block *sb, ext4_group_t grp,
 706                              unsigned long ino, ext4_fsblk_t block,
 707                              const char *fmt, ...)
 708 __releases(bitlock)
 709 __acquires(bitlock)
 710 {
 711         struct va_format vaf;
 712         va_list args;
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         es->s_last_error_ino = cpu_to_le32(ino);
 716         es->s_last_error_block = cpu_to_le64(block);
 717         __save_error_info(sb, function, line);
 718
 719         va_start(args, fmt);
 720
 721         vaf.fmt = fmt;
 722         vaf.va = &args;
 723         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 724                sb->s_id, function, line, grp);
 725         if (ino)
 726                 printk(KERN_CONT "inode %lu: ", ino);
 727         if (block)
 728                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 729         printk(KERN_CONT "%pV\n", &vaf);
 730         va_end(args);
 731
 732         if (test_opt(sb, ERRORS_CONT)) {
 733                 ext4_commit_super(sb, 0);
 734                 return;
 735         }
 736
 737         ext4_unlock_group(sb, grp);
 738         ext4_handle_error(sb);
 739         /*
 740          * We only get here in the ERRORS_RO case; relocking the group
 741          * may be dangerous, but nothing bad will happen since the
 742          * filesystem will have already been marked read/only and the
 743          * journal has been aborted.  We return 1 as a hint to callers
 744          * who might what to use the return value from
 745          * ext4_grp_locked_error() to distinguish between the
 746          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 747          * aggressively from the ext4 function in question, with a
 748          * more appropriate error code.
 749          */
 750         ext4_lock_group(sb, grp);
 751         return;
 752 }
 753
 754 void ext4_update_dynamic_rev(struct super_block *sb)
 755 {
 756         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 757
 758         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 759                 return;
 760
 761         ext4_warning(sb,
 762                      "updating to rev %d because of new feature flag, "
 763                      "running e2fsck is recommended",
 764                      EXT4_DYNAMIC_REV);
 765
 766         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 767         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 768         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 769         /* leave es->s_feature_*compat flags alone */
 770         /* es->s_uuid will be set by e2fsck if empty */
 771
 772         /*
 773          * The rest of the superblock fields should be zero, and if not it
 774          * means they are likely already in use, so leave them alone.  We
 775          * can leave it up to e2fsck to clean up any inconsistencies there.
 776          */
 777 }
 778
 779 /*
 780  * Open the external journal device
 781  */
 782 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 783 {
 784         struct block_device *bdev;
 785         char b[BDEVNAME_SIZE];
 786
 787         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 788         if (IS_ERR(bdev))
 789                 goto fail;
 790         return bdev;
 791
 792 fail:
 793         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 794                         __bdevname(dev, b), PTR_ERR(bdev));
 795         return NULL;
 796 }
 797
 798 /*
 799  * Release the journal device
 800  */
 801 static int ext4_blkdev_put(struct block_device *bdev)
 802 {
 803         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 804 }
 805
 806 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 807 {
 808         struct block_device *bdev;
 809         int ret = -ENODEV;
 810
 811         bdev = sbi->journal_bdev;
 812         if (bdev) {
 813                 ret = ext4_blkdev_put(bdev);
 814                 sbi->journal_bdev = NULL;
 815         }
 816         return ret;
 817 }
 818
 819 static inline struct inode *orphan_list_entry(struct list_head *l)
 820 {
 821         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 822 }
 823
 824 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 825 {
 826         struct list_head *l;
 827
 828         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 829                  le32_to_cpu(sbi->s_es->s_last_orphan));
 830
 831         printk(KERN_ERR "sb_info orphan list:\n");
 832         list_for_each(l, &sbi->s_orphan) {
 833                 struct inode *inode = orphan_list_entry(l);
 834                 printk(KERN_ERR "  "
 835                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 836                        inode->i_sb->s_id, inode->i_ino, inode,
 837                        inode->i_mode, inode->i_nlink,
 838                        NEXT_ORPHAN(inode));
 839         }
 840 }
 841
 842 static void ext4_put_super(struct super_block *sb)
 843 {
 844         struct ext4_sb_info *sbi = EXT4_SB(sb);
 845         struct ext4_super_block *es = sbi->s_es;
 846         int i, err;
 847
 848         ext4_unregister_li_request(sb);
 849         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 850
 851         flush_workqueue(sbi->dio_unwritten_wq);
 852         destroy_workqueue(sbi->dio_unwritten_wq);
 853
 854         if (sbi->s_journal) {
 855                 err = jbd2_journal_destroy(sbi->s_journal);
 856                 sbi->s_journal = NULL;
 857                 if (err < 0)
 858                         ext4_abort(sb, "Couldn't clean up the journal");
 859         }
 860
 861         del_timer(&sbi->s_err_report);
 862         ext4_release_system_zone(sb);
 863         ext4_mb_release(sb);
 864         ext4_ext_release(sb);
 865         ext4_xattr_put_super(sb);
 866
 867         if (!(sb->s_flags & MS_RDONLY)) {
 868                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 869                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 870         }
 871         if (!(sb->s_flags & MS_RDONLY))
 872                 ext4_commit_super(sb, 1);
 873
 874         if (sbi->s_proc) {
 875                 remove_proc_entry("options", sbi->s_proc);
 876                 remove_proc_entry(sb->s_id, ext4_proc_root);
 877         }
 878         kobject_del(&sbi->s_kobj);
 879
 880         for (i = 0; i < sbi->s_gdb_count; i++)
 881                 brelse(sbi->s_group_desc[i]);
 882         ext4_kvfree(sbi->s_group_desc);
 883         ext4_kvfree(sbi->s_flex_groups);
 884         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 885         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 886         percpu_counter_destroy(&sbi->s_dirs_counter);
 887         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 888         brelse(sbi->s_sbh);
 889 #ifdef CONFIG_QUOTA
 890         for (i = 0; i < MAXQUOTAS; i++)
 891                 kfree(sbi->s_qf_names[i]);
 892 #endif
 893
 894         /* Debugging code just in case the in-memory inode orphan list
 895          * isn't empty.  The on-disk one can be non-empty if we've
 896          * detected an error and taken the fs readonly, but the
 897          * in-memory list had better be clean by this point. */
 898         if (!list_empty(&sbi->s_orphan))
 899                 dump_orphan_list(sb, sbi);
 900         J_ASSERT(list_empty(&sbi->s_orphan));
 901
 902         invalidate_bdev(sb->s_bdev);
 903         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 904                 /*
 905                  * Invalidate the journal device's buffers.  We don't want them
 906                  * floating about in memory - the physical journal device may
 907                  * hotswapped, and it breaks the `ro-after' testing code.
 908                  */
 909                 sync_blockdev(sbi->journal_bdev);
 910                 invalidate_bdev(sbi->journal_bdev);
 911                 ext4_blkdev_remove(sbi);
 912         }
 913         if (sbi->s_mmp_tsk)
 914                 kthread_stop(sbi->s_mmp_tsk);
 915         sb->s_fs_info = NULL;
 916         /*
 917          * Now that we are completely done shutting down the
 918          * superblock, we need to actually destroy the kobject.
 919          */
 920         kobject_put(&sbi->s_kobj);
 921         wait_for_completion(&sbi->s_kobj_unregister);
 922         if (sbi->s_chksum_driver)
 923                 crypto_free_shash(sbi->s_chksum_driver);
 924         kfree(sbi->s_blockgroup_lock);
 925         kfree(sbi);
 926 }
 927
 928 static struct kmem_cache *ext4_inode_cachep;
 929
 930 /*
 931  * Called inside transaction, so use GFP_NOFS
 932  */
 933 static struct inode *ext4_alloc_inode(struct super_block *sb)
 934 {
 935         struct ext4_inode_info *ei;
 936
 937         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 938         if (!ei)
 939                 return NULL;
 940
 941         ei->vfs_inode.i_version = 1;
 942         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 943         INIT_LIST_HEAD(&ei->i_prealloc_list);
 944         spin_lock_init(&ei->i_prealloc_lock);
 945         ext4_es_init_tree(&ei->i_es_tree);
 946         rwlock_init(&ei->i_es_lock);
 947         ei->i_reserved_data_blocks = 0;
 948         ei->i_reserved_meta_blocks = 0;
 949         ei->i_allocated_meta_blocks = 0;
 950         ei->i_da_metadata_calc_len = 0;
 951         ei->i_da_metadata_calc_last_lblock = 0;
 952         spin_lock_init(&(ei->i_block_reservation_lock));
 953 #ifdef CONFIG_QUOTA
 954         ei->i_reserved_quota = 0;
 955 #endif
 956         ei->jinode = NULL;
 957         INIT_LIST_HEAD(&ei->i_completed_io_list);
 958         spin_lock_init(&ei->i_completed_io_lock);
 959         ei->i_sync_tid = 0;
 960         ei->i_datasync_tid = 0;
 961         atomic_set(&ei->i_ioend_count, 0);
 962         atomic_set(&ei->i_unwritten, 0);
 963
 964         return &ei->vfs_inode;
 965 }
 966
 967 static int ext4_drop_inode(struct inode *inode)
 968 {
 969         int drop = generic_drop_inode(inode);
 970
 971         trace_ext4_drop_inode(inode, drop);
 972         return drop;
 973 }
 974
 975 static void ext4_i_callback(struct rcu_head *head)
 976 {
 977         struct inode *inode = container_of(head, struct inode, i_rcu);
 978         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 979 }
 980
 981 static void ext4_destroy_inode(struct inode *inode)
 982 {
 983         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 984                 ext4_msg(inode->i_sb, KERN_ERR,
 985                          "Inode %lu (%p): orphan list check failed!",
 986                          inode->i_ino, EXT4_I(inode));
 987                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 988                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 989                                 true);
 990                 dump_stack();
 991         }
 992         call_rcu(&inode->i_rcu, ext4_i_callback);
 993 }
 994
 995 static void init_once(void *foo)
 996 {
 997         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 998
 999         INIT_LIST_HEAD(&ei->i_orphan);
1000         init_rwsem(&ei->xattr_sem);
1001         init_rwsem(&ei->i_data_sem);
1002         inode_init_once(&ei->vfs_inode);
1003 }
1004
1005 static int init_inodecache(void)
1006 {
1007         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1008                                              sizeof(struct ext4_inode_info),
1009                                              0, (SLAB_RECLAIM_ACCOUNT|
1010                                                 SLAB_MEM_SPREAD),
1011                                              init_once);
1012         if (ext4_inode_cachep == NULL)
1013                 return -ENOMEM;
1014         return 0;
1015 }
1016
1017 static void destroy_inodecache(void)
1018 {
1019         /*
1020          * Make sure all delayed rcu free inodes are flushed before we
1021          * destroy cache.
1022          */
1023         rcu_barrier();
1024         kmem_cache_destroy(ext4_inode_cachep);
1025 }
1026
1027 void ext4_clear_inode(struct inode *inode)
1028 {
1029         invalidate_inode_buffers(inode);
1030         clear_inode(inode);
1031         dquot_drop(inode);
1032         ext4_discard_preallocations(inode);
1033         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1034         if (EXT4_I(inode)->jinode) {
1035                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1036                                                EXT4_I(inode)->jinode);
1037                 jbd2_free_inode(EXT4_I(inode)->jinode);
1038                 EXT4_I(inode)->jinode = NULL;
1039         }
1040 }
1041
1042 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1043                                         u64 ino, u32 generation)
1044 {
1045         struct inode *inode;
1046
1047         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1048                 return ERR_PTR(-ESTALE);
1049         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1050                 return ERR_PTR(-ESTALE);
1051
1052         /* iget isn't really right if the inode is currently unallocated!!
1053          *
1054          * ext4_read_inode will return a bad_inode if the inode had been
1055          * deleted, so we should be safe.
1056          *
1057          * Currently we don't know the generation for parent directory, so
1058          * a generation of 0 means "accept any"
1059          */
1060         inode = ext4_iget(sb, ino);
1061         if (IS_ERR(inode))
1062                 return ERR_CAST(inode);
1063         if (generation && inode->i_generation != generation) {
1064                 iput(inode);
1065                 return ERR_PTR(-ESTALE);
1066         }
1067
1068         return inode;
1069 }
1070
1071 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1072                                         int fh_len, int fh_type)
1073 {
1074         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1075                                     ext4_nfs_get_inode);
1076 }
1077
1078 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1079                                         int fh_len, int fh_type)
1080 {
1081         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1082                                     ext4_nfs_get_inode);
1083 }
1084
1085 /*
1086  * Try to release metadata pages (indirect blocks, directories) which are
1087  * mapped via the block device.  Since these pages could have journal heads
1088  * which would prevent try_to_free_buffers() from freeing them, we must use
1089  * jbd2 layer's try_to_free_buffers() function to release them.
1090  */
1091 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1092                                  gfp_t wait)
1093 {
1094         journal_t *journal = EXT4_SB(sb)->s_journal;
1095
1096         WARN_ON(PageChecked(page));
1097         if (!page_has_buffers(page))
1098                 return 0;
1099         if (journal)
1100                 return jbd2_journal_try_to_free_buffers(journal, page,
1101                                                         wait & ~__GFP_WAIT);
1102         return try_to_free_buffers(page);
1103 }
1104
1105 #ifdef CONFIG_QUOTA
1106 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1107 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1108
1109 static int ext4_write_dquot(struct dquot *dquot);
1110 static int ext4_acquire_dquot(struct dquot *dquot);
1111 static int ext4_release_dquot(struct dquot *dquot);
1112 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1113 static int ext4_write_info(struct super_block *sb, int type);
1114 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1115                          struct path *path);
1116 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1117                                  int format_id);
1118 static int ext4_quota_off(struct super_block *sb, int type);
1119 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1120 static int ext4_quota_on_mount(struct super_block *sb, int type);
1121 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1122                                size_t len, loff_t off);
1123 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1124                                 const char *data, size_t len, loff_t off);
1125 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1126                              unsigned int flags);
1127 static int ext4_enable_quotas(struct super_block *sb);
1128
1129 static const struct dquot_operations ext4_quota_operations = {
1130         .get_reserved_space = ext4_get_reserved_space,
1131         .write_dquot    = ext4_write_dquot,
1132         .acquire_dquot  = ext4_acquire_dquot,
1133         .release_dquot  = ext4_release_dquot,
1134         .mark_dirty     = ext4_mark_dquot_dirty,
1135         .write_info     = ext4_write_info,
1136         .alloc_dquot    = dquot_alloc,
1137         .destroy_dquot  = dquot_destroy,
1138 };
1139
1140 static const struct quotactl_ops ext4_qctl_operations = {
1141         .quota_on       = ext4_quota_on,
1142         .quota_off      = ext4_quota_off,
1143         .quota_sync     = dquot_quota_sync,
1144         .get_info       = dquot_get_dqinfo,
1145         .set_info       = dquot_set_dqinfo,
1146         .get_dqblk      = dquot_get_dqblk,
1147         .set_dqblk      = dquot_set_dqblk
1148 };
1149
1150 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1151         .quota_on_meta  = ext4_quota_on_sysfile,
1152         .quota_off      = ext4_quota_off_sysfile,
1153         .quota_sync     = dquot_quota_sync,
1154         .get_info       = dquot_get_dqinfo,
1155         .set_info       = dquot_set_dqinfo,
1156         .get_dqblk      = dquot_get_dqblk,
1157         .set_dqblk      = dquot_set_dqblk
1158 };
1159 #endif
1160
1161 static const struct super_operations ext4_sops = {
1162         .alloc_inode    = ext4_alloc_inode,
1163         .destroy_inode  = ext4_destroy_inode,
1164         .write_inode    = ext4_write_inode,
1165         .dirty_inode    = ext4_dirty_inode,
1166         .drop_inode     = ext4_drop_inode,
1167         .evict_inode    = ext4_evict_inode,
1168         .put_super      = ext4_put_super,
1169         .sync_fs        = ext4_sync_fs,
1170         .freeze_fs      = ext4_freeze,
1171         .unfreeze_fs    = ext4_unfreeze,
1172         .statfs         = ext4_statfs,
1173         .remount_fs     = ext4_remount,
1174         .show_options   = ext4_show_options,
1175 #ifdef CONFIG_QUOTA
1176         .quota_read     = ext4_quota_read,
1177         .quota_write    = ext4_quota_write,
1178 #endif
1179         .bdev_try_to_free_page = bdev_try_to_free_page,
1180 };
1181
1182 static const struct super_operations ext4_nojournal_sops = {
1183         .alloc_inode    = ext4_alloc_inode,
1184         .destroy_inode  = ext4_destroy_inode,
1185         .write_inode    = ext4_write_inode,
1186         .dirty_inode    = ext4_dirty_inode,
1187         .drop_inode     = ext4_drop_inode,
1188         .evict_inode    = ext4_evict_inode,
1189         .put_super      = ext4_put_super,
1190         .statfs         = ext4_statfs,
1191         .remount_fs     = ext4_remount,
1192         .show_options   = ext4_show_options,
1193 #ifdef CONFIG_QUOTA
1194         .quota_read     = ext4_quota_read,
1195         .quota_write    = ext4_quota_write,
1196 #endif
1197         .bdev_try_to_free_page = bdev_try_to_free_page,
1198 };
1199
1200 static const struct export_operations ext4_export_ops = {
1201         .fh_to_dentry = ext4_fh_to_dentry,
1202         .fh_to_parent = ext4_fh_to_parent,
1203         .get_parent = ext4_get_parent,
1204 };
1205
1206 enum {
1207         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1208         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1209         Opt_nouid32, Opt_debug, Opt_removed,
1210         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1211         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1212         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1213         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1214         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1215         Opt_data_err_abort, Opt_data_err_ignore,
1216         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1217         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1218         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1219         Opt_usrquota, Opt_grpquota, Opt_i_version,
1220         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1221         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1222         Opt_inode_readahead_blks, Opt_journal_ioprio,
1223         Opt_dioread_nolock, Opt_dioread_lock,
1224         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1225         Opt_max_dir_size_kb,
1226 };
1227
1228 static const match_table_t tokens = {
1229         {Opt_bsd_df, "bsddf"},
1230         {Opt_minix_df, "minixdf"},
1231         {Opt_grpid, "grpid"},
1232         {Opt_grpid, "bsdgroups"},
1233         {Opt_nogrpid, "nogrpid"},
1234         {Opt_nogrpid, "sysvgroups"},
1235         {Opt_resgid, "resgid=%u"},
1236         {Opt_resuid, "resuid=%u"},
1237         {Opt_sb, "sb=%u"},
1238         {Opt_err_cont, "errors=continue"},
1239         {Opt_err_panic, "errors=panic"},
1240         {Opt_err_ro, "errors=remount-ro"},
1241         {Opt_nouid32, "nouid32"},
1242         {Opt_debug, "debug"},
1243         {Opt_removed, "oldalloc"},
1244         {Opt_removed, "orlov"},
1245         {Opt_user_xattr, "user_xattr"},
1246         {Opt_nouser_xattr, "nouser_xattr"},
1247         {Opt_acl, "acl"},
1248         {Opt_noacl, "noacl"},
1249         {Opt_noload, "norecovery"},
1250         {Opt_noload, "noload"},
1251         {Opt_removed, "nobh"},
1252         {Opt_removed, "bh"},
1253         {Opt_commit, "commit=%u"},
1254         {Opt_min_batch_time, "min_batch_time=%u"},
1255         {Opt_max_batch_time, "max_batch_time=%u"},
1256         {Opt_journal_dev, "journal_dev=%u"},
1257         {Opt_journal_checksum, "journal_checksum"},
1258         {Opt_journal_async_commit, "journal_async_commit"},
1259         {Opt_abort, "abort"},
1260         {Opt_data_journal, "data=journal"},
1261         {Opt_data_ordered, "data=ordered"},
1262         {Opt_data_writeback, "data=writeback"},
1263         {Opt_data_err_abort, "data_err=abort"},
1264         {Opt_data_err_ignore, "data_err=ignore"},
1265         {Opt_offusrjquota, "usrjquota="},
1266         {Opt_usrjquota, "usrjquota=%s"},
1267         {Opt_offgrpjquota, "grpjquota="},
1268         {Opt_grpjquota, "grpjquota=%s"},
1269         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1270         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1271         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1272         {Opt_grpquota, "grpquota"},
1273         {Opt_noquota, "noquota"},
1274         {Opt_quota, "quota"},
1275         {Opt_usrquota, "usrquota"},
1276         {Opt_barrier, "barrier=%u"},
1277         {Opt_barrier, "barrier"},
1278         {Opt_nobarrier, "nobarrier"},
1279         {Opt_i_version, "i_version"},
1280         {Opt_stripe, "stripe=%u"},
1281         {Opt_delalloc, "delalloc"},
1282         {Opt_nodelalloc, "nodelalloc"},
1283         {Opt_mblk_io_submit, "mblk_io_submit"},
1284         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1285         {Opt_block_validity, "block_validity"},
1286         {Opt_noblock_validity, "noblock_validity"},
1287         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1288         {Opt_journal_ioprio, "journal_ioprio=%u"},
1289         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1290         {Opt_auto_da_alloc, "auto_da_alloc"},
1291         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1292         {Opt_dioread_nolock, "dioread_nolock"},
1293         {Opt_dioread_lock, "dioread_lock"},
1294         {Opt_discard, "discard"},
1295         {Opt_nodiscard, "nodiscard"},
1296         {Opt_init_itable, "init_itable=%u"},
1297         {Opt_init_itable, "init_itable"},
1298         {Opt_noinit_itable, "noinit_itable"},
1299         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1300         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1301         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1302         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1303         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1304         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1305         {Opt_err, NULL},
1306 };
1307
1308 static ext4_fsblk_t get_sb_block(void **data)
1309 {
1310         ext4_fsblk_t    sb_block;
1311         char            *options = (char *) *data;
1312
1313         if (!options || strncmp(options, "sb=", 3) != 0)
1314                 return 1;       /* Default location */
1315
1316         options += 3;
1317         /* TODO: use simple_strtoll with >32bit ext4 */
1318         sb_block = simple_strtoul(options, &options, 0);
1319         if (*options && *options != ',') {
1320                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1321                        (char *) *data);
1322                 return 1;
1323         }
1324         if (*options == ',')
1325                 options++;
1326         *data = (void *) options;
1327
1328         return sb_block;
1329 }
1330
1331 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1332 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1333         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1334
1335 #ifdef CONFIG_QUOTA
1336 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1337 {
1338         struct ext4_sb_info *sbi = EXT4_SB(sb);
1339         char *qname;
1340
1341         if (sb_any_quota_loaded(sb) &&
1342                 !sbi->s_qf_names[qtype]) {
1343                 ext4_msg(sb, KERN_ERR,
1344                         "Cannot change journaled "
1345                         "quota options when quota turned on");
1346                 return -1;
1347         }
1348         qname = match_strdup(args);
1349         if (!qname) {
1350                 ext4_msg(sb, KERN_ERR,
1351                         "Not enough memory for storing quotafile name");
1352                 return -1;
1353         }
1354         if (sbi->s_qf_names[qtype] &&
1355                 strcmp(sbi->s_qf_names[qtype], qname)) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "%s quota file already specified", QTYPE2NAME(qtype));
1358                 kfree(qname);
1359                 return -1;
1360         }
1361         sbi->s_qf_names[qtype] = qname;
1362         if (strchr(sbi->s_qf_names[qtype], '/')) {
1363                 ext4_msg(sb, KERN_ERR,
1364                         "quotafile must be on filesystem root");
1365                 kfree(sbi->s_qf_names[qtype]);
1366                 sbi->s_qf_names[qtype] = NULL;
1367                 return -1;
1368         }
1369         set_opt(sb, QUOTA);
1370         return 1;
1371 }
1372
1373 static int clear_qf_name(struct super_block *sb, int qtype)
1374 {
1375
1376         struct ext4_sb_info *sbi = EXT4_SB(sb);
1377
1378         if (sb_any_quota_loaded(sb) &&
1379                 sbi->s_qf_names[qtype]) {
1380                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1381                         " when quota turned on");
1382                 return -1;
1383         }
1384         /*
1385          * The space will be released later when all options are confirmed
1386          * to be correct
1387          */
1388         sbi->s_qf_names[qtype] = NULL;
1389         return 1;
1390 }
1391 #endif
1392
1393 #define MOPT_SET        0x0001
1394 #define MOPT_CLEAR      0x0002
1395 #define MOPT_NOSUPPORT  0x0004
1396 #define MOPT_EXPLICIT   0x0008
1397 #define MOPT_CLEAR_ERR  0x0010
1398 #define MOPT_GTE0       0x0020
1399 #ifdef CONFIG_QUOTA
1400 #define MOPT_Q          0
1401 #define MOPT_QFMT       0x0040
1402 #else
1403 #define MOPT_Q          MOPT_NOSUPPORT
1404 #define MOPT_QFMT       MOPT_NOSUPPORT
1405 #endif
1406 #define MOPT_DATAJ      0x0080
1407
1408 static const struct mount_opts {
1409         int     token;
1410         int     mount_opt;
1411         int     flags;
1412 } ext4_mount_opts[] = {
1413         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1414         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1415         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1416         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1417         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1418         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1419         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1420         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1421         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1422         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1423         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1424         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1425         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1426         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1427         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1428         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1429                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1430         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1431         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1434         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1435         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1436         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1437         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1438         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1439         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1440         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1441         {Opt_commit, 0, MOPT_GTE0},
1442         {Opt_max_batch_time, 0, MOPT_GTE0},
1443         {Opt_min_batch_time, 0, MOPT_GTE0},
1444         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1445         {Opt_init_itable, 0, MOPT_GTE0},
1446         {Opt_stripe, 0, MOPT_GTE0},
1447         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1448         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1449         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1450         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1451         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1452 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1453         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1454         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1455 #else
1456         {Opt_acl, 0, MOPT_NOSUPPORT},
1457         {Opt_noacl, 0, MOPT_NOSUPPORT},
1458 #endif
1459         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1460         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1461         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1462         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1463                                                         MOPT_SET | MOPT_Q},
1464         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1465                                                         MOPT_SET | MOPT_Q},
1466         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1467                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1468         {Opt_usrjquota, 0, MOPT_Q},
1469         {Opt_grpjquota, 0, MOPT_Q},
1470         {Opt_offusrjquota, 0, MOPT_Q},
1471         {Opt_offgrpjquota, 0, MOPT_Q},
1472         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1473         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1474         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1475         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1476         {Opt_err, 0, 0}
1477 };
1478
1479 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1480                             substring_t *args, unsigned long *journal_devnum,
1481                             unsigned int *journal_ioprio, int is_remount)
1482 {
1483         struct ext4_sb_info *sbi = EXT4_SB(sb);
1484         const struct mount_opts *m;
1485         kuid_t uid;
1486         kgid_t gid;
1487         int arg = 0;
1488
1489 #ifdef CONFIG_QUOTA
1490         if (token == Opt_usrjquota)
1491                 return set_qf_name(sb, USRQUOTA, &args[0]);
1492         else if (token == Opt_grpjquota)
1493                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1494         else if (token == Opt_offusrjquota)
1495                 return clear_qf_name(sb, USRQUOTA);
1496         else if (token == Opt_offgrpjquota)
1497                 return clear_qf_name(sb, GRPQUOTA);
1498 #endif
1499         if (args->from && match_int(args, &arg))
1500                 return -1;
1501         switch (token) {
1502         case Opt_noacl:
1503         case Opt_nouser_xattr:
1504                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1505                 break;
1506         case Opt_sb:
1507                 return 1;       /* handled by get_sb_block() */
1508         case Opt_removed:
1509                 ext4_msg(sb, KERN_WARNING,
1510                          "Ignoring removed %s option", opt);
1511                 return 1;
1512         case Opt_resuid:
1513                 uid = make_kuid(current_user_ns(), arg);
1514                 if (!uid_valid(uid)) {
1515                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1516                         return -1;
1517                 }
1518                 sbi->s_resuid = uid;
1519                 return 1;
1520         case Opt_resgid:
1521                 gid = make_kgid(current_user_ns(), arg);
1522                 if (!gid_valid(gid)) {
1523                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1524                         return -1;
1525                 }
1526                 sbi->s_resgid = gid;
1527                 return 1;
1528         case Opt_abort:
1529                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1530                 return 1;
1531         case Opt_i_version:
1532                 sb->s_flags |= MS_I_VERSION;
1533                 return 1;
1534         case Opt_journal_dev:
1535                 if (is_remount) {
1536                         ext4_msg(sb, KERN_ERR,
1537                                  "Cannot specify journal on remount");
1538                         return -1;
1539                 }
1540                 *journal_devnum = arg;
1541                 return 1;
1542         case Opt_journal_ioprio:
1543                 if (arg < 0 || arg > 7)
1544                         return -1;
1545                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1546                 return 1;
1547         }
1548
1549         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1550                 if (token != m->token)
1551                         continue;
1552                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1553                         return -1;
1554                 if (m->flags & MOPT_EXPLICIT)
1555                         set_opt2(sb, EXPLICIT_DELALLOC);
1556                 if (m->flags & MOPT_CLEAR_ERR)
1557                         clear_opt(sb, ERRORS_MASK);
1558                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1559                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1560                                  "options when quota turned on");
1561                         return -1;
1562                 }
1563
1564                 if (m->flags & MOPT_NOSUPPORT) {
1565                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1566                 } else if (token == Opt_commit) {
1567                         if (arg == 0)
1568                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1569                         sbi->s_commit_interval = HZ * arg;
1570                 } else if (token == Opt_max_batch_time) {
1571                         if (arg == 0)
1572                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1573                         sbi->s_max_batch_time = arg;
1574                 } else if (token == Opt_min_batch_time) {
1575                         sbi->s_min_batch_time = arg;
1576                 } else if (token == Opt_inode_readahead_blks) {
1577                         if (arg > (1 << 30))
1578                                 return -1;
1579                         if (arg && !is_power_of_2(arg)) {
1580                                 ext4_msg(sb, KERN_ERR,
1581                                          "EXT4-fs: inode_readahead_blks"
1582                                          " must be a power of 2");
1583                                 return -1;
1584                         }
1585                         sbi->s_inode_readahead_blks = arg;
1586                 } else if (token == Opt_init_itable) {
1587                         set_opt(sb, INIT_INODE_TABLE);
1588                         if (!args->from)
1589                                 arg = EXT4_DEF_LI_WAIT_MULT;
1590                         sbi->s_li_wait_mult = arg;
1591                 } else if (token == Opt_max_dir_size_kb) {
1592                         sbi->s_max_dir_size_kb = arg;
1593                 } else if (token == Opt_stripe) {
1594                         sbi->s_stripe = arg;
1595                 } else if (m->flags & MOPT_DATAJ) {
1596                         if (is_remount) {
1597                                 if (!sbi->s_journal)
1598                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1599                                 else if (test_opt(sb, DATA_FLAGS) !=
1600                                          m->mount_opt) {
1601                                         ext4_msg(sb, KERN_ERR,
1602                                          "Cannot change data mode on remount");
1603                                         return -1;
1604                                 }
1605                         } else {
1606                                 clear_opt(sb, DATA_FLAGS);
1607                                 sbi->s_mount_opt |= m->mount_opt;
1608                         }
1609 #ifdef CONFIG_QUOTA
1610                 } else if (m->flags & MOPT_QFMT) {
1611                         if (sb_any_quota_loaded(sb) &&
1612                             sbi->s_jquota_fmt != m->mount_opt) {
1613                                 ext4_msg(sb, KERN_ERR, "Cannot "
1614                                          "change journaled quota options "
1615                                          "when quota turned on");
1616                                 return -1;
1617                         }
1618                         sbi->s_jquota_fmt = m->mount_opt;
1619 #endif
1620                 } else {
1621                         if (!args->from)
1622                                 arg = 1;
1623                         if (m->flags & MOPT_CLEAR)
1624                                 arg = !arg;
1625                         else if (unlikely(!(m->flags & MOPT_SET))) {
1626                                 ext4_msg(sb, KERN_WARNING,
1627                                          "buggy handling of option %s", opt);
1628                                 WARN_ON(1);
1629                                 return -1;
1630                         }
1631                         if (arg != 0)
1632                                 sbi->s_mount_opt |= m->mount_opt;
1633                         else
1634                                 sbi->s_mount_opt &= ~m->mount_opt;
1635                 }
1636                 return 1;
1637         }
1638         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1639                  "or missing value", opt);
1640         return -1;
1641 }
1642
1643 static int parse_options(char *options, struct super_block *sb,
1644                          unsigned long *journal_devnum,
1645                          unsigned int *journal_ioprio,
1646                          int is_remount)
1647 {
1648         struct ext4_sb_info *sbi = EXT4_SB(sb);
1649         char *p;
1650         substring_t args[MAX_OPT_ARGS];
1651         int token;
1652
1653         if (!options)
1654                 return 1;
1655
1656         while ((p = strsep(&options, ",")) != NULL) {
1657                 if (!*p)
1658                         continue;
1659                 /*
1660                  * Initialize args struct so we know whether arg was
1661                  * found; some options take optional arguments.
1662                  */
1663                 args[0].to = args[0].from = NULL;
1664                 token = match_token(p, tokens, args);
1665                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1666                                      journal_ioprio, is_remount) < 0)
1667                         return 0;
1668         }
1669 #ifdef CONFIG_QUOTA
1670         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1671                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1672                         clear_opt(sb, USRQUOTA);
1673
1674                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1675                         clear_opt(sb, GRPQUOTA);
1676
1677                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1678                         ext4_msg(sb, KERN_ERR, "old and new quota "
1679                                         "format mixing");
1680                         return 0;
1681                 }
1682
1683                 if (!sbi->s_jquota_fmt) {
1684                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1685                                         "not specified");
1686                         return 0;
1687                 }
1688         } else {
1689                 if (sbi->s_jquota_fmt) {
1690                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1691                                         "specified with no journaling "
1692                                         "enabled");
1693                         return 0;
1694                 }
1695         }
1696 #endif
1697         if (test_opt(sb, DIOREAD_NOLOCK)) {
1698                 int blocksize =
1699                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1700
1701                 if (blocksize < PAGE_CACHE_SIZE) {
1702                         ext4_msg(sb, KERN_ERR, "can't mount with "
1703                                  "dioread_nolock if block size != PAGE_SIZE");
1704                         return 0;
1705                 }
1706         }
1707         return 1;
1708 }
1709
1710 static inline void ext4_show_quota_options(struct seq_file *seq,
1711                                            struct super_block *sb)
1712 {
1713 #if defined(CONFIG_QUOTA)
1714         struct ext4_sb_info *sbi = EXT4_SB(sb);
1715
1716         if (sbi->s_jquota_fmt) {
1717                 char *fmtname = "";
1718
1719                 switch (sbi->s_jquota_fmt) {
1720                 case QFMT_VFS_OLD:
1721                         fmtname = "vfsold";
1722                         break;
1723                 case QFMT_VFS_V0:
1724                         fmtname = "vfsv0";
1725                         break;
1726                 case QFMT_VFS_V1:
1727                         fmtname = "vfsv1";
1728                         break;
1729                 }
1730                 seq_printf(seq, ",jqfmt=%s", fmtname);
1731         }
1732
1733         if (sbi->s_qf_names[USRQUOTA])
1734                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1735
1736         if (sbi->s_qf_names[GRPQUOTA])
1737                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1738
1739         if (test_opt(sb, USRQUOTA))
1740                 seq_puts(seq, ",usrquota");
1741
1742         if (test_opt(sb, GRPQUOTA))
1743                 seq_puts(seq, ",grpquota");
1744 #endif
1745 }
1746
1747 static const char *token2str(int token)
1748 {
1749         const struct match_token *t;
1750
1751         for (t = tokens; t->token != Opt_err; t++)
1752                 if (t->token == token && !strchr(t->pattern, '='))
1753                         break;
1754         return t->pattern;
1755 }
1756
1757 /*
1758  * Show an option if
1759  *  - it's set to a non-default value OR
1760  *  - if the per-sb default is different from the global default
1761  */
1762 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1763                               int nodefs)
1764 {
1765         struct ext4_sb_info *sbi = EXT4_SB(sb);
1766         struct ext4_super_block *es = sbi->s_es;
1767         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1768         const struct mount_opts *m;
1769         char sep = nodefs ? '\n' : ',';
1770
1771 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1772 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1773
1774         if (sbi->s_sb_block != 1)
1775                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1776
1777         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1778                 int want_set = m->flags & MOPT_SET;
1779                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1780                     (m->flags & MOPT_CLEAR_ERR))
1781                         continue;
1782                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1783                         continue; /* skip if same as the default */
1784                 if ((want_set &&
1785                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1786                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1787                         continue; /* select Opt_noFoo vs Opt_Foo */
1788                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1789         }
1790
1791         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1792             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1793                 SEQ_OPTS_PRINT("resuid=%u",
1794                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1795         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1796             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1797                 SEQ_OPTS_PRINT("resgid=%u",
1798                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1799         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1800         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1801                 SEQ_OPTS_PUTS("errors=remount-ro");
1802         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1803                 SEQ_OPTS_PUTS("errors=continue");
1804         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1805                 SEQ_OPTS_PUTS("errors=panic");
1806         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1807                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1808         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1809                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1810         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1811                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1812         if (sb->s_flags & MS_I_VERSION)
1813                 SEQ_OPTS_PUTS("i_version");
1814         if (nodefs || sbi->s_stripe)
1815                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1816         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1817                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1818                         SEQ_OPTS_PUTS("data=journal");
1819                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1820                         SEQ_OPTS_PUTS("data=ordered");
1821                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1822                         SEQ_OPTS_PUTS("data=writeback");
1823         }
1824         if (nodefs ||
1825             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1826                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1827                                sbi->s_inode_readahead_blks);
1828
1829         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1830                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1831                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1832         if (nodefs || sbi->s_max_dir_size_kb)
1833                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1834
1835         ext4_show_quota_options(seq, sb);
1836         return 0;
1837 }
1838
1839 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1840 {
1841         return _ext4_show_options(seq, root->d_sb, 0);
1842 }
1843
1844 static int options_seq_show(struct seq_file *seq, void *offset)
1845 {
1846         struct super_block *sb = seq->private;
1847         int rc;
1848
1849         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1850         rc = _ext4_show_options(seq, sb, 1);
1851         seq_puts(seq, "\n");
1852         return rc;
1853 }
1854
1855 static int options_open_fs(struct inode *inode, struct file *file)
1856 {
1857         return single_open(file, options_seq_show, PDE(inode)->data);
1858 }
1859
1860 static const struct file_operations ext4_seq_options_fops = {
1861         .owner = THIS_MODULE,
1862         .open = options_open_fs,
1863         .read = seq_read,
1864         .llseek = seq_lseek,
1865         .release = single_release,
1866 };
1867
1868 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1869                             int read_only)
1870 {
1871         struct ext4_sb_info *sbi = EXT4_SB(sb);
1872         int res = 0;
1873
1874         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1875                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1876                          "forcing read-only mode");
1877                 res = MS_RDONLY;
1878         }
1879         if (read_only)
1880                 goto done;
1881         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1882                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1883                          "running e2fsck is recommended");
1884         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1885                 ext4_msg(sb, KERN_WARNING,
1886                          "warning: mounting fs with errors, "
1887                          "running e2fsck is recommended");
1888         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1889                  le16_to_cpu(es->s_mnt_count) >=
1890                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "warning: maximal mount count reached, "
1893                          "running e2fsck is recommended");
1894         else if (le32_to_cpu(es->s_checkinterval) &&
1895                 (le32_to_cpu(es->s_lastcheck) +
1896                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1897                 ext4_msg(sb, KERN_WARNING,
1898                          "warning: checktime reached, "
1899                          "running e2fsck is recommended");
1900         if (!sbi->s_journal)
1901                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1902         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1903                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1904         le16_add_cpu(&es->s_mnt_count, 1);
1905         es->s_mtime = cpu_to_le32(get_seconds());
1906         ext4_update_dynamic_rev(sb);
1907         if (sbi->s_journal)
1908                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1909
1910         ext4_commit_super(sb, 1);
1911 done:
1912         if (test_opt(sb, DEBUG))
1913                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1914                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1915                         sb->s_blocksize,
1916                         sbi->s_groups_count,
1917                         EXT4_BLOCKS_PER_GROUP(sb),
1918                         EXT4_INODES_PER_GROUP(sb),
1919                         sbi->s_mount_opt, sbi->s_mount_opt2);
1920
1921         cleancache_init_fs(sb);
1922         return res;
1923 }
1924
1925 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1926 {
1927         struct ext4_sb_info *sbi = EXT4_SB(sb);
1928         struct flex_groups *new_groups;
1929         int size;
1930
1931         if (!sbi->s_log_groups_per_flex)
1932                 return 0;
1933
1934         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1935         if (size <= sbi->s_flex_groups_allocated)
1936                 return 0;
1937
1938         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1939         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1940         if (!new_groups) {
1941                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1942                          size / (int) sizeof(struct flex_groups));
1943                 return -ENOMEM;
1944         }
1945
1946         if (sbi->s_flex_groups) {
1947                 memcpy(new_groups, sbi->s_flex_groups,
1948                        (sbi->s_flex_groups_allocated *
1949                         sizeof(struct flex_groups)));
1950                 ext4_kvfree(sbi->s_flex_groups);
1951         }
1952         sbi->s_flex_groups = new_groups;
1953         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1954         return 0;
1955 }
1956
1957 static int ext4_fill_flex_info(struct super_block *sb)
1958 {
1959         struct ext4_sb_info *sbi = EXT4_SB(sb);
1960         struct ext4_group_desc *gdp = NULL;
1961         ext4_group_t flex_group;
1962         unsigned int groups_per_flex = 0;
1963         int i, err;
1964
1965         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1966         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1967                 sbi->s_log_groups_per_flex = 0;
1968                 return 1;
1969         }
1970         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1971
1972         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1973         if (err)
1974                 goto failed;
1975
1976         for (i = 0; i < sbi->s_groups_count; i++) {
1977                 gdp = ext4_get_group_desc(sb, i, NULL);
1978
1979                 flex_group = ext4_flex_group(sbi, i);
1980                 atomic_add(ext4_free_inodes_count(sb, gdp),
1981                            &sbi->s_flex_groups[flex_group].free_inodes);
1982                 atomic_add(ext4_free_group_clusters(sb, gdp),
1983                            &sbi->s_flex_groups[flex_group].free_clusters);
1984                 atomic_add(ext4_used_dirs_count(sb, gdp),
1985                            &sbi->s_flex_groups[flex_group].used_dirs);
1986         }
1987
1988         return 1;
1989 failed:
1990         return 0;
1991 }
1992
1993 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1994                                    struct ext4_group_desc *gdp)
1995 {
1996         int offset;
1997         __u16 crc = 0;
1998         __le32 le_group = cpu_to_le32(block_group);
1999
2000         if ((sbi->s_es->s_feature_ro_compat &
2001              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2002                 /* Use new metadata_csum algorithm */
2003                 __u16 old_csum;
2004                 __u32 csum32;
2005
2006                 old_csum = gdp->bg_checksum;
2007                 gdp->bg_checksum = 0;
2008                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2009                                      sizeof(le_group));
2010                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2011                                      sbi->s_desc_size);
2012                 gdp->bg_checksum = old_csum;
2013
2014                 crc = csum32 & 0xFFFF;
2015                 goto out;
2016         }
2017
2018         /* old crc16 code */
2019         offset = offsetof(struct ext4_group_desc, bg_checksum);
2020
2021         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2022         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2023         crc = crc16(crc, (__u8 *)gdp, offset);
2024         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2025         /* for checksum of struct ext4_group_desc do the rest...*/
2026         if ((sbi->s_es->s_feature_incompat &
2027              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2028             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2029                 crc = crc16(crc, (__u8 *)gdp + offset,
2030                             le16_to_cpu(sbi->s_es->s_desc_size) -
2031                                 offset);
2032
2033 out:
2034         return cpu_to_le16(crc);
2035 }
2036
2037 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2038                                 struct ext4_group_desc *gdp)
2039 {
2040         if (ext4_has_group_desc_csum(sb) &&
2041             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2042                                                       block_group, gdp)))
2043                 return 0;
2044
2045         return 1;
2046 }
2047
2048 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2049                               struct ext4_group_desc *gdp)
2050 {
2051         if (!ext4_has_group_desc_csum(sb))
2052                 return;
2053         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2054 }
2055
2056 /* Called at mount-time, super-block is locked */
2057 static int ext4_check_descriptors(struct super_block *sb,
2058                                   ext4_group_t *first_not_zeroed)
2059 {
2060         struct ext4_sb_info *sbi = EXT4_SB(sb);
2061         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2062         ext4_fsblk_t last_block;
2063         ext4_fsblk_t block_bitmap;
2064         ext4_fsblk_t inode_bitmap;
2065         ext4_fsblk_t inode_table;
2066         int flexbg_flag = 0;
2067         ext4_group_t i, grp = sbi->s_groups_count;
2068
2069         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2070                 flexbg_flag = 1;
2071
2072         ext4_debug("Checking group descriptors");
2073
2074         for (i = 0; i < sbi->s_groups_count; i++) {
2075                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2076
2077                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2078                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2079                 else
2080                         last_block = first_block +
2081                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2082
2083                 if ((grp == sbi->s_groups_count) &&
2084                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2085                         grp = i;
2086
2087                 block_bitmap = ext4_block_bitmap(sb, gdp);
2088                 if (block_bitmap < first_block || block_bitmap > last_block) {
2089                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2090                                "Block bitmap for group %u not in group "
2091                                "(block %llu)!", i, block_bitmap);
2092                         return 0;
2093                 }
2094                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2095                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2096                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2097                                "Inode bitmap for group %u not in group "
2098                                "(block %llu)!", i, inode_bitmap);
2099                         return 0;
2100                 }
2101                 inode_table = ext4_inode_table(sb, gdp);
2102                 if (inode_table < first_block ||
2103                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2104                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2105                                "Inode table for group %u not in group "
2106                                "(block %llu)!", i, inode_table);
2107                         return 0;
2108                 }
2109                 ext4_lock_group(sb, i);
2110                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2111                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2112                                  "Checksum for group %u failed (%u!=%u)",
2113                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2114                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2115                         if (!(sb->s_flags & MS_RDONLY)) {
2116                                 ext4_unlock_group(sb, i);
2117                                 return 0;
2118                         }
2119                 }
2120                 ext4_unlock_group(sb, i);
2121                 if (!flexbg_flag)
2122                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2123         }
2124         if (NULL != first_not_zeroed)
2125                 *first_not_zeroed = grp;
2126
2127         ext4_free_blocks_count_set(sbi->s_es,
2128                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2129         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2130         return 1;
2131 }
2132
2133 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2134  * the superblock) which were deleted from all directories, but held open by
2135  * a process at the time of a crash.  We walk the list and try to delete these
2136  * inodes at recovery time (only with a read-write filesystem).
2137  *
2138  * In order to keep the orphan inode chain consistent during traversal (in
2139  * case of crash during recovery), we link each inode into the superblock
2140  * orphan list_head and handle it the same way as an inode deletion during
2141  * normal operation (which journals the operations for us).
2142  *
2143  * We only do an iget() and an iput() on each inode, which is very safe if we
2144  * accidentally point at an in-use or already deleted inode.  The worst that
2145  * can happen in this case is that we get a "bit already cleared" message from
2146  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2147  * e2fsck was run on this filesystem, and it must have already done the orphan
2148  * inode cleanup for us, so we can safely abort without any further action.
2149  */
2150 static void ext4_orphan_cleanup(struct super_block *sb,
2151                                 struct ext4_super_block *es)
2152 {
2153         unsigned int s_flags = sb->s_flags;
2154         int nr_orphans = 0, nr_truncates = 0;
2155 #ifdef CONFIG_QUOTA
2156         int i;
2157 #endif
2158         if (!es->s_last_orphan) {
2159                 jbd_debug(4, "no orphan inodes to clean up\n");
2160                 return;
2161         }
2162
2163         if (bdev_read_only(sb->s_bdev)) {
2164                 ext4_msg(sb, KERN_ERR, "write access "
2165                         "unavailable, skipping orphan cleanup");
2166                 return;
2167         }
2168
2169         /* Check if feature set would not allow a r/w mount */
2170         if (!ext4_feature_set_ok(sb, 0)) {
2171                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2172                          "unknown ROCOMPAT features");
2173                 return;
2174         }
2175
2176         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2177                 /* don't clear list on RO mount w/ errors */
2178                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2179                         jbd_debug(1, "Errors on filesystem, "
2180                                   "clearing orphan list.\n");
2181                         es->s_last_orphan = 0;
2182                 }
2183                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2184                 return;
2185         }
2186
2187         if (s_flags & MS_RDONLY) {
2188                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2189                 sb->s_flags &= ~MS_RDONLY;
2190         }
2191 #ifdef CONFIG_QUOTA
2192         /* Needed for iput() to work correctly and not trash data */
2193         sb->s_flags |= MS_ACTIVE;
2194         /* Turn on quotas so that they are updated correctly */
2195         for (i = 0; i < MAXQUOTAS; i++) {
2196                 if (EXT4_SB(sb)->s_qf_names[i]) {
2197                         int ret = ext4_quota_on_mount(sb, i);
2198                         if (ret < 0)
2199                                 ext4_msg(sb, KERN_ERR,
2200                                         "Cannot turn on journaled "
2201                                         "quota: error %d", ret);
2202                 }
2203         }
2204 #endif
2205
2206         while (es->s_last_orphan) {
2207                 struct inode *inode;
2208
2209                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2210                 if (IS_ERR(inode)) {
2211                         es->s_last_orphan = 0;
2212                         break;
2213                 }
2214
2215                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2216                 dquot_initialize(inode);
2217                 if (inode->i_nlink) {
2218                         ext4_msg(sb, KERN_DEBUG,
2219                                 "%s: truncating inode %lu to %lld bytes",
2220                                 __func__, inode->i_ino, inode->i_size);
2221                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2222                                   inode->i_ino, inode->i_size);
2223                         mutex_lock(&inode->i_mutex);
2224                         ext4_truncate(inode);
2225                         mutex_unlock(&inode->i_mutex);
2226                         nr_truncates++;
2227                 } else {
2228                         ext4_msg(sb, KERN_DEBUG,
2229                                 "%s: deleting unreferenced inode %lu",
2230                                 __func__, inode->i_ino);
2231                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2232                                   inode->i_ino);
2233                         nr_orphans++;
2234                 }
2235                 iput(inode);  /* The delete magic happens here! */
2236         }
2237
2238 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2239
2240         if (nr_orphans)
2241                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2242                        PLURAL(nr_orphans));
2243         if (nr_truncates)
2244                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2245                        PLURAL(nr_truncates));
2246 #ifdef CONFIG_QUOTA
2247         /* Turn quotas off */
2248         for (i = 0; i < MAXQUOTAS; i++) {
2249                 if (sb_dqopt(sb)->files[i])
2250                         dquot_quota_off(sb, i);
2251         }
2252 #endif
2253         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2254 }
2255
2256 /*
2257  * Maximal extent format file size.
2258  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2259  * extent format containers, within a sector_t, and within i_blocks
2260  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2261  * so that won't be a limiting factor.
2262  *
2263  * However there is other limiting factor. We do store extents in the form
2264  * of starting block and length, hence the resulting length of the extent
2265  * covering maximum file size must fit into on-disk format containers as
2266  * well. Given that length is always by 1 unit bigger than max unit (because
2267  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2268  *
2269  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2270  */
2271 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2272 {
2273         loff_t res;
2274         loff_t upper_limit = MAX_LFS_FILESIZE;
2275
2276         /* small i_blocks in vfs inode? */
2277         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2278                 /*
2279                  * CONFIG_LBDAF is not enabled implies the inode
2280                  * i_block represent total blocks in 512 bytes
2281                  * 32 == size of vfs inode i_blocks * 8
2282                  */
2283                 upper_limit = (1LL << 32) - 1;
2284
2285                 /* total blocks in file system block size */
2286                 upper_limit >>= (blkbits - 9);
2287                 upper_limit <<= blkbits;
2288         }
2289
2290         /*
2291          * 32-bit extent-start container, ee_block. We lower the maxbytes
2292          * by one fs block, so ee_len can cover the extent of maximum file
2293          * size
2294          */
2295         res = (1LL << 32) - 1;
2296         res <<= blkbits;
2297
2298         /* Sanity check against vm- & vfs- imposed limits */
2299         if (res > upper_limit)
2300                 res = upper_limit;
2301
2302         return res;
2303 }
2304
2305 /*
2306  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2307  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2308  * We need to be 1 filesystem block less than the 2^48 sector limit.
2309  */
2310 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2311 {
2312         loff_t res = EXT4_NDIR_BLOCKS;
2313         int meta_blocks;
2314         loff_t upper_limit;
2315         /* This is calculated to be the largest file size for a dense, block
2316          * mapped file such that the file's total number of 512-byte sectors,
2317          * including data and all indirect blocks, does not exceed (2^48 - 1).
2318          *
2319          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2320          * number of 512-byte sectors of the file.
2321          */
2322
2323         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2324                 /*
2325                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2326                  * the inode i_block field represents total file blocks in
2327                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2328                  */
2329                 upper_limit = (1LL << 32) - 1;
2330
2331                 /* total blocks in file system block size */
2332                 upper_limit >>= (bits - 9);
2333
2334         } else {
2335                 /*
2336                  * We use 48 bit ext4_inode i_blocks
2337                  * With EXT4_HUGE_FILE_FL set the i_blocks
2338                  * represent total number of blocks in
2339                  * file system block size
2340                  */
2341                 upper_limit = (1LL << 48) - 1;
2342
2343         }
2344
2345         /* indirect blocks */
2346         meta_blocks = 1;
2347         /* double indirect blocks */
2348         meta_blocks += 1 + (1LL << (bits-2));
2349         /* tripple indirect blocks */
2350         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2351
2352         upper_limit -= meta_blocks;
2353         upper_limit <<= bits;
2354
2355         res += 1LL << (bits-2);
2356         res += 1LL << (2*(bits-2));
2357         res += 1LL << (3*(bits-2));
2358         res <<= bits;
2359         if (res > upper_limit)
2360                 res = upper_limit;
2361
2362         if (res > MAX_LFS_FILESIZE)
2363                 res = MAX_LFS_FILESIZE;
2364
2365         return res;
2366 }
2367
2368 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2369                                    ext4_fsblk_t logical_sb_block, int nr)
2370 {
2371         struct ext4_sb_info *sbi = EXT4_SB(sb);
2372         ext4_group_t bg, first_meta_bg;
2373         int has_super = 0;
2374
2375         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2376
2377         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2378             nr < first_meta_bg)
2379                 return logical_sb_block + nr + 1;
2380         bg = sbi->s_desc_per_block * nr;
2381         if (ext4_bg_has_super(sb, bg))
2382                 has_super = 1;
2383
2384         return (has_super + ext4_group_first_block_no(sb, bg));
2385 }
2386
2387 /**
2388  * ext4_get_stripe_size: Get the stripe size.
2389  * @sbi: In memory super block info
2390  *
2391  * If we have specified it via mount option, then
2392  * use the mount option value. If the value specified at mount time is
2393  * greater than the blocks per group use the super block value.
2394  * If the super block value is greater than blocks per group return 0.
2395  * Allocator needs it be less than blocks per group.
2396  *
2397  */
2398 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2399 {
2400         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2401         unsigned long stripe_width =
2402                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2403         int ret;
2404
2405         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2406                 ret = sbi->s_stripe;
2407         else if (stripe_width <= sbi->s_blocks_per_group)
2408                 ret = stripe_width;
2409         else if (stride <= sbi->s_blocks_per_group)
2410                 ret = stride;
2411         else
2412                 ret = 0;
2413
2414         /*
2415          * If the stripe width is 1, this makes no sense and
2416          * we set it to 0 to turn off stripe handling code.
2417          */
2418         if (ret <= 1)
2419                 ret = 0;
2420
2421         return ret;
2422 }
2423
2424 /* sysfs supprt */
2425
2426 struct ext4_attr {
2427         struct attribute attr;
2428         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2429         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2430                          const char *, size_t);
2431         int offset;
2432 };
2433
2434 static int parse_strtoul(const char *buf,
2435                 unsigned long max, unsigned long *value)
2436 {
2437         char *endp;
2438
2439         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2440         endp = skip_spaces(endp);
2441         if (*endp || *value > max)
2442                 return -EINVAL;
2443
2444         return 0;
2445 }
2446
2447 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2448                                               struct ext4_sb_info *sbi,
2449                                               char *buf)
2450 {
2451         return snprintf(buf, PAGE_SIZE, "%llu\n",
2452                 (s64) EXT4_C2B(sbi,
2453                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2454 }
2455
2456 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2457                                          struct ext4_sb_info *sbi, char *buf)
2458 {
2459         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2460
2461         if (!sb->s_bdev->bd_part)
2462                 return snprintf(buf, PAGE_SIZE, "0\n");
2463         return snprintf(buf, PAGE_SIZE, "%lu\n",
2464                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2465                          sbi->s_sectors_written_start) >> 1);
2466 }
2467
2468 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2469                                           struct ext4_sb_info *sbi, char *buf)
2470 {
2471         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2472
2473         if (!sb->s_bdev->bd_part)
2474                 return snprintf(buf, PAGE_SIZE, "0\n");
2475         return snprintf(buf, PAGE_SIZE, "%llu\n",
2476                         (unsigned long long)(sbi->s_kbytes_written +
2477                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2478                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2479 }
2480
2481 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2482                                           struct ext4_sb_info *sbi,
2483                                           const char *buf, size_t count)
2484 {
2485         unsigned long t;
2486
2487         if (parse_strtoul(buf, 0x40000000, &t))
2488                 return -EINVAL;
2489
2490         if (t && !is_power_of_2(t))
2491                 return -EINVAL;
2492
2493         sbi->s_inode_readahead_blks = t;
2494         return count;
2495 }
2496
2497 static ssize_t sbi_ui_show(struct ext4_attr *a,
2498                            struct ext4_sb_info *sbi, char *buf)
2499 {
2500         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2501
2502         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2503 }
2504
2505 static ssize_t sbi_ui_store(struct ext4_attr *a,
2506                             struct ext4_sb_info *sbi,
2507                             const char *buf, size_t count)
2508 {
2509         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2510         unsigned long t;
2511
2512         if (parse_strtoul(buf, 0xffffffff, &t))
2513                 return -EINVAL;
2514         *ui = t;
2515         return count;
2516 }
2517
2518 static ssize_t trigger_test_error(struct ext4_attr *a,
2519                                   struct ext4_sb_info *sbi,
2520                                   const char *buf, size_t count)
2521 {
2522         int len = count;
2523
2524         if (!capable(CAP_SYS_ADMIN))
2525                 return -EPERM;
2526
2527         if (len && buf[len-1] == '\n')
2528                 len--;
2529
2530         if (len)
2531                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2532         return count;
2533 }
2534
2535 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2536 static struct ext4_attr ext4_attr_##_name = {                   \
2537         .attr = {.name = __stringify(_name), .mode = _mode },   \
2538         .show   = _show,                                        \
2539         .store  = _store,                                       \
2540         .offset = offsetof(struct ext4_sb_info, _elname),       \
2541 }
2542 #define EXT4_ATTR(name, mode, show, store) \
2543 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2544
2545 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2546 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2547 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2548 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2549         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2550 #define ATTR_LIST(name) &ext4_attr_##name.attr
2551
2552 EXT4_RO_ATTR(delayed_allocation_blocks);
2553 EXT4_RO_ATTR(session_write_kbytes);
2554 EXT4_RO_ATTR(lifetime_write_kbytes);
2555 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2556                  inode_readahead_blks_store, s_inode_readahead_blks);
2557 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2558 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2559 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2560 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2561 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2562 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2563 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2564 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2565 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2566 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2567
2568 static struct attribute *ext4_attrs[] = {
2569         ATTR_LIST(delayed_allocation_blocks),
2570         ATTR_LIST(session_write_kbytes),
2571         ATTR_LIST(lifetime_write_kbytes),
2572         ATTR_LIST(inode_readahead_blks),
2573         ATTR_LIST(inode_goal),
2574         ATTR_LIST(mb_stats),
2575         ATTR_LIST(mb_max_to_scan),
2576         ATTR_LIST(mb_min_to_scan),
2577         ATTR_LIST(mb_order2_req),
2578         ATTR_LIST(mb_stream_req),
2579         ATTR_LIST(mb_group_prealloc),
2580         ATTR_LIST(max_writeback_mb_bump),
2581         ATTR_LIST(extent_max_zeroout_kb),
2582         ATTR_LIST(trigger_fs_error),
2583         NULL,
2584 };
2585
2586 /* Features this copy of ext4 supports */
2587 EXT4_INFO_ATTR(lazy_itable_init);
2588 EXT4_INFO_ATTR(batched_discard);
2589 EXT4_INFO_ATTR(meta_bg_resize);
2590
2591 static struct attribute *ext4_feat_attrs[] = {
2592         ATTR_LIST(lazy_itable_init),
2593         ATTR_LIST(batched_discard),
2594         ATTR_LIST(meta_bg_resize),
2595         NULL,
2596 };
2597
2598 static ssize_t ext4_attr_show(struct kobject *kobj,
2599                               struct attribute *attr, char *buf)
2600 {
2601         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2602                                                 s_kobj);
2603         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2604
2605         return a->show ? a->show(a, sbi, buf) : 0;
2606 }
2607
2608 static ssize_t ext4_attr_store(struct kobject *kobj,
2609                                struct attribute *attr,
2610                                const char *buf, size_t len)
2611 {
2612         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2613                                                 s_kobj);
2614         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2615
2616         return a->store ? a->store(a, sbi, buf, len) : 0;
2617 }
2618
2619 static void ext4_sb_release(struct kobject *kobj)
2620 {
2621         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2622                                                 s_kobj);
2623         complete(&sbi->s_kobj_unregister);
2624 }
2625
2626 static const struct sysfs_ops ext4_attr_ops = {
2627         .show   = ext4_attr_show,
2628         .store  = ext4_attr_store,
2629 };
2630
2631 static struct kobj_type ext4_ktype = {
2632         .default_attrs  = ext4_attrs,
2633         .sysfs_ops      = &ext4_attr_ops,
2634         .release        = ext4_sb_release,
2635 };
2636
2637 static void ext4_feat_release(struct kobject *kobj)
2638 {
2639         complete(&ext4_feat->f_kobj_unregister);
2640 }
2641
2642 static struct kobj_type ext4_feat_ktype = {
2643         .default_attrs  = ext4_feat_attrs,
2644         .sysfs_ops      = &ext4_attr_ops,
2645         .release        = ext4_feat_release,
2646 };
2647
2648 /*
2649  * Check whether this filesystem can be mounted based on
2650  * the features present and the RDONLY/RDWR mount requested.
2651  * Returns 1 if this filesystem can be mounted as requested,
2652  * 0 if it cannot be.
2653  */
2654 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2655 {
2656         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2657                 ext4_msg(sb, KERN_ERR,
2658                         "Couldn't mount because of "
2659                         "unsupported optional features (%x)",
2660                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2661                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2662                 return 0;
2663         }
2664
2665         if (readonly)
2666                 return 1;
2667
2668         /* Check that feature set is OK for a read-write mount */
2669         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2670                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2671                          "unsupported optional features (%x)",
2672                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2673                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2674                 return 0;
2675         }
2676         /*
2677          * Large file size enabled file system can only be mounted
2678          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2679          */
2680         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2681                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2682                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2683                                  "cannot be mounted RDWR without "
2684                                  "CONFIG_LBDAF");
2685                         return 0;
2686                 }
2687         }
2688         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2689             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2690                 ext4_msg(sb, KERN_ERR,
2691                          "Can't support bigalloc feature without "
2692                          "extents feature\n");
2693                 return 0;
2694         }
2695
2696 #ifndef CONFIG_QUOTA
2697         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2698             !readonly) {
2699                 ext4_msg(sb, KERN_ERR,
2700                          "Filesystem with quota feature cannot be mounted RDWR "
2701                          "without CONFIG_QUOTA");
2702                 return 0;
2703         }
2704 #endif  /* CONFIG_QUOTA */
2705         return 1;
2706 }
2707
2708 /*
2709  * This function is called once a day if we have errors logged
2710  * on the file system
2711  */
2712 static void print_daily_error_info(unsigned long arg)
2713 {
2714         struct super_block *sb = (struct super_block *) arg;
2715         struct ext4_sb_info *sbi;
2716         struct ext4_super_block *es;
2717
2718         sbi = EXT4_SB(sb);
2719         es = sbi->s_es;
2720
2721         if (es->s_error_count)
2722                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2723                          le32_to_cpu(es->s_error_count));
2724         if (es->s_first_error_time) {
2725                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2726                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2727                        (int) sizeof(es->s_first_error_func),
2728                        es->s_first_error_func,
2729                        le32_to_cpu(es->s_first_error_line));
2730                 if (es->s_first_error_ino)
2731                         printk(": inode %u",
2732                                le32_to_cpu(es->s_first_error_ino));
2733                 if (es->s_first_error_block)
2734                         printk(": block %llu", (unsigned long long)
2735                                le64_to_cpu(es->s_first_error_block));
2736                 printk("\n");
2737         }
2738         if (es->s_last_error_time) {
2739                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2740                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2741                        (int) sizeof(es->s_last_error_func),
2742                        es->s_last_error_func,
2743                        le32_to_cpu(es->s_last_error_line));
2744                 if (es->s_last_error_ino)
2745                         printk(": inode %u",
2746                                le32_to_cpu(es->s_last_error_ino));
2747                 if (es->s_last_error_block)
2748                         printk(": block %llu", (unsigned long long)
2749                                le64_to_cpu(es->s_last_error_block));
2750                 printk("\n");
2751         }
2752         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2753 }
2754
2755 /* Find next suitable group and run ext4_init_inode_table */
2756 static int ext4_run_li_request(struct ext4_li_request *elr)
2757 {
2758         struct ext4_group_desc *gdp = NULL;
2759         ext4_group_t group, ngroups;
2760         struct super_block *sb;
2761         unsigned long timeout = 0;
2762         int ret = 0;
2763
2764         sb = elr->lr_super;
2765         ngroups = EXT4_SB(sb)->s_groups_count;
2766
2767         sb_start_write(sb);
2768         for (group = elr->lr_next_group; group < ngroups; group++) {
2769                 gdp = ext4_get_group_desc(sb, group, NULL);
2770                 if (!gdp) {
2771                         ret = 1;
2772                         break;
2773                 }
2774
2775                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2776                         break;
2777         }
2778
2779         if (group == ngroups)
2780                 ret = 1;
2781
2782         if (!ret) {
2783                 timeout = jiffies;
2784                 ret = ext4_init_inode_table(sb, group,
2785                                             elr->lr_timeout ? 0 : 1);
2786                 if (elr->lr_timeout == 0) {
2787                         timeout = (jiffies - timeout) *
2788                                   elr->lr_sbi->s_li_wait_mult;
2789                         elr->lr_timeout = timeout;
2790                 }
2791                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2792                 elr->lr_next_group = group + 1;
2793         }
2794         sb_end_write(sb);
2795
2796         return ret;
2797 }
2798
2799 /*
2800  * Remove lr_request from the list_request and free the
2801  * request structure. Should be called with li_list_mtx held
2802  */
2803 static void ext4_remove_li_request(struct ext4_li_request *elr)
2804 {
2805         struct ext4_sb_info *sbi;
2806
2807         if (!elr)
2808                 return;
2809
2810         sbi = elr->lr_sbi;
2811
2812         list_del(&elr->lr_request);
2813         sbi->s_li_request = NULL;
2814         kfree(elr);
2815 }
2816
2817 static void ext4_unregister_li_request(struct super_block *sb)
2818 {
2819         mutex_lock(&ext4_li_mtx);
2820         if (!ext4_li_info) {
2821                 mutex_unlock(&ext4_li_mtx);
2822                 return;
2823         }
2824
2825         mutex_lock(&ext4_li_info->li_list_mtx);
2826         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2827         mutex_unlock(&ext4_li_info->li_list_mtx);
2828         mutex_unlock(&ext4_li_mtx);
2829 }
2830
2831 static struct task_struct *ext4_lazyinit_task;
2832
2833 /*
2834  * This is the function where ext4lazyinit thread lives. It walks
2835  * through the request list searching for next scheduled filesystem.
2836  * When such a fs is found, run the lazy initialization request
2837  * (ext4_rn_li_request) and keep track of the time spend in this
2838  * function. Based on that time we compute next schedule time of
2839  * the request. When walking through the list is complete, compute
2840  * next waking time and put itself into sleep.
2841  */
2842 static int ext4_lazyinit_thread(void *arg)
2843 {
2844         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2845         struct list_head *pos, *n;
2846         struct ext4_li_request *elr;
2847         unsigned long next_wakeup, cur;
2848
2849         BUG_ON(NULL == eli);
2850
2851 cont_thread:
2852         while (true) {
2853                 next_wakeup = MAX_JIFFY_OFFSET;
2854
2855                 mutex_lock(&eli->li_list_mtx);
2856                 if (list_empty(&eli->li_request_list)) {
2857                         mutex_unlock(&eli->li_list_mtx);
2858                         goto exit_thread;
2859                 }
2860
2861                 list_for_each_safe(pos, n, &eli->li_request_list) {
2862                         elr = list_entry(pos, struct ext4_li_request,
2863                                          lr_request);
2864
2865                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2866                                 if (ext4_run_li_request(elr) != 0) {
2867                                         /* error, remove the lazy_init job */
2868                                         ext4_remove_li_request(elr);
2869                                         continue;
2870                                 }
2871                         }
2872
2873                         if (time_before(elr->lr_next_sched, next_wakeup))
2874                                 next_wakeup = elr->lr_next_sched;
2875                 }
2876                 mutex_unlock(&eli->li_list_mtx);
2877
2878                 try_to_freeze();
2879
2880                 cur = jiffies;
2881                 if ((time_after_eq(cur, next_wakeup)) ||
2882                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2883                         cond_resched();
2884                         continue;
2885                 }
2886
2887                 schedule_timeout_interruptible(next_wakeup - cur);
2888
2889                 if (kthread_should_stop()) {
2890                         ext4_clear_request_list();
2891                         goto exit_thread;
2892                 }
2893         }
2894
2895 exit_thread:
2896         /*
2897          * It looks like the request list is empty, but we need
2898          * to check it under the li_list_mtx lock, to prevent any
2899          * additions into it, and of course we should lock ext4_li_mtx
2900          * to atomically free the list and ext4_li_info, because at
2901          * this point another ext4 filesystem could be registering
2902          * new one.
2903          */
2904         mutex_lock(&ext4_li_mtx);
2905         mutex_lock(&eli->li_list_mtx);
2906         if (!list_empty(&eli->li_request_list)) {
2907                 mutex_unlock(&eli->li_list_mtx);
2908                 mutex_unlock(&ext4_li_mtx);
2909                 goto cont_thread;
2910         }
2911         mutex_unlock(&eli->li_list_mtx);
2912         kfree(ext4_li_info);
2913         ext4_li_info = NULL;
2914         mutex_unlock(&ext4_li_mtx);
2915
2916         return 0;
2917 }
2918
2919 static void ext4_clear_request_list(void)
2920 {
2921         struct list_head *pos, *n;
2922         struct ext4_li_request *elr;
2923
2924         mutex_lock(&ext4_li_info->li_list_mtx);
2925         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2926                 elr = list_entry(pos, struct ext4_li_request,
2927                                  lr_request);
2928                 ext4_remove_li_request(elr);
2929         }
2930         mutex_unlock(&ext4_li_info->li_list_mtx);
2931 }
2932
2933 static int ext4_run_lazyinit_thread(void)
2934 {
2935         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2936                                          ext4_li_info, "ext4lazyinit");
2937         if (IS_ERR(ext4_lazyinit_task)) {
2938                 int err = PTR_ERR(ext4_lazyinit_task);
2939                 ext4_clear_request_list();
2940                 kfree(ext4_li_info);
2941                 ext4_li_info = NULL;
2942                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2943                                  "initialization thread\n",
2944                                  err);
2945                 return err;
2946         }
2947         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2948         return 0;
2949 }
2950
2951 /*
2952  * Check whether it make sense to run itable init. thread or not.
2953  * If there is at least one uninitialized inode table, return
2954  * corresponding group number, else the loop goes through all
2955  * groups and return total number of groups.
2956  */
2957 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2958 {
2959         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2960         struct ext4_group_desc *gdp = NULL;
2961
2962         for (group = 0; group < ngroups; group++) {
2963                 gdp = ext4_get_group_desc(sb, group, NULL);
2964                 if (!gdp)
2965                         continue;
2966
2967                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2968                         break;
2969         }
2970
2971         return group;
2972 }
2973
2974 static int ext4_li_info_new(void)
2975 {
2976         struct ext4_lazy_init *eli = NULL;
2977
2978         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2979         if (!eli)
2980                 return -ENOMEM;
2981
2982         INIT_LIST_HEAD(&eli->li_request_list);
2983         mutex_init(&eli->li_list_mtx);
2984
2985         eli->li_state |= EXT4_LAZYINIT_QUIT;
2986
2987         ext4_li_info = eli;
2988
2989         return 0;
2990 }
2991
2992 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2993                                             ext4_group_t start)
2994 {
2995         struct ext4_sb_info *sbi = EXT4_SB(sb);
2996         struct ext4_li_request *elr;
2997         unsigned long rnd;
2998
2999         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3000         if (!elr)
3001                 return NULL;
3002
3003         elr->lr_super = sb;
3004         elr->lr_sbi = sbi;
3005         elr->lr_next_group = start;
3006
3007         /*
3008          * Randomize first schedule time of the request to
3009          * spread the inode table initialization requests
3010          * better.
3011          */
3012         get_random_bytes(&rnd, sizeof(rnd));
3013         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3014                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3015
3016         return elr;
3017 }
3018
3019 static int ext4_register_li_request(struct super_block *sb,
3020                                     ext4_group_t first_not_zeroed)
3021 {
3022         struct ext4_sb_info *sbi = EXT4_SB(sb);
3023         struct ext4_li_request *elr;
3024         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3025         int ret = 0;
3026
3027         if (sbi->s_li_request != NULL) {
3028                 /*
3029                  * Reset timeout so it can be computed again, because
3030                  * s_li_wait_mult might have changed.
3031                  */
3032                 sbi->s_li_request->lr_timeout = 0;
3033                 return 0;
3034         }
3035
3036         if (first_not_zeroed == ngroups ||
3037             (sb->s_flags & MS_RDONLY) ||
3038             !test_opt(sb, INIT_INODE_TABLE))
3039                 return 0;
3040
3041         elr = ext4_li_request_new(sb, first_not_zeroed);
3042         if (!elr)
3043                 return -ENOMEM;
3044
3045         mutex_lock(&ext4_li_mtx);
3046
3047         if (NULL == ext4_li_info) {
3048                 ret = ext4_li_info_new();
3049                 if (ret)
3050                         goto out;
3051         }
3052
3053         mutex_lock(&ext4_li_info->li_list_mtx);
3054         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3055         mutex_unlock(&ext4_li_info->li_list_mtx);
3056
3057         sbi->s_li_request = elr;
3058         /*
3059          * set elr to NULL here since it has been inserted to
3060          * the request_list and the removal and free of it is
3061          * handled by ext4_clear_request_list from now on.
3062          */
3063         elr = NULL;
3064
3065         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3066                 ret = ext4_run_lazyinit_thread();
3067                 if (ret)
3068                         goto out;
3069         }
3070 out:
3071         mutex_unlock(&ext4_li_mtx);
3072         if (ret)
3073                 kfree(elr);
3074         return ret;
3075 }
3076
3077 /*
3078  * We do not need to lock anything since this is called on
3079  * module unload.
3080  */
3081 static void ext4_destroy_lazyinit_thread(void)
3082 {
3083         /*
3084          * If thread exited earlier
3085          * there's nothing to be done.
3086          */
3087         if (!ext4_li_info || !ext4_lazyinit_task)
3088                 return;
3089
3090         kthread_stop(ext4_lazyinit_task);
3091 }
3092
3093 static int set_journal_csum_feature_set(struct super_block *sb)
3094 {
3095         int ret = 1;
3096         int compat, incompat;
3097         struct ext4_sb_info *sbi = EXT4_SB(sb);
3098
3099         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3100                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3101                 /* journal checksum v2 */
3102                 compat = 0;
3103                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3104         } else {
3105                 /* journal checksum v1 */
3106                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3107                 incompat = 0;
3108         }
3109
3110         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3111                 ret = jbd2_journal_set_features(sbi->s_journal,
3112                                 compat, 0,
3113                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3114                                 incompat);
3115         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3116                 ret = jbd2_journal_set_features(sbi->s_journal,
3117                                 compat, 0,
3118                                 incompat);
3119                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3120                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3121         } else {
3122                 jbd2_journal_clear_features(sbi->s_journal,
3123                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3124                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3125                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3126         }
3127
3128         return ret;
3129 }
3130
3131 /*
3132  * Note: calculating the overhead so we can be compatible with
3133  * historical BSD practice is quite difficult in the face of
3134  * clusters/bigalloc.  This is because multiple metadata blocks from
3135  * different block group can end up in the same allocation cluster.
3136  * Calculating the exact overhead in the face of clustered allocation
3137  * requires either O(all block bitmaps) in memory or O(number of block
3138  * groups**2) in time.  We will still calculate the superblock for
3139  * older file systems --- and if we come across with a bigalloc file
3140  * system with zero in s_overhead_clusters the estimate will be close to
3141  * correct especially for very large cluster sizes --- but for newer
3142  * file systems, it's better to calculate this figure once at mkfs
3143  * time, and store it in the superblock.  If the superblock value is
3144  * present (even for non-bigalloc file systems), we will use it.
3145  */
3146 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3147                           char *buf)
3148 {
3149         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3150         struct ext4_group_desc  *gdp;
3151         ext4_fsblk_t            first_block, last_block, b;
3152         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3153         int                     s, j, count = 0;
3154
3155         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3156                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3157                         sbi->s_itb_per_group + 2);
3158
3159         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3160                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3161         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3162         for (i = 0; i < ngroups; i++) {
3163                 gdp = ext4_get_group_desc(sb, i, NULL);
3164                 b = ext4_block_bitmap(sb, gdp);
3165                 if (b >= first_block && b <= last_block) {
3166                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3167                         count++;
3168                 }
3169                 b = ext4_inode_bitmap(sb, gdp);
3170                 if (b >= first_block && b <= last_block) {
3171                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3172                         count++;
3173                 }
3174                 b = ext4_inode_table(sb, gdp);
3175                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3176                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3177                                 int c = EXT4_B2C(sbi, b - first_block);
3178                                 ext4_set_bit(c, buf);
3179                                 count++;
3180                         }
3181                 if (i != grp)
3182                         continue;
3183                 s = 0;
3184                 if (ext4_bg_has_super(sb, grp)) {
3185                         ext4_set_bit(s++, buf);
3186                         count++;
3187                 }
3188                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3189                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3190                         count++;
3191                 }
3192         }
3193         if (!count)
3194                 return 0;
3195         return EXT4_CLUSTERS_PER_GROUP(sb) -
3196                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3197 }
3198
3199 /*
3200  * Compute the overhead and stash it in sbi->s_overhead
3201  */
3202 int ext4_calculate_overhead(struct super_block *sb)
3203 {
3204         struct ext4_sb_info *sbi = EXT4_SB(sb);
3205         struct ext4_super_block *es = sbi->s_es;
3206         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3207         ext4_fsblk_t overhead = 0;
3208         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3209
3210         if (!buf)
3211                 return -ENOMEM;
3212
3213         /*
3214          * Compute the overhead (FS structures).  This is constant
3215          * for a given filesystem unless the number of block groups
3216          * changes so we cache the previous value until it does.
3217          */
3218
3219         /*
3220          * All of the blocks before first_data_block are overhead
3221          */
3222         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3223
3224         /*
3225          * Add the overhead found in each block group
3226          */
3227         for (i = 0; i < ngroups; i++) {
3228                 int blks;
3229
3230                 blks = count_overhead(sb, i, buf);
3231                 overhead += blks;
3232                 if (blks)
3233                         memset(buf, 0, PAGE_SIZE);
3234                 cond_resched();
3235         }
3236         /* Add the journal blocks as well */
3237         if (sbi->s_journal)
3238                 overhead += EXT4_B2C(sbi, sbi->s_journal->j_maxlen);
3239
3240         sbi->s_overhead = overhead;
3241         smp_wmb();
3242         free_page((unsigned long) buf);
3243         return 0;
3244 }
3245
3246 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3247 {
3248         char *orig_data = kstrdup(data, GFP_KERNEL);
3249         struct buffer_head *bh;
3250         struct ext4_super_block *es = NULL;
3251         struct ext4_sb_info *sbi;
3252         ext4_fsblk_t block;
3253         ext4_fsblk_t sb_block = get_sb_block(&data);
3254         ext4_fsblk_t logical_sb_block;
3255         unsigned long offset = 0;
3256         unsigned long journal_devnum = 0;
3257         unsigned long def_mount_opts;
3258         struct inode *root;
3259         char *cp;
3260         const char *descr;
3261         int ret = -ENOMEM;
3262         int blocksize, clustersize;
3263         unsigned int db_count;
3264         unsigned int i;
3265         int needs_recovery, has_huge_files, has_bigalloc;
3266         __u64 blocks_count;
3267         int err = 0;
3268         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3269         ext4_group_t first_not_zeroed;
3270
3271         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3272         if (!sbi)
3273                 goto out_free_orig;
3274
3275         sbi->s_blockgroup_lock =
3276                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3277         if (!sbi->s_blockgroup_lock) {
3278                 kfree(sbi);
3279                 goto out_free_orig;
3280         }
3281         sb->s_fs_info = sbi;
3282         sbi->s_sb = sb;
3283         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3284         sbi->s_sb_block = sb_block;
3285         if (sb->s_bdev->bd_part)
3286                 sbi->s_sectors_written_start =
3287                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3288
3289         /* Cleanup superblock name */
3290         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3291                 *cp = '!';
3292
3293         /* -EINVAL is default */
3294         ret = -EINVAL;
3295         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3296         if (!blocksize) {
3297                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3298                 goto out_fail;
3299         }
3300
3301         /*
3302          * The ext4 superblock will not be buffer aligned for other than 1kB
3303          * block sizes.  We need to calculate the offset from buffer start.
3304          */
3305         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3306                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3307                 offset = do_div(logical_sb_block, blocksize);
3308         } else {
3309                 logical_sb_block = sb_block;
3310         }
3311
3312         if (!(bh = sb_bread(sb, logical_sb_block))) {
3313                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3314                 goto out_fail;
3315         }
3316         /*
3317          * Note: s_es must be initialized as soon as possible because
3318          *       some ext4 macro-instructions depend on its value
3319          */
3320         es = (struct ext4_super_block *) (bh->b_data + offset);
3321         sbi->s_es = es;
3322         sb->s_magic = le16_to_cpu(es->s_magic);
3323         if (sb->s_magic != EXT4_SUPER_MAGIC)
3324                 goto cantfind_ext4;
3325         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3326
3327         /* Warn if metadata_csum and gdt_csum are both set. */
3328         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3329                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3330             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3331                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3332                              "redundant flags; please run fsck.");
3333
3334         /* Check for a known checksum algorithm */
3335         if (!ext4_verify_csum_type(sb, es)) {
3336                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3337                          "unknown checksum algorithm.");
3338                 silent = 1;
3339                 goto cantfind_ext4;
3340         }
3341
3342         /* Load the checksum driver */
3343         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3344                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3345                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3346                 if (IS_ERR(sbi->s_chksum_driver)) {
3347                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3348                         ret = PTR_ERR(sbi->s_chksum_driver);
3349                         sbi->s_chksum_driver = NULL;
3350                         goto failed_mount;
3351                 }
3352         }
3353
3354         /* Check superblock checksum */
3355         if (!ext4_superblock_csum_verify(sb, es)) {
3356                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3357                          "invalid superblock checksum.  Run e2fsck?");
3358                 silent = 1;
3359                 goto cantfind_ext4;
3360         }
3361
3362         /* Precompute checksum seed for all metadata */
3363         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3364                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3365                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3366                                                sizeof(es->s_uuid));
3367
3368         /* Set defaults before we parse the mount options */
3369         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3370         set_opt(sb, INIT_INODE_TABLE);
3371         if (def_mount_opts & EXT4_DEFM_DEBUG)
3372                 set_opt(sb, DEBUG);
3373         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3374                 set_opt(sb, GRPID);
3375         if (def_mount_opts & EXT4_DEFM_UID16)
3376                 set_opt(sb, NO_UID32);
3377         /* xattr user namespace & acls are now defaulted on */
3378         set_opt(sb, XATTR_USER);
3379 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3380         set_opt(sb, POSIX_ACL);
3381 #endif
3382         set_opt(sb, MBLK_IO_SUBMIT);
3383         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3384                 set_opt(sb, JOURNAL_DATA);
3385         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3386                 set_opt(sb, ORDERED_DATA);
3387         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3388                 set_opt(sb, WRITEBACK_DATA);
3389
3390         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3391                 set_opt(sb, ERRORS_PANIC);
3392         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3393                 set_opt(sb, ERRORS_CONT);
3394         else
3395                 set_opt(sb, ERRORS_RO);
3396         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3397                 set_opt(sb, BLOCK_VALIDITY);
3398         if (def_mount_opts & EXT4_DEFM_DISCARD)
3399                 set_opt(sb, DISCARD);
3400
3401         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3402         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3403         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3404         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3405         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3406
3407         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3408                 set_opt(sb, BARRIER);
3409
3410         /*
3411          * enable delayed allocation by default
3412          * Use -o nodelalloc to turn it off
3413          */
3414         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3415             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3416                 set_opt(sb, DELALLOC);
3417
3418         /*
3419          * set default s_li_wait_mult for lazyinit, for the case there is
3420          * no mount option specified.
3421          */
3422         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3423
3424         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3425                            &journal_devnum, &journal_ioprio, 0)) {
3426                 ext4_msg(sb, KERN_WARNING,
3427                          "failed to parse options in superblock: %s",
3428                          sbi->s_es->s_mount_opts);
3429         }
3430         sbi->s_def_mount_opt = sbi->s_mount_opt;
3431         if (!parse_options((char *) data, sb, &journal_devnum,
3432                            &journal_ioprio, 0))
3433                 goto failed_mount;
3434
3435         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3436                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3437                             "with data=journal disables delayed "
3438                             "allocation and O_DIRECT support!\n");
3439                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3440                         ext4_msg(sb, KERN_ERR, "can't mount with "
3441                                  "both data=journal and delalloc");
3442                         goto failed_mount;
3443                 }
3444                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3445                         ext4_msg(sb, KERN_ERR, "can't mount with "
3446                                  "both data=journal and delalloc");
3447                         goto failed_mount;
3448                 }
3449                 if (test_opt(sb, DELALLOC))
3450                         clear_opt(sb, DELALLOC);
3451         }
3452
3453         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3454                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3455
3456         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3457             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3458              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3459              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3460                 ext4_msg(sb, KERN_WARNING,
3461                        "feature flags set on rev 0 fs, "
3462                        "running e2fsck is recommended");
3463
3464         if (IS_EXT2_SB(sb)) {
3465                 if (ext2_feature_set_ok(sb))
3466                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3467                                  "using the ext4 subsystem");
3468                 else {
3469                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3470                                  "to feature incompatibilities");
3471                         goto failed_mount;
3472                 }
3473         }
3474
3475         if (IS_EXT3_SB(sb)) {
3476                 if (ext3_feature_set_ok(sb))
3477                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3478                                  "using the ext4 subsystem");
3479                 else {
3480                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3481                                  "to feature incompatibilities");
3482                         goto failed_mount;
3483                 }
3484         }
3485
3486         /*
3487          * Check feature flags regardless of the revision level, since we
3488          * previously didn't change the revision level when setting the flags,
3489          * so there is a chance incompat flags are set on a rev 0 filesystem.
3490          */
3491         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3492                 goto failed_mount;
3493
3494         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3495         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3496             blocksize > EXT4_MAX_BLOCK_SIZE) {
3497                 ext4_msg(sb, KERN_ERR,
3498                        "Unsupported filesystem blocksize %d", blocksize);
3499                 goto failed_mount;
3500         }
3501
3502         if (sb->s_blocksize != blocksize) {
3503                 /* Validate the filesystem blocksize */
3504                 if (!sb_set_blocksize(sb, blocksize)) {
3505                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3506                                         blocksize);
3507                         goto failed_mount;
3508                 }
3509
3510                 brelse(bh);
3511                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3512                 offset = do_div(logical_sb_block, blocksize);
3513                 bh = sb_bread(sb, logical_sb_block);
3514                 if (!bh) {
3515                         ext4_msg(sb, KERN_ERR,
3516                                "Can't read superblock on 2nd try");
3517                         goto failed_mount;
3518                 }
3519                 es = (struct ext4_super_block *)(bh->b_data + offset);
3520                 sbi->s_es = es;
3521                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3522                         ext4_msg(sb, KERN_ERR,
3523                                "Magic mismatch, very weird!");
3524                         goto failed_mount;
3525                 }
3526         }
3527
3528         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3529                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3530         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3531                                                       has_huge_files);
3532         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3533
3534         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3535                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3536                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3537         } else {
3538                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3539                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3540                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3541                     (!is_power_of_2(sbi->s_inode_size)) ||
3542                     (sbi->s_inode_size > blocksize)) {
3543                         ext4_msg(sb, KERN_ERR,
3544                                "unsupported inode size: %d",
3545                                sbi->s_inode_size);
3546                         goto failed_mount;
3547                 }
3548                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3549                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3550         }
3551
3552         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3553         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3554                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3555                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3556                     !is_power_of_2(sbi->s_desc_size)) {
3557                         ext4_msg(sb, KERN_ERR,
3558                                "unsupported descriptor size %lu",
3559                                sbi->s_desc_size);
3560                         goto failed_mount;
3561                 }
3562         } else
3563                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3564
3565         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3566         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3567         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3568                 goto cantfind_ext4;
3569
3570         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3571         if (sbi->s_inodes_per_block == 0)
3572                 goto cantfind_ext4;
3573         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3574                                         sbi->s_inodes_per_block;
3575         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3576         sbi->s_sbh = bh;
3577         sbi->s_mount_state = le16_to_cpu(es->s_state);
3578         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3579         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3580
3581         for (i = 0; i < 4; i++)
3582                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3583         sbi->s_def_hash_version = es->s_def_hash_version;
3584         i = le32_to_cpu(es->s_flags);
3585         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3586                 sbi->s_hash_unsigned = 3;
3587         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3588 #ifdef __CHAR_UNSIGNED__
3589                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3590                 sbi->s_hash_unsigned = 3;
3591 #else
3592                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3593 #endif
3594         }
3595
3596         /* Handle clustersize */
3597         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3598         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3599                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3600         if (has_bigalloc) {
3601                 if (clustersize < blocksize) {
3602                         ext4_msg(sb, KERN_ERR,
3603                                  "cluster size (%d) smaller than "
3604                                  "block size (%d)", clustersize, blocksize);
3605                         goto failed_mount;
3606                 }
3607                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3608                         le32_to_cpu(es->s_log_block_size);
3609                 sbi->s_clusters_per_group =
3610                         le32_to_cpu(es->s_clusters_per_group);
3611                 if (sbi->s_clusters_per_group > blocksize * 8) {
3612                         ext4_msg(sb, KERN_ERR,
3613                                  "#clusters per group too big: %lu",
3614                                  sbi->s_clusters_per_group);
3615                         goto failed_mount;
3616                 }
3617                 if (sbi->s_blocks_per_group !=
3618                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3619                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3620                                  "clusters per group (%lu) inconsistent",
3621                                  sbi->s_blocks_per_group,
3622                                  sbi->s_clusters_per_group);
3623                         goto failed_mount;
3624                 }
3625         } else {
3626                 if (clustersize != blocksize) {
3627                         ext4_warning(sb, "fragment/cluster size (%d) != "
3628                                      "block size (%d)", clustersize,
3629                                      blocksize);
3630                         clustersize = blocksize;
3631                 }
3632                 if (sbi->s_blocks_per_group > blocksize * 8) {
3633                         ext4_msg(sb, KERN_ERR,
3634                                  "#blocks per group too big: %lu",
3635                                  sbi->s_blocks_per_group);
3636                         goto failed_mount;
3637                 }
3638                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3639                 sbi->s_cluster_bits = 0;
3640         }
3641         sbi->s_cluster_ratio = clustersize / blocksize;
3642
3643         if (sbi->s_inodes_per_group > blocksize * 8) {
3644                 ext4_msg(sb, KERN_ERR,
3645                        "#inodes per group too big: %lu",
3646                        sbi->s_inodes_per_group);
3647                 goto failed_mount;
3648         }
3649
3650         /*
3651          * Test whether we have more sectors than will fit in sector_t,
3652          * and whether the max offset is addressable by the page cache.
3653          */
3654         err = generic_check_addressable(sb->s_blocksize_bits,
3655                                         ext4_blocks_count(es));
3656         if (err) {
3657                 ext4_msg(sb, KERN_ERR, "filesystem"
3658                          " too large to mount safely on this system");
3659                 if (sizeof(sector_t) < 8)
3660                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3661                 goto failed_mount;
3662         }
3663
3664         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3665                 goto cantfind_ext4;
3666
3667         /* check blocks count against device size */
3668         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3669         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3670                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3671                        "exceeds size of device (%llu blocks)",
3672                        ext4_blocks_count(es), blocks_count);
3673                 goto failed_mount;
3674         }
3675
3676         /*
3677          * It makes no sense for the first data block to be beyond the end
3678          * of the filesystem.
3679          */
3680         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3681                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3682                          "block %u is beyond end of filesystem (%llu)",
3683                          le32_to_cpu(es->s_first_data_block),
3684                          ext4_blocks_count(es));
3685                 goto failed_mount;
3686         }
3687         blocks_count = (ext4_blocks_count(es) -
3688                         le32_to_cpu(es->s_first_data_block) +
3689                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3690         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3691         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3692                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3693                        "(block count %llu, first data block %u, "
3694                        "blocks per group %lu)", sbi->s_groups_count,
3695                        ext4_blocks_count(es),
3696                        le32_to_cpu(es->s_first_data_block),
3697                        EXT4_BLOCKS_PER_GROUP(sb));
3698                 goto failed_mount;
3699         }
3700         sbi->s_groups_count = blocks_count;
3701         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3702                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3703         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3704                    EXT4_DESC_PER_BLOCK(sb);
3705         sbi->s_group_desc = ext4_kvmalloc(db_count *
3706                                           sizeof(struct buffer_head *),
3707                                           GFP_KERNEL);
3708         if (sbi->s_group_desc == NULL) {
3709                 ext4_msg(sb, KERN_ERR, "not enough memory");
3710                 ret = -ENOMEM;
3711                 goto failed_mount;
3712         }
3713
3714         if (ext4_proc_root)
3715                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3716
3717         if (sbi->s_proc)
3718                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3719                                  &ext4_seq_options_fops, sb);
3720
3721         bgl_lock_init(sbi->s_blockgroup_lock);
3722
3723         for (i = 0; i < db_count; i++) {
3724                 block = descriptor_loc(sb, logical_sb_block, i);
3725                 sbi->s_group_desc[i] = sb_bread(sb, block);
3726                 if (!sbi->s_group_desc[i]) {
3727                         ext4_msg(sb, KERN_ERR,
3728                                "can't read group descriptor %d", i);
3729                         db_count = i;
3730                         goto failed_mount2;
3731                 }
3732         }
3733         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3734                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3735                 goto failed_mount2;
3736         }
3737         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3738                 if (!ext4_fill_flex_info(sb)) {
3739                         ext4_msg(sb, KERN_ERR,
3740                                "unable to initialize "
3741                                "flex_bg meta info!");
3742                         goto failed_mount2;
3743                 }
3744
3745         sbi->s_gdb_count = db_count;
3746         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3747         spin_lock_init(&sbi->s_next_gen_lock);
3748
3749         init_timer(&sbi->s_err_report);
3750         sbi->s_err_report.function = print_daily_error_info;
3751         sbi->s_err_report.data = (unsigned long) sb;
3752
3753         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3754                         ext4_count_free_clusters(sb));
3755         if (!err) {
3756                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3757                                 ext4_count_free_inodes(sb));
3758         }
3759         if (!err) {
3760                 err = percpu_counter_init(&sbi->s_dirs_counter,
3761                                 ext4_count_dirs(sb));
3762         }
3763         if (!err) {
3764                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3765         }
3766         if (err) {
3767                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3768                 goto failed_mount3;
3769         }
3770
3771         sbi->s_stripe = ext4_get_stripe_size(sbi);
3772         sbi->s_max_writeback_mb_bump = 128;
3773         sbi->s_extent_max_zeroout_kb = 32;
3774
3775         /*
3776          * set up enough so that it can read an inode
3777          */
3778         if (!test_opt(sb, NOLOAD) &&
3779             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3780                 sb->s_op = &ext4_sops;
3781         else
3782                 sb->s_op = &ext4_nojournal_sops;
3783         sb->s_export_op = &ext4_export_ops;
3784         sb->s_xattr = ext4_xattr_handlers;
3785 #ifdef CONFIG_QUOTA
3786         sb->s_qcop = &ext4_qctl_operations;
3787         sb->dq_op = &ext4_quota_operations;
3788
3789         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3790                 /* Use qctl operations for hidden quota files. */
3791                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3792         }
3793 #endif
3794         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3795
3796         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3797         mutex_init(&sbi->s_orphan_lock);
3798
3799         sb->s_root = NULL;
3800
3801         needs_recovery = (es->s_last_orphan != 0 ||
3802                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3803                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3804
3805         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3806             !(sb->s_flags & MS_RDONLY))
3807                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3808                         goto failed_mount3;
3809
3810         /*
3811          * The first inode we look at is the journal inode.  Don't try
3812          * root first: it may be modified in the journal!
3813          */
3814         if (!test_opt(sb, NOLOAD) &&
3815             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3816                 if (ext4_load_journal(sb, es, journal_devnum))
3817                         goto failed_mount3;
3818         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3819               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3820                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3821                        "suppressed and not mounted read-only");
3822                 goto failed_mount_wq;
3823         } else {
3824                 clear_opt(sb, DATA_FLAGS);
3825                 sbi->s_journal = NULL;
3826                 needs_recovery = 0;
3827                 goto no_journal;
3828         }
3829
3830         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3831             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3832                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3833                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3834                 goto failed_mount_wq;
3835         }
3836
3837         if (!set_journal_csum_feature_set(sb)) {
3838                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3839                          "feature set");
3840                 goto failed_mount_wq;
3841         }
3842
3843         /* We have now updated the journal if required, so we can
3844          * validate the data journaling mode. */
3845         switch (test_opt(sb, DATA_FLAGS)) {
3846         case 0:
3847                 /* No mode set, assume a default based on the journal
3848                  * capabilities: ORDERED_DATA if the journal can
3849                  * cope, else JOURNAL_DATA
3850                  */
3851                 if (jbd2_journal_check_available_features
3852                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3853                         set_opt(sb, ORDERED_DATA);
3854                 else
3855                         set_opt(sb, JOURNAL_DATA);
3856                 break;
3857
3858         case EXT4_MOUNT_ORDERED_DATA:
3859         case EXT4_MOUNT_WRITEBACK_DATA:
3860                 if (!jbd2_journal_check_available_features
3861                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3862                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3863                                "requested data journaling mode");
3864                         goto failed_mount_wq;
3865                 }
3866         default:
3867                 break;
3868         }
3869         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3870
3871         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3872
3873         /*
3874          * The journal may have updated the bg summary counts, so we
3875          * need to update the global counters.
3876          */
3877         percpu_counter_set(&sbi->s_freeclusters_counter,
3878                            ext4_count_free_clusters(sb));
3879         percpu_counter_set(&sbi->s_freeinodes_counter,
3880                            ext4_count_free_inodes(sb));
3881         percpu_counter_set(&sbi->s_dirs_counter,
3882                            ext4_count_dirs(sb));
3883         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3884
3885 no_journal:
3886         /*
3887          * Get the # of file system overhead blocks from the
3888          * superblock if present.
3889          */
3890         if (es->s_overhead_clusters)
3891                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3892         else {
3893                 err = ext4_calculate_overhead(sb);
3894                 if (err)
3895                         goto failed_mount_wq;
3896         }
3897
3898         /*
3899          * The maximum number of concurrent works can be high and
3900          * concurrency isn't really necessary.  Limit it to 1.
3901          */
3902         EXT4_SB(sb)->dio_unwritten_wq =
3903                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3904         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3905                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3906                 ret = -ENOMEM;
3907                 goto failed_mount_wq;
3908         }
3909
3910         /*
3911          * The jbd2_journal_load will have done any necessary log recovery,
3912          * so we can safely mount the rest of the filesystem now.
3913          */
3914
3915         root = ext4_iget(sb, EXT4_ROOT_INO);
3916         if (IS_ERR(root)) {
3917                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3918                 ret = PTR_ERR(root);
3919                 root = NULL;
3920                 goto failed_mount4;
3921         }
3922         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3923                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3924                 iput(root);
3925                 goto failed_mount4;
3926         }
3927         sb->s_root = d_make_root(root);
3928         if (!sb->s_root) {
3929                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3930                 ret = -ENOMEM;
3931                 goto failed_mount4;
3932         }
3933
3934         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3935                 sb->s_flags |= MS_RDONLY;
3936
3937         /* determine the minimum size of new large inodes, if present */
3938         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3939                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3940                                                      EXT4_GOOD_OLD_INODE_SIZE;
3941                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3942                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3943                         if (sbi->s_want_extra_isize <
3944                             le16_to_cpu(es->s_want_extra_isize))
3945                                 sbi->s_want_extra_isize =
3946                                         le16_to_cpu(es->s_want_extra_isize);
3947                         if (sbi->s_want_extra_isize <
3948                             le16_to_cpu(es->s_min_extra_isize))
3949                                 sbi->s_want_extra_isize =
3950                                         le16_to_cpu(es->s_min_extra_isize);
3951                 }
3952         }
3953         /* Check if enough inode space is available */
3954         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3955                                                         sbi->s_inode_size) {
3956                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3957                                                        EXT4_GOOD_OLD_INODE_SIZE;
3958                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3959                          "available");
3960         }
3961
3962         err = ext4_setup_system_zone(sb);
3963         if (err) {
3964                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3965                          "zone (%d)", err);
3966                 goto failed_mount4a;
3967         }
3968
3969         ext4_ext_init(sb);
3970         err = ext4_mb_init(sb);
3971         if (err) {
3972                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3973                          err);
3974                 goto failed_mount5;
3975         }
3976
3977         err = ext4_register_li_request(sb, first_not_zeroed);
3978         if (err)
3979                 goto failed_mount6;
3980
3981         sbi->s_kobj.kset = ext4_kset;
3982         init_completion(&sbi->s_kobj_unregister);
3983         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3984                                    "%s", sb->s_id);
3985         if (err)
3986                 goto failed_mount7;
3987
3988         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3989         ext4_orphan_cleanup(sb, es);
3990         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3991         if (needs_recovery) {
3992                 ext4_msg(sb, KERN_INFO, "recovery complete");
3993                 ext4_mark_recovery_complete(sb, es);
3994         }
3995         if (EXT4_SB(sb)->s_journal) {
3996                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3997                         descr = " journalled data mode";
3998                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3999                         descr = " ordered data mode";
4000                 else
4001                         descr = " writeback data mode";
4002         } else
4003                 descr = "out journal";
4004
4005 #ifdef CONFIG_QUOTA
4006         /* Enable quota usage during mount. */
4007         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4008             !(sb->s_flags & MS_RDONLY)) {
4009                 err = ext4_enable_quotas(sb);
4010                 if (err)
4011                         goto failed_mount7;
4012         }
4013 #endif  /* CONFIG_QUOTA */
4014
4015         if (test_opt(sb, DISCARD)) {
4016                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4017                 if (!blk_queue_discard(q))
4018                         ext4_msg(sb, KERN_WARNING,
4019                                  "mounting with \"discard\" option, but "
4020                                  "the device does not support discard");
4021         }
4022
4023         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4024                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4025                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4026
4027         if (es->s_error_count)
4028                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4029
4030         kfree(orig_data);
4031         return 0;
4032
4033 cantfind_ext4:
4034         if (!silent)
4035                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4036         goto failed_mount;
4037
4038 failed_mount7:
4039         ext4_unregister_li_request(sb);
4040 failed_mount6:
4041         ext4_mb_release(sb);
4042 failed_mount5:
4043         ext4_ext_release(sb);
4044         ext4_release_system_zone(sb);
4045 failed_mount4a:
4046         dput(sb->s_root);
4047         sb->s_root = NULL;
4048 failed_mount4:
4049         ext4_msg(sb, KERN_ERR, "mount failed");
4050         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4051 failed_mount_wq:
4052         if (sbi->s_journal) {
4053                 jbd2_journal_destroy(sbi->s_journal);
4054                 sbi->s_journal = NULL;
4055         }
4056 failed_mount3:
4057         del_timer(&sbi->s_err_report);
4058         if (sbi->s_flex_groups)
4059                 ext4_kvfree(sbi->s_flex_groups);
4060         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4061         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4062         percpu_counter_destroy(&sbi->s_dirs_counter);
4063         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4064         if (sbi->s_mmp_tsk)
4065                 kthread_stop(sbi->s_mmp_tsk);
4066 failed_mount2:
4067         for (i = 0; i < db_count; i++)
4068                 brelse(sbi->s_group_desc[i]);
4069         ext4_kvfree(sbi->s_group_desc);
4070 failed_mount:
4071         if (sbi->s_chksum_driver)
4072                 crypto_free_shash(sbi->s_chksum_driver);
4073         if (sbi->s_proc) {
4074                 remove_proc_entry("options", sbi->s_proc);
4075                 remove_proc_entry(sb->s_id, ext4_proc_root);
4076         }
4077 #ifdef CONFIG_QUOTA
4078         for (i = 0; i < MAXQUOTAS; i++)
4079                 kfree(sbi->s_qf_names[i]);
4080 #endif
4081         ext4_blkdev_remove(sbi);
4082         brelse(bh);
4083 out_fail:
4084         sb->s_fs_info = NULL;
4085         kfree(sbi->s_blockgroup_lock);
4086         kfree(sbi);
4087 out_free_orig:
4088         kfree(orig_data);
4089         return err ? err : ret;
4090 }
4091
4092 /*
4093  * Setup any per-fs journal parameters now.  We'll do this both on
4094  * initial mount, once the journal has been initialised but before we've
4095  * done any recovery; and again on any subsequent remount.
4096  */
4097 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4098 {
4099         struct ext4_sb_info *sbi = EXT4_SB(sb);
4100
4101         journal->j_commit_interval = sbi->s_commit_interval;
4102         journal->j_min_batch_time = sbi->s_min_batch_time;
4103         journal->j_max_batch_time = sbi->s_max_batch_time;
4104
4105         write_lock(&journal->j_state_lock);
4106         if (test_opt(sb, BARRIER))
4107                 journal->j_flags |= JBD2_BARRIER;
4108         else
4109                 journal->j_flags &= ~JBD2_BARRIER;
4110         if (test_opt(sb, DATA_ERR_ABORT))
4111                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4112         else
4113                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4114         write_unlock(&journal->j_state_lock);
4115 }
4116
4117 static journal_t *ext4_get_journal(struct super_block *sb,
4118                                    unsigned int journal_inum)
4119 {
4120         struct inode *journal_inode;
4121         journal_t *journal;
4122
4123         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4124
4125         /* First, test for the existence of a valid inode on disk.  Bad
4126          * things happen if we iget() an unused inode, as the subsequent
4127          * iput() will try to delete it. */
4128
4129         journal_inode = ext4_iget(sb, journal_inum);
4130         if (IS_ERR(journal_inode)) {
4131                 ext4_msg(sb, KERN_ERR, "no journal found");
4132                 return NULL;
4133         }
4134         if (!journal_inode->i_nlink) {
4135                 make_bad_inode(journal_inode);
4136                 iput(journal_inode);
4137                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4138                 return NULL;
4139         }
4140
4141         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4142                   journal_inode, journal_inode->i_size);
4143         if (!S_ISREG(journal_inode->i_mode)) {
4144                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4145                 iput(journal_inode);
4146                 return NULL;
4147         }
4148
4149         journal = jbd2_journal_init_inode(journal_inode);
4150         if (!journal) {
4151                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4152                 iput(journal_inode);
4153                 return NULL;
4154         }
4155         journal->j_private = sb;
4156         ext4_init_journal_params(sb, journal);
4157         return journal;
4158 }
4159
4160 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4161                                        dev_t j_dev)
4162 {
4163         struct buffer_head *bh;
4164         journal_t *journal;
4165         ext4_fsblk_t start;
4166         ext4_fsblk_t len;
4167         int hblock, blocksize;
4168         ext4_fsblk_t sb_block;
4169         unsigned long offset;
4170         struct ext4_super_block *es;
4171         struct block_device *bdev;
4172
4173         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4174
4175         bdev = ext4_blkdev_get(j_dev, sb);
4176         if (bdev == NULL)
4177                 return NULL;
4178
4179         blocksize = sb->s_blocksize;
4180         hblock = bdev_logical_block_size(bdev);
4181         if (blocksize < hblock) {
4182                 ext4_msg(sb, KERN_ERR,
4183                         "blocksize too small for journal device");
4184                 goto out_bdev;
4185         }
4186
4187         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4188         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4189         set_blocksize(bdev, blocksize);
4190         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4191                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4192                        "external journal");
4193                 goto out_bdev;
4194         }
4195
4196         es = (struct ext4_super_block *) (bh->b_data + offset);
4197         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4198             !(le32_to_cpu(es->s_feature_incompat) &
4199               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4200                 ext4_msg(sb, KERN_ERR, "external journal has "
4201                                         "bad superblock");
4202                 brelse(bh);
4203                 goto out_bdev;
4204         }
4205
4206         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4207                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4208                 brelse(bh);
4209                 goto out_bdev;
4210         }
4211
4212         len = ext4_blocks_count(es);
4213         start = sb_block + 1;
4214         brelse(bh);     /* we're done with the superblock */
4215
4216         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4217                                         start, len, blocksize);
4218         if (!journal) {
4219                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4220                 goto out_bdev;
4221         }
4222         journal->j_private = sb;
4223         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4224         wait_on_buffer(journal->j_sb_buffer);
4225         if (!buffer_uptodate(journal->j_sb_buffer)) {
4226                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4227                 goto out_journal;
4228         }
4229         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4230                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4231                                         "user (unsupported) - %d",
4232                         be32_to_cpu(journal->j_superblock->s_nr_users));
4233                 goto out_journal;
4234         }
4235         EXT4_SB(sb)->journal_bdev = bdev;
4236         ext4_init_journal_params(sb, journal);
4237         return journal;
4238
4239 out_journal:
4240         jbd2_journal_destroy(journal);
4241 out_bdev:
4242         ext4_blkdev_put(bdev);
4243         return NULL;
4244 }
4245
4246 static int ext4_load_journal(struct super_block *sb,
4247                              struct ext4_super_block *es,
4248                              unsigned long journal_devnum)
4249 {
4250         journal_t *journal;
4251         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4252         dev_t journal_dev;
4253         int err = 0;
4254         int really_read_only;
4255
4256         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4257
4258         if (journal_devnum &&
4259             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4260                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4261                         "numbers have changed");
4262                 journal_dev = new_decode_dev(journal_devnum);
4263         } else
4264                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4265
4266         really_read_only = bdev_read_only(sb->s_bdev);
4267
4268         /*
4269          * Are we loading a blank journal or performing recovery after a
4270          * crash?  For recovery, we need to check in advance whether we
4271          * can get read-write access to the device.
4272          */
4273         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4274                 if (sb->s_flags & MS_RDONLY) {
4275                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4276                                         "required on readonly filesystem");
4277                         if (really_read_only) {
4278                                 ext4_msg(sb, KERN_ERR, "write access "
4279                                         "unavailable, cannot proceed");
4280                                 return -EROFS;
4281                         }
4282                         ext4_msg(sb, KERN_INFO, "write access will "
4283                                "be enabled during recovery");
4284                 }
4285         }
4286
4287         if (journal_inum && journal_dev) {
4288                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4289                        "and inode journals!");
4290                 return -EINVAL;
4291         }
4292
4293         if (journal_inum) {
4294                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4295                         return -EINVAL;
4296         } else {
4297                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4298                         return -EINVAL;
4299         }
4300
4301         if (!(journal->j_flags & JBD2_BARRIER))
4302                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4303
4304         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4305                 err = jbd2_journal_wipe(journal, !really_read_only);
4306         if (!err) {
4307                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4308                 if (save)
4309                         memcpy(save, ((char *) es) +
4310                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4311                 err = jbd2_journal_load(journal);
4312                 if (save)
4313                         memcpy(((char *) es) + EXT4_S_ERR_START,
4314                                save, EXT4_S_ERR_LEN);
4315                 kfree(save);
4316         }
4317
4318         if (err) {
4319                 ext4_msg(sb, KERN_ERR, "error loading journal");
4320                 jbd2_journal_destroy(journal);
4321                 return err;
4322         }
4323
4324         EXT4_SB(sb)->s_journal = journal;
4325         ext4_clear_journal_err(sb, es);
4326
4327         if (!really_read_only && journal_devnum &&
4328             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4329                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4330
4331                 /* Make sure we flush the recovery flag to disk. */
4332                 ext4_commit_super(sb, 1);
4333         }
4334
4335         return 0;
4336 }
4337
4338 static int ext4_commit_super(struct super_block *sb, int sync)
4339 {
4340         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4341         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4342         int error = 0;
4343
4344         if (!sbh || block_device_ejected(sb))
4345                 return error;
4346         if (buffer_write_io_error(sbh)) {
4347                 /*
4348                  * Oh, dear.  A previous attempt to write the
4349                  * superblock failed.  This could happen because the
4350                  * USB device was yanked out.  Or it could happen to
4351                  * be a transient write error and maybe the block will
4352                  * be remapped.  Nothing we can do but to retry the
4353                  * write and hope for the best.
4354                  */
4355                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4356                        "superblock detected");
4357                 clear_buffer_write_io_error(sbh);
4358                 set_buffer_uptodate(sbh);
4359         }
4360         /*
4361          * If the file system is mounted read-only, don't update the
4362          * superblock write time.  This avoids updating the superblock
4363          * write time when we are mounting the root file system
4364          * read/only but we need to replay the journal; at that point,
4365          * for people who are east of GMT and who make their clock
4366          * tick in localtime for Windows bug-for-bug compatibility,
4367          * the clock is set in the future, and this will cause e2fsck
4368          * to complain and force a full file system check.
4369          */
4370         if (!(sb->s_flags & MS_RDONLY))
4371                 es->s_wtime = cpu_to_le32(get_seconds());
4372         if (sb->s_bdev->bd_part)
4373                 es->s_kbytes_written =
4374                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4375                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4376                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4377         else
4378                 es->s_kbytes_written =
4379                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4380         ext4_free_blocks_count_set(es,
4381                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4382                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4383         es->s_free_inodes_count =
4384                 cpu_to_le32(percpu_counter_sum_positive(
4385                                 &EXT4_SB(sb)->s_freeinodes_counter));
4386         BUFFER_TRACE(sbh, "marking dirty");
4387         ext4_superblock_csum_set(sb);
4388         mark_buffer_dirty(sbh);
4389         if (sync) {
4390                 error = sync_dirty_buffer(sbh);
4391                 if (error)
4392                         return error;
4393
4394                 error = buffer_write_io_error(sbh);
4395                 if (error) {
4396                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4397                                "superblock");
4398                         clear_buffer_write_io_error(sbh);
4399                         set_buffer_uptodate(sbh);
4400                 }
4401         }
4402         return error;
4403 }
4404
4405 /*
4406  * Have we just finished recovery?  If so, and if we are mounting (or
4407  * remounting) the filesystem readonly, then we will end up with a
4408  * consistent fs on disk.  Record that fact.
4409  */
4410 static void ext4_mark_recovery_complete(struct super_block *sb,
4411                                         struct ext4_super_block *es)
4412 {
4413         journal_t *journal = EXT4_SB(sb)->s_journal;
4414
4415         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4416                 BUG_ON(journal != NULL);
4417                 return;
4418         }
4419         jbd2_journal_lock_updates(journal);
4420         if (jbd2_journal_flush(journal) < 0)
4421                 goto out;
4422
4423         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4424             sb->s_flags & MS_RDONLY) {
4425                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4426                 ext4_commit_super(sb, 1);
4427         }
4428
4429 out:
4430         jbd2_journal_unlock_updates(journal);
4431 }
4432
4433 /*
4434  * If we are mounting (or read-write remounting) a filesystem whose journal
4435  * has recorded an error from a previous lifetime, move that error to the
4436  * main filesystem now.
4437  */
4438 static void ext4_clear_journal_err(struct super_block *sb,
4439                                    struct ext4_super_block *es)
4440 {
4441         journal_t *journal;
4442         int j_errno;
4443         const char *errstr;
4444
4445         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4446
4447         journal = EXT4_SB(sb)->s_journal;
4448
4449         /*
4450          * Now check for any error status which may have been recorded in the
4451          * journal by a prior ext4_error() or ext4_abort()
4452          */
4453
4454         j_errno = jbd2_journal_errno(journal);
4455         if (j_errno) {
4456                 char nbuf[16];
4457
4458                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4459                 ext4_warning(sb, "Filesystem error recorded "
4460                              "from previous mount: %s", errstr);
4461                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4462
4463                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4464                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4465                 ext4_commit_super(sb, 1);
4466
4467                 jbd2_journal_clear_err(journal);
4468                 jbd2_journal_update_sb_errno(journal);
4469         }
4470 }
4471
4472 /*
4473  * Force the running and committing transactions to commit,
4474  * and wait on the commit.
4475  */
4476 int ext4_force_commit(struct super_block *sb)
4477 {
4478         journal_t *journal;
4479         int ret = 0;
4480
4481         if (sb->s_flags & MS_RDONLY)
4482                 return 0;
4483
4484         journal = EXT4_SB(sb)->s_journal;
4485         if (journal)
4486                 ret = ext4_journal_force_commit(journal);
4487
4488         return ret;
4489 }
4490
4491 static int ext4_sync_fs(struct super_block *sb, int wait)
4492 {
4493         int ret = 0;
4494         tid_t target;
4495         struct ext4_sb_info *sbi = EXT4_SB(sb);
4496
4497         trace_ext4_sync_fs(sb, wait);
4498         flush_workqueue(sbi->dio_unwritten_wq);
4499         /*
4500          * Writeback quota in non-journalled quota case - journalled quota has
4501          * no dirty dquots
4502          */
4503         dquot_writeback_dquots(sb, -1);
4504         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4505                 if (wait)
4506                         jbd2_log_wait_commit(sbi->s_journal, target);
4507         }
4508         return ret;
4509 }
4510
4511 /*
4512  * LVM calls this function before a (read-only) snapshot is created.  This
4513  * gives us a chance to flush the journal completely and mark the fs clean.
4514  *
4515  * Note that only this function cannot bring a filesystem to be in a clean
4516  * state independently. It relies on upper layer to stop all data & metadata
4517  * modifications.
4518  */
4519 static int ext4_freeze(struct super_block *sb)
4520 {
4521         int error = 0;
4522         journal_t *journal;
4523
4524         if (sb->s_flags & MS_RDONLY)
4525                 return 0;
4526
4527         journal = EXT4_SB(sb)->s_journal;
4528
4529         /* Now we set up the journal barrier. */
4530         jbd2_journal_lock_updates(journal);
4531
4532         /*
4533          * Don't clear the needs_recovery flag if we failed to flush
4534          * the journal.
4535          */
4536         error = jbd2_journal_flush(journal);
4537         if (error < 0)
4538                 goto out;
4539
4540         /* Journal blocked and flushed, clear needs_recovery flag. */
4541         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4542         error = ext4_commit_super(sb, 1);
4543 out:
4544         /* we rely on upper layer to stop further updates */
4545         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4546         return error;
4547 }
4548
4549 /*
4550  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4551  * flag here, even though the filesystem is not technically dirty yet.
4552  */
4553 static int ext4_unfreeze(struct super_block *sb)
4554 {
4555         if (sb->s_flags & MS_RDONLY)
4556                 return 0;
4557
4558         /* Reset the needs_recovery flag before the fs is unlocked. */
4559         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4560         ext4_commit_super(sb, 1);
4561         return 0;
4562 }
4563
4564 /*
4565  * Structure to save mount options for ext4_remount's benefit
4566  */
4567 struct ext4_mount_options {
4568         unsigned long s_mount_opt;
4569         unsigned long s_mount_opt2;
4570         kuid_t s_resuid;
4571         kgid_t s_resgid;
4572         unsigned long s_commit_interval;
4573         u32 s_min_batch_time, s_max_batch_time;
4574 #ifdef CONFIG_QUOTA
4575         int s_jquota_fmt;
4576         char *s_qf_names[MAXQUOTAS];
4577 #endif
4578 };
4579
4580 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4581 {
4582         struct ext4_super_block *es;
4583         struct ext4_sb_info *sbi = EXT4_SB(sb);
4584         unsigned long old_sb_flags;
4585         struct ext4_mount_options old_opts;
4586         int enable_quota = 0;
4587         ext4_group_t g;
4588         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4589         int err = 0;
4590 #ifdef CONFIG_QUOTA
4591         int i;
4592 #endif
4593         char *orig_data = kstrdup(data, GFP_KERNEL);
4594
4595         /* Store the original options */
4596         old_sb_flags = sb->s_flags;
4597         old_opts.s_mount_opt = sbi->s_mount_opt;
4598         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4599         old_opts.s_resuid = sbi->s_resuid;
4600         old_opts.s_resgid = sbi->s_resgid;
4601         old_opts.s_commit_interval = sbi->s_commit_interval;
4602         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4603         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4604 #ifdef CONFIG_QUOTA
4605         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4606         for (i = 0; i < MAXQUOTAS; i++)
4607                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4608 #endif
4609         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4610                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4611
4612         /*
4613          * Allow the "check" option to be passed as a remount option.
4614          */
4615         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4616                 err = -EINVAL;
4617                 goto restore_opts;
4618         }
4619
4620         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4621                 ext4_abort(sb, "Abort forced by user");
4622
4623         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4624                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4625
4626         es = sbi->s_es;
4627
4628         if (sbi->s_journal) {
4629                 ext4_init_journal_params(sb, sbi->s_journal);
4630                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4631         }
4632
4633         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4634                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4635                         err = -EROFS;
4636                         goto restore_opts;
4637                 }
4638
4639                 if (*flags & MS_RDONLY) {
4640                         err = dquot_suspend(sb, -1);
4641                         if (err < 0)
4642                                 goto restore_opts;
4643
4644                         /*
4645                          * First of all, the unconditional stuff we have to do
4646                          * to disable replay of the journal when we next remount
4647                          */
4648                         sb->s_flags |= MS_RDONLY;
4649
4650                         /*
4651                          * OK, test if we are remounting a valid rw partition
4652                          * readonly, and if so set the rdonly flag and then
4653                          * mark the partition as valid again.
4654                          */
4655                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4656                             (sbi->s_mount_state & EXT4_VALID_FS))
4657                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4658
4659                         if (sbi->s_journal)
4660                                 ext4_mark_recovery_complete(sb, es);
4661                 } else {
4662                         /* Make sure we can mount this feature set readwrite */
4663                         if (!ext4_feature_set_ok(sb, 0)) {
4664                                 err = -EROFS;
4665                                 goto restore_opts;
4666                         }
4667                         /*
4668                          * Make sure the group descriptor checksums
4669                          * are sane.  If they aren't, refuse to remount r/w.
4670                          */
4671                         for (g = 0; g < sbi->s_groups_count; g++) {
4672                                 struct ext4_group_desc *gdp =
4673                                         ext4_get_group_desc(sb, g, NULL);
4674
4675                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4676                                         ext4_msg(sb, KERN_ERR,
4677                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4678                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4679                                                le16_to_cpu(gdp->bg_checksum));
4680                                         err = -EINVAL;
4681                                         goto restore_opts;
4682                                 }
4683                         }
4684
4685                         /*
4686                          * If we have an unprocessed orphan list hanging
4687                          * around from a previously readonly bdev mount,
4688                          * require a full umount/remount for now.
4689                          */
4690                         if (es->s_last_orphan) {
4691                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4692                                        "remount RDWR because of unprocessed "
4693                                        "orphan inode list.  Please "
4694                                        "umount/remount instead");
4695                                 err = -EINVAL;
4696                                 goto restore_opts;
4697                         }
4698
4699                         /*
4700                          * Mounting a RDONLY partition read-write, so reread
4701                          * and store the current valid flag.  (It may have
4702                          * been changed by e2fsck since we originally mounted
4703                          * the partition.)
4704                          */
4705                         if (sbi->s_journal)
4706                                 ext4_clear_journal_err(sb, es);
4707                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4708                         if (!ext4_setup_super(sb, es, 0))
4709                                 sb->s_flags &= ~MS_RDONLY;
4710                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4711                                                      EXT4_FEATURE_INCOMPAT_MMP))
4712                                 if (ext4_multi_mount_protect(sb,
4713                                                 le64_to_cpu(es->s_mmp_block))) {
4714                                         err = -EROFS;
4715                                         goto restore_opts;
4716                                 }
4717                         enable_quota = 1;
4718                 }
4719         }
4720
4721         /*
4722          * Reinitialize lazy itable initialization thread based on
4723          * current settings
4724          */
4725         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4726                 ext4_unregister_li_request(sb);
4727         else {
4728                 ext4_group_t first_not_zeroed;
4729                 first_not_zeroed = ext4_has_uninit_itable(sb);
4730                 ext4_register_li_request(sb, first_not_zeroed);
4731         }
4732
4733         ext4_setup_system_zone(sb);
4734         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4735                 ext4_commit_super(sb, 1);
4736
4737 #ifdef CONFIG_QUOTA
4738         /* Release old quota file names */
4739         for (i = 0; i < MAXQUOTAS; i++)
4740                 if (old_opts.s_qf_names[i] &&
4741                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4742                         kfree(old_opts.s_qf_names[i]);
4743         if (enable_quota) {
4744                 if (sb_any_quota_suspended(sb))
4745                         dquot_resume(sb, -1);
4746                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4747                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4748                         err = ext4_enable_quotas(sb);
4749                         if (err)
4750                                 goto restore_opts;
4751                 }
4752         }
4753 #endif
4754
4755         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4756         kfree(orig_data);
4757         return 0;
4758
4759 restore_opts:
4760         sb->s_flags = old_sb_flags;
4761         sbi->s_mount_opt = old_opts.s_mount_opt;
4762         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4763         sbi->s_resuid = old_opts.s_resuid;
4764         sbi->s_resgid = old_opts.s_resgid;
4765         sbi->s_commit_interval = old_opts.s_commit_interval;
4766         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4767         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4768 #ifdef CONFIG_QUOTA
4769         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4770         for (i = 0; i < MAXQUOTAS; i++) {
4771                 if (sbi->s_qf_names[i] &&
4772                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4773                         kfree(sbi->s_qf_names[i]);
4774                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4775         }
4776 #endif
4777         kfree(orig_data);
4778         return err;
4779 }
4780
4781 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4782 {
4783         struct super_block *sb = dentry->d_sb;
4784         struct ext4_sb_info *sbi = EXT4_SB(sb);
4785         struct ext4_super_block *es = sbi->s_es;
4786         ext4_fsblk_t overhead = 0;
4787         u64 fsid;
4788         s64 bfree;
4789
4790         if (!test_opt(sb, MINIX_DF))
4791                 overhead = sbi->s_overhead;
4792
4793         buf->f_type = EXT4_SUPER_MAGIC;
4794         buf->f_bsize = sb->s_blocksize;
4795         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4796         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4797                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4798         /* prevent underflow in case that few free space is available */
4799         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4800         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4801         if (buf->f_bfree < ext4_r_blocks_count(es))
4802                 buf->f_bavail = 0;
4803         buf->f_files = le32_to_cpu(es->s_inodes_count);
4804         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4805         buf->f_namelen = EXT4_NAME_LEN;
4806         fsid = le64_to_cpup((void *)es->s_uuid) ^
4807                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4808         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4809         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4810
4811         return 0;
4812 }
4813
4814 /* Helper function for writing quotas on sync - we need to start transaction
4815  * before quota file is locked for write. Otherwise the are possible deadlocks:
4816  * Process 1                         Process 2
4817  * ext4_create()                     quota_sync()
4818  *   jbd2_journal_start()                  write_dquot()
4819  *   dquot_initialize()                         down(dqio_mutex)
4820  *     down(dqio_mutex)                    jbd2_journal_start()
4821  *
4822  */
4823
4824 #ifdef CONFIG_QUOTA
4825
4826 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4827 {
4828         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4829 }
4830
4831 static int ext4_write_dquot(struct dquot *dquot)
4832 {
4833         int ret, err;
4834         handle_t *handle;
4835         struct inode *inode;
4836
4837         inode = dquot_to_inode(dquot);
4838         handle = ext4_journal_start(inode,
4839                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4840         if (IS_ERR(handle))
4841                 return PTR_ERR(handle);
4842         ret = dquot_commit(dquot);
4843         err = ext4_journal_stop(handle);
4844         if (!ret)
4845                 ret = err;
4846         return ret;
4847 }
4848
4849 static int ext4_acquire_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_INIT_BLOCKS(dquot->dq_sb));
4856         if (IS_ERR(handle))
4857                 return PTR_ERR(handle);
4858         ret = dquot_acquire(dquot);
4859         err = ext4_journal_stop(handle);
4860         if (!ret)
4861                 ret = err;
4862         return ret;
4863 }
4864
4865 static int ext4_release_dquot(struct dquot *dquot)
4866 {
4867         int ret, err;
4868         handle_t *handle;
4869
4870         handle = ext4_journal_start(dquot_to_inode(dquot),
4871                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4872         if (IS_ERR(handle)) {
4873                 /* Release dquot anyway to avoid endless cycle in dqput() */
4874                 dquot_release(dquot);
4875                 return PTR_ERR(handle);
4876         }
4877         ret = dquot_release(dquot);
4878         err = ext4_journal_stop(handle);
4879         if (!ret)
4880                 ret = err;
4881         return ret;
4882 }
4883
4884 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4885 {
4886         /* Are we journaling quotas? */
4887         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4888             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4889                 dquot_mark_dquot_dirty(dquot);
4890                 return ext4_write_dquot(dquot);
4891         } else {
4892                 return dquot_mark_dquot_dirty(dquot);
4893         }
4894 }
4895
4896 static int ext4_write_info(struct super_block *sb, int type)
4897 {
4898         int ret, err;
4899         handle_t *handle;
4900
4901         /* Data block + inode block */
4902         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4903         if (IS_ERR(handle))
4904                 return PTR_ERR(handle);
4905         ret = dquot_commit_info(sb, type);
4906         err = ext4_journal_stop(handle);
4907         if (!ret)
4908                 ret = err;
4909         return ret;
4910 }
4911
4912 /*
4913  * Turn on quotas during mount time - we need to find
4914  * the quota file and such...
4915  */
4916 static int ext4_quota_on_mount(struct super_block *sb, int type)
4917 {
4918         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4919                                         EXT4_SB(sb)->s_jquota_fmt, type);
4920 }
4921
4922 /*
4923  * Standard function to be called on quota_on
4924  */
4925 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4926                          struct path *path)
4927 {
4928         int err;
4929
4930         if (!test_opt(sb, QUOTA))
4931                 return -EINVAL;
4932
4933         /* Quotafile not on the same filesystem? */
4934         if (path->dentry->d_sb != sb)
4935                 return -EXDEV;
4936         /* Journaling quota? */
4937         if (EXT4_SB(sb)->s_qf_names[type]) {
4938                 /* Quotafile not in fs root? */
4939                 if (path->dentry->d_parent != sb->s_root)
4940                         ext4_msg(sb, KERN_WARNING,
4941                                 "Quota file not on filesystem root. "
4942                                 "Journaled quota will not work");
4943         }
4944
4945         /*
4946          * When we journal data on quota file, we have to flush journal to see
4947          * all updates to the file when we bypass pagecache...
4948          */
4949         if (EXT4_SB(sb)->s_journal &&
4950             ext4_should_journal_data(path->dentry->d_inode)) {
4951                 /*
4952                  * We don't need to lock updates but journal_flush() could
4953                  * otherwise be livelocked...
4954                  */
4955                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4956                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4957                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4958                 if (err)
4959                         return err;
4960         }
4961
4962         return dquot_quota_on(sb, type, format_id, path);
4963 }
4964
4965 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4966                              unsigned int flags)
4967 {
4968         int err;
4969         struct inode *qf_inode;
4970         unsigned long qf_inums[MAXQUOTAS] = {
4971                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4972                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4973         };
4974
4975         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4976
4977         if (!qf_inums[type])
4978                 return -EPERM;
4979
4980         qf_inode = ext4_iget(sb, qf_inums[type]);
4981         if (IS_ERR(qf_inode)) {
4982                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4983                 return PTR_ERR(qf_inode);
4984         }
4985
4986         err = dquot_enable(qf_inode, type, format_id, flags);
4987         iput(qf_inode);
4988
4989         return err;
4990 }
4991
4992 /* Enable usage tracking for all quota types. */
4993 static int ext4_enable_quotas(struct super_block *sb)
4994 {
4995         int type, err = 0;
4996         unsigned long qf_inums[MAXQUOTAS] = {
4997                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4998                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4999         };
5000
5001         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5002         for (type = 0; type < MAXQUOTAS; type++) {
5003                 if (qf_inums[type]) {
5004                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5005                                                 DQUOT_USAGE_ENABLED);
5006                         if (err) {
5007                                 ext4_warning(sb,
5008                                         "Failed to enable quota (type=%d) "
5009                                         "tracking. Please run e2fsck to fix.",
5010                                         type);
5011                                 return err;
5012                         }
5013                 }
5014         }
5015         return 0;
5016 }
5017
5018 /*
5019  * quota_on function that is used when QUOTA feature is set.
5020  */
5021 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5022                                  int format_id)
5023 {
5024         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5025                 return -EINVAL;
5026
5027         /*
5028          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5029          */
5030         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5031 }
5032
5033 static int ext4_quota_off(struct super_block *sb, int type)
5034 {
5035         struct inode *inode = sb_dqopt(sb)->files[type];
5036         handle_t *handle;
5037
5038         /* Force all delayed allocation blocks to be allocated.
5039          * Caller already holds s_umount sem */
5040         if (test_opt(sb, DELALLOC))
5041                 sync_filesystem(sb);
5042
5043         if (!inode)
5044                 goto out;
5045
5046         /* Update modification times of quota files when userspace can
5047          * start looking at them */
5048         handle = ext4_journal_start(inode, 1);
5049         if (IS_ERR(handle))
5050                 goto out;
5051         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5052         ext4_mark_inode_dirty(handle, inode);
5053         ext4_journal_stop(handle);
5054
5055 out:
5056         return dquot_quota_off(sb, type);
5057 }
5058
5059 /*
5060  * quota_off function that is used when QUOTA feature is set.
5061  */
5062 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5063 {
5064         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5065                 return -EINVAL;
5066
5067         /* Disable only the limits. */
5068         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5069 }
5070
5071 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5072  * acquiring the locks... As quota files are never truncated and quota code
5073  * itself serializes the operations (and no one else should touch the files)
5074  * we don't have to be afraid of races */
5075 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5076                                size_t len, loff_t off)
5077 {
5078         struct inode *inode = sb_dqopt(sb)->files[type];
5079         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5080         int err = 0;
5081         int offset = off & (sb->s_blocksize - 1);
5082         int tocopy;
5083         size_t toread;
5084         struct buffer_head *bh;
5085         loff_t i_size = i_size_read(inode);
5086
5087         if (off > i_size)
5088                 return 0;
5089         if (off+len > i_size)
5090                 len = i_size-off;
5091         toread = len;
5092         while (toread > 0) {
5093                 tocopy = sb->s_blocksize - offset < toread ?
5094                                 sb->s_blocksize - offset : toread;
5095                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5096                 if (err)
5097                         return err;
5098                 if (!bh)        /* A hole? */
5099                         memset(data, 0, tocopy);
5100                 else
5101                         memcpy(data, bh->b_data+offset, tocopy);
5102                 brelse(bh);
5103                 offset = 0;
5104                 toread -= tocopy;
5105                 data += tocopy;
5106                 blk++;
5107         }
5108         return len;
5109 }
5110
5111 /* Write to quotafile (we know the transaction is already started and has
5112  * enough credits) */
5113 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5114                                 const char *data, size_t len, loff_t off)
5115 {
5116         struct inode *inode = sb_dqopt(sb)->files[type];
5117         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5118         int err = 0;
5119         int offset = off & (sb->s_blocksize - 1);
5120         struct buffer_head *bh;
5121         handle_t *handle = journal_current_handle();
5122
5123         if (EXT4_SB(sb)->s_journal && !handle) {
5124                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5125                         " cancelled because transaction is not started",
5126                         (unsigned long long)off, (unsigned long long)len);
5127                 return -EIO;
5128         }
5129         /*
5130          * Since we account only one data block in transaction credits,
5131          * then it is impossible to cross a block boundary.
5132          */
5133         if (sb->s_blocksize - offset < len) {
5134                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5135                         " cancelled because not block aligned",
5136                         (unsigned long long)off, (unsigned long long)len);
5137                 return -EIO;
5138         }
5139
5140         bh = ext4_bread(handle, inode, blk, 1, &err);
5141         if (!bh)
5142                 goto out;
5143         err = ext4_journal_get_write_access(handle, bh);
5144         if (err) {
5145                 brelse(bh);
5146                 goto out;
5147         }
5148         lock_buffer(bh);
5149         memcpy(bh->b_data+offset, data, len);
5150         flush_dcache_page(bh->b_page);
5151         unlock_buffer(bh);
5152         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5153         brelse(bh);
5154 out:
5155         if (err)
5156                 return err;
5157         if (inode->i_size < off + len) {
5158                 i_size_write(inode, off + len);
5159                 EXT4_I(inode)->i_disksize = inode->i_size;
5160                 ext4_mark_inode_dirty(handle, inode);
5161         }
5162         return len;
5163 }
5164
5165 #endif
5166
5167 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5168                        const char *dev_name, void *data)
5169 {
5170         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5171 }
5172
5173 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5174 static inline void register_as_ext2(void)
5175 {
5176         int err = register_filesystem(&ext2_fs_type);
5177         if (err)
5178                 printk(KERN_WARNING
5179                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5180 }
5181
5182 static inline void unregister_as_ext2(void)
5183 {
5184         unregister_filesystem(&ext2_fs_type);
5185 }
5186
5187 static inline int ext2_feature_set_ok(struct super_block *sb)
5188 {
5189         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5190                 return 0;
5191         if (sb->s_flags & MS_RDONLY)
5192                 return 1;
5193         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5194                 return 0;
5195         return 1;
5196 }
5197 MODULE_ALIAS("ext2");
5198 #else
5199 static inline void register_as_ext2(void) { }
5200 static inline void unregister_as_ext2(void) { }
5201 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5202 #endif
5203
5204 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5205 static inline void register_as_ext3(void)
5206 {
5207         int err = register_filesystem(&ext3_fs_type);
5208         if (err)
5209                 printk(KERN_WARNING
5210                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5211 }
5212
5213 static inline void unregister_as_ext3(void)
5214 {
5215         unregister_filesystem(&ext3_fs_type);
5216 }
5217
5218 static inline int ext3_feature_set_ok(struct super_block *sb)
5219 {
5220         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5221                 return 0;
5222         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5223                 return 0;
5224         if (sb->s_flags & MS_RDONLY)
5225                 return 1;
5226         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5227                 return 0;
5228         return 1;
5229 }
5230 MODULE_ALIAS("ext3");
5231 #else
5232 static inline void register_as_ext3(void) { }
5233 static inline void unregister_as_ext3(void) { }
5234 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5235 #endif
5236
5237 static struct file_system_type ext4_fs_type = {
5238         .owner          = THIS_MODULE,
5239         .name           = "ext4",
5240         .mount          = ext4_mount,
5241         .kill_sb        = kill_block_super,
5242         .fs_flags       = FS_REQUIRES_DEV,
5243 };
5244
5245 static int __init ext4_init_feat_adverts(void)
5246 {
5247         struct ext4_features *ef;
5248         int ret = -ENOMEM;
5249
5250         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5251         if (!ef)
5252                 goto out;
5253
5254         ef->f_kobj.kset = ext4_kset;
5255         init_completion(&ef->f_kobj_unregister);
5256         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5257                                    "features");
5258         if (ret) {
5259                 kfree(ef);
5260                 goto out;
5261         }
5262
5263         ext4_feat = ef;
5264         ret = 0;
5265 out:
5266         return ret;
5267 }
5268
5269 static void ext4_exit_feat_adverts(void)
5270 {
5271         kobject_put(&ext4_feat->f_kobj);
5272         wait_for_completion(&ext4_feat->f_kobj_unregister);
5273         kfree(ext4_feat);
5274 }
5275
5276 /* Shared across all ext4 file systems */
5277 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5278 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5279
5280 static int __init ext4_init_fs(void)
5281 {
5282         int i, err;
5283
5284         ext4_li_info = NULL;
5285         mutex_init(&ext4_li_mtx);
5286
5287         /* Build-time check for flags consistency */
5288         ext4_check_flag_values();
5289
5290         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5291                 mutex_init(&ext4__aio_mutex[i]);
5292                 init_waitqueue_head(&ext4__ioend_wq[i]);
5293         }
5294
5295         err = ext4_init_es();
5296         if (err)
5297                 return err;
5298
5299         err = ext4_init_pageio();
5300         if (err)
5301                 goto out7;
5302
5303         err = ext4_init_system_zone();
5304         if (err)
5305                 goto out6;
5306         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5307         if (!ext4_kset) {
5308                 err = -ENOMEM;
5309                 goto out5;
5310         }
5311         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5312
5313         err = ext4_init_feat_adverts();
5314         if (err)
5315                 goto out4;
5316
5317         err = ext4_init_mballoc();
5318         if (err)
5319                 goto out3;
5320
5321         err = ext4_init_xattr();
5322         if (err)
5323                 goto out2;
5324         err = init_inodecache();
5325         if (err)
5326                 goto out1;
5327         register_as_ext3();
5328         register_as_ext2();
5329         err = register_filesystem(&ext4_fs_type);
5330         if (err)
5331                 goto out;
5332
5333         return 0;
5334 out:
5335         unregister_as_ext2();
5336         unregister_as_ext3();
5337         destroy_inodecache();
5338 out1:
5339         ext4_exit_xattr();
5340 out2:
5341         ext4_exit_mballoc();
5342 out3:
5343         ext4_exit_feat_adverts();
5344 out4:
5345         if (ext4_proc_root)
5346                 remove_proc_entry("fs/ext4", NULL);
5347         kset_unregister(ext4_kset);
5348 out5:
5349         ext4_exit_system_zone();
5350 out6:
5351         ext4_exit_pageio();
5352 out7:
5353         ext4_exit_es();
5354
5355         return err;
5356 }
5357
5358 static void __exit ext4_exit_fs(void)
5359 {
5360         ext4_destroy_lazyinit_thread();
5361         unregister_as_ext2();
5362         unregister_as_ext3();
5363         unregister_filesystem(&ext4_fs_type);
5364         destroy_inodecache();
5365         ext4_exit_xattr();
5366         ext4_exit_mballoc();
5367         ext4_exit_feat_adverts();
5368         remove_proc_entry("fs/ext4", NULL);
5369         kset_unregister(ext4_kset);
5370         ext4_exit_system_zone();
5371         ext4_exit_pageio();
5372 }
5373
5374 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5375 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5376 MODULE_LICENSE("GPL");
5377 module_init(ext4_init_fs)
5378 module_exit(ext4_exit_fs)