fs/ext4/super.c

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