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         le32_add_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         if (sbi->s_journal) {
 854                 err = jbd2_journal_destroy(sbi->s_journal);
 855                 sbi->s_journal = NULL;
 856                 if (err < 0)
 857                         ext4_abort(sb, "Couldn't clean up the journal");
 858         }
 859
 860         del_timer(&sbi->s_err_report);
 861         ext4_release_system_zone(sb);
 862         ext4_mb_release(sb);
 863         ext4_ext_release(sb);
 864         ext4_xattr_put_super(sb);
 865
 866         if (!(sb->s_flags & MS_RDONLY)) {
 867                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 868                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 869         }
 870         if (!(sb->s_flags & MS_RDONLY))
 871                 ext4_commit_super(sb, 1);
 872
 873         if (sbi->s_proc) {
 874                 remove_proc_entry("options", sbi->s_proc);
 875                 remove_proc_entry(sb->s_id, ext4_proc_root);
 876         }
 877         kobject_del(&sbi->s_kobj);
 878
 879         for (i = 0; i < sbi->s_gdb_count; i++)
 880                 brelse(sbi->s_group_desc[i]);
 881         ext4_kvfree(sbi->s_group_desc);
 882         ext4_kvfree(sbi->s_flex_groups);
 883         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 884         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 885         percpu_counter_destroy(&sbi->s_dirs_counter);
 886         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 887         brelse(sbi->s_sbh);
 888 #ifdef CONFIG_QUOTA
 889         for (i = 0; i < MAXQUOTAS; i++)
 890                 kfree(sbi->s_qf_names[i]);
 891 #endif
 892
 893         /* Debugging code just in case the in-memory inode orphan list
 894          * isn't empty.  The on-disk one can be non-empty if we've
 895          * detected an error and taken the fs readonly, but the
 896          * in-memory list had better be clean by this point. */
 897         if (!list_empty(&sbi->s_orphan))
 898                 dump_orphan_list(sb, sbi);
 899         J_ASSERT(list_empty(&sbi->s_orphan));
 900
 901         invalidate_bdev(sb->s_bdev);
 902         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 903                 /*
 904                  * Invalidate the journal device's buffers.  We don't want them
 905                  * floating about in memory - the physical journal device may
 906                  * hotswapped, and it breaks the `ro-after' testing code.
 907                  */
 908                 sync_blockdev(sbi->journal_bdev);
 909                 invalidate_bdev(sbi->journal_bdev);
 910                 ext4_blkdev_remove(sbi);
 911         }
 912         if (sbi->s_mmp_tsk)
 913                 kthread_stop(sbi->s_mmp_tsk);
 914         sb->s_fs_info = NULL;
 915         /*
 916          * Now that we are completely done shutting down the
 917          * superblock, we need to actually destroy the kobject.
 918          */
 919         kobject_put(&sbi->s_kobj);
 920         wait_for_completion(&sbi->s_kobj_unregister);
 921         if (sbi->s_chksum_driver)
 922                 crypto_free_shash(sbi->s_chksum_driver);
 923         kfree(sbi->s_blockgroup_lock);
 924         kfree(sbi);
 925 }
 926
 927 static struct kmem_cache *ext4_inode_cachep;
 928
 929 /*
 930  * Called inside transaction, so use GFP_NOFS
 931  */
 932 static struct inode *ext4_alloc_inode(struct super_block *sb)
 933 {
 934         struct ext4_inode_info *ei;
 935
 936         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 937         if (!ei)
 938                 return NULL;
 939
 940         ei->vfs_inode.i_version = 1;
 941         ei->vfs_inode.i_data.writeback_index = 0;
 942         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 943         INIT_LIST_HEAD(&ei->i_prealloc_list);
 944         spin_lock_init(&ei->i_prealloc_lock);
 945         ei->i_reserved_data_blocks = 0;
 946         ei->i_reserved_meta_blocks = 0;
 947         ei->i_allocated_meta_blocks = 0;
 948         ei->i_da_metadata_calc_len = 0;
 949         ei->i_da_metadata_calc_last_lblock = 0;
 950         spin_lock_init(&(ei->i_block_reservation_lock));
 951 #ifdef CONFIG_QUOTA
 952         ei->i_reserved_quota = 0;
 953 #endif
 954         ei->jinode = NULL;
 955         INIT_LIST_HEAD(&ei->i_completed_io_list);
 956         spin_lock_init(&ei->i_completed_io_lock);
 957         ei->i_sync_tid = 0;
 958         ei->i_datasync_tid = 0;
 959         atomic_set(&ei->i_ioend_count, 0);
 960         atomic_set(&ei->i_unwritten, 0);
 961
 962         return &ei->vfs_inode;
 963 }
 964
 965 static int ext4_drop_inode(struct inode *inode)
 966 {
 967         int drop = generic_drop_inode(inode);
 968
 969         trace_ext4_drop_inode(inode, drop);
 970         return drop;
 971 }
 972
 973 static void ext4_i_callback(struct rcu_head *head)
 974 {
 975         struct inode *inode = container_of(head, struct inode, i_rcu);
 976         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 977 }
 978
 979 static void ext4_destroy_inode(struct inode *inode)
 980 {
 981         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 982                 ext4_msg(inode->i_sb, KERN_ERR,
 983                          "Inode %lu (%p): orphan list check failed!",
 984                          inode->i_ino, EXT4_I(inode));
 985                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 986                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 987                                 true);
 988                 dump_stack();
 989         }
 990         call_rcu(&inode->i_rcu, ext4_i_callback);
 991 }
 992
 993 static void init_once(void *foo)
 994 {
 995         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 996
 997         INIT_LIST_HEAD(&ei->i_orphan);
 998 #ifdef CONFIG_EXT4_FS_XATTR
 999         init_rwsem(&ei->xattr_sem);
1000 #endif
1001         init_rwsem(&ei->i_data_sem);
1002         inode_init_once(&ei->vfs_inode);
1003 }
1004
1005 static int init_inodecache(void)
1006 {
1007         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1008                                              sizeof(struct ext4_inode_info),
1009                                              0, (SLAB_RECLAIM_ACCOUNT|
1010                                                 SLAB_MEM_SPREAD),
1011                                              init_once);
1012         if (ext4_inode_cachep == NULL)
1013                 return -ENOMEM;
1014         return 0;
1015 }
1016
1017 static void destroy_inodecache(void)
1018 {
1019         /*
1020          * Make sure all delayed rcu free inodes are flushed before we
1021          * destroy cache.
1022          */
1023         rcu_barrier();
1024         kmem_cache_destroy(ext4_inode_cachep);
1025 }
1026
1027 void ext4_clear_inode(struct inode *inode)
1028 {
1029         invalidate_inode_buffers(inode);
1030         clear_inode(inode);
1031         dquot_drop(inode);
1032         ext4_discard_preallocations(inode);
1033         if (EXT4_I(inode)->jinode) {
1034                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1035                                                EXT4_I(inode)->jinode);
1036                 jbd2_free_inode(EXT4_I(inode)->jinode);
1037                 EXT4_I(inode)->jinode = NULL;
1038         }
1039 }
1040
1041 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1042                                         u64 ino, u32 generation)
1043 {
1044         struct inode *inode;
1045
1046         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1047                 return ERR_PTR(-ESTALE);
1048         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1049                 return ERR_PTR(-ESTALE);
1050
1051         /* iget isn't really right if the inode is currently unallocated!!
1052          *
1053          * ext4_read_inode will return a bad_inode if the inode had been
1054          * deleted, so we should be safe.
1055          *
1056          * Currently we don't know the generation for parent directory, so
1057          * a generation of 0 means "accept any"
1058          */
1059         inode = ext4_iget(sb, ino);
1060         if (IS_ERR(inode))
1061                 return ERR_CAST(inode);
1062         if (generation && inode->i_generation != generation) {
1063                 iput(inode);
1064                 return ERR_PTR(-ESTALE);
1065         }
1066
1067         return inode;
1068 }
1069
1070 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1071                                         int fh_len, int fh_type)
1072 {
1073         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1074                                     ext4_nfs_get_inode);
1075 }
1076
1077 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1078                                         int fh_len, int fh_type)
1079 {
1080         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1081                                     ext4_nfs_get_inode);
1082 }
1083
1084 /*
1085  * Try to release metadata pages (indirect blocks, directories) which are
1086  * mapped via the block device.  Since these pages could have journal heads
1087  * which would prevent try_to_free_buffers() from freeing them, we must use
1088  * jbd2 layer's try_to_free_buffers() function to release them.
1089  */
1090 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1091                                  gfp_t wait)
1092 {
1093         journal_t *journal = EXT4_SB(sb)->s_journal;
1094
1095         WARN_ON(PageChecked(page));
1096         if (!page_has_buffers(page))
1097                 return 0;
1098         if (journal)
1099                 return jbd2_journal_try_to_free_buffers(journal, page,
1100                                                         wait & ~__GFP_WAIT);
1101         return try_to_free_buffers(page);
1102 }
1103
1104 #ifdef CONFIG_QUOTA
1105 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1106 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1107
1108 static int ext4_write_dquot(struct dquot *dquot);
1109 static int ext4_acquire_dquot(struct dquot *dquot);
1110 static int ext4_release_dquot(struct dquot *dquot);
1111 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1112 static int ext4_write_info(struct super_block *sb, int type);
1113 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1114                          struct path *path);
1115 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1116                                  int format_id);
1117 static int ext4_quota_off(struct super_block *sb, int type);
1118 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1119 static int ext4_quota_on_mount(struct super_block *sb, int type);
1120 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1121                                size_t len, loff_t off);
1122 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1123                                 const char *data, size_t len, loff_t off);
1124 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1125                              unsigned int flags);
1126 static int ext4_enable_quotas(struct super_block *sb);
1127
1128 static const struct dquot_operations ext4_quota_operations = {
1129         .get_reserved_space = ext4_get_reserved_space,
1130         .write_dquot    = ext4_write_dquot,
1131         .acquire_dquot  = ext4_acquire_dquot,
1132         .release_dquot  = ext4_release_dquot,
1133         .mark_dirty     = ext4_mark_dquot_dirty,
1134         .write_info     = ext4_write_info,
1135         .alloc_dquot    = dquot_alloc,
1136         .destroy_dquot  = dquot_destroy,
1137 };
1138
1139 static const struct quotactl_ops ext4_qctl_operations = {
1140         .quota_on       = ext4_quota_on,
1141         .quota_off      = ext4_quota_off,
1142         .quota_sync     = dquot_quota_sync,
1143         .get_info       = dquot_get_dqinfo,
1144         .set_info       = dquot_set_dqinfo,
1145         .get_dqblk      = dquot_get_dqblk,
1146         .set_dqblk      = dquot_set_dqblk
1147 };
1148
1149 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1150         .quota_on_meta  = ext4_quota_on_sysfile,
1151         .quota_off      = ext4_quota_off_sysfile,
1152         .quota_sync     = dquot_quota_sync,
1153         .get_info       = dquot_get_dqinfo,
1154         .set_info       = dquot_set_dqinfo,
1155         .get_dqblk      = dquot_get_dqblk,
1156         .set_dqblk      = dquot_set_dqblk
1157 };
1158 #endif
1159
1160 static const struct super_operations ext4_sops = {
1161         .alloc_inode    = ext4_alloc_inode,
1162         .destroy_inode  = ext4_destroy_inode,
1163         .write_inode    = ext4_write_inode,
1164         .dirty_inode    = ext4_dirty_inode,
1165         .drop_inode     = ext4_drop_inode,
1166         .evict_inode    = ext4_evict_inode,
1167         .put_super      = ext4_put_super,
1168         .sync_fs        = ext4_sync_fs,
1169         .freeze_fs      = ext4_freeze,
1170         .unfreeze_fs    = ext4_unfreeze,
1171         .statfs         = ext4_statfs,
1172         .remount_fs     = ext4_remount,
1173         .show_options   = ext4_show_options,
1174 #ifdef CONFIG_QUOTA
1175         .quota_read     = ext4_quota_read,
1176         .quota_write    = ext4_quota_write,
1177 #endif
1178         .bdev_try_to_free_page = bdev_try_to_free_page,
1179 };
1180
1181 static const struct super_operations ext4_nojournal_sops = {
1182         .alloc_inode    = ext4_alloc_inode,
1183         .destroy_inode  = ext4_destroy_inode,
1184         .write_inode    = ext4_write_inode,
1185         .dirty_inode    = ext4_dirty_inode,
1186         .drop_inode     = ext4_drop_inode,
1187         .evict_inode    = ext4_evict_inode,
1188         .put_super      = ext4_put_super,
1189         .statfs         = ext4_statfs,
1190         .remount_fs     = ext4_remount,
1191         .show_options   = ext4_show_options,
1192 #ifdef CONFIG_QUOTA
1193         .quota_read     = ext4_quota_read,
1194         .quota_write    = ext4_quota_write,
1195 #endif
1196         .bdev_try_to_free_page = bdev_try_to_free_page,
1197 };
1198
1199 static const struct export_operations ext4_export_ops = {
1200         .fh_to_dentry = ext4_fh_to_dentry,
1201         .fh_to_parent = ext4_fh_to_parent,
1202         .get_parent = ext4_get_parent,
1203 };
1204
1205 enum {
1206         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1207         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1208         Opt_nouid32, Opt_debug, Opt_removed,
1209         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1210         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1211         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1212         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1213         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1214         Opt_data_err_abort, Opt_data_err_ignore,
1215         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1216         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1217         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1218         Opt_usrquota, Opt_grpquota, Opt_i_version,
1219         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1220         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1221         Opt_inode_readahead_blks, Opt_journal_ioprio,
1222         Opt_dioread_nolock, Opt_dioread_lock,
1223         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1224         Opt_max_dir_size_kb,
1225 };
1226
1227 static const match_table_t tokens = {
1228         {Opt_bsd_df, "bsddf"},
1229         {Opt_minix_df, "minixdf"},
1230         {Opt_grpid, "grpid"},
1231         {Opt_grpid, "bsdgroups"},
1232         {Opt_nogrpid, "nogrpid"},
1233         {Opt_nogrpid, "sysvgroups"},
1234         {Opt_resgid, "resgid=%u"},
1235         {Opt_resuid, "resuid=%u"},
1236         {Opt_sb, "sb=%u"},
1237         {Opt_err_cont, "errors=continue"},
1238         {Opt_err_panic, "errors=panic"},
1239         {Opt_err_ro, "errors=remount-ro"},
1240         {Opt_nouid32, "nouid32"},
1241         {Opt_debug, "debug"},
1242         {Opt_removed, "oldalloc"},
1243         {Opt_removed, "orlov"},
1244         {Opt_user_xattr, "user_xattr"},
1245         {Opt_nouser_xattr, "nouser_xattr"},
1246         {Opt_acl, "acl"},
1247         {Opt_noacl, "noacl"},
1248         {Opt_noload, "norecovery"},
1249         {Opt_noload, "noload"},
1250         {Opt_removed, "nobh"},
1251         {Opt_removed, "bh"},
1252         {Opt_commit, "commit=%u"},
1253         {Opt_min_batch_time, "min_batch_time=%u"},
1254         {Opt_max_batch_time, "max_batch_time=%u"},
1255         {Opt_journal_dev, "journal_dev=%u"},
1256         {Opt_journal_checksum, "journal_checksum"},
1257         {Opt_journal_async_commit, "journal_async_commit"},
1258         {Opt_abort, "abort"},
1259         {Opt_data_journal, "data=journal"},
1260         {Opt_data_ordered, "data=ordered"},
1261         {Opt_data_writeback, "data=writeback"},
1262         {Opt_data_err_abort, "data_err=abort"},
1263         {Opt_data_err_ignore, "data_err=ignore"},
1264         {Opt_offusrjquota, "usrjquota="},
1265         {Opt_usrjquota, "usrjquota=%s"},
1266         {Opt_offgrpjquota, "grpjquota="},
1267         {Opt_grpjquota, "grpjquota=%s"},
1268         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1269         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1270         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1271         {Opt_grpquota, "grpquota"},
1272         {Opt_noquota, "noquota"},
1273         {Opt_quota, "quota"},
1274         {Opt_usrquota, "usrquota"},
1275         {Opt_barrier, "barrier=%u"},
1276         {Opt_barrier, "barrier"},
1277         {Opt_nobarrier, "nobarrier"},
1278         {Opt_i_version, "i_version"},
1279         {Opt_stripe, "stripe=%u"},
1280         {Opt_delalloc, "delalloc"},
1281         {Opt_nodelalloc, "nodelalloc"},
1282         {Opt_mblk_io_submit, "mblk_io_submit"},
1283         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1284         {Opt_block_validity, "block_validity"},
1285         {Opt_noblock_validity, "noblock_validity"},
1286         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1287         {Opt_journal_ioprio, "journal_ioprio=%u"},
1288         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1289         {Opt_auto_da_alloc, "auto_da_alloc"},
1290         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1291         {Opt_dioread_nolock, "dioread_nolock"},
1292         {Opt_dioread_lock, "dioread_lock"},
1293         {Opt_discard, "discard"},
1294         {Opt_nodiscard, "nodiscard"},
1295         {Opt_init_itable, "init_itable=%u"},
1296         {Opt_init_itable, "init_itable"},
1297         {Opt_noinit_itable, "noinit_itable"},
1298         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1299         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1300         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1301         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1302         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1303         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1304         {Opt_err, NULL},
1305 };
1306
1307 static ext4_fsblk_t get_sb_block(void **data)
1308 {
1309         ext4_fsblk_t    sb_block;
1310         char            *options = (char *) *data;
1311
1312         if (!options || strncmp(options, "sb=", 3) != 0)
1313                 return 1;       /* Default location */
1314
1315         options += 3;
1316         /* TODO: use simple_strtoll with >32bit ext4 */
1317         sb_block = simple_strtoul(options, &options, 0);
1318         if (*options && *options != ',') {
1319                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1320                        (char *) *data);
1321                 return 1;
1322         }
1323         if (*options == ',')
1324                 options++;
1325         *data = (void *) options;
1326
1327         return sb_block;
1328 }
1329
1330 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1331 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1332         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1333
1334 #ifdef CONFIG_QUOTA
1335 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1336 {
1337         struct ext4_sb_info *sbi = EXT4_SB(sb);
1338         char *qname;
1339
1340         if (sb_any_quota_loaded(sb) &&
1341                 !sbi->s_qf_names[qtype]) {
1342                 ext4_msg(sb, KERN_ERR,
1343                         "Cannot change journaled "
1344                         "quota options when quota turned on");
1345                 return -1;
1346         }
1347         qname = match_strdup(args);
1348         if (!qname) {
1349                 ext4_msg(sb, KERN_ERR,
1350                         "Not enough memory for storing quotafile name");
1351                 return -1;
1352         }
1353         if (sbi->s_qf_names[qtype] &&
1354                 strcmp(sbi->s_qf_names[qtype], qname)) {
1355                 ext4_msg(sb, KERN_ERR,
1356                         "%s quota file already specified", QTYPE2NAME(qtype));
1357                 kfree(qname);
1358                 return -1;
1359         }
1360         sbi->s_qf_names[qtype] = qname;
1361         if (strchr(sbi->s_qf_names[qtype], '/')) {
1362                 ext4_msg(sb, KERN_ERR,
1363                         "quotafile must be on filesystem root");
1364                 kfree(sbi->s_qf_names[qtype]);
1365                 sbi->s_qf_names[qtype] = NULL;
1366                 return -1;
1367         }
1368         set_opt(sb, QUOTA);
1369         return 1;
1370 }
1371
1372 static int clear_qf_name(struct super_block *sb, int qtype)
1373 {
1374
1375         struct ext4_sb_info *sbi = EXT4_SB(sb);
1376
1377         if (sb_any_quota_loaded(sb) &&
1378                 sbi->s_qf_names[qtype]) {
1379                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1380                         " when quota turned on");
1381                 return -1;
1382         }
1383         /*
1384          * The space will be released later when all options are confirmed
1385          * to be correct
1386          */
1387         sbi->s_qf_names[qtype] = NULL;
1388         return 1;
1389 }
1390 #endif
1391
1392 #define MOPT_SET        0x0001
1393 #define MOPT_CLEAR      0x0002
1394 #define MOPT_NOSUPPORT  0x0004
1395 #define MOPT_EXPLICIT   0x0008
1396 #define MOPT_CLEAR_ERR  0x0010
1397 #define MOPT_GTE0       0x0020
1398 #ifdef CONFIG_QUOTA
1399 #define MOPT_Q          0
1400 #define MOPT_QFMT       0x0040
1401 #else
1402 #define MOPT_Q          MOPT_NOSUPPORT
1403 #define MOPT_QFMT       MOPT_NOSUPPORT
1404 #endif
1405 #define MOPT_DATAJ      0x0080
1406
1407 static const struct mount_opts {
1408         int     token;
1409         int     mount_opt;
1410         int     flags;
1411 } ext4_mount_opts[] = {
1412         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1413         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1414         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1415         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1416         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1417         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1418         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1419         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1420         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1421         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1422         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1423         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1424         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1425         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1426         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1427         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1428                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1429         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1430         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1431         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1434         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1435         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1436         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1437         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1438         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1439         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1440         {Opt_commit, 0, MOPT_GTE0},
1441         {Opt_max_batch_time, 0, MOPT_GTE0},
1442         {Opt_min_batch_time, 0, MOPT_GTE0},
1443         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1444         {Opt_init_itable, 0, MOPT_GTE0},
1445         {Opt_stripe, 0, MOPT_GTE0},
1446         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1447         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1448         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1449 #ifdef CONFIG_EXT4_FS_XATTR
1450         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1451         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1452 #else
1453         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1454         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1455 #endif
1456 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1457         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1458         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1459 #else
1460         {Opt_acl, 0, MOPT_NOSUPPORT},
1461         {Opt_noacl, 0, MOPT_NOSUPPORT},
1462 #endif
1463         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1464         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1465         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1466         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1467                                                         MOPT_SET | MOPT_Q},
1468         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1469                                                         MOPT_SET | MOPT_Q},
1470         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1471                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1472         {Opt_usrjquota, 0, MOPT_Q},
1473         {Opt_grpjquota, 0, MOPT_Q},
1474         {Opt_offusrjquota, 0, MOPT_Q},
1475         {Opt_offgrpjquota, 0, MOPT_Q},
1476         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1477         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1478         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1479         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
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_max_dir_size_kb) {
1596                         sbi->s_max_dir_size_kb = arg;
1597                 } else if (token == Opt_stripe) {
1598                         sbi->s_stripe = arg;
1599                 } else if (m->flags & MOPT_DATAJ) {
1600                         if (is_remount) {
1601                                 if (!sbi->s_journal)
1602                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1603                                 else if (test_opt(sb, DATA_FLAGS) !=
1604                                          m->mount_opt) {
1605                                         ext4_msg(sb, KERN_ERR,
1606                                          "Cannot change data mode on remount");
1607                                         return -1;
1608                                 }
1609                         } else {
1610                                 clear_opt(sb, DATA_FLAGS);
1611                                 sbi->s_mount_opt |= m->mount_opt;
1612                         }
1613 #ifdef CONFIG_QUOTA
1614                 } else if (m->flags & MOPT_QFMT) {
1615                         if (sb_any_quota_loaded(sb) &&
1616                             sbi->s_jquota_fmt != m->mount_opt) {
1617                                 ext4_msg(sb, KERN_ERR, "Cannot "
1618                                          "change journaled quota options "
1619                                          "when quota turned on");
1620                                 return -1;
1621                         }
1622                         sbi->s_jquota_fmt = m->mount_opt;
1623 #endif
1624                 } else {
1625                         if (!args->from)
1626                                 arg = 1;
1627                         if (m->flags & MOPT_CLEAR)
1628                                 arg = !arg;
1629                         else if (unlikely(!(m->flags & MOPT_SET))) {
1630                                 ext4_msg(sb, KERN_WARNING,
1631                                          "buggy handling of option %s", opt);
1632                                 WARN_ON(1);
1633                                 return -1;
1634                         }
1635                         if (arg != 0)
1636                                 sbi->s_mount_opt |= m->mount_opt;
1637                         else
1638                                 sbi->s_mount_opt &= ~m->mount_opt;
1639                 }
1640                 return 1;
1641         }
1642         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1643                  "or missing value", opt);
1644         return -1;
1645 }
1646
1647 static int parse_options(char *options, struct super_block *sb,
1648                          unsigned long *journal_devnum,
1649                          unsigned int *journal_ioprio,
1650                          int is_remount)
1651 {
1652 #ifdef CONFIG_QUOTA
1653         struct ext4_sb_info *sbi = EXT4_SB(sb);
1654 #endif
1655         char *p;
1656         substring_t args[MAX_OPT_ARGS];
1657         int token;
1658
1659         if (!options)
1660                 return 1;
1661
1662         while ((p = strsep(&options, ",")) != NULL) {
1663                 if (!*p)
1664                         continue;
1665                 /*
1666                  * Initialize args struct so we know whether arg was
1667                  * found; some options take optional arguments.
1668                  */
1669                 args[0].to = args[0].from = NULL;
1670                 token = match_token(p, tokens, args);
1671                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1672                                      journal_ioprio, is_remount) < 0)
1673                         return 0;
1674         }
1675 #ifdef CONFIG_QUOTA
1676         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1677                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1678                         clear_opt(sb, USRQUOTA);
1679
1680                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1681                         clear_opt(sb, GRPQUOTA);
1682
1683                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1684                         ext4_msg(sb, KERN_ERR, "old and new quota "
1685                                         "format mixing");
1686                         return 0;
1687                 }
1688
1689                 if (!sbi->s_jquota_fmt) {
1690                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1691                                         "not specified");
1692                         return 0;
1693                 }
1694         } else {
1695                 if (sbi->s_jquota_fmt) {
1696                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1697                                         "specified with no journaling "
1698                                         "enabled");
1699                         return 0;
1700                 }
1701         }
1702 #endif
1703         return 1;
1704 }
1705
1706 static inline void ext4_show_quota_options(struct seq_file *seq,
1707                                            struct super_block *sb)
1708 {
1709 #if defined(CONFIG_QUOTA)
1710         struct ext4_sb_info *sbi = EXT4_SB(sb);
1711
1712         if (sbi->s_jquota_fmt) {
1713                 char *fmtname = "";
1714
1715                 switch (sbi->s_jquota_fmt) {
1716                 case QFMT_VFS_OLD:
1717                         fmtname = "vfsold";
1718                         break;
1719                 case QFMT_VFS_V0:
1720                         fmtname = "vfsv0";
1721                         break;
1722                 case QFMT_VFS_V1:
1723                         fmtname = "vfsv1";
1724                         break;
1725                 }
1726                 seq_printf(seq, ",jqfmt=%s", fmtname);
1727         }
1728
1729         if (sbi->s_qf_names[USRQUOTA])
1730                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1731
1732         if (sbi->s_qf_names[GRPQUOTA])
1733                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1734
1735         if (test_opt(sb, USRQUOTA))
1736                 seq_puts(seq, ",usrquota");
1737
1738         if (test_opt(sb, GRPQUOTA))
1739                 seq_puts(seq, ",grpquota");
1740 #endif
1741 }
1742
1743 static const char *token2str(int token)
1744 {
1745         const struct match_token *t;
1746
1747         for (t = tokens; t->token != Opt_err; t++)
1748                 if (t->token == token && !strchr(t->pattern, '='))
1749                         break;
1750         return t->pattern;
1751 }
1752
1753 /*
1754  * Show an option if
1755  *  - it's set to a non-default value OR
1756  *  - if the per-sb default is different from the global default
1757  */
1758 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1759                               int nodefs)
1760 {
1761         struct ext4_sb_info *sbi = EXT4_SB(sb);
1762         struct ext4_super_block *es = sbi->s_es;
1763         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1764         const struct mount_opts *m;
1765         char sep = nodefs ? '\n' : ',';
1766
1767 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1768 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1769
1770         if (sbi->s_sb_block != 1)
1771                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1772
1773         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1774                 int want_set = m->flags & MOPT_SET;
1775                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1776                     (m->flags & MOPT_CLEAR_ERR))
1777                         continue;
1778                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1779                         continue; /* skip if same as the default */
1780                 if ((want_set &&
1781                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1782                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1783                         continue; /* select Opt_noFoo vs Opt_Foo */
1784                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1785         }
1786
1787         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1788             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1789                 SEQ_OPTS_PRINT("resuid=%u",
1790                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1791         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1792             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1793                 SEQ_OPTS_PRINT("resgid=%u",
1794                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1795         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1796         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1797                 SEQ_OPTS_PUTS("errors=remount-ro");
1798         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1799                 SEQ_OPTS_PUTS("errors=continue");
1800         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1801                 SEQ_OPTS_PUTS("errors=panic");
1802         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1803                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1804         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1805                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1806         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1807                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1808         if (sb->s_flags & MS_I_VERSION)
1809                 SEQ_OPTS_PUTS("i_version");
1810         if (nodefs || sbi->s_stripe)
1811                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1812         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1813                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1814                         SEQ_OPTS_PUTS("data=journal");
1815                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1816                         SEQ_OPTS_PUTS("data=ordered");
1817                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1818                         SEQ_OPTS_PUTS("data=writeback");
1819         }
1820         if (nodefs ||
1821             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1822                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1823                                sbi->s_inode_readahead_blks);
1824
1825         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1826                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1827                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1828         if (nodefs || sbi->s_max_dir_size_kb)
1829                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1830
1831         ext4_show_quota_options(seq, sb);
1832         return 0;
1833 }
1834
1835 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1836 {
1837         return _ext4_show_options(seq, root->d_sb, 0);
1838 }
1839
1840 static int options_seq_show(struct seq_file *seq, void *offset)
1841 {
1842         struct super_block *sb = seq->private;
1843         int rc;
1844
1845         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1846         rc = _ext4_show_options(seq, sb, 1);
1847         seq_puts(seq, "\n");
1848         return rc;
1849 }
1850
1851 static int options_open_fs(struct inode *inode, struct file *file)
1852 {
1853         return single_open(file, options_seq_show, PDE(inode)->data);
1854 }
1855
1856 static const struct file_operations ext4_seq_options_fops = {
1857         .owner = THIS_MODULE,
1858         .open = options_open_fs,
1859         .read = seq_read,
1860         .llseek = seq_lseek,
1861         .release = single_release,
1862 };
1863
1864 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1865                             int read_only)
1866 {
1867         struct ext4_sb_info *sbi = EXT4_SB(sb);
1868         int res = 0;
1869
1870         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1871                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1872                          "forcing read-only mode");
1873                 res = MS_RDONLY;
1874         }
1875         if (read_only)
1876                 goto done;
1877         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1878                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1879                          "running e2fsck is recommended");
1880         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1881                 ext4_msg(sb, KERN_WARNING,
1882                          "warning: mounting fs with errors, "
1883                          "running e2fsck is recommended");
1884         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1885                  le16_to_cpu(es->s_mnt_count) >=
1886                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1887                 ext4_msg(sb, KERN_WARNING,
1888                          "warning: maximal mount count reached, "
1889                          "running e2fsck is recommended");
1890         else if (le32_to_cpu(es->s_checkinterval) &&
1891                 (le32_to_cpu(es->s_lastcheck) +
1892                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1893                 ext4_msg(sb, KERN_WARNING,
1894                          "warning: checktime reached, "
1895                          "running e2fsck is recommended");
1896         if (!sbi->s_journal)
1897                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1898         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1899                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1900         le16_add_cpu(&es->s_mnt_count, 1);
1901         es->s_mtime = cpu_to_le32(get_seconds());
1902         ext4_update_dynamic_rev(sb);
1903         if (sbi->s_journal)
1904                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1905
1906         ext4_commit_super(sb, 1);
1907 done:
1908         if (test_opt(sb, DEBUG))
1909                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1910                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1911                         sb->s_blocksize,
1912                         sbi->s_groups_count,
1913                         EXT4_BLOCKS_PER_GROUP(sb),
1914                         EXT4_INODES_PER_GROUP(sb),
1915                         sbi->s_mount_opt, sbi->s_mount_opt2);
1916
1917         cleancache_init_fs(sb);
1918         return res;
1919 }
1920
1921 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1922 {
1923         struct ext4_sb_info *sbi = EXT4_SB(sb);
1924         struct flex_groups *new_groups;
1925         int size;
1926
1927         if (!sbi->s_log_groups_per_flex)
1928                 return 0;
1929
1930         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1931         if (size <= sbi->s_flex_groups_allocated)
1932                 return 0;
1933
1934         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1935         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1936         if (!new_groups) {
1937                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1938                          size / (int) sizeof(struct flex_groups));
1939                 return -ENOMEM;
1940         }
1941
1942         if (sbi->s_flex_groups) {
1943                 memcpy(new_groups, sbi->s_flex_groups,
1944                        (sbi->s_flex_groups_allocated *
1945                         sizeof(struct flex_groups)));
1946                 ext4_kvfree(sbi->s_flex_groups);
1947         }
1948         sbi->s_flex_groups = new_groups;
1949         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1950         return 0;
1951 }
1952
1953 static int ext4_fill_flex_info(struct super_block *sb)
1954 {
1955         struct ext4_sb_info *sbi = EXT4_SB(sb);
1956         struct ext4_group_desc *gdp = NULL;
1957         ext4_group_t flex_group;
1958         unsigned int groups_per_flex = 0;
1959         int i, err;
1960
1961         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1962         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1963                 sbi->s_log_groups_per_flex = 0;
1964                 return 1;
1965         }
1966         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1967
1968         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1969         if (err)
1970                 goto failed;
1971
1972         for (i = 0; i < sbi->s_groups_count; i++) {
1973                 gdp = ext4_get_group_desc(sb, i, NULL);
1974
1975                 flex_group = ext4_flex_group(sbi, i);
1976                 atomic_add(ext4_free_inodes_count(sb, gdp),
1977                            &sbi->s_flex_groups[flex_group].free_inodes);
1978                 atomic_add(ext4_free_group_clusters(sb, gdp),
1979                            &sbi->s_flex_groups[flex_group].free_clusters);
1980                 atomic_add(ext4_used_dirs_count(sb, gdp),
1981                            &sbi->s_flex_groups[flex_group].used_dirs);
1982         }
1983
1984         return 1;
1985 failed:
1986         return 0;
1987 }
1988
1989 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1990                                    struct ext4_group_desc *gdp)
1991 {
1992         int offset;
1993         __u16 crc = 0;
1994         __le32 le_group = cpu_to_le32(block_group);
1995
1996         if ((sbi->s_es->s_feature_ro_compat &
1997              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1998                 /* Use new metadata_csum algorithm */
1999                 __u16 old_csum;
2000                 __u32 csum32;
2001
2002                 old_csum = gdp->bg_checksum;
2003                 gdp->bg_checksum = 0;
2004                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2005                                      sizeof(le_group));
2006                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2007                                      sbi->s_desc_size);
2008                 gdp->bg_checksum = old_csum;
2009
2010                 crc = csum32 & 0xFFFF;
2011                 goto out;
2012         }
2013
2014         /* old crc16 code */
2015         offset = offsetof(struct ext4_group_desc, bg_checksum);
2016
2017         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2018         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2019         crc = crc16(crc, (__u8 *)gdp, offset);
2020         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2021         /* for checksum of struct ext4_group_desc do the rest...*/
2022         if ((sbi->s_es->s_feature_incompat &
2023              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2024             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2025                 crc = crc16(crc, (__u8 *)gdp + offset,
2026                             le16_to_cpu(sbi->s_es->s_desc_size) -
2027                                 offset);
2028
2029 out:
2030         return cpu_to_le16(crc);
2031 }
2032
2033 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2034                                 struct ext4_group_desc *gdp)
2035 {
2036         if (ext4_has_group_desc_csum(sb) &&
2037             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2038                                                       block_group, gdp)))
2039                 return 0;
2040
2041         return 1;
2042 }
2043
2044 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2045                               struct ext4_group_desc *gdp)
2046 {
2047         if (!ext4_has_group_desc_csum(sb))
2048                 return;
2049         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2050 }
2051
2052 /* Called at mount-time, super-block is locked */
2053 static int ext4_check_descriptors(struct super_block *sb,
2054                                   ext4_group_t *first_not_zeroed)
2055 {
2056         struct ext4_sb_info *sbi = EXT4_SB(sb);
2057         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2058         ext4_fsblk_t last_block;
2059         ext4_fsblk_t block_bitmap;
2060         ext4_fsblk_t inode_bitmap;
2061         ext4_fsblk_t inode_table;
2062         int flexbg_flag = 0;
2063         ext4_group_t i, grp = sbi->s_groups_count;
2064
2065         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2066                 flexbg_flag = 1;
2067
2068         ext4_debug("Checking group descriptors");
2069
2070         for (i = 0; i < sbi->s_groups_count; i++) {
2071                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2072
2073                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2074                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2075                 else
2076                         last_block = first_block +
2077                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2078
2079                 if ((grp == sbi->s_groups_count) &&
2080                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2081                         grp = i;
2082
2083                 block_bitmap = ext4_block_bitmap(sb, gdp);
2084                 if (block_bitmap < first_block || block_bitmap > last_block) {
2085                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2086                                "Block bitmap for group %u not in group "
2087                                "(block %llu)!", i, block_bitmap);
2088                         return 0;
2089                 }
2090                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2091                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2092                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2093                                "Inode bitmap for group %u not in group "
2094                                "(block %llu)!", i, inode_bitmap);
2095                         return 0;
2096                 }
2097                 inode_table = ext4_inode_table(sb, gdp);
2098                 if (inode_table < first_block ||
2099                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2100                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2101                                "Inode table for group %u not in group "
2102                                "(block %llu)!", i, inode_table);
2103                         return 0;
2104                 }
2105                 ext4_lock_group(sb, i);
2106                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                  "Checksum for group %u failed (%u!=%u)",
2109                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2110                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2111                         if (!(sb->s_flags & MS_RDONLY)) {
2112                                 ext4_unlock_group(sb, i);
2113                                 return 0;
2114                         }
2115                 }
2116                 ext4_unlock_group(sb, i);
2117                 if (!flexbg_flag)
2118                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2119         }
2120         if (NULL != first_not_zeroed)
2121                 *first_not_zeroed = grp;
2122
2123         ext4_free_blocks_count_set(sbi->s_es,
2124                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2125         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2126         return 1;
2127 }
2128
2129 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2130  * the superblock) which were deleted from all directories, but held open by
2131  * a process at the time of a crash.  We walk the list and try to delete these
2132  * inodes at recovery time (only with a read-write filesystem).
2133  *
2134  * In order to keep the orphan inode chain consistent during traversal (in
2135  * case of crash during recovery), we link each inode into the superblock
2136  * orphan list_head and handle it the same way as an inode deletion during
2137  * normal operation (which journals the operations for us).
2138  *
2139  * We only do an iget() and an iput() on each inode, which is very safe if we
2140  * accidentally point at an in-use or already deleted inode.  The worst that
2141  * can happen in this case is that we get a "bit already cleared" message from
2142  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2143  * e2fsck was run on this filesystem, and it must have already done the orphan
2144  * inode cleanup for us, so we can safely abort without any further action.
2145  */
2146 static void ext4_orphan_cleanup(struct super_block *sb,
2147                                 struct ext4_super_block *es)
2148 {
2149         unsigned int s_flags = sb->s_flags;
2150         int nr_orphans = 0, nr_truncates = 0;
2151 #ifdef CONFIG_QUOTA
2152         int i;
2153 #endif
2154         if (!es->s_last_orphan) {
2155                 jbd_debug(4, "no orphan inodes to clean up\n");
2156                 return;
2157         }
2158
2159         if (bdev_read_only(sb->s_bdev)) {
2160                 ext4_msg(sb, KERN_ERR, "write access "
2161                         "unavailable, skipping orphan cleanup");
2162                 return;
2163         }
2164
2165         /* Check if feature set would not allow a r/w mount */
2166         if (!ext4_feature_set_ok(sb, 0)) {
2167                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2168                          "unknown ROCOMPAT features");
2169                 return;
2170         }
2171
2172         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2173                 /* don't clear list on RO mount w/ errors */
2174                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2175                         jbd_debug(1, "Errors on filesystem, "
2176                                   "clearing orphan list.\n");
2177                         es->s_last_orphan = 0;
2178                 }
2179                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2180                 return;
2181         }
2182
2183         if (s_flags & MS_RDONLY) {
2184                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2185                 sb->s_flags &= ~MS_RDONLY;
2186         }
2187 #ifdef CONFIG_QUOTA
2188         /* Needed for iput() to work correctly and not trash data */
2189         sb->s_flags |= MS_ACTIVE;
2190         /* Turn on quotas so that they are updated correctly */
2191         for (i = 0; i < MAXQUOTAS; i++) {
2192                 if (EXT4_SB(sb)->s_qf_names[i]) {
2193                         int ret = ext4_quota_on_mount(sb, i);
2194                         if (ret < 0)
2195                                 ext4_msg(sb, KERN_ERR,
2196                                         "Cannot turn on journaled "
2197                                         "quota: error %d", ret);
2198                 }
2199         }
2200 #endif
2201
2202         while (es->s_last_orphan) {
2203                 struct inode *inode;
2204
2205                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2206                 if (IS_ERR(inode)) {
2207                         es->s_last_orphan = 0;
2208                         break;
2209                 }
2210
2211                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2212                 dquot_initialize(inode);
2213                 if (inode->i_nlink) {
2214                         ext4_msg(sb, KERN_DEBUG,
2215                                 "%s: truncating inode %lu to %lld bytes",
2216                                 __func__, inode->i_ino, inode->i_size);
2217                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2218                                   inode->i_ino, inode->i_size);
2219                         ext4_truncate(inode);
2220                         nr_truncates++;
2221                 } else {
2222                         ext4_msg(sb, KERN_DEBUG,
2223                                 "%s: deleting unreferenced inode %lu",
2224                                 __func__, inode->i_ino);
2225                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2226                                   inode->i_ino);
2227                         nr_orphans++;
2228                 }
2229                 iput(inode);  /* The delete magic happens here! */
2230         }
2231
2232 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2233
2234         if (nr_orphans)
2235                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2236                        PLURAL(nr_orphans));
2237         if (nr_truncates)
2238                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2239                        PLURAL(nr_truncates));
2240 #ifdef CONFIG_QUOTA
2241         /* Turn quotas off */
2242         for (i = 0; i < MAXQUOTAS; i++) {
2243                 if (sb_dqopt(sb)->files[i])
2244                         dquot_quota_off(sb, i);
2245         }
2246 #endif
2247         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2248 }
2249
2250 /*
2251  * Maximal extent format file size.
2252  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2253  * extent format containers, within a sector_t, and within i_blocks
2254  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2255  * so that won't be a limiting factor.
2256  *
2257  * However there is other limiting factor. We do store extents in the form
2258  * of starting block and length, hence the resulting length of the extent
2259  * covering maximum file size must fit into on-disk format containers as
2260  * well. Given that length is always by 1 unit bigger than max unit (because
2261  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2262  *
2263  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2264  */
2265 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2266 {
2267         loff_t res;
2268         loff_t upper_limit = MAX_LFS_FILESIZE;
2269
2270         /* small i_blocks in vfs inode? */
2271         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2272                 /*
2273                  * CONFIG_LBDAF is not enabled implies the inode
2274                  * i_block represent total blocks in 512 bytes
2275                  * 32 == size of vfs inode i_blocks * 8
2276                  */
2277                 upper_limit = (1LL << 32) - 1;
2278
2279                 /* total blocks in file system block size */
2280                 upper_limit >>= (blkbits - 9);
2281                 upper_limit <<= blkbits;
2282         }
2283
2284         /*
2285          * 32-bit extent-start container, ee_block. We lower the maxbytes
2286          * by one fs block, so ee_len can cover the extent of maximum file
2287          * size
2288          */
2289         res = (1LL << 32) - 1;
2290         res <<= blkbits;
2291
2292         /* Sanity check against vm- & vfs- imposed limits */
2293         if (res > upper_limit)
2294                 res = upper_limit;
2295
2296         return res;
2297 }
2298
2299 /*
2300  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2301  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2302  * We need to be 1 filesystem block less than the 2^48 sector limit.
2303  */
2304 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2305 {
2306         loff_t res = EXT4_NDIR_BLOCKS;
2307         int meta_blocks;
2308         loff_t upper_limit;
2309         /* This is calculated to be the largest file size for a dense, block
2310          * mapped file such that the file's total number of 512-byte sectors,
2311          * including data and all indirect blocks, does not exceed (2^48 - 1).
2312          *
2313          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2314          * number of 512-byte sectors of the file.
2315          */
2316
2317         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2318                 /*
2319                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2320                  * the inode i_block field represents total file blocks in
2321                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2322                  */
2323                 upper_limit = (1LL << 32) - 1;
2324
2325                 /* total blocks in file system block size */
2326                 upper_limit >>= (bits - 9);
2327
2328         } else {
2329                 /*
2330                  * We use 48 bit ext4_inode i_blocks
2331                  * With EXT4_HUGE_FILE_FL set the i_blocks
2332                  * represent total number of blocks in
2333                  * file system block size
2334                  */
2335                 upper_limit = (1LL << 48) - 1;
2336
2337         }
2338
2339         /* indirect blocks */
2340         meta_blocks = 1;
2341         /* double indirect blocks */
2342         meta_blocks += 1 + (1LL << (bits-2));
2343         /* tripple indirect blocks */
2344         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2345
2346         upper_limit -= meta_blocks;
2347         upper_limit <<= bits;
2348
2349         res += 1LL << (bits-2);
2350         res += 1LL << (2*(bits-2));
2351         res += 1LL << (3*(bits-2));
2352         res <<= bits;
2353         if (res > upper_limit)
2354                 res = upper_limit;
2355
2356         if (res > MAX_LFS_FILESIZE)
2357                 res = MAX_LFS_FILESIZE;
2358
2359         return res;
2360 }
2361
2362 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2363                                    ext4_fsblk_t logical_sb_block, int nr)
2364 {
2365         struct ext4_sb_info *sbi = EXT4_SB(sb);
2366         ext4_group_t bg, first_meta_bg;
2367         int has_super = 0;
2368
2369         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2370
2371         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2372             nr < first_meta_bg)
2373                 return logical_sb_block + nr + 1;
2374         bg = sbi->s_desc_per_block * nr;
2375         if (ext4_bg_has_super(sb, bg))
2376                 has_super = 1;
2377
2378         return (has_super + ext4_group_first_block_no(sb, bg));
2379 }
2380
2381 /**
2382  * ext4_get_stripe_size: Get the stripe size.
2383  * @sbi: In memory super block info
2384  *
2385  * If we have specified it via mount option, then
2386  * use the mount option value. If the value specified at mount time is
2387  * greater than the blocks per group use the super block value.
2388  * If the super block value is greater than blocks per group return 0.
2389  * Allocator needs it be less than blocks per group.
2390  *
2391  */
2392 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2393 {
2394         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2395         unsigned long stripe_width =
2396                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2397         int ret;
2398
2399         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2400                 ret = sbi->s_stripe;
2401         else if (stripe_width <= sbi->s_blocks_per_group)
2402                 ret = stripe_width;
2403         else if (stride <= sbi->s_blocks_per_group)
2404                 ret = stride;
2405         else
2406                 ret = 0;
2407
2408         /*
2409          * If the stripe width is 1, this makes no sense and
2410          * we set it to 0 to turn off stripe handling code.
2411          */
2412         if (ret <= 1)
2413                 ret = 0;
2414
2415         return ret;
2416 }
2417
2418 /* sysfs supprt */
2419
2420 struct ext4_attr {
2421         struct attribute attr;
2422         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2423         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2424                          const char *, size_t);
2425         int offset;
2426 };
2427
2428 static int parse_strtoul(const char *buf,
2429                 unsigned long max, unsigned long *value)
2430 {
2431         char *endp;
2432
2433         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2434         endp = skip_spaces(endp);
2435         if (*endp || *value > max)
2436                 return -EINVAL;
2437
2438         return 0;
2439 }
2440
2441 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2442                                               struct ext4_sb_info *sbi,
2443                                               char *buf)
2444 {
2445         return snprintf(buf, PAGE_SIZE, "%llu\n",
2446                 (s64) EXT4_C2B(sbi,
2447                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2448 }
2449
2450 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2451                                          struct ext4_sb_info *sbi, char *buf)
2452 {
2453         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2454
2455         if (!sb->s_bdev->bd_part)
2456                 return snprintf(buf, PAGE_SIZE, "0\n");
2457         return snprintf(buf, PAGE_SIZE, "%lu\n",
2458                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2459                          sbi->s_sectors_written_start) >> 1);
2460 }
2461
2462 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2463                                           struct ext4_sb_info *sbi, char *buf)
2464 {
2465         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2466
2467         if (!sb->s_bdev->bd_part)
2468                 return snprintf(buf, PAGE_SIZE, "0\n");
2469         return snprintf(buf, PAGE_SIZE, "%llu\n",
2470                         (unsigned long long)(sbi->s_kbytes_written +
2471                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2472                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2473 }
2474
2475 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2476                                           struct ext4_sb_info *sbi,
2477                                           const char *buf, size_t count)
2478 {
2479         unsigned long t;
2480
2481         if (parse_strtoul(buf, 0x40000000, &t))
2482                 return -EINVAL;
2483
2484         if (t && !is_power_of_2(t))
2485                 return -EINVAL;
2486
2487         sbi->s_inode_readahead_blks = t;
2488         return count;
2489 }
2490
2491 static ssize_t sbi_ui_show(struct ext4_attr *a,
2492                            struct ext4_sb_info *sbi, char *buf)
2493 {
2494         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2495
2496         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2497 }
2498
2499 static ssize_t sbi_ui_store(struct ext4_attr *a,
2500                             struct ext4_sb_info *sbi,
2501                             const char *buf, size_t count)
2502 {
2503         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2504         unsigned long t;
2505
2506         if (parse_strtoul(buf, 0xffffffff, &t))
2507                 return -EINVAL;
2508         *ui = t;
2509         return count;
2510 }
2511
2512 static ssize_t trigger_test_error(struct ext4_attr *a,
2513                                   struct ext4_sb_info *sbi,
2514                                   const char *buf, size_t count)
2515 {
2516         int len = count;
2517
2518         if (!capable(CAP_SYS_ADMIN))
2519                 return -EPERM;
2520
2521         if (len && buf[len-1] == '\n')
2522                 len--;
2523
2524         if (len)
2525                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2526         return count;
2527 }
2528
2529 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2530 static struct ext4_attr ext4_attr_##_name = {                   \
2531         .attr = {.name = __stringify(_name), .mode = _mode },   \
2532         .show   = _show,                                        \
2533         .store  = _store,                                       \
2534         .offset = offsetof(struct ext4_sb_info, _elname),       \
2535 }
2536 #define EXT4_ATTR(name, mode, show, store) \
2537 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2538
2539 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2540 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2541 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2542 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2543         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2544 #define ATTR_LIST(name) &ext4_attr_##name.attr
2545
2546 EXT4_RO_ATTR(delayed_allocation_blocks);
2547 EXT4_RO_ATTR(session_write_kbytes);
2548 EXT4_RO_ATTR(lifetime_write_kbytes);
2549 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2550                  inode_readahead_blks_store, s_inode_readahead_blks);
2551 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2552 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2553 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2554 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2555 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2556 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2557 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2558 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2559 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2560 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2561
2562 static struct attribute *ext4_attrs[] = {
2563         ATTR_LIST(delayed_allocation_blocks),
2564         ATTR_LIST(session_write_kbytes),
2565         ATTR_LIST(lifetime_write_kbytes),
2566         ATTR_LIST(inode_readahead_blks),
2567         ATTR_LIST(inode_goal),
2568         ATTR_LIST(mb_stats),
2569         ATTR_LIST(mb_max_to_scan),
2570         ATTR_LIST(mb_min_to_scan),
2571         ATTR_LIST(mb_order2_req),
2572         ATTR_LIST(mb_stream_req),
2573         ATTR_LIST(mb_group_prealloc),
2574         ATTR_LIST(max_writeback_mb_bump),
2575         ATTR_LIST(extent_max_zeroout_kb),
2576         ATTR_LIST(trigger_fs_error),
2577         NULL,
2578 };
2579
2580 /* Features this copy of ext4 supports */
2581 EXT4_INFO_ATTR(lazy_itable_init);
2582 EXT4_INFO_ATTR(batched_discard);
2583 EXT4_INFO_ATTR(meta_bg_resize);
2584
2585 static struct attribute *ext4_feat_attrs[] = {
2586         ATTR_LIST(lazy_itable_init),
2587         ATTR_LIST(batched_discard),
2588         ATTR_LIST(meta_bg_resize),
2589         NULL,
2590 };
2591
2592 static ssize_t ext4_attr_show(struct kobject *kobj,
2593                               struct attribute *attr, char *buf)
2594 {
2595         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2596                                                 s_kobj);
2597         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2598
2599         return a->show ? a->show(a, sbi, buf) : 0;
2600 }
2601
2602 static ssize_t ext4_attr_store(struct kobject *kobj,
2603                                struct attribute *attr,
2604                                const char *buf, size_t len)
2605 {
2606         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2607                                                 s_kobj);
2608         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2609
2610         return a->store ? a->store(a, sbi, buf, len) : 0;
2611 }
2612
2613 static void ext4_sb_release(struct kobject *kobj)
2614 {
2615         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2616                                                 s_kobj);
2617         complete(&sbi->s_kobj_unregister);
2618 }
2619
2620 static const struct sysfs_ops ext4_attr_ops = {
2621         .show   = ext4_attr_show,
2622         .store  = ext4_attr_store,
2623 };
2624
2625 static struct kobj_type ext4_ktype = {
2626         .default_attrs  = ext4_attrs,
2627         .sysfs_ops      = &ext4_attr_ops,
2628         .release        = ext4_sb_release,
2629 };
2630
2631 static void ext4_feat_release(struct kobject *kobj)
2632 {
2633         complete(&ext4_feat->f_kobj_unregister);
2634 }
2635
2636 static struct kobj_type ext4_feat_ktype = {
2637         .default_attrs  = ext4_feat_attrs,
2638         .sysfs_ops      = &ext4_attr_ops,
2639         .release        = ext4_feat_release,
2640 };
2641
2642 /*
2643  * Check whether this filesystem can be mounted based on
2644  * the features present and the RDONLY/RDWR mount requested.
2645  * Returns 1 if this filesystem can be mounted as requested,
2646  * 0 if it cannot be.
2647  */
2648 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2649 {
2650         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2651                 ext4_msg(sb, KERN_ERR,
2652                         "Couldn't mount because of "
2653                         "unsupported optional features (%x)",
2654                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2655                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2656                 return 0;
2657         }
2658
2659         if (readonly)
2660                 return 1;
2661
2662         /* Check that feature set is OK for a read-write mount */
2663         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2664                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2665                          "unsupported optional features (%x)",
2666                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2667                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2668                 return 0;
2669         }
2670         /*
2671          * Large file size enabled file system can only be mounted
2672          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2673          */
2674         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2675                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2676                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2677                                  "cannot be mounted RDWR without "
2678                                  "CONFIG_LBDAF");
2679                         return 0;
2680                 }
2681         }
2682         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2683             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2684                 ext4_msg(sb, KERN_ERR,
2685                          "Can't support bigalloc feature without "
2686                          "extents feature\n");
2687                 return 0;
2688         }
2689
2690 #ifndef CONFIG_QUOTA
2691         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2692             !readonly) {
2693                 ext4_msg(sb, KERN_ERR,
2694                          "Filesystem with quota feature cannot be mounted RDWR "
2695                          "without CONFIG_QUOTA");
2696                 return 0;
2697         }
2698 #endif  /* CONFIG_QUOTA */
2699         return 1;
2700 }
2701
2702 /*
2703  * This function is called once a day if we have errors logged
2704  * on the file system
2705  */
2706 static void print_daily_error_info(unsigned long arg)
2707 {
2708         struct super_block *sb = (struct super_block *) arg;
2709         struct ext4_sb_info *sbi;
2710         struct ext4_super_block *es;
2711
2712         sbi = EXT4_SB(sb);
2713         es = sbi->s_es;
2714
2715         if (es->s_error_count)
2716                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2717                          le32_to_cpu(es->s_error_count));
2718         if (es->s_first_error_time) {
2719                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2720                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2721                        (int) sizeof(es->s_first_error_func),
2722                        es->s_first_error_func,
2723                        le32_to_cpu(es->s_first_error_line));
2724                 if (es->s_first_error_ino)
2725                         printk(": inode %u",
2726                                le32_to_cpu(es->s_first_error_ino));
2727                 if (es->s_first_error_block)
2728                         printk(": block %llu", (unsigned long long)
2729                                le64_to_cpu(es->s_first_error_block));
2730                 printk("\n");
2731         }
2732         if (es->s_last_error_time) {
2733                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2734                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2735                        (int) sizeof(es->s_last_error_func),
2736                        es->s_last_error_func,
2737                        le32_to_cpu(es->s_last_error_line));
2738                 if (es->s_last_error_ino)
2739                         printk(": inode %u",
2740                                le32_to_cpu(es->s_last_error_ino));
2741                 if (es->s_last_error_block)
2742                         printk(": block %llu", (unsigned long long)
2743                                le64_to_cpu(es->s_last_error_block));
2744                 printk("\n");
2745         }
2746         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2747 }
2748
2749 /* Find next suitable group and run ext4_init_inode_table */
2750 static int ext4_run_li_request(struct ext4_li_request *elr)
2751 {
2752         struct ext4_group_desc *gdp = NULL;
2753         ext4_group_t group, ngroups;
2754         struct super_block *sb;
2755         unsigned long timeout = 0;
2756         int ret = 0;
2757
2758         sb = elr->lr_super;
2759         ngroups = EXT4_SB(sb)->s_groups_count;
2760
2761         sb_start_write(sb);
2762         for (group = elr->lr_next_group; group < ngroups; group++) {
2763                 gdp = ext4_get_group_desc(sb, group, NULL);
2764                 if (!gdp) {
2765                         ret = 1;
2766                         break;
2767                 }
2768
2769                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2770                         break;
2771         }
2772
2773         if (group == ngroups)
2774                 ret = 1;
2775
2776         if (!ret) {
2777                 timeout = jiffies;
2778                 ret = ext4_init_inode_table(sb, group,
2779                                             elr->lr_timeout ? 0 : 1);
2780                 if (elr->lr_timeout == 0) {
2781                         timeout = (jiffies - timeout) *
2782                                   elr->lr_sbi->s_li_wait_mult;
2783                         elr->lr_timeout = timeout;
2784                 }
2785                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2786                 elr->lr_next_group = group + 1;
2787         }
2788         sb_end_write(sb);
2789
2790         return ret;
2791 }
2792
2793 /*
2794  * Remove lr_request from the list_request and free the
2795  * request structure. Should be called with li_list_mtx held
2796  */
2797 static void ext4_remove_li_request(struct ext4_li_request *elr)
2798 {
2799         struct ext4_sb_info *sbi;
2800
2801         if (!elr)
2802                 return;
2803
2804         sbi = elr->lr_sbi;
2805
2806         list_del(&elr->lr_request);
2807         sbi->s_li_request = NULL;
2808         kfree(elr);
2809 }
2810
2811 static void ext4_unregister_li_request(struct super_block *sb)
2812 {
2813         mutex_lock(&ext4_li_mtx);
2814         if (!ext4_li_info) {
2815                 mutex_unlock(&ext4_li_mtx);
2816                 return;
2817         }
2818
2819         mutex_lock(&ext4_li_info->li_list_mtx);
2820         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2821         mutex_unlock(&ext4_li_info->li_list_mtx);
2822         mutex_unlock(&ext4_li_mtx);
2823 }
2824
2825 static struct task_struct *ext4_lazyinit_task;
2826
2827 /*
2828  * This is the function where ext4lazyinit thread lives. It walks
2829  * through the request list searching for next scheduled filesystem.
2830  * When such a fs is found, run the lazy initialization request
2831  * (ext4_rn_li_request) and keep track of the time spend in this
2832  * function. Based on that time we compute next schedule time of
2833  * the request. When walking through the list is complete, compute
2834  * next waking time and put itself into sleep.
2835  */
2836 static int ext4_lazyinit_thread(void *arg)
2837 {
2838         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2839         struct list_head *pos, *n;
2840         struct ext4_li_request *elr;
2841         unsigned long next_wakeup, cur;
2842
2843         BUG_ON(NULL == eli);
2844
2845 cont_thread:
2846         while (true) {
2847                 next_wakeup = MAX_JIFFY_OFFSET;
2848
2849                 mutex_lock(&eli->li_list_mtx);
2850                 if (list_empty(&eli->li_request_list)) {
2851                         mutex_unlock(&eli->li_list_mtx);
2852                         goto exit_thread;
2853                 }
2854
2855                 list_for_each_safe(pos, n, &eli->li_request_list) {
2856                         elr = list_entry(pos, struct ext4_li_request,
2857                                          lr_request);
2858
2859                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2860                                 if (ext4_run_li_request(elr) != 0) {
2861                                         /* error, remove the lazy_init job */
2862                                         ext4_remove_li_request(elr);
2863                                         continue;
2864                                 }
2865                         }
2866
2867                         if (time_before(elr->lr_next_sched, next_wakeup))
2868                                 next_wakeup = elr->lr_next_sched;
2869                 }
2870                 mutex_unlock(&eli->li_list_mtx);
2871
2872                 try_to_freeze();
2873
2874                 cur = jiffies;
2875                 if ((time_after_eq(cur, next_wakeup)) ||
2876                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2877                         cond_resched();
2878                         continue;
2879                 }
2880
2881                 schedule_timeout_interruptible(next_wakeup - cur);
2882
2883                 if (kthread_should_stop()) {
2884                         ext4_clear_request_list();
2885                         goto exit_thread;
2886                 }
2887         }
2888
2889 exit_thread:
2890         /*
2891          * It looks like the request list is empty, but we need
2892          * to check it under the li_list_mtx lock, to prevent any
2893          * additions into it, and of course we should lock ext4_li_mtx
2894          * to atomically free the list and ext4_li_info, because at
2895          * this point another ext4 filesystem could be registering
2896          * new one.
2897          */
2898         mutex_lock(&ext4_li_mtx);
2899         mutex_lock(&eli->li_list_mtx);
2900         if (!list_empty(&eli->li_request_list)) {
2901                 mutex_unlock(&eli->li_list_mtx);
2902                 mutex_unlock(&ext4_li_mtx);
2903                 goto cont_thread;
2904         }
2905         mutex_unlock(&eli->li_list_mtx);
2906         kfree(ext4_li_info);
2907         ext4_li_info = NULL;
2908         mutex_unlock(&ext4_li_mtx);
2909
2910         return 0;
2911 }
2912
2913 static void ext4_clear_request_list(void)
2914 {
2915         struct list_head *pos, *n;
2916         struct ext4_li_request *elr;
2917
2918         mutex_lock(&ext4_li_info->li_list_mtx);
2919         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2920                 elr = list_entry(pos, struct ext4_li_request,
2921                                  lr_request);
2922                 ext4_remove_li_request(elr);
2923         }
2924         mutex_unlock(&ext4_li_info->li_list_mtx);
2925 }
2926
2927 static int ext4_run_lazyinit_thread(void)
2928 {
2929         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2930                                          ext4_li_info, "ext4lazyinit");
2931         if (IS_ERR(ext4_lazyinit_task)) {
2932                 int err = PTR_ERR(ext4_lazyinit_task);
2933                 ext4_clear_request_list();
2934                 kfree(ext4_li_info);
2935                 ext4_li_info = NULL;
2936                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2937                                  "initialization thread\n",
2938                                  err);
2939                 return err;
2940         }
2941         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2942         return 0;
2943 }
2944
2945 /*
2946  * Check whether it make sense to run itable init. thread or not.
2947  * If there is at least one uninitialized inode table, return
2948  * corresponding group number, else the loop goes through all
2949  * groups and return total number of groups.
2950  */
2951 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2952 {
2953         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2954         struct ext4_group_desc *gdp = NULL;
2955
2956         for (group = 0; group < ngroups; group++) {
2957                 gdp = ext4_get_group_desc(sb, group, NULL);
2958                 if (!gdp)
2959                         continue;
2960
2961                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2962                         break;
2963         }
2964
2965         return group;
2966 }
2967
2968 static int ext4_li_info_new(void)
2969 {
2970         struct ext4_lazy_init *eli = NULL;
2971
2972         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2973         if (!eli)
2974                 return -ENOMEM;
2975
2976         INIT_LIST_HEAD(&eli->li_request_list);
2977         mutex_init(&eli->li_list_mtx);
2978
2979         eli->li_state |= EXT4_LAZYINIT_QUIT;
2980
2981         ext4_li_info = eli;
2982
2983         return 0;
2984 }
2985
2986 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2987                                             ext4_group_t start)
2988 {
2989         struct ext4_sb_info *sbi = EXT4_SB(sb);
2990         struct ext4_li_request *elr;
2991         unsigned long rnd;
2992
2993         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2994         if (!elr)
2995                 return NULL;
2996
2997         elr->lr_super = sb;
2998         elr->lr_sbi = sbi;
2999         elr->lr_next_group = start;
3000
3001         /*
3002          * Randomize first schedule time of the request to
3003          * spread the inode table initialization requests
3004          * better.
3005          */
3006         get_random_bytes(&rnd, sizeof(rnd));
3007         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3008                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3009
3010         return elr;
3011 }
3012
3013 static int ext4_register_li_request(struct super_block *sb,
3014                                     ext4_group_t first_not_zeroed)
3015 {
3016         struct ext4_sb_info *sbi = EXT4_SB(sb);
3017         struct ext4_li_request *elr;
3018         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3019         int ret = 0;
3020
3021         if (sbi->s_li_request != NULL) {
3022                 /*
3023                  * Reset timeout so it can be computed again, because
3024                  * s_li_wait_mult might have changed.
3025                  */
3026                 sbi->s_li_request->lr_timeout = 0;
3027                 return 0;
3028         }
3029
3030         if (first_not_zeroed == ngroups ||
3031             (sb->s_flags & MS_RDONLY) ||
3032             !test_opt(sb, INIT_INODE_TABLE))
3033                 return 0;
3034
3035         elr = ext4_li_request_new(sb, first_not_zeroed);
3036         if (!elr)
3037                 return -ENOMEM;
3038
3039         mutex_lock(&ext4_li_mtx);
3040
3041         if (NULL == ext4_li_info) {
3042                 ret = ext4_li_info_new();
3043                 if (ret)
3044                         goto out;
3045         }
3046
3047         mutex_lock(&ext4_li_info->li_list_mtx);
3048         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3049         mutex_unlock(&ext4_li_info->li_list_mtx);
3050
3051         sbi->s_li_request = elr;
3052         /*
3053          * set elr to NULL here since it has been inserted to
3054          * the request_list and the removal and free of it is
3055          * handled by ext4_clear_request_list from now on.
3056          */
3057         elr = NULL;
3058
3059         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3060                 ret = ext4_run_lazyinit_thread();
3061                 if (ret)
3062                         goto out;
3063         }
3064 out:
3065         mutex_unlock(&ext4_li_mtx);
3066         if (ret)
3067                 kfree(elr);
3068         return ret;
3069 }
3070
3071 /*
3072  * We do not need to lock anything since this is called on
3073  * module unload.
3074  */
3075 static void ext4_destroy_lazyinit_thread(void)
3076 {
3077         /*
3078          * If thread exited earlier
3079          * there's nothing to be done.
3080          */
3081         if (!ext4_li_info || !ext4_lazyinit_task)
3082                 return;
3083
3084         kthread_stop(ext4_lazyinit_task);
3085 }
3086
3087 static int set_journal_csum_feature_set(struct super_block *sb)
3088 {
3089         int ret = 1;
3090         int compat, incompat;
3091         struct ext4_sb_info *sbi = EXT4_SB(sb);
3092
3093         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3094                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3095                 /* journal checksum v2 */
3096                 compat = 0;
3097                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3098         } else {
3099                 /* journal checksum v1 */
3100                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3101                 incompat = 0;
3102         }
3103
3104         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3105                 ret = jbd2_journal_set_features(sbi->s_journal,
3106                                 compat, 0,
3107                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3108                                 incompat);
3109         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3110                 ret = jbd2_journal_set_features(sbi->s_journal,
3111                                 compat, 0,
3112                                 incompat);
3113                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3114                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3115         } else {
3116                 jbd2_journal_clear_features(sbi->s_journal,
3117                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3118                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3119                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3120         }
3121
3122         return ret;
3123 }
3124
3125 /*
3126  * Note: calculating the overhead so we can be compatible with
3127  * historical BSD practice is quite difficult in the face of
3128  * clusters/bigalloc.  This is because multiple metadata blocks from
3129  * different block group can end up in the same allocation cluster.
3130  * Calculating the exact overhead in the face of clustered allocation
3131  * requires either O(all block bitmaps) in memory or O(number of block
3132  * groups**2) in time.  We will still calculate the superblock for
3133  * older file systems --- and if we come across with a bigalloc file
3134  * system with zero in s_overhead_clusters the estimate will be close to
3135  * correct especially for very large cluster sizes --- but for newer
3136  * file systems, it's better to calculate this figure once at mkfs
3137  * time, and store it in the superblock.  If the superblock value is
3138  * present (even for non-bigalloc file systems), we will use it.
3139  */
3140 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3141                           char *buf)
3142 {
3143         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3144         struct ext4_group_desc  *gdp;
3145         ext4_fsblk_t            first_block, last_block, b;
3146         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3147         int                     s, j, count = 0;
3148
3149         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3150                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3151                         sbi->s_itb_per_group + 2);
3152
3153         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3154                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3155         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3156         for (i = 0; i < ngroups; i++) {
3157                 gdp = ext4_get_group_desc(sb, i, NULL);
3158                 b = ext4_block_bitmap(sb, gdp);
3159                 if (b >= first_block && b <= last_block) {
3160                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3161                         count++;
3162                 }
3163                 b = ext4_inode_bitmap(sb, gdp);
3164                 if (b >= first_block && b <= last_block) {
3165                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3166                         count++;
3167                 }
3168                 b = ext4_inode_table(sb, gdp);
3169                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3170                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3171                                 int c = EXT4_B2C(sbi, b - first_block);
3172                                 ext4_set_bit(c, buf);
3173                                 count++;
3174                         }
3175                 if (i != grp)
3176                         continue;
3177                 s = 0;
3178                 if (ext4_bg_has_super(sb, grp)) {
3179                         ext4_set_bit(s++, buf);
3180                         count++;
3181                 }
3182                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3183                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3184                         count++;
3185                 }
3186         }
3187         if (!count)
3188                 return 0;
3189         return EXT4_CLUSTERS_PER_GROUP(sb) -
3190                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3191 }
3192
3193 /*
3194  * Compute the overhead and stash it in sbi->s_overhead
3195  */
3196 int ext4_calculate_overhead(struct super_block *sb)
3197 {
3198         struct ext4_sb_info *sbi = EXT4_SB(sb);
3199         struct ext4_super_block *es = sbi->s_es;
3200         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3201         ext4_fsblk_t overhead = 0;
3202         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3203
3204         memset(buf, 0, PAGE_SIZE);
3205         if (!buf)
3206                 return -ENOMEM;
3207
3208         /*
3209          * Compute the overhead (FS structures).  This is constant
3210          * for a given filesystem unless the number of block groups
3211          * changes so we cache the previous value until it does.
3212          */
3213
3214         /*
3215          * All of the blocks before first_data_block are overhead
3216          */
3217         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3218
3219         /*
3220          * Add the overhead found in each block group
3221          */
3222         for (i = 0; i < ngroups; i++) {
3223                 int blks;
3224
3225                 blks = count_overhead(sb, i, buf);
3226                 overhead += blks;
3227                 if (blks)
3228                         memset(buf, 0, PAGE_SIZE);
3229                 cond_resched();
3230         }
3231         sbi->s_overhead = overhead;
3232         smp_wmb();
3233         free_page((unsigned long) buf);
3234         return 0;
3235 }
3236
3237 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3238 {
3239         char *orig_data = kstrdup(data, GFP_KERNEL);
3240         struct buffer_head *bh;
3241         struct ext4_super_block *es = NULL;
3242         struct ext4_sb_info *sbi;
3243         ext4_fsblk_t block;
3244         ext4_fsblk_t sb_block = get_sb_block(&data);
3245         ext4_fsblk_t logical_sb_block;
3246         unsigned long offset = 0;
3247         unsigned long journal_devnum = 0;
3248         unsigned long def_mount_opts;
3249         struct inode *root;
3250         char *cp;
3251         const char *descr;
3252         int ret = -ENOMEM;
3253         int blocksize, clustersize;
3254         unsigned int db_count;
3255         unsigned int i;
3256         int needs_recovery, has_huge_files, has_bigalloc;
3257         __u64 blocks_count;
3258         int err;
3259         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3260         ext4_group_t first_not_zeroed;
3261
3262         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3263         if (!sbi)
3264                 goto out_free_orig;
3265
3266         sbi->s_blockgroup_lock =
3267                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3268         if (!sbi->s_blockgroup_lock) {
3269                 kfree(sbi);
3270                 goto out_free_orig;
3271         }
3272         sb->s_fs_info = sbi;
3273         sbi->s_sb = sb;
3274         sbi->s_mount_opt = 0;
3275         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3276         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3277         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3278         sbi->s_sb_block = sb_block;
3279         if (sb->s_bdev->bd_part)
3280                 sbi->s_sectors_written_start =
3281                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3282
3283         /* Cleanup superblock name */
3284         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3285                 *cp = '!';
3286
3287         ret = -EINVAL;
3288         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3289         if (!blocksize) {
3290                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3291                 goto out_fail;
3292         }
3293
3294         /*
3295          * The ext4 superblock will not be buffer aligned for other than 1kB
3296          * block sizes.  We need to calculate the offset from buffer start.
3297          */
3298         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3299                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3300                 offset = do_div(logical_sb_block, blocksize);
3301         } else {
3302                 logical_sb_block = sb_block;
3303         }
3304
3305         if (!(bh = sb_bread(sb, logical_sb_block))) {
3306                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3307                 goto out_fail;
3308         }
3309         /*
3310          * Note: s_es must be initialized as soon as possible because
3311          *       some ext4 macro-instructions depend on its value
3312          */
3313         es = (struct ext4_super_block *) (bh->b_data + offset);
3314         sbi->s_es = es;
3315         sb->s_magic = le16_to_cpu(es->s_magic);
3316         if (sb->s_magic != EXT4_SUPER_MAGIC)
3317                 goto cantfind_ext4;
3318         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3319
3320         /* Warn if metadata_csum and gdt_csum are both set. */
3321         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3322                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3323             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3324                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3325                              "redundant flags; please run fsck.");
3326
3327         /* Check for a known checksum algorithm */
3328         if (!ext4_verify_csum_type(sb, es)) {
3329                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3330                          "unknown checksum algorithm.");
3331                 silent = 1;
3332                 goto cantfind_ext4;
3333         }
3334
3335         /* Load the checksum driver */
3336         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3337                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3338                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3339                 if (IS_ERR(sbi->s_chksum_driver)) {
3340                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3341                         ret = PTR_ERR(sbi->s_chksum_driver);
3342                         sbi->s_chksum_driver = NULL;
3343                         goto failed_mount;
3344                 }
3345         }
3346
3347         /* Check superblock checksum */
3348         if (!ext4_superblock_csum_verify(sb, es)) {
3349                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3350                          "invalid superblock checksum.  Run e2fsck?");
3351                 silent = 1;
3352                 goto cantfind_ext4;
3353         }
3354
3355         /* Precompute checksum seed for all metadata */
3356         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3357                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3358                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3359                                                sizeof(es->s_uuid));
3360
3361         /* Set defaults before we parse the mount options */
3362         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3363         set_opt(sb, INIT_INODE_TABLE);
3364         if (def_mount_opts & EXT4_DEFM_DEBUG)
3365                 set_opt(sb, DEBUG);
3366         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3367                 set_opt(sb, GRPID);
3368         if (def_mount_opts & EXT4_DEFM_UID16)
3369                 set_opt(sb, NO_UID32);
3370         /* xattr user namespace & acls are now defaulted on */
3371 #ifdef CONFIG_EXT4_FS_XATTR
3372         set_opt(sb, XATTR_USER);
3373 #endif
3374 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3375         set_opt(sb, POSIX_ACL);
3376 #endif
3377         set_opt(sb, MBLK_IO_SUBMIT);
3378         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3379                 set_opt(sb, JOURNAL_DATA);
3380         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3381                 set_opt(sb, ORDERED_DATA);
3382         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3383                 set_opt(sb, WRITEBACK_DATA);
3384
3385         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3386                 set_opt(sb, ERRORS_PANIC);
3387         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3388                 set_opt(sb, ERRORS_CONT);
3389         else
3390                 set_opt(sb, ERRORS_RO);
3391         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3392                 set_opt(sb, BLOCK_VALIDITY);
3393         if (def_mount_opts & EXT4_DEFM_DISCARD)
3394                 set_opt(sb, DISCARD);
3395
3396         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3397         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3398         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3399         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3400         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3401
3402         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3403                 set_opt(sb, BARRIER);
3404
3405         /*
3406          * enable delayed allocation by default
3407          * Use -o nodelalloc to turn it off
3408          */
3409         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3410             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3411                 set_opt(sb, DELALLOC);
3412
3413         /*
3414          * set default s_li_wait_mult for lazyinit, for the case there is
3415          * no mount option specified.
3416          */
3417         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3418
3419         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3420                            &journal_devnum, &journal_ioprio, 0)) {
3421                 ext4_msg(sb, KERN_WARNING,
3422                          "failed to parse options in superblock: %s",
3423                          sbi->s_es->s_mount_opts);
3424         }
3425         sbi->s_def_mount_opt = sbi->s_mount_opt;
3426         if (!parse_options((char *) data, sb, &journal_devnum,
3427                            &journal_ioprio, 0))
3428                 goto failed_mount;
3429
3430         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3431                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3432                             "with data=journal disables delayed "
3433                             "allocation and O_DIRECT support!\n");
3434                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3435                         ext4_msg(sb, KERN_ERR, "can't mount with "
3436                                  "both data=journal and delalloc");
3437                         goto failed_mount;
3438                 }
3439                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3440                         ext4_msg(sb, KERN_ERR, "can't mount with "
3441                                  "both data=journal and delalloc");
3442                         goto failed_mount;
3443                 }
3444                 if (test_opt(sb, DELALLOC))
3445                         clear_opt(sb, DELALLOC);
3446         }
3447
3448         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3449         if (test_opt(sb, DIOREAD_NOLOCK)) {
3450                 if (blocksize < PAGE_SIZE) {
3451                         ext4_msg(sb, KERN_ERR, "can't mount with "
3452                                  "dioread_nolock if block size != PAGE_SIZE");
3453                         goto failed_mount;
3454                 }
3455         }
3456
3457         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3458                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3459
3460         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3461             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3462              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3463              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3464                 ext4_msg(sb, KERN_WARNING,
3465                        "feature flags set on rev 0 fs, "
3466                        "running e2fsck is recommended");
3467
3468         if (IS_EXT2_SB(sb)) {
3469                 if (ext2_feature_set_ok(sb))
3470                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3471                                  "using the ext4 subsystem");
3472                 else {
3473                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3474                                  "to feature incompatibilities");
3475                         goto failed_mount;
3476                 }
3477         }
3478
3479         if (IS_EXT3_SB(sb)) {
3480                 if (ext3_feature_set_ok(sb))
3481                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3482                                  "using the ext4 subsystem");
3483                 else {
3484                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3485                                  "to feature incompatibilities");
3486                         goto failed_mount;
3487                 }
3488         }
3489
3490         /*
3491          * Check feature flags regardless of the revision level, since we
3492          * previously didn't change the revision level when setting the flags,
3493          * so there is a chance incompat flags are set on a rev 0 filesystem.
3494          */
3495         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3496                 goto failed_mount;
3497
3498         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3499             blocksize > EXT4_MAX_BLOCK_SIZE) {
3500                 ext4_msg(sb, KERN_ERR,
3501                        "Unsupported filesystem blocksize %d", blocksize);
3502                 goto failed_mount;
3503         }
3504
3505         if (sb->s_blocksize != blocksize) {
3506                 /* Validate the filesystem blocksize */
3507                 if (!sb_set_blocksize(sb, blocksize)) {
3508                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3509                                         blocksize);
3510                         goto failed_mount;
3511                 }
3512
3513                 brelse(bh);
3514                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3515                 offset = do_div(logical_sb_block, blocksize);
3516                 bh = sb_bread(sb, logical_sb_block);
3517                 if (!bh) {
3518                         ext4_msg(sb, KERN_ERR,
3519                                "Can't read superblock on 2nd try");
3520                         goto failed_mount;
3521                 }
3522                 es = (struct ext4_super_block *)(bh->b_data + offset);
3523                 sbi->s_es = es;
3524                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "Magic mismatch, very weird!");
3527                         goto failed_mount;
3528                 }
3529         }
3530
3531         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3532                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3533         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3534                                                       has_huge_files);
3535         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3536
3537         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3538                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3539                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3540         } else {
3541                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3542                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3543                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3544                     (!is_power_of_2(sbi->s_inode_size)) ||
3545                     (sbi->s_inode_size > blocksize)) {
3546                         ext4_msg(sb, KERN_ERR,
3547                                "unsupported inode size: %d",
3548                                sbi->s_inode_size);
3549                         goto failed_mount;
3550                 }
3551                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3552                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3553         }
3554
3555         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3556         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3557                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3558                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3559                     !is_power_of_2(sbi->s_desc_size)) {
3560                         ext4_msg(sb, KERN_ERR,
3561                                "unsupported descriptor size %lu",
3562                                sbi->s_desc_size);
3563                         goto failed_mount;
3564                 }
3565         } else
3566                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3567
3568         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3569         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3570         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3571                 goto cantfind_ext4;
3572
3573         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3574         if (sbi->s_inodes_per_block == 0)
3575                 goto cantfind_ext4;
3576         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3577                                         sbi->s_inodes_per_block;
3578         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3579         sbi->s_sbh = bh;
3580         sbi->s_mount_state = le16_to_cpu(es->s_state);
3581         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3582         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3583
3584         for (i = 0; i < 4; i++)
3585                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3586         sbi->s_def_hash_version = es->s_def_hash_version;
3587         i = le32_to_cpu(es->s_flags);
3588         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3589                 sbi->s_hash_unsigned = 3;
3590         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3591 #ifdef __CHAR_UNSIGNED__
3592                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3593                 sbi->s_hash_unsigned = 3;
3594 #else
3595                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3596 #endif
3597         }
3598
3599         /* Handle clustersize */
3600         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3601         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3602                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3603         if (has_bigalloc) {
3604                 if (clustersize < blocksize) {
3605                         ext4_msg(sb, KERN_ERR,
3606                                  "cluster size (%d) smaller than "
3607                                  "block size (%d)", clustersize, blocksize);
3608                         goto failed_mount;
3609                 }
3610                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3611                         le32_to_cpu(es->s_log_block_size);
3612                 sbi->s_clusters_per_group =
3613                         le32_to_cpu(es->s_clusters_per_group);
3614                 if (sbi->s_clusters_per_group > blocksize * 8) {
3615                         ext4_msg(sb, KERN_ERR,
3616                                  "#clusters per group too big: %lu",
3617                                  sbi->s_clusters_per_group);
3618                         goto failed_mount;
3619                 }
3620                 if (sbi->s_blocks_per_group !=
3621                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3622                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3623                                  "clusters per group (%lu) inconsistent",
3624                                  sbi->s_blocks_per_group,
3625                                  sbi->s_clusters_per_group);
3626                         goto failed_mount;
3627                 }
3628         } else {
3629                 if (clustersize != blocksize) {
3630                         ext4_warning(sb, "fragment/cluster size (%d) != "
3631                                      "block size (%d)", clustersize,
3632                                      blocksize);
3633                         clustersize = blocksize;
3634                 }
3635                 if (sbi->s_blocks_per_group > blocksize * 8) {
3636                         ext4_msg(sb, KERN_ERR,
3637                                  "#blocks per group too big: %lu",
3638                                  sbi->s_blocks_per_group);
3639                         goto failed_mount;
3640                 }
3641                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3642                 sbi->s_cluster_bits = 0;
3643         }
3644         sbi->s_cluster_ratio = clustersize / blocksize;
3645
3646         if (sbi->s_inodes_per_group > blocksize * 8) {
3647                 ext4_msg(sb, KERN_ERR,
3648                        "#inodes per group too big: %lu",
3649                        sbi->s_inodes_per_group);
3650                 goto failed_mount;
3651         }
3652
3653         /*
3654          * Test whether we have more sectors than will fit in sector_t,
3655          * and whether the max offset is addressable by the page cache.
3656          */
3657         err = generic_check_addressable(sb->s_blocksize_bits,
3658                                         ext4_blocks_count(es));
3659         if (err) {
3660                 ext4_msg(sb, KERN_ERR, "filesystem"
3661                          " too large to mount safely on this system");
3662                 if (sizeof(sector_t) < 8)
3663                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3664                 ret = err;
3665                 goto failed_mount;
3666         }
3667
3668         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3669                 goto cantfind_ext4;
3670
3671         /* check blocks count against device size */
3672         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3673         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3674                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3675                        "exceeds size of device (%llu blocks)",
3676                        ext4_blocks_count(es), blocks_count);
3677                 goto failed_mount;
3678         }
3679
3680         /*
3681          * It makes no sense for the first data block to be beyond the end
3682          * of the filesystem.
3683          */
3684         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3685                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3686                          "block %u is beyond end of filesystem (%llu)",
3687                          le32_to_cpu(es->s_first_data_block),
3688                          ext4_blocks_count(es));
3689                 goto failed_mount;
3690         }
3691         blocks_count = (ext4_blocks_count(es) -
3692                         le32_to_cpu(es->s_first_data_block) +
3693                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3694         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3695         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3696                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3697                        "(block count %llu, first data block %u, "
3698                        "blocks per group %lu)", sbi->s_groups_count,
3699                        ext4_blocks_count(es),
3700                        le32_to_cpu(es->s_first_data_block),
3701                        EXT4_BLOCKS_PER_GROUP(sb));
3702                 goto failed_mount;
3703         }
3704         sbi->s_groups_count = blocks_count;
3705         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3706                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3707         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3708                    EXT4_DESC_PER_BLOCK(sb);
3709         sbi->s_group_desc = ext4_kvmalloc(db_count *
3710                                           sizeof(struct buffer_head *),
3711                                           GFP_KERNEL);
3712         if (sbi->s_group_desc == NULL) {
3713                 ext4_msg(sb, KERN_ERR, "not enough memory");
3714                 ret = -ENOMEM;
3715                 goto failed_mount;
3716         }
3717
3718         if (ext4_proc_root)
3719                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3720
3721         if (sbi->s_proc)
3722                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3723                                  &ext4_seq_options_fops, sb);
3724
3725         bgl_lock_init(sbi->s_blockgroup_lock);
3726
3727         for (i = 0; i < db_count; i++) {
3728                 block = descriptor_loc(sb, logical_sb_block, i);
3729                 sbi->s_group_desc[i] = sb_bread(sb, block);
3730                 if (!sbi->s_group_desc[i]) {
3731                         ext4_msg(sb, KERN_ERR,
3732                                "can't read group descriptor %d", i);
3733                         db_count = i;
3734                         goto failed_mount2;
3735                 }
3736         }
3737         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3738                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3739                 goto failed_mount2;
3740         }
3741         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3742                 if (!ext4_fill_flex_info(sb)) {
3743                         ext4_msg(sb, KERN_ERR,
3744                                "unable to initialize "
3745                                "flex_bg meta info!");
3746                         goto failed_mount2;
3747                 }
3748
3749         sbi->s_gdb_count = db_count;
3750         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3751         spin_lock_init(&sbi->s_next_gen_lock);
3752
3753         init_timer(&sbi->s_err_report);
3754         sbi->s_err_report.function = print_daily_error_info;
3755         sbi->s_err_report.data = (unsigned long) sb;
3756
3757         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3758                         ext4_count_free_clusters(sb));
3759         if (!err) {
3760                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3761                                 ext4_count_free_inodes(sb));
3762         }
3763         if (!err) {
3764                 err = percpu_counter_init(&sbi->s_dirs_counter,
3765                                 ext4_count_dirs(sb));
3766         }
3767         if (!err) {
3768                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3769         }
3770         if (err) {
3771                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3772                 ret = err;
3773                 goto failed_mount3;
3774         }
3775
3776         sbi->s_stripe = ext4_get_stripe_size(sbi);
3777         sbi->s_max_writeback_mb_bump = 128;
3778         sbi->s_extent_max_zeroout_kb = 32;
3779
3780         /*
3781          * set up enough so that it can read an inode
3782          */
3783         if (!test_opt(sb, NOLOAD) &&
3784             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3785                 sb->s_op = &ext4_sops;
3786         else
3787                 sb->s_op = &ext4_nojournal_sops;
3788         sb->s_export_op = &ext4_export_ops;
3789         sb->s_xattr = ext4_xattr_handlers;
3790 #ifdef CONFIG_QUOTA
3791         sb->s_qcop = &ext4_qctl_operations;
3792         sb->dq_op = &ext4_quota_operations;
3793
3794         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3795                 /* Use qctl operations for hidden quota files. */
3796                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3797         }
3798 #endif
3799         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3800
3801         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3802         mutex_init(&sbi->s_orphan_lock);
3803         sbi->s_resize_flags = 0;
3804
3805         sb->s_root = NULL;
3806
3807         needs_recovery = (es->s_last_orphan != 0 ||
3808                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3809                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3810
3811         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3812             !(sb->s_flags & MS_RDONLY))
3813                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3814                         goto failed_mount3;
3815
3816         /*
3817          * The first inode we look at is the journal inode.  Don't try
3818          * root first: it may be modified in the journal!
3819          */
3820         if (!test_opt(sb, NOLOAD) &&
3821             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3822                 if (ext4_load_journal(sb, es, journal_devnum))
3823                         goto failed_mount3;
3824         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3825               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3826                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3827                        "suppressed and not mounted read-only");
3828                 goto failed_mount_wq;
3829         } else {
3830                 clear_opt(sb, DATA_FLAGS);
3831                 sbi->s_journal = NULL;
3832                 needs_recovery = 0;
3833                 goto no_journal;
3834         }
3835
3836         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3837             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3838                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3839                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3840                 goto failed_mount_wq;
3841         }
3842
3843         if (!set_journal_csum_feature_set(sb)) {
3844                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3845                          "feature set");
3846                 goto failed_mount_wq;
3847         }
3848
3849         /* We have now updated the journal if required, so we can
3850          * validate the data journaling mode. */
3851         switch (test_opt(sb, DATA_FLAGS)) {
3852         case 0:
3853                 /* No mode set, assume a default based on the journal
3854                  * capabilities: ORDERED_DATA if the journal can
3855                  * cope, else JOURNAL_DATA
3856                  */
3857                 if (jbd2_journal_check_available_features
3858                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3859                         set_opt(sb, ORDERED_DATA);
3860                 else
3861                         set_opt(sb, JOURNAL_DATA);
3862                 break;
3863
3864         case EXT4_MOUNT_ORDERED_DATA:
3865         case EXT4_MOUNT_WRITEBACK_DATA:
3866                 if (!jbd2_journal_check_available_features
3867                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3868                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3869                                "requested data journaling mode");
3870                         goto failed_mount_wq;
3871                 }
3872         default:
3873                 break;
3874         }
3875         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3876
3877         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3878
3879         /*
3880          * The journal may have updated the bg summary counts, so we
3881          * need to update the global counters.
3882          */
3883         percpu_counter_set(&sbi->s_freeclusters_counter,
3884                            ext4_count_free_clusters(sb));
3885         percpu_counter_set(&sbi->s_freeinodes_counter,
3886                            ext4_count_free_inodes(sb));
3887         percpu_counter_set(&sbi->s_dirs_counter,
3888                            ext4_count_dirs(sb));
3889         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3890
3891 no_journal:
3892         /*
3893          * Get the # of file system overhead blocks from the
3894          * superblock if present.
3895          */
3896         if (es->s_overhead_clusters)
3897                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3898         else {
3899                 ret = ext4_calculate_overhead(sb);
3900                 if (ret)
3901                         goto failed_mount_wq;
3902         }
3903
3904         /*
3905          * The maximum number of concurrent works can be high and
3906          * concurrency isn't really necessary.  Limit it to 1.
3907          */
3908         EXT4_SB(sb)->dio_unwritten_wq =
3909                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3910         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3911                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3912                 goto failed_mount_wq;
3913         }
3914
3915         /*
3916          * The jbd2_journal_load will have done any necessary log recovery,
3917          * so we can safely mount the rest of the filesystem now.
3918          */
3919
3920         root = ext4_iget(sb, EXT4_ROOT_INO);
3921         if (IS_ERR(root)) {
3922                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3923                 ret = PTR_ERR(root);
3924                 root = NULL;
3925                 goto failed_mount4;
3926         }
3927         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3928                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3929                 iput(root);
3930                 goto failed_mount4;
3931         }
3932         sb->s_root = d_make_root(root);
3933         if (!sb->s_root) {
3934                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3935                 ret = -ENOMEM;
3936                 goto failed_mount4;
3937         }
3938
3939         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3940                 sb->s_flags |= MS_RDONLY;
3941
3942         /* determine the minimum size of new large inodes, if present */
3943         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3944                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3945                                                      EXT4_GOOD_OLD_INODE_SIZE;
3946                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3947                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3948                         if (sbi->s_want_extra_isize <
3949                             le16_to_cpu(es->s_want_extra_isize))
3950                                 sbi->s_want_extra_isize =
3951                                         le16_to_cpu(es->s_want_extra_isize);
3952                         if (sbi->s_want_extra_isize <
3953                             le16_to_cpu(es->s_min_extra_isize))
3954                                 sbi->s_want_extra_isize =
3955                                         le16_to_cpu(es->s_min_extra_isize);
3956                 }
3957         }
3958         /* Check if enough inode space is available */
3959         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3960                                                         sbi->s_inode_size) {
3961                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3962                                                        EXT4_GOOD_OLD_INODE_SIZE;
3963                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3964                          "available");
3965         }
3966
3967         err = ext4_setup_system_zone(sb);
3968         if (err) {
3969                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3970                          "zone (%d)", err);
3971                 goto failed_mount4a;
3972         }
3973
3974         ext4_ext_init(sb);
3975         err = ext4_mb_init(sb);
3976         if (err) {
3977                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3978                          err);
3979                 goto failed_mount5;
3980         }
3981
3982         err = ext4_register_li_request(sb, first_not_zeroed);
3983         if (err)
3984                 goto failed_mount6;
3985
3986         sbi->s_kobj.kset = ext4_kset;
3987         init_completion(&sbi->s_kobj_unregister);
3988         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3989                                    "%s", sb->s_id);
3990         if (err)
3991                 goto failed_mount7;
3992
3993         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3994         ext4_orphan_cleanup(sb, es);
3995         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3996         if (needs_recovery) {
3997                 ext4_msg(sb, KERN_INFO, "recovery complete");
3998                 ext4_mark_recovery_complete(sb, es);
3999         }
4000         if (EXT4_SB(sb)->s_journal) {
4001                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4002                         descr = " journalled data mode";
4003                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4004                         descr = " ordered data mode";
4005                 else
4006                         descr = " writeback data mode";
4007         } else
4008                 descr = "out journal";
4009
4010 #ifdef CONFIG_QUOTA
4011         /* Enable quota usage during mount. */
4012         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4013             !(sb->s_flags & MS_RDONLY)) {
4014                 ret = ext4_enable_quotas(sb);
4015                 if (ret)
4016                         goto failed_mount7;
4017         }
4018 #endif  /* CONFIG_QUOTA */
4019
4020         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4021                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4022                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4023
4024         if (es->s_error_count)
4025                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4026
4027         kfree(orig_data);
4028         return 0;
4029
4030 cantfind_ext4:
4031         if (!silent)
4032                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4033         goto failed_mount;
4034
4035 failed_mount7:
4036         ext4_unregister_li_request(sb);
4037 failed_mount6:
4038         ext4_mb_release(sb);
4039 failed_mount5:
4040         ext4_ext_release(sb);
4041         ext4_release_system_zone(sb);
4042 failed_mount4a:
4043         dput(sb->s_root);
4044         sb->s_root = NULL;
4045 failed_mount4:
4046         ext4_msg(sb, KERN_ERR, "mount failed");
4047         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4048 failed_mount_wq:
4049         if (sbi->s_journal) {
4050                 jbd2_journal_destroy(sbi->s_journal);
4051                 sbi->s_journal = NULL;
4052         }
4053 failed_mount3:
4054         del_timer(&sbi->s_err_report);
4055         if (sbi->s_flex_groups)
4056                 ext4_kvfree(sbi->s_flex_groups);
4057         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4058         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4059         percpu_counter_destroy(&sbi->s_dirs_counter);
4060         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4061         if (sbi->s_mmp_tsk)
4062                 kthread_stop(sbi->s_mmp_tsk);
4063 failed_mount2:
4064         for (i = 0; i < db_count; i++)
4065                 brelse(sbi->s_group_desc[i]);
4066         ext4_kvfree(sbi->s_group_desc);
4067 failed_mount:
4068         if (sbi->s_chksum_driver)
4069                 crypto_free_shash(sbi->s_chksum_driver);
4070         if (sbi->s_proc) {
4071                 remove_proc_entry("options", sbi->s_proc);
4072                 remove_proc_entry(sb->s_id, ext4_proc_root);
4073         }
4074 #ifdef CONFIG_QUOTA
4075         for (i = 0; i < MAXQUOTAS; i++)
4076                 kfree(sbi->s_qf_names[i]);
4077 #endif
4078         ext4_blkdev_remove(sbi);
4079         brelse(bh);
4080 out_fail:
4081         sb->s_fs_info = NULL;
4082         kfree(sbi->s_blockgroup_lock);
4083         kfree(sbi);
4084 out_free_orig:
4085         kfree(orig_data);
4086         return ret;
4087 }
4088
4089 /*
4090  * Setup any per-fs journal parameters now.  We'll do this both on
4091  * initial mount, once the journal has been initialised but before we've
4092  * done any recovery; and again on any subsequent remount.
4093  */
4094 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4095 {
4096         struct ext4_sb_info *sbi = EXT4_SB(sb);
4097
4098         journal->j_commit_interval = sbi->s_commit_interval;
4099         journal->j_min_batch_time = sbi->s_min_batch_time;
4100         journal->j_max_batch_time = sbi->s_max_batch_time;
4101
4102         write_lock(&journal->j_state_lock);
4103         if (test_opt(sb, BARRIER))
4104                 journal->j_flags |= JBD2_BARRIER;
4105         else
4106                 journal->j_flags &= ~JBD2_BARRIER;
4107         if (test_opt(sb, DATA_ERR_ABORT))
4108                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4109         else
4110                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4111         write_unlock(&journal->j_state_lock);
4112 }
4113
4114 static journal_t *ext4_get_journal(struct super_block *sb,
4115                                    unsigned int journal_inum)
4116 {
4117         struct inode *journal_inode;
4118         journal_t *journal;
4119
4120         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4121
4122         /* First, test for the existence of a valid inode on disk.  Bad
4123          * things happen if we iget() an unused inode, as the subsequent
4124          * iput() will try to delete it. */
4125
4126         journal_inode = ext4_iget(sb, journal_inum);
4127         if (IS_ERR(journal_inode)) {
4128                 ext4_msg(sb, KERN_ERR, "no journal found");
4129                 return NULL;
4130         }
4131         if (!journal_inode->i_nlink) {
4132                 make_bad_inode(journal_inode);
4133                 iput(journal_inode);
4134                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4135                 return NULL;
4136         }
4137
4138         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4139                   journal_inode, journal_inode->i_size);
4140         if (!S_ISREG(journal_inode->i_mode)) {
4141                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4142                 iput(journal_inode);
4143                 return NULL;
4144         }
4145
4146         journal = jbd2_journal_init_inode(journal_inode);
4147         if (!journal) {
4148                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4149                 iput(journal_inode);
4150                 return NULL;
4151         }
4152         journal->j_private = sb;
4153         ext4_init_journal_params(sb, journal);
4154         return journal;
4155 }
4156
4157 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4158                                        dev_t j_dev)
4159 {
4160         struct buffer_head *bh;
4161         journal_t *journal;
4162         ext4_fsblk_t start;
4163         ext4_fsblk_t len;
4164         int hblock, blocksize;
4165         ext4_fsblk_t sb_block;
4166         unsigned long offset;
4167         struct ext4_super_block *es;
4168         struct block_device *bdev;
4169
4170         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4171
4172         bdev = ext4_blkdev_get(j_dev, sb);
4173         if (bdev == NULL)
4174                 return NULL;
4175
4176         blocksize = sb->s_blocksize;
4177         hblock = bdev_logical_block_size(bdev);
4178         if (blocksize < hblock) {
4179                 ext4_msg(sb, KERN_ERR,
4180                         "blocksize too small for journal device");
4181                 goto out_bdev;
4182         }
4183
4184         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4185         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4186         set_blocksize(bdev, blocksize);
4187         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4188                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4189                        "external journal");
4190                 goto out_bdev;
4191         }
4192
4193         es = (struct ext4_super_block *) (bh->b_data + offset);
4194         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4195             !(le32_to_cpu(es->s_feature_incompat) &
4196               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4197                 ext4_msg(sb, KERN_ERR, "external journal has "
4198                                         "bad superblock");
4199                 brelse(bh);
4200                 goto out_bdev;
4201         }
4202
4203         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4204                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4205                 brelse(bh);
4206                 goto out_bdev;
4207         }
4208
4209         len = ext4_blocks_count(es);
4210         start = sb_block + 1;
4211         brelse(bh);     /* we're done with the superblock */
4212
4213         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4214                                         start, len, blocksize);
4215         if (!journal) {
4216                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4217                 goto out_bdev;
4218         }
4219         journal->j_private = sb;
4220         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4221         wait_on_buffer(journal->j_sb_buffer);
4222         if (!buffer_uptodate(journal->j_sb_buffer)) {
4223                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4224                 goto out_journal;
4225         }
4226         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4227                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4228                                         "user (unsupported) - %d",
4229                         be32_to_cpu(journal->j_superblock->s_nr_users));
4230                 goto out_journal;
4231         }
4232         EXT4_SB(sb)->journal_bdev = bdev;
4233         ext4_init_journal_params(sb, journal);
4234         return journal;
4235
4236 out_journal:
4237         jbd2_journal_destroy(journal);
4238 out_bdev:
4239         ext4_blkdev_put(bdev);
4240         return NULL;
4241 }
4242
4243 static int ext4_load_journal(struct super_block *sb,
4244                              struct ext4_super_block *es,
4245                              unsigned long journal_devnum)
4246 {
4247         journal_t *journal;
4248         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4249         dev_t journal_dev;
4250         int err = 0;
4251         int really_read_only;
4252
4253         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4254
4255         if (journal_devnum &&
4256             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4257                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4258                         "numbers have changed");
4259                 journal_dev = new_decode_dev(journal_devnum);
4260         } else
4261                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4262
4263         really_read_only = bdev_read_only(sb->s_bdev);
4264
4265         /*
4266          * Are we loading a blank journal or performing recovery after a
4267          * crash?  For recovery, we need to check in advance whether we
4268          * can get read-write access to the device.
4269          */
4270         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4271                 if (sb->s_flags & MS_RDONLY) {
4272                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4273                                         "required on readonly filesystem");
4274                         if (really_read_only) {
4275                                 ext4_msg(sb, KERN_ERR, "write access "
4276                                         "unavailable, cannot proceed");
4277                                 return -EROFS;
4278                         }
4279                         ext4_msg(sb, KERN_INFO, "write access will "
4280                                "be enabled during recovery");
4281                 }
4282         }
4283
4284         if (journal_inum && journal_dev) {
4285                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4286                        "and inode journals!");
4287                 return -EINVAL;
4288         }
4289
4290         if (journal_inum) {
4291                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4292                         return -EINVAL;
4293         } else {
4294                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4295                         return -EINVAL;
4296         }
4297
4298         if (!(journal->j_flags & JBD2_BARRIER))
4299                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4300
4301         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4302                 err = jbd2_journal_wipe(journal, !really_read_only);
4303         if (!err) {
4304                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4305                 if (save)
4306                         memcpy(save, ((char *) es) +
4307                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4308                 err = jbd2_journal_load(journal);
4309                 if (save)
4310                         memcpy(((char *) es) + EXT4_S_ERR_START,
4311                                save, EXT4_S_ERR_LEN);
4312                 kfree(save);
4313         }
4314
4315         if (err) {
4316                 ext4_msg(sb, KERN_ERR, "error loading journal");
4317                 jbd2_journal_destroy(journal);
4318                 return err;
4319         }
4320
4321         EXT4_SB(sb)->s_journal = journal;
4322         ext4_clear_journal_err(sb, es);
4323
4324         if (!really_read_only && journal_devnum &&
4325             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4326                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4327
4328                 /* Make sure we flush the recovery flag to disk. */
4329                 ext4_commit_super(sb, 1);
4330         }
4331
4332         return 0;
4333 }
4334
4335 static int ext4_commit_super(struct super_block *sb, int sync)
4336 {
4337         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4338         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4339         int error = 0;
4340
4341         if (!sbh || block_device_ejected(sb))
4342                 return error;
4343         if (buffer_write_io_error(sbh)) {
4344                 /*
4345                  * Oh, dear.  A previous attempt to write the
4346                  * superblock failed.  This could happen because the
4347                  * USB device was yanked out.  Or it could happen to
4348                  * be a transient write error and maybe the block will
4349                  * be remapped.  Nothing we can do but to retry the
4350                  * write and hope for the best.
4351                  */
4352                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4353                        "superblock detected");
4354                 clear_buffer_write_io_error(sbh);
4355                 set_buffer_uptodate(sbh);
4356         }
4357         /*
4358          * If the file system is mounted read-only, don't update the
4359          * superblock write time.  This avoids updating the superblock
4360          * write time when we are mounting the root file system
4361          * read/only but we need to replay the journal; at that point,
4362          * for people who are east of GMT and who make their clock
4363          * tick in localtime for Windows bug-for-bug compatibility,
4364          * the clock is set in the future, and this will cause e2fsck
4365          * to complain and force a full file system check.
4366          */
4367         if (!(sb->s_flags & MS_RDONLY))
4368                 es->s_wtime = cpu_to_le32(get_seconds());
4369         if (sb->s_bdev->bd_part)
4370                 es->s_kbytes_written =
4371                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4372                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4373                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4374         else
4375                 es->s_kbytes_written =
4376                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4377         ext4_free_blocks_count_set(es,
4378                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4379                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4380         es->s_free_inodes_count =
4381                 cpu_to_le32(percpu_counter_sum_positive(
4382                                 &EXT4_SB(sb)->s_freeinodes_counter));
4383         BUFFER_TRACE(sbh, "marking dirty");
4384         ext4_superblock_csum_set(sb, es);
4385         mark_buffer_dirty(sbh);
4386         if (sync) {
4387                 error = sync_dirty_buffer(sbh);
4388                 if (error)
4389                         return error;
4390
4391                 error = buffer_write_io_error(sbh);
4392                 if (error) {
4393                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4394                                "superblock");
4395                         clear_buffer_write_io_error(sbh);
4396                         set_buffer_uptodate(sbh);
4397                 }
4398         }
4399         return error;
4400 }
4401
4402 /*
4403  * Have we just finished recovery?  If so, and if we are mounting (or
4404  * remounting) the filesystem readonly, then we will end up with a
4405  * consistent fs on disk.  Record that fact.
4406  */
4407 static void ext4_mark_recovery_complete(struct super_block *sb,
4408                                         struct ext4_super_block *es)
4409 {
4410         journal_t *journal = EXT4_SB(sb)->s_journal;
4411
4412         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4413                 BUG_ON(journal != NULL);
4414                 return;
4415         }
4416         jbd2_journal_lock_updates(journal);
4417         if (jbd2_journal_flush(journal) < 0)
4418                 goto out;
4419
4420         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4421             sb->s_flags & MS_RDONLY) {
4422                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4423                 ext4_commit_super(sb, 1);
4424         }
4425
4426 out:
4427         jbd2_journal_unlock_updates(journal);
4428 }
4429
4430 /*
4431  * If we are mounting (or read-write remounting) a filesystem whose journal
4432  * has recorded an error from a previous lifetime, move that error to the
4433  * main filesystem now.
4434  */
4435 static void ext4_clear_journal_err(struct super_block *sb,
4436                                    struct ext4_super_block *es)
4437 {
4438         journal_t *journal;
4439         int j_errno;
4440         const char *errstr;
4441
4442         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4443
4444         journal = EXT4_SB(sb)->s_journal;
4445
4446         /*
4447          * Now check for any error status which may have been recorded in the
4448          * journal by a prior ext4_error() or ext4_abort()
4449          */
4450
4451         j_errno = jbd2_journal_errno(journal);
4452         if (j_errno) {
4453                 char nbuf[16];
4454
4455                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4456                 ext4_warning(sb, "Filesystem error recorded "
4457                              "from previous mount: %s", errstr);
4458                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4459
4460                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4461                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4462                 ext4_commit_super(sb, 1);
4463
4464                 jbd2_journal_clear_err(journal);
4465                 jbd2_journal_update_sb_errno(journal);
4466         }
4467 }
4468
4469 /*
4470  * Force the running and committing transactions to commit,
4471  * and wait on the commit.
4472  */
4473 int ext4_force_commit(struct super_block *sb)
4474 {
4475         journal_t *journal;
4476         int ret = 0;
4477
4478         if (sb->s_flags & MS_RDONLY)
4479                 return 0;
4480
4481         journal = EXT4_SB(sb)->s_journal;
4482         if (journal)
4483                 ret = ext4_journal_force_commit(journal);
4484
4485         return ret;
4486 }
4487
4488 static int ext4_sync_fs(struct super_block *sb, int wait)
4489 {
4490         int ret = 0;
4491         tid_t target;
4492         struct ext4_sb_info *sbi = EXT4_SB(sb);
4493
4494         trace_ext4_sync_fs(sb, wait);
4495         flush_workqueue(sbi->dio_unwritten_wq);
4496         /*
4497          * Writeback quota in non-journalled quota case - journalled quota has
4498          * no dirty dquots
4499          */
4500         dquot_writeback_dquots(sb, -1);
4501         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4502                 if (wait)
4503                         jbd2_log_wait_commit(sbi->s_journal, target);
4504         }
4505         return ret;
4506 }
4507
4508 /*
4509  * LVM calls this function before a (read-only) snapshot is created.  This
4510  * gives us a chance to flush the journal completely and mark the fs clean.
4511  *
4512  * Note that only this function cannot bring a filesystem to be in a clean
4513  * state independently. It relies on upper layer to stop all data & metadata
4514  * modifications.
4515  */
4516 static int ext4_freeze(struct super_block *sb)
4517 {
4518         int error = 0;
4519         journal_t *journal;
4520
4521         if (sb->s_flags & MS_RDONLY)
4522                 return 0;
4523
4524         journal = EXT4_SB(sb)->s_journal;
4525
4526         /* Now we set up the journal barrier. */
4527         jbd2_journal_lock_updates(journal);
4528
4529         /*
4530          * Don't clear the needs_recovery flag if we failed to flush
4531          * the journal.
4532          */
4533         error = jbd2_journal_flush(journal);
4534         if (error < 0)
4535                 goto out;
4536
4537         /* Journal blocked and flushed, clear needs_recovery flag. */
4538         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4539         error = ext4_commit_super(sb, 1);
4540 out:
4541         /* we rely on upper layer to stop further updates */
4542         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4543         return error;
4544 }
4545
4546 /*
4547  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4548  * flag here, even though the filesystem is not technically dirty yet.
4549  */
4550 static int ext4_unfreeze(struct super_block *sb)
4551 {
4552         if (sb->s_flags & MS_RDONLY)
4553                 return 0;
4554
4555         /* Reset the needs_recovery flag before the fs is unlocked. */
4556         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4557         ext4_commit_super(sb, 1);
4558         return 0;
4559 }
4560
4561 /*
4562  * Structure to save mount options for ext4_remount's benefit
4563  */
4564 struct ext4_mount_options {
4565         unsigned long s_mount_opt;
4566         unsigned long s_mount_opt2;
4567         kuid_t s_resuid;
4568         kgid_t s_resgid;
4569         unsigned long s_commit_interval;
4570         u32 s_min_batch_time, s_max_batch_time;
4571 #ifdef CONFIG_QUOTA
4572         int s_jquota_fmt;
4573         char *s_qf_names[MAXQUOTAS];
4574 #endif
4575 };
4576
4577 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4578 {
4579         struct ext4_super_block *es;
4580         struct ext4_sb_info *sbi = EXT4_SB(sb);
4581         unsigned long old_sb_flags;
4582         struct ext4_mount_options old_opts;
4583         int enable_quota = 0;
4584         ext4_group_t g;
4585         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4586         int err = 0;
4587 #ifdef CONFIG_QUOTA
4588         int i;
4589 #endif
4590         char *orig_data = kstrdup(data, GFP_KERNEL);
4591
4592         /* Store the original options */
4593         old_sb_flags = sb->s_flags;
4594         old_opts.s_mount_opt = sbi->s_mount_opt;
4595         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4596         old_opts.s_resuid = sbi->s_resuid;
4597         old_opts.s_resgid = sbi->s_resgid;
4598         old_opts.s_commit_interval = sbi->s_commit_interval;
4599         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4600         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4601 #ifdef CONFIG_QUOTA
4602         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4603         for (i = 0; i < MAXQUOTAS; i++)
4604                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4605 #endif
4606         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4607                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4608
4609         /*
4610          * Allow the "check" option to be passed as a remount option.
4611          */
4612         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4613                 err = -EINVAL;
4614                 goto restore_opts;
4615         }
4616
4617         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4618                 ext4_abort(sb, "Abort forced by user");
4619
4620         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4621                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4622
4623         es = sbi->s_es;
4624
4625         if (sbi->s_journal) {
4626                 ext4_init_journal_params(sb, sbi->s_journal);
4627                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4628         }
4629
4630         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4631                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4632                         err = -EROFS;
4633                         goto restore_opts;
4634                 }
4635
4636                 if (*flags & MS_RDONLY) {
4637                         err = dquot_suspend(sb, -1);
4638                         if (err < 0)
4639                                 goto restore_opts;
4640
4641                         /*
4642                          * First of all, the unconditional stuff we have to do
4643                          * to disable replay of the journal when we next remount
4644                          */
4645                         sb->s_flags |= MS_RDONLY;
4646
4647                         /*
4648                          * OK, test if we are remounting a valid rw partition
4649                          * readonly, and if so set the rdonly flag and then
4650                          * mark the partition as valid again.
4651                          */
4652                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4653                             (sbi->s_mount_state & EXT4_VALID_FS))
4654                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4655
4656                         if (sbi->s_journal)
4657                                 ext4_mark_recovery_complete(sb, es);
4658                 } else {
4659                         /* Make sure we can mount this feature set readwrite */
4660                         if (!ext4_feature_set_ok(sb, 0)) {
4661                                 err = -EROFS;
4662                                 goto restore_opts;
4663                         }
4664                         /*
4665                          * Make sure the group descriptor checksums
4666                          * are sane.  If they aren't, refuse to remount r/w.
4667                          */
4668                         for (g = 0; g < sbi->s_groups_count; g++) {
4669                                 struct ext4_group_desc *gdp =
4670                                         ext4_get_group_desc(sb, g, NULL);
4671
4672                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4673                                         ext4_msg(sb, KERN_ERR,
4674                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4675                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4676                                                le16_to_cpu(gdp->bg_checksum));
4677                                         err = -EINVAL;
4678                                         goto restore_opts;
4679                                 }
4680                         }
4681
4682                         /*
4683                          * If we have an unprocessed orphan list hanging
4684                          * around from a previously readonly bdev mount,
4685                          * require a full umount/remount for now.
4686                          */
4687                         if (es->s_last_orphan) {
4688                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4689                                        "remount RDWR because of unprocessed "
4690                                        "orphan inode list.  Please "
4691                                        "umount/remount instead");
4692                                 err = -EINVAL;
4693                                 goto restore_opts;
4694                         }
4695
4696                         /*
4697                          * Mounting a RDONLY partition read-write, so reread
4698                          * and store the current valid flag.  (It may have
4699                          * been changed by e2fsck since we originally mounted
4700                          * the partition.)
4701                          */
4702                         if (sbi->s_journal)
4703                                 ext4_clear_journal_err(sb, es);
4704                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4705                         if (!ext4_setup_super(sb, es, 0))
4706                                 sb->s_flags &= ~MS_RDONLY;
4707                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4708                                                      EXT4_FEATURE_INCOMPAT_MMP))
4709                                 if (ext4_multi_mount_protect(sb,
4710                                                 le64_to_cpu(es->s_mmp_block))) {
4711                                         err = -EROFS;
4712                                         goto restore_opts;
4713                                 }
4714                         enable_quota = 1;
4715                 }
4716         }
4717
4718         /*
4719          * Reinitialize lazy itable initialization thread based on
4720          * current settings
4721          */
4722         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4723                 ext4_unregister_li_request(sb);
4724         else {
4725                 ext4_group_t first_not_zeroed;
4726                 first_not_zeroed = ext4_has_uninit_itable(sb);
4727                 ext4_register_li_request(sb, first_not_zeroed);
4728         }
4729
4730         ext4_setup_system_zone(sb);
4731         if (sbi->s_journal == NULL)
4732                 ext4_commit_super(sb, 1);
4733
4734 #ifdef CONFIG_QUOTA
4735         /* Release old quota file names */
4736         for (i = 0; i < MAXQUOTAS; i++)
4737                 if (old_opts.s_qf_names[i] &&
4738                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4739                         kfree(old_opts.s_qf_names[i]);
4740         if (enable_quota) {
4741                 if (sb_any_quota_suspended(sb))
4742                         dquot_resume(sb, -1);
4743                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4744                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4745                         err = ext4_enable_quotas(sb);
4746                         if (err)
4747                                 goto restore_opts;
4748                 }
4749         }
4750 #endif
4751
4752         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4753         kfree(orig_data);
4754         return 0;
4755
4756 restore_opts:
4757         sb->s_flags = old_sb_flags;
4758         sbi->s_mount_opt = old_opts.s_mount_opt;
4759         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4760         sbi->s_resuid = old_opts.s_resuid;
4761         sbi->s_resgid = old_opts.s_resgid;
4762         sbi->s_commit_interval = old_opts.s_commit_interval;
4763         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4764         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4765 #ifdef CONFIG_QUOTA
4766         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4767         for (i = 0; i < MAXQUOTAS; i++) {
4768                 if (sbi->s_qf_names[i] &&
4769                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4770                         kfree(sbi->s_qf_names[i]);
4771                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4772         }
4773 #endif
4774         kfree(orig_data);
4775         return err;
4776 }
4777
4778 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4779 {
4780         struct super_block *sb = dentry->d_sb;
4781         struct ext4_sb_info *sbi = EXT4_SB(sb);
4782         struct ext4_super_block *es = sbi->s_es;
4783         ext4_fsblk_t overhead = 0;
4784         u64 fsid;
4785         s64 bfree;
4786
4787         if (!test_opt(sb, MINIX_DF))
4788                 overhead = sbi->s_overhead;
4789
4790         buf->f_type = EXT4_SUPER_MAGIC;
4791         buf->f_bsize = sb->s_blocksize;
4792         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4793         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4794                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4795         /* prevent underflow in case that few free space is available */
4796         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4797         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4798         if (buf->f_bfree < ext4_r_blocks_count(es))
4799                 buf->f_bavail = 0;
4800         buf->f_files = le32_to_cpu(es->s_inodes_count);
4801         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4802         buf->f_namelen = EXT4_NAME_LEN;
4803         fsid = le64_to_cpup((void *)es->s_uuid) ^
4804                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4805         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4806         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4807
4808         return 0;
4809 }
4810
4811 /* Helper function for writing quotas on sync - we need to start transaction
4812  * before quota file is locked for write. Otherwise the are possible deadlocks:
4813  * Process 1                         Process 2
4814  * ext4_create()                     quota_sync()
4815  *   jbd2_journal_start()                  write_dquot()
4816  *   dquot_initialize()                         down(dqio_mutex)
4817  *     down(dqio_mutex)                    jbd2_journal_start()
4818  *
4819  */
4820
4821 #ifdef CONFIG_QUOTA
4822
4823 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4824 {
4825         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4826 }
4827
4828 static int ext4_write_dquot(struct dquot *dquot)
4829 {
4830         int ret, err;
4831         handle_t *handle;
4832         struct inode *inode;
4833
4834         inode = dquot_to_inode(dquot);
4835         handle = ext4_journal_start(inode,
4836                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4837         if (IS_ERR(handle))
4838                 return PTR_ERR(handle);
4839         ret = dquot_commit(dquot);
4840         err = ext4_journal_stop(handle);
4841         if (!ret)
4842                 ret = err;
4843         return ret;
4844 }
4845
4846 static int ext4_acquire_dquot(struct dquot *dquot)
4847 {
4848         int ret, err;
4849         handle_t *handle;
4850
4851         handle = ext4_journal_start(dquot_to_inode(dquot),
4852                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4853         if (IS_ERR(handle))
4854                 return PTR_ERR(handle);
4855         ret = dquot_acquire(dquot);
4856         err = ext4_journal_stop(handle);
4857         if (!ret)
4858                 ret = err;
4859         return ret;
4860 }
4861
4862 static int ext4_release_dquot(struct dquot *dquot)
4863 {
4864         int ret, err;
4865         handle_t *handle;
4866
4867         handle = ext4_journal_start(dquot_to_inode(dquot),
4868                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4869         if (IS_ERR(handle)) {
4870                 /* Release dquot anyway to avoid endless cycle in dqput() */
4871                 dquot_release(dquot);
4872                 return PTR_ERR(handle);
4873         }
4874         ret = dquot_release(dquot);
4875         err = ext4_journal_stop(handle);
4876         if (!ret)
4877                 ret = err;
4878         return ret;
4879 }
4880
4881 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4882 {
4883         /* Are we journaling quotas? */
4884         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4885             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4886                 dquot_mark_dquot_dirty(dquot);
4887                 return ext4_write_dquot(dquot);
4888         } else {
4889                 return dquot_mark_dquot_dirty(dquot);
4890         }
4891 }
4892
4893 static int ext4_write_info(struct super_block *sb, int type)
4894 {
4895         int ret, err;
4896         handle_t *handle;
4897
4898         /* Data block + inode block */
4899         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4900         if (IS_ERR(handle))
4901                 return PTR_ERR(handle);
4902         ret = dquot_commit_info(sb, type);
4903         err = ext4_journal_stop(handle);
4904         if (!ret)
4905                 ret = err;
4906         return ret;
4907 }
4908
4909 /*
4910  * Turn on quotas during mount time - we need to find
4911  * the quota file and such...
4912  */
4913 static int ext4_quota_on_mount(struct super_block *sb, int type)
4914 {
4915         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4916                                         EXT4_SB(sb)->s_jquota_fmt, type);
4917 }
4918
4919 /*
4920  * Standard function to be called on quota_on
4921  */
4922 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4923                          struct path *path)
4924 {
4925         int err;
4926
4927         if (!test_opt(sb, QUOTA))
4928                 return -EINVAL;
4929
4930         /* Quotafile not on the same filesystem? */
4931         if (path->dentry->d_sb != sb)
4932                 return -EXDEV;
4933         /* Journaling quota? */
4934         if (EXT4_SB(sb)->s_qf_names[type]) {
4935                 /* Quotafile not in fs root? */
4936                 if (path->dentry->d_parent != sb->s_root)
4937                         ext4_msg(sb, KERN_WARNING,
4938                                 "Quota file not on filesystem root. "
4939                                 "Journaled quota will not work");
4940         }
4941
4942         /*
4943          * When we journal data on quota file, we have to flush journal to see
4944          * all updates to the file when we bypass pagecache...
4945          */
4946         if (EXT4_SB(sb)->s_journal &&
4947             ext4_should_journal_data(path->dentry->d_inode)) {
4948                 /*
4949                  * We don't need to lock updates but journal_flush() could
4950                  * otherwise be livelocked...
4951                  */
4952                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4953                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4954                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4955                 if (err)
4956                         return err;
4957         }
4958
4959         return dquot_quota_on(sb, type, format_id, path);
4960 }
4961
4962 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4963                              unsigned int flags)
4964 {
4965         int err;
4966         struct inode *qf_inode;
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         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4973
4974         if (!qf_inums[type])
4975                 return -EPERM;
4976
4977         qf_inode = ext4_iget(sb, qf_inums[type]);
4978         if (IS_ERR(qf_inode)) {
4979                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4980                 return PTR_ERR(qf_inode);
4981         }
4982
4983         err = dquot_enable(qf_inode, type, format_id, flags);
4984         iput(qf_inode);
4985
4986         return err;
4987 }
4988
4989 /* Enable usage tracking for all quota types. */
4990 static int ext4_enable_quotas(struct super_block *sb)
4991 {
4992         int type, err = 0;
4993         unsigned long qf_inums[MAXQUOTAS] = {
4994                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4995                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4996         };
4997
4998         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
4999         for (type = 0; type < MAXQUOTAS; type++) {
5000                 if (qf_inums[type]) {
5001                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5002                                                 DQUOT_USAGE_ENABLED);
5003                         if (err) {
5004                                 ext4_warning(sb,
5005                                         "Failed to enable quota (type=%d) "
5006                                         "tracking. Please run e2fsck to fix.",
5007                                         type);
5008                                 return err;
5009                         }
5010                 }
5011         }
5012         return 0;
5013 }
5014
5015 /*
5016  * quota_on function that is used when QUOTA feature is set.
5017  */
5018 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5019                                  int format_id)
5020 {
5021         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5022                 return -EINVAL;
5023
5024         /*
5025          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5026          */
5027         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5028 }
5029
5030 static int ext4_quota_off(struct super_block *sb, int type)
5031 {
5032         struct inode *inode = sb_dqopt(sb)->files[type];
5033         handle_t *handle;
5034
5035         /* Force all delayed allocation blocks to be allocated.
5036          * Caller already holds s_umount sem */
5037         if (test_opt(sb, DELALLOC))
5038                 sync_filesystem(sb);
5039
5040         if (!inode)
5041                 goto out;
5042
5043         /* Update modification times of quota files when userspace can
5044          * start looking at them */
5045         handle = ext4_journal_start(inode, 1);
5046         if (IS_ERR(handle))
5047                 goto out;
5048         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5049         ext4_mark_inode_dirty(handle, inode);
5050         ext4_journal_stop(handle);
5051
5052 out:
5053         return dquot_quota_off(sb, type);
5054 }
5055
5056 /*
5057  * quota_off function that is used when QUOTA feature is set.
5058  */
5059 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5060 {
5061         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5062                 return -EINVAL;
5063
5064         /* Disable only the limits. */
5065         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5066 }
5067
5068 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5069  * acquiring the locks... As quota files are never truncated and quota code
5070  * itself serializes the operations (and no one else should touch the files)
5071  * we don't have to be afraid of races */
5072 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5073                                size_t len, loff_t off)
5074 {
5075         struct inode *inode = sb_dqopt(sb)->files[type];
5076         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5077         int err = 0;
5078         int offset = off & (sb->s_blocksize - 1);
5079         int tocopy;
5080         size_t toread;
5081         struct buffer_head *bh;
5082         loff_t i_size = i_size_read(inode);
5083
5084         if (off > i_size)
5085                 return 0;
5086         if (off+len > i_size)
5087                 len = i_size-off;
5088         toread = len;
5089         while (toread > 0) {
5090                 tocopy = sb->s_blocksize - offset < toread ?
5091                                 sb->s_blocksize - offset : toread;
5092                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5093                 if (err)
5094                         return err;
5095                 if (!bh)        /* A hole? */
5096                         memset(data, 0, tocopy);
5097                 else
5098                         memcpy(data, bh->b_data+offset, tocopy);
5099                 brelse(bh);
5100                 offset = 0;
5101                 toread -= tocopy;
5102                 data += tocopy;
5103                 blk++;
5104         }
5105         return len;
5106 }
5107
5108 /* Write to quotafile (we know the transaction is already started and has
5109  * enough credits) */
5110 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5111                                 const char *data, size_t len, loff_t off)
5112 {
5113         struct inode *inode = sb_dqopt(sb)->files[type];
5114         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5115         int err = 0;
5116         int offset = off & (sb->s_blocksize - 1);
5117         struct buffer_head *bh;
5118         handle_t *handle = journal_current_handle();
5119
5120         if (EXT4_SB(sb)->s_journal && !handle) {
5121                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5122                         " cancelled because transaction is not started",
5123                         (unsigned long long)off, (unsigned long long)len);
5124                 return -EIO;
5125         }
5126         /*
5127          * Since we account only one data block in transaction credits,
5128          * then it is impossible to cross a block boundary.
5129          */
5130         if (sb->s_blocksize - offset < len) {
5131                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5132                         " cancelled because not block aligned",
5133                         (unsigned long long)off, (unsigned long long)len);
5134                 return -EIO;
5135         }
5136
5137         bh = ext4_bread(handle, inode, blk, 1, &err);
5138         if (!bh)
5139                 goto out;
5140         err = ext4_journal_get_write_access(handle, bh);
5141         if (err) {
5142                 brelse(bh);
5143                 goto out;
5144         }
5145         lock_buffer(bh);
5146         memcpy(bh->b_data+offset, data, len);
5147         flush_dcache_page(bh->b_page);
5148         unlock_buffer(bh);
5149         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5150         brelse(bh);
5151 out:
5152         if (err)
5153                 return err;
5154         if (inode->i_size < off + len) {
5155                 i_size_write(inode, off + len);
5156                 EXT4_I(inode)->i_disksize = inode->i_size;
5157                 ext4_mark_inode_dirty(handle, inode);
5158         }
5159         return len;
5160 }
5161
5162 #endif
5163
5164 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5165                        const char *dev_name, void *data)
5166 {
5167         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5168 }
5169
5170 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5171 static inline void register_as_ext2(void)
5172 {
5173         int err = register_filesystem(&ext2_fs_type);
5174         if (err)
5175                 printk(KERN_WARNING
5176                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5177 }
5178
5179 static inline void unregister_as_ext2(void)
5180 {
5181         unregister_filesystem(&ext2_fs_type);
5182 }
5183
5184 static inline int ext2_feature_set_ok(struct super_block *sb)
5185 {
5186         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5187                 return 0;
5188         if (sb->s_flags & MS_RDONLY)
5189                 return 1;
5190         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5191                 return 0;
5192         return 1;
5193 }
5194 MODULE_ALIAS("ext2");
5195 #else
5196 static inline void register_as_ext2(void) { }
5197 static inline void unregister_as_ext2(void) { }
5198 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5199 #endif
5200
5201 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5202 static inline void register_as_ext3(void)
5203 {
5204         int err = register_filesystem(&ext3_fs_type);
5205         if (err)
5206                 printk(KERN_WARNING
5207                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5208 }
5209
5210 static inline void unregister_as_ext3(void)
5211 {
5212         unregister_filesystem(&ext3_fs_type);
5213 }
5214
5215 static inline int ext3_feature_set_ok(struct super_block *sb)
5216 {
5217         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5218                 return 0;
5219         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5220                 return 0;
5221         if (sb->s_flags & MS_RDONLY)
5222                 return 1;
5223         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5224                 return 0;
5225         return 1;
5226 }
5227 MODULE_ALIAS("ext3");
5228 #else
5229 static inline void register_as_ext3(void) { }
5230 static inline void unregister_as_ext3(void) { }
5231 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5232 #endif
5233
5234 static struct file_system_type ext4_fs_type = {
5235         .owner          = THIS_MODULE,
5236         .name           = "ext4",
5237         .mount          = ext4_mount,
5238         .kill_sb        = kill_block_super,
5239         .fs_flags       = FS_REQUIRES_DEV,
5240 };
5241
5242 static int __init ext4_init_feat_adverts(void)
5243 {
5244         struct ext4_features *ef;
5245         int ret = -ENOMEM;
5246
5247         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5248         if (!ef)
5249                 goto out;
5250
5251         ef->f_kobj.kset = ext4_kset;
5252         init_completion(&ef->f_kobj_unregister);
5253         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5254                                    "features");
5255         if (ret) {
5256                 kfree(ef);
5257                 goto out;
5258         }
5259
5260         ext4_feat = ef;
5261         ret = 0;
5262 out:
5263         return ret;
5264 }
5265
5266 static void ext4_exit_feat_adverts(void)
5267 {
5268         kobject_put(&ext4_feat->f_kobj);
5269         wait_for_completion(&ext4_feat->f_kobj_unregister);
5270         kfree(ext4_feat);
5271 }
5272
5273 /* Shared across all ext4 file systems */
5274 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5275 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5276
5277 static int __init ext4_init_fs(void)
5278 {
5279         int i, err;
5280
5281         ext4_li_info = NULL;
5282         mutex_init(&ext4_li_mtx);
5283
5284         ext4_check_flag_values();
5285
5286         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5287                 mutex_init(&ext4__aio_mutex[i]);
5288                 init_waitqueue_head(&ext4__ioend_wq[i]);
5289         }
5290
5291         err = ext4_init_pageio();
5292         if (err)
5293                 return err;
5294         err = ext4_init_system_zone();
5295         if (err)
5296                 goto out6;
5297         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5298         if (!ext4_kset) {
5299                 err = -ENOMEM;
5300                 goto out5;
5301         }
5302         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5303
5304         err = ext4_init_feat_adverts();
5305         if (err)
5306                 goto out4;
5307
5308         err = ext4_init_mballoc();
5309         if (err)
5310                 goto out3;
5311
5312         err = ext4_init_xattr();
5313         if (err)
5314                 goto out2;
5315         err = init_inodecache();
5316         if (err)
5317                 goto out1;
5318         register_as_ext3();
5319         register_as_ext2();
5320         err = register_filesystem(&ext4_fs_type);
5321         if (err)
5322                 goto out;
5323
5324         return 0;
5325 out:
5326         unregister_as_ext2();
5327         unregister_as_ext3();
5328         destroy_inodecache();
5329 out1:
5330         ext4_exit_xattr();
5331 out2:
5332         ext4_exit_mballoc();
5333 out3:
5334         ext4_exit_feat_adverts();
5335 out4:
5336         if (ext4_proc_root)
5337                 remove_proc_entry("fs/ext4", NULL);
5338         kset_unregister(ext4_kset);
5339 out5:
5340         ext4_exit_system_zone();
5341 out6:
5342         ext4_exit_pageio();
5343         return err;
5344 }
5345
5346 static void __exit ext4_exit_fs(void)
5347 {
5348         ext4_destroy_lazyinit_thread();
5349         unregister_as_ext2();
5350         unregister_as_ext3();
5351         unregister_filesystem(&ext4_fs_type);
5352         destroy_inodecache();
5353         ext4_exit_xattr();
5354         ext4_exit_mballoc();
5355         ext4_exit_feat_adverts();
5356         remove_proc_entry("fs/ext4", NULL);
5357         kset_unregister(ext4_kset);
5358         ext4_exit_system_zone();
5359         ext4_exit_pageio();
5360 }
5361
5362 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5363 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5364 MODULE_LICENSE("GPL");
5365 module_init(ext4_init_fs)
5366 module_exit(ext4_exit_fs)