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_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(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 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434
 435 /* Deal with the reporting of failure conditions on a filesystem such as
 436  * inconsistencies detected or read IO failures.
 437  *
 438  * On ext2, we can store the error state of the filesystem in the
 439  * superblock.  That is not possible on ext4, because we may have other
 440  * write ordering constraints on the superblock which prevent us from
 441  * writing it out straight away; and given that the journal is about to
 442  * be aborted, we can't rely on the current, or future, transactions to
 443  * write out the superblock safely.
 444  *
 445  * We'll just use the jbd2_journal_abort() error code to record an error in
 446  * the journal instead.  On recovery, the journal will complain about
 447  * that error until we've noted it down and cleared it.
 448  */
 449
 450 static void ext4_handle_error(struct super_block *sb)
 451 {
 452         if (sb->s_flags & MS_RDONLY)
 453                 return;
 454
 455         if (!test_opt(sb, ERRORS_CONT)) {
 456                 journal_t *journal = EXT4_SB(sb)->s_journal;
 457
 458                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 459                 if (journal)
 460                         jbd2_journal_abort(journal, -EIO);
 461         }
 462         if (test_opt(sb, ERRORS_RO)) {
 463                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 464                 sb->s_flags |= MS_RDONLY;
 465         }
 466         if (test_opt(sb, ERRORS_PANIC)) {
 467                 if (EXT4_SB(sb)->s_journal &&
 468                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 469                         return;
 470                 panic("EXT4-fs (device %s): panic forced after error\n",
 471                         sb->s_id);
 472         }
 473 }
 474
 475 void __ext4_error(struct super_block *sb, const char *function,
 476                   unsigned int line, const char *fmt, ...)
 477 {
 478         struct va_format vaf;
 479         va_list args;
 480
 481         va_start(args, fmt);
 482         vaf.fmt = fmt;
 483         vaf.va = &args;
 484         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 485                sb->s_id, function, line, current->comm, &vaf);
 486         va_end(args);
 487         save_error_info(sb, function, line);
 488
 489         ext4_handle_error(sb);
 490 }
 491
 492 void ext4_error_inode(struct inode *inode, const char *function,
 493                       unsigned int line, ext4_fsblk_t block,
 494                       const char *fmt, ...)
 495 {
 496         va_list args;
 497         struct va_format vaf;
 498         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 499
 500         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 501         es->s_last_error_block = cpu_to_le64(block);
 502         save_error_info(inode->i_sb, function, line);
 503         va_start(args, fmt);
 504         vaf.fmt = fmt;
 505         vaf.va = &args;
 506         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 507                inode->i_sb->s_id, function, line, inode->i_ino);
 508         if (block)
 509                 printk(KERN_CONT "block %llu: ", block);
 510         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 511         va_end(args);
 512
 513         ext4_handle_error(inode->i_sb);
 514 }
 515
 516 void ext4_error_file(struct file *file, const char *function,
 517                      unsigned int line, ext4_fsblk_t block,
 518                      const char *fmt, ...)
 519 {
 520         va_list args;
 521         struct va_format vaf;
 522         struct ext4_super_block *es;
 523         struct inode *inode = file->f_dentry->d_inode;
 524         char pathname[80], *path;
 525
 526         es = EXT4_SB(inode->i_sb)->s_es;
 527         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 528         save_error_info(inode->i_sb, function, line);
 529         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 530         if (IS_ERR(path))
 531                 path = "(unknown)";
 532         printk(KERN_CRIT
 533                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 534                inode->i_sb->s_id, function, line, inode->i_ino);
 535         if (block)
 536                 printk(KERN_CONT "block %llu: ", block);
 537         va_start(args, fmt);
 538         vaf.fmt = fmt;
 539         vaf.va = &args;
 540         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 541         va_end(args);
 542
 543         ext4_handle_error(inode->i_sb);
 544 }
 545
 546 static const char *ext4_decode_error(struct super_block *sb, int errno,
 547                                      char nbuf[16])
 548 {
 549         char *errstr = NULL;
 550
 551         switch (errno) {
 552         case -EIO:
 553                 errstr = "IO failure";
 554                 break;
 555         case -ENOMEM:
 556                 errstr = "Out of memory";
 557                 break;
 558         case -EROFS:
 559                 if (!sb || (EXT4_SB(sb)->s_journal &&
 560                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 561                         errstr = "Journal has aborted";
 562                 else
 563                         errstr = "Readonly filesystem";
 564                 break;
 565         default:
 566                 /* If the caller passed in an extra buffer for unknown
 567                  * errors, textualise them now.  Else we just return
 568                  * NULL. */
 569                 if (nbuf) {
 570                         /* Check for truncated error codes... */
 571                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 572                                 errstr = nbuf;
 573                 }
 574                 break;
 575         }
 576
 577         return errstr;
 578 }
 579
 580 /* __ext4_std_error decodes expected errors from journaling functions
 581  * automatically and invokes the appropriate error response.  */
 582
 583 void __ext4_std_error(struct super_block *sb, const char *function,
 584                       unsigned int line, int errno)
 585 {
 586         char nbuf[16];
 587         const char *errstr;
 588
 589         /* Special case: if the error is EROFS, and we're not already
 590          * inside a transaction, then there's really no point in logging
 591          * an error. */
 592         if (errno == -EROFS && journal_current_handle() == NULL &&
 593             (sb->s_flags & MS_RDONLY))
 594                 return;
 595
 596         errstr = ext4_decode_error(sb, errno, nbuf);
 597         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 598                sb->s_id, function, line, errstr);
 599         save_error_info(sb, function, line);
 600
 601         ext4_handle_error(sb);
 602 }
 603
 604 /*
 605  * ext4_abort is a much stronger failure handler than ext4_error.  The
 606  * abort function may be used to deal with unrecoverable failures such
 607  * as journal IO errors or ENOMEM at a critical moment in log management.
 608  *
 609  * We unconditionally force the filesystem into an ABORT|READONLY state,
 610  * unless the error response on the fs has been set to panic in which
 611  * case we take the easy way out and panic immediately.
 612  */
 613
 614 void __ext4_abort(struct super_block *sb, const char *function,
 615                 unsigned int line, const char *fmt, ...)
 616 {
 617         va_list args;
 618
 619         save_error_info(sb, function, line);
 620         va_start(args, fmt);
 621         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 622                function, line);
 623         vprintk(fmt, args);
 624         printk("\n");
 625         va_end(args);
 626
 627         if ((sb->s_flags & MS_RDONLY) == 0) {
 628                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 629                 sb->s_flags |= MS_RDONLY;
 630                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 631                 if (EXT4_SB(sb)->s_journal)
 632                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 633                 save_error_info(sb, function, line);
 634         }
 635         if (test_opt(sb, ERRORS_PANIC)) {
 636                 if (EXT4_SB(sb)->s_journal &&
 637                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 638                         return;
 639                 panic("EXT4-fs panic from previous error\n");
 640         }
 641 }
 642
 643 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 644 {
 645         struct va_format vaf;
 646         va_list args;
 647
 648         va_start(args, fmt);
 649         vaf.fmt = fmt;
 650         vaf.va = &args;
 651         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_warning(struct super_block *sb, const char *function,
 656                     unsigned int line, const char *fmt, ...)
 657 {
 658         struct va_format vaf;
 659         va_list args;
 660
 661         va_start(args, fmt);
 662         vaf.fmt = fmt;
 663         vaf.va = &args;
 664         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 665                sb->s_id, function, line, &vaf);
 666         va_end(args);
 667 }
 668
 669 void __ext4_grp_locked_error(const char *function, unsigned int line,
 670                              struct super_block *sb, ext4_group_t grp,
 671                              unsigned long ino, ext4_fsblk_t block,
 672                              const char *fmt, ...)
 673 __releases(bitlock)
 674 __acquires(bitlock)
 675 {
 676         struct va_format vaf;
 677         va_list args;
 678         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 679
 680         es->s_last_error_ino = cpu_to_le32(ino);
 681         es->s_last_error_block = cpu_to_le64(block);
 682         __save_error_info(sb, function, line);
 683
 684         va_start(args, fmt);
 685
 686         vaf.fmt = fmt;
 687         vaf.va = &args;
 688         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 689                sb->s_id, function, line, grp);
 690         if (ino)
 691                 printk(KERN_CONT "inode %lu: ", ino);
 692         if (block)
 693                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 694         printk(KERN_CONT "%pV\n", &vaf);
 695         va_end(args);
 696
 697         if (test_opt(sb, ERRORS_CONT)) {
 698                 ext4_commit_super(sb, 0);
 699                 return;
 700         }
 701
 702         ext4_unlock_group(sb, grp);
 703         ext4_handle_error(sb);
 704         /*
 705          * We only get here in the ERRORS_RO case; relocking the group
 706          * may be dangerous, but nothing bad will happen since the
 707          * filesystem will have already been marked read/only and the
 708          * journal has been aborted.  We return 1 as a hint to callers
 709          * who might what to use the return value from
 710          * ext4_grp_locked_error() to distinguish between the
 711          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 712          * aggressively from the ext4 function in question, with a
 713          * more appropriate error code.
 714          */
 715         ext4_lock_group(sb, grp);
 716         return;
 717 }
 718
 719 void ext4_update_dynamic_rev(struct super_block *sb)
 720 {
 721         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 722
 723         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 724                 return;
 725
 726         ext4_warning(sb,
 727                      "updating to rev %d because of new feature flag, "
 728                      "running e2fsck is recommended",
 729                      EXT4_DYNAMIC_REV);
 730
 731         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 732         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 733         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 734         /* leave es->s_feature_*compat flags alone */
 735         /* es->s_uuid will be set by e2fsck if empty */
 736
 737         /*
 738          * The rest of the superblock fields should be zero, and if not it
 739          * means they are likely already in use, so leave them alone.  We
 740          * can leave it up to e2fsck to clean up any inconsistencies there.
 741          */
 742 }
 743
 744 /*
 745  * Open the external journal device
 746  */
 747 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 748 {
 749         struct block_device *bdev;
 750         char b[BDEVNAME_SIZE];
 751
 752         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 753         if (IS_ERR(bdev))
 754                 goto fail;
 755         return bdev;
 756
 757 fail:
 758         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 759                         __bdevname(dev, b), PTR_ERR(bdev));
 760         return NULL;
 761 }
 762
 763 /*
 764  * Release the journal device
 765  */
 766 static int ext4_blkdev_put(struct block_device *bdev)
 767 {
 768         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 769 }
 770
 771 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 772 {
 773         struct block_device *bdev;
 774         int ret = -ENODEV;
 775
 776         bdev = sbi->journal_bdev;
 777         if (bdev) {
 778                 ret = ext4_blkdev_put(bdev);
 779                 sbi->journal_bdev = NULL;
 780         }
 781         return ret;
 782 }
 783
 784 static inline struct inode *orphan_list_entry(struct list_head *l)
 785 {
 786         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 787 }
 788
 789 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 790 {
 791         struct list_head *l;
 792
 793         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 794                  le32_to_cpu(sbi->s_es->s_last_orphan));
 795
 796         printk(KERN_ERR "sb_info orphan list:\n");
 797         list_for_each(l, &sbi->s_orphan) {
 798                 struct inode *inode = orphan_list_entry(l);
 799                 printk(KERN_ERR "  "
 800                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 801                        inode->i_sb->s_id, inode->i_ino, inode,
 802                        inode->i_mode, inode->i_nlink,
 803                        NEXT_ORPHAN(inode));
 804         }
 805 }
 806
 807 static void ext4_put_super(struct super_block *sb)
 808 {
 809         struct ext4_sb_info *sbi = EXT4_SB(sb);
 810         struct ext4_super_block *es = sbi->s_es;
 811         int aborted = 0;
 812         int i, err;
 813
 814         ext4_unregister_li_request(sb);
 815         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 816
 817         flush_workqueue(sbi->dio_unwritten_wq);
 818         destroy_workqueue(sbi->dio_unwritten_wq);
 819
 820         lock_super(sb);
 821         if (sb->s_dirt)
 822                 ext4_commit_super(sb, 1);
 823
 824         if (sbi->s_journal) {
 825                 aborted = is_journal_aborted(sbi->s_journal);
 826                 err = jbd2_journal_destroy(sbi->s_journal);
 827                 sbi->s_journal = NULL;
 828                 if ((err < 0) && !aborted)
 829                         ext4_abort(sb, "Couldn't clean up the journal");
 830         }
 831
 832         del_timer_sync(&sbi->s_err_report);
 833         ext4_release_system_zone(sb);
 834         ext4_mb_release(sb);
 835         ext4_ext_release(sb);
 836         ext4_xattr_put_super(sb);
 837
 838         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 839                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 840                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 841                 ext4_commit_super(sb, 1);
 842         }
 843         if (sbi->s_proc) {
 844                 remove_proc_entry(sb->s_id, ext4_proc_root);
 845         }
 846         kobject_del(&sbi->s_kobj);
 847
 848         for (i = 0; i < sbi->s_gdb_count; i++)
 849                 brelse(sbi->s_group_desc[i]);
 850         ext4_kvfree(sbi->s_group_desc);
 851         ext4_kvfree(sbi->s_flex_groups);
 852         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 853         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 854         percpu_counter_destroy(&sbi->s_dirs_counter);
 855         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 856         brelse(sbi->s_sbh);
 857 #ifdef CONFIG_QUOTA
 858         for (i = 0; i < MAXQUOTAS; i++)
 859                 kfree(sbi->s_qf_names[i]);
 860 #endif
 861
 862         /* Debugging code just in case the in-memory inode orphan list
 863          * isn't empty.  The on-disk one can be non-empty if we've
 864          * detected an error and taken the fs readonly, but the
 865          * in-memory list had better be clean by this point. */
 866         if (!list_empty(&sbi->s_orphan))
 867                 dump_orphan_list(sb, sbi);
 868         J_ASSERT(list_empty(&sbi->s_orphan));
 869
 870         sync_blockdev(sb->s_bdev);
 871         invalidate_bdev(sb->s_bdev);
 872         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 873                 /*
 874                  * Invalidate the journal device's buffers.  We don't want them
 875                  * floating about in memory - the physical journal device may
 876                  * hotswapped, and it breaks the `ro-after' testing code.
 877                  */
 878                 sync_blockdev(sbi->journal_bdev);
 879                 invalidate_bdev(sbi->journal_bdev);
 880                 ext4_blkdev_remove(sbi);
 881         }
 882         if (sbi->s_mmp_tsk)
 883                 kthread_stop(sbi->s_mmp_tsk);
 884         sb->s_fs_info = NULL;
 885         /*
 886          * Now that we are completely done shutting down the
 887          * superblock, we need to actually destroy the kobject.
 888          */
 889         unlock_super(sb);
 890         kobject_put(&sbi->s_kobj);
 891         wait_for_completion(&sbi->s_kobj_unregister);
 892         kfree(sbi->s_blockgroup_lock);
 893         kfree(sbi);
 894 }
 895
 896 static struct kmem_cache *ext4_inode_cachep;
 897
 898 /*
 899  * Called inside transaction, so use GFP_NOFS
 900  */
 901 static struct inode *ext4_alloc_inode(struct super_block *sb)
 902 {
 903         struct ext4_inode_info *ei;
 904
 905         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 906         if (!ei)
 907                 return NULL;
 908
 909         ei->vfs_inode.i_version = 1;
 910         ei->vfs_inode.i_data.writeback_index = 0;
 911         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 912         INIT_LIST_HEAD(&ei->i_prealloc_list);
 913         spin_lock_init(&ei->i_prealloc_lock);
 914         ei->i_reserved_data_blocks = 0;
 915         ei->i_reserved_meta_blocks = 0;
 916         ei->i_allocated_meta_blocks = 0;
 917         ei->i_da_metadata_calc_len = 0;
 918         ei->i_da_metadata_calc_last_lblock = 0;
 919         spin_lock_init(&(ei->i_block_reservation_lock));
 920 #ifdef CONFIG_QUOTA
 921         ei->i_reserved_quota = 0;
 922 #endif
 923         ei->jinode = NULL;
 924         INIT_LIST_HEAD(&ei->i_completed_io_list);
 925         spin_lock_init(&ei->i_completed_io_lock);
 926         ei->cur_aio_dio = NULL;
 927         ei->i_sync_tid = 0;
 928         ei->i_datasync_tid = 0;
 929         atomic_set(&ei->i_ioend_count, 0);
 930         atomic_set(&ei->i_aiodio_unwritten, 0);
 931
 932         return &ei->vfs_inode;
 933 }
 934
 935 static int ext4_drop_inode(struct inode *inode)
 936 {
 937         int drop = generic_drop_inode(inode);
 938
 939         trace_ext4_drop_inode(inode, drop);
 940         return drop;
 941 }
 942
 943 static void ext4_i_callback(struct rcu_head *head)
 944 {
 945         struct inode *inode = container_of(head, struct inode, i_rcu);
 946         INIT_LIST_HEAD(&inode->i_dentry);
 947         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 948 }
 949
 950 static void ext4_destroy_inode(struct inode *inode)
 951 {
 952         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 953                 ext4_msg(inode->i_sb, KERN_ERR,
 954                          "Inode %lu (%p): orphan list check failed!",
 955                          inode->i_ino, EXT4_I(inode));
 956                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 957                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 958                                 true);
 959                 dump_stack();
 960         }
 961         call_rcu(&inode->i_rcu, ext4_i_callback);
 962 }
 963
 964 static void init_once(void *foo)
 965 {
 966         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 967
 968         INIT_LIST_HEAD(&ei->i_orphan);
 969 #ifdef CONFIG_EXT4_FS_XATTR
 970         init_rwsem(&ei->xattr_sem);
 971 #endif
 972         init_rwsem(&ei->i_data_sem);
 973         inode_init_once(&ei->vfs_inode);
 974 }
 975
 976 static int init_inodecache(void)
 977 {
 978         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 979                                              sizeof(struct ext4_inode_info),
 980                                              0, (SLAB_RECLAIM_ACCOUNT|
 981                                                 SLAB_MEM_SPREAD),
 982                                              init_once);
 983         if (ext4_inode_cachep == NULL)
 984                 return -ENOMEM;
 985         return 0;
 986 }
 987
 988 static void destroy_inodecache(void)
 989 {
 990         kmem_cache_destroy(ext4_inode_cachep);
 991 }
 992
 993 void ext4_clear_inode(struct inode *inode)
 994 {
 995         invalidate_inode_buffers(inode);
 996         end_writeback(inode);
 997         dquot_drop(inode);
 998         ext4_discard_preallocations(inode);
 999         if (EXT4_I(inode)->jinode) {
1000                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1001                                                EXT4_I(inode)->jinode);
1002                 jbd2_free_inode(EXT4_I(inode)->jinode);
1003                 EXT4_I(inode)->jinode = NULL;
1004         }
1005 }
1006
1007 static inline void ext4_show_quota_options(struct seq_file *seq,
1008                                            struct super_block *sb)
1009 {
1010 #if defined(CONFIG_QUOTA)
1011         struct ext4_sb_info *sbi = EXT4_SB(sb);
1012
1013         if (sbi->s_jquota_fmt) {
1014                 char *fmtname = "";
1015
1016                 switch (sbi->s_jquota_fmt) {
1017                 case QFMT_VFS_OLD:
1018                         fmtname = "vfsold";
1019                         break;
1020                 case QFMT_VFS_V0:
1021                         fmtname = "vfsv0";
1022                         break;
1023                 case QFMT_VFS_V1:
1024                         fmtname = "vfsv1";
1025                         break;
1026                 }
1027                 seq_printf(seq, ",jqfmt=%s", fmtname);
1028         }
1029
1030         if (sbi->s_qf_names[USRQUOTA])
1031                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1032
1033         if (sbi->s_qf_names[GRPQUOTA])
1034                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1035
1036         if (test_opt(sb, USRQUOTA))
1037                 seq_puts(seq, ",usrquota");
1038
1039         if (test_opt(sb, GRPQUOTA))
1040                 seq_puts(seq, ",grpquota");
1041 #endif
1042 }
1043
1044 /*
1045  * Show an option if
1046  *  - it's set to a non-default value OR
1047  *  - if the per-sb default is different from the global default
1048  */
1049 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1050 {
1051         int def_errors;
1052         unsigned long def_mount_opts;
1053         struct super_block *sb = vfs->mnt_sb;
1054         struct ext4_sb_info *sbi = EXT4_SB(sb);
1055         struct ext4_super_block *es = sbi->s_es;
1056
1057         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1058         def_errors     = le16_to_cpu(es->s_errors);
1059
1060         if (sbi->s_sb_block != 1)
1061                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1062         if (test_opt(sb, MINIX_DF))
1063                 seq_puts(seq, ",minixdf");
1064         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1065                 seq_puts(seq, ",grpid");
1066         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1067                 seq_puts(seq, ",nogrpid");
1068         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1069             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1070                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1071         }
1072         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1073             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1074                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1075         }
1076         if (test_opt(sb, ERRORS_RO)) {
1077                 if (def_errors == EXT4_ERRORS_PANIC ||
1078                     def_errors == EXT4_ERRORS_CONTINUE) {
1079                         seq_puts(seq, ",errors=remount-ro");
1080                 }
1081         }
1082         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1083                 seq_puts(seq, ",errors=continue");
1084         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1085                 seq_puts(seq, ",errors=panic");
1086         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1087                 seq_puts(seq, ",nouid32");
1088         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1089                 seq_puts(seq, ",debug");
1090 #ifdef CONFIG_EXT4_FS_XATTR
1091         if (test_opt(sb, XATTR_USER))
1092                 seq_puts(seq, ",user_xattr");
1093         if (!test_opt(sb, XATTR_USER))
1094                 seq_puts(seq, ",nouser_xattr");
1095 #endif
1096 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1097         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1098                 seq_puts(seq, ",acl");
1099         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1100                 seq_puts(seq, ",noacl");
1101 #endif
1102         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1103                 seq_printf(seq, ",commit=%u",
1104                            (unsigned) (sbi->s_commit_interval / HZ));
1105         }
1106         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1107                 seq_printf(seq, ",min_batch_time=%u",
1108                            (unsigned) sbi->s_min_batch_time);
1109         }
1110         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1111                 seq_printf(seq, ",max_batch_time=%u",
1112                            (unsigned) sbi->s_max_batch_time);
1113         }
1114
1115         /*
1116          * We're changing the default of barrier mount option, so
1117          * let's always display its mount state so it's clear what its
1118          * status is.
1119          */
1120         seq_puts(seq, ",barrier=");
1121         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1122         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1123                 seq_puts(seq, ",journal_async_commit");
1124         else if (test_opt(sb, JOURNAL_CHECKSUM))
1125                 seq_puts(seq, ",journal_checksum");
1126         if (test_opt(sb, I_VERSION))
1127                 seq_puts(seq, ",i_version");
1128         if (!test_opt(sb, DELALLOC) &&
1129             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1130                 seq_puts(seq, ",nodelalloc");
1131
1132         if (!test_opt(sb, MBLK_IO_SUBMIT))
1133                 seq_puts(seq, ",nomblk_io_submit");
1134         if (sbi->s_stripe)
1135                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1136         /*
1137          * journal mode get enabled in different ways
1138          * So just print the value even if we didn't specify it
1139          */
1140         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1141                 seq_puts(seq, ",data=journal");
1142         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1143                 seq_puts(seq, ",data=ordered");
1144         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1145                 seq_puts(seq, ",data=writeback");
1146
1147         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1148                 seq_printf(seq, ",inode_readahead_blks=%u",
1149                            sbi->s_inode_readahead_blks);
1150
1151         if (test_opt(sb, DATA_ERR_ABORT))
1152                 seq_puts(seq, ",data_err=abort");
1153
1154         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1155                 seq_puts(seq, ",noauto_da_alloc");
1156
1157         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1158                 seq_puts(seq, ",discard");
1159
1160         if (test_opt(sb, NOLOAD))
1161                 seq_puts(seq, ",norecovery");
1162
1163         if (test_opt(sb, DIOREAD_NOLOCK))
1164                 seq_puts(seq, ",dioread_nolock");
1165
1166         if (test_opt(sb, BLOCK_VALIDITY) &&
1167             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1168                 seq_puts(seq, ",block_validity");
1169
1170         if (!test_opt(sb, INIT_INODE_TABLE))
1171                 seq_puts(seq, ",noinit_itable");
1172         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1173                 seq_printf(seq, ",init_itable=%u",
1174                            (unsigned) sbi->s_li_wait_mult);
1175
1176         ext4_show_quota_options(seq, sb);
1177
1178         return 0;
1179 }
1180
1181 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1182                                         u64 ino, u32 generation)
1183 {
1184         struct inode *inode;
1185
1186         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1187                 return ERR_PTR(-ESTALE);
1188         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1189                 return ERR_PTR(-ESTALE);
1190
1191         /* iget isn't really right if the inode is currently unallocated!!
1192          *
1193          * ext4_read_inode will return a bad_inode if the inode had been
1194          * deleted, so we should be safe.
1195          *
1196          * Currently we don't know the generation for parent directory, so
1197          * a generation of 0 means "accept any"
1198          */
1199         inode = ext4_iget_normal(sb, ino);
1200         if (IS_ERR(inode))
1201                 return ERR_CAST(inode);
1202         if (generation && inode->i_generation != generation) {
1203                 iput(inode);
1204                 return ERR_PTR(-ESTALE);
1205         }
1206
1207         return inode;
1208 }
1209
1210 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1211                                         int fh_len, int fh_type)
1212 {
1213         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1214                                     ext4_nfs_get_inode);
1215 }
1216
1217 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1218                                         int fh_len, int fh_type)
1219 {
1220         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1221                                     ext4_nfs_get_inode);
1222 }
1223
1224 /*
1225  * Try to release metadata pages (indirect blocks, directories) which are
1226  * mapped via the block device.  Since these pages could have journal heads
1227  * which would prevent try_to_free_buffers() from freeing them, we must use
1228  * jbd2 layer's try_to_free_buffers() function to release them.
1229  */
1230 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1231                                  gfp_t wait)
1232 {
1233         journal_t *journal = EXT4_SB(sb)->s_journal;
1234
1235         WARN_ON(PageChecked(page));
1236         if (!page_has_buffers(page))
1237                 return 0;
1238         if (journal)
1239                 return jbd2_journal_try_to_free_buffers(journal, page,
1240                                                         wait & ~__GFP_WAIT);
1241         return try_to_free_buffers(page);
1242 }
1243
1244 #ifdef CONFIG_QUOTA
1245 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1246 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1247
1248 static int ext4_write_dquot(struct dquot *dquot);
1249 static int ext4_acquire_dquot(struct dquot *dquot);
1250 static int ext4_release_dquot(struct dquot *dquot);
1251 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1252 static int ext4_write_info(struct super_block *sb, int type);
1253 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1254                          struct path *path);
1255 static int ext4_quota_off(struct super_block *sb, int type);
1256 static int ext4_quota_on_mount(struct super_block *sb, int type);
1257 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1258                                size_t len, loff_t off);
1259 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1260                                 const char *data, size_t len, loff_t off);
1261
1262 static const struct dquot_operations ext4_quota_operations = {
1263         .get_reserved_space = ext4_get_reserved_space,
1264         .write_dquot    = ext4_write_dquot,
1265         .acquire_dquot  = ext4_acquire_dquot,
1266         .release_dquot  = ext4_release_dquot,
1267         .mark_dirty     = ext4_mark_dquot_dirty,
1268         .write_info     = ext4_write_info,
1269         .alloc_dquot    = dquot_alloc,
1270         .destroy_dquot  = dquot_destroy,
1271 };
1272
1273 static const struct quotactl_ops ext4_qctl_operations = {
1274         .quota_on       = ext4_quota_on,
1275         .quota_off      = ext4_quota_off,
1276         .quota_sync     = dquot_quota_sync,
1277         .get_info       = dquot_get_dqinfo,
1278         .set_info       = dquot_set_dqinfo,
1279         .get_dqblk      = dquot_get_dqblk,
1280         .set_dqblk      = dquot_set_dqblk
1281 };
1282 #endif
1283
1284 static const struct super_operations ext4_sops = {
1285         .alloc_inode    = ext4_alloc_inode,
1286         .destroy_inode  = ext4_destroy_inode,
1287         .write_inode    = ext4_write_inode,
1288         .dirty_inode    = ext4_dirty_inode,
1289         .drop_inode     = ext4_drop_inode,
1290         .evict_inode    = ext4_evict_inode,
1291         .put_super      = ext4_put_super,
1292         .sync_fs        = ext4_sync_fs,
1293         .freeze_fs      = ext4_freeze,
1294         .unfreeze_fs    = ext4_unfreeze,
1295         .statfs         = ext4_statfs,
1296         .remount_fs     = ext4_remount,
1297         .show_options   = ext4_show_options,
1298 #ifdef CONFIG_QUOTA
1299         .quota_read     = ext4_quota_read,
1300         .quota_write    = ext4_quota_write,
1301 #endif
1302         .bdev_try_to_free_page = bdev_try_to_free_page,
1303 };
1304
1305 static const struct super_operations ext4_nojournal_sops = {
1306         .alloc_inode    = ext4_alloc_inode,
1307         .destroy_inode  = ext4_destroy_inode,
1308         .write_inode    = ext4_write_inode,
1309         .dirty_inode    = ext4_dirty_inode,
1310         .drop_inode     = ext4_drop_inode,
1311         .evict_inode    = ext4_evict_inode,
1312         .write_super    = ext4_write_super,
1313         .put_super      = ext4_put_super,
1314         .statfs         = ext4_statfs,
1315         .remount_fs     = ext4_remount,
1316         .show_options   = ext4_show_options,
1317 #ifdef CONFIG_QUOTA
1318         .quota_read     = ext4_quota_read,
1319         .quota_write    = ext4_quota_write,
1320 #endif
1321         .bdev_try_to_free_page = bdev_try_to_free_page,
1322 };
1323
1324 static const struct export_operations ext4_export_ops = {
1325         .fh_to_dentry = ext4_fh_to_dentry,
1326         .fh_to_parent = ext4_fh_to_parent,
1327         .get_parent = ext4_get_parent,
1328 };
1329
1330 enum {
1331         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1332         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1333         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1334         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1335         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1336         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1337         Opt_journal_update, Opt_journal_dev,
1338         Opt_journal_checksum, Opt_journal_async_commit,
1339         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1340         Opt_data_err_abort, Opt_data_err_ignore,
1341         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1342         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1343         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1344         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1345         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1346         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1347         Opt_inode_readahead_blks, Opt_journal_ioprio,
1348         Opt_dioread_nolock, Opt_dioread_lock,
1349         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1350 };
1351
1352 static const match_table_t tokens = {
1353         {Opt_bsd_df, "bsddf"},
1354         {Opt_minix_df, "minixdf"},
1355         {Opt_grpid, "grpid"},
1356         {Opt_grpid, "bsdgroups"},
1357         {Opt_nogrpid, "nogrpid"},
1358         {Opt_nogrpid, "sysvgroups"},
1359         {Opt_resgid, "resgid=%u"},
1360         {Opt_resuid, "resuid=%u"},
1361         {Opt_sb, "sb=%u"},
1362         {Opt_err_cont, "errors=continue"},
1363         {Opt_err_panic, "errors=panic"},
1364         {Opt_err_ro, "errors=remount-ro"},
1365         {Opt_nouid32, "nouid32"},
1366         {Opt_debug, "debug"},
1367         {Opt_oldalloc, "oldalloc"},
1368         {Opt_orlov, "orlov"},
1369         {Opt_user_xattr, "user_xattr"},
1370         {Opt_nouser_xattr, "nouser_xattr"},
1371         {Opt_acl, "acl"},
1372         {Opt_noacl, "noacl"},
1373         {Opt_noload, "noload"},
1374         {Opt_noload, "norecovery"},
1375         {Opt_nobh, "nobh"},
1376         {Opt_bh, "bh"},
1377         {Opt_commit, "commit=%u"},
1378         {Opt_min_batch_time, "min_batch_time=%u"},
1379         {Opt_max_batch_time, "max_batch_time=%u"},
1380         {Opt_journal_update, "journal=update"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_checksum, "journal_checksum"},
1383         {Opt_journal_async_commit, "journal_async_commit"},
1384         {Opt_abort, "abort"},
1385         {Opt_data_journal, "data=journal"},
1386         {Opt_data_ordered, "data=ordered"},
1387         {Opt_data_writeback, "data=writeback"},
1388         {Opt_data_err_abort, "data_err=abort"},
1389         {Opt_data_err_ignore, "data_err=ignore"},
1390         {Opt_offusrjquota, "usrjquota="},
1391         {Opt_usrjquota, "usrjquota=%s"},
1392         {Opt_offgrpjquota, "grpjquota="},
1393         {Opt_grpjquota, "grpjquota=%s"},
1394         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1395         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1396         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1397         {Opt_grpquota, "grpquota"},
1398         {Opt_noquota, "noquota"},
1399         {Opt_quota, "quota"},
1400         {Opt_usrquota, "usrquota"},
1401         {Opt_barrier, "barrier=%u"},
1402         {Opt_barrier, "barrier"},
1403         {Opt_nobarrier, "nobarrier"},
1404         {Opt_i_version, "i_version"},
1405         {Opt_stripe, "stripe=%u"},
1406         {Opt_resize, "resize"},
1407         {Opt_delalloc, "delalloc"},
1408         {Opt_nodelalloc, "nodelalloc"},
1409         {Opt_mblk_io_submit, "mblk_io_submit"},
1410         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1411         {Opt_block_validity, "block_validity"},
1412         {Opt_noblock_validity, "noblock_validity"},
1413         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1414         {Opt_journal_ioprio, "journal_ioprio=%u"},
1415         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1416         {Opt_auto_da_alloc, "auto_da_alloc"},
1417         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1418         {Opt_dioread_nolock, "dioread_nolock"},
1419         {Opt_dioread_lock, "dioread_lock"},
1420         {Opt_discard, "discard"},
1421         {Opt_nodiscard, "nodiscard"},
1422         {Opt_init_itable, "init_itable=%u"},
1423         {Opt_init_itable, "init_itable"},
1424         {Opt_noinit_itable, "noinit_itable"},
1425         {Opt_err, NULL},
1426 };
1427
1428 static ext4_fsblk_t get_sb_block(void **data)
1429 {
1430         ext4_fsblk_t    sb_block;
1431         char            *options = (char *) *data;
1432
1433         if (!options || strncmp(options, "sb=", 3) != 0)
1434                 return 1;       /* Default location */
1435
1436         options += 3;
1437         /* TODO: use simple_strtoll with >32bit ext4 */
1438         sb_block = simple_strtoul(options, &options, 0);
1439         if (*options && *options != ',') {
1440                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1441                        (char *) *data);
1442                 return 1;
1443         }
1444         if (*options == ',')
1445                 options++;
1446         *data = (void *) options;
1447
1448         return sb_block;
1449 }
1450
1451 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1452 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1453         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1454
1455 #ifdef CONFIG_QUOTA
1456 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1457 {
1458         struct ext4_sb_info *sbi = EXT4_SB(sb);
1459         char *qname;
1460
1461         if (sb_any_quota_loaded(sb) &&
1462                 !sbi->s_qf_names[qtype]) {
1463                 ext4_msg(sb, KERN_ERR,
1464                         "Cannot change journaled "
1465                         "quota options when quota turned on");
1466                 return 0;
1467         }
1468         qname = match_strdup(args);
1469         if (!qname) {
1470                 ext4_msg(sb, KERN_ERR,
1471                         "Not enough memory for storing quotafile name");
1472                 return 0;
1473         }
1474         if (sbi->s_qf_names[qtype] &&
1475                 strcmp(sbi->s_qf_names[qtype], qname)) {
1476                 ext4_msg(sb, KERN_ERR,
1477                         "%s quota file already specified", QTYPE2NAME(qtype));
1478                 kfree(qname);
1479                 return 0;
1480         }
1481         sbi->s_qf_names[qtype] = qname;
1482         if (strchr(sbi->s_qf_names[qtype], '/')) {
1483                 ext4_msg(sb, KERN_ERR,
1484                         "quotafile must be on filesystem root");
1485                 kfree(sbi->s_qf_names[qtype]);
1486                 sbi->s_qf_names[qtype] = NULL;
1487                 return 0;
1488         }
1489         set_opt(sb, QUOTA);
1490         return 1;
1491 }
1492
1493 static int clear_qf_name(struct super_block *sb, int qtype)
1494 {
1495
1496         struct ext4_sb_info *sbi = EXT4_SB(sb);
1497
1498         if (sb_any_quota_loaded(sb) &&
1499                 sbi->s_qf_names[qtype]) {
1500                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1501                         " when quota turned on");
1502                 return 0;
1503         }
1504         /*
1505          * The space will be released later when all options are confirmed
1506          * to be correct
1507          */
1508         sbi->s_qf_names[qtype] = NULL;
1509         return 1;
1510 }
1511 #endif
1512
1513 static int parse_options(char *options, struct super_block *sb,
1514                          unsigned long *journal_devnum,
1515                          unsigned int *journal_ioprio,
1516                          ext4_fsblk_t *n_blocks_count, int is_remount)
1517 {
1518         struct ext4_sb_info *sbi = EXT4_SB(sb);
1519         char *p;
1520         substring_t args[MAX_OPT_ARGS];
1521         int data_opt = 0;
1522         int option;
1523 #ifdef CONFIG_QUOTA
1524         int qfmt;
1525 #endif
1526
1527         if (!options)
1528                 return 1;
1529
1530         while ((p = strsep(&options, ",")) != NULL) {
1531                 int token;
1532                 if (!*p)
1533                         continue;
1534
1535                 /*
1536                  * Initialize args struct so we know whether arg was
1537                  * found; some options take optional arguments.
1538                  */
1539                 args[0].to = args[0].from = NULL;
1540                 token = match_token(p, tokens, args);
1541                 switch (token) {
1542                 case Opt_bsd_df:
1543                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1544                         clear_opt(sb, MINIX_DF);
1545                         break;
1546                 case Opt_minix_df:
1547                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1548                         set_opt(sb, MINIX_DF);
1549
1550                         break;
1551                 case Opt_grpid:
1552                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1553                         set_opt(sb, GRPID);
1554
1555                         break;
1556                 case Opt_nogrpid:
1557                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1558                         clear_opt(sb, GRPID);
1559
1560                         break;
1561                 case Opt_resuid:
1562                         if (match_int(&args[0], &option))
1563                                 return 0;
1564                         sbi->s_resuid = option;
1565                         break;
1566                 case Opt_resgid:
1567                         if (match_int(&args[0], &option))
1568                                 return 0;
1569                         sbi->s_resgid = option;
1570                         break;
1571                 case Opt_sb:
1572                         /* handled by get_sb_block() instead of here */
1573                         /* *sb_block = match_int(&args[0]); */
1574                         break;
1575                 case Opt_err_panic:
1576                         clear_opt(sb, ERRORS_CONT);
1577                         clear_opt(sb, ERRORS_RO);
1578                         set_opt(sb, ERRORS_PANIC);
1579                         break;
1580                 case Opt_err_ro:
1581                         clear_opt(sb, ERRORS_CONT);
1582                         clear_opt(sb, ERRORS_PANIC);
1583                         set_opt(sb, ERRORS_RO);
1584                         break;
1585                 case Opt_err_cont:
1586                         clear_opt(sb, ERRORS_RO);
1587                         clear_opt(sb, ERRORS_PANIC);
1588                         set_opt(sb, ERRORS_CONT);
1589                         break;
1590                 case Opt_nouid32:
1591                         set_opt(sb, NO_UID32);
1592                         break;
1593                 case Opt_debug:
1594                         set_opt(sb, DEBUG);
1595                         break;
1596                 case Opt_oldalloc:
1597                         ext4_msg(sb, KERN_WARNING,
1598                                  "Ignoring deprecated oldalloc option");
1599                         break;
1600                 case Opt_orlov:
1601                         ext4_msg(sb, KERN_WARNING,
1602                                  "Ignoring deprecated orlov option");
1603                         break;
1604 #ifdef CONFIG_EXT4_FS_XATTR
1605                 case Opt_user_xattr:
1606                         set_opt(sb, XATTR_USER);
1607                         break;
1608                 case Opt_nouser_xattr:
1609                         clear_opt(sb, XATTR_USER);
1610                         break;
1611 #else
1612                 case Opt_user_xattr:
1613                 case Opt_nouser_xattr:
1614                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1615                         break;
1616 #endif
1617 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1618                 case Opt_acl:
1619                         set_opt(sb, POSIX_ACL);
1620                         break;
1621                 case Opt_noacl:
1622                         clear_opt(sb, POSIX_ACL);
1623                         break;
1624 #else
1625                 case Opt_acl:
1626                 case Opt_noacl:
1627                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1628                         break;
1629 #endif
1630                 case Opt_journal_update:
1631                         /* @@@ FIXME */
1632                         /* Eventually we will want to be able to create
1633                            a journal file here.  For now, only allow the
1634                            user to specify an existing inode to be the
1635                            journal file. */
1636                         if (is_remount) {
1637                                 ext4_msg(sb, KERN_ERR,
1638                                          "Cannot specify journal on remount");
1639                                 return 0;
1640                         }
1641                         set_opt(sb, UPDATE_JOURNAL);
1642                         break;
1643                 case Opt_journal_dev:
1644                         if (is_remount) {
1645                                 ext4_msg(sb, KERN_ERR,
1646                                         "Cannot specify journal on remount");
1647                                 return 0;
1648                         }
1649                         if (match_int(&args[0], &option))
1650                                 return 0;
1651                         *journal_devnum = option;
1652                         break;
1653                 case Opt_journal_checksum:
1654                         set_opt(sb, JOURNAL_CHECKSUM);
1655                         break;
1656                 case Opt_journal_async_commit:
1657                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1658                         set_opt(sb, JOURNAL_CHECKSUM);
1659                         break;
1660                 case Opt_noload:
1661                         set_opt(sb, NOLOAD);
1662                         break;
1663                 case Opt_commit:
1664                         if (match_int(&args[0], &option))
1665                                 return 0;
1666                         if (option < 0)
1667                                 return 0;
1668                         if (option == 0)
1669                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1670                         sbi->s_commit_interval = HZ * option;
1671                         break;
1672                 case Opt_max_batch_time:
1673                         if (match_int(&args[0], &option))
1674                                 return 0;
1675                         if (option < 0)
1676                                 return 0;
1677                         sbi->s_max_batch_time = option;
1678                         break;
1679                 case Opt_min_batch_time:
1680                         if (match_int(&args[0], &option))
1681                                 return 0;
1682                         if (option < 0)
1683                                 return 0;
1684                         sbi->s_min_batch_time = option;
1685                         break;
1686                 case Opt_data_journal:
1687                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1688                         goto datacheck;
1689                 case Opt_data_ordered:
1690                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1691                         goto datacheck;
1692                 case Opt_data_writeback:
1693                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1694                 datacheck:
1695                         if (is_remount) {
1696                                 if (!sbi->s_journal)
1697                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1698                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1699                                         ext4_msg(sb, KERN_ERR,
1700                                                 "Cannot change data mode on remount");
1701                                         return 0;
1702                                 }
1703                         } else {
1704                                 clear_opt(sb, DATA_FLAGS);
1705                                 sbi->s_mount_opt |= data_opt;
1706                         }
1707                         break;
1708                 case Opt_data_err_abort:
1709                         set_opt(sb, DATA_ERR_ABORT);
1710                         break;
1711                 case Opt_data_err_ignore:
1712                         clear_opt(sb, DATA_ERR_ABORT);
1713                         break;
1714 #ifdef CONFIG_QUOTA
1715                 case Opt_usrjquota:
1716                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1717                                 return 0;
1718                         break;
1719                 case Opt_grpjquota:
1720                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1721                                 return 0;
1722                         break;
1723                 case Opt_offusrjquota:
1724                         if (!clear_qf_name(sb, USRQUOTA))
1725                                 return 0;
1726                         break;
1727                 case Opt_offgrpjquota:
1728                         if (!clear_qf_name(sb, GRPQUOTA))
1729                                 return 0;
1730                         break;
1731
1732                 case Opt_jqfmt_vfsold:
1733                         qfmt = QFMT_VFS_OLD;
1734                         goto set_qf_format;
1735                 case Opt_jqfmt_vfsv0:
1736                         qfmt = QFMT_VFS_V0;
1737                         goto set_qf_format;
1738                 case Opt_jqfmt_vfsv1:
1739                         qfmt = QFMT_VFS_V1;
1740 set_qf_format:
1741                         if (sb_any_quota_loaded(sb) &&
1742                             sbi->s_jquota_fmt != qfmt) {
1743                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1744                                         "journaled quota options when "
1745                                         "quota turned on");
1746                                 return 0;
1747                         }
1748                         sbi->s_jquota_fmt = qfmt;
1749                         break;
1750                 case Opt_quota:
1751                 case Opt_usrquota:
1752                         set_opt(sb, QUOTA);
1753                         set_opt(sb, USRQUOTA);
1754                         break;
1755                 case Opt_grpquota:
1756                         set_opt(sb, QUOTA);
1757                         set_opt(sb, GRPQUOTA);
1758                         break;
1759                 case Opt_noquota:
1760                         if (sb_any_quota_loaded(sb)) {
1761                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1762                                         "options when quota turned on");
1763                                 return 0;
1764                         }
1765                         clear_opt(sb, QUOTA);
1766                         clear_opt(sb, USRQUOTA);
1767                         clear_opt(sb, GRPQUOTA);
1768                         break;
1769 #else
1770                 case Opt_quota:
1771                 case Opt_usrquota:
1772                 case Opt_grpquota:
1773                         ext4_msg(sb, KERN_ERR,
1774                                 "quota options not supported");
1775                         break;
1776                 case Opt_usrjquota:
1777                 case Opt_grpjquota:
1778                 case Opt_offusrjquota:
1779                 case Opt_offgrpjquota:
1780                 case Opt_jqfmt_vfsold:
1781                 case Opt_jqfmt_vfsv0:
1782                 case Opt_jqfmt_vfsv1:
1783                         ext4_msg(sb, KERN_ERR,
1784                                 "journaled quota options not supported");
1785                         break;
1786                 case Opt_noquota:
1787                         break;
1788 #endif
1789                 case Opt_abort:
1790                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1791                         break;
1792                 case Opt_nobarrier:
1793                         clear_opt(sb, BARRIER);
1794                         break;
1795                 case Opt_barrier:
1796                         if (args[0].from) {
1797                                 if (match_int(&args[0], &option))
1798                                         return 0;
1799                         } else
1800                                 option = 1;     /* No argument, default to 1 */
1801                         if (option)
1802                                 set_opt(sb, BARRIER);
1803                         else
1804                                 clear_opt(sb, BARRIER);
1805                         break;
1806                 case Opt_ignore:
1807                         break;
1808                 case Opt_resize:
1809                         if (!is_remount) {
1810                                 ext4_msg(sb, KERN_ERR,
1811                                         "resize option only available "
1812                                         "for remount");
1813                                 return 0;
1814                         }
1815                         if (match_int(&args[0], &option) != 0)
1816                                 return 0;
1817                         *n_blocks_count = option;
1818                         break;
1819                 case Opt_nobh:
1820                         ext4_msg(sb, KERN_WARNING,
1821                                  "Ignoring deprecated nobh option");
1822                         break;
1823                 case Opt_bh:
1824                         ext4_msg(sb, KERN_WARNING,
1825                                  "Ignoring deprecated bh option");
1826                         break;
1827                 case Opt_i_version:
1828                         set_opt(sb, I_VERSION);
1829                         sb->s_flags |= MS_I_VERSION;
1830                         break;
1831                 case Opt_nodelalloc:
1832                         clear_opt(sb, DELALLOC);
1833                         clear_opt2(sb, EXPLICIT_DELALLOC);
1834                         break;
1835                 case Opt_mblk_io_submit:
1836                         set_opt(sb, MBLK_IO_SUBMIT);
1837                         break;
1838                 case Opt_nomblk_io_submit:
1839                         clear_opt(sb, MBLK_IO_SUBMIT);
1840                         break;
1841                 case Opt_stripe:
1842                         if (match_int(&args[0], &option))
1843                                 return 0;
1844                         if (option < 0)
1845                                 return 0;
1846                         sbi->s_stripe = option;
1847                         break;
1848                 case Opt_delalloc:
1849                         set_opt(sb, DELALLOC);
1850                         set_opt2(sb, EXPLICIT_DELALLOC);
1851                         break;
1852                 case Opt_block_validity:
1853                         set_opt(sb, BLOCK_VALIDITY);
1854                         break;
1855                 case Opt_noblock_validity:
1856                         clear_opt(sb, BLOCK_VALIDITY);
1857                         break;
1858                 case Opt_inode_readahead_blks:
1859                         if (match_int(&args[0], &option))
1860                                 return 0;
1861                         if (option < 0 || option > (1 << 30))
1862                                 return 0;
1863                         if (option && !is_power_of_2(option)) {
1864                                 ext4_msg(sb, KERN_ERR,
1865                                          "EXT4-fs: inode_readahead_blks"
1866                                          " must be a power of 2");
1867                                 return 0;
1868                         }
1869                         sbi->s_inode_readahead_blks = option;
1870                         break;
1871                 case Opt_journal_ioprio:
1872                         if (match_int(&args[0], &option))
1873                                 return 0;
1874                         if (option < 0 || option > 7)
1875                                 break;
1876                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1877                                                             option);
1878                         break;
1879                 case Opt_noauto_da_alloc:
1880                         set_opt(sb, NO_AUTO_DA_ALLOC);
1881                         break;
1882                 case Opt_auto_da_alloc:
1883                         if (args[0].from) {
1884                                 if (match_int(&args[0], &option))
1885                                         return 0;
1886                         } else
1887                                 option = 1;     /* No argument, default to 1 */
1888                         if (option)
1889                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1890                         else
1891                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1892                         break;
1893                 case Opt_discard:
1894                         set_opt(sb, DISCARD);
1895                         break;
1896                 case Opt_nodiscard:
1897                         clear_opt(sb, DISCARD);
1898                         break;
1899                 case Opt_dioread_nolock:
1900                         set_opt(sb, DIOREAD_NOLOCK);
1901                         break;
1902                 case Opt_dioread_lock:
1903                         clear_opt(sb, DIOREAD_NOLOCK);
1904                         break;
1905                 case Opt_init_itable:
1906                         set_opt(sb, INIT_INODE_TABLE);
1907                         if (args[0].from) {
1908                                 if (match_int(&args[0], &option))
1909                                         return 0;
1910                         } else
1911                                 option = EXT4_DEF_LI_WAIT_MULT;
1912                         if (option < 0)
1913                                 return 0;
1914                         sbi->s_li_wait_mult = option;
1915                         break;
1916                 case Opt_noinit_itable:
1917                         clear_opt(sb, INIT_INODE_TABLE);
1918                         break;
1919                 default:
1920                         ext4_msg(sb, KERN_ERR,
1921                                "Unrecognized mount option \"%s\" "
1922                                "or missing value", p);
1923                         return 0;
1924                 }
1925         }
1926 #ifdef CONFIG_QUOTA
1927         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1928                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1929                         clear_opt(sb, USRQUOTA);
1930
1931                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1932                         clear_opt(sb, GRPQUOTA);
1933
1934                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1935                         ext4_msg(sb, KERN_ERR, "old and new quota "
1936                                         "format mixing");
1937                         return 0;
1938                 }
1939
1940                 if (!sbi->s_jquota_fmt) {
1941                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1942                                         "not specified");
1943                         return 0;
1944                 }
1945         }
1946 #endif
1947         if (test_opt(sb, DIOREAD_NOLOCK)) {
1948                 int blocksize =
1949                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1950
1951                 if (blocksize < PAGE_CACHE_SIZE) {
1952                         ext4_msg(sb, KERN_ERR, "can't mount with "
1953                                  "dioread_nolock if block size != PAGE_SIZE");
1954                         return 0;
1955                 }
1956         }
1957         return 1;
1958 }
1959
1960 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1961                             int read_only)
1962 {
1963         struct ext4_sb_info *sbi = EXT4_SB(sb);
1964         int res = 0;
1965
1966         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1967                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1968                          "forcing read-only mode");
1969                 res = MS_RDONLY;
1970         }
1971         if (read_only)
1972                 goto done;
1973         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1974                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1975                          "running e2fsck is recommended");
1976         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1977                 ext4_msg(sb, KERN_WARNING,
1978                          "warning: mounting fs with errors, "
1979                          "running e2fsck is recommended");
1980         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1981                  le16_to_cpu(es->s_mnt_count) >=
1982                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1983                 ext4_msg(sb, KERN_WARNING,
1984                          "warning: maximal mount count reached, "
1985                          "running e2fsck is recommended");
1986         else if (le32_to_cpu(es->s_checkinterval) &&
1987                 (le32_to_cpu(es->s_lastcheck) +
1988                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1989                 ext4_msg(sb, KERN_WARNING,
1990                          "warning: checktime reached, "
1991                          "running e2fsck is recommended");
1992         if (!sbi->s_journal)
1993                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1994         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1995                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1996         le16_add_cpu(&es->s_mnt_count, 1);
1997         es->s_mtime = cpu_to_le32(get_seconds());
1998         ext4_update_dynamic_rev(sb);
1999         if (sbi->s_journal)
2000                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2001
2002         ext4_commit_super(sb, 1);
2003 done:
2004         if (test_opt(sb, DEBUG))
2005                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2006                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2007                         sb->s_blocksize,
2008                         sbi->s_groups_count,
2009                         EXT4_BLOCKS_PER_GROUP(sb),
2010                         EXT4_INODES_PER_GROUP(sb),
2011                         sbi->s_mount_opt, sbi->s_mount_opt2);
2012
2013         cleancache_init_fs(sb);
2014         return res;
2015 }
2016
2017 static int ext4_fill_flex_info(struct super_block *sb)
2018 {
2019         struct ext4_sb_info *sbi = EXT4_SB(sb);
2020         struct ext4_group_desc *gdp = NULL;
2021         ext4_group_t flex_group_count;
2022         ext4_group_t flex_group;
2023         unsigned int groups_per_flex = 0;
2024         size_t size;
2025         int i;
2026
2027         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2028         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2029                 sbi->s_log_groups_per_flex = 0;
2030                 return 1;
2031         }
2032         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2033
2034         /* We allocate both existing and potentially added groups */
2035         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2036                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2037                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2038         size = flex_group_count * sizeof(struct flex_groups);
2039         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2040         if (sbi->s_flex_groups == NULL) {
2041                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2042                          flex_group_count);
2043                 goto failed;
2044         }
2045
2046         for (i = 0; i < sbi->s_groups_count; i++) {
2047                 gdp = ext4_get_group_desc(sb, i, NULL);
2048
2049                 flex_group = ext4_flex_group(sbi, i);
2050                 atomic_add(ext4_free_inodes_count(sb, gdp),
2051                            &sbi->s_flex_groups[flex_group].free_inodes);
2052                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2053                              &sbi->s_flex_groups[flex_group].free_clusters);
2054                 atomic_add(ext4_used_dirs_count(sb, gdp),
2055                            &sbi->s_flex_groups[flex_group].used_dirs);
2056         }
2057
2058         return 1;
2059 failed:
2060         return 0;
2061 }
2062
2063 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2064                             struct ext4_group_desc *gdp)
2065 {
2066         __u16 crc = 0;
2067
2068         if (sbi->s_es->s_feature_ro_compat &
2069             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2070                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2071                 __le32 le_group = cpu_to_le32(block_group);
2072
2073                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2074                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2075                 crc = crc16(crc, (__u8 *)gdp, offset);
2076                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2077                 /* for checksum of struct ext4_group_desc do the rest...*/
2078                 if ((sbi->s_es->s_feature_incompat &
2079                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2080                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2081                         crc = crc16(crc, (__u8 *)gdp + offset,
2082                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2083                                         offset);
2084         }
2085
2086         return cpu_to_le16(crc);
2087 }
2088
2089 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2090                                 struct ext4_group_desc *gdp)
2091 {
2092         if ((sbi->s_es->s_feature_ro_compat &
2093              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2094             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2095                 return 0;
2096
2097         return 1;
2098 }
2099
2100 /* Called at mount-time, super-block is locked */
2101 static int ext4_check_descriptors(struct super_block *sb,
2102                                   ext4_fsblk_t sb_block,
2103                                   ext4_group_t *first_not_zeroed)
2104 {
2105         struct ext4_sb_info *sbi = EXT4_SB(sb);
2106         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2107         ext4_fsblk_t last_block;
2108         ext4_fsblk_t block_bitmap;
2109         ext4_fsblk_t inode_bitmap;
2110         ext4_fsblk_t inode_table;
2111         int flexbg_flag = 0;
2112         ext4_group_t i, grp = sbi->s_groups_count;
2113
2114         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2115                 flexbg_flag = 1;
2116
2117         ext4_debug("Checking group descriptors");
2118
2119         for (i = 0; i < sbi->s_groups_count; i++) {
2120                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2121
2122                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2123                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2124                 else
2125                         last_block = first_block +
2126                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2127
2128                 if ((grp == sbi->s_groups_count) &&
2129                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2130                         grp = i;
2131
2132                 block_bitmap = ext4_block_bitmap(sb, gdp);
2133                 if (block_bitmap == sb_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                  "Block bitmap for group %u overlaps "
2136                                  "superblock", i);
2137                 }
2138                 if (block_bitmap < first_block || block_bitmap > last_block) {
2139                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2140                                "Block bitmap for group %u not in group "
2141                                "(block %llu)!", i, block_bitmap);
2142                         return 0;
2143                 }
2144                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2145                 if (inode_bitmap == sb_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                  "Inode bitmap for group %u overlaps "
2148                                  "superblock", i);
2149                 }
2150                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2151                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2152                                "Inode bitmap for group %u not in group "
2153                                "(block %llu)!", i, inode_bitmap);
2154                         return 0;
2155                 }
2156                 inode_table = ext4_inode_table(sb, gdp);
2157                 if (inode_table == sb_block) {
2158                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2159                                  "Inode table for group %u overlaps "
2160                                  "superblock", i);
2161                 }
2162                 if (inode_table < first_block ||
2163                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2164                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2165                                "Inode table for group %u not in group "
2166                                "(block %llu)!", i, inode_table);
2167                         return 0;
2168                 }
2169                 ext4_lock_group(sb, i);
2170                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2171                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2172                                  "Checksum for group %u failed (%u!=%u)",
2173                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2174                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2175                         if (!(sb->s_flags & MS_RDONLY)) {
2176                                 ext4_unlock_group(sb, i);
2177                                 return 0;
2178                         }
2179                 }
2180                 ext4_unlock_group(sb, i);
2181                 if (!flexbg_flag)
2182                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2183         }
2184         if (NULL != first_not_zeroed)
2185                 *first_not_zeroed = grp;
2186
2187         ext4_free_blocks_count_set(sbi->s_es,
2188                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2189         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2190         return 1;
2191 }
2192
2193 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2194  * the superblock) which were deleted from all directories, but held open by
2195  * a process at the time of a crash.  We walk the list and try to delete these
2196  * inodes at recovery time (only with a read-write filesystem).
2197  *
2198  * In order to keep the orphan inode chain consistent during traversal (in
2199  * case of crash during recovery), we link each inode into the superblock
2200  * orphan list_head and handle it the same way as an inode deletion during
2201  * normal operation (which journals the operations for us).
2202  *
2203  * We only do an iget() and an iput() on each inode, which is very safe if we
2204  * accidentally point at an in-use or already deleted inode.  The worst that
2205  * can happen in this case is that we get a "bit already cleared" message from
2206  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2207  * e2fsck was run on this filesystem, and it must have already done the orphan
2208  * inode cleanup for us, so we can safely abort without any further action.
2209  */
2210 static void ext4_orphan_cleanup(struct super_block *sb,
2211                                 struct ext4_super_block *es)
2212 {
2213         unsigned int s_flags = sb->s_flags;
2214         int nr_orphans = 0, nr_truncates = 0;
2215 #ifdef CONFIG_QUOTA
2216         int i;
2217 #endif
2218         if (!es->s_last_orphan) {
2219                 jbd_debug(4, "no orphan inodes to clean up\n");
2220                 return;
2221         }
2222
2223         if (bdev_read_only(sb->s_bdev)) {
2224                 ext4_msg(sb, KERN_ERR, "write access "
2225                         "unavailable, skipping orphan cleanup");
2226                 return;
2227         }
2228
2229         /* Check if feature set would not allow a r/w mount */
2230         if (!ext4_feature_set_ok(sb, 0)) {
2231                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2232                          "unknown ROCOMPAT features");
2233                 return;
2234         }
2235
2236         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2237                 if (es->s_last_orphan)
2238                         jbd_debug(1, "Errors on filesystem, "
2239                                   "clearing orphan list.\n");
2240                 es->s_last_orphan = 0;
2241                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2242                 return;
2243         }
2244
2245         if (s_flags & MS_RDONLY) {
2246                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2247                 sb->s_flags &= ~MS_RDONLY;
2248         }
2249 #ifdef CONFIG_QUOTA
2250         /* Needed for iput() to work correctly and not trash data */
2251         sb->s_flags |= MS_ACTIVE;
2252         /* Turn on quotas so that they are updated correctly */
2253         for (i = 0; i < MAXQUOTAS; i++) {
2254                 if (EXT4_SB(sb)->s_qf_names[i]) {
2255                         int ret = ext4_quota_on_mount(sb, i);
2256                         if (ret < 0)
2257                                 ext4_msg(sb, KERN_ERR,
2258                                         "Cannot turn on journaled "
2259                                         "quota: error %d", ret);
2260                 }
2261         }
2262 #endif
2263
2264         while (es->s_last_orphan) {
2265                 struct inode *inode;
2266
2267                 /*
2268                  * We may have encountered an error during cleanup; if
2269                  * so, skip the rest.
2270                  */
2271                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2272                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2273                         es->s_last_orphan = 0;
2274                         break;
2275                 }
2276
2277                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2278                 if (IS_ERR(inode)) {
2279                         es->s_last_orphan = 0;
2280                         break;
2281                 }
2282
2283                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2284                 dquot_initialize(inode);
2285                 if (inode->i_nlink) {
2286                         ext4_msg(sb, KERN_DEBUG,
2287                                 "%s: truncating inode %lu to %lld bytes",
2288                                 __func__, inode->i_ino, inode->i_size);
2289                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2290                                   inode->i_ino, inode->i_size);
2291                         mutex_lock(&inode->i_mutex);
2292                         ext4_truncate(inode);
2293                         mutex_unlock(&inode->i_mutex);
2294                         nr_truncates++;
2295                 } else {
2296                         ext4_msg(sb, KERN_DEBUG,
2297                                 "%s: deleting unreferenced inode %lu",
2298                                 __func__, inode->i_ino);
2299                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2300                                   inode->i_ino);
2301                         nr_orphans++;
2302                 }
2303                 iput(inode);  /* The delete magic happens here! */
2304         }
2305
2306 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2307
2308         if (nr_orphans)
2309                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2310                        PLURAL(nr_orphans));
2311         if (nr_truncates)
2312                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2313                        PLURAL(nr_truncates));
2314 #ifdef CONFIG_QUOTA
2315         /* Turn quotas off */
2316         for (i = 0; i < MAXQUOTAS; i++) {
2317                 if (sb_dqopt(sb)->files[i])
2318                         dquot_quota_off(sb, i);
2319         }
2320 #endif
2321         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2322 }
2323
2324 /*
2325  * Maximal extent format file size.
2326  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2327  * extent format containers, within a sector_t, and within i_blocks
2328  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2329  * so that won't be a limiting factor.
2330  *
2331  * However there is other limiting factor. We do store extents in the form
2332  * of starting block and length, hence the resulting length of the extent
2333  * covering maximum file size must fit into on-disk format containers as
2334  * well. Given that length is always by 1 unit bigger than max unit (because
2335  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2336  *
2337  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2338  */
2339 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2340 {
2341         loff_t res;
2342         loff_t upper_limit = MAX_LFS_FILESIZE;
2343
2344         /* small i_blocks in vfs inode? */
2345         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2346                 /*
2347                  * CONFIG_LBDAF is not enabled implies the inode
2348                  * i_block represent total blocks in 512 bytes
2349                  * 32 == size of vfs inode i_blocks * 8
2350                  */
2351                 upper_limit = (1LL << 32) - 1;
2352
2353                 /* total blocks in file system block size */
2354                 upper_limit >>= (blkbits - 9);
2355                 upper_limit <<= blkbits;
2356         }
2357
2358         /*
2359          * 32-bit extent-start container, ee_block. We lower the maxbytes
2360          * by one fs block, so ee_len can cover the extent of maximum file
2361          * size
2362          */
2363         res = (1LL << 32) - 1;
2364         res <<= blkbits;
2365
2366         /* Sanity check against vm- & vfs- imposed limits */
2367         if (res > upper_limit)
2368                 res = upper_limit;
2369
2370         return res;
2371 }
2372
2373 /*
2374  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2375  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2376  * We need to be 1 filesystem block less than the 2^48 sector limit.
2377  */
2378 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2379 {
2380         loff_t res = EXT4_NDIR_BLOCKS;
2381         int meta_blocks;
2382         loff_t upper_limit;
2383         /* This is calculated to be the largest file size for a dense, block
2384          * mapped file such that the file's total number of 512-byte sectors,
2385          * including data and all indirect blocks, does not exceed (2^48 - 1).
2386          *
2387          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2388          * number of 512-byte sectors of the file.
2389          */
2390
2391         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2392                 /*
2393                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2394                  * the inode i_block field represents total file blocks in
2395                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2396                  */
2397                 upper_limit = (1LL << 32) - 1;
2398
2399                 /* total blocks in file system block size */
2400                 upper_limit >>= (bits - 9);
2401
2402         } else {
2403                 /*
2404                  * We use 48 bit ext4_inode i_blocks
2405                  * With EXT4_HUGE_FILE_FL set the i_blocks
2406                  * represent total number of blocks in
2407                  * file system block size
2408                  */
2409                 upper_limit = (1LL << 48) - 1;
2410
2411         }
2412
2413         /* indirect blocks */
2414         meta_blocks = 1;
2415         /* double indirect blocks */
2416         meta_blocks += 1 + (1LL << (bits-2));
2417         /* tripple indirect blocks */
2418         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2419
2420         upper_limit -= meta_blocks;
2421         upper_limit <<= bits;
2422
2423         res += 1LL << (bits-2);
2424         res += 1LL << (2*(bits-2));
2425         res += 1LL << (3*(bits-2));
2426         res <<= bits;
2427         if (res > upper_limit)
2428                 res = upper_limit;
2429
2430         if (res > MAX_LFS_FILESIZE)
2431                 res = MAX_LFS_FILESIZE;
2432
2433         return res;
2434 }
2435
2436 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2437                                    ext4_fsblk_t logical_sb_block, int nr)
2438 {
2439         struct ext4_sb_info *sbi = EXT4_SB(sb);
2440         ext4_group_t bg, first_meta_bg;
2441         int has_super = 0;
2442
2443         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2444
2445         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2446             nr < first_meta_bg)
2447                 return logical_sb_block + nr + 1;
2448         bg = sbi->s_desc_per_block * nr;
2449         if (ext4_bg_has_super(sb, bg))
2450                 has_super = 1;
2451
2452         return (has_super + ext4_group_first_block_no(sb, bg));
2453 }
2454
2455 /**
2456  * ext4_get_stripe_size: Get the stripe size.
2457  * @sbi: In memory super block info
2458  *
2459  * If we have specified it via mount option, then
2460  * use the mount option value. If the value specified at mount time is
2461  * greater than the blocks per group use the super block value.
2462  * If the super block value is greater than blocks per group return 0.
2463  * Allocator needs it be less than blocks per group.
2464  *
2465  */
2466 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2467 {
2468         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2469         unsigned long stripe_width =
2470                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2471         int ret;
2472
2473         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2474                 ret = sbi->s_stripe;
2475         else if (stripe_width <= sbi->s_blocks_per_group)
2476                 ret = stripe_width;
2477         else if (stride <= sbi->s_blocks_per_group)
2478                 ret = stride;
2479         else
2480                 ret = 0;
2481
2482         /*
2483          * If the stripe width is 1, this makes no sense and
2484          * we set it to 0 to turn off stripe handling code.
2485          */
2486         if (ret <= 1)
2487                 ret = 0;
2488
2489         return ret;
2490 }
2491
2492 /* sysfs supprt */
2493
2494 struct ext4_attr {
2495         struct attribute attr;
2496         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2497         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2498                          const char *, size_t);
2499         int offset;
2500 };
2501
2502 static int parse_strtoul(const char *buf,
2503                 unsigned long max, unsigned long *value)
2504 {
2505         char *endp;
2506
2507         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2508         endp = skip_spaces(endp);
2509         if (*endp || *value > max)
2510                 return -EINVAL;
2511
2512         return 0;
2513 }
2514
2515 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2516                                               struct ext4_sb_info *sbi,
2517                                               char *buf)
2518 {
2519         return snprintf(buf, PAGE_SIZE, "%llu\n",
2520                 (s64) EXT4_C2B(sbi,
2521                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2522 }
2523
2524 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2525                                          struct ext4_sb_info *sbi, char *buf)
2526 {
2527         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2528
2529         if (!sb->s_bdev->bd_part)
2530                 return snprintf(buf, PAGE_SIZE, "0\n");
2531         return snprintf(buf, PAGE_SIZE, "%lu\n",
2532                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2533                          sbi->s_sectors_written_start) >> 1);
2534 }
2535
2536 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2537                                           struct ext4_sb_info *sbi, char *buf)
2538 {
2539         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2540
2541         if (!sb->s_bdev->bd_part)
2542                 return snprintf(buf, PAGE_SIZE, "0\n");
2543         return snprintf(buf, PAGE_SIZE, "%llu\n",
2544                         (unsigned long long)(sbi->s_kbytes_written +
2545                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2546                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2547 }
2548
2549 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2550                                           struct ext4_sb_info *sbi,
2551                                           const char *buf, size_t count)
2552 {
2553         unsigned long t;
2554
2555         if (parse_strtoul(buf, 0x40000000, &t))
2556                 return -EINVAL;
2557
2558         if (t && !is_power_of_2(t))
2559                 return -EINVAL;
2560
2561         sbi->s_inode_readahead_blks = t;
2562         return count;
2563 }
2564
2565 static ssize_t sbi_ui_show(struct ext4_attr *a,
2566                            struct ext4_sb_info *sbi, char *buf)
2567 {
2568         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2569
2570         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2571 }
2572
2573 static ssize_t sbi_ui_store(struct ext4_attr *a,
2574                             struct ext4_sb_info *sbi,
2575                             const char *buf, size_t count)
2576 {
2577         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2578         unsigned long t;
2579
2580         if (parse_strtoul(buf, 0xffffffff, &t))
2581                 return -EINVAL;
2582         *ui = t;
2583         return count;
2584 }
2585
2586 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2587 static struct ext4_attr ext4_attr_##_name = {                   \
2588         .attr = {.name = __stringify(_name), .mode = _mode },   \
2589         .show   = _show,                                        \
2590         .store  = _store,                                       \
2591         .offset = offsetof(struct ext4_sb_info, _elname),       \
2592 }
2593 #define EXT4_ATTR(name, mode, show, store) \
2594 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2595
2596 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2597 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2598 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2599 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2600         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2601 #define ATTR_LIST(name) &ext4_attr_##name.attr
2602
2603 EXT4_RO_ATTR(delayed_allocation_blocks);
2604 EXT4_RO_ATTR(session_write_kbytes);
2605 EXT4_RO_ATTR(lifetime_write_kbytes);
2606 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2607                  inode_readahead_blks_store, s_inode_readahead_blks);
2608 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2609 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2610 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2611 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2612 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2613 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2614 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2615 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2616
2617 static struct attribute *ext4_attrs[] = {
2618         ATTR_LIST(delayed_allocation_blocks),
2619         ATTR_LIST(session_write_kbytes),
2620         ATTR_LIST(lifetime_write_kbytes),
2621         ATTR_LIST(inode_readahead_blks),
2622         ATTR_LIST(inode_goal),
2623         ATTR_LIST(mb_stats),
2624         ATTR_LIST(mb_max_to_scan),
2625         ATTR_LIST(mb_min_to_scan),
2626         ATTR_LIST(mb_order2_req),
2627         ATTR_LIST(mb_stream_req),
2628         ATTR_LIST(mb_group_prealloc),
2629         ATTR_LIST(max_writeback_mb_bump),
2630         NULL,
2631 };
2632
2633 /* Features this copy of ext4 supports */
2634 EXT4_INFO_ATTR(lazy_itable_init);
2635 EXT4_INFO_ATTR(batched_discard);
2636
2637 static struct attribute *ext4_feat_attrs[] = {
2638         ATTR_LIST(lazy_itable_init),
2639         ATTR_LIST(batched_discard),
2640         NULL,
2641 };
2642
2643 static ssize_t ext4_attr_show(struct kobject *kobj,
2644                               struct attribute *attr, char *buf)
2645 {
2646         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2647                                                 s_kobj);
2648         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2649
2650         return a->show ? a->show(a, sbi, buf) : 0;
2651 }
2652
2653 static ssize_t ext4_attr_store(struct kobject *kobj,
2654                                struct attribute *attr,
2655                                const char *buf, size_t len)
2656 {
2657         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2658                                                 s_kobj);
2659         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2660
2661         return a->store ? a->store(a, sbi, buf, len) : 0;
2662 }
2663
2664 static void ext4_sb_release(struct kobject *kobj)
2665 {
2666         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2667                                                 s_kobj);
2668         complete(&sbi->s_kobj_unregister);
2669 }
2670
2671 static const struct sysfs_ops ext4_attr_ops = {
2672         .show   = ext4_attr_show,
2673         .store  = ext4_attr_store,
2674 };
2675
2676 static struct kobj_type ext4_ktype = {
2677         .default_attrs  = ext4_attrs,
2678         .sysfs_ops      = &ext4_attr_ops,
2679         .release        = ext4_sb_release,
2680 };
2681
2682 static void ext4_feat_release(struct kobject *kobj)
2683 {
2684         complete(&ext4_feat->f_kobj_unregister);
2685 }
2686
2687 static struct kobj_type ext4_feat_ktype = {
2688         .default_attrs  = ext4_feat_attrs,
2689         .sysfs_ops      = &ext4_attr_ops,
2690         .release        = ext4_feat_release,
2691 };
2692
2693 /*
2694  * Check whether this filesystem can be mounted based on
2695  * the features present and the RDONLY/RDWR mount requested.
2696  * Returns 1 if this filesystem can be mounted as requested,
2697  * 0 if it cannot be.
2698  */
2699 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2700 {
2701         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2702                 ext4_msg(sb, KERN_ERR,
2703                         "Couldn't mount because of "
2704                         "unsupported optional features (%x)",
2705                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2706                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2707                 return 0;
2708         }
2709
2710         if (readonly)
2711                 return 1;
2712
2713         /* Check that feature set is OK for a read-write mount */
2714         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2715                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2716                          "unsupported optional features (%x)",
2717                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2718                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2719                 return 0;
2720         }
2721         /*
2722          * Large file size enabled file system can only be mounted
2723          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2724          */
2725         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2726                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2727                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2728                                  "cannot be mounted RDWR without "
2729                                  "CONFIG_LBDAF");
2730                         return 0;
2731                 }
2732         }
2733         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2734             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2735                 ext4_msg(sb, KERN_ERR,
2736                          "Can't support bigalloc feature without "
2737                          "extents feature\n");
2738                 return 0;
2739         }
2740         return 1;
2741 }
2742
2743 /*
2744  * This function is called once a day if we have errors logged
2745  * on the file system
2746  */
2747 static void print_daily_error_info(unsigned long arg)
2748 {
2749         struct super_block *sb = (struct super_block *) arg;
2750         struct ext4_sb_info *sbi;
2751         struct ext4_super_block *es;
2752
2753         sbi = EXT4_SB(sb);
2754         es = sbi->s_es;
2755
2756         if (es->s_error_count)
2757                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2758                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2759                          le32_to_cpu(es->s_error_count));
2760         if (es->s_first_error_time) {
2761                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2762                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2763                        (int) sizeof(es->s_first_error_func),
2764                        es->s_first_error_func,
2765                        le32_to_cpu(es->s_first_error_line));
2766                 if (es->s_first_error_ino)
2767                         printk(": inode %u",
2768                                le32_to_cpu(es->s_first_error_ino));
2769                 if (es->s_first_error_block)
2770                         printk(": block %llu", (unsigned long long)
2771                                le64_to_cpu(es->s_first_error_block));
2772                 printk("\n");
2773         }
2774         if (es->s_last_error_time) {
2775                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2776                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2777                        (int) sizeof(es->s_last_error_func),
2778                        es->s_last_error_func,
2779                        le32_to_cpu(es->s_last_error_line));
2780                 if (es->s_last_error_ino)
2781                         printk(": inode %u",
2782                                le32_to_cpu(es->s_last_error_ino));
2783                 if (es->s_last_error_block)
2784                         printk(": block %llu", (unsigned long long)
2785                                le64_to_cpu(es->s_last_error_block));
2786                 printk("\n");
2787         }
2788         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2789 }
2790
2791 /* Find next suitable group and run ext4_init_inode_table */
2792 static int ext4_run_li_request(struct ext4_li_request *elr)
2793 {
2794         struct ext4_group_desc *gdp = NULL;
2795         ext4_group_t group, ngroups;
2796         struct super_block *sb;
2797         unsigned long timeout = 0;
2798         int ret = 0;
2799
2800         sb = elr->lr_super;
2801         ngroups = EXT4_SB(sb)->s_groups_count;
2802
2803         for (group = elr->lr_next_group; group < ngroups; group++) {
2804                 gdp = ext4_get_group_desc(sb, group, NULL);
2805                 if (!gdp) {
2806                         ret = 1;
2807                         break;
2808                 }
2809
2810                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2811                         break;
2812         }
2813
2814         if (group == ngroups)
2815                 ret = 1;
2816
2817         if (!ret) {
2818                 timeout = jiffies;
2819                 ret = ext4_init_inode_table(sb, group,
2820                                             elr->lr_timeout ? 0 : 1);
2821                 if (elr->lr_timeout == 0) {
2822                         timeout = (jiffies - timeout) *
2823                                   elr->lr_sbi->s_li_wait_mult;
2824                         elr->lr_timeout = timeout;
2825                 }
2826                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2827                 elr->lr_next_group = group + 1;
2828         }
2829
2830         return ret;
2831 }
2832
2833 /*
2834  * Remove lr_request from the list_request and free the
2835  * request structure. Should be called with li_list_mtx held
2836  */
2837 static void ext4_remove_li_request(struct ext4_li_request *elr)
2838 {
2839         struct ext4_sb_info *sbi;
2840
2841         if (!elr)
2842                 return;
2843
2844         sbi = elr->lr_sbi;
2845
2846         list_del(&elr->lr_request);
2847         sbi->s_li_request = NULL;
2848         kfree(elr);
2849 }
2850
2851 static void ext4_unregister_li_request(struct super_block *sb)
2852 {
2853         mutex_lock(&ext4_li_mtx);
2854         if (!ext4_li_info) {
2855                 mutex_unlock(&ext4_li_mtx);
2856                 return;
2857         }
2858
2859         mutex_lock(&ext4_li_info->li_list_mtx);
2860         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2861         mutex_unlock(&ext4_li_info->li_list_mtx);
2862         mutex_unlock(&ext4_li_mtx);
2863 }
2864
2865 static struct task_struct *ext4_lazyinit_task;
2866
2867 /*
2868  * This is the function where ext4lazyinit thread lives. It walks
2869  * through the request list searching for next scheduled filesystem.
2870  * When such a fs is found, run the lazy initialization request
2871  * (ext4_rn_li_request) and keep track of the time spend in this
2872  * function. Based on that time we compute next schedule time of
2873  * the request. When walking through the list is complete, compute
2874  * next waking time and put itself into sleep.
2875  */
2876 static int ext4_lazyinit_thread(void *arg)
2877 {
2878         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2879         struct list_head *pos, *n;
2880         struct ext4_li_request *elr;
2881         unsigned long next_wakeup, cur;
2882
2883         BUG_ON(NULL == eli);
2884
2885 cont_thread:
2886         while (true) {
2887                 next_wakeup = MAX_JIFFY_OFFSET;
2888
2889                 mutex_lock(&eli->li_list_mtx);
2890                 if (list_empty(&eli->li_request_list)) {
2891                         mutex_unlock(&eli->li_list_mtx);
2892                         goto exit_thread;
2893                 }
2894
2895                 list_for_each_safe(pos, n, &eli->li_request_list) {
2896                         elr = list_entry(pos, struct ext4_li_request,
2897                                          lr_request);
2898
2899                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2900                                 if (ext4_run_li_request(elr) != 0) {
2901                                         /* error, remove the lazy_init job */
2902                                         ext4_remove_li_request(elr);
2903                                         continue;
2904                                 }
2905                         }
2906
2907                         if (time_before(elr->lr_next_sched, next_wakeup))
2908                                 next_wakeup = elr->lr_next_sched;
2909                 }
2910                 mutex_unlock(&eli->li_list_mtx);
2911
2912                 if (freezing(current))
2913                         refrigerator();
2914
2915                 cur = jiffies;
2916                 if ((time_after_eq(cur, next_wakeup)) ||
2917                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2918                         cond_resched();
2919                         continue;
2920                 }
2921
2922                 schedule_timeout_interruptible(next_wakeup - cur);
2923
2924                 if (kthread_should_stop()) {
2925                         ext4_clear_request_list();
2926                         goto exit_thread;
2927                 }
2928         }
2929
2930 exit_thread:
2931         /*
2932          * It looks like the request list is empty, but we need
2933          * to check it under the li_list_mtx lock, to prevent any
2934          * additions into it, and of course we should lock ext4_li_mtx
2935          * to atomically free the list and ext4_li_info, because at
2936          * this point another ext4 filesystem could be registering
2937          * new one.
2938          */
2939         mutex_lock(&ext4_li_mtx);
2940         mutex_lock(&eli->li_list_mtx);
2941         if (!list_empty(&eli->li_request_list)) {
2942                 mutex_unlock(&eli->li_list_mtx);
2943                 mutex_unlock(&ext4_li_mtx);
2944                 goto cont_thread;
2945         }
2946         mutex_unlock(&eli->li_list_mtx);
2947         kfree(ext4_li_info);
2948         ext4_li_info = NULL;
2949         mutex_unlock(&ext4_li_mtx);
2950
2951         return 0;
2952 }
2953
2954 static void ext4_clear_request_list(void)
2955 {
2956         struct list_head *pos, *n;
2957         struct ext4_li_request *elr;
2958
2959         mutex_lock(&ext4_li_info->li_list_mtx);
2960         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2961                 elr = list_entry(pos, struct ext4_li_request,
2962                                  lr_request);
2963                 ext4_remove_li_request(elr);
2964         }
2965         mutex_unlock(&ext4_li_info->li_list_mtx);
2966 }
2967
2968 static int ext4_run_lazyinit_thread(void)
2969 {
2970         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2971                                          ext4_li_info, "ext4lazyinit");
2972         if (IS_ERR(ext4_lazyinit_task)) {
2973                 int err = PTR_ERR(ext4_lazyinit_task);
2974                 ext4_clear_request_list();
2975                 kfree(ext4_li_info);
2976                 ext4_li_info = NULL;
2977                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2978                                  "initialization thread\n",
2979                                  err);
2980                 return err;
2981         }
2982         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2983         return 0;
2984 }
2985
2986 /*
2987  * Check whether it make sense to run itable init. thread or not.
2988  * If there is at least one uninitialized inode table, return
2989  * corresponding group number, else the loop goes through all
2990  * groups and return total number of groups.
2991  */
2992 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2993 {
2994         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2995         struct ext4_group_desc *gdp = NULL;
2996
2997         for (group = 0; group < ngroups; group++) {
2998                 gdp = ext4_get_group_desc(sb, group, NULL);
2999                 if (!gdp)
3000                         continue;
3001
3002                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3003                         break;
3004         }
3005
3006         return group;
3007 }
3008
3009 static int ext4_li_info_new(void)
3010 {
3011         struct ext4_lazy_init *eli = NULL;
3012
3013         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3014         if (!eli)
3015                 return -ENOMEM;
3016
3017         INIT_LIST_HEAD(&eli->li_request_list);
3018         mutex_init(&eli->li_list_mtx);
3019
3020         eli->li_state |= EXT4_LAZYINIT_QUIT;
3021
3022         ext4_li_info = eli;
3023
3024         return 0;
3025 }
3026
3027 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3028                                             ext4_group_t start)
3029 {
3030         struct ext4_sb_info *sbi = EXT4_SB(sb);
3031         struct ext4_li_request *elr;
3032         unsigned long rnd;
3033
3034         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3035         if (!elr)
3036                 return NULL;
3037
3038         elr->lr_super = sb;
3039         elr->lr_sbi = sbi;
3040         elr->lr_next_group = start;
3041
3042         /*
3043          * Randomize first schedule time of the request to
3044          * spread the inode table initialization requests
3045          * better.
3046          */
3047         get_random_bytes(&rnd, sizeof(rnd));
3048         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3049                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3050
3051         return elr;
3052 }
3053
3054 static int ext4_register_li_request(struct super_block *sb,
3055                                     ext4_group_t first_not_zeroed)
3056 {
3057         struct ext4_sb_info *sbi = EXT4_SB(sb);
3058         struct ext4_li_request *elr;
3059         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3060         int ret = 0;
3061
3062         if (sbi->s_li_request != NULL) {
3063                 /*
3064                  * Reset timeout so it can be computed again, because
3065                  * s_li_wait_mult might have changed.
3066                  */
3067                 sbi->s_li_request->lr_timeout = 0;
3068                 return 0;
3069         }
3070
3071         if (first_not_zeroed == ngroups ||
3072             (sb->s_flags & MS_RDONLY) ||
3073             !test_opt(sb, INIT_INODE_TABLE))
3074                 return 0;
3075
3076         elr = ext4_li_request_new(sb, first_not_zeroed);
3077         if (!elr)
3078                 return -ENOMEM;
3079
3080         mutex_lock(&ext4_li_mtx);
3081
3082         if (NULL == ext4_li_info) {
3083                 ret = ext4_li_info_new();
3084                 if (ret)
3085                         goto out;
3086         }
3087
3088         mutex_lock(&ext4_li_info->li_list_mtx);
3089         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3090         mutex_unlock(&ext4_li_info->li_list_mtx);
3091
3092         sbi->s_li_request = elr;
3093         /*
3094          * set elr to NULL here since it has been inserted to
3095          * the request_list and the removal and free of it is
3096          * handled by ext4_clear_request_list from now on.
3097          */
3098         elr = NULL;
3099
3100         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3101                 ret = ext4_run_lazyinit_thread();
3102                 if (ret)
3103                         goto out;
3104         }
3105 out:
3106         mutex_unlock(&ext4_li_mtx);
3107         if (ret)
3108                 kfree(elr);
3109         return ret;
3110 }
3111
3112 /*
3113  * We do not need to lock anything since this is called on
3114  * module unload.
3115  */
3116 static void ext4_destroy_lazyinit_thread(void)
3117 {
3118         /*
3119          * If thread exited earlier
3120          * there's nothing to be done.
3121          */
3122         if (!ext4_li_info || !ext4_lazyinit_task)
3123                 return;
3124
3125         kthread_stop(ext4_lazyinit_task);
3126 }
3127
3128 /*
3129  * Note: calculating the overhead so we can be compatible with
3130  * historical BSD practice is quite difficult in the face of
3131  * clusters/bigalloc.  This is because multiple metadata blocks from
3132  * different block group can end up in the same allocation cluster.
3133  * Calculating the exact overhead in the face of clustered allocation
3134  * requires either O(all block bitmaps) in memory or O(number of block
3135  * groups**2) in time.  We will still calculate the superblock for
3136  * older file systems --- and if we come across with a bigalloc file
3137  * system with zero in s_overhead_clusters the estimate will be close to
3138  * correct especially for very large cluster sizes --- but for newer
3139  * file systems, it's better to calculate this figure once at mkfs
3140  * time, and store it in the superblock.  If the superblock value is
3141  * present (even for non-bigalloc file systems), we will use it.
3142  */
3143 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3144                           char *buf)
3145 {
3146         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3147         struct ext4_group_desc  *gdp;
3148         ext4_fsblk_t            first_block, last_block, b;
3149         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3150         int                     s, j, count = 0;
3151
3152         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3153                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3154                         sbi->s_itb_per_group + 2);
3155
3156         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3157                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3158         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3159         for (i = 0; i < ngroups; i++) {
3160                 gdp = ext4_get_group_desc(sb, i, NULL);
3161                 b = ext4_block_bitmap(sb, gdp);
3162                 if (b >= first_block && b <= last_block) {
3163                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3164                         count++;
3165                 }
3166                 b = ext4_inode_bitmap(sb, gdp);
3167                 if (b >= first_block && b <= last_block) {
3168                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3169                         count++;
3170                 }
3171                 b = ext4_inode_table(sb, gdp);
3172                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3173                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3174                                 int c = EXT4_B2C(sbi, b - first_block);
3175                                 ext4_set_bit(c, buf);
3176                                 count++;
3177                         }
3178                 if (i != grp)
3179                         continue;
3180                 s = 0;
3181                 if (ext4_bg_has_super(sb, grp)) {
3182                         ext4_set_bit(s++, buf);
3183                         count++;
3184                 }
3185                 j = ext4_bg_num_gdb(sb, grp);
3186                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3187                         ext4_error(sb, "Invalid number of block group "
3188                                    "descriptor blocks: %d", j);
3189                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3190                 }
3191                 count += j;
3192                 for (; j > 0; j--)
3193                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3194         }
3195         if (!count)
3196                 return 0;
3197         return EXT4_CLUSTERS_PER_GROUP(sb) -
3198                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3199 }
3200
3201 /*
3202  * Compute the overhead and stash it in sbi->s_overhead
3203  */
3204 int ext4_calculate_overhead(struct super_block *sb)
3205 {
3206         struct ext4_sb_info *sbi = EXT4_SB(sb);
3207         struct ext4_super_block *es = sbi->s_es;
3208         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3209         ext4_fsblk_t overhead = 0;
3210         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3211
3212         memset(buf, 0, PAGE_SIZE);
3213         if (!buf)
3214                 return -ENOMEM;
3215
3216         /*
3217          * Compute the overhead (FS structures).  This is constant
3218          * for a given filesystem unless the number of block groups
3219          * changes so we cache the previous value until it does.
3220          */
3221
3222         /*
3223          * All of the blocks before first_data_block are overhead
3224          */
3225         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3226
3227         /*
3228          * Add the overhead found in each block group
3229          */
3230         for (i = 0; i < ngroups; i++) {
3231                 int blks;
3232
3233                 blks = count_overhead(sb, i, buf);
3234                 overhead += blks;
3235                 if (blks)
3236                         memset(buf, 0, PAGE_SIZE);
3237                 cond_resched();
3238         }
3239         sbi->s_overhead = overhead;
3240         smp_wmb();
3241         free_page((unsigned long) buf);
3242         return 0;
3243 }
3244
3245 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3246 {
3247         char *orig_data = kstrdup(data, GFP_KERNEL);
3248         struct buffer_head *bh;
3249         struct ext4_super_block *es = NULL;
3250         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3251         ext4_fsblk_t block;
3252         ext4_fsblk_t sb_block = get_sb_block(&data);
3253         ext4_fsblk_t logical_sb_block;
3254         unsigned long offset = 0;
3255         unsigned long journal_devnum = 0;
3256         unsigned long def_mount_opts;
3257         struct inode *root;
3258         char *cp;
3259         const char *descr;
3260         int ret = -ENOMEM;
3261         int blocksize, clustersize;
3262         unsigned int db_count;
3263         unsigned int i;
3264         int needs_recovery, has_huge_files, has_bigalloc;
3265         __u64 blocks_count;
3266         int err;
3267         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3268         ext4_group_t first_not_zeroed;
3269
3270         if ((data && !orig_data) || !sbi)
3271                 goto out_free_base;
3272
3273         sbi->s_blockgroup_lock =
3274                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3275         if (!sbi->s_blockgroup_lock)
3276                 goto out_free_base;
3277
3278         sb->s_fs_info = sbi;
3279         sbi->s_mount_opt = 0;
3280         sbi->s_resuid = EXT4_DEF_RESUID;
3281         sbi->s_resgid = EXT4_DEF_RESGID;
3282         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3283         sbi->s_sb_block = sb_block;
3284         if (sb->s_bdev->bd_part)
3285                 sbi->s_sectors_written_start =
3286                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3287
3288         /* Cleanup superblock name */
3289         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3290                 *cp = '!';
3291
3292         ret = -EINVAL;
3293         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3294         if (!blocksize) {
3295                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3296                 goto out_fail;
3297         }
3298
3299         /*
3300          * The ext4 superblock will not be buffer aligned for other than 1kB
3301          * block sizes.  We need to calculate the offset from buffer start.
3302          */
3303         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3304                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3305                 offset = do_div(logical_sb_block, blocksize);
3306         } else {
3307                 logical_sb_block = sb_block;
3308         }
3309
3310         if (!(bh = sb_bread(sb, logical_sb_block))) {
3311                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3312                 goto out_fail;
3313         }
3314         /*
3315          * Note: s_es must be initialized as soon as possible because
3316          *       some ext4 macro-instructions depend on its value
3317          */
3318         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3319         sbi->s_es = es;
3320         sb->s_magic = le16_to_cpu(es->s_magic);
3321         if (sb->s_magic != EXT4_SUPER_MAGIC)
3322                 goto cantfind_ext4;
3323         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3324
3325         /* Set defaults before we parse the mount options */
3326         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3327         set_opt(sb, INIT_INODE_TABLE);
3328         if (def_mount_opts & EXT4_DEFM_DEBUG)
3329                 set_opt(sb, DEBUG);
3330         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3331                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3332                         "2.6.38");
3333                 set_opt(sb, GRPID);
3334         }
3335         if (def_mount_opts & EXT4_DEFM_UID16)
3336                 set_opt(sb, NO_UID32);
3337         /* xattr user namespace & acls are now defaulted on */
3338 #ifdef CONFIG_EXT4_FS_XATTR
3339         set_opt(sb, XATTR_USER);
3340 #endif
3341 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3342         set_opt(sb, POSIX_ACL);
3343 #endif
3344         set_opt(sb, MBLK_IO_SUBMIT);
3345         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3346                 set_opt(sb, JOURNAL_DATA);
3347         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3348                 set_opt(sb, ORDERED_DATA);
3349         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3350                 set_opt(sb, WRITEBACK_DATA);
3351
3352         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3353                 set_opt(sb, ERRORS_PANIC);
3354         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3355                 set_opt(sb, ERRORS_CONT);
3356         else
3357                 set_opt(sb, ERRORS_RO);
3358         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3359                 set_opt(sb, BLOCK_VALIDITY);
3360         if (def_mount_opts & EXT4_DEFM_DISCARD)
3361                 set_opt(sb, DISCARD);
3362
3363         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3364         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3365         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3366         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3367         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3368
3369         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3370                 set_opt(sb, BARRIER);
3371
3372         /*
3373          * enable delayed allocation by default
3374          * Use -o nodelalloc to turn it off
3375          */
3376         if (!IS_EXT3_SB(sb) &&
3377             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3378                 set_opt(sb, DELALLOC);
3379
3380         /*
3381          * set default s_li_wait_mult for lazyinit, for the case there is
3382          * no mount option specified.
3383          */
3384         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3385
3386         if (sbi->s_es->s_mount_opts[0]) {
3387                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3388                                               sizeof(sbi->s_es->s_mount_opts),
3389                                               GFP_KERNEL);
3390                 if (!s_mount_opts)
3391                         goto failed_mount;
3392                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3393                                    &journal_ioprio, NULL, 0)) {
3394                         ext4_msg(sb, KERN_WARNING,
3395                                  "failed to parse options in superblock: %s",
3396                                  s_mount_opts);
3397                 }
3398                 kfree(s_mount_opts);
3399         }
3400         if (!parse_options((char *) data, sb, &journal_devnum,
3401                            &journal_ioprio, NULL, 0))
3402                 goto failed_mount;
3403
3404         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3405                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3406                             "with data=journal disables delayed "
3407                             "allocation and O_DIRECT support!\n");
3408                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3409                         ext4_msg(sb, KERN_ERR, "can't mount with "
3410                                  "both data=journal and delalloc");
3411                         goto failed_mount;
3412                 }
3413                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3414                         ext4_msg(sb, KERN_ERR, "can't mount with "
3415                                  "both data=journal and dioread_nolock");
3416                         goto failed_mount;
3417                 }
3418                 if (test_opt(sb, DELALLOC))
3419                         clear_opt(sb, DELALLOC);
3420         }
3421
3422         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3423                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3424
3425         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3426             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3427              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3428              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3429                 ext4_msg(sb, KERN_WARNING,
3430                        "feature flags set on rev 0 fs, "
3431                        "running e2fsck is recommended");
3432
3433         if (IS_EXT2_SB(sb)) {
3434                 if (ext2_feature_set_ok(sb))
3435                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3436                                  "using the ext4 subsystem");
3437                 else {
3438                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3439                                  "to feature incompatibilities");
3440                         goto failed_mount;
3441                 }
3442         }
3443
3444         if (IS_EXT3_SB(sb)) {
3445                 if (ext3_feature_set_ok(sb))
3446                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3447                                  "using the ext4 subsystem");
3448                 else {
3449                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3450                                  "to feature incompatibilities");
3451                         goto failed_mount;
3452                 }
3453         }
3454
3455         /*
3456          * Check feature flags regardless of the revision level, since we
3457          * previously didn't change the revision level when setting the flags,
3458          * so there is a chance incompat flags are set on a rev 0 filesystem.
3459          */
3460         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3461                 goto failed_mount;
3462
3463         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3464         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3465             blocksize > EXT4_MAX_BLOCK_SIZE) {
3466                 ext4_msg(sb, KERN_ERR,
3467                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3468                          blocksize, le32_to_cpu(es->s_log_block_size));
3469                 goto failed_mount;
3470         }
3471         if (le32_to_cpu(es->s_log_block_size) >
3472             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3473                 ext4_msg(sb, KERN_ERR,
3474                          "Invalid log block size: %u",
3475                          le32_to_cpu(es->s_log_block_size));
3476                 goto failed_mount;
3477         }
3478
3479         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3480                 ext4_msg(sb, KERN_ERR,
3481                          "Number of reserved GDT blocks insanely large: %d",
3482                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3483                 goto failed_mount;
3484         }
3485
3486         if (sb->s_blocksize != blocksize) {
3487                 /* Validate the filesystem blocksize */
3488                 if (!sb_set_blocksize(sb, blocksize)) {
3489                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3490                                         blocksize);
3491                         goto failed_mount;
3492                 }
3493
3494                 brelse(bh);
3495                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3496                 offset = do_div(logical_sb_block, blocksize);
3497                 bh = sb_bread(sb, logical_sb_block);
3498                 if (!bh) {
3499                         ext4_msg(sb, KERN_ERR,
3500                                "Can't read superblock on 2nd try");
3501                         goto failed_mount;
3502                 }
3503                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3504                 sbi->s_es = es;
3505                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3506                         ext4_msg(sb, KERN_ERR,
3507                                "Magic mismatch, very weird!");
3508                         goto failed_mount;
3509                 }
3510         }
3511
3512         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3513                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3514         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3515                                                       has_huge_files);
3516         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3517
3518         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3519                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3520                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3521         } else {
3522                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3523                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3524                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3525                     (!is_power_of_2(sbi->s_inode_size)) ||
3526                     (sbi->s_inode_size > blocksize)) {
3527                         ext4_msg(sb, KERN_ERR,
3528                                "unsupported inode size: %d",
3529                                sbi->s_inode_size);
3530                         goto failed_mount;
3531                 }
3532                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3533                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3534         }
3535
3536         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3538                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3539                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3540                     !is_power_of_2(sbi->s_desc_size)) {
3541                         ext4_msg(sb, KERN_ERR,
3542                                "unsupported descriptor size %lu",
3543                                sbi->s_desc_size);
3544                         goto failed_mount;
3545                 }
3546         } else
3547                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3548
3549         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3550         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3551
3552         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3553         if (sbi->s_inodes_per_block == 0)
3554                 goto cantfind_ext4;
3555         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3556             sbi->s_inodes_per_group > blocksize * 8) {
3557                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3558                          sbi->s_blocks_per_group);
3559                 goto failed_mount;
3560         }
3561         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3562                                         sbi->s_inodes_per_block;
3563         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3564         sbi->s_sbh = bh;
3565         sbi->s_mount_state = le16_to_cpu(es->s_state);
3566         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3567         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3568
3569         for (i = 0; i < 4; i++)
3570                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3571         sbi->s_def_hash_version = es->s_def_hash_version;
3572         i = le32_to_cpu(es->s_flags);
3573         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3574                 sbi->s_hash_unsigned = 3;
3575         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3576 #ifdef __CHAR_UNSIGNED__
3577                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3578                 sbi->s_hash_unsigned = 3;
3579 #else
3580                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3581 #endif
3582                 sb->s_dirt = 1;
3583         }
3584
3585         /* Handle clustersize */
3586         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3587         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3588                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3589         if (has_bigalloc) {
3590                 if (clustersize < blocksize) {
3591                         ext4_msg(sb, KERN_ERR,
3592                                  "cluster size (%d) smaller than "
3593                                  "block size (%d)", clustersize, blocksize);
3594                         goto failed_mount;
3595                 }
3596                 if (le32_to_cpu(es->s_log_cluster_size) >
3597                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3598                         ext4_msg(sb, KERN_ERR,
3599                                  "Invalid log cluster size: %u",
3600                                  le32_to_cpu(es->s_log_cluster_size));
3601                         goto failed_mount;
3602                 }
3603                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3604                         le32_to_cpu(es->s_log_block_size);
3605                 sbi->s_clusters_per_group =
3606                         le32_to_cpu(es->s_clusters_per_group);
3607                 if (sbi->s_clusters_per_group > blocksize * 8) {
3608                         ext4_msg(sb, KERN_ERR,
3609                                  "#clusters per group too big: %lu",
3610                                  sbi->s_clusters_per_group);
3611                         goto failed_mount;
3612                 }
3613                 if (sbi->s_blocks_per_group !=
3614                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3615                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3616                                  "clusters per group (%lu) inconsistent",
3617                                  sbi->s_blocks_per_group,
3618                                  sbi->s_clusters_per_group);
3619                         goto failed_mount;
3620                 }
3621         } else {
3622                 if (clustersize != blocksize) {
3623                         ext4_warning(sb, "fragment/cluster size (%d) != "
3624                                      "block size (%d)", clustersize,
3625                                      blocksize);
3626                         clustersize = blocksize;
3627                 }
3628                 if (sbi->s_blocks_per_group > blocksize * 8) {
3629                         ext4_msg(sb, KERN_ERR,
3630                                  "#blocks per group too big: %lu",
3631                                  sbi->s_blocks_per_group);
3632                         goto failed_mount;
3633                 }
3634                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3635                 sbi->s_cluster_bits = 0;
3636         }
3637         sbi->s_cluster_ratio = clustersize / blocksize;
3638
3639         /*
3640          * Test whether we have more sectors than will fit in sector_t,
3641          * and whether the max offset is addressable by the page cache.
3642          */
3643         err = generic_check_addressable(sb->s_blocksize_bits,
3644                                         ext4_blocks_count(es));
3645         if (err) {
3646                 ext4_msg(sb, KERN_ERR, "filesystem"
3647                          " too large to mount safely on this system");
3648                 if (sizeof(sector_t) < 8)
3649                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3650                 ret = err;
3651                 goto failed_mount;
3652         }
3653
3654         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3655                 goto cantfind_ext4;
3656
3657         /* check blocks count against device size */
3658         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3659         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3660                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3661                        "exceeds size of device (%llu blocks)",
3662                        ext4_blocks_count(es), blocks_count);
3663                 goto failed_mount;
3664         }
3665
3666         /*
3667          * It makes no sense for the first data block to be beyond the end
3668          * of the filesystem.
3669          */
3670         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3671                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3672                          "block %u is beyond end of filesystem (%llu)",
3673                          le32_to_cpu(es->s_first_data_block),
3674                          ext4_blocks_count(es));
3675                 goto failed_mount;
3676         }
3677         blocks_count = (ext4_blocks_count(es) -
3678                         le32_to_cpu(es->s_first_data_block) +
3679                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3680         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3681         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3682                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3683                        "(block count %llu, first data block %u, "
3684                        "blocks per group %lu)", sbi->s_groups_count,
3685                        ext4_blocks_count(es),
3686                        le32_to_cpu(es->s_first_data_block),
3687                        EXT4_BLOCKS_PER_GROUP(sb));
3688                 goto failed_mount;
3689         }
3690         sbi->s_groups_count = blocks_count;
3691         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3692                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3693         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3694                    EXT4_DESC_PER_BLOCK(sb);
3695         sbi->s_group_desc = ext4_kvmalloc(db_count *
3696                                           sizeof(struct buffer_head *),
3697                                           GFP_KERNEL);
3698         if (sbi->s_group_desc == NULL) {
3699                 ext4_msg(sb, KERN_ERR, "not enough memory");
3700                 goto failed_mount;
3701         }
3702
3703         if (ext4_proc_root)
3704                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3705
3706         bgl_lock_init(sbi->s_blockgroup_lock);
3707
3708         for (i = 0; i < db_count; i++) {
3709                 block = descriptor_loc(sb, logical_sb_block, i);
3710                 sbi->s_group_desc[i] = sb_bread(sb, block);
3711                 if (!sbi->s_group_desc[i]) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                "can't read group descriptor %d", i);
3714                         db_count = i;
3715                         goto failed_mount2;
3716                 }
3717         }
3718         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3719                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3720                 goto failed_mount2;
3721         }
3722         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3723                 if (!ext4_fill_flex_info(sb)) {
3724                         ext4_msg(sb, KERN_ERR,
3725                                "unable to initialize "
3726                                "flex_bg meta info!");
3727                         goto failed_mount2;
3728                 }
3729
3730         sbi->s_gdb_count = db_count;
3731         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3732         spin_lock_init(&sbi->s_next_gen_lock);
3733
3734         init_timer(&sbi->s_err_report);
3735         sbi->s_err_report.function = print_daily_error_info;
3736         sbi->s_err_report.data = (unsigned long) sb;
3737
3738         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3739                         ext4_count_free_clusters(sb));
3740         if (!err) {
3741                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3742                                 ext4_count_free_inodes(sb));
3743         }
3744         if (!err) {
3745                 err = percpu_counter_init(&sbi->s_dirs_counter,
3746                                 ext4_count_dirs(sb));
3747         }
3748         if (!err) {
3749                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3750         }
3751         if (err) {
3752                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3753                 goto failed_mount3;
3754         }
3755
3756         sbi->s_stripe = ext4_get_stripe_size(sbi);
3757         sbi->s_max_writeback_mb_bump = 128;
3758
3759         /*
3760          * set up enough so that it can read an inode
3761          */
3762         if (!test_opt(sb, NOLOAD) &&
3763             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3764                 sb->s_op = &ext4_sops;
3765         else
3766                 sb->s_op = &ext4_nojournal_sops;
3767         sb->s_export_op = &ext4_export_ops;
3768         sb->s_xattr = ext4_xattr_handlers;
3769 #ifdef CONFIG_QUOTA
3770         sb->s_qcop = &ext4_qctl_operations;
3771         sb->dq_op = &ext4_quota_operations;
3772 #endif
3773         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3774
3775         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3776         mutex_init(&sbi->s_orphan_lock);
3777         sbi->s_resize_flags = 0;
3778
3779         sb->s_root = NULL;
3780
3781         needs_recovery = (es->s_last_orphan != 0 ||
3782                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3783                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3784
3785         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3786             !(sb->s_flags & MS_RDONLY))
3787                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3788                         goto failed_mount3;
3789
3790         /*
3791          * The first inode we look at is the journal inode.  Don't try
3792          * root first: it may be modified in the journal!
3793          */
3794         if (!test_opt(sb, NOLOAD) &&
3795             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3796                 if (ext4_load_journal(sb, es, journal_devnum))
3797                         goto failed_mount3;
3798         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3799               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3800                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3801                        "suppressed and not mounted read-only");
3802                 goto failed_mount_wq;
3803         } else {
3804                 clear_opt(sb, DATA_FLAGS);
3805                 sbi->s_journal = NULL;
3806                 needs_recovery = 0;
3807                 goto no_journal;
3808         }
3809
3810         if (ext4_blocks_count(es) > 0xffffffffULL &&
3811             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3812                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3813                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3814                 goto failed_mount_wq;
3815         }
3816
3817         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3818                 jbd2_journal_set_features(sbi->s_journal,
3819                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3820                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3821         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3822                 jbd2_journal_set_features(sbi->s_journal,
3823                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3824                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3825                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3826         } else {
3827                 jbd2_journal_clear_features(sbi->s_journal,
3828                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3829                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3830         }
3831
3832         /* We have now updated the journal if required, so we can
3833          * validate the data journaling mode. */
3834         switch (test_opt(sb, DATA_FLAGS)) {
3835         case 0:
3836                 /* No mode set, assume a default based on the journal
3837                  * capabilities: ORDERED_DATA if the journal can
3838                  * cope, else JOURNAL_DATA
3839                  */
3840                 if (jbd2_journal_check_available_features
3841                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3842                         set_opt(sb, ORDERED_DATA);
3843                 else
3844                         set_opt(sb, JOURNAL_DATA);
3845                 break;
3846
3847         case EXT4_MOUNT_ORDERED_DATA:
3848         case EXT4_MOUNT_WRITEBACK_DATA:
3849                 if (!jbd2_journal_check_available_features
3850                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3851                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3852                                "requested data journaling mode");
3853                         goto failed_mount_wq;
3854                 }
3855         default:
3856                 break;
3857         }
3858         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3859
3860         /*
3861          * The journal may have updated the bg summary counts, so we
3862          * need to update the global counters.
3863          */
3864         percpu_counter_set(&sbi->s_freeclusters_counter,
3865                            ext4_count_free_clusters(sb));
3866         percpu_counter_set(&sbi->s_freeinodes_counter,
3867                            ext4_count_free_inodes(sb));
3868         percpu_counter_set(&sbi->s_dirs_counter,
3869                            ext4_count_dirs(sb));
3870         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3871
3872 no_journal:
3873         /*
3874          * Get the # of file system overhead blocks from the
3875          * superblock if present.
3876          */
3877         if (es->s_overhead_clusters)
3878                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3879         else {
3880                 ret = ext4_calculate_overhead(sb);
3881                 if (ret)
3882                         goto failed_mount_wq;
3883         }
3884
3885         /*
3886          * The maximum number of concurrent works can be high and
3887          * concurrency isn't really necessary.  Limit it to 1.
3888          */
3889         EXT4_SB(sb)->dio_unwritten_wq =
3890                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3891         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3892                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3893                 goto failed_mount_wq;
3894         }
3895
3896         /*
3897          * The jbd2_journal_load will have done any necessary log recovery,
3898          * so we can safely mount the rest of the filesystem now.
3899          */
3900
3901         root = ext4_iget(sb, EXT4_ROOT_INO);
3902         if (IS_ERR(root)) {
3903                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3904                 ret = PTR_ERR(root);
3905                 root = NULL;
3906                 goto failed_mount4;
3907         }
3908         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3909                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3910                 goto failed_mount4;
3911         }
3912         sb->s_root = d_alloc_root(root);
3913         if (!sb->s_root) {
3914                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3915                 ret = -ENOMEM;
3916                 goto failed_mount4;
3917         }
3918
3919         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3920                 sb->s_flags |= MS_RDONLY;
3921
3922         /* determine the minimum size of new large inodes, if present */
3923         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3924                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3925                                                      EXT4_GOOD_OLD_INODE_SIZE;
3926                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3927                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3928                         if (sbi->s_want_extra_isize <
3929                             le16_to_cpu(es->s_want_extra_isize))
3930                                 sbi->s_want_extra_isize =
3931                                         le16_to_cpu(es->s_want_extra_isize);
3932                         if (sbi->s_want_extra_isize <
3933                             le16_to_cpu(es->s_min_extra_isize))
3934                                 sbi->s_want_extra_isize =
3935                                         le16_to_cpu(es->s_min_extra_isize);
3936                 }
3937         }
3938         /* Check if enough inode space is available */
3939         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3940                                                         sbi->s_inode_size) {
3941                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3942                                                        EXT4_GOOD_OLD_INODE_SIZE;
3943                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3944                          "available");
3945         }
3946
3947         err = ext4_setup_system_zone(sb);
3948         if (err) {
3949                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3950                          "zone (%d)", err);
3951                 goto failed_mount4;
3952         }
3953
3954         ext4_ext_init(sb);
3955         err = ext4_mb_init(sb, needs_recovery);
3956         if (err) {
3957                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3958                          err);
3959                 goto failed_mount5;
3960         }
3961
3962         err = ext4_register_li_request(sb, first_not_zeroed);
3963         if (err)
3964                 goto failed_mount6;
3965
3966         sbi->s_kobj.kset = ext4_kset;
3967         init_completion(&sbi->s_kobj_unregister);
3968         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3969                                    "%s", sb->s_id);
3970         if (err)
3971                 goto failed_mount7;
3972
3973         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3974         ext4_orphan_cleanup(sb, es);
3975         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3976         if (needs_recovery) {
3977                 ext4_msg(sb, KERN_INFO, "recovery complete");
3978                 ext4_mark_recovery_complete(sb, es);
3979         }
3980         if (EXT4_SB(sb)->s_journal) {
3981                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3982                         descr = " journalled data mode";
3983                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3984                         descr = " ordered data mode";
3985                 else
3986                         descr = " writeback data mode";
3987         } else
3988                 descr = "out journal";
3989
3990         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3991                  "Opts: %.*s%s%s", descr,
3992                  (int) sizeof(sbi->s_es->s_mount_opts),
3993                  sbi->s_es->s_mount_opts,
3994                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3995
3996         if (es->s_error_count)
3997                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3998
3999         kfree(orig_data);
4000         return 0;
4001
4002 cantfind_ext4:
4003         if (!silent)
4004                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4005         goto failed_mount;
4006
4007 failed_mount7:
4008         ext4_unregister_li_request(sb);
4009 failed_mount6:
4010         ext4_ext_release(sb);
4011 failed_mount5:
4012         ext4_mb_release(sb);
4013         ext4_release_system_zone(sb);
4014 failed_mount4:
4015         iput(root);
4016         sb->s_root = NULL;
4017         ext4_msg(sb, KERN_ERR, "mount failed");
4018         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4019 failed_mount_wq:
4020         if (sbi->s_journal) {
4021                 jbd2_journal_destroy(sbi->s_journal);
4022                 sbi->s_journal = NULL;
4023         }
4024 failed_mount3:
4025         del_timer_sync(&sbi->s_err_report);
4026         if (sbi->s_flex_groups)
4027                 ext4_kvfree(sbi->s_flex_groups);
4028         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4029         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4030         percpu_counter_destroy(&sbi->s_dirs_counter);
4031         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4032         if (sbi->s_mmp_tsk)
4033                 kthread_stop(sbi->s_mmp_tsk);
4034 failed_mount2:
4035         for (i = 0; i < db_count; i++)
4036                 brelse(sbi->s_group_desc[i]);
4037         ext4_kvfree(sbi->s_group_desc);
4038 failed_mount:
4039         if (sbi->s_proc) {
4040                 remove_proc_entry(sb->s_id, ext4_proc_root);
4041         }
4042 #ifdef CONFIG_QUOTA
4043         for (i = 0; i < MAXQUOTAS; i++)
4044                 kfree(sbi->s_qf_names[i]);
4045 #endif
4046         ext4_blkdev_remove(sbi);
4047         brelse(bh);
4048 out_fail:
4049         sb->s_fs_info = NULL;
4050         kfree(sbi->s_blockgroup_lock);
4051 out_free_base:
4052         kfree(sbi);
4053         kfree(orig_data);
4054         return ret;
4055 }
4056
4057 /*
4058  * Setup any per-fs journal parameters now.  We'll do this both on
4059  * initial mount, once the journal has been initialised but before we've
4060  * done any recovery; and again on any subsequent remount.
4061  */
4062 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4063 {
4064         struct ext4_sb_info *sbi = EXT4_SB(sb);
4065
4066         journal->j_commit_interval = sbi->s_commit_interval;
4067         journal->j_min_batch_time = sbi->s_min_batch_time;
4068         journal->j_max_batch_time = sbi->s_max_batch_time;
4069
4070         write_lock(&journal->j_state_lock);
4071         if (test_opt(sb, BARRIER))
4072                 journal->j_flags |= JBD2_BARRIER;
4073         else
4074                 journal->j_flags &= ~JBD2_BARRIER;
4075         if (test_opt(sb, DATA_ERR_ABORT))
4076                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4077         else
4078                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4079         write_unlock(&journal->j_state_lock);
4080 }
4081
4082 static journal_t *ext4_get_journal(struct super_block *sb,
4083                                    unsigned int journal_inum)
4084 {
4085         struct inode *journal_inode;
4086         journal_t *journal;
4087
4088         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4089
4090         /* First, test for the existence of a valid inode on disk.  Bad
4091          * things happen if we iget() an unused inode, as the subsequent
4092          * iput() will try to delete it. */
4093
4094         journal_inode = ext4_iget(sb, journal_inum);
4095         if (IS_ERR(journal_inode)) {
4096                 ext4_msg(sb, KERN_ERR, "no journal found");
4097                 return NULL;
4098         }
4099         if (!journal_inode->i_nlink) {
4100                 make_bad_inode(journal_inode);
4101                 iput(journal_inode);
4102                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4103                 return NULL;
4104         }
4105
4106         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4107                   journal_inode, journal_inode->i_size);
4108         if (!S_ISREG(journal_inode->i_mode)) {
4109                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4110                 iput(journal_inode);
4111                 return NULL;
4112         }
4113
4114         journal = jbd2_journal_init_inode(journal_inode);
4115         if (!journal) {
4116                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4117                 iput(journal_inode);
4118                 return NULL;
4119         }
4120         journal->j_private = sb;
4121         ext4_init_journal_params(sb, journal);
4122         return journal;
4123 }
4124
4125 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4126                                        dev_t j_dev)
4127 {
4128         struct buffer_head *bh;
4129         journal_t *journal;
4130         ext4_fsblk_t start;
4131         ext4_fsblk_t len;
4132         int hblock, blocksize;
4133         ext4_fsblk_t sb_block;
4134         unsigned long offset;
4135         struct ext4_super_block *es;
4136         struct block_device *bdev;
4137
4138         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4139
4140         bdev = ext4_blkdev_get(j_dev, sb);
4141         if (bdev == NULL)
4142                 return NULL;
4143
4144         blocksize = sb->s_blocksize;
4145         hblock = bdev_logical_block_size(bdev);
4146         if (blocksize < hblock) {
4147                 ext4_msg(sb, KERN_ERR,
4148                         "blocksize too small for journal device");
4149                 goto out_bdev;
4150         }
4151
4152         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4153         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4154         set_blocksize(bdev, blocksize);
4155         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4156                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4157                        "external journal");
4158                 goto out_bdev;
4159         }
4160
4161         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
4162         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4163             !(le32_to_cpu(es->s_feature_incompat) &
4164               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4165                 ext4_msg(sb, KERN_ERR, "external journal has "
4166                                         "bad superblock");
4167                 brelse(bh);
4168                 goto out_bdev;
4169         }
4170
4171         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4172                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4173                 brelse(bh);
4174                 goto out_bdev;
4175         }
4176
4177         len = ext4_blocks_count(es);
4178         start = sb_block + 1;
4179         brelse(bh);     /* we're done with the superblock */
4180
4181         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4182                                         start, len, blocksize);
4183         if (!journal) {
4184                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4185                 goto out_bdev;
4186         }
4187         journal->j_private = sb;
4188         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4189         wait_on_buffer(journal->j_sb_buffer);
4190         if (!buffer_uptodate(journal->j_sb_buffer)) {
4191                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4192                 goto out_journal;
4193         }
4194         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4195                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4196                                         "user (unsupported) - %d",
4197                         be32_to_cpu(journal->j_superblock->s_nr_users));
4198                 goto out_journal;
4199         }
4200         EXT4_SB(sb)->journal_bdev = bdev;
4201         ext4_init_journal_params(sb, journal);
4202         return journal;
4203
4204 out_journal:
4205         jbd2_journal_destroy(journal);
4206 out_bdev:
4207         ext4_blkdev_put(bdev);
4208         return NULL;
4209 }
4210
4211 static int ext4_load_journal(struct super_block *sb,
4212                              struct ext4_super_block *es,
4213                              unsigned long journal_devnum)
4214 {
4215         journal_t *journal;
4216         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4217         dev_t journal_dev;
4218         int err = 0;
4219         int really_read_only;
4220
4221         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4222
4223         if (journal_devnum &&
4224             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4225                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4226                         "numbers have changed");
4227                 journal_dev = new_decode_dev(journal_devnum);
4228         } else
4229                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4230
4231         really_read_only = bdev_read_only(sb->s_bdev);
4232
4233         /*
4234          * Are we loading a blank journal or performing recovery after a
4235          * crash?  For recovery, we need to check in advance whether we
4236          * can get read-write access to the device.
4237          */
4238         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4239                 if (sb->s_flags & MS_RDONLY) {
4240                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4241                                         "required on readonly filesystem");
4242                         if (really_read_only) {
4243                                 ext4_msg(sb, KERN_ERR, "write access "
4244                                         "unavailable, cannot proceed");
4245                                 return -EROFS;
4246                         }
4247                         ext4_msg(sb, KERN_INFO, "write access will "
4248                                "be enabled during recovery");
4249                 }
4250         }
4251
4252         if (journal_inum && journal_dev) {
4253                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4254                        "and inode journals!");
4255                 return -EINVAL;
4256         }
4257
4258         if (journal_inum) {
4259                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4260                         return -EINVAL;
4261         } else {
4262                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4263                         return -EINVAL;
4264         }
4265
4266         if (!(journal->j_flags & JBD2_BARRIER))
4267                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4268
4269         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4270                 err = jbd2_journal_update_format(journal);
4271                 if (err)  {
4272                         ext4_msg(sb, KERN_ERR, "error updating journal");
4273                         jbd2_journal_destroy(journal);
4274                         return err;
4275                 }
4276         }
4277
4278         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4279                 err = jbd2_journal_wipe(journal, !really_read_only);
4280         if (!err) {
4281                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4282                 if (save)
4283                         memcpy(save, ((char *) es) +
4284                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4285                 err = jbd2_journal_load(journal);
4286                 if (save)
4287                         memcpy(((char *) es) + EXT4_S_ERR_START,
4288                                save, EXT4_S_ERR_LEN);
4289                 kfree(save);
4290         }
4291
4292         if (err) {
4293                 ext4_msg(sb, KERN_ERR, "error loading journal");
4294                 jbd2_journal_destroy(journal);
4295                 return err;
4296         }
4297
4298         EXT4_SB(sb)->s_journal = journal;
4299         ext4_clear_journal_err(sb, es);
4300
4301         if (!really_read_only && journal_devnum &&
4302             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4303                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4304
4305                 /* Make sure we flush the recovery flag to disk. */
4306                 ext4_commit_super(sb, 1);
4307         }
4308
4309         return 0;
4310 }
4311
4312 static int ext4_commit_super(struct super_block *sb, int sync)
4313 {
4314         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4315         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4316         int error = 0;
4317
4318         if (!sbh || block_device_ejected(sb))
4319                 return error;
4320         if (buffer_write_io_error(sbh)) {
4321                 /*
4322                  * Oh, dear.  A previous attempt to write the
4323                  * superblock failed.  This could happen because the
4324                  * USB device was yanked out.  Or it could happen to
4325                  * be a transient write error and maybe the block will
4326                  * be remapped.  Nothing we can do but to retry the
4327                  * write and hope for the best.
4328                  */
4329                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4330                        "superblock detected");
4331                 clear_buffer_write_io_error(sbh);
4332                 set_buffer_uptodate(sbh);
4333         }
4334         /*
4335          * If the file system is mounted read-only, don't update the
4336          * superblock write time.  This avoids updating the superblock
4337          * write time when we are mounting the root file system
4338          * read/only but we need to replay the journal; at that point,
4339          * for people who are east of GMT and who make their clock
4340          * tick in localtime for Windows bug-for-bug compatibility,
4341          * the clock is set in the future, and this will cause e2fsck
4342          * to complain and force a full file system check.
4343          */
4344         if (!(sb->s_flags & MS_RDONLY))
4345                 es->s_wtime = cpu_to_le32(get_seconds());
4346         if (sb->s_bdev->bd_part)
4347                 es->s_kbytes_written =
4348                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4349                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4350                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4351         else
4352                 es->s_kbytes_written =
4353                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4354         ext4_free_blocks_count_set(es,
4355                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4356                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4357         es->s_free_inodes_count =
4358                 cpu_to_le32(percpu_counter_sum_positive(
4359                                 &EXT4_SB(sb)->s_freeinodes_counter));
4360         sb->s_dirt = 0;
4361         BUFFER_TRACE(sbh, "marking dirty");
4362         mark_buffer_dirty(sbh);
4363         if (sync) {
4364                 error = sync_dirty_buffer(sbh);
4365                 if (error)
4366                         return error;
4367
4368                 error = buffer_write_io_error(sbh);
4369                 if (error) {
4370                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4371                                "superblock");
4372                         clear_buffer_write_io_error(sbh);
4373                         set_buffer_uptodate(sbh);
4374                 }
4375         }
4376         return error;
4377 }
4378
4379 /*
4380  * Have we just finished recovery?  If so, and if we are mounting (or
4381  * remounting) the filesystem readonly, then we will end up with a
4382  * consistent fs on disk.  Record that fact.
4383  */
4384 static void ext4_mark_recovery_complete(struct super_block *sb,
4385                                         struct ext4_super_block *es)
4386 {
4387         journal_t *journal = EXT4_SB(sb)->s_journal;
4388
4389         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4390                 BUG_ON(journal != NULL);
4391                 return;
4392         }
4393         jbd2_journal_lock_updates(journal);
4394         if (jbd2_journal_flush(journal) < 0)
4395                 goto out;
4396
4397         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4398             sb->s_flags & MS_RDONLY) {
4399                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4400                 ext4_commit_super(sb, 1);
4401         }
4402
4403 out:
4404         jbd2_journal_unlock_updates(journal);
4405 }
4406
4407 /*
4408  * If we are mounting (or read-write remounting) a filesystem whose journal
4409  * has recorded an error from a previous lifetime, move that error to the
4410  * main filesystem now.
4411  */
4412 static void ext4_clear_journal_err(struct super_block *sb,
4413                                    struct ext4_super_block *es)
4414 {
4415         journal_t *journal;
4416         int j_errno;
4417         const char *errstr;
4418
4419         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4420
4421         journal = EXT4_SB(sb)->s_journal;
4422
4423         /*
4424          * Now check for any error status which may have been recorded in the
4425          * journal by a prior ext4_error() or ext4_abort()
4426          */
4427
4428         j_errno = jbd2_journal_errno(journal);
4429         if (j_errno) {
4430                 char nbuf[16];
4431
4432                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4433                 ext4_warning(sb, "Filesystem error recorded "
4434                              "from previous mount: %s", errstr);
4435                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4436
4437                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4438                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4439                 ext4_commit_super(sb, 1);
4440
4441                 jbd2_journal_clear_err(journal);
4442         }
4443 }
4444
4445 /*
4446  * Force the running and committing transactions to commit,
4447  * and wait on the commit.
4448  */
4449 int ext4_force_commit(struct super_block *sb)
4450 {
4451         journal_t *journal;
4452         int ret = 0;
4453
4454         if (sb->s_flags & MS_RDONLY)
4455                 return 0;
4456
4457         journal = EXT4_SB(sb)->s_journal;
4458         if (journal) {
4459                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4460                 ret = ext4_journal_force_commit(journal);
4461         }
4462
4463         return ret;
4464 }
4465
4466 static void ext4_write_super(struct super_block *sb)
4467 {
4468         lock_super(sb);
4469         ext4_commit_super(sb, 1);
4470         unlock_super(sb);
4471 }
4472
4473 static int ext4_sync_fs(struct super_block *sb, int wait)
4474 {
4475         int ret = 0;
4476         tid_t target;
4477         struct ext4_sb_info *sbi = EXT4_SB(sb);
4478
4479         trace_ext4_sync_fs(sb, wait);
4480         flush_workqueue(sbi->dio_unwritten_wq);
4481         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4482                 if (wait)
4483                         jbd2_log_wait_commit(sbi->s_journal, target);
4484         }
4485         return ret;
4486 }
4487
4488 /*
4489  * LVM calls this function before a (read-only) snapshot is created.  This
4490  * gives us a chance to flush the journal completely and mark the fs clean.
4491  *
4492  * Note that only this function cannot bring a filesystem to be in a clean
4493  * state independently, because ext4 prevents a new handle from being started
4494  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4495  * the upper layer.
4496  */
4497 static int ext4_freeze(struct super_block *sb)
4498 {
4499         int error = 0;
4500         journal_t *journal;
4501
4502         if (sb->s_flags & MS_RDONLY)
4503                 return 0;
4504
4505         journal = EXT4_SB(sb)->s_journal;
4506
4507         /* Now we set up the journal barrier. */
4508         jbd2_journal_lock_updates(journal);
4509
4510         /*
4511          * Don't clear the needs_recovery flag if we failed to flush
4512          * the journal.
4513          */
4514         error = jbd2_journal_flush(journal);
4515         if (error < 0)
4516                 goto out;
4517
4518         /* Journal blocked and flushed, clear needs_recovery flag. */
4519         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4520         error = ext4_commit_super(sb, 1);
4521 out:
4522         /* we rely on s_frozen to stop further updates */
4523         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4524         return error;
4525 }
4526
4527 /*
4528  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4529  * flag here, even though the filesystem is not technically dirty yet.
4530  */
4531 static int ext4_unfreeze(struct super_block *sb)
4532 {
4533         if (sb->s_flags & MS_RDONLY)
4534                 return 0;
4535
4536         lock_super(sb);
4537         /* Reset the needs_recovery flag before the fs is unlocked. */
4538         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4539         ext4_commit_super(sb, 1);
4540         unlock_super(sb);
4541         return 0;
4542 }
4543
4544 /*
4545  * Structure to save mount options for ext4_remount's benefit
4546  */
4547 struct ext4_mount_options {
4548         unsigned long s_mount_opt;
4549         unsigned long s_mount_opt2;
4550         uid_t s_resuid;
4551         gid_t s_resgid;
4552         unsigned long s_commit_interval;
4553         u32 s_min_batch_time, s_max_batch_time;
4554 #ifdef CONFIG_QUOTA
4555         int s_jquota_fmt;
4556         char *s_qf_names[MAXQUOTAS];
4557 #endif
4558 };
4559
4560 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4561 {
4562         struct ext4_super_block *es;
4563         struct ext4_sb_info *sbi = EXT4_SB(sb);
4564         ext4_fsblk_t n_blocks_count = 0;
4565         unsigned long old_sb_flags;
4566         struct ext4_mount_options old_opts;
4567         int enable_quota = 0;
4568         ext4_group_t g;
4569         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4570         int err = 0;
4571 #ifdef CONFIG_QUOTA
4572         int i;
4573 #endif
4574         char *orig_data = kstrdup(data, GFP_KERNEL);
4575
4576         /* Store the original options */
4577         lock_super(sb);
4578         old_sb_flags = sb->s_flags;
4579         old_opts.s_mount_opt = sbi->s_mount_opt;
4580         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4581         old_opts.s_resuid = sbi->s_resuid;
4582         old_opts.s_resgid = sbi->s_resgid;
4583         old_opts.s_commit_interval = sbi->s_commit_interval;
4584         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4585         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4586 #ifdef CONFIG_QUOTA
4587         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4588         for (i = 0; i < MAXQUOTAS; i++)
4589                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4590 #endif
4591         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4592                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4593
4594         /*
4595          * Allow the "check" option to be passed as a remount option.
4596          */
4597         if (!parse_options(data, sb, NULL, &journal_ioprio,
4598                            &n_blocks_count, 1)) {
4599                 err = -EINVAL;
4600                 goto restore_opts;
4601         }
4602
4603         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4604                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4605                         ext4_msg(sb, KERN_ERR, "can't mount with "
4606                                  "both data=journal and delalloc");
4607                         err = -EINVAL;
4608                         goto restore_opts;
4609                 }
4610                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4611                         ext4_msg(sb, KERN_ERR, "can't mount with "
4612                                  "both data=journal and dioread_nolock");
4613                         err = -EINVAL;
4614                         goto restore_opts;
4615                 }
4616         }
4617
4618         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4619                 ext4_abort(sb, "Abort forced by user");
4620
4621         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4622                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4623
4624         es = sbi->s_es;
4625
4626         if (sbi->s_journal) {
4627                 ext4_init_journal_params(sb, sbi->s_journal);
4628                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4629         }
4630
4631         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4632                 n_blocks_count > ext4_blocks_count(es)) {
4633                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4634                         err = -EROFS;
4635                         goto restore_opts;
4636                 }
4637
4638                 if (*flags & MS_RDONLY) {
4639                         err = dquot_suspend(sb, -1);
4640                         if (err < 0)
4641                                 goto restore_opts;
4642
4643                         /*
4644                          * First of all, the unconditional stuff we have to do
4645                          * to disable replay of the journal when we next remount
4646                          */
4647                         sb->s_flags |= MS_RDONLY;
4648
4649                         /*
4650                          * OK, test if we are remounting a valid rw partition
4651                          * readonly, and if so set the rdonly flag and then
4652                          * mark the partition as valid again.
4653                          */
4654                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4655                             (sbi->s_mount_state & EXT4_VALID_FS))
4656                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4657
4658                         if (sbi->s_journal)
4659                                 ext4_mark_recovery_complete(sb, es);
4660                 } else {
4661                         /* Make sure we can mount this feature set readwrite */
4662                         if (!ext4_feature_set_ok(sb, 0)) {
4663                                 err = -EROFS;
4664                                 goto restore_opts;
4665                         }
4666                         /*
4667                          * Make sure the group descriptor checksums
4668                          * are sane.  If they aren't, refuse to remount r/w.
4669                          */
4670                         for (g = 0; g < sbi->s_groups_count; g++) {
4671                                 struct ext4_group_desc *gdp =
4672                                         ext4_get_group_desc(sb, g, NULL);
4673
4674                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4675                                         ext4_msg(sb, KERN_ERR,
4676                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4677                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4678                                                le16_to_cpu(gdp->bg_checksum));
4679                                         err = -EINVAL;
4680                                         goto restore_opts;
4681                                 }
4682                         }
4683
4684                         /*
4685                          * If we have an unprocessed orphan list hanging
4686                          * around from a previously readonly bdev mount,
4687                          * require a full umount/remount for now.
4688                          */
4689                         if (es->s_last_orphan) {
4690                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4691                                        "remount RDWR because of unprocessed "
4692                                        "orphan inode list.  Please "
4693                                        "umount/remount instead");
4694                                 err = -EINVAL;
4695                                 goto restore_opts;
4696                         }
4697
4698                         /*
4699                          * Mounting a RDONLY partition read-write, so reread
4700                          * and store the current valid flag.  (It may have
4701                          * been changed by e2fsck since we originally mounted
4702                          * the partition.)
4703                          */
4704                         if (sbi->s_journal)
4705                                 ext4_clear_journal_err(sb, es);
4706                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4707                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4708                                 goto restore_opts;
4709                         if (!ext4_setup_super(sb, es, 0))
4710                                 sb->s_flags &= ~MS_RDONLY;
4711                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4712                                                      EXT4_FEATURE_INCOMPAT_MMP))
4713                                 if (ext4_multi_mount_protect(sb,
4714                                                 le64_to_cpu(es->s_mmp_block))) {
4715                                         err = -EROFS;
4716                                         goto restore_opts;
4717                                 }
4718                         enable_quota = 1;
4719                 }
4720         }
4721
4722         /*
4723          * Reinitialize lazy itable initialization thread based on
4724          * current settings
4725          */
4726         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4727                 ext4_unregister_li_request(sb);
4728         else {
4729                 ext4_group_t first_not_zeroed;
4730                 first_not_zeroed = ext4_has_uninit_itable(sb);
4731                 ext4_register_li_request(sb, first_not_zeroed);
4732         }
4733
4734         ext4_setup_system_zone(sb);
4735         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4736                 ext4_commit_super(sb, 1);
4737
4738 #ifdef CONFIG_QUOTA
4739         /* Release old quota file names */
4740         for (i = 0; i < MAXQUOTAS; i++)
4741                 if (old_opts.s_qf_names[i] &&
4742                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4743                         kfree(old_opts.s_qf_names[i]);
4744 #endif
4745         unlock_super(sb);
4746         if (enable_quota)
4747                 dquot_resume(sb, -1);
4748
4749         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4750         kfree(orig_data);
4751         return 0;
4752
4753 restore_opts:
4754         sb->s_flags = old_sb_flags;
4755         sbi->s_mount_opt = old_opts.s_mount_opt;
4756         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4757         sbi->s_resuid = old_opts.s_resuid;
4758         sbi->s_resgid = old_opts.s_resgid;
4759         sbi->s_commit_interval = old_opts.s_commit_interval;
4760         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4761         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4762 #ifdef CONFIG_QUOTA
4763         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4764         for (i = 0; i < MAXQUOTAS; i++) {
4765                 if (sbi->s_qf_names[i] &&
4766                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4767                         kfree(sbi->s_qf_names[i]);
4768                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4769         }
4770 #endif
4771         unlock_super(sb);
4772         kfree(orig_data);
4773         return err;
4774 }
4775
4776 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4777 {
4778         struct super_block *sb = dentry->d_sb;
4779         struct ext4_sb_info *sbi = EXT4_SB(sb);
4780         struct ext4_super_block *es = sbi->s_es;
4781         ext4_fsblk_t overhead = 0;
4782         u64 fsid;
4783         s64 bfree;
4784
4785         if (!test_opt(sb, MINIX_DF))
4786                 overhead = sbi->s_overhead;
4787
4788         buf->f_type = EXT4_SUPER_MAGIC;
4789         buf->f_bsize = sb->s_blocksize;
4790         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4791         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4792                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4793         /* prevent underflow in case that few free space is available */
4794         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4795         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4796         if (buf->f_bfree < ext4_r_blocks_count(es))
4797                 buf->f_bavail = 0;
4798         buf->f_files = le32_to_cpu(es->s_inodes_count);
4799         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4800         buf->f_namelen = EXT4_NAME_LEN;
4801         fsid = le64_to_cpup((void *)es->s_uuid) ^
4802                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4803         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4804         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4805
4806         return 0;
4807 }
4808
4809 /* Helper function for writing quotas on sync - we need to start transaction
4810  * before quota file is locked for write. Otherwise the are possible deadlocks:
4811  * Process 1                         Process 2
4812  * ext4_create()                     quota_sync()
4813  *   jbd2_journal_start()                  write_dquot()
4814  *   dquot_initialize()                         down(dqio_mutex)
4815  *     down(dqio_mutex)                    jbd2_journal_start()
4816  *
4817  */
4818
4819 #ifdef CONFIG_QUOTA
4820
4821 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4822 {
4823         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4824 }
4825
4826 static int ext4_write_dquot(struct dquot *dquot)
4827 {
4828         int ret, err;
4829         handle_t *handle;
4830         struct inode *inode;
4831
4832         inode = dquot_to_inode(dquot);
4833         handle = ext4_journal_start(inode,
4834                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4835         if (IS_ERR(handle))
4836                 return PTR_ERR(handle);
4837         ret = dquot_commit(dquot);
4838         err = ext4_journal_stop(handle);
4839         if (!ret)
4840                 ret = err;
4841         return ret;
4842 }
4843
4844 static int ext4_acquire_dquot(struct dquot *dquot)
4845 {
4846         int ret, err;
4847         handle_t *handle;
4848
4849         handle = ext4_journal_start(dquot_to_inode(dquot),
4850                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4851         if (IS_ERR(handle))
4852                 return PTR_ERR(handle);
4853         ret = dquot_acquire(dquot);
4854         err = ext4_journal_stop(handle);
4855         if (!ret)
4856                 ret = err;
4857         return ret;
4858 }
4859
4860 static int ext4_release_dquot(struct dquot *dquot)
4861 {
4862         int ret, err;
4863         handle_t *handle;
4864
4865         handle = ext4_journal_start(dquot_to_inode(dquot),
4866                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4867         if (IS_ERR(handle)) {
4868                 /* Release dquot anyway to avoid endless cycle in dqput() */
4869                 dquot_release(dquot);
4870                 return PTR_ERR(handle);
4871         }
4872         ret = dquot_release(dquot);
4873         err = ext4_journal_stop(handle);
4874         if (!ret)
4875                 ret = err;
4876         return ret;
4877 }
4878
4879 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4880 {
4881         /* Are we journaling quotas? */
4882         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4883             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4884                 dquot_mark_dquot_dirty(dquot);
4885                 return ext4_write_dquot(dquot);
4886         } else {
4887                 return dquot_mark_dquot_dirty(dquot);
4888         }
4889 }
4890
4891 static int ext4_write_info(struct super_block *sb, int type)
4892 {
4893         int ret, err;
4894         handle_t *handle;
4895
4896         /* Data block + inode block */
4897         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4898         if (IS_ERR(handle))
4899                 return PTR_ERR(handle);
4900         ret = dquot_commit_info(sb, type);
4901         err = ext4_journal_stop(handle);
4902         if (!ret)
4903                 ret = err;
4904         return ret;
4905 }
4906
4907 /*
4908  * Turn on quotas during mount time - we need to find
4909  * the quota file and such...
4910  */
4911 static int ext4_quota_on_mount(struct super_block *sb, int type)
4912 {
4913         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4914                                         EXT4_SB(sb)->s_jquota_fmt, type);
4915 }
4916
4917 /*
4918  * Standard function to be called on quota_on
4919  */
4920 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4921                          struct path *path)
4922 {
4923         int err;
4924
4925         if (!test_opt(sb, QUOTA))
4926                 return -EINVAL;
4927
4928         /* Quotafile not on the same filesystem? */
4929         if (path->mnt->mnt_sb != sb)
4930                 return -EXDEV;
4931         /* Journaling quota? */
4932         if (EXT4_SB(sb)->s_qf_names[type]) {
4933                 /* Quotafile not in fs root? */
4934                 if (path->dentry->d_parent != sb->s_root)
4935                         ext4_msg(sb, KERN_WARNING,
4936                                 "Quota file not on filesystem root. "
4937                                 "Journaled quota will not work");
4938         }
4939
4940         /*
4941          * When we journal data on quota file, we have to flush journal to see
4942          * all updates to the file when we bypass pagecache...
4943          */
4944         if (EXT4_SB(sb)->s_journal &&
4945             ext4_should_journal_data(path->dentry->d_inode)) {
4946                 /*
4947                  * We don't need to lock updates but journal_flush() could
4948                  * otherwise be livelocked...
4949                  */
4950                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4951                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4952                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4953                 if (err)
4954                         return err;
4955         }
4956
4957         return dquot_quota_on(sb, type, format_id, path);
4958 }
4959
4960 static int ext4_quota_off(struct super_block *sb, int type)
4961 {
4962         struct inode *inode = sb_dqopt(sb)->files[type];
4963         handle_t *handle;
4964
4965         /* Force all delayed allocation blocks to be allocated.
4966          * Caller already holds s_umount sem */
4967         if (test_opt(sb, DELALLOC))
4968                 sync_filesystem(sb);
4969
4970         if (!inode)
4971                 goto out;
4972
4973         /* Update modification times of quota files when userspace can
4974          * start looking at them */
4975         handle = ext4_journal_start(inode, 1);
4976         if (IS_ERR(handle))
4977                 goto out;
4978         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4979         ext4_mark_inode_dirty(handle, inode);
4980         ext4_journal_stop(handle);
4981
4982 out:
4983         return dquot_quota_off(sb, type);
4984 }
4985
4986 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4987  * acquiring the locks... As quota files are never truncated and quota code
4988  * itself serializes the operations (and no one else should touch the files)
4989  * we don't have to be afraid of races */
4990 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4991                                size_t len, loff_t off)
4992 {
4993         struct inode *inode = sb_dqopt(sb)->files[type];
4994         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4995         int err = 0;
4996         int offset = off & (sb->s_blocksize - 1);
4997         int tocopy;
4998         size_t toread;
4999         struct buffer_head *bh;
5000         loff_t i_size = i_size_read(inode);
5001
5002         if (off > i_size)
5003                 return 0;
5004         if (off+len > i_size)
5005                 len = i_size-off;
5006         toread = len;
5007         while (toread > 0) {
5008                 tocopy = sb->s_blocksize - offset < toread ?
5009                                 sb->s_blocksize - offset : toread;
5010                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5011                 if (err)
5012                         return err;
5013                 if (!bh)        /* A hole? */
5014                         memset(data, 0, tocopy);
5015                 else
5016                         memcpy(data, bh->b_data+offset, tocopy);
5017                 brelse(bh);
5018                 offset = 0;
5019                 toread -= tocopy;
5020                 data += tocopy;
5021                 blk++;
5022         }
5023         return len;
5024 }
5025
5026 /* Write to quotafile (we know the transaction is already started and has
5027  * enough credits) */
5028 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5029                                 const char *data, size_t len, loff_t off)
5030 {
5031         struct inode *inode = sb_dqopt(sb)->files[type];
5032         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5033         int err = 0;
5034         int offset = off & (sb->s_blocksize - 1);
5035         struct buffer_head *bh;
5036         handle_t *handle = journal_current_handle();
5037
5038         if (EXT4_SB(sb)->s_journal && !handle) {
5039                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5040                         " cancelled because transaction is not started",
5041                         (unsigned long long)off, (unsigned long long)len);
5042                 return -EIO;
5043         }
5044         /*
5045          * Since we account only one data block in transaction credits,
5046          * then it is impossible to cross a block boundary.
5047          */
5048         if (sb->s_blocksize - offset < len) {
5049                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5050                         " cancelled because not block aligned",
5051                         (unsigned long long)off, (unsigned long long)len);
5052                 return -EIO;
5053         }
5054
5055         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
5056         bh = ext4_bread(handle, inode, blk, 1, &err);
5057         if (!bh)
5058                 goto out;
5059         err = ext4_journal_get_write_access(handle, bh);
5060         if (err) {
5061                 brelse(bh);
5062                 goto out;
5063         }
5064         lock_buffer(bh);
5065         memcpy(bh->b_data+offset, data, len);
5066         flush_dcache_page(bh->b_page);
5067         unlock_buffer(bh);
5068         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5069         brelse(bh);
5070 out:
5071         if (err) {
5072                 mutex_unlock(&inode->i_mutex);
5073                 return err;
5074         }
5075         if (inode->i_size < off + len) {
5076                 i_size_write(inode, off + len);
5077                 EXT4_I(inode)->i_disksize = inode->i_size;
5078                 ext4_mark_inode_dirty(handle, inode);
5079         }
5080         mutex_unlock(&inode->i_mutex);
5081         return len;
5082 }
5083
5084 #endif
5085
5086 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5087                        const char *dev_name, void *data)
5088 {
5089         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5090 }
5091
5092 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5093 static inline void register_as_ext2(void)
5094 {
5095         int err = register_filesystem(&ext2_fs_type);
5096         if (err)
5097                 printk(KERN_WARNING
5098                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5099 }
5100
5101 static inline void unregister_as_ext2(void)
5102 {
5103         unregister_filesystem(&ext2_fs_type);
5104 }
5105
5106 static inline int ext2_feature_set_ok(struct super_block *sb)
5107 {
5108         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5109                 return 0;
5110         if (sb->s_flags & MS_RDONLY)
5111                 return 1;
5112         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5113                 return 0;
5114         return 1;
5115 }
5116 MODULE_ALIAS("ext2");
5117 #else
5118 static inline void register_as_ext2(void) { }
5119 static inline void unregister_as_ext2(void) { }
5120 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5121 #endif
5122
5123 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5124 static inline void register_as_ext3(void)
5125 {
5126         int err = register_filesystem(&ext3_fs_type);
5127         if (err)
5128                 printk(KERN_WARNING
5129                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5130 }
5131
5132 static inline void unregister_as_ext3(void)
5133 {
5134         unregister_filesystem(&ext3_fs_type);
5135 }
5136
5137 static inline int ext3_feature_set_ok(struct super_block *sb)
5138 {
5139         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5140                 return 0;
5141         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5142                 return 0;
5143         if (sb->s_flags & MS_RDONLY)
5144                 return 1;
5145         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5146                 return 0;
5147         return 1;
5148 }
5149 MODULE_ALIAS("ext3");
5150 #else
5151 static inline void register_as_ext3(void) { }
5152 static inline void unregister_as_ext3(void) { }
5153 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5154 #endif
5155
5156 static struct file_system_type ext4_fs_type = {
5157         .owner          = THIS_MODULE,
5158         .name           = "ext4",
5159         .mount          = ext4_mount,
5160         .kill_sb        = kill_block_super,
5161         .fs_flags       = FS_REQUIRES_DEV,
5162 };
5163
5164 static int __init ext4_init_feat_adverts(void)
5165 {
5166         struct ext4_features *ef;
5167         int ret = -ENOMEM;
5168
5169         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5170         if (!ef)
5171                 goto out;
5172
5173         ef->f_kobj.kset = ext4_kset;
5174         init_completion(&ef->f_kobj_unregister);
5175         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5176                                    "features");
5177         if (ret) {
5178                 kfree(ef);
5179                 goto out;
5180         }
5181
5182         ext4_feat = ef;
5183         ret = 0;
5184 out:
5185         return ret;
5186 }
5187
5188 static void ext4_exit_feat_adverts(void)
5189 {
5190         kobject_put(&ext4_feat->f_kobj);
5191         wait_for_completion(&ext4_feat->f_kobj_unregister);
5192         kfree(ext4_feat);
5193 }
5194
5195 /* Shared across all ext4 file systems */
5196 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5197 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5198
5199 static int __init ext4_init_fs(void)
5200 {
5201         int i, err;
5202
5203         ext4_check_flag_values();
5204
5205         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5206                 mutex_init(&ext4__aio_mutex[i]);
5207                 init_waitqueue_head(&ext4__ioend_wq[i]);
5208         }
5209
5210         err = ext4_init_pageio();
5211         if (err)
5212                 return err;
5213         err = ext4_init_system_zone();
5214         if (err)
5215                 goto out6;
5216         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5217         if (!ext4_kset)
5218                 goto out5;
5219         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5220
5221         err = ext4_init_feat_adverts();
5222         if (err)
5223                 goto out4;
5224
5225         err = ext4_init_mballoc();
5226         if (err)
5227                 goto out3;
5228
5229         err = ext4_init_xattr();
5230         if (err)
5231                 goto out2;
5232         err = init_inodecache();
5233         if (err)
5234                 goto out1;
5235         register_as_ext3();
5236         register_as_ext2();
5237         err = register_filesystem(&ext4_fs_type);
5238         if (err)
5239                 goto out;
5240
5241         ext4_li_info = NULL;
5242         mutex_init(&ext4_li_mtx);
5243         return 0;
5244 out:
5245         unregister_as_ext2();
5246         unregister_as_ext3();
5247         destroy_inodecache();
5248 out1:
5249         ext4_exit_xattr();
5250 out2:
5251         ext4_exit_mballoc();
5252 out3:
5253         ext4_exit_feat_adverts();
5254 out4:
5255         if (ext4_proc_root)
5256                 remove_proc_entry("fs/ext4", NULL);
5257         kset_unregister(ext4_kset);
5258 out5:
5259         ext4_exit_system_zone();
5260 out6:
5261         ext4_exit_pageio();
5262         return err;
5263 }
5264
5265 static void __exit ext4_exit_fs(void)
5266 {
5267         ext4_destroy_lazyinit_thread();
5268         unregister_as_ext2();
5269         unregister_as_ext3();
5270         unregister_filesystem(&ext4_fs_type);
5271         destroy_inodecache();
5272         ext4_exit_xattr();
5273         ext4_exit_mballoc();
5274         ext4_exit_feat_adverts();
5275         remove_proc_entry("fs/ext4", NULL);
5276         kset_unregister(ext4_kset);
5277         ext4_exit_system_zone();
5278         ext4_exit_pageio();
5279 }
5280
5281 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5282 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5283 MODULE_LICENSE("GPL");
5284 module_init(ext4_init_fs)
5285 module_exit(ext4_exit_fs)