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                 panic("EXT4-fs (device %s): panic forced after error\n",
 468                         sb->s_id);
 469 }
 470
 471 void __ext4_error(struct super_block *sb, const char *function,
 472                   unsigned int line, const char *fmt, ...)
 473 {
 474         struct va_format vaf;
 475         va_list args;
 476
 477         va_start(args, fmt);
 478         vaf.fmt = fmt;
 479         vaf.va = &args;
 480         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 481                sb->s_id, function, line, current->comm, &vaf);
 482         va_end(args);
 483
 484         ext4_handle_error(sb);
 485 }
 486
 487 void ext4_error_inode(struct inode *inode, const char *function,
 488                       unsigned int line, ext4_fsblk_t block,
 489                       const char *fmt, ...)
 490 {
 491         va_list args;
 492         struct va_format vaf;
 493         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 494
 495         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 496         es->s_last_error_block = cpu_to_le64(block);
 497         save_error_info(inode->i_sb, function, line);
 498         va_start(args, fmt);
 499         vaf.fmt = fmt;
 500         vaf.va = &args;
 501         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 502                inode->i_sb->s_id, function, line, inode->i_ino);
 503         if (block)
 504                 printk(KERN_CONT "block %llu: ", block);
 505         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 506         va_end(args);
 507
 508         ext4_handle_error(inode->i_sb);
 509 }
 510
 511 void ext4_error_file(struct file *file, const char *function,
 512                      unsigned int line, ext4_fsblk_t block,
 513                      const char *fmt, ...)
 514 {
 515         va_list args;
 516         struct va_format vaf;
 517         struct ext4_super_block *es;
 518         struct inode *inode = file->f_dentry->d_inode;
 519         char pathname[80], *path;
 520
 521         es = EXT4_SB(inode->i_sb)->s_es;
 522         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 523         save_error_info(inode->i_sb, function, line);
 524         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 525         if (IS_ERR(path))
 526                 path = "(unknown)";
 527         printk(KERN_CRIT
 528                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 529                inode->i_sb->s_id, function, line, inode->i_ino);
 530         if (block)
 531                 printk(KERN_CONT "block %llu: ", block);
 532         va_start(args, fmt);
 533         vaf.fmt = fmt;
 534         vaf.va = &args;
 535         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 536         va_end(args);
 537
 538         ext4_handle_error(inode->i_sb);
 539 }
 540
 541 static const char *ext4_decode_error(struct super_block *sb, int errno,
 542                                      char nbuf[16])
 543 {
 544         char *errstr = NULL;
 545
 546         switch (errno) {
 547         case -EIO:
 548                 errstr = "IO failure";
 549                 break;
 550         case -ENOMEM:
 551                 errstr = "Out of memory";
 552                 break;
 553         case -EROFS:
 554                 if (!sb || (EXT4_SB(sb)->s_journal &&
 555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 556                         errstr = "Journal has aborted";
 557                 else
 558                         errstr = "Readonly filesystem";
 559                 break;
 560         default:
 561                 /* If the caller passed in an extra buffer for unknown
 562                  * errors, textualise them now.  Else we just return
 563                  * NULL. */
 564                 if (nbuf) {
 565                         /* Check for truncated error codes... */
 566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 567                                 errstr = nbuf;
 568                 }
 569                 break;
 570         }
 571
 572         return errstr;
 573 }
 574
 575 /* __ext4_std_error decodes expected errors from journaling functions
 576  * automatically and invokes the appropriate error response.  */
 577
 578 void __ext4_std_error(struct super_block *sb, const char *function,
 579                       unsigned int line, int errno)
 580 {
 581         char nbuf[16];
 582         const char *errstr;
 583
 584         /* Special case: if the error is EROFS, and we're not already
 585          * inside a transaction, then there's really no point in logging
 586          * an error. */
 587         if (errno == -EROFS && journal_current_handle() == NULL &&
 588             (sb->s_flags & MS_RDONLY))
 589                 return;
 590
 591         errstr = ext4_decode_error(sb, errno, nbuf);
 592         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 593                sb->s_id, function, line, errstr);
 594         save_error_info(sb, function, line);
 595
 596         ext4_handle_error(sb);
 597 }
 598
 599 /*
 600  * ext4_abort is a much stronger failure handler than ext4_error.  The
 601  * abort function may be used to deal with unrecoverable failures such
 602  * as journal IO errors or ENOMEM at a critical moment in log management.
 603  *
 604  * We unconditionally force the filesystem into an ABORT|READONLY state,
 605  * unless the error response on the fs has been set to panic in which
 606  * case we take the easy way out and panic immediately.
 607  */
 608
 609 void __ext4_abort(struct super_block *sb, const char *function,
 610                 unsigned int line, const char *fmt, ...)
 611 {
 612         va_list args;
 613
 614         save_error_info(sb, function, line);
 615         va_start(args, fmt);
 616         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 617                function, line);
 618         vprintk(fmt, args);
 619         printk("\n");
 620         va_end(args);
 621
 622         if ((sb->s_flags & MS_RDONLY) == 0) {
 623                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 624                 sb->s_flags |= MS_RDONLY;
 625                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 626                 if (EXT4_SB(sb)->s_journal)
 627                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 628                 save_error_info(sb, function, line);
 629         }
 630         if (test_opt(sb, ERRORS_PANIC))
 631                 panic("EXT4-fs panic from previous error\n");
 632 }
 633
 634 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 635 {
 636         struct va_format vaf;
 637         va_list args;
 638
 639         va_start(args, fmt);
 640         vaf.fmt = fmt;
 641         vaf.va = &args;
 642         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 643         va_end(args);
 644 }
 645
 646 void __ext4_warning(struct super_block *sb, const char *function,
 647                     unsigned int line, const char *fmt, ...)
 648 {
 649         struct va_format vaf;
 650         va_list args;
 651
 652         va_start(args, fmt);
 653         vaf.fmt = fmt;
 654         vaf.va = &args;
 655         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 656                sb->s_id, function, line, &vaf);
 657         va_end(args);
 658 }
 659
 660 void __ext4_grp_locked_error(const char *function, unsigned int line,
 661                              struct super_block *sb, ext4_group_t grp,
 662                              unsigned long ino, ext4_fsblk_t block,
 663                              const char *fmt, ...)
 664 __releases(bitlock)
 665 __acquires(bitlock)
 666 {
 667         struct va_format vaf;
 668         va_list args;
 669         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 670
 671         es->s_last_error_ino = cpu_to_le32(ino);
 672         es->s_last_error_block = cpu_to_le64(block);
 673         __save_error_info(sb, function, line);
 674
 675         va_start(args, fmt);
 676
 677         vaf.fmt = fmt;
 678         vaf.va = &args;
 679         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 680                sb->s_id, function, line, grp);
 681         if (ino)
 682                 printk(KERN_CONT "inode %lu: ", ino);
 683         if (block)
 684                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 685         printk(KERN_CONT "%pV\n", &vaf);
 686         va_end(args);
 687
 688         if (test_opt(sb, ERRORS_CONT)) {
 689                 ext4_commit_super(sb, 0);
 690                 return;
 691         }
 692
 693         ext4_unlock_group(sb, grp);
 694         ext4_handle_error(sb);
 695         /*
 696          * We only get here in the ERRORS_RO case; relocking the group
 697          * may be dangerous, but nothing bad will happen since the
 698          * filesystem will have already been marked read/only and the
 699          * journal has been aborted.  We return 1 as a hint to callers
 700          * who might what to use the return value from
 701          * ext4_grp_locked_error() to distinguish between the
 702          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 703          * aggressively from the ext4 function in question, with a
 704          * more appropriate error code.
 705          */
 706         ext4_lock_group(sb, grp);
 707         return;
 708 }
 709
 710 void ext4_update_dynamic_rev(struct super_block *sb)
 711 {
 712         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 713
 714         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 715                 return;
 716
 717         ext4_warning(sb,
 718                      "updating to rev %d because of new feature flag, "
 719                      "running e2fsck is recommended",
 720                      EXT4_DYNAMIC_REV);
 721
 722         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 723         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 724         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 725         /* leave es->s_feature_*compat flags alone */
 726         /* es->s_uuid will be set by e2fsck if empty */
 727
 728         /*
 729          * The rest of the superblock fields should be zero, and if not it
 730          * means they are likely already in use, so leave them alone.  We
 731          * can leave it up to e2fsck to clean up any inconsistencies there.
 732          */
 733 }
 734
 735 /*
 736  * Open the external journal device
 737  */
 738 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 739 {
 740         struct block_device *bdev;
 741         char b[BDEVNAME_SIZE];
 742
 743         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 744         if (IS_ERR(bdev))
 745                 goto fail;
 746         return bdev;
 747
 748 fail:
 749         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 750                         __bdevname(dev, b), PTR_ERR(bdev));
 751         return NULL;
 752 }
 753
 754 /*
 755  * Release the journal device
 756  */
 757 static int ext4_blkdev_put(struct block_device *bdev)
 758 {
 759         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 760 }
 761
 762 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 763 {
 764         struct block_device *bdev;
 765         int ret = -ENODEV;
 766
 767         bdev = sbi->journal_bdev;
 768         if (bdev) {
 769                 ret = ext4_blkdev_put(bdev);
 770                 sbi->journal_bdev = NULL;
 771         }
 772         return ret;
 773 }
 774
 775 static inline struct inode *orphan_list_entry(struct list_head *l)
 776 {
 777         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 778 }
 779
 780 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 781 {
 782         struct list_head *l;
 783
 784         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 785                  le32_to_cpu(sbi->s_es->s_last_orphan));
 786
 787         printk(KERN_ERR "sb_info orphan list:\n");
 788         list_for_each(l, &sbi->s_orphan) {
 789                 struct inode *inode = orphan_list_entry(l);
 790                 printk(KERN_ERR "  "
 791                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 792                        inode->i_sb->s_id, inode->i_ino, inode,
 793                        inode->i_mode, inode->i_nlink,
 794                        NEXT_ORPHAN(inode));
 795         }
 796 }
 797
 798 static void ext4_put_super(struct super_block *sb)
 799 {
 800         struct ext4_sb_info *sbi = EXT4_SB(sb);
 801         struct ext4_super_block *es = sbi->s_es;
 802         int i, err;
 803
 804         ext4_unregister_li_request(sb);
 805         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 806
 807         flush_workqueue(sbi->dio_unwritten_wq);
 808         destroy_workqueue(sbi->dio_unwritten_wq);
 809
 810         lock_super(sb);
 811         if (sb->s_dirt)
 812                 ext4_commit_super(sb, 1);
 813
 814         if (sbi->s_journal) {
 815                 err = jbd2_journal_destroy(sbi->s_journal);
 816                 sbi->s_journal = NULL;
 817                 if (err < 0)
 818                         ext4_abort(sb, "Couldn't clean up the journal");
 819         }
 820
 821         del_timer(&sbi->s_err_report);
 822         ext4_release_system_zone(sb);
 823         ext4_mb_release(sb);
 824         ext4_ext_release(sb);
 825         ext4_xattr_put_super(sb);
 826
 827         if (!(sb->s_flags & MS_RDONLY)) {
 828                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 829                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 830                 ext4_commit_super(sb, 1);
 831         }
 832         if (sbi->s_proc) {
 833                 remove_proc_entry(sb->s_id, ext4_proc_root);
 834         }
 835         kobject_del(&sbi->s_kobj);
 836
 837         for (i = 0; i < sbi->s_gdb_count; i++)
 838                 brelse(sbi->s_group_desc[i]);
 839         ext4_kvfree(sbi->s_group_desc);
 840         ext4_kvfree(sbi->s_flex_groups);
 841         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 842         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 843         percpu_counter_destroy(&sbi->s_dirs_counter);
 844         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 845         brelse(sbi->s_sbh);
 846 #ifdef CONFIG_QUOTA
 847         for (i = 0; i < MAXQUOTAS; i++)
 848                 kfree(sbi->s_qf_names[i]);
 849 #endif
 850
 851         /* Debugging code just in case the in-memory inode orphan list
 852          * isn't empty.  The on-disk one can be non-empty if we've
 853          * detected an error and taken the fs readonly, but the
 854          * in-memory list had better be clean by this point. */
 855         if (!list_empty(&sbi->s_orphan))
 856                 dump_orphan_list(sb, sbi);
 857         J_ASSERT(list_empty(&sbi->s_orphan));
 858
 859         invalidate_bdev(sb->s_bdev);
 860         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 861                 /*
 862                  * Invalidate the journal device's buffers.  We don't want them
 863                  * floating about in memory - the physical journal device may
 864                  * hotswapped, and it breaks the `ro-after' testing code.
 865                  */
 866                 sync_blockdev(sbi->journal_bdev);
 867                 invalidate_bdev(sbi->journal_bdev);
 868                 ext4_blkdev_remove(sbi);
 869         }
 870         if (sbi->s_mmp_tsk)
 871                 kthread_stop(sbi->s_mmp_tsk);
 872         sb->s_fs_info = NULL;
 873         /*
 874          * Now that we are completely done shutting down the
 875          * superblock, we need to actually destroy the kobject.
 876          */
 877         unlock_super(sb);
 878         kobject_put(&sbi->s_kobj);
 879         wait_for_completion(&sbi->s_kobj_unregister);
 880         kfree(sbi->s_blockgroup_lock);
 881         kfree(sbi);
 882 }
 883
 884 static struct kmem_cache *ext4_inode_cachep;
 885
 886 /*
 887  * Called inside transaction, so use GFP_NOFS
 888  */
 889 static struct inode *ext4_alloc_inode(struct super_block *sb)
 890 {
 891         struct ext4_inode_info *ei;
 892
 893         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 894         if (!ei)
 895                 return NULL;
 896
 897         ei->vfs_inode.i_version = 1;
 898         ei->vfs_inode.i_data.writeback_index = 0;
 899         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 900         INIT_LIST_HEAD(&ei->i_prealloc_list);
 901         spin_lock_init(&ei->i_prealloc_lock);
 902         ei->i_reserved_data_blocks = 0;
 903         ei->i_reserved_meta_blocks = 0;
 904         ei->i_allocated_meta_blocks = 0;
 905         ei->i_da_metadata_calc_len = 0;
 906         spin_lock_init(&(ei->i_block_reservation_lock));
 907 #ifdef CONFIG_QUOTA
 908         ei->i_reserved_quota = 0;
 909 #endif
 910         ei->jinode = NULL;
 911         INIT_LIST_HEAD(&ei->i_completed_io_list);
 912         spin_lock_init(&ei->i_completed_io_lock);
 913         ei->cur_aio_dio = NULL;
 914         ei->i_sync_tid = 0;
 915         ei->i_datasync_tid = 0;
 916         atomic_set(&ei->i_ioend_count, 0);
 917         atomic_set(&ei->i_aiodio_unwritten, 0);
 918
 919         return &ei->vfs_inode;
 920 }
 921
 922 static int ext4_drop_inode(struct inode *inode)
 923 {
 924         int drop = generic_drop_inode(inode);
 925
 926         trace_ext4_drop_inode(inode, drop);
 927         return drop;
 928 }
 929
 930 static void ext4_i_callback(struct rcu_head *head)
 931 {
 932         struct inode *inode = container_of(head, struct inode, i_rcu);
 933         INIT_LIST_HEAD(&inode->i_dentry);
 934         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 935 }
 936
 937 static void ext4_destroy_inode(struct inode *inode)
 938 {
 939         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 940                 ext4_msg(inode->i_sb, KERN_ERR,
 941                          "Inode %lu (%p): orphan list check failed!",
 942                          inode->i_ino, EXT4_I(inode));
 943                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 944                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 945                                 true);
 946                 dump_stack();
 947         }
 948         call_rcu(&inode->i_rcu, ext4_i_callback);
 949 }
 950
 951 static void init_once(void *foo)
 952 {
 953         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 954
 955         INIT_LIST_HEAD(&ei->i_orphan);
 956 #ifdef CONFIG_EXT4_FS_XATTR
 957         init_rwsem(&ei->xattr_sem);
 958 #endif
 959         init_rwsem(&ei->i_data_sem);
 960         inode_init_once(&ei->vfs_inode);
 961 }
 962
 963 static int init_inodecache(void)
 964 {
 965         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 966                                              sizeof(struct ext4_inode_info),
 967                                              0, (SLAB_RECLAIM_ACCOUNT|
 968                                                 SLAB_MEM_SPREAD),
 969                                              init_once);
 970         if (ext4_inode_cachep == NULL)
 971                 return -ENOMEM;
 972         return 0;
 973 }
 974
 975 static void destroy_inodecache(void)
 976 {
 977         kmem_cache_destroy(ext4_inode_cachep);
 978 }
 979
 980 void ext4_clear_inode(struct inode *inode)
 981 {
 982         invalidate_inode_buffers(inode);
 983         end_writeback(inode);
 984         dquot_drop(inode);
 985         ext4_discard_preallocations(inode);
 986         if (EXT4_I(inode)->jinode) {
 987                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 988                                                EXT4_I(inode)->jinode);
 989                 jbd2_free_inode(EXT4_I(inode)->jinode);
 990                 EXT4_I(inode)->jinode = NULL;
 991         }
 992 }
 993
 994 static inline void ext4_show_quota_options(struct seq_file *seq,
 995                                            struct super_block *sb)
 996 {
 997 #if defined(CONFIG_QUOTA)
 998         struct ext4_sb_info *sbi = EXT4_SB(sb);
 999
1000         if (sbi->s_jquota_fmt) {
1001                 char *fmtname = "";
1002
1003                 switch (sbi->s_jquota_fmt) {
1004                 case QFMT_VFS_OLD:
1005                         fmtname = "vfsold";
1006                         break;
1007                 case QFMT_VFS_V0:
1008                         fmtname = "vfsv0";
1009                         break;
1010                 case QFMT_VFS_V1:
1011                         fmtname = "vfsv1";
1012                         break;
1013                 }
1014                 seq_printf(seq, ",jqfmt=%s", fmtname);
1015         }
1016
1017         if (sbi->s_qf_names[USRQUOTA])
1018                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1019
1020         if (sbi->s_qf_names[GRPQUOTA])
1021                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1022
1023         if (test_opt(sb, USRQUOTA))
1024                 seq_puts(seq, ",usrquota");
1025
1026         if (test_opt(sb, GRPQUOTA))
1027                 seq_puts(seq, ",grpquota");
1028 #endif
1029 }
1030
1031 /*
1032  * Show an option if
1033  *  - it's set to a non-default value OR
1034  *  - if the per-sb default is different from the global default
1035  */
1036 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1037 {
1038         int def_errors;
1039         unsigned long def_mount_opts;
1040         struct super_block *sb = vfs->mnt_sb;
1041         struct ext4_sb_info *sbi = EXT4_SB(sb);
1042         struct ext4_super_block *es = sbi->s_es;
1043
1044         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1045         def_errors     = le16_to_cpu(es->s_errors);
1046
1047         if (sbi->s_sb_block != 1)
1048                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1049         if (test_opt(sb, MINIX_DF))
1050                 seq_puts(seq, ",minixdf");
1051         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1052                 seq_puts(seq, ",grpid");
1053         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1054                 seq_puts(seq, ",nogrpid");
1055         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1056             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1057                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1058         }
1059         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1060             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1061                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1062         }
1063         if (test_opt(sb, ERRORS_RO)) {
1064                 if (def_errors == EXT4_ERRORS_PANIC ||
1065                     def_errors == EXT4_ERRORS_CONTINUE) {
1066                         seq_puts(seq, ",errors=remount-ro");
1067                 }
1068         }
1069         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1070                 seq_puts(seq, ",errors=continue");
1071         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1072                 seq_puts(seq, ",errors=panic");
1073         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1074                 seq_puts(seq, ",nouid32");
1075         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1076                 seq_puts(seq, ",debug");
1077 #ifdef CONFIG_EXT4_FS_XATTR
1078         if (test_opt(sb, XATTR_USER))
1079                 seq_puts(seq, ",user_xattr");
1080         if (!test_opt(sb, XATTR_USER))
1081                 seq_puts(seq, ",nouser_xattr");
1082 #endif
1083 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1084         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1085                 seq_puts(seq, ",acl");
1086         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1087                 seq_puts(seq, ",noacl");
1088 #endif
1089         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1090                 seq_printf(seq, ",commit=%u",
1091                            (unsigned) (sbi->s_commit_interval / HZ));
1092         }
1093         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1094                 seq_printf(seq, ",min_batch_time=%u",
1095                            (unsigned) sbi->s_min_batch_time);
1096         }
1097         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1098                 seq_printf(seq, ",max_batch_time=%u",
1099                            (unsigned) sbi->s_min_batch_time);
1100         }
1101
1102         /*
1103          * We're changing the default of barrier mount option, so
1104          * let's always display its mount state so it's clear what its
1105          * status is.
1106          */
1107         seq_puts(seq, ",barrier=");
1108         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1109         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1110                 seq_puts(seq, ",journal_async_commit");
1111         else if (test_opt(sb, JOURNAL_CHECKSUM))
1112                 seq_puts(seq, ",journal_checksum");
1113         if (test_opt(sb, I_VERSION))
1114                 seq_puts(seq, ",i_version");
1115         if (!test_opt(sb, DELALLOC) &&
1116             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1117                 seq_puts(seq, ",nodelalloc");
1118
1119         if (!test_opt(sb, MBLK_IO_SUBMIT))
1120                 seq_puts(seq, ",nomblk_io_submit");
1121         if (sbi->s_stripe)
1122                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1123         /*
1124          * journal mode get enabled in different ways
1125          * So just print the value even if we didn't specify it
1126          */
1127         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1128                 seq_puts(seq, ",data=journal");
1129         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1130                 seq_puts(seq, ",data=ordered");
1131         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1132                 seq_puts(seq, ",data=writeback");
1133
1134         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1135                 seq_printf(seq, ",inode_readahead_blks=%u",
1136                            sbi->s_inode_readahead_blks);
1137
1138         if (test_opt(sb, DATA_ERR_ABORT))
1139                 seq_puts(seq, ",data_err=abort");
1140
1141         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1142                 seq_puts(seq, ",noauto_da_alloc");
1143
1144         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1145                 seq_puts(seq, ",discard");
1146
1147         if (test_opt(sb, NOLOAD))
1148                 seq_puts(seq, ",norecovery");
1149
1150         if (test_opt(sb, DIOREAD_NOLOCK))
1151                 seq_puts(seq, ",dioread_nolock");
1152
1153         if (test_opt(sb, BLOCK_VALIDITY) &&
1154             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1155                 seq_puts(seq, ",block_validity");
1156
1157         if (!test_opt(sb, INIT_INODE_TABLE))
1158                 seq_puts(seq, ",noinit_itable");
1159         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1160                 seq_printf(seq, ",init_itable=%u",
1161                            (unsigned) sbi->s_li_wait_mult);
1162
1163         ext4_show_quota_options(seq, sb);
1164
1165         return 0;
1166 }
1167
1168 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1169                                         u64 ino, u32 generation)
1170 {
1171         struct inode *inode;
1172
1173         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1174                 return ERR_PTR(-ESTALE);
1175         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1176                 return ERR_PTR(-ESTALE);
1177
1178         /* iget isn't really right if the inode is currently unallocated!!
1179          *
1180          * ext4_read_inode will return a bad_inode if the inode had been
1181          * deleted, so we should be safe.
1182          *
1183          * Currently we don't know the generation for parent directory, so
1184          * a generation of 0 means "accept any"
1185          */
1186         inode = ext4_iget(sb, ino);
1187         if (IS_ERR(inode))
1188                 return ERR_CAST(inode);
1189         if (generation && inode->i_generation != generation) {
1190                 iput(inode);
1191                 return ERR_PTR(-ESTALE);
1192         }
1193
1194         return inode;
1195 }
1196
1197 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1198                                         int fh_len, int fh_type)
1199 {
1200         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1201                                     ext4_nfs_get_inode);
1202 }
1203
1204 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1205                                         int fh_len, int fh_type)
1206 {
1207         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1208                                     ext4_nfs_get_inode);
1209 }
1210
1211 /*
1212  * Try to release metadata pages (indirect blocks, directories) which are
1213  * mapped via the block device.  Since these pages could have journal heads
1214  * which would prevent try_to_free_buffers() from freeing them, we must use
1215  * jbd2 layer's try_to_free_buffers() function to release them.
1216  */
1217 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1218                                  gfp_t wait)
1219 {
1220         journal_t *journal = EXT4_SB(sb)->s_journal;
1221
1222         WARN_ON(PageChecked(page));
1223         if (!page_has_buffers(page))
1224                 return 0;
1225         if (journal)
1226                 return jbd2_journal_try_to_free_buffers(journal, page,
1227                                                         wait & ~__GFP_WAIT);
1228         return try_to_free_buffers(page);
1229 }
1230
1231 #ifdef CONFIG_QUOTA
1232 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1233 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1234
1235 static int ext4_write_dquot(struct dquot *dquot);
1236 static int ext4_acquire_dquot(struct dquot *dquot);
1237 static int ext4_release_dquot(struct dquot *dquot);
1238 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1239 static int ext4_write_info(struct super_block *sb, int type);
1240 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1241                          struct path *path);
1242 static int ext4_quota_off(struct super_block *sb, int type);
1243 static int ext4_quota_on_mount(struct super_block *sb, int type);
1244 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1245                                size_t len, loff_t off);
1246 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1247                                 const char *data, size_t len, loff_t off);
1248
1249 static const struct dquot_operations ext4_quota_operations = {
1250         .get_reserved_space = ext4_get_reserved_space,
1251         .write_dquot    = ext4_write_dquot,
1252         .acquire_dquot  = ext4_acquire_dquot,
1253         .release_dquot  = ext4_release_dquot,
1254         .mark_dirty     = ext4_mark_dquot_dirty,
1255         .write_info     = ext4_write_info,
1256         .alloc_dquot    = dquot_alloc,
1257         .destroy_dquot  = dquot_destroy,
1258 };
1259
1260 static const struct quotactl_ops ext4_qctl_operations = {
1261         .quota_on       = ext4_quota_on,
1262         .quota_off      = ext4_quota_off,
1263         .quota_sync     = dquot_quota_sync,
1264         .get_info       = dquot_get_dqinfo,
1265         .set_info       = dquot_set_dqinfo,
1266         .get_dqblk      = dquot_get_dqblk,
1267         .set_dqblk      = dquot_set_dqblk
1268 };
1269 #endif
1270
1271 static const struct super_operations ext4_sops = {
1272         .alloc_inode    = ext4_alloc_inode,
1273         .destroy_inode  = ext4_destroy_inode,
1274         .write_inode    = ext4_write_inode,
1275         .dirty_inode    = ext4_dirty_inode,
1276         .drop_inode     = ext4_drop_inode,
1277         .evict_inode    = ext4_evict_inode,
1278         .put_super      = ext4_put_super,
1279         .sync_fs        = ext4_sync_fs,
1280         .freeze_fs      = ext4_freeze,
1281         .unfreeze_fs    = ext4_unfreeze,
1282         .statfs         = ext4_statfs,
1283         .remount_fs     = ext4_remount,
1284         .show_options   = ext4_show_options,
1285 #ifdef CONFIG_QUOTA
1286         .quota_read     = ext4_quota_read,
1287         .quota_write    = ext4_quota_write,
1288 #endif
1289         .bdev_try_to_free_page = bdev_try_to_free_page,
1290 };
1291
1292 static const struct super_operations ext4_nojournal_sops = {
1293         .alloc_inode    = ext4_alloc_inode,
1294         .destroy_inode  = ext4_destroy_inode,
1295         .write_inode    = ext4_write_inode,
1296         .dirty_inode    = ext4_dirty_inode,
1297         .drop_inode     = ext4_drop_inode,
1298         .evict_inode    = ext4_evict_inode,
1299         .write_super    = ext4_write_super,
1300         .put_super      = ext4_put_super,
1301         .statfs         = ext4_statfs,
1302         .remount_fs     = ext4_remount,
1303         .show_options   = ext4_show_options,
1304 #ifdef CONFIG_QUOTA
1305         .quota_read     = ext4_quota_read,
1306         .quota_write    = ext4_quota_write,
1307 #endif
1308         .bdev_try_to_free_page = bdev_try_to_free_page,
1309 };
1310
1311 static const struct export_operations ext4_export_ops = {
1312         .fh_to_dentry = ext4_fh_to_dentry,
1313         .fh_to_parent = ext4_fh_to_parent,
1314         .get_parent = ext4_get_parent,
1315 };
1316
1317 enum {
1318         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1319         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1320         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1321         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1322         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1323         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1324         Opt_journal_update, Opt_journal_dev,
1325         Opt_journal_checksum, Opt_journal_async_commit,
1326         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1327         Opt_data_err_abort, Opt_data_err_ignore,
1328         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1329         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1330         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1331         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1332         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1333         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1334         Opt_inode_readahead_blks, Opt_journal_ioprio,
1335         Opt_dioread_nolock, Opt_dioread_lock,
1336         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1337 };
1338
1339 static const match_table_t tokens = {
1340         {Opt_bsd_df, "bsddf"},
1341         {Opt_minix_df, "minixdf"},
1342         {Opt_grpid, "grpid"},
1343         {Opt_grpid, "bsdgroups"},
1344         {Opt_nogrpid, "nogrpid"},
1345         {Opt_nogrpid, "sysvgroups"},
1346         {Opt_resgid, "resgid=%u"},
1347         {Opt_resuid, "resuid=%u"},
1348         {Opt_sb, "sb=%u"},
1349         {Opt_err_cont, "errors=continue"},
1350         {Opt_err_panic, "errors=panic"},
1351         {Opt_err_ro, "errors=remount-ro"},
1352         {Opt_nouid32, "nouid32"},
1353         {Opt_debug, "debug"},
1354         {Opt_oldalloc, "oldalloc"},
1355         {Opt_orlov, "orlov"},
1356         {Opt_user_xattr, "user_xattr"},
1357         {Opt_nouser_xattr, "nouser_xattr"},
1358         {Opt_acl, "acl"},
1359         {Opt_noacl, "noacl"},
1360         {Opt_noload, "noload"},
1361         {Opt_noload, "norecovery"},
1362         {Opt_nobh, "nobh"},
1363         {Opt_bh, "bh"},
1364         {Opt_commit, "commit=%u"},
1365         {Opt_min_batch_time, "min_batch_time=%u"},
1366         {Opt_max_batch_time, "max_batch_time=%u"},
1367         {Opt_journal_update, "journal=update"},
1368         {Opt_journal_dev, "journal_dev=%u"},
1369         {Opt_journal_checksum, "journal_checksum"},
1370         {Opt_journal_async_commit, "journal_async_commit"},
1371         {Opt_abort, "abort"},
1372         {Opt_data_journal, "data=journal"},
1373         {Opt_data_ordered, "data=ordered"},
1374         {Opt_data_writeback, "data=writeback"},
1375         {Opt_data_err_abort, "data_err=abort"},
1376         {Opt_data_err_ignore, "data_err=ignore"},
1377         {Opt_offusrjquota, "usrjquota="},
1378         {Opt_usrjquota, "usrjquota=%s"},
1379         {Opt_offgrpjquota, "grpjquota="},
1380         {Opt_grpjquota, "grpjquota=%s"},
1381         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1382         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1383         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1384         {Opt_grpquota, "grpquota"},
1385         {Opt_noquota, "noquota"},
1386         {Opt_quota, "quota"},
1387         {Opt_usrquota, "usrquota"},
1388         {Opt_barrier, "barrier=%u"},
1389         {Opt_barrier, "barrier"},
1390         {Opt_nobarrier, "nobarrier"},
1391         {Opt_i_version, "i_version"},
1392         {Opt_stripe, "stripe=%u"},
1393         {Opt_resize, "resize"},
1394         {Opt_delalloc, "delalloc"},
1395         {Opt_nodelalloc, "nodelalloc"},
1396         {Opt_mblk_io_submit, "mblk_io_submit"},
1397         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1398         {Opt_block_validity, "block_validity"},
1399         {Opt_noblock_validity, "noblock_validity"},
1400         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1401         {Opt_journal_ioprio, "journal_ioprio=%u"},
1402         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1403         {Opt_auto_da_alloc, "auto_da_alloc"},
1404         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1405         {Opt_dioread_nolock, "dioread_nolock"},
1406         {Opt_dioread_lock, "dioread_lock"},
1407         {Opt_discard, "discard"},
1408         {Opt_nodiscard, "nodiscard"},
1409         {Opt_init_itable, "init_itable=%u"},
1410         {Opt_init_itable, "init_itable"},
1411         {Opt_noinit_itable, "noinit_itable"},
1412         {Opt_err, NULL},
1413 };
1414
1415 static ext4_fsblk_t get_sb_block(void **data)
1416 {
1417         ext4_fsblk_t    sb_block;
1418         char            *options = (char *) *data;
1419
1420         if (!options || strncmp(options, "sb=", 3) != 0)
1421                 return 1;       /* Default location */
1422
1423         options += 3;
1424         /* TODO: use simple_strtoll with >32bit ext4 */
1425         sb_block = simple_strtoul(options, &options, 0);
1426         if (*options && *options != ',') {
1427                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1428                        (char *) *data);
1429                 return 1;
1430         }
1431         if (*options == ',')
1432                 options++;
1433         *data = (void *) options;
1434
1435         return sb_block;
1436 }
1437
1438 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1439 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1440         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1441
1442 #ifdef CONFIG_QUOTA
1443 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1444 {
1445         struct ext4_sb_info *sbi = EXT4_SB(sb);
1446         char *qname;
1447
1448         if (sb_any_quota_loaded(sb) &&
1449                 !sbi->s_qf_names[qtype]) {
1450                 ext4_msg(sb, KERN_ERR,
1451                         "Cannot change journaled "
1452                         "quota options when quota turned on");
1453                 return 0;
1454         }
1455         qname = match_strdup(args);
1456         if (!qname) {
1457                 ext4_msg(sb, KERN_ERR,
1458                         "Not enough memory for storing quotafile name");
1459                 return 0;
1460         }
1461         if (sbi->s_qf_names[qtype] &&
1462                 strcmp(sbi->s_qf_names[qtype], qname)) {
1463                 ext4_msg(sb, KERN_ERR,
1464                         "%s quota file already specified", QTYPE2NAME(qtype));
1465                 kfree(qname);
1466                 return 0;
1467         }
1468         sbi->s_qf_names[qtype] = qname;
1469         if (strchr(sbi->s_qf_names[qtype], '/')) {
1470                 ext4_msg(sb, KERN_ERR,
1471                         "quotafile must be on filesystem root");
1472                 kfree(sbi->s_qf_names[qtype]);
1473                 sbi->s_qf_names[qtype] = NULL;
1474                 return 0;
1475         }
1476         set_opt(sb, QUOTA);
1477         return 1;
1478 }
1479
1480 static int clear_qf_name(struct super_block *sb, int qtype)
1481 {
1482
1483         struct ext4_sb_info *sbi = EXT4_SB(sb);
1484
1485         if (sb_any_quota_loaded(sb) &&
1486                 sbi->s_qf_names[qtype]) {
1487                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1488                         " when quota turned on");
1489                 return 0;
1490         }
1491         /*
1492          * The space will be released later when all options are confirmed
1493          * to be correct
1494          */
1495         sbi->s_qf_names[qtype] = NULL;
1496         return 1;
1497 }
1498 #endif
1499
1500 static int parse_options(char *options, struct super_block *sb,
1501                          unsigned long *journal_devnum,
1502                          unsigned int *journal_ioprio,
1503                          ext4_fsblk_t *n_blocks_count, int is_remount)
1504 {
1505         struct ext4_sb_info *sbi = EXT4_SB(sb);
1506         char *p;
1507         substring_t args[MAX_OPT_ARGS];
1508         int data_opt = 0;
1509         int option;
1510 #ifdef CONFIG_QUOTA
1511         int qfmt;
1512 #endif
1513
1514         if (!options)
1515                 return 1;
1516
1517         while ((p = strsep(&options, ",")) != NULL) {
1518                 int token;
1519                 if (!*p)
1520                         continue;
1521
1522                 /*
1523                  * Initialize args struct so we know whether arg was
1524                  * found; some options take optional arguments.
1525                  */
1526                 args[0].to = args[0].from = NULL;
1527                 token = match_token(p, tokens, args);
1528                 switch (token) {
1529                 case Opt_bsd_df:
1530                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1531                         clear_opt(sb, MINIX_DF);
1532                         break;
1533                 case Opt_minix_df:
1534                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1535                         set_opt(sb, MINIX_DF);
1536
1537                         break;
1538                 case Opt_grpid:
1539                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1540                         set_opt(sb, GRPID);
1541
1542                         break;
1543                 case Opt_nogrpid:
1544                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1545                         clear_opt(sb, GRPID);
1546
1547                         break;
1548                 case Opt_resuid:
1549                         if (match_int(&args[0], &option))
1550                                 return 0;
1551                         sbi->s_resuid = option;
1552                         break;
1553                 case Opt_resgid:
1554                         if (match_int(&args[0], &option))
1555                                 return 0;
1556                         sbi->s_resgid = option;
1557                         break;
1558                 case Opt_sb:
1559                         /* handled by get_sb_block() instead of here */
1560                         /* *sb_block = match_int(&args[0]); */
1561                         break;
1562                 case Opt_err_panic:
1563                         clear_opt(sb, ERRORS_CONT);
1564                         clear_opt(sb, ERRORS_RO);
1565                         set_opt(sb, ERRORS_PANIC);
1566                         break;
1567                 case Opt_err_ro:
1568                         clear_opt(sb, ERRORS_CONT);
1569                         clear_opt(sb, ERRORS_PANIC);
1570                         set_opt(sb, ERRORS_RO);
1571                         break;
1572                 case Opt_err_cont:
1573                         clear_opt(sb, ERRORS_RO);
1574                         clear_opt(sb, ERRORS_PANIC);
1575                         set_opt(sb, ERRORS_CONT);
1576                         break;
1577                 case Opt_nouid32:
1578                         set_opt(sb, NO_UID32);
1579                         break;
1580                 case Opt_debug:
1581                         set_opt(sb, DEBUG);
1582                         break;
1583                 case Opt_oldalloc:
1584                         ext4_msg(sb, KERN_WARNING,
1585                                  "Ignoring deprecated oldalloc option");
1586                         break;
1587                 case Opt_orlov:
1588                         ext4_msg(sb, KERN_WARNING,
1589                                  "Ignoring deprecated orlov option");
1590                         break;
1591 #ifdef CONFIG_EXT4_FS_XATTR
1592                 case Opt_user_xattr:
1593                         set_opt(sb, XATTR_USER);
1594                         break;
1595                 case Opt_nouser_xattr:
1596                         clear_opt(sb, XATTR_USER);
1597                         break;
1598 #else
1599                 case Opt_user_xattr:
1600                 case Opt_nouser_xattr:
1601                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1602                         break;
1603 #endif
1604 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1605                 case Opt_acl:
1606                         set_opt(sb, POSIX_ACL);
1607                         break;
1608                 case Opt_noacl:
1609                         clear_opt(sb, POSIX_ACL);
1610                         break;
1611 #else
1612                 case Opt_acl:
1613                 case Opt_noacl:
1614                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1615                         break;
1616 #endif
1617                 case Opt_journal_update:
1618                         /* @@@ FIXME */
1619                         /* Eventually we will want to be able to create
1620                            a journal file here.  For now, only allow the
1621                            user to specify an existing inode to be the
1622                            journal file. */
1623                         if (is_remount) {
1624                                 ext4_msg(sb, KERN_ERR,
1625                                          "Cannot specify journal on remount");
1626                                 return 0;
1627                         }
1628                         set_opt(sb, UPDATE_JOURNAL);
1629                         break;
1630                 case Opt_journal_dev:
1631                         if (is_remount) {
1632                                 ext4_msg(sb, KERN_ERR,
1633                                         "Cannot specify journal on remount");
1634                                 return 0;
1635                         }
1636                         if (match_int(&args[0], &option))
1637                                 return 0;
1638                         *journal_devnum = option;
1639                         break;
1640                 case Opt_journal_checksum:
1641                         set_opt(sb, JOURNAL_CHECKSUM);
1642                         break;
1643                 case Opt_journal_async_commit:
1644                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1645                         set_opt(sb, JOURNAL_CHECKSUM);
1646                         break;
1647                 case Opt_noload:
1648                         set_opt(sb, NOLOAD);
1649                         break;
1650                 case Opt_commit:
1651                         if (match_int(&args[0], &option))
1652                                 return 0;
1653                         if (option < 0)
1654                                 return 0;
1655                         if (option == 0)
1656                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1657                         sbi->s_commit_interval = HZ * option;
1658                         break;
1659                 case Opt_max_batch_time:
1660                         if (match_int(&args[0], &option))
1661                                 return 0;
1662                         if (option < 0)
1663                                 return 0;
1664                         if (option == 0)
1665                                 option = EXT4_DEF_MAX_BATCH_TIME;
1666                         sbi->s_max_batch_time = option;
1667                         break;
1668                 case Opt_min_batch_time:
1669                         if (match_int(&args[0], &option))
1670                                 return 0;
1671                         if (option < 0)
1672                                 return 0;
1673                         sbi->s_min_batch_time = option;
1674                         break;
1675                 case Opt_data_journal:
1676                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1677                         goto datacheck;
1678                 case Opt_data_ordered:
1679                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1680                         goto datacheck;
1681                 case Opt_data_writeback:
1682                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1683                 datacheck:
1684                         if (is_remount) {
1685                                 if (!sbi->s_journal)
1686                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1687                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1688                                         ext4_msg(sb, KERN_ERR,
1689                                                 "Cannot change data mode on remount");
1690                                         return 0;
1691                                 }
1692                         } else {
1693                                 clear_opt(sb, DATA_FLAGS);
1694                                 sbi->s_mount_opt |= data_opt;
1695                         }
1696                         break;
1697                 case Opt_data_err_abort:
1698                         set_opt(sb, DATA_ERR_ABORT);
1699                         break;
1700                 case Opt_data_err_ignore:
1701                         clear_opt(sb, DATA_ERR_ABORT);
1702                         break;
1703 #ifdef CONFIG_QUOTA
1704                 case Opt_usrjquota:
1705                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1706                                 return 0;
1707                         break;
1708                 case Opt_grpjquota:
1709                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1710                                 return 0;
1711                         break;
1712                 case Opt_offusrjquota:
1713                         if (!clear_qf_name(sb, USRQUOTA))
1714                                 return 0;
1715                         break;
1716                 case Opt_offgrpjquota:
1717                         if (!clear_qf_name(sb, GRPQUOTA))
1718                                 return 0;
1719                         break;
1720
1721                 case Opt_jqfmt_vfsold:
1722                         qfmt = QFMT_VFS_OLD;
1723                         goto set_qf_format;
1724                 case Opt_jqfmt_vfsv0:
1725                         qfmt = QFMT_VFS_V0;
1726                         goto set_qf_format;
1727                 case Opt_jqfmt_vfsv1:
1728                         qfmt = QFMT_VFS_V1;
1729 set_qf_format:
1730                         if (sb_any_quota_loaded(sb) &&
1731                             sbi->s_jquota_fmt != qfmt) {
1732                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1733                                         "journaled quota options when "
1734                                         "quota turned on");
1735                                 return 0;
1736                         }
1737                         sbi->s_jquota_fmt = qfmt;
1738                         break;
1739                 case Opt_quota:
1740                 case Opt_usrquota:
1741                         set_opt(sb, QUOTA);
1742                         set_opt(sb, USRQUOTA);
1743                         break;
1744                 case Opt_grpquota:
1745                         set_opt(sb, QUOTA);
1746                         set_opt(sb, GRPQUOTA);
1747                         break;
1748                 case Opt_noquota:
1749                         if (sb_any_quota_loaded(sb)) {
1750                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1751                                         "options when quota turned on");
1752                                 return 0;
1753                         }
1754                         clear_opt(sb, QUOTA);
1755                         clear_opt(sb, USRQUOTA);
1756                         clear_opt(sb, GRPQUOTA);
1757                         break;
1758 #else
1759                 case Opt_quota:
1760                 case Opt_usrquota:
1761                 case Opt_grpquota:
1762                         ext4_msg(sb, KERN_ERR,
1763                                 "quota options not supported");
1764                         break;
1765                 case Opt_usrjquota:
1766                 case Opt_grpjquota:
1767                 case Opt_offusrjquota:
1768                 case Opt_offgrpjquota:
1769                 case Opt_jqfmt_vfsold:
1770                 case Opt_jqfmt_vfsv0:
1771                 case Opt_jqfmt_vfsv1:
1772                         ext4_msg(sb, KERN_ERR,
1773                                 "journaled quota options not supported");
1774                         break;
1775                 case Opt_noquota:
1776                         break;
1777 #endif
1778                 case Opt_abort:
1779                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1780                         break;
1781                 case Opt_nobarrier:
1782                         clear_opt(sb, BARRIER);
1783                         break;
1784                 case Opt_barrier:
1785                         if (args[0].from) {
1786                                 if (match_int(&args[0], &option))
1787                                         return 0;
1788                         } else
1789                                 option = 1;     /* No argument, default to 1 */
1790                         if (option)
1791                                 set_opt(sb, BARRIER);
1792                         else
1793                                 clear_opt(sb, BARRIER);
1794                         break;
1795                 case Opt_ignore:
1796                         break;
1797                 case Opt_resize:
1798                         if (!is_remount) {
1799                                 ext4_msg(sb, KERN_ERR,
1800                                         "resize option only available "
1801                                         "for remount");
1802                                 return 0;
1803                         }
1804                         if (match_int(&args[0], &option) != 0)
1805                                 return 0;
1806                         *n_blocks_count = option;
1807                         break;
1808                 case Opt_nobh:
1809                         ext4_msg(sb, KERN_WARNING,
1810                                  "Ignoring deprecated nobh option");
1811                         break;
1812                 case Opt_bh:
1813                         ext4_msg(sb, KERN_WARNING,
1814                                  "Ignoring deprecated bh option");
1815                         break;
1816                 case Opt_i_version:
1817                         set_opt(sb, I_VERSION);
1818                         sb->s_flags |= MS_I_VERSION;
1819                         break;
1820                 case Opt_nodelalloc:
1821                         clear_opt(sb, DELALLOC);
1822                         clear_opt2(sb, EXPLICIT_DELALLOC);
1823                         break;
1824                 case Opt_mblk_io_submit:
1825                         set_opt(sb, MBLK_IO_SUBMIT);
1826                         break;
1827                 case Opt_nomblk_io_submit:
1828                         clear_opt(sb, MBLK_IO_SUBMIT);
1829                         break;
1830                 case Opt_stripe:
1831                         if (match_int(&args[0], &option))
1832                                 return 0;
1833                         if (option < 0)
1834                                 return 0;
1835                         sbi->s_stripe = option;
1836                         break;
1837                 case Opt_delalloc:
1838                         set_opt(sb, DELALLOC);
1839                         set_opt2(sb, EXPLICIT_DELALLOC);
1840                         break;
1841                 case Opt_block_validity:
1842                         set_opt(sb, BLOCK_VALIDITY);
1843                         break;
1844                 case Opt_noblock_validity:
1845                         clear_opt(sb, BLOCK_VALIDITY);
1846                         break;
1847                 case Opt_inode_readahead_blks:
1848                         if (match_int(&args[0], &option))
1849                                 return 0;
1850                         if (option < 0 || option > (1 << 30))
1851                                 return 0;
1852                         if (option && !is_power_of_2(option)) {
1853                                 ext4_msg(sb, KERN_ERR,
1854                                          "EXT4-fs: inode_readahead_blks"
1855                                          " must be a power of 2");
1856                                 return 0;
1857                         }
1858                         sbi->s_inode_readahead_blks = option;
1859                         break;
1860                 case Opt_journal_ioprio:
1861                         if (match_int(&args[0], &option))
1862                                 return 0;
1863                         if (option < 0 || option > 7)
1864                                 break;
1865                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1866                                                             option);
1867                         break;
1868                 case Opt_noauto_da_alloc:
1869                         set_opt(sb, NO_AUTO_DA_ALLOC);
1870                         break;
1871                 case Opt_auto_da_alloc:
1872                         if (args[0].from) {
1873                                 if (match_int(&args[0], &option))
1874                                         return 0;
1875                         } else
1876                                 option = 1;     /* No argument, default to 1 */
1877                         if (option)
1878                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1879                         else
1880                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1881                         break;
1882                 case Opt_discard:
1883                         set_opt(sb, DISCARD);
1884                         break;
1885                 case Opt_nodiscard:
1886                         clear_opt(sb, DISCARD);
1887                         break;
1888                 case Opt_dioread_nolock:
1889                         set_opt(sb, DIOREAD_NOLOCK);
1890                         break;
1891                 case Opt_dioread_lock:
1892                         clear_opt(sb, DIOREAD_NOLOCK);
1893                         break;
1894                 case Opt_init_itable:
1895                         set_opt(sb, INIT_INODE_TABLE);
1896                         if (args[0].from) {
1897                                 if (match_int(&args[0], &option))
1898                                         return 0;
1899                         } else
1900                                 option = EXT4_DEF_LI_WAIT_MULT;
1901                         if (option < 0)
1902                                 return 0;
1903                         sbi->s_li_wait_mult = option;
1904                         break;
1905                 case Opt_noinit_itable:
1906                         clear_opt(sb, INIT_INODE_TABLE);
1907                         break;
1908                 default:
1909                         ext4_msg(sb, KERN_ERR,
1910                                "Unrecognized mount option \"%s\" "
1911                                "or missing value", p);
1912                         return 0;
1913                 }
1914         }
1915 #ifdef CONFIG_QUOTA
1916         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1917                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1918                         clear_opt(sb, USRQUOTA);
1919
1920                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1921                         clear_opt(sb, GRPQUOTA);
1922
1923                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1924                         ext4_msg(sb, KERN_ERR, "old and new quota "
1925                                         "format mixing");
1926                         return 0;
1927                 }
1928
1929                 if (!sbi->s_jquota_fmt) {
1930                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1931                                         "not specified");
1932                         return 0;
1933                 }
1934         } else {
1935                 if (sbi->s_jquota_fmt) {
1936                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1937                                         "specified with no journaling "
1938                                         "enabled");
1939                         return 0;
1940                 }
1941         }
1942 #endif
1943         return 1;
1944 }
1945
1946 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1947                             int read_only)
1948 {
1949         struct ext4_sb_info *sbi = EXT4_SB(sb);
1950         int res = 0;
1951
1952         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1953                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1954                          "forcing read-only mode");
1955                 res = MS_RDONLY;
1956         }
1957         if (read_only)
1958                 goto done;
1959         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1960                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1961                          "running e2fsck is recommended");
1962         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1963                 ext4_msg(sb, KERN_WARNING,
1964                          "warning: mounting fs with errors, "
1965                          "running e2fsck is recommended");
1966         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1967                  le16_to_cpu(es->s_mnt_count) >=
1968                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1969                 ext4_msg(sb, KERN_WARNING,
1970                          "warning: maximal mount count reached, "
1971                          "running e2fsck is recommended");
1972         else if (le32_to_cpu(es->s_checkinterval) &&
1973                 (le32_to_cpu(es->s_lastcheck) +
1974                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1975                 ext4_msg(sb, KERN_WARNING,
1976                          "warning: checktime reached, "
1977                          "running e2fsck is recommended");
1978         if (!sbi->s_journal)
1979                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1980         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1981                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1982         le16_add_cpu(&es->s_mnt_count, 1);
1983         es->s_mtime = cpu_to_le32(get_seconds());
1984         ext4_update_dynamic_rev(sb);
1985         if (sbi->s_journal)
1986                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1987
1988         ext4_commit_super(sb, 1);
1989 done:
1990         if (test_opt(sb, DEBUG))
1991                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1992                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1993                         sb->s_blocksize,
1994                         sbi->s_groups_count,
1995                         EXT4_BLOCKS_PER_GROUP(sb),
1996                         EXT4_INODES_PER_GROUP(sb),
1997                         sbi->s_mount_opt, sbi->s_mount_opt2);
1998
1999         cleancache_init_fs(sb);
2000         return res;
2001 }
2002
2003 static int ext4_fill_flex_info(struct super_block *sb)
2004 {
2005         struct ext4_sb_info *sbi = EXT4_SB(sb);
2006         struct ext4_group_desc *gdp = NULL;
2007         ext4_group_t flex_group_count;
2008         ext4_group_t flex_group;
2009         unsigned int groups_per_flex = 0;
2010         size_t size;
2011         int i;
2012
2013         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2014         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2015                 sbi->s_log_groups_per_flex = 0;
2016                 return 1;
2017         }
2018         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2019
2020         /* We allocate both existing and potentially added groups */
2021         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2022                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2023                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2024         size = flex_group_count * sizeof(struct flex_groups);
2025         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2026         if (sbi->s_flex_groups == NULL) {
2027                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2028                          flex_group_count);
2029                 goto failed;
2030         }
2031
2032         for (i = 0; i < sbi->s_groups_count; i++) {
2033                 gdp = ext4_get_group_desc(sb, i, NULL);
2034
2035                 flex_group = ext4_flex_group(sbi, i);
2036                 atomic_add(ext4_free_inodes_count(sb, gdp),
2037                            &sbi->s_flex_groups[flex_group].free_inodes);
2038                 atomic_add(ext4_free_group_clusters(sb, gdp),
2039                            &sbi->s_flex_groups[flex_group].free_clusters);
2040                 atomic_add(ext4_used_dirs_count(sb, gdp),
2041                            &sbi->s_flex_groups[flex_group].used_dirs);
2042         }
2043
2044         return 1;
2045 failed:
2046         return 0;
2047 }
2048
2049 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2050                             struct ext4_group_desc *gdp)
2051 {
2052         __u16 crc = 0;
2053
2054         if (sbi->s_es->s_feature_ro_compat &
2055             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2056                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2057                 __le32 le_group = cpu_to_le32(block_group);
2058
2059                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2060                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2061                 crc = crc16(crc, (__u8 *)gdp, offset);
2062                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2063                 /* for checksum of struct ext4_group_desc do the rest...*/
2064                 if ((sbi->s_es->s_feature_incompat &
2065                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2066                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2067                         crc = crc16(crc, (__u8 *)gdp + offset,
2068                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2069                                         offset);
2070         }
2071
2072         return cpu_to_le16(crc);
2073 }
2074
2075 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2076                                 struct ext4_group_desc *gdp)
2077 {
2078         if ((sbi->s_es->s_feature_ro_compat &
2079              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2080             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2081                 return 0;
2082
2083         return 1;
2084 }
2085
2086 /* Called at mount-time, super-block is locked */
2087 static int ext4_check_descriptors(struct super_block *sb,
2088                                   ext4_group_t *first_not_zeroed)
2089 {
2090         struct ext4_sb_info *sbi = EXT4_SB(sb);
2091         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2092         ext4_fsblk_t last_block;
2093         ext4_fsblk_t block_bitmap;
2094         ext4_fsblk_t inode_bitmap;
2095         ext4_fsblk_t inode_table;
2096         int flexbg_flag = 0;
2097         ext4_group_t i, grp = sbi->s_groups_count;
2098
2099         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2100                 flexbg_flag = 1;
2101
2102         ext4_debug("Checking group descriptors");
2103
2104         for (i = 0; i < sbi->s_groups_count; i++) {
2105                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2106
2107                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2108                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2109                 else
2110                         last_block = first_block +
2111                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2112
2113                 if ((grp == sbi->s_groups_count) &&
2114                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2115                         grp = i;
2116
2117                 block_bitmap = ext4_block_bitmap(sb, gdp);
2118                 if (block_bitmap < first_block || block_bitmap > last_block) {
2119                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2120                                "Block bitmap for group %u not in group "
2121                                "(block %llu)!", i, block_bitmap);
2122                         return 0;
2123                 }
2124                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2125                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2126                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2127                                "Inode bitmap for group %u not in group "
2128                                "(block %llu)!", i, inode_bitmap);
2129                         return 0;
2130                 }
2131                 inode_table = ext4_inode_table(sb, gdp);
2132                 if (inode_table < first_block ||
2133                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                "Inode table for group %u not in group "
2136                                "(block %llu)!", i, inode_table);
2137                         return 0;
2138                 }
2139                 ext4_lock_group(sb, i);
2140                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2141                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2142                                  "Checksum for group %u failed (%u!=%u)",
2143                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2144                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2145                         if (!(sb->s_flags & MS_RDONLY)) {
2146                                 ext4_unlock_group(sb, i);
2147                                 return 0;
2148                         }
2149                 }
2150                 ext4_unlock_group(sb, i);
2151                 if (!flexbg_flag)
2152                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2153         }
2154         if (NULL != first_not_zeroed)
2155                 *first_not_zeroed = grp;
2156
2157         ext4_free_blocks_count_set(sbi->s_es,
2158                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2159         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2160         return 1;
2161 }
2162
2163 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2164  * the superblock) which were deleted from all directories, but held open by
2165  * a process at the time of a crash.  We walk the list and try to delete these
2166  * inodes at recovery time (only with a read-write filesystem).
2167  *
2168  * In order to keep the orphan inode chain consistent during traversal (in
2169  * case of crash during recovery), we link each inode into the superblock
2170  * orphan list_head and handle it the same way as an inode deletion during
2171  * normal operation (which journals the operations for us).
2172  *
2173  * We only do an iget() and an iput() on each inode, which is very safe if we
2174  * accidentally point at an in-use or already deleted inode.  The worst that
2175  * can happen in this case is that we get a "bit already cleared" message from
2176  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2177  * e2fsck was run on this filesystem, and it must have already done the orphan
2178  * inode cleanup for us, so we can safely abort without any further action.
2179  */
2180 static void ext4_orphan_cleanup(struct super_block *sb,
2181                                 struct ext4_super_block *es)
2182 {
2183         unsigned int s_flags = sb->s_flags;
2184         int nr_orphans = 0, nr_truncates = 0;
2185 #ifdef CONFIG_QUOTA
2186         int i;
2187 #endif
2188         if (!es->s_last_orphan) {
2189                 jbd_debug(4, "no orphan inodes to clean up\n");
2190                 return;
2191         }
2192
2193         if (bdev_read_only(sb->s_bdev)) {
2194                 ext4_msg(sb, KERN_ERR, "write access "
2195                         "unavailable, skipping orphan cleanup");
2196                 return;
2197         }
2198
2199         /* Check if feature set would not allow a r/w mount */
2200         if (!ext4_feature_set_ok(sb, 0)) {
2201                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2202                          "unknown ROCOMPAT features");
2203                 return;
2204         }
2205
2206         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2207                 if (es->s_last_orphan)
2208                         jbd_debug(1, "Errors on filesystem, "
2209                                   "clearing orphan list.\n");
2210                 es->s_last_orphan = 0;
2211                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2212                 return;
2213         }
2214
2215         if (s_flags & MS_RDONLY) {
2216                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2217                 sb->s_flags &= ~MS_RDONLY;
2218         }
2219 #ifdef CONFIG_QUOTA
2220         /* Needed for iput() to work correctly and not trash data */
2221         sb->s_flags |= MS_ACTIVE;
2222         /* Turn on quotas so that they are updated correctly */
2223         for (i = 0; i < MAXQUOTAS; i++) {
2224                 if (EXT4_SB(sb)->s_qf_names[i]) {
2225                         int ret = ext4_quota_on_mount(sb, i);
2226                         if (ret < 0)
2227                                 ext4_msg(sb, KERN_ERR,
2228                                         "Cannot turn on journaled "
2229                                         "quota: error %d", ret);
2230                 }
2231         }
2232 #endif
2233
2234         while (es->s_last_orphan) {
2235                 struct inode *inode;
2236
2237                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2238                 if (IS_ERR(inode)) {
2239                         es->s_last_orphan = 0;
2240                         break;
2241                 }
2242
2243                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2244                 dquot_initialize(inode);
2245                 if (inode->i_nlink) {
2246                         ext4_msg(sb, KERN_DEBUG,
2247                                 "%s: truncating inode %lu to %lld bytes",
2248                                 __func__, inode->i_ino, inode->i_size);
2249                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2250                                   inode->i_ino, inode->i_size);
2251                         ext4_truncate(inode);
2252                         nr_truncates++;
2253                 } else {
2254                         ext4_msg(sb, KERN_DEBUG,
2255                                 "%s: deleting unreferenced inode %lu",
2256                                 __func__, inode->i_ino);
2257                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2258                                   inode->i_ino);
2259                         nr_orphans++;
2260                 }
2261                 iput(inode);  /* The delete magic happens here! */
2262         }
2263
2264 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2265
2266         if (nr_orphans)
2267                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2268                        PLURAL(nr_orphans));
2269         if (nr_truncates)
2270                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2271                        PLURAL(nr_truncates));
2272 #ifdef CONFIG_QUOTA
2273         /* Turn quotas off */
2274         for (i = 0; i < MAXQUOTAS; i++) {
2275                 if (sb_dqopt(sb)->files[i])
2276                         dquot_quota_off(sb, i);
2277         }
2278 #endif
2279         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2280 }
2281
2282 /*
2283  * Maximal extent format file size.
2284  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2285  * extent format containers, within a sector_t, and within i_blocks
2286  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2287  * so that won't be a limiting factor.
2288  *
2289  * However there is other limiting factor. We do store extents in the form
2290  * of starting block and length, hence the resulting length of the extent
2291  * covering maximum file size must fit into on-disk format containers as
2292  * well. Given that length is always by 1 unit bigger than max unit (because
2293  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2294  *
2295  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2296  */
2297 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2298 {
2299         loff_t res;
2300         loff_t upper_limit = MAX_LFS_FILESIZE;
2301
2302         /* small i_blocks in vfs inode? */
2303         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2304                 /*
2305                  * CONFIG_LBDAF is not enabled implies the inode
2306                  * i_block represent total blocks in 512 bytes
2307                  * 32 == size of vfs inode i_blocks * 8
2308                  */
2309                 upper_limit = (1LL << 32) - 1;
2310
2311                 /* total blocks in file system block size */
2312                 upper_limit >>= (blkbits - 9);
2313                 upper_limit <<= blkbits;
2314         }
2315
2316         /*
2317          * 32-bit extent-start container, ee_block. We lower the maxbytes
2318          * by one fs block, so ee_len can cover the extent of maximum file
2319          * size
2320          */
2321         res = (1LL << 32) - 1;
2322         res <<= blkbits;
2323
2324         /* Sanity check against vm- & vfs- imposed limits */
2325         if (res > upper_limit)
2326                 res = upper_limit;
2327
2328         return res;
2329 }
2330
2331 /*
2332  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2333  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2334  * We need to be 1 filesystem block less than the 2^48 sector limit.
2335  */
2336 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2337 {
2338         loff_t res = EXT4_NDIR_BLOCKS;
2339         int meta_blocks;
2340         loff_t upper_limit;
2341         /* This is calculated to be the largest file size for a dense, block
2342          * mapped file such that the file's total number of 512-byte sectors,
2343          * including data and all indirect blocks, does not exceed (2^48 - 1).
2344          *
2345          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2346          * number of 512-byte sectors of the file.
2347          */
2348
2349         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2350                 /*
2351                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2352                  * the inode i_block field represents total file blocks in
2353                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2354                  */
2355                 upper_limit = (1LL << 32) - 1;
2356
2357                 /* total blocks in file system block size */
2358                 upper_limit >>= (bits - 9);
2359
2360         } else {
2361                 /*
2362                  * We use 48 bit ext4_inode i_blocks
2363                  * With EXT4_HUGE_FILE_FL set the i_blocks
2364                  * represent total number of blocks in
2365                  * file system block size
2366                  */
2367                 upper_limit = (1LL << 48) - 1;
2368
2369         }
2370
2371         /* indirect blocks */
2372         meta_blocks = 1;
2373         /* double indirect blocks */
2374         meta_blocks += 1 + (1LL << (bits-2));
2375         /* tripple indirect blocks */
2376         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2377
2378         upper_limit -= meta_blocks;
2379         upper_limit <<= bits;
2380
2381         res += 1LL << (bits-2);
2382         res += 1LL << (2*(bits-2));
2383         res += 1LL << (3*(bits-2));
2384         res <<= bits;
2385         if (res > upper_limit)
2386                 res = upper_limit;
2387
2388         if (res > MAX_LFS_FILESIZE)
2389                 res = MAX_LFS_FILESIZE;
2390
2391         return res;
2392 }
2393
2394 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2395                                    ext4_fsblk_t logical_sb_block, int nr)
2396 {
2397         struct ext4_sb_info *sbi = EXT4_SB(sb);
2398         ext4_group_t bg, first_meta_bg;
2399         int has_super = 0;
2400
2401         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2402
2403         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2404             nr < first_meta_bg)
2405                 return logical_sb_block + nr + 1;
2406         bg = sbi->s_desc_per_block * nr;
2407         if (ext4_bg_has_super(sb, bg))
2408                 has_super = 1;
2409
2410         return (has_super + ext4_group_first_block_no(sb, bg));
2411 }
2412
2413 /**
2414  * ext4_get_stripe_size: Get the stripe size.
2415  * @sbi: In memory super block info
2416  *
2417  * If we have specified it via mount option, then
2418  * use the mount option value. If the value specified at mount time is
2419  * greater than the blocks per group use the super block value.
2420  * If the super block value is greater than blocks per group return 0.
2421  * Allocator needs it be less than blocks per group.
2422  *
2423  */
2424 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2425 {
2426         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2427         unsigned long stripe_width =
2428                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2429         int ret;
2430
2431         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2432                 ret = sbi->s_stripe;
2433         else if (stripe_width <= sbi->s_blocks_per_group)
2434                 ret = stripe_width;
2435         else if (stride <= sbi->s_blocks_per_group)
2436                 ret = stride;
2437         else
2438                 ret = 0;
2439
2440         /*
2441          * If the stripe width is 1, this makes no sense and
2442          * we set it to 0 to turn off stripe handling code.
2443          */
2444         if (ret <= 1)
2445                 ret = 0;
2446
2447         return ret;
2448 }
2449
2450 /* sysfs supprt */
2451
2452 struct ext4_attr {
2453         struct attribute attr;
2454         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2455         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2456                          const char *, size_t);
2457         int offset;
2458 };
2459
2460 static int parse_strtoul(const char *buf,
2461                 unsigned long max, unsigned long *value)
2462 {
2463         char *endp;
2464
2465         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2466         endp = skip_spaces(endp);
2467         if (*endp || *value > max)
2468                 return -EINVAL;
2469
2470         return 0;
2471 }
2472
2473 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2474                                               struct ext4_sb_info *sbi,
2475                                               char *buf)
2476 {
2477         return snprintf(buf, PAGE_SIZE, "%llu\n",
2478                 (s64) EXT4_C2B(sbi,
2479                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2480 }
2481
2482 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2483                                          struct ext4_sb_info *sbi, char *buf)
2484 {
2485         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2486
2487         if (!sb->s_bdev->bd_part)
2488                 return snprintf(buf, PAGE_SIZE, "0\n");
2489         return snprintf(buf, PAGE_SIZE, "%lu\n",
2490                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2491                          sbi->s_sectors_written_start) >> 1);
2492 }
2493
2494 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2495                                           struct ext4_sb_info *sbi, char *buf)
2496 {
2497         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2498
2499         if (!sb->s_bdev->bd_part)
2500                 return snprintf(buf, PAGE_SIZE, "0\n");
2501         return snprintf(buf, PAGE_SIZE, "%llu\n",
2502                         (unsigned long long)(sbi->s_kbytes_written +
2503                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2504                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2505 }
2506
2507 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2508                                           struct ext4_sb_info *sbi,
2509                                           const char *buf, size_t count)
2510 {
2511         unsigned long t;
2512
2513         if (parse_strtoul(buf, 0x40000000, &t))
2514                 return -EINVAL;
2515
2516         if (t && !is_power_of_2(t))
2517                 return -EINVAL;
2518
2519         sbi->s_inode_readahead_blks = t;
2520         return count;
2521 }
2522
2523 static ssize_t sbi_ui_show(struct ext4_attr *a,
2524                            struct ext4_sb_info *sbi, char *buf)
2525 {
2526         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2527
2528         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2529 }
2530
2531 static ssize_t sbi_ui_store(struct ext4_attr *a,
2532                             struct ext4_sb_info *sbi,
2533                             const char *buf, size_t count)
2534 {
2535         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2536         unsigned long t;
2537
2538         if (parse_strtoul(buf, 0xffffffff, &t))
2539                 return -EINVAL;
2540         *ui = t;
2541         return count;
2542 }
2543
2544 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2545 static struct ext4_attr ext4_attr_##_name = {                   \
2546         .attr = {.name = __stringify(_name), .mode = _mode },   \
2547         .show   = _show,                                        \
2548         .store  = _store,                                       \
2549         .offset = offsetof(struct ext4_sb_info, _elname),       \
2550 }
2551 #define EXT4_ATTR(name, mode, show, store) \
2552 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2553
2554 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2555 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2556 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2557 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2558         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2559 #define ATTR_LIST(name) &ext4_attr_##name.attr
2560
2561 EXT4_RO_ATTR(delayed_allocation_blocks);
2562 EXT4_RO_ATTR(session_write_kbytes);
2563 EXT4_RO_ATTR(lifetime_write_kbytes);
2564 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2565                  inode_readahead_blks_store, s_inode_readahead_blks);
2566 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2567 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2568 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2569 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2570 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2571 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2572 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2573 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2574
2575 static struct attribute *ext4_attrs[] = {
2576         ATTR_LIST(delayed_allocation_blocks),
2577         ATTR_LIST(session_write_kbytes),
2578         ATTR_LIST(lifetime_write_kbytes),
2579         ATTR_LIST(inode_readahead_blks),
2580         ATTR_LIST(inode_goal),
2581         ATTR_LIST(mb_stats),
2582         ATTR_LIST(mb_max_to_scan),
2583         ATTR_LIST(mb_min_to_scan),
2584         ATTR_LIST(mb_order2_req),
2585         ATTR_LIST(mb_stream_req),
2586         ATTR_LIST(mb_group_prealloc),
2587         ATTR_LIST(max_writeback_mb_bump),
2588         NULL,
2589 };
2590
2591 /* Features this copy of ext4 supports */
2592 EXT4_INFO_ATTR(lazy_itable_init);
2593 EXT4_INFO_ATTR(batched_discard);
2594
2595 static struct attribute *ext4_feat_attrs[] = {
2596         ATTR_LIST(lazy_itable_init),
2597         ATTR_LIST(batched_discard),
2598         NULL,
2599 };
2600
2601 static ssize_t ext4_attr_show(struct kobject *kobj,
2602                               struct attribute *attr, char *buf)
2603 {
2604         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2605                                                 s_kobj);
2606         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2607
2608         return a->show ? a->show(a, sbi, buf) : 0;
2609 }
2610
2611 static ssize_t ext4_attr_store(struct kobject *kobj,
2612                                struct attribute *attr,
2613                                const char *buf, size_t len)
2614 {
2615         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2616                                                 s_kobj);
2617         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2618
2619         return a->store ? a->store(a, sbi, buf, len) : 0;
2620 }
2621
2622 static void ext4_sb_release(struct kobject *kobj)
2623 {
2624         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2625                                                 s_kobj);
2626         complete(&sbi->s_kobj_unregister);
2627 }
2628
2629 static const struct sysfs_ops ext4_attr_ops = {
2630         .show   = ext4_attr_show,
2631         .store  = ext4_attr_store,
2632 };
2633
2634 static struct kobj_type ext4_ktype = {
2635         .default_attrs  = ext4_attrs,
2636         .sysfs_ops      = &ext4_attr_ops,
2637         .release        = ext4_sb_release,
2638 };
2639
2640 static void ext4_feat_release(struct kobject *kobj)
2641 {
2642         complete(&ext4_feat->f_kobj_unregister);
2643 }
2644
2645 static struct kobj_type ext4_feat_ktype = {
2646         .default_attrs  = ext4_feat_attrs,
2647         .sysfs_ops      = &ext4_attr_ops,
2648         .release        = ext4_feat_release,
2649 };
2650
2651 /*
2652  * Check whether this filesystem can be mounted based on
2653  * the features present and the RDONLY/RDWR mount requested.
2654  * Returns 1 if this filesystem can be mounted as requested,
2655  * 0 if it cannot be.
2656  */
2657 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2658 {
2659         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2660                 ext4_msg(sb, KERN_ERR,
2661                         "Couldn't mount because of "
2662                         "unsupported optional features (%x)",
2663                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2664                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2665                 return 0;
2666         }
2667
2668         if (readonly)
2669                 return 1;
2670
2671         /* Check that feature set is OK for a read-write mount */
2672         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2673                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2674                          "unsupported optional features (%x)",
2675                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2676                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2677                 return 0;
2678         }
2679         /*
2680          * Large file size enabled file system can only be mounted
2681          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2682          */
2683         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2684                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2685                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2686                                  "cannot be mounted RDWR without "
2687                                  "CONFIG_LBDAF");
2688                         return 0;
2689                 }
2690         }
2691         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2692             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2693                 ext4_msg(sb, KERN_ERR,
2694                          "Can't support bigalloc feature without "
2695                          "extents feature\n");
2696                 return 0;
2697         }
2698         return 1;
2699 }
2700
2701 /*
2702  * This function is called once a day if we have errors logged
2703  * on the file system
2704  */
2705 static void print_daily_error_info(unsigned long arg)
2706 {
2707         struct super_block *sb = (struct super_block *) arg;
2708         struct ext4_sb_info *sbi;
2709         struct ext4_super_block *es;
2710
2711         sbi = EXT4_SB(sb);
2712         es = sbi->s_es;
2713
2714         if (es->s_error_count)
2715                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2716                          le32_to_cpu(es->s_error_count));
2717         if (es->s_first_error_time) {
2718                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2719                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2720                        (int) sizeof(es->s_first_error_func),
2721                        es->s_first_error_func,
2722                        le32_to_cpu(es->s_first_error_line));
2723                 if (es->s_first_error_ino)
2724                         printk(": inode %u",
2725                                le32_to_cpu(es->s_first_error_ino));
2726                 if (es->s_first_error_block)
2727                         printk(": block %llu", (unsigned long long)
2728                                le64_to_cpu(es->s_first_error_block));
2729                 printk("\n");
2730         }
2731         if (es->s_last_error_time) {
2732                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2733                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2734                        (int) sizeof(es->s_last_error_func),
2735                        es->s_last_error_func,
2736                        le32_to_cpu(es->s_last_error_line));
2737                 if (es->s_last_error_ino)
2738                         printk(": inode %u",
2739                                le32_to_cpu(es->s_last_error_ino));
2740                 if (es->s_last_error_block)
2741                         printk(": block %llu", (unsigned long long)
2742                                le64_to_cpu(es->s_last_error_block));
2743                 printk("\n");
2744         }
2745         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2746 }
2747
2748 /* Find next suitable group and run ext4_init_inode_table */
2749 static int ext4_run_li_request(struct ext4_li_request *elr)
2750 {
2751         struct ext4_group_desc *gdp = NULL;
2752         ext4_group_t group, ngroups;
2753         struct super_block *sb;
2754         unsigned long timeout = 0;
2755         int ret = 0;
2756
2757         sb = elr->lr_super;
2758         ngroups = EXT4_SB(sb)->s_groups_count;
2759
2760         for (group = elr->lr_next_group; group < ngroups; group++) {
2761                 gdp = ext4_get_group_desc(sb, group, NULL);
2762                 if (!gdp) {
2763                         ret = 1;
2764                         break;
2765                 }
2766
2767                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2768                         break;
2769         }
2770
2771         if (group == ngroups)
2772                 ret = 1;
2773
2774         if (!ret) {
2775                 timeout = jiffies;
2776                 ret = ext4_init_inode_table(sb, group,
2777                                             elr->lr_timeout ? 0 : 1);
2778                 if (elr->lr_timeout == 0) {
2779                         timeout = (jiffies - timeout) *
2780                                   elr->lr_sbi->s_li_wait_mult;
2781                         elr->lr_timeout = timeout;
2782                 }
2783                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2784                 elr->lr_next_group = group + 1;
2785         }
2786
2787         return ret;
2788 }
2789
2790 /*
2791  * Remove lr_request from the list_request and free the
2792  * request structure. Should be called with li_list_mtx held
2793  */
2794 static void ext4_remove_li_request(struct ext4_li_request *elr)
2795 {
2796         struct ext4_sb_info *sbi;
2797
2798         if (!elr)
2799                 return;
2800
2801         sbi = elr->lr_sbi;
2802
2803         list_del(&elr->lr_request);
2804         sbi->s_li_request = NULL;
2805         kfree(elr);
2806 }
2807
2808 static void ext4_unregister_li_request(struct super_block *sb)
2809 {
2810         mutex_lock(&ext4_li_mtx);
2811         if (!ext4_li_info) {
2812                 mutex_unlock(&ext4_li_mtx);
2813                 return;
2814         }
2815
2816         mutex_lock(&ext4_li_info->li_list_mtx);
2817         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2818         mutex_unlock(&ext4_li_info->li_list_mtx);
2819         mutex_unlock(&ext4_li_mtx);
2820 }
2821
2822 static struct task_struct *ext4_lazyinit_task;
2823
2824 /*
2825  * This is the function where ext4lazyinit thread lives. It walks
2826  * through the request list searching for next scheduled filesystem.
2827  * When such a fs is found, run the lazy initialization request
2828  * (ext4_rn_li_request) and keep track of the time spend in this
2829  * function. Based on that time we compute next schedule time of
2830  * the request. When walking through the list is complete, compute
2831  * next waking time and put itself into sleep.
2832  */
2833 static int ext4_lazyinit_thread(void *arg)
2834 {
2835         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2836         struct list_head *pos, *n;
2837         struct ext4_li_request *elr;
2838         unsigned long next_wakeup, cur;
2839
2840         BUG_ON(NULL == eli);
2841
2842 cont_thread:
2843         while (true) {
2844                 next_wakeup = MAX_JIFFY_OFFSET;
2845
2846                 mutex_lock(&eli->li_list_mtx);
2847                 if (list_empty(&eli->li_request_list)) {
2848                         mutex_unlock(&eli->li_list_mtx);
2849                         goto exit_thread;
2850                 }
2851
2852                 list_for_each_safe(pos, n, &eli->li_request_list) {
2853                         elr = list_entry(pos, struct ext4_li_request,
2854                                          lr_request);
2855
2856                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2857                                 if (ext4_run_li_request(elr) != 0) {
2858                                         /* error, remove the lazy_init job */
2859                                         ext4_remove_li_request(elr);
2860                                         continue;
2861                                 }
2862                         }
2863
2864                         if (time_before(elr->lr_next_sched, next_wakeup))
2865                                 next_wakeup = elr->lr_next_sched;
2866                 }
2867                 mutex_unlock(&eli->li_list_mtx);
2868
2869                 if (freezing(current))
2870                         refrigerator();
2871
2872                 cur = jiffies;
2873                 if ((time_after_eq(cur, next_wakeup)) ||
2874                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2875                         cond_resched();
2876                         continue;
2877                 }
2878
2879                 schedule_timeout_interruptible(next_wakeup - cur);
2880
2881                 if (kthread_should_stop()) {
2882                         ext4_clear_request_list();
2883                         goto exit_thread;
2884                 }
2885         }
2886
2887 exit_thread:
2888         /*
2889          * It looks like the request list is empty, but we need
2890          * to check it under the li_list_mtx lock, to prevent any
2891          * additions into it, and of course we should lock ext4_li_mtx
2892          * to atomically free the list and ext4_li_info, because at
2893          * this point another ext4 filesystem could be registering
2894          * new one.
2895          */
2896         mutex_lock(&ext4_li_mtx);
2897         mutex_lock(&eli->li_list_mtx);
2898         if (!list_empty(&eli->li_request_list)) {
2899                 mutex_unlock(&eli->li_list_mtx);
2900                 mutex_unlock(&ext4_li_mtx);
2901                 goto cont_thread;
2902         }
2903         mutex_unlock(&eli->li_list_mtx);
2904         kfree(ext4_li_info);
2905         ext4_li_info = NULL;
2906         mutex_unlock(&ext4_li_mtx);
2907
2908         return 0;
2909 }
2910
2911 static void ext4_clear_request_list(void)
2912 {
2913         struct list_head *pos, *n;
2914         struct ext4_li_request *elr;
2915
2916         mutex_lock(&ext4_li_info->li_list_mtx);
2917         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2918                 elr = list_entry(pos, struct ext4_li_request,
2919                                  lr_request);
2920                 ext4_remove_li_request(elr);
2921         }
2922         mutex_unlock(&ext4_li_info->li_list_mtx);
2923 }
2924
2925 static int ext4_run_lazyinit_thread(void)
2926 {
2927         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2928                                          ext4_li_info, "ext4lazyinit");
2929         if (IS_ERR(ext4_lazyinit_task)) {
2930                 int err = PTR_ERR(ext4_lazyinit_task);
2931                 ext4_clear_request_list();
2932                 kfree(ext4_li_info);
2933                 ext4_li_info = NULL;
2934                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2935                                  "initialization thread\n",
2936                                  err);
2937                 return err;
2938         }
2939         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2940         return 0;
2941 }
2942
2943 /*
2944  * Check whether it make sense to run itable init. thread or not.
2945  * If there is at least one uninitialized inode table, return
2946  * corresponding group number, else the loop goes through all
2947  * groups and return total number of groups.
2948  */
2949 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2950 {
2951         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2952         struct ext4_group_desc *gdp = NULL;
2953
2954         for (group = 0; group < ngroups; group++) {
2955                 gdp = ext4_get_group_desc(sb, group, NULL);
2956                 if (!gdp)
2957                         continue;
2958
2959                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2960                         break;
2961         }
2962
2963         return group;
2964 }
2965
2966 static int ext4_li_info_new(void)
2967 {
2968         struct ext4_lazy_init *eli = NULL;
2969
2970         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2971         if (!eli)
2972                 return -ENOMEM;
2973
2974         INIT_LIST_HEAD(&eli->li_request_list);
2975         mutex_init(&eli->li_list_mtx);
2976
2977         eli->li_state |= EXT4_LAZYINIT_QUIT;
2978
2979         ext4_li_info = eli;
2980
2981         return 0;
2982 }
2983
2984 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2985                                             ext4_group_t start)
2986 {
2987         struct ext4_sb_info *sbi = EXT4_SB(sb);
2988         struct ext4_li_request *elr;
2989         unsigned long rnd;
2990
2991         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2992         if (!elr)
2993                 return NULL;
2994
2995         elr->lr_super = sb;
2996         elr->lr_sbi = sbi;
2997         elr->lr_next_group = start;
2998
2999         /*
3000          * Randomize first schedule time of the request to
3001          * spread the inode table initialization requests
3002          * better.
3003          */
3004         get_random_bytes(&rnd, sizeof(rnd));
3005         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3006                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3007
3008         return elr;
3009 }
3010
3011 static int ext4_register_li_request(struct super_block *sb,
3012                                     ext4_group_t first_not_zeroed)
3013 {
3014         struct ext4_sb_info *sbi = EXT4_SB(sb);
3015         struct ext4_li_request *elr;
3016         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3017         int ret = 0;
3018
3019         if (sbi->s_li_request != NULL) {
3020                 /*
3021                  * Reset timeout so it can be computed again, because
3022                  * s_li_wait_mult might have changed.
3023                  */
3024                 sbi->s_li_request->lr_timeout = 0;
3025                 return 0;
3026         }
3027
3028         if (first_not_zeroed == ngroups ||
3029             (sb->s_flags & MS_RDONLY) ||
3030             !test_opt(sb, INIT_INODE_TABLE))
3031                 return 0;
3032
3033         elr = ext4_li_request_new(sb, first_not_zeroed);
3034         if (!elr)
3035                 return -ENOMEM;
3036
3037         mutex_lock(&ext4_li_mtx);
3038
3039         if (NULL == ext4_li_info) {
3040                 ret = ext4_li_info_new();
3041                 if (ret)
3042                         goto out;
3043         }
3044
3045         mutex_lock(&ext4_li_info->li_list_mtx);
3046         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3047         mutex_unlock(&ext4_li_info->li_list_mtx);
3048
3049         sbi->s_li_request = elr;
3050         /*
3051          * set elr to NULL here since it has been inserted to
3052          * the request_list and the removal and free of it is
3053          * handled by ext4_clear_request_list from now on.
3054          */
3055         elr = NULL;
3056
3057         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3058                 ret = ext4_run_lazyinit_thread();
3059                 if (ret)
3060                         goto out;
3061         }
3062 out:
3063         mutex_unlock(&ext4_li_mtx);
3064         if (ret)
3065                 kfree(elr);
3066         return ret;
3067 }
3068
3069 /*
3070  * We do not need to lock anything since this is called on
3071  * module unload.
3072  */
3073 static void ext4_destroy_lazyinit_thread(void)
3074 {
3075         /*
3076          * If thread exited earlier
3077          * there's nothing to be done.
3078          */
3079         if (!ext4_li_info || !ext4_lazyinit_task)
3080                 return;
3081
3082         kthread_stop(ext4_lazyinit_task);
3083 }
3084
3085 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3086 {
3087         char *orig_data = kstrdup(data, GFP_KERNEL);
3088         struct buffer_head *bh;
3089         struct ext4_super_block *es = NULL;
3090         struct ext4_sb_info *sbi;
3091         ext4_fsblk_t block;
3092         ext4_fsblk_t sb_block = get_sb_block(&data);
3093         ext4_fsblk_t logical_sb_block;
3094         unsigned long offset = 0;
3095         unsigned long journal_devnum = 0;
3096         unsigned long def_mount_opts;
3097         struct inode *root;
3098         char *cp;
3099         const char *descr;
3100         int ret = -ENOMEM;
3101         int blocksize, clustersize;
3102         unsigned int db_count;
3103         unsigned int i;
3104         int needs_recovery, has_huge_files, has_bigalloc;
3105         __u64 blocks_count;
3106         int err;
3107         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3108         ext4_group_t first_not_zeroed;
3109
3110         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3111         if (!sbi)
3112                 goto out_free_orig;
3113
3114         sbi->s_blockgroup_lock =
3115                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3116         if (!sbi->s_blockgroup_lock) {
3117                 kfree(sbi);
3118                 goto out_free_orig;
3119         }
3120         sb->s_fs_info = sbi;
3121         sbi->s_mount_opt = 0;
3122         sbi->s_resuid = EXT4_DEF_RESUID;
3123         sbi->s_resgid = EXT4_DEF_RESGID;
3124         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3125         sbi->s_sb_block = sb_block;
3126         if (sb->s_bdev->bd_part)
3127                 sbi->s_sectors_written_start =
3128                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3129
3130         /* Cleanup superblock name */
3131         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3132                 *cp = '!';
3133
3134         ret = -EINVAL;
3135         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3136         if (!blocksize) {
3137                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3138                 goto out_fail;
3139         }
3140
3141         /*
3142          * The ext4 superblock will not be buffer aligned for other than 1kB
3143          * block sizes.  We need to calculate the offset from buffer start.
3144          */
3145         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3146                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3147                 offset = do_div(logical_sb_block, blocksize);
3148         } else {
3149                 logical_sb_block = sb_block;
3150         }
3151
3152         if (!(bh = sb_bread(sb, logical_sb_block))) {
3153                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3154                 goto out_fail;
3155         }
3156         /*
3157          * Note: s_es must be initialized as soon as possible because
3158          *       some ext4 macro-instructions depend on its value
3159          */
3160         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3161         sbi->s_es = es;
3162         sb->s_magic = le16_to_cpu(es->s_magic);
3163         if (sb->s_magic != EXT4_SUPER_MAGIC)
3164                 goto cantfind_ext4;
3165         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3166
3167         /* Set defaults before we parse the mount options */
3168         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3169         set_opt(sb, INIT_INODE_TABLE);
3170         if (def_mount_opts & EXT4_DEFM_DEBUG)
3171                 set_opt(sb, DEBUG);
3172         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3173                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3174                         "2.6.38");
3175                 set_opt(sb, GRPID);
3176         }
3177         if (def_mount_opts & EXT4_DEFM_UID16)
3178                 set_opt(sb, NO_UID32);
3179         /* xattr user namespace & acls are now defaulted on */
3180 #ifdef CONFIG_EXT4_FS_XATTR
3181         set_opt(sb, XATTR_USER);
3182 #endif
3183 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3184         set_opt(sb, POSIX_ACL);
3185 #endif
3186         set_opt(sb, MBLK_IO_SUBMIT);
3187         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3188                 set_opt(sb, JOURNAL_DATA);
3189         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3190                 set_opt(sb, ORDERED_DATA);
3191         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3192                 set_opt(sb, WRITEBACK_DATA);
3193
3194         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3195                 set_opt(sb, ERRORS_PANIC);
3196         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3197                 set_opt(sb, ERRORS_CONT);
3198         else
3199                 set_opt(sb, ERRORS_RO);
3200         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3201                 set_opt(sb, BLOCK_VALIDITY);
3202         if (def_mount_opts & EXT4_DEFM_DISCARD)
3203                 set_opt(sb, DISCARD);
3204
3205         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3206         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3207         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3208         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3209         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3210
3211         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3212                 set_opt(sb, BARRIER);
3213
3214         /*
3215          * enable delayed allocation by default
3216          * Use -o nodelalloc to turn it off
3217          */
3218         if (!IS_EXT3_SB(sb) &&
3219             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3220                 set_opt(sb, DELALLOC);
3221
3222         /*
3223          * set default s_li_wait_mult for lazyinit, for the case there is
3224          * no mount option specified.
3225          */
3226         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3227
3228         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3229                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3230                 ext4_msg(sb, KERN_WARNING,
3231                          "failed to parse options in superblock: %s",
3232                          sbi->s_es->s_mount_opts);
3233         }
3234         if (!parse_options((char *) data, sb, &journal_devnum,
3235                            &journal_ioprio, NULL, 0))
3236                 goto failed_mount;
3237
3238         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3239                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3240                             "with data=journal disables delayed "
3241                             "allocation and O_DIRECT support!\n");
3242                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3243                         ext4_msg(sb, KERN_ERR, "can't mount with "
3244                                  "both data=journal and delalloc");
3245                         goto failed_mount;
3246                 }
3247                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3248                         ext4_msg(sb, KERN_ERR, "can't mount with "
3249                                  "both data=journal and delalloc");
3250                         goto failed_mount;
3251                 }
3252                 if (test_opt(sb, DELALLOC))
3253                         clear_opt(sb, DELALLOC);
3254         }
3255
3256         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3257         if (test_opt(sb, DIOREAD_NOLOCK)) {
3258                 if (blocksize < PAGE_SIZE) {
3259                         ext4_msg(sb, KERN_ERR, "can't mount with "
3260                                  "dioread_nolock if block size != PAGE_SIZE");
3261                         goto failed_mount;
3262                 }
3263         }
3264
3265         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3266                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3267
3268         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3269             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3270              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3271              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3272                 ext4_msg(sb, KERN_WARNING,
3273                        "feature flags set on rev 0 fs, "
3274                        "running e2fsck is recommended");
3275
3276         if (IS_EXT2_SB(sb)) {
3277                 if (ext2_feature_set_ok(sb))
3278                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3279                                  "using the ext4 subsystem");
3280                 else {
3281                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3282                                  "to feature incompatibilities");
3283                         goto failed_mount;
3284                 }
3285         }
3286
3287         if (IS_EXT3_SB(sb)) {
3288                 if (ext3_feature_set_ok(sb))
3289                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3290                                  "using the ext4 subsystem");
3291                 else {
3292                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3293                                  "to feature incompatibilities");
3294                         goto failed_mount;
3295                 }
3296         }
3297
3298         /*
3299          * Check feature flags regardless of the revision level, since we
3300          * previously didn't change the revision level when setting the flags,
3301          * so there is a chance incompat flags are set on a rev 0 filesystem.
3302          */
3303         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3304                 goto failed_mount;
3305
3306         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3307             blocksize > EXT4_MAX_BLOCK_SIZE) {
3308                 ext4_msg(sb, KERN_ERR,
3309                        "Unsupported filesystem blocksize %d", blocksize);
3310                 goto failed_mount;
3311         }
3312
3313         if (sb->s_blocksize != blocksize) {
3314                 /* Validate the filesystem blocksize */
3315                 if (!sb_set_blocksize(sb, blocksize)) {
3316                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3317                                         blocksize);
3318                         goto failed_mount;
3319                 }
3320
3321                 brelse(bh);
3322                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3323                 offset = do_div(logical_sb_block, blocksize);
3324                 bh = sb_bread(sb, logical_sb_block);
3325                 if (!bh) {
3326                         ext4_msg(sb, KERN_ERR,
3327                                "Can't read superblock on 2nd try");
3328                         goto failed_mount;
3329                 }
3330                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3331                 sbi->s_es = es;
3332                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3333                         ext4_msg(sb, KERN_ERR,
3334                                "Magic mismatch, very weird!");
3335                         goto failed_mount;
3336                 }
3337         }
3338
3339         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3340                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3341         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3342                                                       has_huge_files);
3343         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3344
3345         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3346                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3347                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3348         } else {
3349                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3350                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3351                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3352                     (!is_power_of_2(sbi->s_inode_size)) ||
3353                     (sbi->s_inode_size > blocksize)) {
3354                         ext4_msg(sb, KERN_ERR,
3355                                "unsupported inode size: %d",
3356                                sbi->s_inode_size);
3357                         goto failed_mount;
3358                 }
3359                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3360                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3361         }
3362
3363         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3364         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3365                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3366                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3367                     !is_power_of_2(sbi->s_desc_size)) {
3368                         ext4_msg(sb, KERN_ERR,
3369                                "unsupported descriptor size %lu",
3370                                sbi->s_desc_size);
3371                         goto failed_mount;
3372                 }
3373         } else
3374                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3375
3376         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3377         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3378         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3379                 goto cantfind_ext4;
3380
3381         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3382         if (sbi->s_inodes_per_block == 0)
3383                 goto cantfind_ext4;
3384         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3385                                         sbi->s_inodes_per_block;
3386         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3387         sbi->s_sbh = bh;
3388         sbi->s_mount_state = le16_to_cpu(es->s_state);
3389         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3390         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3391
3392         for (i = 0; i < 4; i++)
3393                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3394         sbi->s_def_hash_version = es->s_def_hash_version;
3395         i = le32_to_cpu(es->s_flags);
3396         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3397                 sbi->s_hash_unsigned = 3;
3398         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3399 #ifdef __CHAR_UNSIGNED__
3400                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3401                 sbi->s_hash_unsigned = 3;
3402 #else
3403                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3404 #endif
3405                 sb->s_dirt = 1;
3406         }
3407
3408         /* Handle clustersize */
3409         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3410         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3411                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3412         if (has_bigalloc) {
3413                 if (clustersize < blocksize) {
3414                         ext4_msg(sb, KERN_ERR,
3415                                  "cluster size (%d) smaller than "
3416                                  "block size (%d)", clustersize, blocksize);
3417                         goto failed_mount;
3418                 }
3419                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3420                         le32_to_cpu(es->s_log_block_size);
3421                 sbi->s_clusters_per_group =
3422                         le32_to_cpu(es->s_clusters_per_group);
3423                 if (sbi->s_clusters_per_group > blocksize * 8) {
3424                         ext4_msg(sb, KERN_ERR,
3425                                  "#clusters per group too big: %lu",
3426                                  sbi->s_clusters_per_group);
3427                         goto failed_mount;
3428                 }
3429                 if (sbi->s_blocks_per_group !=
3430                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3431                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3432                                  "clusters per group (%lu) inconsistent",
3433                                  sbi->s_blocks_per_group,
3434                                  sbi->s_clusters_per_group);
3435                         goto failed_mount;
3436                 }
3437         } else {
3438                 if (clustersize != blocksize) {
3439                         ext4_warning(sb, "fragment/cluster size (%d) != "
3440                                      "block size (%d)", clustersize,
3441                                      blocksize);
3442                         clustersize = blocksize;
3443                 }
3444                 if (sbi->s_blocks_per_group > blocksize * 8) {
3445                         ext4_msg(sb, KERN_ERR,
3446                                  "#blocks per group too big: %lu",
3447                                  sbi->s_blocks_per_group);
3448                         goto failed_mount;
3449                 }
3450                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3451                 sbi->s_cluster_bits = 0;
3452         }
3453         sbi->s_cluster_ratio = clustersize / blocksize;
3454
3455         if (sbi->s_inodes_per_group > blocksize * 8) {
3456                 ext4_msg(sb, KERN_ERR,
3457                        "#inodes per group too big: %lu",
3458                        sbi->s_inodes_per_group);
3459                 goto failed_mount;
3460         }
3461
3462         /*
3463          * Test whether we have more sectors than will fit in sector_t,
3464          * and whether the max offset is addressable by the page cache.
3465          */
3466         err = generic_check_addressable(sb->s_blocksize_bits,
3467                                         ext4_blocks_count(es));
3468         if (err) {
3469                 ext4_msg(sb, KERN_ERR, "filesystem"
3470                          " too large to mount safely on this system");
3471                 if (sizeof(sector_t) < 8)
3472                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3473                 ret = err;
3474                 goto failed_mount;
3475         }
3476
3477         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3478                 goto cantfind_ext4;
3479
3480         /* check blocks count against device size */
3481         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3482         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3483                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3484                        "exceeds size of device (%llu blocks)",
3485                        ext4_blocks_count(es), blocks_count);
3486                 goto failed_mount;
3487         }
3488
3489         /*
3490          * It makes no sense for the first data block to be beyond the end
3491          * of the filesystem.
3492          */
3493         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3494                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3495                          "block %u is beyond end of filesystem (%llu)",
3496                          le32_to_cpu(es->s_first_data_block),
3497                          ext4_blocks_count(es));
3498                 goto failed_mount;
3499         }
3500         blocks_count = (ext4_blocks_count(es) -
3501                         le32_to_cpu(es->s_first_data_block) +
3502                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3503         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3504         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3505                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3506                        "(block count %llu, first data block %u, "
3507                        "blocks per group %lu)", sbi->s_groups_count,
3508                        ext4_blocks_count(es),
3509                        le32_to_cpu(es->s_first_data_block),
3510                        EXT4_BLOCKS_PER_GROUP(sb));
3511                 goto failed_mount;
3512         }
3513         sbi->s_groups_count = blocks_count;
3514         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3515                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3516         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3517                    EXT4_DESC_PER_BLOCK(sb);
3518         sbi->s_group_desc = ext4_kvmalloc(db_count *
3519                                           sizeof(struct buffer_head *),
3520                                           GFP_KERNEL);
3521         if (sbi->s_group_desc == NULL) {
3522                 ext4_msg(sb, KERN_ERR, "not enough memory");
3523                 goto failed_mount;
3524         }
3525
3526         if (ext4_proc_root)
3527                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3528
3529         bgl_lock_init(sbi->s_blockgroup_lock);
3530
3531         for (i = 0; i < db_count; i++) {
3532                 block = descriptor_loc(sb, logical_sb_block, i);
3533                 sbi->s_group_desc[i] = sb_bread(sb, block);
3534                 if (!sbi->s_group_desc[i]) {
3535                         ext4_msg(sb, KERN_ERR,
3536                                "can't read group descriptor %d", i);
3537                         db_count = i;
3538                         goto failed_mount2;
3539                 }
3540         }
3541         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3542                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3543                 goto failed_mount2;
3544         }
3545         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3546                 if (!ext4_fill_flex_info(sb)) {
3547                         ext4_msg(sb, KERN_ERR,
3548                                "unable to initialize "
3549                                "flex_bg meta info!");
3550                         goto failed_mount2;
3551                 }
3552
3553         sbi->s_gdb_count = db_count;
3554         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3555         spin_lock_init(&sbi->s_next_gen_lock);
3556
3557         init_timer(&sbi->s_err_report);
3558         sbi->s_err_report.function = print_daily_error_info;
3559         sbi->s_err_report.data = (unsigned long) sb;
3560
3561         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3562                         ext4_count_free_clusters(sb));
3563         if (!err) {
3564                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3565                                 ext4_count_free_inodes(sb));
3566         }
3567         if (!err) {
3568                 err = percpu_counter_init(&sbi->s_dirs_counter,
3569                                 ext4_count_dirs(sb));
3570         }
3571         if (!err) {
3572                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3573         }
3574         if (err) {
3575                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3576                 goto failed_mount3;
3577         }
3578
3579         sbi->s_stripe = ext4_get_stripe_size(sbi);
3580         sbi->s_max_writeback_mb_bump = 128;
3581
3582         /*
3583          * set up enough so that it can read an inode
3584          */
3585         if (!test_opt(sb, NOLOAD) &&
3586             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3587                 sb->s_op = &ext4_sops;
3588         else
3589                 sb->s_op = &ext4_nojournal_sops;
3590         sb->s_export_op = &ext4_export_ops;
3591         sb->s_xattr = ext4_xattr_handlers;
3592 #ifdef CONFIG_QUOTA
3593         sb->s_qcop = &ext4_qctl_operations;
3594         sb->dq_op = &ext4_quota_operations;
3595 #endif
3596         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3597
3598         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3599         mutex_init(&sbi->s_orphan_lock);
3600         sbi->s_resize_flags = 0;
3601
3602         sb->s_root = NULL;
3603
3604         needs_recovery = (es->s_last_orphan != 0 ||
3605                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3606                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3607
3608         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3609             !(sb->s_flags & MS_RDONLY))
3610                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3611                         goto failed_mount3;
3612
3613         /*
3614          * The first inode we look at is the journal inode.  Don't try
3615          * root first: it may be modified in the journal!
3616          */
3617         if (!test_opt(sb, NOLOAD) &&
3618             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3619                 if (ext4_load_journal(sb, es, journal_devnum))
3620                         goto failed_mount3;
3621         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3622               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3623                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3624                        "suppressed and not mounted read-only");
3625                 goto failed_mount_wq;
3626         } else {
3627                 clear_opt(sb, DATA_FLAGS);
3628                 sbi->s_journal = NULL;
3629                 needs_recovery = 0;
3630                 goto no_journal;
3631         }
3632
3633         if (ext4_blocks_count(es) > 0xffffffffULL &&
3634             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3635                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3636                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3637                 goto failed_mount_wq;
3638         }
3639
3640         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3641                 jbd2_journal_set_features(sbi->s_journal,
3642                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3643                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3644         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3645                 jbd2_journal_set_features(sbi->s_journal,
3646                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3647                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3648                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3649         } else {
3650                 jbd2_journal_clear_features(sbi->s_journal,
3651                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3652                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3653         }
3654
3655         /* We have now updated the journal if required, so we can
3656          * validate the data journaling mode. */
3657         switch (test_opt(sb, DATA_FLAGS)) {
3658         case 0:
3659                 /* No mode set, assume a default based on the journal
3660                  * capabilities: ORDERED_DATA if the journal can
3661                  * cope, else JOURNAL_DATA
3662                  */
3663                 if (jbd2_journal_check_available_features
3664                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3665                         set_opt(sb, ORDERED_DATA);
3666                 else
3667                         set_opt(sb, JOURNAL_DATA);
3668                 break;
3669
3670         case EXT4_MOUNT_ORDERED_DATA:
3671         case EXT4_MOUNT_WRITEBACK_DATA:
3672                 if (!jbd2_journal_check_available_features
3673                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3674                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3675                                "requested data journaling mode");
3676                         goto failed_mount_wq;
3677                 }
3678         default:
3679                 break;
3680         }
3681         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3682
3683         /*
3684          * The journal may have updated the bg summary counts, so we
3685          * need to update the global counters.
3686          */
3687         percpu_counter_set(&sbi->s_freeclusters_counter,
3688                            ext4_count_free_clusters(sb));
3689         percpu_counter_set(&sbi->s_freeinodes_counter,
3690                            ext4_count_free_inodes(sb));
3691         percpu_counter_set(&sbi->s_dirs_counter,
3692                            ext4_count_dirs(sb));
3693         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3694
3695 no_journal:
3696         /*
3697          * The maximum number of concurrent works can be high and
3698          * concurrency isn't really necessary.  Limit it to 1.
3699          */
3700         EXT4_SB(sb)->dio_unwritten_wq =
3701                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3702         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3703                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3704                 goto failed_mount_wq;
3705         }
3706
3707         /*
3708          * The jbd2_journal_load will have done any necessary log recovery,
3709          * so we can safely mount the rest of the filesystem now.
3710          */
3711
3712         root = ext4_iget(sb, EXT4_ROOT_INO);
3713         if (IS_ERR(root)) {
3714                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3715                 ret = PTR_ERR(root);
3716                 root = NULL;
3717                 goto failed_mount4;
3718         }
3719         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3720                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3721                 goto failed_mount4;
3722         }
3723         sb->s_root = d_alloc_root(root);
3724         if (!sb->s_root) {
3725                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3726                 ret = -ENOMEM;
3727                 goto failed_mount4;
3728         }
3729
3730         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3731
3732         /* determine the minimum size of new large inodes, if present */
3733         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3734                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3735                                                      EXT4_GOOD_OLD_INODE_SIZE;
3736                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3737                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3738                         if (sbi->s_want_extra_isize <
3739                             le16_to_cpu(es->s_want_extra_isize))
3740                                 sbi->s_want_extra_isize =
3741                                         le16_to_cpu(es->s_want_extra_isize);
3742                         if (sbi->s_want_extra_isize <
3743                             le16_to_cpu(es->s_min_extra_isize))
3744                                 sbi->s_want_extra_isize =
3745                                         le16_to_cpu(es->s_min_extra_isize);
3746                 }
3747         }
3748         /* Check if enough inode space is available */
3749         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3750                                                         sbi->s_inode_size) {
3751                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3752                                                        EXT4_GOOD_OLD_INODE_SIZE;
3753                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3754                          "available");
3755         }
3756
3757         err = ext4_setup_system_zone(sb);
3758         if (err) {
3759                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3760                          "zone (%d)", err);
3761                 goto failed_mount4;
3762         }
3763
3764         ext4_ext_init(sb);
3765         err = ext4_mb_init(sb, needs_recovery);
3766         if (err) {
3767                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3768                          err);
3769                 goto failed_mount5;
3770         }
3771
3772         err = ext4_register_li_request(sb, first_not_zeroed);
3773         if (err)
3774                 goto failed_mount6;
3775
3776         sbi->s_kobj.kset = ext4_kset;
3777         init_completion(&sbi->s_kobj_unregister);
3778         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3779                                    "%s", sb->s_id);
3780         if (err)
3781                 goto failed_mount7;
3782
3783         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3784         ext4_orphan_cleanup(sb, es);
3785         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3786         if (needs_recovery) {
3787                 ext4_msg(sb, KERN_INFO, "recovery complete");
3788                 ext4_mark_recovery_complete(sb, es);
3789         }
3790         if (EXT4_SB(sb)->s_journal) {
3791                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3792                         descr = " journalled data mode";
3793                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3794                         descr = " ordered data mode";
3795                 else
3796                         descr = " writeback data mode";
3797         } else
3798                 descr = "out journal";
3799
3800         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3801                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3802                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3803
3804         if (es->s_error_count)
3805                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3806
3807         kfree(orig_data);
3808         return 0;
3809
3810 cantfind_ext4:
3811         if (!silent)
3812                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3813         goto failed_mount;
3814
3815 failed_mount7:
3816         ext4_unregister_li_request(sb);
3817 failed_mount6:
3818         ext4_ext_release(sb);
3819 failed_mount5:
3820         ext4_mb_release(sb);
3821         ext4_release_system_zone(sb);
3822 failed_mount4:
3823         iput(root);
3824         sb->s_root = NULL;
3825         ext4_msg(sb, KERN_ERR, "mount failed");
3826         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3827 failed_mount_wq:
3828         if (sbi->s_journal) {
3829                 jbd2_journal_destroy(sbi->s_journal);
3830                 sbi->s_journal = NULL;
3831         }
3832 failed_mount3:
3833         del_timer(&sbi->s_err_report);
3834         if (sbi->s_flex_groups)
3835                 ext4_kvfree(sbi->s_flex_groups);
3836         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3837         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3838         percpu_counter_destroy(&sbi->s_dirs_counter);
3839         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3840         if (sbi->s_mmp_tsk)
3841                 kthread_stop(sbi->s_mmp_tsk);
3842 failed_mount2:
3843         for (i = 0; i < db_count; i++)
3844                 brelse(sbi->s_group_desc[i]);
3845         ext4_kvfree(sbi->s_group_desc);
3846 failed_mount:
3847         if (sbi->s_proc) {
3848                 remove_proc_entry(sb->s_id, ext4_proc_root);
3849         }
3850 #ifdef CONFIG_QUOTA
3851         for (i = 0; i < MAXQUOTAS; i++)
3852                 kfree(sbi->s_qf_names[i]);
3853 #endif
3854         ext4_blkdev_remove(sbi);
3855         brelse(bh);
3856 out_fail:
3857         sb->s_fs_info = NULL;
3858         kfree(sbi->s_blockgroup_lock);
3859         kfree(sbi);
3860 out_free_orig:
3861         kfree(orig_data);
3862         return ret;
3863 }
3864
3865 /*
3866  * Setup any per-fs journal parameters now.  We'll do this both on
3867  * initial mount, once the journal has been initialised but before we've
3868  * done any recovery; and again on any subsequent remount.
3869  */
3870 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3871 {
3872         struct ext4_sb_info *sbi = EXT4_SB(sb);
3873
3874         journal->j_commit_interval = sbi->s_commit_interval;
3875         journal->j_min_batch_time = sbi->s_min_batch_time;
3876         journal->j_max_batch_time = sbi->s_max_batch_time;
3877
3878         write_lock(&journal->j_state_lock);
3879         if (test_opt(sb, BARRIER))
3880                 journal->j_flags |= JBD2_BARRIER;
3881         else
3882                 journal->j_flags &= ~JBD2_BARRIER;
3883         if (test_opt(sb, DATA_ERR_ABORT))
3884                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3885         else
3886                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3887         write_unlock(&journal->j_state_lock);
3888 }
3889
3890 static journal_t *ext4_get_journal(struct super_block *sb,
3891                                    unsigned int journal_inum)
3892 {
3893         struct inode *journal_inode;
3894         journal_t *journal;
3895
3896         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3897
3898         /* First, test for the existence of a valid inode on disk.  Bad
3899          * things happen if we iget() an unused inode, as the subsequent
3900          * iput() will try to delete it. */
3901
3902         journal_inode = ext4_iget(sb, journal_inum);
3903         if (IS_ERR(journal_inode)) {
3904                 ext4_msg(sb, KERN_ERR, "no journal found");
3905                 return NULL;
3906         }
3907         if (!journal_inode->i_nlink) {
3908                 make_bad_inode(journal_inode);
3909                 iput(journal_inode);
3910                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3911                 return NULL;
3912         }
3913
3914         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3915                   journal_inode, journal_inode->i_size);
3916         if (!S_ISREG(journal_inode->i_mode)) {
3917                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3918                 iput(journal_inode);
3919                 return NULL;
3920         }
3921
3922         journal = jbd2_journal_init_inode(journal_inode);
3923         if (!journal) {
3924                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3925                 iput(journal_inode);
3926                 return NULL;
3927         }
3928         journal->j_private = sb;
3929         ext4_init_journal_params(sb, journal);
3930         return journal;
3931 }
3932
3933 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3934                                        dev_t j_dev)
3935 {
3936         struct buffer_head *bh;
3937         journal_t *journal;
3938         ext4_fsblk_t start;
3939         ext4_fsblk_t len;
3940         int hblock, blocksize;
3941         ext4_fsblk_t sb_block;
3942         unsigned long offset;
3943         struct ext4_super_block *es;
3944         struct block_device *bdev;
3945
3946         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3947
3948         bdev = ext4_blkdev_get(j_dev, sb);
3949         if (bdev == NULL)
3950                 return NULL;
3951
3952         blocksize = sb->s_blocksize;
3953         hblock = bdev_logical_block_size(bdev);
3954         if (blocksize < hblock) {
3955                 ext4_msg(sb, KERN_ERR,
3956                         "blocksize too small for journal device");
3957                 goto out_bdev;
3958         }
3959
3960         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3961         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3962         set_blocksize(bdev, blocksize);
3963         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3964                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3965                        "external journal");
3966                 goto out_bdev;
3967         }
3968
3969         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3970         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3971             !(le32_to_cpu(es->s_feature_incompat) &
3972               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3973                 ext4_msg(sb, KERN_ERR, "external journal has "
3974                                         "bad superblock");
3975                 brelse(bh);
3976                 goto out_bdev;
3977         }
3978
3979         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3980                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3981                 brelse(bh);
3982                 goto out_bdev;
3983         }
3984
3985         len = ext4_blocks_count(es);
3986         start = sb_block + 1;
3987         brelse(bh);     /* we're done with the superblock */
3988
3989         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3990                                         start, len, blocksize);
3991         if (!journal) {
3992                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3993                 goto out_bdev;
3994         }
3995         journal->j_private = sb;
3996         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3997         wait_on_buffer(journal->j_sb_buffer);
3998         if (!buffer_uptodate(journal->j_sb_buffer)) {
3999                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4000                 goto out_journal;
4001         }
4002         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4003                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4004                                         "user (unsupported) - %d",
4005                         be32_to_cpu(journal->j_superblock->s_nr_users));
4006                 goto out_journal;
4007         }
4008         EXT4_SB(sb)->journal_bdev = bdev;
4009         ext4_init_journal_params(sb, journal);
4010         return journal;
4011
4012 out_journal:
4013         jbd2_journal_destroy(journal);
4014 out_bdev:
4015         ext4_blkdev_put(bdev);
4016         return NULL;
4017 }
4018
4019 static int ext4_load_journal(struct super_block *sb,
4020                              struct ext4_super_block *es,
4021                              unsigned long journal_devnum)
4022 {
4023         journal_t *journal;
4024         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4025         dev_t journal_dev;
4026         int err = 0;
4027         int really_read_only;
4028
4029         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4030
4031         if (journal_devnum &&
4032             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4033                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4034                         "numbers have changed");
4035                 journal_dev = new_decode_dev(journal_devnum);
4036         } else
4037                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4038
4039         really_read_only = bdev_read_only(sb->s_bdev);
4040
4041         /*
4042          * Are we loading a blank journal or performing recovery after a
4043          * crash?  For recovery, we need to check in advance whether we
4044          * can get read-write access to the device.
4045          */
4046         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4047                 if (sb->s_flags & MS_RDONLY) {
4048                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4049                                         "required on readonly filesystem");
4050                         if (really_read_only) {
4051                                 ext4_msg(sb, KERN_ERR, "write access "
4052                                         "unavailable, cannot proceed");
4053                                 return -EROFS;
4054                         }
4055                         ext4_msg(sb, KERN_INFO, "write access will "
4056                                "be enabled during recovery");
4057                 }
4058         }
4059
4060         if (journal_inum && journal_dev) {
4061                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4062                        "and inode journals!");
4063                 return -EINVAL;
4064         }
4065
4066         if (journal_inum) {
4067                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4068                         return -EINVAL;
4069         } else {
4070                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4071                         return -EINVAL;
4072         }
4073
4074         if (!(journal->j_flags & JBD2_BARRIER))
4075                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4076
4077         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4078                 err = jbd2_journal_update_format(journal);
4079                 if (err)  {
4080                         ext4_msg(sb, KERN_ERR, "error updating journal");
4081                         jbd2_journal_destroy(journal);
4082                         return err;
4083                 }
4084         }
4085
4086         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4087                 err = jbd2_journal_wipe(journal, !really_read_only);
4088         if (!err) {
4089                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4090                 if (save)
4091                         memcpy(save, ((char *) es) +
4092                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4093                 err = jbd2_journal_load(journal);
4094                 if (save)
4095                         memcpy(((char *) es) + EXT4_S_ERR_START,
4096                                save, EXT4_S_ERR_LEN);
4097                 kfree(save);
4098         }
4099
4100         if (err) {
4101                 ext4_msg(sb, KERN_ERR, "error loading journal");
4102                 jbd2_journal_destroy(journal);
4103                 return err;
4104         }
4105
4106         EXT4_SB(sb)->s_journal = journal;
4107         ext4_clear_journal_err(sb, es);
4108
4109         if (!really_read_only && journal_devnum &&
4110             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4111                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4112
4113                 /* Make sure we flush the recovery flag to disk. */
4114                 ext4_commit_super(sb, 1);
4115         }
4116
4117         return 0;
4118 }
4119
4120 static int ext4_commit_super(struct super_block *sb, int sync)
4121 {
4122         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4123         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4124         int error = 0;
4125
4126         if (!sbh || block_device_ejected(sb))
4127                 return error;
4128         if (buffer_write_io_error(sbh)) {
4129                 /*
4130                  * Oh, dear.  A previous attempt to write the
4131                  * superblock failed.  This could happen because the
4132                  * USB device was yanked out.  Or it could happen to
4133                  * be a transient write error and maybe the block will
4134                  * be remapped.  Nothing we can do but to retry the
4135                  * write and hope for the best.
4136                  */
4137                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4138                        "superblock detected");
4139                 clear_buffer_write_io_error(sbh);
4140                 set_buffer_uptodate(sbh);
4141         }
4142         /*
4143          * If the file system is mounted read-only, don't update the
4144          * superblock write time.  This avoids updating the superblock
4145          * write time when we are mounting the root file system
4146          * read/only but we need to replay the journal; at that point,
4147          * for people who are east of GMT and who make their clock
4148          * tick in localtime for Windows bug-for-bug compatibility,
4149          * the clock is set in the future, and this will cause e2fsck
4150          * to complain and force a full file system check.
4151          */
4152         if (!(sb->s_flags & MS_RDONLY))
4153                 es->s_wtime = cpu_to_le32(get_seconds());
4154         if (sb->s_bdev->bd_part)
4155                 es->s_kbytes_written =
4156                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4157                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4158                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4159         else
4160                 es->s_kbytes_written =
4161                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4162         ext4_free_blocks_count_set(es,
4163                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4164                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4165         es->s_free_inodes_count =
4166                 cpu_to_le32(percpu_counter_sum_positive(
4167                                 &EXT4_SB(sb)->s_freeinodes_counter));
4168         sb->s_dirt = 0;
4169         BUFFER_TRACE(sbh, "marking dirty");
4170         mark_buffer_dirty(sbh);
4171         if (sync) {
4172                 error = sync_dirty_buffer(sbh);
4173                 if (error)
4174                         return error;
4175
4176                 error = buffer_write_io_error(sbh);
4177                 if (error) {
4178                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4179                                "superblock");
4180                         clear_buffer_write_io_error(sbh);
4181                         set_buffer_uptodate(sbh);
4182                 }
4183         }
4184         return error;
4185 }
4186
4187 /*
4188  * Have we just finished recovery?  If so, and if we are mounting (or
4189  * remounting) the filesystem readonly, then we will end up with a
4190  * consistent fs on disk.  Record that fact.
4191  */
4192 static void ext4_mark_recovery_complete(struct super_block *sb,
4193                                         struct ext4_super_block *es)
4194 {
4195         journal_t *journal = EXT4_SB(sb)->s_journal;
4196
4197         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4198                 BUG_ON(journal != NULL);
4199                 return;
4200         }
4201         jbd2_journal_lock_updates(journal);
4202         if (jbd2_journal_flush(journal) < 0)
4203                 goto out;
4204
4205         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4206             sb->s_flags & MS_RDONLY) {
4207                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4208                 ext4_commit_super(sb, 1);
4209         }
4210
4211 out:
4212         jbd2_journal_unlock_updates(journal);
4213 }
4214
4215 /*
4216  * If we are mounting (or read-write remounting) a filesystem whose journal
4217  * has recorded an error from a previous lifetime, move that error to the
4218  * main filesystem now.
4219  */
4220 static void ext4_clear_journal_err(struct super_block *sb,
4221                                    struct ext4_super_block *es)
4222 {
4223         journal_t *journal;
4224         int j_errno;
4225         const char *errstr;
4226
4227         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4228
4229         journal = EXT4_SB(sb)->s_journal;
4230
4231         /*
4232          * Now check for any error status which may have been recorded in the
4233          * journal by a prior ext4_error() or ext4_abort()
4234          */
4235
4236         j_errno = jbd2_journal_errno(journal);
4237         if (j_errno) {
4238                 char nbuf[16];
4239
4240                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4241                 ext4_warning(sb, "Filesystem error recorded "
4242                              "from previous mount: %s", errstr);
4243                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4244
4245                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4246                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4247                 ext4_commit_super(sb, 1);
4248
4249                 jbd2_journal_clear_err(journal);
4250         }
4251 }
4252
4253 /*
4254  * Force the running and committing transactions to commit,
4255  * and wait on the commit.
4256  */
4257 int ext4_force_commit(struct super_block *sb)
4258 {
4259         journal_t *journal;
4260         int ret = 0;
4261
4262         if (sb->s_flags & MS_RDONLY)
4263                 return 0;
4264
4265         journal = EXT4_SB(sb)->s_journal;
4266         if (journal) {
4267                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4268                 ret = ext4_journal_force_commit(journal);
4269         }
4270
4271         return ret;
4272 }
4273
4274 static void ext4_write_super(struct super_block *sb)
4275 {
4276         lock_super(sb);
4277         ext4_commit_super(sb, 1);
4278         unlock_super(sb);
4279 }
4280
4281 static int ext4_sync_fs(struct super_block *sb, int wait)
4282 {
4283         int ret = 0;
4284         tid_t target;
4285         struct ext4_sb_info *sbi = EXT4_SB(sb);
4286
4287         trace_ext4_sync_fs(sb, wait);
4288         flush_workqueue(sbi->dio_unwritten_wq);
4289         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4290                 if (wait)
4291                         jbd2_log_wait_commit(sbi->s_journal, target);
4292         }
4293         return ret;
4294 }
4295
4296 /*
4297  * LVM calls this function before a (read-only) snapshot is created.  This
4298  * gives us a chance to flush the journal completely and mark the fs clean.
4299  *
4300  * Note that only this function cannot bring a filesystem to be in a clean
4301  * state independently, because ext4 prevents a new handle from being started
4302  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4303  * the upper layer.
4304  */
4305 static int ext4_freeze(struct super_block *sb)
4306 {
4307         int error = 0;
4308         journal_t *journal;
4309
4310         if (sb->s_flags & MS_RDONLY)
4311                 return 0;
4312
4313         journal = EXT4_SB(sb)->s_journal;
4314
4315         /* Now we set up the journal barrier. */
4316         jbd2_journal_lock_updates(journal);
4317
4318         /*
4319          * Don't clear the needs_recovery flag if we failed to flush
4320          * the journal.
4321          */
4322         error = jbd2_journal_flush(journal);
4323         if (error < 0)
4324                 goto out;
4325
4326         /* Journal blocked and flushed, clear needs_recovery flag. */
4327         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4328         error = ext4_commit_super(sb, 1);
4329 out:
4330         /* we rely on s_frozen to stop further updates */
4331         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4332         return error;
4333 }
4334
4335 /*
4336  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4337  * flag here, even though the filesystem is not technically dirty yet.
4338  */
4339 static int ext4_unfreeze(struct super_block *sb)
4340 {
4341         if (sb->s_flags & MS_RDONLY)
4342                 return 0;
4343
4344         lock_super(sb);
4345         /* Reset the needs_recovery flag before the fs is unlocked. */
4346         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4347         ext4_commit_super(sb, 1);
4348         unlock_super(sb);
4349         return 0;
4350 }
4351
4352 /*
4353  * Structure to save mount options for ext4_remount's benefit
4354  */
4355 struct ext4_mount_options {
4356         unsigned long s_mount_opt;
4357         unsigned long s_mount_opt2;
4358         uid_t s_resuid;
4359         gid_t s_resgid;
4360         unsigned long s_commit_interval;
4361         u32 s_min_batch_time, s_max_batch_time;
4362 #ifdef CONFIG_QUOTA
4363         int s_jquota_fmt;
4364         char *s_qf_names[MAXQUOTAS];
4365 #endif
4366 };
4367
4368 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4369 {
4370         struct ext4_super_block *es;
4371         struct ext4_sb_info *sbi = EXT4_SB(sb);
4372         ext4_fsblk_t n_blocks_count = 0;
4373         unsigned long old_sb_flags;
4374         struct ext4_mount_options old_opts;
4375         int enable_quota = 0;
4376         ext4_group_t g;
4377         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4378         int err = 0;
4379 #ifdef CONFIG_QUOTA
4380         int i;
4381 #endif
4382         char *orig_data = kstrdup(data, GFP_KERNEL);
4383
4384         /* Store the original options */
4385         lock_super(sb);
4386         old_sb_flags = sb->s_flags;
4387         old_opts.s_mount_opt = sbi->s_mount_opt;
4388         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4389         old_opts.s_resuid = sbi->s_resuid;
4390         old_opts.s_resgid = sbi->s_resgid;
4391         old_opts.s_commit_interval = sbi->s_commit_interval;
4392         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4393         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4394 #ifdef CONFIG_QUOTA
4395         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4396         for (i = 0; i < MAXQUOTAS; i++)
4397                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4398 #endif
4399         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4400                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4401
4402         /*
4403          * Allow the "check" option to be passed as a remount option.
4404          */
4405         if (!parse_options(data, sb, NULL, &journal_ioprio,
4406                            &n_blocks_count, 1)) {
4407                 err = -EINVAL;
4408                 goto restore_opts;
4409         }
4410
4411         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4412                 ext4_abort(sb, "Abort forced by user");
4413
4414         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4415                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4416
4417         es = sbi->s_es;
4418
4419         if (sbi->s_journal) {
4420                 ext4_init_journal_params(sb, sbi->s_journal);
4421                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4422         }
4423
4424         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4425                 n_blocks_count > ext4_blocks_count(es)) {
4426                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4427                         err = -EROFS;
4428                         goto restore_opts;
4429                 }
4430
4431                 if (*flags & MS_RDONLY) {
4432                         err = dquot_suspend(sb, -1);
4433                         if (err < 0)
4434                                 goto restore_opts;
4435
4436                         /*
4437                          * First of all, the unconditional stuff we have to do
4438                          * to disable replay of the journal when we next remount
4439                          */
4440                         sb->s_flags |= MS_RDONLY;
4441
4442                         /*
4443                          * OK, test if we are remounting a valid rw partition
4444                          * readonly, and if so set the rdonly flag and then
4445                          * mark the partition as valid again.
4446                          */
4447                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4448                             (sbi->s_mount_state & EXT4_VALID_FS))
4449                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4450
4451                         if (sbi->s_journal)
4452                                 ext4_mark_recovery_complete(sb, es);
4453                 } else {
4454                         /* Make sure we can mount this feature set readwrite */
4455                         if (!ext4_feature_set_ok(sb, 0)) {
4456                                 err = -EROFS;
4457                                 goto restore_opts;
4458                         }
4459                         /*
4460                          * Make sure the group descriptor checksums
4461                          * are sane.  If they aren't, refuse to remount r/w.
4462                          */
4463                         for (g = 0; g < sbi->s_groups_count; g++) {
4464                                 struct ext4_group_desc *gdp =
4465                                         ext4_get_group_desc(sb, g, NULL);
4466
4467                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4468                                         ext4_msg(sb, KERN_ERR,
4469                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4470                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4471                                                le16_to_cpu(gdp->bg_checksum));
4472                                         err = -EINVAL;
4473                                         goto restore_opts;
4474                                 }
4475                         }
4476
4477                         /*
4478                          * If we have an unprocessed orphan list hanging
4479                          * around from a previously readonly bdev mount,
4480                          * require a full umount/remount for now.
4481                          */
4482                         if (es->s_last_orphan) {
4483                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4484                                        "remount RDWR because of unprocessed "
4485                                        "orphan inode list.  Please "
4486                                        "umount/remount instead");
4487                                 err = -EINVAL;
4488                                 goto restore_opts;
4489                         }
4490
4491                         /*
4492                          * Mounting a RDONLY partition read-write, so reread
4493                          * and store the current valid flag.  (It may have
4494                          * been changed by e2fsck since we originally mounted
4495                          * the partition.)
4496                          */
4497                         if (sbi->s_journal)
4498                                 ext4_clear_journal_err(sb, es);
4499                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4500                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4501                                 goto restore_opts;
4502                         if (!ext4_setup_super(sb, es, 0))
4503                                 sb->s_flags &= ~MS_RDONLY;
4504                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4505                                                      EXT4_FEATURE_INCOMPAT_MMP))
4506                                 if (ext4_multi_mount_protect(sb,
4507                                                 le64_to_cpu(es->s_mmp_block))) {
4508                                         err = -EROFS;
4509                                         goto restore_opts;
4510                                 }
4511                         enable_quota = 1;
4512                 }
4513         }
4514
4515         /*
4516          * Reinitialize lazy itable initialization thread based on
4517          * current settings
4518          */
4519         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4520                 ext4_unregister_li_request(sb);
4521         else {
4522                 ext4_group_t first_not_zeroed;
4523                 first_not_zeroed = ext4_has_uninit_itable(sb);
4524                 ext4_register_li_request(sb, first_not_zeroed);
4525         }
4526
4527         ext4_setup_system_zone(sb);
4528         if (sbi->s_journal == NULL)
4529                 ext4_commit_super(sb, 1);
4530
4531 #ifdef CONFIG_QUOTA
4532         /* Release old quota file names */
4533         for (i = 0; i < MAXQUOTAS; i++)
4534                 if (old_opts.s_qf_names[i] &&
4535                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4536                         kfree(old_opts.s_qf_names[i]);
4537 #endif
4538         unlock_super(sb);
4539         if (enable_quota)
4540                 dquot_resume(sb, -1);
4541
4542         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4543         kfree(orig_data);
4544         return 0;
4545
4546 restore_opts:
4547         sb->s_flags = old_sb_flags;
4548         sbi->s_mount_opt = old_opts.s_mount_opt;
4549         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4550         sbi->s_resuid = old_opts.s_resuid;
4551         sbi->s_resgid = old_opts.s_resgid;
4552         sbi->s_commit_interval = old_opts.s_commit_interval;
4553         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4554         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4555 #ifdef CONFIG_QUOTA
4556         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4557         for (i = 0; i < MAXQUOTAS; i++) {
4558                 if (sbi->s_qf_names[i] &&
4559                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4560                         kfree(sbi->s_qf_names[i]);
4561                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4562         }
4563 #endif
4564         unlock_super(sb);
4565         kfree(orig_data);
4566         return err;
4567 }
4568
4569 /*
4570  * Note: calculating the overhead so we can be compatible with
4571  * historical BSD practice is quite difficult in the face of
4572  * clusters/bigalloc.  This is because multiple metadata blocks from
4573  * different block group can end up in the same allocation cluster.
4574  * Calculating the exact overhead in the face of clustered allocation
4575  * requires either O(all block bitmaps) in memory or O(number of block
4576  * groups**2) in time.  We will still calculate the superblock for
4577  * older file systems --- and if we come across with a bigalloc file
4578  * system with zero in s_overhead_clusters the estimate will be close to
4579  * correct especially for very large cluster sizes --- but for newer
4580  * file systems, it's better to calculate this figure once at mkfs
4581  * time, and store it in the superblock.  If the superblock value is
4582  * present (even for non-bigalloc file systems), we will use it.
4583  */
4584 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4585 {
4586         struct super_block *sb = dentry->d_sb;
4587         struct ext4_sb_info *sbi = EXT4_SB(sb);
4588         struct ext4_super_block *es = sbi->s_es;
4589         struct ext4_group_desc *gdp;
4590         u64 fsid;
4591         s64 bfree;
4592
4593         if (test_opt(sb, MINIX_DF)) {
4594                 sbi->s_overhead_last = 0;
4595         } else if (es->s_overhead_clusters) {
4596                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4597         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4598                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4599                 ext4_fsblk_t overhead = 0;
4600
4601                 /*
4602                  * Compute the overhead (FS structures).  This is constant
4603                  * for a given filesystem unless the number of block groups
4604                  * changes so we cache the previous value until it does.
4605                  */
4606
4607                 /*
4608                  * All of the blocks before first_data_block are
4609                  * overhead
4610                  */
4611                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4612
4613                 /*
4614                  * Add the overhead found in each block group
4615                  */
4616                 for (i = 0; i < ngroups; i++) {
4617                         gdp = ext4_get_group_desc(sb, i, NULL);
4618                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4619                         cond_resched();
4620                 }
4621                 sbi->s_overhead_last = overhead;
4622                 smp_wmb();
4623                 sbi->s_blocks_last = ext4_blocks_count(es);
4624         }
4625
4626         buf->f_type = EXT4_SUPER_MAGIC;
4627         buf->f_bsize = sb->s_blocksize;
4628         buf->f_blocks = (ext4_blocks_count(es) -
4629                          EXT4_C2B(sbi, sbi->s_overhead_last));
4630         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4631                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4632         /* prevent underflow in case that few free space is available */
4633         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4634         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4635         if (buf->f_bfree < ext4_r_blocks_count(es))
4636                 buf->f_bavail = 0;
4637         buf->f_files = le32_to_cpu(es->s_inodes_count);
4638         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4639         buf->f_namelen = EXT4_NAME_LEN;
4640         fsid = le64_to_cpup((void *)es->s_uuid) ^
4641                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4642         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4643         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4644
4645         return 0;
4646 }
4647
4648 /* Helper function for writing quotas on sync - we need to start transaction
4649  * before quota file is locked for write. Otherwise the are possible deadlocks:
4650  * Process 1                         Process 2
4651  * ext4_create()                     quota_sync()
4652  *   jbd2_journal_start()                  write_dquot()
4653  *   dquot_initialize()                         down(dqio_mutex)
4654  *     down(dqio_mutex)                    jbd2_journal_start()
4655  *
4656  */
4657
4658 #ifdef CONFIG_QUOTA
4659
4660 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4661 {
4662         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4663 }
4664
4665 static int ext4_write_dquot(struct dquot *dquot)
4666 {
4667         int ret, err;
4668         handle_t *handle;
4669         struct inode *inode;
4670
4671         inode = dquot_to_inode(dquot);
4672         handle = ext4_journal_start(inode,
4673                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4674         if (IS_ERR(handle))
4675                 return PTR_ERR(handle);
4676         ret = dquot_commit(dquot);
4677         err = ext4_journal_stop(handle);
4678         if (!ret)
4679                 ret = err;
4680         return ret;
4681 }
4682
4683 static int ext4_acquire_dquot(struct dquot *dquot)
4684 {
4685         int ret, err;
4686         handle_t *handle;
4687
4688         handle = ext4_journal_start(dquot_to_inode(dquot),
4689                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4690         if (IS_ERR(handle))
4691                 return PTR_ERR(handle);
4692         ret = dquot_acquire(dquot);
4693         err = ext4_journal_stop(handle);
4694         if (!ret)
4695                 ret = err;
4696         return ret;
4697 }
4698
4699 static int ext4_release_dquot(struct dquot *dquot)
4700 {
4701         int ret, err;
4702         handle_t *handle;
4703
4704         handle = ext4_journal_start(dquot_to_inode(dquot),
4705                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4706         if (IS_ERR(handle)) {
4707                 /* Release dquot anyway to avoid endless cycle in dqput() */
4708                 dquot_release(dquot);
4709                 return PTR_ERR(handle);
4710         }
4711         ret = dquot_release(dquot);
4712         err = ext4_journal_stop(handle);
4713         if (!ret)
4714                 ret = err;
4715         return ret;
4716 }
4717
4718 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4719 {
4720         /* Are we journaling quotas? */
4721         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4722             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4723                 dquot_mark_dquot_dirty(dquot);
4724                 return ext4_write_dquot(dquot);
4725         } else {
4726                 return dquot_mark_dquot_dirty(dquot);
4727         }
4728 }
4729
4730 static int ext4_write_info(struct super_block *sb, int type)
4731 {
4732         int ret, err;
4733         handle_t *handle;
4734
4735         /* Data block + inode block */
4736         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4737         if (IS_ERR(handle))
4738                 return PTR_ERR(handle);
4739         ret = dquot_commit_info(sb, type);
4740         err = ext4_journal_stop(handle);
4741         if (!ret)
4742                 ret = err;
4743         return ret;
4744 }
4745
4746 /*
4747  * Turn on quotas during mount time - we need to find
4748  * the quota file and such...
4749  */
4750 static int ext4_quota_on_mount(struct super_block *sb, int type)
4751 {
4752         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4753                                         EXT4_SB(sb)->s_jquota_fmt, type);
4754 }
4755
4756 /*
4757  * Standard function to be called on quota_on
4758  */
4759 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4760                          struct path *path)
4761 {
4762         int err;
4763
4764         if (!test_opt(sb, QUOTA))
4765                 return -EINVAL;
4766
4767         /* Quotafile not on the same filesystem? */
4768         if (path->mnt->mnt_sb != sb)
4769                 return -EXDEV;
4770         /* Journaling quota? */
4771         if (EXT4_SB(sb)->s_qf_names[type]) {
4772                 /* Quotafile not in fs root? */
4773                 if (path->dentry->d_parent != sb->s_root)
4774                         ext4_msg(sb, KERN_WARNING,
4775                                 "Quota file not on filesystem root. "
4776                                 "Journaled quota will not work");
4777         }
4778
4779         /*
4780          * When we journal data on quota file, we have to flush journal to see
4781          * all updates to the file when we bypass pagecache...
4782          */
4783         if (EXT4_SB(sb)->s_journal &&
4784             ext4_should_journal_data(path->dentry->d_inode)) {
4785                 /*
4786                  * We don't need to lock updates but journal_flush() could
4787                  * otherwise be livelocked...
4788                  */
4789                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4790                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4791                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4792                 if (err)
4793                         return err;
4794         }
4795
4796         return dquot_quota_on(sb, type, format_id, path);
4797 }
4798
4799 static int ext4_quota_off(struct super_block *sb, int type)
4800 {
4801         struct inode *inode = sb_dqopt(sb)->files[type];
4802         handle_t *handle;
4803
4804         /* Force all delayed allocation blocks to be allocated.
4805          * Caller already holds s_umount sem */
4806         if (test_opt(sb, DELALLOC))
4807                 sync_filesystem(sb);
4808
4809         if (!inode)
4810                 goto out;
4811
4812         /* Update modification times of quota files when userspace can
4813          * start looking at them */
4814         handle = ext4_journal_start(inode, 1);
4815         if (IS_ERR(handle))
4816                 goto out;
4817         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4818         ext4_mark_inode_dirty(handle, inode);
4819         ext4_journal_stop(handle);
4820
4821 out:
4822         return dquot_quota_off(sb, type);
4823 }
4824
4825 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4826  * acquiring the locks... As quota files are never truncated and quota code
4827  * itself serializes the operations (and no one else should touch the files)
4828  * we don't have to be afraid of races */
4829 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4830                                size_t len, loff_t off)
4831 {
4832         struct inode *inode = sb_dqopt(sb)->files[type];
4833         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4834         int err = 0;
4835         int offset = off & (sb->s_blocksize - 1);
4836         int tocopy;
4837         size_t toread;
4838         struct buffer_head *bh;
4839         loff_t i_size = i_size_read(inode);
4840
4841         if (off > i_size)
4842                 return 0;
4843         if (off+len > i_size)
4844                 len = i_size-off;
4845         toread = len;
4846         while (toread > 0) {
4847                 tocopy = sb->s_blocksize - offset < toread ?
4848                                 sb->s_blocksize - offset : toread;
4849                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4850                 if (err)
4851                         return err;
4852                 if (!bh)        /* A hole? */
4853                         memset(data, 0, tocopy);
4854                 else
4855                         memcpy(data, bh->b_data+offset, tocopy);
4856                 brelse(bh);
4857                 offset = 0;
4858                 toread -= tocopy;
4859                 data += tocopy;
4860                 blk++;
4861         }
4862         return len;
4863 }
4864
4865 /* Write to quotafile (we know the transaction is already started and has
4866  * enough credits) */
4867 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4868                                 const char *data, size_t len, loff_t off)
4869 {
4870         struct inode *inode = sb_dqopt(sb)->files[type];
4871         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4872         int err = 0;
4873         int offset = off & (sb->s_blocksize - 1);
4874         struct buffer_head *bh;
4875         handle_t *handle = journal_current_handle();
4876
4877         if (EXT4_SB(sb)->s_journal && !handle) {
4878                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4879                         " cancelled because transaction is not started",
4880                         (unsigned long long)off, (unsigned long long)len);
4881                 return -EIO;
4882         }
4883         /*
4884          * Since we account only one data block in transaction credits,
4885          * then it is impossible to cross a block boundary.
4886          */
4887         if (sb->s_blocksize - offset < len) {
4888                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4889                         " cancelled because not block aligned",
4890                         (unsigned long long)off, (unsigned long long)len);
4891                 return -EIO;
4892         }
4893
4894         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4895         bh = ext4_bread(handle, inode, blk, 1, &err);
4896         if (!bh)
4897                 goto out;
4898         err = ext4_journal_get_write_access(handle, bh);
4899         if (err) {
4900                 brelse(bh);
4901                 goto out;
4902         }
4903         lock_buffer(bh);
4904         memcpy(bh->b_data+offset, data, len);
4905         flush_dcache_page(bh->b_page);
4906         unlock_buffer(bh);
4907         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4908         brelse(bh);
4909 out:
4910         if (err) {
4911                 mutex_unlock(&inode->i_mutex);
4912                 return err;
4913         }
4914         if (inode->i_size < off + len) {
4915                 i_size_write(inode, off + len);
4916                 EXT4_I(inode)->i_disksize = inode->i_size;
4917                 ext4_mark_inode_dirty(handle, inode);
4918         }
4919         mutex_unlock(&inode->i_mutex);
4920         return len;
4921 }
4922
4923 #endif
4924
4925 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4926                        const char *dev_name, void *data)
4927 {
4928         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4929 }
4930
4931 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4932 static inline void register_as_ext2(void)
4933 {
4934         int err = register_filesystem(&ext2_fs_type);
4935         if (err)
4936                 printk(KERN_WARNING
4937                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4938 }
4939
4940 static inline void unregister_as_ext2(void)
4941 {
4942         unregister_filesystem(&ext2_fs_type);
4943 }
4944
4945 static inline int ext2_feature_set_ok(struct super_block *sb)
4946 {
4947         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4948                 return 0;
4949         if (sb->s_flags & MS_RDONLY)
4950                 return 1;
4951         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4952                 return 0;
4953         return 1;
4954 }
4955 MODULE_ALIAS("ext2");
4956 #else
4957 static inline void register_as_ext2(void) { }
4958 static inline void unregister_as_ext2(void) { }
4959 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4960 #endif
4961
4962 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4963 static inline void register_as_ext3(void)
4964 {
4965         int err = register_filesystem(&ext3_fs_type);
4966         if (err)
4967                 printk(KERN_WARNING
4968                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4969 }
4970
4971 static inline void unregister_as_ext3(void)
4972 {
4973         unregister_filesystem(&ext3_fs_type);
4974 }
4975
4976 static inline int ext3_feature_set_ok(struct super_block *sb)
4977 {
4978         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4979                 return 0;
4980         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4981                 return 0;
4982         if (sb->s_flags & MS_RDONLY)
4983                 return 1;
4984         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4985                 return 0;
4986         return 1;
4987 }
4988 MODULE_ALIAS("ext3");
4989 #else
4990 static inline void register_as_ext3(void) { }
4991 static inline void unregister_as_ext3(void) { }
4992 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4993 #endif
4994
4995 static struct file_system_type ext4_fs_type = {
4996         .owner          = THIS_MODULE,
4997         .name           = "ext4",
4998         .mount          = ext4_mount,
4999         .kill_sb        = kill_block_super,
5000         .fs_flags       = FS_REQUIRES_DEV,
5001 };
5002
5003 static int __init ext4_init_feat_adverts(void)
5004 {
5005         struct ext4_features *ef;
5006         int ret = -ENOMEM;
5007
5008         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5009         if (!ef)
5010                 goto out;
5011
5012         ef->f_kobj.kset = ext4_kset;
5013         init_completion(&ef->f_kobj_unregister);
5014         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5015                                    "features");
5016         if (ret) {
5017                 kfree(ef);
5018                 goto out;
5019         }
5020
5021         ext4_feat = ef;
5022         ret = 0;
5023 out:
5024         return ret;
5025 }
5026
5027 static void ext4_exit_feat_adverts(void)
5028 {
5029         kobject_put(&ext4_feat->f_kobj);
5030         wait_for_completion(&ext4_feat->f_kobj_unregister);
5031         kfree(ext4_feat);
5032 }
5033
5034 /* Shared across all ext4 file systems */
5035 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5036 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5037
5038 static int __init ext4_init_fs(void)
5039 {
5040         int i, err;
5041
5042         ext4_check_flag_values();
5043
5044         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5045                 mutex_init(&ext4__aio_mutex[i]);
5046                 init_waitqueue_head(&ext4__ioend_wq[i]);
5047         }
5048
5049         err = ext4_init_pageio();
5050         if (err)
5051                 return err;
5052         err = ext4_init_system_zone();
5053         if (err)
5054                 goto out6;
5055         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5056         if (!ext4_kset)
5057                 goto out5;
5058         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5059
5060         err = ext4_init_feat_adverts();
5061         if (err)
5062                 goto out4;
5063
5064         err = ext4_init_mballoc();
5065         if (err)
5066                 goto out3;
5067
5068         err = ext4_init_xattr();
5069         if (err)
5070                 goto out2;
5071         err = init_inodecache();
5072         if (err)
5073                 goto out1;
5074         register_as_ext3();
5075         register_as_ext2();
5076         err = register_filesystem(&ext4_fs_type);
5077         if (err)
5078                 goto out;
5079
5080         ext4_li_info = NULL;
5081         mutex_init(&ext4_li_mtx);
5082         return 0;
5083 out:
5084         unregister_as_ext2();
5085         unregister_as_ext3();
5086         destroy_inodecache();
5087 out1:
5088         ext4_exit_xattr();
5089 out2:
5090         ext4_exit_mballoc();
5091 out3:
5092         ext4_exit_feat_adverts();
5093 out4:
5094         if (ext4_proc_root)
5095                 remove_proc_entry("fs/ext4", NULL);
5096         kset_unregister(ext4_kset);
5097 out5:
5098         ext4_exit_system_zone();
5099 out6:
5100         ext4_exit_pageio();
5101         return err;
5102 }
5103
5104 static void __exit ext4_exit_fs(void)
5105 {
5106         ext4_destroy_lazyinit_thread();
5107         unregister_as_ext2();
5108         unregister_as_ext3();
5109         unregister_filesystem(&ext4_fs_type);
5110         destroy_inodecache();
5111         ext4_exit_xattr();
5112         ext4_exit_mballoc();
5113         ext4_exit_feat_adverts();
5114         remove_proc_entry("fs/ext4", NULL);
5115         kset_unregister(ext4_kset);
5116         ext4_exit_system_zone();
5117         ext4_exit_pageio();
5118 }
5119
5120 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5121 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5122 MODULE_LICENSE("GPL");
5123 module_init(ext4_init_fs)
5124 module_exit(ext4_exit_fs)