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         save_error_info(sb, function, line);
 484
 485         ext4_handle_error(sb);
 486 }
 487
 488 void ext4_error_inode(struct inode *inode, const char *function,
 489                       unsigned int line, ext4_fsblk_t block,
 490                       const char *fmt, ...)
 491 {
 492         va_list args;
 493         struct va_format vaf;
 494         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 495
 496         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 497         es->s_last_error_block = cpu_to_le64(block);
 498         save_error_info(inode->i_sb, function, line);
 499         va_start(args, fmt);
 500         vaf.fmt = fmt;
 501         vaf.va = &args;
 502         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 503                inode->i_sb->s_id, function, line, inode->i_ino);
 504         if (block)
 505                 printk(KERN_CONT "block %llu: ", block);
 506         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 507         va_end(args);
 508
 509         ext4_handle_error(inode->i_sb);
 510 }
 511
 512 void ext4_error_file(struct file *file, const char *function,
 513                      unsigned int line, ext4_fsblk_t block,
 514                      const char *fmt, ...)
 515 {
 516         va_list args;
 517         struct va_format vaf;
 518         struct ext4_super_block *es;
 519         struct inode *inode = file->f_dentry->d_inode;
 520         char pathname[80], *path;
 521
 522         es = EXT4_SB(inode->i_sb)->s_es;
 523         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 524         save_error_info(inode->i_sb, function, line);
 525         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 526         if (IS_ERR(path))
 527                 path = "(unknown)";
 528         printk(KERN_CRIT
 529                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 530                inode->i_sb->s_id, function, line, inode->i_ino);
 531         if (block)
 532                 printk(KERN_CONT "block %llu: ", block);
 533         va_start(args, fmt);
 534         vaf.fmt = fmt;
 535         vaf.va = &args;
 536         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 537         va_end(args);
 538
 539         ext4_handle_error(inode->i_sb);
 540 }
 541
 542 static const char *ext4_decode_error(struct super_block *sb, int errno,
 543                                      char nbuf[16])
 544 {
 545         char *errstr = NULL;
 546
 547         switch (errno) {
 548         case -EIO:
 549                 errstr = "IO failure";
 550                 break;
 551         case -ENOMEM:
 552                 errstr = "Out of memory";
 553                 break;
 554         case -EROFS:
 555                 if (!sb || (EXT4_SB(sb)->s_journal &&
 556                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 557                         errstr = "Journal has aborted";
 558                 else
 559                         errstr = "Readonly filesystem";
 560                 break;
 561         default:
 562                 /* If the caller passed in an extra buffer for unknown
 563                  * errors, textualise them now.  Else we just return
 564                  * NULL. */
 565                 if (nbuf) {
 566                         /* Check for truncated error codes... */
 567                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 568                                 errstr = nbuf;
 569                 }
 570                 break;
 571         }
 572
 573         return errstr;
 574 }
 575
 576 /* __ext4_std_error decodes expected errors from journaling functions
 577  * automatically and invokes the appropriate error response.  */
 578
 579 void __ext4_std_error(struct super_block *sb, const char *function,
 580                       unsigned int line, int errno)
 581 {
 582         char nbuf[16];
 583         const char *errstr;
 584
 585         /* Special case: if the error is EROFS, and we're not already
 586          * inside a transaction, then there's really no point in logging
 587          * an error. */
 588         if (errno == -EROFS && journal_current_handle() == NULL &&
 589             (sb->s_flags & MS_RDONLY))
 590                 return;
 591
 592         errstr = ext4_decode_error(sb, errno, nbuf);
 593         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 594                sb->s_id, function, line, errstr);
 595         save_error_info(sb, function, line);
 596
 597         ext4_handle_error(sb);
 598 }
 599
 600 /*
 601  * ext4_abort is a much stronger failure handler than ext4_error.  The
 602  * abort function may be used to deal with unrecoverable failures such
 603  * as journal IO errors or ENOMEM at a critical moment in log management.
 604  *
 605  * We unconditionally force the filesystem into an ABORT|READONLY state,
 606  * unless the error response on the fs has been set to panic in which
 607  * case we take the easy way out and panic immediately.
 608  */
 609
 610 void __ext4_abort(struct super_block *sb, const char *function,
 611                 unsigned int line, const char *fmt, ...)
 612 {
 613         va_list args;
 614
 615         save_error_info(sb, function, line);
 616         va_start(args, fmt);
 617         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 618                function, line);
 619         vprintk(fmt, args);
 620         printk("\n");
 621         va_end(args);
 622
 623         if ((sb->s_flags & MS_RDONLY) == 0) {
 624                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 625                 sb->s_flags |= MS_RDONLY;
 626                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 627                 if (EXT4_SB(sb)->s_journal)
 628                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 629                 save_error_info(sb, function, line);
 630         }
 631         if (test_opt(sb, ERRORS_PANIC))
 632                 panic("EXT4-fs panic from previous error\n");
 633 }
 634
 635 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 636 {
 637         struct va_format vaf;
 638         va_list args;
 639
 640         va_start(args, fmt);
 641         vaf.fmt = fmt;
 642         vaf.va = &args;
 643         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 644         va_end(args);
 645 }
 646
 647 void __ext4_warning(struct super_block *sb, const char *function,
 648                     unsigned int line, const char *fmt, ...)
 649 {
 650         struct va_format vaf;
 651         va_list args;
 652
 653         va_start(args, fmt);
 654         vaf.fmt = fmt;
 655         vaf.va = &args;
 656         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 657                sb->s_id, function, line, &vaf);
 658         va_end(args);
 659 }
 660
 661 void __ext4_grp_locked_error(const char *function, unsigned int line,
 662                              struct super_block *sb, ext4_group_t grp,
 663                              unsigned long ino, ext4_fsblk_t block,
 664                              const char *fmt, ...)
 665 __releases(bitlock)
 666 __acquires(bitlock)
 667 {
 668         struct va_format vaf;
 669         va_list args;
 670         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 671
 672         es->s_last_error_ino = cpu_to_le32(ino);
 673         es->s_last_error_block = cpu_to_le64(block);
 674         __save_error_info(sb, function, line);
 675
 676         va_start(args, fmt);
 677
 678         vaf.fmt = fmt;
 679         vaf.va = &args;
 680         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 681                sb->s_id, function, line, grp);
 682         if (ino)
 683                 printk(KERN_CONT "inode %lu: ", ino);
 684         if (block)
 685                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 686         printk(KERN_CONT "%pV\n", &vaf);
 687         va_end(args);
 688
 689         if (test_opt(sb, ERRORS_CONT)) {
 690                 ext4_commit_super(sb, 0);
 691                 return;
 692         }
 693
 694         ext4_unlock_group(sb, grp);
 695         ext4_handle_error(sb);
 696         /*
 697          * We only get here in the ERRORS_RO case; relocking the group
 698          * may be dangerous, but nothing bad will happen since the
 699          * filesystem will have already been marked read/only and the
 700          * journal has been aborted.  We return 1 as a hint to callers
 701          * who might what to use the return value from
 702          * ext4_grp_locked_error() to distinguish between the
 703          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 704          * aggressively from the ext4 function in question, with a
 705          * more appropriate error code.
 706          */
 707         ext4_lock_group(sb, grp);
 708         return;
 709 }
 710
 711 void ext4_update_dynamic_rev(struct super_block *sb)
 712 {
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 716                 return;
 717
 718         ext4_warning(sb,
 719                      "updating to rev %d because of new feature flag, "
 720                      "running e2fsck is recommended",
 721                      EXT4_DYNAMIC_REV);
 722
 723         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 724         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 725         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 726         /* leave es->s_feature_*compat flags alone */
 727         /* es->s_uuid will be set by e2fsck if empty */
 728
 729         /*
 730          * The rest of the superblock fields should be zero, and if not it
 731          * means they are likely already in use, so leave them alone.  We
 732          * can leave it up to e2fsck to clean up any inconsistencies there.
 733          */
 734 }
 735
 736 /*
 737  * Open the external journal device
 738  */
 739 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 740 {
 741         struct block_device *bdev;
 742         char b[BDEVNAME_SIZE];
 743
 744         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 745         if (IS_ERR(bdev))
 746                 goto fail;
 747         return bdev;
 748
 749 fail:
 750         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 751                         __bdevname(dev, b), PTR_ERR(bdev));
 752         return NULL;
 753 }
 754
 755 /*
 756  * Release the journal device
 757  */
 758 static int ext4_blkdev_put(struct block_device *bdev)
 759 {
 760         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 761 }
 762
 763 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 764 {
 765         struct block_device *bdev;
 766         int ret = -ENODEV;
 767
 768         bdev = sbi->journal_bdev;
 769         if (bdev) {
 770                 ret = ext4_blkdev_put(bdev);
 771                 sbi->journal_bdev = NULL;
 772         }
 773         return ret;
 774 }
 775
 776 static inline struct inode *orphan_list_entry(struct list_head *l)
 777 {
 778         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 779 }
 780
 781 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 782 {
 783         struct list_head *l;
 784
 785         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 786                  le32_to_cpu(sbi->s_es->s_last_orphan));
 787
 788         printk(KERN_ERR "sb_info orphan list:\n");
 789         list_for_each(l, &sbi->s_orphan) {
 790                 struct inode *inode = orphan_list_entry(l);
 791                 printk(KERN_ERR "  "
 792                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 793                        inode->i_sb->s_id, inode->i_ino, inode,
 794                        inode->i_mode, inode->i_nlink,
 795                        NEXT_ORPHAN(inode));
 796         }
 797 }
 798
 799 static void ext4_put_super(struct super_block *sb)
 800 {
 801         struct ext4_sb_info *sbi = EXT4_SB(sb);
 802         struct ext4_super_block *es = sbi->s_es;
 803         int i, err;
 804
 805         ext4_unregister_li_request(sb);
 806         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 807
 808         flush_workqueue(sbi->dio_unwritten_wq);
 809         destroy_workqueue(sbi->dio_unwritten_wq);
 810
 811         lock_super(sb);
 812         if (sb->s_dirt)
 813                 ext4_commit_super(sb, 1);
 814
 815         if (sbi->s_journal) {
 816                 err = jbd2_journal_destroy(sbi->s_journal);
 817                 sbi->s_journal = NULL;
 818                 if (err < 0)
 819                         ext4_abort(sb, "Couldn't clean up the journal");
 820         }
 821
 822         del_timer(&sbi->s_err_report);
 823         ext4_release_system_zone(sb);
 824         ext4_mb_release(sb);
 825         ext4_ext_release(sb);
 826         ext4_xattr_put_super(sb);
 827
 828         if (!(sb->s_flags & MS_RDONLY)) {
 829                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 830                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 831                 ext4_commit_super(sb, 1);
 832         }
 833         if (sbi->s_proc) {
 834                 remove_proc_entry(sb->s_id, ext4_proc_root);
 835         }
 836         kobject_del(&sbi->s_kobj);
 837
 838         for (i = 0; i < sbi->s_gdb_count; i++)
 839                 brelse(sbi->s_group_desc[i]);
 840         ext4_kvfree(sbi->s_group_desc);
 841         ext4_kvfree(sbi->s_flex_groups);
 842         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 843         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 844         percpu_counter_destroy(&sbi->s_dirs_counter);
 845         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 846         brelse(sbi->s_sbh);
 847 #ifdef CONFIG_QUOTA
 848         for (i = 0; i < MAXQUOTAS; i++)
 849                 kfree(sbi->s_qf_names[i]);
 850 #endif
 851
 852         /* Debugging code just in case the in-memory inode orphan list
 853          * isn't empty.  The on-disk one can be non-empty if we've
 854          * detected an error and taken the fs readonly, but the
 855          * in-memory list had better be clean by this point. */
 856         if (!list_empty(&sbi->s_orphan))
 857                 dump_orphan_list(sb, sbi);
 858         J_ASSERT(list_empty(&sbi->s_orphan));
 859
 860         invalidate_bdev(sb->s_bdev);
 861         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 862                 /*
 863                  * Invalidate the journal device's buffers.  We don't want them
 864                  * floating about in memory - the physical journal device may
 865                  * hotswapped, and it breaks the `ro-after' testing code.
 866                  */
 867                 sync_blockdev(sbi->journal_bdev);
 868                 invalidate_bdev(sbi->journal_bdev);
 869                 ext4_blkdev_remove(sbi);
 870         }
 871         if (sbi->s_mmp_tsk)
 872                 kthread_stop(sbi->s_mmp_tsk);
 873         sb->s_fs_info = NULL;
 874         /*
 875          * Now that we are completely done shutting down the
 876          * superblock, we need to actually destroy the kobject.
 877          */
 878         unlock_super(sb);
 879         kobject_put(&sbi->s_kobj);
 880         wait_for_completion(&sbi->s_kobj_unregister);
 881         kfree(sbi->s_blockgroup_lock);
 882         kfree(sbi);
 883 }
 884
 885 static struct kmem_cache *ext4_inode_cachep;
 886
 887 /*
 888  * Called inside transaction, so use GFP_NOFS
 889  */
 890 static struct inode *ext4_alloc_inode(struct super_block *sb)
 891 {
 892         struct ext4_inode_info *ei;
 893
 894         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 895         if (!ei)
 896                 return NULL;
 897
 898         ei->vfs_inode.i_version = 1;
 899         ei->vfs_inode.i_data.writeback_index = 0;
 900         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 901         INIT_LIST_HEAD(&ei->i_prealloc_list);
 902         spin_lock_init(&ei->i_prealloc_lock);
 903         ei->i_reserved_data_blocks = 0;
 904         ei->i_reserved_meta_blocks = 0;
 905         ei->i_allocated_meta_blocks = 0;
 906         ei->i_da_metadata_calc_len = 0;
 907         spin_lock_init(&(ei->i_block_reservation_lock));
 908 #ifdef CONFIG_QUOTA
 909         ei->i_reserved_quota = 0;
 910 #endif
 911         ei->jinode = NULL;
 912         INIT_LIST_HEAD(&ei->i_completed_io_list);
 913         spin_lock_init(&ei->i_completed_io_lock);
 914         ei->cur_aio_dio = NULL;
 915         ei->i_sync_tid = 0;
 916         ei->i_datasync_tid = 0;
 917         atomic_set(&ei->i_ioend_count, 0);
 918         atomic_set(&ei->i_aiodio_unwritten, 0);
 919
 920         return &ei->vfs_inode;
 921 }
 922
 923 static int ext4_drop_inode(struct inode *inode)
 924 {
 925         int drop = generic_drop_inode(inode);
 926
 927         trace_ext4_drop_inode(inode, drop);
 928         return drop;
 929 }
 930
 931 static void ext4_i_callback(struct rcu_head *head)
 932 {
 933         struct inode *inode = container_of(head, struct inode, i_rcu);
 934         INIT_LIST_HEAD(&inode->i_dentry);
 935         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 936 }
 937
 938 static void ext4_destroy_inode(struct inode *inode)
 939 {
 940         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 941                 ext4_msg(inode->i_sb, KERN_ERR,
 942                          "Inode %lu (%p): orphan list check failed!",
 943                          inode->i_ino, EXT4_I(inode));
 944                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 945                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 946                                 true);
 947                 dump_stack();
 948         }
 949         call_rcu(&inode->i_rcu, ext4_i_callback);
 950 }
 951
 952 static void init_once(void *foo)
 953 {
 954         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 955
 956         INIT_LIST_HEAD(&ei->i_orphan);
 957 #ifdef CONFIG_EXT4_FS_XATTR
 958         init_rwsem(&ei->xattr_sem);
 959 #endif
 960         init_rwsem(&ei->i_data_sem);
 961         inode_init_once(&ei->vfs_inode);
 962 }
 963
 964 static int init_inodecache(void)
 965 {
 966         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 967                                              sizeof(struct ext4_inode_info),
 968                                              0, (SLAB_RECLAIM_ACCOUNT|
 969                                                 SLAB_MEM_SPREAD),
 970                                              init_once);
 971         if (ext4_inode_cachep == NULL)
 972                 return -ENOMEM;
 973         return 0;
 974 }
 975
 976 static void destroy_inodecache(void)
 977 {
 978         kmem_cache_destroy(ext4_inode_cachep);
 979 }
 980
 981 void ext4_clear_inode(struct inode *inode)
 982 {
 983         invalidate_inode_buffers(inode);
 984         end_writeback(inode);
 985         dquot_drop(inode);
 986         ext4_discard_preallocations(inode);
 987         if (EXT4_I(inode)->jinode) {
 988                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 989                                                EXT4_I(inode)->jinode);
 990                 jbd2_free_inode(EXT4_I(inode)->jinode);
 991                 EXT4_I(inode)->jinode = NULL;
 992         }
 993 }
 994
 995 static inline void ext4_show_quota_options(struct seq_file *seq,
 996                                            struct super_block *sb)
 997 {
 998 #if defined(CONFIG_QUOTA)
 999         struct ext4_sb_info *sbi = EXT4_SB(sb);
1000
1001         if (sbi->s_jquota_fmt) {
1002                 char *fmtname = "";
1003
1004                 switch (sbi->s_jquota_fmt) {
1005                 case QFMT_VFS_OLD:
1006                         fmtname = "vfsold";
1007                         break;
1008                 case QFMT_VFS_V0:
1009                         fmtname = "vfsv0";
1010                         break;
1011                 case QFMT_VFS_V1:
1012                         fmtname = "vfsv1";
1013                         break;
1014                 }
1015                 seq_printf(seq, ",jqfmt=%s", fmtname);
1016         }
1017
1018         if (sbi->s_qf_names[USRQUOTA])
1019                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1020
1021         if (sbi->s_qf_names[GRPQUOTA])
1022                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1023
1024         if (test_opt(sb, USRQUOTA))
1025                 seq_puts(seq, ",usrquota");
1026
1027         if (test_opt(sb, GRPQUOTA))
1028                 seq_puts(seq, ",grpquota");
1029 #endif
1030 }
1031
1032 /*
1033  * Show an option if
1034  *  - it's set to a non-default value OR
1035  *  - if the per-sb default is different from the global default
1036  */
1037 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1038 {
1039         int def_errors;
1040         unsigned long def_mount_opts;
1041         struct super_block *sb = vfs->mnt_sb;
1042         struct ext4_sb_info *sbi = EXT4_SB(sb);
1043         struct ext4_super_block *es = sbi->s_es;
1044
1045         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1046         def_errors     = le16_to_cpu(es->s_errors);
1047
1048         if (sbi->s_sb_block != 1)
1049                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1050         if (test_opt(sb, MINIX_DF))
1051                 seq_puts(seq, ",minixdf");
1052         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1053                 seq_puts(seq, ",grpid");
1054         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1055                 seq_puts(seq, ",nogrpid");
1056         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1057             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1058                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1059         }
1060         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1061             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1062                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1063         }
1064         if (test_opt(sb, ERRORS_RO)) {
1065                 if (def_errors == EXT4_ERRORS_PANIC ||
1066                     def_errors == EXT4_ERRORS_CONTINUE) {
1067                         seq_puts(seq, ",errors=remount-ro");
1068                 }
1069         }
1070         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1071                 seq_puts(seq, ",errors=continue");
1072         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1073                 seq_puts(seq, ",errors=panic");
1074         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1075                 seq_puts(seq, ",nouid32");
1076         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1077                 seq_puts(seq, ",debug");
1078 #ifdef CONFIG_EXT4_FS_XATTR
1079         if (test_opt(sb, XATTR_USER))
1080                 seq_puts(seq, ",user_xattr");
1081         if (!test_opt(sb, XATTR_USER))
1082                 seq_puts(seq, ",nouser_xattr");
1083 #endif
1084 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1085         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1086                 seq_puts(seq, ",acl");
1087         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1088                 seq_puts(seq, ",noacl");
1089 #endif
1090         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1091                 seq_printf(seq, ",commit=%u",
1092                            (unsigned) (sbi->s_commit_interval / HZ));
1093         }
1094         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1095                 seq_printf(seq, ",min_batch_time=%u",
1096                            (unsigned) sbi->s_min_batch_time);
1097         }
1098         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1099                 seq_printf(seq, ",max_batch_time=%u",
1100                            (unsigned) sbi->s_min_batch_time);
1101         }
1102
1103         /*
1104          * We're changing the default of barrier mount option, so
1105          * let's always display its mount state so it's clear what its
1106          * status is.
1107          */
1108         seq_puts(seq, ",barrier=");
1109         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1110         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1111                 seq_puts(seq, ",journal_async_commit");
1112         else if (test_opt(sb, JOURNAL_CHECKSUM))
1113                 seq_puts(seq, ",journal_checksum");
1114         if (test_opt(sb, I_VERSION))
1115                 seq_puts(seq, ",i_version");
1116         if (!test_opt(sb, DELALLOC) &&
1117             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1118                 seq_puts(seq, ",nodelalloc");
1119
1120         if (!test_opt(sb, MBLK_IO_SUBMIT))
1121                 seq_puts(seq, ",nomblk_io_submit");
1122         if (sbi->s_stripe)
1123                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1124         /*
1125          * journal mode get enabled in different ways
1126          * So just print the value even if we didn't specify it
1127          */
1128         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1129                 seq_puts(seq, ",data=journal");
1130         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1131                 seq_puts(seq, ",data=ordered");
1132         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1133                 seq_puts(seq, ",data=writeback");
1134
1135         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1136                 seq_printf(seq, ",inode_readahead_blks=%u",
1137                            sbi->s_inode_readahead_blks);
1138
1139         if (test_opt(sb, DATA_ERR_ABORT))
1140                 seq_puts(seq, ",data_err=abort");
1141
1142         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1143                 seq_puts(seq, ",noauto_da_alloc");
1144
1145         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1146                 seq_puts(seq, ",discard");
1147
1148         if (test_opt(sb, NOLOAD))
1149                 seq_puts(seq, ",norecovery");
1150
1151         if (test_opt(sb, DIOREAD_NOLOCK))
1152                 seq_puts(seq, ",dioread_nolock");
1153
1154         if (test_opt(sb, BLOCK_VALIDITY) &&
1155             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1156                 seq_puts(seq, ",block_validity");
1157
1158         if (!test_opt(sb, INIT_INODE_TABLE))
1159                 seq_puts(seq, ",noinit_itable");
1160         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1161                 seq_printf(seq, ",init_itable=%u",
1162                            (unsigned) sbi->s_li_wait_mult);
1163
1164         ext4_show_quota_options(seq, sb);
1165
1166         return 0;
1167 }
1168
1169 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1170                                         u64 ino, u32 generation)
1171 {
1172         struct inode *inode;
1173
1174         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1175                 return ERR_PTR(-ESTALE);
1176         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1177                 return ERR_PTR(-ESTALE);
1178
1179         /* iget isn't really right if the inode is currently unallocated!!
1180          *
1181          * ext4_read_inode will return a bad_inode if the inode had been
1182          * deleted, so we should be safe.
1183          *
1184          * Currently we don't know the generation for parent directory, so
1185          * a generation of 0 means "accept any"
1186          */
1187         inode = ext4_iget(sb, ino);
1188         if (IS_ERR(inode))
1189                 return ERR_CAST(inode);
1190         if (generation && inode->i_generation != generation) {
1191                 iput(inode);
1192                 return ERR_PTR(-ESTALE);
1193         }
1194
1195         return inode;
1196 }
1197
1198 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1199                                         int fh_len, int fh_type)
1200 {
1201         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1202                                     ext4_nfs_get_inode);
1203 }
1204
1205 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1206                                         int fh_len, int fh_type)
1207 {
1208         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1209                                     ext4_nfs_get_inode);
1210 }
1211
1212 /*
1213  * Try to release metadata pages (indirect blocks, directories) which are
1214  * mapped via the block device.  Since these pages could have journal heads
1215  * which would prevent try_to_free_buffers() from freeing them, we must use
1216  * jbd2 layer's try_to_free_buffers() function to release them.
1217  */
1218 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1219                                  gfp_t wait)
1220 {
1221         journal_t *journal = EXT4_SB(sb)->s_journal;
1222
1223         WARN_ON(PageChecked(page));
1224         if (!page_has_buffers(page))
1225                 return 0;
1226         if (journal)
1227                 return jbd2_journal_try_to_free_buffers(journal, page,
1228                                                         wait & ~__GFP_WAIT);
1229         return try_to_free_buffers(page);
1230 }
1231
1232 #ifdef CONFIG_QUOTA
1233 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1234 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1235
1236 static int ext4_write_dquot(struct dquot *dquot);
1237 static int ext4_acquire_dquot(struct dquot *dquot);
1238 static int ext4_release_dquot(struct dquot *dquot);
1239 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1240 static int ext4_write_info(struct super_block *sb, int type);
1241 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1242                          struct path *path);
1243 static int ext4_quota_off(struct super_block *sb, int type);
1244 static int ext4_quota_on_mount(struct super_block *sb, int type);
1245 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1246                                size_t len, loff_t off);
1247 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1248                                 const char *data, size_t len, loff_t off);
1249
1250 static const struct dquot_operations ext4_quota_operations = {
1251         .get_reserved_space = ext4_get_reserved_space,
1252         .write_dquot    = ext4_write_dquot,
1253         .acquire_dquot  = ext4_acquire_dquot,
1254         .release_dquot  = ext4_release_dquot,
1255         .mark_dirty     = ext4_mark_dquot_dirty,
1256         .write_info     = ext4_write_info,
1257         .alloc_dquot    = dquot_alloc,
1258         .destroy_dquot  = dquot_destroy,
1259 };
1260
1261 static const struct quotactl_ops ext4_qctl_operations = {
1262         .quota_on       = ext4_quota_on,
1263         .quota_off      = ext4_quota_off,
1264         .quota_sync     = dquot_quota_sync,
1265         .get_info       = dquot_get_dqinfo,
1266         .set_info       = dquot_set_dqinfo,
1267         .get_dqblk      = dquot_get_dqblk,
1268         .set_dqblk      = dquot_set_dqblk
1269 };
1270 #endif
1271
1272 static const struct super_operations ext4_sops = {
1273         .alloc_inode    = ext4_alloc_inode,
1274         .destroy_inode  = ext4_destroy_inode,
1275         .write_inode    = ext4_write_inode,
1276         .dirty_inode    = ext4_dirty_inode,
1277         .drop_inode     = ext4_drop_inode,
1278         .evict_inode    = ext4_evict_inode,
1279         .put_super      = ext4_put_super,
1280         .sync_fs        = ext4_sync_fs,
1281         .freeze_fs      = ext4_freeze,
1282         .unfreeze_fs    = ext4_unfreeze,
1283         .statfs         = ext4_statfs,
1284         .remount_fs     = ext4_remount,
1285         .show_options   = ext4_show_options,
1286 #ifdef CONFIG_QUOTA
1287         .quota_read     = ext4_quota_read,
1288         .quota_write    = ext4_quota_write,
1289 #endif
1290         .bdev_try_to_free_page = bdev_try_to_free_page,
1291 };
1292
1293 static const struct super_operations ext4_nojournal_sops = {
1294         .alloc_inode    = ext4_alloc_inode,
1295         .destroy_inode  = ext4_destroy_inode,
1296         .write_inode    = ext4_write_inode,
1297         .dirty_inode    = ext4_dirty_inode,
1298         .drop_inode     = ext4_drop_inode,
1299         .evict_inode    = ext4_evict_inode,
1300         .write_super    = ext4_write_super,
1301         .put_super      = ext4_put_super,
1302         .statfs         = ext4_statfs,
1303         .remount_fs     = ext4_remount,
1304         .show_options   = ext4_show_options,
1305 #ifdef CONFIG_QUOTA
1306         .quota_read     = ext4_quota_read,
1307         .quota_write    = ext4_quota_write,
1308 #endif
1309         .bdev_try_to_free_page = bdev_try_to_free_page,
1310 };
1311
1312 static const struct export_operations ext4_export_ops = {
1313         .fh_to_dentry = ext4_fh_to_dentry,
1314         .fh_to_parent = ext4_fh_to_parent,
1315         .get_parent = ext4_get_parent,
1316 };
1317
1318 enum {
1319         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1320         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1321         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1322         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1323         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1324         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1325         Opt_journal_update, Opt_journal_dev,
1326         Opt_journal_checksum, Opt_journal_async_commit,
1327         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1328         Opt_data_err_abort, Opt_data_err_ignore,
1329         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1330         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1331         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1332         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1333         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1334         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1335         Opt_inode_readahead_blks, Opt_journal_ioprio,
1336         Opt_dioread_nolock, Opt_dioread_lock,
1337         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1338 };
1339
1340 static const match_table_t tokens = {
1341         {Opt_bsd_df, "bsddf"},
1342         {Opt_minix_df, "minixdf"},
1343         {Opt_grpid, "grpid"},
1344         {Opt_grpid, "bsdgroups"},
1345         {Opt_nogrpid, "nogrpid"},
1346         {Opt_nogrpid, "sysvgroups"},
1347         {Opt_resgid, "resgid=%u"},
1348         {Opt_resuid, "resuid=%u"},
1349         {Opt_sb, "sb=%u"},
1350         {Opt_err_cont, "errors=continue"},
1351         {Opt_err_panic, "errors=panic"},
1352         {Opt_err_ro, "errors=remount-ro"},
1353         {Opt_nouid32, "nouid32"},
1354         {Opt_debug, "debug"},
1355         {Opt_oldalloc, "oldalloc"},
1356         {Opt_orlov, "orlov"},
1357         {Opt_user_xattr, "user_xattr"},
1358         {Opt_nouser_xattr, "nouser_xattr"},
1359         {Opt_acl, "acl"},
1360         {Opt_noacl, "noacl"},
1361         {Opt_noload, "noload"},
1362         {Opt_noload, "norecovery"},
1363         {Opt_nobh, "nobh"},
1364         {Opt_bh, "bh"},
1365         {Opt_commit, "commit=%u"},
1366         {Opt_min_batch_time, "min_batch_time=%u"},
1367         {Opt_max_batch_time, "max_batch_time=%u"},
1368         {Opt_journal_update, "journal=update"},
1369         {Opt_journal_dev, "journal_dev=%u"},
1370         {Opt_journal_checksum, "journal_checksum"},
1371         {Opt_journal_async_commit, "journal_async_commit"},
1372         {Opt_abort, "abort"},
1373         {Opt_data_journal, "data=journal"},
1374         {Opt_data_ordered, "data=ordered"},
1375         {Opt_data_writeback, "data=writeback"},
1376         {Opt_data_err_abort, "data_err=abort"},
1377         {Opt_data_err_ignore, "data_err=ignore"},
1378         {Opt_offusrjquota, "usrjquota="},
1379         {Opt_usrjquota, "usrjquota=%s"},
1380         {Opt_offgrpjquota, "grpjquota="},
1381         {Opt_grpjquota, "grpjquota=%s"},
1382         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1383         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1384         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1385         {Opt_grpquota, "grpquota"},
1386         {Opt_noquota, "noquota"},
1387         {Opt_quota, "quota"},
1388         {Opt_usrquota, "usrquota"},
1389         {Opt_barrier, "barrier=%u"},
1390         {Opt_barrier, "barrier"},
1391         {Opt_nobarrier, "nobarrier"},
1392         {Opt_i_version, "i_version"},
1393         {Opt_stripe, "stripe=%u"},
1394         {Opt_resize, "resize"},
1395         {Opt_delalloc, "delalloc"},
1396         {Opt_nodelalloc, "nodelalloc"},
1397         {Opt_mblk_io_submit, "mblk_io_submit"},
1398         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1399         {Opt_block_validity, "block_validity"},
1400         {Opt_noblock_validity, "noblock_validity"},
1401         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1402         {Opt_journal_ioprio, "journal_ioprio=%u"},
1403         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1404         {Opt_auto_da_alloc, "auto_da_alloc"},
1405         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1406         {Opt_dioread_nolock, "dioread_nolock"},
1407         {Opt_dioread_lock, "dioread_lock"},
1408         {Opt_discard, "discard"},
1409         {Opt_nodiscard, "nodiscard"},
1410         {Opt_init_itable, "init_itable=%u"},
1411         {Opt_init_itable, "init_itable"},
1412         {Opt_noinit_itable, "noinit_itable"},
1413         {Opt_err, NULL},
1414 };
1415
1416 static ext4_fsblk_t get_sb_block(void **data)
1417 {
1418         ext4_fsblk_t    sb_block;
1419         char            *options = (char *) *data;
1420
1421         if (!options || strncmp(options, "sb=", 3) != 0)
1422                 return 1;       /* Default location */
1423
1424         options += 3;
1425         /* TODO: use simple_strtoll with >32bit ext4 */
1426         sb_block = simple_strtoul(options, &options, 0);
1427         if (*options && *options != ',') {
1428                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1429                        (char *) *data);
1430                 return 1;
1431         }
1432         if (*options == ',')
1433                 options++;
1434         *data = (void *) options;
1435
1436         return sb_block;
1437 }
1438
1439 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1440 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1441         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1442
1443 #ifdef CONFIG_QUOTA
1444 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1445 {
1446         struct ext4_sb_info *sbi = EXT4_SB(sb);
1447         char *qname;
1448
1449         if (sb_any_quota_loaded(sb) &&
1450                 !sbi->s_qf_names[qtype]) {
1451                 ext4_msg(sb, KERN_ERR,
1452                         "Cannot change journaled "
1453                         "quota options when quota turned on");
1454                 return 0;
1455         }
1456         qname = match_strdup(args);
1457         if (!qname) {
1458                 ext4_msg(sb, KERN_ERR,
1459                         "Not enough memory for storing quotafile name");
1460                 return 0;
1461         }
1462         if (sbi->s_qf_names[qtype] &&
1463                 strcmp(sbi->s_qf_names[qtype], qname)) {
1464                 ext4_msg(sb, KERN_ERR,
1465                         "%s quota file already specified", QTYPE2NAME(qtype));
1466                 kfree(qname);
1467                 return 0;
1468         }
1469         sbi->s_qf_names[qtype] = qname;
1470         if (strchr(sbi->s_qf_names[qtype], '/')) {
1471                 ext4_msg(sb, KERN_ERR,
1472                         "quotafile must be on filesystem root");
1473                 kfree(sbi->s_qf_names[qtype]);
1474                 sbi->s_qf_names[qtype] = NULL;
1475                 return 0;
1476         }
1477         set_opt(sb, QUOTA);
1478         return 1;
1479 }
1480
1481 static int clear_qf_name(struct super_block *sb, int qtype)
1482 {
1483
1484         struct ext4_sb_info *sbi = EXT4_SB(sb);
1485
1486         if (sb_any_quota_loaded(sb) &&
1487                 sbi->s_qf_names[qtype]) {
1488                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1489                         " when quota turned on");
1490                 return 0;
1491         }
1492         /*
1493          * The space will be released later when all options are confirmed
1494          * to be correct
1495          */
1496         sbi->s_qf_names[qtype] = NULL;
1497         return 1;
1498 }
1499 #endif
1500
1501 static int parse_options(char *options, struct super_block *sb,
1502                          unsigned long *journal_devnum,
1503                          unsigned int *journal_ioprio,
1504                          ext4_fsblk_t *n_blocks_count, int is_remount)
1505 {
1506         struct ext4_sb_info *sbi = EXT4_SB(sb);
1507         char *p;
1508         substring_t args[MAX_OPT_ARGS];
1509         int data_opt = 0;
1510         int option;
1511 #ifdef CONFIG_QUOTA
1512         int qfmt;
1513 #endif
1514
1515         if (!options)
1516                 return 1;
1517
1518         while ((p = strsep(&options, ",")) != NULL) {
1519                 int token;
1520                 if (!*p)
1521                         continue;
1522
1523                 /*
1524                  * Initialize args struct so we know whether arg was
1525                  * found; some options take optional arguments.
1526                  */
1527                 args[0].to = args[0].from = NULL;
1528                 token = match_token(p, tokens, args);
1529                 switch (token) {
1530                 case Opt_bsd_df:
1531                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1532                         clear_opt(sb, MINIX_DF);
1533                         break;
1534                 case Opt_minix_df:
1535                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1536                         set_opt(sb, MINIX_DF);
1537
1538                         break;
1539                 case Opt_grpid:
1540                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1541                         set_opt(sb, GRPID);
1542
1543                         break;
1544                 case Opt_nogrpid:
1545                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1546                         clear_opt(sb, GRPID);
1547
1548                         break;
1549                 case Opt_resuid:
1550                         if (match_int(&args[0], &option))
1551                                 return 0;
1552                         sbi->s_resuid = option;
1553                         break;
1554                 case Opt_resgid:
1555                         if (match_int(&args[0], &option))
1556                                 return 0;
1557                         sbi->s_resgid = option;
1558                         break;
1559                 case Opt_sb:
1560                         /* handled by get_sb_block() instead of here */
1561                         /* *sb_block = match_int(&args[0]); */
1562                         break;
1563                 case Opt_err_panic:
1564                         clear_opt(sb, ERRORS_CONT);
1565                         clear_opt(sb, ERRORS_RO);
1566                         set_opt(sb, ERRORS_PANIC);
1567                         break;
1568                 case Opt_err_ro:
1569                         clear_opt(sb, ERRORS_CONT);
1570                         clear_opt(sb, ERRORS_PANIC);
1571                         set_opt(sb, ERRORS_RO);
1572                         break;
1573                 case Opt_err_cont:
1574                         clear_opt(sb, ERRORS_RO);
1575                         clear_opt(sb, ERRORS_PANIC);
1576                         set_opt(sb, ERRORS_CONT);
1577                         break;
1578                 case Opt_nouid32:
1579                         set_opt(sb, NO_UID32);
1580                         break;
1581                 case Opt_debug:
1582                         set_opt(sb, DEBUG);
1583                         break;
1584                 case Opt_oldalloc:
1585                         ext4_msg(sb, KERN_WARNING,
1586                                  "Ignoring deprecated oldalloc option");
1587                         break;
1588                 case Opt_orlov:
1589                         ext4_msg(sb, KERN_WARNING,
1590                                  "Ignoring deprecated orlov option");
1591                         break;
1592 #ifdef CONFIG_EXT4_FS_XATTR
1593                 case Opt_user_xattr:
1594                         set_opt(sb, XATTR_USER);
1595                         break;
1596                 case Opt_nouser_xattr:
1597                         clear_opt(sb, XATTR_USER);
1598                         break;
1599 #else
1600                 case Opt_user_xattr:
1601                 case Opt_nouser_xattr:
1602                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1603                         break;
1604 #endif
1605 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1606                 case Opt_acl:
1607                         set_opt(sb, POSIX_ACL);
1608                         break;
1609                 case Opt_noacl:
1610                         clear_opt(sb, POSIX_ACL);
1611                         break;
1612 #else
1613                 case Opt_acl:
1614                 case Opt_noacl:
1615                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1616                         break;
1617 #endif
1618                 case Opt_journal_update:
1619                         /* @@@ FIXME */
1620                         /* Eventually we will want to be able to create
1621                            a journal file here.  For now, only allow the
1622                            user to specify an existing inode to be the
1623                            journal file. */
1624                         if (is_remount) {
1625                                 ext4_msg(sb, KERN_ERR,
1626                                          "Cannot specify journal on remount");
1627                                 return 0;
1628                         }
1629                         set_opt(sb, UPDATE_JOURNAL);
1630                         break;
1631                 case Opt_journal_dev:
1632                         if (is_remount) {
1633                                 ext4_msg(sb, KERN_ERR,
1634                                         "Cannot specify journal on remount");
1635                                 return 0;
1636                         }
1637                         if (match_int(&args[0], &option))
1638                                 return 0;
1639                         *journal_devnum = option;
1640                         break;
1641                 case Opt_journal_checksum:
1642                         set_opt(sb, JOURNAL_CHECKSUM);
1643                         break;
1644                 case Opt_journal_async_commit:
1645                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1646                         set_opt(sb, JOURNAL_CHECKSUM);
1647                         break;
1648                 case Opt_noload:
1649                         set_opt(sb, NOLOAD);
1650                         break;
1651                 case Opt_commit:
1652                         if (match_int(&args[0], &option))
1653                                 return 0;
1654                         if (option < 0)
1655                                 return 0;
1656                         if (option == 0)
1657                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1658                         sbi->s_commit_interval = HZ * option;
1659                         break;
1660                 case Opt_max_batch_time:
1661                         if (match_int(&args[0], &option))
1662                                 return 0;
1663                         if (option < 0)
1664                                 return 0;
1665                         if (option == 0)
1666                                 option = EXT4_DEF_MAX_BATCH_TIME;
1667                         sbi->s_max_batch_time = option;
1668                         break;
1669                 case Opt_min_batch_time:
1670                         if (match_int(&args[0], &option))
1671                                 return 0;
1672                         if (option < 0)
1673                                 return 0;
1674                         sbi->s_min_batch_time = option;
1675                         break;
1676                 case Opt_data_journal:
1677                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1678                         goto datacheck;
1679                 case Opt_data_ordered:
1680                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1681                         goto datacheck;
1682                 case Opt_data_writeback:
1683                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1684                 datacheck:
1685                         if (is_remount) {
1686                                 if (!sbi->s_journal)
1687                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1688                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1689                                         ext4_msg(sb, KERN_ERR,
1690                                                 "Cannot change data mode on remount");
1691                                         return 0;
1692                                 }
1693                         } else {
1694                                 clear_opt(sb, DATA_FLAGS);
1695                                 sbi->s_mount_opt |= data_opt;
1696                         }
1697                         break;
1698                 case Opt_data_err_abort:
1699                         set_opt(sb, DATA_ERR_ABORT);
1700                         break;
1701                 case Opt_data_err_ignore:
1702                         clear_opt(sb, DATA_ERR_ABORT);
1703                         break;
1704 #ifdef CONFIG_QUOTA
1705                 case Opt_usrjquota:
1706                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1707                                 return 0;
1708                         break;
1709                 case Opt_grpjquota:
1710                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1711                                 return 0;
1712                         break;
1713                 case Opt_offusrjquota:
1714                         if (!clear_qf_name(sb, USRQUOTA))
1715                                 return 0;
1716                         break;
1717                 case Opt_offgrpjquota:
1718                         if (!clear_qf_name(sb, GRPQUOTA))
1719                                 return 0;
1720                         break;
1721
1722                 case Opt_jqfmt_vfsold:
1723                         qfmt = QFMT_VFS_OLD;
1724                         goto set_qf_format;
1725                 case Opt_jqfmt_vfsv0:
1726                         qfmt = QFMT_VFS_V0;
1727                         goto set_qf_format;
1728                 case Opt_jqfmt_vfsv1:
1729                         qfmt = QFMT_VFS_V1;
1730 set_qf_format:
1731                         if (sb_any_quota_loaded(sb) &&
1732                             sbi->s_jquota_fmt != qfmt) {
1733                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1734                                         "journaled quota options when "
1735                                         "quota turned on");
1736                                 return 0;
1737                         }
1738                         sbi->s_jquota_fmt = qfmt;
1739                         break;
1740                 case Opt_quota:
1741                 case Opt_usrquota:
1742                         set_opt(sb, QUOTA);
1743                         set_opt(sb, USRQUOTA);
1744                         break;
1745                 case Opt_grpquota:
1746                         set_opt(sb, QUOTA);
1747                         set_opt(sb, GRPQUOTA);
1748                         break;
1749                 case Opt_noquota:
1750                         if (sb_any_quota_loaded(sb)) {
1751                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1752                                         "options when quota turned on");
1753                                 return 0;
1754                         }
1755                         clear_opt(sb, QUOTA);
1756                         clear_opt(sb, USRQUOTA);
1757                         clear_opt(sb, GRPQUOTA);
1758                         break;
1759 #else
1760                 case Opt_quota:
1761                 case Opt_usrquota:
1762                 case Opt_grpquota:
1763                         ext4_msg(sb, KERN_ERR,
1764                                 "quota options not supported");
1765                         break;
1766                 case Opt_usrjquota:
1767                 case Opt_grpjquota:
1768                 case Opt_offusrjquota:
1769                 case Opt_offgrpjquota:
1770                 case Opt_jqfmt_vfsold:
1771                 case Opt_jqfmt_vfsv0:
1772                 case Opt_jqfmt_vfsv1:
1773                         ext4_msg(sb, KERN_ERR,
1774                                 "journaled quota options not supported");
1775                         break;
1776                 case Opt_noquota:
1777                         break;
1778 #endif
1779                 case Opt_abort:
1780                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1781                         break;
1782                 case Opt_nobarrier:
1783                         clear_opt(sb, BARRIER);
1784                         break;
1785                 case Opt_barrier:
1786                         if (args[0].from) {
1787                                 if (match_int(&args[0], &option))
1788                                         return 0;
1789                         } else
1790                                 option = 1;     /* No argument, default to 1 */
1791                         if (option)
1792                                 set_opt(sb, BARRIER);
1793                         else
1794                                 clear_opt(sb, BARRIER);
1795                         break;
1796                 case Opt_ignore:
1797                         break;
1798                 case Opt_resize:
1799                         if (!is_remount) {
1800                                 ext4_msg(sb, KERN_ERR,
1801                                         "resize option only available "
1802                                         "for remount");
1803                                 return 0;
1804                         }
1805                         if (match_int(&args[0], &option) != 0)
1806                                 return 0;
1807                         *n_blocks_count = option;
1808                         break;
1809                 case Opt_nobh:
1810                         ext4_msg(sb, KERN_WARNING,
1811                                  "Ignoring deprecated nobh option");
1812                         break;
1813                 case Opt_bh:
1814                         ext4_msg(sb, KERN_WARNING,
1815                                  "Ignoring deprecated bh option");
1816                         break;
1817                 case Opt_i_version:
1818                         set_opt(sb, I_VERSION);
1819                         sb->s_flags |= MS_I_VERSION;
1820                         break;
1821                 case Opt_nodelalloc:
1822                         clear_opt(sb, DELALLOC);
1823                         clear_opt2(sb, EXPLICIT_DELALLOC);
1824                         break;
1825                 case Opt_mblk_io_submit:
1826                         set_opt(sb, MBLK_IO_SUBMIT);
1827                         break;
1828                 case Opt_nomblk_io_submit:
1829                         clear_opt(sb, MBLK_IO_SUBMIT);
1830                         break;
1831                 case Opt_stripe:
1832                         if (match_int(&args[0], &option))
1833                                 return 0;
1834                         if (option < 0)
1835                                 return 0;
1836                         sbi->s_stripe = option;
1837                         break;
1838                 case Opt_delalloc:
1839                         set_opt(sb, DELALLOC);
1840                         set_opt2(sb, EXPLICIT_DELALLOC);
1841                         break;
1842                 case Opt_block_validity:
1843                         set_opt(sb, BLOCK_VALIDITY);
1844                         break;
1845                 case Opt_noblock_validity:
1846                         clear_opt(sb, BLOCK_VALIDITY);
1847                         break;
1848                 case Opt_inode_readahead_blks:
1849                         if (match_int(&args[0], &option))
1850                                 return 0;
1851                         if (option < 0 || option > (1 << 30))
1852                                 return 0;
1853                         if (option && !is_power_of_2(option)) {
1854                                 ext4_msg(sb, KERN_ERR,
1855                                          "EXT4-fs: inode_readahead_blks"
1856                                          " must be a power of 2");
1857                                 return 0;
1858                         }
1859                         sbi->s_inode_readahead_blks = option;
1860                         break;
1861                 case Opt_journal_ioprio:
1862                         if (match_int(&args[0], &option))
1863                                 return 0;
1864                         if (option < 0 || option > 7)
1865                                 break;
1866                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1867                                                             option);
1868                         break;
1869                 case Opt_noauto_da_alloc:
1870                         set_opt(sb, NO_AUTO_DA_ALLOC);
1871                         break;
1872                 case Opt_auto_da_alloc:
1873                         if (args[0].from) {
1874                                 if (match_int(&args[0], &option))
1875                                         return 0;
1876                         } else
1877                                 option = 1;     /* No argument, default to 1 */
1878                         if (option)
1879                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1880                         else
1881                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1882                         break;
1883                 case Opt_discard:
1884                         set_opt(sb, DISCARD);
1885                         break;
1886                 case Opt_nodiscard:
1887                         clear_opt(sb, DISCARD);
1888                         break;
1889                 case Opt_dioread_nolock:
1890                         set_opt(sb, DIOREAD_NOLOCK);
1891                         break;
1892                 case Opt_dioread_lock:
1893                         clear_opt(sb, DIOREAD_NOLOCK);
1894                         break;
1895                 case Opt_init_itable:
1896                         set_opt(sb, INIT_INODE_TABLE);
1897                         if (args[0].from) {
1898                                 if (match_int(&args[0], &option))
1899                                         return 0;
1900                         } else
1901                                 option = EXT4_DEF_LI_WAIT_MULT;
1902                         if (option < 0)
1903                                 return 0;
1904                         sbi->s_li_wait_mult = option;
1905                         break;
1906                 case Opt_noinit_itable:
1907                         clear_opt(sb, INIT_INODE_TABLE);
1908                         break;
1909                 default:
1910                         ext4_msg(sb, KERN_ERR,
1911                                "Unrecognized mount option \"%s\" "
1912                                "or missing value", p);
1913                         return 0;
1914                 }
1915         }
1916 #ifdef CONFIG_QUOTA
1917         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1918                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1919                         clear_opt(sb, USRQUOTA);
1920
1921                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1922                         clear_opt(sb, GRPQUOTA);
1923
1924                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1925                         ext4_msg(sb, KERN_ERR, "old and new quota "
1926                                         "format mixing");
1927                         return 0;
1928                 }
1929
1930                 if (!sbi->s_jquota_fmt) {
1931                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1932                                         "not specified");
1933                         return 0;
1934                 }
1935         } else {
1936                 if (sbi->s_jquota_fmt) {
1937                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1938                                         "specified with no journaling "
1939                                         "enabled");
1940                         return 0;
1941                 }
1942         }
1943 #endif
1944         return 1;
1945 }
1946
1947 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1948                             int read_only)
1949 {
1950         struct ext4_sb_info *sbi = EXT4_SB(sb);
1951         int res = 0;
1952
1953         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1954                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1955                          "forcing read-only mode");
1956                 res = MS_RDONLY;
1957         }
1958         if (read_only)
1959                 goto done;
1960         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1961                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1962                          "running e2fsck is recommended");
1963         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1964                 ext4_msg(sb, KERN_WARNING,
1965                          "warning: mounting fs with errors, "
1966                          "running e2fsck is recommended");
1967         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1968                  le16_to_cpu(es->s_mnt_count) >=
1969                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1970                 ext4_msg(sb, KERN_WARNING,
1971                          "warning: maximal mount count reached, "
1972                          "running e2fsck is recommended");
1973         else if (le32_to_cpu(es->s_checkinterval) &&
1974                 (le32_to_cpu(es->s_lastcheck) +
1975                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1976                 ext4_msg(sb, KERN_WARNING,
1977                          "warning: checktime reached, "
1978                          "running e2fsck is recommended");
1979         if (!sbi->s_journal)
1980                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1981         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1982                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1983         le16_add_cpu(&es->s_mnt_count, 1);
1984         es->s_mtime = cpu_to_le32(get_seconds());
1985         ext4_update_dynamic_rev(sb);
1986         if (sbi->s_journal)
1987                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1988
1989         ext4_commit_super(sb, 1);
1990 done:
1991         if (test_opt(sb, DEBUG))
1992                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1993                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1994                         sb->s_blocksize,
1995                         sbi->s_groups_count,
1996                         EXT4_BLOCKS_PER_GROUP(sb),
1997                         EXT4_INODES_PER_GROUP(sb),
1998                         sbi->s_mount_opt, sbi->s_mount_opt2);
1999
2000         cleancache_init_fs(sb);
2001         return res;
2002 }
2003
2004 static int ext4_fill_flex_info(struct super_block *sb)
2005 {
2006         struct ext4_sb_info *sbi = EXT4_SB(sb);
2007         struct ext4_group_desc *gdp = NULL;
2008         ext4_group_t flex_group_count;
2009         ext4_group_t flex_group;
2010         unsigned int groups_per_flex = 0;
2011         size_t size;
2012         int i;
2013
2014         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2015         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2016                 sbi->s_log_groups_per_flex = 0;
2017                 return 1;
2018         }
2019         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2020
2021         /* We allocate both existing and potentially added groups */
2022         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2023                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2024                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2025         size = flex_group_count * sizeof(struct flex_groups);
2026         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2027         if (sbi->s_flex_groups == NULL) {
2028                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2029                          flex_group_count);
2030                 goto failed;
2031         }
2032
2033         for (i = 0; i < sbi->s_groups_count; i++) {
2034                 gdp = ext4_get_group_desc(sb, i, NULL);
2035
2036                 flex_group = ext4_flex_group(sbi, i);
2037                 atomic_add(ext4_free_inodes_count(sb, gdp),
2038                            &sbi->s_flex_groups[flex_group].free_inodes);
2039                 atomic_add(ext4_free_group_clusters(sb, gdp),
2040                            &sbi->s_flex_groups[flex_group].free_clusters);
2041                 atomic_add(ext4_used_dirs_count(sb, gdp),
2042                            &sbi->s_flex_groups[flex_group].used_dirs);
2043         }
2044
2045         return 1;
2046 failed:
2047         return 0;
2048 }
2049
2050 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2051                             struct ext4_group_desc *gdp)
2052 {
2053         __u16 crc = 0;
2054
2055         if (sbi->s_es->s_feature_ro_compat &
2056             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2057                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2058                 __le32 le_group = cpu_to_le32(block_group);
2059
2060                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2061                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2062                 crc = crc16(crc, (__u8 *)gdp, offset);
2063                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2064                 /* for checksum of struct ext4_group_desc do the rest...*/
2065                 if ((sbi->s_es->s_feature_incompat &
2066                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2067                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2068                         crc = crc16(crc, (__u8 *)gdp + offset,
2069                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2070                                         offset);
2071         }
2072
2073         return cpu_to_le16(crc);
2074 }
2075
2076 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2077                                 struct ext4_group_desc *gdp)
2078 {
2079         if ((sbi->s_es->s_feature_ro_compat &
2080              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2081             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2082                 return 0;
2083
2084         return 1;
2085 }
2086
2087 /* Called at mount-time, super-block is locked */
2088 static int ext4_check_descriptors(struct super_block *sb,
2089                                   ext4_group_t *first_not_zeroed)
2090 {
2091         struct ext4_sb_info *sbi = EXT4_SB(sb);
2092         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2093         ext4_fsblk_t last_block;
2094         ext4_fsblk_t block_bitmap;
2095         ext4_fsblk_t inode_bitmap;
2096         ext4_fsblk_t inode_table;
2097         int flexbg_flag = 0;
2098         ext4_group_t i, grp = sbi->s_groups_count;
2099
2100         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2101                 flexbg_flag = 1;
2102
2103         ext4_debug("Checking group descriptors");
2104
2105         for (i = 0; i < sbi->s_groups_count; i++) {
2106                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2107
2108                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2109                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2110                 else
2111                         last_block = first_block +
2112                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2113
2114                 if ((grp == sbi->s_groups_count) &&
2115                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2116                         grp = i;
2117
2118                 block_bitmap = ext4_block_bitmap(sb, gdp);
2119                 if (block_bitmap < first_block || block_bitmap > last_block) {
2120                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2121                                "Block bitmap for group %u not in group "
2122                                "(block %llu)!", i, block_bitmap);
2123                         return 0;
2124                 }
2125                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2126                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2127                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2128                                "Inode bitmap for group %u not in group "
2129                                "(block %llu)!", i, inode_bitmap);
2130                         return 0;
2131                 }
2132                 inode_table = ext4_inode_table(sb, gdp);
2133                 if (inode_table < first_block ||
2134                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2135                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2136                                "Inode table for group %u not in group "
2137                                "(block %llu)!", i, inode_table);
2138                         return 0;
2139                 }
2140                 ext4_lock_group(sb, i);
2141                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2142                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2143                                  "Checksum for group %u failed (%u!=%u)",
2144                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2145                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2146                         if (!(sb->s_flags & MS_RDONLY)) {
2147                                 ext4_unlock_group(sb, i);
2148                                 return 0;
2149                         }
2150                 }
2151                 ext4_unlock_group(sb, i);
2152                 if (!flexbg_flag)
2153                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2154         }
2155         if (NULL != first_not_zeroed)
2156                 *first_not_zeroed = grp;
2157
2158         ext4_free_blocks_count_set(sbi->s_es,
2159                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2160         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2161         return 1;
2162 }
2163
2164 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2165  * the superblock) which were deleted from all directories, but held open by
2166  * a process at the time of a crash.  We walk the list and try to delete these
2167  * inodes at recovery time (only with a read-write filesystem).
2168  *
2169  * In order to keep the orphan inode chain consistent during traversal (in
2170  * case of crash during recovery), we link each inode into the superblock
2171  * orphan list_head and handle it the same way as an inode deletion during
2172  * normal operation (which journals the operations for us).
2173  *
2174  * We only do an iget() and an iput() on each inode, which is very safe if we
2175  * accidentally point at an in-use or already deleted inode.  The worst that
2176  * can happen in this case is that we get a "bit already cleared" message from
2177  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2178  * e2fsck was run on this filesystem, and it must have already done the orphan
2179  * inode cleanup for us, so we can safely abort without any further action.
2180  */
2181 static void ext4_orphan_cleanup(struct super_block *sb,
2182                                 struct ext4_super_block *es)
2183 {
2184         unsigned int s_flags = sb->s_flags;
2185         int nr_orphans = 0, nr_truncates = 0;
2186 #ifdef CONFIG_QUOTA
2187         int i;
2188 #endif
2189         if (!es->s_last_orphan) {
2190                 jbd_debug(4, "no orphan inodes to clean up\n");
2191                 return;
2192         }
2193
2194         if (bdev_read_only(sb->s_bdev)) {
2195                 ext4_msg(sb, KERN_ERR, "write access "
2196                         "unavailable, skipping orphan cleanup");
2197                 return;
2198         }
2199
2200         /* Check if feature set would not allow a r/w mount */
2201         if (!ext4_feature_set_ok(sb, 0)) {
2202                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2203                          "unknown ROCOMPAT features");
2204                 return;
2205         }
2206
2207         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2208                 if (es->s_last_orphan)
2209                         jbd_debug(1, "Errors on filesystem, "
2210                                   "clearing orphan list.\n");
2211                 es->s_last_orphan = 0;
2212                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2213                 return;
2214         }
2215
2216         if (s_flags & MS_RDONLY) {
2217                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2218                 sb->s_flags &= ~MS_RDONLY;
2219         }
2220 #ifdef CONFIG_QUOTA
2221         /* Needed for iput() to work correctly and not trash data */
2222         sb->s_flags |= MS_ACTIVE;
2223         /* Turn on quotas so that they are updated correctly */
2224         for (i = 0; i < MAXQUOTAS; i++) {
2225                 if (EXT4_SB(sb)->s_qf_names[i]) {
2226                         int ret = ext4_quota_on_mount(sb, i);
2227                         if (ret < 0)
2228                                 ext4_msg(sb, KERN_ERR,
2229                                         "Cannot turn on journaled "
2230                                         "quota: error %d", ret);
2231                 }
2232         }
2233 #endif
2234
2235         while (es->s_last_orphan) {
2236                 struct inode *inode;
2237
2238                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2239                 if (IS_ERR(inode)) {
2240                         es->s_last_orphan = 0;
2241                         break;
2242                 }
2243
2244                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2245                 dquot_initialize(inode);
2246                 if (inode->i_nlink) {
2247                         ext4_msg(sb, KERN_DEBUG,
2248                                 "%s: truncating inode %lu to %lld bytes",
2249                                 __func__, inode->i_ino, inode->i_size);
2250                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2251                                   inode->i_ino, inode->i_size);
2252                         ext4_truncate(inode);
2253                         nr_truncates++;
2254                 } else {
2255                         ext4_msg(sb, KERN_DEBUG,
2256                                 "%s: deleting unreferenced inode %lu",
2257                                 __func__, inode->i_ino);
2258                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2259                                   inode->i_ino);
2260                         nr_orphans++;
2261                 }
2262                 iput(inode);  /* The delete magic happens here! */
2263         }
2264
2265 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2266
2267         if (nr_orphans)
2268                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2269                        PLURAL(nr_orphans));
2270         if (nr_truncates)
2271                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2272                        PLURAL(nr_truncates));
2273 #ifdef CONFIG_QUOTA
2274         /* Turn quotas off */
2275         for (i = 0; i < MAXQUOTAS; i++) {
2276                 if (sb_dqopt(sb)->files[i])
2277                         dquot_quota_off(sb, i);
2278         }
2279 #endif
2280         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2281 }
2282
2283 /*
2284  * Maximal extent format file size.
2285  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2286  * extent format containers, within a sector_t, and within i_blocks
2287  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2288  * so that won't be a limiting factor.
2289  *
2290  * However there is other limiting factor. We do store extents in the form
2291  * of starting block and length, hence the resulting length of the extent
2292  * covering maximum file size must fit into on-disk format containers as
2293  * well. Given that length is always by 1 unit bigger than max unit (because
2294  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2295  *
2296  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2297  */
2298 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2299 {
2300         loff_t res;
2301         loff_t upper_limit = MAX_LFS_FILESIZE;
2302
2303         /* small i_blocks in vfs inode? */
2304         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2305                 /*
2306                  * CONFIG_LBDAF is not enabled implies the inode
2307                  * i_block represent total blocks in 512 bytes
2308                  * 32 == size of vfs inode i_blocks * 8
2309                  */
2310                 upper_limit = (1LL << 32) - 1;
2311
2312                 /* total blocks in file system block size */
2313                 upper_limit >>= (blkbits - 9);
2314                 upper_limit <<= blkbits;
2315         }
2316
2317         /*
2318          * 32-bit extent-start container, ee_block. We lower the maxbytes
2319          * by one fs block, so ee_len can cover the extent of maximum file
2320          * size
2321          */
2322         res = (1LL << 32) - 1;
2323         res <<= blkbits;
2324
2325         /* Sanity check against vm- & vfs- imposed limits */
2326         if (res > upper_limit)
2327                 res = upper_limit;
2328
2329         return res;
2330 }
2331
2332 /*
2333  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2334  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2335  * We need to be 1 filesystem block less than the 2^48 sector limit.
2336  */
2337 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2338 {
2339         loff_t res = EXT4_NDIR_BLOCKS;
2340         int meta_blocks;
2341         loff_t upper_limit;
2342         /* This is calculated to be the largest file size for a dense, block
2343          * mapped file such that the file's total number of 512-byte sectors,
2344          * including data and all indirect blocks, does not exceed (2^48 - 1).
2345          *
2346          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2347          * number of 512-byte sectors of the file.
2348          */
2349
2350         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2351                 /*
2352                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2353                  * the inode i_block field represents total file blocks in
2354                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2355                  */
2356                 upper_limit = (1LL << 32) - 1;
2357
2358                 /* total blocks in file system block size */
2359                 upper_limit >>= (bits - 9);
2360
2361         } else {
2362                 /*
2363                  * We use 48 bit ext4_inode i_blocks
2364                  * With EXT4_HUGE_FILE_FL set the i_blocks
2365                  * represent total number of blocks in
2366                  * file system block size
2367                  */
2368                 upper_limit = (1LL << 48) - 1;
2369
2370         }
2371
2372         /* indirect blocks */
2373         meta_blocks = 1;
2374         /* double indirect blocks */
2375         meta_blocks += 1 + (1LL << (bits-2));
2376         /* tripple indirect blocks */
2377         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2378
2379         upper_limit -= meta_blocks;
2380         upper_limit <<= bits;
2381
2382         res += 1LL << (bits-2);
2383         res += 1LL << (2*(bits-2));
2384         res += 1LL << (3*(bits-2));
2385         res <<= bits;
2386         if (res > upper_limit)
2387                 res = upper_limit;
2388
2389         if (res > MAX_LFS_FILESIZE)
2390                 res = MAX_LFS_FILESIZE;
2391
2392         return res;
2393 }
2394
2395 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2396                                    ext4_fsblk_t logical_sb_block, int nr)
2397 {
2398         struct ext4_sb_info *sbi = EXT4_SB(sb);
2399         ext4_group_t bg, first_meta_bg;
2400         int has_super = 0;
2401
2402         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2403
2404         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2405             nr < first_meta_bg)
2406                 return logical_sb_block + nr + 1;
2407         bg = sbi->s_desc_per_block * nr;
2408         if (ext4_bg_has_super(sb, bg))
2409                 has_super = 1;
2410
2411         return (has_super + ext4_group_first_block_no(sb, bg));
2412 }
2413
2414 /**
2415  * ext4_get_stripe_size: Get the stripe size.
2416  * @sbi: In memory super block info
2417  *
2418  * If we have specified it via mount option, then
2419  * use the mount option value. If the value specified at mount time is
2420  * greater than the blocks per group use the super block value.
2421  * If the super block value is greater than blocks per group return 0.
2422  * Allocator needs it be less than blocks per group.
2423  *
2424  */
2425 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2426 {
2427         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2428         unsigned long stripe_width =
2429                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2430         int ret;
2431
2432         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2433                 ret = sbi->s_stripe;
2434         else if (stripe_width <= sbi->s_blocks_per_group)
2435                 ret = stripe_width;
2436         else if (stride <= sbi->s_blocks_per_group)
2437                 ret = stride;
2438         else
2439                 ret = 0;
2440
2441         /*
2442          * If the stripe width is 1, this makes no sense and
2443          * we set it to 0 to turn off stripe handling code.
2444          */
2445         if (ret <= 1)
2446                 ret = 0;
2447
2448         return ret;
2449 }
2450
2451 /* sysfs supprt */
2452
2453 struct ext4_attr {
2454         struct attribute attr;
2455         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2456         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2457                          const char *, size_t);
2458         int offset;
2459 };
2460
2461 static int parse_strtoul(const char *buf,
2462                 unsigned long max, unsigned long *value)
2463 {
2464         char *endp;
2465
2466         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2467         endp = skip_spaces(endp);
2468         if (*endp || *value > max)
2469                 return -EINVAL;
2470
2471         return 0;
2472 }
2473
2474 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2475                                               struct ext4_sb_info *sbi,
2476                                               char *buf)
2477 {
2478         return snprintf(buf, PAGE_SIZE, "%llu\n",
2479                 (s64) EXT4_C2B(sbi,
2480                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2481 }
2482
2483 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2484                                          struct ext4_sb_info *sbi, char *buf)
2485 {
2486         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2487
2488         if (!sb->s_bdev->bd_part)
2489                 return snprintf(buf, PAGE_SIZE, "0\n");
2490         return snprintf(buf, PAGE_SIZE, "%lu\n",
2491                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2492                          sbi->s_sectors_written_start) >> 1);
2493 }
2494
2495 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2496                                           struct ext4_sb_info *sbi, char *buf)
2497 {
2498         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2499
2500         if (!sb->s_bdev->bd_part)
2501                 return snprintf(buf, PAGE_SIZE, "0\n");
2502         return snprintf(buf, PAGE_SIZE, "%llu\n",
2503                         (unsigned long long)(sbi->s_kbytes_written +
2504                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2505                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2506 }
2507
2508 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2509                                           struct ext4_sb_info *sbi,
2510                                           const char *buf, size_t count)
2511 {
2512         unsigned long t;
2513
2514         if (parse_strtoul(buf, 0x40000000, &t))
2515                 return -EINVAL;
2516
2517         if (t && !is_power_of_2(t))
2518                 return -EINVAL;
2519
2520         sbi->s_inode_readahead_blks = t;
2521         return count;
2522 }
2523
2524 static ssize_t sbi_ui_show(struct ext4_attr *a,
2525                            struct ext4_sb_info *sbi, char *buf)
2526 {
2527         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2528
2529         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2530 }
2531
2532 static ssize_t sbi_ui_store(struct ext4_attr *a,
2533                             struct ext4_sb_info *sbi,
2534                             const char *buf, size_t count)
2535 {
2536         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2537         unsigned long t;
2538
2539         if (parse_strtoul(buf, 0xffffffff, &t))
2540                 return -EINVAL;
2541         *ui = t;
2542         return count;
2543 }
2544
2545 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2546 static struct ext4_attr ext4_attr_##_name = {                   \
2547         .attr = {.name = __stringify(_name), .mode = _mode },   \
2548         .show   = _show,                                        \
2549         .store  = _store,                                       \
2550         .offset = offsetof(struct ext4_sb_info, _elname),       \
2551 }
2552 #define EXT4_ATTR(name, mode, show, store) \
2553 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2554
2555 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2556 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2557 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2558 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2559         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2560 #define ATTR_LIST(name) &ext4_attr_##name.attr
2561
2562 EXT4_RO_ATTR(delayed_allocation_blocks);
2563 EXT4_RO_ATTR(session_write_kbytes);
2564 EXT4_RO_ATTR(lifetime_write_kbytes);
2565 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2566                  inode_readahead_blks_store, s_inode_readahead_blks);
2567 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2568 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2569 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2570 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2571 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2572 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2573 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2574 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2575
2576 static struct attribute *ext4_attrs[] = {
2577         ATTR_LIST(delayed_allocation_blocks),
2578         ATTR_LIST(session_write_kbytes),
2579         ATTR_LIST(lifetime_write_kbytes),
2580         ATTR_LIST(inode_readahead_blks),
2581         ATTR_LIST(inode_goal),
2582         ATTR_LIST(mb_stats),
2583         ATTR_LIST(mb_max_to_scan),
2584         ATTR_LIST(mb_min_to_scan),
2585         ATTR_LIST(mb_order2_req),
2586         ATTR_LIST(mb_stream_req),
2587         ATTR_LIST(mb_group_prealloc),
2588         ATTR_LIST(max_writeback_mb_bump),
2589         NULL,
2590 };
2591
2592 /* Features this copy of ext4 supports */
2593 EXT4_INFO_ATTR(lazy_itable_init);
2594 EXT4_INFO_ATTR(batched_discard);
2595
2596 static struct attribute *ext4_feat_attrs[] = {
2597         ATTR_LIST(lazy_itable_init),
2598         ATTR_LIST(batched_discard),
2599         NULL,
2600 };
2601
2602 static ssize_t ext4_attr_show(struct kobject *kobj,
2603                               struct attribute *attr, char *buf)
2604 {
2605         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2606                                                 s_kobj);
2607         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2608
2609         return a->show ? a->show(a, sbi, buf) : 0;
2610 }
2611
2612 static ssize_t ext4_attr_store(struct kobject *kobj,
2613                                struct attribute *attr,
2614                                const char *buf, size_t len)
2615 {
2616         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2617                                                 s_kobj);
2618         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2619
2620         return a->store ? a->store(a, sbi, buf, len) : 0;
2621 }
2622
2623 static void ext4_sb_release(struct kobject *kobj)
2624 {
2625         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2626                                                 s_kobj);
2627         complete(&sbi->s_kobj_unregister);
2628 }
2629
2630 static const struct sysfs_ops ext4_attr_ops = {
2631         .show   = ext4_attr_show,
2632         .store  = ext4_attr_store,
2633 };
2634
2635 static struct kobj_type ext4_ktype = {
2636         .default_attrs  = ext4_attrs,
2637         .sysfs_ops      = &ext4_attr_ops,
2638         .release        = ext4_sb_release,
2639 };
2640
2641 static void ext4_feat_release(struct kobject *kobj)
2642 {
2643         complete(&ext4_feat->f_kobj_unregister);
2644 }
2645
2646 static struct kobj_type ext4_feat_ktype = {
2647         .default_attrs  = ext4_feat_attrs,
2648         .sysfs_ops      = &ext4_attr_ops,
2649         .release        = ext4_feat_release,
2650 };
2651
2652 /*
2653  * Check whether this filesystem can be mounted based on
2654  * the features present and the RDONLY/RDWR mount requested.
2655  * Returns 1 if this filesystem can be mounted as requested,
2656  * 0 if it cannot be.
2657  */
2658 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2659 {
2660         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2661                 ext4_msg(sb, KERN_ERR,
2662                         "Couldn't mount because of "
2663                         "unsupported optional features (%x)",
2664                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2665                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2666                 return 0;
2667         }
2668
2669         if (readonly)
2670                 return 1;
2671
2672         /* Check that feature set is OK for a read-write mount */
2673         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2674                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2675                          "unsupported optional features (%x)",
2676                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2677                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2678                 return 0;
2679         }
2680         /*
2681          * Large file size enabled file system can only be mounted
2682          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2683          */
2684         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2685                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2686                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2687                                  "cannot be mounted RDWR without "
2688                                  "CONFIG_LBDAF");
2689                         return 0;
2690                 }
2691         }
2692         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2693             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2694                 ext4_msg(sb, KERN_ERR,
2695                          "Can't support bigalloc feature without "
2696                          "extents feature\n");
2697                 return 0;
2698         }
2699         return 1;
2700 }
2701
2702 /*
2703  * This function is called once a day if we have errors logged
2704  * on the file system
2705  */
2706 static void print_daily_error_info(unsigned long arg)
2707 {
2708         struct super_block *sb = (struct super_block *) arg;
2709         struct ext4_sb_info *sbi;
2710         struct ext4_super_block *es;
2711
2712         sbi = EXT4_SB(sb);
2713         es = sbi->s_es;
2714
2715         if (es->s_error_count)
2716                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2717                          le32_to_cpu(es->s_error_count));
2718         if (es->s_first_error_time) {
2719                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2720                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2721                        (int) sizeof(es->s_first_error_func),
2722                        es->s_first_error_func,
2723                        le32_to_cpu(es->s_first_error_line));
2724                 if (es->s_first_error_ino)
2725                         printk(": inode %u",
2726                                le32_to_cpu(es->s_first_error_ino));
2727                 if (es->s_first_error_block)
2728                         printk(": block %llu", (unsigned long long)
2729                                le64_to_cpu(es->s_first_error_block));
2730                 printk("\n");
2731         }
2732         if (es->s_last_error_time) {
2733                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2734                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2735                        (int) sizeof(es->s_last_error_func),
2736                        es->s_last_error_func,
2737                        le32_to_cpu(es->s_last_error_line));
2738                 if (es->s_last_error_ino)
2739                         printk(": inode %u",
2740                                le32_to_cpu(es->s_last_error_ino));
2741                 if (es->s_last_error_block)
2742                         printk(": block %llu", (unsigned long long)
2743                                le64_to_cpu(es->s_last_error_block));
2744                 printk("\n");
2745         }
2746         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2747 }
2748
2749 /* Find next suitable group and run ext4_init_inode_table */
2750 static int ext4_run_li_request(struct ext4_li_request *elr)
2751 {
2752         struct ext4_group_desc *gdp = NULL;
2753         ext4_group_t group, ngroups;
2754         struct super_block *sb;
2755         unsigned long timeout = 0;
2756         int ret = 0;
2757
2758         sb = elr->lr_super;
2759         ngroups = EXT4_SB(sb)->s_groups_count;
2760
2761         for (group = elr->lr_next_group; group < ngroups; group++) {
2762                 gdp = ext4_get_group_desc(sb, group, NULL);
2763                 if (!gdp) {
2764                         ret = 1;
2765                         break;
2766                 }
2767
2768                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2769                         break;
2770         }
2771
2772         if (group == ngroups)
2773                 ret = 1;
2774
2775         if (!ret) {
2776                 timeout = jiffies;
2777                 ret = ext4_init_inode_table(sb, group,
2778                                             elr->lr_timeout ? 0 : 1);
2779                 if (elr->lr_timeout == 0) {
2780                         timeout = (jiffies - timeout) *
2781                                   elr->lr_sbi->s_li_wait_mult;
2782                         elr->lr_timeout = timeout;
2783                 }
2784                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2785                 elr->lr_next_group = group + 1;
2786         }
2787
2788         return ret;
2789 }
2790
2791 /*
2792  * Remove lr_request from the list_request and free the
2793  * request structure. Should be called with li_list_mtx held
2794  */
2795 static void ext4_remove_li_request(struct ext4_li_request *elr)
2796 {
2797         struct ext4_sb_info *sbi;
2798
2799         if (!elr)
2800                 return;
2801
2802         sbi = elr->lr_sbi;
2803
2804         list_del(&elr->lr_request);
2805         sbi->s_li_request = NULL;
2806         kfree(elr);
2807 }
2808
2809 static void ext4_unregister_li_request(struct super_block *sb)
2810 {
2811         mutex_lock(&ext4_li_mtx);
2812         if (!ext4_li_info) {
2813                 mutex_unlock(&ext4_li_mtx);
2814                 return;
2815         }
2816
2817         mutex_lock(&ext4_li_info->li_list_mtx);
2818         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2819         mutex_unlock(&ext4_li_info->li_list_mtx);
2820         mutex_unlock(&ext4_li_mtx);
2821 }
2822
2823 static struct task_struct *ext4_lazyinit_task;
2824
2825 /*
2826  * This is the function where ext4lazyinit thread lives. It walks
2827  * through the request list searching for next scheduled filesystem.
2828  * When such a fs is found, run the lazy initialization request
2829  * (ext4_rn_li_request) and keep track of the time spend in this
2830  * function. Based on that time we compute next schedule time of
2831  * the request. When walking through the list is complete, compute
2832  * next waking time and put itself into sleep.
2833  */
2834 static int ext4_lazyinit_thread(void *arg)
2835 {
2836         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2837         struct list_head *pos, *n;
2838         struct ext4_li_request *elr;
2839         unsigned long next_wakeup, cur;
2840
2841         BUG_ON(NULL == eli);
2842
2843 cont_thread:
2844         while (true) {
2845                 next_wakeup = MAX_JIFFY_OFFSET;
2846
2847                 mutex_lock(&eli->li_list_mtx);
2848                 if (list_empty(&eli->li_request_list)) {
2849                         mutex_unlock(&eli->li_list_mtx);
2850                         goto exit_thread;
2851                 }
2852
2853                 list_for_each_safe(pos, n, &eli->li_request_list) {
2854                         elr = list_entry(pos, struct ext4_li_request,
2855                                          lr_request);
2856
2857                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2858                                 if (ext4_run_li_request(elr) != 0) {
2859                                         /* error, remove the lazy_init job */
2860                                         ext4_remove_li_request(elr);
2861                                         continue;
2862                                 }
2863                         }
2864
2865                         if (time_before(elr->lr_next_sched, next_wakeup))
2866                                 next_wakeup = elr->lr_next_sched;
2867                 }
2868                 mutex_unlock(&eli->li_list_mtx);
2869
2870                 if (freezing(current))
2871                         refrigerator();
2872
2873                 cur = jiffies;
2874                 if ((time_after_eq(cur, next_wakeup)) ||
2875                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2876                         cond_resched();
2877                         continue;
2878                 }
2879
2880                 schedule_timeout_interruptible(next_wakeup - cur);
2881
2882                 if (kthread_should_stop()) {
2883                         ext4_clear_request_list();
2884                         goto exit_thread;
2885                 }
2886         }
2887
2888 exit_thread:
2889         /*
2890          * It looks like the request list is empty, but we need
2891          * to check it under the li_list_mtx lock, to prevent any
2892          * additions into it, and of course we should lock ext4_li_mtx
2893          * to atomically free the list and ext4_li_info, because at
2894          * this point another ext4 filesystem could be registering
2895          * new one.
2896          */
2897         mutex_lock(&ext4_li_mtx);
2898         mutex_lock(&eli->li_list_mtx);
2899         if (!list_empty(&eli->li_request_list)) {
2900                 mutex_unlock(&eli->li_list_mtx);
2901                 mutex_unlock(&ext4_li_mtx);
2902                 goto cont_thread;
2903         }
2904         mutex_unlock(&eli->li_list_mtx);
2905         kfree(ext4_li_info);
2906         ext4_li_info = NULL;
2907         mutex_unlock(&ext4_li_mtx);
2908
2909         return 0;
2910 }
2911
2912 static void ext4_clear_request_list(void)
2913 {
2914         struct list_head *pos, *n;
2915         struct ext4_li_request *elr;
2916
2917         mutex_lock(&ext4_li_info->li_list_mtx);
2918         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2919                 elr = list_entry(pos, struct ext4_li_request,
2920                                  lr_request);
2921                 ext4_remove_li_request(elr);
2922         }
2923         mutex_unlock(&ext4_li_info->li_list_mtx);
2924 }
2925
2926 static int ext4_run_lazyinit_thread(void)
2927 {
2928         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2929                                          ext4_li_info, "ext4lazyinit");
2930         if (IS_ERR(ext4_lazyinit_task)) {
2931                 int err = PTR_ERR(ext4_lazyinit_task);
2932                 ext4_clear_request_list();
2933                 kfree(ext4_li_info);
2934                 ext4_li_info = NULL;
2935                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2936                                  "initialization thread\n",
2937                                  err);
2938                 return err;
2939         }
2940         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2941         return 0;
2942 }
2943
2944 /*
2945  * Check whether it make sense to run itable init. thread or not.
2946  * If there is at least one uninitialized inode table, return
2947  * corresponding group number, else the loop goes through all
2948  * groups and return total number of groups.
2949  */
2950 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2951 {
2952         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2953         struct ext4_group_desc *gdp = NULL;
2954
2955         for (group = 0; group < ngroups; group++) {
2956                 gdp = ext4_get_group_desc(sb, group, NULL);
2957                 if (!gdp)
2958                         continue;
2959
2960                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2961                         break;
2962         }
2963
2964         return group;
2965 }
2966
2967 static int ext4_li_info_new(void)
2968 {
2969         struct ext4_lazy_init *eli = NULL;
2970
2971         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2972         if (!eli)
2973                 return -ENOMEM;
2974
2975         INIT_LIST_HEAD(&eli->li_request_list);
2976         mutex_init(&eli->li_list_mtx);
2977
2978         eli->li_state |= EXT4_LAZYINIT_QUIT;
2979
2980         ext4_li_info = eli;
2981
2982         return 0;
2983 }
2984
2985 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2986                                             ext4_group_t start)
2987 {
2988         struct ext4_sb_info *sbi = EXT4_SB(sb);
2989         struct ext4_li_request *elr;
2990         unsigned long rnd;
2991
2992         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2993         if (!elr)
2994                 return NULL;
2995
2996         elr->lr_super = sb;
2997         elr->lr_sbi = sbi;
2998         elr->lr_next_group = start;
2999
3000         /*
3001          * Randomize first schedule time of the request to
3002          * spread the inode table initialization requests
3003          * better.
3004          */
3005         get_random_bytes(&rnd, sizeof(rnd));
3006         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3007                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3008
3009         return elr;
3010 }
3011
3012 static int ext4_register_li_request(struct super_block *sb,
3013                                     ext4_group_t first_not_zeroed)
3014 {
3015         struct ext4_sb_info *sbi = EXT4_SB(sb);
3016         struct ext4_li_request *elr;
3017         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3018         int ret = 0;
3019
3020         if (sbi->s_li_request != NULL) {
3021                 /*
3022                  * Reset timeout so it can be computed again, because
3023                  * s_li_wait_mult might have changed.
3024                  */
3025                 sbi->s_li_request->lr_timeout = 0;
3026                 return 0;
3027         }
3028
3029         if (first_not_zeroed == ngroups ||
3030             (sb->s_flags & MS_RDONLY) ||
3031             !test_opt(sb, INIT_INODE_TABLE))
3032                 return 0;
3033
3034         elr = ext4_li_request_new(sb, first_not_zeroed);
3035         if (!elr)
3036                 return -ENOMEM;
3037
3038         mutex_lock(&ext4_li_mtx);
3039
3040         if (NULL == ext4_li_info) {
3041                 ret = ext4_li_info_new();
3042                 if (ret)
3043                         goto out;
3044         }
3045
3046         mutex_lock(&ext4_li_info->li_list_mtx);
3047         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3048         mutex_unlock(&ext4_li_info->li_list_mtx);
3049
3050         sbi->s_li_request = elr;
3051         /*
3052          * set elr to NULL here since it has been inserted to
3053          * the request_list and the removal and free of it is
3054          * handled by ext4_clear_request_list from now on.
3055          */
3056         elr = NULL;
3057
3058         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3059                 ret = ext4_run_lazyinit_thread();
3060                 if (ret)
3061                         goto out;
3062         }
3063 out:
3064         mutex_unlock(&ext4_li_mtx);
3065         if (ret)
3066                 kfree(elr);
3067         return ret;
3068 }
3069
3070 /*
3071  * We do not need to lock anything since this is called on
3072  * module unload.
3073  */
3074 static void ext4_destroy_lazyinit_thread(void)
3075 {
3076         /*
3077          * If thread exited earlier
3078          * there's nothing to be done.
3079          */
3080         if (!ext4_li_info || !ext4_lazyinit_task)
3081                 return;
3082
3083         kthread_stop(ext4_lazyinit_task);
3084 }
3085
3086 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3087 {
3088         char *orig_data = kstrdup(data, GFP_KERNEL);
3089         struct buffer_head *bh;
3090         struct ext4_super_block *es = NULL;
3091         struct ext4_sb_info *sbi;
3092         ext4_fsblk_t block;
3093         ext4_fsblk_t sb_block = get_sb_block(&data);
3094         ext4_fsblk_t logical_sb_block;
3095         unsigned long offset = 0;
3096         unsigned long journal_devnum = 0;
3097         unsigned long def_mount_opts;
3098         struct inode *root;
3099         char *cp;
3100         const char *descr;
3101         int ret = -ENOMEM;
3102         int blocksize, clustersize;
3103         unsigned int db_count;
3104         unsigned int i;
3105         int needs_recovery, has_huge_files, has_bigalloc;
3106         __u64 blocks_count;
3107         int err;
3108         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3109         ext4_group_t first_not_zeroed;
3110
3111         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3112         if (!sbi)
3113                 goto out_free_orig;
3114
3115         sbi->s_blockgroup_lock =
3116                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3117         if (!sbi->s_blockgroup_lock) {
3118                 kfree(sbi);
3119                 goto out_free_orig;
3120         }
3121         sb->s_fs_info = sbi;
3122         sbi->s_mount_opt = 0;
3123         sbi->s_resuid = EXT4_DEF_RESUID;
3124         sbi->s_resgid = EXT4_DEF_RESGID;
3125         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3126         sbi->s_sb_block = sb_block;
3127         if (sb->s_bdev->bd_part)
3128                 sbi->s_sectors_written_start =
3129                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3130
3131         /* Cleanup superblock name */
3132         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3133                 *cp = '!';
3134
3135         ret = -EINVAL;
3136         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3137         if (!blocksize) {
3138                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3139                 goto out_fail;
3140         }
3141
3142         /*
3143          * The ext4 superblock will not be buffer aligned for other than 1kB
3144          * block sizes.  We need to calculate the offset from buffer start.
3145          */
3146         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3147                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3148                 offset = do_div(logical_sb_block, blocksize);
3149         } else {
3150                 logical_sb_block = sb_block;
3151         }
3152
3153         if (!(bh = sb_bread(sb, logical_sb_block))) {
3154                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3155                 goto out_fail;
3156         }
3157         /*
3158          * Note: s_es must be initialized as soon as possible because
3159          *       some ext4 macro-instructions depend on its value
3160          */
3161         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3162         sbi->s_es = es;
3163         sb->s_magic = le16_to_cpu(es->s_magic);
3164         if (sb->s_magic != EXT4_SUPER_MAGIC)
3165                 goto cantfind_ext4;
3166         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3167
3168         /* Set defaults before we parse the mount options */
3169         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3170         set_opt(sb, INIT_INODE_TABLE);
3171         if (def_mount_opts & EXT4_DEFM_DEBUG)
3172                 set_opt(sb, DEBUG);
3173         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3174                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3175                         "2.6.38");
3176                 set_opt(sb, GRPID);
3177         }
3178         if (def_mount_opts & EXT4_DEFM_UID16)
3179                 set_opt(sb, NO_UID32);
3180         /* xattr user namespace & acls are now defaulted on */
3181 #ifdef CONFIG_EXT4_FS_XATTR
3182         set_opt(sb, XATTR_USER);
3183 #endif
3184 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3185         set_opt(sb, POSIX_ACL);
3186 #endif
3187         set_opt(sb, MBLK_IO_SUBMIT);
3188         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3189                 set_opt(sb, JOURNAL_DATA);
3190         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3191                 set_opt(sb, ORDERED_DATA);
3192         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3193                 set_opt(sb, WRITEBACK_DATA);
3194
3195         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3196                 set_opt(sb, ERRORS_PANIC);
3197         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3198                 set_opt(sb, ERRORS_CONT);
3199         else
3200                 set_opt(sb, ERRORS_RO);
3201         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3202                 set_opt(sb, BLOCK_VALIDITY);
3203         if (def_mount_opts & EXT4_DEFM_DISCARD)
3204                 set_opt(sb, DISCARD);
3205
3206         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3207         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3208         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3209         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3210         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3211
3212         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3213                 set_opt(sb, BARRIER);
3214
3215         /*
3216          * enable delayed allocation by default
3217          * Use -o nodelalloc to turn it off
3218          */
3219         if (!IS_EXT3_SB(sb) &&
3220             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3221                 set_opt(sb, DELALLOC);
3222
3223         /*
3224          * set default s_li_wait_mult for lazyinit, for the case there is
3225          * no mount option specified.
3226          */
3227         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3228
3229         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3230                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3231                 ext4_msg(sb, KERN_WARNING,
3232                          "failed to parse options in superblock: %s",
3233                          sbi->s_es->s_mount_opts);
3234         }
3235         if (!parse_options((char *) data, sb, &journal_devnum,
3236                            &journal_ioprio, NULL, 0))
3237                 goto failed_mount;
3238
3239         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3240                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3241                             "with data=journal disables delayed "
3242                             "allocation and O_DIRECT support!\n");
3243                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3244                         ext4_msg(sb, KERN_ERR, "can't mount with "
3245                                  "both data=journal and delalloc");
3246                         goto failed_mount;
3247                 }
3248                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3249                         ext4_msg(sb, KERN_ERR, "can't mount with "
3250                                  "both data=journal and delalloc");
3251                         goto failed_mount;
3252                 }
3253                 if (test_opt(sb, DELALLOC))
3254                         clear_opt(sb, DELALLOC);
3255         }
3256
3257         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3258         if (test_opt(sb, DIOREAD_NOLOCK)) {
3259                 if (blocksize < PAGE_SIZE) {
3260                         ext4_msg(sb, KERN_ERR, "can't mount with "
3261                                  "dioread_nolock if block size != PAGE_SIZE");
3262                         goto failed_mount;
3263                 }
3264         }
3265
3266         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3267                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3268
3269         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3270             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3271              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3272              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3273                 ext4_msg(sb, KERN_WARNING,
3274                        "feature flags set on rev 0 fs, "
3275                        "running e2fsck is recommended");
3276
3277         if (IS_EXT2_SB(sb)) {
3278                 if (ext2_feature_set_ok(sb))
3279                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3280                                  "using the ext4 subsystem");
3281                 else {
3282                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3283                                  "to feature incompatibilities");
3284                         goto failed_mount;
3285                 }
3286         }
3287
3288         if (IS_EXT3_SB(sb)) {
3289                 if (ext3_feature_set_ok(sb))
3290                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3291                                  "using the ext4 subsystem");
3292                 else {
3293                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3294                                  "to feature incompatibilities");
3295                         goto failed_mount;
3296                 }
3297         }
3298
3299         /*
3300          * Check feature flags regardless of the revision level, since we
3301          * previously didn't change the revision level when setting the flags,
3302          * so there is a chance incompat flags are set on a rev 0 filesystem.
3303          */
3304         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3305                 goto failed_mount;
3306
3307         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3308             blocksize > EXT4_MAX_BLOCK_SIZE) {
3309                 ext4_msg(sb, KERN_ERR,
3310                        "Unsupported filesystem blocksize %d", blocksize);
3311                 goto failed_mount;
3312         }
3313
3314         if (sb->s_blocksize != blocksize) {
3315                 /* Validate the filesystem blocksize */
3316                 if (!sb_set_blocksize(sb, blocksize)) {
3317                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3318                                         blocksize);
3319                         goto failed_mount;
3320                 }
3321
3322                 brelse(bh);
3323                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3324                 offset = do_div(logical_sb_block, blocksize);
3325                 bh = sb_bread(sb, logical_sb_block);
3326                 if (!bh) {
3327                         ext4_msg(sb, KERN_ERR,
3328                                "Can't read superblock on 2nd try");
3329                         goto failed_mount;
3330                 }
3331                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3332                 sbi->s_es = es;
3333                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3334                         ext4_msg(sb, KERN_ERR,
3335                                "Magic mismatch, very weird!");
3336                         goto failed_mount;
3337                 }
3338         }
3339
3340         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3341                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3342         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3343                                                       has_huge_files);
3344         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3345
3346         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3347                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3348                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3349         } else {
3350                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3351                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3352                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3353                     (!is_power_of_2(sbi->s_inode_size)) ||
3354                     (sbi->s_inode_size > blocksize)) {
3355                         ext4_msg(sb, KERN_ERR,
3356                                "unsupported inode size: %d",
3357                                sbi->s_inode_size);
3358                         goto failed_mount;
3359                 }
3360                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3361                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3362         }
3363
3364         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3365         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3366                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3367                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3368                     !is_power_of_2(sbi->s_desc_size)) {
3369                         ext4_msg(sb, KERN_ERR,
3370                                "unsupported descriptor size %lu",
3371                                sbi->s_desc_size);
3372                         goto failed_mount;
3373                 }
3374         } else
3375                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3376
3377         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3378         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3379         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3380                 goto cantfind_ext4;
3381
3382         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3383         if (sbi->s_inodes_per_block == 0)
3384                 goto cantfind_ext4;
3385         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3386                                         sbi->s_inodes_per_block;
3387         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3388         sbi->s_sbh = bh;
3389         sbi->s_mount_state = le16_to_cpu(es->s_state);
3390         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3391         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3392
3393         for (i = 0; i < 4; i++)
3394                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3395         sbi->s_def_hash_version = es->s_def_hash_version;
3396         i = le32_to_cpu(es->s_flags);
3397         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3398                 sbi->s_hash_unsigned = 3;
3399         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3400 #ifdef __CHAR_UNSIGNED__
3401                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3402                 sbi->s_hash_unsigned = 3;
3403 #else
3404                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3405 #endif
3406                 sb->s_dirt = 1;
3407         }
3408
3409         /* Handle clustersize */
3410         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3411         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3412                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3413         if (has_bigalloc) {
3414                 if (clustersize < blocksize) {
3415                         ext4_msg(sb, KERN_ERR,
3416                                  "cluster size (%d) smaller than "
3417                                  "block size (%d)", clustersize, blocksize);
3418                         goto failed_mount;
3419                 }
3420                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3421                         le32_to_cpu(es->s_log_block_size);
3422                 sbi->s_clusters_per_group =
3423                         le32_to_cpu(es->s_clusters_per_group);
3424                 if (sbi->s_clusters_per_group > blocksize * 8) {
3425                         ext4_msg(sb, KERN_ERR,
3426                                  "#clusters per group too big: %lu",
3427                                  sbi->s_clusters_per_group);
3428                         goto failed_mount;
3429                 }
3430                 if (sbi->s_blocks_per_group !=
3431                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3432                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3433                                  "clusters per group (%lu) inconsistent",
3434                                  sbi->s_blocks_per_group,
3435                                  sbi->s_clusters_per_group);
3436                         goto failed_mount;
3437                 }
3438         } else {
3439                 if (clustersize != blocksize) {
3440                         ext4_warning(sb, "fragment/cluster size (%d) != "
3441                                      "block size (%d)", clustersize,
3442                                      blocksize);
3443                         clustersize = blocksize;
3444                 }
3445                 if (sbi->s_blocks_per_group > blocksize * 8) {
3446                         ext4_msg(sb, KERN_ERR,
3447                                  "#blocks per group too big: %lu",
3448                                  sbi->s_blocks_per_group);
3449                         goto failed_mount;
3450                 }
3451                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3452                 sbi->s_cluster_bits = 0;
3453         }
3454         sbi->s_cluster_ratio = clustersize / blocksize;
3455
3456         if (sbi->s_inodes_per_group > blocksize * 8) {
3457                 ext4_msg(sb, KERN_ERR,
3458                        "#inodes per group too big: %lu",
3459                        sbi->s_inodes_per_group);
3460                 goto failed_mount;
3461         }
3462
3463         /*
3464          * Test whether we have more sectors than will fit in sector_t,
3465          * and whether the max offset is addressable by the page cache.
3466          */
3467         err = generic_check_addressable(sb->s_blocksize_bits,
3468                                         ext4_blocks_count(es));
3469         if (err) {
3470                 ext4_msg(sb, KERN_ERR, "filesystem"
3471                          " too large to mount safely on this system");
3472                 if (sizeof(sector_t) < 8)
3473                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3474                 ret = err;
3475                 goto failed_mount;
3476         }
3477
3478         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3479                 goto cantfind_ext4;
3480
3481         /* check blocks count against device size */
3482         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3483         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3484                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3485                        "exceeds size of device (%llu blocks)",
3486                        ext4_blocks_count(es), blocks_count);
3487                 goto failed_mount;
3488         }
3489
3490         /*
3491          * It makes no sense for the first data block to be beyond the end
3492          * of the filesystem.
3493          */
3494         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3495                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3496                          "block %u is beyond end of filesystem (%llu)",
3497                          le32_to_cpu(es->s_first_data_block),
3498                          ext4_blocks_count(es));
3499                 goto failed_mount;
3500         }
3501         blocks_count = (ext4_blocks_count(es) -
3502                         le32_to_cpu(es->s_first_data_block) +
3503                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3504         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3505         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3506                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3507                        "(block count %llu, first data block %u, "
3508                        "blocks per group %lu)", sbi->s_groups_count,
3509                        ext4_blocks_count(es),
3510                        le32_to_cpu(es->s_first_data_block),
3511                        EXT4_BLOCKS_PER_GROUP(sb));
3512                 goto failed_mount;
3513         }
3514         sbi->s_groups_count = blocks_count;
3515         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3516                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3517         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3518                    EXT4_DESC_PER_BLOCK(sb);
3519         sbi->s_group_desc = ext4_kvmalloc(db_count *
3520                                           sizeof(struct buffer_head *),
3521                                           GFP_KERNEL);
3522         if (sbi->s_group_desc == NULL) {
3523                 ext4_msg(sb, KERN_ERR, "not enough memory");
3524                 goto failed_mount;
3525         }
3526
3527         if (ext4_proc_root)
3528                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3529
3530         bgl_lock_init(sbi->s_blockgroup_lock);
3531
3532         for (i = 0; i < db_count; i++) {
3533                 block = descriptor_loc(sb, logical_sb_block, i);
3534                 sbi->s_group_desc[i] = sb_bread(sb, block);
3535                 if (!sbi->s_group_desc[i]) {
3536                         ext4_msg(sb, KERN_ERR,
3537                                "can't read group descriptor %d", i);
3538                         db_count = i;
3539                         goto failed_mount2;
3540                 }
3541         }
3542         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3543                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3544                 goto failed_mount2;
3545         }
3546         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3547                 if (!ext4_fill_flex_info(sb)) {
3548                         ext4_msg(sb, KERN_ERR,
3549                                "unable to initialize "
3550                                "flex_bg meta info!");
3551                         goto failed_mount2;
3552                 }
3553
3554         sbi->s_gdb_count = db_count;
3555         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3556         spin_lock_init(&sbi->s_next_gen_lock);
3557
3558         init_timer(&sbi->s_err_report);
3559         sbi->s_err_report.function = print_daily_error_info;
3560         sbi->s_err_report.data = (unsigned long) sb;
3561
3562         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3563                         ext4_count_free_clusters(sb));
3564         if (!err) {
3565                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3566                                 ext4_count_free_inodes(sb));
3567         }
3568         if (!err) {
3569                 err = percpu_counter_init(&sbi->s_dirs_counter,
3570                                 ext4_count_dirs(sb));
3571         }
3572         if (!err) {
3573                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3574         }
3575         if (err) {
3576                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3577                 goto failed_mount3;
3578         }
3579
3580         sbi->s_stripe = ext4_get_stripe_size(sbi);
3581         sbi->s_max_writeback_mb_bump = 128;
3582
3583         /*
3584          * set up enough so that it can read an inode
3585          */
3586         if (!test_opt(sb, NOLOAD) &&
3587             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3588                 sb->s_op = &ext4_sops;
3589         else
3590                 sb->s_op = &ext4_nojournal_sops;
3591         sb->s_export_op = &ext4_export_ops;
3592         sb->s_xattr = ext4_xattr_handlers;
3593 #ifdef CONFIG_QUOTA
3594         sb->s_qcop = &ext4_qctl_operations;
3595         sb->dq_op = &ext4_quota_operations;
3596 #endif
3597         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3598
3599         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3600         mutex_init(&sbi->s_orphan_lock);
3601         sbi->s_resize_flags = 0;
3602
3603         sb->s_root = NULL;
3604
3605         needs_recovery = (es->s_last_orphan != 0 ||
3606                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3607                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3608
3609         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3610             !(sb->s_flags & MS_RDONLY))
3611                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3612                         goto failed_mount3;
3613
3614         /*
3615          * The first inode we look at is the journal inode.  Don't try
3616          * root first: it may be modified in the journal!
3617          */
3618         if (!test_opt(sb, NOLOAD) &&
3619             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3620                 if (ext4_load_journal(sb, es, journal_devnum))
3621                         goto failed_mount3;
3622         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3623               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3624                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3625                        "suppressed and not mounted read-only");
3626                 goto failed_mount_wq;
3627         } else {
3628                 clear_opt(sb, DATA_FLAGS);
3629                 sbi->s_journal = NULL;
3630                 needs_recovery = 0;
3631                 goto no_journal;
3632         }
3633
3634         if (ext4_blocks_count(es) > 0xffffffffULL &&
3635             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3636                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3637                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3638                 goto failed_mount_wq;
3639         }
3640
3641         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3642                 jbd2_journal_set_features(sbi->s_journal,
3643                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3644                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3645         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3646                 jbd2_journal_set_features(sbi->s_journal,
3647                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3648                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3649                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3650         } else {
3651                 jbd2_journal_clear_features(sbi->s_journal,
3652                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3653                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3654         }
3655
3656         /* We have now updated the journal if required, so we can
3657          * validate the data journaling mode. */
3658         switch (test_opt(sb, DATA_FLAGS)) {
3659         case 0:
3660                 /* No mode set, assume a default based on the journal
3661                  * capabilities: ORDERED_DATA if the journal can
3662                  * cope, else JOURNAL_DATA
3663                  */
3664                 if (jbd2_journal_check_available_features
3665                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3666                         set_opt(sb, ORDERED_DATA);
3667                 else
3668                         set_opt(sb, JOURNAL_DATA);
3669                 break;
3670
3671         case EXT4_MOUNT_ORDERED_DATA:
3672         case EXT4_MOUNT_WRITEBACK_DATA:
3673                 if (!jbd2_journal_check_available_features
3674                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3675                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3676                                "requested data journaling mode");
3677                         goto failed_mount_wq;
3678                 }
3679         default:
3680                 break;
3681         }
3682         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3683
3684         /*
3685          * The journal may have updated the bg summary counts, so we
3686          * need to update the global counters.
3687          */
3688         percpu_counter_set(&sbi->s_freeclusters_counter,
3689                            ext4_count_free_clusters(sb));
3690         percpu_counter_set(&sbi->s_freeinodes_counter,
3691                            ext4_count_free_inodes(sb));
3692         percpu_counter_set(&sbi->s_dirs_counter,
3693                            ext4_count_dirs(sb));
3694         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3695
3696 no_journal:
3697         /*
3698          * The maximum number of concurrent works can be high and
3699          * concurrency isn't really necessary.  Limit it to 1.
3700          */
3701         EXT4_SB(sb)->dio_unwritten_wq =
3702                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3703         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3704                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3705                 goto failed_mount_wq;
3706         }
3707
3708         /*
3709          * The jbd2_journal_load will have done any necessary log recovery,
3710          * so we can safely mount the rest of the filesystem now.
3711          */
3712
3713         root = ext4_iget(sb, EXT4_ROOT_INO);
3714         if (IS_ERR(root)) {
3715                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3716                 ret = PTR_ERR(root);
3717                 root = NULL;
3718                 goto failed_mount4;
3719         }
3720         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3721                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3722                 goto failed_mount4;
3723         }
3724         sb->s_root = d_alloc_root(root);
3725         if (!sb->s_root) {
3726                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3727                 ret = -ENOMEM;
3728                 goto failed_mount4;
3729         }
3730
3731         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3732                 sb->s_flags |= MS_RDONLY;
3733
3734         /* determine the minimum size of new large inodes, if present */
3735         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3736                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3737                                                      EXT4_GOOD_OLD_INODE_SIZE;
3738                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3739                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3740                         if (sbi->s_want_extra_isize <
3741                             le16_to_cpu(es->s_want_extra_isize))
3742                                 sbi->s_want_extra_isize =
3743                                         le16_to_cpu(es->s_want_extra_isize);
3744                         if (sbi->s_want_extra_isize <
3745                             le16_to_cpu(es->s_min_extra_isize))
3746                                 sbi->s_want_extra_isize =
3747                                         le16_to_cpu(es->s_min_extra_isize);
3748                 }
3749         }
3750         /* Check if enough inode space is available */
3751         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3752                                                         sbi->s_inode_size) {
3753                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3754                                                        EXT4_GOOD_OLD_INODE_SIZE;
3755                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3756                          "available");
3757         }
3758
3759         err = ext4_setup_system_zone(sb);
3760         if (err) {
3761                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3762                          "zone (%d)", err);
3763                 goto failed_mount4;
3764         }
3765
3766         ext4_ext_init(sb);
3767         err = ext4_mb_init(sb, needs_recovery);
3768         if (err) {
3769                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3770                          err);
3771                 goto failed_mount5;
3772         }
3773
3774         err = ext4_register_li_request(sb, first_not_zeroed);
3775         if (err)
3776                 goto failed_mount6;
3777
3778         sbi->s_kobj.kset = ext4_kset;
3779         init_completion(&sbi->s_kobj_unregister);
3780         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3781                                    "%s", sb->s_id);
3782         if (err)
3783                 goto failed_mount7;
3784
3785         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3786         ext4_orphan_cleanup(sb, es);
3787         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3788         if (needs_recovery) {
3789                 ext4_msg(sb, KERN_INFO, "recovery complete");
3790                 ext4_mark_recovery_complete(sb, es);
3791         }
3792         if (EXT4_SB(sb)->s_journal) {
3793                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3794                         descr = " journalled data mode";
3795                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3796                         descr = " ordered data mode";
3797                 else
3798                         descr = " writeback data mode";
3799         } else
3800                 descr = "out journal";
3801
3802         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3803                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3804                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3805
3806         if (es->s_error_count)
3807                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3808
3809         kfree(orig_data);
3810         return 0;
3811
3812 cantfind_ext4:
3813         if (!silent)
3814                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3815         goto failed_mount;
3816
3817 failed_mount7:
3818         ext4_unregister_li_request(sb);
3819 failed_mount6:
3820         ext4_ext_release(sb);
3821 failed_mount5:
3822         ext4_mb_release(sb);
3823         ext4_release_system_zone(sb);
3824 failed_mount4:
3825         iput(root);
3826         sb->s_root = NULL;
3827         ext4_msg(sb, KERN_ERR, "mount failed");
3828         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3829 failed_mount_wq:
3830         if (sbi->s_journal) {
3831                 jbd2_journal_destroy(sbi->s_journal);
3832                 sbi->s_journal = NULL;
3833         }
3834 failed_mount3:
3835         del_timer(&sbi->s_err_report);
3836         if (sbi->s_flex_groups)
3837                 ext4_kvfree(sbi->s_flex_groups);
3838         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3839         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3840         percpu_counter_destroy(&sbi->s_dirs_counter);
3841         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3842         if (sbi->s_mmp_tsk)
3843                 kthread_stop(sbi->s_mmp_tsk);
3844 failed_mount2:
3845         for (i = 0; i < db_count; i++)
3846                 brelse(sbi->s_group_desc[i]);
3847         ext4_kvfree(sbi->s_group_desc);
3848 failed_mount:
3849         if (sbi->s_proc) {
3850                 remove_proc_entry(sb->s_id, ext4_proc_root);
3851         }
3852 #ifdef CONFIG_QUOTA
3853         for (i = 0; i < MAXQUOTAS; i++)
3854                 kfree(sbi->s_qf_names[i]);
3855 #endif
3856         ext4_blkdev_remove(sbi);
3857         brelse(bh);
3858 out_fail:
3859         sb->s_fs_info = NULL;
3860         kfree(sbi->s_blockgroup_lock);
3861         kfree(sbi);
3862 out_free_orig:
3863         kfree(orig_data);
3864         return ret;
3865 }
3866
3867 /*
3868  * Setup any per-fs journal parameters now.  We'll do this both on
3869  * initial mount, once the journal has been initialised but before we've
3870  * done any recovery; and again on any subsequent remount.
3871  */
3872 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3873 {
3874         struct ext4_sb_info *sbi = EXT4_SB(sb);
3875
3876         journal->j_commit_interval = sbi->s_commit_interval;
3877         journal->j_min_batch_time = sbi->s_min_batch_time;
3878         journal->j_max_batch_time = sbi->s_max_batch_time;
3879
3880         write_lock(&journal->j_state_lock);
3881         if (test_opt(sb, BARRIER))
3882                 journal->j_flags |= JBD2_BARRIER;
3883         else
3884                 journal->j_flags &= ~JBD2_BARRIER;
3885         if (test_opt(sb, DATA_ERR_ABORT))
3886                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3887         else
3888                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3889         write_unlock(&journal->j_state_lock);
3890 }
3891
3892 static journal_t *ext4_get_journal(struct super_block *sb,
3893                                    unsigned int journal_inum)
3894 {
3895         struct inode *journal_inode;
3896         journal_t *journal;
3897
3898         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3899
3900         /* First, test for the existence of a valid inode on disk.  Bad
3901          * things happen if we iget() an unused inode, as the subsequent
3902          * iput() will try to delete it. */
3903
3904         journal_inode = ext4_iget(sb, journal_inum);
3905         if (IS_ERR(journal_inode)) {
3906                 ext4_msg(sb, KERN_ERR, "no journal found");
3907                 return NULL;
3908         }
3909         if (!journal_inode->i_nlink) {
3910                 make_bad_inode(journal_inode);
3911                 iput(journal_inode);
3912                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3913                 return NULL;
3914         }
3915
3916         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3917                   journal_inode, journal_inode->i_size);
3918         if (!S_ISREG(journal_inode->i_mode)) {
3919                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3920                 iput(journal_inode);
3921                 return NULL;
3922         }
3923
3924         journal = jbd2_journal_init_inode(journal_inode);
3925         if (!journal) {
3926                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3927                 iput(journal_inode);
3928                 return NULL;
3929         }
3930         journal->j_private = sb;
3931         ext4_init_journal_params(sb, journal);
3932         return journal;
3933 }
3934
3935 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3936                                        dev_t j_dev)
3937 {
3938         struct buffer_head *bh;
3939         journal_t *journal;
3940         ext4_fsblk_t start;
3941         ext4_fsblk_t len;
3942         int hblock, blocksize;
3943         ext4_fsblk_t sb_block;
3944         unsigned long offset;
3945         struct ext4_super_block *es;
3946         struct block_device *bdev;
3947
3948         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3949
3950         bdev = ext4_blkdev_get(j_dev, sb);
3951         if (bdev == NULL)
3952                 return NULL;
3953
3954         blocksize = sb->s_blocksize;
3955         hblock = bdev_logical_block_size(bdev);
3956         if (blocksize < hblock) {
3957                 ext4_msg(sb, KERN_ERR,
3958                         "blocksize too small for journal device");
3959                 goto out_bdev;
3960         }
3961
3962         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3963         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3964         set_blocksize(bdev, blocksize);
3965         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3966                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3967                        "external journal");
3968                 goto out_bdev;
3969         }
3970
3971         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3972         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3973             !(le32_to_cpu(es->s_feature_incompat) &
3974               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3975                 ext4_msg(sb, KERN_ERR, "external journal has "
3976                                         "bad superblock");
3977                 brelse(bh);
3978                 goto out_bdev;
3979         }
3980
3981         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3982                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3983                 brelse(bh);
3984                 goto out_bdev;
3985         }
3986
3987         len = ext4_blocks_count(es);
3988         start = sb_block + 1;
3989         brelse(bh);     /* we're done with the superblock */
3990
3991         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3992                                         start, len, blocksize);
3993         if (!journal) {
3994                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3995                 goto out_bdev;
3996         }
3997         journal->j_private = sb;
3998         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3999         wait_on_buffer(journal->j_sb_buffer);
4000         if (!buffer_uptodate(journal->j_sb_buffer)) {
4001                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4002                 goto out_journal;
4003         }
4004         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4005                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4006                                         "user (unsupported) - %d",
4007                         be32_to_cpu(journal->j_superblock->s_nr_users));
4008                 goto out_journal;
4009         }
4010         EXT4_SB(sb)->journal_bdev = bdev;
4011         ext4_init_journal_params(sb, journal);
4012         return journal;
4013
4014 out_journal:
4015         jbd2_journal_destroy(journal);
4016 out_bdev:
4017         ext4_blkdev_put(bdev);
4018         return NULL;
4019 }
4020
4021 static int ext4_load_journal(struct super_block *sb,
4022                              struct ext4_super_block *es,
4023                              unsigned long journal_devnum)
4024 {
4025         journal_t *journal;
4026         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4027         dev_t journal_dev;
4028         int err = 0;
4029         int really_read_only;
4030
4031         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4032
4033         if (journal_devnum &&
4034             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4035                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4036                         "numbers have changed");
4037                 journal_dev = new_decode_dev(journal_devnum);
4038         } else
4039                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4040
4041         really_read_only = bdev_read_only(sb->s_bdev);
4042
4043         /*
4044          * Are we loading a blank journal or performing recovery after a
4045          * crash?  For recovery, we need to check in advance whether we
4046          * can get read-write access to the device.
4047          */
4048         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4049                 if (sb->s_flags & MS_RDONLY) {
4050                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4051                                         "required on readonly filesystem");
4052                         if (really_read_only) {
4053                                 ext4_msg(sb, KERN_ERR, "write access "
4054                                         "unavailable, cannot proceed");
4055                                 return -EROFS;
4056                         }
4057                         ext4_msg(sb, KERN_INFO, "write access will "
4058                                "be enabled during recovery");
4059                 }
4060         }
4061
4062         if (journal_inum && journal_dev) {
4063                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4064                        "and inode journals!");
4065                 return -EINVAL;
4066         }
4067
4068         if (journal_inum) {
4069                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4070                         return -EINVAL;
4071         } else {
4072                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4073                         return -EINVAL;
4074         }
4075
4076         if (!(journal->j_flags & JBD2_BARRIER))
4077                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4078
4079         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4080                 err = jbd2_journal_update_format(journal);
4081                 if (err)  {
4082                         ext4_msg(sb, KERN_ERR, "error updating journal");
4083                         jbd2_journal_destroy(journal);
4084                         return err;
4085                 }
4086         }
4087
4088         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4089                 err = jbd2_journal_wipe(journal, !really_read_only);
4090         if (!err) {
4091                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4092                 if (save)
4093                         memcpy(save, ((char *) es) +
4094                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4095                 err = jbd2_journal_load(journal);
4096                 if (save)
4097                         memcpy(((char *) es) + EXT4_S_ERR_START,
4098                                save, EXT4_S_ERR_LEN);
4099                 kfree(save);
4100         }
4101
4102         if (err) {
4103                 ext4_msg(sb, KERN_ERR, "error loading journal");
4104                 jbd2_journal_destroy(journal);
4105                 return err;
4106         }
4107
4108         EXT4_SB(sb)->s_journal = journal;
4109         ext4_clear_journal_err(sb, es);
4110
4111         if (!really_read_only && journal_devnum &&
4112             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4113                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4114
4115                 /* Make sure we flush the recovery flag to disk. */
4116                 ext4_commit_super(sb, 1);
4117         }
4118
4119         return 0;
4120 }
4121
4122 static int ext4_commit_super(struct super_block *sb, int sync)
4123 {
4124         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4125         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4126         int error = 0;
4127
4128         if (!sbh || block_device_ejected(sb))
4129                 return error;
4130         if (buffer_write_io_error(sbh)) {
4131                 /*
4132                  * Oh, dear.  A previous attempt to write the
4133                  * superblock failed.  This could happen because the
4134                  * USB device was yanked out.  Or it could happen to
4135                  * be a transient write error and maybe the block will
4136                  * be remapped.  Nothing we can do but to retry the
4137                  * write and hope for the best.
4138                  */
4139                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4140                        "superblock detected");
4141                 clear_buffer_write_io_error(sbh);
4142                 set_buffer_uptodate(sbh);
4143         }
4144         /*
4145          * If the file system is mounted read-only, don't update the
4146          * superblock write time.  This avoids updating the superblock
4147          * write time when we are mounting the root file system
4148          * read/only but we need to replay the journal; at that point,
4149          * for people who are east of GMT and who make their clock
4150          * tick in localtime for Windows bug-for-bug compatibility,
4151          * the clock is set in the future, and this will cause e2fsck
4152          * to complain and force a full file system check.
4153          */
4154         if (!(sb->s_flags & MS_RDONLY))
4155                 es->s_wtime = cpu_to_le32(get_seconds());
4156         if (sb->s_bdev->bd_part)
4157                 es->s_kbytes_written =
4158                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4159                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4160                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4161         else
4162                 es->s_kbytes_written =
4163                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4164         ext4_free_blocks_count_set(es,
4165                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4166                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4167         es->s_free_inodes_count =
4168                 cpu_to_le32(percpu_counter_sum_positive(
4169                                 &EXT4_SB(sb)->s_freeinodes_counter));
4170         sb->s_dirt = 0;
4171         BUFFER_TRACE(sbh, "marking dirty");
4172         mark_buffer_dirty(sbh);
4173         if (sync) {
4174                 error = sync_dirty_buffer(sbh);
4175                 if (error)
4176                         return error;
4177
4178                 error = buffer_write_io_error(sbh);
4179                 if (error) {
4180                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4181                                "superblock");
4182                         clear_buffer_write_io_error(sbh);
4183                         set_buffer_uptodate(sbh);
4184                 }
4185         }
4186         return error;
4187 }
4188
4189 /*
4190  * Have we just finished recovery?  If so, and if we are mounting (or
4191  * remounting) the filesystem readonly, then we will end up with a
4192  * consistent fs on disk.  Record that fact.
4193  */
4194 static void ext4_mark_recovery_complete(struct super_block *sb,
4195                                         struct ext4_super_block *es)
4196 {
4197         journal_t *journal = EXT4_SB(sb)->s_journal;
4198
4199         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4200                 BUG_ON(journal != NULL);
4201                 return;
4202         }
4203         jbd2_journal_lock_updates(journal);
4204         if (jbd2_journal_flush(journal) < 0)
4205                 goto out;
4206
4207         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4208             sb->s_flags & MS_RDONLY) {
4209                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4210                 ext4_commit_super(sb, 1);
4211         }
4212
4213 out:
4214         jbd2_journal_unlock_updates(journal);
4215 }
4216
4217 /*
4218  * If we are mounting (or read-write remounting) a filesystem whose journal
4219  * has recorded an error from a previous lifetime, move that error to the
4220  * main filesystem now.
4221  */
4222 static void ext4_clear_journal_err(struct super_block *sb,
4223                                    struct ext4_super_block *es)
4224 {
4225         journal_t *journal;
4226         int j_errno;
4227         const char *errstr;
4228
4229         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4230
4231         journal = EXT4_SB(sb)->s_journal;
4232
4233         /*
4234          * Now check for any error status which may have been recorded in the
4235          * journal by a prior ext4_error() or ext4_abort()
4236          */
4237
4238         j_errno = jbd2_journal_errno(journal);
4239         if (j_errno) {
4240                 char nbuf[16];
4241
4242                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4243                 ext4_warning(sb, "Filesystem error recorded "
4244                              "from previous mount: %s", errstr);
4245                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4246
4247                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4248                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4249                 ext4_commit_super(sb, 1);
4250
4251                 jbd2_journal_clear_err(journal);
4252         }
4253 }
4254
4255 /*
4256  * Force the running and committing transactions to commit,
4257  * and wait on the commit.
4258  */
4259 int ext4_force_commit(struct super_block *sb)
4260 {
4261         journal_t *journal;
4262         int ret = 0;
4263
4264         if (sb->s_flags & MS_RDONLY)
4265                 return 0;
4266
4267         journal = EXT4_SB(sb)->s_journal;
4268         if (journal) {
4269                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4270                 ret = ext4_journal_force_commit(journal);
4271         }
4272
4273         return ret;
4274 }
4275
4276 static void ext4_write_super(struct super_block *sb)
4277 {
4278         lock_super(sb);
4279         ext4_commit_super(sb, 1);
4280         unlock_super(sb);
4281 }
4282
4283 static int ext4_sync_fs(struct super_block *sb, int wait)
4284 {
4285         int ret = 0;
4286         tid_t target;
4287         struct ext4_sb_info *sbi = EXT4_SB(sb);
4288
4289         trace_ext4_sync_fs(sb, wait);
4290         flush_workqueue(sbi->dio_unwritten_wq);
4291         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4292                 if (wait)
4293                         jbd2_log_wait_commit(sbi->s_journal, target);
4294         }
4295         return ret;
4296 }
4297
4298 /*
4299  * LVM calls this function before a (read-only) snapshot is created.  This
4300  * gives us a chance to flush the journal completely and mark the fs clean.
4301  *
4302  * Note that only this function cannot bring a filesystem to be in a clean
4303  * state independently, because ext4 prevents a new handle from being started
4304  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4305  * the upper layer.
4306  */
4307 static int ext4_freeze(struct super_block *sb)
4308 {
4309         int error = 0;
4310         journal_t *journal;
4311
4312         if (sb->s_flags & MS_RDONLY)
4313                 return 0;
4314
4315         journal = EXT4_SB(sb)->s_journal;
4316
4317         /* Now we set up the journal barrier. */
4318         jbd2_journal_lock_updates(journal);
4319
4320         /*
4321          * Don't clear the needs_recovery flag if we failed to flush
4322          * the journal.
4323          */
4324         error = jbd2_journal_flush(journal);
4325         if (error < 0)
4326                 goto out;
4327
4328         /* Journal blocked and flushed, clear needs_recovery flag. */
4329         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4330         error = ext4_commit_super(sb, 1);
4331 out:
4332         /* we rely on s_frozen to stop further updates */
4333         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4334         return error;
4335 }
4336
4337 /*
4338  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4339  * flag here, even though the filesystem is not technically dirty yet.
4340  */
4341 static int ext4_unfreeze(struct super_block *sb)
4342 {
4343         if (sb->s_flags & MS_RDONLY)
4344                 return 0;
4345
4346         lock_super(sb);
4347         /* Reset the needs_recovery flag before the fs is unlocked. */
4348         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4349         ext4_commit_super(sb, 1);
4350         unlock_super(sb);
4351         return 0;
4352 }
4353
4354 /*
4355  * Structure to save mount options for ext4_remount's benefit
4356  */
4357 struct ext4_mount_options {
4358         unsigned long s_mount_opt;
4359         unsigned long s_mount_opt2;
4360         uid_t s_resuid;
4361         gid_t s_resgid;
4362         unsigned long s_commit_interval;
4363         u32 s_min_batch_time, s_max_batch_time;
4364 #ifdef CONFIG_QUOTA
4365         int s_jquota_fmt;
4366         char *s_qf_names[MAXQUOTAS];
4367 #endif
4368 };
4369
4370 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4371 {
4372         struct ext4_super_block *es;
4373         struct ext4_sb_info *sbi = EXT4_SB(sb);
4374         ext4_fsblk_t n_blocks_count = 0;
4375         unsigned long old_sb_flags;
4376         struct ext4_mount_options old_opts;
4377         int enable_quota = 0;
4378         ext4_group_t g;
4379         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4380         int err = 0;
4381 #ifdef CONFIG_QUOTA
4382         int i;
4383 #endif
4384         char *orig_data = kstrdup(data, GFP_KERNEL);
4385
4386         /* Store the original options */
4387         lock_super(sb);
4388         old_sb_flags = sb->s_flags;
4389         old_opts.s_mount_opt = sbi->s_mount_opt;
4390         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4391         old_opts.s_resuid = sbi->s_resuid;
4392         old_opts.s_resgid = sbi->s_resgid;
4393         old_opts.s_commit_interval = sbi->s_commit_interval;
4394         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4395         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4396 #ifdef CONFIG_QUOTA
4397         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4398         for (i = 0; i < MAXQUOTAS; i++)
4399                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4400 #endif
4401         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4402                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4403
4404         /*
4405          * Allow the "check" option to be passed as a remount option.
4406          */
4407         if (!parse_options(data, sb, NULL, &journal_ioprio,
4408                            &n_blocks_count, 1)) {
4409                 err = -EINVAL;
4410                 goto restore_opts;
4411         }
4412
4413         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4414                 ext4_abort(sb, "Abort forced by user");
4415
4416         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4417                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4418
4419         es = sbi->s_es;
4420
4421         if (sbi->s_journal) {
4422                 ext4_init_journal_params(sb, sbi->s_journal);
4423                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4424         }
4425
4426         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4427                 n_blocks_count > ext4_blocks_count(es)) {
4428                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4429                         err = -EROFS;
4430                         goto restore_opts;
4431                 }
4432
4433                 if (*flags & MS_RDONLY) {
4434                         err = dquot_suspend(sb, -1);
4435                         if (err < 0)
4436                                 goto restore_opts;
4437
4438                         /*
4439                          * First of all, the unconditional stuff we have to do
4440                          * to disable replay of the journal when we next remount
4441                          */
4442                         sb->s_flags |= MS_RDONLY;
4443
4444                         /*
4445                          * OK, test if we are remounting a valid rw partition
4446                          * readonly, and if so set the rdonly flag and then
4447                          * mark the partition as valid again.
4448                          */
4449                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4450                             (sbi->s_mount_state & EXT4_VALID_FS))
4451                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4452
4453                         if (sbi->s_journal)
4454                                 ext4_mark_recovery_complete(sb, es);
4455                 } else {
4456                         /* Make sure we can mount this feature set readwrite */
4457                         if (!ext4_feature_set_ok(sb, 0)) {
4458                                 err = -EROFS;
4459                                 goto restore_opts;
4460                         }
4461                         /*
4462                          * Make sure the group descriptor checksums
4463                          * are sane.  If they aren't, refuse to remount r/w.
4464                          */
4465                         for (g = 0; g < sbi->s_groups_count; g++) {
4466                                 struct ext4_group_desc *gdp =
4467                                         ext4_get_group_desc(sb, g, NULL);
4468
4469                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4470                                         ext4_msg(sb, KERN_ERR,
4471                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4472                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4473                                                le16_to_cpu(gdp->bg_checksum));
4474                                         err = -EINVAL;
4475                                         goto restore_opts;
4476                                 }
4477                         }
4478
4479                         /*
4480                          * If we have an unprocessed orphan list hanging
4481                          * around from a previously readonly bdev mount,
4482                          * require a full umount/remount for now.
4483                          */
4484                         if (es->s_last_orphan) {
4485                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4486                                        "remount RDWR because of unprocessed "
4487                                        "orphan inode list.  Please "
4488                                        "umount/remount instead");
4489                                 err = -EINVAL;
4490                                 goto restore_opts;
4491                         }
4492
4493                         /*
4494                          * Mounting a RDONLY partition read-write, so reread
4495                          * and store the current valid flag.  (It may have
4496                          * been changed by e2fsck since we originally mounted
4497                          * the partition.)
4498                          */
4499                         if (sbi->s_journal)
4500                                 ext4_clear_journal_err(sb, es);
4501                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4502                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4503                                 goto restore_opts;
4504                         if (!ext4_setup_super(sb, es, 0))
4505                                 sb->s_flags &= ~MS_RDONLY;
4506                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4507                                                      EXT4_FEATURE_INCOMPAT_MMP))
4508                                 if (ext4_multi_mount_protect(sb,
4509                                                 le64_to_cpu(es->s_mmp_block))) {
4510                                         err = -EROFS;
4511                                         goto restore_opts;
4512                                 }
4513                         enable_quota = 1;
4514                 }
4515         }
4516
4517         /*
4518          * Reinitialize lazy itable initialization thread based on
4519          * current settings
4520          */
4521         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4522                 ext4_unregister_li_request(sb);
4523         else {
4524                 ext4_group_t first_not_zeroed;
4525                 first_not_zeroed = ext4_has_uninit_itable(sb);
4526                 ext4_register_li_request(sb, first_not_zeroed);
4527         }
4528
4529         ext4_setup_system_zone(sb);
4530         if (sbi->s_journal == NULL)
4531                 ext4_commit_super(sb, 1);
4532
4533 #ifdef CONFIG_QUOTA
4534         /* Release old quota file names */
4535         for (i = 0; i < MAXQUOTAS; i++)
4536                 if (old_opts.s_qf_names[i] &&
4537                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4538                         kfree(old_opts.s_qf_names[i]);
4539 #endif
4540         unlock_super(sb);
4541         if (enable_quota)
4542                 dquot_resume(sb, -1);
4543
4544         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4545         kfree(orig_data);
4546         return 0;
4547
4548 restore_opts:
4549         sb->s_flags = old_sb_flags;
4550         sbi->s_mount_opt = old_opts.s_mount_opt;
4551         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4552         sbi->s_resuid = old_opts.s_resuid;
4553         sbi->s_resgid = old_opts.s_resgid;
4554         sbi->s_commit_interval = old_opts.s_commit_interval;
4555         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4556         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4557 #ifdef CONFIG_QUOTA
4558         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4559         for (i = 0; i < MAXQUOTAS; i++) {
4560                 if (sbi->s_qf_names[i] &&
4561                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4562                         kfree(sbi->s_qf_names[i]);
4563                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4564         }
4565 #endif
4566         unlock_super(sb);
4567         kfree(orig_data);
4568         return err;
4569 }
4570
4571 /*
4572  * Note: calculating the overhead so we can be compatible with
4573  * historical BSD practice is quite difficult in the face of
4574  * clusters/bigalloc.  This is because multiple metadata blocks from
4575  * different block group can end up in the same allocation cluster.
4576  * Calculating the exact overhead in the face of clustered allocation
4577  * requires either O(all block bitmaps) in memory or O(number of block
4578  * groups**2) in time.  We will still calculate the superblock for
4579  * older file systems --- and if we come across with a bigalloc file
4580  * system with zero in s_overhead_clusters the estimate will be close to
4581  * correct especially for very large cluster sizes --- but for newer
4582  * file systems, it's better to calculate this figure once at mkfs
4583  * time, and store it in the superblock.  If the superblock value is
4584  * present (even for non-bigalloc file systems), we will use it.
4585  */
4586 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4587 {
4588         struct super_block *sb = dentry->d_sb;
4589         struct ext4_sb_info *sbi = EXT4_SB(sb);
4590         struct ext4_super_block *es = sbi->s_es;
4591         struct ext4_group_desc *gdp;
4592         u64 fsid;
4593         s64 bfree;
4594
4595         if (test_opt(sb, MINIX_DF)) {
4596                 sbi->s_overhead_last = 0;
4597         } else if (es->s_overhead_clusters) {
4598                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4599         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4600                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4601                 ext4_fsblk_t overhead = 0;
4602
4603                 /*
4604                  * Compute the overhead (FS structures).  This is constant
4605                  * for a given filesystem unless the number of block groups
4606                  * changes so we cache the previous value until it does.
4607                  */
4608
4609                 /*
4610                  * All of the blocks before first_data_block are
4611                  * overhead
4612                  */
4613                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4614
4615                 /*
4616                  * Add the overhead found in each block group
4617                  */
4618                 for (i = 0; i < ngroups; i++) {
4619                         gdp = ext4_get_group_desc(sb, i, NULL);
4620                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4621                         cond_resched();
4622                 }
4623                 sbi->s_overhead_last = overhead;
4624                 smp_wmb();
4625                 sbi->s_blocks_last = ext4_blocks_count(es);
4626         }
4627
4628         buf->f_type = EXT4_SUPER_MAGIC;
4629         buf->f_bsize = sb->s_blocksize;
4630         buf->f_blocks = (ext4_blocks_count(es) -
4631                          EXT4_C2B(sbi, sbi->s_overhead_last));
4632         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4633                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4634         /* prevent underflow in case that few free space is available */
4635         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4636         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4637         if (buf->f_bfree < ext4_r_blocks_count(es))
4638                 buf->f_bavail = 0;
4639         buf->f_files = le32_to_cpu(es->s_inodes_count);
4640         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4641         buf->f_namelen = EXT4_NAME_LEN;
4642         fsid = le64_to_cpup((void *)es->s_uuid) ^
4643                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4644         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4645         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4646
4647         return 0;
4648 }
4649
4650 /* Helper function for writing quotas on sync - we need to start transaction
4651  * before quota file is locked for write. Otherwise the are possible deadlocks:
4652  * Process 1                         Process 2
4653  * ext4_create()                     quota_sync()
4654  *   jbd2_journal_start()                  write_dquot()
4655  *   dquot_initialize()                         down(dqio_mutex)
4656  *     down(dqio_mutex)                    jbd2_journal_start()
4657  *
4658  */
4659
4660 #ifdef CONFIG_QUOTA
4661
4662 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4663 {
4664         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4665 }
4666
4667 static int ext4_write_dquot(struct dquot *dquot)
4668 {
4669         int ret, err;
4670         handle_t *handle;
4671         struct inode *inode;
4672
4673         inode = dquot_to_inode(dquot);
4674         handle = ext4_journal_start(inode,
4675                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4676         if (IS_ERR(handle))
4677                 return PTR_ERR(handle);
4678         ret = dquot_commit(dquot);
4679         err = ext4_journal_stop(handle);
4680         if (!ret)
4681                 ret = err;
4682         return ret;
4683 }
4684
4685 static int ext4_acquire_dquot(struct dquot *dquot)
4686 {
4687         int ret, err;
4688         handle_t *handle;
4689
4690         handle = ext4_journal_start(dquot_to_inode(dquot),
4691                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4692         if (IS_ERR(handle))
4693                 return PTR_ERR(handle);
4694         ret = dquot_acquire(dquot);
4695         err = ext4_journal_stop(handle);
4696         if (!ret)
4697                 ret = err;
4698         return ret;
4699 }
4700
4701 static int ext4_release_dquot(struct dquot *dquot)
4702 {
4703         int ret, err;
4704         handle_t *handle;
4705
4706         handle = ext4_journal_start(dquot_to_inode(dquot),
4707                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4708         if (IS_ERR(handle)) {
4709                 /* Release dquot anyway to avoid endless cycle in dqput() */
4710                 dquot_release(dquot);
4711                 return PTR_ERR(handle);
4712         }
4713         ret = dquot_release(dquot);
4714         err = ext4_journal_stop(handle);
4715         if (!ret)
4716                 ret = err;
4717         return ret;
4718 }
4719
4720 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4721 {
4722         /* Are we journaling quotas? */
4723         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4724             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4725                 dquot_mark_dquot_dirty(dquot);
4726                 return ext4_write_dquot(dquot);
4727         } else {
4728                 return dquot_mark_dquot_dirty(dquot);
4729         }
4730 }
4731
4732 static int ext4_write_info(struct super_block *sb, int type)
4733 {
4734         int ret, err;
4735         handle_t *handle;
4736
4737         /* Data block + inode block */
4738         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4739         if (IS_ERR(handle))
4740                 return PTR_ERR(handle);
4741         ret = dquot_commit_info(sb, type);
4742         err = ext4_journal_stop(handle);
4743         if (!ret)
4744                 ret = err;
4745         return ret;
4746 }
4747
4748 /*
4749  * Turn on quotas during mount time - we need to find
4750  * the quota file and such...
4751  */
4752 static int ext4_quota_on_mount(struct super_block *sb, int type)
4753 {
4754         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4755                                         EXT4_SB(sb)->s_jquota_fmt, type);
4756 }
4757
4758 /*
4759  * Standard function to be called on quota_on
4760  */
4761 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4762                          struct path *path)
4763 {
4764         int err;
4765
4766         if (!test_opt(sb, QUOTA))
4767                 return -EINVAL;
4768
4769         /* Quotafile not on the same filesystem? */
4770         if (path->mnt->mnt_sb != sb)
4771                 return -EXDEV;
4772         /* Journaling quota? */
4773         if (EXT4_SB(sb)->s_qf_names[type]) {
4774                 /* Quotafile not in fs root? */
4775                 if (path->dentry->d_parent != sb->s_root)
4776                         ext4_msg(sb, KERN_WARNING,
4777                                 "Quota file not on filesystem root. "
4778                                 "Journaled quota will not work");
4779         }
4780
4781         /*
4782          * When we journal data on quota file, we have to flush journal to see
4783          * all updates to the file when we bypass pagecache...
4784          */
4785         if (EXT4_SB(sb)->s_journal &&
4786             ext4_should_journal_data(path->dentry->d_inode)) {
4787                 /*
4788                  * We don't need to lock updates but journal_flush() could
4789                  * otherwise be livelocked...
4790                  */
4791                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4792                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4793                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4794                 if (err)
4795                         return err;
4796         }
4797
4798         return dquot_quota_on(sb, type, format_id, path);
4799 }
4800
4801 static int ext4_quota_off(struct super_block *sb, int type)
4802 {
4803         struct inode *inode = sb_dqopt(sb)->files[type];
4804         handle_t *handle;
4805
4806         /* Force all delayed allocation blocks to be allocated.
4807          * Caller already holds s_umount sem */
4808         if (test_opt(sb, DELALLOC))
4809                 sync_filesystem(sb);
4810
4811         if (!inode)
4812                 goto out;
4813
4814         /* Update modification times of quota files when userspace can
4815          * start looking at them */
4816         handle = ext4_journal_start(inode, 1);
4817         if (IS_ERR(handle))
4818                 goto out;
4819         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4820         ext4_mark_inode_dirty(handle, inode);
4821         ext4_journal_stop(handle);
4822
4823 out:
4824         return dquot_quota_off(sb, type);
4825 }
4826
4827 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4828  * acquiring the locks... As quota files are never truncated and quota code
4829  * itself serializes the operations (and no one else should touch the files)
4830  * we don't have to be afraid of races */
4831 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4832                                size_t len, loff_t off)
4833 {
4834         struct inode *inode = sb_dqopt(sb)->files[type];
4835         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4836         int err = 0;
4837         int offset = off & (sb->s_blocksize - 1);
4838         int tocopy;
4839         size_t toread;
4840         struct buffer_head *bh;
4841         loff_t i_size = i_size_read(inode);
4842
4843         if (off > i_size)
4844                 return 0;
4845         if (off+len > i_size)
4846                 len = i_size-off;
4847         toread = len;
4848         while (toread > 0) {
4849                 tocopy = sb->s_blocksize - offset < toread ?
4850                                 sb->s_blocksize - offset : toread;
4851                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4852                 if (err)
4853                         return err;
4854                 if (!bh)        /* A hole? */
4855                         memset(data, 0, tocopy);
4856                 else
4857                         memcpy(data, bh->b_data+offset, tocopy);
4858                 brelse(bh);
4859                 offset = 0;
4860                 toread -= tocopy;
4861                 data += tocopy;
4862                 blk++;
4863         }
4864         return len;
4865 }
4866
4867 /* Write to quotafile (we know the transaction is already started and has
4868  * enough credits) */
4869 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4870                                 const char *data, size_t len, loff_t off)
4871 {
4872         struct inode *inode = sb_dqopt(sb)->files[type];
4873         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4874         int err = 0;
4875         int offset = off & (sb->s_blocksize - 1);
4876         struct buffer_head *bh;
4877         handle_t *handle = journal_current_handle();
4878
4879         if (EXT4_SB(sb)->s_journal && !handle) {
4880                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4881                         " cancelled because transaction is not started",
4882                         (unsigned long long)off, (unsigned long long)len);
4883                 return -EIO;
4884         }
4885         /*
4886          * Since we account only one data block in transaction credits,
4887          * then it is impossible to cross a block boundary.
4888          */
4889         if (sb->s_blocksize - offset < len) {
4890                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4891                         " cancelled because not block aligned",
4892                         (unsigned long long)off, (unsigned long long)len);
4893                 return -EIO;
4894         }
4895
4896         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4897         bh = ext4_bread(handle, inode, blk, 1, &err);
4898         if (!bh)
4899                 goto out;
4900         err = ext4_journal_get_write_access(handle, bh);
4901         if (err) {
4902                 brelse(bh);
4903                 goto out;
4904         }
4905         lock_buffer(bh);
4906         memcpy(bh->b_data+offset, data, len);
4907         flush_dcache_page(bh->b_page);
4908         unlock_buffer(bh);
4909         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4910         brelse(bh);
4911 out:
4912         if (err) {
4913                 mutex_unlock(&inode->i_mutex);
4914                 return err;
4915         }
4916         if (inode->i_size < off + len) {
4917                 i_size_write(inode, off + len);
4918                 EXT4_I(inode)->i_disksize = inode->i_size;
4919                 ext4_mark_inode_dirty(handle, inode);
4920         }
4921         mutex_unlock(&inode->i_mutex);
4922         return len;
4923 }
4924
4925 #endif
4926
4927 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4928                        const char *dev_name, void *data)
4929 {
4930         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4931 }
4932
4933 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4934 static inline void register_as_ext2(void)
4935 {
4936         int err = register_filesystem(&ext2_fs_type);
4937         if (err)
4938                 printk(KERN_WARNING
4939                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4940 }
4941
4942 static inline void unregister_as_ext2(void)
4943 {
4944         unregister_filesystem(&ext2_fs_type);
4945 }
4946
4947 static inline int ext2_feature_set_ok(struct super_block *sb)
4948 {
4949         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4950                 return 0;
4951         if (sb->s_flags & MS_RDONLY)
4952                 return 1;
4953         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4954                 return 0;
4955         return 1;
4956 }
4957 MODULE_ALIAS("ext2");
4958 #else
4959 static inline void register_as_ext2(void) { }
4960 static inline void unregister_as_ext2(void) { }
4961 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4962 #endif
4963
4964 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4965 static inline void register_as_ext3(void)
4966 {
4967         int err = register_filesystem(&ext3_fs_type);
4968         if (err)
4969                 printk(KERN_WARNING
4970                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4971 }
4972
4973 static inline void unregister_as_ext3(void)
4974 {
4975         unregister_filesystem(&ext3_fs_type);
4976 }
4977
4978 static inline int ext3_feature_set_ok(struct super_block *sb)
4979 {
4980         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4981                 return 0;
4982         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4983                 return 0;
4984         if (sb->s_flags & MS_RDONLY)
4985                 return 1;
4986         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4987                 return 0;
4988         return 1;
4989 }
4990 MODULE_ALIAS("ext3");
4991 #else
4992 static inline void register_as_ext3(void) { }
4993 static inline void unregister_as_ext3(void) { }
4994 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4995 #endif
4996
4997 static struct file_system_type ext4_fs_type = {
4998         .owner          = THIS_MODULE,
4999         .name           = "ext4",
5000         .mount          = ext4_mount,
5001         .kill_sb        = kill_block_super,
5002         .fs_flags       = FS_REQUIRES_DEV,
5003 };
5004
5005 static int __init ext4_init_feat_adverts(void)
5006 {
5007         struct ext4_features *ef;
5008         int ret = -ENOMEM;
5009
5010         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5011         if (!ef)
5012                 goto out;
5013
5014         ef->f_kobj.kset = ext4_kset;
5015         init_completion(&ef->f_kobj_unregister);
5016         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5017                                    "features");
5018         if (ret) {
5019                 kfree(ef);
5020                 goto out;
5021         }
5022
5023         ext4_feat = ef;
5024         ret = 0;
5025 out:
5026         return ret;
5027 }
5028
5029 static void ext4_exit_feat_adverts(void)
5030 {
5031         kobject_put(&ext4_feat->f_kobj);
5032         wait_for_completion(&ext4_feat->f_kobj_unregister);
5033         kfree(ext4_feat);
5034 }
5035
5036 /* Shared across all ext4 file systems */
5037 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5038 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5039
5040 static int __init ext4_init_fs(void)
5041 {
5042         int i, err;
5043
5044         ext4_check_flag_values();
5045
5046         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5047                 mutex_init(&ext4__aio_mutex[i]);
5048                 init_waitqueue_head(&ext4__ioend_wq[i]);
5049         }
5050
5051         err = ext4_init_pageio();
5052         if (err)
5053                 return err;
5054         err = ext4_init_system_zone();
5055         if (err)
5056                 goto out6;
5057         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5058         if (!ext4_kset)
5059                 goto out5;
5060         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5061
5062         err = ext4_init_feat_adverts();
5063         if (err)
5064                 goto out4;
5065
5066         err = ext4_init_mballoc();
5067         if (err)
5068                 goto out3;
5069
5070         err = ext4_init_xattr();
5071         if (err)
5072                 goto out2;
5073         err = init_inodecache();
5074         if (err)
5075                 goto out1;
5076         register_as_ext3();
5077         register_as_ext2();
5078         err = register_filesystem(&ext4_fs_type);
5079         if (err)
5080                 goto out;
5081
5082         ext4_li_info = NULL;
5083         mutex_init(&ext4_li_mtx);
5084         return 0;
5085 out:
5086         unregister_as_ext2();
5087         unregister_as_ext3();
5088         destroy_inodecache();
5089 out1:
5090         ext4_exit_xattr();
5091 out2:
5092         ext4_exit_mballoc();
5093 out3:
5094         ext4_exit_feat_adverts();
5095 out4:
5096         if (ext4_proc_root)
5097                 remove_proc_entry("fs/ext4", NULL);
5098         kset_unregister(ext4_kset);
5099 out5:
5100         ext4_exit_system_zone();
5101 out6:
5102         ext4_exit_pageio();
5103         return err;
5104 }
5105
5106 static void __exit ext4_exit_fs(void)
5107 {
5108         ext4_destroy_lazyinit_thread();
5109         unregister_as_ext2();
5110         unregister_as_ext3();
5111         unregister_filesystem(&ext4_fs_type);
5112         destroy_inodecache();
5113         ext4_exit_xattr();
5114         ext4_exit_mballoc();
5115         ext4_exit_feat_adverts();
5116         remove_proc_entry("fs/ext4", NULL);
5117         kset_unregister(ext4_kset);
5118         ext4_exit_system_zone();
5119         ext4_exit_pageio();
5120 }
5121
5122 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5123 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5124 MODULE_LICENSE("GPL");
5125 module_init(ext4_init_fs)
5126 module_exit(ext4_exit_fs)