fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434
 435 /* Deal with the reporting of failure conditions on a filesystem such as
 436  * inconsistencies detected or read IO failures.
 437  *
 438  * On ext2, we can store the error state of the filesystem in the
 439  * superblock.  That is not possible on ext4, because we may have other
 440  * write ordering constraints on the superblock which prevent us from
 441  * writing it out straight away; and given that the journal is about to
 442  * be aborted, we can't rely on the current, or future, transactions to
 443  * write out the superblock safely.
 444  *
 445  * We'll just use the jbd2_journal_abort() error code to record an error in
 446  * the journal instead.  On recovery, the journal will complain about
 447  * that error until we've noted it down and cleared it.
 448  */
 449
 450 static void ext4_handle_error(struct super_block *sb)
 451 {
 452         if (sb->s_flags & MS_RDONLY)
 453                 return;
 454
 455         if (!test_opt(sb, ERRORS_CONT)) {
 456                 journal_t *journal = EXT4_SB(sb)->s_journal;
 457
 458                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 459                 if (journal)
 460                         jbd2_journal_abort(journal, -EIO);
 461         }
 462         if (test_opt(sb, ERRORS_RO)) {
 463                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 464                 sb->s_flags |= MS_RDONLY;
 465         }
 466         if (test_opt(sb, ERRORS_PANIC)) {
 467                 if (EXT4_SB(sb)->s_journal &&
 468                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 469                         return;
 470                 panic("EXT4-fs (device %s): panic forced after error\n",
 471                         sb->s_id);
 472         }
 473 }
 474
 475 void __ext4_error(struct super_block *sb, const char *function,
 476                   unsigned int line, const char *fmt, ...)
 477 {
 478         struct va_format vaf;
 479         va_list args;
 480
 481         va_start(args, fmt);
 482         vaf.fmt = fmt;
 483         vaf.va = &args;
 484         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 485                sb->s_id, function, line, current->comm, &vaf);
 486         va_end(args);
 487         save_error_info(sb, function, line);
 488
 489         ext4_handle_error(sb);
 490 }
 491
 492 void ext4_error_inode(struct inode *inode, const char *function,
 493                       unsigned int line, ext4_fsblk_t block,
 494                       const char *fmt, ...)
 495 {
 496         va_list args;
 497         struct va_format vaf;
 498         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 499
 500         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 501         es->s_last_error_block = cpu_to_le64(block);
 502         save_error_info(inode->i_sb, function, line);
 503         va_start(args, fmt);
 504         vaf.fmt = fmt;
 505         vaf.va = &args;
 506         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 507                inode->i_sb->s_id, function, line, inode->i_ino);
 508         if (block)
 509                 printk(KERN_CONT "block %llu: ", block);
 510         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 511         va_end(args);
 512
 513         ext4_handle_error(inode->i_sb);
 514 }
 515
 516 void ext4_error_file(struct file *file, const char *function,
 517                      unsigned int line, ext4_fsblk_t block,
 518                      const char *fmt, ...)
 519 {
 520         va_list args;
 521         struct va_format vaf;
 522         struct ext4_super_block *es;
 523         struct inode *inode = file->f_dentry->d_inode;
 524         char pathname[80], *path;
 525
 526         es = EXT4_SB(inode->i_sb)->s_es;
 527         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 528         save_error_info(inode->i_sb, function, line);
 529         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 530         if (IS_ERR(path))
 531                 path = "(unknown)";
 532         printk(KERN_CRIT
 533                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 534                inode->i_sb->s_id, function, line, inode->i_ino);
 535         if (block)
 536                 printk(KERN_CONT "block %llu: ", block);
 537         va_start(args, fmt);
 538         vaf.fmt = fmt;
 539         vaf.va = &args;
 540         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 541         va_end(args);
 542
 543         ext4_handle_error(inode->i_sb);
 544 }
 545
 546 static const char *ext4_decode_error(struct super_block *sb, int errno,
 547                                      char nbuf[16])
 548 {
 549         char *errstr = NULL;
 550
 551         switch (errno) {
 552         case -EIO:
 553                 errstr = "IO failure";
 554                 break;
 555         case -ENOMEM:
 556                 errstr = "Out of memory";
 557                 break;
 558         case -EROFS:
 559                 if (!sb || (EXT4_SB(sb)->s_journal &&
 560                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 561                         errstr = "Journal has aborted";
 562                 else
 563                         errstr = "Readonly filesystem";
 564                 break;
 565         default:
 566                 /* If the caller passed in an extra buffer for unknown
 567                  * errors, textualise them now.  Else we just return
 568                  * NULL. */
 569                 if (nbuf) {
 570                         /* Check for truncated error codes... */
 571                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 572                                 errstr = nbuf;
 573                 }
 574                 break;
 575         }
 576
 577         return errstr;
 578 }
 579
 580 /* __ext4_std_error decodes expected errors from journaling functions
 581  * automatically and invokes the appropriate error response.  */
 582
 583 void __ext4_std_error(struct super_block *sb, const char *function,
 584                       unsigned int line, int errno)
 585 {
 586         char nbuf[16];
 587         const char *errstr;
 588
 589         /* Special case: if the error is EROFS, and we're not already
 590          * inside a transaction, then there's really no point in logging
 591          * an error. */
 592         if (errno == -EROFS && journal_current_handle() == NULL &&
 593             (sb->s_flags & MS_RDONLY))
 594                 return;
 595
 596         errstr = ext4_decode_error(sb, errno, nbuf);
 597         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 598                sb->s_id, function, line, errstr);
 599         save_error_info(sb, function, line);
 600
 601         ext4_handle_error(sb);
 602 }
 603
 604 /*
 605  * ext4_abort is a much stronger failure handler than ext4_error.  The
 606  * abort function may be used to deal with unrecoverable failures such
 607  * as journal IO errors or ENOMEM at a critical moment in log management.
 608  *
 609  * We unconditionally force the filesystem into an ABORT|READONLY state,
 610  * unless the error response on the fs has been set to panic in which
 611  * case we take the easy way out and panic immediately.
 612  */
 613
 614 void __ext4_abort(struct super_block *sb, const char *function,
 615                 unsigned int line, const char *fmt, ...)
 616 {
 617         va_list args;
 618
 619         save_error_info(sb, function, line);
 620         va_start(args, fmt);
 621         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 622                function, line);
 623         vprintk(fmt, args);
 624         printk("\n");
 625         va_end(args);
 626
 627         if ((sb->s_flags & MS_RDONLY) == 0) {
 628                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 629                 sb->s_flags |= MS_RDONLY;
 630                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 631                 if (EXT4_SB(sb)->s_journal)
 632                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 633                 save_error_info(sb, function, line);
 634         }
 635         if (test_opt(sb, ERRORS_PANIC)) {
 636                 if (EXT4_SB(sb)->s_journal &&
 637                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 638                         return;
 639                 panic("EXT4-fs panic from previous error\n");
 640         }
 641 }
 642
 643 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 644 {
 645         struct va_format vaf;
 646         va_list args;
 647
 648         va_start(args, fmt);
 649         vaf.fmt = fmt;
 650         vaf.va = &args;
 651         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_warning(struct super_block *sb, const char *function,
 656                     unsigned int line, const char *fmt, ...)
 657 {
 658         struct va_format vaf;
 659         va_list args;
 660
 661         va_start(args, fmt);
 662         vaf.fmt = fmt;
 663         vaf.va = &args;
 664         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 665                sb->s_id, function, line, &vaf);
 666         va_end(args);
 667 }
 668
 669 void __ext4_grp_locked_error(const char *function, unsigned int line,
 670                              struct super_block *sb, ext4_group_t grp,
 671                              unsigned long ino, ext4_fsblk_t block,
 672                              const char *fmt, ...)
 673 __releases(bitlock)
 674 __acquires(bitlock)
 675 {
 676         struct va_format vaf;
 677         va_list args;
 678         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 679
 680         es->s_last_error_ino = cpu_to_le32(ino);
 681         es->s_last_error_block = cpu_to_le64(block);
 682         __save_error_info(sb, function, line);
 683
 684         va_start(args, fmt);
 685
 686         vaf.fmt = fmt;
 687         vaf.va = &args;
 688         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 689                sb->s_id, function, line, grp);
 690         if (ino)
 691                 printk(KERN_CONT "inode %lu: ", ino);
 692         if (block)
 693                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 694         printk(KERN_CONT "%pV\n", &vaf);
 695         va_end(args);
 696
 697         if (test_opt(sb, ERRORS_CONT)) {
 698                 ext4_commit_super(sb, 0);
 699                 return;
 700         }
 701
 702         ext4_unlock_group(sb, grp);
 703         ext4_handle_error(sb);
 704         /*
 705          * We only get here in the ERRORS_RO case; relocking the group
 706          * may be dangerous, but nothing bad will happen since the
 707          * filesystem will have already been marked read/only and the
 708          * journal has been aborted.  We return 1 as a hint to callers
 709          * who might what to use the return value from
 710          * ext4_grp_locked_error() to distinguish between the
 711          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 712          * aggressively from the ext4 function in question, with a
 713          * more appropriate error code.
 714          */
 715         ext4_lock_group(sb, grp);
 716         return;
 717 }
 718
 719 void ext4_update_dynamic_rev(struct super_block *sb)
 720 {
 721         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 722
 723         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 724                 return;
 725
 726         ext4_warning(sb,
 727                      "updating to rev %d because of new feature flag, "
 728                      "running e2fsck is recommended",
 729                      EXT4_DYNAMIC_REV);
 730
 731         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 732         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 733         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 734         /* leave es->s_feature_*compat flags alone */
 735         /* es->s_uuid will be set by e2fsck if empty */
 736
 737         /*
 738          * The rest of the superblock fields should be zero, and if not it
 739          * means they are likely already in use, so leave them alone.  We
 740          * can leave it up to e2fsck to clean up any inconsistencies there.
 741          */
 742 }
 743
 744 /*
 745  * Open the external journal device
 746  */
 747 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 748 {
 749         struct block_device *bdev;
 750         char b[BDEVNAME_SIZE];
 751
 752         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 753         if (IS_ERR(bdev))
 754                 goto fail;
 755         return bdev;
 756
 757 fail:
 758         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 759                         __bdevname(dev, b), PTR_ERR(bdev));
 760         return NULL;
 761 }
 762
 763 /*
 764  * Release the journal device
 765  */
 766 static int ext4_blkdev_put(struct block_device *bdev)
 767 {
 768         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 769 }
 770
 771 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 772 {
 773         struct block_device *bdev;
 774         int ret = -ENODEV;
 775
 776         bdev = sbi->journal_bdev;
 777         if (bdev) {
 778                 ret = ext4_blkdev_put(bdev);
 779                 sbi->journal_bdev = NULL;
 780         }
 781         return ret;
 782 }
 783
 784 static inline struct inode *orphan_list_entry(struct list_head *l)
 785 {
 786         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 787 }
 788
 789 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 790 {
 791         struct list_head *l;
 792
 793         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 794                  le32_to_cpu(sbi->s_es->s_last_orphan));
 795
 796         printk(KERN_ERR "sb_info orphan list:\n");
 797         list_for_each(l, &sbi->s_orphan) {
 798                 struct inode *inode = orphan_list_entry(l);
 799                 printk(KERN_ERR "  "
 800                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 801                        inode->i_sb->s_id, inode->i_ino, inode,
 802                        inode->i_mode, inode->i_nlink,
 803                        NEXT_ORPHAN(inode));
 804         }
 805 }
 806
 807 static void ext4_put_super(struct super_block *sb)
 808 {
 809         struct ext4_sb_info *sbi = EXT4_SB(sb);
 810         struct ext4_super_block *es = sbi->s_es;
 811         int i, err;
 812
 813         ext4_unregister_li_request(sb);
 814         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 815
 816         flush_workqueue(sbi->dio_unwritten_wq);
 817         destroy_workqueue(sbi->dio_unwritten_wq);
 818
 819         lock_super(sb);
 820         if (sb->s_dirt)
 821                 ext4_commit_super(sb, 1);
 822
 823         if (sbi->s_journal) {
 824                 err = jbd2_journal_destroy(sbi->s_journal);
 825                 sbi->s_journal = NULL;
 826                 if (err < 0)
 827                         ext4_abort(sb, "Couldn't clean up the journal");
 828         }
 829
 830         del_timer_sync(&sbi->s_err_report);
 831         ext4_release_system_zone(sb);
 832         ext4_mb_release(sb);
 833         ext4_ext_release(sb);
 834         ext4_xattr_put_super(sb);
 835
 836         if (!(sb->s_flags & MS_RDONLY)) {
 837                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 838                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 839                 ext4_commit_super(sb, 1);
 840         }
 841         if (sbi->s_proc) {
 842                 remove_proc_entry(sb->s_id, ext4_proc_root);
 843         }
 844         kobject_del(&sbi->s_kobj);
 845
 846         for (i = 0; i < sbi->s_gdb_count; i++)
 847                 brelse(sbi->s_group_desc[i]);
 848         ext4_kvfree(sbi->s_group_desc);
 849         ext4_kvfree(sbi->s_flex_groups);
 850         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 851         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 852         percpu_counter_destroy(&sbi->s_dirs_counter);
 853         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 854         brelse(sbi->s_sbh);
 855 #ifdef CONFIG_QUOTA
 856         for (i = 0; i < MAXQUOTAS; i++)
 857                 kfree(sbi->s_qf_names[i]);
 858 #endif
 859
 860         /* Debugging code just in case the in-memory inode orphan list
 861          * isn't empty.  The on-disk one can be non-empty if we've
 862          * detected an error and taken the fs readonly, but the
 863          * in-memory list had better be clean by this point. */
 864         if (!list_empty(&sbi->s_orphan))
 865                 dump_orphan_list(sb, sbi);
 866         J_ASSERT(list_empty(&sbi->s_orphan));
 867
 868         sync_blockdev(sb->s_bdev);
 869         invalidate_bdev(sb->s_bdev);
 870         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 871                 /*
 872                  * Invalidate the journal device's buffers.  We don't want them
 873                  * floating about in memory - the physical journal device may
 874                  * hotswapped, and it breaks the `ro-after' testing code.
 875                  */
 876                 sync_blockdev(sbi->journal_bdev);
 877                 invalidate_bdev(sbi->journal_bdev);
 878                 ext4_blkdev_remove(sbi);
 879         }
 880         if (sbi->s_mmp_tsk)
 881                 kthread_stop(sbi->s_mmp_tsk);
 882         sb->s_fs_info = NULL;
 883         /*
 884          * Now that we are completely done shutting down the
 885          * superblock, we need to actually destroy the kobject.
 886          */
 887         unlock_super(sb);
 888         kobject_put(&sbi->s_kobj);
 889         wait_for_completion(&sbi->s_kobj_unregister);
 890         kfree(sbi->s_blockgroup_lock);
 891         kfree(sbi);
 892 }
 893
 894 static struct kmem_cache *ext4_inode_cachep;
 895
 896 /*
 897  * Called inside transaction, so use GFP_NOFS
 898  */
 899 static struct inode *ext4_alloc_inode(struct super_block *sb)
 900 {
 901         struct ext4_inode_info *ei;
 902
 903         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 904         if (!ei)
 905                 return NULL;
 906
 907         ei->vfs_inode.i_version = 1;
 908         ei->vfs_inode.i_data.writeback_index = 0;
 909         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 910         INIT_LIST_HEAD(&ei->i_prealloc_list);
 911         spin_lock_init(&ei->i_prealloc_lock);
 912         ei->i_reserved_data_blocks = 0;
 913         ei->i_reserved_meta_blocks = 0;
 914         ei->i_allocated_meta_blocks = 0;
 915         ei->i_da_metadata_calc_len = 0;
 916         ei->i_da_metadata_calc_last_lblock = 0;
 917         spin_lock_init(&(ei->i_block_reservation_lock));
 918 #ifdef CONFIG_QUOTA
 919         ei->i_reserved_quota = 0;
 920 #endif
 921         ei->jinode = NULL;
 922         INIT_LIST_HEAD(&ei->i_completed_io_list);
 923         spin_lock_init(&ei->i_completed_io_lock);
 924         ei->cur_aio_dio = NULL;
 925         ei->i_sync_tid = 0;
 926         ei->i_datasync_tid = 0;
 927         atomic_set(&ei->i_ioend_count, 0);
 928         atomic_set(&ei->i_aiodio_unwritten, 0);
 929
 930         return &ei->vfs_inode;
 931 }
 932
 933 static int ext4_drop_inode(struct inode *inode)
 934 {
 935         int drop = generic_drop_inode(inode);
 936
 937         trace_ext4_drop_inode(inode, drop);
 938         return drop;
 939 }
 940
 941 static void ext4_i_callback(struct rcu_head *head)
 942 {
 943         struct inode *inode = container_of(head, struct inode, i_rcu);
 944         INIT_LIST_HEAD(&inode->i_dentry);
 945         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 946 }
 947
 948 static void ext4_destroy_inode(struct inode *inode)
 949 {
 950         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 951                 ext4_msg(inode->i_sb, KERN_ERR,
 952                          "Inode %lu (%p): orphan list check failed!",
 953                          inode->i_ino, EXT4_I(inode));
 954                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 955                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 956                                 true);
 957                 dump_stack();
 958         }
 959         call_rcu(&inode->i_rcu, ext4_i_callback);
 960 }
 961
 962 static void init_once(void *foo)
 963 {
 964         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 965
 966         INIT_LIST_HEAD(&ei->i_orphan);
 967 #ifdef CONFIG_EXT4_FS_XATTR
 968         init_rwsem(&ei->xattr_sem);
 969 #endif
 970         init_rwsem(&ei->i_data_sem);
 971         inode_init_once(&ei->vfs_inode);
 972 }
 973
 974 static int init_inodecache(void)
 975 {
 976         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 977                                              sizeof(struct ext4_inode_info),
 978                                              0, (SLAB_RECLAIM_ACCOUNT|
 979                                                 SLAB_MEM_SPREAD),
 980                                              init_once);
 981         if (ext4_inode_cachep == NULL)
 982                 return -ENOMEM;
 983         return 0;
 984 }
 985
 986 static void destroy_inodecache(void)
 987 {
 988         kmem_cache_destroy(ext4_inode_cachep);
 989 }
 990
 991 void ext4_clear_inode(struct inode *inode)
 992 {
 993         invalidate_inode_buffers(inode);
 994         end_writeback(inode);
 995         dquot_drop(inode);
 996         ext4_discard_preallocations(inode);
 997         if (EXT4_I(inode)->jinode) {
 998                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 999                                                EXT4_I(inode)->jinode);
1000                 jbd2_free_inode(EXT4_I(inode)->jinode);
1001                 EXT4_I(inode)->jinode = NULL;
1002         }
1003 }
1004
1005 static inline void ext4_show_quota_options(struct seq_file *seq,
1006                                            struct super_block *sb)
1007 {
1008 #if defined(CONFIG_QUOTA)
1009         struct ext4_sb_info *sbi = EXT4_SB(sb);
1010
1011         if (sbi->s_jquota_fmt) {
1012                 char *fmtname = "";
1013
1014                 switch (sbi->s_jquota_fmt) {
1015                 case QFMT_VFS_OLD:
1016                         fmtname = "vfsold";
1017                         break;
1018                 case QFMT_VFS_V0:
1019                         fmtname = "vfsv0";
1020                         break;
1021                 case QFMT_VFS_V1:
1022                         fmtname = "vfsv1";
1023                         break;
1024                 }
1025                 seq_printf(seq, ",jqfmt=%s", fmtname);
1026         }
1027
1028         if (sbi->s_qf_names[USRQUOTA])
1029                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1030
1031         if (sbi->s_qf_names[GRPQUOTA])
1032                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1033
1034         if (test_opt(sb, USRQUOTA))
1035                 seq_puts(seq, ",usrquota");
1036
1037         if (test_opt(sb, GRPQUOTA))
1038                 seq_puts(seq, ",grpquota");
1039 #endif
1040 }
1041
1042 /*
1043  * Show an option if
1044  *  - it's set to a non-default value OR
1045  *  - if the per-sb default is different from the global default
1046  */
1047 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1048 {
1049         int def_errors;
1050         unsigned long def_mount_opts;
1051         struct super_block *sb = vfs->mnt_sb;
1052         struct ext4_sb_info *sbi = EXT4_SB(sb);
1053         struct ext4_super_block *es = sbi->s_es;
1054
1055         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1056         def_errors     = le16_to_cpu(es->s_errors);
1057
1058         if (sbi->s_sb_block != 1)
1059                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1060         if (test_opt(sb, MINIX_DF))
1061                 seq_puts(seq, ",minixdf");
1062         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1063                 seq_puts(seq, ",grpid");
1064         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1065                 seq_puts(seq, ",nogrpid");
1066         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1067             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1068                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1069         }
1070         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1071             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1072                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1073         }
1074         if (test_opt(sb, ERRORS_RO)) {
1075                 if (def_errors == EXT4_ERRORS_PANIC ||
1076                     def_errors == EXT4_ERRORS_CONTINUE) {
1077                         seq_puts(seq, ",errors=remount-ro");
1078                 }
1079         }
1080         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1081                 seq_puts(seq, ",errors=continue");
1082         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1083                 seq_puts(seq, ",errors=panic");
1084         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1085                 seq_puts(seq, ",nouid32");
1086         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1087                 seq_puts(seq, ",debug");
1088 #ifdef CONFIG_EXT4_FS_XATTR
1089         if (test_opt(sb, XATTR_USER))
1090                 seq_puts(seq, ",user_xattr");
1091         if (!test_opt(sb, XATTR_USER))
1092                 seq_puts(seq, ",nouser_xattr");
1093 #endif
1094 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1095         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1096                 seq_puts(seq, ",acl");
1097         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1098                 seq_puts(seq, ",noacl");
1099 #endif
1100         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1101                 seq_printf(seq, ",commit=%u",
1102                            (unsigned) (sbi->s_commit_interval / HZ));
1103         }
1104         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1105                 seq_printf(seq, ",min_batch_time=%u",
1106                            (unsigned) sbi->s_min_batch_time);
1107         }
1108         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1109                 seq_printf(seq, ",max_batch_time=%u",
1110                            (unsigned) sbi->s_max_batch_time);
1111         }
1112
1113         /*
1114          * We're changing the default of barrier mount option, so
1115          * let's always display its mount state so it's clear what its
1116          * status is.
1117          */
1118         seq_puts(seq, ",barrier=");
1119         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1120         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1121                 seq_puts(seq, ",journal_async_commit");
1122         else if (test_opt(sb, JOURNAL_CHECKSUM))
1123                 seq_puts(seq, ",journal_checksum");
1124         if (test_opt(sb, I_VERSION))
1125                 seq_puts(seq, ",i_version");
1126         if (!test_opt(sb, DELALLOC) &&
1127             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1128                 seq_puts(seq, ",nodelalloc");
1129
1130         if (!test_opt(sb, MBLK_IO_SUBMIT))
1131                 seq_puts(seq, ",nomblk_io_submit");
1132         if (sbi->s_stripe)
1133                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1134         /*
1135          * journal mode get enabled in different ways
1136          * So just print the value even if we didn't specify it
1137          */
1138         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1139                 seq_puts(seq, ",data=journal");
1140         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1141                 seq_puts(seq, ",data=ordered");
1142         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1143                 seq_puts(seq, ",data=writeback");
1144
1145         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1146                 seq_printf(seq, ",inode_readahead_blks=%u",
1147                            sbi->s_inode_readahead_blks);
1148
1149         if (test_opt(sb, DATA_ERR_ABORT))
1150                 seq_puts(seq, ",data_err=abort");
1151
1152         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1153                 seq_puts(seq, ",noauto_da_alloc");
1154
1155         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1156                 seq_puts(seq, ",discard");
1157
1158         if (test_opt(sb, NOLOAD))
1159                 seq_puts(seq, ",norecovery");
1160
1161         if (test_opt(sb, DIOREAD_NOLOCK))
1162                 seq_puts(seq, ",dioread_nolock");
1163
1164         if (test_opt(sb, BLOCK_VALIDITY) &&
1165             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1166                 seq_puts(seq, ",block_validity");
1167
1168         if (!test_opt(sb, INIT_INODE_TABLE))
1169                 seq_puts(seq, ",noinit_itable");
1170         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1171                 seq_printf(seq, ",init_itable=%u",
1172                            (unsigned) sbi->s_li_wait_mult);
1173
1174         ext4_show_quota_options(seq, sb);
1175
1176         return 0;
1177 }
1178
1179 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1180                                         u64 ino, u32 generation)
1181 {
1182         struct inode *inode;
1183
1184         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1185                 return ERR_PTR(-ESTALE);
1186         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1187                 return ERR_PTR(-ESTALE);
1188
1189         /* iget isn't really right if the inode is currently unallocated!!
1190          *
1191          * ext4_read_inode will return a bad_inode if the inode had been
1192          * deleted, so we should be safe.
1193          *
1194          * Currently we don't know the generation for parent directory, so
1195          * a generation of 0 means "accept any"
1196          */
1197         inode = ext4_iget_normal(sb, ino);
1198         if (IS_ERR(inode))
1199                 return ERR_CAST(inode);
1200         if (generation && inode->i_generation != generation) {
1201                 iput(inode);
1202                 return ERR_PTR(-ESTALE);
1203         }
1204
1205         return inode;
1206 }
1207
1208 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1209                                         int fh_len, int fh_type)
1210 {
1211         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1212                                     ext4_nfs_get_inode);
1213 }
1214
1215 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1216                                         int fh_len, int fh_type)
1217 {
1218         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1219                                     ext4_nfs_get_inode);
1220 }
1221
1222 /*
1223  * Try to release metadata pages (indirect blocks, directories) which are
1224  * mapped via the block device.  Since these pages could have journal heads
1225  * which would prevent try_to_free_buffers() from freeing them, we must use
1226  * jbd2 layer's try_to_free_buffers() function to release them.
1227  */
1228 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1229                                  gfp_t wait)
1230 {
1231         journal_t *journal = EXT4_SB(sb)->s_journal;
1232
1233         WARN_ON(PageChecked(page));
1234         if (!page_has_buffers(page))
1235                 return 0;
1236         if (journal)
1237                 return jbd2_journal_try_to_free_buffers(journal, page,
1238                                                         wait & ~__GFP_WAIT);
1239         return try_to_free_buffers(page);
1240 }
1241
1242 #ifdef CONFIG_QUOTA
1243 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1244 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1245
1246 static int ext4_write_dquot(struct dquot *dquot);
1247 static int ext4_acquire_dquot(struct dquot *dquot);
1248 static int ext4_release_dquot(struct dquot *dquot);
1249 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1250 static int ext4_write_info(struct super_block *sb, int type);
1251 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1252                          struct path *path);
1253 static int ext4_quota_off(struct super_block *sb, int type);
1254 static int ext4_quota_on_mount(struct super_block *sb, int type);
1255 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1256                                size_t len, loff_t off);
1257 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1258                                 const char *data, size_t len, loff_t off);
1259
1260 static const struct dquot_operations ext4_quota_operations = {
1261         .get_reserved_space = ext4_get_reserved_space,
1262         .write_dquot    = ext4_write_dquot,
1263         .acquire_dquot  = ext4_acquire_dquot,
1264         .release_dquot  = ext4_release_dquot,
1265         .mark_dirty     = ext4_mark_dquot_dirty,
1266         .write_info     = ext4_write_info,
1267         .alloc_dquot    = dquot_alloc,
1268         .destroy_dquot  = dquot_destroy,
1269 };
1270
1271 static const struct quotactl_ops ext4_qctl_operations = {
1272         .quota_on       = ext4_quota_on,
1273         .quota_off      = ext4_quota_off,
1274         .quota_sync     = dquot_quota_sync,
1275         .get_info       = dquot_get_dqinfo,
1276         .set_info       = dquot_set_dqinfo,
1277         .get_dqblk      = dquot_get_dqblk,
1278         .set_dqblk      = dquot_set_dqblk
1279 };
1280 #endif
1281
1282 static const struct super_operations ext4_sops = {
1283         .alloc_inode    = ext4_alloc_inode,
1284         .destroy_inode  = ext4_destroy_inode,
1285         .write_inode    = ext4_write_inode,
1286         .dirty_inode    = ext4_dirty_inode,
1287         .drop_inode     = ext4_drop_inode,
1288         .evict_inode    = ext4_evict_inode,
1289         .put_super      = ext4_put_super,
1290         .sync_fs        = ext4_sync_fs,
1291         .freeze_fs      = ext4_freeze,
1292         .unfreeze_fs    = ext4_unfreeze,
1293         .statfs         = ext4_statfs,
1294         .remount_fs     = ext4_remount,
1295         .show_options   = ext4_show_options,
1296 #ifdef CONFIG_QUOTA
1297         .quota_read     = ext4_quota_read,
1298         .quota_write    = ext4_quota_write,
1299 #endif
1300         .bdev_try_to_free_page = bdev_try_to_free_page,
1301 };
1302
1303 static const struct super_operations ext4_nojournal_sops = {
1304         .alloc_inode    = ext4_alloc_inode,
1305         .destroy_inode  = ext4_destroy_inode,
1306         .write_inode    = ext4_write_inode,
1307         .dirty_inode    = ext4_dirty_inode,
1308         .drop_inode     = ext4_drop_inode,
1309         .evict_inode    = ext4_evict_inode,
1310         .write_super    = ext4_write_super,
1311         .put_super      = ext4_put_super,
1312         .statfs         = ext4_statfs,
1313         .remount_fs     = ext4_remount,
1314         .show_options   = ext4_show_options,
1315 #ifdef CONFIG_QUOTA
1316         .quota_read     = ext4_quota_read,
1317         .quota_write    = ext4_quota_write,
1318 #endif
1319         .bdev_try_to_free_page = bdev_try_to_free_page,
1320 };
1321
1322 static const struct export_operations ext4_export_ops = {
1323         .fh_to_dentry = ext4_fh_to_dentry,
1324         .fh_to_parent = ext4_fh_to_parent,
1325         .get_parent = ext4_get_parent,
1326 };
1327
1328 enum {
1329         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1330         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1331         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1332         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1333         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1334         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1335         Opt_journal_update, Opt_journal_dev,
1336         Opt_journal_checksum, Opt_journal_async_commit,
1337         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1338         Opt_data_err_abort, Opt_data_err_ignore,
1339         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1340         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1341         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1342         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1343         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1344         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1345         Opt_inode_readahead_blks, Opt_journal_ioprio,
1346         Opt_dioread_nolock, Opt_dioread_lock,
1347         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1348 };
1349
1350 static const match_table_t tokens = {
1351         {Opt_bsd_df, "bsddf"},
1352         {Opt_minix_df, "minixdf"},
1353         {Opt_grpid, "grpid"},
1354         {Opt_grpid, "bsdgroups"},
1355         {Opt_nogrpid, "nogrpid"},
1356         {Opt_nogrpid, "sysvgroups"},
1357         {Opt_resgid, "resgid=%u"},
1358         {Opt_resuid, "resuid=%u"},
1359         {Opt_sb, "sb=%u"},
1360         {Opt_err_cont, "errors=continue"},
1361         {Opt_err_panic, "errors=panic"},
1362         {Opt_err_ro, "errors=remount-ro"},
1363         {Opt_nouid32, "nouid32"},
1364         {Opt_debug, "debug"},
1365         {Opt_oldalloc, "oldalloc"},
1366         {Opt_orlov, "orlov"},
1367         {Opt_user_xattr, "user_xattr"},
1368         {Opt_nouser_xattr, "nouser_xattr"},
1369         {Opt_acl, "acl"},
1370         {Opt_noacl, "noacl"},
1371         {Opt_noload, "noload"},
1372         {Opt_noload, "norecovery"},
1373         {Opt_nobh, "nobh"},
1374         {Opt_bh, "bh"},
1375         {Opt_commit, "commit=%u"},
1376         {Opt_min_batch_time, "min_batch_time=%u"},
1377         {Opt_max_batch_time, "max_batch_time=%u"},
1378         {Opt_journal_update, "journal=update"},
1379         {Opt_journal_dev, "journal_dev=%u"},
1380         {Opt_journal_checksum, "journal_checksum"},
1381         {Opt_journal_async_commit, "journal_async_commit"},
1382         {Opt_abort, "abort"},
1383         {Opt_data_journal, "data=journal"},
1384         {Opt_data_ordered, "data=ordered"},
1385         {Opt_data_writeback, "data=writeback"},
1386         {Opt_data_err_abort, "data_err=abort"},
1387         {Opt_data_err_ignore, "data_err=ignore"},
1388         {Opt_offusrjquota, "usrjquota="},
1389         {Opt_usrjquota, "usrjquota=%s"},
1390         {Opt_offgrpjquota, "grpjquota="},
1391         {Opt_grpjquota, "grpjquota=%s"},
1392         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1393         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1394         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1395         {Opt_grpquota, "grpquota"},
1396         {Opt_noquota, "noquota"},
1397         {Opt_quota, "quota"},
1398         {Opt_usrquota, "usrquota"},
1399         {Opt_barrier, "barrier=%u"},
1400         {Opt_barrier, "barrier"},
1401         {Opt_nobarrier, "nobarrier"},
1402         {Opt_i_version, "i_version"},
1403         {Opt_stripe, "stripe=%u"},
1404         {Opt_resize, "resize"},
1405         {Opt_delalloc, "delalloc"},
1406         {Opt_nodelalloc, "nodelalloc"},
1407         {Opt_mblk_io_submit, "mblk_io_submit"},
1408         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1409         {Opt_block_validity, "block_validity"},
1410         {Opt_noblock_validity, "noblock_validity"},
1411         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1412         {Opt_journal_ioprio, "journal_ioprio=%u"},
1413         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1414         {Opt_auto_da_alloc, "auto_da_alloc"},
1415         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1416         {Opt_dioread_nolock, "dioread_nolock"},
1417         {Opt_dioread_lock, "dioread_lock"},
1418         {Opt_discard, "discard"},
1419         {Opt_nodiscard, "nodiscard"},
1420         {Opt_init_itable, "init_itable=%u"},
1421         {Opt_init_itable, "init_itable"},
1422         {Opt_noinit_itable, "noinit_itable"},
1423         {Opt_err, NULL},
1424 };
1425
1426 static ext4_fsblk_t get_sb_block(void **data)
1427 {
1428         ext4_fsblk_t    sb_block;
1429         char            *options = (char *) *data;
1430
1431         if (!options || strncmp(options, "sb=", 3) != 0)
1432                 return 1;       /* Default location */
1433
1434         options += 3;
1435         /* TODO: use simple_strtoll with >32bit ext4 */
1436         sb_block = simple_strtoul(options, &options, 0);
1437         if (*options && *options != ',') {
1438                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1439                        (char *) *data);
1440                 return 1;
1441         }
1442         if (*options == ',')
1443                 options++;
1444         *data = (void *) options;
1445
1446         return sb_block;
1447 }
1448
1449 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1450 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1451         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1452
1453 #ifdef CONFIG_QUOTA
1454 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1455 {
1456         struct ext4_sb_info *sbi = EXT4_SB(sb);
1457         char *qname;
1458
1459         if (sb_any_quota_loaded(sb) &&
1460                 !sbi->s_qf_names[qtype]) {
1461                 ext4_msg(sb, KERN_ERR,
1462                         "Cannot change journaled "
1463                         "quota options when quota turned on");
1464                 return 0;
1465         }
1466         qname = match_strdup(args);
1467         if (!qname) {
1468                 ext4_msg(sb, KERN_ERR,
1469                         "Not enough memory for storing quotafile name");
1470                 return 0;
1471         }
1472         if (sbi->s_qf_names[qtype] &&
1473                 strcmp(sbi->s_qf_names[qtype], qname)) {
1474                 ext4_msg(sb, KERN_ERR,
1475                         "%s quota file already specified", QTYPE2NAME(qtype));
1476                 kfree(qname);
1477                 return 0;
1478         }
1479         sbi->s_qf_names[qtype] = qname;
1480         if (strchr(sbi->s_qf_names[qtype], '/')) {
1481                 ext4_msg(sb, KERN_ERR,
1482                         "quotafile must be on filesystem root");
1483                 kfree(sbi->s_qf_names[qtype]);
1484                 sbi->s_qf_names[qtype] = NULL;
1485                 return 0;
1486         }
1487         set_opt(sb, QUOTA);
1488         return 1;
1489 }
1490
1491 static int clear_qf_name(struct super_block *sb, int qtype)
1492 {
1493
1494         struct ext4_sb_info *sbi = EXT4_SB(sb);
1495
1496         if (sb_any_quota_loaded(sb) &&
1497                 sbi->s_qf_names[qtype]) {
1498                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1499                         " when quota turned on");
1500                 return 0;
1501         }
1502         /*
1503          * The space will be released later when all options are confirmed
1504          * to be correct
1505          */
1506         sbi->s_qf_names[qtype] = NULL;
1507         return 1;
1508 }
1509 #endif
1510
1511 static int parse_options(char *options, struct super_block *sb,
1512                          unsigned long *journal_devnum,
1513                          unsigned int *journal_ioprio,
1514                          ext4_fsblk_t *n_blocks_count, int is_remount)
1515 {
1516         struct ext4_sb_info *sbi = EXT4_SB(sb);
1517         char *p;
1518         substring_t args[MAX_OPT_ARGS];
1519         int data_opt = 0;
1520         int option;
1521 #ifdef CONFIG_QUOTA
1522         int qfmt;
1523 #endif
1524
1525         if (!options)
1526                 return 1;
1527
1528         while ((p = strsep(&options, ",")) != NULL) {
1529                 int token;
1530                 if (!*p)
1531                         continue;
1532
1533                 /*
1534                  * Initialize args struct so we know whether arg was
1535                  * found; some options take optional arguments.
1536                  */
1537                 args[0].to = args[0].from = NULL;
1538                 token = match_token(p, tokens, args);
1539                 switch (token) {
1540                 case Opt_bsd_df:
1541                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1542                         clear_opt(sb, MINIX_DF);
1543                         break;
1544                 case Opt_minix_df:
1545                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1546                         set_opt(sb, MINIX_DF);
1547
1548                         break;
1549                 case Opt_grpid:
1550                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1551                         set_opt(sb, GRPID);
1552
1553                         break;
1554                 case Opt_nogrpid:
1555                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1556                         clear_opt(sb, GRPID);
1557
1558                         break;
1559                 case Opt_resuid:
1560                         if (match_int(&args[0], &option))
1561                                 return 0;
1562                         sbi->s_resuid = option;
1563                         break;
1564                 case Opt_resgid:
1565                         if (match_int(&args[0], &option))
1566                                 return 0;
1567                         sbi->s_resgid = option;
1568                         break;
1569                 case Opt_sb:
1570                         /* handled by get_sb_block() instead of here */
1571                         /* *sb_block = match_int(&args[0]); */
1572                         break;
1573                 case Opt_err_panic:
1574                         clear_opt(sb, ERRORS_CONT);
1575                         clear_opt(sb, ERRORS_RO);
1576                         set_opt(sb, ERRORS_PANIC);
1577                         break;
1578                 case Opt_err_ro:
1579                         clear_opt(sb, ERRORS_CONT);
1580                         clear_opt(sb, ERRORS_PANIC);
1581                         set_opt(sb, ERRORS_RO);
1582                         break;
1583                 case Opt_err_cont:
1584                         clear_opt(sb, ERRORS_RO);
1585                         clear_opt(sb, ERRORS_PANIC);
1586                         set_opt(sb, ERRORS_CONT);
1587                         break;
1588                 case Opt_nouid32:
1589                         set_opt(sb, NO_UID32);
1590                         break;
1591                 case Opt_debug:
1592                         set_opt(sb, DEBUG);
1593                         break;
1594                 case Opt_oldalloc:
1595                         ext4_msg(sb, KERN_WARNING,
1596                                  "Ignoring deprecated oldalloc option");
1597                         break;
1598                 case Opt_orlov:
1599                         ext4_msg(sb, KERN_WARNING,
1600                                  "Ignoring deprecated orlov option");
1601                         break;
1602 #ifdef CONFIG_EXT4_FS_XATTR
1603                 case Opt_user_xattr:
1604                         set_opt(sb, XATTR_USER);
1605                         break;
1606                 case Opt_nouser_xattr:
1607                         clear_opt(sb, XATTR_USER);
1608                         break;
1609 #else
1610                 case Opt_user_xattr:
1611                 case Opt_nouser_xattr:
1612                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1613                         break;
1614 #endif
1615 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1616                 case Opt_acl:
1617                         set_opt(sb, POSIX_ACL);
1618                         break;
1619                 case Opt_noacl:
1620                         clear_opt(sb, POSIX_ACL);
1621                         break;
1622 #else
1623                 case Opt_acl:
1624                 case Opt_noacl:
1625                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1626                         break;
1627 #endif
1628                 case Opt_journal_update:
1629                         /* @@@ FIXME */
1630                         /* Eventually we will want to be able to create
1631                            a journal file here.  For now, only allow the
1632                            user to specify an existing inode to be the
1633                            journal file. */
1634                         if (is_remount) {
1635                                 ext4_msg(sb, KERN_ERR,
1636                                          "Cannot specify journal on remount");
1637                                 return 0;
1638                         }
1639                         set_opt(sb, UPDATE_JOURNAL);
1640                         break;
1641                 case Opt_journal_dev:
1642                         if (is_remount) {
1643                                 ext4_msg(sb, KERN_ERR,
1644                                         "Cannot specify journal on remount");
1645                                 return 0;
1646                         }
1647                         if (match_int(&args[0], &option))
1648                                 return 0;
1649                         *journal_devnum = option;
1650                         break;
1651                 case Opt_journal_checksum:
1652                         set_opt(sb, JOURNAL_CHECKSUM);
1653                         break;
1654                 case Opt_journal_async_commit:
1655                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1656                         set_opt(sb, JOURNAL_CHECKSUM);
1657                         break;
1658                 case Opt_noload:
1659                         set_opt(sb, NOLOAD);
1660                         break;
1661                 case Opt_commit:
1662                         if (match_int(&args[0], &option))
1663                                 return 0;
1664                         if (option < 0)
1665                                 return 0;
1666                         if (option == 0)
1667                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1668                         sbi->s_commit_interval = HZ * option;
1669                         break;
1670                 case Opt_max_batch_time:
1671                         if (match_int(&args[0], &option))
1672                                 return 0;
1673                         if (option < 0)
1674                                 return 0;
1675                         sbi->s_max_batch_time = option;
1676                         break;
1677                 case Opt_min_batch_time:
1678                         if (match_int(&args[0], &option))
1679                                 return 0;
1680                         if (option < 0)
1681                                 return 0;
1682                         sbi->s_min_batch_time = option;
1683                         break;
1684                 case Opt_data_journal:
1685                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1686                         goto datacheck;
1687                 case Opt_data_ordered:
1688                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1689                         goto datacheck;
1690                 case Opt_data_writeback:
1691                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1692                 datacheck:
1693                         if (is_remount) {
1694                                 if (!sbi->s_journal)
1695                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1696                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1697                                         ext4_msg(sb, KERN_ERR,
1698                                                 "Cannot change data mode on remount");
1699                                         return 0;
1700                                 }
1701                         } else {
1702                                 clear_opt(sb, DATA_FLAGS);
1703                                 sbi->s_mount_opt |= data_opt;
1704                         }
1705                         break;
1706                 case Opt_data_err_abort:
1707                         set_opt(sb, DATA_ERR_ABORT);
1708                         break;
1709                 case Opt_data_err_ignore:
1710                         clear_opt(sb, DATA_ERR_ABORT);
1711                         break;
1712 #ifdef CONFIG_QUOTA
1713                 case Opt_usrjquota:
1714                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1715                                 return 0;
1716                         break;
1717                 case Opt_grpjquota:
1718                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1719                                 return 0;
1720                         break;
1721                 case Opt_offusrjquota:
1722                         if (!clear_qf_name(sb, USRQUOTA))
1723                                 return 0;
1724                         break;
1725                 case Opt_offgrpjquota:
1726                         if (!clear_qf_name(sb, GRPQUOTA))
1727                                 return 0;
1728                         break;
1729
1730                 case Opt_jqfmt_vfsold:
1731                         qfmt = QFMT_VFS_OLD;
1732                         goto set_qf_format;
1733                 case Opt_jqfmt_vfsv0:
1734                         qfmt = QFMT_VFS_V0;
1735                         goto set_qf_format;
1736                 case Opt_jqfmt_vfsv1:
1737                         qfmt = QFMT_VFS_V1;
1738 set_qf_format:
1739                         if (sb_any_quota_loaded(sb) &&
1740                             sbi->s_jquota_fmt != qfmt) {
1741                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1742                                         "journaled quota options when "
1743                                         "quota turned on");
1744                                 return 0;
1745                         }
1746                         sbi->s_jquota_fmt = qfmt;
1747                         break;
1748                 case Opt_quota:
1749                 case Opt_usrquota:
1750                         set_opt(sb, QUOTA);
1751                         set_opt(sb, USRQUOTA);
1752                         break;
1753                 case Opt_grpquota:
1754                         set_opt(sb, QUOTA);
1755                         set_opt(sb, GRPQUOTA);
1756                         break;
1757                 case Opt_noquota:
1758                         if (sb_any_quota_loaded(sb)) {
1759                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1760                                         "options when quota turned on");
1761                                 return 0;
1762                         }
1763                         clear_opt(sb, QUOTA);
1764                         clear_opt(sb, USRQUOTA);
1765                         clear_opt(sb, GRPQUOTA);
1766                         break;
1767 #else
1768                 case Opt_quota:
1769                 case Opt_usrquota:
1770                 case Opt_grpquota:
1771                         ext4_msg(sb, KERN_ERR,
1772                                 "quota options not supported");
1773                         break;
1774                 case Opt_usrjquota:
1775                 case Opt_grpjquota:
1776                 case Opt_offusrjquota:
1777                 case Opt_offgrpjquota:
1778                 case Opt_jqfmt_vfsold:
1779                 case Opt_jqfmt_vfsv0:
1780                 case Opt_jqfmt_vfsv1:
1781                         ext4_msg(sb, KERN_ERR,
1782                                 "journaled quota options not supported");
1783                         break;
1784                 case Opt_noquota:
1785                         break;
1786 #endif
1787                 case Opt_abort:
1788                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1789                         break;
1790                 case Opt_nobarrier:
1791                         clear_opt(sb, BARRIER);
1792                         break;
1793                 case Opt_barrier:
1794                         if (args[0].from) {
1795                                 if (match_int(&args[0], &option))
1796                                         return 0;
1797                         } else
1798                                 option = 1;     /* No argument, default to 1 */
1799                         if (option)
1800                                 set_opt(sb, BARRIER);
1801                         else
1802                                 clear_opt(sb, BARRIER);
1803                         break;
1804                 case Opt_ignore:
1805                         break;
1806                 case Opt_resize:
1807                         if (!is_remount) {
1808                                 ext4_msg(sb, KERN_ERR,
1809                                         "resize option only available "
1810                                         "for remount");
1811                                 return 0;
1812                         }
1813                         if (match_int(&args[0], &option) != 0)
1814                                 return 0;
1815                         *n_blocks_count = option;
1816                         break;
1817                 case Opt_nobh:
1818                         ext4_msg(sb, KERN_WARNING,
1819                                  "Ignoring deprecated nobh option");
1820                         break;
1821                 case Opt_bh:
1822                         ext4_msg(sb, KERN_WARNING,
1823                                  "Ignoring deprecated bh option");
1824                         break;
1825                 case Opt_i_version:
1826                         set_opt(sb, I_VERSION);
1827                         sb->s_flags |= MS_I_VERSION;
1828                         break;
1829                 case Opt_nodelalloc:
1830                         clear_opt(sb, DELALLOC);
1831                         clear_opt2(sb, EXPLICIT_DELALLOC);
1832                         break;
1833                 case Opt_mblk_io_submit:
1834                         set_opt(sb, MBLK_IO_SUBMIT);
1835                         break;
1836                 case Opt_nomblk_io_submit:
1837                         clear_opt(sb, MBLK_IO_SUBMIT);
1838                         break;
1839                 case Opt_stripe:
1840                         if (match_int(&args[0], &option))
1841                                 return 0;
1842                         if (option < 0)
1843                                 return 0;
1844                         sbi->s_stripe = option;
1845                         break;
1846                 case Opt_delalloc:
1847                         set_opt(sb, DELALLOC);
1848                         set_opt2(sb, EXPLICIT_DELALLOC);
1849                         break;
1850                 case Opt_block_validity:
1851                         set_opt(sb, BLOCK_VALIDITY);
1852                         break;
1853                 case Opt_noblock_validity:
1854                         clear_opt(sb, BLOCK_VALIDITY);
1855                         break;
1856                 case Opt_inode_readahead_blks:
1857                         if (match_int(&args[0], &option))
1858                                 return 0;
1859                         if (option < 0 || option > (1 << 30))
1860                                 return 0;
1861                         if (option && !is_power_of_2(option)) {
1862                                 ext4_msg(sb, KERN_ERR,
1863                                          "EXT4-fs: inode_readahead_blks"
1864                                          " must be a power of 2");
1865                                 return 0;
1866                         }
1867                         sbi->s_inode_readahead_blks = option;
1868                         break;
1869                 case Opt_journal_ioprio:
1870                         if (match_int(&args[0], &option))
1871                                 return 0;
1872                         if (option < 0 || option > 7)
1873                                 break;
1874                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1875                                                             option);
1876                         break;
1877                 case Opt_noauto_da_alloc:
1878                         set_opt(sb, NO_AUTO_DA_ALLOC);
1879                         break;
1880                 case Opt_auto_da_alloc:
1881                         if (args[0].from) {
1882                                 if (match_int(&args[0], &option))
1883                                         return 0;
1884                         } else
1885                                 option = 1;     /* No argument, default to 1 */
1886                         if (option)
1887                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1888                         else
1889                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1890                         break;
1891                 case Opt_discard:
1892                         set_opt(sb, DISCARD);
1893                         break;
1894                 case Opt_nodiscard:
1895                         clear_opt(sb, DISCARD);
1896                         break;
1897                 case Opt_dioread_nolock:
1898                         set_opt(sb, DIOREAD_NOLOCK);
1899                         break;
1900                 case Opt_dioread_lock:
1901                         clear_opt(sb, DIOREAD_NOLOCK);
1902                         break;
1903                 case Opt_init_itable:
1904                         set_opt(sb, INIT_INODE_TABLE);
1905                         if (args[0].from) {
1906                                 if (match_int(&args[0], &option))
1907                                         return 0;
1908                         } else
1909                                 option = EXT4_DEF_LI_WAIT_MULT;
1910                         if (option < 0)
1911                                 return 0;
1912                         sbi->s_li_wait_mult = option;
1913                         break;
1914                 case Opt_noinit_itable:
1915                         clear_opt(sb, INIT_INODE_TABLE);
1916                         break;
1917                 default:
1918                         ext4_msg(sb, KERN_ERR,
1919                                "Unrecognized mount option \"%s\" "
1920                                "or missing value", p);
1921                         return 0;
1922                 }
1923         }
1924 #ifdef CONFIG_QUOTA
1925         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1926                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1927                         clear_opt(sb, USRQUOTA);
1928
1929                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1930                         clear_opt(sb, GRPQUOTA);
1931
1932                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1933                         ext4_msg(sb, KERN_ERR, "old and new quota "
1934                                         "format mixing");
1935                         return 0;
1936                 }
1937
1938                 if (!sbi->s_jquota_fmt) {
1939                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1940                                         "not specified");
1941                         return 0;
1942                 }
1943         }
1944 #endif
1945         if (test_opt(sb, DIOREAD_NOLOCK)) {
1946                 int blocksize =
1947                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1948
1949                 if (blocksize < PAGE_CACHE_SIZE) {
1950                         ext4_msg(sb, KERN_ERR, "can't mount with "
1951                                  "dioread_nolock if block size != PAGE_SIZE");
1952                         return 0;
1953                 }
1954         }
1955         return 1;
1956 }
1957
1958 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1959                             int read_only)
1960 {
1961         struct ext4_sb_info *sbi = EXT4_SB(sb);
1962         int res = 0;
1963
1964         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1965                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1966                          "forcing read-only mode");
1967                 res = MS_RDONLY;
1968         }
1969         if (read_only)
1970                 goto done;
1971         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1972                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1973                          "running e2fsck is recommended");
1974         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1975                 ext4_msg(sb, KERN_WARNING,
1976                          "warning: mounting fs with errors, "
1977                          "running e2fsck is recommended");
1978         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1979                  le16_to_cpu(es->s_mnt_count) >=
1980                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1981                 ext4_msg(sb, KERN_WARNING,
1982                          "warning: maximal mount count reached, "
1983                          "running e2fsck is recommended");
1984         else if (le32_to_cpu(es->s_checkinterval) &&
1985                 (le32_to_cpu(es->s_lastcheck) +
1986                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1987                 ext4_msg(sb, KERN_WARNING,
1988                          "warning: checktime reached, "
1989                          "running e2fsck is recommended");
1990         if (!sbi->s_journal)
1991                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1992         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1993                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1994         le16_add_cpu(&es->s_mnt_count, 1);
1995         es->s_mtime = cpu_to_le32(get_seconds());
1996         ext4_update_dynamic_rev(sb);
1997         if (sbi->s_journal)
1998                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1999
2000         ext4_commit_super(sb, 1);
2001 done:
2002         if (test_opt(sb, DEBUG))
2003                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2004                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2005                         sb->s_blocksize,
2006                         sbi->s_groups_count,
2007                         EXT4_BLOCKS_PER_GROUP(sb),
2008                         EXT4_INODES_PER_GROUP(sb),
2009                         sbi->s_mount_opt, sbi->s_mount_opt2);
2010
2011         cleancache_init_fs(sb);
2012         return res;
2013 }
2014
2015 static int ext4_fill_flex_info(struct super_block *sb)
2016 {
2017         struct ext4_sb_info *sbi = EXT4_SB(sb);
2018         struct ext4_group_desc *gdp = NULL;
2019         ext4_group_t flex_group_count;
2020         ext4_group_t flex_group;
2021         unsigned int groups_per_flex = 0;
2022         size_t size;
2023         int i;
2024
2025         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2026         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2027                 sbi->s_log_groups_per_flex = 0;
2028                 return 1;
2029         }
2030         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2031
2032         /* We allocate both existing and potentially added groups */
2033         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2034                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2035                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2036         size = flex_group_count * sizeof(struct flex_groups);
2037         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2038         if (sbi->s_flex_groups == NULL) {
2039                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2040                          flex_group_count);
2041                 goto failed;
2042         }
2043
2044         for (i = 0; i < sbi->s_groups_count; i++) {
2045                 gdp = ext4_get_group_desc(sb, i, NULL);
2046
2047                 flex_group = ext4_flex_group(sbi, i);
2048                 atomic_add(ext4_free_inodes_count(sb, gdp),
2049                            &sbi->s_flex_groups[flex_group].free_inodes);
2050                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2051                              &sbi->s_flex_groups[flex_group].free_clusters);
2052                 atomic_add(ext4_used_dirs_count(sb, gdp),
2053                            &sbi->s_flex_groups[flex_group].used_dirs);
2054         }
2055
2056         return 1;
2057 failed:
2058         return 0;
2059 }
2060
2061 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2062                             struct ext4_group_desc *gdp)
2063 {
2064         __u16 crc = 0;
2065
2066         if (sbi->s_es->s_feature_ro_compat &
2067             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2068                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2069                 __le32 le_group = cpu_to_le32(block_group);
2070
2071                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2072                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2073                 crc = crc16(crc, (__u8 *)gdp, offset);
2074                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2075                 /* for checksum of struct ext4_group_desc do the rest...*/
2076                 if ((sbi->s_es->s_feature_incompat &
2077                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2078                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2079                         crc = crc16(crc, (__u8 *)gdp + offset,
2080                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2081                                         offset);
2082         }
2083
2084         return cpu_to_le16(crc);
2085 }
2086
2087 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2088                                 struct ext4_group_desc *gdp)
2089 {
2090         if ((sbi->s_es->s_feature_ro_compat &
2091              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2092             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2093                 return 0;
2094
2095         return 1;
2096 }
2097
2098 /* Called at mount-time, super-block is locked */
2099 static int ext4_check_descriptors(struct super_block *sb,
2100                                   ext4_group_t *first_not_zeroed)
2101 {
2102         struct ext4_sb_info *sbi = EXT4_SB(sb);
2103         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2104         ext4_fsblk_t last_block;
2105         ext4_fsblk_t block_bitmap;
2106         ext4_fsblk_t inode_bitmap;
2107         ext4_fsblk_t inode_table;
2108         int flexbg_flag = 0;
2109         ext4_group_t i, grp = sbi->s_groups_count;
2110
2111         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2112                 flexbg_flag = 1;
2113
2114         ext4_debug("Checking group descriptors");
2115
2116         for (i = 0; i < sbi->s_groups_count; i++) {
2117                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2118
2119                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2120                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2121                 else
2122                         last_block = first_block +
2123                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2124
2125                 if ((grp == sbi->s_groups_count) &&
2126                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2127                         grp = i;
2128
2129                 block_bitmap = ext4_block_bitmap(sb, gdp);
2130                 if (block_bitmap < first_block || block_bitmap > last_block) {
2131                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2132                                "Block bitmap for group %u not in group "
2133                                "(block %llu)!", i, block_bitmap);
2134                         return 0;
2135                 }
2136                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2137                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2138                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2139                                "Inode bitmap for group %u not in group "
2140                                "(block %llu)!", i, inode_bitmap);
2141                         return 0;
2142                 }
2143                 inode_table = ext4_inode_table(sb, gdp);
2144                 if (inode_table < first_block ||
2145                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                "Inode table for group %u not in group "
2148                                "(block %llu)!", i, inode_table);
2149                         return 0;
2150                 }
2151                 ext4_lock_group(sb, i);
2152                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2153                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2154                                  "Checksum for group %u failed (%u!=%u)",
2155                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2156                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2157                         if (!(sb->s_flags & MS_RDONLY)) {
2158                                 ext4_unlock_group(sb, i);
2159                                 return 0;
2160                         }
2161                 }
2162                 ext4_unlock_group(sb, i);
2163                 if (!flexbg_flag)
2164                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2165         }
2166         if (NULL != first_not_zeroed)
2167                 *first_not_zeroed = grp;
2168
2169         ext4_free_blocks_count_set(sbi->s_es,
2170                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2171         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2172         return 1;
2173 }
2174
2175 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2176  * the superblock) which were deleted from all directories, but held open by
2177  * a process at the time of a crash.  We walk the list and try to delete these
2178  * inodes at recovery time (only with a read-write filesystem).
2179  *
2180  * In order to keep the orphan inode chain consistent during traversal (in
2181  * case of crash during recovery), we link each inode into the superblock
2182  * orphan list_head and handle it the same way as an inode deletion during
2183  * normal operation (which journals the operations for us).
2184  *
2185  * We only do an iget() and an iput() on each inode, which is very safe if we
2186  * accidentally point at an in-use or already deleted inode.  The worst that
2187  * can happen in this case is that we get a "bit already cleared" message from
2188  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2189  * e2fsck was run on this filesystem, and it must have already done the orphan
2190  * inode cleanup for us, so we can safely abort without any further action.
2191  */
2192 static void ext4_orphan_cleanup(struct super_block *sb,
2193                                 struct ext4_super_block *es)
2194 {
2195         unsigned int s_flags = sb->s_flags;
2196         int nr_orphans = 0, nr_truncates = 0;
2197 #ifdef CONFIG_QUOTA
2198         int i;
2199 #endif
2200         if (!es->s_last_orphan) {
2201                 jbd_debug(4, "no orphan inodes to clean up\n");
2202                 return;
2203         }
2204
2205         if (bdev_read_only(sb->s_bdev)) {
2206                 ext4_msg(sb, KERN_ERR, "write access "
2207                         "unavailable, skipping orphan cleanup");
2208                 return;
2209         }
2210
2211         /* Check if feature set would not allow a r/w mount */
2212         if (!ext4_feature_set_ok(sb, 0)) {
2213                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2214                          "unknown ROCOMPAT features");
2215                 return;
2216         }
2217
2218         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2219                 if (es->s_last_orphan)
2220                         jbd_debug(1, "Errors on filesystem, "
2221                                   "clearing orphan list.\n");
2222                 es->s_last_orphan = 0;
2223                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2224                 return;
2225         }
2226
2227         if (s_flags & MS_RDONLY) {
2228                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2229                 sb->s_flags &= ~MS_RDONLY;
2230         }
2231 #ifdef CONFIG_QUOTA
2232         /* Needed for iput() to work correctly and not trash data */
2233         sb->s_flags |= MS_ACTIVE;
2234         /* Turn on quotas so that they are updated correctly */
2235         for (i = 0; i < MAXQUOTAS; i++) {
2236                 if (EXT4_SB(sb)->s_qf_names[i]) {
2237                         int ret = ext4_quota_on_mount(sb, i);
2238                         if (ret < 0)
2239                                 ext4_msg(sb, KERN_ERR,
2240                                         "Cannot turn on journaled "
2241                                         "quota: error %d", ret);
2242                 }
2243         }
2244 #endif
2245
2246         while (es->s_last_orphan) {
2247                 struct inode *inode;
2248
2249                 /*
2250                  * We may have encountered an error during cleanup; if
2251                  * so, skip the rest.
2252                  */
2253                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2254                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2255                         es->s_last_orphan = 0;
2256                         break;
2257                 }
2258
2259                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2260                 if (IS_ERR(inode)) {
2261                         es->s_last_orphan = 0;
2262                         break;
2263                 }
2264
2265                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2266                 dquot_initialize(inode);
2267                 if (inode->i_nlink) {
2268                         ext4_msg(sb, KERN_DEBUG,
2269                                 "%s: truncating inode %lu to %lld bytes",
2270                                 __func__, inode->i_ino, inode->i_size);
2271                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2272                                   inode->i_ino, inode->i_size);
2273                         mutex_lock(&inode->i_mutex);
2274                         ext4_truncate(inode);
2275                         mutex_unlock(&inode->i_mutex);
2276                         nr_truncates++;
2277                 } else {
2278                         ext4_msg(sb, KERN_DEBUG,
2279                                 "%s: deleting unreferenced inode %lu",
2280                                 __func__, inode->i_ino);
2281                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2282                                   inode->i_ino);
2283                         nr_orphans++;
2284                 }
2285                 iput(inode);  /* The delete magic happens here! */
2286         }
2287
2288 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2289
2290         if (nr_orphans)
2291                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2292                        PLURAL(nr_orphans));
2293         if (nr_truncates)
2294                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2295                        PLURAL(nr_truncates));
2296 #ifdef CONFIG_QUOTA
2297         /* Turn quotas off */
2298         for (i = 0; i < MAXQUOTAS; i++) {
2299                 if (sb_dqopt(sb)->files[i])
2300                         dquot_quota_off(sb, i);
2301         }
2302 #endif
2303         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2304 }
2305
2306 /*
2307  * Maximal extent format file size.
2308  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2309  * extent format containers, within a sector_t, and within i_blocks
2310  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2311  * so that won't be a limiting factor.
2312  *
2313  * However there is other limiting factor. We do store extents in the form
2314  * of starting block and length, hence the resulting length of the extent
2315  * covering maximum file size must fit into on-disk format containers as
2316  * well. Given that length is always by 1 unit bigger than max unit (because
2317  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2318  *
2319  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2320  */
2321 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2322 {
2323         loff_t res;
2324         loff_t upper_limit = MAX_LFS_FILESIZE;
2325
2326         /* small i_blocks in vfs inode? */
2327         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2328                 /*
2329                  * CONFIG_LBDAF is not enabled implies the inode
2330                  * i_block represent total blocks in 512 bytes
2331                  * 32 == size of vfs inode i_blocks * 8
2332                  */
2333                 upper_limit = (1LL << 32) - 1;
2334
2335                 /* total blocks in file system block size */
2336                 upper_limit >>= (blkbits - 9);
2337                 upper_limit <<= blkbits;
2338         }
2339
2340         /*
2341          * 32-bit extent-start container, ee_block. We lower the maxbytes
2342          * by one fs block, so ee_len can cover the extent of maximum file
2343          * size
2344          */
2345         res = (1LL << 32) - 1;
2346         res <<= blkbits;
2347
2348         /* Sanity check against vm- & vfs- imposed limits */
2349         if (res > upper_limit)
2350                 res = upper_limit;
2351
2352         return res;
2353 }
2354
2355 /*
2356  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2357  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2358  * We need to be 1 filesystem block less than the 2^48 sector limit.
2359  */
2360 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2361 {
2362         loff_t res = EXT4_NDIR_BLOCKS;
2363         int meta_blocks;
2364         loff_t upper_limit;
2365         /* This is calculated to be the largest file size for a dense, block
2366          * mapped file such that the file's total number of 512-byte sectors,
2367          * including data and all indirect blocks, does not exceed (2^48 - 1).
2368          *
2369          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2370          * number of 512-byte sectors of the file.
2371          */
2372
2373         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2374                 /*
2375                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2376                  * the inode i_block field represents total file blocks in
2377                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2378                  */
2379                 upper_limit = (1LL << 32) - 1;
2380
2381                 /* total blocks in file system block size */
2382                 upper_limit >>= (bits - 9);
2383
2384         } else {
2385                 /*
2386                  * We use 48 bit ext4_inode i_blocks
2387                  * With EXT4_HUGE_FILE_FL set the i_blocks
2388                  * represent total number of blocks in
2389                  * file system block size
2390                  */
2391                 upper_limit = (1LL << 48) - 1;
2392
2393         }
2394
2395         /* indirect blocks */
2396         meta_blocks = 1;
2397         /* double indirect blocks */
2398         meta_blocks += 1 + (1LL << (bits-2));
2399         /* tripple indirect blocks */
2400         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2401
2402         upper_limit -= meta_blocks;
2403         upper_limit <<= bits;
2404
2405         res += 1LL << (bits-2);
2406         res += 1LL << (2*(bits-2));
2407         res += 1LL << (3*(bits-2));
2408         res <<= bits;
2409         if (res > upper_limit)
2410                 res = upper_limit;
2411
2412         if (res > MAX_LFS_FILESIZE)
2413                 res = MAX_LFS_FILESIZE;
2414
2415         return res;
2416 }
2417
2418 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2419                                    ext4_fsblk_t logical_sb_block, int nr)
2420 {
2421         struct ext4_sb_info *sbi = EXT4_SB(sb);
2422         ext4_group_t bg, first_meta_bg;
2423         int has_super = 0;
2424
2425         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2426
2427         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2428             nr < first_meta_bg)
2429                 return logical_sb_block + nr + 1;
2430         bg = sbi->s_desc_per_block * nr;
2431         if (ext4_bg_has_super(sb, bg))
2432                 has_super = 1;
2433
2434         return (has_super + ext4_group_first_block_no(sb, bg));
2435 }
2436
2437 /**
2438  * ext4_get_stripe_size: Get the stripe size.
2439  * @sbi: In memory super block info
2440  *
2441  * If we have specified it via mount option, then
2442  * use the mount option value. If the value specified at mount time is
2443  * greater than the blocks per group use the super block value.
2444  * If the super block value is greater than blocks per group return 0.
2445  * Allocator needs it be less than blocks per group.
2446  *
2447  */
2448 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2449 {
2450         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2451         unsigned long stripe_width =
2452                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2453         int ret;
2454
2455         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2456                 ret = sbi->s_stripe;
2457         else if (stripe_width <= sbi->s_blocks_per_group)
2458                 ret = stripe_width;
2459         else if (stride <= sbi->s_blocks_per_group)
2460                 ret = stride;
2461         else
2462                 ret = 0;
2463
2464         /*
2465          * If the stripe width is 1, this makes no sense and
2466          * we set it to 0 to turn off stripe handling code.
2467          */
2468         if (ret <= 1)
2469                 ret = 0;
2470
2471         return ret;
2472 }
2473
2474 /* sysfs supprt */
2475
2476 struct ext4_attr {
2477         struct attribute attr;
2478         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2479         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2480                          const char *, size_t);
2481         int offset;
2482 };
2483
2484 static int parse_strtoul(const char *buf,
2485                 unsigned long max, unsigned long *value)
2486 {
2487         char *endp;
2488
2489         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2490         endp = skip_spaces(endp);
2491         if (*endp || *value > max)
2492                 return -EINVAL;
2493
2494         return 0;
2495 }
2496
2497 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2498                                               struct ext4_sb_info *sbi,
2499                                               char *buf)
2500 {
2501         return snprintf(buf, PAGE_SIZE, "%llu\n",
2502                 (s64) EXT4_C2B(sbi,
2503                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2504 }
2505
2506 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2507                                          struct ext4_sb_info *sbi, char *buf)
2508 {
2509         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2510
2511         if (!sb->s_bdev->bd_part)
2512                 return snprintf(buf, PAGE_SIZE, "0\n");
2513         return snprintf(buf, PAGE_SIZE, "%lu\n",
2514                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2515                          sbi->s_sectors_written_start) >> 1);
2516 }
2517
2518 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2519                                           struct ext4_sb_info *sbi, char *buf)
2520 {
2521         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2522
2523         if (!sb->s_bdev->bd_part)
2524                 return snprintf(buf, PAGE_SIZE, "0\n");
2525         return snprintf(buf, PAGE_SIZE, "%llu\n",
2526                         (unsigned long long)(sbi->s_kbytes_written +
2527                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2528                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2529 }
2530
2531 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2532                                           struct ext4_sb_info *sbi,
2533                                           const char *buf, size_t count)
2534 {
2535         unsigned long t;
2536
2537         if (parse_strtoul(buf, 0x40000000, &t))
2538                 return -EINVAL;
2539
2540         if (t && !is_power_of_2(t))
2541                 return -EINVAL;
2542
2543         sbi->s_inode_readahead_blks = t;
2544         return count;
2545 }
2546
2547 static ssize_t sbi_ui_show(struct ext4_attr *a,
2548                            struct ext4_sb_info *sbi, char *buf)
2549 {
2550         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2551
2552         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2553 }
2554
2555 static ssize_t sbi_ui_store(struct ext4_attr *a,
2556                             struct ext4_sb_info *sbi,
2557                             const char *buf, size_t count)
2558 {
2559         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2560         unsigned long t;
2561
2562         if (parse_strtoul(buf, 0xffffffff, &t))
2563                 return -EINVAL;
2564         *ui = t;
2565         return count;
2566 }
2567
2568 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2569 static struct ext4_attr ext4_attr_##_name = {                   \
2570         .attr = {.name = __stringify(_name), .mode = _mode },   \
2571         .show   = _show,                                        \
2572         .store  = _store,                                       \
2573         .offset = offsetof(struct ext4_sb_info, _elname),       \
2574 }
2575 #define EXT4_ATTR(name, mode, show, store) \
2576 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2577
2578 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2579 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2580 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2581 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2582         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2583 #define ATTR_LIST(name) &ext4_attr_##name.attr
2584
2585 EXT4_RO_ATTR(delayed_allocation_blocks);
2586 EXT4_RO_ATTR(session_write_kbytes);
2587 EXT4_RO_ATTR(lifetime_write_kbytes);
2588 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2589                  inode_readahead_blks_store, s_inode_readahead_blks);
2590 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2591 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2592 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2593 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2594 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2595 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2596 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2597 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2598
2599 static struct attribute *ext4_attrs[] = {
2600         ATTR_LIST(delayed_allocation_blocks),
2601         ATTR_LIST(session_write_kbytes),
2602         ATTR_LIST(lifetime_write_kbytes),
2603         ATTR_LIST(inode_readahead_blks),
2604         ATTR_LIST(inode_goal),
2605         ATTR_LIST(mb_stats),
2606         ATTR_LIST(mb_max_to_scan),
2607         ATTR_LIST(mb_min_to_scan),
2608         ATTR_LIST(mb_order2_req),
2609         ATTR_LIST(mb_stream_req),
2610         ATTR_LIST(mb_group_prealloc),
2611         ATTR_LIST(max_writeback_mb_bump),
2612         NULL,
2613 };
2614
2615 /* Features this copy of ext4 supports */
2616 EXT4_INFO_ATTR(lazy_itable_init);
2617 EXT4_INFO_ATTR(batched_discard);
2618
2619 static struct attribute *ext4_feat_attrs[] = {
2620         ATTR_LIST(lazy_itable_init),
2621         ATTR_LIST(batched_discard),
2622         NULL,
2623 };
2624
2625 static ssize_t ext4_attr_show(struct kobject *kobj,
2626                               struct attribute *attr, char *buf)
2627 {
2628         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2629                                                 s_kobj);
2630         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2631
2632         return a->show ? a->show(a, sbi, buf) : 0;
2633 }
2634
2635 static ssize_t ext4_attr_store(struct kobject *kobj,
2636                                struct attribute *attr,
2637                                const char *buf, size_t len)
2638 {
2639         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2640                                                 s_kobj);
2641         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2642
2643         return a->store ? a->store(a, sbi, buf, len) : 0;
2644 }
2645
2646 static void ext4_sb_release(struct kobject *kobj)
2647 {
2648         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2649                                                 s_kobj);
2650         complete(&sbi->s_kobj_unregister);
2651 }
2652
2653 static const struct sysfs_ops ext4_attr_ops = {
2654         .show   = ext4_attr_show,
2655         .store  = ext4_attr_store,
2656 };
2657
2658 static struct kobj_type ext4_ktype = {
2659         .default_attrs  = ext4_attrs,
2660         .sysfs_ops      = &ext4_attr_ops,
2661         .release        = ext4_sb_release,
2662 };
2663
2664 static void ext4_feat_release(struct kobject *kobj)
2665 {
2666         complete(&ext4_feat->f_kobj_unregister);
2667 }
2668
2669 static struct kobj_type ext4_feat_ktype = {
2670         .default_attrs  = ext4_feat_attrs,
2671         .sysfs_ops      = &ext4_attr_ops,
2672         .release        = ext4_feat_release,
2673 };
2674
2675 /*
2676  * Check whether this filesystem can be mounted based on
2677  * the features present and the RDONLY/RDWR mount requested.
2678  * Returns 1 if this filesystem can be mounted as requested,
2679  * 0 if it cannot be.
2680  */
2681 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2682 {
2683         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2684                 ext4_msg(sb, KERN_ERR,
2685                         "Couldn't mount because of "
2686                         "unsupported optional features (%x)",
2687                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2688                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2689                 return 0;
2690         }
2691
2692         if (readonly)
2693                 return 1;
2694
2695         /* Check that feature set is OK for a read-write mount */
2696         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2697                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2698                          "unsupported optional features (%x)",
2699                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2700                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2701                 return 0;
2702         }
2703         /*
2704          * Large file size enabled file system can only be mounted
2705          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2706          */
2707         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2708                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2709                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2710                                  "cannot be mounted RDWR without "
2711                                  "CONFIG_LBDAF");
2712                         return 0;
2713                 }
2714         }
2715         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2716             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2717                 ext4_msg(sb, KERN_ERR,
2718                          "Can't support bigalloc feature without "
2719                          "extents feature\n");
2720                 return 0;
2721         }
2722         return 1;
2723 }
2724
2725 /*
2726  * This function is called once a day if we have errors logged
2727  * on the file system
2728  */
2729 static void print_daily_error_info(unsigned long arg)
2730 {
2731         struct super_block *sb = (struct super_block *) arg;
2732         struct ext4_sb_info *sbi;
2733         struct ext4_super_block *es;
2734
2735         sbi = EXT4_SB(sb);
2736         es = sbi->s_es;
2737
2738         if (es->s_error_count)
2739                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2740                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2741                          le32_to_cpu(es->s_error_count));
2742         if (es->s_first_error_time) {
2743                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2744                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2745                        (int) sizeof(es->s_first_error_func),
2746                        es->s_first_error_func,
2747                        le32_to_cpu(es->s_first_error_line));
2748                 if (es->s_first_error_ino)
2749                         printk(": inode %u",
2750                                le32_to_cpu(es->s_first_error_ino));
2751                 if (es->s_first_error_block)
2752                         printk(": block %llu", (unsigned long long)
2753                                le64_to_cpu(es->s_first_error_block));
2754                 printk("\n");
2755         }
2756         if (es->s_last_error_time) {
2757                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2758                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2759                        (int) sizeof(es->s_last_error_func),
2760                        es->s_last_error_func,
2761                        le32_to_cpu(es->s_last_error_line));
2762                 if (es->s_last_error_ino)
2763                         printk(": inode %u",
2764                                le32_to_cpu(es->s_last_error_ino));
2765                 if (es->s_last_error_block)
2766                         printk(": block %llu", (unsigned long long)
2767                                le64_to_cpu(es->s_last_error_block));
2768                 printk("\n");
2769         }
2770         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2771 }
2772
2773 /* Find next suitable group and run ext4_init_inode_table */
2774 static int ext4_run_li_request(struct ext4_li_request *elr)
2775 {
2776         struct ext4_group_desc *gdp = NULL;
2777         ext4_group_t group, ngroups;
2778         struct super_block *sb;
2779         unsigned long timeout = 0;
2780         int ret = 0;
2781
2782         sb = elr->lr_super;
2783         ngroups = EXT4_SB(sb)->s_groups_count;
2784
2785         for (group = elr->lr_next_group; group < ngroups; group++) {
2786                 gdp = ext4_get_group_desc(sb, group, NULL);
2787                 if (!gdp) {
2788                         ret = 1;
2789                         break;
2790                 }
2791
2792                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2793                         break;
2794         }
2795
2796         if (group == ngroups)
2797                 ret = 1;
2798
2799         if (!ret) {
2800                 timeout = jiffies;
2801                 ret = ext4_init_inode_table(sb, group,
2802                                             elr->lr_timeout ? 0 : 1);
2803                 if (elr->lr_timeout == 0) {
2804                         timeout = (jiffies - timeout) *
2805                                   elr->lr_sbi->s_li_wait_mult;
2806                         elr->lr_timeout = timeout;
2807                 }
2808                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2809                 elr->lr_next_group = group + 1;
2810         }
2811
2812         return ret;
2813 }
2814
2815 /*
2816  * Remove lr_request from the list_request and free the
2817  * request structure. Should be called with li_list_mtx held
2818  */
2819 static void ext4_remove_li_request(struct ext4_li_request *elr)
2820 {
2821         struct ext4_sb_info *sbi;
2822
2823         if (!elr)
2824                 return;
2825
2826         sbi = elr->lr_sbi;
2827
2828         list_del(&elr->lr_request);
2829         sbi->s_li_request = NULL;
2830         kfree(elr);
2831 }
2832
2833 static void ext4_unregister_li_request(struct super_block *sb)
2834 {
2835         mutex_lock(&ext4_li_mtx);
2836         if (!ext4_li_info) {
2837                 mutex_unlock(&ext4_li_mtx);
2838                 return;
2839         }
2840
2841         mutex_lock(&ext4_li_info->li_list_mtx);
2842         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2843         mutex_unlock(&ext4_li_info->li_list_mtx);
2844         mutex_unlock(&ext4_li_mtx);
2845 }
2846
2847 static struct task_struct *ext4_lazyinit_task;
2848
2849 /*
2850  * This is the function where ext4lazyinit thread lives. It walks
2851  * through the request list searching for next scheduled filesystem.
2852  * When such a fs is found, run the lazy initialization request
2853  * (ext4_rn_li_request) and keep track of the time spend in this
2854  * function. Based on that time we compute next schedule time of
2855  * the request. When walking through the list is complete, compute
2856  * next waking time and put itself into sleep.
2857  */
2858 static int ext4_lazyinit_thread(void *arg)
2859 {
2860         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2861         struct list_head *pos, *n;
2862         struct ext4_li_request *elr;
2863         unsigned long next_wakeup, cur;
2864
2865         BUG_ON(NULL == eli);
2866
2867 cont_thread:
2868         while (true) {
2869                 next_wakeup = MAX_JIFFY_OFFSET;
2870
2871                 mutex_lock(&eli->li_list_mtx);
2872                 if (list_empty(&eli->li_request_list)) {
2873                         mutex_unlock(&eli->li_list_mtx);
2874                         goto exit_thread;
2875                 }
2876
2877                 list_for_each_safe(pos, n, &eli->li_request_list) {
2878                         elr = list_entry(pos, struct ext4_li_request,
2879                                          lr_request);
2880
2881                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2882                                 if (ext4_run_li_request(elr) != 0) {
2883                                         /* error, remove the lazy_init job */
2884                                         ext4_remove_li_request(elr);
2885                                         continue;
2886                                 }
2887                         }
2888
2889                         if (time_before(elr->lr_next_sched, next_wakeup))
2890                                 next_wakeup = elr->lr_next_sched;
2891                 }
2892                 mutex_unlock(&eli->li_list_mtx);
2893
2894                 if (freezing(current))
2895                         refrigerator();
2896
2897                 cur = jiffies;
2898                 if ((time_after_eq(cur, next_wakeup)) ||
2899                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2900                         cond_resched();
2901                         continue;
2902                 }
2903
2904                 schedule_timeout_interruptible(next_wakeup - cur);
2905
2906                 if (kthread_should_stop()) {
2907                         ext4_clear_request_list();
2908                         goto exit_thread;
2909                 }
2910         }
2911
2912 exit_thread:
2913         /*
2914          * It looks like the request list is empty, but we need
2915          * to check it under the li_list_mtx lock, to prevent any
2916          * additions into it, and of course we should lock ext4_li_mtx
2917          * to atomically free the list and ext4_li_info, because at
2918          * this point another ext4 filesystem could be registering
2919          * new one.
2920          */
2921         mutex_lock(&ext4_li_mtx);
2922         mutex_lock(&eli->li_list_mtx);
2923         if (!list_empty(&eli->li_request_list)) {
2924                 mutex_unlock(&eli->li_list_mtx);
2925                 mutex_unlock(&ext4_li_mtx);
2926                 goto cont_thread;
2927         }
2928         mutex_unlock(&eli->li_list_mtx);
2929         kfree(ext4_li_info);
2930         ext4_li_info = NULL;
2931         mutex_unlock(&ext4_li_mtx);
2932
2933         return 0;
2934 }
2935
2936 static void ext4_clear_request_list(void)
2937 {
2938         struct list_head *pos, *n;
2939         struct ext4_li_request *elr;
2940
2941         mutex_lock(&ext4_li_info->li_list_mtx);
2942         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2943                 elr = list_entry(pos, struct ext4_li_request,
2944                                  lr_request);
2945                 ext4_remove_li_request(elr);
2946         }
2947         mutex_unlock(&ext4_li_info->li_list_mtx);
2948 }
2949
2950 static int ext4_run_lazyinit_thread(void)
2951 {
2952         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2953                                          ext4_li_info, "ext4lazyinit");
2954         if (IS_ERR(ext4_lazyinit_task)) {
2955                 int err = PTR_ERR(ext4_lazyinit_task);
2956                 ext4_clear_request_list();
2957                 kfree(ext4_li_info);
2958                 ext4_li_info = NULL;
2959                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2960                                  "initialization thread\n",
2961                                  err);
2962                 return err;
2963         }
2964         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2965         return 0;
2966 }
2967
2968 /*
2969  * Check whether it make sense to run itable init. thread or not.
2970  * If there is at least one uninitialized inode table, return
2971  * corresponding group number, else the loop goes through all
2972  * groups and return total number of groups.
2973  */
2974 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2975 {
2976         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2977         struct ext4_group_desc *gdp = NULL;
2978
2979         for (group = 0; group < ngroups; group++) {
2980                 gdp = ext4_get_group_desc(sb, group, NULL);
2981                 if (!gdp)
2982                         continue;
2983
2984                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2985                         break;
2986         }
2987
2988         return group;
2989 }
2990
2991 static int ext4_li_info_new(void)
2992 {
2993         struct ext4_lazy_init *eli = NULL;
2994
2995         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2996         if (!eli)
2997                 return -ENOMEM;
2998
2999         INIT_LIST_HEAD(&eli->li_request_list);
3000         mutex_init(&eli->li_list_mtx);
3001
3002         eli->li_state |= EXT4_LAZYINIT_QUIT;
3003
3004         ext4_li_info = eli;
3005
3006         return 0;
3007 }
3008
3009 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3010                                             ext4_group_t start)
3011 {
3012         struct ext4_sb_info *sbi = EXT4_SB(sb);
3013         struct ext4_li_request *elr;
3014         unsigned long rnd;
3015
3016         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3017         if (!elr)
3018                 return NULL;
3019
3020         elr->lr_super = sb;
3021         elr->lr_sbi = sbi;
3022         elr->lr_next_group = start;
3023
3024         /*
3025          * Randomize first schedule time of the request to
3026          * spread the inode table initialization requests
3027          * better.
3028          */
3029         get_random_bytes(&rnd, sizeof(rnd));
3030         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3031                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3032
3033         return elr;
3034 }
3035
3036 static int ext4_register_li_request(struct super_block *sb,
3037                                     ext4_group_t first_not_zeroed)
3038 {
3039         struct ext4_sb_info *sbi = EXT4_SB(sb);
3040         struct ext4_li_request *elr;
3041         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3042         int ret = 0;
3043
3044         if (sbi->s_li_request != NULL) {
3045                 /*
3046                  * Reset timeout so it can be computed again, because
3047                  * s_li_wait_mult might have changed.
3048                  */
3049                 sbi->s_li_request->lr_timeout = 0;
3050                 return 0;
3051         }
3052
3053         if (first_not_zeroed == ngroups ||
3054             (sb->s_flags & MS_RDONLY) ||
3055             !test_opt(sb, INIT_INODE_TABLE))
3056                 return 0;
3057
3058         elr = ext4_li_request_new(sb, first_not_zeroed);
3059         if (!elr)
3060                 return -ENOMEM;
3061
3062         mutex_lock(&ext4_li_mtx);
3063
3064         if (NULL == ext4_li_info) {
3065                 ret = ext4_li_info_new();
3066                 if (ret)
3067                         goto out;
3068         }
3069
3070         mutex_lock(&ext4_li_info->li_list_mtx);
3071         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3072         mutex_unlock(&ext4_li_info->li_list_mtx);
3073
3074         sbi->s_li_request = elr;
3075         /*
3076          * set elr to NULL here since it has been inserted to
3077          * the request_list and the removal and free of it is
3078          * handled by ext4_clear_request_list from now on.
3079          */
3080         elr = NULL;
3081
3082         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3083                 ret = ext4_run_lazyinit_thread();
3084                 if (ret)
3085                         goto out;
3086         }
3087 out:
3088         mutex_unlock(&ext4_li_mtx);
3089         if (ret)
3090                 kfree(elr);
3091         return ret;
3092 }
3093
3094 /*
3095  * We do not need to lock anything since this is called on
3096  * module unload.
3097  */
3098 static void ext4_destroy_lazyinit_thread(void)
3099 {
3100         /*
3101          * If thread exited earlier
3102          * there's nothing to be done.
3103          */
3104         if (!ext4_li_info || !ext4_lazyinit_task)
3105                 return;
3106
3107         kthread_stop(ext4_lazyinit_task);
3108 }
3109
3110 /*
3111  * Note: calculating the overhead so we can be compatible with
3112  * historical BSD practice is quite difficult in the face of
3113  * clusters/bigalloc.  This is because multiple metadata blocks from
3114  * different block group can end up in the same allocation cluster.
3115  * Calculating the exact overhead in the face of clustered allocation
3116  * requires either O(all block bitmaps) in memory or O(number of block
3117  * groups**2) in time.  We will still calculate the superblock for
3118  * older file systems --- and if we come across with a bigalloc file
3119  * system with zero in s_overhead_clusters the estimate will be close to
3120  * correct especially for very large cluster sizes --- but for newer
3121  * file systems, it's better to calculate this figure once at mkfs
3122  * time, and store it in the superblock.  If the superblock value is
3123  * present (even for non-bigalloc file systems), we will use it.
3124  */
3125 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3126                           char *buf)
3127 {
3128         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3129         struct ext4_group_desc  *gdp;
3130         ext4_fsblk_t            first_block, last_block, b;
3131         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3132         int                     s, j, count = 0;
3133
3134         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3135                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3136                         sbi->s_itb_per_group + 2);
3137
3138         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3139                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3140         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3141         for (i = 0; i < ngroups; i++) {
3142                 gdp = ext4_get_group_desc(sb, i, NULL);
3143                 b = ext4_block_bitmap(sb, gdp);
3144                 if (b >= first_block && b <= last_block) {
3145                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3146                         count++;
3147                 }
3148                 b = ext4_inode_bitmap(sb, gdp);
3149                 if (b >= first_block && b <= last_block) {
3150                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3151                         count++;
3152                 }
3153                 b = ext4_inode_table(sb, gdp);
3154                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3155                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3156                                 int c = EXT4_B2C(sbi, b - first_block);
3157                                 ext4_set_bit(c, buf);
3158                                 count++;
3159                         }
3160                 if (i != grp)
3161                         continue;
3162                 s = 0;
3163                 if (ext4_bg_has_super(sb, grp)) {
3164                         ext4_set_bit(s++, buf);
3165                         count++;
3166                 }
3167                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3168                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3169                         count++;
3170                 }
3171         }
3172         if (!count)
3173                 return 0;
3174         return EXT4_CLUSTERS_PER_GROUP(sb) -
3175                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3176 }
3177
3178 /*
3179  * Compute the overhead and stash it in sbi->s_overhead
3180  */
3181 int ext4_calculate_overhead(struct super_block *sb)
3182 {
3183         struct ext4_sb_info *sbi = EXT4_SB(sb);
3184         struct ext4_super_block *es = sbi->s_es;
3185         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3186         ext4_fsblk_t overhead = 0;
3187         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3188
3189         memset(buf, 0, PAGE_SIZE);
3190         if (!buf)
3191                 return -ENOMEM;
3192
3193         /*
3194          * Compute the overhead (FS structures).  This is constant
3195          * for a given filesystem unless the number of block groups
3196          * changes so we cache the previous value until it does.
3197          */
3198
3199         /*
3200          * All of the blocks before first_data_block are overhead
3201          */
3202         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3203
3204         /*
3205          * Add the overhead found in each block group
3206          */
3207         for (i = 0; i < ngroups; i++) {
3208                 int blks;
3209
3210                 blks = count_overhead(sb, i, buf);
3211                 overhead += blks;
3212                 if (blks)
3213                         memset(buf, 0, PAGE_SIZE);
3214                 cond_resched();
3215         }
3216         sbi->s_overhead = overhead;
3217         smp_wmb();
3218         free_page((unsigned long) buf);
3219         return 0;
3220 }
3221
3222 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3223 {
3224         char *orig_data = kstrdup(data, GFP_KERNEL);
3225         struct buffer_head *bh;
3226         struct ext4_super_block *es = NULL;
3227         struct ext4_sb_info *sbi;
3228         ext4_fsblk_t block;
3229         ext4_fsblk_t sb_block = get_sb_block(&data);
3230         ext4_fsblk_t logical_sb_block;
3231         unsigned long offset = 0;
3232         unsigned long journal_devnum = 0;
3233         unsigned long def_mount_opts;
3234         struct inode *root;
3235         char *cp;
3236         const char *descr;
3237         int ret = -ENOMEM;
3238         int blocksize, clustersize;
3239         unsigned int db_count;
3240         unsigned int i;
3241         int needs_recovery, has_huge_files, has_bigalloc;
3242         __u64 blocks_count;
3243         int err;
3244         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3245         ext4_group_t first_not_zeroed;
3246
3247         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3248         if (!sbi)
3249                 goto out_free_orig;
3250
3251         sbi->s_blockgroup_lock =
3252                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3253         if (!sbi->s_blockgroup_lock) {
3254                 kfree(sbi);
3255                 goto out_free_orig;
3256         }
3257         sb->s_fs_info = sbi;
3258         sbi->s_mount_opt = 0;
3259         sbi->s_resuid = EXT4_DEF_RESUID;
3260         sbi->s_resgid = EXT4_DEF_RESGID;
3261         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3262         sbi->s_sb_block = sb_block;
3263         if (sb->s_bdev->bd_part)
3264                 sbi->s_sectors_written_start =
3265                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3266
3267         /* Cleanup superblock name */
3268         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3269                 *cp = '!';
3270
3271         ret = -EINVAL;
3272         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3273         if (!blocksize) {
3274                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3275                 goto out_fail;
3276         }
3277
3278         /*
3279          * The ext4 superblock will not be buffer aligned for other than 1kB
3280          * block sizes.  We need to calculate the offset from buffer start.
3281          */
3282         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3283                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3284                 offset = do_div(logical_sb_block, blocksize);
3285         } else {
3286                 logical_sb_block = sb_block;
3287         }
3288
3289         if (!(bh = sb_bread(sb, logical_sb_block))) {
3290                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3291                 goto out_fail;
3292         }
3293         /*
3294          * Note: s_es must be initialized as soon as possible because
3295          *       some ext4 macro-instructions depend on its value
3296          */
3297         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3298         sbi->s_es = es;
3299         sb->s_magic = le16_to_cpu(es->s_magic);
3300         if (sb->s_magic != EXT4_SUPER_MAGIC)
3301                 goto cantfind_ext4;
3302         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3303
3304         /* Set defaults before we parse the mount options */
3305         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3306         set_opt(sb, INIT_INODE_TABLE);
3307         if (def_mount_opts & EXT4_DEFM_DEBUG)
3308                 set_opt(sb, DEBUG);
3309         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3310                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3311                         "2.6.38");
3312                 set_opt(sb, GRPID);
3313         }
3314         if (def_mount_opts & EXT4_DEFM_UID16)
3315                 set_opt(sb, NO_UID32);
3316         /* xattr user namespace & acls are now defaulted on */
3317 #ifdef CONFIG_EXT4_FS_XATTR
3318         set_opt(sb, XATTR_USER);
3319 #endif
3320 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3321         set_opt(sb, POSIX_ACL);
3322 #endif
3323         set_opt(sb, MBLK_IO_SUBMIT);
3324         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3325                 set_opt(sb, JOURNAL_DATA);
3326         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3327                 set_opt(sb, ORDERED_DATA);
3328         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3329                 set_opt(sb, WRITEBACK_DATA);
3330
3331         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3332                 set_opt(sb, ERRORS_PANIC);
3333         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3334                 set_opt(sb, ERRORS_CONT);
3335         else
3336                 set_opt(sb, ERRORS_RO);
3337         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3338                 set_opt(sb, BLOCK_VALIDITY);
3339         if (def_mount_opts & EXT4_DEFM_DISCARD)
3340                 set_opt(sb, DISCARD);
3341
3342         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3343         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3344         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3345         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3346         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3347
3348         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3349                 set_opt(sb, BARRIER);
3350
3351         /*
3352          * enable delayed allocation by default
3353          * Use -o nodelalloc to turn it off
3354          */
3355         if (!IS_EXT3_SB(sb) &&
3356             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3357                 set_opt(sb, DELALLOC);
3358
3359         /*
3360          * set default s_li_wait_mult for lazyinit, for the case there is
3361          * no mount option specified.
3362          */
3363         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3364
3365         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3366                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3367                 ext4_msg(sb, KERN_WARNING,
3368                          "failed to parse options in superblock: %s",
3369                          sbi->s_es->s_mount_opts);
3370         }
3371         if (!parse_options((char *) data, sb, &journal_devnum,
3372                            &journal_ioprio, NULL, 0))
3373                 goto failed_mount;
3374
3375         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3376                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3377                             "with data=journal disables delayed "
3378                             "allocation and O_DIRECT support!\n");
3379                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3380                         ext4_msg(sb, KERN_ERR, "can't mount with "
3381                                  "both data=journal and delalloc");
3382                         goto failed_mount;
3383                 }
3384                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3385                         ext4_msg(sb, KERN_ERR, "can't mount with "
3386                                  "both data=journal and dioread_nolock");
3387                         goto failed_mount;
3388                 }
3389                 if (test_opt(sb, DELALLOC))
3390                         clear_opt(sb, DELALLOC);
3391         }
3392
3393         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3394                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3395
3396         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3397             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3398              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3399              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3400                 ext4_msg(sb, KERN_WARNING,
3401                        "feature flags set on rev 0 fs, "
3402                        "running e2fsck is recommended");
3403
3404         if (IS_EXT2_SB(sb)) {
3405                 if (ext2_feature_set_ok(sb))
3406                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3407                                  "using the ext4 subsystem");
3408                 else {
3409                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3410                                  "to feature incompatibilities");
3411                         goto failed_mount;
3412                 }
3413         }
3414
3415         if (IS_EXT3_SB(sb)) {
3416                 if (ext3_feature_set_ok(sb))
3417                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3418                                  "using the ext4 subsystem");
3419                 else {
3420                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3421                                  "to feature incompatibilities");
3422                         goto failed_mount;
3423                 }
3424         }
3425
3426         /*
3427          * Check feature flags regardless of the revision level, since we
3428          * previously didn't change the revision level when setting the flags,
3429          * so there is a chance incompat flags are set on a rev 0 filesystem.
3430          */
3431         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3432                 goto failed_mount;
3433
3434         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3435         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3436             blocksize > EXT4_MAX_BLOCK_SIZE) {
3437                 ext4_msg(sb, KERN_ERR,
3438                        "Unsupported filesystem blocksize %d", blocksize);
3439                 goto failed_mount;
3440         }
3441
3442         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3443                 ext4_msg(sb, KERN_ERR,
3444                          "Number of reserved GDT blocks insanely large: %d",
3445                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3446                 goto failed_mount;
3447         }
3448
3449         if (sb->s_blocksize != blocksize) {
3450                 /* Validate the filesystem blocksize */
3451                 if (!sb_set_blocksize(sb, blocksize)) {
3452                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3453                                         blocksize);
3454                         goto failed_mount;
3455                 }
3456
3457                 brelse(bh);
3458                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3459                 offset = do_div(logical_sb_block, blocksize);
3460                 bh = sb_bread(sb, logical_sb_block);
3461                 if (!bh) {
3462                         ext4_msg(sb, KERN_ERR,
3463                                "Can't read superblock on 2nd try");
3464                         goto failed_mount;
3465                 }
3466                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3467                 sbi->s_es = es;
3468                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3469                         ext4_msg(sb, KERN_ERR,
3470                                "Magic mismatch, very weird!");
3471                         goto failed_mount;
3472                 }
3473         }
3474
3475         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3476                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3477         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3478                                                       has_huge_files);
3479         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3480
3481         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3482                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3483                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3484         } else {
3485                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3486                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3487                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3488                     (!is_power_of_2(sbi->s_inode_size)) ||
3489                     (sbi->s_inode_size > blocksize)) {
3490                         ext4_msg(sb, KERN_ERR,
3491                                "unsupported inode size: %d",
3492                                sbi->s_inode_size);
3493                         goto failed_mount;
3494                 }
3495                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3496                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3497         }
3498
3499         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3500         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3501                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3502                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3503                     !is_power_of_2(sbi->s_desc_size)) {
3504                         ext4_msg(sb, KERN_ERR,
3505                                "unsupported descriptor size %lu",
3506                                sbi->s_desc_size);
3507                         goto failed_mount;
3508                 }
3509         } else
3510                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3511
3512         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3513         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3514         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3515                 goto cantfind_ext4;
3516
3517         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3518         if (sbi->s_inodes_per_block == 0)
3519                 goto cantfind_ext4;
3520         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3521                                         sbi->s_inodes_per_block;
3522         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3523         sbi->s_sbh = bh;
3524         sbi->s_mount_state = le16_to_cpu(es->s_state);
3525         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3526         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3527
3528         for (i = 0; i < 4; i++)
3529                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3530         sbi->s_def_hash_version = es->s_def_hash_version;
3531         i = le32_to_cpu(es->s_flags);
3532         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3533                 sbi->s_hash_unsigned = 3;
3534         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3535 #ifdef __CHAR_UNSIGNED__
3536                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3537                 sbi->s_hash_unsigned = 3;
3538 #else
3539                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3540 #endif
3541                 sb->s_dirt = 1;
3542         }
3543
3544         /* Handle clustersize */
3545         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3546         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3547                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3548         if (has_bigalloc) {
3549                 if (clustersize < blocksize) {
3550                         ext4_msg(sb, KERN_ERR,
3551                                  "cluster size (%d) smaller than "
3552                                  "block size (%d)", clustersize, blocksize);
3553                         goto failed_mount;
3554                 }
3555                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3556                         le32_to_cpu(es->s_log_block_size);
3557                 sbi->s_clusters_per_group =
3558                         le32_to_cpu(es->s_clusters_per_group);
3559                 if (sbi->s_clusters_per_group > blocksize * 8) {
3560                         ext4_msg(sb, KERN_ERR,
3561                                  "#clusters per group too big: %lu",
3562                                  sbi->s_clusters_per_group);
3563                         goto failed_mount;
3564                 }
3565                 if (sbi->s_blocks_per_group !=
3566                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3567                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3568                                  "clusters per group (%lu) inconsistent",
3569                                  sbi->s_blocks_per_group,
3570                                  sbi->s_clusters_per_group);
3571                         goto failed_mount;
3572                 }
3573         } else {
3574                 if (clustersize != blocksize) {
3575                         ext4_warning(sb, "fragment/cluster size (%d) != "
3576                                      "block size (%d)", clustersize,
3577                                      blocksize);
3578                         clustersize = blocksize;
3579                 }
3580                 if (sbi->s_blocks_per_group > blocksize * 8) {
3581                         ext4_msg(sb, KERN_ERR,
3582                                  "#blocks per group too big: %lu",
3583                                  sbi->s_blocks_per_group);
3584                         goto failed_mount;
3585                 }
3586                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3587                 sbi->s_cluster_bits = 0;
3588         }
3589         sbi->s_cluster_ratio = clustersize / blocksize;
3590
3591         if (sbi->s_inodes_per_group > blocksize * 8) {
3592                 ext4_msg(sb, KERN_ERR,
3593                        "#inodes per group too big: %lu",
3594                        sbi->s_inodes_per_group);
3595                 goto failed_mount;
3596         }
3597
3598         /*
3599          * Test whether we have more sectors than will fit in sector_t,
3600          * and whether the max offset is addressable by the page cache.
3601          */
3602         err = generic_check_addressable(sb->s_blocksize_bits,
3603                                         ext4_blocks_count(es));
3604         if (err) {
3605                 ext4_msg(sb, KERN_ERR, "filesystem"
3606                          " too large to mount safely on this system");
3607                 if (sizeof(sector_t) < 8)
3608                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3609                 ret = err;
3610                 goto failed_mount;
3611         }
3612
3613         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3614                 goto cantfind_ext4;
3615
3616         /* check blocks count against device size */
3617         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3618         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3619                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3620                        "exceeds size of device (%llu blocks)",
3621                        ext4_blocks_count(es), blocks_count);
3622                 goto failed_mount;
3623         }
3624
3625         /*
3626          * It makes no sense for the first data block to be beyond the end
3627          * of the filesystem.
3628          */
3629         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3630                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3631                          "block %u is beyond end of filesystem (%llu)",