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/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.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/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  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 int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 static int ext4_verify_csum_type(struct super_block *sb,
 144                                  struct ext4_super_block *es)
 145 {
 146         if (!ext4_has_feature_metadata_csum(sb))
 147                 return 1;
 148
 149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 150 }
 151
 152 static __le32 ext4_superblock_csum(struct super_block *sb,
 153                                    struct ext4_super_block *es)
 154 {
 155         struct ext4_sb_info *sbi = EXT4_SB(sb);
 156         int offset = offsetof(struct ext4_super_block, s_checksum);
 157         __u32 csum;
 158
 159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 160
 161         return cpu_to_le32(csum);
 162 }
 163
 164 static int ext4_superblock_csum_verify(struct super_block *sb,
 165                                        struct ext4_super_block *es)
 166 {
 167         if (!ext4_has_metadata_csum(sb))
 168                 return 1;
 169
 170         return es->s_checksum == ext4_superblock_csum(sb, es);
 171 }
 172
 173 void ext4_superblock_csum_set(struct super_block *sb)
 174 {
 175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 176
 177         if (!ext4_has_metadata_csum(sb))
 178                 return;
 179
 180         es->s_checksum = ext4_superblock_csum(sb, es);
 181 }
 182
 183 void *ext4_kvmalloc(size_t size, gfp_t flags)
 184 {
 185         void *ret;
 186
 187         ret = kmalloc(size, flags | __GFP_NOWARN);
 188         if (!ret)
 189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 190         return ret;
 191 }
 192
 193 void *ext4_kvzalloc(size_t size, gfp_t flags)
 194 {
 195         void *ret;
 196
 197         ret = kzalloc(size, flags | __GFP_NOWARN);
 198         if (!ret)
 199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 200         return ret;
 201 }
 202
 203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 209 }
 210
 211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 212                                struct ext4_group_desc *bg)
 213 {
 214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 217 }
 218
 219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le32_to_cpu(bg->bg_inode_table_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 225 }
 226
 227 __u32 ext4_free_group_clusters(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 233 }
 234
 235 __u32 ext4_free_inodes_count(struct super_block *sb,
 236                               struct ext4_group_desc *bg)
 237 {
 238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 241 }
 242
 243 __u32 ext4_used_dirs_count(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 249 }
 250
 251 __u32 ext4_itable_unused_count(struct super_block *sb,
 252                               struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_itable_unused_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 257 }
 258
 259 void ext4_block_bitmap_set(struct super_block *sb,
 260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 261 {
 262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 265 }
 266
 267 void ext4_inode_bitmap_set(struct super_block *sb,
 268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 269 {
 270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 273 }
 274
 275 void ext4_inode_table_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 277 {
 278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 281 }
 282
 283 void ext4_free_group_clusters_set(struct super_block *sb,
 284                                   struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291 void ext4_free_inodes_set(struct super_block *sb,
 292                           struct ext4_group_desc *bg, __u32 count)
 293 {
 294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 297 }
 298
 299 void ext4_used_dirs_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, __u32 count)
 301 {
 302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 305 }
 306
 307 void ext4_itable_unused_set(struct super_block *sb,
 308                           struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315
 316 static void __save_error_info(struct super_block *sb, const char *func,
 317                             unsigned int line)
 318 {
 319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 320
 321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 322         if (bdev_read_only(sb->s_bdev))
 323                 return;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         le32_add_cpu(&es->s_error_count, 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352 /*
 353  * The del_gendisk() function uninitializes the disk-specific data
 354  * structures, including the bdi structure, without telling anyone
 355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 357  * This is a kludge to prevent these oops until we can put in a proper
 358  * hook in del_gendisk() to inform the VFS and file system layers.
 359  */
 360 static int block_device_ejected(struct super_block *sb)
 361 {
 362         struct inode *bd_inode = sb->s_bdev->bd_inode;
 363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 364
 365         return bdi->dev == NULL;
 366 }
 367
 368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 369 {
 370         struct super_block              *sb = journal->j_private;
 371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 372         int                             error = is_journal_aborted(journal);
 373         struct ext4_journal_cb_entry    *jce;
 374
 375         BUG_ON(txn->t_state == T_FINISHED);
 376         spin_lock(&sbi->s_md_lock);
 377         while (!list_empty(&txn->t_private_list)) {
 378                 jce = list_entry(txn->t_private_list.next,
 379                                  struct ext4_journal_cb_entry, jce_list);
 380                 list_del_init(&jce->jce_list);
 381                 spin_unlock(&sbi->s_md_lock);
 382                 jce->jce_func(sb, jce, error);
 383                 spin_lock(&sbi->s_md_lock);
 384         }
 385         spin_unlock(&sbi->s_md_lock);
 386 }
 387
 388 /* Deal with the reporting of failure conditions on a filesystem such as
 389  * inconsistencies detected or read IO failures.
 390  *
 391  * On ext2, we can store the error state of the filesystem in the
 392  * superblock.  That is not possible on ext4, because we may have other
 393  * write ordering constraints on the superblock which prevent us from
 394  * writing it out straight away; and given that the journal is about to
 395  * be aborted, we can't rely on the current, or future, transactions to
 396  * write out the superblock safely.
 397  *
 398  * We'll just use the jbd2_journal_abort() error code to record an error in
 399  * the journal instead.  On recovery, the journal will complain about
 400  * that error until we've noted it down and cleared it.
 401  */
 402
 403 static void ext4_handle_error(struct super_block *sb)
 404 {
 405         if (sb->s_flags & MS_RDONLY)
 406                 return;
 407
 408         if (!test_opt(sb, ERRORS_CONT)) {
 409                 journal_t *journal = EXT4_SB(sb)->s_journal;
 410
 411                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 412                 if (journal)
 413                         jbd2_journal_abort(journal, -EIO);
 414         }
 415         if (test_opt(sb, ERRORS_RO)) {
 416                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 417                 /*
 418                  * Make sure updated value of ->s_mount_flags will be visible
 419                  * before ->s_flags update
 420                  */
 421                 smp_wmb();
 422                 sb->s_flags |= MS_RDONLY;
 423         }
 424         if (test_opt(sb, ERRORS_PANIC)) {
 425                 if (EXT4_SB(sb)->s_journal &&
 426                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 427                         return;
 428                 panic("EXT4-fs (device %s): panic forced after error\n",
 429                         sb->s_id);
 430         }
 431 }
 432
 433 #define ext4_error_ratelimit(sb)                                        \
 434                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 435                              "EXT4-fs error")
 436
 437 void __ext4_error(struct super_block *sb, const char *function,
 438                   unsigned int line, const char *fmt, ...)
 439 {
 440         struct va_format vaf;
 441         va_list args;
 442
 443         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 444                 return;
 445
 446         if (ext4_error_ratelimit(sb)) {
 447                 va_start(args, fmt);
 448                 vaf.fmt = fmt;
 449                 vaf.va = &args;
 450                 printk(KERN_CRIT
 451                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 452                        sb->s_id, function, line, current->comm, &vaf);
 453                 va_end(args);
 454         }
 455         save_error_info(sb, function, line);
 456         ext4_handle_error(sb);
 457 }
 458
 459 void __ext4_error_inode(struct inode *inode, const char *function,
 460                         unsigned int line, ext4_fsblk_t block,
 461                         const char *fmt, ...)
 462 {
 463         va_list args;
 464         struct va_format vaf;
 465         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 466
 467         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 468                 return;
 469
 470         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 471         es->s_last_error_block = cpu_to_le64(block);
 472         if (ext4_error_ratelimit(inode->i_sb)) {
 473                 va_start(args, fmt);
 474                 vaf.fmt = fmt;
 475                 vaf.va = &args;
 476                 if (block)
 477                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 478                                "inode #%lu: block %llu: comm %s: %pV\n",
 479                                inode->i_sb->s_id, function, line, inode->i_ino,
 480                                block, current->comm, &vaf);
 481                 else
 482                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 483                                "inode #%lu: comm %s: %pV\n",
 484                                inode->i_sb->s_id, function, line, inode->i_ino,
 485                                current->comm, &vaf);
 486                 va_end(args);
 487         }
 488         save_error_info(inode->i_sb, function, line);
 489         ext4_handle_error(inode->i_sb);
 490 }
 491
 492 void __ext4_error_file(struct file *file, 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;
 499         struct inode *inode = file_inode(file);
 500         char pathname[80], *path;
 501
 502         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 503                 return;
 504
 505         es = EXT4_SB(inode->i_sb)->s_es;
 506         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 507         if (ext4_error_ratelimit(inode->i_sb)) {
 508                 path = file_path(file, pathname, sizeof(pathname));
 509                 if (IS_ERR(path))
 510                         path = "(unknown)";
 511                 va_start(args, fmt);
 512                 vaf.fmt = fmt;
 513                 vaf.va = &args;
 514                 if (block)
 515                         printk(KERN_CRIT
 516                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 517                                "block %llu: comm %s: path %s: %pV\n",
 518                                inode->i_sb->s_id, function, line, inode->i_ino,
 519                                block, current->comm, path, &vaf);
 520                 else
 521                         printk(KERN_CRIT
 522                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 523                                "comm %s: path %s: %pV\n",
 524                                inode->i_sb->s_id, function, line, inode->i_ino,
 525                                current->comm, path, &vaf);
 526                 va_end(args);
 527         }
 528         save_error_info(inode->i_sb, function, line);
 529         ext4_handle_error(inode->i_sb);
 530 }
 531
 532 const char *ext4_decode_error(struct super_block *sb, int errno,
 533                               char nbuf[16])
 534 {
 535         char *errstr = NULL;
 536
 537         switch (errno) {
 538         case -EFSCORRUPTED:
 539                 errstr = "Corrupt filesystem";
 540                 break;
 541         case -EFSBADCRC:
 542                 errstr = "Filesystem failed CRC";
 543                 break;
 544         case -EIO:
 545                 errstr = "IO failure";
 546                 break;
 547         case -ENOMEM:
 548                 errstr = "Out of memory";
 549                 break;
 550         case -EROFS:
 551                 if (!sb || (EXT4_SB(sb)->s_journal &&
 552                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 553                         errstr = "Journal has aborted";
 554                 else
 555                         errstr = "Readonly filesystem";
 556                 break;
 557         default:
 558                 /* If the caller passed in an extra buffer for unknown
 559                  * errors, textualise them now.  Else we just return
 560                  * NULL. */
 561                 if (nbuf) {
 562                         /* Check for truncated error codes... */
 563                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 564                                 errstr = nbuf;
 565                 }
 566                 break;
 567         }
 568
 569         return errstr;
 570 }
 571
 572 /* __ext4_std_error decodes expected errors from journaling functions
 573  * automatically and invokes the appropriate error response.  */
 574
 575 void __ext4_std_error(struct super_block *sb, const char *function,
 576                       unsigned int line, int errno)
 577 {
 578         char nbuf[16];
 579         const char *errstr;
 580
 581         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 582                 return;
 583
 584         /* Special case: if the error is EROFS, and we're not already
 585          * inside a transaction, then there's really no point in logging
 586          * an error. */
 587         if (errno == -EROFS && journal_current_handle() == NULL &&
 588             (sb->s_flags & MS_RDONLY))
 589                 return;
 590
 591         if (ext4_error_ratelimit(sb)) {
 592                 errstr = ext4_decode_error(sb, errno, nbuf);
 593                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 594                        sb->s_id, function, line, errstr);
 595         }
 596
 597         save_error_info(sb, function, line);
 598         ext4_handle_error(sb);
 599 }
 600
 601 /*
 602  * ext4_abort is a much stronger failure handler than ext4_error.  The
 603  * abort function may be used to deal with unrecoverable failures such
 604  * as journal IO errors or ENOMEM at a critical moment in log management.
 605  *
 606  * We unconditionally force the filesystem into an ABORT|READONLY state,
 607  * unless the error response on the fs has been set to panic in which
 608  * case we take the easy way out and panic immediately.
 609  */
 610
 611 void __ext4_abort(struct super_block *sb, const char *function,
 612                 unsigned int line, const char *fmt, ...)
 613 {
 614         struct va_format vaf;
 615         va_list args;
 616
 617         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 618                 return;
 619
 620         save_error_info(sb, function, line);
 621         va_start(args, fmt);
 622         vaf.fmt = fmt;
 623         vaf.va = &args;
 624         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 625                sb->s_id, function, line, &vaf);
 626         va_end(args);
 627
 628         if ((sb->s_flags & MS_RDONLY) == 0) {
 629                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 630                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 631                 /*
 632                  * Make sure updated value of ->s_mount_flags will be visible
 633                  * before ->s_flags update
 634                  */
 635                 smp_wmb();
 636                 sb->s_flags |= MS_RDONLY;
 637                 if (EXT4_SB(sb)->s_journal)
 638                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 639                 save_error_info(sb, function, line);
 640         }
 641         if (test_opt(sb, ERRORS_PANIC)) {
 642                 if (EXT4_SB(sb)->s_journal &&
 643                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 644                         return;
 645                 panic("EXT4-fs panic from previous error\n");
 646         }
 647 }
 648
 649 void __ext4_msg(struct super_block *sb,
 650                 const char *prefix, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 656                 return;
 657
 658         va_start(args, fmt);
 659         vaf.fmt = fmt;
 660         vaf.va = &args;
 661         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 662         va_end(args);
 663 }
 664
 665 #define ext4_warning_ratelimit(sb)                                      \
 666                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 667                              "EXT4-fs warning")
 668
 669 void __ext4_warning(struct super_block *sb, const char *function,
 670                     unsigned int line, const char *fmt, ...)
 671 {
 672         struct va_format vaf;
 673         va_list args;
 674
 675         if (!ext4_warning_ratelimit(sb))
 676                 return;
 677
 678         va_start(args, fmt);
 679         vaf.fmt = fmt;
 680         vaf.va = &args;
 681         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 682                sb->s_id, function, line, &vaf);
 683         va_end(args);
 684 }
 685
 686 void __ext4_warning_inode(const struct inode *inode, const char *function,
 687                           unsigned int line, const char *fmt, ...)
 688 {
 689         struct va_format vaf;
 690         va_list args;
 691
 692         if (!ext4_warning_ratelimit(inode->i_sb))
 693                 return;
 694
 695         va_start(args, fmt);
 696         vaf.fmt = fmt;
 697         vaf.va = &args;
 698         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 699                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 700                function, line, inode->i_ino, current->comm, &vaf);
 701         va_end(args);
 702 }
 703
 704 void __ext4_grp_locked_error(const char *function, unsigned int line,
 705                              struct super_block *sb, ext4_group_t grp,
 706                              unsigned long ino, ext4_fsblk_t block,
 707                              const char *fmt, ...)
 708 __releases(bitlock)
 709 __acquires(bitlock)
 710 {
 711         struct va_format vaf;
 712         va_list args;
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 716                 return;
 717
 718         es->s_last_error_ino = cpu_to_le32(ino);
 719         es->s_last_error_block = cpu_to_le64(block);
 720         __save_error_info(sb, function, line);
 721
 722         if (ext4_error_ratelimit(sb)) {
 723                 va_start(args, fmt);
 724                 vaf.fmt = fmt;
 725                 vaf.va = &args;
 726                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 727                        sb->s_id, function, line, grp);
 728                 if (ino)
 729                         printk(KERN_CONT "inode %lu: ", ino);
 730                 if (block)
 731                         printk(KERN_CONT "block %llu:",
 732                                (unsigned long long) block);
 733                 printk(KERN_CONT "%pV\n", &vaf);
 734                 va_end(args);
 735         }
 736
 737         if (test_opt(sb, ERRORS_CONT)) {
 738                 ext4_commit_super(sb, 0);
 739                 return;
 740         }
 741
 742         ext4_unlock_group(sb, grp);
 743         ext4_handle_error(sb);
 744         /*
 745          * We only get here in the ERRORS_RO case; relocking the group
 746          * may be dangerous, but nothing bad will happen since the
 747          * filesystem will have already been marked read/only and the
 748          * journal has been aborted.  We return 1 as a hint to callers
 749          * who might what to use the return value from
 750          * ext4_grp_locked_error() to distinguish between the
 751          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 752          * aggressively from the ext4 function in question, with a
 753          * more appropriate error code.
 754          */
 755         ext4_lock_group(sb, grp);
 756         return;
 757 }
 758
 759 void ext4_update_dynamic_rev(struct super_block *sb)
 760 {
 761         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 762
 763         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 764                 return;
 765
 766         ext4_warning(sb,
 767                      "updating to rev %d because of new feature flag, "
 768                      "running e2fsck is recommended",
 769                      EXT4_DYNAMIC_REV);
 770
 771         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 772         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 773         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 774         /* leave es->s_feature_*compat flags alone */
 775         /* es->s_uuid will be set by e2fsck if empty */
 776
 777         /*
 778          * The rest of the superblock fields should be zero, and if not it
 779          * means they are likely already in use, so leave them alone.  We
 780          * can leave it up to e2fsck to clean up any inconsistencies there.
 781          */
 782 }
 783
 784 /*
 785  * Open the external journal device
 786  */
 787 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 788 {
 789         struct block_device *bdev;
 790         char b[BDEVNAME_SIZE];
 791
 792         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 793         if (IS_ERR(bdev))
 794                 goto fail;
 795         return bdev;
 796
 797 fail:
 798         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 799                         __bdevname(dev, b), PTR_ERR(bdev));
 800         return NULL;
 801 }
 802
 803 /*
 804  * Release the journal device
 805  */
 806 static void ext4_blkdev_put(struct block_device *bdev)
 807 {
 808         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 809 }
 810
 811 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 812 {
 813         struct block_device *bdev;
 814         bdev = sbi->journal_bdev;
 815         if (bdev) {
 816                 ext4_blkdev_put(bdev);
 817                 sbi->journal_bdev = NULL;
 818         }
 819 }
 820
 821 static inline struct inode *orphan_list_entry(struct list_head *l)
 822 {
 823         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 824 }
 825
 826 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 827 {
 828         struct list_head *l;
 829
 830         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 831                  le32_to_cpu(sbi->s_es->s_last_orphan));
 832
 833         printk(KERN_ERR "sb_info orphan list:\n");
 834         list_for_each(l, &sbi->s_orphan) {
 835                 struct inode *inode = orphan_list_entry(l);
 836                 printk(KERN_ERR "  "
 837                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 838                        inode->i_sb->s_id, inode->i_ino, inode,
 839                        inode->i_mode, inode->i_nlink,
 840                        NEXT_ORPHAN(inode));
 841         }
 842 }
 843
 844 #ifdef CONFIG_QUOTA
 845 static int ext4_quota_off(struct super_block *sb, int type);
 846
 847 static inline void ext4_quota_off_umount(struct super_block *sb)
 848 {
 849         int type;
 850
 851         if (ext4_has_feature_quota(sb)) {
 852                 dquot_disable(sb, -1,
 853                               DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 854         } else {
 855                 /* Use our quota_off function to clear inode flags etc. */
 856                 for (type = 0; type < EXT4_MAXQUOTAS; type++)
 857                         ext4_quota_off(sb, type);
 858         }
 859 }
 860 #else
 861 static inline void ext4_quota_off_umount(struct super_block *sb)
 862 {
 863 }
 864 #endif
 865
 866 static void ext4_put_super(struct super_block *sb)
 867 {
 868         struct ext4_sb_info *sbi = EXT4_SB(sb);
 869         struct ext4_super_block *es = sbi->s_es;
 870         int aborted = 0;
 871         int i, err;
 872
 873         ext4_unregister_li_request(sb);
 874         ext4_quota_off_umount(sb);
 875
 876         flush_workqueue(sbi->rsv_conversion_wq);
 877         destroy_workqueue(sbi->rsv_conversion_wq);
 878
 879         if (sbi->s_journal) {
 880                 aborted = is_journal_aborted(sbi->s_journal);
 881                 err = jbd2_journal_destroy(sbi->s_journal);
 882                 sbi->s_journal = NULL;
 883                 if ((err < 0) && !aborted)
 884                         ext4_abort(sb, "Couldn't clean up the journal");
 885         }
 886
 887         ext4_unregister_sysfs(sb);
 888         ext4_es_unregister_shrinker(sbi);
 889         del_timer_sync(&sbi->s_err_report);
 890         ext4_release_system_zone(sb);
 891         ext4_mb_release(sb);
 892         ext4_ext_release(sb);
 893
 894         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 895                 ext4_clear_feature_journal_needs_recovery(sb);
 896                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 897         }
 898         if (!(sb->s_flags & MS_RDONLY))
 899                 ext4_commit_super(sb, 1);
 900
 901         for (i = 0; i < sbi->s_gdb_count; i++)
 902                 brelse(sbi->s_group_desc[i]);
 903         kvfree(sbi->s_group_desc);
 904         kvfree(sbi->s_flex_groups);
 905         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 906         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 907         percpu_counter_destroy(&sbi->s_dirs_counter);
 908         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 909         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 910 #ifdef CONFIG_QUOTA
 911         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 912                 kfree(sbi->s_qf_names[i]);
 913 #endif
 914
 915         /* Debugging code just in case the in-memory inode orphan list
 916          * isn't empty.  The on-disk one can be non-empty if we've
 917          * detected an error and taken the fs readonly, but the
 918          * in-memory list had better be clean by this point. */
 919         if (!list_empty(&sbi->s_orphan))
 920                 dump_orphan_list(sb, sbi);
 921         J_ASSERT(list_empty(&sbi->s_orphan));
 922
 923         sync_blockdev(sb->s_bdev);
 924         invalidate_bdev(sb->s_bdev);
 925         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 926                 /*
 927                  * Invalidate the journal device's buffers.  We don't want them
 928                  * floating about in memory - the physical journal device may
 929                  * hotswapped, and it breaks the `ro-after' testing code.
 930                  */
 931                 sync_blockdev(sbi->journal_bdev);
 932                 invalidate_bdev(sbi->journal_bdev);
 933                 ext4_blkdev_remove(sbi);
 934         }
 935         if (sbi->s_mb_cache) {
 936                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 937                 sbi->s_mb_cache = NULL;
 938         }
 939         if (sbi->s_mmp_tsk)
 940                 kthread_stop(sbi->s_mmp_tsk);
 941         brelse(sbi->s_sbh);
 942         sb->s_fs_info = NULL;
 943         /*
 944          * Now that we are completely done shutting down the
 945          * superblock, we need to actually destroy the kobject.
 946          */
 947         kobject_put(&sbi->s_kobj);
 948         wait_for_completion(&sbi->s_kobj_unregister);
 949         if (sbi->s_chksum_driver)
 950                 crypto_free_shash(sbi->s_chksum_driver);
 951         kfree(sbi->s_blockgroup_lock);
 952         kfree(sbi);
 953 }
 954
 955 static struct kmem_cache *ext4_inode_cachep;
 956
 957 /*
 958  * Called inside transaction, so use GFP_NOFS
 959  */
 960 static struct inode *ext4_alloc_inode(struct super_block *sb)
 961 {
 962         struct ext4_inode_info *ei;
 963
 964         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 965         if (!ei)
 966                 return NULL;
 967
 968         ei->vfs_inode.i_version = 1;
 969         spin_lock_init(&ei->i_raw_lock);
 970         INIT_LIST_HEAD(&ei->i_prealloc_list);
 971         spin_lock_init(&ei->i_prealloc_lock);
 972         ext4_es_init_tree(&ei->i_es_tree);
 973         rwlock_init(&ei->i_es_lock);
 974         INIT_LIST_HEAD(&ei->i_es_list);
 975         ei->i_es_all_nr = 0;
 976         ei->i_es_shk_nr = 0;
 977         ei->i_es_shrink_lblk = 0;
 978         ei->i_reserved_data_blocks = 0;
 979         ei->i_reserved_meta_blocks = 0;
 980         ei->i_allocated_meta_blocks = 0;
 981         ei->i_da_metadata_calc_len = 0;
 982         ei->i_da_metadata_calc_last_lblock = 0;
 983         spin_lock_init(&(ei->i_block_reservation_lock));
 984 #ifdef CONFIG_QUOTA
 985         ei->i_reserved_quota = 0;
 986         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 987 #endif
 988         ei->jinode = NULL;
 989         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 990         spin_lock_init(&ei->i_completed_io_lock);
 991         ei->i_sync_tid = 0;
 992         ei->i_datasync_tid = 0;
 993         atomic_set(&ei->i_unwritten, 0);
 994         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 995         return &ei->vfs_inode;
 996 }
 997
 998 static int ext4_drop_inode(struct inode *inode)
 999 {
1000         int drop = generic_drop_inode(inode);
1001
1002         trace_ext4_drop_inode(inode, drop);
1003         return drop;
1004 }
1005
1006 static void ext4_i_callback(struct rcu_head *head)
1007 {
1008         struct inode *inode = container_of(head, struct inode, i_rcu);
1009         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1010 }
1011
1012 static void ext4_destroy_inode(struct inode *inode)
1013 {
1014         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1015                 ext4_msg(inode->i_sb, KERN_ERR,
1016                          "Inode %lu (%p): orphan list check failed!",
1017                          inode->i_ino, EXT4_I(inode));
1018                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1019                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1020                                 true);
1021                 dump_stack();
1022         }
1023         call_rcu(&inode->i_rcu, ext4_i_callback);
1024 }
1025
1026 static void init_once(void *foo)
1027 {
1028         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1029
1030         INIT_LIST_HEAD(&ei->i_orphan);
1031         init_rwsem(&ei->xattr_sem);
1032         init_rwsem(&ei->i_data_sem);
1033         init_rwsem(&ei->i_mmap_sem);
1034         inode_init_once(&ei->vfs_inode);
1035 }
1036
1037 static int __init init_inodecache(void)
1038 {
1039         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1040                                              sizeof(struct ext4_inode_info),
1041                                              0, (SLAB_RECLAIM_ACCOUNT|
1042                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1043                                              init_once);
1044         if (ext4_inode_cachep == NULL)
1045                 return -ENOMEM;
1046         return 0;
1047 }
1048
1049 static void destroy_inodecache(void)
1050 {
1051         /*
1052          * Make sure all delayed rcu free inodes are flushed before we
1053          * destroy cache.
1054          */
1055         rcu_barrier();
1056         kmem_cache_destroy(ext4_inode_cachep);
1057 }
1058
1059 void ext4_clear_inode(struct inode *inode)
1060 {
1061         invalidate_inode_buffers(inode);
1062         clear_inode(inode);
1063         dquot_drop(inode);
1064         ext4_discard_preallocations(inode);
1065         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1066         if (EXT4_I(inode)->jinode) {
1067                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1068                                                EXT4_I(inode)->jinode);
1069                 jbd2_free_inode(EXT4_I(inode)->jinode);
1070                 EXT4_I(inode)->jinode = NULL;
1071         }
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073         fscrypt_put_encryption_info(inode, NULL);
1074 #endif
1075 }
1076
1077 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1078                                         u64 ino, u32 generation)
1079 {
1080         struct inode *inode;
1081
1082         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1083                 return ERR_PTR(-ESTALE);
1084         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1085                 return ERR_PTR(-ESTALE);
1086
1087         /* iget isn't really right if the inode is currently unallocated!!
1088          *
1089          * ext4_read_inode will return a bad_inode if the inode had been
1090          * deleted, so we should be safe.
1091          *
1092          * Currently we don't know the generation for parent directory, so
1093          * a generation of 0 means "accept any"
1094          */
1095         inode = ext4_iget_normal(sb, ino);
1096         if (IS_ERR(inode))
1097                 return ERR_CAST(inode);
1098         if (generation && inode->i_generation != generation) {
1099                 iput(inode);
1100                 return ERR_PTR(-ESTALE);
1101         }
1102
1103         return inode;
1104 }
1105
1106 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1107                                         int fh_len, int fh_type)
1108 {
1109         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1110                                     ext4_nfs_get_inode);
1111 }
1112
1113 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1114                                         int fh_len, int fh_type)
1115 {
1116         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1117                                     ext4_nfs_get_inode);
1118 }
1119
1120 /*
1121  * Try to release metadata pages (indirect blocks, directories) which are
1122  * mapped via the block device.  Since these pages could have journal heads
1123  * which would prevent try_to_free_buffers() from freeing them, we must use
1124  * jbd2 layer's try_to_free_buffers() function to release them.
1125  */
1126 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1127                                  gfp_t wait)
1128 {
1129         journal_t *journal = EXT4_SB(sb)->s_journal;
1130
1131         WARN_ON(PageChecked(page));
1132         if (!page_has_buffers(page))
1133                 return 0;
1134         if (journal)
1135                 return jbd2_journal_try_to_free_buffers(journal, page,
1136                                                 wait & ~__GFP_DIRECT_RECLAIM);
1137         return try_to_free_buffers(page);
1138 }
1139
1140 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1141 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1142 {
1143         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1144                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1145 }
1146
1147 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1148                                                         void *fs_data)
1149 {
1150         handle_t *handle = fs_data;
1151         int res, res2, retries = 0;
1152
1153         res = ext4_convert_inline_data(inode);
1154         if (res)
1155                 return res;
1156
1157         /*
1158          * If a journal handle was specified, then the encryption context is
1159          * being set on a new inode via inheritance and is part of a larger
1160          * transaction to create the inode.  Otherwise the encryption context is
1161          * being set on an existing inode in its own transaction.  Only in the
1162          * latter case should the "retry on ENOSPC" logic be used.
1163          */
1164
1165         if (handle) {
1166                 res = ext4_xattr_set_handle(handle, inode,
1167                                             EXT4_XATTR_INDEX_ENCRYPTION,
1168                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1169                                             ctx, len, 0);
1170                 if (!res) {
1171                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1172                         ext4_clear_inode_state(inode,
1173                                         EXT4_STATE_MAY_INLINE_DATA);
1174                         /*
1175                          * Update inode->i_flags - e.g. S_DAX may get disabled
1176                          */
1177                         ext4_set_inode_flags(inode);
1178                 }
1179                 return res;
1180         }
1181
1182 retry:
1183         handle = ext4_journal_start(inode, EXT4_HT_MISC,
1184                         ext4_jbd2_credits_xattr(inode));
1185         if (IS_ERR(handle))
1186                 return PTR_ERR(handle);
1187
1188         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1189                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1190                                     ctx, len, 0);
1191         if (!res) {
1192                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1193                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1194                 ext4_set_inode_flags(inode);
1195                 res = ext4_mark_inode_dirty(handle, inode);
1196                 if (res)
1197                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1198         }
1199         res2 = ext4_journal_stop(handle);
1200
1201         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1202                 goto retry;
1203         if (!res)
1204                 res = res2;
1205         return res;
1206 }
1207
1208 static int ext4_dummy_context(struct inode *inode)
1209 {
1210         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1211 }
1212
1213 static unsigned ext4_max_namelen(struct inode *inode)
1214 {
1215         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1216                 EXT4_NAME_LEN;
1217 }
1218
1219 static const struct fscrypt_operations ext4_cryptops = {
1220         .key_prefix             = "ext4:",
1221         .get_context            = ext4_get_context,
1222         .set_context            = ext4_set_context,
1223         .dummy_context          = ext4_dummy_context,
1224         .is_encrypted           = ext4_encrypted_inode,
1225         .empty_dir              = ext4_empty_dir,
1226         .max_namelen            = ext4_max_namelen,
1227 };
1228 #else
1229 static const struct fscrypt_operations ext4_cryptops = {
1230         .is_encrypted           = ext4_encrypted_inode,
1231 };
1232 #endif
1233
1234 #ifdef CONFIG_QUOTA
1235 static const char * const quotatypes[] = INITQFNAMES;
1236 #define QTYPE2NAME(t) (quotatypes[t])
1237
1238 static int ext4_write_dquot(struct dquot *dquot);
1239 static int ext4_acquire_dquot(struct dquot *dquot);
1240 static int ext4_release_dquot(struct dquot *dquot);
1241 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1242 static int ext4_write_info(struct super_block *sb, int type);
1243 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1244                          const struct path *path);
1245 static int ext4_quota_on_mount(struct super_block *sb, int type);
1246 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1247                                size_t len, loff_t off);
1248 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1249                                 const char *data, size_t len, loff_t off);
1250 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1251                              unsigned int flags);
1252 static int ext4_enable_quotas(struct super_block *sb);
1253 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1254
1255 static struct dquot **ext4_get_dquots(struct inode *inode)
1256 {
1257         return EXT4_I(inode)->i_dquot;
1258 }
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         .get_projid     = ext4_get_projid,
1270         .get_next_id    = ext4_get_next_id,
1271 };
1272
1273 static const struct quotactl_ops ext4_qctl_operations = {
1274         .quota_on       = ext4_quota_on,
1275         .quota_off      = ext4_quota_off,
1276         .quota_sync     = dquot_quota_sync,
1277         .get_state      = dquot_get_state,
1278         .set_info       = dquot_set_dqinfo,
1279         .get_dqblk      = dquot_get_dqblk,
1280         .set_dqblk      = dquot_set_dqblk,
1281         .get_nextdqblk  = dquot_get_next_dqblk,
1282 };
1283 #endif
1284
1285 static const struct super_operations ext4_sops = {
1286         .alloc_inode    = ext4_alloc_inode,
1287         .destroy_inode  = ext4_destroy_inode,
1288         .write_inode    = ext4_write_inode,
1289         .dirty_inode    = ext4_dirty_inode,
1290         .drop_inode     = ext4_drop_inode,
1291         .evict_inode    = ext4_evict_inode,
1292         .put_super      = ext4_put_super,
1293         .sync_fs        = ext4_sync_fs,
1294         .freeze_fs      = ext4_freeze,
1295         .unfreeze_fs    = ext4_unfreeze,
1296         .statfs         = ext4_statfs,
1297         .remount_fs     = ext4_remount,
1298         .show_options   = ext4_show_options,
1299 #ifdef CONFIG_QUOTA
1300         .quota_read     = ext4_quota_read,
1301         .quota_write    = ext4_quota_write,
1302         .get_dquots     = ext4_get_dquots,
1303 #endif
1304         .bdev_try_to_free_page = bdev_try_to_free_page,
1305 };
1306
1307 static const struct export_operations ext4_export_ops = {
1308         .fh_to_dentry = ext4_fh_to_dentry,
1309         .fh_to_parent = ext4_fh_to_parent,
1310         .get_parent = ext4_get_parent,
1311 };
1312
1313 enum {
1314         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1315         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1316         Opt_nouid32, Opt_debug, Opt_removed,
1317         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1318         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1319         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1320         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1321         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1322         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1323         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1324         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1325         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1326         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1327         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1328         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1329         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1330         Opt_inode_readahead_blks, Opt_journal_ioprio,
1331         Opt_dioread_nolock, Opt_dioread_lock,
1332         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1333         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1334 };
1335
1336 static const match_table_t tokens = {
1337         {Opt_bsd_df, "bsddf"},
1338         {Opt_minix_df, "minixdf"},
1339         {Opt_grpid, "grpid"},
1340         {Opt_grpid, "bsdgroups"},
1341         {Opt_nogrpid, "nogrpid"},
1342         {Opt_nogrpid, "sysvgroups"},
1343         {Opt_resgid, "resgid=%u"},
1344         {Opt_resuid, "resuid=%u"},
1345         {Opt_sb, "sb=%u"},
1346         {Opt_err_cont, "errors=continue"},
1347         {Opt_err_panic, "errors=panic"},
1348         {Opt_err_ro, "errors=remount-ro"},
1349         {Opt_nouid32, "nouid32"},
1350         {Opt_debug, "debug"},
1351         {Opt_removed, "oldalloc"},
1352         {Opt_removed, "orlov"},
1353         {Opt_user_xattr, "user_xattr"},
1354         {Opt_nouser_xattr, "nouser_xattr"},
1355         {Opt_acl, "acl"},
1356         {Opt_noacl, "noacl"},
1357         {Opt_noload, "norecovery"},
1358         {Opt_noload, "noload"},
1359         {Opt_removed, "nobh"},
1360         {Opt_removed, "bh"},
1361         {Opt_commit, "commit=%u"},
1362         {Opt_min_batch_time, "min_batch_time=%u"},
1363         {Opt_max_batch_time, "max_batch_time=%u"},
1364         {Opt_journal_dev, "journal_dev=%u"},
1365         {Opt_journal_path, "journal_path=%s"},
1366         {Opt_journal_checksum, "journal_checksum"},
1367         {Opt_nojournal_checksum, "nojournal_checksum"},
1368         {Opt_journal_async_commit, "journal_async_commit"},
1369         {Opt_abort, "abort"},
1370         {Opt_data_journal, "data=journal"},
1371         {Opt_data_ordered, "data=ordered"},
1372         {Opt_data_writeback, "data=writeback"},
1373         {Opt_data_err_abort, "data_err=abort"},
1374         {Opt_data_err_ignore, "data_err=ignore"},
1375         {Opt_offusrjquota, "usrjquota="},
1376         {Opt_usrjquota, "usrjquota=%s"},
1377         {Opt_offgrpjquota, "grpjquota="},
1378         {Opt_grpjquota, "grpjquota=%s"},
1379         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1380         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1381         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1382         {Opt_grpquota, "grpquota"},
1383         {Opt_noquota, "noquota"},
1384         {Opt_quota, "quota"},
1385         {Opt_usrquota, "usrquota"},
1386         {Opt_prjquota, "prjquota"},
1387         {Opt_barrier, "barrier=%u"},
1388         {Opt_barrier, "barrier"},
1389         {Opt_nobarrier, "nobarrier"},
1390         {Opt_i_version, "i_version"},
1391         {Opt_dax, "dax"},
1392         {Opt_stripe, "stripe=%u"},
1393         {Opt_delalloc, "delalloc"},
1394         {Opt_lazytime, "lazytime"},
1395         {Opt_nolazytime, "nolazytime"},
1396         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1397         {Opt_nodelalloc, "nodelalloc"},
1398         {Opt_removed, "mblk_io_submit"},
1399         {Opt_removed, "nomblk_io_submit"},
1400         {Opt_block_validity, "block_validity"},
1401         {Opt_noblock_validity, "noblock_validity"},
1402         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1403         {Opt_journal_ioprio, "journal_ioprio=%u"},
1404         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1405         {Opt_auto_da_alloc, "auto_da_alloc"},
1406         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1407         {Opt_dioread_nolock, "dioread_nolock"},
1408         {Opt_dioread_lock, "dioread_lock"},
1409         {Opt_discard, "discard"},
1410         {Opt_nodiscard, "nodiscard"},
1411         {Opt_init_itable, "init_itable=%u"},
1412         {Opt_init_itable, "init_itable"},
1413         {Opt_noinit_itable, "noinit_itable"},
1414         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1415         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1416         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1417         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1418         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1419         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1420         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1421         {Opt_err, NULL},
1422 };
1423
1424 static ext4_fsblk_t get_sb_block(void **data)
1425 {
1426         ext4_fsblk_t    sb_block;
1427         char            *options = (char *) *data;
1428
1429         if (!options || strncmp(options, "sb=", 3) != 0)
1430                 return 1;       /* Default location */
1431
1432         options += 3;
1433         /* TODO: use simple_strtoll with >32bit ext4 */
1434         sb_block = simple_strtoul(options, &options, 0);
1435         if (*options && *options != ',') {
1436                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1437                        (char *) *data);
1438                 return 1;
1439         }
1440         if (*options == ',')
1441                 options++;
1442         *data = (void *) options;
1443
1444         return sb_block;
1445 }
1446
1447 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1448 static const char deprecated_msg[] =
1449         "Mount option \"%s\" will be removed by %s\n"
1450         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1451
1452 #ifdef CONFIG_QUOTA
1453 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1454 {
1455         struct ext4_sb_info *sbi = EXT4_SB(sb);
1456         char *qname;
1457         int ret = -1;
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 -1;
1465         }
1466         if (ext4_has_feature_quota(sb)) {
1467                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1468                          "ignored when QUOTA feature is enabled");
1469                 return 1;
1470         }
1471         qname = match_strdup(args);
1472         if (!qname) {
1473                 ext4_msg(sb, KERN_ERR,
1474                         "Not enough memory for storing quotafile name");
1475                 return -1;
1476         }
1477         if (sbi->s_qf_names[qtype]) {
1478                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1479                         ret = 1;
1480                 else
1481                         ext4_msg(sb, KERN_ERR,
1482                                  "%s quota file already specified",
1483                                  QTYPE2NAME(qtype));
1484                 goto errout;
1485         }
1486         if (strchr(qname, '/')) {
1487                 ext4_msg(sb, KERN_ERR,
1488                         "quotafile must be on filesystem root");
1489                 goto errout;
1490         }
1491         sbi->s_qf_names[qtype] = qname;
1492         set_opt(sb, QUOTA);
1493         return 1;
1494 errout:
1495         kfree(qname);
1496         return ret;
1497 }
1498
1499 static int clear_qf_name(struct super_block *sb, int qtype)
1500 {
1501
1502         struct ext4_sb_info *sbi = EXT4_SB(sb);
1503
1504         if (sb_any_quota_loaded(sb) &&
1505                 sbi->s_qf_names[qtype]) {
1506                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1507                         " when quota turned on");
1508                 return -1;
1509         }
1510         kfree(sbi->s_qf_names[qtype]);
1511         sbi->s_qf_names[qtype] = NULL;
1512         return 1;
1513 }
1514 #endif
1515
1516 #define MOPT_SET        0x0001
1517 #define MOPT_CLEAR      0x0002
1518 #define MOPT_NOSUPPORT  0x0004
1519 #define MOPT_EXPLICIT   0x0008
1520 #define MOPT_CLEAR_ERR  0x0010
1521 #define MOPT_GTE0       0x0020
1522 #ifdef CONFIG_QUOTA
1523 #define MOPT_Q          0
1524 #define MOPT_QFMT       0x0040
1525 #else
1526 #define MOPT_Q          MOPT_NOSUPPORT
1527 #define MOPT_QFMT       MOPT_NOSUPPORT
1528 #endif
1529 #define MOPT_DATAJ      0x0080
1530 #define MOPT_NO_EXT2    0x0100
1531 #define MOPT_NO_EXT3    0x0200
1532 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1533 #define MOPT_STRING     0x0400
1534
1535 static const struct mount_opts {
1536         int     token;
1537         int     mount_opt;
1538         int     flags;
1539 } ext4_mount_opts[] = {
1540         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1541         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1542         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1543         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1544         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1545         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1546         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1547          MOPT_EXT4_ONLY | MOPT_SET},
1548         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1549          MOPT_EXT4_ONLY | MOPT_CLEAR},
1550         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1551         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1552         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1553          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1554         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1555          MOPT_EXT4_ONLY | MOPT_CLEAR},
1556         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1557          MOPT_EXT4_ONLY | MOPT_CLEAR},
1558         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1559          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1560         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1561                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1562          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1563         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1564         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1565         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1566         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1567         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1568          MOPT_NO_EXT2},
1569         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1570          MOPT_NO_EXT2},
1571         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1572         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1573         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1574         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1575         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1576         {Opt_commit, 0, MOPT_GTE0},
1577         {Opt_max_batch_time, 0, MOPT_GTE0},
1578         {Opt_min_batch_time, 0, MOPT_GTE0},
1579         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1580         {Opt_init_itable, 0, MOPT_GTE0},
1581         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1582         {Opt_stripe, 0, MOPT_GTE0},
1583         {Opt_resuid, 0, MOPT_GTE0},
1584         {Opt_resgid, 0, MOPT_GTE0},
1585         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1586         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1587         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1588         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1589         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1590         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1591          MOPT_NO_EXT2 | MOPT_DATAJ},
1592         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1593         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1594 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1595         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1596         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1597 #else
1598         {Opt_acl, 0, MOPT_NOSUPPORT},
1599         {Opt_noacl, 0, MOPT_NOSUPPORT},
1600 #endif
1601         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1602         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1603         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1604         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1605         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1606                                                         MOPT_SET | MOPT_Q},
1607         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1608                                                         MOPT_SET | MOPT_Q},
1609         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1610                                                         MOPT_SET | MOPT_Q},
1611         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1612                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1613                                                         MOPT_CLEAR | MOPT_Q},
1614         {Opt_usrjquota, 0, MOPT_Q},
1615         {Opt_grpjquota, 0, MOPT_Q},
1616         {Opt_offusrjquota, 0, MOPT_Q},
1617         {Opt_offgrpjquota, 0, MOPT_Q},
1618         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1619         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1620         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1621         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1622         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1623         {Opt_err, 0, 0}
1624 };
1625
1626 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1627                             substring_t *args, unsigned long *journal_devnum,
1628                             unsigned int *journal_ioprio, int is_remount)
1629 {
1630         struct ext4_sb_info *sbi = EXT4_SB(sb);
1631         const struct mount_opts *m;
1632         kuid_t uid;
1633         kgid_t gid;
1634         int arg = 0;
1635
1636 #ifdef CONFIG_QUOTA
1637         if (token == Opt_usrjquota)
1638                 return set_qf_name(sb, USRQUOTA, &args[0]);
1639         else if (token == Opt_grpjquota)
1640                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1641         else if (token == Opt_offusrjquota)
1642                 return clear_qf_name(sb, USRQUOTA);
1643         else if (token == Opt_offgrpjquota)
1644                 return clear_qf_name(sb, GRPQUOTA);
1645 #endif
1646         switch (token) {
1647         case Opt_noacl:
1648         case Opt_nouser_xattr:
1649                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1650                 break;
1651         case Opt_sb:
1652                 return 1;       /* handled by get_sb_block() */
1653         case Opt_removed:
1654                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1655                 return 1;
1656         case Opt_abort:
1657                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1658                 return 1;
1659         case Opt_i_version:
1660                 sb->s_flags |= MS_I_VERSION;
1661                 return 1;
1662         case Opt_lazytime:
1663                 sb->s_flags |= MS_LAZYTIME;
1664                 return 1;
1665         case Opt_nolazytime:
1666                 sb->s_flags &= ~MS_LAZYTIME;
1667                 return 1;
1668         }
1669
1670         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1671                 if (token == m->token)
1672                         break;
1673
1674         if (m->token == Opt_err) {
1675                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1676                          "or missing value", opt);
1677                 return -1;
1678         }
1679
1680         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1681                 ext4_msg(sb, KERN_ERR,
1682                          "Mount option \"%s\" incompatible with ext2", opt);
1683                 return -1;
1684         }
1685         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1686                 ext4_msg(sb, KERN_ERR,
1687                          "Mount option \"%s\" incompatible with ext3", opt);
1688                 return -1;
1689         }
1690
1691         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1692                 return -1;
1693         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1694                 return -1;
1695         if (m->flags & MOPT_EXPLICIT) {
1696                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1697                         set_opt2(sb, EXPLICIT_DELALLOC);
1698                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1699                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1700                 } else
1701                         return -1;
1702         }
1703         if (m->flags & MOPT_CLEAR_ERR)
1704                 clear_opt(sb, ERRORS_MASK);
1705         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1706                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1707                          "options when quota turned on");
1708                 return -1;
1709         }
1710
1711         if (m->flags & MOPT_NOSUPPORT) {
1712                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1713         } else if (token == Opt_commit) {
1714                 if (arg == 0)
1715                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1716                 sbi->s_commit_interval = HZ * arg;
1717         } else if (token == Opt_debug_want_extra_isize) {
1718                 sbi->s_want_extra_isize = arg;
1719         } else if (token == Opt_max_batch_time) {
1720                 sbi->s_max_batch_time = arg;
1721         } else if (token == Opt_min_batch_time) {
1722                 sbi->s_min_batch_time = arg;
1723         } else if (token == Opt_inode_readahead_blks) {
1724                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1725                         ext4_msg(sb, KERN_ERR,
1726                                  "EXT4-fs: inode_readahead_blks must be "
1727                                  "0 or a power of 2 smaller than 2^31");
1728                         return -1;
1729                 }
1730                 sbi->s_inode_readahead_blks = arg;
1731         } else if (token == Opt_init_itable) {
1732                 set_opt(sb, INIT_INODE_TABLE);
1733                 if (!args->from)
1734                         arg = EXT4_DEF_LI_WAIT_MULT;
1735                 sbi->s_li_wait_mult = arg;
1736         } else if (token == Opt_max_dir_size_kb) {
1737                 sbi->s_max_dir_size_kb = arg;
1738         } else if (token == Opt_stripe) {
1739                 sbi->s_stripe = arg;
1740         } else if (token == Opt_resuid) {
1741                 uid = make_kuid(current_user_ns(), arg);
1742                 if (!uid_valid(uid)) {
1743                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1744                         return -1;
1745                 }
1746                 sbi->s_resuid = uid;
1747         } else if (token == Opt_resgid) {
1748                 gid = make_kgid(current_user_ns(), arg);
1749                 if (!gid_valid(gid)) {
1750                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1751                         return -1;
1752                 }
1753                 sbi->s_resgid = gid;
1754         } else if (token == Opt_journal_dev) {
1755                 if (is_remount) {
1756                         ext4_msg(sb, KERN_ERR,
1757                                  "Cannot specify journal on remount");
1758                         return -1;
1759                 }
1760                 *journal_devnum = arg;
1761         } else if (token == Opt_journal_path) {
1762                 char *journal_path;
1763                 struct inode *journal_inode;
1764                 struct path path;
1765                 int error;
1766
1767                 if (is_remount) {
1768                         ext4_msg(sb, KERN_ERR,
1769                                  "Cannot specify journal on remount");
1770                         return -1;
1771                 }
1772                 journal_path = match_strdup(&args[0]);
1773                 if (!journal_path) {
1774                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1775                                 "journal device string");
1776                         return -1;
1777                 }
1778
1779                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1780                 if (error) {
1781                         ext4_msg(sb, KERN_ERR, "error: could not find "
1782                                 "journal device path: error %d", error);
1783                         kfree(journal_path);
1784                         return -1;
1785                 }
1786
1787                 journal_inode = d_inode(path.dentry);
1788                 if (!S_ISBLK(journal_inode->i_mode)) {
1789                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1790                                 "is not a block device", journal_path);
1791                         path_put(&path);
1792                         kfree(journal_path);
1793                         return -1;
1794                 }
1795
1796                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1797                 path_put(&path);
1798                 kfree(journal_path);
1799         } else if (token == Opt_journal_ioprio) {
1800                 if (arg > 7) {
1801                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1802                                  " (must be 0-7)");
1803                         return -1;
1804                 }
1805                 *journal_ioprio =
1806                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1807         } else if (token == Opt_test_dummy_encryption) {
1808 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1809                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1810                 ext4_msg(sb, KERN_WARNING,
1811                          "Test dummy encryption mode enabled");
1812 #else
1813                 ext4_msg(sb, KERN_WARNING,
1814                          "Test dummy encryption mount option ignored");
1815 #endif
1816         } else if (m->flags & MOPT_DATAJ) {
1817                 if (is_remount) {
1818                         if (!sbi->s_journal)
1819                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1820                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1821                                 ext4_msg(sb, KERN_ERR,
1822                                          "Cannot change data mode on remount");
1823                                 return -1;
1824                         }
1825                 } else {
1826                         clear_opt(sb, DATA_FLAGS);
1827                         sbi->s_mount_opt |= m->mount_opt;
1828                 }
1829 #ifdef CONFIG_QUOTA
1830         } else if (m->flags & MOPT_QFMT) {
1831                 if (sb_any_quota_loaded(sb) &&
1832                     sbi->s_jquota_fmt != m->mount_opt) {
1833                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1834                                  "quota options when quota turned on");
1835                         return -1;
1836                 }
1837                 if (ext4_has_feature_quota(sb)) {
1838                         ext4_msg(sb, KERN_INFO,
1839                                  "Quota format mount options ignored "
1840                                  "when QUOTA feature is enabled");
1841                         return 1;
1842                 }
1843                 sbi->s_jquota_fmt = m->mount_opt;
1844 #endif
1845         } else if (token == Opt_dax) {
1846 #ifdef CONFIG_FS_DAX
1847                 ext4_msg(sb, KERN_WARNING,
1848                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1849                         sbi->s_mount_opt |= m->mount_opt;
1850 #else
1851                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1852                 return -1;
1853 #endif
1854         } else if (token == Opt_data_err_abort) {
1855                 sbi->s_mount_opt |= m->mount_opt;
1856         } else if (token == Opt_data_err_ignore) {
1857                 sbi->s_mount_opt &= ~m->mount_opt;
1858         } else {
1859                 if (!args->from)
1860                         arg = 1;
1861                 if (m->flags & MOPT_CLEAR)
1862                         arg = !arg;
1863                 else if (unlikely(!(m->flags & MOPT_SET))) {
1864                         ext4_msg(sb, KERN_WARNING,
1865                                  "buggy handling of option %s", opt);
1866                         WARN_ON(1);
1867                         return -1;
1868                 }
1869                 if (arg != 0)
1870                         sbi->s_mount_opt |= m->mount_opt;
1871                 else
1872                         sbi->s_mount_opt &= ~m->mount_opt;
1873         }
1874         return 1;
1875 }
1876
1877 static int parse_options(char *options, struct super_block *sb,
1878                          unsigned long *journal_devnum,
1879                          unsigned int *journal_ioprio,
1880                          int is_remount)
1881 {
1882         struct ext4_sb_info *sbi = EXT4_SB(sb);
1883         char *p;
1884         substring_t args[MAX_OPT_ARGS];
1885         int token;
1886
1887         if (!options)
1888                 return 1;
1889
1890         while ((p = strsep(&options, ",")) != NULL) {
1891                 if (!*p)
1892                         continue;
1893                 /*
1894                  * Initialize args struct so we know whether arg was
1895                  * found; some options take optional arguments.
1896                  */
1897                 args[0].to = args[0].from = NULL;
1898                 token = match_token(p, tokens, args);
1899                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1900                                      journal_ioprio, is_remount) < 0)
1901                         return 0;
1902         }
1903 #ifdef CONFIG_QUOTA
1904         /*
1905          * We do the test below only for project quotas. 'usrquota' and
1906          * 'grpquota' mount options are allowed even without quota feature
1907          * to support legacy quotas in quota files.
1908          */
1909         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1910                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1911                          "Cannot enable project quota enforcement.");
1912                 return 0;
1913         }
1914         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1915                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1916                         clear_opt(sb, USRQUOTA);
1917
1918                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1919                         clear_opt(sb, GRPQUOTA);
1920
1921                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1922                         ext4_msg(sb, KERN_ERR, "old and new quota "
1923                                         "format mixing");
1924                         return 0;
1925                 }
1926
1927                 if (!sbi->s_jquota_fmt) {
1928                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1929                                         "not specified");
1930                         return 0;
1931                 }
1932         }
1933 #endif
1934         if (test_opt(sb, DIOREAD_NOLOCK)) {
1935                 int blocksize =
1936                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1937
1938                 if (blocksize < PAGE_SIZE) {
1939                         ext4_msg(sb, KERN_ERR, "can't mount with "
1940                                  "dioread_nolock if block size != PAGE_SIZE");
1941                         return 0;
1942                 }
1943         }
1944         return 1;
1945 }
1946
1947 static inline void ext4_show_quota_options(struct seq_file *seq,
1948                                            struct super_block *sb)
1949 {
1950 #if defined(CONFIG_QUOTA)
1951         struct ext4_sb_info *sbi = EXT4_SB(sb);
1952
1953         if (sbi->s_jquota_fmt) {
1954                 char *fmtname = "";
1955
1956                 switch (sbi->s_jquota_fmt) {
1957                 case QFMT_VFS_OLD:
1958                         fmtname = "vfsold";
1959                         break;
1960                 case QFMT_VFS_V0:
1961                         fmtname = "vfsv0";
1962                         break;
1963                 case QFMT_VFS_V1:
1964                         fmtname = "vfsv1";
1965                         break;
1966                 }
1967                 seq_printf(seq, ",jqfmt=%s", fmtname);
1968         }
1969
1970         if (sbi->s_qf_names[USRQUOTA])
1971                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1972
1973         if (sbi->s_qf_names[GRPQUOTA])
1974                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1975 #endif
1976 }
1977
1978 static const char *token2str(int token)
1979 {
1980         const struct match_token *t;
1981
1982         for (t = tokens; t->token != Opt_err; t++)
1983                 if (t->token == token && !strchr(t->pattern, '='))
1984                         break;
1985         return t->pattern;
1986 }
1987
1988 /*
1989  * Show an option if
1990  *  - it's set to a non-default value OR
1991  *  - if the per-sb default is different from the global default
1992  */
1993 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1994                               int nodefs)
1995 {
1996         struct ext4_sb_info *sbi = EXT4_SB(sb);
1997         struct ext4_super_block *es = sbi->s_es;
1998         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1999         const struct mount_opts *m;
2000         char sep = nodefs ? '\n' : ',';
2001
2002 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2003 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2004
2005         if (sbi->s_sb_block != 1)
2006                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2007
2008         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2009                 int want_set = m->flags & MOPT_SET;
2010                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2011                     (m->flags & MOPT_CLEAR_ERR))
2012                         continue;
2013                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2014                         continue; /* skip if same as the default */
2015                 if ((want_set &&
2016                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2017                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2018                         continue; /* select Opt_noFoo vs Opt_Foo */
2019                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2020         }
2021
2022         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2023             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2024                 SEQ_OPTS_PRINT("resuid=%u",
2025                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2026         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2027             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2028                 SEQ_OPTS_PRINT("resgid=%u",
2029                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2030         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2031         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2032                 SEQ_OPTS_PUTS("errors=remount-ro");
2033         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2034                 SEQ_OPTS_PUTS("errors=continue");
2035         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2036                 SEQ_OPTS_PUTS("errors=panic");
2037         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2038                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2039         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2040                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2041         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2042                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2043         if (sb->s_flags & MS_I_VERSION)
2044                 SEQ_OPTS_PUTS("i_version");
2045         if (nodefs || sbi->s_stripe)
2046                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2047         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2048                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2049                         SEQ_OPTS_PUTS("data=journal");
2050                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2051                         SEQ_OPTS_PUTS("data=ordered");
2052                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2053                         SEQ_OPTS_PUTS("data=writeback");
2054         }
2055         if (nodefs ||
2056             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2057                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2058                                sbi->s_inode_readahead_blks);
2059
2060         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2061                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2062                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2063         if (nodefs || sbi->s_max_dir_size_kb)
2064                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2065         if (test_opt(sb, DATA_ERR_ABORT))
2066                 SEQ_OPTS_PUTS("data_err=abort");
2067
2068         ext4_show_quota_options(seq, sb);
2069         return 0;
2070 }
2071
2072 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2073 {
2074         return _ext4_show_options(seq, root->d_sb, 0);
2075 }
2076
2077 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2078 {
2079         struct super_block *sb = seq->private;
2080         int rc;
2081
2082         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2083         rc = _ext4_show_options(seq, sb, 1);
2084         seq_puts(seq, "\n");
2085         return rc;
2086 }
2087
2088 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2089                             int read_only)
2090 {
2091         struct ext4_sb_info *sbi = EXT4_SB(sb);
2092         int res = 0;
2093
2094         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2095                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2096                          "forcing read-only mode");
2097                 res = MS_RDONLY;
2098         }
2099         if (read_only)
2100                 goto done;
2101         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2102                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2103                          "running e2fsck is recommended");
2104         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2105                 ext4_msg(sb, KERN_WARNING,
2106                          "warning: mounting fs with errors, "
2107                          "running e2fsck is recommended");
2108         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2109                  le16_to_cpu(es->s_mnt_count) >=
2110                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2111                 ext4_msg(sb, KERN_WARNING,
2112                          "warning: maximal mount count reached, "
2113                          "running e2fsck is recommended");
2114         else if (le32_to_cpu(es->s_checkinterval) &&
2115                 (le32_to_cpu(es->s_lastcheck) +
2116                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2117                 ext4_msg(sb, KERN_WARNING,
2118                          "warning: checktime reached, "
2119                          "running e2fsck is recommended");
2120         if (!sbi->s_journal)
2121                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2122         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2123                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2124         le16_add_cpu(&es->s_mnt_count, 1);
2125         es->s_mtime = cpu_to_le32(get_seconds());
2126         ext4_update_dynamic_rev(sb);
2127         if (sbi->s_journal)
2128                 ext4_set_feature_journal_needs_recovery(sb);
2129
2130         ext4_commit_super(sb, 1);
2131 done:
2132         if (test_opt(sb, DEBUG))
2133                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2134                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2135                         sb->s_blocksize,
2136                         sbi->s_groups_count,
2137                         EXT4_BLOCKS_PER_GROUP(sb),
2138                         EXT4_INODES_PER_GROUP(sb),
2139                         sbi->s_mount_opt, sbi->s_mount_opt2);
2140
2141         cleancache_init_fs(sb);
2142         return res;
2143 }
2144
2145 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2146 {
2147         struct ext4_sb_info *sbi = EXT4_SB(sb);
2148         struct flex_groups *new_groups;
2149         int size;
2150
2151         if (!sbi->s_log_groups_per_flex)
2152                 return 0;
2153
2154         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2155         if (size <= sbi->s_flex_groups_allocated)
2156                 return 0;
2157
2158         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2159         new_groups = kvzalloc(size, GFP_KERNEL);
2160         if (!new_groups) {
2161                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2162                          size / (int) sizeof(struct flex_groups));
2163                 return -ENOMEM;
2164         }
2165
2166         if (sbi->s_flex_groups) {
2167                 memcpy(new_groups, sbi->s_flex_groups,
2168                        (sbi->s_flex_groups_allocated *
2169                         sizeof(struct flex_groups)));
2170                 kvfree(sbi->s_flex_groups);
2171         }
2172         sbi->s_flex_groups = new_groups;
2173         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2174         return 0;
2175 }
2176
2177 static int ext4_fill_flex_info(struct super_block *sb)
2178 {
2179         struct ext4_sb_info *sbi = EXT4_SB(sb);
2180         struct ext4_group_desc *gdp = NULL;
2181         ext4_group_t flex_group;
2182         int i, err;
2183
2184         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2185         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2186                 sbi->s_log_groups_per_flex = 0;
2187                 return 1;
2188         }
2189
2190         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2191         if (err)
2192                 goto failed;
2193
2194         for (i = 0; i < sbi->s_groups_count; i++) {
2195                 gdp = ext4_get_group_desc(sb, i, NULL);
2196
2197                 flex_group = ext4_flex_group(sbi, i);
2198                 atomic_add(ext4_free_inodes_count(sb, gdp),
2199                            &sbi->s_flex_groups[flex_group].free_inodes);
2200                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2201                              &sbi->s_flex_groups[flex_group].free_clusters);
2202                 atomic_add(ext4_used_dirs_count(sb, gdp),
2203                            &sbi->s_flex_groups[flex_group].used_dirs);
2204         }
2205
2206         return 1;
2207 failed:
2208         return 0;
2209 }
2210
2211 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2212                                    struct ext4_group_desc *gdp)
2213 {
2214         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2215         __u16 crc = 0;
2216         __le32 le_group = cpu_to_le32(block_group);
2217         struct ext4_sb_info *sbi = EXT4_SB(sb);
2218
2219         if (ext4_has_metadata_csum(sbi->s_sb)) {
2220                 /* Use new metadata_csum algorithm */
2221                 __u32 csum32;
2222                 __u16 dummy_csum = 0;
2223
2224                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2225                                      sizeof(le_group));
2226                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2227                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2228                                      sizeof(dummy_csum));
2229                 offset += sizeof(dummy_csum);
2230                 if (offset < sbi->s_desc_size)
2231                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2232                                              sbi->s_desc_size - offset);
2233
2234                 crc = csum32 & 0xFFFF;
2235                 goto out;
2236         }
2237
2238         /* old crc16 code */
2239         if (!ext4_has_feature_gdt_csum(sb))
2240                 return 0;
2241
2242         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2243         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2244         crc = crc16(crc, (__u8 *)gdp, offset);
2245         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2246         /* for checksum of struct ext4_group_desc do the rest...*/
2247         if (ext4_has_feature_64bit(sb) &&
2248             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2249                 crc = crc16(crc, (__u8 *)gdp + offset,
2250                             le16_to_cpu(sbi->s_es->s_desc_size) -
2251                                 offset);
2252
2253 out:
2254         return cpu_to_le16(crc);
2255 }
2256
2257 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2258                                 struct ext4_group_desc *gdp)
2259 {
2260         if (ext4_has_group_desc_csum(sb) &&
2261             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2262                 return 0;
2263
2264         return 1;
2265 }
2266
2267 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2268                               struct ext4_group_desc *gdp)
2269 {
2270         if (!ext4_has_group_desc_csum(sb))
2271                 return;
2272         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2273 }
2274
2275 /* Called at mount-time, super-block is locked */
2276 static int ext4_check_descriptors(struct super_block *sb,
2277                                   ext4_fsblk_t sb_block,
2278                                   ext4_group_t *first_not_zeroed)
2279 {
2280         struct ext4_sb_info *sbi = EXT4_SB(sb);
2281         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2282         ext4_fsblk_t last_block;
2283         ext4_fsblk_t block_bitmap;
2284         ext4_fsblk_t inode_bitmap;
2285         ext4_fsblk_t inode_table;
2286         int flexbg_flag = 0;
2287         ext4_group_t i, grp = sbi->s_groups_count;
2288
2289         if (ext4_has_feature_flex_bg(sb))
2290                 flexbg_flag = 1;
2291
2292         ext4_debug("Checking group descriptors");
2293
2294         for (i = 0; i < sbi->s_groups_count; i++) {
2295                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2296
2297                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2298                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2299                 else
2300                         last_block = first_block +
2301                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2302
2303                 if ((grp == sbi->s_groups_count) &&
2304                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2305                         grp = i;
2306
2307                 block_bitmap = ext4_block_bitmap(sb, gdp);
2308                 if (block_bitmap == sb_block) {
2309                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2310                                  "Block bitmap for group %u overlaps "
2311                                  "superblock", i);
2312                 }
2313                 if (block_bitmap < first_block || block_bitmap > last_block) {
2314                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2315                                "Block bitmap for group %u not in group "
2316                                "(block %llu)!", i, block_bitmap);
2317                         return 0;
2318                 }
2319                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2320                 if (inode_bitmap == sb_block) {
2321                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2322                                  "Inode bitmap for group %u overlaps "
2323                                  "superblock", i);
2324                 }
2325                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2326                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2327                                "Inode bitmap for group %u not in group "
2328                                "(block %llu)!", i, inode_bitmap);
2329                         return 0;
2330                 }
2331                 inode_table = ext4_inode_table(sb, gdp);
2332                 if (inode_table == sb_block) {
2333                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2334                                  "Inode table for group %u overlaps "
2335                                  "superblock", i);
2336                 }
2337                 if (inode_table < first_block ||
2338                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2339                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2340                                "Inode table for group %u not in group "
2341                                "(block %llu)!", i, inode_table);
2342                         return 0;
2343                 }
2344                 ext4_lock_group(sb, i);
2345                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2346                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2347                                  "Checksum for group %u failed (%u!=%u)",
2348                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2349                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2350                         if (!(sb->s_flags & MS_RDONLY)) {
2351                                 ext4_unlock_group(sb, i);
2352                                 return 0;
2353                         }
2354                 }
2355                 ext4_unlock_group(sb, i);
2356                 if (!flexbg_flag)
2357                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2358         }
2359         if (NULL != first_not_zeroed)
2360                 *first_not_zeroed = grp;
2361         return 1;
2362 }
2363
2364 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2365  * the superblock) which were deleted from all directories, but held open by
2366  * a process at the time of a crash.  We walk the list and try to delete these
2367  * inodes at recovery time (only with a read-write filesystem).
2368  *
2369  * In order to keep the orphan inode chain consistent during traversal (in
2370  * case of crash during recovery), we link each inode into the superblock
2371  * orphan list_head and handle it the same way as an inode deletion during
2372  * normal operation (which journals the operations for us).
2373  *
2374  * We only do an iget() and an iput() on each inode, which is very safe if we
2375  * accidentally point at an in-use or already deleted inode.  The worst that
2376  * can happen in this case is that we get a "bit already cleared" message from
2377  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2378  * e2fsck was run on this filesystem, and it must have already done the orphan
2379  * inode cleanup for us, so we can safely abort without any further action.
2380  */
2381 static void ext4_orphan_cleanup(struct super_block *sb,
2382                                 struct ext4_super_block *es)
2383 {
2384         unsigned int s_flags = sb->s_flags;
2385         int ret, nr_orphans = 0, nr_truncates = 0;
2386 #ifdef CONFIG_QUOTA
2387         int i;
2388 #endif
2389         if (!es->s_last_orphan) {
2390                 jbd_debug(4, "no orphan inodes to clean up\n");
2391                 return;
2392         }
2393
2394         if (bdev_read_only(sb->s_bdev)) {
2395                 ext4_msg(sb, KERN_ERR, "write access "
2396                         "unavailable, skipping orphan cleanup");
2397                 return;
2398         }
2399
2400         /* Check if feature set would not allow a r/w mount */
2401         if (!ext4_feature_set_ok(sb, 0)) {
2402                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2403                          "unknown ROCOMPAT features");
2404                 return;
2405         }
2406
2407         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2408                 /* don't clear list on RO mount w/ errors */
2409                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2410                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2411                                   "clearing orphan list.\n");
2412                         es->s_last_orphan = 0;
2413                 }
2414                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2415                 return;
2416         }
2417
2418         if (s_flags & MS_RDONLY) {
2419                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2420                 sb->s_flags &= ~MS_RDONLY;
2421         }
2422 #ifdef CONFIG_QUOTA
2423         /* Needed for iput() to work correctly and not trash data */
2424         sb->s_flags |= MS_ACTIVE;
2425         /* Turn on quotas so that they are updated correctly */
2426         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2427                 if (EXT4_SB(sb)->s_qf_names[i]) {
2428                         int ret = ext4_quota_on_mount(sb, i);
2429                         if (ret < 0)
2430                                 ext4_msg(sb, KERN_ERR,
2431                                         "Cannot turn on journaled "
2432                                         "quota: error %d", ret);
2433                 }
2434         }
2435 #endif
2436
2437         while (es->s_last_orphan) {
2438                 struct inode *inode;
2439
2440                 /*
2441                  * We may have encountered an error during cleanup; if
2442                  * so, skip the rest.
2443                  */
2444                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2445                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2446                         es->s_last_orphan = 0;
2447                         break;
2448                 }
2449
2450                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2451                 if (IS_ERR(inode)) {
2452                         es->s_last_orphan = 0;
2453                         break;
2454                 }
2455
2456                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2457                 dquot_initialize(inode);
2458                 if (inode->i_nlink) {
2459                         if (test_opt(sb, DEBUG))
2460                                 ext4_msg(sb, KERN_DEBUG,
2461                                         "%s: truncating inode %lu to %lld bytes",
2462                                         __func__, inode->i_ino, inode->i_size);
2463                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2464                                   inode->i_ino, inode->i_size);
2465                         inode_lock(inode);
2466                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2467                         ret = ext4_truncate(inode);
2468                         if (ret)
2469                                 ext4_std_error(inode->i_sb, ret);
2470                         inode_unlock(inode);
2471                         nr_truncates++;
2472                 } else {
2473                         if (test_opt(sb, DEBUG))
2474                                 ext4_msg(sb, KERN_DEBUG,
2475                                         "%s: deleting unreferenced inode %lu",
2476                                         __func__, inode->i_ino);
2477                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2478                                   inode->i_ino);
2479                         nr_orphans++;
2480                 }
2481                 iput(inode);  /* The delete magic happens here! */
2482         }
2483
2484 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2485
2486         if (nr_orphans)
2487                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2488                        PLURAL(nr_orphans));
2489         if (nr_truncates)
2490                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2491                        PLURAL(nr_truncates));
2492 #ifdef CONFIG_QUOTA
2493         /* Turn quotas off */
2494         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2495                 if (sb_dqopt(sb)->files[i])
2496                         dquot_quota_off(sb, i);
2497         }
2498 #endif
2499         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2500 }
2501
2502 /*
2503  * Maximal extent format file size.
2504  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2505  * extent format containers, within a sector_t, and within i_blocks
2506  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2507  * so that won't be a limiting factor.
2508  *
2509  * However there is other limiting factor. We do store extents in the form
2510  * of starting block and length, hence the resulting length of the extent
2511  * covering maximum file size must fit into on-disk format containers as
2512  * well. Given that length is always by 1 unit bigger than max unit (because
2513  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2514  *
2515  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2516  */
2517 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2518 {
2519         loff_t res;
2520         loff_t upper_limit = MAX_LFS_FILESIZE;
2521
2522         /* small i_blocks in vfs inode? */
2523         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2524                 /*
2525                  * CONFIG_LBDAF is not enabled implies the inode
2526                  * i_block represent total blocks in 512 bytes
2527                  * 32 == size of vfs inode i_blocks * 8
2528                  */
2529                 upper_limit = (1LL << 32) - 1;
2530
2531                 /* total blocks in file system block size */
2532                 upper_limit >>= (blkbits - 9);
2533                 upper_limit <<= blkbits;
2534         }
2535
2536         /*
2537          * 32-bit extent-start container, ee_block. We lower the maxbytes
2538          * by one fs block, so ee_len can cover the extent of maximum file
2539          * size
2540          */
2541         res = (1LL << 32) - 1;
2542         res <<= blkbits;
2543
2544         /* Sanity check against vm- & vfs- imposed limits */
2545         if (res > upper_limit)
2546                 res = upper_limit;
2547
2548         return res;
2549 }
2550
2551 /*
2552  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2553  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2554  * We need to be 1 filesystem block less than the 2^48 sector limit.
2555  */
2556 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2557 {
2558         loff_t res = EXT4_NDIR_BLOCKS;
2559         int meta_blocks;
2560         loff_t upper_limit;
2561         /* This is calculated to be the largest file size for a dense, block
2562          * mapped file such that the file's total number of 512-byte sectors,
2563          * including data and all indirect blocks, does not exceed (2^48 - 1).
2564          *
2565          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2566          * number of 512-byte sectors of the file.
2567          */
2568
2569         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2570                 /*
2571                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2572                  * the inode i_block field represents total file blocks in
2573                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2574                  */
2575                 upper_limit = (1LL << 32) - 1;
2576
2577                 /* total blocks in file system block size */
2578                 upper_limit >>= (bits - 9);
2579
2580         } else {
2581                 /*
2582                  * We use 48 bit ext4_inode i_blocks
2583                  * With EXT4_HUGE_FILE_FL set the i_blocks
2584                  * represent total number of blocks in
2585                  * file system block size
2586                  */
2587                 upper_limit = (1LL << 48) - 1;
2588
2589         }
2590
2591         /* indirect blocks */
2592         meta_blocks = 1;
2593         /* double indirect blocks */
2594         meta_blocks += 1 + (1LL << (bits-2));
2595         /* tripple indirect blocks */
2596         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2597
2598         upper_limit -= meta_blocks;
2599         upper_limit <<= bits;
2600
2601         res += 1LL << (bits-2);
2602         res += 1LL << (2*(bits-2));
2603         res += 1LL << (3*(bits-2));
2604         res <<= bits;
2605         if (res > upper_limit)
2606                 res = upper_limit;
2607
2608         if (res > MAX_LFS_FILESIZE)
2609                 res = MAX_LFS_FILESIZE;
2610
2611         return res;
2612 }
2613
2614 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2615                                    ext4_fsblk_t logical_sb_block, int nr)
2616 {
2617         struct ext4_sb_info *sbi = EXT4_SB(sb);
2618         ext4_group_t bg, first_meta_bg;
2619         int has_super = 0;
2620
2621         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2622
2623         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2624                 return logical_sb_block + nr + 1;
2625         bg = sbi->s_desc_per_block * nr;
2626         if (ext4_bg_has_super(sb, bg))
2627                 has_super = 1;
2628
2629         /*
2630          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2631          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2632          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2633          * compensate.
2634          */
2635         if (sb->s_blocksize == 1024 && nr == 0 &&
2636             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2637                 has_super++;
2638
2639         return (has_super + ext4_group_first_block_no(sb, bg));
2640 }
2641
2642 /**
2643  * ext4_get_stripe_size: Get the stripe size.
2644  * @sbi: In memory super block info
2645  *
2646  * If we have specified it via mount option, then
2647  * use the mount option value. If the value specified at mount time is
2648  * greater than the blocks per group use the super block value.
2649  * If the super block value is greater than blocks per group return 0.
2650  * Allocator needs it be less than blocks per group.
2651  *
2652  */
2653 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2654 {
2655         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2656         unsigned long stripe_width =
2657                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2658         int ret;
2659
2660         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2661                 ret = sbi->s_stripe;
2662         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2663                 ret = stripe_width;
2664         else if (stride && stride <= sbi->s_blocks_per_group)
2665                 ret = stride;
2666         else
2667                 ret = 0;
2668
2669         /*
2670          * If the stripe width is 1, this makes no sense and
2671          * we set it to 0 to turn off stripe handling code.
2672          */
2673         if (ret <= 1)
2674                 ret = 0;
2675
2676         return ret;
2677 }
2678
2679 /*
2680  * Check whether this filesystem can be mounted based on
2681  * the features present and the RDONLY/RDWR mount requested.
2682  * Returns 1 if this filesystem can be mounted as requested,
2683  * 0 if it cannot be.
2684  */
2685 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2686 {
2687         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2688                 ext4_msg(sb, KERN_ERR,
2689                         "Couldn't mount because of "
2690                         "unsupported optional features (%x)",
2691                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2692                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2693                 return 0;
2694         }
2695
2696         if (readonly)
2697                 return 1;
2698
2699         if (ext4_has_feature_readonly(sb)) {
2700                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2701                 sb->s_flags |= MS_RDONLY;
2702                 return 1;
2703         }
2704
2705         /* Check that feature set is OK for a read-write mount */
2706         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2707                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2708                          "unsupported optional features (%x)",
2709                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2710                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2711                 return 0;
2712         }
2713         /*
2714          * Large file size enabled file system can only be mounted
2715          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2716          */
2717         if (ext4_has_feature_huge_file(sb)) {
2718                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2719                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2720                                  "cannot be mounted RDWR without "
2721                                  "CONFIG_LBDAF");
2722                         return 0;
2723                 }
2724         }
2725         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2726                 ext4_msg(sb, KERN_ERR,
2727                          "Can't support bigalloc feature without "
2728                          "extents feature\n");
2729                 return 0;
2730         }
2731
2732 #ifndef CONFIG_QUOTA
2733         if (ext4_has_feature_quota(sb) && !readonly) {
2734                 ext4_msg(sb, KERN_ERR,
2735                          "Filesystem with quota feature cannot be mounted RDWR "
2736                          "without CONFIG_QUOTA");
2737                 return 0;
2738         }
2739         if (ext4_has_feature_project(sb) && !readonly) {
2740                 ext4_msg(sb, KERN_ERR,
2741                          "Filesystem with project quota feature cannot be mounted RDWR "
2742                          "without CONFIG_QUOTA");
2743                 return 0;
2744         }
2745 #endif  /* CONFIG_QUOTA */
2746         return 1;
2747 }
2748
2749 /*
2750  * This function is called once a day if we have errors logged
2751  * on the file system
2752  */
2753 static void print_daily_error_info(unsigned long arg)
2754 {
2755         struct super_block *sb = (struct super_block *) arg;
2756         struct ext4_sb_info *sbi;
2757         struct ext4_super_block *es;
2758
2759         sbi = EXT4_SB(sb);
2760         es = sbi->s_es;
2761
2762         if (es->s_error_count)
2763                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2764                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2765                          le32_to_cpu(es->s_error_count));
2766         if (es->s_first_error_time) {
2767                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2768                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2769                        (int) sizeof(es->s_first_error_func),
2770                        es->s_first_error_func,
2771                        le32_to_cpu(es->s_first_error_line));
2772                 if (es->s_first_error_ino)
2773                         printk(KERN_CONT ": inode %u",
2774                                le32_to_cpu(es->s_first_error_ino));
2775                 if (es->s_first_error_block)
2776                         printk(KERN_CONT ": block %llu", (unsigned long long)
2777                                le64_to_cpu(es->s_first_error_block));
2778                 printk(KERN_CONT "\n");
2779         }
2780         if (es->s_last_error_time) {
2781                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2782                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2783                        (int) sizeof(es->s_last_error_func),
2784                        es->s_last_error_func,
2785                        le32_to_cpu(es->s_last_error_line));
2786                 if (es->s_last_error_ino)
2787                         printk(KERN_CONT ": inode %u",
2788                                le32_to_cpu(es->s_last_error_ino));
2789                 if (es->s_last_error_block)
2790                         printk(KERN_CONT ": block %llu", (unsigned long long)
2791                                le64_to_cpu(es->s_last_error_block));
2792                 printk(KERN_CONT "\n");
2793         }
2794         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2795 }
2796
2797 /* Find next suitable group and run ext4_init_inode_table */
2798 static int ext4_run_li_request(struct ext4_li_request *elr)
2799 {
2800         struct ext4_group_desc *gdp = NULL;
2801         ext4_group_t group, ngroups;
2802         struct super_block *sb;
2803         unsigned long timeout = 0;
2804         int ret = 0;
2805
2806         sb = elr->lr_super;
2807         ngroups = EXT4_SB(sb)->s_groups_count;
2808
2809         for (group = elr->lr_next_group; group < ngroups; group++) {
2810                 gdp = ext4_get_group_desc(sb, group, NULL);
2811                 if (!gdp) {
2812                         ret = 1;
2813                         break;
2814                 }
2815
2816                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2817                         break;
2818         }
2819
2820         if (group >= ngroups)
2821                 ret = 1;
2822
2823         if (!ret) {
2824                 timeout = jiffies;
2825                 ret = ext4_init_inode_table(sb, group,
2826                                             elr->lr_timeout ? 0 : 1);
2827                 if (elr->lr_timeout == 0) {
2828                         timeout = (jiffies - timeout) *
2829                                   elr->lr_sbi->s_li_wait_mult;
2830                         elr->lr_timeout = timeout;
2831                 }
2832                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2833                 elr->lr_next_group = group + 1;
2834         }
2835         return ret;
2836 }
2837
2838 /*
2839  * Remove lr_request from the list_request and free the
2840  * request structure. Should be called with li_list_mtx held
2841  */
2842 static void ext4_remove_li_request(struct ext4_li_request *elr)
2843 {
2844         struct ext4_sb_info *sbi;
2845
2846         if (!elr)
2847                 return;
2848
2849         sbi = elr->lr_sbi;
2850
2851         list_del(&elr->lr_request);
2852         sbi->s_li_request = NULL;
2853         kfree(elr);
2854 }
2855
2856 static void ext4_unregister_li_request(struct super_block *sb)
2857 {
2858         mutex_lock(&ext4_li_mtx);
2859         if (!ext4_li_info) {
2860                 mutex_unlock(&ext4_li_mtx);
2861                 return;
2862         }
2863
2864         mutex_lock(&ext4_li_info->li_list_mtx);
2865         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2866         mutex_unlock(&ext4_li_info->li_list_mtx);
2867         mutex_unlock(&ext4_li_mtx);
2868 }
2869
2870 static struct task_struct *ext4_lazyinit_task;
2871
2872 /*
2873  * This is the function where ext4lazyinit thread lives. It walks
2874  * through the request list searching for next scheduled filesystem.
2875  * When such a fs is found, run the lazy initialization request
2876  * (ext4_rn_li_request) and keep track of the time spend in this
2877  * function. Based on that time we compute next schedule time of
2878  * the request. When walking through the list is complete, compute
2879  * next waking time and put itself into sleep.
2880  */
2881 static int ext4_lazyinit_thread(void *arg)
2882 {
2883         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2884         struct list_head *pos, *n;
2885         struct ext4_li_request *elr;
2886         unsigned long next_wakeup, cur;
2887
2888         BUG_ON(NULL == eli);
2889
2890 cont_thread:
2891         while (true) {
2892                 next_wakeup = MAX_JIFFY_OFFSET;
2893
2894                 mutex_lock(&eli->li_list_mtx);
2895                 if (list_empty(&eli->li_request_list)) {
2896                         mutex_unlock(&eli->li_list_mtx);
2897                         goto exit_thread;
2898                 }
2899                 list_for_each_safe(pos, n, &eli->li_request_list) {
2900                         int err = 0;
2901                         int progress = 0;
2902                         elr = list_entry(pos, struct ext4_li_request,
2903                                          lr_request);
2904
2905                         if (time_before(jiffies, elr->lr_next_sched)) {
2906                                 if (time_before(elr->lr_next_sched, next_wakeup))
2907                                         next_wakeup = elr->lr_next_sched;
2908                                 continue;
2909                         }
2910                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2911                                 if (sb_start_write_trylock(elr->lr_super)) {
2912                                         progress = 1;
2913                                         /*
2914                                          * We hold sb->s_umount, sb can not
2915                                          * be removed from the list, it is
2916                                          * now safe to drop li_list_mtx
2917                                          */
2918                                         mutex_unlock(&eli->li_list_mtx);
2919                                         err = ext4_run_li_request(elr);
2920                                         sb_end_write(elr->lr_super);
2921                                         mutex_lock(&eli->li_list_mtx);
2922                                         n = pos->next;
2923                                 }
2924                                 up_read((&elr->lr_super->s_umount));
2925                         }
2926                         /* error, remove the lazy_init job */
2927                         if (err) {
2928                                 ext4_remove_li_request(elr);
2929                                 continue;
2930                         }
2931                         if (!progress) {
2932                                 elr->lr_next_sched = jiffies +
2933                                         (prandom_u32()
2934                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2935                         }
2936                         if (time_before(elr->lr_next_sched, next_wakeup))
2937                                 next_wakeup = elr->lr_next_sched;
2938                 }
2939                 mutex_unlock(&eli->li_list_mtx);
2940
2941                 try_to_freeze();
2942
2943                 cur = jiffies;
2944                 if ((time_after_eq(cur, next_wakeup)) ||
2945                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2946                         cond_resched();
2947                         continue;
2948                 }
2949
2950                 schedule_timeout_interruptible(next_wakeup - cur);
2951
2952                 if (kthread_should_stop()) {
2953                         ext4_clear_request_list();
2954                         goto exit_thread;
2955                 }
2956         }
2957
2958 exit_thread:
2959         /*
2960          * It looks like the request list is empty, but we need
2961          * to check it under the li_list_mtx lock, to prevent any
2962          * additions into it, and of course we should lock ext4_li_mtx
2963          * to atomically free the list and ext4_li_info, because at
2964          * this point another ext4 filesystem could be registering
2965          * new one.
2966          */
2967         mutex_lock(&ext4_li_mtx);
2968         mutex_lock(&eli->li_list_mtx);
2969         if (!list_empty(&eli->li_request_list)) {
2970                 mutex_unlock(&eli->li_list_mtx);
2971                 mutex_unlock(&ext4_li_mtx);
2972                 goto cont_thread;
2973         }
2974         mutex_unlock(&eli->li_list_mtx);
2975         kfree(ext4_li_info);
2976         ext4_li_info = NULL;
2977         mutex_unlock(&ext4_li_mtx);
2978
2979         return 0;
2980 }
2981
2982 static void ext4_clear_request_list(void)
2983 {
2984         struct list_head *pos, *n;
2985         struct ext4_li_request *elr;
2986
2987         mutex_lock(&ext4_li_info->li_list_mtx);
2988         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2989                 elr = list_entry(pos, struct ext4_li_request,
2990                                  lr_request);
2991                 ext4_remove_li_request(elr);
2992         }
2993         mutex_unlock(&ext4_li_info->li_list_mtx);
2994 }
2995
2996 static int ext4_run_lazyinit_thread(void)
2997 {
2998         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2999                                          ext4_li_info, "ext4lazyinit");
3000         if (IS_ERR(ext4_lazyinit_task)) {
3001                 int err = PTR_ERR(ext4_lazyinit_task);
3002                 ext4_clear_request_list();
3003                 kfree(ext4_li_info);
3004                 ext4_li_info = NULL;
3005                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3006                                  "initialization thread\n",
3007                                  err);
3008                 return err;
3009         }
3010         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3011         return 0;
3012 }
3013
3014 /*
3015  * Check whether it make sense to run itable init. thread or not.
3016  * If there is at least one uninitialized inode table, return
3017  * corresponding group number, else the loop goes through all
3018  * groups and return total number of groups.
3019  */
3020 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3021 {
3022         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3023         struct ext4_group_desc *gdp = NULL;
3024
3025         for (group = 0; group < ngroups; group++) {
3026                 gdp = ext4_get_group_desc(sb, group, NULL);
3027                 if (!gdp)
3028                         continue;
3029
3030                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3031                         break;
3032         }
3033
3034         return group;
3035 }
3036
3037 static int ext4_li_info_new(void)
3038 {
3039         struct ext4_lazy_init *eli = NULL;
3040
3041         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3042         if (!eli)
3043                 return -ENOMEM;
3044
3045         INIT_LIST_HEAD(&eli->li_request_list);
3046         mutex_init(&eli->li_list_mtx);
3047
3048         eli->li_state |= EXT4_LAZYINIT_QUIT;
3049
3050         ext4_li_info = eli;
3051
3052         return 0;
3053 }
3054
3055 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3056                                             ext4_group_t start)
3057 {
3058         struct ext4_sb_info *sbi = EXT4_SB(sb);
3059         struct ext4_li_request *elr;
3060
3061         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3062         if (!elr)
3063                 return NULL;
3064
3065         elr->lr_super = sb;
3066         elr->lr_sbi = sbi;
3067         elr->lr_next_group = start;
3068
3069         /*
3070          * Randomize first schedule time of the request to
3071          * spread the inode table initialization requests
3072          * better.
3073          */
3074         elr->lr_next_sched = jiffies + (prandom_u32() %
3075                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3076         return elr;
3077 }
3078
3079 int ext4_register_li_request(struct super_block *sb,
3080                              ext4_group_t first_not_zeroed)
3081 {
3082         struct ext4_sb_info *sbi = EXT4_SB(sb);
3083         struct ext4_li_request *elr = NULL;
3084         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3085         int ret = 0;
3086
3087         mutex_lock(&ext4_li_mtx);
3088         if (sbi->s_li_request != NULL) {
3089                 /*
3090                  * Reset timeout so it can be computed again, because
3091                  * s_li_wait_mult might have changed.
3092                  */
3093                 sbi->s_li_request->lr_timeout = 0;
3094                 goto out;
3095         }
3096
3097         if (first_not_zeroed == ngroups ||
3098             (sb->s_flags & MS_RDONLY) ||
3099             !test_opt(sb, INIT_INODE_TABLE))
3100                 goto out;
3101
3102         elr = ext4_li_request_new(sb, first_not_zeroed);
3103         if (!elr) {
3104                 ret = -ENOMEM;
3105                 goto out;
3106         }
3107
3108         if (NULL == ext4_li_info) {
3109                 ret = ext4_li_info_new();
3110                 if (ret)
3111                         goto out;
3112         }
3113
3114         mutex_lock(&ext4_li_info->li_list_mtx);
3115         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3116         mutex_unlock(&ext4_li_info->li_list_mtx);
3117
3118         sbi->s_li_request = elr;
3119         /*
3120          * set elr to NULL here since it has been inserted to
3121          * the request_list and the removal and free of it is
3122          * handled by ext4_clear_request_list from now on.
3123          */
3124         elr = NULL;
3125
3126         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3127                 ret = ext4_run_lazyinit_thread();
3128                 if (ret)
3129                         goto out;
3130         }
3131 out:
3132         mutex_unlock(&ext4_li_mtx);
3133         if (ret)
3134                 kfree(elr);
3135         return ret;
3136 }
3137
3138 /*
3139  * We do not need to lock anything since this is called on
3140  * module unload.
3141  */
3142 static void ext4_destroy_lazyinit_thread(void)
3143 {
3144         /*
3145          * If thread exited earlier
3146          * there's nothing to be done.
3147          */
3148         if (!ext4_li_info || !ext4_lazyinit_task)
3149                 return;
3150
3151         kthread_stop(ext4_lazyinit_task);
3152 }
3153
3154 static int set_journal_csum_feature_set(struct super_block *sb)
3155 {
3156         int ret = 1;
3157         int compat, incompat;
3158         struct ext4_sb_info *sbi = EXT4_SB(sb);
3159
3160         if (ext4_has_metadata_csum(sb)) {
3161                 /* journal checksum v3 */
3162                 compat = 0;
3163                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3164         } else {
3165                 /* journal checksum v1 */
3166                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3167                 incompat = 0;
3168         }
3169
3170         jbd2_journal_clear_features(sbi->s_journal,
3171                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3172                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3173                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3174         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3175                 ret = jbd2_journal_set_features(sbi->s_journal,
3176                                 compat, 0,
3177                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3178                                 incompat);
3179         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3180                 ret = jbd2_journal_set_features(sbi->s_journal,
3181                                 compat, 0,
3182                                 incompat);
3183                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3184                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3185         } else {
3186                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3187                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3188         }
3189
3190         return ret;
3191 }
3192
3193 /*
3194  * Note: calculating the overhead so we can be compatible with
3195  * historical BSD practice is quite difficult in the face of
3196  * clusters/bigalloc.  This is because multiple metadata blocks from
3197  * different block group can end up in the same allocation cluster.
3198  * Calculating the exact overhead in the face of clustered allocation
3199  * requires either O(all block bitmaps) in memory or O(number of block
3200  * groups**2) in time.  We will still calculate the superblock for
3201  * older file systems --- and if we come across with a bigalloc file
3202  * system with zero in s_overhead_clusters the estimate will be close to
3203  * correct especially for very large cluster sizes --- but for newer
3204  * file systems, it's better to calculate this figure once at mkfs
3205  * time, and store it in the superblock.  If the superblock value is
3206  * present (even for non-bigalloc file systems), we will use it.
3207  */
3208 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3209                           char *buf)
3210 {
3211         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3212         struct ext4_group_desc  *gdp;
3213         ext4_fsblk_t            first_block, last_block, b;
3214         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3215         int                     s, j, count = 0;
3216
3217         if (!ext4_has_feature_bigalloc(sb))
3218                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3219                         sbi->s_itb_per_group + 2);
3220
3221         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3222                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3223         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3224         for (i = 0; i < ngroups; i++) {
3225                 gdp = ext4_get_group_desc(sb, i, NULL);
3226                 b = ext4_block_bitmap(sb, gdp);
3227                 if (b >= first_block && b <= last_block) {
3228                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3229                         count++;
3230                 }
3231                 b = ext4_inode_bitmap(sb, gdp);
3232                 if (b >= first_block && b <= last_block) {
3233                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3234                         count++;
3235                 }
3236                 b = ext4_inode_table(sb, gdp);
3237                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3238                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3239                                 int c = EXT4_B2C(sbi, b - first_block);
3240                                 ext4_set_bit(c, buf);
3241                                 count++;
3242                         }
3243                 if (i != grp)
3244                         continue;
3245                 s = 0;
3246                 if (ext4_bg_has_super(sb, grp)) {
3247                         ext4_set_bit(s++, buf);
3248                         count++;
3249                 }
3250                 j = ext4_bg_num_gdb(sb, grp);
3251                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3252                         ext4_error(sb, "Invalid number of block group "
3253                                    "descriptor blocks: %d", j);
3254                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3255                 }
3256                 count += j;
3257                 for (; j > 0; j--)
3258                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3259         }
3260         if (!count)
3261                 return 0;
3262         return EXT4_CLUSTERS_PER_GROUP(sb) -
3263                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3264 }
3265
3266 /*
3267  * Compute the overhead and stash it in sbi->s_overhead
3268  */
3269 int ext4_calculate_overhead(struct super_block *sb)
3270 {
3271         struct ext4_sb_info *sbi = EXT4_SB(sb);
3272         struct ext4_super_block *es = sbi->s_es;
3273         struct inode *j_inode;
3274         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3275         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3276         ext4_fsblk_t overhead = 0;
3277         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3278
3279         if (!buf)
3280                 return -ENOMEM;
3281
3282         /*
3283          * Compute the overhead (FS structures).  This is constant
3284          * for a given filesystem unless the number of block groups
3285          * changes so we cache the previous value until it does.
3286          */
3287
3288         /*
3289          * All of the blocks before first_data_block are overhead
3290          */
3291         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3292
3293         /*
3294          * Add the overhead found in each block group
3295          */
3296         for (i = 0; i < ngroups; i++) {
3297                 int blks;
3298
3299                 blks = count_overhead(sb, i, buf);
3300                 overhead += blks;
3301                 if (blks)
3302                         memset(buf, 0, PAGE_SIZE);
3303                 cond_resched();
3304         }
3305
3306         /*
3307          * Add the internal journal blocks whether the journal has been
3308          * loaded or not
3309          */
3310         if (sbi->s_journal && !sbi->journal_bdev)
3311                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3312         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3313                 j_inode = ext4_get_journal_inode(sb, j_inum);
3314                 if (j_inode) {
3315                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3316                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3317                         iput(j_inode);
3318                 } else {
3319                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3320                 }
3321         }
3322         sbi->s_overhead = overhead;
3323         smp_wmb();
3324         free_page((unsigned long) buf);
3325         return 0;
3326 }
3327
3328 static void ext4_set_resv_clusters(struct super_block *sb)
3329 {
3330         ext4_fsblk_t resv_clusters;
3331         struct ext4_sb_info *sbi = EXT4_SB(sb);
3332
3333         /*
3334          * There's no need to reserve anything when we aren't using extents.
3335          * The space estimates are exact, there are no unwritten extents,
3336          * hole punching doesn't need new metadata... This is needed especially
3337          * to keep ext2/3 backward compatibility.
3338          */
3339         if (!ext4_has_feature_extents(sb))
3340                 return;
3341         /*
3342          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3343          * This should cover the situations where we can not afford to run
3344          * out of space like for example punch hole, or converting
3345          * unwritten extents in delalloc path. In most cases such
3346          * allocation would require 1, or 2 blocks, higher numbers are
3347          * very rare.
3348          */
3349         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3350                          sbi->s_cluster_bits);
3351
3352         do_div(resv_clusters, 50);
3353         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3354
3355         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3356 }
3357
3358 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3359 {
3360         char *orig_data = kstrdup(data, GFP_KERNEL);
3361         struct buffer_head *bh;
3362         struct ext4_super_block *es = NULL;
3363         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3364         ext4_fsblk_t block;
3365         ext4_fsblk_t sb_block = get_sb_block(&data);
3366         ext4_fsblk_t logical_sb_block;
3367         unsigned long offset = 0;
3368         unsigned long journal_devnum = 0;
3369         unsigned long def_mount_opts;
3370         struct inode *root;
3371         const char *descr;
3372         int ret = -ENOMEM;
3373         int blocksize, clustersize;
3374         unsigned int db_count;
3375         unsigned int i;
3376         int needs_recovery, has_huge_files, has_bigalloc;
3377         __u64 blocks_count;
3378         int err = 0;
3379         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3380         ext4_group_t first_not_zeroed;
3381
3382         if ((data && !orig_data) || !sbi)
3383                 goto out_free_base;
3384
3385         sbi->s_blockgroup_lock =
3386                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3387         if (!sbi->s_blockgroup_lock)
3388                 goto out_free_base;
3389
3390         sb->s_fs_info = sbi;
3391         sbi->s_sb = sb;
3392         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3393         sbi->s_sb_block = sb_block;
3394         if (sb->s_bdev->bd_part)
3395                 sbi->s_sectors_written_start =
3396                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3397
3398         /* Cleanup superblock name */
3399         strreplace(sb->s_id, '/', '!');
3400
3401         /* -EINVAL is default */
3402         ret = -EINVAL;
3403         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3404         if (!blocksize) {
3405                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3406                 goto out_fail;
3407         }
3408
3409         /*
3410          * The ext4 superblock will not be buffer aligned for other than 1kB
3411          * block sizes.  We need to calculate the offset from buffer start.
3412          */
3413         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3414                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3415                 offset = do_div(logical_sb_block, blocksize);
3416         } else {
3417                 logical_sb_block = sb_block;
3418         }
3419
3420         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3421                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3422                 goto out_fail;
3423         }
3424         /*
3425          * Note: s_es must be initialized as soon as possible because
3426          *       some ext4 macro-instructions depend on its value
3427          */
3428         es = (struct ext4_super_block *) (bh->b_data + offset);
3429         sbi->s_es = es;
3430         sb->s_magic = le16_to_cpu(es->s_magic);
3431         if (sb->s_magic != EXT4_SUPER_MAGIC)
3432                 goto cantfind_ext4;
3433         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3434
3435         /* Warn if metadata_csum and gdt_csum are both set. */
3436         if (ext4_has_feature_metadata_csum(sb) &&
3437             ext4_has_feature_gdt_csum(sb))
3438                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3439                              "redundant flags; please run fsck.");
3440
3441         /* Check for a known checksum algorithm */
3442         if (!ext4_verify_csum_type(sb, es)) {
3443                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3444                          "unknown checksum algorithm.");
3445                 silent = 1;
3446                 goto cantfind_ext4;
3447         }
3448
3449         /* Load the checksum driver */
3450         if (ext4_has_feature_metadata_csum(sb)) {
3451                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3452                 if (IS_ERR(sbi->s_chksum_driver)) {
3453                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3454                         ret = PTR_ERR(sbi->s_chksum_driver);
3455                         sbi->s_chksum_driver = NULL;
3456                         goto failed_mount;
3457                 }
3458         }
3459
3460         /* Check superblock checksum */
3461         if (!ext4_superblock_csum_verify(sb, es)) {
3462                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3463                          "invalid superblock checksum.  Run e2fsck?");
3464                 silent = 1;
3465                 ret = -EFSBADCRC;
3466                 goto cantfind_ext4;
3467         }
3468
3469         /* Precompute checksum seed for all metadata */
3470         if (ext4_has_feature_csum_seed(sb))
3471                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3472         else if (ext4_has_metadata_csum(sb))
3473                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3474                                                sizeof(es->s_uuid));
3475
3476         /* Set defaults before we parse the mount options */
3477         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3478         set_opt(sb, INIT_INODE_TABLE);
3479         if (def_mount_opts & EXT4_DEFM_DEBUG)
3480                 set_opt(sb, DEBUG);
3481         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3482                 set_opt(sb, GRPID);
3483         if (def_mount_opts & EXT4_DEFM_UID16)
3484                 set_opt(sb, NO_UID32);
3485         /* xattr user namespace & acls are now defaulted on */
3486         set_opt(sb, XATTR_USER);
3487 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3488         set_opt(sb, POSIX_ACL);
3489 #endif
3490         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3491         if (ext4_has_metadata_csum(sb))
3492                 set_opt(sb, JOURNAL_CHECKSUM);
3493
3494         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3495                 set_opt(sb, JOURNAL_DATA);
3496         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3497                 set_opt(sb, ORDERED_DATA);
3498         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3499                 set_opt(sb, WRITEBACK_DATA);
3500
3501         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3502                 set_opt(sb, ERRORS_PANIC);
3503         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3504                 set_opt(sb, ERRORS_CONT);
3505         else
3506                 set_opt(sb, ERRORS_RO);
3507         /* block_validity enabled by default; disable with noblock_validity */
3508         set_opt(sb, BLOCK_VALIDITY);
3509         if (def_mount_opts & EXT4_DEFM_DISCARD)
3510                 set_opt(sb, DISCARD);
3511
3512         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3513         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3514         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3515         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3516         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3517
3518         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3519                 set_opt(sb, BARRIER);
3520
3521         /*
3522          * enable delayed allocation by default
3523          * Use -o nodelalloc to turn it off
3524          */
3525         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3526             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3527                 set_opt(sb, DELALLOC);
3528
3529         /*
3530          * set default s_li_wait_mult for lazyinit, for the case there is
3531          * no mount option specified.
3532          */
3533         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3534
3535         if (sbi->s_es->s_mount_opts[0]) {
3536                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3537                                               sizeof(sbi->s_es->s_mount_opts),
3538                                               GFP_KERNEL);
3539                 if (!s_mount_opts)
3540                         goto failed_mount;
3541                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3542                                    &journal_ioprio, 0)) {
3543                         ext4_msg(sb, KERN_WARNING,
3544                                  "failed to parse options in superblock: %s",
3545                                  s_mount_opts);
3546                 }
3547                 kfree(s_mount_opts);
3548         }
3549         sbi->s_def_mount_opt = sbi->s_mount_opt;
3550         if (!parse_options((char *) data, sb, &journal_devnum,
3551                            &journal_ioprio, 0))
3552                 goto failed_mount;
3553
3554         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3555                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3556                             "with data=journal disables delayed "
3557                             "allocation and O_DIRECT support!\n");
3558                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3559                         ext4_msg(sb, KERN_ERR, "can't mount with "
3560                                  "both data=journal and delalloc");
3561                         goto failed_mount;
3562                 }
3563                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3564                         ext4_msg(sb, KERN_ERR, "can't mount with "
3565                                  "both data=journal and dioread_nolock");
3566                         goto failed_mount;
3567                 }
3568                 if (test_opt(sb, DAX)) {
3569                         ext4_msg(sb, KERN_ERR, "can't mount with "
3570                                  "both data=journal and dax");
3571                         goto failed_mount;
3572                 }
3573                 if (ext4_has_feature_encrypt(sb)) {
3574                         ext4_msg(sb, KERN_WARNING,
3575                                  "encrypted files will use data=ordered "
3576                                  "instead of data journaling mode");
3577                 }
3578                 if (test_opt(sb, DELALLOC))
3579                         clear_opt(sb, DELALLOC);
3580         } else {
3581                 sb->s_iflags |= SB_I_CGROUPWB;
3582         }
3583
3584         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3585                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3586
3587         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3588             (ext4_has_compat_features(sb) ||
3589              ext4_has_ro_compat_features(sb) ||
3590              ext4_has_incompat_features(sb)))
3591                 ext4_msg(sb, KERN_WARNING,
3592                        "feature flags set on rev 0 fs, "
3593                        "running e2fsck is recommended");
3594
3595         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3596                 set_opt2(sb, HURD_COMPAT);
3597                 if (ext4_has_feature_64bit(sb)) {
3598                         ext4_msg(sb, KERN_ERR,
3599                                  "The Hurd can't support 64-bit file systems");
3600                         goto failed_mount;
3601                 }
3602         }
3603
3604         if (IS_EXT2_SB(sb)) {
3605                 if (ext2_feature_set_ok(sb))
3606                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3607                                  "using the ext4 subsystem");
3608                 else {
3609                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3610                                  "to feature incompatibilities");
3611                         goto failed_mount;
3612                 }
3613         }
3614
3615         if (IS_EXT3_SB(sb)) {
3616                 if (ext3_feature_set_ok(sb))
3617                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3618                                  "using the ext4 subsystem");
3619                 else {
3620                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3621                                  "to feature incompatibilities");
3622                         goto failed_mount;
3623                 }
3624         }
3625
3626         /*
3627          * Check feature flags regardless of the revision level, since we
3628          * previously didn't change the revision level when setting the flags,
3629          * so there is a chance incompat flags are set on a rev 0 filesystem.
3630          */
3631         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3632                 goto failed_mount;
3633
3634         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3635         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3636             blocksize > EXT4_MAX_BLOCK_SIZE) {
3637                 ext4_msg(sb, KERN_ERR,
3638                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3639                          blocksize, le32_to_cpu(es->s_log_block_size));
3640                 goto failed_mount;
3641         }
3642         if (le32_to_cpu(es->s_log_block_size) >
3643             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3644                 ext4_msg(sb, KERN_ERR,
3645                          "Invalid log block size: %u",
3646                          le32_to_cpu(es->s_log_block_size));
3647                 goto failed_mount;
3648         }
3649
3650         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3651                 ext4_msg(sb, KERN_ERR,
3652                          "Number of reserved GDT blocks insanely large: %d",
3653                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3654                 goto failed_mount;
3655         }
3656
3657         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3658                 err = bdev_dax_supported(sb, blocksize);
3659                 if (err)
3660                         goto failed_mount;
3661         }
3662
3663         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3664                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3665                          es->s_encryption_level);
3666                 goto failed_mount;
3667         }
3668
3669         if (sb->s_blocksize != blocksize) {
3670                 /* Validate the filesystem blocksize */
3671                 if (!sb_set_blocksize(sb, blocksize)) {
3672                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3673                                         blocksize);
3674                         goto failed_mount;
3675                 }
3676
3677                 brelse(bh);
3678                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3679                 offset = do_div(logical_sb_block, blocksize);
3680                 bh = sb_bread_unmovable(sb, logical_sb_block);
3681                 if (!bh) {
3682                         ext4_msg(sb, KERN_ERR,
3683                                "Can't read superblock on 2nd try");
3684                         goto failed_mount;
3685                 }
3686                 es = (struct ext4_super_block *)(bh->b_data + offset);
3687                 sbi->s_es = es;
3688                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3689                         ext4_msg(sb, KERN_ERR,
3690                                "Magic mismatch, very weird!");
3691                         goto failed_mount;
3692                 }
3693         }
3694
3695         has_huge_files = ext4_has_feature_huge_file(sb);
3696         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3697                                                       has_huge_files);
3698         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3699
3700         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3701                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3702                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3703         } else {
3704                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3705                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3706                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3707                     (!is_power_of_2(sbi->s_inode_size)) ||
3708                     (sbi->s_inode_size > blocksize)) {
3709                         ext4_msg(sb, KERN_ERR,
3710                                "unsupported inode size: %d",
3711                                sbi->s_inode_size);
3712                         goto failed_mount;
3713                 }
3714                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3715                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3716         }
3717
3718         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3719         if (ext4_has_feature_64bit(sb)) {
3720                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3721                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3722                     !is_power_of_2(sbi->s_desc_size)) {
3723                         ext4_msg(sb, KERN_ERR,
3724                                "unsupported descriptor size %lu",
3725                                sbi->s_desc_size);
3726                         goto failed_mount;
3727                 }
3728         } else
3729                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3730
3731         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3732         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3733
3734         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3735         if (sbi->s_inodes_per_block == 0)
3736                 goto cantfind_ext4;
3737         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3738             sbi->s_inodes_per_group > blocksize * 8) {
3739                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3740                          sbi->s_blocks_per_group);
3741                 goto failed_mount;
3742         }
3743         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3744                                         sbi->s_inodes_per_block;
3745         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3746         sbi->s_sbh = bh;
3747         sbi->s_mount_state = le16_to_cpu(es->s_state);
3748         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3749         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3750
3751         for (i = 0; i < 4; i++)
3752                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3753         sbi->s_def_hash_version = es->s_def_hash_version;
3754         if (ext4_has_feature_dir_index(sb)) {
3755                 i = le32_to_cpu(es->s_flags);
3756                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3757                         sbi->s_hash_unsigned = 3;
3758                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3759 #ifdef __CHAR_UNSIGNED__
3760                         if (!(sb->s_flags & MS_RDONLY))
3761                                 es->s_flags |=
3762                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3763                         sbi->s_hash_unsigned = 3;
3764 #else
3765                         if (!(sb->s_flags & MS_RDONLY))
3766                                 es->s_flags |=
3767                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3768 #endif
3769                 }
3770         }
3771
3772         /* Handle clustersize */
3773         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3774         has_bigalloc = ext4_has_feature_bigalloc(sb);
3775         if (has_bigalloc) {
3776                 if (clustersize < blocksize) {
3777                         ext4_msg(sb, KERN_ERR,
3778                                  "cluster size (%d) smaller than "
3779                                  "block size (%d)", clustersize, blocksize);
3780                         goto failed_mount;
3781                 }
3782                 if (le32_to_cpu(es->s_log_cluster_size) >
3783                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3784                         ext4_msg(sb, KERN_ERR,
3785                                  "Invalid log cluster size: %u",
3786                                  le32_to_cpu(es->s_log_cluster_size));
3787                         goto failed_mount;
3788                 }
3789                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3790                         le32_to_cpu(es->s_log_block_size);
3791                 sbi->s_clusters_per_group =
3792                         le32_to_cpu(es->s_clusters_per_group);
3793                 if (sbi->s_clusters_per_group > blocksize * 8) {
3794                         ext4_msg(sb, KERN_ERR,
3795                                  "#clusters per group too big: %lu",
3796                                  sbi->s_clusters_per_group);
3797                         goto failed_mount;
3798                 }
3799                 if (sbi->s_blocks_per_group !=
3800                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3801                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3802                                  "clusters per group (%lu) inconsistent",
3803                                  sbi->s_blocks_per_group,
3804                                  sbi->s_clusters_per_group);
3805                         goto failed_mount;
3806                 }
3807         } else {
3808                 if (clustersize != blocksize) {
3809                         ext4_warning(sb, "fragment/cluster size (%d) != "
3810                                      "block size (%d)", clustersize,
3811                                      blocksize);
3812                         clustersize = blocksize;
3813                 }
3814                 if (sbi->s_blocks_per_group > blocksize * 8) {
3815                         ext4_msg(sb, KERN_ERR,
3816                                  "#blocks per group too big: %lu",
3817                                  sbi->s_blocks_per_group);
3818                         goto failed_mount;
3819                 }
3820                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3821                 sbi->s_cluster_bits = 0;
3822         }
3823         sbi->s_cluster_ratio = clustersize / blocksize;
3824
3825         /* Do we have standard group size of clustersize * 8 blocks ? */
3826         if (sbi->s_blocks_per_group == clustersize << 3)
3827                 set_opt2(sb, STD_GROUP_SIZE);
3828
3829         /*
3830          * Test whether we have more sectors than will fit in sector_t,
3831          * and whether the max offset is addressable by the page cache.
3832          */
3833         err = generic_check_addressable(sb->s_blocksize_bits,
3834                                         ext4_blocks_count(es));
3835         if (err) {
3836                 ext4_msg(sb, KERN_ERR, "filesystem"
3837                          " too large to mount safely on this system");
3838                 if (sizeof(sector_t) < 8)
3839                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3840                 goto failed_mount;
3841         }
3842
3843         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3844                 goto cantfind_ext4;
3845
3846         /* check blocks count against device size */
3847         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3848         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3849                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3850                        "exceeds size of device (%llu blocks)",
3851                        ext4_blocks_count(es), blocks_count);
3852                 goto failed_mount;
3853         }
3854
3855         /*
3856          * It makes no sense for the first data block to be beyond the end
3857          * of the filesystem.
3858          */
3859         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3860                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3861                          "block %u is beyond end of filesystem (%llu)",
3862                          le32_to_cpu(es->s_first_data_block),
3863                          ext4_blocks_count(es));
3864                 goto failed_mount;
3865         }
3866         blocks_count = (ext4_blocks_count(es) -
3867                         le32_to_cpu(es->s_first_data_block) +
3868                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3869         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3870         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3871                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3872                        "(block count %llu, first data block %u, "
3873                        "blocks per group %lu)", sbi->s_groups_count,
3874                        ext4_blocks_count(es),
3875                        le32_to_cpu(es->s_first_data_block),
3876                        EXT4_BLOCKS_PER_GROUP(sb));
3877                 goto failed_mount;
3878         }
3879         sbi->s_groups_count = blocks_count;
3880         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3881                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3882         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3883                    EXT4_DESC_PER_BLOCK(sb);
3884         if (ext4_has_feature_meta_bg(sb)) {
3885                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3886                         ext4_msg(sb, KERN_WARNING,
3887                                  "first meta block group too large: %u "
3888                                  "(group descriptor block count %u)",
3889                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3890                         goto failed_mount;
3891                 }
3892         }
3893         sbi->s_group_desc = kvmalloc(db_count *
3894                                           sizeof(struct buffer_head *),
3895                                           GFP_KERNEL);
3896         if (sbi->s_group_desc == NULL) {
3897                 ext4_msg(sb, KERN_ERR, "not enough memory");
3898                 ret = -ENOMEM;
3899                 goto failed_mount;
3900         }
3901
3902         bgl_lock_init(sbi->s_blockgroup_lock);
3903
3904         /* Pre-read the descriptors into the buffer cache */
3905         for (i = 0; i < db_count; i++) {
3906                 block = descriptor_loc(sb, logical_sb_block, i);
3907                 sb_breadahead(sb, block);
3908         }
3909
3910         for (i = 0; i < db_count; i++) {
3911                 block = descriptor_loc(sb, logical_sb_block, i);
3912                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3913                 if (!sbi->s_group_desc[i]) {
3914                         ext4_msg(sb, KERN_ERR,
3915                                "can't read group descriptor %d", i);
3916                         db_count = i;
3917                         goto failed_mount2;
3918                 }
3919         }
3920         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3921                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3922                 ret = -EFSCORRUPTED;
3923                 goto failed_mount2;
3924         }
3925
3926         sbi->s_gdb_count = db_count;
3927         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3928         spin_lock_init(&sbi->s_next_gen_lock);
3929
3930         setup_timer(&sbi->s_err_report, print_daily_error_info,
3931                 (unsigned long) sb);
3932
3933         /* Register extent status tree shrinker */
3934         if (ext4_es_register_shrinker(sbi))
3935                 goto failed_mount3;
3936
3937         sbi->s_stripe = ext4_get_stripe_size(sbi);
3938         sbi->s_extent_max_zeroout_kb = 32;
3939
3940         /*
3941          * set up enough so that it can read an inode
3942          */
3943         sb->s_op = &ext4_sops;
3944         sb->s_export_op = &ext4_export_ops;
3945         sb->s_xattr = ext4_xattr_handlers;
3946         sb->s_cop = &ext4_cryptops;
3947 #ifdef CONFIG_QUOTA
3948         sb->dq_op = &ext4_quota_operations;
3949         if (ext4_has_feature_quota(sb))
3950                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3951         else
3952                 sb->s_qcop = &ext4_qctl_operations;
3953         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3954 #endif
3955         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3956
3957         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3958         mutex_init(&sbi->s_orphan_lock);
3959
3960         sb->s_root = NULL;
3961
3962         needs_recovery = (es->s_last_orphan != 0 ||
3963                           ext4_has_feature_journal_needs_recovery(sb));
3964
3965         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3966                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3967                         goto failed_mount3a;
3968
3969         /*
3970          * The first inode we look at is the journal inode.  Don't try
3971          * root first: it may be modified in the journal!
3972          */
3973         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3974                 err = ext4_load_journal(sb, es, journal_devnum);
3975                 if (err)
3976                         goto failed_mount3a;
3977         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3978                    ext4_has_feature_journal_needs_recovery(sb)) {
3979                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3980                        "suppressed and not mounted read-only");
3981                 goto failed_mount_wq;
3982         } else {
3983                 /* Nojournal mode, all journal mount options are illegal */
3984                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3985                         ext4_msg(sb, KERN_ERR, "can't mount with "
3986                                  "journal_checksum, fs mounted w/o journal");
3987                         goto failed_mount_wq;
3988                 }
3989                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3990                         ext4_msg(sb, KERN_ERR, "can't mount with "
3991                                  "journal_async_commit, fs mounted w/o journal");
3992                         goto failed_mount_wq;
3993                 }
3994                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3995                         ext4_msg(sb, KERN_ERR, "can't mount with "
3996                                  "commit=%lu, fs mounted w/o journal",
3997                                  sbi->s_commit_interval / HZ);
3998                         goto failed_mount_wq;
3999                 }
4000                 if (EXT4_MOUNT_DATA_FLAGS &
4001                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4002                         ext4_msg(sb, KERN_ERR, "can't mount with "
4003                                  "data=, fs mounted w/o journal");
4004                         goto failed_mount_wq;
4005                 }
4006                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4007                 clear_opt(sb, JOURNAL_CHECKSUM);
4008                 clear_opt(sb, DATA_FLAGS);
4009                 sbi->s_journal = NULL;
4010                 needs_recovery = 0;
4011                 goto no_journal;
4012         }
4013
4014         if (ext4_has_feature_64bit(sb) &&
4015             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4016                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4017                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4018                 goto failed_mount_wq;
4019         }
4020
4021         if (!set_journal_csum_feature_set(sb)) {
4022                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4023                          "feature set");
4024                 goto failed_mount_wq;
4025         }
4026
4027         /* We have now updated the journal if required, so we can
4028          * validate the data journaling mode. */
4029         switch (test_opt(sb, DATA_FLAGS)) {
4030         case 0:
4031                 /* No mode set, assume a default based on the journal
4032                  * capabilities: ORDERED_DATA if the journal can
4033                  * cope, else JOURNAL_DATA
4034                  */
4035                 if (jbd2_journal_check_available_features
4036                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4037                         set_opt(sb, ORDERED_DATA);
4038                 else
4039                         set_opt(sb, JOURNAL_DATA);
4040                 break;
4041
4042         case EXT4_MOUNT_ORDERED_DATA:
4043         case EXT4_MOUNT_WRITEBACK_DATA:
4044                 if (!jbd2_journal_check_available_features
4045                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4046                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4047                                "requested data journaling mode");
4048                         goto failed_mount_wq;
4049                 }
4050         default:
4051                 break;
4052         }
4053
4054         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4055             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4056                 ext4_msg(sb, KERN_ERR, "can't mount with "
4057                         "journal_async_commit in data=ordered mode");
4058                 goto failed_mount_wq;
4059         }
4060
4061         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4062
4063         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4064
4065 no_journal:
4066         sbi->s_mb_cache = ext4_xattr_create_cache();
4067         if (!sbi->s_mb_cache) {
4068                 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4069                 goto failed_mount_wq;
4070         }
4071
4072         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4073             (blocksize != PAGE_SIZE)) {
4074                 ext4_msg(sb, KERN_ERR,
4075                          "Unsupported blocksize for fs encryption");
4076                 goto failed_mount_wq;
4077         }
4078
4079         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4080             !ext4_has_feature_encrypt(sb)) {
4081                 ext4_set_feature_encrypt(sb);
4082                 ext4_commit_super(sb, 1);
4083         }
4084
4085         /*
4086          * Get the # of file system overhead blocks from the
4087          * superblock if present.
4088          */
4089         if (es->s_overhead_clusters)
4090                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4091         else {
4092                 err = ext4_calculate_overhead(sb);
4093                 if (err)
4094                         goto failed_mount_wq;
4095         }
4096
4097         /*
4098          * The maximum number of concurrent works can be high and
4099          * concurrency isn't really necessary.  Limit it to 1.
4100          */
4101         EXT4_SB(sb)->rsv_conversion_wq =
4102                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4103         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4104                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4105                 ret = -ENOMEM;
4106                 goto failed_mount4;
4107         }
4108
4109         /*
4110          * The jbd2_journal_load will have done any necessary log recovery,
4111          * so we can safely mount the rest of the filesystem now.
4112          */
4113
4114         root = ext4_iget(sb, EXT4_ROOT_INO);
4115         if (IS_ERR(root)) {
4116                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4117                 ret = PTR_ERR(root);
4118                 root = NULL;
4119                 goto failed_mount4;
4120         }
4121         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4122                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4123                 iput(root);
4124                 goto failed_mount4;
4125         }
4126         sb->s_root = d_make_root(root);
4127         if (!sb->s_root) {
4128                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4129                 ret = -ENOMEM;
4130                 goto failed_mount4;
4131         }
4132
4133         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4134                 sb->s_flags |= MS_RDONLY;
4135
4136         /* determine the minimum size of new large inodes, if present */
4137         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4138             sbi->s_want_extra_isize == 0) {
4139                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4140                                                      EXT4_GOOD_OLD_INODE_SIZE;
4141                 if (ext4_has_feature_extra_isize(sb)) {
4142                         if (sbi->s_want_extra_isize <
4143                             le16_to_cpu(es->s_want_extra_isize))
4144                                 sbi->s_want_extra_isize =
4145                                         le16_to_cpu(es->s_want_extra_isize);
4146                         if (sbi->s_want_extra_isize <
4147                             le16_to_cpu(es->s_min_extra_isize))
4148                                 sbi->s_want_extra_isize =
4149                                         le16_to_cpu(es->s_min_extra_isize);
4150                 }
4151         }
4152         /* Check if enough inode space is available */
4153         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4154                                                         sbi->s_inode_size) {
4155                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4156                                                        EXT4_GOOD_OLD_INODE_SIZE;
4157                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4158                          "available");
4159         }
4160
4161         ext4_set_resv_clusters(sb);
4162
4163         err = ext4_setup_system_zone(sb);
4164         if (err) {
4165                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4166                          "zone (%d)", err);
4167                 goto failed_mount4a;
4168         }
4169
4170         ext4_ext_init(sb);
4171         err = ext4_mb_init(sb);
4172         if (err) {
4173                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4174                          err);
4175                 goto failed_mount5;
4176         }
4177
4178         block = ext4_count_free_clusters(sb);
4179         ext4_free_blocks_count_set(sbi->s_es,
4180                                    EXT4_C2B(sbi, block));
4181         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4182                                   GFP_KERNEL);
4183         if (!err) {
4184                 unsigned long freei = ext4_count_free_inodes(sb);
4185                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4186                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4187                                           GFP_KERNEL);
4188         }
4189         if (!err)
4190                 err = percpu_counter_init(&sbi->s_dirs_counter,
4191                                           ext4_count_dirs(sb), GFP_KERNEL);
4192         if (!err)
4193                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4194                                           GFP_KERNEL);
4195         if (!err)
4196                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4197
4198         if (err) {
4199                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4200                 goto failed_mount6;
4201         }
4202
4203         if (ext4_has_feature_flex_bg(sb))
4204                 if (!ext4_fill_flex_info(sb)) {
4205                         ext4_msg(sb, KERN_ERR,
4206                                "unable to initialize "
4207                                "flex_bg meta info!");
4208                         goto failed_mount6;
4209                 }
4210
4211         err = ext4_register_li_request(sb, first_not_zeroed);
4212         if (err)
4213                 goto failed_mount6;
4214
4215         err = ext4_register_sysfs(sb);
4216         if (err)
4217                 goto failed_mount7;
4218
4219 #ifdef CONFIG_QUOTA
4220         /* Enable quota usage during mount. */
4221         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4222                 err = ext4_enable_quotas(sb);
4223                 if (err)
4224                         goto failed_mount8;
4225         }
4226 #endif  /* CONFIG_QUOTA */
4227
4228         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4229         ext4_orphan_cleanup(sb, es);
4230         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4231         if (needs_recovery) {
4232                 ext4_msg(sb, KERN_INFO, "recovery complete");
4233                 ext4_mark_recovery_complete(sb, es);
4234         }
4235         if (EXT4_SB(sb)->s_journal) {
4236                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4237                         descr = " journalled data mode";
4238                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4239                         descr = " ordered data mode";
4240                 else
4241                         descr = " writeback data mode";
4242         } else
4243                 descr = "out journal";
4244
4245         if (test_opt(sb, DISCARD)) {
4246                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4247                 if (!blk_queue_discard(q))
4248                         ext4_msg(sb, KERN_WARNING,
4249                                  "mounting with \"discard\" option, but "
4250                                  "the device does not support discard");
4251         }
4252
4253         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4254                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4255                          "Opts: %.*s%s%s", descr,
4256                          (int) sizeof(sbi->s_es->s_mount_opts),
4257                          sbi->s_es->s_mount_opts,
4258                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4259
4260         if (es->s_error_count)
4261                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4262
4263         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4264         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4265         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4266         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4267
4268         kfree(orig_data);
4269         return 0;
4270
4271 cantfind_ext4:
4272         if (!silent)
4273                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4274         goto failed_mount;
4275
4276 #ifdef CONFIG_QUOTA
4277 failed_mount8:
4278         ext4_unregister_sysfs(sb);
4279 #endif
4280 failed_mount7:
4281         ext4_unregister_li_request(sb);
4282 failed_mount6:
4283         ext4_mb_release(sb);
4284         if (sbi->s_flex_groups)
4285                 kvfree(sbi->s_flex_groups);
4286         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4287         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4288         percpu_counter_destroy(&sbi->s_dirs_counter);
4289         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4290 failed_mount5:
4291         ext4_ext_release(sb);
4292         ext4_release_system_zone(sb);
4293 failed_mount4a:
4294         dput(sb->s_root);
4295         sb->s_root = NULL;
4296 failed_mount4:
4297         ext4_msg(sb, KERN_ERR, "mount failed");
4298         if (EXT4_SB(sb)->rsv_conversion_wq)
4299                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4300 failed_mount_wq:
4301         if (sbi->s_mb_cache) {
4302                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4303                 sbi->s_mb_cache = NULL;
4304         }
4305         if (sbi->s_journal) {
4306                 jbd2_journal_destroy(sbi->s_journal);
4307                 sbi->s_journal = NULL;
4308         }
4309 failed_mount3a:
4310         ext4_es_unregister_shrinker(sbi);
4311 failed_mount3:
4312         del_timer_sync(&sbi->s_err_report);
4313         if (sbi->s_mmp_tsk)
4314                 kthread_stop(sbi->s_mmp_tsk);
4315 failed_mount2:
4316         for (i = 0; i < db_count; i++)
4317                 brelse(sbi->s_group_desc[i]);
4318         kvfree(sbi->s_group_desc);
4319 failed_mount:
4320         if (sbi->s_chksum_driver)
4321                 crypto_free_shash(sbi->s_chksum_driver);
4322 #ifdef CONFIG_QUOTA
4323         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4324                 kfree(sbi->s_qf_names[i]);
4325 #endif
4326         ext4_blkdev_remove(sbi);
4327         brelse(bh);
4328 out_fail:
4329         sb->s_fs_info = NULL;
4330         kfree(sbi->s_blockgroup_lock);
4331 out_free_base:
4332         kfree(sbi);
4333         kfree(orig_data);
4334         return err ? err : ret;
4335 }
4336
4337 /*
4338  * Setup any per-fs journal parameters now.  We'll do this both on
4339  * initial mount, once the journal has been initialised but before we've
4340  * done any recovery; and again on any subsequent remount.
4341  */
4342 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4343 {
4344         struct ext4_sb_info *sbi = EXT4_SB(sb);
4345
4346         journal->j_commit_interval = sbi->s_commit_interval;
4347         journal->j_min_batch_time = sbi->s_min_batch_time;
4348         journal->j_max_batch_time = sbi->s_max_batch_time;
4349
4350         write_lock(&journal->j_state_lock);
4351         if (test_opt(sb, BARRIER))
4352                 journal->j_flags |= JBD2_BARRIER;
4353         else
4354                 journal->j_flags &= ~JBD2_BARRIER;
4355         if (test_opt(sb, DATA_ERR_ABORT))
4356                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4357         else
4358                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4359         write_unlock(&journal->j_state_lock);
4360 }
4361
4362 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4363                                              unsigned int journal_inum)
4364 {
4365         struct inode *journal_inode;
4366
4367         /*
4368          * Test for the existence of a valid inode on disk.  Bad things
4369          * happen if we iget() an unused inode, as the subsequent iput()
4370          * will try to delete it.
4371          */
4372         journal_inode = ext4_iget(sb, journal_inum);
4373         if (IS_ERR(journal_inode)) {
4374                 ext4_msg(sb, KERN_ERR, "no journal found");
4375                 return NULL;
4376         }
4377         if (!journal_inode->i_nlink) {
4378                 make_bad_inode(journal_inode);
4379                 iput(journal_inode);
4380                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4381                 return NULL;
4382         }
4383
4384         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4385                   journal_inode, journal_inode->i_size);
4386         if (!S_ISREG(journal_inode->i_mode)) {
4387                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4388                 iput(journal_inode);
4389                 return NULL;
4390         }
4391         return journal_inode;
4392 }
4393
4394 static journal_t *ext4_get_journal(struct super_block *sb,
4395                                    unsigned int journal_inum)
4396 {
4397         struct inode *journal_inode;
4398         journal_t *journal;
4399
4400         BUG_ON(!ext4_has_feature_journal(sb));
4401
4402         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4403         if (!journal_inode)
4404                 return NULL;
4405
4406         journal = jbd2_journal_init_inode(journal_inode);
4407         if (!journal) {
4408                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4409                 iput(journal_inode);
4410                 return NULL;
4411         }
4412         journal->j_private = sb;
4413         ext4_init_journal_params(sb, journal);
4414         return journal;
4415 }
4416
4417 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4418                                        dev_t j_dev)
4419 {
4420         struct buffer_head *bh;
4421         journal_t *journal;
4422         ext4_fsblk_t start;
4423         ext4_fsblk_t len;
4424         int hblock, blocksize;
4425         ext4_fsblk_t sb_block;
4426         unsigned long offset;
4427         struct ext4_super_block *es;
4428         struct block_device *bdev;
4429
4430         BUG_ON(!ext4_has_feature_journal(sb));
4431
4432         bdev = ext4_blkdev_get(j_dev, sb);
4433         if (bdev == NULL)
4434                 return NULL;
4435
4436         blocksize = sb->s_blocksize;
4437         hblock = bdev_logical_block_size(bdev);
4438         if (blocksize < hblock) {
4439                 ext4_msg(sb, KERN_ERR,
4440                         "blocksize too small for journal device");
4441                 goto out_bdev;
4442         }
4443
4444         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4445         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4446         set_blocksize(bdev, blocksize);
4447         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4448                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4449                        "external journal");
4450                 goto out_bdev;
4451         }
4452
4453         es = (struct ext4_super_block *) (bh->b_data + offset);
4454         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4455             !(le32_to_cpu(es->s_feature_incompat) &
4456               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4457                 ext4_msg(sb, KERN_ERR, "external journal has "
4458                                         "bad superblock");
4459                 brelse(bh);
4460                 goto out_bdev;
4461         }
4462
4463         if ((le32_to_cpu(es->s_feature_ro_compat) &
4464              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4465             es->s_checksum != ext4_superblock_csum(sb, es)) {
4466                 ext4_msg(sb, KERN_ERR, "external journal has "
4467                                        "corrupt superblock");
4468                 brelse(bh);
4469                 goto out_bdev;
4470         }
4471
4472         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4473                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4474                 brelse(bh);
4475                 goto out_bdev;
4476         }
4477
4478         len = ext4_blocks_count(es);
4479         start = sb_block + 1;
4480         brelse(bh);     /* we're done with the superblock */
4481
4482         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4483                                         start, len, blocksize);
4484         if (!journal) {
4485                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4486                 goto out_bdev;
4487         }
4488         journal->j_private = sb;
4489         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4490         wait_on_buffer(journal->j_sb_buffer);
4491         if (!buffer_uptodate(journal->j_sb_buffer)) {
4492                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4493                 goto out_journal;
4494         }
4495         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4496                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4497                                         "user (unsupported) - %d",
4498                         be32_to_cpu(journal->j_superblock->s_nr_users));
4499                 goto out_journal;
4500         }
4501         EXT4_SB(sb)->journal_bdev = bdev;
4502         ext4_init_journal_params(sb, journal);
4503         return journal;
4504
4505 out_journal:
4506         jbd2_journal_destroy(journal);
4507 out_bdev:
4508         ext4_blkdev_put(bdev);
4509         return NULL;
4510 }
4511
4512 static int ext4_load_journal(struct super_block *sb,
4513                              struct ext4_super_block *es,
4514                              unsigned long journal_devnum)
4515 {
4516         journal_t *journal;
4517         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4518         dev_t journal_dev;
4519         int err = 0;
4520         int really_read_only;
4521
4522         BUG_ON(!ext4_has_feature_journal(sb));
4523
4524         if (journal_devnum &&
4525             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4526                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4527                         "numbers have changed");
4528                 journal_dev = new_decode_dev(journal_devnum);
4529         } else
4530                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4531
4532         really_read_only = bdev_read_only(sb->s_bdev);
4533
4534         /*
4535          * Are we loading a blank journal or performing recovery after a
4536          * crash?  For recovery, we need to check in advance whether we
4537          * can get read-write access to the device.
4538          */
4539         if (ext4_has_feature_journal_needs_recovery(sb)) {
4540                 if (sb->s_flags & MS_RDONLY) {
4541                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4542                                         "required on readonly filesystem");
4543                         if (really_read_only) {
4544                                 ext4_msg(sb, KERN_ERR, "write access "
4545                                         "unavailable, cannot proceed");
4546                                 return -EROFS;
4547                         }
4548                         ext4_msg(sb, KERN_INFO, "write access will "
4549                                "be enabled during recovery");
4550                 }
4551         }
4552
4553         if (journal_inum && journal_dev) {
4554                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4555                        "and inode journals!");
4556                 return -EINVAL;
4557         }
4558
4559         if (journal_inum) {
4560                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4561                         return -EINVAL;
4562         } else {
4563                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4564                         return -EINVAL;
4565         }
4566
4567         if (!(journal->j_flags & JBD2_BARRIER))
4568                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4569
4570         if (!ext4_has_feature_journal_needs_recovery(sb))
4571                 err = jbd2_journal_wipe(journal, !really_read_only);
4572         if (!err) {
4573                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4574                 if (save)
4575                         memcpy(save, ((char *) es) +
4576                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4577                 err = jbd2_journal_load(journal);
4578                 if (save)
4579                         memcpy(((char *) es) + EXT4_S_ERR_START,
4580                                save, EXT4_S_ERR_LEN);
4581                 kfree(save);
4582         }
4583
4584         if (err) {
4585                 ext4_msg(sb, KERN_ERR, "error loading journal");
4586                 jbd2_journal_destroy(journal);
4587                 return err;
4588         }
4589
4590         EXT4_SB(sb)->s_journal = journal;
4591         ext4_clear_journal_err(sb, es);
4592
4593         if (!really_read_only && journal_devnum &&
4594             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4595                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4596
4597                 /* Make sure we flush the recovery flag to disk. */
4598                 ext4_commit_super(sb, 1);
4599         }
4600
4601         return 0;
4602 }
4603
4604 static int ext4_commit_super(struct super_block *sb, int sync)
4605 {
4606         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4607         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4608         int error = 0;
4609
4610         if (!sbh || block_device_ejected(sb))
4611                 return error;
4612         /*
4613          * If the file system is mounted read-only, don't update the
4614          * superblock write time.  This avoids updating the superblock
4615          * write time when we are mounting the root file system
4616          * read/only but we need to replay the journal; at that point,
4617          * for people who are east of GMT and who make their clock
4618          * tick in localtime for Windows bug-for-bug compatibility,
4619          * the clock is set in the future, and this will cause e2fsck
4620          * to complain and force a full file system check.
4621          */
4622         if (!(sb->s_flags & MS_RDONLY))
4623                 es->s_wtime = cpu_to_le32(get_seconds());
4624         if (sb->s_bdev->bd_part)
4625                 es->s_kbytes_written =
4626                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4627                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4628                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4629         else
4630                 es->s_kbytes_written =
4631                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4632         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4633                 ext4_free_blocks_count_set(es,
4634                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4635                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4636         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4637                 es->s_free_inodes_count =
4638                         cpu_to_le32(percpu_counter_sum_positive(
4639                                 &EXT4_SB(sb)->s_freeinodes_counter));
4640         BUFFER_TRACE(sbh, "marking dirty");
4641         ext4_superblock_csum_set(sb);
4642         if (sync)
4643                 lock_buffer(sbh);
4644         if (buffer_write_io_error(sbh)) {
4645                 /*
4646                  * Oh, dear.  A previous attempt to write the
4647                  * superblock failed.  This could happen because the
4648                  * USB device was yanked out.  Or it could happen to
4649                  * be a transient write error and maybe the block will
4650                  * be remapped.  Nothing we can do but to retry the
4651                  * write and hope for the best.
4652                  */
4653                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4654                        "superblock detected");
4655                 clear_buffer_write_io_error(sbh);
4656                 set_buffer_uptodate(sbh);
4657         }
4658         mark_buffer_dirty(sbh);
4659         if (sync) {
4660                 unlock_buffer(sbh);
4661                 error = __sync_dirty_buffer(sbh,
4662                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4663                 if (error)
4664                         return error;
4665
4666                 error = buffer_write_io_error(sbh);
4667                 if (error) {
4668                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4669                                "superblock");
4670                         clear_buffer_write_io_error(sbh);
4671                         set_buffer_uptodate(sbh);
4672                 }
4673         }
4674         return error;
4675 }
4676
4677 /*
4678  * Have we just finished recovery?  If so, and if we are mounting (or
4679  * remounting) the filesystem readonly, then we will end up with a
4680  * consistent fs on disk.  Record that fact.
4681  */
4682 static void ext4_mark_recovery_complete(struct super_block *sb,
4683                                         struct ext4_super_block *es)
4684 {
4685         journal_t *journal = EXT4_SB(sb)->s_journal;
4686
4687         if (!ext4_has_feature_journal(sb)) {
4688                 BUG_ON(journal != NULL);
4689                 return;
4690         }
4691         jbd2_journal_lock_updates(journal);
4692         if (jbd2_journal_flush(journal) < 0)
4693                 goto out;
4694
4695         if (ext4_has_feature_journal_needs_recovery(sb) &&
4696             sb->s_flags & MS_RDONLY) {
4697                 ext4_clear_feature_journal_needs_recovery(sb);
4698                 ext4_commit_super(sb, 1);
4699         }
4700
4701 out:
4702         jbd2_journal_unlock_updates(journal);
4703 }
4704
4705 /*
4706  * If we are mounting (or read-write remounting) a filesystem whose journal
4707  * has recorded an error from a previous lifetime, move that error to the
4708  * main filesystem now.
4709  */
4710 static void ext4_clear_journal_err(struct super_block *sb,
4711                                    struct ext4_super_block *es)
4712 {
4713         journal_t *journal;
4714         int j_errno;
4715         const char *errstr;
4716
4717         BUG_ON(!ext4_has_feature_journal(sb));
4718
4719         journal = EXT4_SB(sb)->s_journal;
4720
4721         /*
4722          * Now check for any error status which may have been recorded in the
4723          * journal by a prior ext4_error() or ext4_abort()
4724          */
4725
4726         j_errno = jbd2_journal_errno(journal);
4727         if (j_errno) {
4728                 char nbuf[16];
4729
4730                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4731                 ext4_warning(sb, "Filesystem error recorded "
4732                              "from previous mount: %s", errstr);
4733                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4734
4735                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4736                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4737                 ext4_commit_super(sb, 1);
4738
4739                 jbd2_journal_clear_err(journal);
4740                 jbd2_journal_update_sb_errno(journal);
4741         }
4742 }
4743
4744 /*
4745  * Force the running and committing transactions to commit,
4746  * and wait on the commit.
4747  */
4748 int ext4_force_commit(struct super_block *sb)
4749 {
4750         journal_t *journal;
4751
4752         if (sb->s_flags & MS_RDONLY)
4753                 return 0;
4754
4755         journal = EXT4_SB(sb)->s_journal;
4756         return ext4_journal_force_commit(journal);
4757 }
4758
4759 static int ext4_sync_fs(struct super_block *sb, int wait)
4760 {
4761         int ret = 0;
4762         tid_t target;
4763         bool needs_barrier = false;
4764         struct ext4_sb_info *sbi = EXT4_SB(sb);
4765
4766         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4767                 return 0;
4768
4769         trace_ext4_sync_fs(sb, wait);
4770         flush_workqueue(sbi->rsv_conversion_wq);
4771         /*
4772          * Writeback quota in non-journalled quota case - journalled quota has
4773          * no dirty dquots
4774          */
4775         dquot_writeback_dquots(sb, -1);
4776         /*
4777          * Data writeback is possible w/o journal transaction, so barrier must
4778          * being sent at the end of the function. But we can skip it if
4779          * transaction_commit will do it for us.
4780          */
4781         if (sbi->s_journal) {
4782                 target = jbd2_get_latest_transaction(sbi->s_journal);
4783                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4784                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4785                         needs_barrier = true;
4786
4787                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4788                         if (wait)
4789                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4790                                                            target);
4791                 }
4792         } else if (wait && test_opt(sb, BARRIER))
4793                 needs_barrier = true;
4794         if (needs_barrier) {
4795                 int err;
4796                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4797                 if (!ret)
4798                         ret = err;
4799         }
4800
4801         return ret;
4802 }
4803
4804 /*
4805  * LVM calls this function before a (read-only) snapshot is created.  This
4806  * gives us a chance to flush the journal completely and mark the fs clean.
4807  *
4808  * Note that only this function cannot bring a filesystem to be in a clean
4809  * state independently. It relies on upper layer to stop all data & metadata
4810  * modifications.
4811  */
4812 static int ext4_freeze(struct super_block *sb)
4813 {
4814         int error = 0;
4815         journal_t *journal;
4816
4817         if (sb->s_flags & MS_RDONLY)
4818                 return 0;
4819
4820         journal = EXT4_SB(sb)->s_journal;
4821
4822         if (journal) {
4823                 /* Now we set up the journal barrier. */
4824                 jbd2_journal_lock_updates(journal);
4825
4826                 /*
4827                  * Don't clear the needs_recovery flag if we failed to
4828                  * flush the journal.
4829                  */
4830                 error = jbd2_journal_flush(journal);
4831                 if (error < 0)
4832                         goto out;
4833
4834                 /* Journal blocked and flushed, clear needs_recovery flag. */
4835                 ext4_clear_feature_journal_needs_recovery(sb);
4836         }
4837
4838         error = ext4_commit_super(sb, 1);
4839 out:
4840         if (journal)
4841                 /* we rely on upper layer to stop further updates */
4842                 jbd2_journal_unlock_updates(journal);
4843         return error;
4844 }
4845
4846 /*
4847  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4848  * flag here, even though the filesystem is not technically dirty yet.
4849  */
4850 static int ext4_unfreeze(struct super_block *sb)
4851 {
4852         if ((sb->s_flags & MS_RDONLY) || ext4_forced_shutdown(EXT4_SB(sb)))
4853                 return 0;
4854
4855         if (EXT4_SB(sb)->s_journal) {
4856                 /* Reset the needs_recovery flag before the fs is unlocked. */
4857                 ext4_set_feature_journal_needs_recovery(sb);
4858         }
4859
4860         ext4_commit_super(sb, 1);
4861         return 0;
4862 }
4863
4864 /*
4865  * Structure to save mount options for ext4_remount's benefit
4866  */
4867 struct ext4_mount_options {
4868         unsigned long s_mount_opt;
4869         unsigned long s_mount_opt2;
4870         kuid_t s_resuid;
4871         kgid_t s_resgid;
4872         unsigned long s_commit_interval;
4873         u32 s_min_batch_time, s_max_batch_time;
4874 #ifdef CONFIG_QUOTA
4875         int s_jquota_fmt;
4876         char *s_qf_names[EXT4_MAXQUOTAS];
4877 #endif
4878 };
4879
4880 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4881 {
4882         struct ext4_super_block *es;
4883         struct ext4_sb_info *sbi = EXT4_SB(sb);
4884         unsigned long old_sb_flags;
4885         struct ext4_mount_options old_opts;
4886         int enable_quota = 0;
4887         ext4_group_t g;
4888         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4889         int err = 0;
4890 #ifdef CONFIG_QUOTA
4891         int i, j;
4892 #endif
4893         char *orig_data = kstrdup(data, GFP_KERNEL);
4894
4895         /* Store the original options */
4896         old_sb_flags = sb->s_flags;
4897         old_opts.s_mount_opt = sbi->s_mount_opt;
4898         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4899         old_opts.s_resuid = sbi->s_resuid;
4900         old_opts.s_resgid = sbi->s_resgid;
4901         old_opts.s_commit_interval = sbi->s_commit_interval;
4902         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4903         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4904 #ifdef CONFIG_QUOTA
4905         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4906         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4907                 if (sbi->s_qf_names[i]) {
4908                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4909                                                          GFP_KERNEL);
4910                         if (!old_opts.s_qf_names[i]) {
4911                                 for (j = 0; j < i; j++)
4912                                         kfree(old_opts.s_qf_names[j]);
4913                                 kfree(orig_data);
4914                                 return -ENOMEM;
4915                         }
4916                 } else
4917                         old_opts.s_qf_names[i] = NULL;
4918 #endif
4919         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4920                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4921
4922         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4923                 err = -EINVAL;
4924                 goto restore_opts;
4925         }
4926
4927         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4928             test_opt(sb, JOURNAL_CHECKSUM)) {
4929                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4930                          "during remount not supported; ignoring");
4931                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4932         }
4933
4934         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4935                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4936                         ext4_msg(sb, KERN_ERR, "can't mount with "
4937                                  "both data=journal and delalloc");
4938                         err = -EINVAL;
4939                         goto restore_opts;
4940                 }
4941                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4942                         ext4_msg(sb, KERN_ERR, "can't mount with "
4943                                  "both data=journal and dioread_nolock");
4944                         err = -EINVAL;
4945                         goto restore_opts;
4946                 }
4947                 if (test_opt(sb, DAX)) {
4948                         ext4_msg(sb, KERN_ERR, "can't mount with "
4949                                  "both data=journal and dax");
4950                         err = -EINVAL;
4951                         goto restore_opts;
4952                 }
4953         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
4954                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4955                         ext4_msg(sb, KERN_ERR, "can't mount with "
4956                                 "journal_async_commit in data=ordered mode");
4957                         err = -EINVAL;
4958                         goto restore_opts;
4959                 }
4960         }
4961
4962         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4963                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4964                         "dax flag with busy inodes while remounting");
4965                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4966         }
4967
4968         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4969                 ext4_abort(sb, "Abort forced by user");
4970
4971         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4972                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4973
4974         es = sbi->s_es;
4975
4976         if (sbi->s_journal) {
4977                 ext4_init_journal_params(sb, sbi->s_journal);
4978                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4979         }
4980
4981         if (*flags & MS_LAZYTIME)
4982                 sb->s_flags |= MS_LAZYTIME;
4983
4984         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4985                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4986                         err = -EROFS;
4987                         goto restore_opts;
4988                 }
4989
4990                 if (*flags & MS_RDONLY) {
4991                         err = sync_filesystem(sb);
4992                         if (err < 0)
4993                                 goto restore_opts;
4994                         err = dquot_suspend(sb, -1);
4995                         if (err < 0)
4996                                 goto restore_opts;
4997
4998                         /*
4999                          * First of all, the unconditional stuff we have to do
5000                          * to disable replay of the journal when we next remount
5001                          */
5002                         sb->s_flags |= MS_RDONLY;
5003
5004                         /*
5005                          * OK, test if we are remounting a valid rw partition
5006                          * readonly, and if so set the rdonly flag and then
5007                          * mark the partition as valid again.
5008                          */
5009                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5010                             (sbi->s_mount_state & EXT4_VALID_FS))
5011                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5012
5013                         if (sbi->s_journal)
5014                                 ext4_mark_recovery_complete(sb, es);
5015                 } else {
5016                         /* Make sure we can mount this feature set readwrite */
5017                         if (ext4_has_feature_readonly(sb) ||
5018                             !ext4_feature_set_ok(sb, 0)) {
5019                                 err = -EROFS;
5020                                 goto restore_opts;
5021                         }
5022                         /*
5023                          * Make sure the group descriptor checksums
5024                          * are sane.  If they aren't, refuse to remount r/w.
5025                          */
5026                         for (g = 0; g < sbi->s_groups_count; g++) {
5027                                 struct ext4_group_desc *gdp =
5028                                         ext4_get_group_desc(sb, g, NULL);
5029
5030                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5031                                         ext4_msg(sb, KERN_ERR,
5032                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5033                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5034                                                le16_to_cpu(gdp->bg_checksum));
5035                                         err = -EFSBADCRC;
5036                                         goto restore_opts;
5037                                 }
5038                         }
5039
5040                         /*
5041                          * If we have an unprocessed orphan list hanging
5042                          * around from a previously readonly bdev mount,
5043                          * require a full umount/remount for now.
5044                          */
5045                         if (es->s_last_orphan) {
5046                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5047                                        "remount RDWR because of unprocessed "
5048                                        "orphan inode list.  Please "
5049                                        "umount/remount instead");
5050                                 err = -EINVAL;
5051                                 goto restore_opts;
5052                         }
5053
5054                         /*
5055                          * Mounting a RDONLY partition read-write, so reread
5056                          * and store the current valid flag.  (It may have
5057                          * been changed by e2fsck since we originally mounted
5058                          * the partition.)
5059                          */
5060                         if (sbi->s_journal)
5061                                 ext4_clear_journal_err(sb, es);
5062                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5063                         if (!ext4_setup_super(sb, es, 0))
5064                                 sb->s_flags &= ~MS_RDONLY;
5065                         if (ext4_has_feature_mmp(sb))
5066                                 if (ext4_multi_mount_protect(sb,
5067                                                 le64_to_cpu(es->s_mmp_block))) {
5068                                         err = -EROFS;
5069                                         goto restore_opts;
5070                                 }
5071                         enable_quota = 1;
5072                 }
5073         }
5074
5075         /*
5076          * Reinitialize lazy itable initialization thread based on
5077          * current settings
5078          */
5079         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5080                 ext4_unregister_li_request(sb);
5081         else {
5082                 ext4_group_t first_not_zeroed;
5083                 first_not_zeroed = ext4_has_uninit_itable(sb);
5084                 ext4_register_li_request(sb, first_not_zeroed);
5085         }
5086
5087         ext4_setup_system_zone(sb);
5088         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5089                 ext4_commit_super(sb, 1);
5090
5091 #ifdef CONFIG_QUOTA
5092         /* Release old quota file names */
5093         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5094                 kfree(old_opts.s_qf_names[i]);
5095         if (enable_quota) {
5096                 if (sb_any_quota_suspended(sb))
5097                         dquot_resume(sb, -1);
5098                 else if (ext4_has_feature_quota(sb)) {
5099                         err = ext4_enable_quotas(sb);
5100                         if (err)
5101                                 goto restore_opts;
5102                 }
5103         }
5104 #endif
5105
5106         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5107         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5108         kfree(orig_data);
5109         return 0;
5110
5111 restore_opts:
5112         sb->s_flags = old_sb_flags;
5113         sbi->s_mount_opt = old_opts.s_mount_opt;
5114         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5115         sbi->s_resuid = old_opts.s_resuid;
5116         sbi->s_resgid = old_opts.s_resgid;
5117         sbi->s_commit_interval = old_opts.s_commit_interval;
5118         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5119         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5120 #ifdef CONFIG_QUOTA
5121         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5122         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5123                 kfree(sbi->s_qf_names[i]);
5124                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5125         }
5126 #endif
5127         kfree(orig_data);
5128         return err;
5129 }
5130
5131 #ifdef CONFIG_QUOTA
5132 static int ext4_statfs_project(struct super_block *sb,
5133                                kprojid_t projid, struct kstatfs *buf)
5134 {
5135         struct kqid qid;
5136         struct dquot *dquot;
5137         u64 limit;
5138         u64 curblock;
5139
5140         qid = make_kqid_projid(projid);
5141         dquot = dqget(sb, qid);
5142         if (IS_ERR(dquot))
5143                 return PTR_ERR(dquot);
5144         spin_lock(&dq_data_lock);
5145
5146         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5147                  dquot->dq_dqb.dqb_bsoftlimit :
5148                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5149         if (limit && buf->f_blocks > limit) {
5150                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5151                 buf->f_blocks = limit;
5152                 buf->f_bfree = buf->f_bavail =
5153                         (buf->f_blocks > curblock) ?
5154                          (buf->f_blocks - curblock) : 0;
5155         }
5156
5157         limit = dquot->dq_dqb.dqb_isoftlimit ?
5158                 dquot->dq_dqb.dqb_isoftlimit :
5159                 dquot->dq_dqb.dqb_ihardlimit;
5160         if (limit && buf->f_files > limit) {
5161                 buf->f_files = limit;
5162                 buf->f_ffree =
5163                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5164                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5165         }
5166
5167         spin_unlock(&dq_data_lock);
5168         dqput(dquot);
5169         return 0;
5170 }
5171 #endif
5172
5173 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5174 {
5175         struct super_block *sb = dentry->d_sb;
5176         struct ext4_sb_info *sbi = EXT4_SB(sb);
5177         struct ext4_super_block *es = sbi->s_es;
5178         ext4_fsblk_t overhead = 0, resv_blocks;
5179         u64 fsid;
5180         s64 bfree;
5181         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5182
5183         if (!test_opt(sb, MINIX_DF))
5184                 overhead = sbi->s_overhead;
5185
5186         buf->f_type = EXT4_SUPER_MAGIC;
5187         buf->f_bsize = sb->s_blocksize;
5188         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5189         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5190                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5191         /* prevent underflow in case that few free space is available */
5192         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5193         buf->f_bavail = buf->f_bfree -
5194                         (ext4_r_blocks_count(es) + resv_blocks);
5195         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5196                 buf->f_bavail = 0;
5197         buf->f_files = le32_to_cpu(es->s_inodes_count);
5198         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5199         buf->f_namelen = EXT4_NAME_LEN;
5200         fsid = le64_to_cpup((void *)es->s_uuid) ^
5201                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5202         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5203         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5204
5205 #ifdef CONFIG_QUOTA
5206         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5207             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5208                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5209 #endif
5210         return 0;
5211 }
5212
5213 /* Helper function for writing quotas on sync - we need to start transaction
5214  * before quota file is locked for write. Otherwise the are possible deadlocks:
5215  * Process 1                         Process 2
5216  * ext4_create()                     quota_sync()
5217  *   jbd2_journal_start()                  write_dquot()
5218  *   dquot_initialize()                         down(dqio_mutex)
5219  *     down(dqio_mutex)                    jbd2_journal_start()
5220  *
5221  */
5222
5223 #ifdef CONFIG_QUOTA
5224
5225 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5226 {
5227         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5228 }
5229
5230 static int ext4_write_dquot(struct dquot *dquot)
5231 {
5232         int ret, err;
5233         handle_t *handle;
5234         struct inode *inode;
5235
5236         inode = dquot_to_inode(dquot);
5237         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5238                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5239         if (IS_ERR(handle))
5240                 return PTR_ERR(handle);
5241         ret = dquot_commit(dquot);
5242         err = ext4_journal_stop(handle);
5243         if (!ret)
5244                 ret = err;
5245         return ret;
5246 }
5247
5248 static int ext4_acquire_dquot(struct dquot *dquot)
5249 {
5250         int ret, err;
5251         handle_t *handle;
5252
5253         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5254                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5255         if (IS_ERR(handle))
5256                 return PTR_ERR(handle);
5257         ret = dquot_acquire(dquot);
5258         err = ext4_journal_stop(handle);
5259         if (!ret)
5260                 ret = err;
5261         return ret;
5262 }
5263
5264 static int ext4_release_dquot(struct dquot *dquot)
5265 {
5266         int ret, err;
5267         handle_t *handle;
5268
5269         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5270                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5271         if (IS_ERR(handle)) {
5272                 /* Release dquot anyway to avoid endless cycle in dqput() */
5273                 dquot_release(dquot);
5274                 return PTR_ERR(handle);
5275         }
5276         ret = dquot_release(dquot);
5277         err = ext4_journal_stop(handle);
5278         if (!ret)
5279                 ret = err;
5280         return ret;
5281 }
5282
5283 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5284 {
5285         struct super_block *sb = dquot->dq_sb;
5286         struct ext4_sb_info *sbi = EXT4_SB(sb);
5287
5288         /* Are we journaling quotas? */
5289         if (ext4_has_feature_quota(sb) ||
5290             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5291                 dquot_mark_dquot_dirty(dquot);
5292                 return ext4_write_dquot(dquot);
5293         } else {
5294                 return dquot_mark_dquot_dirty(dquot);
5295         }
5296 }
5297
5298 static int ext4_write_info(struct super_block *sb, int type)
5299 {
5300         int ret, err;
5301         handle_t *handle;
5302
5303         /* Data block + inode block */
5304         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5305         if (IS_ERR(handle))
5306                 return PTR_ERR(handle);
5307         ret = dquot_commit_info(sb, type);
5308         err = ext4_journal_stop(handle);
5309         if (!ret)
5310                 ret = err;
5311         return ret;
5312 }
5313
5314 /*
5315  * Turn on quotas during mount time - we need to find
5316  * the quota file and such...
5317  */
5318 static int ext4_quota_on_mount(struct super_block *sb, int type)
5319 {
5320         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5321                                         EXT4_SB(sb)->s_jquota_fmt, type);
5322 }
5323
5324 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5325 {
5326         struct ext4_inode_info *ei = EXT4_I(inode);
5327
5328         /* The first argument of lockdep_set_subclass has to be
5329          * *exactly* the same as the argument to init_rwsem() --- in
5330          * this case, in init_once() --- or lockdep gets unhappy
5331          * because the name of the lock is set using the
5332          * stringification of the argument to init_rwsem().
5333          */
5334         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5335         lockdep_set_subclass(&ei->i_data_sem, subclass);
5336 }
5337
5338 /*
5339  * Standard function to be called on quota_on
5340  */
5341 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5342                          const struct path *path)
5343 {
5344         int err;
5345
5346         if (!test_opt(sb, QUOTA))
5347                 return -EINVAL;
5348
5349         /* Quotafile not on the same filesystem? */
5350         if (path->dentry->d_sb != sb)
5351                 return -EXDEV;
5352         /* Journaling quota? */
5353         if (EXT4_SB(sb)->s_qf_names[type]) {
5354                 /* Quotafile not in fs root? */
5355                 if (path->dentry->d_parent != sb->s_root)
5356                         ext4_msg(sb, KERN_WARNING,
5357                                 "Quota file not on filesystem root. "
5358                                 "Journaled quota will not work");
5359         }
5360
5361         /*
5362          * When we journal data on quota file, we have to flush journal to see
5363          * all updates to the file when we bypass pagecache...
5364          */
5365         if (EXT4_SB(sb)->s_journal &&
5366             ext4_should_journal_data(d_inode(path->dentry))) {
5367                 /*
5368                  * We don't need to lock updates but journal_flush() could
5369                  * otherwise be livelocked...
5370                  */
5371                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5372                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5373                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5374                 if (err)
5375                         return err;
5376         }
5377
5378         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5379         err = dquot_quota_on(sb, type, format_id, path);
5380         if (err) {
5381                 lockdep_set_quota_inode(path->dentry->d_inode,
5382                                              I_DATA_SEM_NORMAL);
5383         } else {
5384                 struct inode *inode = d_inode(path->dentry);
5385                 handle_t *handle;
5386
5387                 /*
5388                  * Set inode flags to prevent userspace from messing with quota
5389                  * files. If this fails, we return success anyway since quotas
5390                  * are already enabled and this is not a hard failure.
5391                  */
5392                 inode_lock(inode);
5393                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5394                 if (IS_ERR(handle))
5395                         goto unlock_inode;
5396                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5397                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5398                                 S_NOATIME | S_IMMUTABLE);
5399                 ext4_mark_inode_dirty(handle, inode);
5400                 ext4_journal_stop(handle);
5401         unlock_inode:
5402                 inode_unlock(inode);
5403         }
5404         return err;
5405 }
5406
5407 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5408                              unsigned int flags)
5409 {
5410         int err;
5411         struct inode *qf_inode;
5412         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5413                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5414                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5415                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5416         };
5417
5418         BUG_ON(!ext4_has_feature_quota(sb));
5419
5420         if (!qf_inums[type])
5421                 return -EPERM;
5422
5423         qf_inode = ext4_iget(sb, qf_inums[type]);
5424         if (IS_ERR(qf_inode)) {
5425                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5426                 return PTR_ERR(qf_inode);
5427         }
5428
5429         /* Don't account quota for quota files to avoid recursion */
5430         qf_inode->i_flags |= S_NOQUOTA;
5431         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5432         err = dquot_enable(qf_inode, type, format_id, flags);
5433         iput(qf_inode);
5434         if (err)
5435                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5436
5437         return err;
5438 }
5439
5440 /* Enable usage tracking for all quota types. */
5441 static int ext4_enable_quotas(struct super_block *sb)
5442 {
5443         int type, err = 0;
5444         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5445                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5446                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5447                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5448         };
5449         bool quota_mopt[EXT4_MAXQUOTAS] = {
5450                 test_opt(sb, USRQUOTA),
5451                 test_opt(sb, GRPQUOTA),
5452                 test_opt(sb, PRJQUOTA),
5453         };
5454
5455         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5456         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5457                 if (qf_inums[type]) {
5458                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5459                                 DQUOT_USAGE_ENABLED |
5460                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5461                         if (err) {
5462                                 ext4_warning(sb,
5463                                         "Failed to enable quota tracking "
5464                                         "(type=%d, err=%d). Please run "
5465                                         "e2fsck to fix.", type, err);
5466                                 return err;
5467                         }
5468                 }
5469         }
5470         return 0;
5471 }
5472
5473 static int ext4_quota_off(struct super_block *sb, int type)
5474 {
5475         struct inode *inode = sb_dqopt(sb)->files[type];
5476         handle_t *handle;
5477         int err;
5478
5479         /* Force all delayed allocation blocks to be allocated.
5480          * Caller already holds s_umount sem */
5481         if (test_opt(sb, DELALLOC))
5482                 sync_filesystem(sb);
5483
5484         if (!inode || !igrab(inode))
5485                 goto out;
5486
5487         err = dquot_quota_off(sb, type);
5488         if (err)
5489                 goto out_put;
5490
5491         inode_lock(inode);
5492         /*
5493          * Update modification times of quota files when userspace can
5494          * start looking at them. If we fail, we return success anyway since
5495          * this is not a hard failure and quotas are already disabled.
5496          */
5497         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5498         if (IS_ERR(handle))
5499                 goto out_unlock;
5500         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5501         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5502         inode->i_mtime = inode->i_ctime = current_time(inode);
5503         ext4_mark_inode_dirty(handle, inode);
5504         ext4_journal_stop(handle);
5505 out_unlock:
5506         inode_unlock(inode);
5507 out_put:
5508         iput(inode);
5509         return err;
5510 out:
5511         return dquot_quota_off(sb, type);
5512 }
5513
5514 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5515  * acquiring the locks... As quota files are never truncated and quota code
5516  * itself serializes the operations (and no one else should touch the files)
5517  * we don't have to be afraid of races */
5518 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5519                                size_t len, loff_t off)
5520 {
5521         struct inode *inode = sb_dqopt(sb)->files[type];
5522         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5523         int offset = off & (sb->s_blocksize - 1);
5524         int tocopy;
5525         size_t toread;
5526         struct buffer_head *bh;
5527         loff_t i_size = i_size_read(inode);
5528
5529         if (off > i_size)
5530                 return 0;
5531         if (off+len > i_size)
5532                 len = i_size-off;
5533         toread = len;
5534         while (toread > 0) {
5535                 tocopy = sb->s_blocksize - offset < toread ?
5536                                 sb->s_blocksize - offset : toread;
5537                 bh = ext4_bread(NULL, inode, blk, 0);
5538                 if (IS_ERR(bh))
5539                         return PTR_ERR(bh);
5540                 if (!bh)        /* A hole? */
5541                         memset(data, 0, tocopy);
5542                 else
5543                         memcpy(data, bh->b_data+offset, tocopy);
5544                 brelse(bh);
5545                 offset = 0;
5546                 toread -= tocopy;
5547                 data += tocopy;
5548                 blk++;
5549         }
5550         return len;
5551 }
5552
5553 /* Write to quotafile (we know the transaction is already started and has
5554  * enough credits) */
5555 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5556                                 const char *data, size_t len, loff_t off)
5557 {
5558         struct inode *inode = sb_dqopt(sb)->files[type];
5559         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5560         int err, offset = off & (sb->s_blocksize - 1);
5561         int retries = 0;
5562         struct buffer_head *bh;
5563         handle_t *handle = journal_current_handle();
5564
5565         if (EXT4_SB(sb)->s_journal && !handle) {
5566                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5567                         " cancelled because transaction is not started",
5568                         (unsigned long long)off, (unsigned long long)len);
5569                 return -EIO;
5570         }
5571         /*
5572          * Since we account only one data block in transaction credits,
5573          * then it is impossible to cross a block boundary.
5574          */
5575         if (sb->s_blocksize - offset < len) {
5576                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5577                         " cancelled because not block aligned",
5578                         (unsigned long long)off, (unsigned long long)len);
5579                 return -EIO;
5580         }
5581
5582         do {
5583                 bh = ext4_bread(handle, inode, blk,
5584                                 EXT4_GET_BLOCKS_CREATE |
5585                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5586         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5587                  ext4_should_retry_alloc(inode->i_sb, &retries));
5588         if (IS_ERR(bh))
5589                 return PTR_ERR(bh);
5590         if (!bh)
5591                 goto out;
5592         BUFFER_TRACE(bh, "get write access");
5593         err = ext4_journal_get_write_access(handle, bh);
5594         if (err) {
5595                 brelse(bh);
5596                 return err;
5597         }
5598         lock_buffer(bh);
5599         memcpy(bh->b_data+offset, data, len);
5600         flush_dcache_page(bh->b_page);
5601         unlock_buffer(bh);
5602         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5603         brelse(bh);
5604 out:
5605         if (inode->i_size < off + len) {
5606                 i_size_write(inode, off + len);
5607                 EXT4_I(inode)->i_disksize = inode->i_size;
5608                 ext4_mark_inode_dirty(handle, inode);
5609         }
5610         return len;
5611 }
5612
5613 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5614 {
5615         const struct quota_format_ops   *ops;
5616
5617         if (!sb_has_quota_loaded(sb, qid->type))
5618                 return -ESRCH;
5619         ops = sb_dqopt(sb)->ops[qid->type];
5620         if (!ops || !ops->get_next_id)
5621                 return -ENOSYS;
5622         return dquot_get_next_id(sb, qid);
5623 }
5624 #endif
5625
5626 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5627                        const char *dev_name, void *data)
5628 {
5629         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5630 }
5631
5632 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5633 static inline void register_as_ext2(void)
5634 {
5635         int err = register_filesystem(&ext2_fs_type);
5636         if (err)
5637                 printk(KERN_WARNING
5638                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5639 }
5640
5641 static inline void unregister_as_ext2(void)
5642 {
5643         unregister_filesystem(&ext2_fs_type);
5644 }
5645
5646 static inline int ext2_feature_set_ok(struct super_block *sb)
5647 {
5648         if (ext4_has_unknown_ext2_incompat_features(sb))
5649                 return 0;
5650         if (sb->s_flags & MS_RDONLY)
5651                 return 1;
5652         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5653                 return 0;
5654         return 1;
5655 }
5656 #else
5657 static inline void register_as_ext2(void) { }
5658 static inline void unregister_as_ext2(void) { }
5659 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5660 #endif
5661
5662 static inline void register_as_ext3(void)
5663 {
5664         int err = register_filesystem(&ext3_fs_type);
5665         if (err)
5666                 printk(KERN_WARNING
5667                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5668 }
5669
5670 static inline void unregister_as_ext3(void)
5671 {
5672         unregister_filesystem(&ext3_fs_type);
5673 }
5674
5675 static inline int ext3_feature_set_ok(struct super_block *sb)
5676 {
5677         if (ext4_has_unknown_ext3_incompat_features(sb))
5678                 return 0;
5679         if (!ext4_has_feature_journal(sb))
5680                 return 0;
5681         if (sb->s_flags & MS_RDONLY)
5682                 return 1;
5683         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5684                 return 0;
5685         return 1;
5686 }
5687
5688 static struct file_system_type ext4_fs_type = {
5689         .owner          = THIS_MODULE,
5690         .name           = "ext4",
5691         .mount          = ext4_mount,
5692         .kill_sb        = kill_block_super,
5693         .fs_flags       = FS_REQUIRES_DEV,
5694 };
5695 MODULE_ALIAS_FS("ext4");
5696
5697 /* Shared across all ext4 file systems */
5698 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5699
5700 static int __init ext4_init_fs(void)
5701 {
5702         int i, err;
5703
5704         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5705         ext4_li_info = NULL;
5706         mutex_init(&ext4_li_mtx);
5707
5708         /* Build-time check for flags consistency */
5709         ext4_check_flag_values();
5710
5711         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5712                 init_waitqueue_head(&ext4__ioend_wq[i]);
5713
5714         err = ext4_init_es();
5715         if (err)
5716                 return err;
5717
5718         err = ext4_init_pageio();
5719         if (err)
5720                 goto out5;
5721
5722         err = ext4_init_system_zone();
5723         if (err)
5724                 goto out4;
5725
5726         err = ext4_init_sysfs();
5727         if (err)
5728                 goto out3;
5729
5730         err = ext4_init_mballoc();
5731         if (err)
5732                 goto out2;
5733         err = init_inodecache();
5734         if (err)
5735                 goto out1;
5736         register_as_ext3();
5737         register_as_ext2();
5738         err = register_filesystem(&ext4_fs_type);
5739         if (err)
5740                 goto out;
5741
5742         return 0;
5743 out:
5744         unregister_as_ext2();
5745         unregister_as_ext3();
5746         destroy_inodecache();
5747 out1:
5748         ext4_exit_mballoc();
5749 out2:
5750         ext4_exit_sysfs();
5751 out3:
5752         ext4_exit_system_zone();
5753 out4:
5754         ext4_exit_pageio();
5755 out5:
5756         ext4_exit_es();
5757
5758         return err;
5759 }
5760
5761 static void __exit ext4_exit_fs(void)
5762 {
5763         ext4_destroy_lazyinit_thread();
5764         unregister_as_ext2();
5765         unregister_as_ext3();
5766         unregister_filesystem(&ext4_fs_type);
5767         destroy_inodecache();
5768         ext4_exit_mballoc();
5769         ext4_exit_sysfs();
5770         ext4_exit_system_zone();
5771         ext4_exit_pageio();
5772         ext4_exit_es();
5773 }
5774
5775 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5776 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5777 MODULE_LICENSE("GPL");
5778 module_init(ext4_init_fs)
5779 module_exit(ext4_exit_fs)