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
 377         ext4_process_freed_data(sb, txn->t_tid);
 378
 379         spin_lock(&sbi->s_md_lock);
 380         while (!list_empty(&txn->t_private_list)) {
 381                 jce = list_entry(txn->t_private_list.next,
 382                                  struct ext4_journal_cb_entry, jce_list);
 383                 list_del_init(&jce->jce_list);
 384                 spin_unlock(&sbi->s_md_lock);
 385                 jce->jce_func(sb, jce, error);
 386                 spin_lock(&sbi->s_md_lock);
 387         }
 388         spin_unlock(&sbi->s_md_lock);
 389 }
 390
 391 /* Deal with the reporting of failure conditions on a filesystem such as
 392  * inconsistencies detected or read IO failures.
 393  *
 394  * On ext2, we can store the error state of the filesystem in the
 395  * superblock.  That is not possible on ext4, because we may have other
 396  * write ordering constraints on the superblock which prevent us from
 397  * writing it out straight away; and given that the journal is about to
 398  * be aborted, we can't rely on the current, or future, transactions to
 399  * write out the superblock safely.
 400  *
 401  * We'll just use the jbd2_journal_abort() error code to record an error in
 402  * the journal instead.  On recovery, the journal will complain about
 403  * that error until we've noted it down and cleared it.
 404  */
 405
 406 static void ext4_handle_error(struct super_block *sb)
 407 {
 408         if (sb->s_flags & MS_RDONLY)
 409                 return;
 410
 411         if (!test_opt(sb, ERRORS_CONT)) {
 412                 journal_t *journal = EXT4_SB(sb)->s_journal;
 413
 414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 415                 if (journal)
 416                         jbd2_journal_abort(journal, -EIO);
 417         }
 418         if (test_opt(sb, ERRORS_RO)) {
 419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 420                 /*
 421                  * Make sure updated value of ->s_mount_flags will be visible
 422                  * before ->s_flags update
 423                  */
 424                 smp_wmb();
 425                 sb->s_flags |= MS_RDONLY;
 426         }
 427         if (test_opt(sb, ERRORS_PANIC)) {
 428                 if (EXT4_SB(sb)->s_journal &&
 429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 430                         return;
 431                 panic("EXT4-fs (device %s): panic forced after error\n",
 432                         sb->s_id);
 433         }
 434 }
 435
 436 #define ext4_error_ratelimit(sb)                                        \
 437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 438                              "EXT4-fs error")
 439
 440 void __ext4_error(struct super_block *sb, const char *function,
 441                   unsigned int line, const char *fmt, ...)
 442 {
 443         struct va_format vaf;
 444         va_list args;
 445
 446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 447                 return;
 448
 449         if (ext4_error_ratelimit(sb)) {
 450                 va_start(args, fmt);
 451                 vaf.fmt = fmt;
 452                 vaf.va = &args;
 453                 printk(KERN_CRIT
 454                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 455                        sb->s_id, function, line, current->comm, &vaf);
 456                 va_end(args);
 457         }
 458         save_error_info(sb, function, line);
 459         ext4_handle_error(sb);
 460 }
 461
 462 void __ext4_error_inode(struct inode *inode, const char *function,
 463                         unsigned int line, ext4_fsblk_t block,
 464                         const char *fmt, ...)
 465 {
 466         va_list args;
 467         struct va_format vaf;
 468         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 469
 470         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 471                 return;
 472
 473         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 474         es->s_last_error_block = cpu_to_le64(block);
 475         if (ext4_error_ratelimit(inode->i_sb)) {
 476                 va_start(args, fmt);
 477                 vaf.fmt = fmt;
 478                 vaf.va = &args;
 479                 if (block)
 480                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 481                                "inode #%lu: block %llu: comm %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                block, current->comm, &vaf);
 484                 else
 485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 486                                "inode #%lu: comm %s: %pV\n",
 487                                inode->i_sb->s_id, function, line, inode->i_ino,
 488                                current->comm, &vaf);
 489                 va_end(args);
 490         }
 491         save_error_info(inode->i_sb, function, line);
 492         ext4_handle_error(inode->i_sb);
 493 }
 494
 495 void __ext4_error_file(struct file *file, const char *function,
 496                        unsigned int line, ext4_fsblk_t block,
 497                        const char *fmt, ...)
 498 {
 499         va_list args;
 500         struct va_format vaf;
 501         struct ext4_super_block *es;
 502         struct inode *inode = file_inode(file);
 503         char pathname[80], *path;
 504
 505         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 506                 return;
 507
 508         es = EXT4_SB(inode->i_sb)->s_es;
 509         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 510         if (ext4_error_ratelimit(inode->i_sb)) {
 511                 path = file_path(file, pathname, sizeof(pathname));
 512                 if (IS_ERR(path))
 513                         path = "(unknown)";
 514                 va_start(args, fmt);
 515                 vaf.fmt = fmt;
 516                 vaf.va = &args;
 517                 if (block)
 518                         printk(KERN_CRIT
 519                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 520                                "block %llu: comm %s: path %s: %pV\n",
 521                                inode->i_sb->s_id, function, line, inode->i_ino,
 522                                block, current->comm, path, &vaf);
 523                 else
 524                         printk(KERN_CRIT
 525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 526                                "comm %s: path %s: %pV\n",
 527                                inode->i_sb->s_id, function, line, inode->i_ino,
 528                                current->comm, path, &vaf);
 529                 va_end(args);
 530         }
 531         save_error_info(inode->i_sb, function, line);
 532         ext4_handle_error(inode->i_sb);
 533 }
 534
 535 const char *ext4_decode_error(struct super_block *sb, int errno,
 536                               char nbuf[16])
 537 {
 538         char *errstr = NULL;
 539
 540         switch (errno) {
 541         case -EFSCORRUPTED:
 542                 errstr = "Corrupt filesystem";
 543                 break;
 544         case -EFSBADCRC:
 545                 errstr = "Filesystem failed CRC";
 546                 break;
 547         case -EIO:
 548                 errstr = "IO failure";
 549                 break;
 550         case -ENOMEM:
 551                 errstr = "Out of memory";
 552                 break;
 553         case -EROFS:
 554                 if (!sb || (EXT4_SB(sb)->s_journal &&
 555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 556                         errstr = "Journal has aborted";
 557                 else
 558                         errstr = "Readonly filesystem";
 559                 break;
 560         default:
 561                 /* If the caller passed in an extra buffer for unknown
 562                  * errors, textualise them now.  Else we just return
 563                  * NULL. */
 564                 if (nbuf) {
 565                         /* Check for truncated error codes... */
 566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 567                                 errstr = nbuf;
 568                 }
 569                 break;
 570         }
 571
 572         return errstr;
 573 }
 574
 575 /* __ext4_std_error decodes expected errors from journaling functions
 576  * automatically and invokes the appropriate error response.  */
 577
 578 void __ext4_std_error(struct super_block *sb, const char *function,
 579                       unsigned int line, int errno)
 580 {
 581         char nbuf[16];
 582         const char *errstr;
 583
 584         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 585                 return;
 586
 587         /* Special case: if the error is EROFS, and we're not already
 588          * inside a transaction, then there's really no point in logging
 589          * an error. */
 590         if (errno == -EROFS && journal_current_handle() == NULL &&
 591             (sb->s_flags & MS_RDONLY))
 592                 return;
 593
 594         if (ext4_error_ratelimit(sb)) {
 595                 errstr = ext4_decode_error(sb, errno, nbuf);
 596                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 597                        sb->s_id, function, line, errstr);
 598         }
 599
 600         save_error_info(sb, function, line);
 601         ext4_handle_error(sb);
 602 }
 603
 604 /*
 605  * ext4_abort is a much stronger failure handler than ext4_error.  The
 606  * abort function may be used to deal with unrecoverable failures such
 607  * as journal IO errors or ENOMEM at a critical moment in log management.
 608  *
 609  * We unconditionally force the filesystem into an ABORT|READONLY state,
 610  * unless the error response on the fs has been set to panic in which
 611  * case we take the easy way out and panic immediately.
 612  */
 613
 614 void __ext4_abort(struct super_block *sb, const char *function,
 615                 unsigned int line, const char *fmt, ...)
 616 {
 617         struct va_format vaf;
 618         va_list args;
 619
 620         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 621                 return;
 622
 623         save_error_info(sb, function, line);
 624         va_start(args, fmt);
 625         vaf.fmt = fmt;
 626         vaf.va = &args;
 627         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 628                sb->s_id, function, line, &vaf);
 629         va_end(args);
 630
 631         if ((sb->s_flags & MS_RDONLY) == 0) {
 632                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 633                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 634                 /*
 635                  * Make sure updated value of ->s_mount_flags will be visible
 636                  * before ->s_flags update
 637                  */
 638                 smp_wmb();
 639                 sb->s_flags |= MS_RDONLY;
 640                 if (EXT4_SB(sb)->s_journal)
 641                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 642                 save_error_info(sb, function, line);
 643         }
 644         if (test_opt(sb, ERRORS_PANIC)) {
 645                 if (EXT4_SB(sb)->s_journal &&
 646                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 647                         return;
 648                 panic("EXT4-fs panic from previous error\n");
 649         }
 650 }
 651
 652 void __ext4_msg(struct super_block *sb,
 653                 const char *prefix, const char *fmt, ...)
 654 {
 655         struct va_format vaf;
 656         va_list args;
 657
 658         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 659                 return;
 660
 661         va_start(args, fmt);
 662         vaf.fmt = fmt;
 663         vaf.va = &args;
 664         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 665         va_end(args);
 666 }
 667
 668 #define ext4_warning_ratelimit(sb)                                      \
 669                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 670                              "EXT4-fs warning")
 671
 672 void __ext4_warning(struct super_block *sb, const char *function,
 673                     unsigned int line, const char *fmt, ...)
 674 {
 675         struct va_format vaf;
 676         va_list args;
 677
 678         if (!ext4_warning_ratelimit(sb))
 679                 return;
 680
 681         va_start(args, fmt);
 682         vaf.fmt = fmt;
 683         vaf.va = &args;
 684         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 685                sb->s_id, function, line, &vaf);
 686         va_end(args);
 687 }
 688
 689 void __ext4_warning_inode(const struct inode *inode, const char *function,
 690                           unsigned int line, const char *fmt, ...)
 691 {
 692         struct va_format vaf;
 693         va_list args;
 694
 695         if (!ext4_warning_ratelimit(inode->i_sb))
 696                 return;
 697
 698         va_start(args, fmt);
 699         vaf.fmt = fmt;
 700         vaf.va = &args;
 701         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 702                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 703                function, line, inode->i_ino, current->comm, &vaf);
 704         va_end(args);
 705 }
 706
 707 void __ext4_grp_locked_error(const char *function, unsigned int line,
 708                              struct super_block *sb, ext4_group_t grp,
 709                              unsigned long ino, ext4_fsblk_t block,
 710                              const char *fmt, ...)
 711 __releases(bitlock)
 712 __acquires(bitlock)
 713 {
 714         struct va_format vaf;
 715         va_list args;
 716         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 717
 718         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 719                 return;
 720
 721         es->s_last_error_ino = cpu_to_le32(ino);
 722         es->s_last_error_block = cpu_to_le64(block);
 723         __save_error_info(sb, function, line);
 724
 725         if (ext4_error_ratelimit(sb)) {
 726                 va_start(args, fmt);
 727                 vaf.fmt = fmt;
 728                 vaf.va = &args;
 729                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 730                        sb->s_id, function, line, grp);
 731                 if (ino)
 732                         printk(KERN_CONT "inode %lu: ", ino);
 733                 if (block)
 734                         printk(KERN_CONT "block %llu:",
 735                                (unsigned long long) block);
 736                 printk(KERN_CONT "%pV\n", &vaf);
 737                 va_end(args);
 738         }
 739
 740         if (test_opt(sb, ERRORS_CONT)) {
 741                 ext4_commit_super(sb, 0);
 742                 return;
 743         }
 744
 745         ext4_unlock_group(sb, grp);
 746         ext4_handle_error(sb);
 747         /*
 748          * We only get here in the ERRORS_RO case; relocking the group
 749          * may be dangerous, but nothing bad will happen since the
 750          * filesystem will have already been marked read/only and the
 751          * journal has been aborted.  We return 1 as a hint to callers
 752          * who might what to use the return value from
 753          * ext4_grp_locked_error() to distinguish between the
 754          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 755          * aggressively from the ext4 function in question, with a
 756          * more appropriate error code.
 757          */
 758         ext4_lock_group(sb, grp);
 759         return;
 760 }
 761
 762 void ext4_update_dynamic_rev(struct super_block *sb)
 763 {
 764         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 765
 766         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 767                 return;
 768
 769         ext4_warning(sb,
 770                      "updating to rev %d because of new feature flag, "
 771                      "running e2fsck is recommended",
 772                      EXT4_DYNAMIC_REV);
 773
 774         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 775         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 776         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 777         /* leave es->s_feature_*compat flags alone */
 778         /* es->s_uuid will be set by e2fsck if empty */
 779
 780         /*
 781          * The rest of the superblock fields should be zero, and if not it
 782          * means they are likely already in use, so leave them alone.  We
 783          * can leave it up to e2fsck to clean up any inconsistencies there.
 784          */
 785 }
 786
 787 /*
 788  * Open the external journal device
 789  */
 790 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 791 {
 792         struct block_device *bdev;
 793         char b[BDEVNAME_SIZE];
 794
 795         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 796         if (IS_ERR(bdev))
 797                 goto fail;
 798         return bdev;
 799
 800 fail:
 801         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 802                         __bdevname(dev, b), PTR_ERR(bdev));
 803         return NULL;
 804 }
 805
 806 /*
 807  * Release the journal device
 808  */
 809 static void ext4_blkdev_put(struct block_device *bdev)
 810 {
 811         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 812 }
 813
 814 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 815 {
 816         struct block_device *bdev;
 817         bdev = sbi->journal_bdev;
 818         if (bdev) {
 819                 ext4_blkdev_put(bdev);
 820                 sbi->journal_bdev = NULL;
 821         }
 822 }
 823
 824 static inline struct inode *orphan_list_entry(struct list_head *l)
 825 {
 826         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 827 }
 828
 829 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 830 {
 831         struct list_head *l;
 832
 833         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 834                  le32_to_cpu(sbi->s_es->s_last_orphan));
 835
 836         printk(KERN_ERR "sb_info orphan list:\n");
 837         list_for_each(l, &sbi->s_orphan) {
 838                 struct inode *inode = orphan_list_entry(l);
 839                 printk(KERN_ERR "  "
 840                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 841                        inode->i_sb->s_id, inode->i_ino, inode,
 842                        inode->i_mode, inode->i_nlink,
 843                        NEXT_ORPHAN(inode));
 844         }
 845 }
 846
 847 #ifdef CONFIG_QUOTA
 848 static int ext4_quota_off(struct super_block *sb, int type);
 849
 850 static inline void ext4_quota_off_umount(struct super_block *sb)
 851 {
 852         int type;
 853
 854         /* Use our quota_off function to clear inode flags etc. */
 855         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 856                 ext4_quota_off(sb, type);
 857 }
 858 #else
 859 static inline void ext4_quota_off_umount(struct super_block *sb)
 860 {
 861 }
 862 #endif
 863
 864 static void ext4_put_super(struct super_block *sb)
 865 {
 866         struct ext4_sb_info *sbi = EXT4_SB(sb);
 867         struct ext4_super_block *es = sbi->s_es;
 868         int aborted = 0;
 869         int i, err;
 870
 871         ext4_unregister_li_request(sb);
 872         ext4_quota_off_umount(sb);
 873
 874         flush_workqueue(sbi->rsv_conversion_wq);
 875         destroy_workqueue(sbi->rsv_conversion_wq);
 876
 877         if (sbi->s_journal) {
 878                 aborted = is_journal_aborted(sbi->s_journal);
 879                 err = jbd2_journal_destroy(sbi->s_journal);
 880                 sbi->s_journal = NULL;
 881                 if ((err < 0) && !aborted)
 882                         ext4_abort(sb, "Couldn't clean up the journal");
 883         }
 884
 885         ext4_unregister_sysfs(sb);
 886         ext4_es_unregister_shrinker(sbi);
 887         del_timer_sync(&sbi->s_err_report);
 888         ext4_release_system_zone(sb);
 889         ext4_mb_release(sb);
 890         ext4_ext_release(sb);
 891
 892         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 893                 ext4_clear_feature_journal_needs_recovery(sb);
 894                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 895         }
 896         if (!(sb->s_flags & MS_RDONLY))
 897                 ext4_commit_super(sb, 1);
 898
 899         for (i = 0; i < sbi->s_gdb_count; i++)
 900                 brelse(sbi->s_group_desc[i]);
 901         kvfree(sbi->s_group_desc);
 902         kvfree(sbi->s_flex_groups);
 903         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 904         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 905         percpu_counter_destroy(&sbi->s_dirs_counter);
 906         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 907         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 908 #ifdef CONFIG_QUOTA
 909         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 910                 kfree(sbi->s_qf_names[i]);
 911 #endif
 912
 913         /* Debugging code just in case the in-memory inode orphan list
 914          * isn't empty.  The on-disk one can be non-empty if we've
 915          * detected an error and taken the fs readonly, but the
 916          * in-memory list had better be clean by this point. */
 917         if (!list_empty(&sbi->s_orphan))
 918                 dump_orphan_list(sb, sbi);
 919         J_ASSERT(list_empty(&sbi->s_orphan));
 920
 921         sync_blockdev(sb->s_bdev);
 922         invalidate_bdev(sb->s_bdev);
 923         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 924                 /*
 925                  * Invalidate the journal device's buffers.  We don't want them
 926                  * floating about in memory - the physical journal device may
 927                  * hotswapped, and it breaks the `ro-after' testing code.
 928                  */
 929                 sync_blockdev(sbi->journal_bdev);
 930                 invalidate_bdev(sbi->journal_bdev);
 931                 ext4_blkdev_remove(sbi);
 932         }
 933         if (sbi->s_ea_inode_cache) {
 934                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 935                 sbi->s_ea_inode_cache = NULL;
 936         }
 937         if (sbi->s_ea_block_cache) {
 938                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 939                 sbi->s_ea_block_cache = NULL;
 940         }
 941         if (sbi->s_mmp_tsk)
 942                 kthread_stop(sbi->s_mmp_tsk);
 943         brelse(sbi->s_sbh);
 944         sb->s_fs_info = NULL;
 945         /*
 946          * Now that we are completely done shutting down the
 947          * superblock, we need to actually destroy the kobject.
 948          */
 949         kobject_put(&sbi->s_kobj);
 950         wait_for_completion(&sbi->s_kobj_unregister);
 951         if (sbi->s_chksum_driver)
 952                 crypto_free_shash(sbi->s_chksum_driver);
 953         kfree(sbi->s_blockgroup_lock);
 954         kfree(sbi);
 955 }
 956
 957 static struct kmem_cache *ext4_inode_cachep;
 958
 959 /*
 960  * Called inside transaction, so use GFP_NOFS
 961  */
 962 static struct inode *ext4_alloc_inode(struct super_block *sb)
 963 {
 964         struct ext4_inode_info *ei;
 965
 966         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 967         if (!ei)
 968                 return NULL;
 969
 970         ei->vfs_inode.i_version = 1;
 971         spin_lock_init(&ei->i_raw_lock);
 972         INIT_LIST_HEAD(&ei->i_prealloc_list);
 973         spin_lock_init(&ei->i_prealloc_lock);
 974         ext4_es_init_tree(&ei->i_es_tree);
 975         rwlock_init(&ei->i_es_lock);
 976         INIT_LIST_HEAD(&ei->i_es_list);
 977         ei->i_es_all_nr = 0;
 978         ei->i_es_shk_nr = 0;
 979         ei->i_es_shrink_lblk = 0;
 980         ei->i_reserved_data_blocks = 0;
 981         ei->i_reserved_meta_blocks = 0;
 982         ei->i_allocated_meta_blocks = 0;
 983         ei->i_da_metadata_calc_len = 0;
 984         ei->i_da_metadata_calc_last_lblock = 0;
 985         spin_lock_init(&(ei->i_block_reservation_lock));
 986 #ifdef CONFIG_QUOTA
 987         ei->i_reserved_quota = 0;
 988         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 989 #endif
 990         ei->jinode = NULL;
 991         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 992         spin_lock_init(&ei->i_completed_io_lock);
 993         ei->i_sync_tid = 0;
 994         ei->i_datasync_tid = 0;
 995         atomic_set(&ei->i_unwritten, 0);
 996         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 997         return &ei->vfs_inode;
 998 }
 999
1000 static int ext4_drop_inode(struct inode *inode)
1001 {
1002         int drop = generic_drop_inode(inode);
1003
1004         trace_ext4_drop_inode(inode, drop);
1005         return drop;
1006 }
1007
1008 static void ext4_i_callback(struct rcu_head *head)
1009 {
1010         struct inode *inode = container_of(head, struct inode, i_rcu);
1011         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1012 }
1013
1014 static void ext4_destroy_inode(struct inode *inode)
1015 {
1016         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1017                 ext4_msg(inode->i_sb, KERN_ERR,
1018                          "Inode %lu (%p): orphan list check failed!",
1019                          inode->i_ino, EXT4_I(inode));
1020                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1021                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1022                                 true);
1023                 dump_stack();
1024         }
1025         call_rcu(&inode->i_rcu, ext4_i_callback);
1026 }
1027
1028 static void init_once(void *foo)
1029 {
1030         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1031
1032         INIT_LIST_HEAD(&ei->i_orphan);
1033         init_rwsem(&ei->xattr_sem);
1034         init_rwsem(&ei->i_data_sem);
1035         init_rwsem(&ei->i_mmap_sem);
1036         inode_init_once(&ei->vfs_inode);
1037 }
1038
1039 static int __init init_inodecache(void)
1040 {
1041         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1042                                              sizeof(struct ext4_inode_info),
1043                                              0, (SLAB_RECLAIM_ACCOUNT|
1044                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1045                                              init_once);
1046         if (ext4_inode_cachep == NULL)
1047                 return -ENOMEM;
1048         return 0;
1049 }
1050
1051 static void destroy_inodecache(void)
1052 {
1053         /*
1054          * Make sure all delayed rcu free inodes are flushed before we
1055          * destroy cache.
1056          */
1057         rcu_barrier();
1058         kmem_cache_destroy(ext4_inode_cachep);
1059 }
1060
1061 void ext4_clear_inode(struct inode *inode)
1062 {
1063         invalidate_inode_buffers(inode);
1064         clear_inode(inode);
1065         dquot_drop(inode);
1066         ext4_discard_preallocations(inode);
1067         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1068         if (EXT4_I(inode)->jinode) {
1069                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1070                                                EXT4_I(inode)->jinode);
1071                 jbd2_free_inode(EXT4_I(inode)->jinode);
1072                 EXT4_I(inode)->jinode = NULL;
1073         }
1074 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1075         fscrypt_put_encryption_info(inode, NULL);
1076 #endif
1077 }
1078
1079 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1080                                         u64 ino, u32 generation)
1081 {
1082         struct inode *inode;
1083
1084         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1085                 return ERR_PTR(-ESTALE);
1086         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1087                 return ERR_PTR(-ESTALE);
1088
1089         /* iget isn't really right if the inode is currently unallocated!!
1090          *
1091          * ext4_read_inode will return a bad_inode if the inode had been
1092          * deleted, so we should be safe.
1093          *
1094          * Currently we don't know the generation for parent directory, so
1095          * a generation of 0 means "accept any"
1096          */
1097         inode = ext4_iget_normal(sb, ino);
1098         if (IS_ERR(inode))
1099                 return ERR_CAST(inode);
1100         if (generation && inode->i_generation != generation) {
1101                 iput(inode);
1102                 return ERR_PTR(-ESTALE);
1103         }
1104
1105         return inode;
1106 }
1107
1108 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1109                                         int fh_len, int fh_type)
1110 {
1111         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1112                                     ext4_nfs_get_inode);
1113 }
1114
1115 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1116                                         int fh_len, int fh_type)
1117 {
1118         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1119                                     ext4_nfs_get_inode);
1120 }
1121
1122 /*
1123  * Try to release metadata pages (indirect blocks, directories) which are
1124  * mapped via the block device.  Since these pages could have journal heads
1125  * which would prevent try_to_free_buffers() from freeing them, we must use
1126  * jbd2 layer's try_to_free_buffers() function to release them.
1127  */
1128 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1129                                  gfp_t wait)
1130 {
1131         journal_t *journal = EXT4_SB(sb)->s_journal;
1132
1133         WARN_ON(PageChecked(page));
1134         if (!page_has_buffers(page))
1135                 return 0;
1136         if (journal)
1137                 return jbd2_journal_try_to_free_buffers(journal, page,
1138                                                 wait & ~__GFP_DIRECT_RECLAIM);
1139         return try_to_free_buffers(page);
1140 }
1141
1142 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1143 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1144 {
1145         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1146                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1147 }
1148
1149 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1150                                                         void *fs_data)
1151 {
1152         handle_t *handle = fs_data;
1153         int res, res2, credits, retries = 0;
1154
1155         /*
1156          * Encrypting the root directory is not allowed because e2fsck expects
1157          * lost+found to exist and be unencrypted, and encrypting the root
1158          * directory would imply encrypting the lost+found directory as well as
1159          * the filename "lost+found" itself.
1160          */
1161         if (inode->i_ino == EXT4_ROOT_INO)
1162                 return -EPERM;
1163
1164         res = ext4_convert_inline_data(inode);
1165         if (res)
1166                 return res;
1167
1168         /*
1169          * If a journal handle was specified, then the encryption context is
1170          * being set on a new inode via inheritance and is part of a larger
1171          * transaction to create the inode.  Otherwise the encryption context is
1172          * being set on an existing inode in its own transaction.  Only in the
1173          * latter case should the "retry on ENOSPC" logic be used.
1174          */
1175
1176         if (handle) {
1177                 res = ext4_xattr_set_handle(handle, inode,
1178                                             EXT4_XATTR_INDEX_ENCRYPTION,
1179                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1180                                             ctx, len, 0);
1181                 if (!res) {
1182                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1183                         ext4_clear_inode_state(inode,
1184                                         EXT4_STATE_MAY_INLINE_DATA);
1185                         /*
1186                          * Update inode->i_flags - e.g. S_DAX may get disabled
1187                          */
1188                         ext4_set_inode_flags(inode);
1189                 }
1190                 return res;
1191         }
1192
1193         res = dquot_initialize(inode);
1194         if (res)
1195                 return res;
1196 retry:
1197         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1198                                      &credits);
1199         if (res)
1200                 return res;
1201
1202         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1203         if (IS_ERR(handle))
1204                 return PTR_ERR(handle);
1205
1206         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1207                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1208                                     ctx, len, 0);
1209         if (!res) {
1210                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1211                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1212                 ext4_set_inode_flags(inode);
1213                 res = ext4_mark_inode_dirty(handle, inode);
1214                 if (res)
1215                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1216         }
1217         res2 = ext4_journal_stop(handle);
1218
1219         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1220                 goto retry;
1221         if (!res)
1222                 res = res2;
1223         return res;
1224 }
1225
1226 static bool ext4_dummy_context(struct inode *inode)
1227 {
1228         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1229 }
1230
1231 static unsigned ext4_max_namelen(struct inode *inode)
1232 {
1233         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1234                 EXT4_NAME_LEN;
1235 }
1236
1237 static const struct fscrypt_operations ext4_cryptops = {
1238         .key_prefix             = "ext4:",
1239         .get_context            = ext4_get_context,
1240         .set_context            = ext4_set_context,
1241         .dummy_context          = ext4_dummy_context,
1242         .is_encrypted           = ext4_encrypted_inode,
1243         .empty_dir              = ext4_empty_dir,
1244         .max_namelen            = ext4_max_namelen,
1245 };
1246 #else
1247 static const struct fscrypt_operations ext4_cryptops = {
1248         .is_encrypted           = ext4_encrypted_inode,
1249 };
1250 #endif
1251
1252 #ifdef CONFIG_QUOTA
1253 static const char * const quotatypes[] = INITQFNAMES;
1254 #define QTYPE2NAME(t) (quotatypes[t])
1255
1256 static int ext4_write_dquot(struct dquot *dquot);
1257 static int ext4_acquire_dquot(struct dquot *dquot);
1258 static int ext4_release_dquot(struct dquot *dquot);
1259 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1260 static int ext4_write_info(struct super_block *sb, int type);
1261 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1262                          const struct path *path);
1263 static int ext4_quota_on_mount(struct super_block *sb, int type);
1264 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1265                                size_t len, loff_t off);
1266 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1267                                 const char *data, size_t len, loff_t off);
1268 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1269                              unsigned int flags);
1270 static int ext4_enable_quotas(struct super_block *sb);
1271 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1272
1273 static struct dquot **ext4_get_dquots(struct inode *inode)
1274 {
1275         return EXT4_I(inode)->i_dquot;
1276 }
1277
1278 static const struct dquot_operations ext4_quota_operations = {
1279         .get_reserved_space     = ext4_get_reserved_space,
1280         .write_dquot            = ext4_write_dquot,
1281         .acquire_dquot          = ext4_acquire_dquot,
1282         .release_dquot          = ext4_release_dquot,
1283         .mark_dirty             = ext4_mark_dquot_dirty,
1284         .write_info             = ext4_write_info,
1285         .alloc_dquot            = dquot_alloc,
1286         .destroy_dquot          = dquot_destroy,
1287         .get_projid             = ext4_get_projid,
1288         .get_inode_usage        = ext4_get_inode_usage,
1289         .get_next_id            = ext4_get_next_id,
1290 };
1291
1292 static const struct quotactl_ops ext4_qctl_operations = {
1293         .quota_on       = ext4_quota_on,
1294         .quota_off      = ext4_quota_off,
1295         .quota_sync     = dquot_quota_sync,
1296         .get_state      = dquot_get_state,
1297         .set_info       = dquot_set_dqinfo,
1298         .get_dqblk      = dquot_get_dqblk,
1299         .set_dqblk      = dquot_set_dqblk,
1300         .get_nextdqblk  = dquot_get_next_dqblk,
1301 };
1302 #endif
1303
1304 static const struct super_operations ext4_sops = {
1305         .alloc_inode    = ext4_alloc_inode,
1306         .destroy_inode  = ext4_destroy_inode,
1307         .write_inode    = ext4_write_inode,
1308         .dirty_inode    = ext4_dirty_inode,
1309         .drop_inode     = ext4_drop_inode,
1310         .evict_inode    = ext4_evict_inode,
1311         .put_super      = ext4_put_super,
1312         .sync_fs        = ext4_sync_fs,
1313         .freeze_fs      = ext4_freeze,
1314         .unfreeze_fs    = ext4_unfreeze,
1315         .statfs         = ext4_statfs,
1316         .remount_fs     = ext4_remount,
1317         .show_options   = ext4_show_options,
1318 #ifdef CONFIG_QUOTA
1319         .quota_read     = ext4_quota_read,
1320         .quota_write    = ext4_quota_write,
1321         .get_dquots     = ext4_get_dquots,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_removed,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1339         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1348         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1349         Opt_inode_readahead_blks, Opt_journal_ioprio,
1350         Opt_dioread_nolock, Opt_dioread_lock,
1351         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1352         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1353 };
1354
1355 static const match_table_t tokens = {
1356         {Opt_bsd_df, "bsddf"},
1357         {Opt_minix_df, "minixdf"},
1358         {Opt_grpid, "grpid"},
1359         {Opt_grpid, "bsdgroups"},
1360         {Opt_nogrpid, "nogrpid"},
1361         {Opt_nogrpid, "sysvgroups"},
1362         {Opt_resgid, "resgid=%u"},
1363         {Opt_resuid, "resuid=%u"},
1364         {Opt_sb, "sb=%u"},
1365         {Opt_err_cont, "errors=continue"},
1366         {Opt_err_panic, "errors=panic"},
1367         {Opt_err_ro, "errors=remount-ro"},
1368         {Opt_nouid32, "nouid32"},
1369         {Opt_debug, "debug"},
1370         {Opt_removed, "oldalloc"},
1371         {Opt_removed, "orlov"},
1372         {Opt_user_xattr, "user_xattr"},
1373         {Opt_nouser_xattr, "nouser_xattr"},
1374         {Opt_acl, "acl"},
1375         {Opt_noacl, "noacl"},
1376         {Opt_noload, "norecovery"},
1377         {Opt_noload, "noload"},
1378         {Opt_removed, "nobh"},
1379         {Opt_removed, "bh"},
1380         {Opt_commit, "commit=%u"},
1381         {Opt_min_batch_time, "min_batch_time=%u"},
1382         {Opt_max_batch_time, "max_batch_time=%u"},
1383         {Opt_journal_dev, "journal_dev=%u"},
1384         {Opt_journal_path, "journal_path=%s"},
1385         {Opt_journal_checksum, "journal_checksum"},
1386         {Opt_nojournal_checksum, "nojournal_checksum"},
1387         {Opt_journal_async_commit, "journal_async_commit"},
1388         {Opt_abort, "abort"},
1389         {Opt_data_journal, "data=journal"},
1390         {Opt_data_ordered, "data=ordered"},
1391         {Opt_data_writeback, "data=writeback"},
1392         {Opt_data_err_abort, "data_err=abort"},
1393         {Opt_data_err_ignore, "data_err=ignore"},
1394         {Opt_offusrjquota, "usrjquota="},
1395         {Opt_usrjquota, "usrjquota=%s"},
1396         {Opt_offgrpjquota, "grpjquota="},
1397         {Opt_grpjquota, "grpjquota=%s"},
1398         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1399         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1400         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1401         {Opt_grpquota, "grpquota"},
1402         {Opt_noquota, "noquota"},
1403         {Opt_quota, "quota"},
1404         {Opt_usrquota, "usrquota"},
1405         {Opt_prjquota, "prjquota"},
1406         {Opt_barrier, "barrier=%u"},
1407         {Opt_barrier, "barrier"},
1408         {Opt_nobarrier, "nobarrier"},
1409         {Opt_i_version, "i_version"},
1410         {Opt_dax, "dax"},
1411         {Opt_stripe, "stripe=%u"},
1412         {Opt_delalloc, "delalloc"},
1413         {Opt_lazytime, "lazytime"},
1414         {Opt_nolazytime, "nolazytime"},
1415         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1416         {Opt_nodelalloc, "nodelalloc"},
1417         {Opt_removed, "mblk_io_submit"},
1418         {Opt_removed, "nomblk_io_submit"},
1419         {Opt_block_validity, "block_validity"},
1420         {Opt_noblock_validity, "noblock_validity"},
1421         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1422         {Opt_journal_ioprio, "journal_ioprio=%u"},
1423         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1424         {Opt_auto_da_alloc, "auto_da_alloc"},
1425         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1426         {Opt_dioread_nolock, "dioread_nolock"},
1427         {Opt_dioread_lock, "dioread_lock"},
1428         {Opt_discard, "discard"},
1429         {Opt_nodiscard, "nodiscard"},
1430         {Opt_init_itable, "init_itable=%u"},
1431         {Opt_init_itable, "init_itable"},
1432         {Opt_noinit_itable, "noinit_itable"},
1433         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1434         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1435         {Opt_nombcache, "nombcache"},
1436         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1437         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1438         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1439         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1440         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1441         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1442         {Opt_err, NULL},
1443 };
1444
1445 static ext4_fsblk_t get_sb_block(void **data)
1446 {
1447         ext4_fsblk_t    sb_block;
1448         char            *options = (char *) *data;
1449
1450         if (!options || strncmp(options, "sb=", 3) != 0)
1451                 return 1;       /* Default location */
1452
1453         options += 3;
1454         /* TODO: use simple_strtoll with >32bit ext4 */
1455         sb_block = simple_strtoul(options, &options, 0);
1456         if (*options && *options != ',') {
1457                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1458                        (char *) *data);
1459                 return 1;
1460         }
1461         if (*options == ',')
1462                 options++;
1463         *data = (void *) options;
1464
1465         return sb_block;
1466 }
1467
1468 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1469 static const char deprecated_msg[] =
1470         "Mount option \"%s\" will be removed by %s\n"
1471         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1472
1473 #ifdef CONFIG_QUOTA
1474 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1475 {
1476         struct ext4_sb_info *sbi = EXT4_SB(sb);
1477         char *qname;
1478         int ret = -1;
1479
1480         if (sb_any_quota_loaded(sb) &&
1481                 !sbi->s_qf_names[qtype]) {
1482                 ext4_msg(sb, KERN_ERR,
1483                         "Cannot change journaled "
1484                         "quota options when quota turned on");
1485                 return -1;
1486         }
1487         if (ext4_has_feature_quota(sb)) {
1488                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1489                          "ignored when QUOTA feature is enabled");
1490                 return 1;
1491         }
1492         qname = match_strdup(args);
1493         if (!qname) {
1494                 ext4_msg(sb, KERN_ERR,
1495                         "Not enough memory for storing quotafile name");
1496                 return -1;
1497         }
1498         if (sbi->s_qf_names[qtype]) {
1499                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1500                         ret = 1;
1501                 else
1502                         ext4_msg(sb, KERN_ERR,
1503                                  "%s quota file already specified",
1504                                  QTYPE2NAME(qtype));
1505                 goto errout;
1506         }
1507         if (strchr(qname, '/')) {
1508                 ext4_msg(sb, KERN_ERR,
1509                         "quotafile must be on filesystem root");
1510                 goto errout;
1511         }
1512         sbi->s_qf_names[qtype] = qname;
1513         set_opt(sb, QUOTA);
1514         return 1;
1515 errout:
1516         kfree(qname);
1517         return ret;
1518 }
1519
1520 static int clear_qf_name(struct super_block *sb, int qtype)
1521 {
1522
1523         struct ext4_sb_info *sbi = EXT4_SB(sb);
1524
1525         if (sb_any_quota_loaded(sb) &&
1526                 sbi->s_qf_names[qtype]) {
1527                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1528                         " when quota turned on");
1529                 return -1;
1530         }
1531         kfree(sbi->s_qf_names[qtype]);
1532         sbi->s_qf_names[qtype] = NULL;
1533         return 1;
1534 }
1535 #endif
1536
1537 #define MOPT_SET        0x0001
1538 #define MOPT_CLEAR      0x0002
1539 #define MOPT_NOSUPPORT  0x0004
1540 #define MOPT_EXPLICIT   0x0008
1541 #define MOPT_CLEAR_ERR  0x0010
1542 #define MOPT_GTE0       0x0020
1543 #ifdef CONFIG_QUOTA
1544 #define MOPT_Q          0
1545 #define MOPT_QFMT       0x0040
1546 #else
1547 #define MOPT_Q          MOPT_NOSUPPORT
1548 #define MOPT_QFMT       MOPT_NOSUPPORT
1549 #endif
1550 #define MOPT_DATAJ      0x0080
1551 #define MOPT_NO_EXT2    0x0100
1552 #define MOPT_NO_EXT3    0x0200
1553 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1554 #define MOPT_STRING     0x0400
1555
1556 static const struct mount_opts {
1557         int     token;
1558         int     mount_opt;
1559         int     flags;
1560 } ext4_mount_opts[] = {
1561         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1562         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1563         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1564         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1565         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1566         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1567         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1568          MOPT_EXT4_ONLY | MOPT_SET},
1569         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1570          MOPT_EXT4_ONLY | MOPT_CLEAR},
1571         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1572         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1573         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1574          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1575         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1576          MOPT_EXT4_ONLY | MOPT_CLEAR},
1577         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1578          MOPT_EXT4_ONLY | MOPT_CLEAR},
1579         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1580          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1581         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1582                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1583          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1584         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1585         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1586         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1587         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1588         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1589          MOPT_NO_EXT2},
1590         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1591          MOPT_NO_EXT2},
1592         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1593         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1594         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1595         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1596         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1597         {Opt_commit, 0, MOPT_GTE0},
1598         {Opt_max_batch_time, 0, MOPT_GTE0},
1599         {Opt_min_batch_time, 0, MOPT_GTE0},
1600         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1601         {Opt_init_itable, 0, MOPT_GTE0},
1602         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1603         {Opt_stripe, 0, MOPT_GTE0},
1604         {Opt_resuid, 0, MOPT_GTE0},
1605         {Opt_resgid, 0, MOPT_GTE0},
1606         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1607         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1608         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1609         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1610         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1611         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1612          MOPT_NO_EXT2 | MOPT_DATAJ},
1613         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1614         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1615 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1616         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1617         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1618 #else
1619         {Opt_acl, 0, MOPT_NOSUPPORT},
1620         {Opt_noacl, 0, MOPT_NOSUPPORT},
1621 #endif
1622         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1623         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1624         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1625         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1626         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1627                                                         MOPT_SET | MOPT_Q},
1628         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1629                                                         MOPT_SET | MOPT_Q},
1630         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1631                                                         MOPT_SET | MOPT_Q},
1632         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1633                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1634                                                         MOPT_CLEAR | MOPT_Q},
1635         {Opt_usrjquota, 0, MOPT_Q},
1636         {Opt_grpjquota, 0, MOPT_Q},
1637         {Opt_offusrjquota, 0, MOPT_Q},
1638         {Opt_offgrpjquota, 0, MOPT_Q},
1639         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1640         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1641         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1642         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1643         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1644         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1645         {Opt_err, 0, 0}
1646 };
1647
1648 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1649                             substring_t *args, unsigned long *journal_devnum,
1650                             unsigned int *journal_ioprio, int is_remount)
1651 {
1652         struct ext4_sb_info *sbi = EXT4_SB(sb);
1653         const struct mount_opts *m;
1654         kuid_t uid;
1655         kgid_t gid;
1656         int arg = 0;
1657
1658 #ifdef CONFIG_QUOTA
1659         if (token == Opt_usrjquota)
1660                 return set_qf_name(sb, USRQUOTA, &args[0]);
1661         else if (token == Opt_grpjquota)
1662                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1663         else if (token == Opt_offusrjquota)
1664                 return clear_qf_name(sb, USRQUOTA);
1665         else if (token == Opt_offgrpjquota)
1666                 return clear_qf_name(sb, GRPQUOTA);
1667 #endif
1668         switch (token) {
1669         case Opt_noacl:
1670         case Opt_nouser_xattr:
1671                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1672                 break;
1673         case Opt_sb:
1674                 return 1;       /* handled by get_sb_block() */
1675         case Opt_removed:
1676                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1677                 return 1;
1678         case Opt_abort:
1679                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1680                 return 1;
1681         case Opt_i_version:
1682                 sb->s_flags |= MS_I_VERSION;
1683                 return 1;
1684         case Opt_lazytime:
1685                 sb->s_flags |= MS_LAZYTIME;
1686                 return 1;
1687         case Opt_nolazytime:
1688                 sb->s_flags &= ~MS_LAZYTIME;
1689                 return 1;
1690         }
1691
1692         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1693                 if (token == m->token)
1694                         break;
1695
1696         if (m->token == Opt_err) {
1697                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1698                          "or missing value", opt);
1699                 return -1;
1700         }
1701
1702         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1703                 ext4_msg(sb, KERN_ERR,
1704                          "Mount option \"%s\" incompatible with ext2", opt);
1705                 return -1;
1706         }
1707         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1708                 ext4_msg(sb, KERN_ERR,
1709                          "Mount option \"%s\" incompatible with ext3", opt);
1710                 return -1;
1711         }
1712
1713         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1714                 return -1;
1715         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1716                 return -1;
1717         if (m->flags & MOPT_EXPLICIT) {
1718                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1719                         set_opt2(sb, EXPLICIT_DELALLOC);
1720                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1721                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1722                 } else
1723                         return -1;
1724         }
1725         if (m->flags & MOPT_CLEAR_ERR)
1726                 clear_opt(sb, ERRORS_MASK);
1727         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1728                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1729                          "options when quota turned on");
1730                 return -1;
1731         }
1732
1733         if (m->flags & MOPT_NOSUPPORT) {
1734                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1735         } else if (token == Opt_commit) {
1736                 if (arg == 0)
1737                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1738                 sbi->s_commit_interval = HZ * arg;
1739         } else if (token == Opt_debug_want_extra_isize) {
1740                 sbi->s_want_extra_isize = arg;
1741         } else if (token == Opt_max_batch_time) {
1742                 sbi->s_max_batch_time = arg;
1743         } else if (token == Opt_min_batch_time) {
1744                 sbi->s_min_batch_time = arg;
1745         } else if (token == Opt_inode_readahead_blks) {
1746                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1747                         ext4_msg(sb, KERN_ERR,
1748                                  "EXT4-fs: inode_readahead_blks must be "
1749                                  "0 or a power of 2 smaller than 2^31");
1750                         return -1;
1751                 }
1752                 sbi->s_inode_readahead_blks = arg;
1753         } else if (token == Opt_init_itable) {
1754                 set_opt(sb, INIT_INODE_TABLE);
1755                 if (!args->from)
1756                         arg = EXT4_DEF_LI_WAIT_MULT;
1757                 sbi->s_li_wait_mult = arg;
1758         } else if (token == Opt_max_dir_size_kb) {
1759                 sbi->s_max_dir_size_kb = arg;
1760         } else if (token == Opt_stripe) {
1761                 sbi->s_stripe = arg;
1762         } else if (token == Opt_resuid) {
1763                 uid = make_kuid(current_user_ns(), arg);
1764                 if (!uid_valid(uid)) {
1765                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1766                         return -1;
1767                 }
1768                 sbi->s_resuid = uid;
1769         } else if (token == Opt_resgid) {
1770                 gid = make_kgid(current_user_ns(), arg);
1771                 if (!gid_valid(gid)) {
1772                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1773                         return -1;
1774                 }
1775                 sbi->s_resgid = gid;
1776         } else if (token == Opt_journal_dev) {
1777                 if (is_remount) {
1778                         ext4_msg(sb, KERN_ERR,
1779                                  "Cannot specify journal on remount");
1780                         return -1;
1781                 }
1782                 *journal_devnum = arg;
1783         } else if (token == Opt_journal_path) {
1784                 char *journal_path;
1785                 struct inode *journal_inode;
1786                 struct path path;
1787                 int error;
1788
1789                 if (is_remount) {
1790                         ext4_msg(sb, KERN_ERR,
1791                                  "Cannot specify journal on remount");
1792                         return -1;
1793                 }
1794                 journal_path = match_strdup(&args[0]);
1795                 if (!journal_path) {
1796                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1797                                 "journal device string");
1798                         return -1;
1799                 }
1800
1801                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1802                 if (error) {
1803                         ext4_msg(sb, KERN_ERR, "error: could not find "
1804                                 "journal device path: error %d", error);
1805                         kfree(journal_path);
1806                         return -1;
1807                 }
1808
1809                 journal_inode = d_inode(path.dentry);
1810                 if (!S_ISBLK(journal_inode->i_mode)) {
1811                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1812                                 "is not a block device", journal_path);
1813                         path_put(&path);
1814                         kfree(journal_path);
1815                         return -1;
1816                 }
1817
1818                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1819                 path_put(&path);
1820                 kfree(journal_path);
1821         } else if (token == Opt_journal_ioprio) {
1822                 if (arg > 7) {
1823                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1824                                  " (must be 0-7)");
1825                         return -1;
1826                 }
1827                 *journal_ioprio =
1828                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1829         } else if (token == Opt_test_dummy_encryption) {
1830 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1831                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1832                 ext4_msg(sb, KERN_WARNING,
1833                          "Test dummy encryption mode enabled");
1834 #else
1835                 ext4_msg(sb, KERN_WARNING,
1836                          "Test dummy encryption mount option ignored");
1837 #endif
1838         } else if (m->flags & MOPT_DATAJ) {
1839                 if (is_remount) {
1840                         if (!sbi->s_journal)
1841                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1842                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1843                                 ext4_msg(sb, KERN_ERR,
1844                                          "Cannot change data mode on remount");
1845                                 return -1;
1846                         }
1847                 } else {
1848                         clear_opt(sb, DATA_FLAGS);
1849                         sbi->s_mount_opt |= m->mount_opt;
1850                 }
1851 #ifdef CONFIG_QUOTA
1852         } else if (m->flags & MOPT_QFMT) {
1853                 if (sb_any_quota_loaded(sb) &&
1854                     sbi->s_jquota_fmt != m->mount_opt) {
1855                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1856                                  "quota options when quota turned on");
1857                         return -1;
1858                 }
1859                 if (ext4_has_feature_quota(sb)) {
1860                         ext4_msg(sb, KERN_INFO,
1861                                  "Quota format mount options ignored "
1862                                  "when QUOTA feature is enabled");
1863                         return 1;
1864                 }
1865                 sbi->s_jquota_fmt = m->mount_opt;
1866 #endif
1867         } else if (token == Opt_dax) {
1868 #ifdef CONFIG_FS_DAX
1869                 ext4_msg(sb, KERN_WARNING,
1870                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1871                         sbi->s_mount_opt |= m->mount_opt;
1872 #else
1873                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1874                 return -1;
1875 #endif
1876         } else if (token == Opt_data_err_abort) {
1877                 sbi->s_mount_opt |= m->mount_opt;
1878         } else if (token == Opt_data_err_ignore) {
1879                 sbi->s_mount_opt &= ~m->mount_opt;
1880         } else {
1881                 if (!args->from)
1882                         arg = 1;
1883                 if (m->flags & MOPT_CLEAR)
1884                         arg = !arg;
1885                 else if (unlikely(!(m->flags & MOPT_SET))) {
1886                         ext4_msg(sb, KERN_WARNING,
1887                                  "buggy handling of option %s", opt);
1888                         WARN_ON(1);
1889                         return -1;
1890                 }
1891                 if (arg != 0)
1892                         sbi->s_mount_opt |= m->mount_opt;
1893                 else
1894                         sbi->s_mount_opt &= ~m->mount_opt;
1895         }
1896         return 1;
1897 }
1898
1899 static int parse_options(char *options, struct super_block *sb,
1900                          unsigned long *journal_devnum,
1901                          unsigned int *journal_ioprio,
1902                          int is_remount)
1903 {
1904         struct ext4_sb_info *sbi = EXT4_SB(sb);
1905         char *p;
1906         substring_t args[MAX_OPT_ARGS];
1907         int token;
1908
1909         if (!options)
1910                 return 1;
1911
1912         while ((p = strsep(&options, ",")) != NULL) {
1913                 if (!*p)
1914                         continue;
1915                 /*
1916                  * Initialize args struct so we know whether arg was
1917                  * found; some options take optional arguments.
1918                  */
1919                 args[0].to = args[0].from = NULL;
1920                 token = match_token(p, tokens, args);
1921                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1922                                      journal_ioprio, is_remount) < 0)
1923                         return 0;
1924         }
1925 #ifdef CONFIG_QUOTA
1926         /*
1927          * We do the test below only for project quotas. 'usrquota' and
1928          * 'grpquota' mount options are allowed even without quota feature
1929          * to support legacy quotas in quota files.
1930          */
1931         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1932                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1933                          "Cannot enable project quota enforcement.");
1934                 return 0;
1935         }
1936         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1937                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1938                         clear_opt(sb, USRQUOTA);
1939
1940                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1941                         clear_opt(sb, GRPQUOTA);
1942
1943                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1944                         ext4_msg(sb, KERN_ERR, "old and new quota "
1945                                         "format mixing");
1946                         return 0;
1947                 }
1948
1949                 if (!sbi->s_jquota_fmt) {
1950                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1951                                         "not specified");
1952                         return 0;
1953                 }
1954         }
1955 #endif
1956         if (test_opt(sb, DIOREAD_NOLOCK)) {
1957                 int blocksize =
1958                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1959
1960                 if (blocksize < PAGE_SIZE) {
1961                         ext4_msg(sb, KERN_ERR, "can't mount with "
1962                                  "dioread_nolock if block size != PAGE_SIZE");
1963                         return 0;
1964                 }
1965         }
1966         return 1;
1967 }
1968
1969 static inline void ext4_show_quota_options(struct seq_file *seq,
1970                                            struct super_block *sb)
1971 {
1972 #if defined(CONFIG_QUOTA)
1973         struct ext4_sb_info *sbi = EXT4_SB(sb);
1974
1975         if (sbi->s_jquota_fmt) {
1976                 char *fmtname = "";
1977
1978                 switch (sbi->s_jquota_fmt) {
1979                 case QFMT_VFS_OLD:
1980                         fmtname = "vfsold";
1981                         break;
1982                 case QFMT_VFS_V0:
1983                         fmtname = "vfsv0";
1984                         break;
1985                 case QFMT_VFS_V1:
1986                         fmtname = "vfsv1";
1987                         break;
1988                 }
1989                 seq_printf(seq, ",jqfmt=%s", fmtname);
1990         }
1991
1992         if (sbi->s_qf_names[USRQUOTA])
1993                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1994
1995         if (sbi->s_qf_names[GRPQUOTA])
1996                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1997 #endif
1998 }
1999
2000 static const char *token2str(int token)
2001 {
2002         const struct match_token *t;
2003
2004         for (t = tokens; t->token != Opt_err; t++)
2005                 if (t->token == token && !strchr(t->pattern, '='))
2006                         break;
2007         return t->pattern;
2008 }
2009
2010 /*
2011  * Show an option if
2012  *  - it's set to a non-default value OR
2013  *  - if the per-sb default is different from the global default
2014  */
2015 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2016                               int nodefs)
2017 {
2018         struct ext4_sb_info *sbi = EXT4_SB(sb);
2019         struct ext4_super_block *es = sbi->s_es;
2020         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2021         const struct mount_opts *m;
2022         char sep = nodefs ? '\n' : ',';
2023
2024 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2025 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2026
2027         if (sbi->s_sb_block != 1)
2028                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2029
2030         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2031                 int want_set = m->flags & MOPT_SET;
2032                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2033                     (m->flags & MOPT_CLEAR_ERR))
2034                         continue;
2035                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2036                         continue; /* skip if same as the default */
2037                 if ((want_set &&
2038                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2039                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2040                         continue; /* select Opt_noFoo vs Opt_Foo */
2041                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2042         }
2043
2044         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2045             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2046                 SEQ_OPTS_PRINT("resuid=%u",
2047                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2048         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2049             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2050                 SEQ_OPTS_PRINT("resgid=%u",
2051                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2052         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2053         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2054                 SEQ_OPTS_PUTS("errors=remount-ro");
2055         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2056                 SEQ_OPTS_PUTS("errors=continue");
2057         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2058                 SEQ_OPTS_PUTS("errors=panic");
2059         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2060                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2061         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2062                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2063         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2064                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2065         if (sb->s_flags & MS_I_VERSION)
2066                 SEQ_OPTS_PUTS("i_version");
2067         if (nodefs || sbi->s_stripe)
2068                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2069         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2070                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2071                         SEQ_OPTS_PUTS("data=journal");
2072                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2073                         SEQ_OPTS_PUTS("data=ordered");
2074                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2075                         SEQ_OPTS_PUTS("data=writeback");
2076         }
2077         if (nodefs ||
2078             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2079                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2080                                sbi->s_inode_readahead_blks);
2081
2082         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2083                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2084                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2085         if (nodefs || sbi->s_max_dir_size_kb)
2086                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2087         if (test_opt(sb, DATA_ERR_ABORT))
2088                 SEQ_OPTS_PUTS("data_err=abort");
2089
2090         ext4_show_quota_options(seq, sb);
2091         return 0;
2092 }
2093
2094 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2095 {
2096         return _ext4_show_options(seq, root->d_sb, 0);
2097 }
2098
2099 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2100 {
2101         struct super_block *sb = seq->private;
2102         int rc;
2103
2104         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2105         rc = _ext4_show_options(seq, sb, 1);
2106         seq_puts(seq, "\n");
2107         return rc;
2108 }
2109
2110 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2111                             int read_only)
2112 {
2113         struct ext4_sb_info *sbi = EXT4_SB(sb);
2114         int res = 0;
2115
2116         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2117                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2118                          "forcing read-only mode");
2119                 res = MS_RDONLY;
2120         }
2121         if (read_only)
2122                 goto done;
2123         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2124                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2125                          "running e2fsck is recommended");
2126         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2127                 ext4_msg(sb, KERN_WARNING,
2128                          "warning: mounting fs with errors, "
2129                          "running e2fsck is recommended");
2130         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2131                  le16_to_cpu(es->s_mnt_count) >=
2132                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2133                 ext4_msg(sb, KERN_WARNING,
2134                          "warning: maximal mount count reached, "
2135                          "running e2fsck is recommended");
2136         else if (le32_to_cpu(es->s_checkinterval) &&
2137                 (le32_to_cpu(es->s_lastcheck) +
2138                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2139                 ext4_msg(sb, KERN_WARNING,
2140                          "warning: checktime reached, "
2141                          "running e2fsck is recommended");
2142         if (!sbi->s_journal)
2143                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2144         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2145                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2146         le16_add_cpu(&es->s_mnt_count, 1);
2147         es->s_mtime = cpu_to_le32(get_seconds());
2148         ext4_update_dynamic_rev(sb);
2149         if (sbi->s_journal)
2150                 ext4_set_feature_journal_needs_recovery(sb);
2151
2152         ext4_commit_super(sb, 1);
2153 done:
2154         if (test_opt(sb, DEBUG))
2155                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2156                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2157                         sb->s_blocksize,
2158                         sbi->s_groups_count,
2159                         EXT4_BLOCKS_PER_GROUP(sb),
2160                         EXT4_INODES_PER_GROUP(sb),
2161                         sbi->s_mount_opt, sbi->s_mount_opt2);
2162
2163         cleancache_init_fs(sb);
2164         return res;
2165 }
2166
2167 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2168 {
2169         struct ext4_sb_info *sbi = EXT4_SB(sb);
2170         struct flex_groups *new_groups;
2171         int size;
2172
2173         if (!sbi->s_log_groups_per_flex)
2174                 return 0;
2175
2176         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2177         if (size <= sbi->s_flex_groups_allocated)
2178                 return 0;
2179
2180         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2181         new_groups = kvzalloc(size, GFP_KERNEL);
2182         if (!new_groups) {
2183                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2184                          size / (int) sizeof(struct flex_groups));
2185                 return -ENOMEM;
2186         }
2187
2188         if (sbi->s_flex_groups) {
2189                 memcpy(new_groups, sbi->s_flex_groups,
2190                        (sbi->s_flex_groups_allocated *
2191                         sizeof(struct flex_groups)));
2192                 kvfree(sbi->s_flex_groups);
2193         }
2194         sbi->s_flex_groups = new_groups;
2195         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2196         return 0;
2197 }
2198
2199 static int ext4_fill_flex_info(struct super_block *sb)
2200 {
2201         struct ext4_sb_info *sbi = EXT4_SB(sb);
2202         struct ext4_group_desc *gdp = NULL;
2203         ext4_group_t flex_group;
2204         int i, err;
2205
2206         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2207         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2208                 sbi->s_log_groups_per_flex = 0;
2209                 return 1;
2210         }
2211
2212         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2213         if (err)
2214                 goto failed;
2215
2216         for (i = 0; i < sbi->s_groups_count; i++) {
2217                 gdp = ext4_get_group_desc(sb, i, NULL);
2218
2219                 flex_group = ext4_flex_group(sbi, i);
2220                 atomic_add(ext4_free_inodes_count(sb, gdp),
2221                            &sbi->s_flex_groups[flex_group].free_inodes);
2222                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2223                              &sbi->s_flex_groups[flex_group].free_clusters);
2224                 atomic_add(ext4_used_dirs_count(sb, gdp),
2225                            &sbi->s_flex_groups[flex_group].used_dirs);
2226         }
2227
2228         return 1;
2229 failed:
2230         return 0;
2231 }
2232
2233 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2234                                    struct ext4_group_desc *gdp)
2235 {
2236         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2237         __u16 crc = 0;
2238         __le32 le_group = cpu_to_le32(block_group);
2239         struct ext4_sb_info *sbi = EXT4_SB(sb);
2240
2241         if (ext4_has_metadata_csum(sbi->s_sb)) {
2242                 /* Use new metadata_csum algorithm */
2243                 __u32 csum32;
2244                 __u16 dummy_csum = 0;
2245
2246                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2247                                      sizeof(le_group));
2248                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2249                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2250                                      sizeof(dummy_csum));
2251                 offset += sizeof(dummy_csum);
2252                 if (offset < sbi->s_desc_size)
2253                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2254                                              sbi->s_desc_size - offset);
2255
2256                 crc = csum32 & 0xFFFF;
2257                 goto out;
2258         }
2259
2260         /* old crc16 code */
2261         if (!ext4_has_feature_gdt_csum(sb))
2262                 return 0;
2263
2264         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2265         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2266         crc = crc16(crc, (__u8 *)gdp, offset);
2267         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2268         /* for checksum of struct ext4_group_desc do the rest...*/
2269         if (ext4_has_feature_64bit(sb) &&
2270             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2271                 crc = crc16(crc, (__u8 *)gdp + offset,
2272                             le16_to_cpu(sbi->s_es->s_desc_size) -
2273                                 offset);
2274
2275 out:
2276         return cpu_to_le16(crc);
2277 }
2278
2279 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2280                                 struct ext4_group_desc *gdp)
2281 {
2282         if (ext4_has_group_desc_csum(sb) &&
2283             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2284                 return 0;
2285
2286         return 1;
2287 }
2288
2289 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2290                               struct ext4_group_desc *gdp)
2291 {
2292         if (!ext4_has_group_desc_csum(sb))
2293                 return;
2294         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2295 }
2296
2297 /* Called at mount-time, super-block is locked */
2298 static int ext4_check_descriptors(struct super_block *sb,
2299                                   ext4_fsblk_t sb_block,
2300                                   ext4_group_t *first_not_zeroed)
2301 {
2302         struct ext4_sb_info *sbi = EXT4_SB(sb);
2303         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2304         ext4_fsblk_t last_block;
2305         ext4_fsblk_t block_bitmap;
2306         ext4_fsblk_t inode_bitmap;
2307         ext4_fsblk_t inode_table;
2308         int flexbg_flag = 0;
2309         ext4_group_t i, grp = sbi->s_groups_count;
2310
2311         if (ext4_has_feature_flex_bg(sb))
2312                 flexbg_flag = 1;
2313
2314         ext4_debug("Checking group descriptors");
2315
2316         for (i = 0; i < sbi->s_groups_count; i++) {
2317                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2318
2319                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2320                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2321                 else
2322                         last_block = first_block +
2323                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2324
2325                 if ((grp == sbi->s_groups_count) &&
2326                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2327                         grp = i;
2328
2329                 block_bitmap = ext4_block_bitmap(sb, gdp);
2330                 if (block_bitmap == sb_block) {
2331                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2332                                  "Block bitmap for group %u overlaps "
2333                                  "superblock", i);
2334                 }
2335                 if (block_bitmap < first_block || block_bitmap > last_block) {
2336                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2337                                "Block bitmap for group %u not in group "
2338                                "(block %llu)!", i, block_bitmap);
2339                         return 0;
2340                 }
2341                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2342                 if (inode_bitmap == sb_block) {
2343                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2344                                  "Inode bitmap for group %u overlaps "
2345                                  "superblock", i);
2346                 }
2347                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2348                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2349                                "Inode bitmap for group %u not in group "
2350                                "(block %llu)!", i, inode_bitmap);
2351                         return 0;
2352                 }
2353                 inode_table = ext4_inode_table(sb, gdp);
2354                 if (inode_table == sb_block) {
2355                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2356                                  "Inode table for group %u overlaps "
2357                                  "superblock", i);
2358                 }
2359                 if (inode_table < first_block ||
2360                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2361                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2362                                "Inode table for group %u not in group "
2363                                "(block %llu)!", i, inode_table);
2364                         return 0;
2365                 }
2366                 ext4_lock_group(sb, i);
2367                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2368                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2369                                  "Checksum for group %u failed (%u!=%u)",
2370                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2371                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2372                         if (!(sb->s_flags & MS_RDONLY)) {
2373                                 ext4_unlock_group(sb, i);
2374                                 return 0;
2375                         }
2376                 }
2377                 ext4_unlock_group(sb, i);
2378                 if (!flexbg_flag)
2379                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2380         }
2381         if (NULL != first_not_zeroed)
2382                 *first_not_zeroed = grp;
2383         return 1;
2384 }
2385
2386 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2387  * the superblock) which were deleted from all directories, but held open by
2388  * a process at the time of a crash.  We walk the list and try to delete these
2389  * inodes at recovery time (only with a read-write filesystem).
2390  *
2391  * In order to keep the orphan inode chain consistent during traversal (in
2392  * case of crash during recovery), we link each inode into the superblock
2393  * orphan list_head and handle it the same way as an inode deletion during
2394  * normal operation (which journals the operations for us).
2395  *
2396  * We only do an iget() and an iput() on each inode, which is very safe if we
2397  * accidentally point at an in-use or already deleted inode.  The worst that
2398  * can happen in this case is that we get a "bit already cleared" message from
2399  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2400  * e2fsck was run on this filesystem, and it must have already done the orphan
2401  * inode cleanup for us, so we can safely abort without any further action.
2402  */
2403 static void ext4_orphan_cleanup(struct super_block *sb,
2404                                 struct ext4_super_block *es)
2405 {
2406         unsigned int s_flags = sb->s_flags;
2407         int ret, nr_orphans = 0, nr_truncates = 0;
2408 #ifdef CONFIG_QUOTA
2409         int i;
2410 #endif
2411         if (!es->s_last_orphan) {
2412                 jbd_debug(4, "no orphan inodes to clean up\n");
2413                 return;
2414         }
2415
2416         if (bdev_read_only(sb->s_bdev)) {
2417                 ext4_msg(sb, KERN_ERR, "write access "
2418                         "unavailable, skipping orphan cleanup");
2419                 return;
2420         }
2421
2422         /* Check if feature set would not allow a r/w mount */
2423         if (!ext4_feature_set_ok(sb, 0)) {
2424                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2425                          "unknown ROCOMPAT features");
2426                 return;
2427         }
2428
2429         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2430                 /* don't clear list on RO mount w/ errors */
2431                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2432                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2433                                   "clearing orphan list.\n");
2434                         es->s_last_orphan = 0;
2435                 }
2436                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2437                 return;
2438         }
2439
2440         if (s_flags & MS_RDONLY) {
2441                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2442                 sb->s_flags &= ~MS_RDONLY;
2443         }
2444 #ifdef CONFIG_QUOTA
2445         /* Needed for iput() to work correctly and not trash data */
2446         sb->s_flags |= MS_ACTIVE;
2447         /* Turn on quotas so that they are updated correctly */
2448         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2449                 if (EXT4_SB(sb)->s_qf_names[i]) {
2450                         int ret = ext4_quota_on_mount(sb, i);
2451                         if (ret < 0)
2452                                 ext4_msg(sb, KERN_ERR,
2453                                         "Cannot turn on journaled "
2454                                         "quota: error %d", ret);
2455                 }
2456         }
2457 #endif
2458
2459         while (es->s_last_orphan) {
2460                 struct inode *inode;
2461
2462                 /*
2463                  * We may have encountered an error during cleanup; if
2464                  * so, skip the rest.
2465                  */
2466                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2467                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2468                         es->s_last_orphan = 0;
2469                         break;
2470                 }
2471
2472                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2473                 if (IS_ERR(inode)) {
2474                         es->s_last_orphan = 0;
2475                         break;
2476                 }
2477
2478                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2479                 dquot_initialize(inode);
2480                 if (inode->i_nlink) {
2481                         if (test_opt(sb, DEBUG))
2482                                 ext4_msg(sb, KERN_DEBUG,
2483                                         "%s: truncating inode %lu to %lld bytes",
2484                                         __func__, inode->i_ino, inode->i_size);
2485                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2486                                   inode->i_ino, inode->i_size);
2487                         inode_lock(inode);
2488                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2489                         ret = ext4_truncate(inode);
2490                         if (ret)
2491                                 ext4_std_error(inode->i_sb, ret);
2492                         inode_unlock(inode);
2493                         nr_truncates++;
2494                 } else {
2495                         if (test_opt(sb, DEBUG))
2496                                 ext4_msg(sb, KERN_DEBUG,
2497                                         "%s: deleting unreferenced inode %lu",
2498                                         __func__, inode->i_ino);
2499                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2500                                   inode->i_ino);
2501                         nr_orphans++;
2502                 }
2503                 iput(inode);  /* The delete magic happens here! */
2504         }
2505
2506 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2507
2508         if (nr_orphans)
2509                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2510                        PLURAL(nr_orphans));
2511         if (nr_truncates)
2512                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2513                        PLURAL(nr_truncates));
2514 #ifdef CONFIG_QUOTA
2515         /* Turn quotas off */
2516         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2517                 if (sb_dqopt(sb)->files[i])
2518                         dquot_quota_off(sb, i);
2519         }
2520 #endif
2521         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2522 }
2523
2524 /*
2525  * Maximal extent format file size.
2526  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2527  * extent format containers, within a sector_t, and within i_blocks
2528  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2529  * so that won't be a limiting factor.
2530  *
2531  * However there is other limiting factor. We do store extents in the form
2532  * of starting block and length, hence the resulting length of the extent
2533  * covering maximum file size must fit into on-disk format containers as
2534  * well. Given that length is always by 1 unit bigger than max unit (because
2535  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2536  *
2537  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2538  */
2539 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2540 {
2541         loff_t res;
2542         loff_t upper_limit = MAX_LFS_FILESIZE;
2543
2544         /* small i_blocks in vfs inode? */
2545         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2546                 /*
2547                  * CONFIG_LBDAF is not enabled implies the inode
2548                  * i_block represent total blocks in 512 bytes
2549                  * 32 == size of vfs inode i_blocks * 8
2550                  */
2551                 upper_limit = (1LL << 32) - 1;
2552
2553                 /* total blocks in file system block size */
2554                 upper_limit >>= (blkbits - 9);
2555                 upper_limit <<= blkbits;
2556         }
2557
2558         /*
2559          * 32-bit extent-start container, ee_block. We lower the maxbytes
2560          * by one fs block, so ee_len can cover the extent of maximum file
2561          * size
2562          */
2563         res = (1LL << 32) - 1;
2564         res <<= blkbits;
2565
2566         /* Sanity check against vm- & vfs- imposed limits */
2567         if (res > upper_limit)
2568                 res = upper_limit;
2569
2570         return res;
2571 }
2572
2573 /*
2574  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2575  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2576  * We need to be 1 filesystem block less than the 2^48 sector limit.
2577  */
2578 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2579 {
2580         loff_t res = EXT4_NDIR_BLOCKS;
2581         int meta_blocks;
2582         loff_t upper_limit;
2583         /* This is calculated to be the largest file size for a dense, block
2584          * mapped file such that the file's total number of 512-byte sectors,
2585          * including data and all indirect blocks, does not exceed (2^48 - 1).
2586          *
2587          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2588          * number of 512-byte sectors of the file.
2589          */
2590
2591         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2592                 /*
2593                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2594                  * the inode i_block field represents total file blocks in
2595                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2596                  */
2597                 upper_limit = (1LL << 32) - 1;
2598
2599                 /* total blocks in file system block size */
2600                 upper_limit >>= (bits - 9);
2601
2602         } else {
2603                 /*
2604                  * We use 48 bit ext4_inode i_blocks
2605                  * With EXT4_HUGE_FILE_FL set the i_blocks
2606                  * represent total number of blocks in
2607                  * file system block size
2608                  */
2609                 upper_limit = (1LL << 48) - 1;
2610
2611         }
2612
2613         /* indirect blocks */
2614         meta_blocks = 1;
2615         /* double indirect blocks */
2616         meta_blocks += 1 + (1LL << (bits-2));
2617         /* tripple indirect blocks */
2618         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2619
2620         upper_limit -= meta_blocks;
2621         upper_limit <<= bits;
2622
2623         res += 1LL << (bits-2);
2624         res += 1LL << (2*(bits-2));
2625         res += 1LL << (3*(bits-2));
2626         res <<= bits;
2627         if (res > upper_limit)
2628                 res = upper_limit;
2629
2630         if (res > MAX_LFS_FILESIZE)
2631                 res = MAX_LFS_FILESIZE;
2632
2633         return res;
2634 }
2635
2636 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2637                                    ext4_fsblk_t logical_sb_block, int nr)
2638 {
2639         struct ext4_sb_info *sbi = EXT4_SB(sb);
2640         ext4_group_t bg, first_meta_bg;
2641         int has_super = 0;
2642
2643         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2644
2645         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2646                 return logical_sb_block + nr + 1;
2647         bg = sbi->s_desc_per_block * nr;
2648         if (ext4_bg_has_super(sb, bg))
2649                 has_super = 1;
2650
2651         /*
2652          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2653          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2654          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2655          * compensate.
2656          */
2657         if (sb->s_blocksize == 1024 && nr == 0 &&
2658             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2659                 has_super++;
2660
2661         return (has_super + ext4_group_first_block_no(sb, bg));
2662 }
2663
2664 /**
2665  * ext4_get_stripe_size: Get the stripe size.
2666  * @sbi: In memory super block info
2667  *
2668  * If we have specified it via mount option, then
2669  * use the mount option value. If the value specified at mount time is
2670  * greater than the blocks per group use the super block value.
2671  * If the super block value is greater than blocks per group return 0.
2672  * Allocator needs it be less than blocks per group.
2673  *
2674  */
2675 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2676 {
2677         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2678         unsigned long stripe_width =
2679                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2680         int ret;
2681
2682         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2683                 ret = sbi->s_stripe;
2684         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2685                 ret = stripe_width;
2686         else if (stride && stride <= sbi->s_blocks_per_group)
2687                 ret = stride;
2688         else
2689                 ret = 0;
2690
2691         /*
2692          * If the stripe width is 1, this makes no sense and
2693          * we set it to 0 to turn off stripe handling code.
2694          */
2695         if (ret <= 1)
2696                 ret = 0;
2697
2698         return ret;
2699 }
2700
2701 /*
2702  * Check whether this filesystem can be mounted based on
2703  * the features present and the RDONLY/RDWR mount requested.
2704  * Returns 1 if this filesystem can be mounted as requested,
2705  * 0 if it cannot be.
2706  */
2707 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2708 {
2709         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2710                 ext4_msg(sb, KERN_ERR,
2711                         "Couldn't mount because of "
2712                         "unsupported optional features (%x)",
2713                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2714                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2715                 return 0;
2716         }
2717
2718         if (readonly)
2719                 return 1;
2720
2721         if (ext4_has_feature_readonly(sb)) {
2722                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2723                 sb->s_flags |= MS_RDONLY;
2724                 return 1;
2725         }
2726
2727         /* Check that feature set is OK for a read-write mount */
2728         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2729                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2730                          "unsupported optional features (%x)",
2731                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2732                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2733                 return 0;
2734         }
2735         /*
2736          * Large file size enabled file system can only be mounted
2737          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2738          */
2739         if (ext4_has_feature_huge_file(sb)) {
2740                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2741                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2742                                  "cannot be mounted RDWR without "
2743                                  "CONFIG_LBDAF");
2744                         return 0;
2745                 }
2746         }
2747         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2748                 ext4_msg(sb, KERN_ERR,
2749                          "Can't support bigalloc feature without "
2750                          "extents feature\n");
2751                 return 0;
2752         }
2753
2754 #ifndef CONFIG_QUOTA
2755         if (ext4_has_feature_quota(sb) && !readonly) {
2756                 ext4_msg(sb, KERN_ERR,
2757                          "Filesystem with quota feature cannot be mounted RDWR "
2758                          "without CONFIG_QUOTA");
2759                 return 0;
2760         }
2761         if (ext4_has_feature_project(sb) && !readonly) {
2762                 ext4_msg(sb, KERN_ERR,
2763                          "Filesystem with project quota feature cannot be mounted RDWR "
2764                          "without CONFIG_QUOTA");
2765                 return 0;
2766         }
2767 #endif  /* CONFIG_QUOTA */
2768         return 1;
2769 }
2770
2771 /*
2772  * This function is called once a day if we have errors logged
2773  * on the file system
2774  */
2775 static void print_daily_error_info(unsigned long arg)
2776 {
2777         struct super_block *sb = (struct super_block *) arg;
2778         struct ext4_sb_info *sbi;
2779         struct ext4_super_block *es;
2780
2781         sbi = EXT4_SB(sb);
2782         es = sbi->s_es;
2783
2784         if (es->s_error_count)
2785                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2786                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2787                          le32_to_cpu(es->s_error_count));
2788         if (es->s_first_error_time) {
2789                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2790                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2791                        (int) sizeof(es->s_first_error_func),
2792                        es->s_first_error_func,
2793                        le32_to_cpu(es->s_first_error_line));
2794                 if (es->s_first_error_ino)
2795                         printk(KERN_CONT ": inode %u",
2796                                le32_to_cpu(es->s_first_error_ino));
2797                 if (es->s_first_error_block)
2798                         printk(KERN_CONT ": block %llu", (unsigned long long)
2799                                le64_to_cpu(es->s_first_error_block));
2800                 printk(KERN_CONT "\n");
2801         }
2802         if (es->s_last_error_time) {
2803                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2804                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2805                        (int) sizeof(es->s_last_error_func),
2806                        es->s_last_error_func,
2807                        le32_to_cpu(es->s_last_error_line));
2808                 if (es->s_last_error_ino)
2809                         printk(KERN_CONT ": inode %u",
2810                                le32_to_cpu(es->s_last_error_ino));
2811                 if (es->s_last_error_block)
2812                         printk(KERN_CONT ": block %llu", (unsigned long long)
2813                                le64_to_cpu(es->s_last_error_block));
2814                 printk(KERN_CONT "\n");
2815         }
2816         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2817 }
2818
2819 /* Find next suitable group and run ext4_init_inode_table */
2820 static int ext4_run_li_request(struct ext4_li_request *elr)
2821 {
2822         struct ext4_group_desc *gdp = NULL;
2823         ext4_group_t group, ngroups;
2824         struct super_block *sb;
2825         unsigned long timeout = 0;
2826         int ret = 0;
2827
2828         sb = elr->lr_super;
2829         ngroups = EXT4_SB(sb)->s_groups_count;
2830
2831         for (group = elr->lr_next_group; group < ngroups; group++) {
2832                 gdp = ext4_get_group_desc(sb, group, NULL);
2833                 if (!gdp) {
2834                         ret = 1;
2835                         break;
2836                 }
2837
2838                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2839                         break;
2840         }
2841
2842         if (group >= ngroups)
2843                 ret = 1;
2844
2845         if (!ret) {
2846                 timeout = jiffies;
2847                 ret = ext4_init_inode_table(sb, group,
2848                                             elr->lr_timeout ? 0 : 1);
2849                 if (elr->lr_timeout == 0) {
2850                         timeout = (jiffies - timeout) *
2851                                   elr->lr_sbi->s_li_wait_mult;
2852                         elr->lr_timeout = timeout;
2853                 }
2854                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2855                 elr->lr_next_group = group + 1;
2856         }
2857         return ret;
2858 }
2859
2860 /*
2861  * Remove lr_request from the list_request and free the
2862  * request structure. Should be called with li_list_mtx held
2863  */
2864 static void ext4_remove_li_request(struct ext4_li_request *elr)
2865 {
2866         struct ext4_sb_info *sbi;
2867
2868         if (!elr)
2869                 return;
2870
2871         sbi = elr->lr_sbi;
2872
2873         list_del(&elr->lr_request);
2874         sbi->s_li_request = NULL;
2875         kfree(elr);
2876 }
2877
2878 static void ext4_unregister_li_request(struct super_block *sb)
2879 {
2880         mutex_lock(&ext4_li_mtx);
2881         if (!ext4_li_info) {
2882                 mutex_unlock(&ext4_li_mtx);
2883                 return;
2884         }
2885
2886         mutex_lock(&ext4_li_info->li_list_mtx);
2887         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2888         mutex_unlock(&ext4_li_info->li_list_mtx);
2889         mutex_unlock(&ext4_li_mtx);
2890 }
2891
2892 static struct task_struct *ext4_lazyinit_task;
2893
2894 /*
2895  * This is the function where ext4lazyinit thread lives. It walks
2896  * through the request list searching for next scheduled filesystem.
2897  * When such a fs is found, run the lazy initialization request
2898  * (ext4_rn_li_request) and keep track of the time spend in this
2899  * function. Based on that time we compute next schedule time of
2900  * the request. When walking through the list is complete, compute
2901  * next waking time and put itself into sleep.
2902  */
2903 static int ext4_lazyinit_thread(void *arg)
2904 {
2905         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2906         struct list_head *pos, *n;
2907         struct ext4_li_request *elr;
2908         unsigned long next_wakeup, cur;
2909
2910         BUG_ON(NULL == eli);
2911
2912 cont_thread:
2913         while (true) {
2914                 next_wakeup = MAX_JIFFY_OFFSET;
2915
2916                 mutex_lock(&eli->li_list_mtx);
2917                 if (list_empty(&eli->li_request_list)) {
2918                         mutex_unlock(&eli->li_list_mtx);
2919                         goto exit_thread;
2920                 }
2921                 list_for_each_safe(pos, n, &eli->li_request_list) {
2922                         int err = 0;
2923                         int progress = 0;
2924                         elr = list_entry(pos, struct ext4_li_request,
2925                                          lr_request);
2926
2927                         if (time_before(jiffies, elr->lr_next_sched)) {
2928                                 if (time_before(elr->lr_next_sched, next_wakeup))
2929                                         next_wakeup = elr->lr_next_sched;
2930                                 continue;
2931                         }
2932                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2933                                 if (sb_start_write_trylock(elr->lr_super)) {
2934                                         progress = 1;
2935                                         /*
2936                                          * We hold sb->s_umount, sb can not
2937                                          * be removed from the list, it is
2938                                          * now safe to drop li_list_mtx
2939                                          */
2940                                         mutex_unlock(&eli->li_list_mtx);
2941                                         err = ext4_run_li_request(elr);
2942                                         sb_end_write(elr->lr_super);
2943                                         mutex_lock(&eli->li_list_mtx);
2944                                         n = pos->next;
2945                                 }
2946                                 up_read((&elr->lr_super->s_umount));
2947                         }
2948                         /* error, remove the lazy_init job */
2949                         if (err) {
2950                                 ext4_remove_li_request(elr);
2951                                 continue;
2952                         }
2953                         if (!progress) {
2954                                 elr->lr_next_sched = jiffies +
2955                                         (prandom_u32()
2956                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2957                         }
2958                         if (time_before(elr->lr_next_sched, next_wakeup))
2959                                 next_wakeup = elr->lr_next_sched;
2960                 }
2961                 mutex_unlock(&eli->li_list_mtx);
2962
2963                 try_to_freeze();
2964
2965                 cur = jiffies;
2966                 if ((time_after_eq(cur, next_wakeup)) ||
2967                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2968                         cond_resched();
2969                         continue;
2970                 }
2971
2972                 schedule_timeout_interruptible(next_wakeup - cur);
2973
2974                 if (kthread_should_stop()) {
2975                         ext4_clear_request_list();
2976                         goto exit_thread;
2977                 }
2978         }
2979
2980 exit_thread:
2981         /*
2982          * It looks like the request list is empty, but we need
2983          * to check it under the li_list_mtx lock, to prevent any
2984          * additions into it, and of course we should lock ext4_li_mtx
2985          * to atomically free the list and ext4_li_info, because at
2986          * this point another ext4 filesystem could be registering
2987          * new one.
2988          */
2989         mutex_lock(&ext4_li_mtx);
2990         mutex_lock(&eli->li_list_mtx);
2991         if (!list_empty(&eli->li_request_list)) {
2992                 mutex_unlock(&eli->li_list_mtx);
2993                 mutex_unlock(&ext4_li_mtx);
2994                 goto cont_thread;
2995         }
2996         mutex_unlock(&eli->li_list_mtx);
2997         kfree(ext4_li_info);
2998         ext4_li_info = NULL;
2999         mutex_unlock(&ext4_li_mtx);
3000
3001         return 0;
3002 }
3003
3004 static void ext4_clear_request_list(void)
3005 {
3006         struct list_head *pos, *n;
3007         struct ext4_li_request *elr;
3008
3009         mutex_lock(&ext4_li_info->li_list_mtx);
3010         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3011                 elr = list_entry(pos, struct ext4_li_request,
3012                                  lr_request);
3013                 ext4_remove_li_request(elr);
3014         }
3015         mutex_unlock(&ext4_li_info->li_list_mtx);
3016 }
3017
3018 static int ext4_run_lazyinit_thread(void)
3019 {
3020         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3021                                          ext4_li_info, "ext4lazyinit");
3022         if (IS_ERR(ext4_lazyinit_task)) {
3023                 int err = PTR_ERR(ext4_lazyinit_task);
3024                 ext4_clear_request_list();
3025                 kfree(ext4_li_info);
3026                 ext4_li_info = NULL;
3027                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3028                                  "initialization thread\n",
3029                                  err);
3030                 return err;
3031         }
3032         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3033         return 0;
3034 }
3035
3036 /*
3037  * Check whether it make sense to run itable init. thread or not.
3038  * If there is at least one uninitialized inode table, return
3039  * corresponding group number, else the loop goes through all
3040  * groups and return total number of groups.
3041  */
3042 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3043 {
3044         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3045         struct ext4_group_desc *gdp = NULL;
3046
3047         for (group = 0; group < ngroups; group++) {
3048                 gdp = ext4_get_group_desc(sb, group, NULL);
3049                 if (!gdp)
3050                         continue;
3051
3052                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3053                         break;
3054         }
3055
3056         return group;
3057 }
3058
3059 static int ext4_li_info_new(void)
3060 {
3061         struct ext4_lazy_init *eli = NULL;
3062
3063         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3064         if (!eli)
3065                 return -ENOMEM;
3066
3067         INIT_LIST_HEAD(&eli->li_request_list);
3068         mutex_init(&eli->li_list_mtx);
3069
3070         eli->li_state |= EXT4_LAZYINIT_QUIT;
3071
3072         ext4_li_info = eli;
3073
3074         return 0;
3075 }
3076
3077 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3078                                             ext4_group_t start)
3079 {
3080         struct ext4_sb_info *sbi = EXT4_SB(sb);
3081         struct ext4_li_request *elr;
3082
3083         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3084         if (!elr)
3085                 return NULL;
3086
3087         elr->lr_super = sb;
3088         elr->lr_sbi = sbi;
3089         elr->lr_next_group = start;
3090
3091         /*
3092          * Randomize first schedule time of the request to
3093          * spread the inode table initialization requests
3094          * better.
3095          */
3096         elr->lr_next_sched = jiffies + (prandom_u32() %
3097                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3098         return elr;
3099 }
3100
3101 int ext4_register_li_request(struct super_block *sb,
3102                              ext4_group_t first_not_zeroed)
3103 {
3104         struct ext4_sb_info *sbi = EXT4_SB(sb);
3105         struct ext4_li_request *elr = NULL;
3106         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3107         int ret = 0;
3108
3109         mutex_lock(&ext4_li_mtx);
3110         if (sbi->s_li_request != NULL) {
3111                 /*
3112                  * Reset timeout so it can be computed again, because
3113                  * s_li_wait_mult might have changed.
3114                  */
3115                 sbi->s_li_request->lr_timeout = 0;
3116                 goto out;
3117         }
3118
3119         if (first_not_zeroed == ngroups ||
3120             (sb->s_flags & MS_RDONLY) ||
3121             !test_opt(sb, INIT_INODE_TABLE))
3122                 goto out;
3123
3124         elr = ext4_li_request_new(sb, first_not_zeroed);
3125         if (!elr) {
3126                 ret = -ENOMEM;
3127                 goto out;
3128         }
3129
3130         if (NULL == ext4_li_info) {
3131                 ret = ext4_li_info_new();
3132                 if (ret)
3133                         goto out;
3134         }
3135
3136         mutex_lock(&ext4_li_info->li_list_mtx);
3137         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3138         mutex_unlock(&ext4_li_info->li_list_mtx);
3139
3140         sbi->s_li_request = elr;
3141         /*
3142          * set elr to NULL here since it has been inserted to
3143          * the request_list and the removal and free of it is
3144          * handled by ext4_clear_request_list from now on.
3145          */
3146         elr = NULL;
3147
3148         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3149                 ret = ext4_run_lazyinit_thread();
3150                 if (ret)
3151                         goto out;
3152         }
3153 out:
3154         mutex_unlock(&ext4_li_mtx);
3155         if (ret)
3156                 kfree(elr);
3157         return ret;
3158 }
3159
3160 /*
3161  * We do not need to lock anything since this is called on
3162  * module unload.
3163  */
3164 static void ext4_destroy_lazyinit_thread(void)
3165 {
3166         /*
3167          * If thread exited earlier
3168          * there's nothing to be done.
3169          */
3170         if (!ext4_li_info || !ext4_lazyinit_task)
3171                 return;
3172
3173         kthread_stop(ext4_lazyinit_task);
3174 }
3175
3176 static int set_journal_csum_feature_set(struct super_block *sb)
3177 {
3178         int ret = 1;
3179         int compat, incompat;
3180         struct ext4_sb_info *sbi = EXT4_SB(sb);
3181
3182         if (ext4_has_metadata_csum(sb)) {
3183                 /* journal checksum v3 */
3184                 compat = 0;
3185                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3186         } else {
3187                 /* journal checksum v1 */
3188                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3189                 incompat = 0;
3190         }
3191
3192         jbd2_journal_clear_features(sbi->s_journal,
3193                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3194                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3195                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3196         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3197                 ret = jbd2_journal_set_features(sbi->s_journal,
3198                                 compat, 0,
3199                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3200                                 incompat);
3201         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3202                 ret = jbd2_journal_set_features(sbi->s_journal,
3203                                 compat, 0,
3204                                 incompat);
3205                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3206                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3207         } else {
3208                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3209                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3210         }
3211
3212         return ret;
3213 }
3214
3215 /*
3216  * Note: calculating the overhead so we can be compatible with
3217  * historical BSD practice is quite difficult in the face of
3218  * clusters/bigalloc.  This is because multiple metadata blocks from
3219  * different block group can end up in the same allocation cluster.
3220  * Calculating the exact overhead in the face of clustered allocation
3221  * requires either O(all block bitmaps) in memory or O(number of block
3222  * groups**2) in time.  We will still calculate the superblock for
3223  * older file systems --- and if we come across with a bigalloc file
3224  * system with zero in s_overhead_clusters the estimate will be close to
3225  * correct especially for very large cluster sizes --- but for newer
3226  * file systems, it's better to calculate this figure once at mkfs
3227  * time, and store it in the superblock.  If the superblock value is
3228  * present (even for non-bigalloc file systems), we will use it.
3229  */
3230 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3231                           char *buf)
3232 {
3233         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3234         struct ext4_group_desc  *gdp;
3235         ext4_fsblk_t            first_block, last_block, b;
3236         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3237         int                     s, j, count = 0;
3238
3239         if (!ext4_has_feature_bigalloc(sb))
3240                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3241                         sbi->s_itb_per_group + 2);
3242
3243         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3244                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3245         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3246         for (i = 0; i < ngroups; i++) {
3247                 gdp = ext4_get_group_desc(sb, i, NULL);
3248                 b = ext4_block_bitmap(sb, gdp);
3249                 if (b >= first_block && b <= last_block) {
3250                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3251                         count++;
3252                 }
3253                 b = ext4_inode_bitmap(sb, gdp);
3254                 if (b >= first_block && b <= last_block) {
3255                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3256                         count++;
3257                 }
3258                 b = ext4_inode_table(sb, gdp);
3259                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3260                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3261                                 int c = EXT4_B2C(sbi, b - first_block);
3262                                 ext4_set_bit(c, buf);
3263                                 count++;
3264                         }
3265                 if (i != grp)
3266                         continue;
3267                 s = 0;
3268                 if (ext4_bg_has_super(sb, grp)) {
3269                         ext4_set_bit(s++, buf);
3270                         count++;
3271                 }
3272                 j = ext4_bg_num_gdb(sb, grp);
3273                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3274                         ext4_error(sb, "Invalid number of block group "
3275                                    "descriptor blocks: %d", j);
3276                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3277                 }
3278                 count += j;
3279                 for (; j > 0; j--)
3280                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3281         }
3282         if (!count)
3283                 return 0;
3284         return EXT4_CLUSTERS_PER_GROUP(sb) -
3285                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3286 }
3287
3288 /*
3289  * Compute the overhead and stash it in sbi->s_overhead
3290  */
3291 int ext4_calculate_overhead(struct super_block *sb)
3292 {
3293         struct ext4_sb_info *sbi = EXT4_SB(sb);
3294         struct ext4_super_block *es = sbi->s_es;
3295         struct inode *j_inode;
3296         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3297         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3298         ext4_fsblk_t overhead = 0;
3299         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3300
3301         if (!buf)
3302                 return -ENOMEM;
3303
3304         /*
3305          * Compute the overhead (FS structures).  This is constant
3306          * for a given filesystem unless the number of block groups
3307          * changes so we cache the previous value until it does.
3308          */
3309
3310         /*
3311          * All of the blocks before first_data_block are overhead
3312          */
3313         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3314
3315         /*
3316          * Add the overhead found in each block group
3317          */
3318         for (i = 0; i < ngroups; i++) {
3319                 int blks;
3320
3321                 blks = count_overhead(sb, i, buf);
3322                 overhead += blks;
3323                 if (blks)
3324                         memset(buf, 0, PAGE_SIZE);
3325                 cond_resched();
3326         }
3327
3328         /*
3329          * Add the internal journal blocks whether the journal has been
3330          * loaded or not
3331          */
3332         if (sbi->s_journal && !sbi->journal_bdev)
3333                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3334         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3335                 j_inode = ext4_get_journal_inode(sb, j_inum);
3336                 if (j_inode) {
3337                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3338                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3339                         iput(j_inode);
3340                 } else {
3341                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3342                 }
3343         }
3344         sbi->s_overhead = overhead;
3345         smp_wmb();
3346         free_page((unsigned long) buf);
3347         return 0;
3348 }
3349
3350 static void ext4_set_resv_clusters(struct super_block *sb)
3351 {
3352         ext4_fsblk_t resv_clusters;
3353         struct ext4_sb_info *sbi = EXT4_SB(sb);
3354
3355         /*
3356          * There's no need to reserve anything when we aren't using extents.
3357          * The space estimates are exact, there are no unwritten extents,
3358          * hole punching doesn't need new metadata... This is needed especially
3359          * to keep ext2/3 backward compatibility.
3360          */
3361         if (!ext4_has_feature_extents(sb))
3362                 return;
3363         /*
3364          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3365          * This should cover the situations where we can not afford to run
3366          * out of space like for example punch hole, or converting
3367          * unwritten extents in delalloc path. In most cases such
3368          * allocation would require 1, or 2 blocks, higher numbers are
3369          * very rare.
3370          */
3371         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3372                          sbi->s_cluster_bits);
3373
3374         do_div(resv_clusters, 50);
3375         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3376
3377         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3378 }
3379
3380 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3381 {
3382         char *orig_data = kstrdup(data, GFP_KERNEL);
3383         struct buffer_head *bh;
3384         struct ext4_super_block *es = NULL;
3385         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3386         ext4_fsblk_t block;
3387         ext4_fsblk_t sb_block = get_sb_block(&data);
3388         ext4_fsblk_t logical_sb_block;
3389         unsigned long offset = 0;
3390         unsigned long journal_devnum = 0;
3391         unsigned long def_mount_opts;
3392         struct inode *root;
3393         const char *descr;
3394         int ret = -ENOMEM;
3395         int blocksize, clustersize;
3396         unsigned int db_count;
3397         unsigned int i;
3398         int needs_recovery, has_huge_files, has_bigalloc;
3399         __u64 blocks_count;
3400         int err = 0;
3401         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3402         ext4_group_t first_not_zeroed;
3403
3404         if ((data && !orig_data) || !sbi)
3405                 goto out_free_base;
3406
3407         sbi->s_blockgroup_lock =
3408                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3409         if (!sbi->s_blockgroup_lock)
3410                 goto out_free_base;
3411
3412         sb->s_fs_info = sbi;
3413         sbi->s_sb = sb;
3414         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3415         sbi->s_sb_block = sb_block;
3416         if (sb->s_bdev->bd_part)
3417                 sbi->s_sectors_written_start =
3418                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3419
3420         /* Cleanup superblock name */
3421         strreplace(sb->s_id, '/', '!');
3422
3423         /* -EINVAL is default */
3424         ret = -EINVAL;
3425         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3426         if (!blocksize) {
3427                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3428                 goto out_fail;
3429         }
3430
3431         /*
3432          * The ext4 superblock will not be buffer aligned for other than 1kB
3433          * block sizes.  We need to calculate the offset from buffer start.
3434          */
3435         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3436                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3437                 offset = do_div(logical_sb_block, blocksize);
3438         } else {
3439                 logical_sb_block = sb_block;
3440         }
3441
3442         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3443                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3444                 goto out_fail;
3445         }
3446         /*
3447          * Note: s_es must be initialized as soon as possible because
3448          *       some ext4 macro-instructions depend on its value
3449          */
3450         es = (struct ext4_super_block *) (bh->b_data + offset);
3451         sbi->s_es = es;
3452         sb->s_magic = le16_to_cpu(es->s_magic);
3453         if (sb->s_magic != EXT4_SUPER_MAGIC)
3454                 goto cantfind_ext4;
3455         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3456
3457         /* Warn if metadata_csum and gdt_csum are both set. */
3458         if (ext4_has_feature_metadata_csum(sb) &&
3459             ext4_has_feature_gdt_csum(sb))
3460                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3461                              "redundant flags; please run fsck.");
3462
3463         /* Check for a known checksum algorithm */
3464         if (!ext4_verify_csum_type(sb, es)) {
3465                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3466                          "unknown checksum algorithm.");
3467                 silent = 1;
3468                 goto cantfind_ext4;
3469         }
3470
3471         /* Load the checksum driver */
3472         if (ext4_has_feature_metadata_csum(sb) ||
3473             ext4_has_feature_ea_inode(sb)) {
3474                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3475                 if (IS_ERR(sbi->s_chksum_driver)) {
3476                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3477                         ret = PTR_ERR(sbi->s_chksum_driver);
3478                         sbi->s_chksum_driver = NULL;
3479                         goto failed_mount;
3480                 }
3481         }
3482
3483         /* Check superblock checksum */
3484         if (!ext4_superblock_csum_verify(sb, es)) {
3485                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3486                          "invalid superblock checksum.  Run e2fsck?");
3487                 silent = 1;
3488                 ret = -EFSBADCRC;
3489                 goto cantfind_ext4;
3490         }
3491
3492         /* Precompute checksum seed for all metadata */
3493         if (ext4_has_feature_csum_seed(sb))
3494                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3495         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3496                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3497                                                sizeof(es->s_uuid));
3498
3499         /* Set defaults before we parse the mount options */
3500         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3501         set_opt(sb, INIT_INODE_TABLE);
3502         if (def_mount_opts & EXT4_DEFM_DEBUG)
3503                 set_opt(sb, DEBUG);
3504         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3505                 set_opt(sb, GRPID);
3506         if (def_mount_opts & EXT4_DEFM_UID16)
3507                 set_opt(sb, NO_UID32);
3508         /* xattr user namespace & acls are now defaulted on */
3509         set_opt(sb, XATTR_USER);
3510 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3511         set_opt(sb, POSIX_ACL);
3512 #endif
3513         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3514         if (ext4_has_metadata_csum(sb))
3515                 set_opt(sb, JOURNAL_CHECKSUM);
3516
3517         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3518                 set_opt(sb, JOURNAL_DATA);
3519         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3520                 set_opt(sb, ORDERED_DATA);
3521         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3522                 set_opt(sb, WRITEBACK_DATA);
3523
3524         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3525                 set_opt(sb, ERRORS_PANIC);
3526         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3527                 set_opt(sb, ERRORS_CONT);
3528         else
3529                 set_opt(sb, ERRORS_RO);
3530         /* block_validity enabled by default; disable with noblock_validity */
3531         set_opt(sb, BLOCK_VALIDITY);
3532         if (def_mount_opts & EXT4_DEFM_DISCARD)
3533                 set_opt(sb, DISCARD);
3534
3535         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3536         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3537         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3538         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3539         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3540
3541         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3542                 set_opt(sb, BARRIER);
3543
3544         /*
3545          * enable delayed allocation by default
3546          * Use -o nodelalloc to turn it off
3547          */
3548         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3549             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3550                 set_opt(sb, DELALLOC);
3551
3552         /*
3553          * set default s_li_wait_mult for lazyinit, for the case there is
3554          * no mount option specified.
3555          */
3556         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3557
3558         if (sbi->s_es->s_mount_opts[0]) {
3559                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3560                                               sizeof(sbi->s_es->s_mount_opts),
3561                                               GFP_KERNEL);
3562                 if (!s_mount_opts)
3563                         goto failed_mount;
3564                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3565                                    &journal_ioprio, 0)) {
3566                         ext4_msg(sb, KERN_WARNING,
3567                                  "failed to parse options in superblock: %s",
3568                                  s_mount_opts);
3569                 }
3570                 kfree(s_mount_opts);
3571         }
3572         sbi->s_def_mount_opt = sbi->s_mount_opt;
3573         if (!parse_options((char *) data, sb, &journal_devnum,
3574                            &journal_ioprio, 0))
3575                 goto failed_mount;
3576
3577         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3578                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3579                             "with data=journal disables delayed "
3580                             "allocation and O_DIRECT support!\n");
3581                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3582                         ext4_msg(sb, KERN_ERR, "can't mount with "
3583                                  "both data=journal and delalloc");
3584                         goto failed_mount;
3585                 }
3586                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3587                         ext4_msg(sb, KERN_ERR, "can't mount with "
3588                                  "both data=journal and dioread_nolock");
3589                         goto failed_mount;
3590                 }
3591                 if (test_opt(sb, DAX)) {
3592                         ext4_msg(sb, KERN_ERR, "can't mount with "
3593                                  "both data=journal and dax");
3594                         goto failed_mount;
3595                 }
3596                 if (ext4_has_feature_encrypt(sb)) {
3597                         ext4_msg(sb, KERN_WARNING,
3598                                  "encrypted files will use data=ordered "
3599                                  "instead of data journaling mode");
3600                 }
3601                 if (test_opt(sb, DELALLOC))
3602                         clear_opt(sb, DELALLOC);
3603         } else {
3604                 sb->s_iflags |= SB_I_CGROUPWB;
3605         }
3606
3607         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3608                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3609
3610         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3611             (ext4_has_compat_features(sb) ||
3612              ext4_has_ro_compat_features(sb) ||
3613              ext4_has_incompat_features(sb)))
3614                 ext4_msg(sb, KERN_WARNING,
3615                        "feature flags set on rev 0 fs, "
3616                        "running e2fsck is recommended");
3617
3618         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3619                 set_opt2(sb, HURD_COMPAT);
3620                 if (ext4_has_feature_64bit(sb)) {
3621                         ext4_msg(sb, KERN_ERR,
3622                                  "The Hurd can't support 64-bit file systems");
3623                         goto failed_mount;
3624                 }
3625
3626                 /*
3627                  * ea_inode feature uses l_i_version field which is not
3628                  * available in HURD_COMPAT mode.
3629                  */
3630                 if (ext4_has_feature_ea_inode(sb)) {
3631                         ext4_msg(sb, KERN_ERR,
3632                                  "ea_inode feature is not supported for Hurd");
3633                         goto failed_mount;
3634                 }
3635         }
3636
3637         if (IS_EXT2_SB(sb)) {
3638                 if (ext2_feature_set_ok(sb))
3639                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3640                                  "using the ext4 subsystem");
3641                 else {
3642                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3643                                  "to feature incompatibilities");
3644                         goto failed_mount;
3645                 }
3646         }
3647
3648         if (IS_EXT3_SB(sb)) {
3649                 if (ext3_feature_set_ok(sb))
3650                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3651                                  "using the ext4 subsystem");
3652                 else {
3653                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3654                                  "to feature incompatibilities");
3655                         goto failed_mount;
3656                 }
3657         }
3658
3659         /*
3660          * Check feature flags regardless of the revision level, since we
3661          * previously didn't change the revision level when setting the flags,
3662          * so there is a chance incompat flags are set on a rev 0 filesystem.
3663          */
3664         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3665                 goto failed_mount;
3666
3667         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3668         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3669             blocksize > EXT4_MAX_BLOCK_SIZE) {
3670                 ext4_msg(sb, KERN_ERR,
3671                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3672                          blocksize, le32_to_cpu(es->s_log_block_size));
3673                 goto failed_mount;
3674         }
3675         if (le32_to_cpu(es->s_log_block_size) >
3676             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3677                 ext4_msg(sb, KERN_ERR,
3678                          "Invalid log block size: %u",
3679                          le32_to_cpu(es->s_log_block_size));
3680                 goto failed_mount;
3681         }
3682
3683         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3684                 ext4_msg(sb, KERN_ERR,
3685                          "Number of reserved GDT blocks insanely large: %d",
3686                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3687                 goto failed_mount;
3688         }
3689
3690         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3691                 err = bdev_dax_supported(sb, blocksize);
3692                 if (err)
3693                         goto failed_mount;
3694         }
3695
3696         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3697                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3698                          es->s_encryption_level);
3699                 goto failed_mount;
3700         }
3701
3702         if (sb->s_blocksize != blocksize) {
3703                 /* Validate the filesystem blocksize */
3704                 if (!sb_set_blocksize(sb, blocksize)) {
3705                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3706                                         blocksize);
3707                         goto failed_mount;
3708                 }
3709
3710                 brelse(bh);
3711                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3712                 offset = do_div(logical_sb_block, blocksize);
3713                 bh = sb_bread_unmovable(sb, logical_sb_block);
3714                 if (!bh) {
3715                         ext4_msg(sb, KERN_ERR,
3716                                "Can't read superblock on 2nd try");
3717                         goto failed_mount;
3718                 }
3719                 es = (struct ext4_super_block *)(bh->b_data + offset);
3720                 sbi->s_es = es;
3721                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3722                         ext4_msg(sb, KERN_ERR,
3723                                "Magic mismatch, very weird!");
3724                         goto failed_mount;
3725                 }
3726         }
3727
3728         has_huge_files = ext4_has_feature_huge_file(sb);
3729         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3730                                                       has_huge_files);
3731         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3732
3733         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3734                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3735                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3736         } else {
3737                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3738                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3739                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3740                     (!is_power_of_2(sbi->s_inode_size)) ||
3741                     (sbi->s_inode_size > blocksize)) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                "unsupported inode size: %d",
3744                                sbi->s_inode_size);
3745                         goto failed_mount;
3746                 }
3747                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3748                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3749         }
3750
3751         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3752         if (ext4_has_feature_64bit(sb)) {
3753                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3754                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3755                     !is_power_of_2(sbi->s_desc_size)) {
3756                         ext4_msg(sb, KERN_ERR,
3757                                "unsupported descriptor size %lu",
3758                                sbi->s_desc_size);
3759                         goto failed_mount;
3760                 }
3761         } else
3762                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3763
3764         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3765         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3766
3767         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3768         if (sbi->s_inodes_per_block == 0)
3769                 goto cantfind_ext4;
3770         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3771             sbi->s_inodes_per_group > blocksize * 8) {
3772                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3773                          sbi->s_blocks_per_group);
3774                 goto failed_mount;
3775         }
3776         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3777                                         sbi->s_inodes_per_block;
3778         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3779         sbi->s_sbh = bh;
3780         sbi->s_mount_state = le16_to_cpu(es->s_state);
3781         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3782         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3783
3784         for (i = 0; i < 4; i++)
3785                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3786         sbi->s_def_hash_version = es->s_def_hash_version;
3787         if (ext4_has_feature_dir_index(sb)) {
3788                 i = le32_to_cpu(es->s_flags);
3789                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3790                         sbi->s_hash_unsigned = 3;
3791                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3792 #ifdef __CHAR_UNSIGNED__
3793                         if (!(sb->s_flags & MS_RDONLY))
3794                                 es->s_flags |=
3795                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3796                         sbi->s_hash_unsigned = 3;
3797 #else
3798                         if (!(sb->s_flags & MS_RDONLY))
3799                                 es->s_flags |=
3800                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3801 #endif
3802                 }
3803         }
3804
3805         /* Handle clustersize */
3806         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3807         has_bigalloc = ext4_has_feature_bigalloc(sb);
3808         if (has_bigalloc) {
3809                 if (clustersize < blocksize) {
3810                         ext4_msg(sb, KERN_ERR,
3811                                  "cluster size (%d) smaller than "
3812                                  "block size (%d)", clustersize, blocksize);
3813                         goto failed_mount;
3814                 }
3815                 if (le32_to_cpu(es->s_log_cluster_size) >
3816                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3817                         ext4_msg(sb, KERN_ERR,
3818                                  "Invalid log cluster size: %u",
3819                                  le32_to_cpu(es->s_log_cluster_size));
3820                         goto failed_mount;
3821                 }
3822                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3823                         le32_to_cpu(es->s_log_block_size);
3824                 sbi->s_clusters_per_group =
3825                         le32_to_cpu(es->s_clusters_per_group);
3826                 if (sbi->s_clusters_per_group > blocksize * 8) {
3827                         ext4_msg(sb, KERN_ERR,
3828                                  "#clusters per group too big: %lu",
3829                                  sbi->s_clusters_per_group);
3830                         goto failed_mount;
3831                 }
3832                 if (sbi->s_blocks_per_group !=
3833                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3834                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3835                                  "clusters per group (%lu) inconsistent",
3836                                  sbi->s_blocks_per_group,
3837                                  sbi->s_clusters_per_group);
3838                         goto failed_mount;
3839                 }
3840         } else {
3841                 if (clustersize != blocksize) {
3842                         ext4_warning(sb, "fragment/cluster size (%d) != "
3843                                      "block size (%d)", clustersize,
3844                                      blocksize);
3845                         clustersize = blocksize;
3846                 }
3847                 if (sbi->s_blocks_per_group > blocksize * 8) {
3848                         ext4_msg(sb, KERN_ERR,
3849                                  "#blocks per group too big: %lu",
3850                                  sbi->s_blocks_per_group);
3851                         goto failed_mount;
3852                 }
3853                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3854                 sbi->s_cluster_bits = 0;
3855         }
3856         sbi->s_cluster_ratio = clustersize / blocksize;
3857
3858         /* Do we have standard group size of clustersize * 8 blocks ? */
3859         if (sbi->s_blocks_per_group == clustersize << 3)
3860                 set_opt2(sb, STD_GROUP_SIZE);
3861
3862         /*
3863          * Test whether we have more sectors than will fit in sector_t,
3864          * and whether the max offset is addressable by the page cache.
3865          */
3866         err = generic_check_addressable(sb->s_blocksize_bits,
3867                                         ext4_blocks_count(es));
3868         if (err) {
3869                 ext4_msg(sb, KERN_ERR, "filesystem"
3870                          " too large to mount safely on this system");
3871                 if (sizeof(sector_t) < 8)
3872                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3873                 goto failed_mount;
3874         }
3875
3876         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3877                 goto cantfind_ext4;
3878
3879         /* check blocks count against device size */
3880         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3881         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3882                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3883                        "exceeds size of device (%llu blocks)",
3884                        ext4_blocks_count(es), blocks_count);
3885                 goto failed_mount;
3886         }
3887
3888         /*
3889          * It makes no sense for the first data block to be beyond the end
3890          * of the filesystem.
3891          */
3892         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3893                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3894                          "block %u is beyond end of filesystem (%llu)",
3895                          le32_to_cpu(es->s_first_data_block),
3896                          ext4_blocks_count(es));
3897                 goto failed_mount;
3898         }
3899         blocks_count = (ext4_blocks_count(es) -
3900                         le32_to_cpu(es->s_first_data_block) +
3901                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3902         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3903         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3904                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3905                        "(block count %llu, first data block %u, "
3906                        "blocks per group %lu)", sbi->s_groups_count,
3907                        ext4_blocks_count(es),
3908                        le32_to_cpu(es->s_first_data_block),
3909                        EXT4_BLOCKS_PER_GROUP(sb));
3910                 goto failed_mount;
3911         }
3912         sbi->s_groups_count = blocks_count;
3913         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3914                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3915         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3916                    EXT4_DESC_PER_BLOCK(sb);
3917         if (ext4_has_feature_meta_bg(sb)) {
3918                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3919                         ext4_msg(sb, KERN_WARNING,
3920                                  "first meta block group too large: %u "
3921                                  "(group descriptor block count %u)",
3922                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3923                         goto failed_mount;
3924                 }
3925         }
3926         sbi->s_group_desc = kvmalloc(db_count *
3927                                           sizeof(struct buffer_head *),
3928                                           GFP_KERNEL);
3929         if (sbi->s_group_desc == NULL) {
3930                 ext4_msg(sb, KERN_ERR, "not enough memory");
3931                 ret = -ENOMEM;
3932                 goto failed_mount;
3933         }
3934
3935         bgl_lock_init(sbi->s_blockgroup_lock);
3936
3937         /* Pre-read the descriptors into the buffer cache */
3938         for (i = 0; i < db_count; i++) {
3939                 block = descriptor_loc(sb, logical_sb_block, i);
3940                 sb_breadahead(sb, block);
3941         }
3942
3943         for (i = 0; i < db_count; i++) {
3944                 block = descriptor_loc(sb, logical_sb_block, i);
3945                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3946                 if (!sbi->s_group_desc[i]) {
3947                         ext4_msg(sb, KERN_ERR,
3948                                "can't read group descriptor %d", i);
3949                         db_count = i;
3950                         goto failed_mount2;
3951                 }
3952         }
3953         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3954                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3955                 ret = -EFSCORRUPTED;
3956                 goto failed_mount2;
3957         }
3958
3959         sbi->s_gdb_count = db_count;
3960         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3961         spin_lock_init(&sbi->s_next_gen_lock);
3962
3963         setup_timer(&sbi->s_err_report, print_daily_error_info,
3964                 (unsigned long) sb);
3965
3966         /* Register extent status tree shrinker */
3967         if (ext4_es_register_shrinker(sbi))
3968                 goto failed_mount3;
3969
3970         sbi->s_stripe = ext4_get_stripe_size(sbi);
3971         sbi->s_extent_max_zeroout_kb = 32;
3972
3973         /*
3974          * set up enough so that it can read an inode
3975          */
3976         sb->s_op = &ext4_sops;
3977         sb->s_export_op = &ext4_export_ops;
3978         sb->s_xattr = ext4_xattr_handlers;
3979         sb->s_cop = &ext4_cryptops;
3980 #ifdef CONFIG_QUOTA
3981         sb->dq_op = &ext4_quota_operations;
3982         if (ext4_has_feature_quota(sb))
3983                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3984         else
3985                 sb->s_qcop = &ext4_qctl_operations;
3986         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3987 #endif
3988         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3989
3990         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3991         mutex_init(&sbi->s_orphan_lock);
3992
3993         sb->s_root = NULL;
3994
3995         needs_recovery = (es->s_last_orphan != 0 ||
3996                           ext4_has_feature_journal_needs_recovery(sb));
3997
3998         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3999                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4000                         goto failed_mount3a;
4001
4002         /*
4003          * The first inode we look at is the journal inode.  Don't try
4004          * root first: it may be modified in the journal!
4005          */
4006         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4007                 err = ext4_load_journal(sb, es, journal_devnum);
4008                 if (err)
4009                         goto failed_mount3a;
4010         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
4011                    ext4_has_feature_journal_needs_recovery(sb)) {
4012                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4013                        "suppressed and not mounted read-only");
4014                 goto failed_mount_wq;
4015         } else {
4016                 /* Nojournal mode, all journal mount options are illegal */
4017                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4018                         ext4_msg(sb, KERN_ERR, "can't mount with "
4019                                  "journal_checksum, fs mounted w/o journal");
4020                         goto failed_mount_wq;
4021                 }
4022                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4023                         ext4_msg(sb, KERN_ERR, "can't mount with "
4024                                  "journal_async_commit, fs mounted w/o journal");
4025                         goto failed_mount_wq;
4026                 }
4027                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4028                         ext4_msg(sb, KERN_ERR, "can't mount with "
4029                                  "commit=%lu, fs mounted w/o journal",
4030                                  sbi->s_commit_interval / HZ);
4031                         goto failed_mount_wq;
4032                 }
4033                 if (EXT4_MOUNT_DATA_FLAGS &
4034                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4035                         ext4_msg(sb, KERN_ERR, "can't mount with "
4036                                  "data=, fs mounted w/o journal");
4037                         goto failed_mount_wq;
4038                 }
4039                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4040                 clear_opt(sb, JOURNAL_CHECKSUM);
4041                 clear_opt(sb, DATA_FLAGS);
4042                 sbi->s_journal = NULL;
4043                 needs_recovery = 0;
4044                 goto no_journal;
4045         }
4046
4047         if (ext4_has_feature_64bit(sb) &&
4048             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4049                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4050                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4051                 goto failed_mount_wq;
4052         }
4053
4054         if (!set_journal_csum_feature_set(sb)) {
4055                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4056                          "feature set");
4057                 goto failed_mount_wq;
4058         }
4059
4060         /* We have now updated the journal if required, so we can
4061          * validate the data journaling mode. */
4062         switch (test_opt(sb, DATA_FLAGS)) {
4063         case 0:
4064                 /* No mode set, assume a default based on the journal
4065                  * capabilities: ORDERED_DATA if the journal can
4066                  * cope, else JOURNAL_DATA
4067                  */
4068                 if (jbd2_journal_check_available_features
4069                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4070                         set_opt(sb, ORDERED_DATA);
4071                 else
4072                         set_opt(sb, JOURNAL_DATA);
4073                 break;
4074
4075         case EXT4_MOUNT_ORDERED_DATA:
4076         case EXT4_MOUNT_WRITEBACK_DATA:
4077                 if (!jbd2_journal_check_available_features
4078                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4079                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4080                                "requested data journaling mode");
4081                         goto failed_mount_wq;
4082                 }
4083         default:
4084                 break;
4085         }
4086
4087         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4088             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4089                 ext4_msg(sb, KERN_ERR, "can't mount with "
4090                         "journal_async_commit in data=ordered mode");
4091                 goto failed_mount_wq;
4092         }
4093
4094         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4095
4096         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4097
4098 no_journal:
4099         if (!test_opt(sb, NO_MBCACHE)) {
4100                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4101                 if (!sbi->s_ea_block_cache) {
4102                         ext4_msg(sb, KERN_ERR,
4103                                  "Failed to create ea_block_cache");
4104                         goto failed_mount_wq;
4105                 }
4106
4107                 if (ext4_has_feature_ea_inode(sb)) {
4108                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4109                         if (!sbi->s_ea_inode_cache) {
4110                                 ext4_msg(sb, KERN_ERR,
4111                                          "Failed to create ea_inode_cache");
4112                                 goto failed_mount_wq;
4113                         }
4114                 }
4115         }
4116
4117         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4118             (blocksize != PAGE_SIZE)) {
4119                 ext4_msg(sb, KERN_ERR,
4120                          "Unsupported blocksize for fs encryption");
4121                 goto failed_mount_wq;
4122         }
4123
4124         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4125             !ext4_has_feature_encrypt(sb)) {
4126                 ext4_set_feature_encrypt(sb);
4127                 ext4_commit_super(sb, 1);
4128         }
4129
4130         /*
4131          * Get the # of file system overhead blocks from the
4132          * superblock if present.
4133          */
4134         if (es->s_overhead_clusters)
4135                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4136         else {
4137                 err = ext4_calculate_overhead(sb);
4138                 if (err)
4139                         goto failed_mount_wq;
4140         }
4141
4142         /*
4143          * The maximum number of concurrent works can be high and
4144          * concurrency isn't really necessary.  Limit it to 1.
4145          */
4146         EXT4_SB(sb)->rsv_conversion_wq =
4147                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4148         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4149                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4150                 ret = -ENOMEM;
4151                 goto failed_mount4;
4152         }
4153
4154         /*
4155          * The jbd2_journal_load will have done any necessary log recovery,
4156          * so we can safely mount the rest of the filesystem now.
4157          */
4158
4159         root = ext4_iget(sb, EXT4_ROOT_INO);
4160         if (IS_ERR(root)) {
4161                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4162                 ret = PTR_ERR(root);
4163                 root = NULL;
4164                 goto failed_mount4;
4165         }
4166         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4167                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4168                 iput(root);
4169                 goto failed_mount4;
4170         }
4171         sb->s_root = d_make_root(root);
4172         if (!sb->s_root) {
4173                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4174                 ret = -ENOMEM;
4175                 goto failed_mount4;
4176         }
4177
4178         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4179                 sb->s_flags |= MS_RDONLY;
4180
4181         /* determine the minimum size of new large inodes, if present */
4182         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4183             sbi->s_want_extra_isize == 0) {
4184                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4185                                                      EXT4_GOOD_OLD_INODE_SIZE;
4186                 if (ext4_has_feature_extra_isize(sb)) {
4187                         if (sbi->s_want_extra_isize <
4188                             le16_to_cpu(es->s_want_extra_isize))
4189                                 sbi->s_want_extra_isize =
4190                                         le16_to_cpu(es->s_want_extra_isize);
4191                         if (sbi->s_want_extra_isize <
4192                             le16_to_cpu(es->s_min_extra_isize))
4193                                 sbi->s_want_extra_isize =
4194                                         le16_to_cpu(es->s_min_extra_isize);
4195                 }
4196         }
4197         /* Check if enough inode space is available */
4198         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4199                                                         sbi->s_inode_size) {
4200                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4201                                                        EXT4_GOOD_OLD_INODE_SIZE;
4202                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4203                          "available");
4204         }
4205
4206         ext4_set_resv_clusters(sb);
4207
4208         err = ext4_setup_system_zone(sb);
4209         if (err) {
4210                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4211                          "zone (%d)", err);
4212                 goto failed_mount4a;
4213         }
4214
4215         ext4_ext_init(sb);
4216         err = ext4_mb_init(sb);
4217         if (err) {
4218                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4219                          err);
4220                 goto failed_mount5;
4221         }
4222
4223         block = ext4_count_free_clusters(sb);
4224         ext4_free_blocks_count_set(sbi->s_es,
4225                                    EXT4_C2B(sbi, block));
4226         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4227                                   GFP_KERNEL);
4228         if (!err) {
4229                 unsigned long freei = ext4_count_free_inodes(sb);
4230                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4231                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4232                                           GFP_KERNEL);
4233         }
4234         if (!err)
4235                 err = percpu_counter_init(&sbi->s_dirs_counter,
4236                                           ext4_count_dirs(sb), GFP_KERNEL);
4237         if (!err)
4238                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4239                                           GFP_KERNEL);
4240         if (!err)
4241                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4242
4243         if (err) {
4244                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4245                 goto failed_mount6;
4246         }
4247
4248         if (ext4_has_feature_flex_bg(sb))
4249                 if (!ext4_fill_flex_info(sb)) {
4250                         ext4_msg(sb, KERN_ERR,
4251                                "unable to initialize "
4252                                "flex_bg meta info!");
4253                         goto failed_mount6;
4254                 }
4255
4256         err = ext4_register_li_request(sb, first_not_zeroed);
4257         if (err)
4258                 goto failed_mount6;
4259
4260         err = ext4_register_sysfs(sb);
4261         if (err)
4262                 goto failed_mount7;
4263
4264 #ifdef CONFIG_QUOTA
4265         /* Enable quota usage during mount. */
4266         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4267                 err = ext4_enable_quotas(sb);
4268                 if (err)
4269                         goto failed_mount8;
4270         }
4271 #endif  /* CONFIG_QUOTA */
4272
4273         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4274         ext4_orphan_cleanup(sb, es);
4275         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4276         if (needs_recovery) {
4277                 ext4_msg(sb, KERN_INFO, "recovery complete");
4278                 ext4_mark_recovery_complete(sb, es);
4279         }
4280         if (EXT4_SB(sb)->s_journal) {
4281                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4282                         descr = " journalled data mode";
4283                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4284                         descr = " ordered data mode";
4285                 else
4286                         descr = " writeback data mode";
4287         } else
4288                 descr = "out journal";
4289
4290         if (test_opt(sb, DISCARD)) {
4291                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4292                 if (!blk_queue_discard(q))
4293                         ext4_msg(sb, KERN_WARNING,
4294                                  "mounting with \"discard\" option, but "
4295                                  "the device does not support discard");
4296         }
4297
4298         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4299                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4300                          "Opts: %.*s%s%s", descr,
4301                          (int) sizeof(sbi->s_es->s_mount_opts),
4302                          sbi->s_es->s_mount_opts,
4303                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4304
4305         if (es->s_error_count)
4306                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4307
4308         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4309         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4310         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4311         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4312
4313         kfree(orig_data);
4314         return 0;
4315
4316 cantfind_ext4:
4317         if (!silent)
4318                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4319         goto failed_mount;
4320
4321 #ifdef CONFIG_QUOTA
4322 failed_mount8:
4323         ext4_unregister_sysfs(sb);
4324 #endif
4325 failed_mount7:
4326         ext4_unregister_li_request(sb);
4327 failed_mount6:
4328         ext4_mb_release(sb);
4329         if (sbi->s_flex_groups)
4330                 kvfree(sbi->s_flex_groups);
4331         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4332         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4333         percpu_counter_destroy(&sbi->s_dirs_counter);
4334         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4335 failed_mount5:
4336         ext4_ext_release(sb);
4337         ext4_release_system_zone(sb);
4338 failed_mount4a:
4339         dput(sb->s_root);
4340         sb->s_root = NULL;
4341 failed_mount4:
4342         ext4_msg(sb, KERN_ERR, "mount failed");
4343         if (EXT4_SB(sb)->rsv_conversion_wq)
4344                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4345 failed_mount_wq:
4346         if (sbi->s_ea_inode_cache) {
4347                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4348                 sbi->s_ea_inode_cache = NULL;
4349         }
4350         if (sbi->s_ea_block_cache) {
4351                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4352                 sbi->s_ea_block_cache = NULL;
4353         }
4354         if (sbi->s_journal) {
4355                 jbd2_journal_destroy(sbi->s_journal);
4356                 sbi->s_journal = NULL;
4357         }
4358 failed_mount3a:
4359         ext4_es_unregister_shrinker(sbi);
4360 failed_mount3:
4361         del_timer_sync(&sbi->s_err_report);
4362         if (sbi->s_mmp_tsk)
4363                 kthread_stop(sbi->s_mmp_tsk);
4364 failed_mount2:
4365         for (i = 0; i < db_count; i++)
4366                 brelse(sbi->s_group_desc[i]);
4367         kvfree(sbi->s_group_desc);
4368 failed_mount:
4369         if (sbi->s_chksum_driver)
4370                 crypto_free_shash(sbi->s_chksum_driver);
4371 #ifdef CONFIG_QUOTA
4372         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4373                 kfree(sbi->s_qf_names[i]);
4374 #endif
4375         ext4_blkdev_remove(sbi);
4376         brelse(bh);
4377 out_fail:
4378         sb->s_fs_info = NULL;
4379         kfree(sbi->s_blockgroup_lock);
4380 out_free_base:
4381         kfree(sbi);
4382         kfree(orig_data);
4383         return err ? err : ret;
4384 }
4385
4386 /*
4387  * Setup any per-fs journal parameters now.  We'll do this both on
4388  * initial mount, once the journal has been initialised but before we've
4389  * done any recovery; and again on any subsequent remount.
4390  */
4391 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4392 {
4393         struct ext4_sb_info *sbi = EXT4_SB(sb);
4394
4395         journal->j_commit_interval = sbi->s_commit_interval;
4396         journal->j_min_batch_time = sbi->s_min_batch_time;
4397         journal->j_max_batch_time = sbi->s_max_batch_time;
4398
4399         write_lock(&journal->j_state_lock);
4400         if (test_opt(sb, BARRIER))
4401                 journal->j_flags |= JBD2_BARRIER;
4402         else
4403                 journal->j_flags &= ~JBD2_BARRIER;
4404         if (test_opt(sb, DATA_ERR_ABORT))
4405                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4406         else
4407                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4408         write_unlock(&journal->j_state_lock);
4409 }
4410
4411 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4412                                              unsigned int journal_inum)
4413 {
4414         struct inode *journal_inode;
4415
4416         /*
4417          * Test for the existence of a valid inode on disk.  Bad things
4418          * happen if we iget() an unused inode, as the subsequent iput()
4419          * will try to delete it.
4420          */
4421         journal_inode = ext4_iget(sb, journal_inum);
4422         if (IS_ERR(journal_inode)) {
4423                 ext4_msg(sb, KERN_ERR, "no journal found");
4424                 return NULL;
4425         }
4426         if (!journal_inode->i_nlink) {
4427                 make_bad_inode(journal_inode);
4428                 iput(journal_inode);
4429                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4430                 return NULL;
4431         }
4432
4433         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4434                   journal_inode, journal_inode->i_size);
4435         if (!S_ISREG(journal_inode->i_mode)) {
4436                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4437                 iput(journal_inode);
4438                 return NULL;
4439         }
4440         return journal_inode;
4441 }
4442
4443 static journal_t *ext4_get_journal(struct super_block *sb,
4444                                    unsigned int journal_inum)
4445 {
4446         struct inode *journal_inode;
4447         journal_t *journal;
4448
4449         BUG_ON(!ext4_has_feature_journal(sb));
4450
4451         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4452         if (!journal_inode)
4453                 return NULL;
4454
4455         journal = jbd2_journal_init_inode(journal_inode);
4456         if (!journal) {
4457                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4458                 iput(journal_inode);
4459                 return NULL;
4460         }
4461         journal->j_private = sb;
4462         ext4_init_journal_params(sb, journal);
4463         return journal;
4464 }
4465
4466 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4467                                        dev_t j_dev)
4468 {
4469         struct buffer_head *bh;
4470         journal_t *journal;
4471         ext4_fsblk_t start;
4472         ext4_fsblk_t len;
4473         int hblock, blocksize;
4474         ext4_fsblk_t sb_block;
4475         unsigned long offset;
4476         struct ext4_super_block *es;
4477         struct block_device *bdev;
4478
4479         BUG_ON(!ext4_has_feature_journal(sb));
4480
4481         bdev = ext4_blkdev_get(j_dev, sb);
4482         if (bdev == NULL)
4483                 return NULL;
4484
4485         blocksize = sb->s_blocksize;
4486         hblock = bdev_logical_block_size(bdev);
4487         if (blocksize < hblock) {
4488                 ext4_msg(sb, KERN_ERR,
4489                         "blocksize too small for journal device");
4490                 goto out_bdev;
4491         }
4492
4493         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4494         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4495         set_blocksize(bdev, blocksize);
4496         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4497                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4498                        "external journal");
4499                 goto out_bdev;
4500         }
4501
4502         es = (struct ext4_super_block *) (bh->b_data + offset);
4503         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4504             !(le32_to_cpu(es->s_feature_incompat) &
4505               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4506                 ext4_msg(sb, KERN_ERR, "external journal has "
4507                                         "bad superblock");
4508                 brelse(bh);
4509                 goto out_bdev;
4510         }
4511
4512         if ((le32_to_cpu(es->s_feature_ro_compat) &
4513              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4514             es->s_checksum != ext4_superblock_csum(sb, es)) {
4515                 ext4_msg(sb, KERN_ERR, "external journal has "
4516                                        "corrupt superblock");
4517                 brelse(bh);
4518                 goto out_bdev;
4519         }
4520
4521         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4522                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4523                 brelse(bh);
4524                 goto out_bdev;
4525         }
4526
4527         len = ext4_blocks_count(es);
4528         start = sb_block + 1;
4529         brelse(bh);     /* we're done with the superblock */
4530
4531         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4532                                         start, len, blocksize);
4533         if (!journal) {
4534                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4535                 goto out_bdev;
4536         }
4537         journal->j_private = sb;
4538         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4539         wait_on_buffer(journal->j_sb_buffer);
4540         if (!buffer_uptodate(journal->j_sb_buffer)) {
4541                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4542                 goto out_journal;
4543         }
4544         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4545                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4546                                         "user (unsupported) - %d",
4547                         be32_to_cpu(journal->j_superblock->s_nr_users));
4548                 goto out_journal;
4549         }
4550         EXT4_SB(sb)->journal_bdev = bdev;
4551         ext4_init_journal_params(sb, journal);
4552         return journal;
4553
4554 out_journal:
4555         jbd2_journal_destroy(journal);
4556 out_bdev:
4557         ext4_blkdev_put(bdev);
4558         return NULL;
4559 }
4560
4561 static int ext4_load_journal(struct super_block *sb,
4562                              struct ext4_super_block *es,
4563                              unsigned long journal_devnum)
4564 {
4565         journal_t *journal;
4566         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4567         dev_t journal_dev;
4568         int err = 0;
4569         int really_read_only;
4570
4571         BUG_ON(!ext4_has_feature_journal(sb));
4572
4573         if (journal_devnum &&
4574             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4575                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4576                         "numbers have changed");
4577                 journal_dev = new_decode_dev(journal_devnum);
4578         } else
4579                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4580
4581         really_read_only = bdev_read_only(sb->s_bdev);
4582
4583         /*
4584          * Are we loading a blank journal or performing recovery after a
4585          * crash?  For recovery, we need to check in advance whether we
4586          * can get read-write access to the device.
4587          */
4588         if (ext4_has_feature_journal_needs_recovery(sb)) {
4589                 if (sb->s_flags & MS_RDONLY) {
4590                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4591                                         "required on readonly filesystem");
4592                         if (really_read_only) {
4593                                 ext4_msg(sb, KERN_ERR, "write access "
4594                                         "unavailable, cannot proceed");
4595                                 return -EROFS;
4596                         }
4597                         ext4_msg(sb, KERN_INFO, "write access will "
4598                                "be enabled during recovery");
4599                 }
4600         }
4601
4602         if (journal_inum && journal_dev) {
4603                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4604                        "and inode journals!");
4605                 return -EINVAL;
4606         }
4607
4608         if (journal_inum) {
4609                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4610                         return -EINVAL;
4611         } else {
4612                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4613                         return -EINVAL;
4614         }
4615
4616         if (!(journal->j_flags & JBD2_BARRIER))
4617                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4618
4619         if (!ext4_has_feature_journal_needs_recovery(sb))
4620                 err = jbd2_journal_wipe(journal, !really_read_only);
4621         if (!err) {
4622                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4623                 if (save)
4624                         memcpy(save, ((char *) es) +
4625                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4626                 err = jbd2_journal_load(journal);
4627                 if (save)
4628                         memcpy(((char *) es) + EXT4_S_ERR_START,
4629                                save, EXT4_S_ERR_LEN);
4630                 kfree(save);
4631         }
4632
4633         if (err) {
4634                 ext4_msg(sb, KERN_ERR, "error loading journal");
4635                 jbd2_journal_destroy(journal);
4636                 return err;
4637         }
4638
4639         EXT4_SB(sb)->s_journal = journal;
4640         ext4_clear_journal_err(sb, es);
4641
4642         if (!really_read_only && journal_devnum &&
4643             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4644                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4645
4646                 /* Make sure we flush the recovery flag to disk. */
4647                 ext4_commit_super(sb, 1);
4648         }
4649
4650         return 0;
4651 }
4652
4653 static int ext4_commit_super(struct super_block *sb, int sync)
4654 {
4655         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4656         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4657         int error = 0;
4658
4659         if (!sbh || block_device_ejected(sb))
4660                 return error;
4661         /*
4662          * If the file system is mounted read-only, don't update the
4663          * superblock write time.  This avoids updating the superblock
4664          * write time when we are mounting the root file system
4665          * read/only but we need to replay the journal; at that point,
4666          * for people who are east of GMT and who make their clock
4667          * tick in localtime for Windows bug-for-bug compatibility,
4668          * the clock is set in the future, and this will cause e2fsck
4669          * to complain and force a full file system check.
4670          */
4671         if (!(sb->s_flags & MS_RDONLY))
4672                 es->s_wtime = cpu_to_le32(get_seconds());
4673         if (sb->s_bdev->bd_part)
4674                 es->s_kbytes_written =
4675                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4676                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4677                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4678         else
4679                 es->s_kbytes_written =
4680                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4681         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4682                 ext4_free_blocks_count_set(es,
4683                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4684                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4685         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4686                 es->s_free_inodes_count =
4687                         cpu_to_le32(percpu_counter_sum_positive(
4688                                 &EXT4_SB(sb)->s_freeinodes_counter));
4689         BUFFER_TRACE(sbh, "marking dirty");
4690         ext4_superblock_csum_set(sb);
4691         if (sync)
4692                 lock_buffer(sbh);
4693         if (buffer_write_io_error(sbh)) {
4694                 /*
4695                  * Oh, dear.  A previous attempt to write the
4696                  * superblock failed.  This could happen because the
4697                  * USB device was yanked out.  Or it could happen to
4698                  * be a transient write error and maybe the block will
4699                  * be remapped.  Nothing we can do but to retry the
4700                  * write and hope for the best.
4701                  */
4702                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4703                        "superblock detected");
4704                 clear_buffer_write_io_error(sbh);
4705                 set_buffer_uptodate(sbh);
4706         }
4707         mark_buffer_dirty(sbh);
4708         if (sync) {
4709                 unlock_buffer(sbh);
4710                 error = __sync_dirty_buffer(sbh,
4711                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4712                 if (error)
4713                         return error;
4714
4715                 error = buffer_write_io_error(sbh);
4716                 if (error) {
4717                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4718                                "superblock");
4719                         clear_buffer_write_io_error(sbh);
4720                         set_buffer_uptodate(sbh);
4721                 }
4722         }
4723         return error;
4724 }
4725
4726 /*
4727  * Have we just finished recovery?  If so, and if we are mounting (or
4728  * remounting) the filesystem readonly, then we will end up with a
4729  * consistent fs on disk.  Record that fact.
4730  */
4731 static void ext4_mark_recovery_complete(struct super_block *sb,
4732                                         struct ext4_super_block *es)
4733 {
4734         journal_t *journal = EXT4_SB(sb)->s_journal;
4735
4736         if (!ext4_has_feature_journal(sb)) {
4737                 BUG_ON(journal != NULL);
4738                 return;
4739         }
4740         jbd2_journal_lock_updates(journal);
4741         if (jbd2_journal_flush(journal) < 0)
4742                 goto out;
4743
4744         if (ext4_has_feature_journal_needs_recovery(sb) &&
4745             sb->s_flags & MS_RDONLY) {
4746                 ext4_clear_feature_journal_needs_recovery(sb);
4747                 ext4_commit_super(sb, 1);
4748         }
4749
4750 out:
4751         jbd2_journal_unlock_updates(journal);
4752 }
4753
4754 /*
4755  * If we are mounting (or read-write remounting) a filesystem whose journal
4756  * has recorded an error from a previous lifetime, move that error to the
4757  * main filesystem now.
4758  */
4759 static void ext4_clear_journal_err(struct super_block *sb,
4760                                    struct ext4_super_block *es)
4761 {
4762         journal_t *journal;
4763         int j_errno;
4764         const char *errstr;
4765
4766         BUG_ON(!ext4_has_feature_journal(sb));
4767
4768         journal = EXT4_SB(sb)->s_journal;
4769
4770         /*
4771          * Now check for any error status which may have been recorded in the
4772          * journal by a prior ext4_error() or ext4_abort()
4773          */
4774
4775         j_errno = jbd2_journal_errno(journal);
4776         if (j_errno) {
4777                 char nbuf[16];
4778
4779                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4780                 ext4_warning(sb, "Filesystem error recorded "
4781                              "from previous mount: %s", errstr);
4782                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4783
4784                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4785                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4786                 ext4_commit_super(sb, 1);
4787
4788                 jbd2_journal_clear_err(journal);
4789                 jbd2_journal_update_sb_errno(journal);
4790         }
4791 }
4792
4793 /*
4794  * Force the running and committing transactions to commit,
4795  * and wait on the commit.
4796  */
4797 int ext4_force_commit(struct super_block *sb)
4798 {
4799         journal_t *journal;
4800
4801         if (sb->s_flags & MS_RDONLY)
4802                 return 0;
4803
4804         journal = EXT4_SB(sb)->s_journal;
4805         return ext4_journal_force_commit(journal);
4806 }
4807
4808 static int ext4_sync_fs(struct super_block *sb, int wait)
4809 {
4810         int ret = 0;
4811         tid_t target;
4812         bool needs_barrier = false;
4813         struct ext4_sb_info *sbi = EXT4_SB(sb);
4814
4815         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4816                 return 0;
4817
4818         trace_ext4_sync_fs(sb, wait);
4819         flush_workqueue(sbi->rsv_conversion_wq);
4820         /*
4821          * Writeback quota in non-journalled quota case - journalled quota has
4822          * no dirty dquots
4823          */
4824         dquot_writeback_dquots(sb, -1);
4825         /*
4826          * Data writeback is possible w/o journal transaction, so barrier must
4827          * being sent at the end of the function. But we can skip it if
4828          * transaction_commit will do it for us.
4829          */
4830         if (sbi->s_journal) {
4831                 target = jbd2_get_latest_transaction(sbi->s_journal);
4832                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4833                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4834                         needs_barrier = true;
4835
4836                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4837                         if (wait)
4838                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4839                                                            target);
4840                 }
4841         } else if (wait && test_opt(sb, BARRIER))
4842                 needs_barrier = true;
4843         if (needs_barrier) {
4844                 int err;
4845                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4846                 if (!ret)
4847                         ret = err;
4848         }
4849
4850         return ret;
4851 }
4852
4853 /*
4854  * LVM calls this function before a (read-only) snapshot is created.  This
4855  * gives us a chance to flush the journal completely and mark the fs clean.
4856  *
4857  * Note that only this function cannot bring a filesystem to be in a clean
4858  * state independently. It relies on upper layer to stop all data & metadata
4859  * modifications.
4860  */
4861 static int ext4_freeze(struct super_block *sb)
4862 {
4863         int error = 0;
4864         journal_t *journal;
4865
4866         if (sb->s_flags & MS_RDONLY)
4867                 return 0;
4868
4869         journal = EXT4_SB(sb)->s_journal;
4870
4871         if (journal) {
4872                 /* Now we set up the journal barrier. */
4873                 jbd2_journal_lock_updates(journal);
4874
4875                 /*
4876                  * Don't clear the needs_recovery flag if we failed to
4877                  * flush the journal.
4878                  */
4879                 error = jbd2_journal_flush(journal);
4880                 if (error < 0)
4881                         goto out;
4882
4883                 /* Journal blocked and flushed, clear needs_recovery flag. */
4884                 ext4_clear_feature_journal_needs_recovery(sb);
4885         }
4886
4887         error = ext4_commit_super(sb, 1);
4888 out:
4889         if (journal)
4890                 /* we rely on upper layer to stop further updates */
4891                 jbd2_journal_unlock_updates(journal);
4892         return error;
4893 }
4894
4895 /*
4896  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4897  * flag here, even though the filesystem is not technically dirty yet.
4898  */
4899 static int ext4_unfreeze(struct super_block *sb)
4900 {
4901         if ((sb->s_flags & MS_RDONLY) || ext4_forced_shutdown(EXT4_SB(sb)))
4902                 return 0;
4903
4904         if (EXT4_SB(sb)->s_journal) {
4905                 /* Reset the needs_recovery flag before the fs is unlocked. */
4906                 ext4_set_feature_journal_needs_recovery(sb);
4907         }
4908
4909         ext4_commit_super(sb, 1);
4910         return 0;
4911 }
4912
4913 /*
4914  * Structure to save mount options for ext4_remount's benefit
4915  */
4916 struct ext4_mount_options {
4917         unsigned long s_mount_opt;
4918         unsigned long s_mount_opt2;
4919         kuid_t s_resuid;
4920         kgid_t s_resgid;
4921         unsigned long s_commit_interval;
4922         u32 s_min_batch_time, s_max_batch_time;
4923 #ifdef CONFIG_QUOTA
4924         int s_jquota_fmt;
4925         char *s_qf_names[EXT4_MAXQUOTAS];
4926 #endif
4927 };
4928
4929 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4930 {
4931         struct ext4_super_block *es;
4932         struct ext4_sb_info *sbi = EXT4_SB(sb);
4933         unsigned long old_sb_flags;
4934         struct ext4_mount_options old_opts;
4935         int enable_quota = 0;
4936         ext4_group_t g;
4937         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4938         int err = 0;
4939 #ifdef CONFIG_QUOTA
4940         int i, j;
4941 #endif
4942         char *orig_data = kstrdup(data, GFP_KERNEL);
4943
4944         /* Store the original options */
4945         old_sb_flags = sb->s_flags;
4946         old_opts.s_mount_opt = sbi->s_mount_opt;
4947         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4948         old_opts.s_resuid = sbi->s_resuid;
4949         old_opts.s_resgid = sbi->s_resgid;
4950         old_opts.s_commit_interval = sbi->s_commit_interval;
4951         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4952         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4953 #ifdef CONFIG_QUOTA
4954         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4955         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4956                 if (sbi->s_qf_names[i]) {
4957                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4958                                                          GFP_KERNEL);
4959                         if (!old_opts.s_qf_names[i]) {
4960                                 for (j = 0; j < i; j++)
4961                                         kfree(old_opts.s_qf_names[j]);
4962                                 kfree(orig_data);
4963                                 return -ENOMEM;
4964                         }
4965                 } else
4966                         old_opts.s_qf_names[i] = NULL;
4967 #endif
4968         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4969                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4970
4971         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4972                 err = -EINVAL;
4973                 goto restore_opts;
4974         }
4975
4976         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4977             test_opt(sb, JOURNAL_CHECKSUM)) {
4978                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4979                          "during remount not supported; ignoring");
4980                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4981         }
4982
4983         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4984                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4985                         ext4_msg(sb, KERN_ERR, "can't mount with "
4986                                  "both data=journal and delalloc");
4987                         err = -EINVAL;
4988                         goto restore_opts;
4989                 }
4990                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4991                         ext4_msg(sb, KERN_ERR, "can't mount with "
4992                                  "both data=journal and dioread_nolock");
4993                         err = -EINVAL;
4994                         goto restore_opts;
4995                 }
4996                 if (test_opt(sb, DAX)) {
4997                         ext4_msg(sb, KERN_ERR, "can't mount with "
4998                                  "both data=journal and dax");
4999                         err = -EINVAL;
5000                         goto restore_opts;
5001                 }
5002         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5003                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5004                         ext4_msg(sb, KERN_ERR, "can't mount with "
5005                                 "journal_async_commit in data=ordered mode");
5006                         err = -EINVAL;
5007                         goto restore_opts;
5008                 }
5009         }
5010
5011         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5012                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5013                 err = -EINVAL;
5014                 goto restore_opts;
5015         }
5016
5017         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5018                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5019                         "dax flag with busy inodes while remounting");
5020                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5021         }
5022
5023         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5024                 ext4_abort(sb, "Abort forced by user");
5025
5026         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
5027                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
5028
5029         es = sbi->s_es;
5030
5031         if (sbi->s_journal) {
5032                 ext4_init_journal_params(sb, sbi->s_journal);
5033                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5034         }
5035
5036         if (*flags & MS_LAZYTIME)
5037                 sb->s_flags |= MS_LAZYTIME;
5038
5039         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
5040                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5041                         err = -EROFS;
5042                         goto restore_opts;
5043                 }
5044
5045                 if (*flags & MS_RDONLY) {
5046                         err = sync_filesystem(sb);
5047                         if (err < 0)
5048                                 goto restore_opts;
5049                         err = dquot_suspend(sb, -1);
5050                         if (err < 0)
5051                                 goto restore_opts;
5052
5053                         /*
5054                          * First of all, the unconditional stuff we have to do
5055                          * to disable replay of the journal when we next remount
5056                          */
5057                         sb->s_flags |= MS_RDONLY;
5058
5059                         /*
5060                          * OK, test if we are remounting a valid rw partition
5061                          * readonly, and if so set the rdonly flag and then
5062                          * mark the partition as valid again.
5063                          */
5064                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5065                             (sbi->s_mount_state & EXT4_VALID_FS))
5066                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5067
5068                         if (sbi->s_journal)
5069                                 ext4_mark_recovery_complete(sb, es);
5070                 } else {
5071                         /* Make sure we can mount this feature set readwrite */
5072                         if (ext4_has_feature_readonly(sb) ||
5073                             !ext4_feature_set_ok(sb, 0)) {
5074                                 err = -EROFS;
5075                                 goto restore_opts;
5076                         }
5077                         /*
5078                          * Make sure the group descriptor checksums
5079                          * are sane.  If they aren't, refuse to remount r/w.
5080                          */
5081                         for (g = 0; g < sbi->s_groups_count; g++) {
5082                                 struct ext4_group_desc *gdp =
5083                                         ext4_get_group_desc(sb, g, NULL);
5084
5085                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5086                                         ext4_msg(sb, KERN_ERR,
5087                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5088                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5089                                                le16_to_cpu(gdp->bg_checksum));
5090                                         err = -EFSBADCRC;
5091                                         goto restore_opts;
5092                                 }
5093                         }
5094
5095                         /*
5096                          * If we have an unprocessed orphan list hanging
5097                          * around from a previously readonly bdev mount,
5098                          * require a full umount/remount for now.
5099                          */
5100                         if (es->s_last_orphan) {
5101                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5102                                        "remount RDWR because of unprocessed "
5103                                        "orphan inode list.  Please "
5104                                        "umount/remount instead");
5105                                 err = -EINVAL;
5106                                 goto restore_opts;
5107                         }
5108
5109                         /*
5110                          * Mounting a RDONLY partition read-write, so reread
5111                          * and store the current valid flag.  (It may have
5112                          * been changed by e2fsck since we originally mounted
5113                          * the partition.)
5114                          */
5115                         if (sbi->s_journal)
5116                                 ext4_clear_journal_err(sb, es);
5117                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5118                         if (!ext4_setup_super(sb, es, 0))
5119                                 sb->s_flags &= ~MS_RDONLY;
5120                         if (ext4_has_feature_mmp(sb))
5121                                 if (ext4_multi_mount_protect(sb,
5122                                                 le64_to_cpu(es->s_mmp_block))) {
5123                                         err = -EROFS;
5124                                         goto restore_opts;
5125                                 }
5126                         enable_quota = 1;
5127                 }
5128         }
5129
5130         /*
5131          * Reinitialize lazy itable initialization thread based on
5132          * current settings
5133          */
5134         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5135                 ext4_unregister_li_request(sb);
5136         else {
5137                 ext4_group_t first_not_zeroed;
5138                 first_not_zeroed = ext4_has_uninit_itable(sb);
5139                 ext4_register_li_request(sb, first_not_zeroed);
5140         }
5141
5142         ext4_setup_system_zone(sb);
5143         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5144                 ext4_commit_super(sb, 1);
5145
5146 #ifdef CONFIG_QUOTA
5147         /* Release old quota file names */
5148         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5149                 kfree(old_opts.s_qf_names[i]);
5150         if (enable_quota) {
5151                 if (sb_any_quota_suspended(sb))
5152                         dquot_resume(sb, -1);
5153                 else if (ext4_has_feature_quota(sb)) {
5154                         err = ext4_enable_quotas(sb);
5155                         if (err)
5156                                 goto restore_opts;
5157                 }
5158         }
5159 #endif
5160
5161         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5162         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5163         kfree(orig_data);
5164         return 0;
5165
5166 restore_opts:
5167         sb->s_flags = old_sb_flags;
5168         sbi->s_mount_opt = old_opts.s_mount_opt;
5169         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5170         sbi->s_resuid = old_opts.s_resuid;
5171         sbi->s_resgid = old_opts.s_resgid;
5172         sbi->s_commit_interval = old_opts.s_commit_interval;
5173         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5174         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5175 #ifdef CONFIG_QUOTA
5176         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5177         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5178                 kfree(sbi->s_qf_names[i]);
5179                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5180         }
5181 #endif
5182         kfree(orig_data);
5183         return err;
5184 }
5185
5186 #ifdef CONFIG_QUOTA
5187 static int ext4_statfs_project(struct super_block *sb,
5188                                kprojid_t projid, struct kstatfs *buf)
5189 {
5190         struct kqid qid;
5191         struct dquot *dquot;
5192         u64 limit;
5193         u64 curblock;
5194
5195         qid = make_kqid_projid(projid);
5196         dquot = dqget(sb, qid);
5197         if (IS_ERR(dquot))
5198                 return PTR_ERR(dquot);
5199         spin_lock(&dq_data_lock);
5200
5201         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5202                  dquot->dq_dqb.dqb_bsoftlimit :
5203                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5204         if (limit && buf->f_blocks > limit) {
5205                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5206                 buf->f_blocks = limit;
5207                 buf->f_bfree = buf->f_bavail =
5208                         (buf->f_blocks > curblock) ?
5209                          (buf->f_blocks - curblock) : 0;
5210         }
5211
5212         limit = dquot->dq_dqb.dqb_isoftlimit ?
5213                 dquot->dq_dqb.dqb_isoftlimit :
5214                 dquot->dq_dqb.dqb_ihardlimit;
5215         if (limit && buf->f_files > limit) {
5216                 buf->f_files = limit;
5217                 buf->f_ffree =
5218                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5219                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5220         }
5221
5222         spin_unlock(&dq_data_lock);
5223         dqput(dquot);
5224         return 0;
5225 }
5226 #endif
5227
5228 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5229 {
5230         struct super_block *sb = dentry->d_sb;
5231         struct ext4_sb_info *sbi = EXT4_SB(sb);
5232         struct ext4_super_block *es = sbi->s_es;
5233         ext4_fsblk_t overhead = 0, resv_blocks;
5234         u64 fsid;
5235         s64 bfree;
5236         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5237
5238         if (!test_opt(sb, MINIX_DF))
5239                 overhead = sbi->s_overhead;
5240
5241         buf->f_type = EXT4_SUPER_MAGIC;
5242         buf->f_bsize = sb->s_blocksize;
5243         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5244         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5245                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5246         /* prevent underflow in case that few free space is available */
5247         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5248         buf->f_bavail = buf->f_bfree -
5249                         (ext4_r_blocks_count(es) + resv_blocks);
5250         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5251                 buf->f_bavail = 0;
5252         buf->f_files = le32_to_cpu(es->s_inodes_count);
5253         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5254         buf->f_namelen = EXT4_NAME_LEN;
5255         fsid = le64_to_cpup((void *)es->s_uuid) ^
5256                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5257         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5258         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5259
5260 #ifdef CONFIG_QUOTA
5261         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5262             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5263                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5264 #endif
5265         return 0;
5266 }
5267
5268 /* Helper function for writing quotas on sync - we need to start transaction
5269  * before quota file is locked for write. Otherwise the are possible deadlocks:
5270  * Process 1                         Process 2
5271  * ext4_create()                     quota_sync()
5272  *   jbd2_journal_start()                  write_dquot()
5273  *   dquot_initialize()                         down(dqio_mutex)
5274  *     down(dqio_mutex)                    jbd2_journal_start()
5275  *
5276  */
5277
5278 #ifdef CONFIG_QUOTA
5279
5280 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5281 {
5282         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5283 }
5284
5285 static int ext4_write_dquot(struct dquot *dquot)
5286 {
5287         int ret, err;
5288         handle_t *handle;
5289         struct inode *inode;
5290
5291         inode = dquot_to_inode(dquot);
5292         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5293                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5294         if (IS_ERR(handle))
5295                 return PTR_ERR(handle);
5296         ret = dquot_commit(dquot);
5297         err = ext4_journal_stop(handle);
5298         if (!ret)
5299                 ret = err;
5300         return ret;
5301 }
5302
5303 static int ext4_acquire_dquot(struct dquot *dquot)
5304 {
5305         int ret, err;
5306         handle_t *handle;
5307
5308         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5309                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5310         if (IS_ERR(handle))
5311                 return PTR_ERR(handle);
5312         ret = dquot_acquire(dquot);
5313         err = ext4_journal_stop(handle);
5314         if (!ret)
5315                 ret = err;
5316         return ret;
5317 }
5318
5319 static int ext4_release_dquot(struct dquot *dquot)
5320 {
5321         int ret, err;
5322         handle_t *handle;
5323
5324         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5325                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5326         if (IS_ERR(handle)) {
5327                 /* Release dquot anyway to avoid endless cycle in dqput() */
5328                 dquot_release(dquot);
5329                 return PTR_ERR(handle);
5330         }
5331         ret = dquot_release(dquot);
5332         err = ext4_journal_stop(handle);
5333         if (!ret)
5334                 ret = err;
5335         return ret;
5336 }
5337
5338 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5339 {
5340         struct super_block *sb = dquot->dq_sb;
5341         struct ext4_sb_info *sbi = EXT4_SB(sb);
5342
5343         /* Are we journaling quotas? */
5344         if (ext4_has_feature_quota(sb) ||
5345             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5346                 dquot_mark_dquot_dirty(dquot);
5347                 return ext4_write_dquot(dquot);
5348         } else {
5349                 return dquot_mark_dquot_dirty(dquot);
5350         }
5351 }
5352
5353 static int ext4_write_info(struct super_block *sb, int type)
5354 {
5355         int ret, err;
5356         handle_t *handle;
5357
5358         /* Data block + inode block */
5359         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5360         if (IS_ERR(handle))
5361                 return PTR_ERR(handle);
5362         ret = dquot_commit_info(sb, type);
5363         err = ext4_journal_stop(handle);
5364         if (!ret)
5365                 ret = err;
5366         return ret;
5367 }
5368
5369 /*
5370  * Turn on quotas during mount time - we need to find
5371  * the quota file and such...
5372  */
5373 static int ext4_quota_on_mount(struct super_block *sb, int type)
5374 {
5375         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5376                                         EXT4_SB(sb)->s_jquota_fmt, type);
5377 }
5378
5379 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5380 {
5381         struct ext4_inode_info *ei = EXT4_I(inode);
5382
5383         /* The first argument of lockdep_set_subclass has to be
5384          * *exactly* the same as the argument to init_rwsem() --- in
5385          * this case, in init_once() --- or lockdep gets unhappy
5386          * because the name of the lock is set using the
5387          * stringification of the argument to init_rwsem().
5388          */
5389         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5390         lockdep_set_subclass(&ei->i_data_sem, subclass);
5391 }
5392
5393 /*
5394  * Standard function to be called on quota_on
5395  */
5396 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5397                          const struct path *path)
5398 {
5399         int err;
5400
5401         if (!test_opt(sb, QUOTA))
5402                 return -EINVAL;
5403
5404         /* Quotafile not on the same filesystem? */
5405         if (path->dentry->d_sb != sb)
5406                 return -EXDEV;
5407         /* Journaling quota? */
5408         if (EXT4_SB(sb)->s_qf_names[type]) {
5409                 /* Quotafile not in fs root? */
5410                 if (path->dentry->d_parent != sb->s_root)
5411                         ext4_msg(sb, KERN_WARNING,
5412                                 "Quota file not on filesystem root. "
5413                                 "Journaled quota will not work");
5414         }
5415
5416         /*
5417          * When we journal data on quota file, we have to flush journal to see
5418          * all updates to the file when we bypass pagecache...
5419          */
5420         if (EXT4_SB(sb)->s_journal &&
5421             ext4_should_journal_data(d_inode(path->dentry))) {
5422                 /*
5423                  * We don't need to lock updates but journal_flush() could
5424                  * otherwise be livelocked...
5425                  */
5426                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5427                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5428                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5429                 if (err)
5430                         return err;
5431         }
5432
5433         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5434         err = dquot_quota_on(sb, type, format_id, path);
5435         if (err) {
5436                 lockdep_set_quota_inode(path->dentry->d_inode,
5437                                              I_DATA_SEM_NORMAL);
5438         } else {
5439                 struct inode *inode = d_inode(path->dentry);
5440                 handle_t *handle;
5441
5442                 /*
5443                  * Set inode flags to prevent userspace from messing with quota
5444                  * files. If this fails, we return success anyway since quotas
5445                  * are already enabled and this is not a hard failure.
5446                  */
5447                 inode_lock(inode);
5448                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5449                 if (IS_ERR(handle))
5450                         goto unlock_inode;
5451                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5452                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5453                                 S_NOATIME | S_IMMUTABLE);
5454                 ext4_mark_inode_dirty(handle, inode);
5455                 ext4_journal_stop(handle);
5456         unlock_inode:
5457                 inode_unlock(inode);
5458         }
5459         return err;
5460 }
5461
5462 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5463                              unsigned int flags)
5464 {
5465         int err;
5466         struct inode *qf_inode;
5467         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5468                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5469                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5470                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5471         };
5472
5473         BUG_ON(!ext4_has_feature_quota(sb));
5474
5475         if (!qf_inums[type])
5476                 return -EPERM;
5477
5478         qf_inode = ext4_iget(sb, qf_inums[type]);
5479         if (IS_ERR(qf_inode)) {
5480                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5481                 return PTR_ERR(qf_inode);
5482         }
5483
5484         /* Don't account quota for quota files to avoid recursion */
5485         qf_inode->i_flags |= S_NOQUOTA;
5486         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5487         err = dquot_enable(qf_inode, type, format_id, flags);
5488         iput(qf_inode);
5489         if (err)
5490                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5491
5492         return err;
5493 }
5494
5495 /* Enable usage tracking for all quota types. */
5496 static int ext4_enable_quotas(struct super_block *sb)
5497 {
5498         int type, err = 0;
5499         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5500                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5501                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5502                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5503         };
5504         bool quota_mopt[EXT4_MAXQUOTAS] = {
5505                 test_opt(sb, USRQUOTA),
5506                 test_opt(sb, GRPQUOTA),
5507                 test_opt(sb, PRJQUOTA),
5508         };
5509
5510         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5511         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5512                 if (qf_inums[type]) {
5513                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5514                                 DQUOT_USAGE_ENABLED |
5515                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5516                         if (err) {
5517                                 ext4_warning(sb,
5518                                         "Failed to enable quota tracking "
5519                                         "(type=%d, err=%d). Please run "
5520                                         "e2fsck to fix.", type, err);
5521                                 return err;
5522                         }
5523                 }
5524         }
5525         return 0;
5526 }
5527
5528 static int ext4_quota_off(struct super_block *sb, int type)
5529 {
5530         struct inode *inode = sb_dqopt(sb)->files[type];
5531         handle_t *handle;
5532         int err;
5533
5534         /* Force all delayed allocation blocks to be allocated.
5535          * Caller already holds s_umount sem */
5536         if (test_opt(sb, DELALLOC))
5537                 sync_filesystem(sb);
5538
5539         if (!inode || !igrab(inode))
5540                 goto out;
5541
5542         err = dquot_quota_off(sb, type);
5543         if (err || ext4_has_feature_quota(sb))
5544                 goto out_put;
5545
5546         inode_lock(inode);
5547         /*
5548          * Update modification times of quota files when userspace can
5549          * start looking at them. If we fail, we return success anyway since
5550          * this is not a hard failure and quotas are already disabled.
5551          */
5552         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5553         if (IS_ERR(handle))
5554                 goto out_unlock;
5555         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5556         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5557         inode->i_mtime = inode->i_ctime = current_time(inode);
5558         ext4_mark_inode_dirty(handle, inode);
5559         ext4_journal_stop(handle);
5560 out_unlock:
5561         inode_unlock(inode);
5562 out_put:
5563         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5564         iput(inode);
5565         return err;
5566 out:
5567         return dquot_quota_off(sb, type);
5568 }
5569
5570 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5571  * acquiring the locks... As quota files are never truncated and quota code
5572  * itself serializes the operations (and no one else should touch the files)
5573  * we don't have to be afraid of races */
5574 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5575                                size_t len, loff_t off)
5576 {
5577         struct inode *inode = sb_dqopt(sb)->files[type];
5578         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5579         int offset = off & (sb->s_blocksize - 1);
5580         int tocopy;
5581         size_t toread;
5582         struct buffer_head *bh;
5583         loff_t i_size = i_size_read(inode);
5584
5585         if (off > i_size)
5586                 return 0;
5587         if (off+len > i_size)
5588                 len = i_size-off;
5589         toread = len;
5590         while (toread > 0) {
5591                 tocopy = sb->s_blocksize - offset < toread ?
5592                                 sb->s_blocksize - offset : toread;
5593                 bh = ext4_bread(NULL, inode, blk, 0);
5594                 if (IS_ERR(bh))
5595                         return PTR_ERR(bh);
5596                 if (!bh)        /* A hole? */
5597                         memset(data, 0, tocopy);
5598                 else
5599                         memcpy(data, bh->b_data+offset, tocopy);
5600                 brelse(bh);
5601                 offset = 0;
5602                 toread -= tocopy;
5603                 data += tocopy;
5604                 blk++;
5605         }
5606         return len;
5607 }
5608
5609 /* Write to quotafile (we know the transaction is already started and has
5610  * enough credits) */
5611 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5612                                 const char *data, size_t len, loff_t off)
5613 {
5614         struct inode *inode = sb_dqopt(sb)->files[type];
5615         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5616         int err, offset = off & (sb->s_blocksize - 1);
5617         int retries = 0;
5618         struct buffer_head *bh;
5619         handle_t *handle = journal_current_handle();
5620
5621         if (EXT4_SB(sb)->s_journal && !handle) {
5622                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5623                         " cancelled because transaction is not started",
5624                         (unsigned long long)off, (unsigned long long)len);
5625                 return -EIO;
5626         }
5627         /*
5628          * Since we account only one data block in transaction credits,
5629          * then it is impossible to cross a block boundary.
5630          */
5631         if (sb->s_blocksize - offset < len) {
5632                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5633                         " cancelled because not block aligned",
5634                         (unsigned long long)off, (unsigned long long)len);
5635                 return -EIO;
5636         }
5637
5638         do {
5639                 bh = ext4_bread(handle, inode, blk,
5640                                 EXT4_GET_BLOCKS_CREATE |
5641                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5642         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5643                  ext4_should_retry_alloc(inode->i_sb, &retries));
5644         if (IS_ERR(bh))
5645                 return PTR_ERR(bh);
5646         if (!bh)
5647                 goto out;
5648         BUFFER_TRACE(bh, "get write access");
5649         err = ext4_journal_get_write_access(handle, bh);
5650         if (err) {
5651                 brelse(bh);
5652                 return err;
5653         }
5654         lock_buffer(bh);
5655         memcpy(bh->b_data+offset, data, len);
5656         flush_dcache_page(bh->b_page);
5657         unlock_buffer(bh);
5658         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5659         brelse(bh);
5660 out:
5661         if (inode->i_size < off + len) {
5662                 i_size_write(inode, off + len);
5663                 EXT4_I(inode)->i_disksize = inode->i_size;
5664                 ext4_mark_inode_dirty(handle, inode);
5665         }
5666         return len;
5667 }
5668
5669 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5670 {
5671         const struct quota_format_ops   *ops;
5672
5673         if (!sb_has_quota_loaded(sb, qid->type))
5674                 return -ESRCH;
5675         ops = sb_dqopt(sb)->ops[qid->type];
5676         if (!ops || !ops->get_next_id)
5677                 return -ENOSYS;
5678         return dquot_get_next_id(sb, qid);
5679 }
5680 #endif
5681
5682 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5683                        const char *dev_name, void *data)
5684 {
5685         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5686 }
5687
5688 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5689 static inline void register_as_ext2(void)
5690 {
5691         int err = register_filesystem(&ext2_fs_type);
5692         if (err)
5693                 printk(KERN_WARNING
5694                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5695 }
5696
5697 static inline void unregister_as_ext2(void)
5698 {
5699         unregister_filesystem(&ext2_fs_type);
5700 }
5701
5702 static inline int ext2_feature_set_ok(struct super_block *sb)
5703 {
5704         if (ext4_has_unknown_ext2_incompat_features(sb))
5705                 return 0;
5706         if (sb->s_flags & MS_RDONLY)
5707                 return 1;
5708         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5709                 return 0;
5710         return 1;
5711 }
5712 #else
5713 static inline void register_as_ext2(void) { }
5714 static inline void unregister_as_ext2(void) { }
5715 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5716 #endif
5717
5718 static inline void register_as_ext3(void)
5719 {
5720         int err = register_filesystem(&ext3_fs_type);
5721         if (err)
5722                 printk(KERN_WARNING
5723                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5724 }
5725
5726 static inline void unregister_as_ext3(void)
5727 {
5728         unregister_filesystem(&ext3_fs_type);
5729 }
5730
5731 static inline int ext3_feature_set_ok(struct super_block *sb)
5732 {
5733         if (ext4_has_unknown_ext3_incompat_features(sb))
5734                 return 0;
5735         if (!ext4_has_feature_journal(sb))
5736                 return 0;
5737         if (sb->s_flags & MS_RDONLY)
5738                 return 1;
5739         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5740                 return 0;
5741         return 1;
5742 }
5743
5744 static struct file_system_type ext4_fs_type = {
5745         .owner          = THIS_MODULE,
5746         .name           = "ext4",
5747         .mount          = ext4_mount,
5748         .kill_sb        = kill_block_super,
5749         .fs_flags       = FS_REQUIRES_DEV,
5750 };
5751 MODULE_ALIAS_FS("ext4");
5752
5753 /* Shared across all ext4 file systems */
5754 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5755
5756 static int __init ext4_init_fs(void)
5757 {
5758         int i, err;
5759
5760         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5761         ext4_li_info = NULL;
5762         mutex_init(&ext4_li_mtx);
5763
5764         /* Build-time check for flags consistency */
5765         ext4_check_flag_values();
5766
5767         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5768                 init_waitqueue_head(&ext4__ioend_wq[i]);
5769
5770         err = ext4_init_es();
5771         if (err)
5772                 return err;
5773
5774         err = ext4_init_pageio();
5775         if (err)
5776                 goto out5;
5777
5778         err = ext4_init_system_zone();
5779         if (err)
5780                 goto out4;
5781
5782         err = ext4_init_sysfs();
5783         if (err)
5784                 goto out3;
5785
5786         err = ext4_init_mballoc();
5787         if (err)
5788                 goto out2;
5789         err = init_inodecache();
5790         if (err)
5791                 goto out1;
5792         register_as_ext3();
5793         register_as_ext2();
5794         err = register_filesystem(&ext4_fs_type);
5795         if (err)
5796                 goto out;
5797
5798         return 0;
5799 out:
5800         unregister_as_ext2();
5801         unregister_as_ext3();
5802         destroy_inodecache();
5803 out1:
5804         ext4_exit_mballoc();
5805 out2:
5806         ext4_exit_sysfs();
5807 out3:
5808         ext4_exit_system_zone();
5809 out4:
5810         ext4_exit_pageio();
5811 out5:
5812         ext4_exit_es();
5813
5814         return err;
5815 }
5816
5817 static void __exit ext4_exit_fs(void)
5818 {
5819         ext4_destroy_lazyinit_thread();
5820         unregister_as_ext2();
5821         unregister_as_ext3();
5822         unregister_filesystem(&ext4_fs_type);
5823         destroy_inodecache();
5824         ext4_exit_mballoc();
5825         ext4_exit_sysfs();
5826         ext4_exit_system_zone();
5827         ext4_exit_pageio();
5828         ext4_exit_es();
5829 }
5830
5831 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5832 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5833 MODULE_LICENSE("GPL");
5834 module_init(ext4_init_fs)
5835 module_exit(ext4_exit_fs)