fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 static int ext4_verify_csum_type(struct super_block *sb,
 144                                  struct ext4_super_block *es)
 145 {
 146         if (!ext4_has_feature_metadata_csum(sb))
 147                 return 1;
 148
 149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 150 }
 151
 152 static __le32 ext4_superblock_csum(struct super_block *sb,
 153                                    struct ext4_super_block *es)
 154 {
 155         struct ext4_sb_info *sbi = EXT4_SB(sb);
 156         int offset = offsetof(struct ext4_super_block, s_checksum);
 157         __u32 csum;
 158
 159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 160
 161         return cpu_to_le32(csum);
 162 }
 163
 164 static int ext4_superblock_csum_verify(struct super_block *sb,
 165                                        struct ext4_super_block *es)
 166 {
 167         if (!ext4_has_metadata_csum(sb))
 168                 return 1;
 169
 170         return es->s_checksum == ext4_superblock_csum(sb, es);
 171 }
 172
 173 void ext4_superblock_csum_set(struct super_block *sb)
 174 {
 175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 176
 177         if (!ext4_has_metadata_csum(sb))
 178                 return;
 179
 180         es->s_checksum = ext4_superblock_csum(sb, es);
 181 }
 182
 183 void *ext4_kvmalloc(size_t size, gfp_t flags)
 184 {
 185         void *ret;
 186
 187         ret = kmalloc(size, flags | __GFP_NOWARN);
 188         if (!ret)
 189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 190         return ret;
 191 }
 192
 193 void *ext4_kvzalloc(size_t size, gfp_t flags)
 194 {
 195         void *ret;
 196
 197         ret = kzalloc(size, flags | __GFP_NOWARN);
 198         if (!ret)
 199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 200         return ret;
 201 }
 202
 203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 209 }
 210
 211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 212                                struct ext4_group_desc *bg)
 213 {
 214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 217 }
 218
 219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le32_to_cpu(bg->bg_inode_table_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 225 }
 226
 227 __u32 ext4_free_group_clusters(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 233 }
 234
 235 __u32 ext4_free_inodes_count(struct super_block *sb,
 236                               struct ext4_group_desc *bg)
 237 {
 238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 241 }
 242
 243 __u32 ext4_used_dirs_count(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 249 }
 250
 251 __u32 ext4_itable_unused_count(struct super_block *sb,
 252                               struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_itable_unused_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 257 }
 258
 259 void ext4_block_bitmap_set(struct super_block *sb,
 260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 261 {
 262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 265 }
 266
 267 void ext4_inode_bitmap_set(struct super_block *sb,
 268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 269 {
 270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 273 }
 274
 275 void ext4_inode_table_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 277 {
 278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 281 }
 282
 283 void ext4_free_group_clusters_set(struct super_block *sb,
 284                                   struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291 void ext4_free_inodes_set(struct super_block *sb,
 292                           struct ext4_group_desc *bg, __u32 count)
 293 {
 294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 297 }
 298
 299 void ext4_used_dirs_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, __u32 count)
 301 {
 302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 305 }
 306
 307 void ext4_itable_unused_set(struct super_block *sb,
 308                           struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315
 316 static void __save_error_info(struct super_block *sb, const char *func,
 317                             unsigned int line)
 318 {
 319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 320
 321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 322         if (bdev_read_only(sb->s_bdev))
 323                 return;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         le32_add_cpu(&es->s_error_count, 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352 /*
 353  * The del_gendisk() function uninitializes the disk-specific data
 354  * structures, including the bdi structure, without telling anyone
 355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 357  * This is a kludge to prevent these oops until we can put in a proper
 358  * hook in del_gendisk() to inform the VFS and file system layers.
 359  */
 360 static int block_device_ejected(struct super_block *sb)
 361 {
 362         struct inode *bd_inode = sb->s_bdev->bd_inode;
 363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 364
 365         return bdi->dev == NULL;
 366 }
 367
 368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 369 {
 370         struct super_block              *sb = journal->j_private;
 371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 372         int                             error = is_journal_aborted(journal);
 373         struct ext4_journal_cb_entry    *jce;
 374
 375         BUG_ON(txn->t_state == T_FINISHED);
 376         spin_lock(&sbi->s_md_lock);
 377         while (!list_empty(&txn->t_private_list)) {
 378                 jce = list_entry(txn->t_private_list.next,
 379                                  struct ext4_journal_cb_entry, jce_list);
 380                 list_del_init(&jce->jce_list);
 381                 spin_unlock(&sbi->s_md_lock);
 382                 jce->jce_func(sb, jce, error);
 383                 spin_lock(&sbi->s_md_lock);
 384         }
 385         spin_unlock(&sbi->s_md_lock);
 386 }
 387
 388 /* Deal with the reporting of failure conditions on a filesystem such as
 389  * inconsistencies detected or read IO failures.
 390  *
 391  * On ext2, we can store the error state of the filesystem in the
 392  * superblock.  That is not possible on ext4, because we may have other
 393  * write ordering constraints on the superblock which prevent us from
 394  * writing it out straight away; and given that the journal is about to
 395  * be aborted, we can't rely on the current, or future, transactions to
 396  * write out the superblock safely.
 397  *
 398  * We'll just use the jbd2_journal_abort() error code to record an error in
 399  * the journal instead.  On recovery, the journal will complain about
 400  * that error until we've noted it down and cleared it.
 401  */
 402
 403 static void ext4_handle_error(struct super_block *sb)
 404 {
 405         if (sb->s_flags & MS_RDONLY)
 406                 return;
 407
 408         if (!test_opt(sb, ERRORS_CONT)) {
 409                 journal_t *journal = EXT4_SB(sb)->s_journal;
 410
 411                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 412                 if (journal)
 413                         jbd2_journal_abort(journal, -EIO);
 414         }
 415         if (test_opt(sb, ERRORS_RO)) {
 416                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 417                 /*
 418                  * Make sure updated value of ->s_mount_flags will be visible
 419                  * before ->s_flags update
 420                  */
 421                 smp_wmb();
 422                 sb->s_flags |= MS_RDONLY;
 423         }
 424         if (test_opt(sb, ERRORS_PANIC)) {
 425                 if (EXT4_SB(sb)->s_journal &&
 426                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 427                         return;
 428                 panic("EXT4-fs (device %s): panic forced after error\n",
 429                         sb->s_id);
 430         }
 431 }
 432
 433 #define ext4_error_ratelimit(sb)                                        \
 434                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 435                              "EXT4-fs error")
 436
 437 void __ext4_error(struct super_block *sb, const char *function,
 438                   unsigned int line, const char *fmt, ...)
 439 {
 440         struct va_format vaf;
 441         va_list args;
 442
 443         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 444                 return;
 445
 446         if (ext4_error_ratelimit(sb)) {
 447                 va_start(args, fmt);
 448                 vaf.fmt = fmt;
 449                 vaf.va = &args;
 450                 printk(KERN_CRIT
 451                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 452                        sb->s_id, function, line, current->comm, &vaf);
 453                 va_end(args);
 454         }
 455         save_error_info(sb, function, line);
 456         ext4_handle_error(sb);
 457 }
 458
 459 void __ext4_error_inode(struct inode *inode, const char *function,
 460                         unsigned int line, ext4_fsblk_t block,
 461                         const char *fmt, ...)
 462 {
 463         va_list args;
 464         struct va_format vaf;
 465         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 466
 467         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 468                 return;
 469
 470         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 471         es->s_last_error_block = cpu_to_le64(block);
 472         if (ext4_error_ratelimit(inode->i_sb)) {
 473                 va_start(args, fmt);
 474                 vaf.fmt = fmt;
 475                 vaf.va = &args;
 476                 if (block)
 477                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 478                                "inode #%lu: block %llu: comm %s: %pV\n",
 479                                inode->i_sb->s_id, function, line, inode->i_ino,
 480                                block, current->comm, &vaf);
 481                 else
 482                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 483                                "inode #%lu: comm %s: %pV\n",
 484                                inode->i_sb->s_id, function, line, inode->i_ino,
 485                                current->comm, &vaf);
 486                 va_end(args);
 487         }
 488         save_error_info(inode->i_sb, function, line);
 489         ext4_handle_error(inode->i_sb);
 490 }
 491
 492 void __ext4_error_file(struct file *file, const char *function,
 493                        unsigned int line, ext4_fsblk_t block,
 494                        const char *fmt, ...)
 495 {
 496         va_list args;
 497         struct va_format vaf;
 498         struct ext4_super_block *es;
 499         struct inode *inode = file_inode(file);
 500         char pathname[80], *path;
 501
 502         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 503                 return;
 504
 505         es = EXT4_SB(inode->i_sb)->s_es;
 506         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 507         if (ext4_error_ratelimit(inode->i_sb)) {
 508                 path = file_path(file, pathname, sizeof(pathname));
 509                 if (IS_ERR(path))
 510                         path = "(unknown)";
 511                 va_start(args, fmt);
 512                 vaf.fmt = fmt;
 513                 vaf.va = &args;
 514                 if (block)
 515                         printk(KERN_CRIT
 516                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 517                                "block %llu: comm %s: path %s: %pV\n",
 518                                inode->i_sb->s_id, function, line, inode->i_ino,
 519                                block, current->comm, path, &vaf);
 520                 else
 521                         printk(KERN_CRIT
 522                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 523                                "comm %s: path %s: %pV\n",
 524                                inode->i_sb->s_id, function, line, inode->i_ino,
 525                                current->comm, path, &vaf);
 526                 va_end(args);
 527         }
 528         save_error_info(inode->i_sb, function, line);
 529         ext4_handle_error(inode->i_sb);
 530 }
 531
 532 const char *ext4_decode_error(struct super_block *sb, int errno,
 533                               char nbuf[16])
 534 {
 535         char *errstr = NULL;
 536
 537         switch (errno) {
 538         case -EFSCORRUPTED:
 539                 errstr = "Corrupt filesystem";
 540                 break;
 541         case -EFSBADCRC:
 542                 errstr = "Filesystem failed CRC";
 543                 break;
 544         case -EIO:
 545                 errstr = "IO failure";
 546                 break;
 547         case -ENOMEM:
 548                 errstr = "Out of memory";
 549                 break;
 550         case -EROFS:
 551                 if (!sb || (EXT4_SB(sb)->s_journal &&
 552                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 553                         errstr = "Journal has aborted";
 554                 else
 555                         errstr = "Readonly filesystem";
 556                 break;
 557         default:
 558                 /* If the caller passed in an extra buffer for unknown
 559                  * errors, textualise them now.  Else we just return
 560                  * NULL. */
 561                 if (nbuf) {
 562                         /* Check for truncated error codes... */
 563                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 564                                 errstr = nbuf;
 565                 }
 566                 break;
 567         }
 568
 569         return errstr;
 570 }
 571
 572 /* __ext4_std_error decodes expected errors from journaling functions
 573  * automatically and invokes the appropriate error response.  */
 574
 575 void __ext4_std_error(struct super_block *sb, const char *function,
 576                       unsigned int line, int errno)
 577 {
 578         char nbuf[16];
 579         const char *errstr;
 580
 581         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 582                 return;
 583
 584         /* Special case: if the error is EROFS, and we're not already
 585          * inside a transaction, then there's really no point in logging
 586          * an error. */
 587         if (errno == -EROFS && journal_current_handle() == NULL &&
 588             (sb->s_flags & MS_RDONLY))
 589                 return;
 590
 591         if (ext4_error_ratelimit(sb)) {
 592                 errstr = ext4_decode_error(sb, errno, nbuf);
 593                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 594                        sb->s_id, function, line, errstr);
 595         }
 596
 597         save_error_info(sb, function, line);
 598         ext4_handle_error(sb);
 599 }
 600
 601 /*
 602  * ext4_abort is a much stronger failure handler than ext4_error.  The
 603  * abort function may be used to deal with unrecoverable failures such
 604  * as journal IO errors or ENOMEM at a critical moment in log management.
 605  *
 606  * We unconditionally force the filesystem into an ABORT|READONLY state,
 607  * unless the error response on the fs has been set to panic in which
 608  * case we take the easy way out and panic immediately.
 609  */
 610
 611 void __ext4_abort(struct super_block *sb, const char *function,
 612                 unsigned int line, const char *fmt, ...)
 613 {
 614         struct va_format vaf;
 615         va_list args;
 616
 617         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 618                 return;
 619
 620         save_error_info(sb, function, line);
 621         va_start(args, fmt);
 622         vaf.fmt = fmt;
 623         vaf.va = &args;
 624         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 625                sb->s_id, function, line, &vaf);
 626         va_end(args);
 627
 628         if ((sb->s_flags & MS_RDONLY) == 0) {
 629                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 630                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 631                 /*
 632                  * Make sure updated value of ->s_mount_flags will be visible
 633                  * before ->s_flags update
 634                  */
 635                 smp_wmb();
 636                 sb->s_flags |= MS_RDONLY;
 637                 if (EXT4_SB(sb)->s_journal)
 638                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 639                 save_error_info(sb, function, line);
 640         }
 641         if (test_opt(sb, ERRORS_PANIC)) {
 642                 if (EXT4_SB(sb)->s_journal &&
 643                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 644                         return;
 645                 panic("EXT4-fs panic from previous error\n");
 646         }
 647 }
 648
 649 void __ext4_msg(struct super_block *sb,
 650                 const char *prefix, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 656                 return;
 657
 658         va_start(args, fmt);
 659         vaf.fmt = fmt;
 660         vaf.va = &args;
 661         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 662         va_end(args);
 663 }
 664
 665 #define ext4_warning_ratelimit(sb)                                      \
 666                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 667                              "EXT4-fs warning")
 668
 669 void __ext4_warning(struct super_block *sb, const char *function,
 670                     unsigned int line, const char *fmt, ...)
 671 {
 672         struct va_format vaf;
 673         va_list args;
 674
 675         if (!ext4_warning_ratelimit(sb))
 676                 return;
 677
 678         va_start(args, fmt);
 679         vaf.fmt = fmt;
 680         vaf.va = &args;
 681         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 682                sb->s_id, function, line, &vaf);
 683         va_end(args);
 684 }
 685
 686 void __ext4_warning_inode(const struct inode *inode, const char *function,
 687                           unsigned int line, const char *fmt, ...)
 688 {
 689         struct va_format vaf;
 690         va_list args;
 691
 692         if (!ext4_warning_ratelimit(inode->i_sb))
 693                 return;
 694
 695         va_start(args, fmt);
 696         vaf.fmt = fmt;
 697         vaf.va = &args;
 698         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 699                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 700                function, line, inode->i_ino, current->comm, &vaf);
 701         va_end(args);
 702 }
 703
 704 void __ext4_grp_locked_error(const char *function, unsigned int line,
 705                              struct super_block *sb, ext4_group_t grp,
 706                              unsigned long ino, ext4_fsblk_t block,
 707                              const char *fmt, ...)
 708 __releases(bitlock)
 709 __acquires(bitlock)
 710 {
 711         struct va_format vaf;
 712         va_list args;
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 716                 return;
 717
 718         es->s_last_error_ino = cpu_to_le32(ino);
 719         es->s_last_error_block = cpu_to_le64(block);
 720         __save_error_info(sb, function, line);
 721
 722         if (ext4_error_ratelimit(sb)) {
 723                 va_start(args, fmt);
 724                 vaf.fmt = fmt;
 725                 vaf.va = &args;
 726                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 727                        sb->s_id, function, line, grp);
 728                 if (ino)
 729                         printk(KERN_CONT "inode %lu: ", ino);
 730                 if (block)
 731                         printk(KERN_CONT "block %llu:",
 732                                (unsigned long long) block);
 733                 printk(KERN_CONT "%pV\n", &vaf);
 734                 va_end(args);
 735         }
 736
 737         if (test_opt(sb, ERRORS_CONT)) {
 738                 ext4_commit_super(sb, 0);
 739                 return;
 740         }
 741
 742         ext4_unlock_group(sb, grp);
 743         ext4_handle_error(sb);
 744         /*
 745          * We only get here in the ERRORS_RO case; relocking the group
 746          * may be dangerous, but nothing bad will happen since the
 747          * filesystem will have already been marked read/only and the
 748          * journal has been aborted.  We return 1 as a hint to callers
 749          * who might what to use the return value from
 750          * ext4_grp_locked_error() to distinguish between the
 751          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 752          * aggressively from the ext4 function in question, with a
 753          * more appropriate error code.
 754          */
 755         ext4_lock_group(sb, grp);
 756         return;
 757 }
 758
 759 void ext4_update_dynamic_rev(struct super_block *sb)
 760 {
 761         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 762
 763         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 764                 return;
 765
 766         ext4_warning(sb,
 767                      "updating to rev %d because of new feature flag, "
 768                      "running e2fsck is recommended",
 769                      EXT4_DYNAMIC_REV);
 770
 771         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 772         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 773         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 774         /* leave es->s_feature_*compat flags alone */
 775         /* es->s_uuid will be set by e2fsck if empty */
 776
 777         /*
 778          * The rest of the superblock fields should be zero, and if not it
 779          * means they are likely already in use, so leave them alone.  We
 780          * can leave it up to e2fsck to clean up any inconsistencies there.
 781          */
 782 }
 783
 784 /*
 785  * Open the external journal device
 786  */
 787 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 788 {
 789         struct block_device *bdev;
 790         char b[BDEVNAME_SIZE];
 791
 792         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 793         if (IS_ERR(bdev))
 794                 goto fail;
 795         return bdev;
 796
 797 fail:
 798         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 799                         __bdevname(dev, b), PTR_ERR(bdev));
 800         return NULL;
 801 }
 802
 803 /*
 804  * Release the journal device
 805  */
 806 static void ext4_blkdev_put(struct block_device *bdev)
 807 {
 808         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 809 }
 810
 811 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 812 {
 813         struct block_device *bdev;
 814         bdev = sbi->journal_bdev;
 815         if (bdev) {
 816                 ext4_blkdev_put(bdev);
 817                 sbi->journal_bdev = NULL;
 818         }
 819 }
 820
 821 static inline struct inode *orphan_list_entry(struct list_head *l)
 822 {
 823         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 824 }
 825
 826 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 827 {
 828         struct list_head *l;
 829
 830         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 831                  le32_to_cpu(sbi->s_es->s_last_orphan));
 832
 833         printk(KERN_ERR "sb_info orphan list:\n");
 834         list_for_each(l, &sbi->s_orphan) {
 835                 struct inode *inode = orphan_list_entry(l);
 836                 printk(KERN_ERR "  "
 837                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 838                        inode->i_sb->s_id, inode->i_ino, inode,
 839                        inode->i_mode, inode->i_nlink,
 840                        NEXT_ORPHAN(inode));
 841         }
 842 }
 843
 844 #ifdef CONFIG_QUOTA
 845 static int ext4_quota_off(struct super_block *sb, int type);
 846
 847 static inline void ext4_quota_off_umount(struct super_block *sb)
 848 {
 849         int type;
 850
 851         /* Use our quota_off function to clear inode flags etc. */
 852         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 853                 ext4_quota_off(sb, type);
 854 }
 855 #else
 856 static inline void ext4_quota_off_umount(struct super_block *sb)
 857 {
 858 }
 859 #endif
 860
 861 static void ext4_put_super(struct super_block *sb)
 862 {
 863         struct ext4_sb_info *sbi = EXT4_SB(sb);
 864         struct ext4_super_block *es = sbi->s_es;
 865         int aborted = 0;
 866         int i, err;
 867
 868         ext4_unregister_li_request(sb);
 869         ext4_quota_off_umount(sb);
 870
 871         flush_workqueue(sbi->rsv_conversion_wq);
 872         destroy_workqueue(sbi->rsv_conversion_wq);
 873
 874         if (sbi->s_journal) {
 875                 aborted = is_journal_aborted(sbi->s_journal);
 876                 err = jbd2_journal_destroy(sbi->s_journal);
 877                 sbi->s_journal = NULL;
 878                 if ((err < 0) && !aborted)
 879                         ext4_abort(sb, "Couldn't clean up the journal");
 880         }
 881
 882         ext4_unregister_sysfs(sb);
 883         ext4_es_unregister_shrinker(sbi);
 884         del_timer_sync(&sbi->s_err_report);
 885         ext4_release_system_zone(sb);
 886         ext4_mb_release(sb);
 887         ext4_ext_release(sb);
 888
 889         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 890                 ext4_clear_feature_journal_needs_recovery(sb);
 891                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 892         }
 893         if (!(sb->s_flags & MS_RDONLY))
 894                 ext4_commit_super(sb, 1);
 895
 896         for (i = 0; i < sbi->s_gdb_count; i++)
 897                 brelse(sbi->s_group_desc[i]);
 898         kvfree(sbi->s_group_desc);
 899         kvfree(sbi->s_flex_groups);
 900         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 901         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 902         percpu_counter_destroy(&sbi->s_dirs_counter);
 903         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 904         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 905 #ifdef CONFIG_QUOTA
 906         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 907                 kfree(sbi->s_qf_names[i]);
 908 #endif
 909
 910         /* Debugging code just in case the in-memory inode orphan list
 911          * isn't empty.  The on-disk one can be non-empty if we've
 912          * detected an error and taken the fs readonly, but the
 913          * in-memory list had better be clean by this point. */
 914         if (!list_empty(&sbi->s_orphan))
 915                 dump_orphan_list(sb, sbi);
 916         J_ASSERT(list_empty(&sbi->s_orphan));
 917
 918         sync_blockdev(sb->s_bdev);
 919         invalidate_bdev(sb->s_bdev);
 920         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 921                 /*
 922                  * Invalidate the journal device's buffers.  We don't want them
 923                  * floating about in memory - the physical journal device may
 924                  * hotswapped, and it breaks the `ro-after' testing code.
 925                  */
 926                 sync_blockdev(sbi->journal_bdev);
 927                 invalidate_bdev(sbi->journal_bdev);
 928                 ext4_blkdev_remove(sbi);
 929         }
 930         if (sbi->s_mb_cache) {
 931                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 932                 sbi->s_mb_cache = NULL;
 933         }
 934         if (sbi->s_mmp_tsk)
 935                 kthread_stop(sbi->s_mmp_tsk);
 936         brelse(sbi->s_sbh);
 937         sb->s_fs_info = NULL;
 938         /*
 939          * Now that we are completely done shutting down the
 940          * superblock, we need to actually destroy the kobject.
 941          */
 942         kobject_put(&sbi->s_kobj);
 943         wait_for_completion(&sbi->s_kobj_unregister);
 944         if (sbi->s_chksum_driver)
 945                 crypto_free_shash(sbi->s_chksum_driver);
 946         kfree(sbi->s_blockgroup_lock);
 947         kfree(sbi);
 948 }
 949
 950 static struct kmem_cache *ext4_inode_cachep;
 951
 952 /*
 953  * Called inside transaction, so use GFP_NOFS
 954  */
 955 static struct inode *ext4_alloc_inode(struct super_block *sb)
 956 {
 957         struct ext4_inode_info *ei;
 958
 959         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 960         if (!ei)
 961                 return NULL;
 962
 963         ei->vfs_inode.i_version = 1;
 964         spin_lock_init(&ei->i_raw_lock);
 965         INIT_LIST_HEAD(&ei->i_prealloc_list);
 966         spin_lock_init(&ei->i_prealloc_lock);
 967         ext4_es_init_tree(&ei->i_es_tree);
 968         rwlock_init(&ei->i_es_lock);
 969         INIT_LIST_HEAD(&ei->i_es_list);
 970         ei->i_es_all_nr = 0;
 971         ei->i_es_shk_nr = 0;
 972         ei->i_es_shrink_lblk = 0;
 973         ei->i_reserved_data_blocks = 0;
 974         ei->i_reserved_meta_blocks = 0;
 975         ei->i_allocated_meta_blocks = 0;
 976         ei->i_da_metadata_calc_len = 0;
 977         ei->i_da_metadata_calc_last_lblock = 0;
 978         spin_lock_init(&(ei->i_block_reservation_lock));
 979 #ifdef CONFIG_QUOTA
 980         ei->i_reserved_quota = 0;
 981         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 982 #endif
 983         ei->jinode = NULL;
 984         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 985         spin_lock_init(&ei->i_completed_io_lock);
 986         ei->i_sync_tid = 0;
 987         ei->i_datasync_tid = 0;
 988         atomic_set(&ei->i_unwritten, 0);
 989         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 990         return &ei->vfs_inode;
 991 }
 992
 993 static int ext4_drop_inode(struct inode *inode)
 994 {
 995         int drop = generic_drop_inode(inode);
 996
 997         trace_ext4_drop_inode(inode, drop);
 998         return drop;
 999 }
1000
1001 static void ext4_i_callback(struct rcu_head *head)
1002 {
1003         struct inode *inode = container_of(head, struct inode, i_rcu);
1004         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1005 }
1006
1007 static void ext4_destroy_inode(struct inode *inode)
1008 {
1009         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1010                 ext4_msg(inode->i_sb, KERN_ERR,
1011                          "Inode %lu (%p): orphan list check failed!",
1012                          inode->i_ino, EXT4_I(inode));
1013                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1014                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1015                                 true);
1016                 dump_stack();
1017         }
1018         call_rcu(&inode->i_rcu, ext4_i_callback);
1019 }
1020
1021 static void init_once(void *foo)
1022 {
1023         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1024
1025         INIT_LIST_HEAD(&ei->i_orphan);
1026         init_rwsem(&ei->xattr_sem);
1027         init_rwsem(&ei->i_data_sem);
1028         init_rwsem(&ei->i_mmap_sem);
1029         inode_init_once(&ei->vfs_inode);
1030 }
1031
1032 static int __init init_inodecache(void)
1033 {
1034         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1035                                              sizeof(struct ext4_inode_info),
1036                                              0, (SLAB_RECLAIM_ACCOUNT|
1037                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1038                                              init_once);
1039         if (ext4_inode_cachep == NULL)
1040                 return -ENOMEM;
1041         return 0;
1042 }
1043
1044 static void destroy_inodecache(void)
1045 {
1046         /*
1047          * Make sure all delayed rcu free inodes are flushed before we
1048          * destroy cache.
1049          */
1050         rcu_barrier();
1051         kmem_cache_destroy(ext4_inode_cachep);
1052 }
1053
1054 void ext4_clear_inode(struct inode *inode)
1055 {
1056         invalidate_inode_buffers(inode);
1057         clear_inode(inode);
1058         dquot_drop(inode);
1059         ext4_discard_preallocations(inode);
1060         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1061         if (EXT4_I(inode)->jinode) {
1062                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1063                                                EXT4_I(inode)->jinode);
1064                 jbd2_free_inode(EXT4_I(inode)->jinode);
1065                 EXT4_I(inode)->jinode = NULL;
1066         }
1067 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1068         fscrypt_put_encryption_info(inode, NULL);
1069 #endif
1070 }
1071
1072 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1073                                         u64 ino, u32 generation)
1074 {
1075         struct inode *inode;
1076
1077         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1078                 return ERR_PTR(-ESTALE);
1079         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1080                 return ERR_PTR(-ESTALE);
1081
1082         /* iget isn't really right if the inode is currently unallocated!!
1083          *
1084          * ext4_read_inode will return a bad_inode if the inode had been
1085          * deleted, so we should be safe.
1086          *
1087          * Currently we don't know the generation for parent directory, so
1088          * a generation of 0 means "accept any"
1089          */
1090         inode = ext4_iget_normal(sb, ino);
1091         if (IS_ERR(inode))
1092                 return ERR_CAST(inode);
1093         if (generation && inode->i_generation != generation) {
1094                 iput(inode);
1095                 return ERR_PTR(-ESTALE);
1096         }
1097
1098         return inode;
1099 }
1100
1101 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1102                                         int fh_len, int fh_type)
1103 {
1104         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1105                                     ext4_nfs_get_inode);
1106 }
1107
1108 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1109                                         int fh_len, int fh_type)
1110 {
1111         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1112                                     ext4_nfs_get_inode);
1113 }
1114
1115 /*
1116  * Try to release metadata pages (indirect blocks, directories) which are
1117  * mapped via the block device.  Since these pages could have journal heads
1118  * which would prevent try_to_free_buffers() from freeing them, we must use
1119  * jbd2 layer's try_to_free_buffers() function to release them.
1120  */
1121 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1122                                  gfp_t wait)
1123 {
1124         journal_t *journal = EXT4_SB(sb)->s_journal;
1125
1126         WARN_ON(PageChecked(page));
1127         if (!page_has_buffers(page))
1128                 return 0;
1129         if (journal)
1130                 return jbd2_journal_try_to_free_buffers(journal, page,
1131                                                 wait & ~__GFP_DIRECT_RECLAIM);
1132         return try_to_free_buffers(page);
1133 }
1134
1135 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1136 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1137 {
1138         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1139                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1140 }
1141
1142 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1143                                                         void *fs_data)
1144 {
1145         handle_t *handle = fs_data;
1146         int res, res2, retries = 0;
1147
1148         res = ext4_convert_inline_data(inode);
1149         if (res)
1150                 return res;
1151
1152         /*
1153          * If a journal handle was specified, then the encryption context is
1154          * being set on a new inode via inheritance and is part of a larger
1155          * transaction to create the inode.  Otherwise the encryption context is
1156          * being set on an existing inode in its own transaction.  Only in the
1157          * latter case should the "retry on ENOSPC" logic be used.
1158          */
1159
1160         if (handle) {
1161                 res = ext4_xattr_set_handle(handle, inode,
1162                                             EXT4_XATTR_INDEX_ENCRYPTION,
1163                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1164                                             ctx, len, 0);
1165                 if (!res) {
1166                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1167                         ext4_clear_inode_state(inode,
1168                                         EXT4_STATE_MAY_INLINE_DATA);
1169                         /*
1170                          * Update inode->i_flags - e.g. S_DAX may get disabled
1171                          */
1172                         ext4_set_inode_flags(inode);
1173                 }
1174                 return res;
1175         }
1176
1177         res = dquot_initialize(inode);
1178         if (res)
1179                 return res;
1180 retry:
1181         handle = ext4_journal_start(inode, EXT4_HT_MISC,
1182                         ext4_jbd2_credits_xattr(inode));
1183         if (IS_ERR(handle))
1184                 return PTR_ERR(handle);
1185
1186         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1187                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1188                                     ctx, len, 0);
1189         if (!res) {
1190                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1191                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1192                 ext4_set_inode_flags(inode);
1193                 res = ext4_mark_inode_dirty(handle, inode);
1194                 if (res)
1195                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1196         }
1197         res2 = ext4_journal_stop(handle);
1198
1199         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1200                 goto retry;
1201         if (!res)
1202                 res = res2;
1203         return res;
1204 }
1205
1206 static int ext4_dummy_context(struct inode *inode)
1207 {
1208         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1209 }
1210
1211 static unsigned ext4_max_namelen(struct inode *inode)
1212 {
1213         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1214                 EXT4_NAME_LEN;
1215 }
1216
1217 static const struct fscrypt_operations ext4_cryptops = {
1218         .key_prefix             = "ext4:",
1219         .get_context            = ext4_get_context,
1220         .set_context            = ext4_set_context,
1221         .dummy_context          = ext4_dummy_context,
1222         .is_encrypted           = ext4_encrypted_inode,
1223         .empty_dir              = ext4_empty_dir,
1224         .max_namelen            = ext4_max_namelen,
1225 };
1226 #else
1227 static const struct fscrypt_operations ext4_cryptops = {
1228         .is_encrypted           = ext4_encrypted_inode,
1229 };
1230 #endif
1231
1232 #ifdef CONFIG_QUOTA
1233 static const char * const quotatypes[] = INITQFNAMES;
1234 #define QTYPE2NAME(t) (quotatypes[t])
1235
1236 static int ext4_write_dquot(struct dquot *dquot);
1237 static int ext4_acquire_dquot(struct dquot *dquot);
1238 static int ext4_release_dquot(struct dquot *dquot);
1239 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1240 static int ext4_write_info(struct super_block *sb, int type);
1241 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1242                          const struct path *path);
1243 static int ext4_quota_on_mount(struct super_block *sb, int type);
1244 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1245                                size_t len, loff_t off);
1246 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1247                                 const char *data, size_t len, loff_t off);
1248 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1249                              unsigned int flags);
1250 static int ext4_enable_quotas(struct super_block *sb);
1251 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1252
1253 static struct dquot **ext4_get_dquots(struct inode *inode)
1254 {
1255         return EXT4_I(inode)->i_dquot;
1256 }
1257
1258 static const struct dquot_operations ext4_quota_operations = {
1259         .get_reserved_space = ext4_get_reserved_space,
1260         .write_dquot    = ext4_write_dquot,
1261         .acquire_dquot  = ext4_acquire_dquot,
1262         .release_dquot  = ext4_release_dquot,
1263         .mark_dirty     = ext4_mark_dquot_dirty,
1264         .write_info     = ext4_write_info,
1265         .alloc_dquot    = dquot_alloc,
1266         .destroy_dquot  = dquot_destroy,
1267         .get_projid     = ext4_get_projid,
1268         .get_next_id    = ext4_get_next_id,
1269 };
1270
1271 static const struct quotactl_ops ext4_qctl_operations = {
1272         .quota_on       = ext4_quota_on,
1273         .quota_off      = ext4_quota_off,
1274         .quota_sync     = dquot_quota_sync,
1275         .get_state      = dquot_get_state,
1276         .set_info       = dquot_set_dqinfo,
1277         .get_dqblk      = dquot_get_dqblk,
1278         .set_dqblk      = dquot_set_dqblk,
1279         .get_nextdqblk  = dquot_get_next_dqblk,
1280 };
1281 #endif
1282
1283 static const struct super_operations ext4_sops = {
1284         .alloc_inode    = ext4_alloc_inode,
1285         .destroy_inode  = ext4_destroy_inode,
1286         .write_inode    = ext4_write_inode,
1287         .dirty_inode    = ext4_dirty_inode,
1288         .drop_inode     = ext4_drop_inode,
1289         .evict_inode    = ext4_evict_inode,
1290         .put_super      = ext4_put_super,
1291         .sync_fs        = ext4_sync_fs,
1292         .freeze_fs      = ext4_freeze,
1293         .unfreeze_fs    = ext4_unfreeze,
1294         .statfs         = ext4_statfs,
1295         .remount_fs     = ext4_remount,
1296         .show_options   = ext4_show_options,
1297 #ifdef CONFIG_QUOTA
1298         .quota_read     = ext4_quota_read,
1299         .quota_write    = ext4_quota_write,
1300         .get_dquots     = ext4_get_dquots,
1301 #endif
1302         .bdev_try_to_free_page = bdev_try_to_free_page,
1303 };
1304
1305 static const struct export_operations ext4_export_ops = {
1306         .fh_to_dentry = ext4_fh_to_dentry,
1307         .fh_to_parent = ext4_fh_to_parent,
1308         .get_parent = ext4_get_parent,
1309 };
1310
1311 enum {
1312         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1313         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1314         Opt_nouid32, Opt_debug, Opt_removed,
1315         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1316         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1317         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1318         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1319         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1320         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1321         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1322         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1323         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1324         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1325         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1326         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1327         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1328         Opt_inode_readahead_blks, Opt_journal_ioprio,
1329         Opt_dioread_nolock, Opt_dioread_lock,
1330         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1331         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1332 };
1333
1334 static const match_table_t tokens = {
1335         {Opt_bsd_df, "bsddf"},
1336         {Opt_minix_df, "minixdf"},
1337         {Opt_grpid, "grpid"},
1338         {Opt_grpid, "bsdgroups"},
1339         {Opt_nogrpid, "nogrpid"},
1340         {Opt_nogrpid, "sysvgroups"},
1341         {Opt_resgid, "resgid=%u"},
1342         {Opt_resuid, "resuid=%u"},
1343         {Opt_sb, "sb=%u"},
1344         {Opt_err_cont, "errors=continue"},
1345         {Opt_err_panic, "errors=panic"},
1346         {Opt_err_ro, "errors=remount-ro"},
1347         {Opt_nouid32, "nouid32"},
1348         {Opt_debug, "debug"},
1349         {Opt_removed, "oldalloc"},
1350         {Opt_removed, "orlov"},
1351         {Opt_user_xattr, "user_xattr"},
1352         {Opt_nouser_xattr, "nouser_xattr"},
1353         {Opt_acl, "acl"},
1354         {Opt_noacl, "noacl"},
1355         {Opt_noload, "norecovery"},
1356         {Opt_noload, "noload"},
1357         {Opt_removed, "nobh"},
1358         {Opt_removed, "bh"},
1359         {Opt_commit, "commit=%u"},
1360         {Opt_min_batch_time, "min_batch_time=%u"},
1361         {Opt_max_batch_time, "max_batch_time=%u"},
1362         {Opt_journal_dev, "journal_dev=%u"},
1363         {Opt_journal_path, "journal_path=%s"},
1364         {Opt_journal_checksum, "journal_checksum"},
1365         {Opt_nojournal_checksum, "nojournal_checksum"},
1366         {Opt_journal_async_commit, "journal_async_commit"},
1367         {Opt_abort, "abort"},
1368         {Opt_data_journal, "data=journal"},
1369         {Opt_data_ordered, "data=ordered"},
1370         {Opt_data_writeback, "data=writeback"},
1371         {Opt_data_err_abort, "data_err=abort"},
1372         {Opt_data_err_ignore, "data_err=ignore"},
1373         {Opt_offusrjquota, "usrjquota="},
1374         {Opt_usrjquota, "usrjquota=%s"},
1375         {Opt_offgrpjquota, "grpjquota="},
1376         {Opt_grpjquota, "grpjquota=%s"},
1377         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1378         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1379         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1380         {Opt_grpquota, "grpquota"},
1381         {Opt_noquota, "noquota"},
1382         {Opt_quota, "quota"},
1383         {Opt_usrquota, "usrquota"},
1384         {Opt_prjquota, "prjquota"},
1385         {Opt_barrier, "barrier=%u"},
1386         {Opt_barrier, "barrier"},
1387         {Opt_nobarrier, "nobarrier"},
1388         {Opt_i_version, "i_version"},
1389         {Opt_dax, "dax"},
1390         {Opt_stripe, "stripe=%u"},
1391         {Opt_delalloc, "delalloc"},
1392         {Opt_lazytime, "lazytime"},
1393         {Opt_nolazytime, "nolazytime"},
1394         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1395         {Opt_nodelalloc, "nodelalloc"},
1396         {Opt_removed, "mblk_io_submit"},
1397         {Opt_removed, "nomblk_io_submit"},
1398         {Opt_block_validity, "block_validity"},
1399         {Opt_noblock_validity, "noblock_validity"},
1400         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1401         {Opt_journal_ioprio, "journal_ioprio=%u"},
1402         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1403         {Opt_auto_da_alloc, "auto_da_alloc"},
1404         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1405         {Opt_dioread_nolock, "dioread_nolock"},
1406         {Opt_dioread_lock, "dioread_lock"},
1407         {Opt_discard, "discard"},
1408         {Opt_nodiscard, "nodiscard"},
1409         {Opt_init_itable, "init_itable=%u"},
1410         {Opt_init_itable, "init_itable"},
1411         {Opt_noinit_itable, "noinit_itable"},
1412         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1413         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1414         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1415         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1416         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1417         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1418         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1419         {Opt_err, NULL},
1420 };
1421
1422 static ext4_fsblk_t get_sb_block(void **data)
1423 {
1424         ext4_fsblk_t    sb_block;
1425         char            *options = (char *) *data;
1426
1427         if (!options || strncmp(options, "sb=", 3) != 0)
1428                 return 1;       /* Default location */
1429
1430         options += 3;
1431         /* TODO: use simple_strtoll with >32bit ext4 */
1432         sb_block = simple_strtoul(options, &options, 0);
1433         if (*options && *options != ',') {
1434                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1435                        (char *) *data);
1436                 return 1;
1437         }
1438         if (*options == ',')
1439                 options++;
1440         *data = (void *) options;
1441
1442         return sb_block;
1443 }
1444
1445 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1446 static const char deprecated_msg[] =
1447         "Mount option \"%s\" will be removed by %s\n"
1448         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1449
1450 #ifdef CONFIG_QUOTA
1451 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1452 {
1453         struct ext4_sb_info *sbi = EXT4_SB(sb);
1454         char *qname;
1455         int ret = -1;
1456
1457         if (sb_any_quota_loaded(sb) &&
1458                 !sbi->s_qf_names[qtype]) {
1459                 ext4_msg(sb, KERN_ERR,
1460                         "Cannot change journaled "
1461                         "quota options when quota turned on");
1462                 return -1;
1463         }
1464         if (ext4_has_feature_quota(sb)) {
1465                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1466                          "ignored when QUOTA feature is enabled");
1467                 return 1;
1468         }
1469         qname = match_strdup(args);
1470         if (!qname) {
1471                 ext4_msg(sb, KERN_ERR,
1472                         "Not enough memory for storing quotafile name");
1473                 return -1;
1474         }
1475         if (sbi->s_qf_names[qtype]) {
1476                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1477                         ret = 1;
1478                 else
1479                         ext4_msg(sb, KERN_ERR,
1480                                  "%s quota file already specified",
1481                                  QTYPE2NAME(qtype));
1482                 goto errout;
1483         }
1484         if (strchr(qname, '/')) {
1485                 ext4_msg(sb, KERN_ERR,
1486                         "quotafile must be on filesystem root");
1487                 goto errout;
1488         }
1489         sbi->s_qf_names[qtype] = qname;
1490         set_opt(sb, QUOTA);
1491         return 1;
1492 errout:
1493         kfree(qname);
1494         return ret;
1495 }
1496
1497 static int clear_qf_name(struct super_block *sb, int qtype)
1498 {
1499
1500         struct ext4_sb_info *sbi = EXT4_SB(sb);
1501
1502         if (sb_any_quota_loaded(sb) &&
1503                 sbi->s_qf_names[qtype]) {
1504                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1505                         " when quota turned on");
1506                 return -1;
1507         }
1508         kfree(sbi->s_qf_names[qtype]);
1509         sbi->s_qf_names[qtype] = NULL;
1510         return 1;
1511 }
1512 #endif
1513
1514 #define MOPT_SET        0x0001
1515 #define MOPT_CLEAR      0x0002
1516 #define MOPT_NOSUPPORT  0x0004
1517 #define MOPT_EXPLICIT   0x0008
1518 #define MOPT_CLEAR_ERR  0x0010
1519 #define MOPT_GTE0       0x0020
1520 #ifdef CONFIG_QUOTA
1521 #define MOPT_Q          0
1522 #define MOPT_QFMT       0x0040
1523 #else
1524 #define MOPT_Q          MOPT_NOSUPPORT
1525 #define MOPT_QFMT       MOPT_NOSUPPORT
1526 #endif
1527 #define MOPT_DATAJ      0x0080
1528 #define MOPT_NO_EXT2    0x0100
1529 #define MOPT_NO_EXT3    0x0200
1530 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1531 #define MOPT_STRING     0x0400
1532
1533 static const struct mount_opts {
1534         int     token;
1535         int     mount_opt;
1536         int     flags;
1537 } ext4_mount_opts[] = {
1538         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1539         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1540         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1541         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1542         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1543         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1544         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1545          MOPT_EXT4_ONLY | MOPT_SET},
1546         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1547          MOPT_EXT4_ONLY | MOPT_CLEAR},
1548         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1549         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1550         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1551          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1552         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1553          MOPT_EXT4_ONLY | MOPT_CLEAR},
1554         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1555          MOPT_EXT4_ONLY | MOPT_CLEAR},
1556         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1557          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1558         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1559                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1560          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1561         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1562         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1563         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1564         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1565         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1566          MOPT_NO_EXT2},
1567         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1568          MOPT_NO_EXT2},
1569         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1570         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1571         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1572         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1573         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1574         {Opt_commit, 0, MOPT_GTE0},
1575         {Opt_max_batch_time, 0, MOPT_GTE0},
1576         {Opt_min_batch_time, 0, MOPT_GTE0},
1577         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1578         {Opt_init_itable, 0, MOPT_GTE0},
1579         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1580         {Opt_stripe, 0, MOPT_GTE0},
1581         {Opt_resuid, 0, MOPT_GTE0},
1582         {Opt_resgid, 0, MOPT_GTE0},
1583         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1584         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1585         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1586         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1587         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1588         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1589          MOPT_NO_EXT2 | MOPT_DATAJ},
1590         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1591         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1592 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1593         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1594         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1595 #else
1596         {Opt_acl, 0, MOPT_NOSUPPORT},
1597         {Opt_noacl, 0, MOPT_NOSUPPORT},
1598 #endif
1599         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1600         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1601         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1602         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1603         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1604                                                         MOPT_SET | MOPT_Q},
1605         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1606                                                         MOPT_SET | MOPT_Q},
1607         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1608                                                         MOPT_SET | MOPT_Q},
1609         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1610                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1611                                                         MOPT_CLEAR | MOPT_Q},
1612         {Opt_usrjquota, 0, MOPT_Q},
1613         {Opt_grpjquota, 0, MOPT_Q},
1614         {Opt_offusrjquota, 0, MOPT_Q},
1615         {Opt_offgrpjquota, 0, MOPT_Q},
1616         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1617         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1618         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1619         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1620         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1621         {Opt_err, 0, 0}
1622 };
1623
1624 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1625                             substring_t *args, unsigned long *journal_devnum,
1626                             unsigned int *journal_ioprio, int is_remount)
1627 {
1628         struct ext4_sb_info *sbi = EXT4_SB(sb);
1629         const struct mount_opts *m;
1630         kuid_t uid;
1631         kgid_t gid;
1632         int arg = 0;
1633
1634 #ifdef CONFIG_QUOTA
1635         if (token == Opt_usrjquota)
1636                 return set_qf_name(sb, USRQUOTA, &args[0]);
1637         else if (token == Opt_grpjquota)
1638                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1639         else if (token == Opt_offusrjquota)
1640                 return clear_qf_name(sb, USRQUOTA);
1641         else if (token == Opt_offgrpjquota)
1642                 return clear_qf_name(sb, GRPQUOTA);
1643 #endif
1644         switch (token) {
1645         case Opt_noacl:
1646         case Opt_nouser_xattr:
1647                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1648                 break;
1649         case Opt_sb:
1650                 return 1;       /* handled by get_sb_block() */
1651         case Opt_removed:
1652                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1653                 return 1;
1654         case Opt_abort:
1655                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1656                 return 1;
1657         case Opt_i_version:
1658                 sb->s_flags |= MS_I_VERSION;
1659                 return 1;
1660         case Opt_lazytime:
1661                 sb->s_flags |= MS_LAZYTIME;
1662                 return 1;
1663         case Opt_nolazytime:
1664                 sb->s_flags &= ~MS_LAZYTIME;
1665                 return 1;
1666         }
1667
1668         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1669                 if (token == m->token)
1670                         break;
1671
1672         if (m->token == Opt_err) {
1673                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1674                          "or missing value", opt);
1675                 return -1;
1676         }
1677
1678         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1679                 ext4_msg(sb, KERN_ERR,
1680                          "Mount option \"%s\" incompatible with ext2", opt);
1681                 return -1;
1682         }
1683         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1684                 ext4_msg(sb, KERN_ERR,
1685                          "Mount option \"%s\" incompatible with ext3", opt);
1686                 return -1;
1687         }
1688
1689         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1690                 return -1;
1691         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1692                 return -1;
1693         if (m->flags & MOPT_EXPLICIT) {
1694                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1695                         set_opt2(sb, EXPLICIT_DELALLOC);
1696                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1697                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1698                 } else
1699                         return -1;
1700         }
1701         if (m->flags & MOPT_CLEAR_ERR)
1702                 clear_opt(sb, ERRORS_MASK);
1703         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1704                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1705                          "options when quota turned on");
1706                 return -1;
1707         }
1708
1709         if (m->flags & MOPT_NOSUPPORT) {
1710                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1711         } else if (token == Opt_commit) {
1712                 if (arg == 0)
1713                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1714                 sbi->s_commit_interval = HZ * arg;
1715         } else if (token == Opt_debug_want_extra_isize) {
1716                 sbi->s_want_extra_isize = arg;
1717         } else if (token == Opt_max_batch_time) {
1718                 sbi->s_max_batch_time = arg;
1719         } else if (token == Opt_min_batch_time) {
1720                 sbi->s_min_batch_time = arg;
1721         } else if (token == Opt_inode_readahead_blks) {
1722                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1723                         ext4_msg(sb, KERN_ERR,
1724                                  "EXT4-fs: inode_readahead_blks must be "
1725                                  "0 or a power of 2 smaller than 2^31");
1726                         return -1;
1727                 }
1728                 sbi->s_inode_readahead_blks = arg;
1729         } else if (token == Opt_init_itable) {
1730                 set_opt(sb, INIT_INODE_TABLE);
1731                 if (!args->from)
1732                         arg = EXT4_DEF_LI_WAIT_MULT;
1733                 sbi->s_li_wait_mult = arg;
1734         } else if (token == Opt_max_dir_size_kb) {
1735                 sbi->s_max_dir_size_kb = arg;
1736         } else if (token == Opt_stripe) {
1737                 sbi->s_stripe = arg;
1738         } else if (token == Opt_resuid) {
1739                 uid = make_kuid(current_user_ns(), arg);
1740                 if (!uid_valid(uid)) {
1741                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1742                         return -1;
1743                 }
1744                 sbi->s_resuid = uid;
1745         } else if (token == Opt_resgid) {
1746                 gid = make_kgid(current_user_ns(), arg);
1747                 if (!gid_valid(gid)) {
1748                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1749                         return -1;
1750                 }
1751                 sbi->s_resgid = gid;
1752         } else if (token == Opt_journal_dev) {
1753                 if (is_remount) {
1754                         ext4_msg(sb, KERN_ERR,
1755                                  "Cannot specify journal on remount");
1756                         return -1;
1757                 }
1758                 *journal_devnum = arg;
1759         } else if (token == Opt_journal_path) {
1760                 char *journal_path;
1761                 struct inode *journal_inode;
1762                 struct path path;
1763                 int error;
1764
1765                 if (is_remount) {
1766                         ext4_msg(sb, KERN_ERR,
1767                                  "Cannot specify journal on remount");
1768                         return -1;
1769                 }
1770                 journal_path = match_strdup(&args[0]);
1771                 if (!journal_path) {
1772                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1773                                 "journal device string");
1774                         return -1;
1775                 }
1776
1777                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1778                 if (error) {
1779                         ext4_msg(sb, KERN_ERR, "error: could not find "
1780                                 "journal device path: error %d", error);
1781                         kfree(journal_path);
1782                         return -1;
1783                 }
1784
1785                 journal_inode = d_inode(path.dentry);
1786                 if (!S_ISBLK(journal_inode->i_mode)) {
1787                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1788                                 "is not a block device", journal_path);
1789                         path_put(&path);
1790                         kfree(journal_path);
1791                         return -1;
1792                 }
1793
1794                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1795                 path_put(&path);
1796                 kfree(journal_path);
1797         } else if (token == Opt_journal_ioprio) {
1798                 if (arg > 7) {
1799                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1800                                  " (must be 0-7)");
1801                         return -1;
1802                 }
1803                 *journal_ioprio =
1804                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1805         } else if (token == Opt_test_dummy_encryption) {
1806 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1807                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1808                 ext4_msg(sb, KERN_WARNING,
1809                          "Test dummy encryption mode enabled");
1810 #else
1811                 ext4_msg(sb, KERN_WARNING,
1812                          "Test dummy encryption mount option ignored");
1813 #endif
1814         } else if (m->flags & MOPT_DATAJ) {
1815                 if (is_remount) {
1816                         if (!sbi->s_journal)
1817                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1818                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1819                                 ext4_msg(sb, KERN_ERR,
1820                                          "Cannot change data mode on remount");
1821                                 return -1;
1822                         }
1823                 } else {
1824                         clear_opt(sb, DATA_FLAGS);
1825                         sbi->s_mount_opt |= m->mount_opt;
1826                 }
1827 #ifdef CONFIG_QUOTA
1828         } else if (m->flags & MOPT_QFMT) {
1829                 if (sb_any_quota_loaded(sb) &&
1830                     sbi->s_jquota_fmt != m->mount_opt) {
1831                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1832                                  "quota options when quota turned on");
1833                         return -1;
1834                 }
1835                 if (ext4_has_feature_quota(sb)) {
1836                         ext4_msg(sb, KERN_INFO,
1837                                  "Quota format mount options ignored "
1838                                  "when QUOTA feature is enabled");
1839                         return 1;
1840                 }
1841                 sbi->s_jquota_fmt = m->mount_opt;
1842 #endif
1843         } else if (token == Opt_dax) {
1844 #ifdef CONFIG_FS_DAX
1845                 ext4_msg(sb, KERN_WARNING,
1846                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1847                         sbi->s_mount_opt |= m->mount_opt;
1848 #else
1849                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1850                 return -1;
1851 #endif
1852         } else if (token == Opt_data_err_abort) {
1853                 sbi->s_mount_opt |= m->mount_opt;
1854         } else if (token == Opt_data_err_ignore) {
1855                 sbi->s_mount_opt &= ~m->mount_opt;
1856         } else {
1857                 if (!args->from)
1858                         arg = 1;
1859                 if (m->flags & MOPT_CLEAR)
1860                         arg = !arg;
1861                 else if (unlikely(!(m->flags & MOPT_SET))) {
1862                         ext4_msg(sb, KERN_WARNING,
1863                                  "buggy handling of option %s", opt);
1864                         WARN_ON(1);
1865                         return -1;
1866                 }
1867                 if (arg != 0)
1868                         sbi->s_mount_opt |= m->mount_opt;
1869                 else
1870                         sbi->s_mount_opt &= ~m->mount_opt;
1871         }
1872         return 1;
1873 }
1874
1875 static int parse_options(char *options, struct super_block *sb,
1876                          unsigned long *journal_devnum,
1877                          unsigned int *journal_ioprio,
1878                          int is_remount)
1879 {
1880         struct ext4_sb_info *sbi = EXT4_SB(sb);
1881         char *p;
1882         substring_t args[MAX_OPT_ARGS];
1883         int token;
1884
1885         if (!options)
1886                 return 1;
1887
1888         while ((p = strsep(&options, ",")) != NULL) {
1889                 if (!*p)
1890                         continue;
1891                 /*
1892                  * Initialize args struct so we know whether arg was
1893                  * found; some options take optional arguments.
1894                  */
1895                 args[0].to = args[0].from = NULL;
1896                 token = match_token(p, tokens, args);
1897                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1898                                      journal_ioprio, is_remount) < 0)
1899                         return 0;
1900         }
1901 #ifdef CONFIG_QUOTA
1902         /*
1903          * We do the test below only for project quotas. 'usrquota' and
1904          * 'grpquota' mount options are allowed even without quota feature
1905          * to support legacy quotas in quota files.
1906          */
1907         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1908                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1909                          "Cannot enable project quota enforcement.");
1910                 return 0;
1911         }
1912         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1913                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1914                         clear_opt(sb, USRQUOTA);
1915
1916                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1917                         clear_opt(sb, GRPQUOTA);
1918
1919                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1920                         ext4_msg(sb, KERN_ERR, "old and new quota "
1921                                         "format mixing");
1922                         return 0;
1923                 }
1924
1925                 if (!sbi->s_jquota_fmt) {
1926                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1927                                         "not specified");
1928                         return 0;
1929                 }
1930         }
1931 #endif
1932         if (test_opt(sb, DIOREAD_NOLOCK)) {
1933                 int blocksize =
1934                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1935
1936                 if (blocksize < PAGE_SIZE) {
1937                         ext4_msg(sb, KERN_ERR, "can't mount with "
1938                                  "dioread_nolock if block size != PAGE_SIZE");
1939                         return 0;
1940                 }
1941         }
1942         return 1;
1943 }
1944
1945 static inline void ext4_show_quota_options(struct seq_file *seq,
1946                                            struct super_block *sb)
1947 {
1948 #if defined(CONFIG_QUOTA)
1949         struct ext4_sb_info *sbi = EXT4_SB(sb);
1950
1951         if (sbi->s_jquota_fmt) {
1952                 char *fmtname = "";
1953
1954                 switch (sbi->s_jquota_fmt) {
1955                 case QFMT_VFS_OLD:
1956                         fmtname = "vfsold";
1957                         break;
1958                 case QFMT_VFS_V0:
1959                         fmtname = "vfsv0";
1960                         break;
1961                 case QFMT_VFS_V1:
1962                         fmtname = "vfsv1";
1963                         break;
1964                 }
1965                 seq_printf(seq, ",jqfmt=%s", fmtname);
1966         }
1967
1968         if (sbi->s_qf_names[USRQUOTA])
1969                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1970
1971         if (sbi->s_qf_names[GRPQUOTA])
1972                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1973 #endif
1974 }
1975
1976 static const char *token2str(int token)
1977 {
1978         const struct match_token *t;
1979
1980         for (t = tokens; t->token != Opt_err; t++)
1981                 if (t->token == token && !strchr(t->pattern, '='))
1982                         break;
1983         return t->pattern;
1984 }
1985
1986 /*
1987  * Show an option if
1988  *  - it's set to a non-default value OR
1989  *  - if the per-sb default is different from the global default
1990  */
1991 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1992                               int nodefs)
1993 {
1994         struct ext4_sb_info *sbi = EXT4_SB(sb);
1995         struct ext4_super_block *es = sbi->s_es;
1996         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1997         const struct mount_opts *m;
1998         char sep = nodefs ? '\n' : ',';
1999
2000 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2001 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2002
2003         if (sbi->s_sb_block != 1)
2004                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2005
2006         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2007                 int want_set = m->flags & MOPT_SET;
2008                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2009                     (m->flags & MOPT_CLEAR_ERR))
2010                         continue;
2011                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2012                         continue; /* skip if same as the default */
2013                 if ((want_set &&
2014                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2015                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2016                         continue; /* select Opt_noFoo vs Opt_Foo */
2017                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2018         }
2019
2020         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2021             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2022                 SEQ_OPTS_PRINT("resuid=%u",
2023                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2024         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2025             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2026                 SEQ_OPTS_PRINT("resgid=%u",
2027                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2028         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2029         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2030                 SEQ_OPTS_PUTS("errors=remount-ro");
2031         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2032                 SEQ_OPTS_PUTS("errors=continue");
2033         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2034                 SEQ_OPTS_PUTS("errors=panic");
2035         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2036                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2037         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2038                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2039         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2040                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2041         if (sb->s_flags & MS_I_VERSION)
2042                 SEQ_OPTS_PUTS("i_version");
2043         if (nodefs || sbi->s_stripe)
2044                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2045         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2046                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2047                         SEQ_OPTS_PUTS("data=journal");
2048                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2049                         SEQ_OPTS_PUTS("data=ordered");
2050                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2051                         SEQ_OPTS_PUTS("data=writeback");
2052         }
2053         if (nodefs ||
2054             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2055                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2056                                sbi->s_inode_readahead_blks);
2057
2058         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2059                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2060                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2061         if (nodefs || sbi->s_max_dir_size_kb)
2062                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2063         if (test_opt(sb, DATA_ERR_ABORT))
2064                 SEQ_OPTS_PUTS("data_err=abort");
2065
2066         ext4_show_quota_options(seq, sb);
2067         return 0;
2068 }
2069
2070 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2071 {
2072         return _ext4_show_options(seq, root->d_sb, 0);
2073 }
2074
2075 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2076 {
2077         struct super_block *sb = seq->private;
2078         int rc;
2079
2080         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2081         rc = _ext4_show_options(seq, sb, 1);
2082         seq_puts(seq, "\n");
2083         return rc;
2084 }
2085
2086 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2087                             int read_only)
2088 {
2089         struct ext4_sb_info *sbi = EXT4_SB(sb);
2090         int res = 0;
2091
2092         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2093                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2094                          "forcing read-only mode");
2095                 res = MS_RDONLY;
2096         }
2097         if (read_only)
2098                 goto done;
2099         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2100                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2101                          "running e2fsck is recommended");
2102         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2103                 ext4_msg(sb, KERN_WARNING,
2104                          "warning: mounting fs with errors, "
2105                          "running e2fsck is recommended");
2106         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2107                  le16_to_cpu(es->s_mnt_count) >=
2108                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2109                 ext4_msg(sb, KERN_WARNING,
2110                          "warning: maximal mount count reached, "
2111                          "running e2fsck is recommended");
2112         else if (le32_to_cpu(es->s_checkinterval) &&
2113                 (le32_to_cpu(es->s_lastcheck) +
2114                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2115                 ext4_msg(sb, KERN_WARNING,
2116                          "warning: checktime reached, "
2117                          "running e2fsck is recommended");
2118         if (!sbi->s_journal)
2119                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2120         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2121                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2122         le16_add_cpu(&es->s_mnt_count, 1);
2123         es->s_mtime = cpu_to_le32(get_seconds());
2124         ext4_update_dynamic_rev(sb);
2125         if (sbi->s_journal)
2126                 ext4_set_feature_journal_needs_recovery(sb);
2127
2128         ext4_commit_super(sb, 1);
2129 done:
2130         if (test_opt(sb, DEBUG))
2131                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2132                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2133                         sb->s_blocksize,
2134                         sbi->s_groups_count,
2135                         EXT4_BLOCKS_PER_GROUP(sb),
2136                         EXT4_INODES_PER_GROUP(sb),
2137                         sbi->s_mount_opt, sbi->s_mount_opt2);
2138
2139         cleancache_init_fs(sb);
2140         return res;
2141 }
2142
2143 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2144 {
2145         struct ext4_sb_info *sbi = EXT4_SB(sb);
2146         struct flex_groups *new_groups;
2147         int size;
2148
2149         if (!sbi->s_log_groups_per_flex)
2150                 return 0;
2151
2152         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2153         if (size <= sbi->s_flex_groups_allocated)
2154                 return 0;
2155
2156         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2157         new_groups = kvzalloc(size, GFP_KERNEL);
2158         if (!new_groups) {
2159                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2160                          size / (int) sizeof(struct flex_groups));
2161                 return -ENOMEM;
2162         }
2163
2164         if (sbi->s_flex_groups) {
2165                 memcpy(new_groups, sbi->s_flex_groups,
2166                        (sbi->s_flex_groups_allocated *
2167                         sizeof(struct flex_groups)));
2168                 kvfree(sbi->s_flex_groups);
2169         }
2170         sbi->s_flex_groups = new_groups;
2171         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2172         return 0;
2173 }
2174
2175 static int ext4_fill_flex_info(struct super_block *sb)
2176 {
2177         struct ext4_sb_info *sbi = EXT4_SB(sb);
2178         struct ext4_group_desc *gdp = NULL;
2179         ext4_group_t flex_group;
2180         int i, err;
2181
2182         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2183         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2184                 sbi->s_log_groups_per_flex = 0;
2185                 return 1;
2186         }
2187
2188         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2189         if (err)
2190                 goto failed;
2191
2192         for (i = 0; i < sbi->s_groups_count; i++) {
2193                 gdp = ext4_get_group_desc(sb, i, NULL);
2194
2195                 flex_group = ext4_flex_group(sbi, i);
2196                 atomic_add(ext4_free_inodes_count(sb, gdp),
2197                            &sbi->s_flex_groups[flex_group].free_inodes);
2198                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2199                              &sbi->s_flex_groups[flex_group].free_clusters);
2200                 atomic_add(ext4_used_dirs_count(sb, gdp),
2201                            &sbi->s_flex_groups[flex_group].used_dirs);
2202         }
2203
2204         return 1;
2205 failed:
2206         return 0;
2207 }
2208
2209 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2210                                    struct ext4_group_desc *gdp)
2211 {
2212         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2213         __u16 crc = 0;
2214         __le32 le_group = cpu_to_le32(block_group);
2215         struct ext4_sb_info *sbi = EXT4_SB(sb);
2216
2217         if (ext4_has_metadata_csum(sbi->s_sb)) {
2218                 /* Use new metadata_csum algorithm */
2219                 __u32 csum32;
2220                 __u16 dummy_csum = 0;
2221
2222                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2223                                      sizeof(le_group));
2224                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2225                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2226                                      sizeof(dummy_csum));
2227                 offset += sizeof(dummy_csum);
2228                 if (offset < sbi->s_desc_size)
2229                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2230                                              sbi->s_desc_size - offset);
2231
2232                 crc = csum32 & 0xFFFF;
2233                 goto out;
2234         }
2235
2236         /* old crc16 code */
2237         if (!ext4_has_feature_gdt_csum(sb))
2238                 return 0;
2239
2240         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2241         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2242         crc = crc16(crc, (__u8 *)gdp, offset);
2243         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2244         /* for checksum of struct ext4_group_desc do the rest...*/
2245         if (ext4_has_feature_64bit(sb) &&
2246             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2247                 crc = crc16(crc, (__u8 *)gdp + offset,
2248                             le16_to_cpu(sbi->s_es->s_desc_size) -
2249                                 offset);
2250
2251 out:
2252         return cpu_to_le16(crc);
2253 }
2254
2255 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2256                                 struct ext4_group_desc *gdp)
2257 {
2258         if (ext4_has_group_desc_csum(sb) &&
2259             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2260                 return 0;
2261
2262         return 1;
2263 }
2264
2265 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2266                               struct ext4_group_desc *gdp)
2267 {
2268         if (!ext4_has_group_desc_csum(sb))
2269                 return;
2270         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2271 }
2272
2273 /* Called at mount-time, super-block is locked */
2274 static int ext4_check_descriptors(struct super_block *sb,
2275                                   ext4_fsblk_t sb_block,
2276                                   ext4_group_t *first_not_zeroed)
2277 {
2278         struct ext4_sb_info *sbi = EXT4_SB(sb);
2279         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2280         ext4_fsblk_t last_block;
2281         ext4_fsblk_t block_bitmap;
2282         ext4_fsblk_t inode_bitmap;
2283         ext4_fsblk_t inode_table;
2284         int flexbg_flag = 0;
2285         ext4_group_t i, grp = sbi->s_groups_count;
2286
2287         if (ext4_has_feature_flex_bg(sb))
2288                 flexbg_flag = 1;
2289
2290         ext4_debug("Checking group descriptors");
2291
2292         for (i = 0; i < sbi->s_groups_count; i++) {
2293                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2294
2295                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2296                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2297                 else
2298                         last_block = first_block +
2299                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2300
2301                 if ((grp == sbi->s_groups_count) &&
2302                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2303                         grp = i;
2304
2305                 block_bitmap = ext4_block_bitmap(sb, gdp);
2306                 if (block_bitmap == sb_block) {
2307                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2308                                  "Block bitmap for group %u overlaps "
2309                                  "superblock", i);
2310                 }
2311                 if (block_bitmap < first_block || block_bitmap > last_block) {
2312                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2313                                "Block bitmap for group %u not in group "
2314                                "(block %llu)!", i, block_bitmap);
2315                         return 0;
2316                 }
2317                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2318                 if (inode_bitmap == sb_block) {
2319                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2320                                  "Inode bitmap for group %u overlaps "
2321                                  "superblock", i);
2322                 }
2323                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2324                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2325                                "Inode bitmap for group %u not in group "
2326                                "(block %llu)!", i, inode_bitmap);
2327                         return 0;
2328                 }
2329                 inode_table = ext4_inode_table(sb, gdp);
2330                 if (inode_table == sb_block) {
2331                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2332                                  "Inode table for group %u overlaps "
2333                                  "superblock", i);
2334                 }
2335                 if (inode_table < first_block ||
2336                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2337                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2338                                "Inode table for group %u not in group "
2339                                "(block %llu)!", i, inode_table);
2340                         return 0;
2341                 }
2342                 ext4_lock_group(sb, i);
2343                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2344                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2345                                  "Checksum for group %u failed (%u!=%u)",
2346                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2347                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2348                         if (!(sb->s_flags & MS_RDONLY)) {
2349                                 ext4_unlock_group(sb, i);
2350                                 return 0;
2351                         }
2352                 }
2353                 ext4_unlock_group(sb, i);
2354                 if (!flexbg_flag)
2355                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2356         }
2357         if (NULL != first_not_zeroed)
2358                 *first_not_zeroed = grp;
2359         return 1;
2360 }
2361
2362 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2363  * the superblock) which were deleted from all directories, but held open by
2364  * a process at the time of a crash.  We walk the list and try to delete these
2365  * inodes at recovery time (only with a read-write filesystem).
2366  *
2367  * In order to keep the orphan inode chain consistent during traversal (in
2368  * case of crash during recovery), we link each inode into the superblock
2369  * orphan list_head and handle it the same way as an inode deletion during
2370  * normal operation (which journals the operations for us).
2371  *
2372  * We only do an iget() and an iput() on each inode, which is very safe if we
2373  * accidentally point at an in-use or already deleted inode.  The worst that
2374  * can happen in this case is that we get a "bit already cleared" message from
2375  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2376  * e2fsck was run on this filesystem, and it must have already done the orphan
2377  * inode cleanup for us, so we can safely abort without any further action.
2378  */
2379 static void ext4_orphan_cleanup(struct super_block *sb,
2380                                 struct ext4_super_block *es)
2381 {
2382         unsigned int s_flags = sb->s_flags;
2383         int ret, nr_orphans = 0, nr_truncates = 0;
2384 #ifdef CONFIG_QUOTA
2385         int i;
2386 #endif
2387         if (!es->s_last_orphan) {
2388                 jbd_debug(4, "no orphan inodes to clean up\n");
2389                 return;
2390         }
2391
2392         if (bdev_read_only(sb->s_bdev)) {
2393                 ext4_msg(sb, KERN_ERR, "write access "
2394                         "unavailable, skipping orphan cleanup");
2395                 return;
2396         }
2397
2398         /* Check if feature set would not allow a r/w mount */
2399         if (!ext4_feature_set_ok(sb, 0)) {
2400                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2401                          "unknown ROCOMPAT features");
2402                 return;
2403         }
2404
2405         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2406                 /* don't clear list on RO mount w/ errors */
2407                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2408                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2409                                   "clearing orphan list.\n");
2410                         es->s_last_orphan = 0;
2411                 }
2412                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2413                 return;
2414         }
2415
2416         if (s_flags & MS_RDONLY) {
2417                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2418                 sb->s_flags &= ~MS_RDONLY;
2419         }
2420 #ifdef CONFIG_QUOTA
2421         /* Needed for iput() to work correctly and not trash data */
2422         sb->s_flags |= MS_ACTIVE;
2423         /* Turn on quotas so that they are updated correctly */
2424         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2425                 if (EXT4_SB(sb)->s_qf_names[i]) {
2426                         int ret = ext4_quota_on_mount(sb, i);
2427                         if (ret < 0)
2428                                 ext4_msg(sb, KERN_ERR,
2429                                         "Cannot turn on journaled "
2430                                         "quota: error %d", ret);
2431                 }
2432         }
2433 #endif
2434
2435         while (es->s_last_orphan) {
2436                 struct inode *inode;
2437
2438                 /*
2439                  * We may have encountered an error during cleanup; if
2440                  * so, skip the rest.
2441                  */
2442                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2443                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2444                         es->s_last_orphan = 0;
2445                         break;
2446                 }
2447
2448                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2449                 if (IS_ERR(inode)) {
2450                         es->s_last_orphan = 0;
2451                         break;
2452                 }
2453
2454                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2455                 dquot_initialize(inode);
2456                 if (inode->i_nlink) {
2457                         if (test_opt(sb, DEBUG))
2458                                 ext4_msg(sb, KERN_DEBUG,
2459                                         "%s: truncating inode %lu to %lld bytes",
2460                                         __func__, inode->i_ino, inode->i_size);
2461                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2462                                   inode->i_ino, inode->i_size);
2463                         inode_lock(inode);
2464                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2465                         ret = ext4_truncate(inode);
2466                         if (ret)
2467                                 ext4_std_error(inode->i_sb, ret);
2468                         inode_unlock(inode);
2469                         nr_truncates++;
2470                 } else {
2471                         if (test_opt(sb, DEBUG))
2472                                 ext4_msg(sb, KERN_DEBUG,
2473                                         "%s: deleting unreferenced inode %lu",
2474                                         __func__, inode->i_ino);
2475                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2476                                   inode->i_ino);
2477                         nr_orphans++;
2478                 }
2479                 iput(inode);  /* The delete magic happens here! */
2480         }
2481
2482 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2483
2484         if (nr_orphans)
2485                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2486                        PLURAL(nr_orphans));
2487         if (nr_truncates)
2488                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2489                        PLURAL(nr_truncates));
2490 #ifdef CONFIG_QUOTA
2491         /* Turn quotas off */
2492         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2493                 if (sb_dqopt(sb)->files[i])
2494                         dquot_quota_off(sb, i);
2495         }
2496 #endif
2497         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2498 }
2499
2500 /*
2501  * Maximal extent format file size.
2502  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2503  * extent format containers, within a sector_t, and within i_blocks
2504  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2505  * so that won't be a limiting factor.
2506  *
2507  * However there is other limiting factor. We do store extents in the form
2508  * of starting block and length, hence the resulting length of the extent
2509  * covering maximum file size must fit into on-disk format containers as
2510  * well. Given that length is always by 1 unit bigger than max unit (because
2511  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2512  *
2513  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2514  */
2515 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2516 {
2517         loff_t res;
2518         loff_t upper_limit = MAX_LFS_FILESIZE;
2519
2520         /* small i_blocks in vfs inode? */
2521         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2522                 /*
2523                  * CONFIG_LBDAF is not enabled implies the inode
2524                  * i_block represent total blocks in 512 bytes
2525                  * 32 == size of vfs inode i_blocks * 8
2526                  */
2527                 upper_limit = (1LL << 32) - 1;
2528
2529                 /* total blocks in file system block size */
2530                 upper_limit >>= (blkbits - 9);
2531                 upper_limit <<= blkbits;
2532         }
2533
2534         /*
2535          * 32-bit extent-start container, ee_block. We lower the maxbytes
2536          * by one fs block, so ee_len can cover the extent of maximum file
2537          * size
2538          */
2539         res = (1LL << 32) - 1;
2540         res <<= blkbits;
2541
2542         /* Sanity check against vm- & vfs- imposed limits */
2543         if (res > upper_limit)
2544                 res = upper_limit;
2545
2546         return res;
2547 }
2548
2549 /*
2550  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2551  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2552  * We need to be 1 filesystem block less than the 2^48 sector limit.
2553  */
2554 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2555 {
2556         loff_t res = EXT4_NDIR_BLOCKS;
2557         int meta_blocks;
2558         loff_t upper_limit;
2559         /* This is calculated to be the largest file size for a dense, block
2560          * mapped file such that the file's total number of 512-byte sectors,
2561          * including data and all indirect blocks, does not exceed (2^48 - 1).
2562          *
2563          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2564          * number of 512-byte sectors of the file.
2565          */
2566
2567         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2568                 /*
2569                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2570                  * the inode i_block field represents total file blocks in
2571                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2572                  */
2573                 upper_limit = (1LL << 32) - 1;
2574
2575                 /* total blocks in file system block size */
2576                 upper_limit >>= (bits - 9);
2577
2578         } else {
2579                 /*
2580                  * We use 48 bit ext4_inode i_blocks
2581                  * With EXT4_HUGE_FILE_FL set the i_blocks
2582                  * represent total number of blocks in
2583                  * file system block size
2584                  */
2585                 upper_limit = (1LL << 48) - 1;
2586
2587         }
2588
2589         /* indirect blocks */
2590         meta_blocks = 1;
2591         /* double indirect blocks */
2592         meta_blocks += 1 + (1LL << (bits-2));
2593         /* tripple indirect blocks */
2594         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2595
2596         upper_limit -= meta_blocks;
2597         upper_limit <<= bits;
2598
2599         res += 1LL << (bits-2);
2600         res += 1LL << (2*(bits-2));
2601         res += 1LL << (3*(bits-2));
2602         res <<= bits;
2603         if (res > upper_limit)
2604                 res = upper_limit;
2605
2606         if (res > MAX_LFS_FILESIZE)
2607                 res = MAX_LFS_FILESIZE;
2608
2609         return res;
2610 }
2611
2612 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2613                                    ext4_fsblk_t logical_sb_block, int nr)
2614 {
2615         struct ext4_sb_info *sbi = EXT4_SB(sb);
2616         ext4_group_t bg, first_meta_bg;
2617         int has_super = 0;
2618
2619         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2620
2621         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2622                 return logical_sb_block + nr + 1;
2623         bg = sbi->s_desc_per_block * nr;
2624         if (ext4_bg_has_super(sb, bg))
2625                 has_super = 1;
2626
2627         /*
2628          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2629          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2630          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2631          * compensate.
2632          */
2633         if (sb->s_blocksize == 1024 && nr == 0 &&
2634             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2635                 has_super++;
2636
2637         return (has_super + ext4_group_first_block_no(sb, bg));
2638 }
2639
2640 /**
2641  * ext4_get_stripe_size: Get the stripe size.
2642  * @sbi: In memory super block info
2643  *
2644  * If we have specified it via mount option, then
2645  * use the mount option value. If the value specified at mount time is
2646  * greater than the blocks per group use the super block value.
2647  * If the super block value is greater than blocks per group return 0.
2648  * Allocator needs it be less than blocks per group.
2649  *
2650  */
2651 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2652 {
2653         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2654         unsigned long stripe_width =
2655                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2656         int ret;
2657
2658         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2659                 ret = sbi->s_stripe;
2660         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2661                 ret = stripe_width;
2662         else if (stride && stride <= sbi->s_blocks_per_group)
2663                 ret = stride;
2664         else
2665                 ret = 0;
2666
2667         /*
2668          * If the stripe width is 1, this makes no sense and
2669          * we set it to 0 to turn off stripe handling code.
2670          */
2671         if (ret <= 1)
2672                 ret = 0;
2673
2674         return ret;
2675 }
2676
2677 /*
2678  * Check whether this filesystem can be mounted based on
2679  * the features present and the RDONLY/RDWR mount requested.
2680  * Returns 1 if this filesystem can be mounted as requested,
2681  * 0 if it cannot be.
2682  */
2683 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2684 {
2685         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2686                 ext4_msg(sb, KERN_ERR,
2687                         "Couldn't mount because of "
2688                         "unsupported optional features (%x)",
2689                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2690                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2691                 return 0;
2692         }
2693
2694         if (readonly)
2695                 return 1;
2696
2697         if (ext4_has_feature_readonly(sb)) {
2698                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2699                 sb->s_flags |= MS_RDONLY;
2700                 return 1;
2701         }
2702
2703         /* Check that feature set is OK for a read-write mount */
2704         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2705                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2706                          "unsupported optional features (%x)",
2707                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2708                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2709                 return 0;
2710         }
2711         /*
2712          * Large file size enabled file system can only be mounted
2713          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2714          */
2715         if (ext4_has_feature_huge_file(sb)) {
2716                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2717                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2718                                  "cannot be mounted RDWR without "
2719                                  "CONFIG_LBDAF");
2720                         return 0;
2721                 }
2722         }
2723         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2724                 ext4_msg(sb, KERN_ERR,
2725                          "Can't support bigalloc feature without "
2726                          "extents feature\n");
2727                 return 0;
2728         }
2729
2730 #ifndef CONFIG_QUOTA
2731         if (ext4_has_feature_quota(sb) && !readonly) {
2732                 ext4_msg(sb, KERN_ERR,
2733                          "Filesystem with quota feature cannot be mounted RDWR "
2734                          "without CONFIG_QUOTA");
2735                 return 0;
2736         }
2737         if (ext4_has_feature_project(sb) && !readonly) {
2738                 ext4_msg(sb, KERN_ERR,
2739                          "Filesystem with project quota feature cannot be mounted RDWR "
2740                          "without CONFIG_QUOTA");
2741                 return 0;
2742         }
2743 #endif  /* CONFIG_QUOTA */
2744         return 1;
2745 }
2746
2747 /*
2748  * This function is called once a day if we have errors logged
2749  * on the file system
2750  */
2751 static void print_daily_error_info(unsigned long arg)
2752 {
2753         struct super_block *sb = (struct super_block *) arg;
2754         struct ext4_sb_info *sbi;
2755         struct ext4_super_block *es;
2756
2757         sbi = EXT4_SB(sb);
2758         es = sbi->s_es;
2759
2760         if (es->s_error_count)
2761                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2762                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2763                          le32_to_cpu(es->s_error_count));
2764         if (es->s_first_error_time) {
2765                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2766                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2767                        (int) sizeof(es->s_first_error_func),
2768                        es->s_first_error_func,
2769                        le32_to_cpu(es->s_first_error_line));
2770                 if (es->s_first_error_ino)
2771                         printk(KERN_CONT ": inode %u",
2772                                le32_to_cpu(es->s_first_error_ino));
2773                 if (es->s_first_error_block)
2774                         printk(KERN_CONT ": block %llu", (unsigned long long)
2775                                le64_to_cpu(es->s_first_error_block));
2776                 printk(KERN_CONT "\n");
2777         }
2778         if (es->s_last_error_time) {
2779                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2780                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2781                        (int) sizeof(es->s_last_error_func),
2782                        es->s_last_error_func,
2783                        le32_to_cpu(es->s_last_error_line));
2784                 if (es->s_last_error_ino)
2785                         printk(KERN_CONT ": inode %u",
2786                                le32_to_cpu(es->s_last_error_ino));
2787                 if (es->s_last_error_block)
2788                         printk(KERN_CONT ": block %llu", (unsigned long long)
2789                                le64_to_cpu(es->s_last_error_block));
2790                 printk(KERN_CONT "\n");
2791         }
2792         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2793 }
2794
2795 /* Find next suitable group and run ext4_init_inode_table */
2796 static int ext4_run_li_request(struct ext4_li_request *elr)
2797 {
2798         struct ext4_group_desc *gdp = NULL;
2799         ext4_group_t group, ngroups;
2800         struct super_block *sb;
2801         unsigned long timeout = 0;
2802         int ret = 0;
2803
2804         sb = elr->lr_super;
2805         ngroups = EXT4_SB(sb)->s_groups_count;
2806
2807         for (group = elr->lr_next_group; group < ngroups; group++) {
2808                 gdp = ext4_get_group_desc(sb, group, NULL);
2809                 if (!gdp) {
2810                         ret = 1;
2811                         break;
2812                 }
2813
2814                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2815                         break;
2816         }
2817
2818         if (group >= ngroups)
2819                 ret = 1;
2820
2821         if (!ret) {
2822                 timeout = jiffies;
2823                 ret = ext4_init_inode_table(sb, group,
2824                                             elr->lr_timeout ? 0 : 1);
2825                 if (elr->lr_timeout == 0) {
2826                         timeout = (jiffies - timeout) *
2827                                   elr->lr_sbi->s_li_wait_mult;
2828                         elr->lr_timeout = timeout;
2829                 }
2830                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2831                 elr->lr_next_group = group + 1;
2832         }
2833         return ret;
2834 }
2835
2836 /*
2837  * Remove lr_request from the list_request and free the
2838  * request structure. Should be called with li_list_mtx held
2839  */
2840 static void ext4_remove_li_request(struct ext4_li_request *elr)
2841 {
2842         struct ext4_sb_info *sbi;
2843
2844         if (!elr)
2845                 return;
2846
2847         sbi = elr->lr_sbi;
2848
2849         list_del(&elr->lr_request);
2850         sbi->s_li_request = NULL;
2851         kfree(elr);
2852 }
2853
2854 static void ext4_unregister_li_request(struct super_block *sb)
2855 {
2856         mutex_lock(&ext4_li_mtx);
2857         if (!ext4_li_info) {
2858                 mutex_unlock(&ext4_li_mtx);
2859                 return;
2860         }
2861
2862         mutex_lock(&ext4_li_info->li_list_mtx);
2863         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2864         mutex_unlock(&ext4_li_info->li_list_mtx);
2865         mutex_unlock(&ext4_li_mtx);
2866 }
2867
2868 static struct task_struct *ext4_lazyinit_task;
2869
2870 /*
2871  * This is the function where ext4lazyinit thread lives. It walks
2872  * through the request list searching for next scheduled filesystem.
2873  * When such a fs is found, run the lazy initialization request
2874  * (ext4_rn_li_request) and keep track of the time spend in this
2875  * function. Based on that time we compute next schedule time of
2876  * the request. When walking through the list is complete, compute
2877  * next waking time and put itself into sleep.
2878  */
2879 static int ext4_lazyinit_thread(void *arg)
2880 {
2881         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2882         struct list_head *pos, *n;
2883         struct ext4_li_request *elr;
2884         unsigned long next_wakeup, cur;
2885
2886         BUG_ON(NULL == eli);
2887
2888 cont_thread:
2889         while (true) {
2890                 next_wakeup = MAX_JIFFY_OFFSET;
2891
2892                 mutex_lock(&eli->li_list_mtx);
2893                 if (list_empty(&eli->li_request_list)) {
2894                         mutex_unlock(&eli->li_list_mtx);
2895                         goto exit_thread;
2896                 }
2897                 list_for_each_safe(pos, n, &eli->li_request_list) {
2898                         int err = 0;
2899                         int progress = 0;
2900                         elr = list_entry(pos, struct ext4_li_request,
2901                                          lr_request);
2902
2903                         if (time_before(jiffies, elr->lr_next_sched)) {
2904                                 if (time_before(elr->lr_next_sched, next_wakeup))
2905                                         next_wakeup = elr->lr_next_sched;
2906                                 continue;
2907                         }
2908                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2909                                 if (sb_start_write_trylock(elr->lr_super)) {
2910                                         progress = 1;
2911                                         /*
2912                                          * We hold sb->s_umount, sb can not
2913                                          * be removed from the list, it is
2914                                          * now safe to drop li_list_mtx
2915                                          */
2916                                         mutex_unlock(&eli->li_list_mtx);
2917                                         err = ext4_run_li_request(elr);
2918                                         sb_end_write(elr->lr_super);
2919                                         mutex_lock(&eli->li_list_mtx);
2920                                         n = pos->next;
2921                                 }
2922                                 up_read((&elr->lr_super->s_umount));
2923                         }
2924                         /* error, remove the lazy_init job */
2925                         if (err) {
2926                                 ext4_remove_li_request(elr);
2927                                 continue;
2928                         }
2929                         if (!progress) {
2930                                 elr->lr_next_sched = jiffies +
2931                                         (prandom_u32()
2932                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2933                         }
2934                         if (time_before(elr->lr_next_sched, next_wakeup))
2935                                 next_wakeup = elr->lr_next_sched;
2936                 }
2937                 mutex_unlock(&eli->li_list_mtx);
2938
2939                 try_to_freeze();
2940
2941                 cur = jiffies;
2942                 if ((time_after_eq(cur, next_wakeup)) ||
2943                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2944                         cond_resched();
2945                         continue;
2946                 }
2947
2948                 schedule_timeout_interruptible(next_wakeup - cur);
2949
2950                 if (kthread_should_stop()) {
2951                         ext4_clear_request_list();
2952                         goto exit_thread;
2953                 }
2954         }
2955
2956 exit_thread:
2957         /*
2958          * It looks like the request list is empty, but we need
2959          * to check it under the li_list_mtx lock, to prevent any
2960          * additions into it, and of course we should lock ext4_li_mtx
2961          * to atomically free the list and ext4_li_info, because at
2962          * this point another ext4 filesystem could be registering
2963          * new one.
2964          */
2965         mutex_lock(&ext4_li_mtx);
2966         mutex_lock(&eli->li_list_mtx);
2967         if (!list_empty(&eli->li_request_list)) {
2968                 mutex_unlock(&eli->li_list_mtx);
2969                 mutex_unlock(&ext4_li_mtx);
2970                 goto cont_thread;
2971         }
2972         mutex_unlock(&eli->li_list_mtx);
2973         kfree(ext4_li_info);
2974         ext4_li_info = NULL;
2975         mutex_unlock(&ext4_li_mtx);
2976
2977         return 0;
2978 }
2979
2980 static void ext4_clear_request_list(void)
2981 {
2982         struct list_head *pos, *n;
2983         struct ext4_li_request *elr;
2984
2985         mutex_lock(&ext4_li_info->li_list_mtx);
2986         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2987                 elr = list_entry(pos, struct ext4_li_request,
2988                                  lr_request);
2989                 ext4_remove_li_request(elr);
2990         }
2991         mutex_unlock(&ext4_li_info->li_list_mtx);
2992 }
2993
2994 static int ext4_run_lazyinit_thread(void)
2995 {
2996         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2997                                          ext4_li_info, "ext4lazyinit");
2998         if (IS_ERR(ext4_lazyinit_task)) {
2999                 int err = PTR_ERR(ext4_lazyinit_task);
3000                 ext4_clear_request_list();
3001                 kfree(ext4_li_info);
3002                 ext4_li_info = NULL;
3003                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3004                                  "initialization thread\n",
3005                                  err);
3006                 return err;
3007         }
3008         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3009         return 0;
3010 }
3011
3012 /*
3013  * Check whether it make sense to run itable init. thread or not.
3014  * If there is at least one uninitialized inode table, return
3015  * corresponding group number, else the loop goes through all
3016  * groups and return total number of groups.
3017  */
3018 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3019 {
3020         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3021         struct ext4_group_desc *gdp = NULL;
3022
3023         for (group = 0; group < ngroups; group++) {
3024                 gdp = ext4_get_group_desc(sb, group, NULL);
3025                 if (!gdp)
3026                         continue;
3027
3028                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3029                         break;
3030         }
3031
3032         return group;
3033 }
3034
3035 static int ext4_li_info_new(void)
3036 {
3037         struct ext4_lazy_init *eli = NULL;
3038
3039         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3040         if (!eli)
3041                 return -ENOMEM;
3042
3043         INIT_LIST_HEAD(&eli->li_request_list);
3044         mutex_init(&eli->li_list_mtx);
3045
3046         eli->li_state |= EXT4_LAZYINIT_QUIT;
3047
3048         ext4_li_info = eli;
3049
3050         return 0;
3051 }
3052
3053 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3054                                             ext4_group_t start)
3055 {
3056         struct ext4_sb_info *sbi = EXT4_SB(sb);
3057         struct ext4_li_request *elr;
3058
3059         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3060         if (!elr)
3061                 return NULL;
3062
3063         elr->lr_super = sb;
3064         elr->lr_sbi = sbi;
3065         elr->lr_next_group = start;
3066
3067         /*
3068          * Randomize first schedule time of the request to
3069          * spread the inode table initialization requests
3070          * better.
3071          */
3072         elr->lr_next_sched = jiffies + (prandom_u32() %
3073                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3074         return elr;
3075 }
3076
3077 int ext4_register_li_request(struct super_block *sb,
3078                              ext4_group_t first_not_zeroed)
3079 {
3080         struct ext4_sb_info *sbi = EXT4_SB(sb);
3081         struct ext4_li_request *elr = NULL;
3082         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3083         int ret = 0;
3084
3085         mutex_lock(&ext4_li_mtx);
3086         if (sbi->s_li_request != NULL) {
3087                 /*
3088                  * Reset timeout so it can be computed again, because
3089                  * s_li_wait_mult might have changed.
3090                  */
3091                 sbi->s_li_request->lr_timeout = 0;
3092                 goto out;
3093         }
3094
3095         if (first_not_zeroed == ngroups ||
3096             (sb->s_flags & MS_RDONLY) ||
3097             !test_opt(sb, INIT_INODE_TABLE))
3098                 goto out;
3099
3100         elr = ext4_li_request_new(sb, first_not_zeroed);
3101         if (!elr) {
3102                 ret = -ENOMEM;
3103                 goto out;
3104         }
3105
3106         if (NULL == ext4_li_info) {
3107                 ret = ext4_li_info_new();
3108                 if (ret)
3109                         goto out;
3110         }
3111
3112         mutex_lock(&ext4_li_info->li_list_mtx);
3113         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3114         mutex_unlock(&ext4_li_info->li_list_mtx);
3115
3116         sbi->s_li_request = elr;
3117         /*
3118          * set elr to NULL here since it has been inserted to
3119          * the request_list and the removal and free of it is
3120          * handled by ext4_clear_request_list from now on.
3121          */
3122         elr = NULL;
3123
3124         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3125                 ret = ext4_run_lazyinit_thread();
3126                 if (ret)
3127                         goto out;
3128         }
3129 out:
3130         mutex_unlock(&ext4_li_mtx);
3131         if (ret)
3132                 kfree(elr);
3133         return ret;
3134 }
3135
3136 /*
3137  * We do not need to lock anything since this is called on
3138  * module unload.
3139  */
3140 static void ext4_destroy_lazyinit_thread(void)
3141 {
3142         /*
3143          * If thread exited earlier
3144          * there's nothing to be done.
3145          */
3146         if (!ext4_li_info || !ext4_lazyinit_task)
3147                 return;
3148
3149         kthread_stop(ext4_lazyinit_task);
3150 }
3151
3152 static int set_journal_csum_feature_set(struct super_block *sb)
3153 {
3154         int ret = 1;
3155         int compat, incompat;
3156         struct ext4_sb_info *sbi = EXT4_SB(sb);
3157
3158         if (ext4_has_metadata_csum(sb)) {
3159                 /* journal checksum v3 */
3160                 compat = 0;
3161                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3162         } else {
3163                 /* journal checksum v1 */
3164                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3165                 incompat = 0;
3166         }
3167
3168         jbd2_journal_clear_features(sbi->s_journal,
3169                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3170                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3171                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3172         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3173                 ret = jbd2_journal_set_features(sbi->s_journal,
3174                                 compat, 0,
3175                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3176                                 incompat);
3177         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3178                 ret = jbd2_journal_set_features(sbi->s_journal,
3179                                 compat, 0,
3180                                 incompat);
3181                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3182                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3183         } else {
3184                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3185                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3186         }
3187
3188         return ret;
3189 }
3190
3191 /*
3192  * Note: calculating the overhead so we can be compatible with
3193  * historical BSD practice is quite difficult in the face of
3194  * clusters/bigalloc.  This is because multiple metadata blocks from
3195  * different block group can end up in the same allocation cluster.
3196  * Calculating the exact overhead in the face of clustered allocation
3197  * requires either O(all block bitmaps) in memory or O(number of block
3198  * groups**2) in time.  We will still calculate the superblock for
3199  * older file systems --- and if we come across with a bigalloc file
3200  * system with zero in s_overhead_clusters the estimate will be close to
3201  * correct especially for very large cluster sizes --- but for newer
3202  * file systems, it's better to calculate this figure once at mkfs
3203  * time, and store it in the superblock.  If the superblock value is
3204  * present (even for non-bigalloc file systems), we will use it.
3205  */
3206 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3207                           char *buf)
3208 {
3209         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3210         struct ext4_group_desc  *gdp;
3211         ext4_fsblk_t            first_block, last_block, b;
3212         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3213         int                     s, j, count = 0;
3214
3215         if (!ext4_has_feature_bigalloc(sb))
3216                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3217                         sbi->s_itb_per_group + 2);
3218
3219         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3220                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3221         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3222         for (i = 0; i < ngroups; i++) {
3223                 gdp = ext4_get_group_desc(sb, i, NULL);
3224                 b = ext4_block_bitmap(sb, gdp);
3225                 if (b >= first_block && b <= last_block) {
3226                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3227                         count++;
3228                 }
3229                 b = ext4_inode_bitmap(sb, gdp);
3230                 if (b >= first_block && b <= last_block) {
3231                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3232                         count++;
3233                 }
3234                 b = ext4_inode_table(sb, gdp);
3235                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3236                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3237                                 int c = EXT4_B2C(sbi, b - first_block);
3238                                 ext4_set_bit(c, buf);
3239                                 count++;
3240                         }
3241                 if (i != grp)
3242                         continue;
3243                 s = 0;
3244                 if (ext4_bg_has_super(sb, grp)) {
3245                         ext4_set_bit(s++, buf);
3246                         count++;
3247                 }
3248                 j = ext4_bg_num_gdb(sb, grp);
3249                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3250                         ext4_error(sb, "Invalid number of block group "
3251                                    "descriptor blocks: %d", j);
3252                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3253                 }
3254                 count += j;
3255                 for (; j > 0; j--)
3256                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3257         }
3258         if (!count)
3259                 return 0;
3260         return EXT4_CLUSTERS_PER_GROUP(sb) -
3261                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3262 }
3263
3264 /*
3265  * Compute the overhead and stash it in sbi->s_overhead
3266  */
3267 int ext4_calculate_overhead(struct super_block *sb)
3268 {
3269         struct ext4_sb_info *sbi = EXT4_SB(sb);
3270         struct ext4_super_block *es = sbi->s_es;
3271         struct inode *j_inode;
3272         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3273         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3274         ext4_fsblk_t overhead = 0;
3275         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3276
3277         if (!buf)
3278                 return -ENOMEM;
3279
3280         /*
3281          * Compute the overhead (FS structures).  This is constant
3282          * for a given filesystem unless the number of block groups
3283          * changes so we cache the previous value until it does.
3284          */
3285
3286         /*
3287          * All of the blocks before first_data_block are overhead
3288          */
3289         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3290
3291         /*
3292          * Add the overhead found in each block group
3293          */
3294         for (i = 0; i < ngroups; i++) {
3295                 int blks;
3296
3297                 blks = count_overhead(sb, i, buf);
3298                 overhead += blks;
3299                 if (blks)
3300                         memset(buf, 0, PAGE_SIZE);
3301                 cond_resched();
3302         }
3303
3304         /*
3305          * Add the internal journal blocks whether the journal has been
3306          * loaded or not
3307          */
3308         if (sbi->s_journal && !sbi->journal_bdev)
3309                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3310         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3311                 j_inode = ext4_get_journal_inode(sb, j_inum);
3312                 if (j_inode) {
3313                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3314                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3315                         iput(j_inode);
3316                 } else {
3317                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3318                 }
3319         }
3320         sbi->s_overhead = overhead;
3321         smp_wmb();
3322         free_page((unsigned long) buf);
3323         return 0;
3324 }
3325
3326 static void ext4_set_resv_clusters(struct super_block *sb)
3327 {
3328         ext4_fsblk_t resv_clusters;
3329         struct ext4_sb_info *sbi = EXT4_SB(sb);
3330
3331         /*
3332          * There's no need to reserve anything when we aren't using extents.
3333          * The space estimates are exact, there are no unwritten extents,
3334          * hole punching doesn't need new metadata... This is needed especially
3335          * to keep ext2/3 backward compatibility.
3336          */
3337         if (!ext4_has_feature_extents(sb))
3338                 return;
3339         /*
3340          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3341          * This should cover the situations where we can not afford to run
3342          * out of space like for example punch hole, or converting
3343          * unwritten extents in delalloc path. In most cases such
3344          * allocation would require 1, or 2 blocks, higher numbers are
3345          * very rare.
3346          */
3347         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3348                          sbi->s_cluster_bits);
3349
3350         do_div(resv_clusters, 50);
3351         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3352
3353         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3354 }
3355
3356 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3357 {
3358         char *orig_data = kstrdup(data, GFP_KERNEL);
3359         struct buffer_head *bh;
3360         struct ext4_super_block *es = NULL;
3361         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3362         ext4_fsblk_t block;
3363         ext4_fsblk_t sb_block = get_sb_block(&data);
3364         ext4_fsblk_t logical_sb_block;
3365         unsigned long offset = 0;
3366         unsigned long journal_devnum = 0;
3367         unsigned long def_mount_opts;
3368         struct inode *root;
3369         const char *descr;
3370         int ret = -ENOMEM;
3371         int blocksize, clustersize;
3372         unsigned int db_count;
3373         unsigned int i;
3374         int needs_recovery, has_huge_files, has_bigalloc;
3375         __u64 blocks_count;
3376         int err = 0;
3377         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3378         ext4_group_t first_not_zeroed;
3379
3380         if ((data && !orig_data) || !sbi)
3381                 goto out_free_base;
3382
3383         sbi->s_blockgroup_lock =
3384                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3385         if (!sbi->s_blockgroup_lock)
3386                 goto out_free_base;
3387
3388         sb->s_fs_info = sbi;
3389         sbi->s_sb = sb;
3390         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3391         sbi->s_sb_block = sb_block;
3392         if (sb->s_bdev->bd_part)
3393                 sbi->s_sectors_written_start =
3394                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3395
3396         /* Cleanup superblock name */
3397         strreplace(sb->s_id, '/', '!');
3398
3399         /* -EINVAL is default */
3400         ret = -EINVAL;
3401         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3402         if (!blocksize) {
3403                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3404                 goto out_fail;
3405         }
3406
3407         /*
3408          * The ext4 superblock will not be buffer aligned for other than 1kB
3409          * block sizes.  We need to calculate the offset from buffer start.
3410          */
3411         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3412                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3413                 offset = do_div(logical_sb_block, blocksize);
3414         } else {
3415                 logical_sb_block = sb_block;
3416         }
3417
3418         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3419                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3420                 goto out_fail;
3421         }
3422         /*
3423          * Note: s_es must be initialized as soon as possible because
3424          *       some ext4 macro-instructions depend on its value
3425          */
3426         es = (struct ext4_super_block *) (bh->b_data + offset);
3427         sbi->s_es = es;
3428         sb->s_magic = le16_to_cpu(es->s_magic);
3429         if (sb->s_magic != EXT4_SUPER_MAGIC)
3430                 goto cantfind_ext4;
3431         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3432
3433         /* Warn if metadata_csum and gdt_csum are both set. */
3434         if (ext4_has_feature_metadata_csum(sb) &&
3435             ext4_has_feature_gdt_csum(sb))
3436                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3437                              "redundant flags; please run fsck.");
3438
3439         /* Check for a known checksum algorithm */
3440         if (!ext4_verify_csum_type(sb, es)) {
3441                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3442                          "unknown checksum algorithm.");
3443                 silent = 1;
3444                 goto cantfind_ext4;
3445         }
3446
3447         /* Load the checksum driver */
3448         if (ext4_has_feature_metadata_csum(sb)) {
3449                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3450                 if (IS_ERR(sbi->s_chksum_driver)) {
3451                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3452                         ret = PTR_ERR(sbi->s_chksum_driver);
3453                         sbi->s_chksum_driver = NULL;
3454                         goto failed_mount;
3455                 }
3456         }
3457
3458         /* Check superblock checksum */
3459         if (!ext4_superblock_csum_verify(sb, es)) {
3460                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3461                          "invalid superblock checksum.  Run e2fsck?");
3462                 silent = 1;
3463                 ret = -EFSBADCRC;
3464                 goto cantfind_ext4;
3465         }
3466
3467         /* Precompute checksum seed for all metadata */
3468         if (ext4_has_feature_csum_seed(sb))
3469                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3470         else if (ext4_has_metadata_csum(sb))
3471                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3472                                                sizeof(es->s_uuid));
3473
3474         /* Set defaults before we parse the mount options */
3475         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3476         set_opt(sb, INIT_INODE_TABLE);
3477         if (def_mount_opts & EXT4_DEFM_DEBUG)
3478                 set_opt(sb, DEBUG);
3479         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3480                 set_opt(sb, GRPID);
3481         if (def_mount_opts & EXT4_DEFM_UID16)
3482                 set_opt(sb, NO_UID32);
3483         /* xattr user namespace & acls are now defaulted on */
3484         set_opt(sb, XATTR_USER);
3485 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3486         set_opt(sb, POSIX_ACL);
3487 #endif
3488         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3489         if (ext4_has_metadata_csum(sb))
3490                 set_opt(sb, JOURNAL_CHECKSUM);
3491
3492         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3493                 set_opt(sb, JOURNAL_DATA);
3494         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3495                 set_opt(sb, ORDERED_DATA);
3496         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3497                 set_opt(sb, WRITEBACK_DATA);
3498
3499         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3500                 set_opt(sb, ERRORS_PANIC);
3501         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3502                 set_opt(sb, ERRORS_CONT);
3503         else
3504                 set_opt(sb, ERRORS_RO);
3505         /* block_validity enabled by default; disable with noblock_validity */
3506         set_opt(sb, BLOCK_VALIDITY);
3507         if (def_mount_opts & EXT4_DEFM_DISCARD)
3508                 set_opt(sb, DISCARD);
3509
3510         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3511         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3512         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3513         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3514         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3515
3516         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3517                 set_opt(sb, BARRIER);
3518
3519         /*
3520          * enable delayed allocation by default
3521          * Use -o nodelalloc to turn it off
3522          */
3523         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3524             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3525                 set_opt(sb, DELALLOC);
3526
3527         /*
3528          * set default s_li_wait_mult for lazyinit, for the case there is
3529          * no mount option specified.
3530          */
3531         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3532
3533         if (sbi->s_es->s_mount_opts[0]) {
3534                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3535                                               sizeof(sbi->s_es->s_mount_opts),
3536                                               GFP_KERNEL);
3537                 if (!s_mount_opts)
3538                         goto failed_mount;
3539                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3540                                    &journal_ioprio, 0)) {
3541                         ext4_msg(sb, KERN_WARNING,
3542                                  "failed to parse options in superblock: %s",
3543                                  s_mount_opts);
3544                 }
3545                 kfree(s_mount_opts);
3546         }
3547         sbi->s_def_mount_opt = sbi->s_mount_opt;
3548         if (!parse_options((char *) data, sb, &journal_devnum,
3549                            &journal_ioprio, 0))
3550                 goto failed_mount;
3551
3552         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3553                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3554                             "with data=journal disables delayed "
3555                             "allocation and O_DIRECT support!\n");
3556                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3557                         ext4_msg(sb, KERN_ERR, "can't mount with "
3558                                  "both data=journal and delalloc");
3559                         goto failed_mount;
3560                 }
3561                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3562                         ext4_msg(sb, KERN_ERR, "can't mount with "
3563                                  "both data=journal and dioread_nolock");
3564                         goto failed_mount;
3565                 }
3566                 if (test_opt(sb, DAX)) {
3567                         ext4_msg(sb, KERN_ERR, "can't mount with "
3568                                  "both data=journal and dax");
3569                         goto failed_mount;
3570                 }
3571                 if (ext4_has_feature_encrypt(sb)) {
3572                         ext4_msg(sb, KERN_WARNING,
3573                                  "encrypted files will use data=ordered "
3574                                  "instead of data journaling mode");
3575                 }
3576                 if (test_opt(sb, DELALLOC))
3577                         clear_opt(sb, DELALLOC);
3578         } else {
3579                 sb->s_iflags |= SB_I_CGROUPWB;
3580         }
3581
3582         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3583                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3584
3585         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3586             (ext4_has_compat_features(sb) ||
3587              ext4_has_ro_compat_features(sb) ||
3588              ext4_has_incompat_features(sb)))
3589                 ext4_msg(sb, KERN_WARNING,
3590                        "feature flags set on rev 0 fs, "
3591                        "running e2fsck is recommended");
3592
3593         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3594                 set_opt2(sb, HURD_COMPAT);
3595                 if (ext4_has_feature_64bit(sb)) {
3596                         ext4_msg(sb, KERN_ERR,
3597                                  "The Hurd can't support 64-bit file systems");
3598                         goto failed_mount;
3599                 }
3600         }
3601
3602         if (IS_EXT2_SB(sb)) {
3603                 if (ext2_feature_set_ok(sb))
3604                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3605                                  "using the ext4 subsystem");
3606                 else {
3607                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3608                                  "to feature incompatibilities");
3609                         goto failed_mount;
3610                 }
3611         }
3612
3613         if (IS_EXT3_SB(sb)) {
3614                 if (ext3_feature_set_ok(sb))
3615                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3616                                  "using the ext4 subsystem");
3617                 else {
3618                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3619                                  "to feature incompatibilities");
3620                         goto failed_mount;
3621                 }
3622         }
3623
3624         /*
3625          * Check feature flags regardless of the revision level, since we
3626          * previously didn't change the revision level when setting the flags,
3627          * so there is a chance incompat flags are set on a rev 0 filesystem.
3628          */
3629         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3630                 goto failed_mount;
3631
3632         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3633         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3634             blocksize > EXT4_MAX_BLOCK_SIZE) {
3635                 ext4_msg(sb, KERN_ERR,
3636                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3637                          blocksize, le32_to_cpu(es->s_log_block_size));
3638                 goto failed_mount;
3639         }
3640         if (le32_to_cpu(es->s_log_block_size) >
3641             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3642                 ext4_msg(sb, KERN_ERR,
3643                          "Invalid log block size: %u",
3644                          le32_to_cpu(es->s_log_block_size));
3645                 goto failed_mount;
3646         }
3647
3648         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3649                 ext4_msg(sb, KERN_ERR,
3650                          "Number of reserved GDT blocks insanely large: %d",
3651                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3652                 goto failed_mount;
3653         }
3654
3655         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3656                 err = bdev_dax_supported(sb, blocksize);
3657                 if (err)
3658                         goto failed_mount;
3659         }
3660
3661         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3662                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3663                          es->s_encryption_level);
3664                 goto failed_mount;
3665         }
3666
3667         if (sb->s_blocksize != blocksize) {
3668                 /* Validate the filesystem blocksize */
3669                 if (!sb_set_blocksize(sb, blocksize)) {
3670                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3671                                         blocksize);
3672                         goto failed_mount;
3673                 }
3674
3675                 brelse(bh);
3676                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3677                 offset = do_div(logical_sb_block, blocksize);
3678                 bh = sb_bread_unmovable(sb, logical_sb_block);
3679                 if (!bh) {
3680                         ext4_msg(sb, KERN_ERR,
3681                                "Can't read superblock on 2nd try");
3682                         goto failed_mount;
3683                 }
3684                 es = (struct ext4_super_block *)(bh->b_data + offset);
3685                 sbi->s_es = es;
3686                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3687                         ext4_msg(sb, KERN_ERR,
3688                                "Magic mismatch, very weird!");
3689                         goto failed_mount;
3690                 }
3691         }
3692
3693         has_huge_files = ext4_has_feature_huge_file(sb);
3694         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3695                                                       has_huge_files);
3696         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3697
3698         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3699                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3700                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3701         } else {
3702                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3703                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3704                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3705                     (!is_power_of_2(sbi->s_inode_size)) ||
3706                     (sbi->s_inode_size > blocksize)) {
3707                         ext4_msg(sb, KERN_ERR,
3708                                "unsupported inode size: %d",
3709                                sbi->s_inode_size);
3710                         goto failed_mount;
3711                 }
3712                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3713                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3714         }
3715
3716         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3717         if (ext4_has_feature_64bit(sb)) {
3718                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3719                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3720                     !is_power_of_2(sbi->s_desc_size)) {
3721                         ext4_msg(sb, KERN_ERR,
3722                                "unsupported descriptor size %lu",
3723                                sbi->s_desc_size);
3724                         goto failed_mount;
3725                 }
3726         } else
3727                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3728
3729         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3730         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3731
3732         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3733         if (sbi->s_inodes_per_block == 0)
3734                 goto cantfind_ext4;
3735         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3736             sbi->s_inodes_per_group > blocksize * 8) {
3737                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3738                          sbi->s_blocks_per_group);
3739                 goto failed_mount;
3740         }
3741         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3742                                         sbi->s_inodes_per_block;
3743         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3744         sbi->s_sbh = bh;
3745         sbi->s_mount_state = le16_to_cpu(es->s_state);
3746         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3747         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3748
3749         for (i = 0; i < 4; i++)
3750                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3751         sbi->s_def_hash_version = es->s_def_hash_version;
3752         if (ext4_has_feature_dir_index(sb)) {
3753                 i = le32_to_cpu(es->s_flags);
3754                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3755                         sbi->s_hash_unsigned = 3;
3756                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3757 #ifdef __CHAR_UNSIGNED__
3758                         if (!(sb->s_flags & MS_RDONLY))
3759                                 es->s_flags |=
3760                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3761                         sbi->s_hash_unsigned = 3;
3762 #else
3763                         if (!(sb->s_flags & MS_RDONLY))
3764                                 es->s_flags |=
3765                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3766 #endif
3767                 }
3768         }
3769
3770         /* Handle clustersize */
3771         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3772         has_bigalloc = ext4_has_feature_bigalloc(sb);
3773         if (has_bigalloc) {
3774                 if (clustersize < blocksize) {
3775                         ext4_msg(sb, KERN_ERR,
3776                                  "cluster size (%d) smaller than "
3777                                  "block size (%d)", clustersize, blocksize);
3778                         goto failed_mount;
3779                 }
3780                 if (le32_to_cpu(es->s_log_cluster_size) >
3781                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3782                         ext4_msg(sb, KERN_ERR,
3783                                  "Invalid log cluster size: %u",
3784                                  le32_to_cpu(es->s_log_cluster_size));
3785                         goto failed_mount;
3786                 }
3787                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3788                         le32_to_cpu(es->s_log_block_size);
3789                 sbi->s_clusters_per_group =
3790                         le32_to_cpu(es->s_clusters_per_group);
3791                 if (sbi->s_clusters_per_group > blocksize * 8) {
3792                         ext4_msg(sb, KERN_ERR,
3793                                  "#clusters per group too big: %lu",
3794                                  sbi->s_clusters_per_group);
3795                         goto failed_mount;
3796                 }
3797                 if (sbi->s_blocks_per_group !=
3798                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3799                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3800                                  "clusters per group (%lu) inconsistent",
3801                                  sbi->s_blocks_per_group,
3802                                  sbi->s_clusters_per_group);
3803                         goto failed_mount;
3804                 }
3805         } else {
3806                 if (clustersize != blocksize) {
3807                         ext4_warning(sb, "fragment/cluster size (%d) != "
3808                                      "block size (%d)", clustersize,
3809                                      blocksize);
3810                         clustersize = blocksize;
3811                 }
3812                 if (sbi->s_blocks_per_group > blocksize * 8) {
3813                         ext4_msg(sb, KERN_ERR,
3814                                  "#blocks per group too big: %lu",
3815                                  sbi->s_blocks_per_group);
3816                         goto failed_mount;
3817                 }
3818                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3819                 sbi->s_cluster_bits = 0;
3820         }
3821         sbi->s_cluster_ratio = clustersize / blocksize;
3822
3823         /* Do we have standard group size of clustersize * 8 blocks ? */
3824         if (sbi->s_blocks_per_group == clustersize << 3)
3825                 set_opt2(sb, STD_GROUP_SIZE);
3826
3827         /*
3828          * Test whether we have more sectors than will fit in sector_t,
3829          * and whether the max offset is addressable by the page cache.
3830          */
3831         err = generic_check_addressable(sb->s_blocksize_bits,
3832                                         ext4_blocks_count(es));
3833         if (err) {
3834                 ext4_msg(sb, KERN_ERR, "filesystem"
3835                          " too large to mount safely on this system");
3836                 if (sizeof(sector_t) < 8)
3837                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3838                 goto failed_mount;
3839         }
3840
3841         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3842                 goto cantfind_ext4;
3843
3844         /* check blocks count against device size */
3845         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3846         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3847                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3848                        "exceeds size of device (%llu blocks)",
3849                        ext4_blocks_count(es), blocks_count);
3850                 goto failed_mount;
3851         }
3852
3853         /*
3854          * It makes no sense for the first data block to be beyond the end
3855          * of the filesystem.
3856          */
3857         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3858                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3859                          "block %u is beyond end of filesystem (%llu)",
3860                          le32_to_cpu(es->s_first_data_block),
3861                          ext4_blocks_count(es));
3862                 goto failed_mount;
3863         }
3864         blocks_count = (ext4_blocks_count(es) -
3865                         le32_to_cpu(es->s_first_data_block) +
3866                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3867         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3868         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3869                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3870                        "(block count %llu, first data block %u, "
3871                        "blocks per group %lu)", sbi->s_groups_count,
3872                        ext4_blocks_count(es),
3873                        le32_to_cpu(es->s_first_data_block),
3874                        EXT4_BLOCKS_PER_GROUP(sb));
3875                 goto failed_mount;
3876         }
3877         sbi->s_groups_count = blocks_count;
3878         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3879                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3880         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3881                    EXT4_DESC_PER_BLOCK(sb);
3882         if (ext4_has_feature_meta_bg(sb)) {
3883                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3884                         ext4_msg(sb, KERN_WARNING,
3885                                  "first meta block group too large: %u "
3886                                  "(group descriptor block count %u)",
3887                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3888                         goto failed_mount;
3889                 }
3890         }
3891         sbi->s_group_desc = kvmalloc(db_count *
3892                                           sizeof(struct buffer_head *),
3893                                           GFP_KERNEL);
3894         if (sbi->s_group_desc == NULL) {
3895                 ext4_msg(sb, KERN_ERR, "not enough memory");
3896                 ret = -ENOMEM;
3897                 goto failed_mount;
3898         }
3899
3900         bgl_lock_init(sbi->s_blockgroup_lock);
3901
3902         /* Pre-read the descriptors into the buffer cache */
3903         for (i = 0; i < db_count; i++) {
3904                 block = descriptor_loc(sb, logical_sb_block, i);
3905                 sb_breadahead(sb, block);
3906         }
3907
3908         for (i = 0; i < db_count; i++) {
3909                 block = descriptor_loc(sb, logical_sb_block, i);
3910                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3911                 if (!sbi->s_group_desc[i]) {
3912                         ext4_msg(sb, KERN_ERR,
3913                                "can't read group descriptor %d", i);
3914                         db_count = i;
3915                         goto failed_mount2;
3916                 }
3917         }
3918         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3919                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3920                 ret = -EFSCORRUPTED;
3921                 goto failed_mount2;
3922         }
3923
3924         sbi->s_gdb_count = db_count;
3925         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3926         spin_lock_init(&sbi->s_next_gen_lock);
3927
3928         setup_timer(&sbi->s_err_report, print_daily_error_info,
3929                 (unsigned long) sb);
3930
3931         /* Register extent status tree shrinker */
3932         if (ext4_es_register_shrinker(sbi))
3933                 goto failed_mount3;
3934
3935         sbi->s_stripe = ext4_get_stripe_size(sbi);
3936         sbi->s_extent_max_zeroout_kb = 32;
3937
3938         /*
3939          * set up enough so that it can read an inode
3940          */
3941         sb->s_op = &ext4_sops;
3942         sb->s_export_op = &ext4_export_ops;
3943         sb->s_xattr = ext4_xattr_handlers;
3944         sb->s_cop = &ext4_cryptops;
3945 #ifdef CONFIG_QUOTA
3946         sb->dq_op = &ext4_quota_operations;
3947         if (ext4_has_feature_quota(sb))
3948                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3949         else
3950                 sb->s_qcop = &ext4_qctl_operations;
3951         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3952 #endif
3953         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3954
3955         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3956         mutex_init(&sbi->s_orphan_lock);
3957
3958         sb->s_root = NULL;
3959
3960         needs_recovery = (es->s_last_orphan != 0 ||
3961                           ext4_has_feature_journal_needs_recovery(sb));
3962
3963         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3964                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3965                         goto failed_mount3a;
3966
3967         /*
3968          * The first inode we look at is the journal inode.  Don't try
3969          * root first: it may be modified in the journal!
3970          */
3971         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3972                 err = ext4_load_journal(sb, es, journal_devnum);
3973                 if (err)
3974                         goto failed_mount3a;
3975         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3976                    ext4_has_feature_journal_needs_recovery(sb)) {
3977                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3978                        "suppressed and not mounted read-only");
3979                 goto failed_mount_wq;
3980         } else {
3981                 /* Nojournal mode, all journal mount options are illegal */
3982                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3983                         ext4_msg(sb, KERN_ERR, "can't mount with "
3984                                  "journal_checksum, fs mounted w/o journal");
3985                         goto failed_mount_wq;
3986                 }
3987                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3988                         ext4_msg(sb, KERN_ERR, "can't mount with "
3989                                  "journal_async_commit, fs mounted w/o journal");
3990                         goto failed_mount_wq;
3991                 }
3992                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3993                         ext4_msg(sb, KERN_ERR, "can't mount with "
3994                                  "commit=%lu, fs mounted w/o journal",
3995                                  sbi->s_commit_interval / HZ);
3996                         goto failed_mount_wq;
3997                 }
3998                 if (EXT4_MOUNT_DATA_FLAGS &
3999                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4000                         ext4_msg(sb, KERN_ERR, "can't mount with "
4001                                  "data=, fs mounted w/o journal");
4002                         goto failed_mount_wq;
4003                 }
4004                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4005                 clear_opt(sb, JOURNAL_CHECKSUM);
4006                 clear_opt(sb, DATA_FLAGS);
4007                 sbi->s_journal = NULL;
4008                 needs_recovery = 0;
4009                 goto no_journal;
4010         }
4011
4012         if (ext4_has_feature_64bit(sb) &&
4013             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4014                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4015                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4016                 goto failed_mount_wq;
4017         }
4018
4019         if (!set_journal_csum_feature_set(sb)) {
4020                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4021                          "feature set");
4022                 goto failed_mount_wq;
4023         }
4024
4025         /* We have now updated the journal if required, so we can
4026          * validate the data journaling mode. */
4027         switch (test_opt(sb, DATA_FLAGS)) {
4028         case 0:
4029                 /* No mode set, assume a default based on the journal
4030                  * capabilities: ORDERED_DATA if the journal can
4031                  * cope, else JOURNAL_DATA
4032                  */
4033                 if (jbd2_journal_check_available_features
4034                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4035                         set_opt(sb, ORDERED_DATA);
4036                 else
4037                         set_opt(sb, JOURNAL_DATA);
4038                 break;
4039
4040         case EXT4_MOUNT_ORDERED_DATA:
4041         case EXT4_MOUNT_WRITEBACK_DATA:
4042                 if (!jbd2_journal_check_available_features
4043                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4044                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4045                                "requested data journaling mode");
4046                         goto failed_mount_wq;
4047                 }
4048         default:
4049                 break;
4050         }
4051
4052         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4053             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4054                 ext4_msg(sb, KERN_ERR, "can't mount with "
4055                         "journal_async_commit in data=ordered mode");
4056                 goto failed_mount_wq;
4057         }
4058
4059         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4060
4061         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4062
4063 no_journal:
4064         sbi->s_mb_cache = ext4_xattr_create_cache();
4065         if (!sbi->s_mb_cache) {
4066                 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4067                 goto failed_mount_wq;
4068         }
4069
4070         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4071             (blocksize != PAGE_SIZE)) {
4072                 ext4_msg(sb, KERN_ERR,
4073                          "Unsupported blocksize for fs encryption");
4074                 goto failed_mount_wq;
4075         }
4076
4077         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
4078             !ext4_has_feature_encrypt(sb)) {
4079                 ext4_set_feature_encrypt(sb);
4080                 ext4_commit_super(sb, 1);
4081         }
4082
4083         /*
4084          * Get the # of file system overhead blocks from the
4085          * superblock if present.
4086          */
4087         if (es->s_overhead_clusters)
4088                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4089         else {
4090                 err = ext4_calculate_overhead(sb);
4091                 if (err)
4092                         goto failed_mount_wq;
4093         }
4094
4095         /*
4096          * The maximum number of concurrent works can be high and
4097          * concurrency isn't really necessary.  Limit it to 1.
4098          */
4099         EXT4_SB(sb)->rsv_conversion_wq =
4100                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4101         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4102                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4103                 ret = -ENOMEM;
4104                 goto failed_mount4;
4105         }
4106
4107         /*
4108          * The jbd2_journal_load will have done any necessary log recovery,
4109          * so we can safely mount the rest of the filesystem now.
4110          */
4111
4112         root = ext4_iget(sb, EXT4_ROOT_INO);
4113         if (IS_ERR(root)) {
4114                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4115                 ret = PTR_ERR(root);
4116                 root = NULL;
4117                 goto failed_mount4;
4118         }
4119         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4120                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4121                 iput(root);
4122                 goto failed_mount4;
4123         }
4124         sb->s_root = d_make_root(root);
4125         if (!sb->s_root) {
4126                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4127                 ret = -ENOMEM;
4128                 goto failed_mount4;
4129         }
4130
4131         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4132                 sb->s_flags |= MS_RDONLY;
4133
4134         /* determine the minimum size of new large inodes, if present */
4135         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4136             sbi->s_want_extra_isize == 0) {
4137                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4138                                                      EXT4_GOOD_OLD_INODE_SIZE;
4139                 if (ext4_has_feature_extra_isize(sb)) {
4140                         if (sbi->s_want_extra_isize <
4141                             le16_to_cpu(es->s_want_extra_isize))
4142                                 sbi->s_want_extra_isize =
4143                                         le16_to_cpu(es->s_want_extra_isize);
4144                         if (sbi->s_want_extra_isize <
4145                             le16_to_cpu(es->s_min_extra_isize))
4146                                 sbi->s_want_extra_isize =
4147                                         le16_to_cpu(es->s_min_extra_isize);
4148                 }
4149         }
4150         /* Check if enough inode space is available */
4151         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4152                                                         sbi->s_inode_size) {
4153                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4154                                                        EXT4_GOOD_OLD_INODE_SIZE;
4155                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4156                          "available");
4157         }
4158
4159         ext4_set_resv_clusters(sb);
4160
4161         err = ext4_setup_system_zone(sb);
4162         if (err) {
4163                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4164                          "zone (%d)", err);
4165                 goto failed_mount4a;
4166         }
4167
4168         ext4_ext_init(sb);
4169         err = ext4_mb_init(sb);
4170         if (err) {
4171                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4172                          err);
4173                 goto failed_mount5;
4174         }
4175
4176         block = ext4_count_free_clusters(sb);
4177         ext4_free_blocks_count_set(sbi->s_es,
4178                                    EXT4_C2B(sbi, block));
4179         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4180                                   GFP_KERNEL);
4181         if (!err) {
4182                 unsigned long freei = ext4_count_free_inodes(sb);
4183                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4184                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4185                                           GFP_KERNEL);
4186         }
4187         if (!err)
4188                 err = percpu_counter_init(&sbi->s_dirs_counter,
4189                                           ext4_count_dirs(sb), GFP_KERNEL);
4190         if (!err)
4191                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4192                                           GFP_KERNEL);
4193         if (!err)
4194                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4195
4196         if (err) {
4197                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4198                 goto failed_mount6;
4199         }
4200
4201         if (ext4_has_feature_flex_bg(sb))
4202                 if (!ext4_fill_flex_info(sb)) {
4203                         ext4_msg(sb, KERN_ERR,
4204                                "unable to initialize "
4205                                "flex_bg meta info!");
4206                         goto failed_mount6;
4207                 }
4208
4209         err = ext4_register_li_request(sb, first_not_zeroed);
4210         if (err)
4211                 goto failed_mount6;
4212
4213         err = ext4_register_sysfs(sb);
4214         if (err)
4215                 goto failed_mount7;
4216
4217 #ifdef CONFIG_QUOTA
4218         /* Enable quota usage during mount. */
4219         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4220                 err = ext4_enable_quotas(sb);
4221                 if (err)
4222                         goto failed_mount8;
4223         }
4224 #endif  /* CONFIG_QUOTA */
4225
4226         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4227         ext4_orphan_cleanup(sb, es);
4228         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4229         if (needs_recovery) {
4230                 ext4_msg(sb, KERN_INFO, "recovery complete");
4231                 ext4_mark_recovery_complete(sb, es);
4232         }
4233         if (EXT4_SB(sb)->s_journal) {
4234                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4235                         descr = " journalled data mode";
4236                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4237                         descr = " ordered data mode";
4238                 else
4239                         descr = " writeback data mode";
4240         } else
4241                 descr = "out journal";
4242
4243         if (test_opt(sb, DISCARD)) {
4244                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4245                 if (!blk_queue_discard(q))
4246                         ext4_msg(sb, KERN_WARNING,
4247                                  "mounting with \"discard\" option, but "
4248                                  "the device does not support discard");
4249         }
4250
4251         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4252                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4253                          "Opts: %.*s%s%s", descr,
4254                          (int) sizeof(sbi->s_es->s_mount_opts),
4255                          sbi->s_es->s_mount_opts,
4256                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4257
4258         if (es->s_error_count)
4259                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4260
4261         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4262         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4263         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4264         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4265
4266         kfree(orig_data);
4267         return 0;
4268
4269 cantfind_ext4:
4270         if (!silent)
4271                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4272         goto failed_mount;
4273
4274 #ifdef CONFIG_QUOTA
4275 failed_mount8:
4276         ext4_unregister_sysfs(sb);
4277 #endif
4278 failed_mount7:
4279         ext4_unregister_li_request(sb);
4280 failed_mount6:
4281         ext4_mb_release(sb);
4282         if (sbi->s_flex_groups)
4283                 kvfree(sbi->s_flex_groups);
4284         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4285         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4286         percpu_counter_destroy(&sbi->s_dirs_counter);
4287         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4288 failed_mount5:
4289         ext4_ext_release(sb);
4290         ext4_release_system_zone(sb);
4291 failed_mount4a:
4292         dput(sb->s_root);
4293         sb->s_root = NULL;
4294 failed_mount4:
4295         ext4_msg(sb, KERN_ERR, "mount failed");
4296         if (EXT4_SB(sb)->rsv_conversion_wq)
4297                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4298 failed_mount_wq:
4299         if (sbi->s_mb_cache) {
4300                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4301                 sbi->s_mb_cache = NULL;
4302         }
4303         if (sbi->s_journal) {
4304                 jbd2_journal_destroy(sbi->s_journal);
4305                 sbi->s_journal = NULL;
4306         }
4307 failed_mount3a:
4308         ext4_es_unregister_shrinker(sbi);
4309 failed_mount3:
4310         del_timer_sync(&sbi->s_err_report);
4311         if (sbi->s_mmp_tsk)
4312                 kthread_stop(sbi->s_mmp_tsk);
4313 failed_mount2:
4314         for (i = 0; i < db_count; i++)
4315                 brelse(sbi->s_group_desc[i]);
4316         kvfree(sbi->s_group_desc);
4317 failed_mount:
4318         if (sbi->s_chksum_driver)
4319                 crypto_free_shash(sbi->s_chksum_driver);
4320 #ifdef CONFIG_QUOTA
4321         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4322                 kfree(sbi->s_qf_names[i]);
4323 #endif
4324         ext4_blkdev_remove(sbi);
4325         brelse(bh);
4326 out_fail:
4327         sb->s_fs_info = NULL;
4328         kfree(sbi->s_blockgroup_lock);
4329 out_free_base:
4330         kfree(sbi);
4331         kfree(orig_data);
4332         return err ? err : ret;
4333 }
4334
4335 /*
4336  * Setup any per-fs journal parameters now.  We'll do this both on
4337  * initial mount, once the journal has been initialised but before we've
4338  * done any recovery; and again on any subsequent remount.
4339  */
4340 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4341 {
4342         struct ext4_sb_info *sbi = EXT4_SB(sb);
4343
4344         journal->j_commit_interval = sbi->s_commit_interval;
4345         journal->j_min_batch_time = sbi->s_min_batch_time;
4346         journal->j_max_batch_time = sbi->s_max_batch_time;
4347
4348         write_lock(&journal->j_state_lock);
4349         if (test_opt(sb, BARRIER))
4350                 journal->j_flags |= JBD2_BARRIER;
4351         else
4352                 journal->j_flags &= ~JBD2_BARRIER;
4353         if (test_opt(sb, DATA_ERR_ABORT))
4354                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4355         else
4356                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4357         write_unlock(&journal->j_state_lock);
4358 }
4359
4360 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4361                                              unsigned int journal_inum)
4362 {
4363         struct inode *journal_inode;
4364
4365         /*
4366          * Test for the existence of a valid inode on disk.  Bad things
4367          * happen if we iget() an unused inode, as the subsequent iput()
4368          * will try to delete it.
4369          */
4370         journal_inode = ext4_iget(sb, journal_inum);
4371         if (IS_ERR(journal_inode)) {
4372                 ext4_msg(sb, KERN_ERR, "no journal found");
4373                 return NULL;
4374         }
4375         if (!journal_inode->i_nlink) {
4376                 make_bad_inode(journal_inode);
4377                 iput(journal_inode);
4378                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4379                 return NULL;
4380         }
4381
4382         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4383                   journal_inode, journal_inode->i_size);
4384         if (!S_ISREG(journal_inode->i_mode)) {
4385                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4386                 iput(journal_inode);
4387                 return NULL;
4388         }
4389         return journal_inode;
4390 }
4391
4392 static journal_t *ext4_get_journal(struct super_block *sb,
4393                                    unsigned int journal_inum)
4394 {
4395         struct inode *journal_inode;
4396         journal_t *journal;
4397
4398         BUG_ON(!ext4_has_feature_journal(sb));
4399
4400         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4401         if (!journal_inode)
4402                 return NULL;
4403
4404         journal = jbd2_journal_init_inode(journal_inode);
4405         if (!journal) {
4406                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4407                 iput(journal_inode);
4408                 return NULL;
4409         }
4410         journal->j_private = sb;
4411         ext4_init_journal_params(sb, journal);
4412         return journal;
4413 }
4414
4415 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4416                                        dev_t j_dev)
4417 {
4418         struct buffer_head *bh;
4419         journal_t *journal;
4420         ext4_fsblk_t start;
4421         ext4_fsblk_t len;
4422         int hblock, blocksize;
4423         ext4_fsblk_t sb_block;
4424         unsigned long offset;
4425         struct ext4_super_block *es;
4426         struct block_device *bdev;
4427
4428         BUG_ON(!ext4_has_feature_journal(sb));
4429
4430         bdev = ext4_blkdev_get(j_dev, sb);
4431         if (bdev == NULL)
4432                 return NULL;
4433
4434         blocksize = sb->s_blocksize;
4435         hblock = bdev_logical_block_size(bdev);
4436         if (blocksize < hblock) {
4437                 ext4_msg(sb, KERN_ERR,
4438                         "blocksize too small for journal device");
4439                 goto out_bdev;
4440         }
4441
4442         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4443         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4444         set_blocksize(bdev, blocksize);
4445         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4446                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4447                        "external journal");
4448                 goto out_bdev;
4449         }
4450
4451         es = (struct ext4_super_block *) (bh->b_data + offset);
4452         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4453             !(le32_to_cpu(es->s_feature_incompat) &
4454               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4455                 ext4_msg(sb, KERN_ERR, "external journal has "
4456                                         "bad superblock");
4457                 brelse(bh);
4458                 goto out_bdev;
4459         }
4460
4461         if ((le32_to_cpu(es->s_feature_ro_compat) &
4462              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4463             es->s_checksum != ext4_superblock_csum(sb, es)) {
4464                 ext4_msg(sb, KERN_ERR, "external journal has "
4465                                        "corrupt superblock");
4466                 brelse(bh);
4467                 goto out_bdev;
4468         }
4469
4470         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4471                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4472                 brelse(bh);
4473                 goto out_bdev;
4474         }
4475
4476         len = ext4_blocks_count(es);
4477         start = sb_block + 1;
4478         brelse(bh);     /* we're done with the superblock */
4479
4480         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4481                                         start, len, blocksize);
4482         if (!journal) {
4483                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4484                 goto out_bdev;
4485         }
4486         journal->j_private = sb;
4487         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4488         wait_on_buffer(journal->j_sb_buffer);
4489         if (!buffer_uptodate(journal->j_sb_buffer)) {
4490                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4491                 goto out_journal;
4492         }
4493         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4494                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4495                                         "user (unsupported) - %d",
4496                         be32_to_cpu(journal->j_superblock->s_nr_users));
4497                 goto out_journal;
4498         }
4499         EXT4_SB(sb)->journal_bdev = bdev;
4500         ext4_init_journal_params(sb, journal);
4501         return journal;
4502
4503 out_journal:
4504         jbd2_journal_destroy(journal);
4505 out_bdev:
4506         ext4_blkdev_put(bdev);
4507         return NULL;
4508 }
4509
4510 static int ext4_load_journal(struct super_block *sb,
4511                              struct ext4_super_block *es,
4512                              unsigned long journal_devnum)
4513 {
4514         journal_t *journal;
4515         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4516         dev_t journal_dev;
4517         int err = 0;
4518         int really_read_only;
4519
4520         BUG_ON(!ext4_has_feature_journal(sb));
4521
4522         if (journal_devnum &&
4523             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4524                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4525                         "numbers have changed");
4526                 journal_dev = new_decode_dev(journal_devnum);
4527         } else
4528                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4529
4530         really_read_only = bdev_read_only(sb->s_bdev);
4531
4532         /*
4533          * Are we loading a blank journal or performing recovery after a
4534          * crash?  For recovery, we need to check in advance whether we
4535          * can get read-write access to the device.
4536          */
4537         if (ext4_has_feature_journal_needs_recovery(sb)) {
4538                 if (sb->s_flags & MS_RDONLY) {
4539                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4540                                         "required on readonly filesystem");
4541                         if (really_read_only) {
4542                                 ext4_msg(sb, KERN_ERR, "write access "
4543                                         "unavailable, cannot proceed");
4544                                 return -EROFS;
4545                         }
4546                         ext4_msg(sb, KERN_INFO, "write access will "
4547                                "be enabled during recovery");
4548                 }
4549         }
4550
4551         if (journal_inum && journal_dev) {
4552                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4553                        "and inode journals!");
4554                 return -EINVAL;
4555         }
4556
4557         if (journal_inum) {
4558                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4559                         return -EINVAL;
4560         } else {
4561                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4562                         return -EINVAL;
4563         }
4564
4565         if (!(journal->j_flags & JBD2_BARRIER))
4566                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4567
4568         if (!ext4_has_feature_journal_needs_recovery(sb))
4569                 err = jbd2_journal_wipe(journal, !really_read_only);
4570         if (!err) {
4571                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4572                 if (save)
4573                         memcpy(save, ((char *) es) +
4574                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4575                 err = jbd2_journal_load(journal);
4576                 if (save)
4577                         memcpy(((char *) es) + EXT4_S_ERR_START,
4578                                save, EXT4_S_ERR_LEN);
4579                 kfree(save);
4580         }
4581
4582         if (err) {
4583                 ext4_msg(sb, KERN_ERR, "error loading journal");
4584                 jbd2_journal_destroy(journal);
4585                 return err;
4586         }
4587
4588         EXT4_SB(sb)->s_journal = journal;
4589         ext4_clear_journal_err(sb, es);
4590
4591         if (!really_read_only && journal_devnum &&
4592             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4593                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4594
4595                 /* Make sure we flush the recovery flag to disk. */
4596                 ext4_commit_super(sb, 1);
4597         }
4598
4599         return 0;
4600 }
4601
4602 static int ext4_commit_super(struct super_block *sb, int sync)
4603 {
4604         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4605         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4606         int error = 0;
4607
4608         if (!sbh || block_device_ejected(sb))
4609                 return error;
4610         /*
4611          * If the file system is mounted read-only, don't update the
4612          * superblock write time.  This avoids updating the superblock
4613          * write time when we are mounting the root file system
4614          * read/only but we need to replay the journal; at that point,
4615          * for people who are east of GMT and who make their clock
4616          * tick in localtime for Windows bug-for-bug compatibility,
4617          * the clock is set in the future, and this will cause e2fsck
4618          * to complain and force a full file system check.
4619          */
4620         if (!(sb->s_flags & MS_RDONLY))
4621                 es->s_wtime = cpu_to_le32(get_seconds());
4622         if (sb->s_bdev->bd_part)
4623                 es->s_kbytes_written =
4624                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4625                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4626                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4627         else
4628                 es->s_kbytes_written =
4629                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4630         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4631                 ext4_free_blocks_count_set(es,
4632                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4633                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4634         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4635                 es->s_free_inodes_count =
4636                         cpu_to_le32(percpu_counter_sum_positive(
4637                                 &EXT4_SB(sb)->s_freeinodes_counter));
4638         BUFFER_TRACE(sbh, "marking dirty");
4639         ext4_superblock_csum_set(sb);
4640         if (sync)
4641                 lock_buffer(sbh);
4642         if (buffer_write_io_error(sbh)) {
4643                 /*
4644                  * Oh, dear.  A previous attempt to write the
4645                  * superblock failed.  This could happen because the
4646                  * USB device was yanked out.  Or it could happen to
4647                  * be a transient write error and maybe the block will
4648                  * be remapped.  Nothing we can do but to retry the
4649                  * write and hope for the best.
4650                  */
4651                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4652                        "superblock detected");
4653                 clear_buffer_write_io_error(sbh);
4654                 set_buffer_uptodate(sbh);
4655         }
4656         mark_buffer_dirty(sbh);
4657         if (sync) {
4658                 unlock_buffer(sbh);
4659                 error = __sync_dirty_buffer(sbh,
4660                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4661                 if (error)
4662                         return error;
4663
4664                 error = buffer_write_io_error(sbh);
4665                 if (error) {
4666                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4667                                "superblock");
4668                         clear_buffer_write_io_error(sbh);
4669                         set_buffer_uptodate(sbh);
4670                 }
4671         }
4672         return error;
4673 }
4674
4675 /*
4676  * Have we just finished recovery?  If so, and if we are mounting (or
4677  * remounting) the filesystem readonly, then we will end up with a
4678  * consistent fs on disk.  Record that fact.
4679  */
4680 static void ext4_mark_recovery_complete(struct super_block *sb,
4681                                         struct ext4_super_block *es)
4682 {
4683         journal_t *journal = EXT4_SB(sb)->s_journal;
4684
4685         if (!ext4_has_feature_journal(sb)) {
4686                 BUG_ON(journal != NULL);
4687                 return;
4688         }
4689         jbd2_journal_lock_updates(journal);
4690         if (jbd2_journal_flush(journal) < 0)
4691                 goto out;
4692
4693         if (ext4_has_feature_journal_needs_recovery(sb) &&
4694             sb->s_flags & MS_RDONLY) {
4695                 ext4_clear_feature_journal_needs_recovery(sb);
4696                 ext4_commit_super(sb, 1);
4697         }
4698
4699 out:
4700         jbd2_journal_unlock_updates(journal);
4701 }
4702
4703 /*
4704  * If we are mounting (or read-write remounting) a filesystem whose journal
4705  * has recorded an error from a previous lifetime, move that error to the
4706  * main filesystem now.
4707  */
4708 static void ext4_clear_journal_err(struct super_block *sb,
4709                                    struct ext4_super_block *es)
4710 {
4711         journal_t *journal;
4712         int j_errno;
4713         const char *errstr;
4714
4715         BUG_ON(!ext4_has_feature_journal(sb));
4716
4717         journal = EXT4_SB(sb)->s_journal;
4718
4719         /*
4720          * Now check for any error status which may have been recorded in the
4721          * journal by a prior ext4_error() or ext4_abort()
4722          */
4723
4724         j_errno = jbd2_journal_errno(journal);
4725         if (j_errno) {
4726                 char nbuf[16];
4727
4728                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4729                 ext4_warning(sb, "Filesystem error recorded "
4730                              "from previous mount: %s", errstr);
4731                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4732
4733                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4734                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4735                 ext4_commit_super(sb, 1);
4736
4737                 jbd2_journal_clear_err(journal);
4738                 jbd2_journal_update_sb_errno(journal);
4739         }
4740 }
4741
4742 /*
4743  * Force the running and committing transactions to commit,
4744  * and wait on the commit.
4745  */
4746 int ext4_force_commit(struct super_block *sb)
4747 {
4748         journal_t *journal;
4749
4750         if (sb->s_flags & MS_RDONLY)
4751                 return 0;
4752
4753         journal = EXT4_SB(sb)->s_journal;
4754         return ext4_journal_force_commit(journal);
4755 }
4756
4757 static int ext4_sync_fs(struct super_block *sb, int wait)
4758 {
4759         int ret = 0;
4760         tid_t target;
4761         bool needs_barrier = false;
4762         struct ext4_sb_info *sbi = EXT4_SB(sb);
4763
4764         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4765                 return 0;
4766
4767         trace_ext4_sync_fs(sb, wait);
4768         flush_workqueue(sbi->rsv_conversion_wq);
4769         /*
4770          * Writeback quota in non-journalled quota case - journalled quota has
4771          * no dirty dquots
4772          */
4773         dquot_writeback_dquots(sb, -1);
4774         /*
4775          * Data writeback is possible w/o journal transaction, so barrier must
4776          * being sent at the end of the function. But we can skip it if
4777          * transaction_commit will do it for us.
4778          */
4779         if (sbi->s_journal) {
4780                 target = jbd2_get_latest_transaction(sbi->s_journal);
4781                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4782                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4783                         needs_barrier = true;
4784
4785                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4786                         if (wait)
4787                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4788                                                            target);
4789                 }
4790         } else if (wait && test_opt(sb, BARRIER))
4791                 needs_barrier = true;
4792         if (needs_barrier) {
4793                 int err;
4794                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4795                 if (!ret)
4796                         ret = err;
4797         }
4798
4799         return ret;
4800 }
4801
4802 /*
4803  * LVM calls this function before a (read-only) snapshot is created.  This
4804  * gives us a chance to flush the journal completely and mark the fs clean.
4805  *
4806  * Note that only this function cannot bring a filesystem to be in a clean
4807  * state independently. It relies on upper layer to stop all data & metadata
4808  * modifications.
4809  */
4810 static int ext4_freeze(struct super_block *sb)
4811 {
4812         int error = 0;
4813         journal_t *journal;
4814
4815         if (sb->s_flags & MS_RDONLY)
4816                 return 0;
4817
4818         journal = EXT4_SB(sb)->s_journal;
4819
4820         if (journal) {
4821                 /* Now we set up the journal barrier. */
4822                 jbd2_journal_lock_updates(journal);
4823
4824                 /*
4825                  * Don't clear the needs_recovery flag if we failed to
4826                  * flush the journal.
4827                  */
4828                 error = jbd2_journal_flush(journal);
4829                 if (error < 0)
4830                         goto out;
4831
4832                 /* Journal blocked and flushed, clear needs_recovery flag. */
4833                 ext4_clear_feature_journal_needs_recovery(sb);
4834         }
4835
4836         error = ext4_commit_super(sb, 1);
4837 out:
4838         if (journal)
4839                 /* we rely on upper layer to stop further updates */
4840                 jbd2_journal_unlock_updates(journal);
4841         return error;
4842 }
4843
4844 /*
4845  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4846  * flag here, even though the filesystem is not technically dirty yet.
4847  */
4848 static int ext4_unfreeze(struct super_block *sb)
4849 {
4850         if ((sb->s_flags & MS_RDONLY) || ext4_forced_shutdown(EXT4_SB(sb)))
4851                 return 0;
4852
4853         if (EXT4_SB(sb)->s_journal) {
4854                 /* Reset the needs_recovery flag before the fs is unlocked. */
4855                 ext4_set_feature_journal_needs_recovery(sb);
4856         }
4857
4858         ext4_commit_super(sb, 1);
4859         return 0;
4860 }
4861
4862 /*
4863  * Structure to save mount options for ext4_remount's benefit
4864  */
4865 struct ext4_mount_options {
4866         unsigned long s_mount_opt;
4867         unsigned long s_mount_opt2;
4868         kuid_t s_resuid;
4869         kgid_t s_resgid;
4870         unsigned long s_commit_interval;
4871         u32 s_min_batch_time, s_max_batch_time;
4872 #ifdef CONFIG_QUOTA
4873         int s_jquota_fmt;
4874         char *s_qf_names[EXT4_MAXQUOTAS];
4875 #endif
4876 };
4877
4878 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4879 {
4880         struct ext4_super_block *es;
4881         struct ext4_sb_info *sbi = EXT4_SB(sb);
4882         unsigned long old_sb_flags;
4883         struct ext4_mount_options old_opts;
4884         int enable_quota = 0;
4885         ext4_group_t g;
4886         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4887         int err = 0;
4888 #ifdef CONFIG_QUOTA
4889         int i, j;
4890 #endif
4891         char *orig_data = kstrdup(data, GFP_KERNEL);
4892
4893         /* Store the original options */
4894         old_sb_flags = sb->s_flags;
4895         old_opts.s_mount_opt = sbi->s_mount_opt;
4896         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4897         old_opts.s_resuid = sbi->s_resuid;
4898         old_opts.s_resgid = sbi->s_resgid;
4899         old_opts.s_commit_interval = sbi->s_commit_interval;
4900         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4901         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4902 #ifdef CONFIG_QUOTA
4903         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4904         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4905                 if (sbi->s_qf_names[i]) {
4906                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4907                                                          GFP_KERNEL);
4908                         if (!old_opts.s_qf_names[i]) {
4909                                 for (j = 0; j < i; j++)
4910                                         kfree(old_opts.s_qf_names[j]);
4911                                 kfree(orig_data);
4912                                 return -ENOMEM;
4913                         }
4914                 } else
4915                         old_opts.s_qf_names[i] = NULL;
4916 #endif
4917         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4918                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4919
4920         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4921                 err = -EINVAL;
4922                 goto restore_opts;
4923         }
4924
4925         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4926             test_opt(sb, JOURNAL_CHECKSUM)) {
4927                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4928                          "during remount not supported; ignoring");
4929                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4930         }
4931
4932         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4933                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4934                         ext4_msg(sb, KERN_ERR, "can't mount with "
4935                                  "both data=journal and delalloc");
4936                         err = -EINVAL;
4937                         goto restore_opts;
4938                 }
4939                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4940                         ext4_msg(sb, KERN_ERR, "can't mount with "
4941                                  "both data=journal and dioread_nolock");
4942                         err = -EINVAL;
4943                         goto restore_opts;
4944                 }
4945                 if (test_opt(sb, DAX)) {
4946                         ext4_msg(sb, KERN_ERR, "can't mount with "
4947                                  "both data=journal and dax");
4948                         err = -EINVAL;
4949                         goto restore_opts;
4950                 }
4951         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
4952                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4953                         ext4_msg(sb, KERN_ERR, "can't mount with "
4954                                 "journal_async_commit in data=ordered mode");
4955                         err = -EINVAL;
4956                         goto restore_opts;
4957                 }
4958         }
4959
4960         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4961                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4962                         "dax flag with busy inodes while remounting");
4963                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4964         }
4965
4966         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4967                 ext4_abort(sb, "Abort forced by user");
4968
4969         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4970                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4971
4972         es = sbi->s_es;
4973
4974         if (sbi->s_journal) {
4975                 ext4_init_journal_params(sb, sbi->s_journal);
4976                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4977         }
4978
4979         if (*flags & MS_LAZYTIME)
4980                 sb->s_flags |= MS_LAZYTIME;
4981
4982         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4983                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4984                         err = -EROFS;
4985                         goto restore_opts;
4986                 }
4987
4988                 if (*flags & MS_RDONLY) {
4989                         err = sync_filesystem(sb);
4990                         if (err < 0)
4991                                 goto restore_opts;
4992                         err = dquot_suspend(sb, -1);
4993                         if (err < 0)
4994                                 goto restore_opts;
4995
4996                         /*
4997                          * First of all, the unconditional stuff we have to do
4998                          * to disable replay of the journal when we next remount
4999                          */
5000                         sb->s_flags |= MS_RDONLY;
5001
5002                         /*
5003                          * OK, test if we are remounting a valid rw partition
5004                          * readonly, and if so set the rdonly flag and then
5005                          * mark the partition as valid again.
5006                          */
5007                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5008                             (sbi->s_mount_state & EXT4_VALID_FS))
5009                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5010
5011                         if (sbi->s_journal)
5012                                 ext4_mark_recovery_complete(sb, es);
5013                 } else {
5014                         /* Make sure we can mount this feature set readwrite */
5015                         if (ext4_has_feature_readonly(sb) ||
5016                             !ext4_feature_set_ok(sb, 0)) {
5017                                 err = -EROFS;
5018                                 goto restore_opts;
5019                         }
5020                         /*
5021                          * Make sure the group descriptor checksums
5022                          * are sane.  If they aren't, refuse to remount r/w.
5023                          */
5024                         for (g = 0; g < sbi->s_groups_count; g++) {
5025                                 struct ext4_group_desc *gdp =
5026                                         ext4_get_group_desc(sb, g, NULL);
5027
5028                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5029                                         ext4_msg(sb, KERN_ERR,
5030                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5031                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5032                                                le16_to_cpu(gdp->bg_checksum));
5033                                         err = -EFSBADCRC;
5034                                         goto restore_opts;
5035                                 }
5036                         }
5037
5038                         /*
5039                          * If we have an unprocessed orphan list hanging
5040                          * around from a previously readonly bdev mount,
5041                          * require a full umount/remount for now.
5042                          */
5043                         if (es->s_last_orphan) {
5044                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5045                                        "remount RDWR because of unprocessed "
5046                                        "orphan inode list.  Please "
5047                                        "umount/remount instead");
5048                                 err = -EINVAL;
5049                                 goto restore_opts;
5050                         }
5051
5052                         /*
5053                          * Mounting a RDONLY partition read-write, so reread
5054                          * and store the current valid flag.  (It may have
5055                          * been changed by e2fsck since we originally mounted
5056                          * the partition.)
5057                          */
5058                         if (sbi->s_journal)
5059                                 ext4_clear_journal_err(sb, es);
5060                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5061                         if (!ext4_setup_super(sb, es, 0))
5062                                 sb->s_flags &= ~MS_RDONLY;
5063                         if (ext4_has_feature_mmp(sb))
5064                                 if (ext4_multi_mount_protect(sb,
5065                                                 le64_to_cpu(es->s_mmp_block))) {
5066                                         err = -EROFS;
5067                                         goto restore_opts;
5068                                 }
5069                         enable_quota = 1;
5070                 }
5071         }
5072
5073         /*
5074          * Reinitialize lazy itable initialization thread based on
5075          * current settings
5076          */
5077         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5078                 ext4_unregister_li_request(sb);
5079         else {
5080                 ext4_group_t first_not_zeroed;
5081                 first_not_zeroed = ext4_has_uninit_itable(sb);
5082                 ext4_register_li_request(sb, first_not_zeroed);
5083         }
5084
5085         ext4_setup_system_zone(sb);
5086         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5087                 ext4_commit_super(sb, 1);
5088
5089 #ifdef CONFIG_QUOTA
5090         /* Release old quota file names */
5091         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5092                 kfree(old_opts.s_qf_names[i]);
5093         if (enable_quota) {
5094                 if (sb_any_quota_suspended(sb))
5095                         dquot_resume(sb, -1);
5096                 else if (ext4_has_feature_quota(sb)) {
5097                         err = ext4_enable_quotas(sb);
5098                         if (err)
5099                                 goto restore_opts;
5100                 }
5101         }
5102 #endif
5103
5104         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5105         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5106         kfree(orig_data);
5107         return 0;
5108
5109 restore_opts:
5110         sb->s_flags = old_sb_flags;
5111         sbi->s_mount_opt = old_opts.s_mount_opt;
5112         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5113         sbi->s_resuid = old_opts.s_resuid;
5114         sbi->s_resgid = old_opts.s_resgid;
5115         sbi->s_commit_interval = old_opts.s_commit_interval;
5116         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5117         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5118 #ifdef CONFIG_QUOTA
5119         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5120         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5121                 kfree(sbi->s_qf_names[i]);
5122                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5123         }
5124 #endif
5125         kfree(orig_data);
5126         return err;
5127 }
5128
5129 #ifdef CONFIG_QUOTA
5130 static int ext4_statfs_project(struct super_block *sb,
5131                                kprojid_t projid, struct kstatfs *buf)
5132 {
5133         struct kqid qid;
5134         struct dquot *dquot;
5135         u64 limit;
5136         u64 curblock;
5137
5138         qid = make_kqid_projid(projid);
5139         dquot = dqget(sb, qid);
5140         if (IS_ERR(dquot))
5141                 return PTR_ERR(dquot);
5142         spin_lock(&dq_data_lock);
5143
5144         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5145                  dquot->dq_dqb.dqb_bsoftlimit :
5146                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5147         if (limit && buf->f_blocks > limit) {
5148                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5149                 buf->f_blocks = limit;
5150                 buf->f_bfree = buf->f_bavail =
5151                         (buf->f_blocks > curblock) ?
5152                          (buf->f_blocks - curblock) : 0;
5153         }
5154
5155         limit = dquot->dq_dqb.dqb_isoftlimit ?
5156                 dquot->dq_dqb.dqb_isoftlimit :
5157                 dquot->dq_dqb.dqb_ihardlimit;
5158         if (limit && buf->f_files > limit) {
5159                 buf->f_files = limit;
5160                 buf->f_ffree =
5161                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5162                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5163         }
5164
5165         spin_unlock(&dq_data_lock);
5166         dqput(dquot);
5167         return 0;
5168 }
5169 #endif
5170
5171 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5172 {
5173         struct super_block *sb = dentry->d_sb;
5174         struct ext4_sb_info *sbi = EXT4_SB(sb);
5175         struct ext4_super_block *es = sbi->s_es;
5176         ext4_fsblk_t overhead = 0, resv_blocks;
5177         u64 fsid;
5178         s64 bfree;
5179         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5180
5181         if (!test_opt(sb, MINIX_DF))
5182                 overhead = sbi->s_overhead;
5183
5184         buf->f_type = EXT4_SUPER_MAGIC;
5185         buf->f_bsize = sb->s_blocksize;
5186         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5187         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5188                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5189         /* prevent underflow in case that few free space is available */
5190         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5191         buf->f_bavail = buf->f_bfree -
5192                         (ext4_r_blocks_count(es) + resv_blocks);
5193         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5194                 buf->f_bavail = 0;
5195         buf->f_files = le32_to_cpu(es->s_inodes_count);
5196         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5197         buf->f_namelen = EXT4_NAME_LEN;
5198         fsid = le64_to_cpup((void *)es->s_uuid) ^
5199                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5200         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5201         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5202
5203 #ifdef CONFIG_QUOTA
5204         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5205             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5206                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5207 #endif
5208         return 0;
5209 }
5210
5211 /* Helper function for writing quotas on sync - we need to start transaction
5212  * before quota file is locked for write. Otherwise the are possible deadlocks:
5213  * Process 1                         Process 2
5214  * ext4_create()                     quota_sync()
5215  *   jbd2_journal_start()                  write_dquot()
5216  *   dquot_initialize()                         down(dqio_mutex)
5217  *     down(dqio_mutex)                    jbd2_journal_start()
5218  *
5219  */
5220
5221 #ifdef CONFIG_QUOTA
5222
5223 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5224 {
5225         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5226 }
5227
5228 static int ext4_write_dquot(struct dquot *dquot)
5229 {
5230         int ret, err;
5231         handle_t *handle;
5232         struct inode *inode;
5233
5234         inode = dquot_to_inode(dquot);
5235         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5236                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5237         if (IS_ERR(handle))
5238                 return PTR_ERR(handle);
5239         ret = dquot_commit(dquot);
5240         err = ext4_journal_stop(handle);
5241         if (!ret)
5242                 ret = err;
5243         return ret;
5244 }
5245
5246 static int ext4_acquire_dquot(struct dquot *dquot)
5247 {
5248         int ret, err;
5249         handle_t *handle;
5250
5251         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5252                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5253         if (IS_ERR(handle))
5254                 return PTR_ERR(handle);
5255         ret = dquot_acquire(dquot);
5256         err = ext4_journal_stop(handle);
5257         if (!ret)
5258                 ret = err;
5259         return ret;
5260 }
5261
5262 static int ext4_release_dquot(struct dquot *dquot)
5263 {
5264         int ret, err;
5265         handle_t *handle;
5266
5267         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5268                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5269         if (IS_ERR(handle)) {
5270                 /* Release dquot anyway to avoid endless cycle in dqput() */
5271                 dquot_release(dquot);
5272                 return PTR_ERR(handle);
5273         }
5274         ret = dquot_release(dquot);
5275         err = ext4_journal_stop(handle);
5276         if (!ret)
5277                 ret = err;
5278         return ret;
5279 }
5280
5281 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5282 {
5283         struct super_block *sb = dquot->dq_sb;
5284         struct ext4_sb_info *sbi = EXT4_SB(sb);
5285
5286         /* Are we journaling quotas? */
5287         if (ext4_has_feature_quota(sb) ||
5288             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5289                 dquot_mark_dquot_dirty(dquot);
5290                 return ext4_write_dquot(dquot);
5291         } else {
5292                 return dquot_mark_dquot_dirty(dquot);
5293         }
5294 }
5295
5296 static int ext4_write_info(struct super_block *sb, int type)
5297 {
5298         int ret, err;
5299         handle_t *handle;
5300
5301         /* Data block + inode block */
5302         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5303         if (IS_ERR(handle))
5304                 return PTR_ERR(handle);
5305         ret = dquot_commit_info(sb, type);
5306         err = ext4_journal_stop(handle);
5307         if (!ret)
5308                 ret = err;
5309         return ret;
5310 }
5311
5312 /*
5313  * Turn on quotas during mount time - we need to find
5314  * the quota file and such...
5315  */
5316 static int ext4_quota_on_mount(struct super_block *sb, int type)
5317 {
5318         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5319                                         EXT4_SB(sb)->s_jquota_fmt, type);
5320 }
5321
5322 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5323 {
5324         struct ext4_inode_info *ei = EXT4_I(inode);
5325
5326         /* The first argument of lockdep_set_subclass has to be
5327          * *exactly* the same as the argument to init_rwsem() --- in
5328          * this case, in init_once() --- or lockdep gets unhappy
5329          * because the name of the lock is set using the
5330          * stringification of the argument to init_rwsem().
5331          */
5332         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5333         lockdep_set_subclass(&ei->i_data_sem, subclass);
5334 }
5335
5336 /*
5337  * Standard function to be called on quota_on
5338  */
5339 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5340                          const struct path *path)
5341 {
5342         int err;
5343
5344         if (!test_opt(sb, QUOTA))
5345                 return -EINVAL;
5346
5347         /* Quotafile not on the same filesystem? */
5348         if (path->dentry->d_sb != sb)
5349                 return -EXDEV;
5350         /* Journaling quota? */
5351         if (EXT4_SB(sb)->s_qf_names[type]) {
5352                 /* Quotafile not in fs root? */
5353                 if (path->dentry->d_parent != sb->s_root)
5354                         ext4_msg(sb, KERN_WARNING,
5355                                 "Quota file not on filesystem root. "
5356                                 "Journaled quota will not work");
5357         }
5358
5359         /*
5360          * When we journal data on quota file, we have to flush journal to see
5361          * all updates to the file when we bypass pagecache...
5362          */
5363         if (EXT4_SB(sb)->s_journal &&
5364             ext4_should_journal_data(d_inode(path->dentry))) {
5365                 /*
5366                  * We don't need to lock updates but journal_flush() could
5367                  * otherwise be livelocked...
5368                  */
5369                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5370                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5371                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5372                 if (err)
5373                         return err;
5374         }
5375
5376         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5377         err = dquot_quota_on(sb, type, format_id, path);
5378         if (err) {
5379                 lockdep_set_quota_inode(path->dentry->d_inode,
5380                                              I_DATA_SEM_NORMAL);
5381         } else {
5382                 struct inode *inode = d_inode(path->dentry);
5383                 handle_t *handle;
5384
5385                 /*
5386                  * Set inode flags to prevent userspace from messing with quota
5387                  * files. If this fails, we return success anyway since quotas
5388                  * are already enabled and this is not a hard failure.
5389                  */
5390                 inode_lock(inode);
5391                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5392                 if (IS_ERR(handle))
5393                         goto unlock_inode;
5394                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5395                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5396                                 S_NOATIME | S_IMMUTABLE);
5397                 ext4_mark_inode_dirty(handle, inode);
5398                 ext4_journal_stop(handle);
5399         unlock_inode:
5400                 inode_unlock(inode);
5401         }
5402         return err;
5403 }
5404
5405 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5406                              unsigned int flags)
5407 {
5408         int err;
5409         struct inode *qf_inode;
5410         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5411                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5412                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5413                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5414         };
5415
5416         BUG_ON(!ext4_has_feature_quota(sb));
5417
5418         if (!qf_inums[type])
5419                 return -EPERM;
5420
5421         qf_inode = ext4_iget(sb, qf_inums[type]);
5422         if (IS_ERR(qf_inode)) {
5423                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5424                 return PTR_ERR(qf_inode);
5425         }
5426
5427         /* Don't account quota for quota files to avoid recursion */
5428         qf_inode->i_flags |= S_NOQUOTA;
5429         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5430         err = dquot_enable(qf_inode, type, format_id, flags);
5431         iput(qf_inode);
5432         if (err)
5433                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5434
5435         return err;
5436 }
5437
5438 /* Enable usage tracking for all quota types. */
5439 static int ext4_enable_quotas(struct super_block *sb)
5440 {
5441         int type, err = 0;
5442         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5443                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5444                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5445                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5446         };
5447         bool quota_mopt[EXT4_MAXQUOTAS] = {
5448                 test_opt(sb, USRQUOTA),
5449                 test_opt(sb, GRPQUOTA),
5450                 test_opt(sb, PRJQUOTA),
5451         };
5452
5453         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5454         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5455                 if (qf_inums[type]) {
5456                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5457                                 DQUOT_USAGE_ENABLED |
5458                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5459                         if (err) {
5460                                 ext4_warning(sb,
5461                                         "Failed to enable quota tracking "
5462                                         "(type=%d, err=%d). Please run "
5463                                         "e2fsck to fix.", type, err);
5464                                 return err;
5465                         }
5466                 }
5467         }
5468         return 0;
5469 }
5470
5471 static int ext4_quota_off(struct super_block *sb, int type)
5472 {
5473         struct inode *inode = sb_dqopt(sb)->files[type];
5474         handle_t *handle;
5475         int err;
5476
5477         /* Force all delayed allocation blocks to be allocated.
5478          * Caller already holds s_umount sem */
5479         if (test_opt(sb, DELALLOC))
5480                 sync_filesystem(sb);
5481
5482         if (!inode || !igrab(inode))
5483                 goto out;
5484
5485         err = dquot_quota_off(sb, type);
5486         if (err || ext4_has_feature_quota(sb))
5487                 goto out_put;
5488
5489         inode_lock(inode);
5490         /*
5491          * Update modification times of quota files when userspace can
5492          * start looking at them. If we fail, we return success anyway since
5493          * this is not a hard failure and quotas are already disabled.
5494          */
5495         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5496         if (IS_ERR(handle))
5497                 goto out_unlock;
5498         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5499         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5500         inode->i_mtime = inode->i_ctime = current_time(inode);
5501         ext4_mark_inode_dirty(handle, inode);
5502         ext4_journal_stop(handle);
5503 out_unlock:
5504         inode_unlock(inode);
5505 out_put:
5506         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5507         iput(inode);
5508         return err;
5509 out:
5510         return dquot_quota_off(sb, type);
5511 }
5512
5513 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5514  * acquiring the locks... As quota files are never truncated and quota code
5515  * itself serializes the operations (and no one else should touch the files)
5516  * we don't have to be afraid of races */
5517 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5518                                size_t len, loff_t off)
5519 {
5520         struct inode *inode = sb_dqopt(sb)->files[type];
5521         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5522         int offset = off & (sb->s_blocksize - 1);
5523         int tocopy;
5524         size_t toread;
5525         struct buffer_head *bh;
5526         loff_t i_size = i_size_read(inode);
5527
5528         if (off > i_size)
5529                 return 0;
5530         if (off+len > i_size)
5531                 len = i_size-off;
5532         toread = len;
5533         while (toread > 0) {
5534                 tocopy = sb->s_blocksize - offset < toread ?
5535                                 sb->s_blocksize - offset : toread;
5536                 bh = ext4_bread(NULL, inode, blk, 0);
5537                 if (IS_ERR(bh))
5538                         return PTR_ERR(bh);
5539                 if (!bh)        /* A hole? */
5540                         memset(data, 0, tocopy);
5541                 else
5542                         memcpy(data, bh->b_data+offset, tocopy);
5543                 brelse(bh);
5544                 offset = 0;
5545                 toread -= tocopy;
5546                 data += tocopy;
5547                 blk++;
5548         }
5549         return len;
5550 }
5551
5552 /* Write to quotafile (we know the transaction is already started and has
5553  * enough credits) */
5554 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5555                                 const char *data, size_t len, loff_t off)
5556 {
5557         struct inode *inode = sb_dqopt(sb)->files[type];
5558         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5559         int err, offset = off & (sb->s_blocksize - 1);
5560         int retries = 0;
5561         struct buffer_head *bh;
5562         handle_t *handle = journal_current_handle();
5563
5564         if (EXT4_SB(sb)->s_journal && !handle) {
5565                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5566                         " cancelled because transaction is not started",
5567                         (unsigned long long)off, (unsigned long long)len);
5568                 return -EIO;
5569         }
5570         /*
5571          * Since we account only one data block in transaction credits,
5572          * then it is impossible to cross a block boundary.
5573          */
5574         if (sb->s_blocksize - offset < len) {
5575                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5576                         " cancelled because not block aligned",
5577                         (unsigned long long)off, (unsigned long long)len);
5578                 return -EIO;
5579         }
5580
5581         do {
5582                 bh = ext4_bread(handle, inode, blk,
5583                                 EXT4_GET_BLOCKS_CREATE |
5584                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5585         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5586                  ext4_should_retry_alloc(inode->i_sb, &retries));
5587         if (IS_ERR(bh))
5588                 return PTR_ERR(bh);
5589         if (!bh)
5590                 goto out;
5591         BUFFER_TRACE(bh, "get write access");
5592         err = ext4_journal_get_write_access(handle, bh);
5593         if (err) {
5594                 brelse(bh);
5595                 return err;
5596         }
5597         lock_buffer(bh);
5598         memcpy(bh->b_data+offset, data, len);
5599         flush_dcache_page(bh->b_page);
5600         unlock_buffer(bh);
5601         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5602         brelse(bh);
5603 out:
5604         if (inode->i_size < off + len) {
5605                 i_size_write(inode, off + len);
5606                 EXT4_I(inode)->i_disksize = inode->i_size;
5607                 ext4_mark_inode_dirty(handle, inode);
5608         }
5609         return len;
5610 }
5611
5612 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5613 {
5614         const struct quota_format_ops   *ops;
5615
5616         if (!sb_has_quota_loaded(sb, qid->type))
5617                 return -ESRCH;
5618         ops = sb_dqopt(sb)->ops[qid->type];
5619         if (!ops || !ops->get_next_id)
5620                 return -ENOSYS;
5621         return dquot_get_next_id(sb, qid);
5622 }
5623 #endif
5624
5625 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5626                        const char *dev_name, void *data)
5627 {
5628         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5629 }
5630
5631 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5632 static inline void register_as_ext2(void)
5633 {
5634         int err = register_filesystem(&ext2_fs_type);
5635         if (err)
5636                 printk(KERN_WARNING
5637                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5638 }
5639
5640 static inline void unregister_as_ext2(void)
5641 {
5642         unregister_filesystem(&ext2_fs_type);
5643 }
5644
5645 static inline int ext2_feature_set_ok(struct super_block *sb)
5646 {
5647         if (ext4_has_unknown_ext2_incompat_features(sb))
5648                 return 0;
5649         if (sb->s_flags & MS_RDONLY)
5650                 return 1;
5651         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5652                 return 0;
5653         return 1;
5654 }
5655 #else
5656 static inline void register_as_ext2(void) { }
5657 static inline void unregister_as_ext2(void) { }
5658 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5659 #endif
5660
5661 static inline void register_as_ext3(void)
5662 {
5663         int err = register_filesystem(&ext3_fs_type);
5664         if (err)
5665                 printk(KERN_WARNING
5666                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5667 }
5668
5669 static inline void unregister_as_ext3(void)
5670 {
5671         unregister_filesystem(&ext3_fs_type);
5672 }
5673
5674 static inline int ext3_feature_set_ok(struct super_block *sb)
5675 {
5676         if (ext4_has_unknown_ext3_incompat_features(sb))
5677                 return 0;
5678         if (!ext4_has_feature_journal(sb))
5679                 return 0;
5680         if (sb->s_flags & MS_RDONLY)
5681                 return 1;
5682         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5683                 return 0;
5684         return 1;
5685 }
5686
5687 static struct file_system_type ext4_fs_type = {
5688         .owner          = THIS_MODULE,
5689         .name           = "ext4",
5690         .mount          = ext4_mount,
5691         .kill_sb        = kill_block_super,
5692         .fs_flags       = FS_REQUIRES_DEV,
5693 };
5694 MODULE_ALIAS_FS("ext4");
5695
5696 /* Shared across all ext4 file systems */
5697 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5698
5699 static int __init ext4_init_fs(void)
5700 {
5701         int i, err;
5702
5703         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5704         ext4_li_info = NULL;
5705         mutex_init(&ext4_li_mtx);
5706
5707         /* Build-time check for flags consistency */
5708         ext4_check_flag_values();
5709
5710         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5711                 init_waitqueue_head(&ext4__ioend_wq[i]);
5712
5713         err = ext4_init_es();
5714         if (err)
5715                 return err;
5716
5717         err = ext4_init_pageio();
5718         if (err)
5719                 goto out5;
5720
5721         err = ext4_init_system_zone();
5722         if (err)
5723                 goto out4;
5724
5725         err = ext4_init_sysfs();
5726         if (err)
5727                 goto out3;
5728
5729         err = ext4_init_mballoc();
5730         if (err)
5731                 goto out2;
5732         err = init_inodecache();
5733         if (err)
5734                 goto out1;
5735         register_as_ext3();
5736         register_as_ext2();
5737         err = register_filesystem(&ext4_fs_type);
5738         if (err)
5739                 goto out;
5740
5741         return 0;
5742 out:
5743         unregister_as_ext2();
5744         unregister_as_ext3();
5745         destroy_inodecache();
5746 out1:
5747         ext4_exit_mballoc();
5748 out2:
5749         ext4_exit_sysfs();
5750 out3:
5751         ext4_exit_system_zone();
5752 out4:
5753         ext4_exit_pageio();
5754 out5:
5755         ext4_exit_es();
5756
5757         return err;
5758 }
5759
5760 static void __exit ext4_exit_fs(void)
5761 {
5762         ext4_destroy_lazyinit_thread();
5763         unregister_as_ext2();
5764         unregister_as_ext3();
5765         unregister_filesystem(&ext4_fs_type);
5766         destroy_inodecache();
5767         ext4_exit_mballoc();
5768         ext4_exit_sysfs();
5769         ext4_exit_system_zone();
5770         ext4_exit_pageio();
5771         ext4_exit_es();
5772 }
5773
5774 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5775 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5776 MODULE_LICENSE("GPL");
5777 module_init(ext4_init_fs)
5778 module_exit(ext4_exit_fs)