fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 static int ext4_verify_csum_type(struct super_block *sb,
 144                                  struct ext4_super_block *es)
 145 {
 146         if (!ext4_has_feature_metadata_csum(sb))
 147                 return 1;
 148
 149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 150 }
 151
 152 static __le32 ext4_superblock_csum(struct super_block *sb,
 153                                    struct ext4_super_block *es)
 154 {
 155         struct ext4_sb_info *sbi = EXT4_SB(sb);
 156         int offset = offsetof(struct ext4_super_block, s_checksum);
 157         __u32 csum;
 158
 159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 160
 161         return cpu_to_le32(csum);
 162 }
 163
 164 static int ext4_superblock_csum_verify(struct super_block *sb,
 165                                        struct ext4_super_block *es)
 166 {
 167         if (!ext4_has_metadata_csum(sb))
 168                 return 1;
 169
 170         return es->s_checksum == ext4_superblock_csum(sb, es);
 171 }
 172
 173 void ext4_superblock_csum_set(struct super_block *sb)
 174 {
 175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 176
 177         if (!ext4_has_metadata_csum(sb))
 178                 return;
 179
 180         es->s_checksum = ext4_superblock_csum(sb, es);
 181 }
 182
 183 void *ext4_kvmalloc(size_t size, gfp_t flags)
 184 {
 185         void *ret;
 186
 187         ret = kmalloc(size, flags | __GFP_NOWARN);
 188         if (!ret)
 189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 190         return ret;
 191 }
 192
 193 void *ext4_kvzalloc(size_t size, gfp_t flags)
 194 {
 195         void *ret;
 196
 197         ret = kzalloc(size, flags | __GFP_NOWARN);
 198         if (!ret)
 199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 200         return ret;
 201 }
 202
 203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 209 }
 210
 211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 212                                struct ext4_group_desc *bg)
 213 {
 214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 217 }
 218
 219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le32_to_cpu(bg->bg_inode_table_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 225 }
 226
 227 __u32 ext4_free_group_clusters(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 233 }
 234
 235 __u32 ext4_free_inodes_count(struct super_block *sb,
 236                               struct ext4_group_desc *bg)
 237 {
 238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 241 }
 242
 243 __u32 ext4_used_dirs_count(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 249 }
 250
 251 __u32 ext4_itable_unused_count(struct super_block *sb,
 252                               struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_itable_unused_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 257 }
 258
 259 void ext4_block_bitmap_set(struct super_block *sb,
 260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 261 {
 262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 265 }
 266
 267 void ext4_inode_bitmap_set(struct super_block *sb,
 268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 269 {
 270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 273 }
 274
 275 void ext4_inode_table_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 277 {
 278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 281 }
 282
 283 void ext4_free_group_clusters_set(struct super_block *sb,
 284                                   struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291 void ext4_free_inodes_set(struct super_block *sb,
 292                           struct ext4_group_desc *bg, __u32 count)
 293 {
 294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 297 }
 298
 299 void ext4_used_dirs_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, __u32 count)
 301 {
 302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 305 }
 306
 307 void ext4_itable_unused_set(struct super_block *sb,
 308                           struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315
 316 static void __save_error_info(struct super_block *sb, const char *func,
 317                             unsigned int line)
 318 {
 319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 320
 321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 322         if (bdev_read_only(sb->s_bdev))
 323                 return;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         le32_add_cpu(&es->s_error_count, 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352 /*
 353  * The del_gendisk() function uninitializes the disk-specific data
 354  * structures, including the bdi structure, without telling anyone
 355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 357  * This is a kludge to prevent these oops until we can put in a proper
 358  * hook in del_gendisk() to inform the VFS and file system layers.
 359  */
 360 static int block_device_ejected(struct super_block *sb)
 361 {
 362         struct inode *bd_inode = sb->s_bdev->bd_inode;
 363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 364
 365         return bdi->dev == NULL;
 366 }
 367
 368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 369 {
 370         struct super_block              *sb = journal->j_private;
 371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 372         int                             error = is_journal_aborted(journal);
 373         struct ext4_journal_cb_entry    *jce;
 374
 375         BUG_ON(txn->t_state == T_FINISHED);
 376
 377         ext4_process_freed_data(sb, txn->t_tid);
 378
 379         spin_lock(&sbi->s_md_lock);
 380         while (!list_empty(&txn->t_private_list)) {
 381                 jce = list_entry(txn->t_private_list.next,
 382                                  struct ext4_journal_cb_entry, jce_list);
 383                 list_del_init(&jce->jce_list);
 384                 spin_unlock(&sbi->s_md_lock);
 385                 jce->jce_func(sb, jce, error);
 386                 spin_lock(&sbi->s_md_lock);
 387         }
 388         spin_unlock(&sbi->s_md_lock);
 389 }
 390
 391 /* Deal with the reporting of failure conditions on a filesystem such as
 392  * inconsistencies detected or read IO failures.
 393  *
 394  * On ext2, we can store the error state of the filesystem in the
 395  * superblock.  That is not possible on ext4, because we may have other
 396  * write ordering constraints on the superblock which prevent us from
 397  * writing it out straight away; and given that the journal is about to
 398  * be aborted, we can't rely on the current, or future, transactions to
 399  * write out the superblock safely.
 400  *
 401  * We'll just use the jbd2_journal_abort() error code to record an error in
 402  * the journal instead.  On recovery, the journal will complain about
 403  * that error until we've noted it down and cleared it.
 404  */
 405
 406 static void ext4_handle_error(struct super_block *sb)
 407 {
 408         if (sb_rdonly(sb))
 409                 return;
 410
 411         if (!test_opt(sb, ERRORS_CONT)) {
 412                 journal_t *journal = EXT4_SB(sb)->s_journal;
 413
 414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 415                 if (journal)
 416                         jbd2_journal_abort(journal, -EIO);
 417         }
 418         if (test_opt(sb, ERRORS_RO)) {
 419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 420                 /*
 421                  * Make sure updated value of ->s_mount_flags will be visible
 422                  * before ->s_flags update
 423                  */
 424                 smp_wmb();
 425                 sb->s_flags |= MS_RDONLY;
 426         }
 427         if (test_opt(sb, ERRORS_PANIC)) {
 428                 if (EXT4_SB(sb)->s_journal &&
 429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 430                         return;
 431                 panic("EXT4-fs (device %s): panic forced after error\n",
 432                         sb->s_id);
 433         }
 434 }
 435
 436 #define ext4_error_ratelimit(sb)                                        \
 437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 438                              "EXT4-fs error")
 439
 440 void __ext4_error(struct super_block *sb, const char *function,
 441                   unsigned int line, const char *fmt, ...)
 442 {
 443         struct va_format vaf;
 444         va_list args;
 445
 446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 447                 return;
 448
 449         if (ext4_error_ratelimit(sb)) {
 450                 va_start(args, fmt);
 451                 vaf.fmt = fmt;
 452                 vaf.va = &args;
 453                 printk(KERN_CRIT
 454                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 455                        sb->s_id, function, line, current->comm, &vaf);
 456                 va_end(args);
 457         }
 458         save_error_info(sb, function, line);
 459         ext4_handle_error(sb);
 460 }
 461
 462 void __ext4_error_inode(struct inode *inode, const char *function,
 463                         unsigned int line, ext4_fsblk_t block,
 464                         const char *fmt, ...)
 465 {
 466         va_list args;
 467         struct va_format vaf;
 468         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 469
 470         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 471                 return;
 472
 473         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 474         es->s_last_error_block = cpu_to_le64(block);
 475         if (ext4_error_ratelimit(inode->i_sb)) {
 476                 va_start(args, fmt);
 477                 vaf.fmt = fmt;
 478                 vaf.va = &args;
 479                 if (block)
 480                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 481                                "inode #%lu: block %llu: comm %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                block, current->comm, &vaf);
 484                 else
 485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 486                                "inode #%lu: comm %s: %pV\n",
 487                                inode->i_sb->s_id, function, line, inode->i_ino,
 488                                current->comm, &vaf);
 489                 va_end(args);
 490         }
 491         save_error_info(inode->i_sb, function, line);
 492         ext4_handle_error(inode->i_sb);
 493 }
 494
 495 void __ext4_error_file(struct file *file, const char *function,
 496                        unsigned int line, ext4_fsblk_t block,
 497                        const char *fmt, ...)
 498 {
 499         va_list args;
 500         struct va_format vaf;
 501         struct ext4_super_block *es;
 502         struct inode *inode = file_inode(file);
 503         char pathname[80], *path;
 504
 505         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 506                 return;
 507
 508         es = EXT4_SB(inode->i_sb)->s_es;
 509         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 510         if (ext4_error_ratelimit(inode->i_sb)) {
 511                 path = file_path(file, pathname, sizeof(pathname));
 512                 if (IS_ERR(path))
 513                         path = "(unknown)";
 514                 va_start(args, fmt);
 515                 vaf.fmt = fmt;
 516                 vaf.va = &args;
 517                 if (block)
 518                         printk(KERN_CRIT
 519                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 520                                "block %llu: comm %s: path %s: %pV\n",
 521                                inode->i_sb->s_id, function, line, inode->i_ino,
 522                                block, current->comm, path, &vaf);
 523                 else
 524                         printk(KERN_CRIT
 525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 526                                "comm %s: path %s: %pV\n",
 527                                inode->i_sb->s_id, function, line, inode->i_ino,
 528                                current->comm, path, &vaf);
 529                 va_end(args);
 530         }
 531         save_error_info(inode->i_sb, function, line);
 532         ext4_handle_error(inode->i_sb);
 533 }
 534
 535 const char *ext4_decode_error(struct super_block *sb, int errno,
 536                               char nbuf[16])
 537 {
 538         char *errstr = NULL;
 539
 540         switch (errno) {
 541         case -EFSCORRUPTED:
 542                 errstr = "Corrupt filesystem";
 543                 break;
 544         case -EFSBADCRC:
 545                 errstr = "Filesystem failed CRC";
 546                 break;
 547         case -EIO:
 548                 errstr = "IO failure";
 549                 break;
 550         case -ENOMEM:
 551                 errstr = "Out of memory";
 552                 break;
 553         case -EROFS:
 554                 if (!sb || (EXT4_SB(sb)->s_journal &&
 555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 556                         errstr = "Journal has aborted";
 557                 else
 558                         errstr = "Readonly filesystem";
 559                 break;
 560         default:
 561                 /* If the caller passed in an extra buffer for unknown
 562                  * errors, textualise them now.  Else we just return
 563                  * NULL. */
 564                 if (nbuf) {
 565                         /* Check for truncated error codes... */
 566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 567                                 errstr = nbuf;
 568                 }
 569                 break;
 570         }
 571
 572         return errstr;
 573 }
 574
 575 /* __ext4_std_error decodes expected errors from journaling functions
 576  * automatically and invokes the appropriate error response.  */
 577
 578 void __ext4_std_error(struct super_block *sb, const char *function,
 579                       unsigned int line, int errno)
 580 {
 581         char nbuf[16];
 582         const char *errstr;
 583
 584         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 585                 return;
 586
 587         /* Special case: if the error is EROFS, and we're not already
 588          * inside a transaction, then there's really no point in logging
 589          * an error. */
 590         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 591                 return;
 592
 593         if (ext4_error_ratelimit(sb)) {
 594                 errstr = ext4_decode_error(sb, errno, nbuf);
 595                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 596                        sb->s_id, function, line, errstr);
 597         }
 598
 599         save_error_info(sb, function, line);
 600         ext4_handle_error(sb);
 601 }
 602
 603 /*
 604  * ext4_abort is a much stronger failure handler than ext4_error.  The
 605  * abort function may be used to deal with unrecoverable failures such
 606  * as journal IO errors or ENOMEM at a critical moment in log management.
 607  *
 608  * We unconditionally force the filesystem into an ABORT|READONLY state,
 609  * unless the error response on the fs has been set to panic in which
 610  * case we take the easy way out and panic immediately.
 611  */
 612
 613 void __ext4_abort(struct super_block *sb, const char *function,
 614                 unsigned int line, const char *fmt, ...)
 615 {
 616         struct va_format vaf;
 617         va_list args;
 618
 619         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 620                 return;
 621
 622         save_error_info(sb, function, line);
 623         va_start(args, fmt);
 624         vaf.fmt = fmt;
 625         vaf.va = &args;
 626         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 627                sb->s_id, function, line, &vaf);
 628         va_end(args);
 629
 630         if (sb_rdonly(sb) == 0) {
 631                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 632                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 633                 /*
 634                  * Make sure updated value of ->s_mount_flags will be visible
 635                  * before ->s_flags update
 636                  */
 637                 smp_wmb();
 638                 sb->s_flags |= MS_RDONLY;
 639                 if (EXT4_SB(sb)->s_journal)
 640                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 641                 save_error_info(sb, function, line);
 642         }
 643         if (test_opt(sb, ERRORS_PANIC)) {
 644                 if (EXT4_SB(sb)->s_journal &&
 645                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 646                         return;
 647                 panic("EXT4-fs panic from previous error\n");
 648         }
 649 }
 650
 651 void __ext4_msg(struct super_block *sb,
 652                 const char *prefix, const char *fmt, ...)
 653 {
 654         struct va_format vaf;
 655         va_list args;
 656
 657         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 658                 return;
 659
 660         va_start(args, fmt);
 661         vaf.fmt = fmt;
 662         vaf.va = &args;
 663         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 664         va_end(args);
 665 }
 666
 667 #define ext4_warning_ratelimit(sb)                                      \
 668                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 669                              "EXT4-fs warning")
 670
 671 void __ext4_warning(struct super_block *sb, const char *function,
 672                     unsigned int line, const char *fmt, ...)
 673 {
 674         struct va_format vaf;
 675         va_list args;
 676
 677         if (!ext4_warning_ratelimit(sb))
 678                 return;
 679
 680         va_start(args, fmt);
 681         vaf.fmt = fmt;
 682         vaf.va = &args;
 683         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 684                sb->s_id, function, line, &vaf);
 685         va_end(args);
 686 }
 687
 688 void __ext4_warning_inode(const struct inode *inode, const char *function,
 689                           unsigned int line, const char *fmt, ...)
 690 {
 691         struct va_format vaf;
 692         va_list args;
 693
 694         if (!ext4_warning_ratelimit(inode->i_sb))
 695                 return;
 696
 697         va_start(args, fmt);
 698         vaf.fmt = fmt;
 699         vaf.va = &args;
 700         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 701                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 702                function, line, inode->i_ino, current->comm, &vaf);
 703         va_end(args);
 704 }
 705
 706 void __ext4_grp_locked_error(const char *function, unsigned int line,
 707                              struct super_block *sb, ext4_group_t grp,
 708                              unsigned long ino, ext4_fsblk_t block,
 709                              const char *fmt, ...)
 710 __releases(bitlock)
 711 __acquires(bitlock)
 712 {
 713         struct va_format vaf;
 714         va_list args;
 715         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 716
 717         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 718                 return;
 719
 720         es->s_last_error_ino = cpu_to_le32(ino);
 721         es->s_last_error_block = cpu_to_le64(block);
 722         __save_error_info(sb, function, line);
 723
 724         if (ext4_error_ratelimit(sb)) {
 725                 va_start(args, fmt);
 726                 vaf.fmt = fmt;
 727                 vaf.va = &args;
 728                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 729                        sb->s_id, function, line, grp);
 730                 if (ino)
 731                         printk(KERN_CONT "inode %lu: ", ino);
 732                 if (block)
 733                         printk(KERN_CONT "block %llu:",
 734                                (unsigned long long) block);
 735                 printk(KERN_CONT "%pV\n", &vaf);
 736                 va_end(args);
 737         }
 738
 739         if (test_opt(sb, ERRORS_CONT)) {
 740                 ext4_commit_super(sb, 0);
 741                 return;
 742         }
 743
 744         ext4_unlock_group(sb, grp);
 745         ext4_handle_error(sb);
 746         /*
 747          * We only get here in the ERRORS_RO case; relocking the group
 748          * may be dangerous, but nothing bad will happen since the
 749          * filesystem will have already been marked read/only and the
 750          * journal has been aborted.  We return 1 as a hint to callers
 751          * who might what to use the return value from
 752          * ext4_grp_locked_error() to distinguish between the
 753          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 754          * aggressively from the ext4 function in question, with a
 755          * more appropriate error code.
 756          */
 757         ext4_lock_group(sb, grp);
 758         return;
 759 }
 760
 761 void ext4_update_dynamic_rev(struct super_block *sb)
 762 {
 763         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 764
 765         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 766                 return;
 767
 768         ext4_warning(sb,
 769                      "updating to rev %d because of new feature flag, "
 770                      "running e2fsck is recommended",
 771                      EXT4_DYNAMIC_REV);
 772
 773         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 774         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 775         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 776         /* leave es->s_feature_*compat flags alone */
 777         /* es->s_uuid will be set by e2fsck if empty */
 778
 779         /*
 780          * The rest of the superblock fields should be zero, and if not it
 781          * means they are likely already in use, so leave them alone.  We
 782          * can leave it up to e2fsck to clean up any inconsistencies there.
 783          */
 784 }
 785
 786 /*
 787  * Open the external journal device
 788  */
 789 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 790 {
 791         struct block_device *bdev;
 792         char b[BDEVNAME_SIZE];
 793
 794         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 795         if (IS_ERR(bdev))
 796                 goto fail;
 797         return bdev;
 798
 799 fail:
 800         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 801                         __bdevname(dev, b), PTR_ERR(bdev));
 802         return NULL;
 803 }
 804
 805 /*
 806  * Release the journal device
 807  */
 808 static void ext4_blkdev_put(struct block_device *bdev)
 809 {
 810         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 811 }
 812
 813 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 814 {
 815         struct block_device *bdev;
 816         bdev = sbi->journal_bdev;
 817         if (bdev) {
 818                 ext4_blkdev_put(bdev);
 819                 sbi->journal_bdev = NULL;
 820         }
 821 }
 822
 823 static inline struct inode *orphan_list_entry(struct list_head *l)
 824 {
 825         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 826 }
 827
 828 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 829 {
 830         struct list_head *l;
 831
 832         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 833                  le32_to_cpu(sbi->s_es->s_last_orphan));
 834
 835         printk(KERN_ERR "sb_info orphan list:\n");
 836         list_for_each(l, &sbi->s_orphan) {
 837                 struct inode *inode = orphan_list_entry(l);
 838                 printk(KERN_ERR "  "
 839                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 840                        inode->i_sb->s_id, inode->i_ino, inode,
 841                        inode->i_mode, inode->i_nlink,
 842                        NEXT_ORPHAN(inode));
 843         }
 844 }
 845
 846 #ifdef CONFIG_QUOTA
 847 static int ext4_quota_off(struct super_block *sb, int type);
 848
 849 static inline void ext4_quota_off_umount(struct super_block *sb)
 850 {
 851         int type;
 852
 853         /* Use our quota_off function to clear inode flags etc. */
 854         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 855                 ext4_quota_off(sb, type);
 856 }
 857 #else
 858 static inline void ext4_quota_off_umount(struct super_block *sb)
 859 {
 860 }
 861 #endif
 862
 863 static void ext4_put_super(struct super_block *sb)
 864 {
 865         struct ext4_sb_info *sbi = EXT4_SB(sb);
 866         struct ext4_super_block *es = sbi->s_es;
 867         int aborted = 0;
 868         int i, err;
 869
 870         ext4_unregister_li_request(sb);
 871         ext4_quota_off_umount(sb);
 872
 873         flush_workqueue(sbi->rsv_conversion_wq);
 874         destroy_workqueue(sbi->rsv_conversion_wq);
 875
 876         if (sbi->s_journal) {
 877                 aborted = is_journal_aborted(sbi->s_journal);
 878                 err = jbd2_journal_destroy(sbi->s_journal);
 879                 sbi->s_journal = NULL;
 880                 if ((err < 0) && !aborted)
 881                         ext4_abort(sb, "Couldn't clean up the journal");
 882         }
 883
 884         ext4_unregister_sysfs(sb);
 885         ext4_es_unregister_shrinker(sbi);
 886         del_timer_sync(&sbi->s_err_report);
 887         ext4_release_system_zone(sb);
 888         ext4_mb_release(sb);
 889         ext4_ext_release(sb);
 890
 891         if (!sb_rdonly(sb) && !aborted) {
 892                 ext4_clear_feature_journal_needs_recovery(sb);
 893                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 894         }
 895         if (!sb_rdonly(sb))
 896                 ext4_commit_super(sb, 1);
 897
 898         for (i = 0; i < sbi->s_gdb_count; i++)
 899                 brelse(sbi->s_group_desc[i]);
 900         kvfree(sbi->s_group_desc);
 901         kvfree(sbi->s_flex_groups);
 902         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 903         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 904         percpu_counter_destroy(&sbi->s_dirs_counter);
 905         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 906         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 907 #ifdef CONFIG_QUOTA
 908         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 909                 kfree(sbi->s_qf_names[i]);
 910 #endif
 911
 912         /* Debugging code just in case the in-memory inode orphan list
 913          * isn't empty.  The on-disk one can be non-empty if we've
 914          * detected an error and taken the fs readonly, but the
 915          * in-memory list had better be clean by this point. */
 916         if (!list_empty(&sbi->s_orphan))
 917                 dump_orphan_list(sb, sbi);
 918         J_ASSERT(list_empty(&sbi->s_orphan));
 919
 920         sync_blockdev(sb->s_bdev);
 921         invalidate_bdev(sb->s_bdev);
 922         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 923                 /*
 924                  * Invalidate the journal device's buffers.  We don't want them
 925                  * floating about in memory - the physical journal device may
 926                  * hotswapped, and it breaks the `ro-after' testing code.
 927                  */
 928                 sync_blockdev(sbi->journal_bdev);
 929                 invalidate_bdev(sbi->journal_bdev);
 930                 ext4_blkdev_remove(sbi);
 931         }
 932         if (sbi->s_ea_inode_cache) {
 933                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 934                 sbi->s_ea_inode_cache = NULL;
 935         }
 936         if (sbi->s_ea_block_cache) {
 937                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 938                 sbi->s_ea_block_cache = NULL;
 939         }
 940         if (sbi->s_mmp_tsk)
 941                 kthread_stop(sbi->s_mmp_tsk);
 942         brelse(sbi->s_sbh);
 943         sb->s_fs_info = NULL;
 944         /*
 945          * Now that we are completely done shutting down the
 946          * superblock, we need to actually destroy the kobject.
 947          */
 948         kobject_put(&sbi->s_kobj);
 949         wait_for_completion(&sbi->s_kobj_unregister);
 950         if (sbi->s_chksum_driver)
 951                 crypto_free_shash(sbi->s_chksum_driver);
 952         kfree(sbi->s_blockgroup_lock);
 953         fs_put_dax(sbi->s_daxdev);
 954         kfree(sbi);
 955 }
 956
 957 static struct kmem_cache *ext4_inode_cachep;
 958
 959 /*
 960  * Called inside transaction, so use GFP_NOFS
 961  */
 962 static struct inode *ext4_alloc_inode(struct super_block *sb)
 963 {
 964         struct ext4_inode_info *ei;
 965
 966         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 967         if (!ei)
 968                 return NULL;
 969
 970         ei->vfs_inode.i_version = 1;
 971         spin_lock_init(&ei->i_raw_lock);
 972         INIT_LIST_HEAD(&ei->i_prealloc_list);
 973         spin_lock_init(&ei->i_prealloc_lock);
 974         ext4_es_init_tree(&ei->i_es_tree);
 975         rwlock_init(&ei->i_es_lock);
 976         INIT_LIST_HEAD(&ei->i_es_list);
 977         ei->i_es_all_nr = 0;
 978         ei->i_es_shk_nr = 0;
 979         ei->i_es_shrink_lblk = 0;
 980         ei->i_reserved_data_blocks = 0;
 981         ei->i_da_metadata_calc_len = 0;
 982         ei->i_da_metadata_calc_last_lblock = 0;
 983         spin_lock_init(&(ei->i_block_reservation_lock));
 984 #ifdef CONFIG_QUOTA
 985         ei->i_reserved_quota = 0;
 986         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 987 #endif
 988         ei->jinode = NULL;
 989         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 990         spin_lock_init(&ei->i_completed_io_lock);
 991         ei->i_sync_tid = 0;
 992         ei->i_datasync_tid = 0;
 993         atomic_set(&ei->i_unwritten, 0);
 994         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 995         return &ei->vfs_inode;
 996 }
 997
 998 static int ext4_drop_inode(struct inode *inode)
 999 {
1000         int drop = generic_drop_inode(inode);
1001
1002         trace_ext4_drop_inode(inode, drop);
1003         return drop;
1004 }
1005
1006 static void ext4_i_callback(struct rcu_head *head)
1007 {
1008         struct inode *inode = container_of(head, struct inode, i_rcu);
1009         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1010 }
1011
1012 static void ext4_destroy_inode(struct inode *inode)
1013 {
1014         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1015                 ext4_msg(inode->i_sb, KERN_ERR,
1016                          "Inode %lu (%p): orphan list check failed!",
1017                          inode->i_ino, EXT4_I(inode));
1018                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1019                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1020                                 true);
1021                 dump_stack();
1022         }
1023         call_rcu(&inode->i_rcu, ext4_i_callback);
1024 }
1025
1026 static void init_once(void *foo)
1027 {
1028         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1029
1030         INIT_LIST_HEAD(&ei->i_orphan);
1031         init_rwsem(&ei->xattr_sem);
1032         init_rwsem(&ei->i_data_sem);
1033         init_rwsem(&ei->i_mmap_sem);
1034         inode_init_once(&ei->vfs_inode);
1035 }
1036
1037 static int __init init_inodecache(void)
1038 {
1039         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1040                                              sizeof(struct ext4_inode_info),
1041                                              0, (SLAB_RECLAIM_ACCOUNT|
1042                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1043                                              init_once);
1044         if (ext4_inode_cachep == NULL)
1045                 return -ENOMEM;
1046         return 0;
1047 }
1048
1049 static void destroy_inodecache(void)
1050 {
1051         /*
1052          * Make sure all delayed rcu free inodes are flushed before we
1053          * destroy cache.
1054          */
1055         rcu_barrier();
1056         kmem_cache_destroy(ext4_inode_cachep);
1057 }
1058
1059 void ext4_clear_inode(struct inode *inode)
1060 {
1061         invalidate_inode_buffers(inode);
1062         clear_inode(inode);
1063         dquot_drop(inode);
1064         ext4_discard_preallocations(inode);
1065         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1066         if (EXT4_I(inode)->jinode) {
1067                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1068                                                EXT4_I(inode)->jinode);
1069                 jbd2_free_inode(EXT4_I(inode)->jinode);
1070                 EXT4_I(inode)->jinode = NULL;
1071         }
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073         fscrypt_put_encryption_info(inode, NULL);
1074 #endif
1075 }
1076
1077 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1078                                         u64 ino, u32 generation)
1079 {
1080         struct inode *inode;
1081
1082         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1083                 return ERR_PTR(-ESTALE);
1084         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1085                 return ERR_PTR(-ESTALE);
1086
1087         /* iget isn't really right if the inode is currently unallocated!!
1088          *
1089          * ext4_read_inode will return a bad_inode if the inode had been
1090          * deleted, so we should be safe.
1091          *
1092          * Currently we don't know the generation for parent directory, so
1093          * a generation of 0 means "accept any"
1094          */
1095         inode = ext4_iget_normal(sb, ino);
1096         if (IS_ERR(inode))
1097                 return ERR_CAST(inode);
1098         if (generation && inode->i_generation != generation) {
1099                 iput(inode);
1100                 return ERR_PTR(-ESTALE);
1101         }
1102
1103         return inode;
1104 }
1105
1106 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1107                                         int fh_len, int fh_type)
1108 {
1109         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1110                                     ext4_nfs_get_inode);
1111 }
1112
1113 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1114                                         int fh_len, int fh_type)
1115 {
1116         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1117                                     ext4_nfs_get_inode);
1118 }
1119
1120 /*
1121  * Try to release metadata pages (indirect blocks, directories) which are
1122  * mapped via the block device.  Since these pages could have journal heads
1123  * which would prevent try_to_free_buffers() from freeing them, we must use
1124  * jbd2 layer's try_to_free_buffers() function to release them.
1125  */
1126 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1127                                  gfp_t wait)
1128 {
1129         journal_t *journal = EXT4_SB(sb)->s_journal;
1130
1131         WARN_ON(PageChecked(page));
1132         if (!page_has_buffers(page))
1133                 return 0;
1134         if (journal)
1135                 return jbd2_journal_try_to_free_buffers(journal, page,
1136                                                 wait & ~__GFP_DIRECT_RECLAIM);
1137         return try_to_free_buffers(page);
1138 }
1139
1140 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1141 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1142 {
1143         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1144                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1145 }
1146
1147 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1148                                                         void *fs_data)
1149 {
1150         handle_t *handle = fs_data;
1151         int res, res2, credits, retries = 0;
1152
1153         /*
1154          * Encrypting the root directory is not allowed because e2fsck expects
1155          * lost+found to exist and be unencrypted, and encrypting the root
1156          * directory would imply encrypting the lost+found directory as well as
1157          * the filename "lost+found" itself.
1158          */
1159         if (inode->i_ino == EXT4_ROOT_INO)
1160                 return -EPERM;
1161
1162         res = ext4_convert_inline_data(inode);
1163         if (res)
1164                 return res;
1165
1166         /*
1167          * If a journal handle was specified, then the encryption context is
1168          * being set on a new inode via inheritance and is part of a larger
1169          * transaction to create the inode.  Otherwise the encryption context is
1170          * being set on an existing inode in its own transaction.  Only in the
1171          * latter case should the "retry on ENOSPC" logic be used.
1172          */
1173
1174         if (handle) {
1175                 res = ext4_xattr_set_handle(handle, inode,
1176                                             EXT4_XATTR_INDEX_ENCRYPTION,
1177                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1178                                             ctx, len, 0);
1179                 if (!res) {
1180                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1181                         ext4_clear_inode_state(inode,
1182                                         EXT4_STATE_MAY_INLINE_DATA);
1183                         /*
1184                          * Update inode->i_flags - e.g. S_DAX may get disabled
1185                          */
1186                         ext4_set_inode_flags(inode);
1187                 }
1188                 return res;
1189         }
1190
1191         res = dquot_initialize(inode);
1192         if (res)
1193                 return res;
1194 retry:
1195         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1196                                      &credits);
1197         if (res)
1198                 return res;
1199
1200         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1201         if (IS_ERR(handle))
1202                 return PTR_ERR(handle);
1203
1204         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1205                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1206                                     ctx, len, 0);
1207         if (!res) {
1208                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1209                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1210                 ext4_set_inode_flags(inode);
1211                 res = ext4_mark_inode_dirty(handle, inode);
1212                 if (res)
1213                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1214         }
1215         res2 = ext4_journal_stop(handle);
1216
1217         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1218                 goto retry;
1219         if (!res)
1220                 res = res2;
1221         return res;
1222 }
1223
1224 static bool ext4_dummy_context(struct inode *inode)
1225 {
1226         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1227 }
1228
1229 static unsigned ext4_max_namelen(struct inode *inode)
1230 {
1231         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1232                 EXT4_NAME_LEN;
1233 }
1234
1235 static const struct fscrypt_operations ext4_cryptops = {
1236         .key_prefix             = "ext4:",
1237         .get_context            = ext4_get_context,
1238         .set_context            = ext4_set_context,
1239         .dummy_context          = ext4_dummy_context,
1240         .is_encrypted           = ext4_encrypted_inode,
1241         .empty_dir              = ext4_empty_dir,
1242         .max_namelen            = ext4_max_namelen,
1243 };
1244 #else
1245 static const struct fscrypt_operations ext4_cryptops = {
1246         .is_encrypted           = ext4_encrypted_inode,
1247 };
1248 #endif
1249
1250 #ifdef CONFIG_QUOTA
1251 static const char * const quotatypes[] = INITQFNAMES;
1252 #define QTYPE2NAME(t) (quotatypes[t])
1253
1254 static int ext4_write_dquot(struct dquot *dquot);
1255 static int ext4_acquire_dquot(struct dquot *dquot);
1256 static int ext4_release_dquot(struct dquot *dquot);
1257 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1258 static int ext4_write_info(struct super_block *sb, int type);
1259 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1260                          const struct path *path);
1261 static int ext4_quota_on_mount(struct super_block *sb, int type);
1262 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1263                                size_t len, loff_t off);
1264 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1265                                 const char *data, size_t len, loff_t off);
1266 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1267                              unsigned int flags);
1268 static int ext4_enable_quotas(struct super_block *sb);
1269 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1270
1271 static struct dquot **ext4_get_dquots(struct inode *inode)
1272 {
1273         return EXT4_I(inode)->i_dquot;
1274 }
1275
1276 static const struct dquot_operations ext4_quota_operations = {
1277         .get_reserved_space     = ext4_get_reserved_space,
1278         .write_dquot            = ext4_write_dquot,
1279         .acquire_dquot          = ext4_acquire_dquot,
1280         .release_dquot          = ext4_release_dquot,
1281         .mark_dirty             = ext4_mark_dquot_dirty,
1282         .write_info             = ext4_write_info,
1283         .alloc_dquot            = dquot_alloc,
1284         .destroy_dquot          = dquot_destroy,
1285         .get_projid             = ext4_get_projid,
1286         .get_inode_usage        = ext4_get_inode_usage,
1287         .get_next_id            = ext4_get_next_id,
1288 };
1289
1290 static const struct quotactl_ops ext4_qctl_operations = {
1291         .quota_on       = ext4_quota_on,
1292         .quota_off      = ext4_quota_off,
1293         .quota_sync     = dquot_quota_sync,
1294         .get_state      = dquot_get_state,
1295         .set_info       = dquot_set_dqinfo,
1296         .get_dqblk      = dquot_get_dqblk,
1297         .set_dqblk      = dquot_set_dqblk,
1298         .get_nextdqblk  = dquot_get_next_dqblk,
1299 };
1300 #endif
1301
1302 static const struct super_operations ext4_sops = {
1303         .alloc_inode    = ext4_alloc_inode,
1304         .destroy_inode  = ext4_destroy_inode,
1305         .write_inode    = ext4_write_inode,
1306         .dirty_inode    = ext4_dirty_inode,
1307         .drop_inode     = ext4_drop_inode,
1308         .evict_inode    = ext4_evict_inode,
1309         .put_super      = ext4_put_super,
1310         .sync_fs        = ext4_sync_fs,
1311         .freeze_fs      = ext4_freeze,
1312         .unfreeze_fs    = ext4_unfreeze,
1313         .statfs         = ext4_statfs,
1314         .remount_fs     = ext4_remount,
1315         .show_options   = ext4_show_options,
1316 #ifdef CONFIG_QUOTA
1317         .quota_read     = ext4_quota_read,
1318         .quota_write    = ext4_quota_write,
1319         .get_dquots     = ext4_get_dquots,
1320 #endif
1321         .bdev_try_to_free_page = bdev_try_to_free_page,
1322 };
1323
1324 static const struct export_operations ext4_export_ops = {
1325         .fh_to_dentry = ext4_fh_to_dentry,
1326         .fh_to_parent = ext4_fh_to_parent,
1327         .get_parent = ext4_get_parent,
1328 };
1329
1330 enum {
1331         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1332         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1333         Opt_nouid32, Opt_debug, Opt_removed,
1334         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1335         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1336         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1337         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1338         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1339         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1340         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1341         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1342         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1343         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1344         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1345         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1346         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1347         Opt_inode_readahead_blks, Opt_journal_ioprio,
1348         Opt_dioread_nolock, Opt_dioread_lock,
1349         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1350         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_removed, "oldalloc"},
1369         {Opt_removed, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "norecovery"},
1375         {Opt_noload, "noload"},
1376         {Opt_removed, "nobh"},
1377         {Opt_removed, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_path, "journal_path=%s"},
1383         {Opt_journal_checksum, "journal_checksum"},
1384         {Opt_nojournal_checksum, "nojournal_checksum"},
1385         {Opt_journal_async_commit, "journal_async_commit"},
1386         {Opt_abort, "abort"},
1387         {Opt_data_journal, "data=journal"},
1388         {Opt_data_ordered, "data=ordered"},
1389         {Opt_data_writeback, "data=writeback"},
1390         {Opt_data_err_abort, "data_err=abort"},
1391         {Opt_data_err_ignore, "data_err=ignore"},
1392         {Opt_offusrjquota, "usrjquota="},
1393         {Opt_usrjquota, "usrjquota=%s"},
1394         {Opt_offgrpjquota, "grpjquota="},
1395         {Opt_grpjquota, "grpjquota=%s"},
1396         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1397         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1398         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1399         {Opt_grpquota, "grpquota"},
1400         {Opt_noquota, "noquota"},
1401         {Opt_quota, "quota"},
1402         {Opt_usrquota, "usrquota"},
1403         {Opt_prjquota, "prjquota"},
1404         {Opt_barrier, "barrier=%u"},
1405         {Opt_barrier, "barrier"},
1406         {Opt_nobarrier, "nobarrier"},
1407         {Opt_i_version, "i_version"},
1408         {Opt_dax, "dax"},
1409         {Opt_stripe, "stripe=%u"},
1410         {Opt_delalloc, "delalloc"},
1411         {Opt_lazytime, "lazytime"},
1412         {Opt_nolazytime, "nolazytime"},
1413         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1414         {Opt_nodelalloc, "nodelalloc"},
1415         {Opt_removed, "mblk_io_submit"},
1416         {Opt_removed, "nomblk_io_submit"},
1417         {Opt_block_validity, "block_validity"},
1418         {Opt_noblock_validity, "noblock_validity"},
1419         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1420         {Opt_journal_ioprio, "journal_ioprio=%u"},
1421         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1422         {Opt_auto_da_alloc, "auto_da_alloc"},
1423         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1424         {Opt_dioread_nolock, "dioread_nolock"},
1425         {Opt_dioread_lock, "dioread_lock"},
1426         {Opt_discard, "discard"},
1427         {Opt_nodiscard, "nodiscard"},
1428         {Opt_init_itable, "init_itable=%u"},
1429         {Opt_init_itable, "init_itable"},
1430         {Opt_noinit_itable, "noinit_itable"},
1431         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1432         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1433         {Opt_nombcache, "nombcache"},
1434         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1435         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1436         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1437         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1438         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1439         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1440         {Opt_err, NULL},
1441 };
1442
1443 static ext4_fsblk_t get_sb_block(void **data)
1444 {
1445         ext4_fsblk_t    sb_block;
1446         char            *options = (char *) *data;
1447
1448         if (!options || strncmp(options, "sb=", 3) != 0)
1449                 return 1;       /* Default location */
1450
1451         options += 3;
1452         /* TODO: use simple_strtoll with >32bit ext4 */
1453         sb_block = simple_strtoul(options, &options, 0);
1454         if (*options && *options != ',') {
1455                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1456                        (char *) *data);
1457                 return 1;
1458         }
1459         if (*options == ',')
1460                 options++;
1461         *data = (void *) options;
1462
1463         return sb_block;
1464 }
1465
1466 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1467 static const char deprecated_msg[] =
1468         "Mount option \"%s\" will be removed by %s\n"
1469         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1470
1471 #ifdef CONFIG_QUOTA
1472 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1473 {
1474         struct ext4_sb_info *sbi = EXT4_SB(sb);
1475         char *qname;
1476         int ret = -1;
1477
1478         if (sb_any_quota_loaded(sb) &&
1479                 !sbi->s_qf_names[qtype]) {
1480                 ext4_msg(sb, KERN_ERR,
1481                         "Cannot change journaled "
1482                         "quota options when quota turned on");
1483                 return -1;
1484         }
1485         if (ext4_has_feature_quota(sb)) {
1486                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1487                          "ignored when QUOTA feature is enabled");
1488                 return 1;
1489         }
1490         qname = match_strdup(args);
1491         if (!qname) {
1492                 ext4_msg(sb, KERN_ERR,
1493                         "Not enough memory for storing quotafile name");
1494                 return -1;
1495         }
1496         if (sbi->s_qf_names[qtype]) {
1497                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1498                         ret = 1;
1499                 else
1500                         ext4_msg(sb, KERN_ERR,
1501                                  "%s quota file already specified",
1502                                  QTYPE2NAME(qtype));
1503                 goto errout;
1504         }
1505         if (strchr(qname, '/')) {
1506                 ext4_msg(sb, KERN_ERR,
1507                         "quotafile must be on filesystem root");
1508                 goto errout;
1509         }
1510         sbi->s_qf_names[qtype] = qname;
1511         set_opt(sb, QUOTA);
1512         return 1;
1513 errout:
1514         kfree(qname);
1515         return ret;
1516 }
1517
1518 static int clear_qf_name(struct super_block *sb, int qtype)
1519 {
1520
1521         struct ext4_sb_info *sbi = EXT4_SB(sb);
1522
1523         if (sb_any_quota_loaded(sb) &&
1524                 sbi->s_qf_names[qtype]) {
1525                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1526                         " when quota turned on");
1527                 return -1;
1528         }
1529         kfree(sbi->s_qf_names[qtype]);
1530         sbi->s_qf_names[qtype] = NULL;
1531         return 1;
1532 }
1533 #endif
1534
1535 #define MOPT_SET        0x0001
1536 #define MOPT_CLEAR      0x0002
1537 #define MOPT_NOSUPPORT  0x0004
1538 #define MOPT_EXPLICIT   0x0008
1539 #define MOPT_CLEAR_ERR  0x0010
1540 #define MOPT_GTE0       0x0020
1541 #ifdef CONFIG_QUOTA
1542 #define MOPT_Q          0
1543 #define MOPT_QFMT       0x0040
1544 #else
1545 #define MOPT_Q          MOPT_NOSUPPORT
1546 #define MOPT_QFMT       MOPT_NOSUPPORT
1547 #endif
1548 #define MOPT_DATAJ      0x0080
1549 #define MOPT_NO_EXT2    0x0100
1550 #define MOPT_NO_EXT3    0x0200
1551 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1552 #define MOPT_STRING     0x0400
1553
1554 static const struct mount_opts {
1555         int     token;
1556         int     mount_opt;
1557         int     flags;
1558 } ext4_mount_opts[] = {
1559         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1560         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1561         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1562         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1563         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1564         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1565         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1566          MOPT_EXT4_ONLY | MOPT_SET},
1567         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1568          MOPT_EXT4_ONLY | MOPT_CLEAR},
1569         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1570         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1571         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1572          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1573         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1574          MOPT_EXT4_ONLY | MOPT_CLEAR},
1575         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1576          MOPT_EXT4_ONLY | MOPT_CLEAR},
1577         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1578          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1579         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1580                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1581          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1582         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1583         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1584         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1585         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1586         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1587          MOPT_NO_EXT2},
1588         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1589          MOPT_NO_EXT2},
1590         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1591         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1592         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1593         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1594         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1595         {Opt_commit, 0, MOPT_GTE0},
1596         {Opt_max_batch_time, 0, MOPT_GTE0},
1597         {Opt_min_batch_time, 0, MOPT_GTE0},
1598         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1599         {Opt_init_itable, 0, MOPT_GTE0},
1600         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1601         {Opt_stripe, 0, MOPT_GTE0},
1602         {Opt_resuid, 0, MOPT_GTE0},
1603         {Opt_resgid, 0, MOPT_GTE0},
1604         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1605         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1606         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1607         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1608         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1609         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1610          MOPT_NO_EXT2 | MOPT_DATAJ},
1611         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1612         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1613 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1614         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1615         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1616 #else
1617         {Opt_acl, 0, MOPT_NOSUPPORT},
1618         {Opt_noacl, 0, MOPT_NOSUPPORT},
1619 #endif
1620         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1621         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1622         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1623         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1624         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1625                                                         MOPT_SET | MOPT_Q},
1626         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1627                                                         MOPT_SET | MOPT_Q},
1628         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1629                                                         MOPT_SET | MOPT_Q},
1630         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1631                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1632                                                         MOPT_CLEAR | MOPT_Q},
1633         {Opt_usrjquota, 0, MOPT_Q},
1634         {Opt_grpjquota, 0, MOPT_Q},
1635         {Opt_offusrjquota, 0, MOPT_Q},
1636         {Opt_offgrpjquota, 0, MOPT_Q},
1637         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1638         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1639         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1640         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1641         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1642         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1643         {Opt_err, 0, 0}
1644 };
1645
1646 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1647                             substring_t *args, unsigned long *journal_devnum,
1648                             unsigned int *journal_ioprio, int is_remount)
1649 {
1650         struct ext4_sb_info *sbi = EXT4_SB(sb);
1651         const struct mount_opts *m;
1652         kuid_t uid;
1653         kgid_t gid;
1654         int arg = 0;
1655
1656 #ifdef CONFIG_QUOTA
1657         if (token == Opt_usrjquota)
1658                 return set_qf_name(sb, USRQUOTA, &args[0]);
1659         else if (token == Opt_grpjquota)
1660                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1661         else if (token == Opt_offusrjquota)
1662                 return clear_qf_name(sb, USRQUOTA);
1663         else if (token == Opt_offgrpjquota)
1664                 return clear_qf_name(sb, GRPQUOTA);
1665 #endif
1666         switch (token) {
1667         case Opt_noacl:
1668         case Opt_nouser_xattr:
1669                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1670                 break;
1671         case Opt_sb:
1672                 return 1;       /* handled by get_sb_block() */
1673         case Opt_removed:
1674                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1675                 return 1;
1676         case Opt_abort:
1677                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1678                 return 1;
1679         case Opt_i_version:
1680                 sb->s_flags |= SB_I_VERSION;
1681                 return 1;
1682         case Opt_lazytime:
1683                 sb->s_flags |= MS_LAZYTIME;
1684                 return 1;
1685         case Opt_nolazytime:
1686                 sb->s_flags &= ~MS_LAZYTIME;
1687                 return 1;
1688         }
1689
1690         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1691                 if (token == m->token)
1692                         break;
1693
1694         if (m->token == Opt_err) {
1695                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1696                          "or missing value", opt);
1697                 return -1;
1698         }
1699
1700         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1701                 ext4_msg(sb, KERN_ERR,
1702                          "Mount option \"%s\" incompatible with ext2", opt);
1703                 return -1;
1704         }
1705         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1706                 ext4_msg(sb, KERN_ERR,
1707                          "Mount option \"%s\" incompatible with ext3", opt);
1708                 return -1;
1709         }
1710
1711         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1712                 return -1;
1713         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1714                 return -1;
1715         if (m->flags & MOPT_EXPLICIT) {
1716                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1717                         set_opt2(sb, EXPLICIT_DELALLOC);
1718                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1719                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1720                 } else
1721                         return -1;
1722         }
1723         if (m->flags & MOPT_CLEAR_ERR)
1724                 clear_opt(sb, ERRORS_MASK);
1725         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1726                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1727                          "options when quota turned on");
1728                 return -1;
1729         }
1730
1731         if (m->flags & MOPT_NOSUPPORT) {
1732                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1733         } else if (token == Opt_commit) {
1734                 if (arg == 0)
1735                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1736                 sbi->s_commit_interval = HZ * arg;
1737         } else if (token == Opt_debug_want_extra_isize) {
1738                 sbi->s_want_extra_isize = arg;
1739         } else if (token == Opt_max_batch_time) {
1740                 sbi->s_max_batch_time = arg;
1741         } else if (token == Opt_min_batch_time) {
1742                 sbi->s_min_batch_time = arg;
1743         } else if (token == Opt_inode_readahead_blks) {
1744                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1745                         ext4_msg(sb, KERN_ERR,
1746                                  "EXT4-fs: inode_readahead_blks must be "
1747                                  "0 or a power of 2 smaller than 2^31");
1748                         return -1;
1749                 }
1750                 sbi->s_inode_readahead_blks = arg;
1751         } else if (token == Opt_init_itable) {
1752                 set_opt(sb, INIT_INODE_TABLE);
1753                 if (!args->from)
1754                         arg = EXT4_DEF_LI_WAIT_MULT;
1755                 sbi->s_li_wait_mult = arg;
1756         } else if (token == Opt_max_dir_size_kb) {
1757                 sbi->s_max_dir_size_kb = arg;
1758         } else if (token == Opt_stripe) {
1759                 sbi->s_stripe = arg;
1760         } else if (token == Opt_resuid) {
1761                 uid = make_kuid(current_user_ns(), arg);
1762                 if (!uid_valid(uid)) {
1763                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1764                         return -1;
1765                 }
1766                 sbi->s_resuid = uid;
1767         } else if (token == Opt_resgid) {
1768                 gid = make_kgid(current_user_ns(), arg);
1769                 if (!gid_valid(gid)) {
1770                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1771                         return -1;
1772                 }
1773                 sbi->s_resgid = gid;
1774         } else if (token == Opt_journal_dev) {
1775                 if (is_remount) {
1776                         ext4_msg(sb, KERN_ERR,
1777                                  "Cannot specify journal on remount");
1778                         return -1;
1779                 }
1780                 *journal_devnum = arg;
1781         } else if (token == Opt_journal_path) {
1782                 char *journal_path;
1783                 struct inode *journal_inode;
1784                 struct path path;
1785                 int error;
1786
1787                 if (is_remount) {
1788                         ext4_msg(sb, KERN_ERR,
1789                                  "Cannot specify journal on remount");
1790                         return -1;
1791                 }
1792                 journal_path = match_strdup(&args[0]);
1793                 if (!journal_path) {
1794                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1795                                 "journal device string");
1796                         return -1;
1797                 }
1798
1799                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1800                 if (error) {
1801                         ext4_msg(sb, KERN_ERR, "error: could not find "
1802                                 "journal device path: error %d", error);
1803                         kfree(journal_path);
1804                         return -1;
1805                 }
1806
1807                 journal_inode = d_inode(path.dentry);
1808                 if (!S_ISBLK(journal_inode->i_mode)) {
1809                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1810                                 "is not a block device", journal_path);
1811                         path_put(&path);
1812                         kfree(journal_path);
1813                         return -1;
1814                 }
1815
1816                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1817                 path_put(&path);
1818                 kfree(journal_path);
1819         } else if (token == Opt_journal_ioprio) {
1820                 if (arg > 7) {
1821                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1822                                  " (must be 0-7)");
1823                         return -1;
1824                 }
1825                 *journal_ioprio =
1826                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1827         } else if (token == Opt_test_dummy_encryption) {
1828 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1829                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1830                 ext4_msg(sb, KERN_WARNING,
1831                          "Test dummy encryption mode enabled");
1832 #else
1833                 ext4_msg(sb, KERN_WARNING,
1834                          "Test dummy encryption mount option ignored");
1835 #endif
1836         } else if (m->flags & MOPT_DATAJ) {
1837                 if (is_remount) {
1838                         if (!sbi->s_journal)
1839                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1840                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1841                                 ext4_msg(sb, KERN_ERR,
1842                                          "Cannot change data mode on remount");
1843                                 return -1;
1844                         }
1845                 } else {
1846                         clear_opt(sb, DATA_FLAGS);
1847                         sbi->s_mount_opt |= m->mount_opt;
1848                 }
1849 #ifdef CONFIG_QUOTA
1850         } else if (m->flags & MOPT_QFMT) {
1851                 if (sb_any_quota_loaded(sb) &&
1852                     sbi->s_jquota_fmt != m->mount_opt) {
1853                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1854                                  "quota options when quota turned on");
1855                         return -1;
1856                 }
1857                 if (ext4_has_feature_quota(sb)) {
1858                         ext4_msg(sb, KERN_INFO,
1859                                  "Quota format mount options ignored "
1860                                  "when QUOTA feature is enabled");
1861                         return 1;
1862                 }
1863                 sbi->s_jquota_fmt = m->mount_opt;
1864 #endif
1865         } else if (token == Opt_dax) {
1866 #ifdef CONFIG_FS_DAX
1867                 ext4_msg(sb, KERN_WARNING,
1868                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1869                         sbi->s_mount_opt |= m->mount_opt;
1870 #else
1871                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1872                 return -1;
1873 #endif
1874         } else if (token == Opt_data_err_abort) {
1875                 sbi->s_mount_opt |= m->mount_opt;
1876         } else if (token == Opt_data_err_ignore) {
1877                 sbi->s_mount_opt &= ~m->mount_opt;
1878         } else {
1879                 if (!args->from)
1880                         arg = 1;
1881                 if (m->flags & MOPT_CLEAR)
1882                         arg = !arg;
1883                 else if (unlikely(!(m->flags & MOPT_SET))) {
1884                         ext4_msg(sb, KERN_WARNING,
1885                                  "buggy handling of option %s", opt);
1886                         WARN_ON(1);
1887                         return -1;
1888                 }
1889                 if (arg != 0)
1890                         sbi->s_mount_opt |= m->mount_opt;
1891                 else
1892                         sbi->s_mount_opt &= ~m->mount_opt;
1893         }
1894         return 1;
1895 }
1896
1897 static int parse_options(char *options, struct super_block *sb,
1898                          unsigned long *journal_devnum,
1899                          unsigned int *journal_ioprio,
1900                          int is_remount)
1901 {
1902         struct ext4_sb_info *sbi = EXT4_SB(sb);
1903         char *p;
1904         substring_t args[MAX_OPT_ARGS];
1905         int token;
1906
1907         if (!options)
1908                 return 1;
1909
1910         while ((p = strsep(&options, ",")) != NULL) {
1911                 if (!*p)
1912                         continue;
1913                 /*
1914                  * Initialize args struct so we know whether arg was
1915                  * found; some options take optional arguments.
1916                  */
1917                 args[0].to = args[0].from = NULL;
1918                 token = match_token(p, tokens, args);
1919                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1920                                      journal_ioprio, is_remount) < 0)
1921                         return 0;
1922         }
1923 #ifdef CONFIG_QUOTA
1924         /*
1925          * We do the test below only for project quotas. 'usrquota' and
1926          * 'grpquota' mount options are allowed even without quota feature
1927          * to support legacy quotas in quota files.
1928          */
1929         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1930                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1931                          "Cannot enable project quota enforcement.");
1932                 return 0;
1933         }
1934         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1935                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1936                         clear_opt(sb, USRQUOTA);
1937
1938                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1939                         clear_opt(sb, GRPQUOTA);
1940
1941                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1942                         ext4_msg(sb, KERN_ERR, "old and new quota "
1943                                         "format mixing");
1944                         return 0;
1945                 }
1946
1947                 if (!sbi->s_jquota_fmt) {
1948                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1949                                         "not specified");
1950                         return 0;
1951                 }
1952         }
1953 #endif
1954         if (test_opt(sb, DIOREAD_NOLOCK)) {
1955                 int blocksize =
1956                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1957
1958                 if (blocksize < PAGE_SIZE) {
1959                         ext4_msg(sb, KERN_ERR, "can't mount with "
1960                                  "dioread_nolock if block size != PAGE_SIZE");
1961                         return 0;
1962                 }
1963         }
1964         return 1;
1965 }
1966
1967 static inline void ext4_show_quota_options(struct seq_file *seq,
1968                                            struct super_block *sb)
1969 {
1970 #if defined(CONFIG_QUOTA)
1971         struct ext4_sb_info *sbi = EXT4_SB(sb);
1972
1973         if (sbi->s_jquota_fmt) {
1974                 char *fmtname = "";
1975
1976                 switch (sbi->s_jquota_fmt) {
1977                 case QFMT_VFS_OLD:
1978                         fmtname = "vfsold";
1979                         break;
1980                 case QFMT_VFS_V0:
1981                         fmtname = "vfsv0";
1982                         break;
1983                 case QFMT_VFS_V1:
1984                         fmtname = "vfsv1";
1985                         break;
1986                 }
1987                 seq_printf(seq, ",jqfmt=%s", fmtname);
1988         }
1989
1990         if (sbi->s_qf_names[USRQUOTA])
1991                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1992
1993         if (sbi->s_qf_names[GRPQUOTA])
1994                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1995 #endif
1996 }
1997
1998 static const char *token2str(int token)
1999 {
2000         const struct match_token *t;
2001
2002         for (t = tokens; t->token != Opt_err; t++)
2003                 if (t->token == token && !strchr(t->pattern, '='))
2004                         break;
2005         return t->pattern;
2006 }
2007
2008 /*
2009  * Show an option if
2010  *  - it's set to a non-default value OR
2011  *  - if the per-sb default is different from the global default
2012  */
2013 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2014                               int nodefs)
2015 {
2016         struct ext4_sb_info *sbi = EXT4_SB(sb);
2017         struct ext4_super_block *es = sbi->s_es;
2018         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2019         const struct mount_opts *m;
2020         char sep = nodefs ? '\n' : ',';
2021
2022 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2023 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2024
2025         if (sbi->s_sb_block != 1)
2026                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2027
2028         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2029                 int want_set = m->flags & MOPT_SET;
2030                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2031                     (m->flags & MOPT_CLEAR_ERR))
2032                         continue;
2033                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2034                         continue; /* skip if same as the default */
2035                 if ((want_set &&
2036                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2037                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2038                         continue; /* select Opt_noFoo vs Opt_Foo */
2039                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2040         }
2041
2042         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2043             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2044                 SEQ_OPTS_PRINT("resuid=%u",
2045                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2046         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2047             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2048                 SEQ_OPTS_PRINT("resgid=%u",
2049                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2050         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2051         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2052                 SEQ_OPTS_PUTS("errors=remount-ro");
2053         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2054                 SEQ_OPTS_PUTS("errors=continue");
2055         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2056                 SEQ_OPTS_PUTS("errors=panic");
2057         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2058                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2059         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2060                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2061         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2062                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2063         if (sb->s_flags & SB_I_VERSION)
2064                 SEQ_OPTS_PUTS("i_version");
2065         if (nodefs || sbi->s_stripe)
2066                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2067         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2068                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2069                         SEQ_OPTS_PUTS("data=journal");
2070                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2071                         SEQ_OPTS_PUTS("data=ordered");
2072                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2073                         SEQ_OPTS_PUTS("data=writeback");
2074         }
2075         if (nodefs ||
2076             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2077                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2078                                sbi->s_inode_readahead_blks);
2079
2080         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2081                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2082                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2083         if (nodefs || sbi->s_max_dir_size_kb)
2084                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2085         if (test_opt(sb, DATA_ERR_ABORT))
2086                 SEQ_OPTS_PUTS("data_err=abort");
2087
2088         ext4_show_quota_options(seq, sb);
2089         return 0;
2090 }
2091
2092 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2093 {
2094         return _ext4_show_options(seq, root->d_sb, 0);
2095 }
2096
2097 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2098 {
2099         struct super_block *sb = seq->private;
2100         int rc;
2101
2102         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2103         rc = _ext4_show_options(seq, sb, 1);
2104         seq_puts(seq, "\n");
2105         return rc;
2106 }
2107
2108 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2109                             int read_only)
2110 {
2111         struct ext4_sb_info *sbi = EXT4_SB(sb);
2112         int res = 0;
2113
2114         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2115                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2116                          "forcing read-only mode");
2117                 res = MS_RDONLY;
2118         }
2119         if (read_only)
2120                 goto done;
2121         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2122                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2123                          "running e2fsck is recommended");
2124         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2125                 ext4_msg(sb, KERN_WARNING,
2126                          "warning: mounting fs with errors, "
2127                          "running e2fsck is recommended");
2128         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2129                  le16_to_cpu(es->s_mnt_count) >=
2130                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2131                 ext4_msg(sb, KERN_WARNING,
2132                          "warning: maximal mount count reached, "
2133                          "running e2fsck is recommended");
2134         else if (le32_to_cpu(es->s_checkinterval) &&
2135                 (le32_to_cpu(es->s_lastcheck) +
2136                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2137                 ext4_msg(sb, KERN_WARNING,
2138                          "warning: checktime reached, "
2139                          "running e2fsck is recommended");
2140         if (!sbi->s_journal)
2141                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2142         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2143                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2144         le16_add_cpu(&es->s_mnt_count, 1);
2145         es->s_mtime = cpu_to_le32(get_seconds());
2146         ext4_update_dynamic_rev(sb);
2147         if (sbi->s_journal)
2148                 ext4_set_feature_journal_needs_recovery(sb);
2149
2150         ext4_commit_super(sb, 1);
2151 done:
2152         if (test_opt(sb, DEBUG))
2153                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2154                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2155                         sb->s_blocksize,
2156                         sbi->s_groups_count,
2157                         EXT4_BLOCKS_PER_GROUP(sb),
2158                         EXT4_INODES_PER_GROUP(sb),
2159                         sbi->s_mount_opt, sbi->s_mount_opt2);
2160
2161         cleancache_init_fs(sb);
2162         return res;
2163 }
2164
2165 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2166 {
2167         struct ext4_sb_info *sbi = EXT4_SB(sb);
2168         struct flex_groups *new_groups;
2169         int size;
2170
2171         if (!sbi->s_log_groups_per_flex)
2172                 return 0;
2173
2174         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2175         if (size <= sbi->s_flex_groups_allocated)
2176                 return 0;
2177
2178         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2179         new_groups = kvzalloc(size, GFP_KERNEL);
2180         if (!new_groups) {
2181                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2182                          size / (int) sizeof(struct flex_groups));
2183                 return -ENOMEM;
2184         }
2185
2186         if (sbi->s_flex_groups) {
2187                 memcpy(new_groups, sbi->s_flex_groups,
2188                        (sbi->s_flex_groups_allocated *
2189                         sizeof(struct flex_groups)));
2190                 kvfree(sbi->s_flex_groups);
2191         }
2192         sbi->s_flex_groups = new_groups;
2193         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2194         return 0;
2195 }
2196
2197 static int ext4_fill_flex_info(struct super_block *sb)
2198 {
2199         struct ext4_sb_info *sbi = EXT4_SB(sb);
2200         struct ext4_group_desc *gdp = NULL;
2201         ext4_group_t flex_group;
2202         int i, err;
2203
2204         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2205         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2206                 sbi->s_log_groups_per_flex = 0;
2207                 return 1;
2208         }
2209
2210         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2211         if (err)
2212                 goto failed;
2213
2214         for (i = 0; i < sbi->s_groups_count; i++) {
2215                 gdp = ext4_get_group_desc(sb, i, NULL);
2216
2217                 flex_group = ext4_flex_group(sbi, i);
2218                 atomic_add(ext4_free_inodes_count(sb, gdp),
2219                            &sbi->s_flex_groups[flex_group].free_inodes);
2220                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2221                              &sbi->s_flex_groups[flex_group].free_clusters);
2222                 atomic_add(ext4_used_dirs_count(sb, gdp),
2223                            &sbi->s_flex_groups[flex_group].used_dirs);
2224         }
2225
2226         return 1;
2227 failed:
2228         return 0;
2229 }
2230
2231 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2232                                    struct ext4_group_desc *gdp)
2233 {
2234         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2235         __u16 crc = 0;
2236         __le32 le_group = cpu_to_le32(block_group);
2237         struct ext4_sb_info *sbi = EXT4_SB(sb);
2238
2239         if (ext4_has_metadata_csum(sbi->s_sb)) {
2240                 /* Use new metadata_csum algorithm */
2241                 __u32 csum32;
2242                 __u16 dummy_csum = 0;
2243
2244                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2245                                      sizeof(le_group));
2246                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2247                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2248                                      sizeof(dummy_csum));
2249                 offset += sizeof(dummy_csum);
2250                 if (offset < sbi->s_desc_size)
2251                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2252                                              sbi->s_desc_size - offset);
2253
2254                 crc = csum32 & 0xFFFF;
2255                 goto out;
2256         }
2257
2258         /* old crc16 code */
2259         if (!ext4_has_feature_gdt_csum(sb))
2260                 return 0;
2261
2262         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2263         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2264         crc = crc16(crc, (__u8 *)gdp, offset);
2265         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2266         /* for checksum of struct ext4_group_desc do the rest...*/
2267         if (ext4_has_feature_64bit(sb) &&
2268             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2269                 crc = crc16(crc, (__u8 *)gdp + offset,
2270                             le16_to_cpu(sbi->s_es->s_desc_size) -
2271                                 offset);
2272
2273 out:
2274         return cpu_to_le16(crc);
2275 }
2276
2277 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2278                                 struct ext4_group_desc *gdp)
2279 {
2280         if (ext4_has_group_desc_csum(sb) &&
2281             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2282                 return 0;
2283
2284         return 1;
2285 }
2286
2287 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2288                               struct ext4_group_desc *gdp)
2289 {
2290         if (!ext4_has_group_desc_csum(sb))
2291                 return;
2292         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2293 }
2294
2295 /* Called at mount-time, super-block is locked */
2296 static int ext4_check_descriptors(struct super_block *sb,
2297                                   ext4_fsblk_t sb_block,
2298                                   ext4_group_t *first_not_zeroed)
2299 {
2300         struct ext4_sb_info *sbi = EXT4_SB(sb);
2301         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2302         ext4_fsblk_t last_block;
2303         ext4_fsblk_t block_bitmap;
2304         ext4_fsblk_t inode_bitmap;
2305         ext4_fsblk_t inode_table;
2306         int flexbg_flag = 0;
2307         ext4_group_t i, grp = sbi->s_groups_count;
2308
2309         if (ext4_has_feature_flex_bg(sb))
2310                 flexbg_flag = 1;
2311
2312         ext4_debug("Checking group descriptors");
2313
2314         for (i = 0; i < sbi->s_groups_count; i++) {
2315                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2316
2317                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2318                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2319                 else
2320                         last_block = first_block +
2321                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2322
2323                 if ((grp == sbi->s_groups_count) &&
2324                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2325                         grp = i;
2326
2327                 block_bitmap = ext4_block_bitmap(sb, gdp);
2328                 if (block_bitmap == sb_block) {
2329                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2330                                  "Block bitmap for group %u overlaps "
2331                                  "superblock", i);
2332                 }
2333                 if (block_bitmap < first_block || block_bitmap > last_block) {
2334                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2335                                "Block bitmap for group %u not in group "
2336                                "(block %llu)!", i, block_bitmap);
2337                         return 0;
2338                 }
2339                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2340                 if (inode_bitmap == sb_block) {
2341                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2342                                  "Inode bitmap for group %u overlaps "
2343                                  "superblock", i);
2344                 }
2345                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2346                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2347                                "Inode bitmap for group %u not in group "
2348                                "(block %llu)!", i, inode_bitmap);
2349                         return 0;
2350                 }
2351                 inode_table = ext4_inode_table(sb, gdp);
2352                 if (inode_table == sb_block) {
2353                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2354                                  "Inode table for group %u overlaps "
2355                                  "superblock", i);
2356                 }
2357                 if (inode_table < first_block ||
2358                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2359                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2360                                "Inode table for group %u not in group "
2361                                "(block %llu)!", i, inode_table);
2362                         return 0;
2363                 }
2364                 ext4_lock_group(sb, i);
2365                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2366                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2367                                  "Checksum for group %u failed (%u!=%u)",
2368                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2369                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2370                         if (!sb_rdonly(sb)) {
2371                                 ext4_unlock_group(sb, i);
2372                                 return 0;
2373                         }
2374                 }
2375                 ext4_unlock_group(sb, i);
2376                 if (!flexbg_flag)
2377                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2378         }
2379         if (NULL != first_not_zeroed)
2380                 *first_not_zeroed = grp;
2381         return 1;
2382 }
2383
2384 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2385  * the superblock) which were deleted from all directories, but held open by
2386  * a process at the time of a crash.  We walk the list and try to delete these
2387  * inodes at recovery time (only with a read-write filesystem).
2388  *
2389  * In order to keep the orphan inode chain consistent during traversal (in
2390  * case of crash during recovery), we link each inode into the superblock
2391  * orphan list_head and handle it the same way as an inode deletion during
2392  * normal operation (which journals the operations for us).
2393  *
2394  * We only do an iget() and an iput() on each inode, which is very safe if we
2395  * accidentally point at an in-use or already deleted inode.  The worst that
2396  * can happen in this case is that we get a "bit already cleared" message from
2397  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2398  * e2fsck was run on this filesystem, and it must have already done the orphan
2399  * inode cleanup for us, so we can safely abort without any further action.
2400  */
2401 static void ext4_orphan_cleanup(struct super_block *sb,
2402                                 struct ext4_super_block *es)
2403 {
2404         unsigned int s_flags = sb->s_flags;
2405         int ret, nr_orphans = 0, nr_truncates = 0;
2406 #ifdef CONFIG_QUOTA
2407         int quota_update = 0;
2408         int i;
2409 #endif
2410         if (!es->s_last_orphan) {
2411                 jbd_debug(4, "no orphan inodes to clean up\n");
2412                 return;
2413         }
2414
2415         if (bdev_read_only(sb->s_bdev)) {
2416                 ext4_msg(sb, KERN_ERR, "write access "
2417                         "unavailable, skipping orphan cleanup");
2418                 return;
2419         }
2420
2421         /* Check if feature set would not allow a r/w mount */
2422         if (!ext4_feature_set_ok(sb, 0)) {
2423                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2424                          "unknown ROCOMPAT features");
2425                 return;
2426         }
2427
2428         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2429                 /* don't clear list on RO mount w/ errors */
2430                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2431                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2432                                   "clearing orphan list.\n");
2433                         es->s_last_orphan = 0;
2434                 }
2435                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2436                 return;
2437         }
2438
2439         if (s_flags & MS_RDONLY) {
2440                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2441                 sb->s_flags &= ~MS_RDONLY;
2442         }
2443 #ifdef CONFIG_QUOTA
2444         /* Needed for iput() to work correctly and not trash data */
2445         sb->s_flags |= MS_ACTIVE;
2446
2447         /*
2448          * Turn on quotas which were not enabled for read-only mounts if
2449          * filesystem has quota feature, so that they are updated correctly.
2450          */
2451         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2452                 int ret = ext4_enable_quotas(sb);
2453
2454                 if (!ret)
2455                         quota_update = 1;
2456                 else
2457                         ext4_msg(sb, KERN_ERR,
2458                                 "Cannot turn on quotas: error %d", ret);
2459         }
2460
2461         /* Turn on journaled quotas used for old sytle */
2462         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2463                 if (EXT4_SB(sb)->s_qf_names[i]) {
2464                         int ret = ext4_quota_on_mount(sb, i);
2465
2466                         if (!ret)
2467                                 quota_update = 1;
2468                         else
2469                                 ext4_msg(sb, KERN_ERR,
2470                                         "Cannot turn on journaled "
2471                                         "quota: type %d: error %d", i, ret);
2472                 }
2473         }
2474 #endif
2475
2476         while (es->s_last_orphan) {
2477                 struct inode *inode;
2478
2479                 /*
2480                  * We may have encountered an error during cleanup; if
2481                  * so, skip the rest.
2482                  */
2483                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2484                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2485                         es->s_last_orphan = 0;
2486                         break;
2487                 }
2488
2489                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2490                 if (IS_ERR(inode)) {
2491                         es->s_last_orphan = 0;
2492                         break;
2493                 }
2494
2495                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2496                 dquot_initialize(inode);
2497                 if (inode->i_nlink) {
2498                         if (test_opt(sb, DEBUG))
2499                                 ext4_msg(sb, KERN_DEBUG,
2500                                         "%s: truncating inode %lu to %lld bytes",
2501                                         __func__, inode->i_ino, inode->i_size);
2502                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2503                                   inode->i_ino, inode->i_size);
2504                         inode_lock(inode);
2505                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2506                         ret = ext4_truncate(inode);
2507                         if (ret)
2508                                 ext4_std_error(inode->i_sb, ret);
2509                         inode_unlock(inode);
2510                         nr_truncates++;
2511                 } else {
2512                         if (test_opt(sb, DEBUG))
2513                                 ext4_msg(sb, KERN_DEBUG,
2514                                         "%s: deleting unreferenced inode %lu",
2515                                         __func__, inode->i_ino);
2516                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2517                                   inode->i_ino);
2518                         nr_orphans++;
2519                 }
2520                 iput(inode);  /* The delete magic happens here! */
2521         }
2522
2523 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2524
2525         if (nr_orphans)
2526                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2527                        PLURAL(nr_orphans));
2528         if (nr_truncates)
2529                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2530                        PLURAL(nr_truncates));
2531 #ifdef CONFIG_QUOTA
2532         /* Turn off quotas if they were enabled for orphan cleanup */
2533         if (quota_update) {
2534                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2535                         if (sb_dqopt(sb)->files[i])
2536                                 dquot_quota_off(sb, i);
2537                 }
2538         }
2539 #endif
2540         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2541 }
2542
2543 /*
2544  * Maximal extent format file size.
2545  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2546  * extent format containers, within a sector_t, and within i_blocks
2547  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2548  * so that won't be a limiting factor.
2549  *
2550  * However there is other limiting factor. We do store extents in the form
2551  * of starting block and length, hence the resulting length of the extent
2552  * covering maximum file size must fit into on-disk format containers as
2553  * well. Given that length is always by 1 unit bigger than max unit (because
2554  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2555  *
2556  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2557  */
2558 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2559 {
2560         loff_t res;
2561         loff_t upper_limit = MAX_LFS_FILESIZE;
2562
2563         /* small i_blocks in vfs inode? */
2564         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2565                 /*
2566                  * CONFIG_LBDAF is not enabled implies the inode
2567                  * i_block represent total blocks in 512 bytes
2568                  * 32 == size of vfs inode i_blocks * 8
2569                  */
2570                 upper_limit = (1LL << 32) - 1;
2571
2572                 /* total blocks in file system block size */
2573                 upper_limit >>= (blkbits - 9);
2574                 upper_limit <<= blkbits;
2575         }
2576
2577         /*
2578          * 32-bit extent-start container, ee_block. We lower the maxbytes
2579          * by one fs block, so ee_len can cover the extent of maximum file
2580          * size
2581          */
2582         res = (1LL << 32) - 1;
2583         res <<= blkbits;
2584
2585         /* Sanity check against vm- & vfs- imposed limits */
2586         if (res > upper_limit)
2587                 res = upper_limit;
2588
2589         return res;
2590 }
2591
2592 /*
2593  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2594  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2595  * We need to be 1 filesystem block less than the 2^48 sector limit.
2596  */
2597 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2598 {
2599         loff_t res = EXT4_NDIR_BLOCKS;
2600         int meta_blocks;
2601         loff_t upper_limit;
2602         /* This is calculated to be the largest file size for a dense, block
2603          * mapped file such that the file's total number of 512-byte sectors,
2604          * including data and all indirect blocks, does not exceed (2^48 - 1).
2605          *
2606          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2607          * number of 512-byte sectors of the file.
2608          */
2609
2610         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2611                 /*
2612                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2613                  * the inode i_block field represents total file blocks in
2614                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2615                  */
2616                 upper_limit = (1LL << 32) - 1;
2617
2618                 /* total blocks in file system block size */
2619                 upper_limit >>= (bits - 9);
2620
2621         } else {
2622                 /*
2623                  * We use 48 bit ext4_inode i_blocks
2624                  * With EXT4_HUGE_FILE_FL set the i_blocks
2625                  * represent total number of blocks in
2626                  * file system block size
2627                  */
2628                 upper_limit = (1LL << 48) - 1;
2629
2630         }
2631
2632         /* indirect blocks */
2633         meta_blocks = 1;
2634         /* double indirect blocks */
2635         meta_blocks += 1 + (1LL << (bits-2));
2636         /* tripple indirect blocks */
2637         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2638
2639         upper_limit -= meta_blocks;
2640         upper_limit <<= bits;
2641
2642         res += 1LL << (bits-2);
2643         res += 1LL << (2*(bits-2));
2644         res += 1LL << (3*(bits-2));
2645         res <<= bits;
2646         if (res > upper_limit)
2647                 res = upper_limit;
2648
2649         if (res > MAX_LFS_FILESIZE)
2650                 res = MAX_LFS_FILESIZE;
2651
2652         return res;
2653 }
2654
2655 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2656                                    ext4_fsblk_t logical_sb_block, int nr)
2657 {
2658         struct ext4_sb_info *sbi = EXT4_SB(sb);
2659         ext4_group_t bg, first_meta_bg;
2660         int has_super = 0;
2661
2662         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2663
2664         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2665                 return logical_sb_block + nr + 1;
2666         bg = sbi->s_desc_per_block * nr;
2667         if (ext4_bg_has_super(sb, bg))
2668                 has_super = 1;
2669
2670         /*
2671          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2672          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2673          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2674          * compensate.
2675          */
2676         if (sb->s_blocksize == 1024 && nr == 0 &&
2677             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2678                 has_super++;
2679
2680         return (has_super + ext4_group_first_block_no(sb, bg));
2681 }
2682
2683 /**
2684  * ext4_get_stripe_size: Get the stripe size.
2685  * @sbi: In memory super block info
2686  *
2687  * If we have specified it via mount option, then
2688  * use the mount option value. If the value specified at mount time is
2689  * greater than the blocks per group use the super block value.
2690  * If the super block value is greater than blocks per group return 0.
2691  * Allocator needs it be less than blocks per group.
2692  *
2693  */
2694 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2695 {
2696         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2697         unsigned long stripe_width =
2698                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2699         int ret;
2700
2701         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2702                 ret = sbi->s_stripe;
2703         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2704                 ret = stripe_width;
2705         else if (stride && stride <= sbi->s_blocks_per_group)
2706                 ret = stride;
2707         else
2708                 ret = 0;
2709
2710         /*
2711          * If the stripe width is 1, this makes no sense and
2712          * we set it to 0 to turn off stripe handling code.
2713          */
2714         if (ret <= 1)
2715                 ret = 0;
2716
2717         return ret;
2718 }
2719
2720 /*
2721  * Check whether this filesystem can be mounted based on
2722  * the features present and the RDONLY/RDWR mount requested.
2723  * Returns 1 if this filesystem can be mounted as requested,
2724  * 0 if it cannot be.
2725  */
2726 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2727 {
2728         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2729                 ext4_msg(sb, KERN_ERR,
2730                         "Couldn't mount because of "
2731                         "unsupported optional features (%x)",
2732                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2733                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2734                 return 0;
2735         }
2736
2737         if (readonly)
2738                 return 1;
2739
2740         if (ext4_has_feature_readonly(sb)) {
2741                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2742                 sb->s_flags |= MS_RDONLY;
2743                 return 1;
2744         }
2745
2746         /* Check that feature set is OK for a read-write mount */
2747         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2748                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2749                          "unsupported optional features (%x)",
2750                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2751                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2752                 return 0;
2753         }
2754         /*
2755          * Large file size enabled file system can only be mounted
2756          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2757          */
2758         if (ext4_has_feature_huge_file(sb)) {
2759                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2760                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2761                                  "cannot be mounted RDWR without "
2762                                  "CONFIG_LBDAF");
2763                         return 0;
2764                 }
2765         }
2766         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2767                 ext4_msg(sb, KERN_ERR,
2768                          "Can't support bigalloc feature without "
2769                          "extents feature\n");
2770                 return 0;
2771         }
2772
2773 #ifndef CONFIG_QUOTA
2774         if (ext4_has_feature_quota(sb) && !readonly) {
2775                 ext4_msg(sb, KERN_ERR,
2776                          "Filesystem with quota feature cannot be mounted RDWR "
2777                          "without CONFIG_QUOTA");
2778                 return 0;
2779         }
2780         if (ext4_has_feature_project(sb) && !readonly) {
2781                 ext4_msg(sb, KERN_ERR,
2782                          "Filesystem with project quota feature cannot be mounted RDWR "
2783                          "without CONFIG_QUOTA");
2784                 return 0;
2785         }
2786 #endif  /* CONFIG_QUOTA */
2787         return 1;
2788 }
2789
2790 /*
2791  * This function is called once a day if we have errors logged
2792  * on the file system
2793  */
2794 static void print_daily_error_info(unsigned long arg)
2795 {
2796         struct super_block *sb = (struct super_block *) arg;
2797         struct ext4_sb_info *sbi;
2798         struct ext4_super_block *es;
2799
2800         sbi = EXT4_SB(sb);
2801         es = sbi->s_es;
2802
2803         if (es->s_error_count)
2804                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2805                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2806                          le32_to_cpu(es->s_error_count));
2807         if (es->s_first_error_time) {
2808                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2809                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2810                        (int) sizeof(es->s_first_error_func),
2811                        es->s_first_error_func,
2812                        le32_to_cpu(es->s_first_error_line));
2813                 if (es->s_first_error_ino)
2814                         printk(KERN_CONT ": inode %u",
2815                                le32_to_cpu(es->s_first_error_ino));
2816                 if (es->s_first_error_block)
2817                         printk(KERN_CONT ": block %llu", (unsigned long long)
2818                                le64_to_cpu(es->s_first_error_block));
2819                 printk(KERN_CONT "\n");
2820         }
2821         if (es->s_last_error_time) {
2822                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2823                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2824                        (int) sizeof(es->s_last_error_func),
2825                        es->s_last_error_func,
2826                        le32_to_cpu(es->s_last_error_line));
2827                 if (es->s_last_error_ino)
2828                         printk(KERN_CONT ": inode %u",
2829                                le32_to_cpu(es->s_last_error_ino));
2830                 if (es->s_last_error_block)
2831                         printk(KERN_CONT ": block %llu", (unsigned long long)
2832                                le64_to_cpu(es->s_last_error_block));
2833                 printk(KERN_CONT "\n");
2834         }
2835         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2836 }
2837
2838 /* Find next suitable group and run ext4_init_inode_table */
2839 static int ext4_run_li_request(struct ext4_li_request *elr)
2840 {
2841         struct ext4_group_desc *gdp = NULL;
2842         ext4_group_t group, ngroups;
2843         struct super_block *sb;
2844         unsigned long timeout = 0;
2845         int ret = 0;
2846
2847         sb = elr->lr_super;
2848         ngroups = EXT4_SB(sb)->s_groups_count;
2849
2850         for (group = elr->lr_next_group; group < ngroups; group++) {
2851                 gdp = ext4_get_group_desc(sb, group, NULL);
2852                 if (!gdp) {
2853                         ret = 1;
2854                         break;
2855                 }
2856
2857                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2858                         break;
2859         }
2860
2861         if (group >= ngroups)
2862                 ret = 1;
2863
2864         if (!ret) {
2865                 timeout = jiffies;
2866                 ret = ext4_init_inode_table(sb, group,
2867                                             elr->lr_timeout ? 0 : 1);
2868                 if (elr->lr_timeout == 0) {
2869                         timeout = (jiffies - timeout) *
2870                                   elr->lr_sbi->s_li_wait_mult;
2871                         elr->lr_timeout = timeout;
2872                 }
2873                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2874                 elr->lr_next_group = group + 1;
2875         }
2876         return ret;
2877 }
2878
2879 /*
2880  * Remove lr_request from the list_request and free the
2881  * request structure. Should be called with li_list_mtx held
2882  */
2883 static void ext4_remove_li_request(struct ext4_li_request *elr)
2884 {
2885         struct ext4_sb_info *sbi;
2886
2887         if (!elr)
2888                 return;
2889
2890         sbi = elr->lr_sbi;
2891
2892         list_del(&elr->lr_request);
2893         sbi->s_li_request = NULL;
2894         kfree(elr);
2895 }
2896
2897 static void ext4_unregister_li_request(struct super_block *sb)
2898 {
2899         mutex_lock(&ext4_li_mtx);
2900         if (!ext4_li_info) {
2901                 mutex_unlock(&ext4_li_mtx);
2902                 return;
2903         }
2904
2905         mutex_lock(&ext4_li_info->li_list_mtx);
2906         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2907         mutex_unlock(&ext4_li_info->li_list_mtx);
2908         mutex_unlock(&ext4_li_mtx);
2909 }
2910
2911 static struct task_struct *ext4_lazyinit_task;
2912
2913 /*
2914  * This is the function where ext4lazyinit thread lives. It walks
2915  * through the request list searching for next scheduled filesystem.
2916  * When such a fs is found, run the lazy initialization request
2917  * (ext4_rn_li_request) and keep track of the time spend in this
2918  * function. Based on that time we compute next schedule time of
2919  * the request. When walking through the list is complete, compute
2920  * next waking time and put itself into sleep.
2921  */
2922 static int ext4_lazyinit_thread(void *arg)
2923 {
2924         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2925         struct list_head *pos, *n;
2926         struct ext4_li_request *elr;
2927         unsigned long next_wakeup, cur;
2928
2929         BUG_ON(NULL == eli);
2930
2931 cont_thread:
2932         while (true) {
2933                 next_wakeup = MAX_JIFFY_OFFSET;
2934
2935                 mutex_lock(&eli->li_list_mtx);
2936                 if (list_empty(&eli->li_request_list)) {
2937                         mutex_unlock(&eli->li_list_mtx);
2938                         goto exit_thread;
2939                 }
2940                 list_for_each_safe(pos, n, &eli->li_request_list) {
2941                         int err = 0;
2942                         int progress = 0;
2943                         elr = list_entry(pos, struct ext4_li_request,
2944                                          lr_request);
2945
2946                         if (time_before(jiffies, elr->lr_next_sched)) {
2947                                 if (time_before(elr->lr_next_sched, next_wakeup))
2948                                         next_wakeup = elr->lr_next_sched;
2949                                 continue;
2950                         }
2951                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2952                                 if (sb_start_write_trylock(elr->lr_super)) {
2953                                         progress = 1;
2954                                         /*
2955                                          * We hold sb->s_umount, sb can not
2956                                          * be removed from the list, it is
2957                                          * now safe to drop li_list_mtx
2958                                          */
2959                                         mutex_unlock(&eli->li_list_mtx);
2960                                         err = ext4_run_li_request(elr);
2961                                         sb_end_write(elr->lr_super);
2962                                         mutex_lock(&eli->li_list_mtx);
2963                                         n = pos->next;
2964                                 }
2965                                 up_read((&elr->lr_super->s_umount));
2966                         }
2967                         /* error, remove the lazy_init job */
2968                         if (err) {
2969                                 ext4_remove_li_request(elr);
2970                                 continue;
2971                         }
2972                         if (!progress) {
2973                                 elr->lr_next_sched = jiffies +
2974                                         (prandom_u32()
2975                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2976                         }
2977                         if (time_before(elr->lr_next_sched, next_wakeup))
2978                                 next_wakeup = elr->lr_next_sched;
2979                 }
2980                 mutex_unlock(&eli->li_list_mtx);
2981
2982                 try_to_freeze();
2983
2984                 cur = jiffies;
2985                 if ((time_after_eq(cur, next_wakeup)) ||
2986                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2987                         cond_resched();
2988                         continue;
2989                 }
2990
2991                 schedule_timeout_interruptible(next_wakeup - cur);
2992
2993                 if (kthread_should_stop()) {
2994                         ext4_clear_request_list();
2995                         goto exit_thread;
2996                 }
2997         }
2998
2999 exit_thread:
3000         /*
3001          * It looks like the request list is empty, but we need
3002          * to check it under the li_list_mtx lock, to prevent any
3003          * additions into it, and of course we should lock ext4_li_mtx
3004          * to atomically free the list and ext4_li_info, because at
3005          * this point another ext4 filesystem could be registering
3006          * new one.
3007          */
3008         mutex_lock(&ext4_li_mtx);
3009         mutex_lock(&eli->li_list_mtx);
3010         if (!list_empty(&eli->li_request_list)) {
3011                 mutex_unlock(&eli->li_list_mtx);
3012                 mutex_unlock(&ext4_li_mtx);
3013                 goto cont_thread;
3014         }
3015         mutex_unlock(&eli->li_list_mtx);
3016         kfree(ext4_li_info);
3017         ext4_li_info = NULL;
3018         mutex_unlock(&ext4_li_mtx);
3019
3020         return 0;
3021 }
3022
3023 static void ext4_clear_request_list(void)
3024 {
3025         struct list_head *pos, *n;
3026         struct ext4_li_request *elr;
3027
3028         mutex_lock(&ext4_li_info->li_list_mtx);
3029         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3030                 elr = list_entry(pos, struct ext4_li_request,
3031                                  lr_request);
3032                 ext4_remove_li_request(elr);
3033         }
3034         mutex_unlock(&ext4_li_info->li_list_mtx);
3035 }
3036
3037 static int ext4_run_lazyinit_thread(void)
3038 {
3039         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3040                                          ext4_li_info, "ext4lazyinit");
3041         if (IS_ERR(ext4_lazyinit_task)) {
3042                 int err = PTR_ERR(ext4_lazyinit_task);
3043                 ext4_clear_request_list();
3044                 kfree(ext4_li_info);
3045                 ext4_li_info = NULL;
3046                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3047                                  "initialization thread\n",
3048                                  err);
3049                 return err;
3050         }
3051         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3052         return 0;
3053 }
3054
3055 /*
3056  * Check whether it make sense to run itable init. thread or not.
3057  * If there is at least one uninitialized inode table, return
3058  * corresponding group number, else the loop goes through all
3059  * groups and return total number of groups.
3060  */
3061 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3062 {
3063         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3064         struct ext4_group_desc *gdp = NULL;
3065
3066         for (group = 0; group < ngroups; group++) {
3067                 gdp = ext4_get_group_desc(sb, group, NULL);
3068                 if (!gdp)
3069                         continue;
3070
3071                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3072                         break;
3073         }
3074
3075         return group;
3076 }
3077
3078 static int ext4_li_info_new(void)
3079 {
3080         struct ext4_lazy_init *eli = NULL;
3081
3082         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3083         if (!eli)
3084                 return -ENOMEM;
3085
3086         INIT_LIST_HEAD(&eli->li_request_list);
3087         mutex_init(&eli->li_list_mtx);
3088
3089         eli->li_state |= EXT4_LAZYINIT_QUIT;
3090
3091         ext4_li_info = eli;
3092
3093         return 0;
3094 }
3095
3096 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3097                                             ext4_group_t start)
3098 {
3099         struct ext4_sb_info *sbi = EXT4_SB(sb);
3100         struct ext4_li_request *elr;
3101
3102         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3103         if (!elr)
3104                 return NULL;
3105
3106         elr->lr_super = sb;
3107         elr->lr_sbi = sbi;
3108         elr->lr_next_group = start;
3109
3110         /*
3111          * Randomize first schedule time of the request to
3112          * spread the inode table initialization requests
3113          * better.
3114          */
3115         elr->lr_next_sched = jiffies + (prandom_u32() %
3116                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3117         return elr;
3118 }
3119
3120 int ext4_register_li_request(struct super_block *sb,
3121                              ext4_group_t first_not_zeroed)
3122 {
3123         struct ext4_sb_info *sbi = EXT4_SB(sb);
3124         struct ext4_li_request *elr = NULL;
3125         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3126         int ret = 0;
3127
3128         mutex_lock(&ext4_li_mtx);
3129         if (sbi->s_li_request != NULL) {
3130                 /*
3131                  * Reset timeout so it can be computed again, because
3132                  * s_li_wait_mult might have changed.
3133                  */
3134                 sbi->s_li_request->lr_timeout = 0;
3135                 goto out;
3136         }
3137
3138         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3139             !test_opt(sb, INIT_INODE_TABLE))
3140                 goto out;
3141
3142         elr = ext4_li_request_new(sb, first_not_zeroed);
3143         if (!elr) {
3144                 ret = -ENOMEM;
3145                 goto out;
3146         }
3147
3148         if (NULL == ext4_li_info) {
3149                 ret = ext4_li_info_new();
3150                 if (ret)
3151                         goto out;
3152         }
3153
3154         mutex_lock(&ext4_li_info->li_list_mtx);
3155         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3156         mutex_unlock(&ext4_li_info->li_list_mtx);
3157
3158         sbi->s_li_request = elr;
3159         /*
3160          * set elr to NULL here since it has been inserted to
3161          * the request_list and the removal and free of it is
3162          * handled by ext4_clear_request_list from now on.
3163          */
3164         elr = NULL;
3165
3166         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3167                 ret = ext4_run_lazyinit_thread();
3168                 if (ret)
3169                         goto out;
3170         }
3171 out:
3172         mutex_unlock(&ext4_li_mtx);
3173         if (ret)
3174                 kfree(elr);
3175         return ret;
3176 }
3177
3178 /*
3179  * We do not need to lock anything since this is called on
3180  * module unload.
3181  */
3182 static void ext4_destroy_lazyinit_thread(void)
3183 {
3184         /*
3185          * If thread exited earlier
3186          * there's nothing to be done.
3187          */
3188         if (!ext4_li_info || !ext4_lazyinit_task)
3189                 return;
3190
3191         kthread_stop(ext4_lazyinit_task);
3192 }
3193
3194 static int set_journal_csum_feature_set(struct super_block *sb)
3195 {
3196         int ret = 1;
3197         int compat, incompat;
3198         struct ext4_sb_info *sbi = EXT4_SB(sb);
3199
3200         if (ext4_has_metadata_csum(sb)) {
3201                 /* journal checksum v3 */
3202                 compat = 0;
3203                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3204         } else {
3205                 /* journal checksum v1 */
3206                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3207                 incompat = 0;
3208         }
3209
3210         jbd2_journal_clear_features(sbi->s_journal,
3211                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3212                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3213                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3214         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3215                 ret = jbd2_journal_set_features(sbi->s_journal,
3216                                 compat, 0,
3217                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3218                                 incompat);
3219         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3220                 ret = jbd2_journal_set_features(sbi->s_journal,
3221                                 compat, 0,
3222                                 incompat);
3223                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3224                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3225         } else {
3226                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3227                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3228         }
3229
3230         return ret;
3231 }
3232
3233 /*
3234  * Note: calculating the overhead so we can be compatible with
3235  * historical BSD practice is quite difficult in the face of
3236  * clusters/bigalloc.  This is because multiple metadata blocks from
3237  * different block group can end up in the same allocation cluster.
3238  * Calculating the exact overhead in the face of clustered allocation
3239  * requires either O(all block bitmaps) in memory or O(number of block
3240  * groups**2) in time.  We will still calculate the superblock for
3241  * older file systems --- and if we come across with a bigalloc file
3242  * system with zero in s_overhead_clusters the estimate will be close to
3243  * correct especially for very large cluster sizes --- but for newer
3244  * file systems, it's better to calculate this figure once at mkfs
3245  * time, and store it in the superblock.  If the superblock value is
3246  * present (even for non-bigalloc file systems), we will use it.
3247  */
3248 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3249                           char *buf)
3250 {
3251         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3252         struct ext4_group_desc  *gdp;
3253         ext4_fsblk_t            first_block, last_block, b;
3254         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3255         int                     s, j, count = 0;
3256
3257         if (!ext4_has_feature_bigalloc(sb))
3258                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3259                         sbi->s_itb_per_group + 2);
3260
3261         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3262                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3263         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3264         for (i = 0; i < ngroups; i++) {
3265                 gdp = ext4_get_group_desc(sb, i, NULL);
3266                 b = ext4_block_bitmap(sb, gdp);
3267                 if (b >= first_block && b <= last_block) {
3268                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3269                         count++;
3270                 }
3271                 b = ext4_inode_bitmap(sb, gdp);
3272                 if (b >= first_block && b <= last_block) {
3273                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3274                         count++;
3275                 }
3276                 b = ext4_inode_table(sb, gdp);
3277                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3278                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3279                                 int c = EXT4_B2C(sbi, b - first_block);
3280                                 ext4_set_bit(c, buf);
3281                                 count++;
3282                         }
3283                 if (i != grp)
3284                         continue;
3285                 s = 0;
3286                 if (ext4_bg_has_super(sb, grp)) {
3287                         ext4_set_bit(s++, buf);
3288                         count++;
3289                 }
3290                 j = ext4_bg_num_gdb(sb, grp);
3291                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3292                         ext4_error(sb, "Invalid number of block group "
3293                                    "descriptor blocks: %d", j);
3294                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3295                 }
3296                 count += j;
3297                 for (; j > 0; j--)
3298                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3299         }
3300         if (!count)
3301                 return 0;
3302         return EXT4_CLUSTERS_PER_GROUP(sb) -
3303                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3304 }
3305
3306 /*
3307  * Compute the overhead and stash it in sbi->s_overhead
3308  */
3309 int ext4_calculate_overhead(struct super_block *sb)
3310 {
3311         struct ext4_sb_info *sbi = EXT4_SB(sb);
3312         struct ext4_super_block *es = sbi->s_es;
3313         struct inode *j_inode;
3314         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3315         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3316         ext4_fsblk_t overhead = 0;
3317         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3318
3319         if (!buf)
3320                 return -ENOMEM;
3321
3322         /*
3323          * Compute the overhead (FS structures).  This is constant
3324          * for a given filesystem unless the number of block groups
3325          * changes so we cache the previous value until it does.
3326          */
3327
3328         /*
3329          * All of the blocks before first_data_block are overhead
3330          */
3331         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3332
3333         /*
3334          * Add the overhead found in each block group
3335          */
3336         for (i = 0; i < ngroups; i++) {
3337                 int blks;
3338
3339                 blks = count_overhead(sb, i, buf);
3340                 overhead += blks;
3341                 if (blks)
3342                         memset(buf, 0, PAGE_SIZE);
3343                 cond_resched();
3344         }
3345
3346         /*
3347          * Add the internal journal blocks whether the journal has been
3348          * loaded or not
3349          */
3350         if (sbi->s_journal && !sbi->journal_bdev)
3351                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3352         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3353                 j_inode = ext4_get_journal_inode(sb, j_inum);
3354                 if (j_inode) {
3355                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3356                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3357                         iput(j_inode);
3358                 } else {
3359                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3360                 }
3361         }
3362         sbi->s_overhead = overhead;
3363         smp_wmb();
3364         free_page((unsigned long) buf);
3365         return 0;
3366 }
3367
3368 static void ext4_set_resv_clusters(struct super_block *sb)
3369 {
3370         ext4_fsblk_t resv_clusters;
3371         struct ext4_sb_info *sbi = EXT4_SB(sb);
3372
3373         /*
3374          * There's no need to reserve anything when we aren't using extents.
3375          * The space estimates are exact, there are no unwritten extents,
3376          * hole punching doesn't need new metadata... This is needed especially
3377          * to keep ext2/3 backward compatibility.
3378          */
3379         if (!ext4_has_feature_extents(sb))
3380                 return;
3381         /*
3382          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3383          * This should cover the situations where we can not afford to run
3384          * out of space like for example punch hole, or converting
3385          * unwritten extents in delalloc path. In most cases such
3386          * allocation would require 1, or 2 blocks, higher numbers are
3387          * very rare.
3388          */
3389         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3390                          sbi->s_cluster_bits);
3391
3392         do_div(resv_clusters, 50);
3393         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3394
3395         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3396 }
3397
3398 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3399 {
3400         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3401         char *orig_data = kstrdup(data, GFP_KERNEL);
3402         struct buffer_head *bh;
3403         struct ext4_super_block *es = NULL;
3404         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3405         ext4_fsblk_t block;
3406         ext4_fsblk_t sb_block = get_sb_block(&data);
3407         ext4_fsblk_t logical_sb_block;
3408         unsigned long offset = 0;
3409         unsigned long journal_devnum = 0;
3410         unsigned long def_mount_opts;
3411         struct inode *root;
3412         const char *descr;
3413         int ret = -ENOMEM;
3414         int blocksize, clustersize;
3415         unsigned int db_count;
3416         unsigned int i;
3417         int needs_recovery, has_huge_files, has_bigalloc;
3418         __u64 blocks_count;
3419         int err = 0;
3420         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3421         ext4_group_t first_not_zeroed;
3422
3423         if ((data && !orig_data) || !sbi)
3424                 goto out_free_base;
3425
3426         sbi->s_daxdev = dax_dev;
3427         sbi->s_blockgroup_lock =
3428                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3429         if (!sbi->s_blockgroup_lock)
3430                 goto out_free_base;
3431
3432         sb->s_fs_info = sbi;
3433         sbi->s_sb = sb;
3434         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3435         sbi->s_sb_block = sb_block;
3436         if (sb->s_bdev->bd_part)
3437                 sbi->s_sectors_written_start =
3438                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3439
3440         /* Cleanup superblock name */
3441         strreplace(sb->s_id, '/', '!');
3442
3443         /* -EINVAL is default */
3444         ret = -EINVAL;
3445         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3446         if (!blocksize) {
3447                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3448                 goto out_fail;
3449         }
3450
3451         /*
3452          * The ext4 superblock will not be buffer aligned for other than 1kB
3453          * block sizes.  We need to calculate the offset from buffer start.
3454          */
3455         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3456                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3457                 offset = do_div(logical_sb_block, blocksize);
3458         } else {
3459                 logical_sb_block = sb_block;
3460         }
3461
3462         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3463                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3464                 goto out_fail;
3465         }
3466         /*
3467          * Note: s_es must be initialized as soon as possible because
3468          *       some ext4 macro-instructions depend on its value
3469          */
3470         es = (struct ext4_super_block *) (bh->b_data + offset);
3471         sbi->s_es = es;
3472         sb->s_magic = le16_to_cpu(es->s_magic);
3473         if (sb->s_magic != EXT4_SUPER_MAGIC)
3474                 goto cantfind_ext4;
3475         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3476
3477         /* Warn if metadata_csum and gdt_csum are both set. */
3478         if (ext4_has_feature_metadata_csum(sb) &&
3479             ext4_has_feature_gdt_csum(sb))
3480                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3481                              "redundant flags; please run fsck.");
3482
3483         /* Check for a known checksum algorithm */
3484         if (!ext4_verify_csum_type(sb, es)) {
3485                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3486                          "unknown checksum algorithm.");
3487                 silent = 1;
3488                 goto cantfind_ext4;
3489         }
3490
3491         /* Load the checksum driver */
3492         if (ext4_has_feature_metadata_csum(sb) ||
3493             ext4_has_feature_ea_inode(sb)) {
3494                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3495                 if (IS_ERR(sbi->s_chksum_driver)) {
3496                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3497                         ret = PTR_ERR(sbi->s_chksum_driver);
3498                         sbi->s_chksum_driver = NULL;
3499                         goto failed_mount;
3500                 }
3501         }
3502
3503         /* Check superblock checksum */
3504         if (!ext4_superblock_csum_verify(sb, es)) {
3505                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3506                          "invalid superblock checksum.  Run e2fsck?");
3507                 silent = 1;
3508                 ret = -EFSBADCRC;
3509                 goto cantfind_ext4;
3510         }
3511
3512         /* Precompute checksum seed for all metadata */
3513         if (ext4_has_feature_csum_seed(sb))
3514                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3515         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3516                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3517                                                sizeof(es->s_uuid));
3518
3519         /* Set defaults before we parse the mount options */
3520         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3521         set_opt(sb, INIT_INODE_TABLE);
3522         if (def_mount_opts & EXT4_DEFM_DEBUG)
3523                 set_opt(sb, DEBUG);
3524         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3525                 set_opt(sb, GRPID);
3526         if (def_mount_opts & EXT4_DEFM_UID16)
3527                 set_opt(sb, NO_UID32);
3528         /* xattr user namespace & acls are now defaulted on */
3529         set_opt(sb, XATTR_USER);
3530 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3531         set_opt(sb, POSIX_ACL);
3532 #endif
3533         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3534         if (ext4_has_metadata_csum(sb))
3535                 set_opt(sb, JOURNAL_CHECKSUM);
3536
3537         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3538                 set_opt(sb, JOURNAL_DATA);
3539         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3540                 set_opt(sb, ORDERED_DATA);
3541         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3542                 set_opt(sb, WRITEBACK_DATA);
3543
3544         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3545                 set_opt(sb, ERRORS_PANIC);
3546         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3547                 set_opt(sb, ERRORS_CONT);
3548         else
3549                 set_opt(sb, ERRORS_RO);
3550         /* block_validity enabled by default; disable with noblock_validity */
3551         set_opt(sb, BLOCK_VALIDITY);
3552         if (def_mount_opts & EXT4_DEFM_DISCARD)
3553                 set_opt(sb, DISCARD);
3554
3555         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3556         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3557         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3558         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3559         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3560
3561         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3562                 set_opt(sb, BARRIER);
3563
3564         /*
3565          * enable delayed allocation by default
3566          * Use -o nodelalloc to turn it off
3567          */
3568         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3569             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3570                 set_opt(sb, DELALLOC);
3571
3572         /*
3573          * set default s_li_wait_mult for lazyinit, for the case there is
3574          * no mount option specified.
3575          */
3576         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3577
3578         if (sbi->s_es->s_mount_opts[0]) {
3579                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3580                                               sizeof(sbi->s_es->s_mount_opts),
3581                                               GFP_KERNEL);
3582                 if (!s_mount_opts)
3583                         goto failed_mount;
3584                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3585                                    &journal_ioprio, 0)) {
3586                         ext4_msg(sb, KERN_WARNING,
3587                                  "failed to parse options in superblock: %s",
3588                                  s_mount_opts);
3589                 }
3590                 kfree(s_mount_opts);
3591         }
3592         sbi->s_def_mount_opt = sbi->s_mount_opt;
3593         if (!parse_options((char *) data, sb, &journal_devnum,
3594                            &journal_ioprio, 0))
3595                 goto failed_mount;
3596
3597         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3598                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3599                             "with data=journal disables delayed "
3600                             "allocation and O_DIRECT support!\n");
3601                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3602                         ext4_msg(sb, KERN_ERR, "can't mount with "
3603                                  "both data=journal and delalloc");
3604                         goto failed_mount;
3605                 }
3606                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3607                         ext4_msg(sb, KERN_ERR, "can't mount with "
3608                                  "both data=journal and dioread_nolock");
3609                         goto failed_mount;
3610                 }
3611                 if (test_opt(sb, DAX)) {
3612                         ext4_msg(sb, KERN_ERR, "can't mount with "
3613                                  "both data=journal and dax");
3614                         goto failed_mount;
3615                 }
3616                 if (ext4_has_feature_encrypt(sb)) {
3617                         ext4_msg(sb, KERN_WARNING,
3618                                  "encrypted files will use data=ordered "
3619                                  "instead of data journaling mode");
3620                 }
3621                 if (test_opt(sb, DELALLOC))
3622                         clear_opt(sb, DELALLOC);
3623         } else {
3624                 sb->s_iflags |= SB_I_CGROUPWB;
3625         }
3626
3627         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3628                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3629
3630         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3631             (ext4_has_compat_features(sb) ||
3632              ext4_has_ro_compat_features(sb) ||
3633              ext4_has_incompat_features(sb)))
3634                 ext4_msg(sb, KERN_WARNING,
3635                        "feature flags set on rev 0 fs, "
3636                        "running e2fsck is recommended");
3637
3638         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3639                 set_opt2(sb, HURD_COMPAT);
3640                 if (ext4_has_feature_64bit(sb)) {
3641                         ext4_msg(sb, KERN_ERR,
3642                                  "The Hurd can't support 64-bit file systems");
3643                         goto failed_mount;
3644                 }
3645
3646                 /*
3647                  * ea_inode feature uses l_i_version field which is not
3648                  * available in HURD_COMPAT mode.
3649                  */
3650                 if (ext4_has_feature_ea_inode(sb)) {
3651                         ext4_msg(sb, KERN_ERR,
3652                                  "ea_inode feature is not supported for Hurd");
3653                         goto failed_mount;
3654                 }
3655         }
3656
3657         if (IS_EXT2_SB(sb)) {
3658                 if (ext2_feature_set_ok(sb))
3659                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3660                                  "using the ext4 subsystem");
3661                 else {
3662                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3663                                  "to feature incompatibilities");
3664                         goto failed_mount;
3665                 }
3666         }
3667
3668         if (IS_EXT3_SB(sb)) {
3669                 if (ext3_feature_set_ok(sb))
3670                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3671                                  "using the ext4 subsystem");
3672                 else {
3673                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3674                                  "to feature incompatibilities");
3675                         goto failed_mount;
3676                 }
3677         }
3678
3679         /*
3680          * Check feature flags regardless of the revision level, since we
3681          * previously didn't change the revision level when setting the flags,
3682          * so there is a chance incompat flags are set on a rev 0 filesystem.
3683          */
3684         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3685                 goto failed_mount;
3686
3687         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3688         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3689             blocksize > EXT4_MAX_BLOCK_SIZE) {
3690                 ext4_msg(sb, KERN_ERR,
3691                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3692                          blocksize, le32_to_cpu(es->s_log_block_size));
3693                 goto failed_mount;
3694         }
3695         if (le32_to_cpu(es->s_log_block_size) >
3696             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3697                 ext4_msg(sb, KERN_ERR,
3698                          "Invalid log block size: %u",
3699                          le32_to_cpu(es->s_log_block_size));
3700                 goto failed_mount;
3701         }
3702
3703         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3704                 ext4_msg(sb, KERN_ERR,
3705                          "Number of reserved GDT blocks insanely large: %d",
3706                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3707                 goto failed_mount;
3708         }
3709
3710         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3711                 err = bdev_dax_supported(sb, blocksize);
3712                 if (err)
3713                         goto failed_mount;
3714         }
3715
3716         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3717                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3718                          es->s_encryption_level);
3719                 goto failed_mount;
3720         }
3721
3722         if (sb->s_blocksize != blocksize) {
3723                 /* Validate the filesystem blocksize */
3724                 if (!sb_set_blocksize(sb, blocksize)) {
3725                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3726                                         blocksize);
3727                         goto failed_mount;
3728                 }
3729
3730                 brelse(bh);
3731                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3732                 offset = do_div(logical_sb_block, blocksize);
3733                 bh = sb_bread_unmovable(sb, logical_sb_block);
3734                 if (!bh) {
3735                         ext4_msg(sb, KERN_ERR,
3736                                "Can't read superblock on 2nd try");
3737                         goto failed_mount;
3738                 }
3739                 es = (struct ext4_super_block *)(bh->b_data + offset);
3740                 sbi->s_es = es;
3741                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                "Magic mismatch, very weird!");
3744                         goto failed_mount;
3745                 }
3746         }
3747
3748         has_huge_files = ext4_has_feature_huge_file(sb);
3749         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3750                                                       has_huge_files);
3751         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3752
3753         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3754                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3755                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3756         } else {
3757                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3758                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3759                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3760                     (!is_power_of_2(sbi->s_inode_size)) ||
3761                     (sbi->s_inode_size > blocksize)) {
3762                         ext4_msg(sb, KERN_ERR,
3763                                "unsupported inode size: %d",
3764                                sbi->s_inode_size);
3765                         goto failed_mount;
3766                 }
3767                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3768                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3769         }
3770
3771         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3772         if (ext4_has_feature_64bit(sb)) {
3773                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3774                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3775                     !is_power_of_2(sbi->s_desc_size)) {
3776                         ext4_msg(sb, KERN_ERR,
3777                                "unsupported descriptor size %lu",
3778                                sbi->s_desc_size);
3779                         goto failed_mount;
3780                 }
3781         } else
3782                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3783
3784         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3785         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3786
3787         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3788         if (sbi->s_inodes_per_block == 0)
3789                 goto cantfind_ext4;
3790         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3791             sbi->s_inodes_per_group > blocksize * 8) {
3792                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3793                          sbi->s_blocks_per_group);
3794                 goto failed_mount;
3795         }
3796         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3797                                         sbi->s_inodes_per_block;
3798         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3799         sbi->s_sbh = bh;
3800         sbi->s_mount_state = le16_to_cpu(es->s_state);
3801         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3802         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3803
3804         for (i = 0; i < 4; i++)
3805                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3806         sbi->s_def_hash_version = es->s_def_hash_version;
3807         if (ext4_has_feature_dir_index(sb)) {
3808                 i = le32_to_cpu(es->s_flags);
3809                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3810                         sbi->s_hash_unsigned = 3;
3811                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3812 #ifdef __CHAR_UNSIGNED__
3813                         if (!sb_rdonly(sb))
3814                                 es->s_flags |=
3815                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3816                         sbi->s_hash_unsigned = 3;
3817 #else
3818                         if (!sb_rdonly(sb))
3819                                 es->s_flags |=
3820                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3821 #endif
3822                 }
3823         }
3824
3825         /* Handle clustersize */
3826         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3827         has_bigalloc = ext4_has_feature_bigalloc(sb);
3828         if (has_bigalloc) {
3829                 if (clustersize < blocksize) {
3830                         ext4_msg(sb, KERN_ERR,
3831                                  "cluster size (%d) smaller than "
3832                                  "block size (%d)", clustersize, blocksize);
3833                         goto failed_mount;
3834                 }
3835                 if (le32_to_cpu(es->s_log_cluster_size) >
3836                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3837                         ext4_msg(sb, KERN_ERR,
3838                                  "Invalid log cluster size: %u",
3839                                  le32_to_cpu(es->s_log_cluster_size));
3840                         goto failed_mount;
3841                 }
3842                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3843                         le32_to_cpu(es->s_log_block_size);
3844                 sbi->s_clusters_per_group =
3845                         le32_to_cpu(es->s_clusters_per_group);
3846                 if (sbi->s_clusters_per_group > blocksize * 8) {
3847                         ext4_msg(sb, KERN_ERR,
3848                                  "#clusters per group too big: %lu",
3849                                  sbi->s_clusters_per_group);
3850                         goto failed_mount;
3851                 }
3852                 if (sbi->s_blocks_per_group !=
3853                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3854                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3855                                  "clusters per group (%lu) inconsistent",
3856                                  sbi->s_blocks_per_group,
3857                                  sbi->s_clusters_per_group);
3858                         goto failed_mount;
3859                 }
3860         } else {
3861                 if (clustersize != blocksize) {
3862                         ext4_warning(sb, "fragment/cluster size (%d) != "
3863                                      "block size (%d)", clustersize,
3864                                      blocksize);
3865                         clustersize = blocksize;
3866                 }
3867                 if (sbi->s_blocks_per_group > blocksize * 8) {
3868                         ext4_msg(sb, KERN_ERR,
3869                                  "#blocks per group too big: %lu",
3870                                  sbi->s_blocks_per_group);
3871                         goto failed_mount;
3872                 }
3873                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3874                 sbi->s_cluster_bits = 0;
3875         }
3876         sbi->s_cluster_ratio = clustersize / blocksize;
3877
3878         /* Do we have standard group size of clustersize * 8 blocks ? */
3879         if (sbi->s_blocks_per_group == clustersize << 3)
3880                 set_opt2(sb, STD_GROUP_SIZE);
3881
3882         /*
3883          * Test whether we have more sectors than will fit in sector_t,
3884          * and whether the max offset is addressable by the page cache.
3885          */
3886         err = generic_check_addressable(sb->s_blocksize_bits,
3887                                         ext4_blocks_count(es));
3888         if (err) {
3889                 ext4_msg(sb, KERN_ERR, "filesystem"
3890                          " too large to mount safely on this system");
3891                 if (sizeof(sector_t) < 8)
3892                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3893                 goto failed_mount;
3894         }
3895
3896         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3897                 goto cantfind_ext4;
3898
3899         /* check blocks count against device size */
3900         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3901         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3902                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3903                        "exceeds size of device (%llu blocks)",
3904                        ext4_blocks_count(es), blocks_count);
3905                 goto failed_mount;
3906         }
3907
3908         /*
3909          * It makes no sense for the first data block to be beyond the end
3910          * of the filesystem.
3911          */
3912         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3913                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3914                          "block %u is beyond end of filesystem (%llu)",
3915                          le32_to_cpu(es->s_first_data_block),
3916                          ext4_blocks_count(es));
3917                 goto failed_mount;
3918         }
3919         blocks_count = (ext4_blocks_count(es) -
3920                         le32_to_cpu(es->s_first_data_block) +
3921                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3922         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3923         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3924                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3925                        "(block count %llu, first data block %u, "
3926                        "blocks per group %lu)", sbi->s_groups_count,
3927                        ext4_blocks_count(es),
3928                        le32_to_cpu(es->s_first_data_block),
3929                        EXT4_BLOCKS_PER_GROUP(sb));
3930                 goto failed_mount;
3931         }
3932         sbi->s_groups_count = blocks_count;
3933         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3934                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3935         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3936                    EXT4_DESC_PER_BLOCK(sb);
3937         if (ext4_has_feature_meta_bg(sb)) {
3938                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3939                         ext4_msg(sb, KERN_WARNING,
3940                                  "first meta block group too large: %u "
3941                                  "(group descriptor block count %u)",
3942                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3943                         goto failed_mount;
3944                 }
3945         }
3946         sbi->s_group_desc = kvmalloc(db_count *
3947                                           sizeof(struct buffer_head *),
3948                                           GFP_KERNEL);
3949         if (sbi->s_group_desc == NULL) {
3950                 ext4_msg(sb, KERN_ERR, "not enough memory");
3951                 ret = -ENOMEM;
3952                 goto failed_mount;
3953         }
3954
3955         bgl_lock_init(sbi->s_blockgroup_lock);
3956
3957         /* Pre-read the descriptors into the buffer cache */
3958         for (i = 0; i < db_count; i++) {
3959                 block = descriptor_loc(sb, logical_sb_block, i);
3960                 sb_breadahead(sb, block);
3961         }
3962
3963         for (i = 0; i < db_count; i++) {
3964                 block = descriptor_loc(sb, logical_sb_block, i);
3965                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3966                 if (!sbi->s_group_desc[i]) {
3967                         ext4_msg(sb, KERN_ERR,
3968                                "can't read group descriptor %d", i);
3969                         db_count = i;
3970                         goto failed_mount2;
3971                 }
3972         }
3973         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3974                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3975                 ret = -EFSCORRUPTED;
3976                 goto failed_mount2;
3977         }
3978
3979         sbi->s_gdb_count = db_count;
3980         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3981         spin_lock_init(&sbi->s_next_gen_lock);
3982
3983         setup_timer(&sbi->s_err_report, print_daily_error_info,
3984                 (unsigned long) sb);
3985
3986         /* Register extent status tree shrinker */
3987         if (ext4_es_register_shrinker(sbi))
3988                 goto failed_mount3;
3989
3990         sbi->s_stripe = ext4_get_stripe_size(sbi);
3991         sbi->s_extent_max_zeroout_kb = 32;
3992
3993         /*
3994          * set up enough so that it can read an inode
3995          */
3996         sb->s_op = &ext4_sops;
3997         sb->s_export_op = &ext4_export_ops;
3998         sb->s_xattr = ext4_xattr_handlers;
3999         sb->s_cop = &ext4_cryptops;
4000 #ifdef CONFIG_QUOTA
4001         sb->dq_op = &ext4_quota_operations;
4002         if (ext4_has_feature_quota(sb))
4003                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4004         else
4005                 sb->s_qcop = &ext4_qctl_operations;
4006         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4007 #endif
4008         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4009
4010         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4011         mutex_init(&sbi->s_orphan_lock);
4012
4013         sb->s_root = NULL;
4014
4015         needs_recovery = (es->s_last_orphan != 0 ||
4016                           ext4_has_feature_journal_needs_recovery(sb));
4017
4018         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4019                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4020                         goto failed_mount3a;
4021
4022         /*
4023          * The first inode we look at is the journal inode.  Don't try
4024          * root first: it may be modified in the journal!
4025          */
4026         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4027                 err = ext4_load_journal(sb, es, journal_devnum);
4028                 if (err)
4029                         goto failed_mount3a;
4030         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4031                    ext4_has_feature_journal_needs_recovery(sb)) {
4032                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4033                        "suppressed and not mounted read-only");
4034                 goto failed_mount_wq;
4035         } else {
4036                 /* Nojournal mode, all journal mount options are illegal */
4037                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4038                         ext4_msg(sb, KERN_ERR, "can't mount with "
4039                                  "journal_checksum, fs mounted w/o journal");
4040                         goto failed_mount_wq;
4041                 }
4042                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4043                         ext4_msg(sb, KERN_ERR, "can't mount with "
4044                                  "journal_async_commit, fs mounted w/o journal");
4045                         goto failed_mount_wq;
4046                 }
4047                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4048                         ext4_msg(sb, KERN_ERR, "can't mount with "
4049                                  "commit=%lu, fs mounted w/o journal",
4050                                  sbi->s_commit_interval / HZ);
4051                         goto failed_mount_wq;
4052                 }
4053                 if (EXT4_MOUNT_DATA_FLAGS &
4054                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4055                         ext4_msg(sb, KERN_ERR, "can't mount with "
4056                                  "data=, fs mounted w/o journal");
4057                         goto failed_mount_wq;
4058                 }
4059                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4060                 clear_opt(sb, JOURNAL_CHECKSUM);
4061                 clear_opt(sb, DATA_FLAGS);
4062                 sbi->s_journal = NULL;
4063                 needs_recovery = 0;
4064                 goto no_journal;
4065         }
4066
4067         if (ext4_has_feature_64bit(sb) &&
4068             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4069                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4070                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4071                 goto failed_mount_wq;
4072         }
4073
4074         if (!set_journal_csum_feature_set(sb)) {
4075                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4076                          "feature set");
4077                 goto failed_mount_wq;
4078         }
4079
4080         /* We have now updated the journal if required, so we can
4081          * validate the data journaling mode. */
4082         switch (test_opt(sb, DATA_FLAGS)) {
4083         case 0:
4084                 /* No mode set, assume a default based on the journal
4085                  * capabilities: ORDERED_DATA if the journal can
4086                  * cope, else JOURNAL_DATA
4087                  */
4088                 if (jbd2_journal_check_available_features
4089                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4090                         set_opt(sb, ORDERED_DATA);
4091                 else
4092                         set_opt(sb, JOURNAL_DATA);
4093                 break;
4094
4095         case EXT4_MOUNT_ORDERED_DATA:
4096         case EXT4_MOUNT_WRITEBACK_DATA:
4097                 if (!jbd2_journal_check_available_features
4098                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4099                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4100                                "requested data journaling mode");
4101                         goto failed_mount_wq;
4102                 }
4103         default:
4104                 break;
4105         }
4106
4107         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4108             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4109                 ext4_msg(sb, KERN_ERR, "can't mount with "
4110                         "journal_async_commit in data=ordered mode");
4111                 goto failed_mount_wq;
4112         }
4113
4114         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4115
4116         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4117
4118 no_journal:
4119         if (!test_opt(sb, NO_MBCACHE)) {
4120                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4121                 if (!sbi->s_ea_block_cache) {
4122                         ext4_msg(sb, KERN_ERR,
4123                                  "Failed to create ea_block_cache");
4124                         goto failed_mount_wq;
4125                 }
4126
4127                 if (ext4_has_feature_ea_inode(sb)) {
4128                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4129                         if (!sbi->s_ea_inode_cache) {
4130                                 ext4_msg(sb, KERN_ERR,
4131                                          "Failed to create ea_inode_cache");
4132                                 goto failed_mount_wq;
4133                         }
4134                 }
4135         }
4136
4137         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4138             (blocksize != PAGE_SIZE)) {
4139                 ext4_msg(sb, KERN_ERR,
4140                          "Unsupported blocksize for fs encryption");
4141                 goto failed_mount_wq;
4142         }
4143
4144         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4145             !ext4_has_feature_encrypt(sb)) {
4146                 ext4_set_feature_encrypt(sb);
4147                 ext4_commit_super(sb, 1);
4148         }
4149
4150         /*
4151          * Get the # of file system overhead blocks from the
4152          * superblock if present.
4153          */
4154         if (es->s_overhead_clusters)
4155                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4156         else {
4157                 err = ext4_calculate_overhead(sb);
4158                 if (err)
4159                         goto failed_mount_wq;
4160         }
4161
4162         /*
4163          * The maximum number of concurrent works can be high and
4164          * concurrency isn't really necessary.  Limit it to 1.
4165          */
4166         EXT4_SB(sb)->rsv_conversion_wq =
4167                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4168         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4169                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4170                 ret = -ENOMEM;
4171                 goto failed_mount4;
4172         }
4173
4174         /*
4175          * The jbd2_journal_load will have done any necessary log recovery,
4176          * so we can safely mount the rest of the filesystem now.
4177          */
4178
4179         root = ext4_iget(sb, EXT4_ROOT_INO);
4180         if (IS_ERR(root)) {
4181                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4182                 ret = PTR_ERR(root);
4183                 root = NULL;
4184                 goto failed_mount4;
4185         }
4186         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4187                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4188                 iput(root);
4189                 goto failed_mount4;
4190         }
4191         sb->s_root = d_make_root(root);
4192         if (!sb->s_root) {
4193                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4194                 ret = -ENOMEM;
4195                 goto failed_mount4;
4196         }
4197
4198         if (ext4_setup_super(sb, es, sb_rdonly(sb)))
4199                 sb->s_flags |= MS_RDONLY;
4200
4201         /* determine the minimum size of new large inodes, if present */
4202         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4203             sbi->s_want_extra_isize == 0) {
4204                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4205                                                      EXT4_GOOD_OLD_INODE_SIZE;
4206                 if (ext4_has_feature_extra_isize(sb)) {
4207                         if (sbi->s_want_extra_isize <
4208                             le16_to_cpu(es->s_want_extra_isize))
4209                                 sbi->s_want_extra_isize =
4210                                         le16_to_cpu(es->s_want_extra_isize);
4211                         if (sbi->s_want_extra_isize <
4212                             le16_to_cpu(es->s_min_extra_isize))
4213                                 sbi->s_want_extra_isize =
4214                                         le16_to_cpu(es->s_min_extra_isize);
4215                 }
4216         }
4217         /* Check if enough inode space is available */
4218         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4219                                                         sbi->s_inode_size) {
4220                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4221                                                        EXT4_GOOD_OLD_INODE_SIZE;
4222                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4223                          "available");
4224         }
4225
4226         ext4_set_resv_clusters(sb);
4227
4228         err = ext4_setup_system_zone(sb);
4229         if (err) {
4230                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4231                          "zone (%d)", err);
4232                 goto failed_mount4a;
4233         }
4234
4235         ext4_ext_init(sb);
4236         err = ext4_mb_init(sb);
4237         if (err) {
4238                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4239                          err);
4240                 goto failed_mount5;
4241         }
4242
4243         block = ext4_count_free_clusters(sb);
4244         ext4_free_blocks_count_set(sbi->s_es,
4245                                    EXT4_C2B(sbi, block));
4246         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4247                                   GFP_KERNEL);
4248         if (!err) {
4249                 unsigned long freei = ext4_count_free_inodes(sb);
4250                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4251                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4252                                           GFP_KERNEL);
4253         }
4254         if (!err)
4255                 err = percpu_counter_init(&sbi->s_dirs_counter,
4256                                           ext4_count_dirs(sb), GFP_KERNEL);
4257         if (!err)
4258                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4259                                           GFP_KERNEL);
4260         if (!err)
4261                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4262
4263         if (err) {
4264                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4265                 goto failed_mount6;
4266         }
4267
4268         if (ext4_has_feature_flex_bg(sb))
4269                 if (!ext4_fill_flex_info(sb)) {
4270                         ext4_msg(sb, KERN_ERR,
4271                                "unable to initialize "
4272                                "flex_bg meta info!");
4273                         goto failed_mount6;
4274                 }
4275
4276         err = ext4_register_li_request(sb, first_not_zeroed);
4277         if (err)
4278                 goto failed_mount6;
4279
4280         err = ext4_register_sysfs(sb);
4281         if (err)
4282                 goto failed_mount7;
4283
4284 #ifdef CONFIG_QUOTA
4285         /* Enable quota usage during mount. */
4286         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4287                 err = ext4_enable_quotas(sb);
4288                 if (err)
4289                         goto failed_mount8;
4290         }
4291 #endif  /* CONFIG_QUOTA */
4292
4293         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4294         ext4_orphan_cleanup(sb, es);
4295         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4296         if (needs_recovery) {
4297                 ext4_msg(sb, KERN_INFO, "recovery complete");
4298                 ext4_mark_recovery_complete(sb, es);
4299         }
4300         if (EXT4_SB(sb)->s_journal) {
4301                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4302                         descr = " journalled data mode";
4303                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4304                         descr = " ordered data mode";
4305                 else
4306                         descr = " writeback data mode";
4307         } else
4308                 descr = "out journal";
4309
4310         if (test_opt(sb, DISCARD)) {
4311                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4312                 if (!blk_queue_discard(q))
4313                         ext4_msg(sb, KERN_WARNING,
4314                                  "mounting with \"discard\" option, but "
4315                                  "the device does not support discard");
4316         }
4317
4318         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4319                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4320                          "Opts: %.*s%s%s", descr,
4321                          (int) sizeof(sbi->s_es->s_mount_opts),
4322                          sbi->s_es->s_mount_opts,
4323                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4324
4325         if (es->s_error_count)
4326                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4327
4328         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4329         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4330         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4331         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4332
4333         kfree(orig_data);
4334         return 0;
4335
4336 cantfind_ext4:
4337         if (!silent)
4338                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4339         goto failed_mount;
4340
4341 #ifdef CONFIG_QUOTA
4342 failed_mount8:
4343         ext4_unregister_sysfs(sb);
4344 #endif
4345 failed_mount7:
4346         ext4_unregister_li_request(sb);
4347 failed_mount6:
4348         ext4_mb_release(sb);
4349         if (sbi->s_flex_groups)
4350                 kvfree(sbi->s_flex_groups);
4351         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4352         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4353         percpu_counter_destroy(&sbi->s_dirs_counter);
4354         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4355 failed_mount5:
4356         ext4_ext_release(sb);
4357         ext4_release_system_zone(sb);
4358 failed_mount4a:
4359         dput(sb->s_root);
4360         sb->s_root = NULL;
4361 failed_mount4:
4362         ext4_msg(sb, KERN_ERR, "mount failed");
4363         if (EXT4_SB(sb)->rsv_conversion_wq)
4364                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4365 failed_mount_wq:
4366         if (sbi->s_ea_inode_cache) {
4367                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4368                 sbi->s_ea_inode_cache = NULL;
4369         }
4370         if (sbi->s_ea_block_cache) {
4371                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4372                 sbi->s_ea_block_cache = NULL;
4373         }
4374         if (sbi->s_journal) {
4375                 jbd2_journal_destroy(sbi->s_journal);
4376                 sbi->s_journal = NULL;
4377         }
4378 failed_mount3a:
4379         ext4_es_unregister_shrinker(sbi);
4380 failed_mount3:
4381         del_timer_sync(&sbi->s_err_report);
4382         if (sbi->s_mmp_tsk)
4383                 kthread_stop(sbi->s_mmp_tsk);
4384 failed_mount2:
4385         for (i = 0; i < db_count; i++)
4386                 brelse(sbi->s_group_desc[i]);
4387         kvfree(sbi->s_group_desc);
4388 failed_mount:
4389         if (sbi->s_chksum_driver)
4390                 crypto_free_shash(sbi->s_chksum_driver);
4391 #ifdef CONFIG_QUOTA
4392         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4393                 kfree(sbi->s_qf_names[i]);
4394 #endif
4395         ext4_blkdev_remove(sbi);
4396         brelse(bh);
4397 out_fail:
4398         sb->s_fs_info = NULL;
4399         kfree(sbi->s_blockgroup_lock);
4400 out_free_base:
4401         kfree(sbi);
4402         kfree(orig_data);
4403         fs_put_dax(dax_dev);
4404         return err ? err : ret;
4405 }
4406
4407 /*
4408  * Setup any per-fs journal parameters now.  We'll do this both on
4409  * initial mount, once the journal has been initialised but before we've
4410  * done any recovery; and again on any subsequent remount.
4411  */
4412 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4413 {
4414         struct ext4_sb_info *sbi = EXT4_SB(sb);
4415
4416         journal->j_commit_interval = sbi->s_commit_interval;
4417         journal->j_min_batch_time = sbi->s_min_batch_time;
4418         journal->j_max_batch_time = sbi->s_max_batch_time;
4419
4420         write_lock(&journal->j_state_lock);
4421         if (test_opt(sb, BARRIER))
4422                 journal->j_flags |= JBD2_BARRIER;
4423         else
4424                 journal->j_flags &= ~JBD2_BARRIER;
4425         if (test_opt(sb, DATA_ERR_ABORT))
4426                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4427         else
4428                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4429         write_unlock(&journal->j_state_lock);
4430 }
4431
4432 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4433                                              unsigned int journal_inum)
4434 {
4435         struct inode *journal_inode;
4436
4437         /*
4438          * Test for the existence of a valid inode on disk.  Bad things
4439          * happen if we iget() an unused inode, as the subsequent iput()
4440          * will try to delete it.
4441          */
4442         journal_inode = ext4_iget(sb, journal_inum);
4443         if (IS_ERR(journal_inode)) {
4444                 ext4_msg(sb, KERN_ERR, "no journal found");
4445                 return NULL;
4446         }
4447         if (!journal_inode->i_nlink) {
4448                 make_bad_inode(journal_inode);
4449                 iput(journal_inode);
4450                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4451                 return NULL;
4452         }
4453
4454         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4455                   journal_inode, journal_inode->i_size);
4456         if (!S_ISREG(journal_inode->i_mode)) {
4457                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4458                 iput(journal_inode);
4459                 return NULL;
4460         }
4461         return journal_inode;
4462 }
4463
4464 static journal_t *ext4_get_journal(struct super_block *sb,
4465                                    unsigned int journal_inum)
4466 {
4467         struct inode *journal_inode;
4468         journal_t *journal;
4469
4470         BUG_ON(!ext4_has_feature_journal(sb));
4471
4472         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4473         if (!journal_inode)
4474                 return NULL;
4475
4476         journal = jbd2_journal_init_inode(journal_inode);
4477         if (!journal) {
4478                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4479                 iput(journal_inode);
4480                 return NULL;
4481         }
4482         journal->j_private = sb;
4483         ext4_init_journal_params(sb, journal);
4484         return journal;
4485 }
4486
4487 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4488                                        dev_t j_dev)
4489 {
4490         struct buffer_head *bh;
4491         journal_t *journal;
4492         ext4_fsblk_t start;
4493         ext4_fsblk_t len;
4494         int hblock, blocksize;
4495         ext4_fsblk_t sb_block;
4496         unsigned long offset;
4497         struct ext4_super_block *es;
4498         struct block_device *bdev;
4499
4500         BUG_ON(!ext4_has_feature_journal(sb));
4501
4502         bdev = ext4_blkdev_get(j_dev, sb);
4503         if (bdev == NULL)
4504                 return NULL;
4505
4506         blocksize = sb->s_blocksize;
4507         hblock = bdev_logical_block_size(bdev);
4508         if (blocksize < hblock) {
4509                 ext4_msg(sb, KERN_ERR,
4510                         "blocksize too small for journal device");
4511                 goto out_bdev;
4512         }
4513
4514         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4515         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4516         set_blocksize(bdev, blocksize);
4517         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4518                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4519                        "external journal");
4520                 goto out_bdev;
4521         }
4522
4523         es = (struct ext4_super_block *) (bh->b_data + offset);
4524         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4525             !(le32_to_cpu(es->s_feature_incompat) &
4526               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4527                 ext4_msg(sb, KERN_ERR, "external journal has "
4528                                         "bad superblock");
4529                 brelse(bh);
4530                 goto out_bdev;
4531         }
4532
4533         if ((le32_to_cpu(es->s_feature_ro_compat) &
4534              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4535             es->s_checksum != ext4_superblock_csum(sb, es)) {
4536                 ext4_msg(sb, KERN_ERR, "external journal has "
4537                                        "corrupt superblock");
4538                 brelse(bh);
4539                 goto out_bdev;
4540         }
4541
4542         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4543                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4544                 brelse(bh);
4545                 goto out_bdev;
4546         }
4547
4548         len = ext4_blocks_count(es);
4549         start = sb_block + 1;
4550         brelse(bh);     /* we're done with the superblock */
4551
4552         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4553                                         start, len, blocksize);
4554         if (!journal) {
4555                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4556                 goto out_bdev;
4557         }
4558         journal->j_private = sb;
4559         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4560         wait_on_buffer(journal->j_sb_buffer);
4561         if (!buffer_uptodate(journal->j_sb_buffer)) {
4562                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4563                 goto out_journal;
4564         }
4565         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4566                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4567                                         "user (unsupported) - %d",
4568                         be32_to_cpu(journal->j_superblock->s_nr_users));
4569                 goto out_journal;
4570         }
4571         EXT4_SB(sb)->journal_bdev = bdev;
4572         ext4_init_journal_params(sb, journal);
4573         return journal;
4574
4575 out_journal:
4576         jbd2_journal_destroy(journal);
4577 out_bdev:
4578         ext4_blkdev_put(bdev);
4579         return NULL;
4580 }
4581
4582 static int ext4_load_journal(struct super_block *sb,
4583                              struct ext4_super_block *es,
4584                              unsigned long journal_devnum)
4585 {
4586         journal_t *journal;
4587         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4588         dev_t journal_dev;
4589         int err = 0;
4590         int really_read_only;
4591
4592         BUG_ON(!ext4_has_feature_journal(sb));
4593
4594         if (journal_devnum &&
4595             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4596                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4597                         "numbers have changed");
4598                 journal_dev = new_decode_dev(journal_devnum);
4599         } else
4600                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4601
4602         really_read_only = bdev_read_only(sb->s_bdev);
4603
4604         /*
4605          * Are we loading a blank journal or performing recovery after a
4606          * crash?  For recovery, we need to check in advance whether we
4607          * can get read-write access to the device.
4608          */
4609         if (ext4_has_feature_journal_needs_recovery(sb)) {
4610                 if (sb_rdonly(sb)) {
4611                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4612                                         "required on readonly filesystem");
4613                         if (really_read_only) {
4614                                 ext4_msg(sb, KERN_ERR, "write access "
4615                                         "unavailable, cannot proceed");
4616                                 return -EROFS;
4617                         }
4618                         ext4_msg(sb, KERN_INFO, "write access will "
4619                                "be enabled during recovery");
4620                 }
4621         }
4622
4623         if (journal_inum && journal_dev) {
4624                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4625                        "and inode journals!");
4626                 return -EINVAL;
4627         }
4628
4629         if (journal_inum) {
4630                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4631                         return -EINVAL;
4632         } else {
4633                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4634                         return -EINVAL;
4635         }
4636
4637         if (!(journal->j_flags & JBD2_BARRIER))
4638                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4639
4640         if (!ext4_has_feature_journal_needs_recovery(sb))
4641                 err = jbd2_journal_wipe(journal, !really_read_only);
4642         if (!err) {
4643                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4644                 if (save)
4645                         memcpy(save, ((char *) es) +
4646                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4647                 err = jbd2_journal_load(journal);
4648                 if (save)
4649                         memcpy(((char *) es) + EXT4_S_ERR_START,
4650                                save, EXT4_S_ERR_LEN);
4651                 kfree(save);
4652         }
4653
4654         if (err) {
4655                 ext4_msg(sb, KERN_ERR, "error loading journal");
4656                 jbd2_journal_destroy(journal);
4657                 return err;
4658         }
4659
4660         EXT4_SB(sb)->s_journal = journal;
4661         ext4_clear_journal_err(sb, es);
4662
4663         if (!really_read_only && journal_devnum &&
4664             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4665                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4666
4667                 /* Make sure we flush the recovery flag to disk. */
4668                 ext4_commit_super(sb, 1);
4669         }
4670
4671         return 0;
4672 }
4673
4674 static int ext4_commit_super(struct super_block *sb, int sync)
4675 {
4676         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4677         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4678         int error = 0;
4679
4680         if (!sbh || block_device_ejected(sb))
4681                 return error;
4682         /*
4683          * If the file system is mounted read-only, don't update the
4684          * superblock write time.  This avoids updating the superblock
4685          * write time when we are mounting the root file system
4686          * read/only but we need to replay the journal; at that point,
4687          * for people who are east of GMT and who make their clock
4688          * tick in localtime for Windows bug-for-bug compatibility,
4689          * the clock is set in the future, and this will cause e2fsck
4690          * to complain and force a full file system check.
4691          */
4692         if (!(sb->s_flags & MS_RDONLY))
4693                 es->s_wtime = cpu_to_le32(get_seconds());
4694         if (sb->s_bdev->bd_part)
4695                 es->s_kbytes_written =
4696                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4697                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4698                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4699         else
4700                 es->s_kbytes_written =
4701                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4702         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4703                 ext4_free_blocks_count_set(es,
4704                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4705                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4706         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4707                 es->s_free_inodes_count =
4708                         cpu_to_le32(percpu_counter_sum_positive(
4709                                 &EXT4_SB(sb)->s_freeinodes_counter));
4710         BUFFER_TRACE(sbh, "marking dirty");
4711         ext4_superblock_csum_set(sb);
4712         if (sync)
4713                 lock_buffer(sbh);
4714         if (buffer_write_io_error(sbh)) {
4715                 /*
4716                  * Oh, dear.  A previous attempt to write the
4717                  * superblock failed.  This could happen because the
4718                  * USB device was yanked out.  Or it could happen to
4719                  * be a transient write error and maybe the block will
4720                  * be remapped.  Nothing we can do but to retry the
4721                  * write and hope for the best.
4722                  */
4723                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4724                        "superblock detected");
4725                 clear_buffer_write_io_error(sbh);
4726                 set_buffer_uptodate(sbh);
4727         }
4728         mark_buffer_dirty(sbh);
4729         if (sync) {
4730                 unlock_buffer(sbh);
4731                 error = __sync_dirty_buffer(sbh,
4732                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4733                 if (error)
4734                         return error;
4735
4736                 error = buffer_write_io_error(sbh);
4737                 if (error) {
4738                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4739                                "superblock");
4740                         clear_buffer_write_io_error(sbh);
4741                         set_buffer_uptodate(sbh);
4742                 }
4743         }
4744         return error;
4745 }
4746
4747 /*
4748  * Have we just finished recovery?  If so, and if we are mounting (or
4749  * remounting) the filesystem readonly, then we will end up with a
4750  * consistent fs on disk.  Record that fact.
4751  */
4752 static void ext4_mark_recovery_complete(struct super_block *sb,
4753                                         struct ext4_super_block *es)
4754 {
4755         journal_t *journal = EXT4_SB(sb)->s_journal;
4756
4757         if (!ext4_has_feature_journal(sb)) {
4758                 BUG_ON(journal != NULL);
4759                 return;
4760         }
4761         jbd2_journal_lock_updates(journal);
4762         if (jbd2_journal_flush(journal) < 0)
4763                 goto out;
4764
4765         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4766                 ext4_clear_feature_journal_needs_recovery(sb);
4767                 ext4_commit_super(sb, 1);
4768         }
4769
4770 out:
4771         jbd2_journal_unlock_updates(journal);
4772 }
4773
4774 /*
4775  * If we are mounting (or read-write remounting) a filesystem whose journal
4776  * has recorded an error from a previous lifetime, move that error to the
4777  * main filesystem now.
4778  */
4779 static void ext4_clear_journal_err(struct super_block *sb,
4780                                    struct ext4_super_block *es)
4781 {
4782         journal_t *journal;
4783         int j_errno;
4784         const char *errstr;
4785
4786         BUG_ON(!ext4_has_feature_journal(sb));
4787
4788         journal = EXT4_SB(sb)->s_journal;
4789
4790         /*
4791          * Now check for any error status which may have been recorded in the
4792          * journal by a prior ext4_error() or ext4_abort()
4793          */
4794
4795         j_errno = jbd2_journal_errno(journal);
4796         if (j_errno) {
4797                 char nbuf[16];
4798
4799                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4800                 ext4_warning(sb, "Filesystem error recorded "
4801                              "from previous mount: %s", errstr);
4802                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4803
4804                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4805                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4806                 ext4_commit_super(sb, 1);
4807
4808                 jbd2_journal_clear_err(journal);
4809                 jbd2_journal_update_sb_errno(journal);
4810         }
4811 }
4812
4813 /*
4814  * Force the running and committing transactions to commit,
4815  * and wait on the commit.
4816  */
4817 int ext4_force_commit(struct super_block *sb)
4818 {
4819         journal_t *journal;
4820
4821         if (sb_rdonly(sb))
4822                 return 0;
4823
4824         journal = EXT4_SB(sb)->s_journal;
4825         return ext4_journal_force_commit(journal);
4826 }
4827
4828 static int ext4_sync_fs(struct super_block *sb, int wait)
4829 {
4830         int ret = 0;
4831         tid_t target;
4832         bool needs_barrier = false;
4833         struct ext4_sb_info *sbi = EXT4_SB(sb);
4834
4835         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4836                 return 0;
4837
4838         trace_ext4_sync_fs(sb, wait);
4839         flush_workqueue(sbi->rsv_conversion_wq);
4840         /*
4841          * Writeback quota in non-journalled quota case - journalled quota has
4842          * no dirty dquots
4843          */
4844         dquot_writeback_dquots(sb, -1);
4845         /*
4846          * Data writeback is possible w/o journal transaction, so barrier must
4847          * being sent at the end of the function. But we can skip it if
4848          * transaction_commit will do it for us.
4849          */
4850         if (sbi->s_journal) {
4851                 target = jbd2_get_latest_transaction(sbi->s_journal);
4852                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4853                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4854                         needs_barrier = true;
4855
4856                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4857                         if (wait)
4858                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4859                                                            target);
4860                 }
4861         } else if (wait && test_opt(sb, BARRIER))
4862                 needs_barrier = true;
4863         if (needs_barrier) {
4864                 int err;
4865                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4866                 if (!ret)
4867                         ret = err;
4868         }
4869
4870         return ret;
4871 }
4872
4873 /*
4874  * LVM calls this function before a (read-only) snapshot is created.  This
4875  * gives us a chance to flush the journal completely and mark the fs clean.
4876  *
4877  * Note that only this function cannot bring a filesystem to be in a clean
4878  * state independently. It relies on upper layer to stop all data & metadata
4879  * modifications.
4880  */
4881 static int ext4_freeze(struct super_block *sb)
4882 {
4883         int error = 0;
4884         journal_t *journal;
4885
4886         if (sb_rdonly(sb))
4887                 return 0;
4888
4889         journal = EXT4_SB(sb)->s_journal;
4890
4891         if (journal) {
4892                 /* Now we set up the journal barrier. */
4893                 jbd2_journal_lock_updates(journal);
4894
4895                 /*
4896                  * Don't clear the needs_recovery flag if we failed to
4897                  * flush the journal.
4898                  */
4899                 error = jbd2_journal_flush(journal);
4900                 if (error < 0)
4901                         goto out;
4902
4903                 /* Journal blocked and flushed, clear needs_recovery flag. */
4904                 ext4_clear_feature_journal_needs_recovery(sb);
4905         }
4906
4907         error = ext4_commit_super(sb, 1);
4908 out:
4909         if (journal)
4910                 /* we rely on upper layer to stop further updates */
4911                 jbd2_journal_unlock_updates(journal);
4912         return error;
4913 }
4914
4915 /*
4916  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4917  * flag here, even though the filesystem is not technically dirty yet.
4918  */
4919 static int ext4_unfreeze(struct super_block *sb)
4920 {
4921         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
4922                 return 0;
4923
4924         if (EXT4_SB(sb)->s_journal) {
4925                 /* Reset the needs_recovery flag before the fs is unlocked. */
4926                 ext4_set_feature_journal_needs_recovery(sb);
4927         }
4928
4929         ext4_commit_super(sb, 1);
4930         return 0;
4931 }
4932
4933 /*
4934  * Structure to save mount options for ext4_remount's benefit
4935  */
4936 struct ext4_mount_options {
4937         unsigned long s_mount_opt;
4938         unsigned long s_mount_opt2;
4939         kuid_t s_resuid;
4940         kgid_t s_resgid;
4941         unsigned long s_commit_interval;
4942         u32 s_min_batch_time, s_max_batch_time;
4943 #ifdef CONFIG_QUOTA
4944         int s_jquota_fmt;
4945         char *s_qf_names[EXT4_MAXQUOTAS];
4946 #endif
4947 };
4948
4949 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4950 {
4951         struct ext4_super_block *es;
4952         struct ext4_sb_info *sbi = EXT4_SB(sb);
4953         unsigned long old_sb_flags;
4954         struct ext4_mount_options old_opts;
4955         int enable_quota = 0;
4956         ext4_group_t g;
4957         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4958         int err = 0;
4959 #ifdef CONFIG_QUOTA
4960         int i, j;
4961 #endif
4962         char *orig_data = kstrdup(data, GFP_KERNEL);
4963
4964         /* Store the original options */
4965         old_sb_flags = sb->s_flags;
4966         old_opts.s_mount_opt = sbi->s_mount_opt;
4967         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4968         old_opts.s_resuid = sbi->s_resuid;
4969         old_opts.s_resgid = sbi->s_resgid;
4970         old_opts.s_commit_interval = sbi->s_commit_interval;
4971         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4972         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4973 #ifdef CONFIG_QUOTA
4974         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4975         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4976                 if (sbi->s_qf_names[i]) {
4977                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4978                                                          GFP_KERNEL);
4979                         if (!old_opts.s_qf_names[i]) {
4980                                 for (j = 0; j < i; j++)
4981                                         kfree(old_opts.s_qf_names[j]);
4982                                 kfree(orig_data);
4983                                 return -ENOMEM;
4984                         }
4985                 } else
4986                         old_opts.s_qf_names[i] = NULL;
4987 #endif
4988         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4989                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4990
4991         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4992                 err = -EINVAL;
4993                 goto restore_opts;
4994         }
4995
4996         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4997             test_opt(sb, JOURNAL_CHECKSUM)) {
4998                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4999                          "during remount not supported; ignoring");
5000                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5001         }
5002
5003         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5004                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5005                         ext4_msg(sb, KERN_ERR, "can't mount with "
5006                                  "both data=journal and delalloc");
5007                         err = -EINVAL;
5008                         goto restore_opts;
5009                 }
5010                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5011                         ext4_msg(sb, KERN_ERR, "can't mount with "
5012                                  "both data=journal and dioread_nolock");
5013                         err = -EINVAL;
5014                         goto restore_opts;
5015                 }
5016                 if (test_opt(sb, DAX)) {
5017                         ext4_msg(sb, KERN_ERR, "can't mount with "
5018                                  "both data=journal and dax");
5019                         err = -EINVAL;
5020                         goto restore_opts;
5021                 }
5022         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5023                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5024                         ext4_msg(sb, KERN_ERR, "can't mount with "
5025                                 "journal_async_commit in data=ordered mode");
5026                         err = -EINVAL;
5027                         goto restore_opts;
5028                 }
5029         }
5030
5031         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5032                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5033                 err = -EINVAL;
5034                 goto restore_opts;
5035         }
5036
5037         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5038                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5039                         "dax flag with busy inodes while remounting");
5040                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5041         }
5042
5043         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5044                 ext4_abort(sb, "Abort forced by user");
5045
5046         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
5047                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
5048
5049         es = sbi->s_es;
5050
5051         if (sbi->s_journal) {
5052                 ext4_init_journal_params(sb, sbi->s_journal);
5053                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5054         }
5055
5056         if (*flags & MS_LAZYTIME)
5057                 sb->s_flags |= MS_LAZYTIME;
5058
5059         if ((bool)(*flags & MS_RDONLY) != sb_rdonly(sb)) {
5060                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5061                         err = -EROFS;
5062                         goto restore_opts;
5063                 }
5064
5065                 if (*flags & MS_RDONLY) {
5066                         err = sync_filesystem(sb);
5067                         if (err < 0)
5068                                 goto restore_opts;
5069                         err = dquot_suspend(sb, -1);
5070                         if (err < 0)
5071                                 goto restore_opts;
5072
5073                         /*
5074                          * First of all, the unconditional stuff we have to do
5075                          * to disable replay of the journal when we next remount
5076                          */
5077                         sb->s_flags |= MS_RDONLY;
5078
5079                         /*
5080                          * OK, test if we are remounting a valid rw partition
5081                          * readonly, and if so set the rdonly flag and then
5082                          * mark the partition as valid again.
5083                          */
5084                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5085                             (sbi->s_mount_state & EXT4_VALID_FS))
5086                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5087
5088                         if (sbi->s_journal)
5089                                 ext4_mark_recovery_complete(sb, es);
5090                 } else {
5091                         /* Make sure we can mount this feature set readwrite */
5092                         if (ext4_has_feature_readonly(sb) ||
5093                             !ext4_feature_set_ok(sb, 0)) {
5094                                 err = -EROFS;
5095                                 goto restore_opts;
5096                         }
5097                         /*
5098                          * Make sure the group descriptor checksums
5099                          * are sane.  If they aren't, refuse to remount r/w.
5100                          */
5101                         for (g = 0; g < sbi->s_groups_count; g++) {
5102                                 struct ext4_group_desc *gdp =
5103                                         ext4_get_group_desc(sb, g, NULL);
5104
5105                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5106                                         ext4_msg(sb, KERN_ERR,
5107                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5108                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5109                                                le16_to_cpu(gdp->bg_checksum));
5110                                         err = -EFSBADCRC;
5111                                         goto restore_opts;
5112                                 }
5113                         }
5114
5115                         /*
5116                          * If we have an unprocessed orphan list hanging
5117                          * around from a previously readonly bdev mount,
5118                          * require a full umount/remount for now.
5119                          */
5120                         if (es->s_last_orphan) {
5121                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5122                                        "remount RDWR because of unprocessed "
5123                                        "orphan inode list.  Please "
5124                                        "umount/remount instead");
5125                                 err = -EINVAL;
5126                                 goto restore_opts;
5127                         }
5128
5129                         /*
5130                          * Mounting a RDONLY partition read-write, so reread
5131                          * and store the current valid flag.  (It may have
5132                          * been changed by e2fsck since we originally mounted
5133                          * the partition.)
5134                          */
5135                         if (sbi->s_journal)
5136                                 ext4_clear_journal_err(sb, es);
5137                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5138                         if (!ext4_setup_super(sb, es, 0))
5139                                 sb->s_flags &= ~MS_RDONLY;
5140                         if (ext4_has_feature_mmp(sb))
5141                                 if (ext4_multi_mount_protect(sb,
5142                                                 le64_to_cpu(es->s_mmp_block))) {
5143                                         err = -EROFS;
5144                                         goto restore_opts;
5145                                 }
5146                         enable_quota = 1;
5147                 }
5148         }
5149
5150         /*
5151          * Reinitialize lazy itable initialization thread based on
5152          * current settings
5153          */
5154         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5155                 ext4_unregister_li_request(sb);
5156         else {
5157                 ext4_group_t first_not_zeroed;
5158                 first_not_zeroed = ext4_has_uninit_itable(sb);
5159                 ext4_register_li_request(sb, first_not_zeroed);
5160         }
5161
5162         ext4_setup_system_zone(sb);
5163         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5164                 ext4_commit_super(sb, 1);
5165
5166 #ifdef CONFIG_QUOTA
5167         /* Release old quota file names */
5168         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5169                 kfree(old_opts.s_qf_names[i]);
5170         if (enable_quota) {
5171                 if (sb_any_quota_suspended(sb))
5172                         dquot_resume(sb, -1);
5173                 else if (ext4_has_feature_quota(sb)) {
5174                         err = ext4_enable_quotas(sb);
5175                         if (err)
5176                                 goto restore_opts;
5177                 }
5178         }
5179 #endif
5180
5181         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5182         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5183         kfree(orig_data);
5184         return 0;
5185
5186 restore_opts:
5187         sb->s_flags = old_sb_flags;
5188         sbi->s_mount_opt = old_opts.s_mount_opt;
5189         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5190         sbi->s_resuid = old_opts.s_resuid;
5191         sbi->s_resgid = old_opts.s_resgid;
5192         sbi->s_commit_interval = old_opts.s_commit_interval;
5193         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5194         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5195 #ifdef CONFIG_QUOTA
5196         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5197         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5198                 kfree(sbi->s_qf_names[i]);
5199                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5200         }
5201 #endif
5202         kfree(orig_data);
5203         return err;
5204 }
5205
5206 #ifdef CONFIG_QUOTA
5207 static int ext4_statfs_project(struct super_block *sb,
5208                                kprojid_t projid, struct kstatfs *buf)
5209 {
5210         struct kqid qid;
5211         struct dquot *dquot;
5212         u64 limit;
5213         u64 curblock;
5214
5215         qid = make_kqid_projid(projid);
5216         dquot = dqget(sb, qid);
5217         if (IS_ERR(dquot))
5218                 return PTR_ERR(dquot);
5219         spin_lock(&dquot->dq_dqb_lock);
5220
5221         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5222                  dquot->dq_dqb.dqb_bsoftlimit :
5223                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5224         if (limit && buf->f_blocks > limit) {
5225                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5226                 buf->f_blocks = limit;
5227                 buf->f_bfree = buf->f_bavail =
5228                         (buf->f_blocks > curblock) ?
5229                          (buf->f_blocks - curblock) : 0;
5230         }
5231
5232         limit = dquot->dq_dqb.dqb_isoftlimit ?
5233                 dquot->dq_dqb.dqb_isoftlimit :
5234                 dquot->dq_dqb.dqb_ihardlimit;
5235         if (limit && buf->f_files > limit) {
5236                 buf->f_files = limit;
5237                 buf->f_ffree =
5238                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5239                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5240         }
5241
5242         spin_unlock(&dquot->dq_dqb_lock);
5243         dqput(dquot);
5244         return 0;
5245 }
5246 #endif
5247
5248 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5249 {
5250         struct super_block *sb = dentry->d_sb;
5251         struct ext4_sb_info *sbi = EXT4_SB(sb);
5252         struct ext4_super_block *es = sbi->s_es;
5253         ext4_fsblk_t overhead = 0, resv_blocks;
5254         u64 fsid;
5255         s64 bfree;
5256         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5257
5258         if (!test_opt(sb, MINIX_DF))
5259                 overhead = sbi->s_overhead;
5260
5261         buf->f_type = EXT4_SUPER_MAGIC;
5262         buf->f_bsize = sb->s_blocksize;
5263         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5264         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5265                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5266         /* prevent underflow in case that few free space is available */
5267         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5268         buf->f_bavail = buf->f_bfree -
5269                         (ext4_r_blocks_count(es) + resv_blocks);
5270         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5271                 buf->f_bavail = 0;
5272         buf->f_files = le32_to_cpu(es->s_inodes_count);
5273         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5274         buf->f_namelen = EXT4_NAME_LEN;
5275         fsid = le64_to_cpup((void *)es->s_uuid) ^
5276                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5277         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5278         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5279
5280 #ifdef CONFIG_QUOTA
5281         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5282             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5283                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5284 #endif
5285         return 0;
5286 }
5287
5288
5289 #ifdef CONFIG_QUOTA
5290
5291 /*
5292  * Helper functions so that transaction is started before we acquire dqio_sem
5293  * to keep correct lock ordering of transaction > dqio_sem
5294  */
5295 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5296 {
5297         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5298 }
5299
5300 static int ext4_write_dquot(struct dquot *dquot)
5301 {
5302         int ret, err;
5303         handle_t *handle;
5304         struct inode *inode;
5305
5306         inode = dquot_to_inode(dquot);
5307         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5308                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5309         if (IS_ERR(handle))
5310                 return PTR_ERR(handle);
5311         ret = dquot_commit(dquot);
5312         err = ext4_journal_stop(handle);
5313         if (!ret)
5314                 ret = err;
5315         return ret;
5316 }
5317
5318 static int ext4_acquire_dquot(struct dquot *dquot)
5319 {
5320         int ret, err;
5321         handle_t *handle;
5322
5323         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5324                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5325         if (IS_ERR(handle))
5326                 return PTR_ERR(handle);
5327         ret = dquot_acquire(dquot);
5328         err = ext4_journal_stop(handle);
5329         if (!ret)
5330                 ret = err;
5331         return ret;
5332 }
5333
5334 static int ext4_release_dquot(struct dquot *dquot)
5335 {
5336         int ret, err;
5337         handle_t *handle;
5338
5339         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5340                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5341         if (IS_ERR(handle)) {
5342                 /* Release dquot anyway to avoid endless cycle in dqput() */
5343                 dquot_release(dquot);
5344                 return PTR_ERR(handle);
5345         }
5346         ret = dquot_release(dquot);
5347         err = ext4_journal_stop(handle);
5348         if (!ret)
5349                 ret = err;
5350         return ret;
5351 }
5352
5353 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5354 {
5355         struct super_block *sb = dquot->dq_sb;
5356         struct ext4_sb_info *sbi = EXT4_SB(sb);
5357
5358         /* Are we journaling quotas? */
5359         if (ext4_has_feature_quota(sb) ||
5360             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5361                 dquot_mark_dquot_dirty(dquot);
5362                 return ext4_write_dquot(dquot);
5363         } else {
5364                 return dquot_mark_dquot_dirty(dquot);
5365         }
5366 }
5367
5368 static int ext4_write_info(struct super_block *sb, int type)
5369 {
5370         int ret, err;
5371         handle_t *handle;
5372
5373         /* Data block + inode block */
5374         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5375         if (IS_ERR(handle))
5376                 return PTR_ERR(handle);
5377         ret = dquot_commit_info(sb, type);
5378         err = ext4_journal_stop(handle);
5379         if (!ret)
5380                 ret = err;
5381         return ret;
5382 }
5383
5384 /*
5385  * Turn on quotas during mount time - we need to find
5386  * the quota file and such...
5387  */
5388 static int ext4_quota_on_mount(struct super_block *sb, int type)
5389 {
5390         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5391                                         EXT4_SB(sb)->s_jquota_fmt, type);
5392 }
5393
5394 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5395 {
5396         struct ext4_inode_info *ei = EXT4_I(inode);
5397
5398         /* The first argument of lockdep_set_subclass has to be
5399          * *exactly* the same as the argument to init_rwsem() --- in
5400          * this case, in init_once() --- or lockdep gets unhappy
5401          * because the name of the lock is set using the
5402          * stringification of the argument to init_rwsem().
5403          */
5404         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5405         lockdep_set_subclass(&ei->i_data_sem, subclass);
5406 }
5407
5408 /*
5409  * Standard function to be called on quota_on
5410  */
5411 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5412                          const struct path *path)
5413 {
5414         int err;
5415
5416         if (!test_opt(sb, QUOTA))
5417                 return -EINVAL;
5418
5419         /* Quotafile not on the same filesystem? */
5420         if (path->dentry->d_sb != sb)
5421                 return -EXDEV;
5422         /* Journaling quota? */
5423         if (EXT4_SB(sb)->s_qf_names[type]) {
5424                 /* Quotafile not in fs root? */
5425                 if (path->dentry->d_parent != sb->s_root)
5426                         ext4_msg(sb, KERN_WARNING,
5427                                 "Quota file not on filesystem root. "
5428                                 "Journaled quota will not work");
5429                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5430         } else {
5431                 /*
5432                  * Clear the flag just in case mount options changed since
5433                  * last time.
5434                  */
5435                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5436         }
5437
5438         /*
5439          * When we journal data on quota file, we have to flush journal to see
5440          * all updates to the file when we bypass pagecache...
5441          */
5442         if (EXT4_SB(sb)->s_journal &&
5443             ext4_should_journal_data(d_inode(path->dentry))) {
5444                 /*
5445                  * We don't need to lock updates but journal_flush() could
5446                  * otherwise be livelocked...
5447                  */
5448                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5449                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5450                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5451                 if (err)
5452                         return err;
5453         }
5454
5455         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5456         err = dquot_quota_on(sb, type, format_id, path);
5457         if (err) {
5458                 lockdep_set_quota_inode(path->dentry->d_inode,
5459                                              I_DATA_SEM_NORMAL);
5460         } else {
5461                 struct inode *inode = d_inode(path->dentry);
5462                 handle_t *handle;
5463
5464                 /*
5465                  * Set inode flags to prevent userspace from messing with quota
5466                  * files. If this fails, we return success anyway since quotas
5467                  * are already enabled and this is not a hard failure.
5468                  */
5469                 inode_lock(inode);
5470                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5471                 if (IS_ERR(handle))
5472                         goto unlock_inode;
5473                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5474                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5475                                 S_NOATIME | S_IMMUTABLE);
5476                 ext4_mark_inode_dirty(handle, inode);
5477                 ext4_journal_stop(handle);
5478         unlock_inode:
5479                 inode_unlock(inode);
5480         }
5481         return err;
5482 }
5483
5484 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5485                              unsigned int flags)
5486 {
5487         int err;
5488         struct inode *qf_inode;
5489         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5490                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5491                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5492                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5493         };
5494
5495         BUG_ON(!ext4_has_feature_quota(sb));
5496
5497         if (!qf_inums[type])
5498                 return -EPERM;
5499
5500         qf_inode = ext4_iget(sb, qf_inums[type]);
5501         if (IS_ERR(qf_inode)) {
5502                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5503                 return PTR_ERR(qf_inode);
5504         }
5505
5506         /* Don't account quota for quota files to avoid recursion */
5507         qf_inode->i_flags |= S_NOQUOTA;
5508         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5509         err = dquot_enable(qf_inode, type, format_id, flags);
5510         iput(qf_inode);
5511         if (err)
5512                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5513
5514         return err;
5515 }
5516
5517 /* Enable usage tracking for all quota types. */
5518 static int ext4_enable_quotas(struct super_block *sb)
5519 {
5520         int type, err = 0;
5521         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5522                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5523                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5524                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5525         };
5526         bool quota_mopt[EXT4_MAXQUOTAS] = {
5527                 test_opt(sb, USRQUOTA),
5528                 test_opt(sb, GRPQUOTA),
5529                 test_opt(sb, PRJQUOTA),
5530         };
5531
5532         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5533         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5534                 if (qf_inums[type]) {
5535                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5536                                 DQUOT_USAGE_ENABLED |
5537                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5538                         if (err) {
5539                                 for (type--; type >= 0; type--)
5540                                         dquot_quota_off(sb, type);
5541
5542                                 ext4_warning(sb,
5543                                         "Failed to enable quota tracking "
5544                                         "(type=%d, err=%d). Please run "
5545                                         "e2fsck to fix.", type, err);
5546                                 return err;
5547                         }
5548                 }
5549         }
5550         return 0;
5551 }
5552
5553 static int ext4_quota_off(struct super_block *sb, int type)
5554 {
5555         struct inode *inode = sb_dqopt(sb)->files[type];
5556         handle_t *handle;
5557         int err;
5558
5559         /* Force all delayed allocation blocks to be allocated.
5560          * Caller already holds s_umount sem */
5561         if (test_opt(sb, DELALLOC))
5562                 sync_filesystem(sb);
5563
5564         if (!inode || !igrab(inode))
5565                 goto out;
5566
5567         err = dquot_quota_off(sb, type);
5568         if (err || ext4_has_feature_quota(sb))
5569                 goto out_put;
5570
5571         inode_lock(inode);
5572         /*
5573          * Update modification times of quota files when userspace can
5574          * start looking at them. If we fail, we return success anyway since
5575          * this is not a hard failure and quotas are already disabled.
5576          */
5577         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5578         if (IS_ERR(handle))
5579                 goto out_unlock;
5580         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5581         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5582         inode->i_mtime = inode->i_ctime = current_time(inode);
5583         ext4_mark_inode_dirty(handle, inode);
5584         ext4_journal_stop(handle);
5585 out_unlock:
5586         inode_unlock(inode);
5587 out_put:
5588         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5589         iput(inode);
5590         return err;
5591 out:
5592         return dquot_quota_off(sb, type);
5593 }
5594
5595 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5596  * acquiring the locks... As quota files are never truncated and quota code
5597  * itself serializes the operations (and no one else should touch the files)
5598  * we don't have to be afraid of races */
5599 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5600                                size_t len, loff_t off)
5601 {
5602         struct inode *inode = sb_dqopt(sb)->files[type];
5603         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5604         int offset = off & (sb->s_blocksize - 1);
5605         int tocopy;
5606         size_t toread;
5607         struct buffer_head *bh;
5608         loff_t i_size = i_size_read(inode);
5609
5610         if (off > i_size)
5611                 return 0;
5612         if (off+len > i_size)
5613                 len = i_size-off;
5614         toread = len;
5615         while (toread > 0) {
5616                 tocopy = sb->s_blocksize - offset < toread ?
5617                                 sb->s_blocksize - offset : toread;
5618                 bh = ext4_bread(NULL, inode, blk, 0);
5619                 if (IS_ERR(bh))
5620                         return PTR_ERR(bh);
5621                 if (!bh)        /* A hole? */
5622                         memset(data, 0, tocopy);
5623                 else
5624                         memcpy(data, bh->b_data+offset, tocopy);
5625                 brelse(bh);
5626                 offset = 0;
5627                 toread -= tocopy;
5628                 data += tocopy;
5629                 blk++;
5630         }
5631         return len;
5632 }
5633
5634 /* Write to quotafile (we know the transaction is already started and has
5635  * enough credits) */
5636 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5637                                 const char *data, size_t len, loff_t off)
5638 {
5639         struct inode *inode = sb_dqopt(sb)->files[type];
5640         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5641         int err, offset = off & (sb->s_blocksize - 1);
5642         int retries = 0;
5643         struct buffer_head *bh;
5644         handle_t *handle = journal_current_handle();
5645
5646         if (EXT4_SB(sb)->s_journal && !handle) {
5647                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5648                         " cancelled because transaction is not started",
5649                         (unsigned long long)off, (unsigned long long)len);
5650                 return -EIO;
5651         }
5652         /*
5653          * Since we account only one data block in transaction credits,
5654          * then it is impossible to cross a block boundary.
5655          */
5656         if (sb->s_blocksize - offset < len) {
5657                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5658                         " cancelled because not block aligned",
5659                         (unsigned long long)off, (unsigned long long)len);
5660                 return -EIO;
5661         }
5662
5663         do {
5664                 bh = ext4_bread(handle, inode, blk,
5665                                 EXT4_GET_BLOCKS_CREATE |
5666                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5667         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5668                  ext4_should_retry_alloc(inode->i_sb, &retries));
5669         if (IS_ERR(bh))
5670                 return PTR_ERR(bh);
5671         if (!bh)
5672                 goto out;
5673         BUFFER_TRACE(bh, "get write access");
5674         err = ext4_journal_get_write_access(handle, bh);
5675         if (err) {
5676                 brelse(bh);
5677                 return err;
5678         }
5679         lock_buffer(bh);
5680         memcpy(bh->b_data+offset, data, len);
5681         flush_dcache_page(bh->b_page);
5682         unlock_buffer(bh);
5683         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5684         brelse(bh);
5685 out:
5686         if (inode->i_size < off + len) {
5687                 i_size_write(inode, off + len);
5688                 EXT4_I(inode)->i_disksize = inode->i_size;
5689                 ext4_mark_inode_dirty(handle, inode);
5690         }
5691         return len;
5692 }
5693
5694 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5695 {
5696         const struct quota_format_ops   *ops;
5697
5698         if (!sb_has_quota_loaded(sb, qid->type))
5699                 return -ESRCH;
5700         ops = sb_dqopt(sb)->ops[qid->type];
5701         if (!ops || !ops->get_next_id)
5702                 return -ENOSYS;
5703         return dquot_get_next_id(sb, qid);
5704 }
5705 #endif
5706
5707 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5708                        const char *dev_name, void *data)
5709 {
5710         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5711 }
5712
5713 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5714 static inline void register_as_ext2(void)
5715 {
5716         int err = register_filesystem(&ext2_fs_type);
5717         if (err)
5718                 printk(KERN_WARNING
5719                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5720 }
5721
5722 static inline void unregister_as_ext2(void)
5723 {
5724         unregister_filesystem(&ext2_fs_type);
5725 }
5726
5727 static inline int ext2_feature_set_ok(struct super_block *sb)
5728 {
5729         if (ext4_has_unknown_ext2_incompat_features(sb))
5730                 return 0;
5731         if (sb_rdonly(sb))
5732                 return 1;
5733         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5734                 return 0;
5735         return 1;
5736 }
5737 #else
5738 static inline void register_as_ext2(void) { }
5739 static inline void unregister_as_ext2(void) { }
5740 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5741 #endif
5742
5743 static inline void register_as_ext3(void)
5744 {
5745         int err = register_filesystem(&ext3_fs_type);
5746         if (err)
5747                 printk(KERN_WARNING
5748                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5749 }
5750
5751 static inline void unregister_as_ext3(void)
5752 {
5753         unregister_filesystem(&ext3_fs_type);
5754 }
5755
5756 static inline int ext3_feature_set_ok(struct super_block *sb)
5757 {
5758         if (ext4_has_unknown_ext3_incompat_features(sb))
5759                 return 0;
5760         if (!ext4_has_feature_journal(sb))
5761                 return 0;
5762         if (sb_rdonly(sb))
5763                 return 1;
5764         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5765                 return 0;
5766         return 1;
5767 }
5768
5769 static struct file_system_type ext4_fs_type = {
5770         .owner          = THIS_MODULE,
5771         .name           = "ext4",
5772         .mount          = ext4_mount,
5773         .kill_sb        = kill_block_super,
5774         .fs_flags       = FS_REQUIRES_DEV,
5775 };
5776 MODULE_ALIAS_FS("ext4");
5777
5778 /* Shared across all ext4 file systems */
5779 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5780
5781 static int __init ext4_init_fs(void)
5782 {
5783         int i, err;
5784
5785         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5786         ext4_li_info = NULL;
5787         mutex_init(&ext4_li_mtx);
5788
5789         /* Build-time check for flags consistency */
5790         ext4_check_flag_values();
5791
5792         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5793                 init_waitqueue_head(&ext4__ioend_wq[i]);
5794
5795         err = ext4_init_es();
5796         if (err)
5797                 return err;
5798
5799         err = ext4_init_pageio();
5800         if (err)
5801                 goto out5;
5802
5803         err = ext4_init_system_zone();
5804         if (err)
5805                 goto out4;
5806
5807         err = ext4_init_sysfs();
5808         if (err)
5809                 goto out3;
5810
5811         err = ext4_init_mballoc();
5812         if (err)
5813                 goto out2;
5814         err = init_inodecache();
5815         if (err)
5816                 goto out1;
5817         register_as_ext3();
5818         register_as_ext2();
5819         err = register_filesystem(&ext4_fs_type);
5820         if (err)
5821                 goto out;
5822
5823         return 0;
5824 out:
5825         unregister_as_ext2();
5826         unregister_as_ext3();
5827         destroy_inodecache();
5828 out1:
5829         ext4_exit_mballoc();
5830 out2:
5831         ext4_exit_sysfs();
5832 out3:
5833         ext4_exit_system_zone();
5834 out4:
5835         ext4_exit_pageio();
5836 out5:
5837         ext4_exit_es();
5838
5839         return err;
5840 }
5841
5842 static void __exit ext4_exit_fs(void)
5843 {
5844         ext4_destroy_lazyinit_thread();
5845         unregister_as_ext2();
5846         unregister_as_ext3();
5847         unregister_filesystem(&ext4_fs_type);
5848         destroy_inodecache();
5849         ext4_exit_mballoc();
5850         ext4_exit_sysfs();
5851         ext4_exit_system_zone();
5852         ext4_exit_pageio();
5853         ext4_exit_es();
5854 }
5855
5856 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5857 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5858 MODULE_LICENSE("GPL");
5859 module_init(ext4_init_fs)
5860 module_exit(ext4_exit_fs)