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 |= SB_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 |= SB_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         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1163                 return -EINVAL;
1164
1165         res = ext4_convert_inline_data(inode);
1166         if (res)
1167                 return res;
1168
1169         /*
1170          * If a journal handle was specified, then the encryption context is
1171          * being set on a new inode via inheritance and is part of a larger
1172          * transaction to create the inode.  Otherwise the encryption context is
1173          * being set on an existing inode in its own transaction.  Only in the
1174          * latter case should the "retry on ENOSPC" logic be used.
1175          */
1176
1177         if (handle) {
1178                 res = ext4_xattr_set_handle(handle, inode,
1179                                             EXT4_XATTR_INDEX_ENCRYPTION,
1180                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1181                                             ctx, len, 0);
1182                 if (!res) {
1183                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1184                         ext4_clear_inode_state(inode,
1185                                         EXT4_STATE_MAY_INLINE_DATA);
1186                         /*
1187                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1188                          * S_DAX may be disabled
1189                          */
1190                         ext4_set_inode_flags(inode);
1191                 }
1192                 return res;
1193         }
1194
1195         res = dquot_initialize(inode);
1196         if (res)
1197                 return res;
1198 retry:
1199         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1200                                      &credits);
1201         if (res)
1202                 return res;
1203
1204         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1205         if (IS_ERR(handle))
1206                 return PTR_ERR(handle);
1207
1208         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1209                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1210                                     ctx, len, 0);
1211         if (!res) {
1212                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1213                 /*
1214                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1215                  * S_DAX may be disabled
1216                  */
1217                 ext4_set_inode_flags(inode);
1218                 res = ext4_mark_inode_dirty(handle, inode);
1219                 if (res)
1220                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1221         }
1222         res2 = ext4_journal_stop(handle);
1223
1224         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1225                 goto retry;
1226         if (!res)
1227                 res = res2;
1228         return res;
1229 }
1230
1231 static bool ext4_dummy_context(struct inode *inode)
1232 {
1233         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1234 }
1235
1236 static unsigned ext4_max_namelen(struct inode *inode)
1237 {
1238         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1239                 EXT4_NAME_LEN;
1240 }
1241
1242 static const struct fscrypt_operations ext4_cryptops = {
1243         .key_prefix             = "ext4:",
1244         .get_context            = ext4_get_context,
1245         .set_context            = ext4_set_context,
1246         .dummy_context          = ext4_dummy_context,
1247         .empty_dir              = ext4_empty_dir,
1248         .max_namelen            = ext4_max_namelen,
1249 };
1250 #endif
1251
1252 #ifdef CONFIG_QUOTA
1253 static const char * const quotatypes[] = INITQFNAMES;
1254 #define QTYPE2NAME(t) (quotatypes[t])
1255
1256 static int ext4_write_dquot(struct dquot *dquot);
1257 static int ext4_acquire_dquot(struct dquot *dquot);
1258 static int ext4_release_dquot(struct dquot *dquot);
1259 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1260 static int ext4_write_info(struct super_block *sb, int type);
1261 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1262                          const struct path *path);
1263 static int ext4_quota_on_mount(struct super_block *sb, int type);
1264 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1265                                size_t len, loff_t off);
1266 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1267                                 const char *data, size_t len, loff_t off);
1268 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1269                              unsigned int flags);
1270 static int ext4_enable_quotas(struct super_block *sb);
1271 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1272
1273 static struct dquot **ext4_get_dquots(struct inode *inode)
1274 {
1275         return EXT4_I(inode)->i_dquot;
1276 }
1277
1278 static const struct dquot_operations ext4_quota_operations = {
1279         .get_reserved_space     = ext4_get_reserved_space,
1280         .write_dquot            = ext4_write_dquot,
1281         .acquire_dquot          = ext4_acquire_dquot,
1282         .release_dquot          = ext4_release_dquot,
1283         .mark_dirty             = ext4_mark_dquot_dirty,
1284         .write_info             = ext4_write_info,
1285         .alloc_dquot            = dquot_alloc,
1286         .destroy_dquot          = dquot_destroy,
1287         .get_projid             = ext4_get_projid,
1288         .get_inode_usage        = ext4_get_inode_usage,
1289         .get_next_id            = ext4_get_next_id,
1290 };
1291
1292 static const struct quotactl_ops ext4_qctl_operations = {
1293         .quota_on       = ext4_quota_on,
1294         .quota_off      = ext4_quota_off,
1295         .quota_sync     = dquot_quota_sync,
1296         .get_state      = dquot_get_state,
1297         .set_info       = dquot_set_dqinfo,
1298         .get_dqblk      = dquot_get_dqblk,
1299         .set_dqblk      = dquot_set_dqblk,
1300         .get_nextdqblk  = dquot_get_next_dqblk,
1301 };
1302 #endif
1303
1304 static const struct super_operations ext4_sops = {
1305         .alloc_inode    = ext4_alloc_inode,
1306         .destroy_inode  = ext4_destroy_inode,
1307         .write_inode    = ext4_write_inode,
1308         .dirty_inode    = ext4_dirty_inode,
1309         .drop_inode     = ext4_drop_inode,
1310         .evict_inode    = ext4_evict_inode,
1311         .put_super      = ext4_put_super,
1312         .sync_fs        = ext4_sync_fs,
1313         .freeze_fs      = ext4_freeze,
1314         .unfreeze_fs    = ext4_unfreeze,
1315         .statfs         = ext4_statfs,
1316         .remount_fs     = ext4_remount,
1317         .show_options   = ext4_show_options,
1318 #ifdef CONFIG_QUOTA
1319         .quota_read     = ext4_quota_read,
1320         .quota_write    = ext4_quota_write,
1321         .get_dquots     = ext4_get_dquots,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_removed,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1339         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1348         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1349         Opt_inode_readahead_blks, Opt_journal_ioprio,
1350         Opt_dioread_nolock, Opt_dioread_lock,
1351         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1352         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1353 };
1354
1355 static const match_table_t tokens = {
1356         {Opt_bsd_df, "bsddf"},
1357         {Opt_minix_df, "minixdf"},
1358         {Opt_grpid, "grpid"},
1359         {Opt_grpid, "bsdgroups"},
1360         {Opt_nogrpid, "nogrpid"},
1361         {Opt_nogrpid, "sysvgroups"},
1362         {Opt_resgid, "resgid=%u"},
1363         {Opt_resuid, "resuid=%u"},
1364         {Opt_sb, "sb=%u"},
1365         {Opt_err_cont, "errors=continue"},
1366         {Opt_err_panic, "errors=panic"},
1367         {Opt_err_ro, "errors=remount-ro"},
1368         {Opt_nouid32, "nouid32"},
1369         {Opt_debug, "debug"},
1370         {Opt_removed, "oldalloc"},
1371         {Opt_removed, "orlov"},
1372         {Opt_user_xattr, "user_xattr"},
1373         {Opt_nouser_xattr, "nouser_xattr"},
1374         {Opt_acl, "acl"},
1375         {Opt_noacl, "noacl"},
1376         {Opt_noload, "norecovery"},
1377         {Opt_noload, "noload"},
1378         {Opt_removed, "nobh"},
1379         {Opt_removed, "bh"},
1380         {Opt_commit, "commit=%u"},
1381         {Opt_min_batch_time, "min_batch_time=%u"},
1382         {Opt_max_batch_time, "max_batch_time=%u"},
1383         {Opt_journal_dev, "journal_dev=%u"},
1384         {Opt_journal_path, "journal_path=%s"},
1385         {Opt_journal_checksum, "journal_checksum"},
1386         {Opt_nojournal_checksum, "nojournal_checksum"},
1387         {Opt_journal_async_commit, "journal_async_commit"},
1388         {Opt_abort, "abort"},
1389         {Opt_data_journal, "data=journal"},
1390         {Opt_data_ordered, "data=ordered"},
1391         {Opt_data_writeback, "data=writeback"},
1392         {Opt_data_err_abort, "data_err=abort"},
1393         {Opt_data_err_ignore, "data_err=ignore"},
1394         {Opt_offusrjquota, "usrjquota="},
1395         {Opt_usrjquota, "usrjquota=%s"},
1396         {Opt_offgrpjquota, "grpjquota="},
1397         {Opt_grpjquota, "grpjquota=%s"},
1398         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1399         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1400         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1401         {Opt_grpquota, "grpquota"},
1402         {Opt_noquota, "noquota"},
1403         {Opt_quota, "quota"},
1404         {Opt_usrquota, "usrquota"},
1405         {Opt_prjquota, "prjquota"},
1406         {Opt_barrier, "barrier=%u"},
1407         {Opt_barrier, "barrier"},
1408         {Opt_nobarrier, "nobarrier"},
1409         {Opt_i_version, "i_version"},
1410         {Opt_dax, "dax"},
1411         {Opt_stripe, "stripe=%u"},
1412         {Opt_delalloc, "delalloc"},
1413         {Opt_lazytime, "lazytime"},
1414         {Opt_nolazytime, "nolazytime"},
1415         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1416         {Opt_nodelalloc, "nodelalloc"},
1417         {Opt_removed, "mblk_io_submit"},
1418         {Opt_removed, "nomblk_io_submit"},
1419         {Opt_block_validity, "block_validity"},
1420         {Opt_noblock_validity, "noblock_validity"},
1421         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1422         {Opt_journal_ioprio, "journal_ioprio=%u"},
1423         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1424         {Opt_auto_da_alloc, "auto_da_alloc"},
1425         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1426         {Opt_dioread_nolock, "dioread_nolock"},
1427         {Opt_dioread_lock, "dioread_lock"},
1428         {Opt_discard, "discard"},
1429         {Opt_nodiscard, "nodiscard"},
1430         {Opt_init_itable, "init_itable=%u"},
1431         {Opt_init_itable, "init_itable"},
1432         {Opt_noinit_itable, "noinit_itable"},
1433         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1434         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1435         {Opt_nombcache, "nombcache"},
1436         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1437         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1438         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1439         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1440         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1441         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1442         {Opt_err, NULL},
1443 };
1444
1445 static ext4_fsblk_t get_sb_block(void **data)
1446 {
1447         ext4_fsblk_t    sb_block;
1448         char            *options = (char *) *data;
1449
1450         if (!options || strncmp(options, "sb=", 3) != 0)
1451                 return 1;       /* Default location */
1452
1453         options += 3;
1454         /* TODO: use simple_strtoll with >32bit ext4 */
1455         sb_block = simple_strtoul(options, &options, 0);
1456         if (*options && *options != ',') {
1457                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1458                        (char *) *data);
1459                 return 1;
1460         }
1461         if (*options == ',')
1462                 options++;
1463         *data = (void *) options;
1464
1465         return sb_block;
1466 }
1467
1468 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1469 static const char deprecated_msg[] =
1470         "Mount option \"%s\" will be removed by %s\n"
1471         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1472
1473 #ifdef CONFIG_QUOTA
1474 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1475 {
1476         struct ext4_sb_info *sbi = EXT4_SB(sb);
1477         char *qname;
1478         int ret = -1;
1479
1480         if (sb_any_quota_loaded(sb) &&
1481                 !sbi->s_qf_names[qtype]) {
1482                 ext4_msg(sb, KERN_ERR,
1483                         "Cannot change journaled "
1484                         "quota options when quota turned on");
1485                 return -1;
1486         }
1487         if (ext4_has_feature_quota(sb)) {
1488                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1489                          "ignored when QUOTA feature is enabled");
1490                 return 1;
1491         }
1492         qname = match_strdup(args);
1493         if (!qname) {
1494                 ext4_msg(sb, KERN_ERR,
1495                         "Not enough memory for storing quotafile name");
1496                 return -1;
1497         }
1498         if (sbi->s_qf_names[qtype]) {
1499                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1500                         ret = 1;
1501                 else
1502                         ext4_msg(sb, KERN_ERR,
1503                                  "%s quota file already specified",
1504                                  QTYPE2NAME(qtype));
1505                 goto errout;
1506         }
1507         if (strchr(qname, '/')) {
1508                 ext4_msg(sb, KERN_ERR,
1509                         "quotafile must be on filesystem root");
1510                 goto errout;
1511         }
1512         sbi->s_qf_names[qtype] = qname;
1513         set_opt(sb, QUOTA);
1514         return 1;
1515 errout:
1516         kfree(qname);
1517         return ret;
1518 }
1519
1520 static int clear_qf_name(struct super_block *sb, int qtype)
1521 {
1522
1523         struct ext4_sb_info *sbi = EXT4_SB(sb);
1524
1525         if (sb_any_quota_loaded(sb) &&
1526                 sbi->s_qf_names[qtype]) {
1527                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1528                         " when quota turned on");
1529                 return -1;
1530         }
1531         kfree(sbi->s_qf_names[qtype]);
1532         sbi->s_qf_names[qtype] = NULL;
1533         return 1;
1534 }
1535 #endif
1536
1537 #define MOPT_SET        0x0001
1538 #define MOPT_CLEAR      0x0002
1539 #define MOPT_NOSUPPORT  0x0004
1540 #define MOPT_EXPLICIT   0x0008
1541 #define MOPT_CLEAR_ERR  0x0010
1542 #define MOPT_GTE0       0x0020
1543 #ifdef CONFIG_QUOTA
1544 #define MOPT_Q          0
1545 #define MOPT_QFMT       0x0040
1546 #else
1547 #define MOPT_Q          MOPT_NOSUPPORT
1548 #define MOPT_QFMT       MOPT_NOSUPPORT
1549 #endif
1550 #define MOPT_DATAJ      0x0080
1551 #define MOPT_NO_EXT2    0x0100
1552 #define MOPT_NO_EXT3    0x0200
1553 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1554 #define MOPT_STRING     0x0400
1555
1556 static const struct mount_opts {
1557         int     token;
1558         int     mount_opt;
1559         int     flags;
1560 } ext4_mount_opts[] = {
1561         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1562         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1563         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1564         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1565         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1566         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1567         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1568          MOPT_EXT4_ONLY | MOPT_SET},
1569         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1570          MOPT_EXT4_ONLY | MOPT_CLEAR},
1571         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1572         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1573         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1574          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1575         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1576          MOPT_EXT4_ONLY | MOPT_CLEAR},
1577         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1578          MOPT_EXT4_ONLY | MOPT_CLEAR},
1579         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1580          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1581         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1582                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1583          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1584         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1585         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1586         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1587         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1588         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1589          MOPT_NO_EXT2},
1590         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1591          MOPT_NO_EXT2},
1592         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1593         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1594         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1595         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1596         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1597         {Opt_commit, 0, MOPT_GTE0},
1598         {Opt_max_batch_time, 0, MOPT_GTE0},
1599         {Opt_min_batch_time, 0, MOPT_GTE0},
1600         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1601         {Opt_init_itable, 0, MOPT_GTE0},
1602         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1603         {Opt_stripe, 0, MOPT_GTE0},
1604         {Opt_resuid, 0, MOPT_GTE0},
1605         {Opt_resgid, 0, MOPT_GTE0},
1606         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1607         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1608         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1609         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1610         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1611         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1612          MOPT_NO_EXT2 | MOPT_DATAJ},
1613         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1614         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1615 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1616         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1617         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1618 #else
1619         {Opt_acl, 0, MOPT_NOSUPPORT},
1620         {Opt_noacl, 0, MOPT_NOSUPPORT},
1621 #endif
1622         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1623         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1624         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1625         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1626         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1627                                                         MOPT_SET | MOPT_Q},
1628         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1629                                                         MOPT_SET | MOPT_Q},
1630         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1631                                                         MOPT_SET | MOPT_Q},
1632         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1633                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1634                                                         MOPT_CLEAR | MOPT_Q},
1635         {Opt_usrjquota, 0, MOPT_Q},
1636         {Opt_grpjquota, 0, MOPT_Q},
1637         {Opt_offusrjquota, 0, MOPT_Q},
1638         {Opt_offgrpjquota, 0, MOPT_Q},
1639         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1640         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1641         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1642         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1643         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1644         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1645         {Opt_err, 0, 0}
1646 };
1647
1648 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1649                             substring_t *args, unsigned long *journal_devnum,
1650                             unsigned int *journal_ioprio, int is_remount)
1651 {
1652         struct ext4_sb_info *sbi = EXT4_SB(sb);
1653         const struct mount_opts *m;
1654         kuid_t uid;
1655         kgid_t gid;
1656         int arg = 0;
1657
1658 #ifdef CONFIG_QUOTA
1659         if (token == Opt_usrjquota)
1660                 return set_qf_name(sb, USRQUOTA, &args[0]);
1661         else if (token == Opt_grpjquota)
1662                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1663         else if (token == Opt_offusrjquota)
1664                 return clear_qf_name(sb, USRQUOTA);
1665         else if (token == Opt_offgrpjquota)
1666                 return clear_qf_name(sb, GRPQUOTA);
1667 #endif
1668         switch (token) {
1669         case Opt_noacl:
1670         case Opt_nouser_xattr:
1671                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1672                 break;
1673         case Opt_sb:
1674                 return 1;       /* handled by get_sb_block() */
1675         case Opt_removed:
1676                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1677                 return 1;
1678         case Opt_abort:
1679                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1680                 return 1;
1681         case Opt_i_version:
1682                 sb->s_flags |= SB_I_VERSION;
1683                 return 1;
1684         case Opt_lazytime:
1685                 sb->s_flags |= SB_LAZYTIME;
1686                 return 1;
1687         case Opt_nolazytime:
1688                 sb->s_flags &= ~SB_LAZYTIME;
1689                 return 1;
1690         }
1691
1692         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1693                 if (token == m->token)
1694                         break;
1695
1696         if (m->token == Opt_err) {
1697                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1698                          "or missing value", opt);
1699                 return -1;
1700         }
1701
1702         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1703                 ext4_msg(sb, KERN_ERR,
1704                          "Mount option \"%s\" incompatible with ext2", opt);
1705                 return -1;
1706         }
1707         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1708                 ext4_msg(sb, KERN_ERR,
1709                          "Mount option \"%s\" incompatible with ext3", opt);
1710                 return -1;
1711         }
1712
1713         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1714                 return -1;
1715         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1716                 return -1;
1717         if (m->flags & MOPT_EXPLICIT) {
1718                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1719                         set_opt2(sb, EXPLICIT_DELALLOC);
1720                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1721                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1722                 } else
1723                         return -1;
1724         }
1725         if (m->flags & MOPT_CLEAR_ERR)
1726                 clear_opt(sb, ERRORS_MASK);
1727         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1728                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1729                          "options when quota turned on");
1730                 return -1;
1731         }
1732
1733         if (m->flags & MOPT_NOSUPPORT) {
1734                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1735         } else if (token == Opt_commit) {
1736                 if (arg == 0)
1737                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1738                 sbi->s_commit_interval = HZ * arg;
1739         } else if (token == Opt_debug_want_extra_isize) {
1740                 sbi->s_want_extra_isize = arg;
1741         } else if (token == Opt_max_batch_time) {
1742                 sbi->s_max_batch_time = arg;
1743         } else if (token == Opt_min_batch_time) {
1744                 sbi->s_min_batch_time = arg;
1745         } else if (token == Opt_inode_readahead_blks) {
1746                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1747                         ext4_msg(sb, KERN_ERR,
1748                                  "EXT4-fs: inode_readahead_blks must be "
1749                                  "0 or a power of 2 smaller than 2^31");
1750                         return -1;
1751                 }
1752                 sbi->s_inode_readahead_blks = arg;
1753         } else if (token == Opt_init_itable) {
1754                 set_opt(sb, INIT_INODE_TABLE);
1755                 if (!args->from)
1756                         arg = EXT4_DEF_LI_WAIT_MULT;
1757                 sbi->s_li_wait_mult = arg;
1758         } else if (token == Opt_max_dir_size_kb) {
1759                 sbi->s_max_dir_size_kb = arg;
1760         } else if (token == Opt_stripe) {
1761                 sbi->s_stripe = arg;
1762         } else if (token == Opt_resuid) {
1763                 uid = make_kuid(current_user_ns(), arg);
1764                 if (!uid_valid(uid)) {
1765                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1766                         return -1;
1767                 }
1768                 sbi->s_resuid = uid;
1769         } else if (token == Opt_resgid) {
1770                 gid = make_kgid(current_user_ns(), arg);
1771                 if (!gid_valid(gid)) {
1772                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1773                         return -1;
1774                 }
1775                 sbi->s_resgid = gid;
1776         } else if (token == Opt_journal_dev) {
1777                 if (is_remount) {
1778                         ext4_msg(sb, KERN_ERR,
1779                                  "Cannot specify journal on remount");
1780                         return -1;
1781                 }
1782                 *journal_devnum = arg;
1783         } else if (token == Opt_journal_path) {
1784                 char *journal_path;
1785                 struct inode *journal_inode;
1786                 struct path path;
1787                 int error;
1788
1789                 if (is_remount) {
1790                         ext4_msg(sb, KERN_ERR,
1791                                  "Cannot specify journal on remount");
1792                         return -1;
1793                 }
1794                 journal_path = match_strdup(&args[0]);
1795                 if (!journal_path) {
1796                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1797                                 "journal device string");
1798                         return -1;
1799                 }
1800
1801                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1802                 if (error) {
1803                         ext4_msg(sb, KERN_ERR, "error: could not find "
1804                                 "journal device path: error %d", error);
1805                         kfree(journal_path);
1806                         return -1;
1807                 }
1808
1809                 journal_inode = d_inode(path.dentry);
1810                 if (!S_ISBLK(journal_inode->i_mode)) {
1811                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1812                                 "is not a block device", journal_path);
1813                         path_put(&path);
1814                         kfree(journal_path);
1815                         return -1;
1816                 }
1817
1818                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1819                 path_put(&path);
1820                 kfree(journal_path);
1821         } else if (token == Opt_journal_ioprio) {
1822                 if (arg > 7) {
1823                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1824                                  " (must be 0-7)");
1825                         return -1;
1826                 }
1827                 *journal_ioprio =
1828                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1829         } else if (token == Opt_test_dummy_encryption) {
1830 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1831                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1832                 ext4_msg(sb, KERN_WARNING,
1833                          "Test dummy encryption mode enabled");
1834 #else
1835                 ext4_msg(sb, KERN_WARNING,
1836                          "Test dummy encryption mount option ignored");
1837 #endif
1838         } else if (m->flags & MOPT_DATAJ) {
1839                 if (is_remount) {
1840                         if (!sbi->s_journal)
1841                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1842                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1843                                 ext4_msg(sb, KERN_ERR,
1844                                          "Cannot change data mode on remount");
1845                                 return -1;
1846                         }
1847                 } else {
1848                         clear_opt(sb, DATA_FLAGS);
1849                         sbi->s_mount_opt |= m->mount_opt;
1850                 }
1851 #ifdef CONFIG_QUOTA
1852         } else if (m->flags & MOPT_QFMT) {
1853                 if (sb_any_quota_loaded(sb) &&
1854                     sbi->s_jquota_fmt != m->mount_opt) {
1855                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1856                                  "quota options when quota turned on");
1857                         return -1;
1858                 }
1859                 if (ext4_has_feature_quota(sb)) {
1860                         ext4_msg(sb, KERN_INFO,
1861                                  "Quota format mount options ignored "
1862                                  "when QUOTA feature is enabled");
1863                         return 1;
1864                 }
1865                 sbi->s_jquota_fmt = m->mount_opt;
1866 #endif
1867         } else if (token == Opt_dax) {
1868 #ifdef CONFIG_FS_DAX
1869                 ext4_msg(sb, KERN_WARNING,
1870                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1871                         sbi->s_mount_opt |= m->mount_opt;
1872 #else
1873                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1874                 return -1;
1875 #endif
1876         } else if (token == Opt_data_err_abort) {
1877                 sbi->s_mount_opt |= m->mount_opt;
1878         } else if (token == Opt_data_err_ignore) {
1879                 sbi->s_mount_opt &= ~m->mount_opt;
1880         } else {
1881                 if (!args->from)
1882                         arg = 1;
1883                 if (m->flags & MOPT_CLEAR)
1884                         arg = !arg;
1885                 else if (unlikely(!(m->flags & MOPT_SET))) {
1886                         ext4_msg(sb, KERN_WARNING,
1887                                  "buggy handling of option %s", opt);
1888                         WARN_ON(1);
1889                         return -1;
1890                 }
1891                 if (arg != 0)
1892                         sbi->s_mount_opt |= m->mount_opt;
1893                 else
1894                         sbi->s_mount_opt &= ~m->mount_opt;
1895         }
1896         return 1;
1897 }
1898
1899 static int parse_options(char *options, struct super_block *sb,
1900                          unsigned long *journal_devnum,
1901                          unsigned int *journal_ioprio,
1902                          int is_remount)
1903 {
1904         struct ext4_sb_info *sbi = EXT4_SB(sb);
1905         char *p;
1906         substring_t args[MAX_OPT_ARGS];
1907         int token;
1908
1909         if (!options)
1910                 return 1;
1911
1912         while ((p = strsep(&options, ",")) != NULL) {
1913                 if (!*p)
1914                         continue;
1915                 /*
1916                  * Initialize args struct so we know whether arg was
1917                  * found; some options take optional arguments.
1918                  */
1919                 args[0].to = args[0].from = NULL;
1920                 token = match_token(p, tokens, args);
1921                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1922                                      journal_ioprio, is_remount) < 0)
1923                         return 0;
1924         }
1925 #ifdef CONFIG_QUOTA
1926         /*
1927          * We do the test below only for project quotas. 'usrquota' and
1928          * 'grpquota' mount options are allowed even without quota feature
1929          * to support legacy quotas in quota files.
1930          */
1931         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1932                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1933                          "Cannot enable project quota enforcement.");
1934                 return 0;
1935         }
1936         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1937                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1938                         clear_opt(sb, USRQUOTA);
1939
1940                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1941                         clear_opt(sb, GRPQUOTA);
1942
1943                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1944                         ext4_msg(sb, KERN_ERR, "old and new quota "
1945                                         "format mixing");
1946                         return 0;
1947                 }
1948
1949                 if (!sbi->s_jquota_fmt) {
1950                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1951                                         "not specified");
1952                         return 0;
1953                 }
1954         }
1955 #endif
1956         if (test_opt(sb, DIOREAD_NOLOCK)) {
1957                 int blocksize =
1958                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1959
1960                 if (blocksize < PAGE_SIZE) {
1961                         ext4_msg(sb, KERN_ERR, "can't mount with "
1962                                  "dioread_nolock if block size != PAGE_SIZE");
1963                         return 0;
1964                 }
1965         }
1966         return 1;
1967 }
1968
1969 static inline void ext4_show_quota_options(struct seq_file *seq,
1970                                            struct super_block *sb)
1971 {
1972 #if defined(CONFIG_QUOTA)
1973         struct ext4_sb_info *sbi = EXT4_SB(sb);
1974
1975         if (sbi->s_jquota_fmt) {
1976                 char *fmtname = "";
1977
1978                 switch (sbi->s_jquota_fmt) {
1979                 case QFMT_VFS_OLD:
1980                         fmtname = "vfsold";
1981                         break;
1982                 case QFMT_VFS_V0:
1983                         fmtname = "vfsv0";
1984                         break;
1985                 case QFMT_VFS_V1:
1986                         fmtname = "vfsv1";
1987                         break;
1988                 }
1989                 seq_printf(seq, ",jqfmt=%s", fmtname);
1990         }
1991
1992         if (sbi->s_qf_names[USRQUOTA])
1993                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1994
1995         if (sbi->s_qf_names[GRPQUOTA])
1996                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1997 #endif
1998 }
1999
2000 static const char *token2str(int token)
2001 {
2002         const struct match_token *t;
2003
2004         for (t = tokens; t->token != Opt_err; t++)
2005                 if (t->token == token && !strchr(t->pattern, '='))
2006                         break;
2007         return t->pattern;
2008 }
2009
2010 /*
2011  * Show an option if
2012  *  - it's set to a non-default value OR
2013  *  - if the per-sb default is different from the global default
2014  */
2015 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2016                               int nodefs)
2017 {
2018         struct ext4_sb_info *sbi = EXT4_SB(sb);
2019         struct ext4_super_block *es = sbi->s_es;
2020         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2021         const struct mount_opts *m;
2022         char sep = nodefs ? '\n' : ',';
2023
2024 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2025 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2026
2027         if (sbi->s_sb_block != 1)
2028                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2029
2030         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2031                 int want_set = m->flags & MOPT_SET;
2032                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2033                     (m->flags & MOPT_CLEAR_ERR))
2034                         continue;
2035                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2036                         continue; /* skip if same as the default */
2037                 if ((want_set &&
2038                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2039                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2040                         continue; /* select Opt_noFoo vs Opt_Foo */
2041                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2042         }
2043
2044         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2045             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2046                 SEQ_OPTS_PRINT("resuid=%u",
2047                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2048         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2049             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2050                 SEQ_OPTS_PRINT("resgid=%u",
2051                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2052         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2053         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2054                 SEQ_OPTS_PUTS("errors=remount-ro");
2055         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2056                 SEQ_OPTS_PUTS("errors=continue");
2057         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2058                 SEQ_OPTS_PUTS("errors=panic");
2059         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2060                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2061         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2062                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2063         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2064                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2065         if (sb->s_flags & SB_I_VERSION)
2066                 SEQ_OPTS_PUTS("i_version");
2067         if (nodefs || sbi->s_stripe)
2068                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2069         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2070                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2071                         SEQ_OPTS_PUTS("data=journal");
2072                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2073                         SEQ_OPTS_PUTS("data=ordered");
2074                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2075                         SEQ_OPTS_PUTS("data=writeback");
2076         }
2077         if (nodefs ||
2078             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2079                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2080                                sbi->s_inode_readahead_blks);
2081
2082         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2083                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2084                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2085         if (nodefs || sbi->s_max_dir_size_kb)
2086                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2087         if (test_opt(sb, DATA_ERR_ABORT))
2088                 SEQ_OPTS_PUTS("data_err=abort");
2089
2090         ext4_show_quota_options(seq, sb);
2091         return 0;
2092 }
2093
2094 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2095 {
2096         return _ext4_show_options(seq, root->d_sb, 0);
2097 }
2098
2099 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2100 {
2101         struct super_block *sb = seq->private;
2102         int rc;
2103
2104         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2105         rc = _ext4_show_options(seq, sb, 1);
2106         seq_puts(seq, "\n");
2107         return rc;
2108 }
2109
2110 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2111                             int read_only)
2112 {
2113         struct ext4_sb_info *sbi = EXT4_SB(sb);
2114         int res = 0;
2115
2116         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2117                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2118                          "forcing read-only mode");
2119                 res = SB_RDONLY;
2120         }
2121         if (read_only)
2122                 goto done;
2123         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2124                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2125                          "running e2fsck is recommended");
2126         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2127                 ext4_msg(sb, KERN_WARNING,
2128                          "warning: mounting fs with errors, "
2129                          "running e2fsck is recommended");
2130         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2131                  le16_to_cpu(es->s_mnt_count) >=
2132                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2133                 ext4_msg(sb, KERN_WARNING,
2134                          "warning: maximal mount count reached, "
2135                          "running e2fsck is recommended");
2136         else if (le32_to_cpu(es->s_checkinterval) &&
2137                 (le32_to_cpu(es->s_lastcheck) +
2138                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2139                 ext4_msg(sb, KERN_WARNING,
2140                          "warning: checktime reached, "
2141                          "running e2fsck is recommended");
2142         if (!sbi->s_journal)
2143                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2144         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2145                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2146         le16_add_cpu(&es->s_mnt_count, 1);
2147         es->s_mtime = cpu_to_le32(get_seconds());
2148         ext4_update_dynamic_rev(sb);
2149         if (sbi->s_journal)
2150                 ext4_set_feature_journal_needs_recovery(sb);
2151
2152         ext4_commit_super(sb, 1);
2153 done:
2154         if (test_opt(sb, DEBUG))
2155                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2156                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2157                         sb->s_blocksize,
2158                         sbi->s_groups_count,
2159                         EXT4_BLOCKS_PER_GROUP(sb),
2160                         EXT4_INODES_PER_GROUP(sb),
2161                         sbi->s_mount_opt, sbi->s_mount_opt2);
2162
2163         cleancache_init_fs(sb);
2164         return res;
2165 }
2166
2167 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2168 {
2169         struct ext4_sb_info *sbi = EXT4_SB(sb);
2170         struct flex_groups *new_groups;
2171         int size;
2172
2173         if (!sbi->s_log_groups_per_flex)
2174                 return 0;
2175
2176         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2177         if (size <= sbi->s_flex_groups_allocated)
2178                 return 0;
2179
2180         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2181         new_groups = kvzalloc(size, GFP_KERNEL);
2182         if (!new_groups) {
2183                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2184                          size / (int) sizeof(struct flex_groups));
2185                 return -ENOMEM;
2186         }
2187
2188         if (sbi->s_flex_groups) {
2189                 memcpy(new_groups, sbi->s_flex_groups,
2190                        (sbi->s_flex_groups_allocated *
2191                         sizeof(struct flex_groups)));
2192                 kvfree(sbi->s_flex_groups);
2193         }
2194         sbi->s_flex_groups = new_groups;
2195         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2196         return 0;
2197 }
2198
2199 static int ext4_fill_flex_info(struct super_block *sb)
2200 {
2201         struct ext4_sb_info *sbi = EXT4_SB(sb);
2202         struct ext4_group_desc *gdp = NULL;
2203         ext4_group_t flex_group;
2204         int i, err;
2205
2206         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2207         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2208                 sbi->s_log_groups_per_flex = 0;
2209                 return 1;
2210         }
2211
2212         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2213         if (err)
2214                 goto failed;
2215
2216         for (i = 0; i < sbi->s_groups_count; i++) {
2217                 gdp = ext4_get_group_desc(sb, i, NULL);
2218
2219                 flex_group = ext4_flex_group(sbi, i);
2220                 atomic_add(ext4_free_inodes_count(sb, gdp),
2221                            &sbi->s_flex_groups[flex_group].free_inodes);
2222                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2223                              &sbi->s_flex_groups[flex_group].free_clusters);
2224                 atomic_add(ext4_used_dirs_count(sb, gdp),
2225                            &sbi->s_flex_groups[flex_group].used_dirs);
2226         }
2227
2228         return 1;
2229 failed:
2230         return 0;
2231 }
2232
2233 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2234                                    struct ext4_group_desc *gdp)
2235 {
2236         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2237         __u16 crc = 0;
2238         __le32 le_group = cpu_to_le32(block_group);
2239         struct ext4_sb_info *sbi = EXT4_SB(sb);
2240
2241         if (ext4_has_metadata_csum(sbi->s_sb)) {
2242                 /* Use new metadata_csum algorithm */
2243                 __u32 csum32;
2244                 __u16 dummy_csum = 0;
2245
2246                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2247                                      sizeof(le_group));
2248                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2249                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2250                                      sizeof(dummy_csum));
2251                 offset += sizeof(dummy_csum);
2252                 if (offset < sbi->s_desc_size)
2253                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2254                                              sbi->s_desc_size - offset);
2255
2256                 crc = csum32 & 0xFFFF;
2257                 goto out;
2258         }
2259
2260         /* old crc16 code */
2261         if (!ext4_has_feature_gdt_csum(sb))
2262                 return 0;
2263
2264         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2265         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2266         crc = crc16(crc, (__u8 *)gdp, offset);
2267         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2268         /* for checksum of struct ext4_group_desc do the rest...*/
2269         if (ext4_has_feature_64bit(sb) &&
2270             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2271                 crc = crc16(crc, (__u8 *)gdp + offset,
2272                             le16_to_cpu(sbi->s_es->s_desc_size) -
2273                                 offset);
2274
2275 out:
2276         return cpu_to_le16(crc);
2277 }
2278
2279 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2280                                 struct ext4_group_desc *gdp)
2281 {
2282         if (ext4_has_group_desc_csum(sb) &&
2283             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2284                 return 0;
2285
2286         return 1;
2287 }
2288
2289 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2290                               struct ext4_group_desc *gdp)
2291 {
2292         if (!ext4_has_group_desc_csum(sb))
2293                 return;
2294         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2295 }
2296
2297 /* Called at mount-time, super-block is locked */
2298 static int ext4_check_descriptors(struct super_block *sb,
2299                                   ext4_fsblk_t sb_block,
2300                                   ext4_group_t *first_not_zeroed)
2301 {
2302         struct ext4_sb_info *sbi = EXT4_SB(sb);
2303         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2304         ext4_fsblk_t last_block;
2305         ext4_fsblk_t block_bitmap;
2306         ext4_fsblk_t inode_bitmap;
2307         ext4_fsblk_t inode_table;
2308         int flexbg_flag = 0;
2309         ext4_group_t i, grp = sbi->s_groups_count;
2310
2311         if (ext4_has_feature_flex_bg(sb))
2312                 flexbg_flag = 1;
2313
2314         ext4_debug("Checking group descriptors");
2315
2316         for (i = 0; i < sbi->s_groups_count; i++) {
2317                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2318
2319                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2320                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2321                 else
2322                         last_block = first_block +
2323                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2324
2325                 if ((grp == sbi->s_groups_count) &&
2326                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2327                         grp = i;
2328
2329                 block_bitmap = ext4_block_bitmap(sb, gdp);
2330                 if (block_bitmap == sb_block) {
2331                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2332                                  "Block bitmap for group %u overlaps "
2333                                  "superblock", i);
2334                 }
2335                 if (block_bitmap < first_block || block_bitmap > last_block) {
2336                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2337                                "Block bitmap for group %u not in group "
2338                                "(block %llu)!", i, block_bitmap);
2339                         return 0;
2340                 }
2341                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2342                 if (inode_bitmap == sb_block) {
2343                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2344                                  "Inode bitmap for group %u overlaps "
2345                                  "superblock", i);
2346                 }
2347                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2348                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2349                                "Inode bitmap for group %u not in group "
2350                                "(block %llu)!", i, inode_bitmap);
2351                         return 0;
2352                 }
2353                 inode_table = ext4_inode_table(sb, gdp);
2354                 if (inode_table == sb_block) {
2355                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2356                                  "Inode table for group %u overlaps "
2357                                  "superblock", i);
2358                 }
2359                 if (inode_table < first_block ||
2360                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2361                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2362                                "Inode table for group %u not in group "
2363                                "(block %llu)!", i, inode_table);
2364                         return 0;
2365                 }
2366                 ext4_lock_group(sb, i);
2367                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2368                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2369                                  "Checksum for group %u failed (%u!=%u)",
2370                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2371                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2372                         if (!sb_rdonly(sb)) {
2373                                 ext4_unlock_group(sb, i);
2374                                 return 0;
2375                         }
2376                 }
2377                 ext4_unlock_group(sb, i);
2378                 if (!flexbg_flag)
2379                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2380         }
2381         if (NULL != first_not_zeroed)
2382                 *first_not_zeroed = grp;
2383         return 1;
2384 }
2385
2386 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2387  * the superblock) which were deleted from all directories, but held open by
2388  * a process at the time of a crash.  We walk the list and try to delete these
2389  * inodes at recovery time (only with a read-write filesystem).
2390  *
2391  * In order to keep the orphan inode chain consistent during traversal (in
2392  * case of crash during recovery), we link each inode into the superblock
2393  * orphan list_head and handle it the same way as an inode deletion during
2394  * normal operation (which journals the operations for us).
2395  *
2396  * We only do an iget() and an iput() on each inode, which is very safe if we
2397  * accidentally point at an in-use or already deleted inode.  The worst that
2398  * can happen in this case is that we get a "bit already cleared" message from
2399  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2400  * e2fsck was run on this filesystem, and it must have already done the orphan
2401  * inode cleanup for us, so we can safely abort without any further action.
2402  */
2403 static void ext4_orphan_cleanup(struct super_block *sb,
2404                                 struct ext4_super_block *es)
2405 {
2406         unsigned int s_flags = sb->s_flags;
2407         int ret, nr_orphans = 0, nr_truncates = 0;
2408 #ifdef CONFIG_QUOTA
2409         int quota_update = 0;
2410         int i;
2411 #endif
2412         if (!es->s_last_orphan) {
2413                 jbd_debug(4, "no orphan inodes to clean up\n");
2414                 return;
2415         }
2416
2417         if (bdev_read_only(sb->s_bdev)) {
2418                 ext4_msg(sb, KERN_ERR, "write access "
2419                         "unavailable, skipping orphan cleanup");
2420                 return;
2421         }
2422
2423         /* Check if feature set would not allow a r/w mount */
2424         if (!ext4_feature_set_ok(sb, 0)) {
2425                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2426                          "unknown ROCOMPAT features");
2427                 return;
2428         }
2429
2430         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2431                 /* don't clear list on RO mount w/ errors */
2432                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2433                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2434                                   "clearing orphan list.\n");
2435                         es->s_last_orphan = 0;
2436                 }
2437                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2438                 return;
2439         }
2440
2441         if (s_flags & SB_RDONLY) {
2442                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2443                 sb->s_flags &= ~SB_RDONLY;
2444         }
2445 #ifdef CONFIG_QUOTA
2446         /* Needed for iput() to work correctly and not trash data */
2447         sb->s_flags |= SB_ACTIVE;
2448
2449         /*
2450          * Turn on quotas which were not enabled for read-only mounts if
2451          * filesystem has quota feature, so that they are updated correctly.
2452          */
2453         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2454                 int ret = ext4_enable_quotas(sb);
2455
2456                 if (!ret)
2457                         quota_update = 1;
2458                 else
2459                         ext4_msg(sb, KERN_ERR,
2460                                 "Cannot turn on quotas: error %d", ret);
2461         }
2462
2463         /* Turn on journaled quotas used for old sytle */
2464         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2465                 if (EXT4_SB(sb)->s_qf_names[i]) {
2466                         int ret = ext4_quota_on_mount(sb, i);
2467
2468                         if (!ret)
2469                                 quota_update = 1;
2470                         else
2471                                 ext4_msg(sb, KERN_ERR,
2472                                         "Cannot turn on journaled "
2473                                         "quota: type %d: error %d", i, ret);
2474                 }
2475         }
2476 #endif
2477
2478         while (es->s_last_orphan) {
2479                 struct inode *inode;
2480
2481                 /*
2482                  * We may have encountered an error during cleanup; if
2483                  * so, skip the rest.
2484                  */
2485                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2486                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2487                         es->s_last_orphan = 0;
2488                         break;
2489                 }
2490
2491                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2492                 if (IS_ERR(inode)) {
2493                         es->s_last_orphan = 0;
2494                         break;
2495                 }
2496
2497                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2498                 dquot_initialize(inode);
2499                 if (inode->i_nlink) {
2500                         if (test_opt(sb, DEBUG))
2501                                 ext4_msg(sb, KERN_DEBUG,
2502                                         "%s: truncating inode %lu to %lld bytes",
2503                                         __func__, inode->i_ino, inode->i_size);
2504                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2505                                   inode->i_ino, inode->i_size);
2506                         inode_lock(inode);
2507                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2508                         ret = ext4_truncate(inode);
2509                         if (ret)
2510                                 ext4_std_error(inode->i_sb, ret);
2511                         inode_unlock(inode);
2512                         nr_truncates++;
2513                 } else {
2514                         if (test_opt(sb, DEBUG))
2515                                 ext4_msg(sb, KERN_DEBUG,
2516                                         "%s: deleting unreferenced inode %lu",
2517                                         __func__, inode->i_ino);
2518                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2519                                   inode->i_ino);
2520                         nr_orphans++;
2521                 }
2522                 iput(inode);  /* The delete magic happens here! */
2523         }
2524
2525 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2526
2527         if (nr_orphans)
2528                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2529                        PLURAL(nr_orphans));
2530         if (nr_truncates)
2531                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2532                        PLURAL(nr_truncates));
2533 #ifdef CONFIG_QUOTA
2534         /* Turn off quotas if they were enabled for orphan cleanup */
2535         if (quota_update) {
2536                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2537                         if (sb_dqopt(sb)->files[i])
2538                                 dquot_quota_off(sb, i);
2539                 }
2540         }
2541 #endif
2542         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2543 }
2544
2545 /*
2546  * Maximal extent format file size.
2547  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2548  * extent format containers, within a sector_t, and within i_blocks
2549  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2550  * so that won't be a limiting factor.
2551  *
2552  * However there is other limiting factor. We do store extents in the form
2553  * of starting block and length, hence the resulting length of the extent
2554  * covering maximum file size must fit into on-disk format containers as
2555  * well. Given that length is always by 1 unit bigger than max unit (because
2556  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2557  *
2558  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2559  */
2560 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2561 {
2562         loff_t res;
2563         loff_t upper_limit = MAX_LFS_FILESIZE;
2564
2565         /* small i_blocks in vfs inode? */
2566         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2567                 /*
2568                  * CONFIG_LBDAF is not enabled implies the inode
2569                  * i_block represent total blocks in 512 bytes
2570                  * 32 == size of vfs inode i_blocks * 8
2571                  */
2572                 upper_limit = (1LL << 32) - 1;
2573
2574                 /* total blocks in file system block size */
2575                 upper_limit >>= (blkbits - 9);
2576                 upper_limit <<= blkbits;
2577         }
2578
2579         /*
2580          * 32-bit extent-start container, ee_block. We lower the maxbytes
2581          * by one fs block, so ee_len can cover the extent of maximum file
2582          * size
2583          */
2584         res = (1LL << 32) - 1;
2585         res <<= blkbits;
2586
2587         /* Sanity check against vm- & vfs- imposed limits */
2588         if (res > upper_limit)
2589                 res = upper_limit;
2590
2591         return res;
2592 }
2593
2594 /*
2595  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2596  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2597  * We need to be 1 filesystem block less than the 2^48 sector limit.
2598  */
2599 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2600 {
2601         loff_t res = EXT4_NDIR_BLOCKS;
2602         int meta_blocks;
2603         loff_t upper_limit;
2604         /* This is calculated to be the largest file size for a dense, block
2605          * mapped file such that the file's total number of 512-byte sectors,
2606          * including data and all indirect blocks, does not exceed (2^48 - 1).
2607          *
2608          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2609          * number of 512-byte sectors of the file.
2610          */
2611
2612         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2613                 /*
2614                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2615                  * the inode i_block field represents total file blocks in
2616                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2617                  */
2618                 upper_limit = (1LL << 32) - 1;
2619
2620                 /* total blocks in file system block size */
2621                 upper_limit >>= (bits - 9);
2622
2623         } else {
2624                 /*
2625                  * We use 48 bit ext4_inode i_blocks
2626                  * With EXT4_HUGE_FILE_FL set the i_blocks
2627                  * represent total number of blocks in
2628                  * file system block size
2629                  */
2630                 upper_limit = (1LL << 48) - 1;
2631
2632         }
2633
2634         /* indirect blocks */
2635         meta_blocks = 1;
2636         /* double indirect blocks */
2637         meta_blocks += 1 + (1LL << (bits-2));
2638         /* tripple indirect blocks */
2639         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2640
2641         upper_limit -= meta_blocks;
2642         upper_limit <<= bits;
2643
2644         res += 1LL << (bits-2);
2645         res += 1LL << (2*(bits-2));
2646         res += 1LL << (3*(bits-2));
2647         res <<= bits;
2648         if (res > upper_limit)
2649                 res = upper_limit;
2650
2651         if (res > MAX_LFS_FILESIZE)
2652                 res = MAX_LFS_FILESIZE;
2653
2654         return res;
2655 }
2656
2657 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2658                                    ext4_fsblk_t logical_sb_block, int nr)
2659 {
2660         struct ext4_sb_info *sbi = EXT4_SB(sb);
2661         ext4_group_t bg, first_meta_bg;
2662         int has_super = 0;
2663
2664         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2665
2666         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2667                 return logical_sb_block + nr + 1;
2668         bg = sbi->s_desc_per_block * nr;
2669         if (ext4_bg_has_super(sb, bg))
2670                 has_super = 1;
2671
2672         /*
2673          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2674          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2675          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2676          * compensate.
2677          */
2678         if (sb->s_blocksize == 1024 && nr == 0 &&
2679             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2680                 has_super++;
2681
2682         return (has_super + ext4_group_first_block_no(sb, bg));
2683 }
2684
2685 /**
2686  * ext4_get_stripe_size: Get the stripe size.
2687  * @sbi: In memory super block info
2688  *
2689  * If we have specified it via mount option, then
2690  * use the mount option value. If the value specified at mount time is
2691  * greater than the blocks per group use the super block value.
2692  * If the super block value is greater than blocks per group return 0.
2693  * Allocator needs it be less than blocks per group.
2694  *
2695  */
2696 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2697 {
2698         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2699         unsigned long stripe_width =
2700                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2701         int ret;
2702
2703         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2704                 ret = sbi->s_stripe;
2705         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2706                 ret = stripe_width;
2707         else if (stride && stride <= sbi->s_blocks_per_group)
2708                 ret = stride;
2709         else
2710                 ret = 0;
2711
2712         /*
2713          * If the stripe width is 1, this makes no sense and
2714          * we set it to 0 to turn off stripe handling code.
2715          */
2716         if (ret <= 1)
2717                 ret = 0;
2718
2719         return ret;
2720 }
2721
2722 /*
2723  * Check whether this filesystem can be mounted based on
2724  * the features present and the RDONLY/RDWR mount requested.
2725  * Returns 1 if this filesystem can be mounted as requested,
2726  * 0 if it cannot be.
2727  */
2728 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2729 {
2730         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2731                 ext4_msg(sb, KERN_ERR,
2732                         "Couldn't mount because of "
2733                         "unsupported optional features (%x)",
2734                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2735                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2736                 return 0;
2737         }
2738
2739         if (readonly)
2740                 return 1;
2741
2742         if (ext4_has_feature_readonly(sb)) {
2743                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2744                 sb->s_flags |= SB_RDONLY;
2745                 return 1;
2746         }
2747
2748         /* Check that feature set is OK for a read-write mount */
2749         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2750                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2751                          "unsupported optional features (%x)",
2752                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2753                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2754                 return 0;
2755         }
2756         /*
2757          * Large file size enabled file system can only be mounted
2758          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2759          */
2760         if (ext4_has_feature_huge_file(sb)) {
2761                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2762                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2763                                  "cannot be mounted RDWR without "
2764                                  "CONFIG_LBDAF");
2765                         return 0;
2766                 }
2767         }
2768         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2769                 ext4_msg(sb, KERN_ERR,
2770                          "Can't support bigalloc feature without "
2771                          "extents feature\n");
2772                 return 0;
2773         }
2774
2775 #ifndef CONFIG_QUOTA
2776         if (ext4_has_feature_quota(sb) && !readonly) {
2777                 ext4_msg(sb, KERN_ERR,
2778                          "Filesystem with quota feature cannot be mounted RDWR "
2779                          "without CONFIG_QUOTA");
2780                 return 0;
2781         }
2782         if (ext4_has_feature_project(sb) && !readonly) {
2783                 ext4_msg(sb, KERN_ERR,
2784                          "Filesystem with project quota feature cannot be mounted RDWR "
2785                          "without CONFIG_QUOTA");
2786                 return 0;
2787         }
2788 #endif  /* CONFIG_QUOTA */
2789         return 1;
2790 }
2791
2792 /*
2793  * This function is called once a day if we have errors logged
2794  * on the file system
2795  */
2796 static void print_daily_error_info(struct timer_list *t)
2797 {
2798         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2799         struct super_block *sb = sbi->s_sb;
2800         struct ext4_super_block *es = sbi->s_es;
2801
2802         if (es->s_error_count)
2803                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2804                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2805                          le32_to_cpu(es->s_error_count));
2806         if (es->s_first_error_time) {
2807                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2808                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2809                        (int) sizeof(es->s_first_error_func),
2810                        es->s_first_error_func,
2811                        le32_to_cpu(es->s_first_error_line));
2812                 if (es->s_first_error_ino)
2813                         printk(KERN_CONT ": inode %u",
2814                                le32_to_cpu(es->s_first_error_ino));
2815                 if (es->s_first_error_block)
2816                         printk(KERN_CONT ": block %llu", (unsigned long long)
2817                                le64_to_cpu(es->s_first_error_block));
2818                 printk(KERN_CONT "\n");
2819         }
2820         if (es->s_last_error_time) {
2821                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2822                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2823                        (int) sizeof(es->s_last_error_func),
2824                        es->s_last_error_func,
2825                        le32_to_cpu(es->s_last_error_line));
2826                 if (es->s_last_error_ino)
2827                         printk(KERN_CONT ": inode %u",
2828                                le32_to_cpu(es->s_last_error_ino));
2829                 if (es->s_last_error_block)
2830                         printk(KERN_CONT ": block %llu", (unsigned long long)
2831                                le64_to_cpu(es->s_last_error_block));
2832                 printk(KERN_CONT "\n");
2833         }
2834         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2835 }
2836
2837 /* Find next suitable group and run ext4_init_inode_table */
2838 static int ext4_run_li_request(struct ext4_li_request *elr)
2839 {
2840         struct ext4_group_desc *gdp = NULL;
2841         ext4_group_t group, ngroups;
2842         struct super_block *sb;
2843         unsigned long timeout = 0;
2844         int ret = 0;
2845
2846         sb = elr->lr_super;
2847         ngroups = EXT4_SB(sb)->s_groups_count;
2848
2849         for (group = elr->lr_next_group; group < ngroups; group++) {
2850                 gdp = ext4_get_group_desc(sb, group, NULL);
2851                 if (!gdp) {
2852                         ret = 1;
2853                         break;
2854                 }
2855
2856                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2857                         break;
2858         }
2859
2860         if (group >= ngroups)
2861                 ret = 1;
2862
2863         if (!ret) {
2864                 timeout = jiffies;
2865                 ret = ext4_init_inode_table(sb, group,
2866                                             elr->lr_timeout ? 0 : 1);
2867                 if (elr->lr_timeout == 0) {
2868                         timeout = (jiffies - timeout) *
2869                                   elr->lr_sbi->s_li_wait_mult;
2870                         elr->lr_timeout = timeout;
2871                 }
2872                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2873                 elr->lr_next_group = group + 1;
2874         }
2875         return ret;
2876 }
2877
2878 /*
2879  * Remove lr_request from the list_request and free the
2880  * request structure. Should be called with li_list_mtx held
2881  */
2882 static void ext4_remove_li_request(struct ext4_li_request *elr)
2883 {
2884         struct ext4_sb_info *sbi;
2885
2886         if (!elr)
2887                 return;
2888
2889         sbi = elr->lr_sbi;
2890
2891         list_del(&elr->lr_request);
2892         sbi->s_li_request = NULL;
2893         kfree(elr);
2894 }
2895
2896 static void ext4_unregister_li_request(struct super_block *sb)
2897 {
2898         mutex_lock(&ext4_li_mtx);
2899         if (!ext4_li_info) {
2900                 mutex_unlock(&ext4_li_mtx);
2901                 return;
2902         }
2903
2904         mutex_lock(&ext4_li_info->li_list_mtx);
2905         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2906         mutex_unlock(&ext4_li_info->li_list_mtx);
2907         mutex_unlock(&ext4_li_mtx);
2908 }
2909
2910 static struct task_struct *ext4_lazyinit_task;
2911
2912 /*
2913  * This is the function where ext4lazyinit thread lives. It walks
2914  * through the request list searching for next scheduled filesystem.
2915  * When such a fs is found, run the lazy initialization request
2916  * (ext4_rn_li_request) and keep track of the time spend in this
2917  * function. Based on that time we compute next schedule time of
2918  * the request. When walking through the list is complete, compute
2919  * next waking time and put itself into sleep.
2920  */
2921 static int ext4_lazyinit_thread(void *arg)
2922 {
2923         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2924         struct list_head *pos, *n;
2925         struct ext4_li_request *elr;
2926         unsigned long next_wakeup, cur;
2927
2928         BUG_ON(NULL == eli);
2929
2930 cont_thread:
2931         while (true) {
2932                 next_wakeup = MAX_JIFFY_OFFSET;
2933
2934                 mutex_lock(&eli->li_list_mtx);
2935                 if (list_empty(&eli->li_request_list)) {
2936                         mutex_unlock(&eli->li_list_mtx);
2937                         goto exit_thread;
2938                 }
2939                 list_for_each_safe(pos, n, &eli->li_request_list) {
2940                         int err = 0;
2941                         int progress = 0;
2942                         elr = list_entry(pos, struct ext4_li_request,
2943                                          lr_request);
2944
2945                         if (time_before(jiffies, elr->lr_next_sched)) {
2946                                 if (time_before(elr->lr_next_sched, next_wakeup))
2947                                         next_wakeup = elr->lr_next_sched;
2948                                 continue;
2949                         }
2950                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2951                                 if (sb_start_write_trylock(elr->lr_super)) {
2952                                         progress = 1;
2953                                         /*
2954                                          * We hold sb->s_umount, sb can not
2955                                          * be removed from the list, it is
2956                                          * now safe to drop li_list_mtx
2957                                          */
2958                                         mutex_unlock(&eli->li_list_mtx);
2959                                         err = ext4_run_li_request(elr);
2960                                         sb_end_write(elr->lr_super);
2961                                         mutex_lock(&eli->li_list_mtx);
2962                                         n = pos->next;
2963                                 }
2964                                 up_read((&elr->lr_super->s_umount));
2965                         }
2966                         /* error, remove the lazy_init job */
2967                         if (err) {
2968                                 ext4_remove_li_request(elr);
2969                                 continue;
2970                         }
2971                         if (!progress) {
2972                                 elr->lr_next_sched = jiffies +
2973                                         (prandom_u32()
2974                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2975                         }
2976                         if (time_before(elr->lr_next_sched, next_wakeup))
2977                                 next_wakeup = elr->lr_next_sched;
2978                 }
2979                 mutex_unlock(&eli->li_list_mtx);
2980
2981                 try_to_freeze();
2982
2983                 cur = jiffies;
2984                 if ((time_after_eq(cur, next_wakeup)) ||
2985                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2986                         cond_resched();
2987                         continue;
2988                 }
2989
2990                 schedule_timeout_interruptible(next_wakeup - cur);
2991
2992                 if (kthread_should_stop()) {
2993                         ext4_clear_request_list();
2994                         goto exit_thread;
2995                 }
2996         }
2997
2998 exit_thread:
2999         /*
3000          * It looks like the request list is empty, but we need
3001          * to check it under the li_list_mtx lock, to prevent any
3002          * additions into it, and of course we should lock ext4_li_mtx
3003          * to atomically free the list and ext4_li_info, because at
3004          * this point another ext4 filesystem could be registering
3005          * new one.
3006          */
3007         mutex_lock(&ext4_li_mtx);
3008         mutex_lock(&eli->li_list_mtx);
3009         if (!list_empty(&eli->li_request_list)) {
3010                 mutex_unlock(&eli->li_list_mtx);
3011                 mutex_unlock(&ext4_li_mtx);
3012                 goto cont_thread;
3013         }
3014         mutex_unlock(&eli->li_list_mtx);
3015         kfree(ext4_li_info);
3016         ext4_li_info = NULL;
3017         mutex_unlock(&ext4_li_mtx);
3018
3019         return 0;
3020 }
3021
3022 static void ext4_clear_request_list(void)
3023 {
3024         struct list_head *pos, *n;
3025         struct ext4_li_request *elr;
3026
3027         mutex_lock(&ext4_li_info->li_list_mtx);
3028         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3029                 elr = list_entry(pos, struct ext4_li_request,
3030                                  lr_request);
3031                 ext4_remove_li_request(elr);
3032         }
3033         mutex_unlock(&ext4_li_info->li_list_mtx);
3034 }
3035
3036 static int ext4_run_lazyinit_thread(void)
3037 {
3038         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3039                                          ext4_li_info, "ext4lazyinit");
3040         if (IS_ERR(ext4_lazyinit_task)) {
3041                 int err = PTR_ERR(ext4_lazyinit_task);
3042                 ext4_clear_request_list();
3043                 kfree(ext4_li_info);
3044                 ext4_li_info = NULL;
3045                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3046                                  "initialization thread\n",
3047                                  err);
3048                 return err;
3049         }
3050         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3051         return 0;
3052 }
3053
3054 /*
3055  * Check whether it make sense to run itable init. thread or not.
3056  * If there is at least one uninitialized inode table, return
3057  * corresponding group number, else the loop goes through all
3058  * groups and return total number of groups.
3059  */
3060 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3061 {
3062         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3063         struct ext4_group_desc *gdp = NULL;
3064
3065         for (group = 0; group < ngroups; group++) {
3066                 gdp = ext4_get_group_desc(sb, group, NULL);
3067                 if (!gdp)
3068                         continue;
3069
3070                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3071                         break;
3072         }
3073
3074         return group;
3075 }
3076
3077 static int ext4_li_info_new(void)
3078 {
3079         struct ext4_lazy_init *eli = NULL;
3080
3081         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3082         if (!eli)
3083                 return -ENOMEM;
3084
3085         INIT_LIST_HEAD(&eli->li_request_list);
3086         mutex_init(&eli->li_list_mtx);
3087
3088         eli->li_state |= EXT4_LAZYINIT_QUIT;
3089
3090         ext4_li_info = eli;
3091
3092         return 0;
3093 }
3094
3095 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3096                                             ext4_group_t start)
3097 {
3098         struct ext4_sb_info *sbi = EXT4_SB(sb);
3099         struct ext4_li_request *elr;
3100
3101         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3102         if (!elr)
3103                 return NULL;
3104
3105         elr->lr_super = sb;
3106         elr->lr_sbi = sbi;
3107         elr->lr_next_group = start;
3108
3109         /*
3110          * Randomize first schedule time of the request to
3111          * spread the inode table initialization requests
3112          * better.
3113          */
3114         elr->lr_next_sched = jiffies + (prandom_u32() %
3115                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3116         return elr;
3117 }
3118
3119 int ext4_register_li_request(struct super_block *sb,
3120                              ext4_group_t first_not_zeroed)
3121 {
3122         struct ext4_sb_info *sbi = EXT4_SB(sb);
3123         struct ext4_li_request *elr = NULL;
3124         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3125         int ret = 0;
3126
3127         mutex_lock(&ext4_li_mtx);
3128         if (sbi->s_li_request != NULL) {
3129                 /*
3130                  * Reset timeout so it can be computed again, because
3131                  * s_li_wait_mult might have changed.
3132                  */
3133                 sbi->s_li_request->lr_timeout = 0;
3134                 goto out;
3135         }
3136
3137         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3138             !test_opt(sb, INIT_INODE_TABLE))
3139                 goto out;
3140
3141         elr = ext4_li_request_new(sb, first_not_zeroed);
3142         if (!elr) {
3143                 ret = -ENOMEM;
3144                 goto out;
3145         }
3146
3147         if (NULL == ext4_li_info) {
3148                 ret = ext4_li_info_new();
3149                 if (ret)
3150                         goto out;
3151         }
3152
3153         mutex_lock(&ext4_li_info->li_list_mtx);
3154         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3155         mutex_unlock(&ext4_li_info->li_list_mtx);
3156
3157         sbi->s_li_request = elr;
3158         /*
3159          * set elr to NULL here since it has been inserted to
3160          * the request_list and the removal and free of it is
3161          * handled by ext4_clear_request_list from now on.
3162          */
3163         elr = NULL;
3164
3165         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3166                 ret = ext4_run_lazyinit_thread();
3167                 if (ret)
3168                         goto out;
3169         }
3170 out:
3171         mutex_unlock(&ext4_li_mtx);
3172         if (ret)
3173                 kfree(elr);
3174         return ret;
3175 }
3176
3177 /*
3178  * We do not need to lock anything since this is called on
3179  * module unload.
3180  */
3181 static void ext4_destroy_lazyinit_thread(void)
3182 {
3183         /*
3184          * If thread exited earlier
3185          * there's nothing to be done.
3186          */
3187         if (!ext4_li_info || !ext4_lazyinit_task)
3188                 return;
3189
3190         kthread_stop(ext4_lazyinit_task);
3191 }
3192
3193 static int set_journal_csum_feature_set(struct super_block *sb)
3194 {
3195         int ret = 1;
3196         int compat, incompat;
3197         struct ext4_sb_info *sbi = EXT4_SB(sb);
3198
3199         if (ext4_has_metadata_csum(sb)) {
3200                 /* journal checksum v3 */
3201                 compat = 0;
3202                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3203         } else {
3204                 /* journal checksum v1 */
3205                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3206                 incompat = 0;
3207         }
3208
3209         jbd2_journal_clear_features(sbi->s_journal,
3210                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3211                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3212                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3213         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3214                 ret = jbd2_journal_set_features(sbi->s_journal,
3215                                 compat, 0,
3216                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3217                                 incompat);
3218         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3219                 ret = jbd2_journal_set_features(sbi->s_journal,
3220                                 compat, 0,
3221                                 incompat);
3222                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3223                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3224         } else {
3225                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3226                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3227         }
3228
3229         return ret;
3230 }
3231
3232 /*
3233  * Note: calculating the overhead so we can be compatible with
3234  * historical BSD practice is quite difficult in the face of
3235  * clusters/bigalloc.  This is because multiple metadata blocks from
3236  * different block group can end up in the same allocation cluster.
3237  * Calculating the exact overhead in the face of clustered allocation
3238  * requires either O(all block bitmaps) in memory or O(number of block
3239  * groups**2) in time.  We will still calculate the superblock for
3240  * older file systems --- and if we come across with a bigalloc file
3241  * system with zero in s_overhead_clusters the estimate will be close to
3242  * correct especially for very large cluster sizes --- but for newer
3243  * file systems, it's better to calculate this figure once at mkfs
3244  * time, and store it in the superblock.  If the superblock value is
3245  * present (even for non-bigalloc file systems), we will use it.
3246  */
3247 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3248                           char *buf)
3249 {
3250         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3251         struct ext4_group_desc  *gdp;
3252         ext4_fsblk_t            first_block, last_block, b;
3253         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3254         int                     s, j, count = 0;
3255
3256         if (!ext4_has_feature_bigalloc(sb))
3257                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3258                         sbi->s_itb_per_group + 2);
3259
3260         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3261                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3262         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3263         for (i = 0; i < ngroups; i++) {
3264                 gdp = ext4_get_group_desc(sb, i, NULL);
3265                 b = ext4_block_bitmap(sb, gdp);
3266                 if (b >= first_block && b <= last_block) {
3267                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3268                         count++;
3269                 }
3270                 b = ext4_inode_bitmap(sb, gdp);
3271                 if (b >= first_block && b <= last_block) {
3272                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3273                         count++;
3274                 }
3275                 b = ext4_inode_table(sb, gdp);
3276                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3277                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3278                                 int c = EXT4_B2C(sbi, b - first_block);
3279                                 ext4_set_bit(c, buf);
3280                                 count++;
3281                         }
3282                 if (i != grp)
3283                         continue;
3284                 s = 0;
3285                 if (ext4_bg_has_super(sb, grp)) {
3286                         ext4_set_bit(s++, buf);
3287                         count++;
3288                 }
3289                 j = ext4_bg_num_gdb(sb, grp);
3290                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3291                         ext4_error(sb, "Invalid number of block group "
3292                                    "descriptor blocks: %d", j);
3293                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3294                 }
3295                 count += j;
3296                 for (; j > 0; j--)
3297                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3298         }
3299         if (!count)
3300                 return 0;
3301         return EXT4_CLUSTERS_PER_GROUP(sb) -
3302                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3303 }
3304
3305 /*
3306  * Compute the overhead and stash it in sbi->s_overhead
3307  */
3308 int ext4_calculate_overhead(struct super_block *sb)
3309 {
3310         struct ext4_sb_info *sbi = EXT4_SB(sb);
3311         struct ext4_super_block *es = sbi->s_es;
3312         struct inode *j_inode;
3313         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3314         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3315         ext4_fsblk_t overhead = 0;
3316         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3317
3318         if (!buf)
3319                 return -ENOMEM;
3320
3321         /*
3322          * Compute the overhead (FS structures).  This is constant
3323          * for a given filesystem unless the number of block groups
3324          * changes so we cache the previous value until it does.
3325          */
3326
3327         /*
3328          * All of the blocks before first_data_block are overhead
3329          */
3330         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3331
3332         /*
3333          * Add the overhead found in each block group
3334          */
3335         for (i = 0; i < ngroups; i++) {
3336                 int blks;
3337
3338                 blks = count_overhead(sb, i, buf);
3339                 overhead += blks;
3340                 if (blks)
3341                         memset(buf, 0, PAGE_SIZE);
3342                 cond_resched();
3343         }
3344
3345         /*
3346          * Add the internal journal blocks whether the journal has been
3347          * loaded or not
3348          */
3349         if (sbi->s_journal && !sbi->journal_bdev)
3350                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3351         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3352                 j_inode = ext4_get_journal_inode(sb, j_inum);
3353                 if (j_inode) {
3354                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3355                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3356                         iput(j_inode);
3357                 } else {
3358                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3359                 }
3360         }
3361         sbi->s_overhead = overhead;
3362         smp_wmb();
3363         free_page((unsigned long) buf);
3364         return 0;
3365 }
3366
3367 static void ext4_set_resv_clusters(struct super_block *sb)
3368 {
3369         ext4_fsblk_t resv_clusters;
3370         struct ext4_sb_info *sbi = EXT4_SB(sb);
3371
3372         /*
3373          * There's no need to reserve anything when we aren't using extents.
3374          * The space estimates are exact, there are no unwritten extents,
3375          * hole punching doesn't need new metadata... This is needed especially
3376          * to keep ext2/3 backward compatibility.
3377          */
3378         if (!ext4_has_feature_extents(sb))
3379                 return;
3380         /*
3381          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3382          * This should cover the situations where we can not afford to run
3383          * out of space like for example punch hole, or converting
3384          * unwritten extents in delalloc path. In most cases such
3385          * allocation would require 1, or 2 blocks, higher numbers are
3386          * very rare.
3387          */
3388         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3389                          sbi->s_cluster_bits);
3390
3391         do_div(resv_clusters, 50);
3392         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3393
3394         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3395 }
3396
3397 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3398 {
3399         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3400         char *orig_data = kstrdup(data, GFP_KERNEL);
3401         struct buffer_head *bh;
3402         struct ext4_super_block *es = NULL;
3403         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3404         ext4_fsblk_t block;
3405         ext4_fsblk_t sb_block = get_sb_block(&data);
3406         ext4_fsblk_t logical_sb_block;
3407         unsigned long offset = 0;
3408         unsigned long journal_devnum = 0;
3409         unsigned long def_mount_opts;
3410         struct inode *root;
3411         const char *descr;
3412         int ret = -ENOMEM;
3413         int blocksize, clustersize;
3414         unsigned int db_count;
3415         unsigned int i;
3416         int needs_recovery, has_huge_files, has_bigalloc;
3417         __u64 blocks_count;
3418         int err = 0;
3419         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3420         ext4_group_t first_not_zeroed;
3421
3422         if ((data && !orig_data) || !sbi)
3423                 goto out_free_base;
3424
3425         sbi->s_daxdev = dax_dev;
3426         sbi->s_blockgroup_lock =
3427                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3428         if (!sbi->s_blockgroup_lock)
3429                 goto out_free_base;
3430
3431         sb->s_fs_info = sbi;
3432         sbi->s_sb = sb;
3433         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3434         sbi->s_sb_block = sb_block;
3435         if (sb->s_bdev->bd_part)
3436                 sbi->s_sectors_written_start =
3437                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3438
3439         /* Cleanup superblock name */
3440         strreplace(sb->s_id, '/', '!');
3441
3442         /* -EINVAL is default */
3443         ret = -EINVAL;
3444         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3445         if (!blocksize) {
3446                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3447                 goto out_fail;
3448         }
3449
3450         /*
3451          * The ext4 superblock will not be buffer aligned for other than 1kB
3452          * block sizes.  We need to calculate the offset from buffer start.
3453          */
3454         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3455                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3456                 offset = do_div(logical_sb_block, blocksize);
3457         } else {
3458                 logical_sb_block = sb_block;
3459         }
3460
3461         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3462                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3463                 goto out_fail;
3464         }
3465         /*
3466          * Note: s_es must be initialized as soon as possible because
3467          *       some ext4 macro-instructions depend on its value
3468          */
3469         es = (struct ext4_super_block *) (bh->b_data + offset);
3470         sbi->s_es = es;
3471         sb->s_magic = le16_to_cpu(es->s_magic);
3472         if (sb->s_magic != EXT4_SUPER_MAGIC)
3473                 goto cantfind_ext4;
3474         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3475
3476         /* Warn if metadata_csum and gdt_csum are both set. */
3477         if (ext4_has_feature_metadata_csum(sb) &&
3478             ext4_has_feature_gdt_csum(sb))
3479                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3480                              "redundant flags; please run fsck.");
3481
3482         /* Check for a known checksum algorithm */
3483         if (!ext4_verify_csum_type(sb, es)) {
3484                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3485                          "unknown checksum algorithm.");
3486                 silent = 1;
3487                 goto cantfind_ext4;
3488         }
3489
3490         /* Load the checksum driver */
3491         if (ext4_has_feature_metadata_csum(sb) ||
3492             ext4_has_feature_ea_inode(sb)) {
3493                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3494                 if (IS_ERR(sbi->s_chksum_driver)) {
3495                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3496                         ret = PTR_ERR(sbi->s_chksum_driver);
3497                         sbi->s_chksum_driver = NULL;
3498                         goto failed_mount;
3499                 }
3500         }
3501
3502         /* Check superblock checksum */
3503         if (!ext4_superblock_csum_verify(sb, es)) {
3504                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3505                          "invalid superblock checksum.  Run e2fsck?");
3506                 silent = 1;
3507                 ret = -EFSBADCRC;
3508                 goto cantfind_ext4;
3509         }
3510
3511         /* Precompute checksum seed for all metadata */
3512         if (ext4_has_feature_csum_seed(sb))
3513                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3514         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3515                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3516                                                sizeof(es->s_uuid));
3517
3518         /* Set defaults before we parse the mount options */
3519         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3520         set_opt(sb, INIT_INODE_TABLE);
3521         if (def_mount_opts & EXT4_DEFM_DEBUG)
3522                 set_opt(sb, DEBUG);
3523         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3524                 set_opt(sb, GRPID);
3525         if (def_mount_opts & EXT4_DEFM_UID16)
3526                 set_opt(sb, NO_UID32);
3527         /* xattr user namespace & acls are now defaulted on */
3528         set_opt(sb, XATTR_USER);
3529 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3530         set_opt(sb, POSIX_ACL);
3531 #endif
3532         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3533         if (ext4_has_metadata_csum(sb))
3534                 set_opt(sb, JOURNAL_CHECKSUM);
3535
3536         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3537                 set_opt(sb, JOURNAL_DATA);
3538         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3539                 set_opt(sb, ORDERED_DATA);
3540         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3541                 set_opt(sb, WRITEBACK_DATA);
3542
3543         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3544                 set_opt(sb, ERRORS_PANIC);
3545         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3546                 set_opt(sb, ERRORS_CONT);
3547         else
3548                 set_opt(sb, ERRORS_RO);
3549         /* block_validity enabled by default; disable with noblock_validity */
3550         set_opt(sb, BLOCK_VALIDITY);
3551         if (def_mount_opts & EXT4_DEFM_DISCARD)
3552                 set_opt(sb, DISCARD);
3553
3554         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3555         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3556         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3557         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3558         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3559
3560         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3561                 set_opt(sb, BARRIER);
3562
3563         /*
3564          * enable delayed allocation by default
3565          * Use -o nodelalloc to turn it off
3566          */
3567         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3568             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3569                 set_opt(sb, DELALLOC);
3570
3571         /*
3572          * set default s_li_wait_mult for lazyinit, for the case there is
3573          * no mount option specified.
3574          */
3575         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3576
3577         if (sbi->s_es->s_mount_opts[0]) {
3578                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3579                                               sizeof(sbi->s_es->s_mount_opts),
3580                                               GFP_KERNEL);
3581                 if (!s_mount_opts)
3582                         goto failed_mount;
3583                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3584                                    &journal_ioprio, 0)) {
3585                         ext4_msg(sb, KERN_WARNING,
3586                                  "failed to parse options in superblock: %s",
3587                                  s_mount_opts);
3588                 }
3589                 kfree(s_mount_opts);
3590         }
3591         sbi->s_def_mount_opt = sbi->s_mount_opt;
3592         if (!parse_options((char *) data, sb, &journal_devnum,
3593                            &journal_ioprio, 0))
3594                 goto failed_mount;
3595
3596         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3597                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3598                             "with data=journal disables delayed "
3599                             "allocation and O_DIRECT support!\n");
3600                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3601                         ext4_msg(sb, KERN_ERR, "can't mount with "
3602                                  "both data=journal and delalloc");
3603                         goto failed_mount;
3604                 }
3605                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3606                         ext4_msg(sb, KERN_ERR, "can't mount with "
3607                                  "both data=journal and dioread_nolock");
3608                         goto failed_mount;
3609                 }
3610                 if (test_opt(sb, DAX)) {
3611                         ext4_msg(sb, KERN_ERR, "can't mount with "
3612                                  "both data=journal and dax");
3613                         goto failed_mount;
3614                 }
3615                 if (ext4_has_feature_encrypt(sb)) {
3616                         ext4_msg(sb, KERN_WARNING,
3617                                  "encrypted files will use data=ordered "
3618                                  "instead of data journaling mode");
3619                 }
3620                 if (test_opt(sb, DELALLOC))
3621                         clear_opt(sb, DELALLOC);
3622         } else {
3623                 sb->s_iflags |= SB_I_CGROUPWB;
3624         }
3625
3626         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3627                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3628
3629         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3630             (ext4_has_compat_features(sb) ||
3631              ext4_has_ro_compat_features(sb) ||
3632              ext4_has_incompat_features(sb)))
3633                 ext4_msg(sb, KERN_WARNING,
3634                        "feature flags set on rev 0 fs, "
3635                        "running e2fsck is recommended");
3636
3637         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3638                 set_opt2(sb, HURD_COMPAT);
3639                 if (ext4_has_feature_64bit(sb)) {
3640                         ext4_msg(sb, KERN_ERR,
3641                                  "The Hurd can't support 64-bit file systems");
3642                         goto failed_mount;
3643                 }
3644
3645                 /*
3646                  * ea_inode feature uses l_i_version field which is not
3647                  * available in HURD_COMPAT mode.
3648                  */
3649                 if (ext4_has_feature_ea_inode(sb)) {
3650                         ext4_msg(sb, KERN_ERR,
3651                                  "ea_inode feature is not supported for Hurd");
3652                         goto failed_mount;
3653                 }
3654         }
3655
3656         if (IS_EXT2_SB(sb)) {
3657                 if (ext2_feature_set_ok(sb))
3658                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3659                                  "using the ext4 subsystem");
3660                 else {
3661                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3662                                  "to feature incompatibilities");
3663                         goto failed_mount;
3664                 }
3665         }
3666
3667         if (IS_EXT3_SB(sb)) {
3668                 if (ext3_feature_set_ok(sb))
3669                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3670                                  "using the ext4 subsystem");
3671                 else {
3672                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3673                                  "to feature incompatibilities");
3674                         goto failed_mount;
3675                 }
3676         }
3677
3678         /*
3679          * Check feature flags regardless of the revision level, since we
3680          * previously didn't change the revision level when setting the flags,
3681          * so there is a chance incompat flags are set on a rev 0 filesystem.
3682          */
3683         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3684                 goto failed_mount;
3685
3686         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3687         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3688             blocksize > EXT4_MAX_BLOCK_SIZE) {
3689                 ext4_msg(sb, KERN_ERR,
3690                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3691                          blocksize, le32_to_cpu(es->s_log_block_size));
3692                 goto failed_mount;
3693         }
3694         if (le32_to_cpu(es->s_log_block_size) >
3695             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3696                 ext4_msg(sb, KERN_ERR,
3697                          "Invalid log block size: %u",
3698                          le32_to_cpu(es->s_log_block_size));
3699                 goto failed_mount;
3700         }
3701
3702         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3703                 ext4_msg(sb, KERN_ERR,
3704                          "Number of reserved GDT blocks insanely large: %d",
3705                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3706                 goto failed_mount;
3707         }
3708
3709         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3710                 if (ext4_has_feature_inline_data(sb)) {
3711                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3712                                         " that may contain inline data");
3713                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3714                 }
3715                 err = bdev_dax_supported(sb, blocksize);
3716                 if (err) {
3717                         ext4_msg(sb, KERN_ERR,
3718                                 "DAX unsupported by block device. Turning off DAX.");
3719                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3720                 }
3721         }
3722
3723         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3724                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3725                          es->s_encryption_level);
3726                 goto failed_mount;
3727         }
3728
3729         if (sb->s_blocksize != blocksize) {
3730                 /* Validate the filesystem blocksize */
3731                 if (!sb_set_blocksize(sb, blocksize)) {
3732                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3733                                         blocksize);
3734                         goto failed_mount;
3735                 }
3736
3737                 brelse(bh);
3738                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3739                 offset = do_div(logical_sb_block, blocksize);
3740                 bh = sb_bread_unmovable(sb, logical_sb_block);
3741                 if (!bh) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                "Can't read superblock on 2nd try");
3744                         goto failed_mount;
3745                 }
3746                 es = (struct ext4_super_block *)(bh->b_data + offset);
3747                 sbi->s_es = es;
3748                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3749                         ext4_msg(sb, KERN_ERR,
3750                                "Magic mismatch, very weird!");
3751                         goto failed_mount;
3752                 }
3753         }
3754
3755         has_huge_files = ext4_has_feature_huge_file(sb);
3756         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3757                                                       has_huge_files);
3758         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3759
3760         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3761                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3762                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3763         } else {
3764                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3765                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3766                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3767                     (!is_power_of_2(sbi->s_inode_size)) ||
3768                     (sbi->s_inode_size > blocksize)) {
3769                         ext4_msg(sb, KERN_ERR,
3770                                "unsupported inode size: %d",
3771                                sbi->s_inode_size);
3772                         goto failed_mount;
3773                 }
3774                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3775                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3776         }
3777
3778         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3779         if (ext4_has_feature_64bit(sb)) {
3780                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3781                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3782                     !is_power_of_2(sbi->s_desc_size)) {
3783                         ext4_msg(sb, KERN_ERR,
3784                                "unsupported descriptor size %lu",
3785                                sbi->s_desc_size);
3786                         goto failed_mount;
3787                 }
3788         } else
3789                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3790
3791         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3792         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3793
3794         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3795         if (sbi->s_inodes_per_block == 0)
3796                 goto cantfind_ext4;
3797         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3798             sbi->s_inodes_per_group > blocksize * 8) {
3799                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3800                          sbi->s_blocks_per_group);
3801                 goto failed_mount;
3802         }
3803         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3804                                         sbi->s_inodes_per_block;
3805         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3806         sbi->s_sbh = bh;
3807         sbi->s_mount_state = le16_to_cpu(es->s_state);
3808         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3809         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3810
3811         for (i = 0; i < 4; i++)
3812                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3813         sbi->s_def_hash_version = es->s_def_hash_version;
3814         if (ext4_has_feature_dir_index(sb)) {
3815                 i = le32_to_cpu(es->s_flags);
3816                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3817                         sbi->s_hash_unsigned = 3;
3818                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3819 #ifdef __CHAR_UNSIGNED__
3820                         if (!sb_rdonly(sb))
3821                                 es->s_flags |=
3822                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3823                         sbi->s_hash_unsigned = 3;
3824 #else
3825                         if (!sb_rdonly(sb))
3826                                 es->s_flags |=
3827                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3828 #endif
3829                 }
3830         }
3831
3832         /* Handle clustersize */
3833         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3834         has_bigalloc = ext4_has_feature_bigalloc(sb);
3835         if (has_bigalloc) {
3836                 if (clustersize < blocksize) {
3837                         ext4_msg(sb, KERN_ERR,
3838                                  "cluster size (%d) smaller than "
3839                                  "block size (%d)", clustersize, blocksize);
3840                         goto failed_mount;
3841                 }
3842                 if (le32_to_cpu(es->s_log_cluster_size) >
3843                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3844                         ext4_msg(sb, KERN_ERR,
3845                                  "Invalid log cluster size: %u",
3846                                  le32_to_cpu(es->s_log_cluster_size));
3847                         goto failed_mount;
3848                 }
3849                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3850                         le32_to_cpu(es->s_log_block_size);
3851                 sbi->s_clusters_per_group =
3852                         le32_to_cpu(es->s_clusters_per_group);
3853                 if (sbi->s_clusters_per_group > blocksize * 8) {
3854                         ext4_msg(sb, KERN_ERR,
3855                                  "#clusters per group too big: %lu",
3856                                  sbi->s_clusters_per_group);
3857                         goto failed_mount;
3858                 }
3859                 if (sbi->s_blocks_per_group !=
3860                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3861                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3862                                  "clusters per group (%lu) inconsistent",
3863                                  sbi->s_blocks_per_group,
3864                                  sbi->s_clusters_per_group);
3865                         goto failed_mount;
3866                 }
3867         } else {
3868                 if (clustersize != blocksize) {
3869                         ext4_warning(sb, "fragment/cluster size (%d) != "
3870                                      "block size (%d)", clustersize,
3871                                      blocksize);
3872                         clustersize = blocksize;
3873                 }
3874                 if (sbi->s_blocks_per_group > blocksize * 8) {
3875                         ext4_msg(sb, KERN_ERR,
3876                                  "#blocks per group too big: %lu",
3877                                  sbi->s_blocks_per_group);
3878                         goto failed_mount;
3879                 }
3880                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3881                 sbi->s_cluster_bits = 0;
3882         }
3883         sbi->s_cluster_ratio = clustersize / blocksize;
3884
3885         /* Do we have standard group size of clustersize * 8 blocks ? */
3886         if (sbi->s_blocks_per_group == clustersize << 3)
3887                 set_opt2(sb, STD_GROUP_SIZE);
3888
3889         /*
3890          * Test whether we have more sectors than will fit in sector_t,
3891          * and whether the max offset is addressable by the page cache.
3892          */
3893         err = generic_check_addressable(sb->s_blocksize_bits,
3894                                         ext4_blocks_count(es));
3895         if (err) {
3896                 ext4_msg(sb, KERN_ERR, "filesystem"
3897                          " too large to mount safely on this system");
3898                 if (sizeof(sector_t) < 8)
3899                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3900                 goto failed_mount;
3901         }
3902
3903         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3904                 goto cantfind_ext4;
3905
3906         /* check blocks count against device size */
3907         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3908         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3909                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3910                        "exceeds size of device (%llu blocks)",
3911                        ext4_blocks_count(es), blocks_count);
3912                 goto failed_mount;
3913         }
3914
3915         /*
3916          * It makes no sense for the first data block to be beyond the end
3917          * of the filesystem.
3918          */
3919         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3920                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3921                          "block %u is beyond end of filesystem (%llu)",
3922                          le32_to_cpu(es->s_first_data_block),
3923                          ext4_blocks_count(es));
3924                 goto failed_mount;
3925         }
3926         blocks_count = (ext4_blocks_count(es) -
3927                         le32_to_cpu(es->s_first_data_block) +
3928                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3929         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3930         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3931                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3932                        "(block count %llu, first data block %u, "
3933                        "blocks per group %lu)", sbi->s_groups_count,
3934                        ext4_blocks_count(es),
3935                        le32_to_cpu(es->s_first_data_block),
3936                        EXT4_BLOCKS_PER_GROUP(sb));
3937                 goto failed_mount;
3938         }
3939         sbi->s_groups_count = blocks_count;
3940         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3941                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3942         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3943                    EXT4_DESC_PER_BLOCK(sb);
3944         if (ext4_has_feature_meta_bg(sb)) {
3945                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3946                         ext4_msg(sb, KERN_WARNING,
3947                                  "first meta block group too large: %u "
3948                                  "(group descriptor block count %u)",
3949                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3950                         goto failed_mount;
3951                 }
3952         }
3953         sbi->s_group_desc = kvmalloc(db_count *
3954                                           sizeof(struct buffer_head *),
3955                                           GFP_KERNEL);
3956         if (sbi->s_group_desc == NULL) {
3957                 ext4_msg(sb, KERN_ERR, "not enough memory");
3958                 ret = -ENOMEM;
3959                 goto failed_mount;
3960         }
3961
3962         bgl_lock_init(sbi->s_blockgroup_lock);
3963
3964         /* Pre-read the descriptors into the buffer cache */
3965         for (i = 0; i < db_count; i++) {
3966                 block = descriptor_loc(sb, logical_sb_block, i);
3967                 sb_breadahead(sb, block);
3968         }
3969
3970         for (i = 0; i < db_count; i++) {
3971                 block = descriptor_loc(sb, logical_sb_block, i);
3972                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3973                 if (!sbi->s_group_desc[i]) {
3974                         ext4_msg(sb, KERN_ERR,
3975                                "can't read group descriptor %d", i);
3976                         db_count = i;
3977                         goto failed_mount2;
3978                 }
3979         }
3980         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3981                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3982                 ret = -EFSCORRUPTED;
3983                 goto failed_mount2;
3984         }
3985
3986         sbi->s_gdb_count = db_count;
3987
3988         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
3989
3990         /* Register extent status tree shrinker */
3991         if (ext4_es_register_shrinker(sbi))
3992                 goto failed_mount3;
3993
3994         sbi->s_stripe = ext4_get_stripe_size(sbi);
3995         sbi->s_extent_max_zeroout_kb = 32;
3996
3997         /*
3998          * set up enough so that it can read an inode
3999          */
4000         sb->s_op = &ext4_sops;
4001         sb->s_export_op = &ext4_export_ops;
4002         sb->s_xattr = ext4_xattr_handlers;
4003 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4004         sb->s_cop = &ext4_cryptops;
4005 #endif
4006 #ifdef CONFIG_QUOTA
4007         sb->dq_op = &ext4_quota_operations;
4008         if (ext4_has_feature_quota(sb))
4009                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4010         else
4011                 sb->s_qcop = &ext4_qctl_operations;
4012         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4013 #endif
4014         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4015
4016         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4017         mutex_init(&sbi->s_orphan_lock);
4018
4019         sb->s_root = NULL;
4020
4021         needs_recovery = (es->s_last_orphan != 0 ||
4022                           ext4_has_feature_journal_needs_recovery(sb));
4023
4024         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4025                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4026                         goto failed_mount3a;
4027
4028         /*
4029          * The first inode we look at is the journal inode.  Don't try
4030          * root first: it may be modified in the journal!
4031          */
4032         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4033                 err = ext4_load_journal(sb, es, journal_devnum);
4034                 if (err)
4035                         goto failed_mount3a;
4036         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4037                    ext4_has_feature_journal_needs_recovery(sb)) {
4038                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4039                        "suppressed and not mounted read-only");
4040                 goto failed_mount_wq;
4041         } else {
4042                 /* Nojournal mode, all journal mount options are illegal */
4043                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4044                         ext4_msg(sb, KERN_ERR, "can't mount with "
4045                                  "journal_checksum, fs mounted w/o journal");
4046                         goto failed_mount_wq;
4047                 }
4048                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4049                         ext4_msg(sb, KERN_ERR, "can't mount with "
4050                                  "journal_async_commit, fs mounted w/o journal");
4051                         goto failed_mount_wq;
4052                 }
4053                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4054                         ext4_msg(sb, KERN_ERR, "can't mount with "
4055                                  "commit=%lu, fs mounted w/o journal",
4056                                  sbi->s_commit_interval / HZ);
4057                         goto failed_mount_wq;
4058                 }
4059                 if (EXT4_MOUNT_DATA_FLAGS &
4060                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4061                         ext4_msg(sb, KERN_ERR, "can't mount with "
4062                                  "data=, fs mounted w/o journal");
4063                         goto failed_mount_wq;
4064                 }
4065                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4066                 clear_opt(sb, JOURNAL_CHECKSUM);
4067                 clear_opt(sb, DATA_FLAGS);
4068                 sbi->s_journal = NULL;
4069                 needs_recovery = 0;
4070                 goto no_journal;
4071         }
4072
4073         if (ext4_has_feature_64bit(sb) &&
4074             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4075                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4076                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4077                 goto failed_mount_wq;
4078         }
4079
4080         if (!set_journal_csum_feature_set(sb)) {
4081                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4082                          "feature set");
4083                 goto failed_mount_wq;
4084         }
4085
4086         /* We have now updated the journal if required, so we can
4087          * validate the data journaling mode. */
4088         switch (test_opt(sb, DATA_FLAGS)) {
4089         case 0:
4090                 /* No mode set, assume a default based on the journal
4091                  * capabilities: ORDERED_DATA if the journal can
4092                  * cope, else JOURNAL_DATA
4093                  */
4094                 if (jbd2_journal_check_available_features
4095                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4096                         set_opt(sb, ORDERED_DATA);
4097                 else
4098                         set_opt(sb, JOURNAL_DATA);
4099                 break;
4100
4101         case EXT4_MOUNT_ORDERED_DATA:
4102         case EXT4_MOUNT_WRITEBACK_DATA:
4103                 if (!jbd2_journal_check_available_features
4104                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4105                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4106                                "requested data journaling mode");
4107                         goto failed_mount_wq;
4108                 }
4109         default:
4110                 break;
4111         }
4112
4113         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4114             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4115                 ext4_msg(sb, KERN_ERR, "can't mount with "
4116                         "journal_async_commit in data=ordered mode");
4117                 goto failed_mount_wq;
4118         }
4119
4120         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4121
4122         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4123
4124 no_journal:
4125         if (!test_opt(sb, NO_MBCACHE)) {
4126                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4127                 if (!sbi->s_ea_block_cache) {
4128                         ext4_msg(sb, KERN_ERR,
4129                                  "Failed to create ea_block_cache");
4130                         goto failed_mount_wq;
4131                 }
4132
4133                 if (ext4_has_feature_ea_inode(sb)) {
4134                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4135                         if (!sbi->s_ea_inode_cache) {
4136                                 ext4_msg(sb, KERN_ERR,
4137                                          "Failed to create ea_inode_cache");
4138                                 goto failed_mount_wq;
4139                         }
4140                 }
4141         }
4142
4143         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4144             (blocksize != PAGE_SIZE)) {
4145                 ext4_msg(sb, KERN_ERR,
4146                          "Unsupported blocksize for fs encryption");
4147                 goto failed_mount_wq;
4148         }
4149
4150         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4151             !ext4_has_feature_encrypt(sb)) {
4152                 ext4_set_feature_encrypt(sb);
4153                 ext4_commit_super(sb, 1);
4154         }
4155
4156         /*
4157          * Get the # of file system overhead blocks from the
4158          * superblock if present.
4159          */
4160         if (es->s_overhead_clusters)
4161                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4162         else {
4163                 err = ext4_calculate_overhead(sb);
4164                 if (err)
4165                         goto failed_mount_wq;
4166         }
4167
4168         /*
4169          * The maximum number of concurrent works can be high and
4170          * concurrency isn't really necessary.  Limit it to 1.
4171          */
4172         EXT4_SB(sb)->rsv_conversion_wq =
4173                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4174         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4175                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4176                 ret = -ENOMEM;
4177                 goto failed_mount4;
4178         }
4179
4180         /*
4181          * The jbd2_journal_load will have done any necessary log recovery,
4182          * so we can safely mount the rest of the filesystem now.
4183          */
4184
4185         root = ext4_iget(sb, EXT4_ROOT_INO);
4186         if (IS_ERR(root)) {
4187                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4188                 ret = PTR_ERR(root);
4189                 root = NULL;
4190                 goto failed_mount4;
4191         }
4192         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4193                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4194                 iput(root);
4195                 goto failed_mount4;
4196         }
4197         sb->s_root = d_make_root(root);
4198         if (!sb->s_root) {
4199                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4200                 ret = -ENOMEM;
4201                 goto failed_mount4;
4202         }
4203
4204         if (ext4_setup_super(sb, es, sb_rdonly(sb)))
4205                 sb->s_flags |= SB_RDONLY;
4206
4207         /* determine the minimum size of new large inodes, if present */
4208         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4209             sbi->s_want_extra_isize == 0) {
4210                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4211                                                      EXT4_GOOD_OLD_INODE_SIZE;
4212                 if (ext4_has_feature_extra_isize(sb)) {
4213                         if (sbi->s_want_extra_isize <
4214                             le16_to_cpu(es->s_want_extra_isize))
4215                                 sbi->s_want_extra_isize =
4216                                         le16_to_cpu(es->s_want_extra_isize);
4217                         if (sbi->s_want_extra_isize <
4218                             le16_to_cpu(es->s_min_extra_isize))
4219                                 sbi->s_want_extra_isize =
4220                                         le16_to_cpu(es->s_min_extra_isize);
4221                 }
4222         }
4223         /* Check if enough inode space is available */
4224         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4225                                                         sbi->s_inode_size) {
4226                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4227                                                        EXT4_GOOD_OLD_INODE_SIZE;
4228                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4229                          "available");
4230         }
4231
4232         ext4_set_resv_clusters(sb);
4233
4234         err = ext4_setup_system_zone(sb);
4235         if (err) {
4236                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4237                          "zone (%d)", err);
4238                 goto failed_mount4a;
4239         }
4240
4241         ext4_ext_init(sb);
4242         err = ext4_mb_init(sb);
4243         if (err) {
4244                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4245                          err);
4246                 goto failed_mount5;
4247         }
4248
4249         block = ext4_count_free_clusters(sb);
4250         ext4_free_blocks_count_set(sbi->s_es,
4251                                    EXT4_C2B(sbi, block));
4252         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4253                                   GFP_KERNEL);
4254         if (!err) {
4255                 unsigned long freei = ext4_count_free_inodes(sb);
4256                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4257                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4258                                           GFP_KERNEL);
4259         }
4260         if (!err)
4261                 err = percpu_counter_init(&sbi->s_dirs_counter,
4262                                           ext4_count_dirs(sb), GFP_KERNEL);
4263         if (!err)
4264                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4265                                           GFP_KERNEL);
4266         if (!err)
4267                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4268
4269         if (err) {
4270                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4271                 goto failed_mount6;
4272         }
4273
4274         if (ext4_has_feature_flex_bg(sb))
4275                 if (!ext4_fill_flex_info(sb)) {
4276                         ext4_msg(sb, KERN_ERR,
4277                                "unable to initialize "
4278                                "flex_bg meta info!");
4279                         goto failed_mount6;
4280                 }
4281
4282         err = ext4_register_li_request(sb, first_not_zeroed);
4283         if (err)
4284                 goto failed_mount6;
4285
4286         err = ext4_register_sysfs(sb);
4287         if (err)
4288                 goto failed_mount7;
4289
4290 #ifdef CONFIG_QUOTA
4291         /* Enable quota usage during mount. */
4292         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4293                 err = ext4_enable_quotas(sb);
4294                 if (err)
4295                         goto failed_mount8;
4296         }
4297 #endif  /* CONFIG_QUOTA */
4298
4299         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4300         ext4_orphan_cleanup(sb, es);
4301         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4302         if (needs_recovery) {
4303                 ext4_msg(sb, KERN_INFO, "recovery complete");
4304                 ext4_mark_recovery_complete(sb, es);
4305         }
4306         if (EXT4_SB(sb)->s_journal) {
4307                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4308                         descr = " journalled data mode";
4309                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4310                         descr = " ordered data mode";
4311                 else
4312                         descr = " writeback data mode";
4313         } else
4314                 descr = "out journal";
4315
4316         if (test_opt(sb, DISCARD)) {
4317                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4318                 if (!blk_queue_discard(q))
4319                         ext4_msg(sb, KERN_WARNING,
4320                                  "mounting with \"discard\" option, but "
4321                                  "the device does not support discard");
4322         }
4323
4324         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4325                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4326                          "Opts: %.*s%s%s", descr,
4327                          (int) sizeof(sbi->s_es->s_mount_opts),
4328                          sbi->s_es->s_mount_opts,
4329                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4330
4331         if (es->s_error_count)
4332                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4333
4334         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4335         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4336         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4337         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4338
4339         kfree(orig_data);
4340         return 0;
4341
4342 cantfind_ext4:
4343         if (!silent)
4344                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4345         goto failed_mount;
4346
4347 #ifdef CONFIG_QUOTA
4348 failed_mount8:
4349         ext4_unregister_sysfs(sb);
4350 #endif
4351 failed_mount7:
4352         ext4_unregister_li_request(sb);
4353 failed_mount6:
4354         ext4_mb_release(sb);
4355         if (sbi->s_flex_groups)
4356                 kvfree(sbi->s_flex_groups);
4357         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4358         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4359         percpu_counter_destroy(&sbi->s_dirs_counter);
4360         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4361 failed_mount5:
4362         ext4_ext_release(sb);
4363         ext4_release_system_zone(sb);
4364 failed_mount4a:
4365         dput(sb->s_root);
4366         sb->s_root = NULL;
4367 failed_mount4:
4368         ext4_msg(sb, KERN_ERR, "mount failed");
4369         if (EXT4_SB(sb)->rsv_conversion_wq)
4370                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4371 failed_mount_wq:
4372         if (sbi->s_ea_inode_cache) {
4373                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4374                 sbi->s_ea_inode_cache = NULL;
4375         }
4376         if (sbi->s_ea_block_cache) {
4377                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4378                 sbi->s_ea_block_cache = NULL;
4379         }
4380         if (sbi->s_journal) {
4381                 jbd2_journal_destroy(sbi->s_journal);
4382                 sbi->s_journal = NULL;
4383         }
4384 failed_mount3a:
4385         ext4_es_unregister_shrinker(sbi);
4386 failed_mount3:
4387         del_timer_sync(&sbi->s_err_report);
4388         if (sbi->s_mmp_tsk)
4389                 kthread_stop(sbi->s_mmp_tsk);
4390 failed_mount2:
4391         for (i = 0; i < db_count; i++)
4392                 brelse(sbi->s_group_desc[i]);
4393         kvfree(sbi->s_group_desc);
4394 failed_mount:
4395         if (sbi->s_chksum_driver)
4396                 crypto_free_shash(sbi->s_chksum_driver);
4397 #ifdef CONFIG_QUOTA
4398         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4399                 kfree(sbi->s_qf_names[i]);
4400 #endif
4401         ext4_blkdev_remove(sbi);
4402         brelse(bh);
4403 out_fail:
4404         sb->s_fs_info = NULL;
4405         kfree(sbi->s_blockgroup_lock);
4406 out_free_base:
4407         kfree(sbi);
4408         kfree(orig_data);
4409         fs_put_dax(dax_dev);
4410         return err ? err : ret;
4411 }
4412
4413 /*
4414  * Setup any per-fs journal parameters now.  We'll do this both on
4415  * initial mount, once the journal has been initialised but before we've
4416  * done any recovery; and again on any subsequent remount.
4417  */
4418 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4419 {
4420         struct ext4_sb_info *sbi = EXT4_SB(sb);
4421
4422         journal->j_commit_interval = sbi->s_commit_interval;
4423         journal->j_min_batch_time = sbi->s_min_batch_time;
4424         journal->j_max_batch_time = sbi->s_max_batch_time;
4425
4426         write_lock(&journal->j_state_lock);
4427         if (test_opt(sb, BARRIER))
4428                 journal->j_flags |= JBD2_BARRIER;
4429         else
4430                 journal->j_flags &= ~JBD2_BARRIER;
4431         if (test_opt(sb, DATA_ERR_ABORT))
4432                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4433         else
4434                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4435         write_unlock(&journal->j_state_lock);
4436 }
4437
4438 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4439                                              unsigned int journal_inum)
4440 {
4441         struct inode *journal_inode;
4442
4443         /*
4444          * Test for the existence of a valid inode on disk.  Bad things
4445          * happen if we iget() an unused inode, as the subsequent iput()
4446          * will try to delete it.
4447          */
4448         journal_inode = ext4_iget(sb, journal_inum);
4449         if (IS_ERR(journal_inode)) {
4450                 ext4_msg(sb, KERN_ERR, "no journal found");
4451                 return NULL;
4452         }
4453         if (!journal_inode->i_nlink) {
4454                 make_bad_inode(journal_inode);
4455                 iput(journal_inode);
4456                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4457                 return NULL;
4458         }
4459
4460         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4461                   journal_inode, journal_inode->i_size);
4462         if (!S_ISREG(journal_inode->i_mode)) {
4463                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4464                 iput(journal_inode);
4465                 return NULL;
4466         }
4467         return journal_inode;
4468 }
4469
4470 static journal_t *ext4_get_journal(struct super_block *sb,
4471                                    unsigned int journal_inum)
4472 {
4473         struct inode *journal_inode;
4474         journal_t *journal;
4475
4476         BUG_ON(!ext4_has_feature_journal(sb));
4477
4478         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4479         if (!journal_inode)
4480                 return NULL;
4481
4482         journal = jbd2_journal_init_inode(journal_inode);
4483         if (!journal) {
4484                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4485                 iput(journal_inode);
4486                 return NULL;
4487         }
4488         journal->j_private = sb;
4489         ext4_init_journal_params(sb, journal);
4490         return journal;
4491 }
4492
4493 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4494                                        dev_t j_dev)
4495 {
4496         struct buffer_head *bh;
4497         journal_t *journal;
4498         ext4_fsblk_t start;
4499         ext4_fsblk_t len;
4500         int hblock, blocksize;
4501         ext4_fsblk_t sb_block;
4502         unsigned long offset;
4503         struct ext4_super_block *es;
4504         struct block_device *bdev;
4505
4506         BUG_ON(!ext4_has_feature_journal(sb));
4507
4508         bdev = ext4_blkdev_get(j_dev, sb);
4509         if (bdev == NULL)
4510                 return NULL;
4511
4512         blocksize = sb->s_blocksize;
4513         hblock = bdev_logical_block_size(bdev);
4514         if (blocksize < hblock) {
4515                 ext4_msg(sb, KERN_ERR,
4516                         "blocksize too small for journal device");
4517                 goto out_bdev;
4518         }
4519
4520         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4521         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4522         set_blocksize(bdev, blocksize);
4523         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4524                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4525                        "external journal");
4526                 goto out_bdev;
4527         }
4528
4529         es = (struct ext4_super_block *) (bh->b_data + offset);
4530         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4531             !(le32_to_cpu(es->s_feature_incompat) &
4532               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4533                 ext4_msg(sb, KERN_ERR, "external journal has "
4534                                         "bad superblock");
4535                 brelse(bh);
4536                 goto out_bdev;
4537         }
4538
4539         if ((le32_to_cpu(es->s_feature_ro_compat) &
4540              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4541             es->s_checksum != ext4_superblock_csum(sb, es)) {
4542                 ext4_msg(sb, KERN_ERR, "external journal has "
4543                                        "corrupt superblock");
4544                 brelse(bh);
4545                 goto out_bdev;
4546         }
4547
4548         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4549                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4550                 brelse(bh);
4551                 goto out_bdev;
4552         }
4553
4554         len = ext4_blocks_count(es);
4555         start = sb_block + 1;
4556         brelse(bh);     /* we're done with the superblock */
4557
4558         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4559                                         start, len, blocksize);
4560         if (!journal) {
4561                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4562                 goto out_bdev;
4563         }
4564         journal->j_private = sb;
4565         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4566         wait_on_buffer(journal->j_sb_buffer);
4567         if (!buffer_uptodate(journal->j_sb_buffer)) {
4568                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4569                 goto out_journal;
4570         }
4571         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4572                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4573                                         "user (unsupported) - %d",
4574                         be32_to_cpu(journal->j_superblock->s_nr_users));
4575                 goto out_journal;
4576         }
4577         EXT4_SB(sb)->journal_bdev = bdev;
4578         ext4_init_journal_params(sb, journal);
4579         return journal;
4580
4581 out_journal:
4582         jbd2_journal_destroy(journal);
4583 out_bdev:
4584         ext4_blkdev_put(bdev);
4585         return NULL;
4586 }
4587
4588 static int ext4_load_journal(struct super_block *sb,
4589                              struct ext4_super_block *es,
4590                              unsigned long journal_devnum)
4591 {
4592         journal_t *journal;
4593         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4594         dev_t journal_dev;
4595         int err = 0;
4596         int really_read_only;
4597
4598         BUG_ON(!ext4_has_feature_journal(sb));
4599
4600         if (journal_devnum &&
4601             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4602                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4603                         "numbers have changed");
4604                 journal_dev = new_decode_dev(journal_devnum);
4605         } else
4606                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4607
4608         really_read_only = bdev_read_only(sb->s_bdev);
4609
4610         /*
4611          * Are we loading a blank journal or performing recovery after a
4612          * crash?  For recovery, we need to check in advance whether we
4613          * can get read-write access to the device.
4614          */
4615         if (ext4_has_feature_journal_needs_recovery(sb)) {
4616                 if (sb_rdonly(sb)) {
4617                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4618                                         "required on readonly filesystem");
4619                         if (really_read_only) {
4620                                 ext4_msg(sb, KERN_ERR, "write access "
4621                                         "unavailable, cannot proceed "
4622                                         "(try mounting with noload)");
4623                                 return -EROFS;
4624                         }
4625                         ext4_msg(sb, KERN_INFO, "write access will "
4626                                "be enabled during recovery");
4627                 }
4628         }
4629
4630         if (journal_inum && journal_dev) {
4631                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4632                        "and inode journals!");
4633                 return -EINVAL;
4634         }
4635
4636         if (journal_inum) {
4637                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4638                         return -EINVAL;
4639         } else {
4640                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4641                         return -EINVAL;
4642         }
4643
4644         if (!(journal->j_flags & JBD2_BARRIER))
4645                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4646
4647         if (!ext4_has_feature_journal_needs_recovery(sb))
4648                 err = jbd2_journal_wipe(journal, !really_read_only);
4649         if (!err) {
4650                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4651                 if (save)
4652                         memcpy(save, ((char *) es) +
4653                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4654                 err = jbd2_journal_load(journal);
4655                 if (save)
4656                         memcpy(((char *) es) + EXT4_S_ERR_START,
4657                                save, EXT4_S_ERR_LEN);
4658                 kfree(save);
4659         }
4660
4661         if (err) {
4662                 ext4_msg(sb, KERN_ERR, "error loading journal");
4663                 jbd2_journal_destroy(journal);
4664                 return err;
4665         }
4666
4667         EXT4_SB(sb)->s_journal = journal;
4668         ext4_clear_journal_err(sb, es);
4669
4670         if (!really_read_only && journal_devnum &&
4671             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4672                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4673
4674                 /* Make sure we flush the recovery flag to disk. */
4675                 ext4_commit_super(sb, 1);
4676         }
4677
4678         return 0;
4679 }
4680
4681 static int ext4_commit_super(struct super_block *sb, int sync)
4682 {
4683         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4684         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4685         int error = 0;
4686
4687         if (!sbh || block_device_ejected(sb))
4688                 return error;
4689         /*
4690          * If the file system is mounted read-only, don't update the
4691          * superblock write time.  This avoids updating the superblock
4692          * write time when we are mounting the root file system
4693          * read/only but we need to replay the journal; at that point,
4694          * for people who are east of GMT and who make their clock
4695          * tick in localtime for Windows bug-for-bug compatibility,
4696          * the clock is set in the future, and this will cause e2fsck
4697          * to complain and force a full file system check.
4698          */
4699         if (!(sb->s_flags & SB_RDONLY))
4700                 es->s_wtime = cpu_to_le32(get_seconds());
4701         if (sb->s_bdev->bd_part)
4702                 es->s_kbytes_written =
4703                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4704                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4705                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4706         else
4707                 es->s_kbytes_written =
4708                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4709         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4710                 ext4_free_blocks_count_set(es,
4711                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4712                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4713         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4714                 es->s_free_inodes_count =
4715                         cpu_to_le32(percpu_counter_sum_positive(
4716                                 &EXT4_SB(sb)->s_freeinodes_counter));
4717         BUFFER_TRACE(sbh, "marking dirty");
4718         ext4_superblock_csum_set(sb);
4719         if (sync)
4720                 lock_buffer(sbh);
4721         if (buffer_write_io_error(sbh)) {
4722                 /*
4723                  * Oh, dear.  A previous attempt to write the
4724                  * superblock failed.  This could happen because the
4725                  * USB device was yanked out.  Or it could happen to
4726                  * be a transient write error and maybe the block will
4727                  * be remapped.  Nothing we can do but to retry the
4728                  * write and hope for the best.
4729                  */
4730                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4731                        "superblock detected");
4732                 clear_buffer_write_io_error(sbh);
4733                 set_buffer_uptodate(sbh);
4734         }
4735         mark_buffer_dirty(sbh);
4736         if (sync) {
4737                 unlock_buffer(sbh);
4738                 error = __sync_dirty_buffer(sbh,
4739                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4740                 if (error)
4741                         return error;
4742
4743                 error = buffer_write_io_error(sbh);
4744                 if (error) {
4745                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4746                                "superblock");
4747                         clear_buffer_write_io_error(sbh);
4748                         set_buffer_uptodate(sbh);
4749                 }
4750         }
4751         return error;
4752 }
4753
4754 /*
4755  * Have we just finished recovery?  If so, and if we are mounting (or
4756  * remounting) the filesystem readonly, then we will end up with a
4757  * consistent fs on disk.  Record that fact.
4758  */
4759 static void ext4_mark_recovery_complete(struct super_block *sb,
4760                                         struct ext4_super_block *es)
4761 {
4762         journal_t *journal = EXT4_SB(sb)->s_journal;
4763
4764         if (!ext4_has_feature_journal(sb)) {
4765                 BUG_ON(journal != NULL);
4766                 return;
4767         }
4768         jbd2_journal_lock_updates(journal);
4769         if (jbd2_journal_flush(journal) < 0)
4770                 goto out;
4771
4772         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4773                 ext4_clear_feature_journal_needs_recovery(sb);
4774                 ext4_commit_super(sb, 1);
4775         }
4776
4777 out:
4778         jbd2_journal_unlock_updates(journal);
4779 }
4780
4781 /*
4782  * If we are mounting (or read-write remounting) a filesystem whose journal
4783  * has recorded an error from a previous lifetime, move that error to the
4784  * main filesystem now.
4785  */
4786 static void ext4_clear_journal_err(struct super_block *sb,
4787                                    struct ext4_super_block *es)
4788 {
4789         journal_t *journal;
4790         int j_errno;
4791         const char *errstr;
4792
4793         BUG_ON(!ext4_has_feature_journal(sb));
4794
4795         journal = EXT4_SB(sb)->s_journal;
4796
4797         /*
4798          * Now check for any error status which may have been recorded in the
4799          * journal by a prior ext4_error() or ext4_abort()
4800          */
4801
4802         j_errno = jbd2_journal_errno(journal);
4803         if (j_errno) {
4804                 char nbuf[16];
4805
4806                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4807                 ext4_warning(sb, "Filesystem error recorded "
4808                              "from previous mount: %s", errstr);
4809                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4810
4811                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4812                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4813                 ext4_commit_super(sb, 1);
4814
4815                 jbd2_journal_clear_err(journal);
4816                 jbd2_journal_update_sb_errno(journal);
4817         }
4818 }
4819
4820 /*
4821  * Force the running and committing transactions to commit,
4822  * and wait on the commit.
4823  */
4824 int ext4_force_commit(struct super_block *sb)
4825 {
4826         journal_t *journal;
4827
4828         if (sb_rdonly(sb))
4829                 return 0;
4830
4831         journal = EXT4_SB(sb)->s_journal;
4832         return ext4_journal_force_commit(journal);
4833 }
4834
4835 static int ext4_sync_fs(struct super_block *sb, int wait)
4836 {
4837         int ret = 0;
4838         tid_t target;
4839         bool needs_barrier = false;
4840         struct ext4_sb_info *sbi = EXT4_SB(sb);
4841
4842         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4843                 return 0;
4844
4845         trace_ext4_sync_fs(sb, wait);
4846         flush_workqueue(sbi->rsv_conversion_wq);
4847         /*
4848          * Writeback quota in non-journalled quota case - journalled quota has
4849          * no dirty dquots
4850          */
4851         dquot_writeback_dquots(sb, -1);
4852         /*
4853          * Data writeback is possible w/o journal transaction, so barrier must
4854          * being sent at the end of the function. But we can skip it if
4855          * transaction_commit will do it for us.
4856          */
4857         if (sbi->s_journal) {
4858                 target = jbd2_get_latest_transaction(sbi->s_journal);
4859                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4860                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4861                         needs_barrier = true;
4862
4863                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4864                         if (wait)
4865                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4866                                                            target);
4867                 }
4868         } else if (wait && test_opt(sb, BARRIER))
4869                 needs_barrier = true;
4870         if (needs_barrier) {
4871                 int err;
4872                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4873                 if (!ret)
4874                         ret = err;
4875         }
4876
4877         return ret;
4878 }
4879
4880 /*
4881  * LVM calls this function before a (read-only) snapshot is created.  This
4882  * gives us a chance to flush the journal completely and mark the fs clean.
4883  *
4884  * Note that only this function cannot bring a filesystem to be in a clean
4885  * state independently. It relies on upper layer to stop all data & metadata
4886  * modifications.
4887  */
4888 static int ext4_freeze(struct super_block *sb)
4889 {
4890         int error = 0;
4891         journal_t *journal;
4892
4893         if (sb_rdonly(sb))
4894                 return 0;
4895
4896         journal = EXT4_SB(sb)->s_journal;
4897
4898         if (journal) {
4899                 /* Now we set up the journal barrier. */
4900                 jbd2_journal_lock_updates(journal);
4901
4902                 /*
4903                  * Don't clear the needs_recovery flag if we failed to
4904                  * flush the journal.
4905                  */
4906                 error = jbd2_journal_flush(journal);
4907                 if (error < 0)
4908                         goto out;
4909
4910                 /* Journal blocked and flushed, clear needs_recovery flag. */
4911                 ext4_clear_feature_journal_needs_recovery(sb);
4912         }
4913
4914         error = ext4_commit_super(sb, 1);
4915 out:
4916         if (journal)
4917                 /* we rely on upper layer to stop further updates */
4918                 jbd2_journal_unlock_updates(journal);
4919         return error;
4920 }
4921
4922 /*
4923  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4924  * flag here, even though the filesystem is not technically dirty yet.
4925  */
4926 static int ext4_unfreeze(struct super_block *sb)
4927 {
4928         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
4929                 return 0;
4930
4931         if (EXT4_SB(sb)->s_journal) {
4932                 /* Reset the needs_recovery flag before the fs is unlocked. */
4933                 ext4_set_feature_journal_needs_recovery(sb);
4934         }
4935
4936         ext4_commit_super(sb, 1);
4937         return 0;
4938 }
4939
4940 /*
4941  * Structure to save mount options for ext4_remount's benefit
4942  */
4943 struct ext4_mount_options {
4944         unsigned long s_mount_opt;
4945         unsigned long s_mount_opt2;
4946         kuid_t s_resuid;
4947         kgid_t s_resgid;
4948         unsigned long s_commit_interval;
4949         u32 s_min_batch_time, s_max_batch_time;
4950 #ifdef CONFIG_QUOTA
4951         int s_jquota_fmt;
4952         char *s_qf_names[EXT4_MAXQUOTAS];
4953 #endif
4954 };
4955
4956 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4957 {
4958         struct ext4_super_block *es;
4959         struct ext4_sb_info *sbi = EXT4_SB(sb);
4960         unsigned long old_sb_flags;
4961         struct ext4_mount_options old_opts;
4962         int enable_quota = 0;
4963         ext4_group_t g;
4964         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4965         int err = 0;
4966 #ifdef CONFIG_QUOTA
4967         int i, j;
4968 #endif
4969         char *orig_data = kstrdup(data, GFP_KERNEL);
4970
4971         /* Store the original options */
4972         old_sb_flags = sb->s_flags;
4973         old_opts.s_mount_opt = sbi->s_mount_opt;
4974         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4975         old_opts.s_resuid = sbi->s_resuid;
4976         old_opts.s_resgid = sbi->s_resgid;
4977         old_opts.s_commit_interval = sbi->s_commit_interval;
4978         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4979         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4980 #ifdef CONFIG_QUOTA
4981         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4982         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4983                 if (sbi->s_qf_names[i]) {
4984                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4985                                                          GFP_KERNEL);
4986                         if (!old_opts.s_qf_names[i]) {
4987                                 for (j = 0; j < i; j++)
4988                                         kfree(old_opts.s_qf_names[j]);
4989                                 kfree(orig_data);
4990                                 return -ENOMEM;
4991                         }
4992                 } else
4993                         old_opts.s_qf_names[i] = NULL;
4994 #endif
4995         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4996                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4997
4998         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4999                 err = -EINVAL;
5000                 goto restore_opts;
5001         }
5002
5003         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5004             test_opt(sb, JOURNAL_CHECKSUM)) {
5005                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5006                          "during remount not supported; ignoring");
5007                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5008         }
5009
5010         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5011                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5012                         ext4_msg(sb, KERN_ERR, "can't mount with "
5013                                  "both data=journal and delalloc");
5014                         err = -EINVAL;
5015                         goto restore_opts;
5016                 }
5017                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5018                         ext4_msg(sb, KERN_ERR, "can't mount with "
5019                                  "both data=journal and dioread_nolock");
5020                         err = -EINVAL;
5021                         goto restore_opts;
5022                 }
5023                 if (test_opt(sb, DAX)) {
5024                         ext4_msg(sb, KERN_ERR, "can't mount with "
5025                                  "both data=journal and dax");
5026                         err = -EINVAL;
5027                         goto restore_opts;
5028                 }
5029         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5030                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5031                         ext4_msg(sb, KERN_ERR, "can't mount with "
5032                                 "journal_async_commit in data=ordered mode");
5033                         err = -EINVAL;
5034                         goto restore_opts;
5035                 }
5036         }
5037
5038         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5039                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5040                 err = -EINVAL;
5041                 goto restore_opts;
5042         }
5043
5044         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5045                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5046                         "dax flag with busy inodes while remounting");
5047                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5048         }
5049
5050         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5051                 ext4_abort(sb, "Abort forced by user");
5052
5053         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5054                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5055
5056         es = sbi->s_es;
5057
5058         if (sbi->s_journal) {
5059                 ext4_init_journal_params(sb, sbi->s_journal);
5060                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5061         }
5062
5063         if (*flags & SB_LAZYTIME)
5064                 sb->s_flags |= SB_LAZYTIME;
5065
5066         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5067                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5068                         err = -EROFS;
5069                         goto restore_opts;
5070                 }
5071
5072                 if (*flags & SB_RDONLY) {
5073                         err = sync_filesystem(sb);
5074                         if (err < 0)
5075                                 goto restore_opts;
5076                         err = dquot_suspend(sb, -1);
5077                         if (err < 0)
5078                                 goto restore_opts;
5079
5080                         /*
5081                          * First of all, the unconditional stuff we have to do
5082                          * to disable replay of the journal when we next remount
5083                          */
5084                         sb->s_flags |= SB_RDONLY;
5085
5086                         /*
5087                          * OK, test if we are remounting a valid rw partition
5088                          * readonly, and if so set the rdonly flag and then
5089                          * mark the partition as valid again.
5090                          */
5091                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5092                             (sbi->s_mount_state & EXT4_VALID_FS))
5093                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5094
5095                         if (sbi->s_journal)
5096                                 ext4_mark_recovery_complete(sb, es);
5097                 } else {
5098                         /* Make sure we can mount this feature set readwrite */
5099                         if (ext4_has_feature_readonly(sb) ||
5100                             !ext4_feature_set_ok(sb, 0)) {
5101                                 err = -EROFS;
5102                                 goto restore_opts;
5103                         }
5104                         /*
5105                          * Make sure the group descriptor checksums
5106                          * are sane.  If they aren't, refuse to remount r/w.
5107                          */
5108                         for (g = 0; g < sbi->s_groups_count; g++) {
5109                                 struct ext4_group_desc *gdp =
5110                                         ext4_get_group_desc(sb, g, NULL);
5111
5112                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5113                                         ext4_msg(sb, KERN_ERR,
5114                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5115                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5116                                                le16_to_cpu(gdp->bg_checksum));
5117                                         err = -EFSBADCRC;
5118                                         goto restore_opts;
5119                                 }
5120                         }
5121
5122                         /*
5123                          * If we have an unprocessed orphan list hanging
5124                          * around from a previously readonly bdev mount,
5125                          * require a full umount/remount for now.
5126                          */
5127                         if (es->s_last_orphan) {
5128                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5129                                        "remount RDWR because of unprocessed "
5130                                        "orphan inode list.  Please "
5131                                        "umount/remount instead");
5132                                 err = -EINVAL;
5133                                 goto restore_opts;
5134                         }
5135
5136                         /*
5137                          * Mounting a RDONLY partition read-write, so reread
5138                          * and store the current valid flag.  (It may have
5139                          * been changed by e2fsck since we originally mounted
5140                          * the partition.)
5141                          */
5142                         if (sbi->s_journal)
5143                                 ext4_clear_journal_err(sb, es);
5144                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5145                         if (!ext4_setup_super(sb, es, 0))
5146                                 sb->s_flags &= ~SB_RDONLY;
5147                         if (ext4_has_feature_mmp(sb))
5148                                 if (ext4_multi_mount_protect(sb,
5149                                                 le64_to_cpu(es->s_mmp_block))) {
5150                                         err = -EROFS;
5151                                         goto restore_opts;
5152                                 }
5153                         enable_quota = 1;
5154                 }
5155         }
5156
5157         /*
5158          * Reinitialize lazy itable initialization thread based on
5159          * current settings
5160          */
5161         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5162                 ext4_unregister_li_request(sb);
5163         else {
5164                 ext4_group_t first_not_zeroed;
5165                 first_not_zeroed = ext4_has_uninit_itable(sb);
5166                 ext4_register_li_request(sb, first_not_zeroed);
5167         }
5168
5169         ext4_setup_system_zone(sb);
5170         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY))
5171                 ext4_commit_super(sb, 1);
5172
5173 #ifdef CONFIG_QUOTA
5174         /* Release old quota file names */
5175         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5176                 kfree(old_opts.s_qf_names[i]);
5177         if (enable_quota) {
5178                 if (sb_any_quota_suspended(sb))
5179                         dquot_resume(sb, -1);
5180                 else if (ext4_has_feature_quota(sb)) {
5181                         err = ext4_enable_quotas(sb);
5182                         if (err)
5183                                 goto restore_opts;
5184                 }
5185         }
5186 #endif
5187
5188         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5189         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5190         kfree(orig_data);
5191         return 0;
5192
5193 restore_opts:
5194         sb->s_flags = old_sb_flags;
5195         sbi->s_mount_opt = old_opts.s_mount_opt;
5196         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5197         sbi->s_resuid = old_opts.s_resuid;
5198         sbi->s_resgid = old_opts.s_resgid;
5199         sbi->s_commit_interval = old_opts.s_commit_interval;
5200         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5201         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5202 #ifdef CONFIG_QUOTA
5203         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5204         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5205                 kfree(sbi->s_qf_names[i]);
5206                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5207         }
5208 #endif
5209         kfree(orig_data);
5210         return err;
5211 }
5212
5213 #ifdef CONFIG_QUOTA
5214 static int ext4_statfs_project(struct super_block *sb,
5215                                kprojid_t projid, struct kstatfs *buf)
5216 {
5217         struct kqid qid;
5218         struct dquot *dquot;
5219         u64 limit;
5220         u64 curblock;
5221
5222         qid = make_kqid_projid(projid);
5223         dquot = dqget(sb, qid);
5224         if (IS_ERR(dquot))
5225                 return PTR_ERR(dquot);
5226         spin_lock(&dquot->dq_dqb_lock);
5227
5228         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5229                  dquot->dq_dqb.dqb_bsoftlimit :
5230                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5231         if (limit && buf->f_blocks > limit) {
5232                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5233                 buf->f_blocks = limit;
5234                 buf->f_bfree = buf->f_bavail =
5235                         (buf->f_blocks > curblock) ?
5236                          (buf->f_blocks - curblock) : 0;
5237         }
5238
5239         limit = dquot->dq_dqb.dqb_isoftlimit ?
5240                 dquot->dq_dqb.dqb_isoftlimit :
5241                 dquot->dq_dqb.dqb_ihardlimit;
5242         if (limit && buf->f_files > limit) {
5243                 buf->f_files = limit;
5244                 buf->f_ffree =
5245                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5246                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5247         }
5248
5249         spin_unlock(&dquot->dq_dqb_lock);
5250         dqput(dquot);
5251         return 0;
5252 }
5253 #endif
5254
5255 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5256 {
5257         struct super_block *sb = dentry->d_sb;
5258         struct ext4_sb_info *sbi = EXT4_SB(sb);
5259         struct ext4_super_block *es = sbi->s_es;
5260         ext4_fsblk_t overhead = 0, resv_blocks;
5261         u64 fsid;
5262         s64 bfree;
5263         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5264
5265         if (!test_opt(sb, MINIX_DF))
5266                 overhead = sbi->s_overhead;
5267
5268         buf->f_type = EXT4_SUPER_MAGIC;
5269         buf->f_bsize = sb->s_blocksize;
5270         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5271         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5272                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5273         /* prevent underflow in case that few free space is available */
5274         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5275         buf->f_bavail = buf->f_bfree -
5276                         (ext4_r_blocks_count(es) + resv_blocks);
5277         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5278                 buf->f_bavail = 0;
5279         buf->f_files = le32_to_cpu(es->s_inodes_count);
5280         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5281         buf->f_namelen = EXT4_NAME_LEN;
5282         fsid = le64_to_cpup((void *)es->s_uuid) ^
5283                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5284         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5285         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5286
5287 #ifdef CONFIG_QUOTA
5288         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5289             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5290                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5291 #endif
5292         return 0;
5293 }
5294
5295
5296 #ifdef CONFIG_QUOTA
5297
5298 /*
5299  * Helper functions so that transaction is started before we acquire dqio_sem
5300  * to keep correct lock ordering of transaction > dqio_sem
5301  */
5302 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5303 {
5304         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5305 }
5306
5307 static int ext4_write_dquot(struct dquot *dquot)
5308 {
5309         int ret, err;
5310         handle_t *handle;
5311         struct inode *inode;
5312
5313         inode = dquot_to_inode(dquot);
5314         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5315                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5316         if (IS_ERR(handle))
5317                 return PTR_ERR(handle);
5318         ret = dquot_commit(dquot);
5319         err = ext4_journal_stop(handle);
5320         if (!ret)
5321                 ret = err;
5322         return ret;
5323 }
5324
5325 static int ext4_acquire_dquot(struct dquot *dquot)
5326 {
5327         int ret, err;
5328         handle_t *handle;
5329
5330         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5331                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5332         if (IS_ERR(handle))
5333                 return PTR_ERR(handle);
5334         ret = dquot_acquire(dquot);
5335         err = ext4_journal_stop(handle);
5336         if (!ret)
5337                 ret = err;
5338         return ret;
5339 }
5340
5341 static int ext4_release_dquot(struct dquot *dquot)
5342 {
5343         int ret, err;
5344         handle_t *handle;
5345
5346         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5347                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5348         if (IS_ERR(handle)) {
5349                 /* Release dquot anyway to avoid endless cycle in dqput() */
5350                 dquot_release(dquot);
5351                 return PTR_ERR(handle);
5352         }
5353         ret = dquot_release(dquot);
5354         err = ext4_journal_stop(handle);
5355         if (!ret)
5356                 ret = err;
5357         return ret;
5358 }
5359
5360 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5361 {
5362         struct super_block *sb = dquot->dq_sb;
5363         struct ext4_sb_info *sbi = EXT4_SB(sb);
5364
5365         /* Are we journaling quotas? */
5366         if (ext4_has_feature_quota(sb) ||
5367             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5368                 dquot_mark_dquot_dirty(dquot);
5369                 return ext4_write_dquot(dquot);
5370         } else {
5371                 return dquot_mark_dquot_dirty(dquot);
5372         }
5373 }
5374
5375 static int ext4_write_info(struct super_block *sb, int type)
5376 {
5377         int ret, err;
5378         handle_t *handle;
5379
5380         /* Data block + inode block */
5381         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5382         if (IS_ERR(handle))
5383                 return PTR_ERR(handle);
5384         ret = dquot_commit_info(sb, type);
5385         err = ext4_journal_stop(handle);
5386         if (!ret)
5387                 ret = err;
5388         return ret;
5389 }
5390
5391 /*
5392  * Turn on quotas during mount time - we need to find
5393  * the quota file and such...
5394  */
5395 static int ext4_quota_on_mount(struct super_block *sb, int type)
5396 {
5397         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5398                                         EXT4_SB(sb)->s_jquota_fmt, type);
5399 }
5400
5401 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5402 {
5403         struct ext4_inode_info *ei = EXT4_I(inode);
5404
5405         /* The first argument of lockdep_set_subclass has to be
5406          * *exactly* the same as the argument to init_rwsem() --- in
5407          * this case, in init_once() --- or lockdep gets unhappy
5408          * because the name of the lock is set using the
5409          * stringification of the argument to init_rwsem().
5410          */
5411         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5412         lockdep_set_subclass(&ei->i_data_sem, subclass);
5413 }
5414
5415 /*
5416  * Standard function to be called on quota_on
5417  */
5418 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5419                          const struct path *path)
5420 {
5421         int err;
5422
5423         if (!test_opt(sb, QUOTA))
5424                 return -EINVAL;
5425
5426         /* Quotafile not on the same filesystem? */
5427         if (path->dentry->d_sb != sb)
5428                 return -EXDEV;
5429         /* Journaling quota? */
5430         if (EXT4_SB(sb)->s_qf_names[type]) {
5431                 /* Quotafile not in fs root? */
5432                 if (path->dentry->d_parent != sb->s_root)
5433                         ext4_msg(sb, KERN_WARNING,
5434                                 "Quota file not on filesystem root. "
5435                                 "Journaled quota will not work");
5436                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5437         } else {
5438                 /*
5439                  * Clear the flag just in case mount options changed since
5440                  * last time.
5441                  */
5442                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5443         }
5444
5445         /*
5446          * When we journal data on quota file, we have to flush journal to see
5447          * all updates to the file when we bypass pagecache...
5448          */
5449         if (EXT4_SB(sb)->s_journal &&
5450             ext4_should_journal_data(d_inode(path->dentry))) {
5451                 /*
5452                  * We don't need to lock updates but journal_flush() could
5453                  * otherwise be livelocked...
5454                  */
5455                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5456                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5457                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5458                 if (err)
5459                         return err;
5460         }
5461
5462         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5463         err = dquot_quota_on(sb, type, format_id, path);
5464         if (err) {
5465                 lockdep_set_quota_inode(path->dentry->d_inode,
5466                                              I_DATA_SEM_NORMAL);
5467         } else {
5468                 struct inode *inode = d_inode(path->dentry);
5469                 handle_t *handle;
5470
5471                 /*
5472                  * Set inode flags to prevent userspace from messing with quota
5473                  * files. If this fails, we return success anyway since quotas
5474                  * are already enabled and this is not a hard failure.
5475                  */
5476                 inode_lock(inode);
5477                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5478                 if (IS_ERR(handle))
5479                         goto unlock_inode;
5480                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5481                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5482                                 S_NOATIME | S_IMMUTABLE);
5483                 ext4_mark_inode_dirty(handle, inode);
5484                 ext4_journal_stop(handle);
5485         unlock_inode:
5486                 inode_unlock(inode);
5487         }
5488         return err;
5489 }
5490
5491 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5492                              unsigned int flags)
5493 {
5494         int err;
5495         struct inode *qf_inode;
5496         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5497                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5498                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5499                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5500         };
5501
5502         BUG_ON(!ext4_has_feature_quota(sb));
5503
5504         if (!qf_inums[type])
5505                 return -EPERM;
5506
5507         qf_inode = ext4_iget(sb, qf_inums[type]);
5508         if (IS_ERR(qf_inode)) {
5509                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5510                 return PTR_ERR(qf_inode);
5511         }
5512
5513         /* Don't account quota for quota files to avoid recursion */
5514         qf_inode->i_flags |= S_NOQUOTA;
5515         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5516         err = dquot_enable(qf_inode, type, format_id, flags);
5517         iput(qf_inode);
5518         if (err)
5519                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5520
5521         return err;
5522 }
5523
5524 /* Enable usage tracking for all quota types. */
5525 static int ext4_enable_quotas(struct super_block *sb)
5526 {
5527         int type, err = 0;
5528         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5529                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5530                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5531                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5532         };
5533         bool quota_mopt[EXT4_MAXQUOTAS] = {
5534                 test_opt(sb, USRQUOTA),
5535                 test_opt(sb, GRPQUOTA),
5536                 test_opt(sb, PRJQUOTA),
5537         };
5538
5539         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5540         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5541                 if (qf_inums[type]) {
5542                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5543                                 DQUOT_USAGE_ENABLED |
5544                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5545                         if (err) {
5546                                 for (type--; type >= 0; type--)
5547                                         dquot_quota_off(sb, type);
5548
5549                                 ext4_warning(sb,
5550                                         "Failed to enable quota tracking "
5551                                         "(type=%d, err=%d). Please run "
5552                                         "e2fsck to fix.", type, err);
5553                                 return err;
5554                         }
5555                 }
5556         }
5557         return 0;
5558 }
5559
5560 static int ext4_quota_off(struct super_block *sb, int type)
5561 {
5562         struct inode *inode = sb_dqopt(sb)->files[type];
5563         handle_t *handle;
5564         int err;
5565
5566         /* Force all delayed allocation blocks to be allocated.
5567          * Caller already holds s_umount sem */
5568         if (test_opt(sb, DELALLOC))
5569                 sync_filesystem(sb);
5570
5571         if (!inode || !igrab(inode))
5572                 goto out;
5573
5574         err = dquot_quota_off(sb, type);
5575         if (err || ext4_has_feature_quota(sb))
5576                 goto out_put;
5577
5578         inode_lock(inode);
5579         /*
5580          * Update modification times of quota files when userspace can
5581          * start looking at them. If we fail, we return success anyway since
5582          * this is not a hard failure and quotas are already disabled.
5583          */
5584         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5585         if (IS_ERR(handle))
5586                 goto out_unlock;
5587         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5588         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5589         inode->i_mtime = inode->i_ctime = current_time(inode);
5590         ext4_mark_inode_dirty(handle, inode);
5591         ext4_journal_stop(handle);
5592 out_unlock:
5593         inode_unlock(inode);
5594 out_put:
5595         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5596         iput(inode);
5597         return err;
5598 out:
5599         return dquot_quota_off(sb, type);
5600 }
5601
5602 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5603  * acquiring the locks... As quota files are never truncated and quota code
5604  * itself serializes the operations (and no one else should touch the files)
5605  * we don't have to be afraid of races */
5606 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5607                                size_t len, loff_t off)
5608 {
5609         struct inode *inode = sb_dqopt(sb)->files[type];
5610         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5611         int offset = off & (sb->s_blocksize - 1);
5612         int tocopy;
5613         size_t toread;
5614         struct buffer_head *bh;
5615         loff_t i_size = i_size_read(inode);
5616
5617         if (off > i_size)
5618                 return 0;
5619         if (off+len > i_size)
5620                 len = i_size-off;
5621         toread = len;
5622         while (toread > 0) {
5623                 tocopy = sb->s_blocksize - offset < toread ?
5624                                 sb->s_blocksize - offset : toread;
5625                 bh = ext4_bread(NULL, inode, blk, 0);
5626                 if (IS_ERR(bh))
5627                         return PTR_ERR(bh);
5628                 if (!bh)        /* A hole? */
5629                         memset(data, 0, tocopy);
5630                 else
5631                         memcpy(data, bh->b_data+offset, tocopy);
5632                 brelse(bh);
5633                 offset = 0;
5634                 toread -= tocopy;
5635                 data += tocopy;
5636                 blk++;
5637         }
5638         return len;
5639 }
5640
5641 /* Write to quotafile (we know the transaction is already started and has
5642  * enough credits) */
5643 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5644                                 const char *data, size_t len, loff_t off)
5645 {
5646         struct inode *inode = sb_dqopt(sb)->files[type];
5647         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5648         int err, offset = off & (sb->s_blocksize - 1);
5649         int retries = 0;
5650         struct buffer_head *bh;
5651         handle_t *handle = journal_current_handle();
5652
5653         if (EXT4_SB(sb)->s_journal && !handle) {
5654                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5655                         " cancelled because transaction is not started",
5656                         (unsigned long long)off, (unsigned long long)len);
5657                 return -EIO;
5658         }
5659         /*
5660          * Since we account only one data block in transaction credits,
5661          * then it is impossible to cross a block boundary.
5662          */
5663         if (sb->s_blocksize - offset < len) {
5664                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5665                         " cancelled because not block aligned",
5666                         (unsigned long long)off, (unsigned long long)len);
5667                 return -EIO;
5668         }
5669
5670         do {
5671                 bh = ext4_bread(handle, inode, blk,
5672                                 EXT4_GET_BLOCKS_CREATE |
5673                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5674         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5675                  ext4_should_retry_alloc(inode->i_sb, &retries));
5676         if (IS_ERR(bh))
5677                 return PTR_ERR(bh);
5678         if (!bh)
5679                 goto out;
5680         BUFFER_TRACE(bh, "get write access");
5681         err = ext4_journal_get_write_access(handle, bh);
5682         if (err) {
5683                 brelse(bh);
5684                 return err;
5685         }
5686         lock_buffer(bh);
5687         memcpy(bh->b_data+offset, data, len);
5688         flush_dcache_page(bh->b_page);
5689         unlock_buffer(bh);
5690         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5691         brelse(bh);
5692 out:
5693         if (inode->i_size < off + len) {
5694                 i_size_write(inode, off + len);
5695                 EXT4_I(inode)->i_disksize = inode->i_size;
5696                 ext4_mark_inode_dirty(handle, inode);
5697         }
5698         return len;
5699 }
5700
5701 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5702 {
5703         const struct quota_format_ops   *ops;
5704
5705         if (!sb_has_quota_loaded(sb, qid->type))
5706                 return -ESRCH;
5707         ops = sb_dqopt(sb)->ops[qid->type];
5708         if (!ops || !ops->get_next_id)
5709                 return -ENOSYS;
5710         return dquot_get_next_id(sb, qid);
5711 }
5712 #endif
5713
5714 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5715                        const char *dev_name, void *data)
5716 {
5717         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5718 }
5719
5720 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5721 static inline void register_as_ext2(void)
5722 {
5723         int err = register_filesystem(&ext2_fs_type);
5724         if (err)
5725                 printk(KERN_WARNING
5726                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5727 }
5728
5729 static inline void unregister_as_ext2(void)
5730 {
5731         unregister_filesystem(&ext2_fs_type);
5732 }
5733
5734 static inline int ext2_feature_set_ok(struct super_block *sb)
5735 {
5736         if (ext4_has_unknown_ext2_incompat_features(sb))
5737                 return 0;
5738         if (sb_rdonly(sb))
5739                 return 1;
5740         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5741                 return 0;
5742         return 1;
5743 }
5744 #else
5745 static inline void register_as_ext2(void) { }
5746 static inline void unregister_as_ext2(void) { }
5747 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5748 #endif
5749
5750 static inline void register_as_ext3(void)
5751 {
5752         int err = register_filesystem(&ext3_fs_type);
5753         if (err)
5754                 printk(KERN_WARNING
5755                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5756 }
5757
5758 static inline void unregister_as_ext3(void)
5759 {
5760         unregister_filesystem(&ext3_fs_type);
5761 }
5762
5763 static inline int ext3_feature_set_ok(struct super_block *sb)
5764 {
5765         if (ext4_has_unknown_ext3_incompat_features(sb))
5766                 return 0;
5767         if (!ext4_has_feature_journal(sb))
5768                 return 0;
5769         if (sb_rdonly(sb))
5770                 return 1;
5771         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5772                 return 0;
5773         return 1;
5774 }
5775
5776 static struct file_system_type ext4_fs_type = {
5777         .owner          = THIS_MODULE,
5778         .name           = "ext4",
5779         .mount          = ext4_mount,
5780         .kill_sb        = kill_block_super,
5781         .fs_flags       = FS_REQUIRES_DEV,
5782 };
5783 MODULE_ALIAS_FS("ext4");
5784
5785 /* Shared across all ext4 file systems */
5786 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5787
5788 static int __init ext4_init_fs(void)
5789 {
5790         int i, err;
5791
5792         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5793         ext4_li_info = NULL;
5794         mutex_init(&ext4_li_mtx);
5795
5796         /* Build-time check for flags consistency */
5797         ext4_check_flag_values();
5798
5799         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5800                 init_waitqueue_head(&ext4__ioend_wq[i]);
5801
5802         err = ext4_init_es();
5803         if (err)
5804                 return err;
5805
5806         err = ext4_init_pageio();
5807         if (err)
5808                 goto out5;
5809
5810         err = ext4_init_system_zone();
5811         if (err)
5812                 goto out4;
5813
5814         err = ext4_init_sysfs();
5815         if (err)
5816                 goto out3;
5817
5818         err = ext4_init_mballoc();
5819         if (err)
5820                 goto out2;
5821         err = init_inodecache();
5822         if (err)
5823                 goto out1;
5824         register_as_ext3();
5825         register_as_ext2();
5826         err = register_filesystem(&ext4_fs_type);
5827         if (err)
5828                 goto out;
5829
5830         return 0;
5831 out:
5832         unregister_as_ext2();
5833         unregister_as_ext3();
5834         destroy_inodecache();
5835 out1:
5836         ext4_exit_mballoc();
5837 out2:
5838         ext4_exit_sysfs();
5839 out3:
5840         ext4_exit_system_zone();
5841 out4:
5842         ext4_exit_pageio();
5843 out5:
5844         ext4_exit_es();
5845
5846         return err;
5847 }
5848
5849 static void __exit ext4_exit_fs(void)
5850 {
5851         ext4_destroy_lazyinit_thread();
5852         unregister_as_ext2();
5853         unregister_as_ext3();
5854         unregister_filesystem(&ext4_fs_type);
5855         destroy_inodecache();
5856         ext4_exit_mballoc();
5857         ext4_exit_sysfs();
5858         ext4_exit_system_zone();
5859         ext4_exit_pageio();
5860         ext4_exit_es();
5861 }
5862
5863 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5864 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5865 MODULE_LICENSE("GPL");
5866 module_init(ext4_init_fs)
5867 module_exit(ext4_exit_fs)