fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/dax.h>
  41 #include <linux/cleancache.h>
  42 #include <linux/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53 #include "fsmap.h"
  54
  55 #define CREATE_TRACE_POINTS
  56 #include <trace/events/ext4.h>
  57
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ratelimit_state ext4_mount_msg_ratelimit;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  65 static int ext4_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static int ext4_freeze(struct super_block *sb);
  75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  76                        const char *dev_name, void *data);
  77 static inline int ext2_feature_set_ok(struct super_block *sb);
  78 static inline int ext3_feature_set_ok(struct super_block *sb);
  79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  80 static void ext4_destroy_lazyinit_thread(void);
  81 static void ext4_unregister_li_request(struct super_block *sb);
  82 static void ext4_clear_request_list(void);
  83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  84                                             unsigned int journal_inum);
  85
  86 /*
  87  * Lock ordering
  88  *
  89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  91  *
  92  * page fault path:
  93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  94  *   page lock -> i_data_sem (rw)
  95  *
  96  * buffered write path:
  97  * sb_start_write -> i_mutex -> mmap_sem
  98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
  99  *   i_data_sem (rw)
 100  *
 101  * truncate:
 102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 103  *   i_mmap_rwsem (w) -> page lock
 104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
 105  *   transaction start -> i_data_sem (rw)
 106  *
 107  * direct IO:
 108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
 109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
 110  *   transaction start -> i_data_sem (rw)
 111  *
 112  * writepages:
 113  * transaction start -> page lock(s) -> i_data_sem (rw)
 114  */
 115
 116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
 117 static struct file_system_type ext2_fs_type = {
 118         .owner          = THIS_MODULE,
 119         .name           = "ext2",
 120         .mount          = ext4_mount,
 121         .kill_sb        = kill_block_super,
 122         .fs_flags       = FS_REQUIRES_DEV,
 123 };
 124 MODULE_ALIAS_FS("ext2");
 125 MODULE_ALIAS("ext2");
 126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
 127 #else
 128 #define IS_EXT2_SB(sb) (0)
 129 #endif
 130
 131
 132 static struct file_system_type ext3_fs_type = {
 133         .owner          = THIS_MODULE,
 134         .name           = "ext3",
 135         .mount          = ext4_mount,
 136         .kill_sb        = kill_block_super,
 137         .fs_flags       = FS_REQUIRES_DEV,
 138 };
 139 MODULE_ALIAS_FS("ext3");
 140 MODULE_ALIAS("ext3");
 141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 142
 143 static int ext4_verify_csum_type(struct super_block *sb,
 144                                  struct ext4_super_block *es)
 145 {
 146         if (!ext4_has_feature_metadata_csum(sb))
 147                 return 1;
 148
 149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 150 }
 151
 152 static __le32 ext4_superblock_csum(struct super_block *sb,
 153                                    struct ext4_super_block *es)
 154 {
 155         struct ext4_sb_info *sbi = EXT4_SB(sb);
 156         int offset = offsetof(struct ext4_super_block, s_checksum);
 157         __u32 csum;
 158
 159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 160
 161         return cpu_to_le32(csum);
 162 }
 163
 164 static int ext4_superblock_csum_verify(struct super_block *sb,
 165                                        struct ext4_super_block *es)
 166 {
 167         if (!ext4_has_metadata_csum(sb))
 168                 return 1;
 169
 170         return es->s_checksum == ext4_superblock_csum(sb, es);
 171 }
 172
 173 void ext4_superblock_csum_set(struct super_block *sb)
 174 {
 175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 176
 177         if (!ext4_has_metadata_csum(sb))
 178                 return;
 179
 180         es->s_checksum = ext4_superblock_csum(sb, es);
 181 }
 182
 183 void *ext4_kvmalloc(size_t size, gfp_t flags)
 184 {
 185         void *ret;
 186
 187         ret = kmalloc(size, flags | __GFP_NOWARN);
 188         if (!ret)
 189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 190         return ret;
 191 }
 192
 193 void *ext4_kvzalloc(size_t size, gfp_t flags)
 194 {
 195         void *ret;
 196
 197         ret = kzalloc(size, flags | __GFP_NOWARN);
 198         if (!ret)
 199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 200         return ret;
 201 }
 202
 203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 204                                struct ext4_group_desc *bg)
 205 {
 206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 209 }
 210
 211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 212                                struct ext4_group_desc *bg)
 213 {
 214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 217 }
 218
 219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 220                               struct ext4_group_desc *bg)
 221 {
 222         return le32_to_cpu(bg->bg_inode_table_lo) |
 223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 225 }
 226
 227 __u32 ext4_free_group_clusters(struct super_block *sb,
 228                                struct ext4_group_desc *bg)
 229 {
 230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 233 }
 234
 235 __u32 ext4_free_inodes_count(struct super_block *sb,
 236                               struct ext4_group_desc *bg)
 237 {
 238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 241 }
 242
 243 __u32 ext4_used_dirs_count(struct super_block *sb,
 244                               struct ext4_group_desc *bg)
 245 {
 246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 249 }
 250
 251 __u32 ext4_itable_unused_count(struct super_block *sb,
 252                               struct ext4_group_desc *bg)
 253 {
 254         return le16_to_cpu(bg->bg_itable_unused_lo) |
 255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 257 }
 258
 259 void ext4_block_bitmap_set(struct super_block *sb,
 260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 261 {
 262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 265 }
 266
 267 void ext4_inode_bitmap_set(struct super_block *sb,
 268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 269 {
 270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 273 }
 274
 275 void ext4_inode_table_set(struct super_block *sb,
 276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 277 {
 278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 281 }
 282
 283 void ext4_free_group_clusters_set(struct super_block *sb,
 284                                   struct ext4_group_desc *bg, __u32 count)
 285 {
 286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 289 }
 290
 291 void ext4_free_inodes_set(struct super_block *sb,
 292                           struct ext4_group_desc *bg, __u32 count)
 293 {
 294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 297 }
 298
 299 void ext4_used_dirs_set(struct super_block *sb,
 300                           struct ext4_group_desc *bg, __u32 count)
 301 {
 302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 305 }
 306
 307 void ext4_itable_unused_set(struct super_block *sb,
 308                           struct ext4_group_desc *bg, __u32 count)
 309 {
 310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 313 }
 314
 315
 316 static void __save_error_info(struct super_block *sb, const char *func,
 317                             unsigned int line)
 318 {
 319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 320
 321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 322         if (bdev_read_only(sb->s_bdev))
 323                 return;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         le32_add_cpu(&es->s_error_count, 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352 /*
 353  * The del_gendisk() function uninitializes the disk-specific data
 354  * structures, including the bdi structure, without telling anyone
 355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 357  * This is a kludge to prevent these oops until we can put in a proper
 358  * hook in del_gendisk() to inform the VFS and file system layers.
 359  */
 360 static int block_device_ejected(struct super_block *sb)
 361 {
 362         struct inode *bd_inode = sb->s_bdev->bd_inode;
 363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 364
 365         return bdi->dev == NULL;
 366 }
 367
 368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 369 {
 370         struct super_block              *sb = journal->j_private;
 371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 372         int                             error = is_journal_aborted(journal);
 373         struct ext4_journal_cb_entry    *jce;
 374
 375         BUG_ON(txn->t_state == T_FINISHED);
 376
 377         ext4_process_freed_data(sb, txn->t_tid);
 378
 379         spin_lock(&sbi->s_md_lock);
 380         while (!list_empty(&txn->t_private_list)) {
 381                 jce = list_entry(txn->t_private_list.next,
 382                                  struct ext4_journal_cb_entry, jce_list);
 383                 list_del_init(&jce->jce_list);
 384                 spin_unlock(&sbi->s_md_lock);
 385                 jce->jce_func(sb, jce, error);
 386                 spin_lock(&sbi->s_md_lock);
 387         }
 388         spin_unlock(&sbi->s_md_lock);
 389 }
 390
 391 /* Deal with the reporting of failure conditions on a filesystem such as
 392  * inconsistencies detected or read IO failures.
 393  *
 394  * On ext2, we can store the error state of the filesystem in the
 395  * superblock.  That is not possible on ext4, because we may have other
 396  * write ordering constraints on the superblock which prevent us from
 397  * writing it out straight away; and given that the journal is about to
 398  * be aborted, we can't rely on the current, or future, transactions to
 399  * write out the superblock safely.
 400  *
 401  * We'll just use the jbd2_journal_abort() error code to record an error in
 402  * the journal instead.  On recovery, the journal will complain about
 403  * that error until we've noted it down and cleared it.
 404  */
 405
 406 static void ext4_handle_error(struct super_block *sb)
 407 {
 408         if (sb_rdonly(sb))
 409                 return;
 410
 411         if (!test_opt(sb, ERRORS_CONT)) {
 412                 journal_t *journal = EXT4_SB(sb)->s_journal;
 413
 414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 415                 if (journal)
 416                         jbd2_journal_abort(journal, -EIO);
 417         }
 418         if (test_opt(sb, ERRORS_RO)) {
 419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 420                 /*
 421                  * Make sure updated value of ->s_mount_flags will be visible
 422                  * before ->s_flags update
 423                  */
 424                 smp_wmb();
 425                 sb->s_flags |= MS_RDONLY;
 426         }
 427         if (test_opt(sb, ERRORS_PANIC)) {
 428                 if (EXT4_SB(sb)->s_journal &&
 429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 430                         return;
 431                 panic("EXT4-fs (device %s): panic forced after error\n",
 432                         sb->s_id);
 433         }
 434 }
 435
 436 #define ext4_error_ratelimit(sb)                                        \
 437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 438                              "EXT4-fs error")
 439
 440 void __ext4_error(struct super_block *sb, const char *function,
 441                   unsigned int line, const char *fmt, ...)
 442 {
 443         struct va_format vaf;
 444         va_list args;
 445
 446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 447                 return;
 448
 449         if (ext4_error_ratelimit(sb)) {
 450                 va_start(args, fmt);
 451                 vaf.fmt = fmt;
 452                 vaf.va = &args;
 453                 printk(KERN_CRIT
 454                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 455                        sb->s_id, function, line, current->comm, &vaf);
 456                 va_end(args);
 457         }
 458         save_error_info(sb, function, line);
 459         ext4_handle_error(sb);
 460 }
 461
 462 void __ext4_error_inode(struct inode *inode, const char *function,
 463                         unsigned int line, ext4_fsblk_t block,
 464                         const char *fmt, ...)
 465 {
 466         va_list args;
 467         struct va_format vaf;
 468         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 469
 470         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 471                 return;
 472
 473         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 474         es->s_last_error_block = cpu_to_le64(block);
 475         if (ext4_error_ratelimit(inode->i_sb)) {
 476                 va_start(args, fmt);
 477                 vaf.fmt = fmt;
 478                 vaf.va = &args;
 479                 if (block)
 480                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 481                                "inode #%lu: block %llu: comm %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                block, current->comm, &vaf);
 484                 else
 485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 486                                "inode #%lu: comm %s: %pV\n",
 487                                inode->i_sb->s_id, function, line, inode->i_ino,
 488                                current->comm, &vaf);
 489                 va_end(args);
 490         }
 491         save_error_info(inode->i_sb, function, line);
 492         ext4_handle_error(inode->i_sb);
 493 }
 494
 495 void __ext4_error_file(struct file *file, const char *function,
 496                        unsigned int line, ext4_fsblk_t block,
 497                        const char *fmt, ...)
 498 {
 499         va_list args;
 500         struct va_format vaf;
 501         struct ext4_super_block *es;
 502         struct inode *inode = file_inode(file);
 503         char pathname[80], *path;
 504
 505         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 506                 return;
 507
 508         es = EXT4_SB(inode->i_sb)->s_es;
 509         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 510         if (ext4_error_ratelimit(inode->i_sb)) {
 511                 path = file_path(file, pathname, sizeof(pathname));
 512                 if (IS_ERR(path))
 513                         path = "(unknown)";
 514                 va_start(args, fmt);
 515                 vaf.fmt = fmt;
 516                 vaf.va = &args;
 517                 if (block)
 518                         printk(KERN_CRIT
 519                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 520                                "block %llu: comm %s: path %s: %pV\n",
 521                                inode->i_sb->s_id, function, line, inode->i_ino,
 522                                block, current->comm, path, &vaf);
 523                 else
 524                         printk(KERN_CRIT
 525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 526                                "comm %s: path %s: %pV\n",
 527                                inode->i_sb->s_id, function, line, inode->i_ino,
 528                                current->comm, path, &vaf);
 529                 va_end(args);
 530         }
 531         save_error_info(inode->i_sb, function, line);
 532         ext4_handle_error(inode->i_sb);
 533 }
 534
 535 const char *ext4_decode_error(struct super_block *sb, int errno,
 536                               char nbuf[16])
 537 {
 538         char *errstr = NULL;
 539
 540         switch (errno) {
 541         case -EFSCORRUPTED:
 542                 errstr = "Corrupt filesystem";
 543                 break;
 544         case -EFSBADCRC:
 545                 errstr = "Filesystem failed CRC";
 546                 break;
 547         case -EIO:
 548                 errstr = "IO failure";
 549                 break;
 550         case -ENOMEM:
 551                 errstr = "Out of memory";
 552                 break;
 553         case -EROFS:
 554                 if (!sb || (EXT4_SB(sb)->s_journal &&
 555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 556                         errstr = "Journal has aborted";
 557                 else
 558                         errstr = "Readonly filesystem";
 559                 break;
 560         default:
 561                 /* If the caller passed in an extra buffer for unknown
 562                  * errors, textualise them now.  Else we just return
 563                  * NULL. */
 564                 if (nbuf) {
 565                         /* Check for truncated error codes... */
 566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 567                                 errstr = nbuf;
 568                 }
 569                 break;
 570         }
 571
 572         return errstr;
 573 }
 574
 575 /* __ext4_std_error decodes expected errors from journaling functions
 576  * automatically and invokes the appropriate error response.  */
 577
 578 void __ext4_std_error(struct super_block *sb, const char *function,
 579                       unsigned int line, int errno)
 580 {
 581         char nbuf[16];
 582         const char *errstr;
 583
 584         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 585                 return;
 586
 587         /* Special case: if the error is EROFS, and we're not already
 588          * inside a transaction, then there's really no point in logging
 589          * an error. */
 590         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 591                 return;
 592
 593         if (ext4_error_ratelimit(sb)) {
 594                 errstr = ext4_decode_error(sb, errno, nbuf);
 595                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 596                        sb->s_id, function, line, errstr);
 597         }
 598
 599         save_error_info(sb, function, line);
 600         ext4_handle_error(sb);
 601 }
 602
 603 /*
 604  * ext4_abort is a much stronger failure handler than ext4_error.  The
 605  * abort function may be used to deal with unrecoverable failures such
 606  * as journal IO errors or ENOMEM at a critical moment in log management.
 607  *
 608  * We unconditionally force the filesystem into an ABORT|READONLY state,
 609  * unless the error response on the fs has been set to panic in which
 610  * case we take the easy way out and panic immediately.
 611  */
 612
 613 void __ext4_abort(struct super_block *sb, const char *function,
 614                 unsigned int line, const char *fmt, ...)
 615 {
 616         struct va_format vaf;
 617         va_list args;
 618
 619         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 620                 return;
 621
 622         save_error_info(sb, function, line);
 623         va_start(args, fmt);
 624         vaf.fmt = fmt;
 625         vaf.va = &args;
 626         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 627                sb->s_id, function, line, &vaf);
 628         va_end(args);
 629
 630         if (sb_rdonly(sb) == 0) {
 631                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 632                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 633                 /*
 634                  * Make sure updated value of ->s_mount_flags will be visible
 635                  * before ->s_flags update
 636                  */
 637                 smp_wmb();
 638                 sb->s_flags |= MS_RDONLY;
 639                 if (EXT4_SB(sb)->s_journal)
 640                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 641                 save_error_info(sb, function, line);
 642         }
 643         if (test_opt(sb, ERRORS_PANIC)) {
 644                 if (EXT4_SB(sb)->s_journal &&
 645                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 646                         return;
 647                 panic("EXT4-fs panic from previous error\n");
 648         }
 649 }
 650
 651 void __ext4_msg(struct super_block *sb,
 652                 const char *prefix, const char *fmt, ...)
 653 {
 654         struct va_format vaf;
 655         va_list args;
 656
 657         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 658                 return;
 659
 660         va_start(args, fmt);
 661         vaf.fmt = fmt;
 662         vaf.va = &args;
 663         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 664         va_end(args);
 665 }
 666
 667 #define ext4_warning_ratelimit(sb)                                      \
 668                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 669                              "EXT4-fs warning")
 670
 671 void __ext4_warning(struct super_block *sb, const char *function,
 672                     unsigned int line, const char *fmt, ...)
 673 {
 674         struct va_format vaf;
 675         va_list args;
 676
 677         if (!ext4_warning_ratelimit(sb))
 678                 return;
 679
 680         va_start(args, fmt);
 681         vaf.fmt = fmt;
 682         vaf.va = &args;
 683         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 684                sb->s_id, function, line, &vaf);
 685         va_end(args);
 686 }
 687
 688 void __ext4_warning_inode(const struct inode *inode, const char *function,
 689                           unsigned int line, const char *fmt, ...)
 690 {
 691         struct va_format vaf;
 692         va_list args;
 693
 694         if (!ext4_warning_ratelimit(inode->i_sb))
 695                 return;
 696
 697         va_start(args, fmt);
 698         vaf.fmt = fmt;
 699         vaf.va = &args;
 700         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 701                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 702                function, line, inode->i_ino, current->comm, &vaf);
 703         va_end(args);
 704 }
 705
 706 void __ext4_grp_locked_error(const char *function, unsigned int line,
 707                              struct super_block *sb, ext4_group_t grp,
 708                              unsigned long ino, ext4_fsblk_t block,
 709                              const char *fmt, ...)
 710 __releases(bitlock)
 711 __acquires(bitlock)
 712 {
 713         struct va_format vaf;
 714         va_list args;
 715         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 716
 717         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 718                 return;
 719
 720         es->s_last_error_ino = cpu_to_le32(ino);
 721         es->s_last_error_block = cpu_to_le64(block);
 722         __save_error_info(sb, function, line);
 723
 724         if (ext4_error_ratelimit(sb)) {
 725                 va_start(args, fmt);
 726                 vaf.fmt = fmt;
 727                 vaf.va = &args;
 728                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 729                        sb->s_id, function, line, grp);
 730                 if (ino)
 731                         printk(KERN_CONT "inode %lu: ", ino);
 732                 if (block)
 733                         printk(KERN_CONT "block %llu:",
 734                                (unsigned long long) block);
 735                 printk(KERN_CONT "%pV\n", &vaf);
 736                 va_end(args);
 737         }
 738
 739         if (test_opt(sb, ERRORS_CONT)) {
 740                 ext4_commit_super(sb, 0);
 741                 return;
 742         }
 743
 744         ext4_unlock_group(sb, grp);
 745         ext4_commit_super(sb, 1);
 746         ext4_handle_error(sb);
 747         /*
 748          * We only get here in the ERRORS_RO case; relocking the group
 749          * may be dangerous, but nothing bad will happen since the
 750          * filesystem will have already been marked read/only and the
 751          * journal has been aborted.  We return 1 as a hint to callers
 752          * who might what to use the return value from
 753          * ext4_grp_locked_error() to distinguish between the
 754          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 755          * aggressively from the ext4 function in question, with a
 756          * more appropriate error code.
 757          */
 758         ext4_lock_group(sb, grp);
 759         return;
 760 }
 761
 762 void ext4_update_dynamic_rev(struct super_block *sb)
 763 {
 764         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 765
 766         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 767                 return;
 768
 769         ext4_warning(sb,
 770                      "updating to rev %d because of new feature flag, "
 771                      "running e2fsck is recommended",
 772                      EXT4_DYNAMIC_REV);
 773
 774         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 775         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 776         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 777         /* leave es->s_feature_*compat flags alone */
 778         /* es->s_uuid will be set by e2fsck if empty */
 779
 780         /*
 781          * The rest of the superblock fields should be zero, and if not it
 782          * means they are likely already in use, so leave them alone.  We
 783          * can leave it up to e2fsck to clean up any inconsistencies there.
 784          */
 785 }
 786
 787 /*
 788  * Open the external journal device
 789  */
 790 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 791 {
 792         struct block_device *bdev;
 793         char b[BDEVNAME_SIZE];
 794
 795         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 796         if (IS_ERR(bdev))
 797                 goto fail;
 798         return bdev;
 799
 800 fail:
 801         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 802                         __bdevname(dev, b), PTR_ERR(bdev));
 803         return NULL;
 804 }
 805
 806 /*
 807  * Release the journal device
 808  */
 809 static void ext4_blkdev_put(struct block_device *bdev)
 810 {
 811         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 812 }
 813
 814 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 815 {
 816         struct block_device *bdev;
 817         bdev = sbi->journal_bdev;
 818         if (bdev) {
 819                 ext4_blkdev_put(bdev);
 820                 sbi->journal_bdev = NULL;
 821         }
 822 }
 823
 824 static inline struct inode *orphan_list_entry(struct list_head *l)
 825 {
 826         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 827 }
 828
 829 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 830 {
 831         struct list_head *l;
 832
 833         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 834                  le32_to_cpu(sbi->s_es->s_last_orphan));
 835
 836         printk(KERN_ERR "sb_info orphan list:\n");
 837         list_for_each(l, &sbi->s_orphan) {
 838                 struct inode *inode = orphan_list_entry(l);
 839                 printk(KERN_ERR "  "
 840                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 841                        inode->i_sb->s_id, inode->i_ino, inode,
 842                        inode->i_mode, inode->i_nlink,
 843                        NEXT_ORPHAN(inode));
 844         }
 845 }
 846
 847 #ifdef CONFIG_QUOTA
 848 static int ext4_quota_off(struct super_block *sb, int type);
 849
 850 static inline void ext4_quota_off_umount(struct super_block *sb)
 851 {
 852         int type;
 853
 854         /* Use our quota_off function to clear inode flags etc. */
 855         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 856                 ext4_quota_off(sb, type);
 857 }
 858 #else
 859 static inline void ext4_quota_off_umount(struct super_block *sb)
 860 {
 861 }
 862 #endif
 863
 864 static void ext4_put_super(struct super_block *sb)
 865 {
 866         struct ext4_sb_info *sbi = EXT4_SB(sb);
 867         struct ext4_super_block *es = sbi->s_es;
 868         int aborted = 0;
 869         int i, err;
 870
 871         ext4_unregister_li_request(sb);
 872         ext4_quota_off_umount(sb);
 873
 874         flush_workqueue(sbi->rsv_conversion_wq);
 875         destroy_workqueue(sbi->rsv_conversion_wq);
 876
 877         if (sbi->s_journal) {
 878                 aborted = is_journal_aborted(sbi->s_journal);
 879                 err = jbd2_journal_destroy(sbi->s_journal);
 880                 sbi->s_journal = NULL;
 881                 if ((err < 0) && !aborted)
 882                         ext4_abort(sb, "Couldn't clean up the journal");
 883         }
 884
 885         ext4_unregister_sysfs(sb);
 886         ext4_es_unregister_shrinker(sbi);
 887         del_timer_sync(&sbi->s_err_report);
 888         ext4_release_system_zone(sb);
 889         ext4_mb_release(sb);
 890         ext4_ext_release(sb);
 891
 892         if (!sb_rdonly(sb) && !aborted) {
 893                 ext4_clear_feature_journal_needs_recovery(sb);
 894                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 895         }
 896         if (!sb_rdonly(sb))
 897                 ext4_commit_super(sb, 1);
 898
 899         for (i = 0; i < sbi->s_gdb_count; i++)
 900                 brelse(sbi->s_group_desc[i]);
 901         kvfree(sbi->s_group_desc);
 902         kvfree(sbi->s_flex_groups);
 903         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 904         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 905         percpu_counter_destroy(&sbi->s_dirs_counter);
 906         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 907         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 908 #ifdef CONFIG_QUOTA
 909         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 910                 kfree(sbi->s_qf_names[i]);
 911 #endif
 912
 913         /* Debugging code just in case the in-memory inode orphan list
 914          * isn't empty.  The on-disk one can be non-empty if we've
 915          * detected an error and taken the fs readonly, but the
 916          * in-memory list had better be clean by this point. */
 917         if (!list_empty(&sbi->s_orphan))
 918                 dump_orphan_list(sb, sbi);
 919         J_ASSERT(list_empty(&sbi->s_orphan));
 920
 921         sync_blockdev(sb->s_bdev);
 922         invalidate_bdev(sb->s_bdev);
 923         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 924                 /*
 925                  * Invalidate the journal device's buffers.  We don't want them
 926                  * floating about in memory - the physical journal device may
 927                  * hotswapped, and it breaks the `ro-after' testing code.
 928                  */
 929                 sync_blockdev(sbi->journal_bdev);
 930                 invalidate_bdev(sbi->journal_bdev);
 931                 ext4_blkdev_remove(sbi);
 932         }
 933         if (sbi->s_ea_inode_cache) {
 934                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 935                 sbi->s_ea_inode_cache = NULL;
 936         }
 937         if (sbi->s_ea_block_cache) {
 938                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 939                 sbi->s_ea_block_cache = NULL;
 940         }
 941         if (sbi->s_mmp_tsk)
 942                 kthread_stop(sbi->s_mmp_tsk);
 943         brelse(sbi->s_sbh);
 944         sb->s_fs_info = NULL;
 945         /*
 946          * Now that we are completely done shutting down the
 947          * superblock, we need to actually destroy the kobject.
 948          */
 949         kobject_put(&sbi->s_kobj);
 950         wait_for_completion(&sbi->s_kobj_unregister);
 951         if (sbi->s_chksum_driver)
 952                 crypto_free_shash(sbi->s_chksum_driver);
 953         kfree(sbi->s_blockgroup_lock);
 954         fs_put_dax(sbi->s_daxdev);
 955         kfree(sbi);
 956 }
 957
 958 static struct kmem_cache *ext4_inode_cachep;
 959
 960 /*
 961  * Called inside transaction, so use GFP_NOFS
 962  */
 963 static struct inode *ext4_alloc_inode(struct super_block *sb)
 964 {
 965         struct ext4_inode_info *ei;
 966
 967         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 968         if (!ei)
 969                 return NULL;
 970
 971         ei->vfs_inode.i_version = 1;
 972         spin_lock_init(&ei->i_raw_lock);
 973         INIT_LIST_HEAD(&ei->i_prealloc_list);
 974         spin_lock_init(&ei->i_prealloc_lock);
 975         ext4_es_init_tree(&ei->i_es_tree);
 976         rwlock_init(&ei->i_es_lock);
 977         INIT_LIST_HEAD(&ei->i_es_list);
 978         ei->i_es_all_nr = 0;
 979         ei->i_es_shk_nr = 0;
 980         ei->i_es_shrink_lblk = 0;
 981         ei->i_reserved_data_blocks = 0;
 982         ei->i_da_metadata_calc_len = 0;
 983         ei->i_da_metadata_calc_last_lblock = 0;
 984         spin_lock_init(&(ei->i_block_reservation_lock));
 985 #ifdef CONFIG_QUOTA
 986         ei->i_reserved_quota = 0;
 987         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 988 #endif
 989         ei->jinode = NULL;
 990         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 991         spin_lock_init(&ei->i_completed_io_lock);
 992         ei->i_sync_tid = 0;
 993         ei->i_datasync_tid = 0;
 994         atomic_set(&ei->i_unwritten, 0);
 995         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 996         return &ei->vfs_inode;
 997 }
 998
 999 static int ext4_drop_inode(struct inode *inode)
1000 {
1001         int drop = generic_drop_inode(inode);
1002
1003         trace_ext4_drop_inode(inode, drop);
1004         return drop;
1005 }
1006
1007 static void ext4_i_callback(struct rcu_head *head)
1008 {
1009         struct inode *inode = container_of(head, struct inode, i_rcu);
1010         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1011 }
1012
1013 static void ext4_destroy_inode(struct inode *inode)
1014 {
1015         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1016                 ext4_msg(inode->i_sb, KERN_ERR,
1017                          "Inode %lu (%p): orphan list check failed!",
1018                          inode->i_ino, EXT4_I(inode));
1019                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1020                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1021                                 true);
1022                 dump_stack();
1023         }
1024         call_rcu(&inode->i_rcu, ext4_i_callback);
1025 }
1026
1027 static void init_once(void *foo)
1028 {
1029         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1030
1031         INIT_LIST_HEAD(&ei->i_orphan);
1032         init_rwsem(&ei->xattr_sem);
1033         init_rwsem(&ei->i_data_sem);
1034         init_rwsem(&ei->i_mmap_sem);
1035         inode_init_once(&ei->vfs_inode);
1036 }
1037
1038 static int __init init_inodecache(void)
1039 {
1040         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1041                                              sizeof(struct ext4_inode_info),
1042                                              0, (SLAB_RECLAIM_ACCOUNT|
1043                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1044                                              init_once);
1045         if (ext4_inode_cachep == NULL)
1046                 return -ENOMEM;
1047         return 0;
1048 }
1049
1050 static void destroy_inodecache(void)
1051 {
1052         /*
1053          * Make sure all delayed rcu free inodes are flushed before we
1054          * destroy cache.
1055          */
1056         rcu_barrier();
1057         kmem_cache_destroy(ext4_inode_cachep);
1058 }
1059
1060 void ext4_clear_inode(struct inode *inode)
1061 {
1062         invalidate_inode_buffers(inode);
1063         clear_inode(inode);
1064         dquot_drop(inode);
1065         ext4_discard_preallocations(inode);
1066         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1067         if (EXT4_I(inode)->jinode) {
1068                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1069                                                EXT4_I(inode)->jinode);
1070                 jbd2_free_inode(EXT4_I(inode)->jinode);
1071                 EXT4_I(inode)->jinode = NULL;
1072         }
1073 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1074         fscrypt_put_encryption_info(inode, NULL);
1075 #endif
1076 }
1077
1078 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1079                                         u64 ino, u32 generation)
1080 {
1081         struct inode *inode;
1082
1083         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1084                 return ERR_PTR(-ESTALE);
1085         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1086                 return ERR_PTR(-ESTALE);
1087
1088         /* iget isn't really right if the inode is currently unallocated!!
1089          *
1090          * ext4_read_inode will return a bad_inode if the inode had been
1091          * deleted, so we should be safe.
1092          *
1093          * Currently we don't know the generation for parent directory, so
1094          * a generation of 0 means "accept any"
1095          */
1096         inode = ext4_iget_normal(sb, ino);
1097         if (IS_ERR(inode))
1098                 return ERR_CAST(inode);
1099         if (generation && inode->i_generation != generation) {
1100                 iput(inode);
1101                 return ERR_PTR(-ESTALE);
1102         }
1103
1104         return inode;
1105 }
1106
1107 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1108                                         int fh_len, int fh_type)
1109 {
1110         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1111                                     ext4_nfs_get_inode);
1112 }
1113
1114 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1115                                         int fh_len, int fh_type)
1116 {
1117         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1118                                     ext4_nfs_get_inode);
1119 }
1120
1121 /*
1122  * Try to release metadata pages (indirect blocks, directories) which are
1123  * mapped via the block device.  Since these pages could have journal heads
1124  * which would prevent try_to_free_buffers() from freeing them, we must use
1125  * jbd2 layer's try_to_free_buffers() function to release them.
1126  */
1127 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1128                                  gfp_t wait)
1129 {
1130         journal_t *journal = EXT4_SB(sb)->s_journal;
1131
1132         WARN_ON(PageChecked(page));
1133         if (!page_has_buffers(page))
1134                 return 0;
1135         if (journal)
1136                 return jbd2_journal_try_to_free_buffers(journal, page,
1137                                                 wait & ~__GFP_DIRECT_RECLAIM);
1138         return try_to_free_buffers(page);
1139 }
1140
1141 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1142 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1143 {
1144         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1145                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1146 }
1147
1148 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1149                                                         void *fs_data)
1150 {
1151         handle_t *handle = fs_data;
1152         int res, res2, credits, retries = 0;
1153
1154         /*
1155          * Encrypting the root directory is not allowed because e2fsck expects
1156          * lost+found to exist and be unencrypted, and encrypting the root
1157          * directory would imply encrypting the lost+found directory as well as
1158          * the filename "lost+found" itself.
1159          */
1160         if (inode->i_ino == EXT4_ROOT_INO)
1161                 return -EPERM;
1162
1163         res = ext4_convert_inline_data(inode);
1164         if (res)
1165                 return res;
1166
1167         /*
1168          * If a journal handle was specified, then the encryption context is
1169          * being set on a new inode via inheritance and is part of a larger
1170          * transaction to create the inode.  Otherwise the encryption context is
1171          * being set on an existing inode in its own transaction.  Only in the
1172          * latter case should the "retry on ENOSPC" logic be used.
1173          */
1174
1175         if (handle) {
1176                 res = ext4_xattr_set_handle(handle, inode,
1177                                             EXT4_XATTR_INDEX_ENCRYPTION,
1178                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1179                                             ctx, len, 0);
1180                 if (!res) {
1181                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1182                         ext4_clear_inode_state(inode,
1183                                         EXT4_STATE_MAY_INLINE_DATA);
1184                         /*
1185                          * Update inode->i_flags - e.g. S_DAX may get disabled
1186                          */
1187                         ext4_set_inode_flags(inode);
1188                 }
1189                 return res;
1190         }
1191
1192         res = dquot_initialize(inode);
1193         if (res)
1194                 return res;
1195 retry:
1196         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1197                                      &credits);
1198         if (res)
1199                 return res;
1200
1201         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1202         if (IS_ERR(handle))
1203                 return PTR_ERR(handle);
1204
1205         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1206                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1207                                     ctx, len, 0);
1208         if (!res) {
1209                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1210                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1211                 ext4_set_inode_flags(inode);
1212                 res = ext4_mark_inode_dirty(handle, inode);
1213                 if (res)
1214                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1215         }
1216         res2 = ext4_journal_stop(handle);
1217
1218         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1219                 goto retry;
1220         if (!res)
1221                 res = res2;
1222         return res;
1223 }
1224
1225 static bool ext4_dummy_context(struct inode *inode)
1226 {
1227         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1228 }
1229
1230 static unsigned ext4_max_namelen(struct inode *inode)
1231 {
1232         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1233                 EXT4_NAME_LEN;
1234 }
1235
1236 static const struct fscrypt_operations ext4_cryptops = {
1237         .key_prefix             = "ext4:",
1238         .get_context            = ext4_get_context,
1239         .set_context            = ext4_set_context,
1240         .dummy_context          = ext4_dummy_context,
1241         .is_encrypted           = ext4_encrypted_inode,
1242         .empty_dir              = ext4_empty_dir,
1243         .max_namelen            = ext4_max_namelen,
1244 };
1245 #else
1246 static const struct fscrypt_operations ext4_cryptops = {
1247         .is_encrypted           = ext4_encrypted_inode,
1248 };
1249 #endif
1250
1251 #ifdef CONFIG_QUOTA
1252 static const char * const quotatypes[] = INITQFNAMES;
1253 #define QTYPE2NAME(t) (quotatypes[t])
1254
1255 static int ext4_write_dquot(struct dquot *dquot);
1256 static int ext4_acquire_dquot(struct dquot *dquot);
1257 static int ext4_release_dquot(struct dquot *dquot);
1258 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1259 static int ext4_write_info(struct super_block *sb, int type);
1260 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1261                          const struct path *path);
1262 static int ext4_quota_on_mount(struct super_block *sb, int type);
1263 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1264                                size_t len, loff_t off);
1265 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1266                                 const char *data, size_t len, loff_t off);
1267 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1268                              unsigned int flags);
1269 static int ext4_enable_quotas(struct super_block *sb);
1270 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1271
1272 static struct dquot **ext4_get_dquots(struct inode *inode)
1273 {
1274         return EXT4_I(inode)->i_dquot;
1275 }
1276
1277 static const struct dquot_operations ext4_quota_operations = {
1278         .get_reserved_space     = ext4_get_reserved_space,
1279         .write_dquot            = ext4_write_dquot,
1280         .acquire_dquot          = ext4_acquire_dquot,
1281         .release_dquot          = ext4_release_dquot,
1282         .mark_dirty             = ext4_mark_dquot_dirty,
1283         .write_info             = ext4_write_info,
1284         .alloc_dquot            = dquot_alloc,
1285         .destroy_dquot          = dquot_destroy,
1286         .get_projid             = ext4_get_projid,
1287         .get_inode_usage        = ext4_get_inode_usage,
1288         .get_next_id            = ext4_get_next_id,
1289 };
1290
1291 static const struct quotactl_ops ext4_qctl_operations = {
1292         .quota_on       = ext4_quota_on,
1293         .quota_off      = ext4_quota_off,
1294         .quota_sync     = dquot_quota_sync,
1295         .get_state      = dquot_get_state,
1296         .set_info       = dquot_set_dqinfo,
1297         .get_dqblk      = dquot_get_dqblk,
1298         .set_dqblk      = dquot_set_dqblk,
1299         .get_nextdqblk  = dquot_get_next_dqblk,
1300 };
1301 #endif
1302
1303 static const struct super_operations ext4_sops = {
1304         .alloc_inode    = ext4_alloc_inode,
1305         .destroy_inode  = ext4_destroy_inode,
1306         .write_inode    = ext4_write_inode,
1307         .dirty_inode    = ext4_dirty_inode,
1308         .drop_inode     = ext4_drop_inode,
1309         .evict_inode    = ext4_evict_inode,
1310         .put_super      = ext4_put_super,
1311         .sync_fs        = ext4_sync_fs,
1312         .freeze_fs      = ext4_freeze,
1313         .unfreeze_fs    = ext4_unfreeze,
1314         .statfs         = ext4_statfs,
1315         .remount_fs     = ext4_remount,
1316         .show_options   = ext4_show_options,
1317 #ifdef CONFIG_QUOTA
1318         .quota_read     = ext4_quota_read,
1319         .quota_write    = ext4_quota_write,
1320         .get_dquots     = ext4_get_dquots,
1321 #endif
1322         .bdev_try_to_free_page = bdev_try_to_free_page,
1323 };
1324
1325 static const struct export_operations ext4_export_ops = {
1326         .fh_to_dentry = ext4_fh_to_dentry,
1327         .fh_to_parent = ext4_fh_to_parent,
1328         .get_parent = ext4_get_parent,
1329 };
1330
1331 enum {
1332         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1333         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1334         Opt_nouid32, Opt_debug, Opt_removed,
1335         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1336         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1337         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1338         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1339         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1340         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1341         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1342         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1343         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1344         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1345         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1346         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1347         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1348         Opt_inode_readahead_blks, Opt_journal_ioprio,
1349         Opt_dioread_nolock, Opt_dioread_lock,
1350         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1351         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1352 };
1353
1354 static const match_table_t tokens = {
1355         {Opt_bsd_df, "bsddf"},
1356         {Opt_minix_df, "minixdf"},
1357         {Opt_grpid, "grpid"},
1358         {Opt_grpid, "bsdgroups"},
1359         {Opt_nogrpid, "nogrpid"},
1360         {Opt_nogrpid, "sysvgroups"},
1361         {Opt_resgid, "resgid=%u"},
1362         {Opt_resuid, "resuid=%u"},
1363         {Opt_sb, "sb=%u"},
1364         {Opt_err_cont, "errors=continue"},
1365         {Opt_err_panic, "errors=panic"},
1366         {Opt_err_ro, "errors=remount-ro"},
1367         {Opt_nouid32, "nouid32"},
1368         {Opt_debug, "debug"},
1369         {Opt_removed, "oldalloc"},
1370         {Opt_removed, "orlov"},
1371         {Opt_user_xattr, "user_xattr"},
1372         {Opt_nouser_xattr, "nouser_xattr"},
1373         {Opt_acl, "acl"},
1374         {Opt_noacl, "noacl"},
1375         {Opt_noload, "norecovery"},
1376         {Opt_noload, "noload"},
1377         {Opt_removed, "nobh"},
1378         {Opt_removed, "bh"},
1379         {Opt_commit, "commit=%u"},
1380         {Opt_min_batch_time, "min_batch_time=%u"},
1381         {Opt_max_batch_time, "max_batch_time=%u"},
1382         {Opt_journal_dev, "journal_dev=%u"},
1383         {Opt_journal_path, "journal_path=%s"},
1384         {Opt_journal_checksum, "journal_checksum"},
1385         {Opt_nojournal_checksum, "nojournal_checksum"},
1386         {Opt_journal_async_commit, "journal_async_commit"},
1387         {Opt_abort, "abort"},
1388         {Opt_data_journal, "data=journal"},
1389         {Opt_data_ordered, "data=ordered"},
1390         {Opt_data_writeback, "data=writeback"},
1391         {Opt_data_err_abort, "data_err=abort"},
1392         {Opt_data_err_ignore, "data_err=ignore"},
1393         {Opt_offusrjquota, "usrjquota="},
1394         {Opt_usrjquota, "usrjquota=%s"},
1395         {Opt_offgrpjquota, "grpjquota="},
1396         {Opt_grpjquota, "grpjquota=%s"},
1397         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1398         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1399         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1400         {Opt_grpquota, "grpquota"},
1401         {Opt_noquota, "noquota"},
1402         {Opt_quota, "quota"},
1403         {Opt_usrquota, "usrquota"},
1404         {Opt_prjquota, "prjquota"},
1405         {Opt_barrier, "barrier=%u"},
1406         {Opt_barrier, "barrier"},
1407         {Opt_nobarrier, "nobarrier"},
1408         {Opt_i_version, "i_version"},
1409         {Opt_dax, "dax"},
1410         {Opt_stripe, "stripe=%u"},
1411         {Opt_delalloc, "delalloc"},
1412         {Opt_lazytime, "lazytime"},
1413         {Opt_nolazytime, "nolazytime"},
1414         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1415         {Opt_nodelalloc, "nodelalloc"},
1416         {Opt_removed, "mblk_io_submit"},
1417         {Opt_removed, "nomblk_io_submit"},
1418         {Opt_block_validity, "block_validity"},
1419         {Opt_noblock_validity, "noblock_validity"},
1420         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1421         {Opt_journal_ioprio, "journal_ioprio=%u"},
1422         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1423         {Opt_auto_da_alloc, "auto_da_alloc"},
1424         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1425         {Opt_dioread_nolock, "dioread_nolock"},
1426         {Opt_dioread_lock, "dioread_lock"},
1427         {Opt_discard, "discard"},
1428         {Opt_nodiscard, "nodiscard"},
1429         {Opt_init_itable, "init_itable=%u"},
1430         {Opt_init_itable, "init_itable"},
1431         {Opt_noinit_itable, "noinit_itable"},
1432         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1433         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1434         {Opt_nombcache, "nombcache"},
1435         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1436         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1437         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1438         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1439         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1440         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1441         {Opt_err, NULL},
1442 };
1443
1444 static ext4_fsblk_t get_sb_block(void **data)
1445 {
1446         ext4_fsblk_t    sb_block;
1447         char            *options = (char *) *data;
1448
1449         if (!options || strncmp(options, "sb=", 3) != 0)
1450                 return 1;       /* Default location */
1451
1452         options += 3;
1453         /* TODO: use simple_strtoll with >32bit ext4 */
1454         sb_block = simple_strtoul(options, &options, 0);
1455         if (*options && *options != ',') {
1456                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1457                        (char *) *data);
1458                 return 1;
1459         }
1460         if (*options == ',')
1461                 options++;
1462         *data = (void *) options;
1463
1464         return sb_block;
1465 }
1466
1467 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1468 static const char deprecated_msg[] =
1469         "Mount option \"%s\" will be removed by %s\n"
1470         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1471
1472 #ifdef CONFIG_QUOTA
1473 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1474 {
1475         struct ext4_sb_info *sbi = EXT4_SB(sb);
1476         char *qname;
1477         int ret = -1;
1478
1479         if (sb_any_quota_loaded(sb) &&
1480                 !sbi->s_qf_names[qtype]) {
1481                 ext4_msg(sb, KERN_ERR,
1482                         "Cannot change journaled "
1483                         "quota options when quota turned on");
1484                 return -1;
1485         }
1486         if (ext4_has_feature_quota(sb)) {
1487                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1488                          "ignored when QUOTA feature is enabled");
1489                 return 1;
1490         }
1491         qname = match_strdup(args);
1492         if (!qname) {
1493                 ext4_msg(sb, KERN_ERR,
1494                         "Not enough memory for storing quotafile name");
1495                 return -1;
1496         }
1497         if (sbi->s_qf_names[qtype]) {
1498                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1499                         ret = 1;
1500                 else
1501                         ext4_msg(sb, KERN_ERR,
1502                                  "%s quota file already specified",
1503                                  QTYPE2NAME(qtype));
1504                 goto errout;
1505         }
1506         if (strchr(qname, '/')) {
1507                 ext4_msg(sb, KERN_ERR,
1508                         "quotafile must be on filesystem root");
1509                 goto errout;
1510         }
1511         sbi->s_qf_names[qtype] = qname;
1512         set_opt(sb, QUOTA);
1513         return 1;
1514 errout:
1515         kfree(qname);
1516         return ret;
1517 }
1518
1519 static int clear_qf_name(struct super_block *sb, int qtype)
1520 {
1521
1522         struct ext4_sb_info *sbi = EXT4_SB(sb);
1523
1524         if (sb_any_quota_loaded(sb) &&
1525                 sbi->s_qf_names[qtype]) {
1526                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1527                         " when quota turned on");
1528                 return -1;
1529         }
1530         kfree(sbi->s_qf_names[qtype]);
1531         sbi->s_qf_names[qtype] = NULL;
1532         return 1;
1533 }
1534 #endif
1535
1536 #define MOPT_SET        0x0001
1537 #define MOPT_CLEAR      0x0002
1538 #define MOPT_NOSUPPORT  0x0004
1539 #define MOPT_EXPLICIT   0x0008
1540 #define MOPT_CLEAR_ERR  0x0010
1541 #define MOPT_GTE0       0x0020
1542 #ifdef CONFIG_QUOTA
1543 #define MOPT_Q          0
1544 #define MOPT_QFMT       0x0040
1545 #else
1546 #define MOPT_Q          MOPT_NOSUPPORT
1547 #define MOPT_QFMT       MOPT_NOSUPPORT
1548 #endif
1549 #define MOPT_DATAJ      0x0080
1550 #define MOPT_NO_EXT2    0x0100
1551 #define MOPT_NO_EXT3    0x0200
1552 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1553 #define MOPT_STRING     0x0400
1554
1555 static const struct mount_opts {
1556         int     token;
1557         int     mount_opt;
1558         int     flags;
1559 } ext4_mount_opts[] = {
1560         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1561         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1562         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1563         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1564         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1565         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1566         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1567          MOPT_EXT4_ONLY | MOPT_SET},
1568         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1569          MOPT_EXT4_ONLY | MOPT_CLEAR},
1570         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1571         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1572         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1573          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1574         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1575          MOPT_EXT4_ONLY | MOPT_CLEAR},
1576         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1577          MOPT_EXT4_ONLY | MOPT_CLEAR},
1578         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1579          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1580         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1581                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1582          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1583         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1584         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1585         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1586         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1587         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1588          MOPT_NO_EXT2},
1589         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1590          MOPT_NO_EXT2},
1591         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1592         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1593         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1594         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1595         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1596         {Opt_commit, 0, MOPT_GTE0},
1597         {Opt_max_batch_time, 0, MOPT_GTE0},
1598         {Opt_min_batch_time, 0, MOPT_GTE0},
1599         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1600         {Opt_init_itable, 0, MOPT_GTE0},
1601         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1602         {Opt_stripe, 0, MOPT_GTE0},
1603         {Opt_resuid, 0, MOPT_GTE0},
1604         {Opt_resgid, 0, MOPT_GTE0},
1605         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1606         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1607         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1608         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1609         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1610         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1611          MOPT_NO_EXT2 | MOPT_DATAJ},
1612         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1613         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1614 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1615         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1616         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1617 #else
1618         {Opt_acl, 0, MOPT_NOSUPPORT},
1619         {Opt_noacl, 0, MOPT_NOSUPPORT},
1620 #endif
1621         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1622         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1623         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1624         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1625         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1626                                                         MOPT_SET | MOPT_Q},
1627         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1628                                                         MOPT_SET | MOPT_Q},
1629         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1630                                                         MOPT_SET | MOPT_Q},
1631         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1632                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1633                                                         MOPT_CLEAR | MOPT_Q},
1634         {Opt_usrjquota, 0, MOPT_Q},
1635         {Opt_grpjquota, 0, MOPT_Q},
1636         {Opt_offusrjquota, 0, MOPT_Q},
1637         {Opt_offgrpjquota, 0, MOPT_Q},
1638         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1639         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1640         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1641         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1642         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1643         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1644         {Opt_err, 0, 0}
1645 };
1646
1647 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1648                             substring_t *args, unsigned long *journal_devnum,
1649                             unsigned int *journal_ioprio, int is_remount)
1650 {
1651         struct ext4_sb_info *sbi = EXT4_SB(sb);
1652         const struct mount_opts *m;
1653         kuid_t uid;
1654         kgid_t gid;
1655         int arg = 0;
1656
1657 #ifdef CONFIG_QUOTA
1658         if (token == Opt_usrjquota)
1659                 return set_qf_name(sb, USRQUOTA, &args[0]);
1660         else if (token == Opt_grpjquota)
1661                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1662         else if (token == Opt_offusrjquota)
1663                 return clear_qf_name(sb, USRQUOTA);
1664         else if (token == Opt_offgrpjquota)
1665                 return clear_qf_name(sb, GRPQUOTA);
1666 #endif
1667         switch (token) {
1668         case Opt_noacl:
1669         case Opt_nouser_xattr:
1670                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1671                 break;
1672         case Opt_sb:
1673                 return 1;       /* handled by get_sb_block() */
1674         case Opt_removed:
1675                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1676                 return 1;
1677         case Opt_abort:
1678                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1679                 return 1;
1680         case Opt_i_version:
1681                 sb->s_flags |= SB_I_VERSION;
1682                 return 1;
1683         case Opt_lazytime:
1684                 sb->s_flags |= MS_LAZYTIME;
1685                 return 1;
1686         case Opt_nolazytime:
1687                 sb->s_flags &= ~MS_LAZYTIME;
1688                 return 1;
1689         }
1690
1691         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1692                 if (token == m->token)
1693                         break;
1694
1695         if (m->token == Opt_err) {
1696                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1697                          "or missing value", opt);
1698                 return -1;
1699         }
1700
1701         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1702                 ext4_msg(sb, KERN_ERR,
1703                          "Mount option \"%s\" incompatible with ext2", opt);
1704                 return -1;
1705         }
1706         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1707                 ext4_msg(sb, KERN_ERR,
1708                          "Mount option \"%s\" incompatible with ext3", opt);
1709                 return -1;
1710         }
1711
1712         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1713                 return -1;
1714         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1715                 return -1;
1716         if (m->flags & MOPT_EXPLICIT) {
1717                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1718                         set_opt2(sb, EXPLICIT_DELALLOC);
1719                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1720                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1721                 } else
1722                         return -1;
1723         }
1724         if (m->flags & MOPT_CLEAR_ERR)
1725                 clear_opt(sb, ERRORS_MASK);
1726         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1727                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1728                          "options when quota turned on");
1729                 return -1;
1730         }
1731
1732         if (m->flags & MOPT_NOSUPPORT) {
1733                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1734         } else if (token == Opt_commit) {
1735                 if (arg == 0)
1736                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1737                 sbi->s_commit_interval = HZ * arg;
1738         } else if (token == Opt_debug_want_extra_isize) {
1739                 sbi->s_want_extra_isize = arg;
1740         } else if (token == Opt_max_batch_time) {
1741                 sbi->s_max_batch_time = arg;
1742         } else if (token == Opt_min_batch_time) {
1743                 sbi->s_min_batch_time = arg;
1744         } else if (token == Opt_inode_readahead_blks) {
1745                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1746                         ext4_msg(sb, KERN_ERR,
1747                                  "EXT4-fs: inode_readahead_blks must be "
1748                                  "0 or a power of 2 smaller than 2^31");
1749                         return -1;
1750                 }
1751                 sbi->s_inode_readahead_blks = arg;
1752         } else if (token == Opt_init_itable) {
1753                 set_opt(sb, INIT_INODE_TABLE);
1754                 if (!args->from)
1755                         arg = EXT4_DEF_LI_WAIT_MULT;
1756                 sbi->s_li_wait_mult = arg;
1757         } else if (token == Opt_max_dir_size_kb) {
1758                 sbi->s_max_dir_size_kb = arg;
1759         } else if (token == Opt_stripe) {
1760                 sbi->s_stripe = arg;
1761         } else if (token == Opt_resuid) {
1762                 uid = make_kuid(current_user_ns(), arg);
1763                 if (!uid_valid(uid)) {
1764                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1765                         return -1;
1766                 }
1767                 sbi->s_resuid = uid;
1768         } else if (token == Opt_resgid) {
1769                 gid = make_kgid(current_user_ns(), arg);
1770                 if (!gid_valid(gid)) {
1771                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1772                         return -1;
1773                 }
1774                 sbi->s_resgid = gid;
1775         } else if (token == Opt_journal_dev) {
1776                 if (is_remount) {
1777                         ext4_msg(sb, KERN_ERR,
1778                                  "Cannot specify journal on remount");
1779                         return -1;
1780                 }
1781                 *journal_devnum = arg;
1782         } else if (token == Opt_journal_path) {
1783                 char *journal_path;
1784                 struct inode *journal_inode;
1785                 struct path path;
1786                 int error;
1787
1788                 if (is_remount) {
1789                         ext4_msg(sb, KERN_ERR,
1790                                  "Cannot specify journal on remount");
1791                         return -1;
1792                 }
1793                 journal_path = match_strdup(&args[0]);
1794                 if (!journal_path) {
1795                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1796                                 "journal device string");
1797                         return -1;
1798                 }
1799
1800                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1801                 if (error) {
1802                         ext4_msg(sb, KERN_ERR, "error: could not find "
1803                                 "journal device path: error %d", error);
1804                         kfree(journal_path);
1805                         return -1;
1806                 }
1807
1808                 journal_inode = d_inode(path.dentry);
1809                 if (!S_ISBLK(journal_inode->i_mode)) {
1810                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1811                                 "is not a block device", journal_path);
1812                         path_put(&path);
1813                         kfree(journal_path);
1814                         return -1;
1815                 }
1816
1817                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1818                 path_put(&path);
1819                 kfree(journal_path);
1820         } else if (token == Opt_journal_ioprio) {
1821                 if (arg > 7) {
1822                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1823                                  " (must be 0-7)");
1824                         return -1;
1825                 }
1826                 *journal_ioprio =
1827                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1828         } else if (token == Opt_test_dummy_encryption) {
1829 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1830                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1831                 ext4_msg(sb, KERN_WARNING,
1832                          "Test dummy encryption mode enabled");
1833 #else
1834                 ext4_msg(sb, KERN_WARNING,
1835                          "Test dummy encryption mount option ignored");
1836 #endif
1837         } else if (m->flags & MOPT_DATAJ) {
1838                 if (is_remount) {
1839                         if (!sbi->s_journal)
1840                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1841                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1842                                 ext4_msg(sb, KERN_ERR,
1843                                          "Cannot change data mode on remount");
1844                                 return -1;
1845                         }
1846                 } else {
1847                         clear_opt(sb, DATA_FLAGS);
1848                         sbi->s_mount_opt |= m->mount_opt;
1849                 }
1850 #ifdef CONFIG_QUOTA
1851         } else if (m->flags & MOPT_QFMT) {
1852                 if (sb_any_quota_loaded(sb) &&
1853                     sbi->s_jquota_fmt != m->mount_opt) {
1854                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1855                                  "quota options when quota turned on");
1856                         return -1;
1857                 }
1858                 if (ext4_has_feature_quota(sb)) {
1859                         ext4_msg(sb, KERN_INFO,
1860                                  "Quota format mount options ignored "
1861                                  "when QUOTA feature is enabled");
1862                         return 1;
1863                 }
1864                 sbi->s_jquota_fmt = m->mount_opt;
1865 #endif
1866         } else if (token == Opt_dax) {
1867 #ifdef CONFIG_FS_DAX
1868                 ext4_msg(sb, KERN_WARNING,
1869                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1870                         sbi->s_mount_opt |= m->mount_opt;
1871 #else
1872                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1873                 return -1;
1874 #endif
1875         } else if (token == Opt_data_err_abort) {
1876                 sbi->s_mount_opt |= m->mount_opt;
1877         } else if (token == Opt_data_err_ignore) {
1878                 sbi->s_mount_opt &= ~m->mount_opt;
1879         } else {
1880                 if (!args->from)
1881                         arg = 1;
1882                 if (m->flags & MOPT_CLEAR)
1883                         arg = !arg;
1884                 else if (unlikely(!(m->flags & MOPT_SET))) {
1885                         ext4_msg(sb, KERN_WARNING,
1886                                  "buggy handling of option %s", opt);
1887                         WARN_ON(1);
1888                         return -1;
1889                 }
1890                 if (arg != 0)
1891                         sbi->s_mount_opt |= m->mount_opt;
1892                 else
1893                         sbi->s_mount_opt &= ~m->mount_opt;
1894         }
1895         return 1;
1896 }
1897
1898 static int parse_options(char *options, struct super_block *sb,
1899                          unsigned long *journal_devnum,
1900                          unsigned int *journal_ioprio,
1901                          int is_remount)
1902 {
1903         struct ext4_sb_info *sbi = EXT4_SB(sb);
1904         char *p;
1905         substring_t args[MAX_OPT_ARGS];
1906         int token;
1907
1908         if (!options)
1909                 return 1;
1910
1911         while ((p = strsep(&options, ",")) != NULL) {
1912                 if (!*p)
1913                         continue;
1914                 /*
1915                  * Initialize args struct so we know whether arg was
1916                  * found; some options take optional arguments.
1917                  */
1918                 args[0].to = args[0].from = NULL;
1919                 token = match_token(p, tokens, args);
1920                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1921                                      journal_ioprio, is_remount) < 0)
1922                         return 0;
1923         }
1924 #ifdef CONFIG_QUOTA
1925         /*
1926          * We do the test below only for project quotas. 'usrquota' and
1927          * 'grpquota' mount options are allowed even without quota feature
1928          * to support legacy quotas in quota files.
1929          */
1930         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1931                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1932                          "Cannot enable project quota enforcement.");
1933                 return 0;
1934         }
1935         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1936                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1937                         clear_opt(sb, USRQUOTA);
1938
1939                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1940                         clear_opt(sb, GRPQUOTA);
1941
1942                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1943                         ext4_msg(sb, KERN_ERR, "old and new quota "
1944                                         "format mixing");
1945                         return 0;
1946                 }
1947
1948                 if (!sbi->s_jquota_fmt) {
1949                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1950                                         "not specified");
1951                         return 0;
1952                 }
1953         }
1954 #endif
1955         if (test_opt(sb, DIOREAD_NOLOCK)) {
1956                 int blocksize =
1957                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1958
1959                 if (blocksize < PAGE_SIZE) {
1960                         ext4_msg(sb, KERN_ERR, "can't mount with "
1961                                  "dioread_nolock if block size != PAGE_SIZE");
1962                         return 0;
1963                 }
1964         }
1965         return 1;
1966 }
1967
1968 static inline void ext4_show_quota_options(struct seq_file *seq,
1969                                            struct super_block *sb)
1970 {
1971 #if defined(CONFIG_QUOTA)
1972         struct ext4_sb_info *sbi = EXT4_SB(sb);
1973
1974         if (sbi->s_jquota_fmt) {
1975                 char *fmtname = "";
1976
1977                 switch (sbi->s_jquota_fmt) {
1978                 case QFMT_VFS_OLD:
1979                         fmtname = "vfsold";
1980                         break;
1981                 case QFMT_VFS_V0:
1982                         fmtname = "vfsv0";
1983                         break;
1984                 case QFMT_VFS_V1:
1985                         fmtname = "vfsv1";
1986                         break;
1987                 }
1988                 seq_printf(seq, ",jqfmt=%s", fmtname);
1989         }
1990
1991         if (sbi->s_qf_names[USRQUOTA])
1992                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1993
1994         if (sbi->s_qf_names[GRPQUOTA])
1995                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1996 #endif
1997 }
1998
1999 static const char *token2str(int token)
2000 {
2001         const struct match_token *t;
2002
2003         for (t = tokens; t->token != Opt_err; t++)
2004                 if (t->token == token && !strchr(t->pattern, '='))
2005                         break;
2006         return t->pattern;
2007 }
2008
2009 /*
2010  * Show an option if
2011  *  - it's set to a non-default value OR
2012  *  - if the per-sb default is different from the global default
2013  */
2014 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2015                               int nodefs)
2016 {
2017         struct ext4_sb_info *sbi = EXT4_SB(sb);
2018         struct ext4_super_block *es = sbi->s_es;
2019         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2020         const struct mount_opts *m;
2021         char sep = nodefs ? '\n' : ',';
2022
2023 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2024 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2025
2026         if (sbi->s_sb_block != 1)
2027                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2028
2029         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2030                 int want_set = m->flags & MOPT_SET;
2031                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2032                     (m->flags & MOPT_CLEAR_ERR))
2033                         continue;
2034                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2035                         continue; /* skip if same as the default */
2036                 if ((want_set &&
2037                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2038                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2039                         continue; /* select Opt_noFoo vs Opt_Foo */
2040                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2041         }
2042
2043         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2044             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2045                 SEQ_OPTS_PRINT("resuid=%u",
2046                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2047         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2048             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2049                 SEQ_OPTS_PRINT("resgid=%u",
2050                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2051         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2052         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2053                 SEQ_OPTS_PUTS("errors=remount-ro");
2054         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2055                 SEQ_OPTS_PUTS("errors=continue");
2056         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2057                 SEQ_OPTS_PUTS("errors=panic");
2058         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2059                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2060         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2061                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2062         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2063                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2064         if (sb->s_flags & SB_I_VERSION)
2065                 SEQ_OPTS_PUTS("i_version");
2066         if (nodefs || sbi->s_stripe)
2067                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2068         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2069                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2070                         SEQ_OPTS_PUTS("data=journal");
2071                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2072                         SEQ_OPTS_PUTS("data=ordered");
2073                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2074                         SEQ_OPTS_PUTS("data=writeback");
2075         }
2076         if (nodefs ||
2077             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2078                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2079                                sbi->s_inode_readahead_blks);
2080
2081         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2082                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2083                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2084         if (nodefs || sbi->s_max_dir_size_kb)
2085                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2086         if (test_opt(sb, DATA_ERR_ABORT))
2087                 SEQ_OPTS_PUTS("data_err=abort");
2088
2089         ext4_show_quota_options(seq, sb);
2090         return 0;
2091 }
2092
2093 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2094 {
2095         return _ext4_show_options(seq, root->d_sb, 0);
2096 }
2097
2098 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2099 {
2100         struct super_block *sb = seq->private;
2101         int rc;
2102
2103         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2104         rc = _ext4_show_options(seq, sb, 1);
2105         seq_puts(seq, "\n");
2106         return rc;
2107 }
2108
2109 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2110                             int read_only)
2111 {
2112         struct ext4_sb_info *sbi = EXT4_SB(sb);
2113         int res = 0;
2114
2115         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2116                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2117                          "forcing read-only mode");
2118                 res = MS_RDONLY;
2119         }
2120         if (read_only)
2121                 goto done;
2122         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2123                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2124                          "running e2fsck is recommended");
2125         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2126                 ext4_msg(sb, KERN_WARNING,
2127                          "warning: mounting fs with errors, "
2128                          "running e2fsck is recommended");
2129         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2130                  le16_to_cpu(es->s_mnt_count) >=
2131                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2132                 ext4_msg(sb, KERN_WARNING,
2133                          "warning: maximal mount count reached, "
2134                          "running e2fsck is recommended");
2135         else if (le32_to_cpu(es->s_checkinterval) &&
2136                 (le32_to_cpu(es->s_lastcheck) +
2137                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2138                 ext4_msg(sb, KERN_WARNING,
2139                          "warning: checktime reached, "
2140                          "running e2fsck is recommended");
2141         if (!sbi->s_journal)
2142                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2143         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2144                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2145         le16_add_cpu(&es->s_mnt_count, 1);
2146         es->s_mtime = cpu_to_le32(get_seconds());
2147         ext4_update_dynamic_rev(sb);
2148         if (sbi->s_journal)
2149                 ext4_set_feature_journal_needs_recovery(sb);
2150
2151         ext4_commit_super(sb, 1);
2152 done:
2153         if (test_opt(sb, DEBUG))
2154                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2155                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2156                         sb->s_blocksize,
2157                         sbi->s_groups_count,
2158                         EXT4_BLOCKS_PER_GROUP(sb),
2159                         EXT4_INODES_PER_GROUP(sb),
2160                         sbi->s_mount_opt, sbi->s_mount_opt2);
2161
2162         cleancache_init_fs(sb);
2163         return res;
2164 }
2165
2166 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2167 {
2168         struct ext4_sb_info *sbi = EXT4_SB(sb);
2169         struct flex_groups *new_groups;
2170         int size;
2171
2172         if (!sbi->s_log_groups_per_flex)
2173                 return 0;
2174
2175         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2176         if (size <= sbi->s_flex_groups_allocated)
2177                 return 0;
2178
2179         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2180         new_groups = kvzalloc(size, GFP_KERNEL);
2181         if (!new_groups) {
2182                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2183                          size / (int) sizeof(struct flex_groups));
2184                 return -ENOMEM;
2185         }
2186
2187         if (sbi->s_flex_groups) {
2188                 memcpy(new_groups, sbi->s_flex_groups,
2189                        (sbi->s_flex_groups_allocated *
2190                         sizeof(struct flex_groups)));
2191                 kvfree(sbi->s_flex_groups);
2192         }
2193         sbi->s_flex_groups = new_groups;
2194         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2195         return 0;
2196 }
2197
2198 static int ext4_fill_flex_info(struct super_block *sb)
2199 {
2200         struct ext4_sb_info *sbi = EXT4_SB(sb);
2201         struct ext4_group_desc *gdp = NULL;
2202         ext4_group_t flex_group;
2203         int i, err;
2204
2205         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2206         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2207                 sbi->s_log_groups_per_flex = 0;
2208                 return 1;
2209         }
2210
2211         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2212         if (err)
2213                 goto failed;
2214
2215         for (i = 0; i < sbi->s_groups_count; i++) {
2216                 gdp = ext4_get_group_desc(sb, i, NULL);
2217
2218                 flex_group = ext4_flex_group(sbi, i);
2219                 atomic_add(ext4_free_inodes_count(sb, gdp),
2220                            &sbi->s_flex_groups[flex_group].free_inodes);
2221                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2222                              &sbi->s_flex_groups[flex_group].free_clusters);
2223                 atomic_add(ext4_used_dirs_count(sb, gdp),
2224                            &sbi->s_flex_groups[flex_group].used_dirs);
2225         }
2226
2227         return 1;
2228 failed:
2229         return 0;
2230 }
2231
2232 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2233                                    struct ext4_group_desc *gdp)
2234 {
2235         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2236         __u16 crc = 0;
2237         __le32 le_group = cpu_to_le32(block_group);
2238         struct ext4_sb_info *sbi = EXT4_SB(sb);
2239
2240         if (ext4_has_metadata_csum(sbi->s_sb)) {
2241                 /* Use new metadata_csum algorithm */
2242                 __u32 csum32;
2243                 __u16 dummy_csum = 0;
2244
2245                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2246                                      sizeof(le_group));
2247                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2248                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2249                                      sizeof(dummy_csum));
2250                 offset += sizeof(dummy_csum);
2251                 if (offset < sbi->s_desc_size)
2252                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2253                                              sbi->s_desc_size - offset);
2254
2255                 crc = csum32 & 0xFFFF;
2256                 goto out;
2257         }
2258
2259         /* old crc16 code */
2260         if (!ext4_has_feature_gdt_csum(sb))
2261                 return 0;
2262
2263         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2264         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2265         crc = crc16(crc, (__u8 *)gdp, offset);
2266         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2267         /* for checksum of struct ext4_group_desc do the rest...*/
2268         if (ext4_has_feature_64bit(sb) &&
2269             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2270                 crc = crc16(crc, (__u8 *)gdp + offset,
2271                             le16_to_cpu(sbi->s_es->s_desc_size) -
2272                                 offset);
2273
2274 out:
2275         return cpu_to_le16(crc);
2276 }
2277
2278 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2279                                 struct ext4_group_desc *gdp)
2280 {
2281         if (ext4_has_group_desc_csum(sb) &&
2282             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2283                 return 0;
2284
2285         return 1;
2286 }
2287
2288 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2289                               struct ext4_group_desc *gdp)
2290 {
2291         if (!ext4_has_group_desc_csum(sb))
2292                 return;
2293         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2294 }
2295
2296 /* Called at mount-time, super-block is locked */
2297 static int ext4_check_descriptors(struct super_block *sb,
2298                                   ext4_fsblk_t sb_block,
2299                                   ext4_group_t *first_not_zeroed)
2300 {
2301         struct ext4_sb_info *sbi = EXT4_SB(sb);
2302         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2303         ext4_fsblk_t last_block;
2304         ext4_fsblk_t block_bitmap;
2305         ext4_fsblk_t inode_bitmap;
2306         ext4_fsblk_t inode_table;
2307         int flexbg_flag = 0;
2308         ext4_group_t i, grp = sbi->s_groups_count;
2309
2310         if (ext4_has_feature_flex_bg(sb))
2311                 flexbg_flag = 1;
2312
2313         ext4_debug("Checking group descriptors");
2314
2315         for (i = 0; i < sbi->s_groups_count; i++) {
2316                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2317
2318                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2319                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2320                 else
2321                         last_block = first_block +
2322                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2323
2324                 if ((grp == sbi->s_groups_count) &&
2325                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2326                         grp = i;
2327
2328                 block_bitmap = ext4_block_bitmap(sb, gdp);
2329                 if (block_bitmap == sb_block) {
2330                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2331                                  "Block bitmap for group %u overlaps "
2332                                  "superblock", i);
2333                         if (!sb_rdonly(sb))
2334                                 return 0;
2335                 }
2336                 if (block_bitmap < first_block || block_bitmap > last_block) {
2337                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2338                                "Block bitmap for group %u not in group "
2339                                "(block %llu)!", i, block_bitmap);
2340                         return 0;
2341                 }
2342                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2343                 if (inode_bitmap == sb_block) {
2344                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2345                                  "Inode bitmap for group %u overlaps "
2346                                  "superblock", i);
2347                         if (!sb_rdonly(sb))
2348                                 return 0;
2349                 }
2350                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2351                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2352                                "Inode bitmap for group %u not in group "
2353                                "(block %llu)!", i, inode_bitmap);
2354                         return 0;
2355                 }
2356                 inode_table = ext4_inode_table(sb, gdp);
2357                 if (inode_table == sb_block) {
2358                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2359                                  "Inode table for group %u overlaps "
2360                                  "superblock", i);
2361                         if (!sb_rdonly(sb))
2362                                 return 0;
2363                 }
2364                 if (inode_table < first_block ||
2365                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2366                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2367                                "Inode table for group %u not in group "
2368                                "(block %llu)!", i, inode_table);
2369                         return 0;
2370                 }
2371                 ext4_lock_group(sb, i);
2372                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2373                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2374                                  "Checksum for group %u failed (%u!=%u)",
2375                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2376                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2377                         if (!sb_rdonly(sb)) {
2378                                 ext4_unlock_group(sb, i);
2379                                 return 0;
2380                         }
2381                 }
2382                 ext4_unlock_group(sb, i);
2383                 if (!flexbg_flag)
2384                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2385         }
2386         if (NULL != first_not_zeroed)
2387                 *first_not_zeroed = grp;
2388         return 1;
2389 }
2390
2391 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2392  * the superblock) which were deleted from all directories, but held open by
2393  * a process at the time of a crash.  We walk the list and try to delete these
2394  * inodes at recovery time (only with a read-write filesystem).
2395  *
2396  * In order to keep the orphan inode chain consistent during traversal (in
2397  * case of crash during recovery), we link each inode into the superblock
2398  * orphan list_head and handle it the same way as an inode deletion during
2399  * normal operation (which journals the operations for us).
2400  *
2401  * We only do an iget() and an iput() on each inode, which is very safe if we
2402  * accidentally point at an in-use or already deleted inode.  The worst that
2403  * can happen in this case is that we get a "bit already cleared" message from
2404  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2405  * e2fsck was run on this filesystem, and it must have already done the orphan
2406  * inode cleanup for us, so we can safely abort without any further action.
2407  */
2408 static void ext4_orphan_cleanup(struct super_block *sb,
2409                                 struct ext4_super_block *es)
2410 {
2411         unsigned int s_flags = sb->s_flags;
2412         int ret, nr_orphans = 0, nr_truncates = 0;
2413 #ifdef CONFIG_QUOTA
2414         int quota_update = 0;
2415         int i;
2416 #endif
2417         if (!es->s_last_orphan) {
2418                 jbd_debug(4, "no orphan inodes to clean up\n");
2419                 return;
2420         }
2421
2422         if (bdev_read_only(sb->s_bdev)) {
2423                 ext4_msg(sb, KERN_ERR, "write access "
2424                         "unavailable, skipping orphan cleanup");
2425                 return;
2426         }
2427
2428         /* Check if feature set would not allow a r/w mount */
2429         if (!ext4_feature_set_ok(sb, 0)) {
2430                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2431                          "unknown ROCOMPAT features");
2432                 return;
2433         }
2434
2435         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2436                 /* don't clear list on RO mount w/ errors */
2437                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2438                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2439                                   "clearing orphan list.\n");
2440                         es->s_last_orphan = 0;
2441                 }
2442                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2443                 return;
2444         }
2445
2446         if (s_flags & MS_RDONLY) {
2447                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2448                 sb->s_flags &= ~MS_RDONLY;
2449         }
2450 #ifdef CONFIG_QUOTA
2451         /* Needed for iput() to work correctly and not trash data */
2452         sb->s_flags |= MS_ACTIVE;
2453
2454         /*
2455          * Turn on quotas which were not enabled for read-only mounts if
2456          * filesystem has quota feature, so that they are updated correctly.
2457          */
2458         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2459                 int ret = ext4_enable_quotas(sb);
2460
2461                 if (!ret)
2462                         quota_update = 1;
2463                 else
2464                         ext4_msg(sb, KERN_ERR,
2465                                 "Cannot turn on quotas: error %d", ret);
2466         }
2467
2468         /* Turn on journaled quotas used for old sytle */
2469         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2470                 if (EXT4_SB(sb)->s_qf_names[i]) {
2471                         int ret = ext4_quota_on_mount(sb, i);
2472
2473                         if (!ret)
2474                                 quota_update = 1;
2475                         else
2476                                 ext4_msg(sb, KERN_ERR,
2477                                         "Cannot turn on journaled "
2478                                         "quota: type %d: error %d", i, ret);
2479                 }
2480         }
2481 #endif
2482
2483         while (es->s_last_orphan) {
2484                 struct inode *inode;
2485
2486                 /*
2487                  * We may have encountered an error during cleanup; if
2488                  * so, skip the rest.
2489                  */
2490                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2491                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2492                         es->s_last_orphan = 0;
2493                         break;
2494                 }
2495
2496                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2497                 if (IS_ERR(inode)) {
2498                         es->s_last_orphan = 0;
2499                         break;
2500                 }
2501
2502                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2503                 dquot_initialize(inode);
2504                 if (inode->i_nlink) {
2505                         if (test_opt(sb, DEBUG))
2506                                 ext4_msg(sb, KERN_DEBUG,
2507                                         "%s: truncating inode %lu to %lld bytes",
2508                                         __func__, inode->i_ino, inode->i_size);
2509                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2510                                   inode->i_ino, inode->i_size);
2511                         inode_lock(inode);
2512                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2513                         ret = ext4_truncate(inode);
2514                         if (ret)
2515                                 ext4_std_error(inode->i_sb, ret);
2516                         inode_unlock(inode);
2517                         nr_truncates++;
2518                 } else {
2519                         if (test_opt(sb, DEBUG))
2520                                 ext4_msg(sb, KERN_DEBUG,
2521                                         "%s: deleting unreferenced inode %lu",
2522                                         __func__, inode->i_ino);
2523                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2524                                   inode->i_ino);
2525                         nr_orphans++;
2526                 }
2527                 iput(inode);  /* The delete magic happens here! */
2528         }
2529
2530 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2531
2532         if (nr_orphans)
2533                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2534                        PLURAL(nr_orphans));
2535         if (nr_truncates)
2536                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2537                        PLURAL(nr_truncates));
2538 #ifdef CONFIG_QUOTA
2539         /* Turn off quotas if they were enabled for orphan cleanup */
2540         if (quota_update) {
2541                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2542                         if (sb_dqopt(sb)->files[i])
2543                                 dquot_quota_off(sb, i);
2544                 }
2545         }
2546 #endif
2547         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2548 }
2549
2550 /*
2551  * Maximal extent format file size.
2552  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2553  * extent format containers, within a sector_t, and within i_blocks
2554  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2555  * so that won't be a limiting factor.
2556  *
2557  * However there is other limiting factor. We do store extents in the form
2558  * of starting block and length, hence the resulting length of the extent
2559  * covering maximum file size must fit into on-disk format containers as
2560  * well. Given that length is always by 1 unit bigger than max unit (because
2561  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2562  *
2563  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2564  */
2565 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2566 {
2567         loff_t res;
2568         loff_t upper_limit = MAX_LFS_FILESIZE;
2569
2570         /* small i_blocks in vfs inode? */
2571         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2572                 /*
2573                  * CONFIG_LBDAF is not enabled implies the inode
2574                  * i_block represent total blocks in 512 bytes
2575                  * 32 == size of vfs inode i_blocks * 8
2576                  */
2577                 upper_limit = (1LL << 32) - 1;
2578
2579                 /* total blocks in file system block size */
2580                 upper_limit >>= (blkbits - 9);
2581                 upper_limit <<= blkbits;
2582         }
2583
2584         /*
2585          * 32-bit extent-start container, ee_block. We lower the maxbytes
2586          * by one fs block, so ee_len can cover the extent of maximum file
2587          * size
2588          */
2589         res = (1LL << 32) - 1;
2590         res <<= blkbits;
2591
2592         /* Sanity check against vm- & vfs- imposed limits */
2593         if (res > upper_limit)
2594                 res = upper_limit;
2595
2596         return res;
2597 }
2598
2599 /*
2600  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2601  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2602  * We need to be 1 filesystem block less than the 2^48 sector limit.
2603  */
2604 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2605 {
2606         loff_t res = EXT4_NDIR_BLOCKS;
2607         int meta_blocks;
2608         loff_t upper_limit;
2609         /* This is calculated to be the largest file size for a dense, block
2610          * mapped file such that the file's total number of 512-byte sectors,
2611          * including data and all indirect blocks, does not exceed (2^48 - 1).
2612          *
2613          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2614          * number of 512-byte sectors of the file.
2615          */
2616
2617         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2618                 /*
2619                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2620                  * the inode i_block field represents total file blocks in
2621                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2622                  */
2623                 upper_limit = (1LL << 32) - 1;
2624
2625                 /* total blocks in file system block size */
2626                 upper_limit >>= (bits - 9);
2627
2628         } else {
2629                 /*
2630                  * We use 48 bit ext4_inode i_blocks
2631                  * With EXT4_HUGE_FILE_FL set the i_blocks
2632                  * represent total number of blocks in
2633                  * file system block size
2634                  */
2635                 upper_limit = (1LL << 48) - 1;
2636
2637         }
2638
2639         /* indirect blocks */
2640         meta_blocks = 1;
2641         /* double indirect blocks */
2642         meta_blocks += 1 + (1LL << (bits-2));
2643         /* tripple indirect blocks */
2644         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2645
2646         upper_limit -= meta_blocks;
2647         upper_limit <<= bits;
2648
2649         res += 1LL << (bits-2);
2650         res += 1LL << (2*(bits-2));
2651         res += 1LL << (3*(bits-2));
2652         res <<= bits;
2653         if (res > upper_limit)
2654                 res = upper_limit;
2655
2656         if (res > MAX_LFS_FILESIZE)
2657                 res = MAX_LFS_FILESIZE;
2658
2659         return res;
2660 }
2661
2662 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2663                                    ext4_fsblk_t logical_sb_block, int nr)
2664 {
2665         struct ext4_sb_info *sbi = EXT4_SB(sb);
2666         ext4_group_t bg, first_meta_bg;
2667         int has_super = 0;
2668
2669         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2670
2671         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2672                 return logical_sb_block + nr + 1;
2673         bg = sbi->s_desc_per_block * nr;
2674         if (ext4_bg_has_super(sb, bg))
2675                 has_super = 1;
2676
2677         /*
2678          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2679          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2680          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2681          * compensate.
2682          */
2683         if (sb->s_blocksize == 1024 && nr == 0 &&
2684             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2685                 has_super++;
2686
2687         return (has_super + ext4_group_first_block_no(sb, bg));
2688 }
2689
2690 /**
2691  * ext4_get_stripe_size: Get the stripe size.
2692  * @sbi: In memory super block info
2693  *
2694  * If we have specified it via mount option, then
2695  * use the mount option value. If the value specified at mount time is
2696  * greater than the blocks per group use the super block value.
2697  * If the super block value is greater than blocks per group return 0.
2698  * Allocator needs it be less than blocks per group.
2699  *
2700  */
2701 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2702 {
2703         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2704         unsigned long stripe_width =
2705                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2706         int ret;
2707
2708         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2709                 ret = sbi->s_stripe;
2710         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2711                 ret = stripe_width;
2712         else if (stride && stride <= sbi->s_blocks_per_group)
2713                 ret = stride;
2714         else
2715                 ret = 0;
2716
2717         /*
2718          * If the stripe width is 1, this makes no sense and
2719          * we set it to 0 to turn off stripe handling code.
2720          */
2721         if (ret <= 1)
2722                 ret = 0;
2723
2724         return ret;
2725 }
2726
2727 /*
2728  * Check whether this filesystem can be mounted based on
2729  * the features present and the RDONLY/RDWR mount requested.
2730  * Returns 1 if this filesystem can be mounted as requested,
2731  * 0 if it cannot be.
2732  */
2733 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2734 {
2735         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2736                 ext4_msg(sb, KERN_ERR,
2737                         "Couldn't mount because of "
2738                         "unsupported optional features (%x)",
2739                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2740                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2741                 return 0;
2742         }
2743
2744         if (readonly)
2745                 return 1;
2746
2747         if (ext4_has_feature_readonly(sb)) {
2748                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2749                 sb->s_flags |= MS_RDONLY;
2750                 return 1;
2751         }
2752
2753         /* Check that feature set is OK for a read-write mount */
2754         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2755                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2756                          "unsupported optional features (%x)",
2757                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2758                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2759                 return 0;
2760         }
2761         /*
2762          * Large file size enabled file system can only be mounted
2763          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2764          */
2765         if (ext4_has_feature_huge_file(sb)) {
2766                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2767                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2768                                  "cannot be mounted RDWR without "
2769                                  "CONFIG_LBDAF");
2770                         return 0;
2771                 }
2772         }
2773         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2774                 ext4_msg(sb, KERN_ERR,
2775                          "Can't support bigalloc feature without "
2776                          "extents feature\n");
2777                 return 0;
2778         }
2779
2780 #ifndef CONFIG_QUOTA
2781         if (ext4_has_feature_quota(sb) && !readonly) {
2782                 ext4_msg(sb, KERN_ERR,
2783                          "Filesystem with quota feature cannot be mounted RDWR "
2784                          "without CONFIG_QUOTA");
2785                 return 0;
2786         }
2787         if (ext4_has_feature_project(sb) && !readonly) {
2788                 ext4_msg(sb, KERN_ERR,
2789                          "Filesystem with project quota feature cannot be mounted RDWR "
2790                          "without CONFIG_QUOTA");
2791                 return 0;
2792         }
2793 #endif  /* CONFIG_QUOTA */
2794         return 1;
2795 }
2796
2797 /*
2798  * This function is called once a day if we have errors logged
2799  * on the file system
2800  */
2801 static void print_daily_error_info(unsigned long arg)
2802 {
2803         struct super_block *sb = (struct super_block *) arg;
2804         struct ext4_sb_info *sbi;
2805         struct ext4_super_block *es;
2806
2807         sbi = EXT4_SB(sb);
2808         es = sbi->s_es;
2809
2810         if (es->s_error_count)
2811                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2812                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2813                          le32_to_cpu(es->s_error_count));
2814         if (es->s_first_error_time) {
2815                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2816                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2817                        (int) sizeof(es->s_first_error_func),
2818                        es->s_first_error_func,
2819                        le32_to_cpu(es->s_first_error_line));
2820                 if (es->s_first_error_ino)
2821                         printk(KERN_CONT ": inode %u",
2822                                le32_to_cpu(es->s_first_error_ino));
2823                 if (es->s_first_error_block)
2824                         printk(KERN_CONT ": block %llu", (unsigned long long)
2825                                le64_to_cpu(es->s_first_error_block));
2826                 printk(KERN_CONT "\n");
2827         }
2828         if (es->s_last_error_time) {
2829                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2830                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2831                        (int) sizeof(es->s_last_error_func),
2832                        es->s_last_error_func,
2833                        le32_to_cpu(es->s_last_error_line));
2834                 if (es->s_last_error_ino)
2835                         printk(KERN_CONT ": inode %u",
2836                                le32_to_cpu(es->s_last_error_ino));
2837                 if (es->s_last_error_block)
2838                         printk(KERN_CONT ": block %llu", (unsigned long long)
2839                                le64_to_cpu(es->s_last_error_block));
2840                 printk(KERN_CONT "\n");
2841         }
2842         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2843 }
2844
2845 /* Find next suitable group and run ext4_init_inode_table */
2846 static int ext4_run_li_request(struct ext4_li_request *elr)
2847 {
2848         struct ext4_group_desc *gdp = NULL;
2849         ext4_group_t group, ngroups;
2850         struct super_block *sb;
2851         unsigned long timeout = 0;
2852         int ret = 0;
2853
2854         sb = elr->lr_super;
2855         ngroups = EXT4_SB(sb)->s_groups_count;
2856
2857         for (group = elr->lr_next_group; group < ngroups; group++) {
2858                 gdp = ext4_get_group_desc(sb, group, NULL);
2859                 if (!gdp) {
2860                         ret = 1;
2861                         break;
2862                 }
2863
2864                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2865                         break;
2866         }
2867
2868         if (group >= ngroups)
2869                 ret = 1;
2870
2871         if (!ret) {
2872                 timeout = jiffies;
2873                 ret = ext4_init_inode_table(sb, group,
2874                                             elr->lr_timeout ? 0 : 1);
2875                 if (elr->lr_timeout == 0) {
2876                         timeout = (jiffies - timeout) *
2877                                   elr->lr_sbi->s_li_wait_mult;
2878                         elr->lr_timeout = timeout;
2879                 }
2880                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2881                 elr->lr_next_group = group + 1;
2882         }
2883         return ret;
2884 }
2885
2886 /*
2887  * Remove lr_request from the list_request and free the
2888  * request structure. Should be called with li_list_mtx held
2889  */
2890 static void ext4_remove_li_request(struct ext4_li_request *elr)
2891 {
2892         struct ext4_sb_info *sbi;
2893
2894         if (!elr)
2895                 return;
2896
2897         sbi = elr->lr_sbi;
2898
2899         list_del(&elr->lr_request);
2900         sbi->s_li_request = NULL;
2901         kfree(elr);
2902 }
2903
2904 static void ext4_unregister_li_request(struct super_block *sb)
2905 {
2906         mutex_lock(&ext4_li_mtx);
2907         if (!ext4_li_info) {
2908                 mutex_unlock(&ext4_li_mtx);
2909                 return;
2910         }
2911
2912         mutex_lock(&ext4_li_info->li_list_mtx);
2913         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2914         mutex_unlock(&ext4_li_info->li_list_mtx);
2915         mutex_unlock(&ext4_li_mtx);
2916 }
2917
2918 static struct task_struct *ext4_lazyinit_task;
2919
2920 /*
2921  * This is the function where ext4lazyinit thread lives. It walks
2922  * through the request list searching for next scheduled filesystem.
2923  * When such a fs is found, run the lazy initialization request
2924  * (ext4_rn_li_request) and keep track of the time spend in this
2925  * function. Based on that time we compute next schedule time of
2926  * the request. When walking through the list is complete, compute
2927  * next waking time and put itself into sleep.
2928  */
2929 static int ext4_lazyinit_thread(void *arg)
2930 {
2931         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2932         struct list_head *pos, *n;
2933         struct ext4_li_request *elr;
2934         unsigned long next_wakeup, cur;
2935
2936         BUG_ON(NULL == eli);
2937
2938 cont_thread:
2939         while (true) {
2940                 next_wakeup = MAX_JIFFY_OFFSET;
2941
2942                 mutex_lock(&eli->li_list_mtx);
2943                 if (list_empty(&eli->li_request_list)) {
2944                         mutex_unlock(&eli->li_list_mtx);
2945                         goto exit_thread;
2946                 }
2947                 list_for_each_safe(pos, n, &eli->li_request_list) {
2948                         int err = 0;
2949                         int progress = 0;
2950                         elr = list_entry(pos, struct ext4_li_request,
2951                                          lr_request);
2952
2953                         if (time_before(jiffies, elr->lr_next_sched)) {
2954                                 if (time_before(elr->lr_next_sched, next_wakeup))
2955                                         next_wakeup = elr->lr_next_sched;
2956                                 continue;
2957                         }
2958                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2959                                 if (sb_start_write_trylock(elr->lr_super)) {
2960                                         progress = 1;
2961                                         /*
2962                                          * We hold sb->s_umount, sb can not
2963                                          * be removed from the list, it is
2964                                          * now safe to drop li_list_mtx
2965                                          */
2966                                         mutex_unlock(&eli->li_list_mtx);
2967                                         err = ext4_run_li_request(elr);
2968                                         sb_end_write(elr->lr_super);
2969                                         mutex_lock(&eli->li_list_mtx);
2970                                         n = pos->next;
2971                                 }
2972                                 up_read((&elr->lr_super->s_umount));
2973                         }
2974                         /* error, remove the lazy_init job */
2975                         if (err) {
2976                                 ext4_remove_li_request(elr);
2977                                 continue;
2978                         }
2979                         if (!progress) {
2980                                 elr->lr_next_sched = jiffies +
2981                                         (prandom_u32()
2982                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2983                         }
2984                         if (time_before(elr->lr_next_sched, next_wakeup))
2985                                 next_wakeup = elr->lr_next_sched;
2986                 }
2987                 mutex_unlock(&eli->li_list_mtx);
2988
2989                 try_to_freeze();
2990
2991                 cur = jiffies;
2992                 if ((time_after_eq(cur, next_wakeup)) ||
2993                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2994                         cond_resched();
2995                         continue;
2996                 }
2997
2998                 schedule_timeout_interruptible(next_wakeup - cur);
2999
3000                 if (kthread_should_stop()) {
3001                         ext4_clear_request_list();
3002                         goto exit_thread;
3003                 }
3004         }
3005
3006 exit_thread:
3007         /*
3008          * It looks like the request list is empty, but we need
3009          * to check it under the li_list_mtx lock, to prevent any
3010          * additions into it, and of course we should lock ext4_li_mtx
3011          * to atomically free the list and ext4_li_info, because at
3012          * this point another ext4 filesystem could be registering
3013          * new one.
3014          */
3015         mutex_lock(&ext4_li_mtx);
3016         mutex_lock(&eli->li_list_mtx);
3017         if (!list_empty(&eli->li_request_list)) {
3018                 mutex_unlock(&eli->li_list_mtx);
3019                 mutex_unlock(&ext4_li_mtx);
3020                 goto cont_thread;
3021         }
3022         mutex_unlock(&eli->li_list_mtx);
3023         kfree(ext4_li_info);
3024         ext4_li_info = NULL;
3025         mutex_unlock(&ext4_li_mtx);
3026
3027         return 0;
3028 }
3029
3030 static void ext4_clear_request_list(void)
3031 {
3032         struct list_head *pos, *n;
3033         struct ext4_li_request *elr;
3034
3035         mutex_lock(&ext4_li_info->li_list_mtx);
3036         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3037                 elr = list_entry(pos, struct ext4_li_request,
3038                                  lr_request);
3039                 ext4_remove_li_request(elr);
3040         }
3041         mutex_unlock(&ext4_li_info->li_list_mtx);
3042 }
3043
3044 static int ext4_run_lazyinit_thread(void)
3045 {
3046         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3047                                          ext4_li_info, "ext4lazyinit");
3048         if (IS_ERR(ext4_lazyinit_task)) {
3049                 int err = PTR_ERR(ext4_lazyinit_task);
3050                 ext4_clear_request_list();
3051                 kfree(ext4_li_info);
3052                 ext4_li_info = NULL;
3053                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3054                                  "initialization thread\n",
3055                                  err);
3056                 return err;
3057         }
3058         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3059         return 0;
3060 }
3061
3062 /*
3063  * Check whether it make sense to run itable init. thread or not.
3064  * If there is at least one uninitialized inode table, return
3065  * corresponding group number, else the loop goes through all
3066  * groups and return total number of groups.
3067  */
3068 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3069 {
3070         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3071         struct ext4_group_desc *gdp = NULL;
3072
3073         for (group = 0; group < ngroups; group++) {
3074                 gdp = ext4_get_group_desc(sb, group, NULL);
3075                 if (!gdp)
3076                         continue;
3077
3078                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3079                         break;
3080         }
3081
3082         return group;
3083 }
3084
3085 static int ext4_li_info_new(void)
3086 {
3087         struct ext4_lazy_init *eli = NULL;
3088
3089         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3090         if (!eli)
3091                 return -ENOMEM;
3092
3093         INIT_LIST_HEAD(&eli->li_request_list);
3094         mutex_init(&eli->li_list_mtx);
3095
3096         eli->li_state |= EXT4_LAZYINIT_QUIT;
3097
3098         ext4_li_info = eli;
3099
3100         return 0;
3101 }
3102
3103 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3104                                             ext4_group_t start)
3105 {
3106         struct ext4_sb_info *sbi = EXT4_SB(sb);
3107         struct ext4_li_request *elr;
3108
3109         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3110         if (!elr)
3111                 return NULL;
3112
3113         elr->lr_super = sb;
3114         elr->lr_sbi = sbi;
3115         elr->lr_next_group = start;
3116
3117         /*
3118          * Randomize first schedule time of the request to
3119          * spread the inode table initialization requests
3120          * better.
3121          */
3122         elr->lr_next_sched = jiffies + (prandom_u32() %
3123                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3124         return elr;
3125 }
3126
3127 int ext4_register_li_request(struct super_block *sb,
3128                              ext4_group_t first_not_zeroed)
3129 {
3130         struct ext4_sb_info *sbi = EXT4_SB(sb);
3131         struct ext4_li_request *elr = NULL;
3132         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3133         int ret = 0;
3134
3135         mutex_lock(&ext4_li_mtx);
3136         if (sbi->s_li_request != NULL) {
3137                 /*
3138                  * Reset timeout so it can be computed again, because
3139                  * s_li_wait_mult might have changed.
3140                  */
3141                 sbi->s_li_request->lr_timeout = 0;
3142                 goto out;
3143         }
3144
3145         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3146             !test_opt(sb, INIT_INODE_TABLE))
3147                 goto out;
3148
3149         elr = ext4_li_request_new(sb, first_not_zeroed);
3150         if (!elr) {
3151                 ret = -ENOMEM;
3152                 goto out;
3153         }
3154
3155         if (NULL == ext4_li_info) {
3156                 ret = ext4_li_info_new();
3157                 if (ret)
3158                         goto out;
3159         }
3160
3161         mutex_lock(&ext4_li_info->li_list_mtx);
3162         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3163         mutex_unlock(&ext4_li_info->li_list_mtx);
3164
3165         sbi->s_li_request = elr;
3166         /*
3167          * set elr to NULL here since it has been inserted to
3168          * the request_list and the removal and free of it is
3169          * handled by ext4_clear_request_list from now on.
3170          */
3171         elr = NULL;
3172
3173         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3174                 ret = ext4_run_lazyinit_thread();
3175                 if (ret)
3176                         goto out;
3177         }
3178 out:
3179         mutex_unlock(&ext4_li_mtx);
3180         if (ret)
3181                 kfree(elr);
3182         return ret;
3183 }
3184
3185 /*
3186  * We do not need to lock anything since this is called on
3187  * module unload.
3188  */
3189 static void ext4_destroy_lazyinit_thread(void)
3190 {
3191         /*
3192          * If thread exited earlier
3193          * there's nothing to be done.
3194          */
3195         if (!ext4_li_info || !ext4_lazyinit_task)
3196                 return;
3197
3198         kthread_stop(ext4_lazyinit_task);
3199 }
3200
3201 static int set_journal_csum_feature_set(struct super_block *sb)
3202 {
3203         int ret = 1;
3204         int compat, incompat;
3205         struct ext4_sb_info *sbi = EXT4_SB(sb);
3206
3207         if (ext4_has_metadata_csum(sb)) {
3208                 /* journal checksum v3 */
3209                 compat = 0;
3210                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3211         } else {
3212                 /* journal checksum v1 */
3213                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3214                 incompat = 0;
3215         }
3216
3217         jbd2_journal_clear_features(sbi->s_journal,
3218                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3219                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3220                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3221         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3222                 ret = jbd2_journal_set_features(sbi->s_journal,
3223                                 compat, 0,
3224                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3225                                 incompat);
3226         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3227                 ret = jbd2_journal_set_features(sbi->s_journal,
3228                                 compat, 0,
3229                                 incompat);
3230                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3231                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3232         } else {
3233                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3234                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3235         }
3236
3237         return ret;
3238 }
3239
3240 /*
3241  * Note: calculating the overhead so we can be compatible with
3242  * historical BSD practice is quite difficult in the face of
3243  * clusters/bigalloc.  This is because multiple metadata blocks from
3244  * different block group can end up in the same allocation cluster.
3245  * Calculating the exact overhead in the face of clustered allocation
3246  * requires either O(all block bitmaps) in memory or O(number of block
3247  * groups**2) in time.  We will still calculate the superblock for
3248  * older file systems --- and if we come across with a bigalloc file
3249  * system with zero in s_overhead_clusters the estimate will be close to
3250  * correct especially for very large cluster sizes --- but for newer
3251  * file systems, it's better to calculate this figure once at mkfs
3252  * time, and store it in the superblock.  If the superblock value is
3253  * present (even for non-bigalloc file systems), we will use it.
3254  */
3255 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3256                           char *buf)
3257 {
3258         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3259         struct ext4_group_desc  *gdp;
3260         ext4_fsblk_t            first_block, last_block, b;
3261         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3262         int                     s, j, count = 0;
3263
3264         if (!ext4_has_feature_bigalloc(sb))
3265                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3266                         sbi->s_itb_per_group + 2);
3267
3268         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3269                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3270         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3271         for (i = 0; i < ngroups; i++) {
3272                 gdp = ext4_get_group_desc(sb, i, NULL);
3273                 b = ext4_block_bitmap(sb, gdp);
3274                 if (b >= first_block && b <= last_block) {
3275                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3276                         count++;
3277                 }
3278                 b = ext4_inode_bitmap(sb, gdp);
3279                 if (b >= first_block && b <= last_block) {
3280                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3281                         count++;
3282                 }
3283                 b = ext4_inode_table(sb, gdp);
3284                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3285                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3286                                 int c = EXT4_B2C(sbi, b - first_block);
3287                                 ext4_set_bit(c, buf);
3288                                 count++;
3289                         }
3290                 if (i != grp)
3291                         continue;
3292                 s = 0;
3293                 if (ext4_bg_has_super(sb, grp)) {
3294                         ext4_set_bit(s++, buf);
3295                         count++;
3296                 }
3297                 j = ext4_bg_num_gdb(sb, grp);
3298                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3299                         ext4_error(sb, "Invalid number of block group "
3300                                    "descriptor blocks: %d", j);
3301                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3302                 }
3303                 count += j;
3304                 for (; j > 0; j--)
3305                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3306         }
3307         if (!count)
3308                 return 0;
3309         return EXT4_CLUSTERS_PER_GROUP(sb) -
3310                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3311 }
3312
3313 /*
3314  * Compute the overhead and stash it in sbi->s_overhead
3315  */
3316 int ext4_calculate_overhead(struct super_block *sb)
3317 {
3318         struct ext4_sb_info *sbi = EXT4_SB(sb);
3319         struct ext4_super_block *es = sbi->s_es;
3320         struct inode *j_inode;
3321         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3322         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3323         ext4_fsblk_t overhead = 0;
3324         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3325
3326         if (!buf)
3327                 return -ENOMEM;
3328
3329         /*
3330          * Compute the overhead (FS structures).  This is constant
3331          * for a given filesystem unless the number of block groups
3332          * changes so we cache the previous value until it does.
3333          */
3334
3335         /*
3336          * All of the blocks before first_data_block are overhead
3337          */
3338         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3339
3340         /*
3341          * Add the overhead found in each block group
3342          */
3343         for (i = 0; i < ngroups; i++) {
3344                 int blks;
3345
3346                 blks = count_overhead(sb, i, buf);
3347                 overhead += blks;
3348                 if (blks)
3349                         memset(buf, 0, PAGE_SIZE);
3350                 cond_resched();
3351         }
3352
3353         /*
3354          * Add the internal journal blocks whether the journal has been
3355          * loaded or not
3356          */
3357         if (sbi->s_journal && !sbi->journal_bdev)
3358                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3359         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3360                 j_inode = ext4_get_journal_inode(sb, j_inum);
3361                 if (j_inode) {
3362                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3363                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3364                         iput(j_inode);
3365                 } else {
3366                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3367                 }
3368         }
3369         sbi->s_overhead = overhead;
3370         smp_wmb();
3371         free_page((unsigned long) buf);
3372         return 0;
3373 }
3374
3375 static void ext4_set_resv_clusters(struct super_block *sb)
3376 {
3377         ext4_fsblk_t resv_clusters;
3378         struct ext4_sb_info *sbi = EXT4_SB(sb);
3379
3380         /*
3381          * There's no need to reserve anything when we aren't using extents.
3382          * The space estimates are exact, there are no unwritten extents,
3383          * hole punching doesn't need new metadata... This is needed especially
3384          * to keep ext2/3 backward compatibility.
3385          */
3386         if (!ext4_has_feature_extents(sb))
3387                 return;
3388         /*
3389          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3390          * This should cover the situations where we can not afford to run
3391          * out of space like for example punch hole, or converting
3392          * unwritten extents in delalloc path. In most cases such
3393          * allocation would require 1, or 2 blocks, higher numbers are
3394          * very rare.
3395          */
3396         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3397                          sbi->s_cluster_bits);
3398
3399         do_div(resv_clusters, 50);
3400         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3401
3402         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3403 }
3404
3405 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3406 {
3407         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3408         char *orig_data = kstrdup(data, GFP_KERNEL);
3409         struct buffer_head *bh;
3410         struct ext4_super_block *es = NULL;
3411         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3412         ext4_fsblk_t block;
3413         ext4_fsblk_t sb_block = get_sb_block(&data);
3414         ext4_fsblk_t logical_sb_block;
3415         unsigned long offset = 0;
3416         unsigned long journal_devnum = 0;
3417         unsigned long def_mount_opts;
3418         struct inode *root;
3419         const char *descr;
3420         int ret = -ENOMEM;
3421         int blocksize, clustersize;
3422         unsigned int db_count;
3423         unsigned int i;
3424         int needs_recovery, has_huge_files, has_bigalloc;
3425         __u64 blocks_count;
3426         int err = 0;
3427         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3428         ext4_group_t first_not_zeroed;
3429
3430         if ((data && !orig_data) || !sbi)
3431                 goto out_free_base;
3432
3433         sbi->s_daxdev = dax_dev;
3434         sbi->s_blockgroup_lock =
3435                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3436         if (!sbi->s_blockgroup_lock)
3437                 goto out_free_base;
3438
3439         sb->s_fs_info = sbi;
3440         sbi->s_sb = sb;
3441         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3442         sbi->s_sb_block = sb_block;
3443         if (sb->s_bdev->bd_part)
3444                 sbi->s_sectors_written_start =
3445                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3446
3447         /* Cleanup superblock name */
3448         strreplace(sb->s_id, '/', '!');
3449
3450         /* -EINVAL is default */
3451         ret = -EINVAL;
3452         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3453         if (!blocksize) {
3454                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3455                 goto out_fail;
3456         }
3457
3458         /*
3459          * The ext4 superblock will not be buffer aligned for other than 1kB
3460          * block sizes.  We need to calculate the offset from buffer start.
3461          */
3462         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3463                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3464                 offset = do_div(logical_sb_block, blocksize);
3465         } else {
3466                 logical_sb_block = sb_block;
3467         }
3468
3469         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3470                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3471                 goto out_fail;
3472         }
3473         /*
3474          * Note: s_es must be initialized as soon as possible because
3475          *       some ext4 macro-instructions depend on its value
3476          */
3477         es = (struct ext4_super_block *) (bh->b_data + offset);
3478         sbi->s_es = es;
3479         sb->s_magic = le16_to_cpu(es->s_magic);
3480         if (sb->s_magic != EXT4_SUPER_MAGIC)
3481                 goto cantfind_ext4;
3482         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3483
3484         /* Warn if metadata_csum and gdt_csum are both set. */
3485         if (ext4_has_feature_metadata_csum(sb) &&
3486             ext4_has_feature_gdt_csum(sb))
3487                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3488                              "redundant flags; please run fsck.");
3489
3490         /* Check for a known checksum algorithm */
3491         if (!ext4_verify_csum_type(sb, es)) {
3492                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3493                          "unknown checksum algorithm.");
3494                 silent = 1;
3495                 goto cantfind_ext4;
3496         }
3497
3498         /* Load the checksum driver */
3499         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3500         if (IS_ERR(sbi->s_chksum_driver)) {
3501                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3502                 ret = PTR_ERR(sbi->s_chksum_driver);
3503                 sbi->s_chksum_driver = NULL;
3504                 goto failed_mount;
3505         }
3506
3507         /* Check superblock checksum */
3508         if (!ext4_superblock_csum_verify(sb, es)) {
3509                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3510                          "invalid superblock checksum.  Run e2fsck?");
3511                 silent = 1;
3512                 ret = -EFSBADCRC;
3513                 goto cantfind_ext4;
3514         }
3515
3516         /* Precompute checksum seed for all metadata */
3517         if (ext4_has_feature_csum_seed(sb))
3518                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3519         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3520                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3521                                                sizeof(es->s_uuid));
3522
3523         /* Set defaults before we parse the mount options */
3524         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3525         set_opt(sb, INIT_INODE_TABLE);
3526         if (def_mount_opts & EXT4_DEFM_DEBUG)
3527                 set_opt(sb, DEBUG);
3528         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3529                 set_opt(sb, GRPID);
3530         if (def_mount_opts & EXT4_DEFM_UID16)
3531                 set_opt(sb, NO_UID32);
3532         /* xattr user namespace & acls are now defaulted on */
3533         set_opt(sb, XATTR_USER);
3534 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3535         set_opt(sb, POSIX_ACL);
3536 #endif
3537         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3538         if (ext4_has_metadata_csum(sb))
3539                 set_opt(sb, JOURNAL_CHECKSUM);
3540
3541         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3542                 set_opt(sb, JOURNAL_DATA);
3543         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3544                 set_opt(sb, ORDERED_DATA);
3545         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3546                 set_opt(sb, WRITEBACK_DATA);
3547
3548         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3549                 set_opt(sb, ERRORS_PANIC);
3550         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3551                 set_opt(sb, ERRORS_CONT);
3552         else
3553                 set_opt(sb, ERRORS_RO);
3554         /* block_validity enabled by default; disable with noblock_validity */
3555         set_opt(sb, BLOCK_VALIDITY);
3556         if (def_mount_opts & EXT4_DEFM_DISCARD)
3557                 set_opt(sb, DISCARD);
3558
3559         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3560         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3561         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3562         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3563         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3564
3565         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3566                 set_opt(sb, BARRIER);
3567
3568         /*
3569          * enable delayed allocation by default
3570          * Use -o nodelalloc to turn it off
3571          */
3572         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3573             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3574                 set_opt(sb, DELALLOC);
3575
3576         /*
3577          * set default s_li_wait_mult for lazyinit, for the case there is
3578          * no mount option specified.
3579          */
3580         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3581
3582         if (sbi->s_es->s_mount_opts[0]) {
3583                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3584                                               sizeof(sbi->s_es->s_mount_opts),
3585                                               GFP_KERNEL);
3586                 if (!s_mount_opts)
3587                         goto failed_mount;
3588                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3589                                    &journal_ioprio, 0)) {
3590                         ext4_msg(sb, KERN_WARNING,
3591                                  "failed to parse options in superblock: %s",
3592                                  s_mount_opts);
3593                 }
3594                 kfree(s_mount_opts);
3595         }
3596         sbi->s_def_mount_opt = sbi->s_mount_opt;
3597         if (!parse_options((char *) data, sb, &journal_devnum,
3598                            &journal_ioprio, 0))
3599                 goto failed_mount;
3600
3601         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3602                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3603                             "with data=journal disables delayed "
3604                             "allocation and O_DIRECT support!\n");
3605                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3606                         ext4_msg(sb, KERN_ERR, "can't mount with "
3607                                  "both data=journal and delalloc");
3608                         goto failed_mount;
3609                 }
3610                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3611                         ext4_msg(sb, KERN_ERR, "can't mount with "
3612                                  "both data=journal and dioread_nolock");
3613                         goto failed_mount;
3614                 }
3615                 if (test_opt(sb, DAX)) {
3616                         ext4_msg(sb, KERN_ERR, "can't mount with "
3617                                  "both data=journal and dax");
3618                         goto failed_mount;
3619                 }
3620                 if (ext4_has_feature_encrypt(sb)) {
3621                         ext4_msg(sb, KERN_WARNING,
3622                                  "encrypted files will use data=ordered "
3623                                  "instead of data journaling mode");
3624                 }
3625                 if (test_opt(sb, DELALLOC))
3626                         clear_opt(sb, DELALLOC);
3627         } else {
3628                 sb->s_iflags |= SB_I_CGROUPWB;
3629         }
3630
3631         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3632                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3633
3634         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3635             (ext4_has_compat_features(sb) ||
3636              ext4_has_ro_compat_features(sb) ||
3637              ext4_has_incompat_features(sb)))
3638                 ext4_msg(sb, KERN_WARNING,
3639                        "feature flags set on rev 0 fs, "
3640                        "running e2fsck is recommended");
3641
3642         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3643                 set_opt2(sb, HURD_COMPAT);
3644                 if (ext4_has_feature_64bit(sb)) {
3645                         ext4_msg(sb, KERN_ERR,
3646                                  "The Hurd can't support 64-bit file systems");
3647                         goto failed_mount;
3648                 }
3649
3650                 /*
3651                  * ea_inode feature uses l_i_version field which is not
3652                  * available in HURD_COMPAT mode.
3653                  */
3654                 if (ext4_has_feature_ea_inode(sb)) {
3655                         ext4_msg(sb, KERN_ERR,
3656                                  "ea_inode feature is not supported for Hurd");
3657                         goto failed_mount;
3658                 }
3659         }
3660
3661         if (IS_EXT2_SB(sb)) {
3662                 if (ext2_feature_set_ok(sb))
3663                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3664                                  "using the ext4 subsystem");
3665                 else {
3666                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3667                                  "to feature incompatibilities");
3668                         goto failed_mount;
3669                 }
3670         }
3671
3672         if (IS_EXT3_SB(sb)) {
3673                 if (ext3_feature_set_ok(sb))
3674                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3675                                  "using the ext4 subsystem");
3676                 else {
3677                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3678                                  "to feature incompatibilities");
3679                         goto failed_mount;
3680                 }
3681         }
3682
3683         /*
3684          * Check feature flags regardless of the revision level, since we
3685          * previously didn't change the revision level when setting the flags,
3686          * so there is a chance incompat flags are set on a rev 0 filesystem.
3687          */
3688         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3689                 goto failed_mount;
3690
3691         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3692         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3693             blocksize > EXT4_MAX_BLOCK_SIZE) {
3694                 ext4_msg(sb, KERN_ERR,
3695                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3696                          blocksize, le32_to_cpu(es->s_log_block_size));
3697                 goto failed_mount;
3698         }
3699         if (le32_to_cpu(es->s_log_block_size) >
3700             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3701                 ext4_msg(sb, KERN_ERR,
3702                          "Invalid log block size: %u",
3703                          le32_to_cpu(es->s_log_block_size));
3704                 goto failed_mount;
3705         }
3706
3707         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3708                 ext4_msg(sb, KERN_ERR,
3709                          "Number of reserved GDT blocks insanely large: %d",
3710                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3711                 goto failed_mount;
3712         }
3713
3714         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3715                 if (ext4_has_feature_inline_data(sb)) {
3716                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3717                                         " that may contain inline data");
3718                         goto failed_mount;
3719                 }
3720                 err = bdev_dax_supported(sb, blocksize);
3721                 if (err)
3722                         goto failed_mount;
3723         }
3724
3725         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3726                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3727                          es->s_encryption_level);
3728                 goto failed_mount;
3729         }
3730
3731         if (sb->s_blocksize != blocksize) {
3732                 /* Validate the filesystem blocksize */
3733                 if (!sb_set_blocksize(sb, blocksize)) {
3734                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3735                                         blocksize);
3736                         goto failed_mount;
3737                 }
3738
3739                 brelse(bh);
3740                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3741                 offset = do_div(logical_sb_block, blocksize);
3742                 bh = sb_bread_unmovable(sb, logical_sb_block);
3743                 if (!bh) {
3744                         ext4_msg(sb, KERN_ERR,
3745                                "Can't read superblock on 2nd try");
3746                         goto failed_mount;
3747                 }
3748                 es = (struct ext4_super_block *)(bh->b_data + offset);
3749                 sbi->s_es = es;
3750                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3751                         ext4_msg(sb, KERN_ERR,
3752                                "Magic mismatch, very weird!");
3753                         goto failed_mount;
3754                 }
3755         }
3756
3757         has_huge_files = ext4_has_feature_huge_file(sb);
3758         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3759                                                       has_huge_files);
3760         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3761
3762         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3763                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3764                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3765         } else {
3766                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3767                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3768                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3769                     (!is_power_of_2(sbi->s_inode_size)) ||
3770                     (sbi->s_inode_size > blocksize)) {
3771                         ext4_msg(sb, KERN_ERR,
3772                                "unsupported inode size: %d",
3773                                sbi->s_inode_size);
3774                         goto failed_mount;
3775                 }
3776                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3777                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3778         }
3779
3780         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3781         if (ext4_has_feature_64bit(sb)) {
3782                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3783                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3784                     !is_power_of_2(sbi->s_desc_size)) {
3785                         ext4_msg(sb, KERN_ERR,
3786                                "unsupported descriptor size %lu",
3787                                sbi->s_desc_size);
3788                         goto failed_mount;
3789                 }
3790         } else
3791                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3792
3793         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3794         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3795
3796         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3797         if (sbi->s_inodes_per_block == 0)
3798                 goto cantfind_ext4;
3799         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3800             sbi->s_inodes_per_group > blocksize * 8) {
3801                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3802                          sbi->s_blocks_per_group);
3803                 goto failed_mount;
3804         }
3805         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3806                                         sbi->s_inodes_per_block;
3807         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3808         sbi->s_sbh = bh;
3809         sbi->s_mount_state = le16_to_cpu(es->s_state);
3810         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3811         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3812
3813         for (i = 0; i < 4; i++)
3814                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3815         sbi->s_def_hash_version = es->s_def_hash_version;
3816         if (ext4_has_feature_dir_index(sb)) {
3817                 i = le32_to_cpu(es->s_flags);
3818                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3819                         sbi->s_hash_unsigned = 3;
3820                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3821 #ifdef __CHAR_UNSIGNED__
3822                         if (!sb_rdonly(sb))
3823                                 es->s_flags |=
3824                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3825                         sbi->s_hash_unsigned = 3;
3826 #else
3827                         if (!sb_rdonly(sb))
3828                                 es->s_flags |=
3829                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3830 #endif
3831                 }
3832         }
3833
3834         /* Handle clustersize */
3835         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3836         has_bigalloc = ext4_has_feature_bigalloc(sb);
3837         if (has_bigalloc) {
3838                 if (clustersize < blocksize) {
3839                         ext4_msg(sb, KERN_ERR,
3840                                  "cluster size (%d) smaller than "
3841                                  "block size (%d)", clustersize, blocksize);
3842                         goto failed_mount;
3843                 }
3844                 if (le32_to_cpu(es->s_log_cluster_size) >
3845                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3846                         ext4_msg(sb, KERN_ERR,
3847                                  "Invalid log cluster size: %u",
3848                                  le32_to_cpu(es->s_log_cluster_size));
3849                         goto failed_mount;
3850                 }
3851                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3852                         le32_to_cpu(es->s_log_block_size);
3853                 sbi->s_clusters_per_group =
3854                         le32_to_cpu(es->s_clusters_per_group);
3855                 if (sbi->s_clusters_per_group > blocksize * 8) {
3856                         ext4_msg(sb, KERN_ERR,
3857                                  "#clusters per group too big: %lu",
3858                                  sbi->s_clusters_per_group);
3859                         goto failed_mount;
3860                 }
3861                 if (sbi->s_blocks_per_group !=
3862                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3863                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3864                                  "clusters per group (%lu) inconsistent",
3865                                  sbi->s_blocks_per_group,
3866                                  sbi->s_clusters_per_group);
3867                         goto failed_mount;
3868                 }
3869         } else {
3870                 if (clustersize != blocksize) {
3871                         ext4_warning(sb, "fragment/cluster size (%d) != "
3872                                      "block size (%d)", clustersize,
3873                                      blocksize);
3874                         clustersize = blocksize;
3875                 }
3876                 if (sbi->s_blocks_per_group > blocksize * 8) {
3877                         ext4_msg(sb, KERN_ERR,
3878                                  "#blocks per group too big: %lu",
3879                                  sbi->s_blocks_per_group);
3880                         goto failed_mount;
3881                 }
3882                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3883                 sbi->s_cluster_bits = 0;
3884         }
3885         sbi->s_cluster_ratio = clustersize / blocksize;
3886
3887         /* Do we have standard group size of clustersize * 8 blocks ? */
3888         if (sbi->s_blocks_per_group == clustersize << 3)
3889                 set_opt2(sb, STD_GROUP_SIZE);
3890
3891         /*
3892          * Test whether we have more sectors than will fit in sector_t,
3893          * and whether the max offset is addressable by the page cache.
3894          */
3895         err = generic_check_addressable(sb->s_blocksize_bits,
3896                                         ext4_blocks_count(es));
3897         if (err) {
3898                 ext4_msg(sb, KERN_ERR, "filesystem"
3899                          " too large to mount safely on this system");
3900                 if (sizeof(sector_t) < 8)
3901                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3902                 goto failed_mount;
3903         }
3904
3905         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3906                 goto cantfind_ext4;
3907
3908         /* check blocks count against device size */
3909         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3910         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3911                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3912                        "exceeds size of device (%llu blocks)",
3913                        ext4_blocks_count(es), blocks_count);
3914                 goto failed_mount;
3915         }
3916
3917         /*
3918          * It makes no sense for the first data block to be beyond the end
3919          * of the filesystem.
3920          */
3921         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3922                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3923                          "block %u is beyond end of filesystem (%llu)",
3924                          le32_to_cpu(es->s_first_data_block),
3925                          ext4_blocks_count(es));
3926                 goto failed_mount;
3927         }
3928         blocks_count = (ext4_blocks_count(es) -
3929                         le32_to_cpu(es->s_first_data_block) +
3930                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3931         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3932         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3933                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3934                        "(block count %llu, first data block %u, "
3935                        "blocks per group %lu)", sbi->s_groups_count,
3936                        ext4_blocks_count(es),
3937                        le32_to_cpu(es->s_first_data_block),
3938                        EXT4_BLOCKS_PER_GROUP(sb));
3939                 goto failed_mount;
3940         }
3941         sbi->s_groups_count = blocks_count;
3942         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3943                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3944         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3945                    EXT4_DESC_PER_BLOCK(sb);
3946         if (ext4_has_feature_meta_bg(sb)) {
3947                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3948                         ext4_msg(sb, KERN_WARNING,
3949                                  "first meta block group too large: %u "
3950                                  "(group descriptor block count %u)",
3951                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3952                         goto failed_mount;
3953                 }
3954         }
3955         sbi->s_group_desc = kvmalloc(db_count *
3956                                           sizeof(struct buffer_head *),
3957                                           GFP_KERNEL);
3958         if (sbi->s_group_desc == NULL) {
3959                 ext4_msg(sb, KERN_ERR, "not enough memory");
3960                 ret = -ENOMEM;
3961                 goto failed_mount;
3962         }
3963
3964         bgl_lock_init(sbi->s_blockgroup_lock);
3965
3966         /* Pre-read the descriptors into the buffer cache */
3967         for (i = 0; i < db_count; i++) {
3968                 block = descriptor_loc(sb, logical_sb_block, i);
3969                 sb_breadahead(sb, block);
3970         }
3971
3972         for (i = 0; i < db_count; i++) {
3973                 block = descriptor_loc(sb, logical_sb_block, i);
3974                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3975                 if (!sbi->s_group_desc[i]) {
3976                         ext4_msg(sb, KERN_ERR,
3977                                "can't read group descriptor %d", i);
3978                         db_count = i;
3979                         goto failed_mount2;
3980                 }
3981         }
3982         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3983                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3984                 ret = -EFSCORRUPTED;
3985                 goto failed_mount2;
3986         }
3987
3988         sbi->s_gdb_count = db_count;
3989         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3990         spin_lock_init(&sbi->s_next_gen_lock);
3991
3992         setup_timer(&sbi->s_err_report, print_daily_error_info,
3993                 (unsigned long) sb);
3994
3995         /* Register extent status tree shrinker */
3996         if (ext4_es_register_shrinker(sbi))
3997                 goto failed_mount3;
3998
3999         sbi->s_stripe = ext4_get_stripe_size(sbi);
4000         sbi->s_extent_max_zeroout_kb = 32;
4001
4002         /*
4003          * set up enough so that it can read an inode
4004          */
4005         sb->s_op = &ext4_sops;
4006         sb->s_export_op = &ext4_export_ops;
4007         sb->s_xattr = ext4_xattr_handlers;
4008         sb->s_cop = &ext4_cryptops;
4009 #ifdef CONFIG_QUOTA
4010         sb->dq_op = &ext4_quota_operations;
4011         if (ext4_has_feature_quota(sb))
4012                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4013         else
4014                 sb->s_qcop = &ext4_qctl_operations;
4015         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4016 #endif
4017         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4018
4019         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4020         mutex_init(&sbi->s_orphan_lock);
4021
4022         sb->s_root = NULL;
4023
4024         needs_recovery = (es->s_last_orphan != 0 ||
4025                           ext4_has_feature_journal_needs_recovery(sb));
4026
4027         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4028                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4029                         goto failed_mount3a;
4030
4031         /*
4032          * The first inode we look at is the journal inode.  Don't try
4033          * root first: it may be modified in the journal!
4034          */
4035         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4036                 err = ext4_load_journal(sb, es, journal_devnum);
4037                 if (err)
4038                         goto failed_mount3a;
4039         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4040                    ext4_has_feature_journal_needs_recovery(sb)) {
4041                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4042                        "suppressed and not mounted read-only");
4043                 goto failed_mount_wq;
4044         } else {
4045                 /* Nojournal mode, all journal mount options are illegal */
4046                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4047                         ext4_msg(sb, KERN_ERR, "can't mount with "
4048                                  "journal_checksum, fs mounted w/o journal");
4049                         goto failed_mount_wq;
4050                 }
4051                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4052                         ext4_msg(sb, KERN_ERR, "can't mount with "
4053                                  "journal_async_commit, fs mounted w/o journal");
4054                         goto failed_mount_wq;
4055                 }
4056                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4057                         ext4_msg(sb, KERN_ERR, "can't mount with "
4058                                  "commit=%lu, fs mounted w/o journal",
4059                                  sbi->s_commit_interval / HZ);
4060                         goto failed_mount_wq;
4061                 }
4062                 if (EXT4_MOUNT_DATA_FLAGS &
4063                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4064                         ext4_msg(sb, KERN_ERR, "can't mount with "
4065                                  "data=, fs mounted w/o journal");
4066                         goto failed_mount_wq;
4067                 }
4068                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4069                 clear_opt(sb, JOURNAL_CHECKSUM);
4070                 clear_opt(sb, DATA_FLAGS);
4071                 sbi->s_journal = NULL;
4072                 needs_recovery = 0;
4073                 goto no_journal;
4074         }
4075
4076         if (ext4_has_feature_64bit(sb) &&
4077             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4078                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4079                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4080                 goto failed_mount_wq;
4081         }
4082
4083         if (!set_journal_csum_feature_set(sb)) {
4084                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4085                          "feature set");
4086                 goto failed_mount_wq;
4087         }
4088
4089         /* We have now updated the journal if required, so we can
4090          * validate the data journaling mode. */
4091         switch (test_opt(sb, DATA_FLAGS)) {
4092         case 0:
4093                 /* No mode set, assume a default based on the journal
4094                  * capabilities: ORDERED_DATA if the journal can
4095                  * cope, else JOURNAL_DATA
4096                  */
4097                 if (jbd2_journal_check_available_features
4098                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4099                         set_opt(sb, ORDERED_DATA);
4100                 else
4101                         set_opt(sb, JOURNAL_DATA);
4102                 break;
4103
4104         case EXT4_MOUNT_ORDERED_DATA:
4105         case EXT4_MOUNT_WRITEBACK_DATA:
4106                 if (!jbd2_journal_check_available_features
4107                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4108                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4109                                "requested data journaling mode");
4110                         goto failed_mount_wq;
4111                 }
4112         default:
4113                 break;
4114         }
4115
4116         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4117             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4118                 ext4_msg(sb, KERN_ERR, "can't mount with "
4119                         "journal_async_commit in data=ordered mode");
4120                 goto failed_mount_wq;
4121         }
4122
4123         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4124
4125         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4126
4127 no_journal:
4128         if (!test_opt(sb, NO_MBCACHE)) {
4129                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4130                 if (!sbi->s_ea_block_cache) {
4131                         ext4_msg(sb, KERN_ERR,
4132                                  "Failed to create ea_block_cache");
4133                         goto failed_mount_wq;
4134                 }
4135
4136                 if (ext4_has_feature_ea_inode(sb)) {
4137                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4138                         if (!sbi->s_ea_inode_cache) {
4139                                 ext4_msg(sb, KERN_ERR,
4140                                          "Failed to create ea_inode_cache");
4141                                 goto failed_mount_wq;
4142                         }
4143                 }
4144         }
4145
4146         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4147             (blocksize != PAGE_SIZE)) {
4148                 ext4_msg(sb, KERN_ERR,
4149                          "Unsupported blocksize for fs encryption");
4150                 goto failed_mount_wq;
4151         }
4152
4153         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4154             !ext4_has_feature_encrypt(sb)) {
4155                 ext4_set_feature_encrypt(sb);
4156                 ext4_commit_super(sb, 1);
4157         }
4158
4159         /*
4160          * Get the # of file system overhead blocks from the
4161          * superblock if present.
4162          */
4163         if (es->s_overhead_clusters)
4164                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4165         else {
4166                 err = ext4_calculate_overhead(sb);
4167                 if (err)
4168                         goto failed_mount_wq;
4169         }
4170
4171         /*
4172          * The maximum number of concurrent works can be high and
4173          * concurrency isn't really necessary.  Limit it to 1.
4174          */
4175         EXT4_SB(sb)->rsv_conversion_wq =
4176                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4177         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4178                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4179                 ret = -ENOMEM;
4180                 goto failed_mount4;
4181         }
4182
4183         /*
4184          * The jbd2_journal_load will have done any necessary log recovery,
4185          * so we can safely mount the rest of the filesystem now.
4186          */
4187
4188         root = ext4_iget(sb, EXT4_ROOT_INO);
4189         if (IS_ERR(root)) {
4190                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4191                 ret = PTR_ERR(root);
4192                 root = NULL;
4193                 goto failed_mount4;
4194         }
4195         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4196                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4197                 iput(root);
4198                 goto failed_mount4;
4199         }
4200         sb->s_root = d_make_root(root);
4201         if (!sb->s_root) {
4202                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4203                 ret = -ENOMEM;
4204                 goto failed_mount4;
4205         }
4206
4207         if (ext4_setup_super(sb, es, sb_rdonly(sb)))
4208                 sb->s_flags |= MS_RDONLY;
4209
4210         /* determine the minimum size of new large inodes, if present */
4211         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4212             sbi->s_want_extra_isize == 0) {
4213                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4214                                                      EXT4_GOOD_OLD_INODE_SIZE;
4215                 if (ext4_has_feature_extra_isize(sb)) {
4216                         if (sbi->s_want_extra_isize <
4217                             le16_to_cpu(es->s_want_extra_isize))
4218                                 sbi->s_want_extra_isize =
4219                                         le16_to_cpu(es->s_want_extra_isize);
4220                         if (sbi->s_want_extra_isize <
4221                             le16_to_cpu(es->s_min_extra_isize))
4222                                 sbi->s_want_extra_isize =
4223                                         le16_to_cpu(es->s_min_extra_isize);
4224                 }
4225         }
4226         /* Check if enough inode space is available */
4227         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4228                                                         sbi->s_inode_size) {
4229                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4230                                                        EXT4_GOOD_OLD_INODE_SIZE;
4231                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4232                          "available");
4233         }
4234
4235         ext4_set_resv_clusters(sb);
4236
4237         err = ext4_setup_system_zone(sb);
4238         if (err) {
4239                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4240                          "zone (%d)", err);
4241                 goto failed_mount4a;
4242         }
4243
4244         ext4_ext_init(sb);
4245         err = ext4_mb_init(sb);
4246         if (err) {
4247                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4248                          err);
4249                 goto failed_mount5;
4250         }
4251
4252         block = ext4_count_free_clusters(sb);
4253         ext4_free_blocks_count_set(sbi->s_es,
4254                                    EXT4_C2B(sbi, block));
4255         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4256                                   GFP_KERNEL);
4257         if (!err) {
4258                 unsigned long freei = ext4_count_free_inodes(sb);
4259                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4260                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4261                                           GFP_KERNEL);
4262         }
4263         if (!err)
4264                 err = percpu_counter_init(&sbi->s_dirs_counter,
4265                                           ext4_count_dirs(sb), GFP_KERNEL);
4266         if (!err)
4267                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4268                                           GFP_KERNEL);
4269         if (!err)
4270                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4271
4272         if (err) {
4273                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4274                 goto failed_mount6;
4275         }
4276
4277         if (ext4_has_feature_flex_bg(sb))
4278                 if (!ext4_fill_flex_info(sb)) {
4279                         ext4_msg(sb, KERN_ERR,
4280                                "unable to initialize "
4281                                "flex_bg meta info!");
4282                         goto failed_mount6;
4283                 }
4284
4285         err = ext4_register_li_request(sb, first_not_zeroed);
4286         if (err)
4287                 goto failed_mount6;
4288
4289         err = ext4_register_sysfs(sb);
4290         if (err)
4291                 goto failed_mount7;
4292
4293 #ifdef CONFIG_QUOTA
4294         /* Enable quota usage during mount. */
4295         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4296                 err = ext4_enable_quotas(sb);
4297                 if (err)
4298                         goto failed_mount8;
4299         }
4300 #endif  /* CONFIG_QUOTA */
4301
4302         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4303         ext4_orphan_cleanup(sb, es);
4304         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4305         if (needs_recovery) {
4306                 ext4_msg(sb, KERN_INFO, "recovery complete");
4307                 ext4_mark_recovery_complete(sb, es);
4308         }
4309         if (EXT4_SB(sb)->s_journal) {
4310                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4311                         descr = " journalled data mode";
4312                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4313                         descr = " ordered data mode";
4314                 else
4315                         descr = " writeback data mode";
4316         } else
4317                 descr = "out journal";
4318
4319         if (test_opt(sb, DISCARD)) {
4320                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4321                 if (!blk_queue_discard(q))
4322                         ext4_msg(sb, KERN_WARNING,
4323                                  "mounting with \"discard\" option, but "
4324                                  "the device does not support discard");
4325         }
4326
4327         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4328                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4329                          "Opts: %.*s%s%s", descr,
4330                          (int) sizeof(sbi->s_es->s_mount_opts),
4331                          sbi->s_es->s_mount_opts,
4332                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4333
4334         if (es->s_error_count)
4335                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4336
4337         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4338         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4339         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4340         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4341
4342         kfree(orig_data);
4343         return 0;
4344
4345 cantfind_ext4:
4346         if (!silent)
4347                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4348         goto failed_mount;
4349
4350 #ifdef CONFIG_QUOTA
4351 failed_mount8:
4352         ext4_unregister_sysfs(sb);
4353 #endif
4354 failed_mount7:
4355         ext4_unregister_li_request(sb);
4356 failed_mount6:
4357         ext4_mb_release(sb);
4358         if (sbi->s_flex_groups)
4359                 kvfree(sbi->s_flex_groups);
4360         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4361         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4362         percpu_counter_destroy(&sbi->s_dirs_counter);
4363         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4364 failed_mount5:
4365         ext4_ext_release(sb);
4366         ext4_release_system_zone(sb);
4367 failed_mount4a:
4368         dput(sb->s_root);
4369         sb->s_root = NULL;
4370 failed_mount4:
4371         ext4_msg(sb, KERN_ERR, "mount failed");
4372         if (EXT4_SB(sb)->rsv_conversion_wq)
4373                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4374 failed_mount_wq:
4375         if (sbi->s_ea_inode_cache) {
4376                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4377                 sbi->s_ea_inode_cache = NULL;
4378         }
4379         if (sbi->s_ea_block_cache) {
4380                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4381                 sbi->s_ea_block_cache = NULL;
4382         }
4383         if (sbi->s_journal) {
4384                 jbd2_journal_destroy(sbi->s_journal);
4385                 sbi->s_journal = NULL;
4386         }
4387 failed_mount3a:
4388         ext4_es_unregister_shrinker(sbi);
4389 failed_mount3:
4390         del_timer_sync(&sbi->s_err_report);
4391         if (sbi->s_mmp_tsk)
4392                 kthread_stop(sbi->s_mmp_tsk);
4393 failed_mount2:
4394         for (i = 0; i < db_count; i++)
4395                 brelse(sbi->s_group_desc[i]);
4396         kvfree(sbi->s_group_desc);
4397 failed_mount:
4398         if (sbi->s_chksum_driver)
4399                 crypto_free_shash(sbi->s_chksum_driver);
4400 #ifdef CONFIG_QUOTA
4401         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4402                 kfree(sbi->s_qf_names[i]);
4403 #endif
4404         ext4_blkdev_remove(sbi);
4405         brelse(bh);
4406 out_fail:
4407         sb->s_fs_info = NULL;
4408         kfree(sbi->s_blockgroup_lock);
4409 out_free_base:
4410         kfree(sbi);
4411         kfree(orig_data);
4412         fs_put_dax(dax_dev);
4413         return err ? err : ret;
4414 }
4415
4416 /*
4417  * Setup any per-fs journal parameters now.  We'll do this both on
4418  * initial mount, once the journal has been initialised but before we've
4419  * done any recovery; and again on any subsequent remount.
4420  */
4421 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4422 {
4423         struct ext4_sb_info *sbi = EXT4_SB(sb);
4424
4425         journal->j_commit_interval = sbi->s_commit_interval;
4426         journal->j_min_batch_time = sbi->s_min_batch_time;
4427         journal->j_max_batch_time = sbi->s_max_batch_time;
4428
4429         write_lock(&journal->j_state_lock);
4430         if (test_opt(sb, BARRIER))
4431                 journal->j_flags |= JBD2_BARRIER;
4432         else
4433                 journal->j_flags &= ~JBD2_BARRIER;
4434         if (test_opt(sb, DATA_ERR_ABORT))
4435                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4436         else
4437                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4438         write_unlock(&journal->j_state_lock);
4439 }
4440
4441 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4442                                              unsigned int journal_inum)
4443 {
4444         struct inode *journal_inode;
4445
4446         /*
4447          * Test for the existence of a valid inode on disk.  Bad things
4448          * happen if we iget() an unused inode, as the subsequent iput()
4449          * will try to delete it.
4450          */
4451         journal_inode = ext4_iget(sb, journal_inum);
4452         if (IS_ERR(journal_inode)) {
4453                 ext4_msg(sb, KERN_ERR, "no journal found");
4454                 return NULL;
4455         }
4456         if (!journal_inode->i_nlink) {
4457                 make_bad_inode(journal_inode);
4458                 iput(journal_inode);
4459                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4460                 return NULL;
4461         }
4462
4463         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4464                   journal_inode, journal_inode->i_size);
4465         if (!S_ISREG(journal_inode->i_mode)) {
4466                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4467                 iput(journal_inode);
4468                 return NULL;
4469         }
4470         return journal_inode;
4471 }
4472
4473 static journal_t *ext4_get_journal(struct super_block *sb,
4474                                    unsigned int journal_inum)
4475 {
4476         struct inode *journal_inode;
4477         journal_t *journal;
4478
4479         BUG_ON(!ext4_has_feature_journal(sb));
4480
4481         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4482         if (!journal_inode)
4483                 return NULL;
4484
4485         journal = jbd2_journal_init_inode(journal_inode);
4486         if (!journal) {
4487                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4488                 iput(journal_inode);
4489                 return NULL;
4490         }
4491         journal->j_private = sb;
4492         ext4_init_journal_params(sb, journal);
4493         return journal;
4494 }
4495
4496 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4497                                        dev_t j_dev)
4498 {
4499         struct buffer_head *bh;
4500         journal_t *journal;
4501         ext4_fsblk_t start;
4502         ext4_fsblk_t len;
4503         int hblock, blocksize;
4504         ext4_fsblk_t sb_block;
4505         unsigned long offset;
4506         struct ext4_super_block *es;
4507         struct block_device *bdev;
4508
4509         BUG_ON(!ext4_has_feature_journal(sb));
4510
4511         bdev = ext4_blkdev_get(j_dev, sb);
4512         if (bdev == NULL)
4513                 return NULL;
4514
4515         blocksize = sb->s_blocksize;
4516         hblock = bdev_logical_block_size(bdev);
4517         if (blocksize < hblock) {
4518                 ext4_msg(sb, KERN_ERR,
4519                         "blocksize too small for journal device");
4520                 goto out_bdev;
4521         }
4522
4523         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4524         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4525         set_blocksize(bdev, blocksize);
4526         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4527                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4528                        "external journal");
4529                 goto out_bdev;
4530         }
4531
4532         es = (struct ext4_super_block *) (bh->b_data + offset);
4533         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4534             !(le32_to_cpu(es->s_feature_incompat) &
4535               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4536                 ext4_msg(sb, KERN_ERR, "external journal has "
4537                                         "bad superblock");
4538                 brelse(bh);
4539                 goto out_bdev;
4540         }
4541
4542         if ((le32_to_cpu(es->s_feature_ro_compat) &
4543              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4544             es->s_checksum != ext4_superblock_csum(sb, es)) {
4545                 ext4_msg(sb, KERN_ERR, "external journal has "
4546                                        "corrupt superblock");
4547                 brelse(bh);
4548                 goto out_bdev;
4549         }
4550
4551         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4552                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4553                 brelse(bh);
4554                 goto out_bdev;
4555         }
4556
4557         len = ext4_blocks_count(es);
4558         start = sb_block + 1;
4559         brelse(bh);     /* we're done with the superblock */
4560
4561         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4562                                         start, len, blocksize);
4563         if (!journal) {
4564                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4565                 goto out_bdev;
4566         }
4567         journal->j_private = sb;
4568         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4569         wait_on_buffer(journal->j_sb_buffer);
4570         if (!buffer_uptodate(journal->j_sb_buffer)) {
4571                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4572                 goto out_journal;
4573         }
4574         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4575                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4576                                         "user (unsupported) - %d",
4577                         be32_to_cpu(journal->j_superblock->s_nr_users));
4578                 goto out_journal;
4579         }
4580         EXT4_SB(sb)->journal_bdev = bdev;
4581         ext4_init_journal_params(sb, journal);
4582         return journal;
4583
4584 out_journal:
4585         jbd2_journal_destroy(journal);
4586 out_bdev:
4587         ext4_blkdev_put(bdev);
4588         return NULL;
4589 }
4590
4591 static int ext4_load_journal(struct super_block *sb,
4592                              struct ext4_super_block *es,
4593                              unsigned long journal_devnum)
4594 {
4595         journal_t *journal;
4596         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4597         dev_t journal_dev;
4598         int err = 0;
4599         int really_read_only;
4600
4601         BUG_ON(!ext4_has_feature_journal(sb));
4602
4603         if (journal_devnum &&
4604             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4605                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4606                         "numbers have changed");
4607                 journal_dev = new_decode_dev(journal_devnum);
4608         } else
4609                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4610
4611         really_read_only = bdev_read_only(sb->s_bdev);
4612
4613         /*
4614          * Are we loading a blank journal or performing recovery after a
4615          * crash?  For recovery, we need to check in advance whether we
4616          * can get read-write access to the device.
4617          */
4618         if (ext4_has_feature_journal_needs_recovery(sb)) {
4619                 if (sb_rdonly(sb)) {
4620                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4621                                         "required on readonly filesystem");
4622                         if (really_read_only) {
4623                                 ext4_msg(sb, KERN_ERR, "write access "
4624                                         "unavailable, cannot proceed");
4625                                 return -EROFS;
4626                         }
4627                         ext4_msg(sb, KERN_INFO, "write access will "
4628                                "be enabled during recovery");
4629                 }
4630         }
4631
4632         if (journal_inum && journal_dev) {
4633                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4634                        "and inode journals!");
4635                 return -EINVAL;
4636         }
4637
4638         if (journal_inum) {
4639                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4640                         return -EINVAL;
4641         } else {
4642                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4643                         return -EINVAL;
4644         }
4645
4646         if (!(journal->j_flags & JBD2_BARRIER))
4647                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4648
4649         if (!ext4_has_feature_journal_needs_recovery(sb))
4650                 err = jbd2_journal_wipe(journal, !really_read_only);
4651         if (!err) {
4652                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4653                 if (save)
4654                         memcpy(save, ((char *) es) +
4655                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4656                 err = jbd2_journal_load(journal);
4657                 if (save)
4658                         memcpy(((char *) es) + EXT4_S_ERR_START,
4659                                save, EXT4_S_ERR_LEN);
4660                 kfree(save);
4661         }
4662
4663         if (err) {
4664                 ext4_msg(sb, KERN_ERR, "error loading journal");
4665                 jbd2_journal_destroy(journal);
4666                 return err;
4667         }
4668
4669         EXT4_SB(sb)->s_journal = journal;
4670         ext4_clear_journal_err(sb, es);
4671
4672         if (!really_read_only && journal_devnum &&
4673             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4674                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4675
4676                 /* Make sure we flush the recovery flag to disk. */
4677                 ext4_commit_super(sb, 1);
4678         }
4679
4680         return 0;
4681 }
4682
4683 static int ext4_commit_super(struct super_block *sb, int sync)
4684 {
4685         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4686         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4687         int error = 0;
4688
4689         if (!sbh || block_device_ejected(sb))
4690                 return error;
4691         /*
4692          * If the file system is mounted read-only, don't update the
4693          * superblock write time.  This avoids updating the superblock
4694          * write time when we are mounting the root file system
4695          * read/only but we need to replay the journal; at that point,
4696          * for people who are east of GMT and who make their clock
4697          * tick in localtime for Windows bug-for-bug compatibility,
4698          * the clock is set in the future, and this will cause e2fsck
4699          * to complain and force a full file system check.
4700          */
4701         if (!(sb->s_flags & MS_RDONLY))
4702                 es->s_wtime = cpu_to_le32(get_seconds());
4703         if (sb->s_bdev->bd_part)
4704                 es->s_kbytes_written =
4705                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4706                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4707                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4708         else
4709                 es->s_kbytes_written =
4710                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4711         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4712                 ext4_free_blocks_count_set(es,
4713                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4714                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4715         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4716                 es->s_free_inodes_count =
4717                         cpu_to_le32(percpu_counter_sum_positive(
4718                                 &EXT4_SB(sb)->s_freeinodes_counter));
4719         BUFFER_TRACE(sbh, "marking dirty");
4720         ext4_superblock_csum_set(sb);
4721         if (sync)
4722                 lock_buffer(sbh);
4723         if (buffer_write_io_error(sbh)) {
4724                 /*
4725                  * Oh, dear.  A previous attempt to write the
4726                  * superblock failed.  This could happen because the
4727                  * USB device was yanked out.  Or it could happen to
4728                  * be a transient write error and maybe the block will
4729                  * be remapped.  Nothing we can do but to retry the
4730                  * write and hope for the best.
4731                  */
4732                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4733                        "superblock detected");
4734                 clear_buffer_write_io_error(sbh);
4735                 set_buffer_uptodate(sbh);
4736         }
4737         mark_buffer_dirty(sbh);
4738         if (sync) {
4739                 unlock_buffer(sbh);
4740                 error = __sync_dirty_buffer(sbh,
4741                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4742                 if (error)
4743                         return error;
4744
4745                 error = buffer_write_io_error(sbh);
4746                 if (error) {
4747                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4748                                "superblock");
4749                         clear_buffer_write_io_error(sbh);
4750                         set_buffer_uptodate(sbh);
4751                 }
4752         }
4753         return error;
4754 }
4755
4756 /*
4757  * Have we just finished recovery?  If so, and if we are mounting (or
4758  * remounting) the filesystem readonly, then we will end up with a
4759  * consistent fs on disk.  Record that fact.
4760  */
4761 static void ext4_mark_recovery_complete(struct super_block *sb,
4762                                         struct ext4_super_block *es)
4763 {
4764         journal_t *journal = EXT4_SB(sb)->s_journal;
4765
4766         if (!ext4_has_feature_journal(sb)) {
4767                 BUG_ON(journal != NULL);
4768                 return;
4769         }
4770         jbd2_journal_lock_updates(journal);
4771         if (jbd2_journal_flush(journal) < 0)
4772                 goto out;
4773
4774         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4775                 ext4_clear_feature_journal_needs_recovery(sb);
4776                 ext4_commit_super(sb, 1);
4777         }
4778
4779 out:
4780         jbd2_journal_unlock_updates(journal);
4781 }
4782
4783 /*
4784  * If we are mounting (or read-write remounting) a filesystem whose journal
4785  * has recorded an error from a previous lifetime, move that error to the
4786  * main filesystem now.
4787  */
4788 static void ext4_clear_journal_err(struct super_block *sb,
4789                                    struct ext4_super_block *es)
4790 {
4791         journal_t *journal;
4792         int j_errno;
4793         const char *errstr;
4794
4795         BUG_ON(!ext4_has_feature_journal(sb));
4796
4797         journal = EXT4_SB(sb)->s_journal;
4798
4799         /*
4800          * Now check for any error status which may have been recorded in the
4801          * journal by a prior ext4_error() or ext4_abort()
4802          */
4803
4804         j_errno = jbd2_journal_errno(journal);
4805         if (j_errno) {
4806                 char nbuf[16];
4807
4808                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4809                 ext4_warning(sb, "Filesystem error recorded "
4810                              "from previous mount: %s", errstr);
4811                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4812
4813                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4814                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4815                 ext4_commit_super(sb, 1);
4816
4817                 jbd2_journal_clear_err(journal);
4818                 jbd2_journal_update_sb_errno(journal);
4819         }
4820 }
4821
4822 /*
4823  * Force the running and committing transactions to commit,
4824  * and wait on the commit.
4825  */
4826 int ext4_force_commit(struct super_block *sb)
4827 {
4828         journal_t *journal;
4829
4830         if (sb_rdonly(sb))
4831                 return 0;
4832
4833         journal = EXT4_SB(sb)->s_journal;
4834         return ext4_journal_force_commit(journal);
4835 }
4836
4837 static int ext4_sync_fs(struct super_block *sb, int wait)
4838 {
4839         int ret = 0;
4840         tid_t target;
4841         bool needs_barrier = false;
4842         struct ext4_sb_info *sbi = EXT4_SB(sb);
4843
4844         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
4845                 return 0;
4846
4847         trace_ext4_sync_fs(sb, wait);
4848         flush_workqueue(sbi->rsv_conversion_wq);
4849         /*
4850          * Writeback quota in non-journalled quota case - journalled quota has
4851          * no dirty dquots
4852          */
4853         dquot_writeback_dquots(sb, -1);
4854         /*
4855          * Data writeback is possible w/o journal transaction, so barrier must
4856          * being sent at the end of the function. But we can skip it if
4857          * transaction_commit will do it for us.
4858          */
4859         if (sbi->s_journal) {
4860                 target = jbd2_get_latest_transaction(sbi->s_journal);
4861                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4862                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4863                         needs_barrier = true;
4864
4865                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4866                         if (wait)
4867                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4868                                                            target);
4869                 }
4870         } else if (wait && test_opt(sb, BARRIER))
4871                 needs_barrier = true;
4872         if (needs_barrier) {
4873                 int err;
4874                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4875                 if (!ret)
4876                         ret = err;
4877         }
4878
4879         return ret;
4880 }
4881
4882 /*
4883  * LVM calls this function before a (read-only) snapshot is created.  This
4884  * gives us a chance to flush the journal completely and mark the fs clean.
4885  *
4886  * Note that only this function cannot bring a filesystem to be in a clean
4887  * state independently. It relies on upper layer to stop all data & metadata
4888  * modifications.
4889  */
4890 static int ext4_freeze(struct super_block *sb)
4891 {
4892         int error = 0;
4893         journal_t *journal;
4894
4895         if (sb_rdonly(sb))
4896                 return 0;
4897
4898         journal = EXT4_SB(sb)->s_journal;
4899
4900         if (journal) {
4901                 /* Now we set up the journal barrier. */
4902                 jbd2_journal_lock_updates(journal);
4903
4904                 /*
4905                  * Don't clear the needs_recovery flag if we failed to
4906                  * flush the journal.
4907                  */
4908                 error = jbd2_journal_flush(journal);
4909                 if (error < 0)
4910                         goto out;
4911
4912                 /* Journal blocked and flushed, clear needs_recovery flag. */
4913                 ext4_clear_feature_journal_needs_recovery(sb);
4914         }
4915
4916         error = ext4_commit_super(sb, 1);
4917 out:
4918         if (journal)
4919                 /* we rely on upper layer to stop further updates */
4920                 jbd2_journal_unlock_updates(journal);
4921         return error;
4922 }
4923
4924 /*
4925  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4926  * flag here, even though the filesystem is not technically dirty yet.
4927  */
4928 static int ext4_unfreeze(struct super_block *sb)
4929 {
4930         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
4931                 return 0;
4932
4933         if (EXT4_SB(sb)->s_journal) {
4934                 /* Reset the needs_recovery flag before the fs is unlocked. */
4935                 ext4_set_feature_journal_needs_recovery(sb);
4936         }
4937
4938         ext4_commit_super(sb, 1);
4939         return 0;
4940 }
4941
4942 /*
4943  * Structure to save mount options for ext4_remount's benefit
4944  */
4945 struct ext4_mount_options {
4946         unsigned long s_mount_opt;
4947         unsigned long s_mount_opt2;
4948         kuid_t s_resuid;
4949         kgid_t s_resgid;
4950         unsigned long s_commit_interval;
4951         u32 s_min_batch_time, s_max_batch_time;
4952 #ifdef CONFIG_QUOTA
4953         int s_jquota_fmt;
4954         char *s_qf_names[EXT4_MAXQUOTAS];
4955 #endif
4956 };
4957
4958 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4959 {
4960         struct ext4_super_block *es;
4961         struct ext4_sb_info *sbi = EXT4_SB(sb);
4962         unsigned long old_sb_flags;
4963         struct ext4_mount_options old_opts;
4964         int enable_quota = 0;
4965         ext4_group_t g;
4966         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4967         int err = 0;
4968 #ifdef CONFIG_QUOTA
4969         int i, j;
4970 #endif
4971         char *orig_data = kstrdup(data, GFP_KERNEL);
4972
4973         /* Store the original options */
4974         old_sb_flags = sb->s_flags;
4975         old_opts.s_mount_opt = sbi->s_mount_opt;
4976         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4977         old_opts.s_resuid = sbi->s_resuid;
4978         old_opts.s_resgid = sbi->s_resgid;
4979         old_opts.s_commit_interval = sbi->s_commit_interval;
4980         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4981         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4982 #ifdef CONFIG_QUOTA
4983         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4984         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4985                 if (sbi->s_qf_names[i]) {
4986                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4987                                                          GFP_KERNEL);
4988                         if (!old_opts.s_qf_names[i]) {
4989                                 for (j = 0; j < i; j++)
4990                                         kfree(old_opts.s_qf_names[j]);
4991                                 kfree(orig_data);
4992                                 return -ENOMEM;
4993                         }
4994                 } else
4995                         old_opts.s_qf_names[i] = NULL;
4996 #endif
4997         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4998                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4999
5000         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5001                 err = -EINVAL;
5002                 goto restore_opts;
5003         }
5004
5005         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5006             test_opt(sb, JOURNAL_CHECKSUM)) {
5007                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5008                          "during remount not supported; ignoring");
5009                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5010         }
5011
5012         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5013                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5014                         ext4_msg(sb, KERN_ERR, "can't mount with "
5015                                  "both data=journal and delalloc");
5016                         err = -EINVAL;
5017                         goto restore_opts;
5018                 }
5019                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5020                         ext4_msg(sb, KERN_ERR, "can't mount with "
5021                                  "both data=journal and dioread_nolock");
5022                         err = -EINVAL;
5023                         goto restore_opts;
5024                 }
5025                 if (test_opt(sb, DAX)) {
5026                         ext4_msg(sb, KERN_ERR, "can't mount with "
5027                                  "both data=journal and dax");
5028                         err = -EINVAL;
5029                         goto restore_opts;
5030                 }
5031         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5032                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5033                         ext4_msg(sb, KERN_ERR, "can't mount with "
5034                                 "journal_async_commit in data=ordered mode");
5035                         err = -EINVAL;
5036                         goto restore_opts;
5037                 }
5038         }
5039
5040         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5041                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5042                 err = -EINVAL;
5043                 goto restore_opts;
5044         }
5045
5046         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5047                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5048                         "dax flag with busy inodes while remounting");
5049                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5050         }
5051
5052         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5053                 ext4_abort(sb, "Abort forced by user");
5054
5055         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
5056                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
5057
5058         es = sbi->s_es;
5059
5060         if (sbi->s_journal) {
5061                 ext4_init_journal_params(sb, sbi->s_journal);
5062                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5063         }
5064
5065         if (*flags & MS_LAZYTIME)
5066                 sb->s_flags |= MS_LAZYTIME;
5067
5068         if ((bool)(*flags & MS_RDONLY) != sb_rdonly(sb)) {
5069                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5070                         err = -EROFS;
5071                         goto restore_opts;
5072                 }
5073
5074                 if (*flags & MS_RDONLY) {
5075                         err = sync_filesystem(sb);
5076                         if (err < 0)
5077                                 goto restore_opts;
5078                         err = dquot_suspend(sb, -1);
5079                         if (err < 0)
5080                                 goto restore_opts;
5081
5082                         /*
5083                          * First of all, the unconditional stuff we have to do
5084                          * to disable replay of the journal when we next remount
5085                          */
5086                         sb->s_flags |= MS_RDONLY;
5087
5088                         /*
5089                          * OK, test if we are remounting a valid rw partition
5090                          * readonly, and if so set the rdonly flag and then
5091                          * mark the partition as valid again.
5092                          */
5093                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5094                             (sbi->s_mount_state & EXT4_VALID_FS))
5095                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5096
5097                         if (sbi->s_journal)
5098                                 ext4_mark_recovery_complete(sb, es);
5099                 } else {
5100                         /* Make sure we can mount this feature set readwrite */
5101                         if (ext4_has_feature_readonly(sb) ||
5102                             !ext4_feature_set_ok(sb, 0)) {
5103                                 err = -EROFS;
5104                                 goto restore_opts;
5105                         }
5106                         /*
5107                          * Make sure the group descriptor checksums
5108                          * are sane.  If they aren't, refuse to remount r/w.
5109                          */
5110                         for (g = 0; g < sbi->s_groups_count; g++) {
5111                                 struct ext4_group_desc *gdp =
5112                                         ext4_get_group_desc(sb, g, NULL);
5113
5114                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5115                                         ext4_msg(sb, KERN_ERR,
5116                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5117                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5118                                                le16_to_cpu(gdp->bg_checksum));
5119                                         err = -EFSBADCRC;
5120                                         goto restore_opts;
5121                                 }
5122                         }
5123
5124                         /*
5125                          * If we have an unprocessed orphan list hanging
5126                          * around from a previously readonly bdev mount,
5127                          * require a full umount/remount for now.
5128                          */
5129                         if (es->s_last_orphan) {
5130                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5131                                        "remount RDWR because of unprocessed "
5132                                        "orphan inode list.  Please "
5133                                        "umount/remount instead");
5134                                 err = -EINVAL;
5135                                 goto restore_opts;
5136                         }
5137
5138                         /*
5139                          * Mounting a RDONLY partition read-write, so reread
5140                          * and store the current valid flag.  (It may have
5141                          * been changed by e2fsck since we originally mounted
5142                          * the partition.)
5143                          */
5144                         if (sbi->s_journal)
5145                                 ext4_clear_journal_err(sb, es);
5146                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5147                         if (!ext4_setup_super(sb, es, 0))
5148                                 sb->s_flags &= ~MS_RDONLY;
5149                         if (ext4_has_feature_mmp(sb))
5150                                 if (ext4_multi_mount_protect(sb,
5151                                                 le64_to_cpu(es->s_mmp_block))) {
5152                                         err = -EROFS;
5153                                         goto restore_opts;
5154                                 }
5155                         enable_quota = 1;
5156                 }
5157         }
5158
5159         /*
5160          * Reinitialize lazy itable initialization thread based on
5161          * current settings
5162          */
5163         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5164                 ext4_unregister_li_request(sb);
5165         else {
5166                 ext4_group_t first_not_zeroed;
5167                 first_not_zeroed = ext4_has_uninit_itable(sb);
5168                 ext4_register_li_request(sb, first_not_zeroed);
5169         }
5170
5171         ext4_setup_system_zone(sb);
5172         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5173                 ext4_commit_super(sb, 1);
5174
5175 #ifdef CONFIG_QUOTA
5176         /* Release old quota file names */
5177         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5178                 kfree(old_opts.s_qf_names[i]);
5179         if (enable_quota) {
5180                 if (sb_any_quota_suspended(sb))
5181                         dquot_resume(sb, -1);
5182                 else if (ext4_has_feature_quota(sb)) {
5183                         err = ext4_enable_quotas(sb);
5184                         if (err)
5185                                 goto restore_opts;
5186                 }
5187         }
5188 #endif
5189
5190         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5191         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5192         kfree(orig_data);
5193         return 0;
5194
5195 restore_opts:
5196         sb->s_flags = old_sb_flags;
5197         sbi->s_mount_opt = old_opts.s_mount_opt;
5198         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5199         sbi->s_resuid = old_opts.s_resuid;
5200         sbi->s_resgid = old_opts.s_resgid;
5201         sbi->s_commit_interval = old_opts.s_commit_interval;
5202         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5203         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5204 #ifdef CONFIG_QUOTA
5205         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5206         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5207                 kfree(sbi->s_qf_names[i]);
5208                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5209         }
5210 #endif
5211         kfree(orig_data);
5212         return err;
5213 }
5214
5215 #ifdef CONFIG_QUOTA
5216 static int ext4_statfs_project(struct super_block *sb,
5217                                kprojid_t projid, struct kstatfs *buf)
5218 {
5219         struct kqid qid;
5220         struct dquot *dquot;
5221         u64 limit;
5222         u64 curblock;
5223
5224         qid = make_kqid_projid(projid);
5225         dquot = dqget(sb, qid);
5226         if (IS_ERR(dquot))
5227                 return PTR_ERR(dquot);
5228         spin_lock(&dquot->dq_dqb_lock);
5229
5230         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5231                  dquot->dq_dqb.dqb_bsoftlimit :
5232                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5233         if (limit && buf->f_blocks > limit) {
5234                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5235                 buf->f_blocks = limit;
5236                 buf->f_bfree = buf->f_bavail =
5237                         (buf->f_blocks > curblock) ?
5238                          (buf->f_blocks - curblock) : 0;
5239         }
5240
5241         limit = dquot->dq_dqb.dqb_isoftlimit ?
5242                 dquot->dq_dqb.dqb_isoftlimit :
5243                 dquot->dq_dqb.dqb_ihardlimit;
5244         if (limit && buf->f_files > limit) {
5245                 buf->f_files = limit;
5246                 buf->f_ffree =
5247                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5248                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5249         }
5250
5251         spin_unlock(&dquot->dq_dqb_lock);
5252         dqput(dquot);
5253         return 0;
5254 }
5255 #endif
5256
5257 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5258 {
5259         struct super_block *sb = dentry->d_sb;
5260         struct ext4_sb_info *sbi = EXT4_SB(sb);
5261         struct ext4_super_block *es = sbi->s_es;
5262         ext4_fsblk_t overhead = 0, resv_blocks;
5263         u64 fsid;
5264         s64 bfree;
5265         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5266
5267         if (!test_opt(sb, MINIX_DF))
5268                 overhead = sbi->s_overhead;
5269
5270         buf->f_type = EXT4_SUPER_MAGIC;
5271         buf->f_bsize = sb->s_blocksize;
5272         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5273         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5274                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5275         /* prevent underflow in case that few free space is available */
5276         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5277         buf->f_bavail = buf->f_bfree -
5278                         (ext4_r_blocks_count(es) + resv_blocks);
5279         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5280                 buf->f_bavail = 0;
5281         buf->f_files = le32_to_cpu(es->s_inodes_count);
5282         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5283         buf->f_namelen = EXT4_NAME_LEN;
5284         fsid = le64_to_cpup((void *)es->s_uuid) ^
5285                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5286         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5287         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5288
5289 #ifdef CONFIG_QUOTA
5290         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5291             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5292                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5293 #endif
5294         return 0;
5295 }
5296
5297
5298 #ifdef CONFIG_QUOTA
5299
5300 /*
5301  * Helper functions so that transaction is started before we acquire dqio_sem
5302  * to keep correct lock ordering of transaction > dqio_sem
5303  */
5304 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5305 {
5306         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5307 }
5308
5309 static int ext4_write_dquot(struct dquot *dquot)
5310 {
5311         int ret, err;
5312         handle_t *handle;
5313         struct inode *inode;
5314
5315         inode = dquot_to_inode(dquot);
5316         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5317                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5318         if (IS_ERR(handle))
5319                 return PTR_ERR(handle);
5320         ret = dquot_commit(dquot);
5321         err = ext4_journal_stop(handle);
5322         if (!ret)
5323                 ret = err;
5324         return ret;
5325 }
5326
5327 static int ext4_acquire_dquot(struct dquot *dquot)
5328 {
5329         int ret, err;
5330         handle_t *handle;
5331
5332         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5333                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5334         if (IS_ERR(handle))
5335                 return PTR_ERR(handle);
5336         ret = dquot_acquire(dquot);
5337         err = ext4_journal_stop(handle);
5338         if (!ret)
5339                 ret = err;
5340         return ret;
5341 }
5342
5343 static int ext4_release_dquot(struct dquot *dquot)
5344 {
5345         int ret, err;
5346         handle_t *handle;
5347
5348         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5349                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5350         if (IS_ERR(handle)) {
5351                 /* Release dquot anyway to avoid endless cycle in dqput() */
5352                 dquot_release(dquot);
5353                 return PTR_ERR(handle);
5354         }
5355         ret = dquot_release(dquot);
5356         err = ext4_journal_stop(handle);
5357         if (!ret)
5358                 ret = err;
5359         return ret;
5360 }
5361
5362 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5363 {
5364         struct super_block *sb = dquot->dq_sb;
5365         struct ext4_sb_info *sbi = EXT4_SB(sb);
5366
5367         /* Are we journaling quotas? */
5368         if (ext4_has_feature_quota(sb) ||
5369             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5370                 dquot_mark_dquot_dirty(dquot);
5371                 return ext4_write_dquot(dquot);
5372         } else {
5373                 return dquot_mark_dquot_dirty(dquot);
5374         }
5375 }
5376
5377 static int ext4_write_info(struct super_block *sb, int type)
5378 {
5379         int ret, err;
5380         handle_t *handle;
5381
5382         /* Data block + inode block */
5383         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5384         if (IS_ERR(handle))
5385                 return PTR_ERR(handle);
5386         ret = dquot_commit_info(sb, type);
5387         err = ext4_journal_stop(handle);
5388         if (!ret)
5389                 ret = err;
5390         return ret;
5391 }
5392
5393 /*
5394  * Turn on quotas during mount time - we need to find
5395  * the quota file and such...
5396  */
5397 static int ext4_quota_on_mount(struct super_block *sb, int type)
5398 {
5399         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5400                                         EXT4_SB(sb)->s_jquota_fmt, type);
5401 }
5402
5403 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5404 {
5405         struct ext4_inode_info *ei = EXT4_I(inode);
5406
5407         /* The first argument of lockdep_set_subclass has to be
5408          * *exactly* the same as the argument to init_rwsem() --- in
5409          * this case, in init_once() --- or lockdep gets unhappy
5410          * because the name of the lock is set using the
5411          * stringification of the argument to init_rwsem().
5412          */
5413         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5414         lockdep_set_subclass(&ei->i_data_sem, subclass);
5415 }
5416
5417 /*
5418  * Standard function to be called on quota_on
5419  */
5420 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5421                          const struct path *path)
5422 {
5423         int err;
5424
5425         if (!test_opt(sb, QUOTA))
5426                 return -EINVAL;
5427
5428         /* Quotafile not on the same filesystem? */
5429         if (path->dentry->d_sb != sb)
5430                 return -EXDEV;
5431         /* Journaling quota? */
5432         if (EXT4_SB(sb)->s_qf_names[type]) {
5433                 /* Quotafile not in fs root? */
5434                 if (path->dentry->d_parent != sb->s_root)
5435                         ext4_msg(sb, KERN_WARNING,
5436                                 "Quota file not on filesystem root. "
5437                                 "Journaled quota will not work");
5438                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5439         } else {
5440                 /*
5441                  * Clear the flag just in case mount options changed since
5442                  * last time.
5443                  */
5444                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5445         }
5446
5447         /*
5448          * When we journal data on quota file, we have to flush journal to see
5449          * all updates to the file when we bypass pagecache...
5450          */
5451         if (EXT4_SB(sb)->s_journal &&
5452             ext4_should_journal_data(d_inode(path->dentry))) {
5453                 /*
5454                  * We don't need to lock updates but journal_flush() could
5455                  * otherwise be livelocked...
5456                  */
5457                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5458                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5459                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5460                 if (err)
5461                         return err;
5462         }
5463
5464         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5465         err = dquot_quota_on(sb, type, format_id, path);
5466         if (err) {
5467                 lockdep_set_quota_inode(path->dentry->d_inode,
5468                                              I_DATA_SEM_NORMAL);
5469         } else {
5470                 struct inode *inode = d_inode(path->dentry);
5471                 handle_t *handle;
5472
5473                 /*
5474                  * Set inode flags to prevent userspace from messing with quota
5475                  * files. If this fails, we return success anyway since quotas
5476                  * are already enabled and this is not a hard failure.
5477                  */
5478                 inode_lock(inode);
5479                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5480                 if (IS_ERR(handle))
5481                         goto unlock_inode;
5482                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5483                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5484                                 S_NOATIME | S_IMMUTABLE);
5485                 ext4_mark_inode_dirty(handle, inode);
5486                 ext4_journal_stop(handle);
5487         unlock_inode:
5488                 inode_unlock(inode);
5489         }
5490         return err;
5491 }
5492
5493 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5494                              unsigned int flags)
5495 {
5496         int err;
5497         struct inode *qf_inode;
5498         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5499                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5500                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5501                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5502         };
5503
5504         BUG_ON(!ext4_has_feature_quota(sb));
5505
5506         if (!qf_inums[type])
5507                 return -EPERM;
5508
5509         qf_inode = ext4_iget(sb, qf_inums[type]);
5510         if (IS_ERR(qf_inode)) {
5511                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5512                 return PTR_ERR(qf_inode);
5513         }
5514
5515         /* Don't account quota for quota files to avoid recursion */
5516         qf_inode->i_flags |= S_NOQUOTA;
5517         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5518         err = dquot_enable(qf_inode, type, format_id, flags);
5519         iput(qf_inode);
5520         if (err)
5521                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5522
5523         return err;
5524 }
5525
5526 /* Enable usage tracking for all quota types. */
5527 static int ext4_enable_quotas(struct super_block *sb)
5528 {
5529         int type, err = 0;
5530         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5531                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5532                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5533                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5534         };
5535         bool quota_mopt[EXT4_MAXQUOTAS] = {
5536                 test_opt(sb, USRQUOTA),
5537                 test_opt(sb, GRPQUOTA),
5538                 test_opt(sb, PRJQUOTA),
5539         };
5540
5541         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5542         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5543                 if (qf_inums[type]) {
5544                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5545                                 DQUOT_USAGE_ENABLED |
5546                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5547                         if (err) {
5548                                 for (type--; type >= 0; type--)
5549                                         dquot_quota_off(sb, type);
5550
5551                                 ext4_warning(sb,
5552                                         "Failed to enable quota tracking "
5553                                         "(type=%d, err=%d). Please run "
5554                                         "e2fsck to fix.", type, err);
5555                                 return err;
5556                         }
5557                 }
5558         }
5559         return 0;
5560 }
5561
5562 static int ext4_quota_off(struct super_block *sb, int type)
5563 {
5564         struct inode *inode = sb_dqopt(sb)->files[type];
5565         handle_t *handle;
5566         int err;
5567
5568         /* Force all delayed allocation blocks to be allocated.
5569          * Caller already holds s_umount sem */
5570         if (test_opt(sb, DELALLOC))
5571                 sync_filesystem(sb);
5572
5573         if (!inode || !igrab(inode))
5574                 goto out;
5575
5576         err = dquot_quota_off(sb, type);
5577         if (err || ext4_has_feature_quota(sb))
5578                 goto out_put;
5579
5580         inode_lock(inode);
5581         /*
5582          * Update modification times of quota files when userspace can
5583          * start looking at them. If we fail, we return success anyway since
5584          * this is not a hard failure and quotas are already disabled.
5585          */
5586         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5587         if (IS_ERR(handle))
5588                 goto out_unlock;
5589         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5590         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5591         inode->i_mtime = inode->i_ctime = current_time(inode);
5592         ext4_mark_inode_dirty(handle, inode);
5593         ext4_journal_stop(handle);
5594 out_unlock:
5595         inode_unlock(inode);
5596 out_put:
5597         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5598         iput(inode);
5599         return err;
5600 out:
5601         return dquot_quota_off(sb, type);
5602 }
5603
5604 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5605  * acquiring the locks... As quota files are never truncated and quota code
5606  * itself serializes the operations (and no one else should touch the files)
5607  * we don't have to be afraid of races */
5608 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5609                                size_t len, loff_t off)
5610 {
5611         struct inode *inode = sb_dqopt(sb)->files[type];
5612         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5613         int offset = off & (sb->s_blocksize - 1);
5614         int tocopy;
5615         size_t toread;
5616         struct buffer_head *bh;
5617         loff_t i_size = i_size_read(inode);
5618
5619         if (off > i_size)
5620                 return 0;
5621         if (off+len > i_size)
5622                 len = i_size-off;
5623         toread = len;
5624         while (toread > 0) {
5625                 tocopy = sb->s_blocksize - offset < toread ?
5626                                 sb->s_blocksize - offset : toread;
5627                 bh = ext4_bread(NULL, inode, blk, 0);
5628                 if (IS_ERR(bh))
5629                         return PTR_ERR(bh);
5630                 if (!bh)        /* A hole? */
5631                         memset(data, 0, tocopy);
5632                 else
5633                         memcpy(data, bh->b_data+offset, tocopy);
5634                 brelse(bh);
5635                 offset = 0;
5636                 toread -= tocopy;
5637                 data += tocopy;
5638                 blk++;
5639         }
5640         return len;
5641 }
5642
5643 /* Write to quotafile (we know the transaction is already started and has
5644  * enough credits) */
5645 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5646                                 const char *data, size_t len, loff_t off)
5647 {
5648         struct inode *inode = sb_dqopt(sb)->files[type];
5649         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5650         int err, offset = off & (sb->s_blocksize - 1);
5651         int retries = 0;
5652         struct buffer_head *bh;
5653         handle_t *handle = journal_current_handle();
5654
5655         if (EXT4_SB(sb)->s_journal && !handle) {
5656                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5657                         " cancelled because transaction is not started",
5658                         (unsigned long long)off, (unsigned long long)len);
5659                 return -EIO;
5660         }
5661         /*
5662          * Since we account only one data block in transaction credits,
5663          * then it is impossible to cross a block boundary.
5664          */
5665         if (sb->s_blocksize - offset < len) {
5666                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5667                         " cancelled because not block aligned",
5668                         (unsigned long long)off, (unsigned long long)len);
5669                 return -EIO;
5670         }
5671
5672         do {
5673                 bh = ext4_bread(handle, inode, blk,
5674                                 EXT4_GET_BLOCKS_CREATE |
5675                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5676         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5677                  ext4_should_retry_alloc(inode->i_sb, &retries));
5678         if (IS_ERR(bh))
5679                 return PTR_ERR(bh);
5680         if (!bh)
5681                 goto out;
5682         BUFFER_TRACE(bh, "get write access");
5683         err = ext4_journal_get_write_access(handle, bh);
5684         if (err) {
5685                 brelse(bh);
5686                 return err;
5687         }
5688         lock_buffer(bh);
5689         memcpy(bh->b_data+offset, data, len);
5690         flush_dcache_page(bh->b_page);
5691         unlock_buffer(bh);
5692         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5693         brelse(bh);
5694 out:
5695         if (inode->i_size < off + len) {
5696                 i_size_write(inode, off + len);
5697                 EXT4_I(inode)->i_disksize = inode->i_size;
5698                 ext4_mark_inode_dirty(handle, inode);
5699         }
5700         return len;
5701 }
5702
5703 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5704 {
5705         const struct quota_format_ops   *ops;
5706
5707         if (!sb_has_quota_loaded(sb, qid->type))
5708                 return -ESRCH;
5709         ops = sb_dqopt(sb)->ops[qid->type];
5710         if (!ops || !ops->get_next_id)
5711                 return -ENOSYS;
5712         return dquot_get_next_id(sb, qid);
5713 }
5714 #endif
5715
5716 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5717                        const char *dev_name, void *data)
5718 {
5719         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5720 }
5721
5722 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5723 static inline void register_as_ext2(void)
5724 {
5725         int err = register_filesystem(&ext2_fs_type);
5726         if (err)
5727                 printk(KERN_WARNING
5728                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5729 }
5730
5731 static inline void unregister_as_ext2(void)
5732 {
5733         unregister_filesystem(&ext2_fs_type);
5734 }
5735
5736 static inline int ext2_feature_set_ok(struct super_block *sb)
5737 {
5738         if (ext4_has_unknown_ext2_incompat_features(sb))
5739                 return 0;
5740         if (sb_rdonly(sb))
5741                 return 1;
5742         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5743                 return 0;
5744         return 1;
5745 }
5746 #else
5747 static inline void register_as_ext2(void) { }
5748 static inline void unregister_as_ext2(void) { }
5749 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5750 #endif
5751
5752 static inline void register_as_ext3(void)
5753 {
5754         int err = register_filesystem(&ext3_fs_type);
5755         if (err)
5756                 printk(KERN_WARNING
5757                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5758 }
5759
5760 static inline void unregister_as_ext3(void)
5761 {
5762         unregister_filesystem(&ext3_fs_type);
5763 }
5764
5765 static inline int ext3_feature_set_ok(struct super_block *sb)
5766 {
5767         if (ext4_has_unknown_ext3_incompat_features(sb))
5768                 return 0;
5769         if (!ext4_has_feature_journal(sb))
5770                 return 0;
5771         if (sb_rdonly(sb))
5772                 return 1;
5773         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5774                 return 0;
5775         return 1;
5776 }
5777
5778 static struct file_system_type ext4_fs_type = {
5779         .owner          = THIS_MODULE,
5780         .name           = "ext4",
5781         .mount          = ext4_mount,
5782         .kill_sb        = kill_block_super,
5783         .fs_flags       = FS_REQUIRES_DEV,
5784 };
5785 MODULE_ALIAS_FS("ext4");
5786
5787 /* Shared across all ext4 file systems */
5788 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5789
5790 static int __init ext4_init_fs(void)
5791 {
5792         int i, err;
5793
5794         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5795         ext4_li_info = NULL;
5796         mutex_init(&ext4_li_mtx);
5797
5798         /* Build-time check for flags consistency */
5799         ext4_check_flag_values();
5800
5801         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5802                 init_waitqueue_head(&ext4__ioend_wq[i]);
5803
5804         err = ext4_init_es();
5805         if (err)
5806                 return err;
5807
5808         err = ext4_init_pageio();
5809         if (err)
5810                 goto out5;
5811
5812         err = ext4_init_system_zone();
5813         if (err)
5814                 goto out4;
5815
5816         err = ext4_init_sysfs();
5817         if (err)
5818                 goto out3;
5819
5820         err = ext4_init_mballoc();
5821         if (err)
5822                 goto out2;
5823         err = init_inodecache();
5824         if (err)
5825                 goto out1;
5826         register_as_ext3();
5827         register_as_ext2();
5828         err = register_filesystem(&ext4_fs_type);
5829         if (err)
5830                 goto out;
5831
5832         return 0;
5833 out:
5834         unregister_as_ext2();
5835         unregister_as_ext3();
5836         destroy_inodecache();
5837 out1:
5838         ext4_exit_mballoc();
5839 out2:
5840         ext4_exit_sysfs();
5841 out3:
5842         ext4_exit_system_zone();
5843 out4:
5844         ext4_exit_pageio();
5845 out5:
5846         ext4_exit_es();
5847
5848         return err;
5849 }
5850
5851 static void __exit ext4_exit_fs(void)
5852 {
5853         ext4_destroy_lazyinit_thread();
5854         unregister_as_ext2();
5855         unregister_as_ext3();
5856         unregister_filesystem(&ext4_fs_type);
5857         destroy_inodecache();
5858         ext4_exit_mballoc();
5859         ext4_exit_sysfs();
5860         ext4_exit_system_zone();
5861         ext4_exit_pageio();
5862         ext4_exit_es();
5863 }
5864
5865 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5866 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5867 MODULE_LICENSE("GPL");
5868 module_init(ext4_init_fs)
5869 module_exit(ext4_exit_fs)