fs/ext4/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/super.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card ([email protected])
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller ([email protected]), 1995
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/parser.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/dax.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/uaccess.h>
  44 #include <linux/iversion.h>
  45
  46 #include <linux/kthread.h>
  47 #include <linux/freezer.h>
  48
  49 #include "ext4.h"
  50 #include "ext4_extents.h"       /* Needed for trace points definition */
  51 #include "ext4_jbd2.h"
  52 #include "xattr.h"
  53 #include "acl.h"
  54 #include "mballoc.h"
  55 #include "fsmap.h"
  56
  57 #define CREATE_TRACE_POINTS
  58 #include <trace/events/ext4.h>
  59
  60 static struct ext4_lazy_init *ext4_li_info;
  61 static struct mutex ext4_li_mtx;
  62 static struct ratelimit_state ext4_mount_msg_ratelimit;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static struct inode *ext4_get_journal_inode(struct super_block *sb,
  86                                             unsigned int journal_inum);
  87
  88 /*
  89  * Lock ordering
  90  *
  91  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
  92  * i_mmap_rwsem (inode->i_mmap_rwsem)!
  93  *
  94  * page fault path:
  95  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
  96  *   page lock -> i_data_sem (rw)
  97  *
  98  * buffered write path:
  99  * sb_start_write -> i_mutex -> mmap_sem
 100  * sb_start_write -> i_mutex -> transaction start -> page lock ->
 101  *   i_data_sem (rw)
 102  *
 103  * truncate:
 104  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> i_mmap_rwsem (w) -> page lock
 105  * sb_start_write -> i_mutex -> i_mmap_sem (w) -> transaction start ->
 106  *   i_data_sem (rw)
 107  *
 108  * direct IO:
 109  * sb_start_write -> i_mutex -> mmap_sem
 110  * sb_start_write -> i_mutex -> 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 (test_opt(sb, WARN_ON_ERROR))
 409                 WARN_ON_ONCE(1);
 410
 411         if (sb_rdonly(sb))
 412                 return;
 413
 414         if (!test_opt(sb, ERRORS_CONT)) {
 415                 journal_t *journal = EXT4_SB(sb)->s_journal;
 416
 417                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 418                 if (journal)
 419                         jbd2_journal_abort(journal, -EIO);
 420         }
 421         if (test_opt(sb, ERRORS_RO)) {
 422                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 423                 /*
 424                  * Make sure updated value of ->s_mount_flags will be visible
 425                  * before ->s_flags update
 426                  */
 427                 smp_wmb();
 428                 sb->s_flags |= SB_RDONLY;
 429         }
 430         if (test_opt(sb, ERRORS_PANIC)) {
 431                 if (EXT4_SB(sb)->s_journal &&
 432                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 433                         return;
 434                 panic("EXT4-fs (device %s): panic forced after error\n",
 435                         sb->s_id);
 436         }
 437 }
 438
 439 #define ext4_error_ratelimit(sb)                                        \
 440                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 441                              "EXT4-fs error")
 442
 443 void __ext4_error(struct super_block *sb, const char *function,
 444                   unsigned int line, const char *fmt, ...)
 445 {
 446         struct va_format vaf;
 447         va_list args;
 448
 449         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 450                 return;
 451
 452         trace_ext4_error(sb, function, line);
 453         if (ext4_error_ratelimit(sb)) {
 454                 va_start(args, fmt);
 455                 vaf.fmt = fmt;
 456                 vaf.va = &args;
 457                 printk(KERN_CRIT
 458                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 459                        sb->s_id, function, line, current->comm, &vaf);
 460                 va_end(args);
 461         }
 462         save_error_info(sb, function, line);
 463         ext4_handle_error(sb);
 464 }
 465
 466 void __ext4_error_inode(struct inode *inode, const char *function,
 467                         unsigned int line, ext4_fsblk_t block,
 468                         const char *fmt, ...)
 469 {
 470         va_list args;
 471         struct va_format vaf;
 472         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 473
 474         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 475                 return;
 476
 477         trace_ext4_error(inode->i_sb, function, line);
 478         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 479         es->s_last_error_block = cpu_to_le64(block);
 480         if (ext4_error_ratelimit(inode->i_sb)) {
 481                 va_start(args, fmt);
 482                 vaf.fmt = fmt;
 483                 vaf.va = &args;
 484                 if (block)
 485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 486                                "inode #%lu: block %llu: comm %s: %pV\n",
 487                                inode->i_sb->s_id, function, line, inode->i_ino,
 488                                block, current->comm, &vaf);
 489                 else
 490                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 491                                "inode #%lu: comm %s: %pV\n",
 492                                inode->i_sb->s_id, function, line, inode->i_ino,
 493                                current->comm, &vaf);
 494                 va_end(args);
 495         }
 496         save_error_info(inode->i_sb, function, line);
 497         ext4_handle_error(inode->i_sb);
 498 }
 499
 500 void __ext4_error_file(struct file *file, const char *function,
 501                        unsigned int line, ext4_fsblk_t block,
 502                        const char *fmt, ...)
 503 {
 504         va_list args;
 505         struct va_format vaf;
 506         struct ext4_super_block *es;
 507         struct inode *inode = file_inode(file);
 508         char pathname[80], *path;
 509
 510         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 511                 return;
 512
 513         trace_ext4_error(inode->i_sb, function, line);
 514         es = EXT4_SB(inode->i_sb)->s_es;
 515         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 516         if (ext4_error_ratelimit(inode->i_sb)) {
 517                 path = file_path(file, pathname, sizeof(pathname));
 518                 if (IS_ERR(path))
 519                         path = "(unknown)";
 520                 va_start(args, fmt);
 521                 vaf.fmt = fmt;
 522                 vaf.va = &args;
 523                 if (block)
 524                         printk(KERN_CRIT
 525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 526                                "block %llu: comm %s: path %s: %pV\n",
 527                                inode->i_sb->s_id, function, line, inode->i_ino,
 528                                block, current->comm, path, &vaf);
 529                 else
 530                         printk(KERN_CRIT
 531                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 532                                "comm %s: path %s: %pV\n",
 533                                inode->i_sb->s_id, function, line, inode->i_ino,
 534                                current->comm, path, &vaf);
 535                 va_end(args);
 536         }
 537         save_error_info(inode->i_sb, function, line);
 538         ext4_handle_error(inode->i_sb);
 539 }
 540
 541 const char *ext4_decode_error(struct super_block *sb, int errno,
 542                               char nbuf[16])
 543 {
 544         char *errstr = NULL;
 545
 546         switch (errno) {
 547         case -EFSCORRUPTED:
 548                 errstr = "Corrupt filesystem";
 549                 break;
 550         case -EFSBADCRC:
 551                 errstr = "Filesystem failed CRC";
 552                 break;
 553         case -EIO:
 554                 errstr = "IO failure";
 555                 break;
 556         case -ENOMEM:
 557                 errstr = "Out of memory";
 558                 break;
 559         case -EROFS:
 560                 if (!sb || (EXT4_SB(sb)->s_journal &&
 561                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 562                         errstr = "Journal has aborted";
 563                 else
 564                         errstr = "Readonly filesystem";
 565                 break;
 566         default:
 567                 /* If the caller passed in an extra buffer for unknown
 568                  * errors, textualise them now.  Else we just return
 569                  * NULL. */
 570                 if (nbuf) {
 571                         /* Check for truncated error codes... */
 572                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 573                                 errstr = nbuf;
 574                 }
 575                 break;
 576         }
 577
 578         return errstr;
 579 }
 580
 581 /* __ext4_std_error decodes expected errors from journaling functions
 582  * automatically and invokes the appropriate error response.  */
 583
 584 void __ext4_std_error(struct super_block *sb, const char *function,
 585                       unsigned int line, int errno)
 586 {
 587         char nbuf[16];
 588         const char *errstr;
 589
 590         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 591                 return;
 592
 593         /* Special case: if the error is EROFS, and we're not already
 594          * inside a transaction, then there's really no point in logging
 595          * an error. */
 596         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 597                 return;
 598
 599         if (ext4_error_ratelimit(sb)) {
 600                 errstr = ext4_decode_error(sb, errno, nbuf);
 601                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 602                        sb->s_id, function, line, errstr);
 603         }
 604
 605         save_error_info(sb, function, line);
 606         ext4_handle_error(sb);
 607 }
 608
 609 /*
 610  * ext4_abort is a much stronger failure handler than ext4_error.  The
 611  * abort function may be used to deal with unrecoverable failures such
 612  * as journal IO errors or ENOMEM at a critical moment in log management.
 613  *
 614  * We unconditionally force the filesystem into an ABORT|READONLY state,
 615  * unless the error response on the fs has been set to panic in which
 616  * case we take the easy way out and panic immediately.
 617  */
 618
 619 void __ext4_abort(struct super_block *sb, const char *function,
 620                 unsigned int line, const char *fmt, ...)
 621 {
 622         struct va_format vaf;
 623         va_list args;
 624
 625         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 626                 return;
 627
 628         save_error_info(sb, function, line);
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635
 636         if (sb_rdonly(sb) == 0) {
 637                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 638                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 639                 /*
 640                  * Make sure updated value of ->s_mount_flags will be visible
 641                  * before ->s_flags update
 642                  */
 643                 smp_wmb();
 644                 sb->s_flags |= SB_RDONLY;
 645                 if (EXT4_SB(sb)->s_journal)
 646                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 647                 save_error_info(sb, function, line);
 648         }
 649         if (test_opt(sb, ERRORS_PANIC)) {
 650                 if (EXT4_SB(sb)->s_journal &&
 651                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 652                         return;
 653                 panic("EXT4-fs panic from previous error\n");
 654         }
 655 }
 656
 657 void __ext4_msg(struct super_block *sb,
 658                 const char *prefix, const char *fmt, ...)
 659 {
 660         struct va_format vaf;
 661         va_list args;
 662
 663         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 664                 return;
 665
 666         va_start(args, fmt);
 667         vaf.fmt = fmt;
 668         vaf.va = &args;
 669         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 670         va_end(args);
 671 }
 672
 673 #define ext4_warning_ratelimit(sb)                                      \
 674                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 675                              "EXT4-fs warning")
 676
 677 void __ext4_warning(struct super_block *sb, const char *function,
 678                     unsigned int line, const char *fmt, ...)
 679 {
 680         struct va_format vaf;
 681         va_list args;
 682
 683         if (!ext4_warning_ratelimit(sb))
 684                 return;
 685
 686         va_start(args, fmt);
 687         vaf.fmt = fmt;
 688         vaf.va = &args;
 689         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 690                sb->s_id, function, line, &vaf);
 691         va_end(args);
 692 }
 693
 694 void __ext4_warning_inode(const struct inode *inode, const char *function,
 695                           unsigned int line, const char *fmt, ...)
 696 {
 697         struct va_format vaf;
 698         va_list args;
 699
 700         if (!ext4_warning_ratelimit(inode->i_sb))
 701                 return;
 702
 703         va_start(args, fmt);
 704         vaf.fmt = fmt;
 705         vaf.va = &args;
 706         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 707                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 708                function, line, inode->i_ino, current->comm, &vaf);
 709         va_end(args);
 710 }
 711
 712 void __ext4_grp_locked_error(const char *function, unsigned int line,
 713                              struct super_block *sb, ext4_group_t grp,
 714                              unsigned long ino, ext4_fsblk_t block,
 715                              const char *fmt, ...)
 716 __releases(bitlock)
 717 __acquires(bitlock)
 718 {
 719         struct va_format vaf;
 720         va_list args;
 721         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 722
 723         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 724                 return;
 725
 726         trace_ext4_error(sb, function, line);
 727         es->s_last_error_ino = cpu_to_le32(ino);
 728         es->s_last_error_block = cpu_to_le64(block);
 729         __save_error_info(sb, function, line);
 730
 731         if (ext4_error_ratelimit(sb)) {
 732                 va_start(args, fmt);
 733                 vaf.fmt = fmt;
 734                 vaf.va = &args;
 735                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 736                        sb->s_id, function, line, grp);
 737                 if (ino)
 738                         printk(KERN_CONT "inode %lu: ", ino);
 739                 if (block)
 740                         printk(KERN_CONT "block %llu:",
 741                                (unsigned long long) block);
 742                 printk(KERN_CONT "%pV\n", &vaf);
 743                 va_end(args);
 744         }
 745
 746         if (test_opt(sb, WARN_ON_ERROR))
 747                 WARN_ON_ONCE(1);
 748
 749         if (test_opt(sb, ERRORS_CONT)) {
 750                 ext4_commit_super(sb, 0);
 751                 return;
 752         }
 753
 754         ext4_unlock_group(sb, grp);
 755         ext4_commit_super(sb, 1);
 756         ext4_handle_error(sb);
 757         /*
 758          * We only get here in the ERRORS_RO case; relocking the group
 759          * may be dangerous, but nothing bad will happen since the
 760          * filesystem will have already been marked read/only and the
 761          * journal has been aborted.  We return 1 as a hint to callers
 762          * who might what to use the return value from
 763          * ext4_grp_locked_error() to distinguish between the
 764          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 765          * aggressively from the ext4 function in question, with a
 766          * more appropriate error code.
 767          */
 768         ext4_lock_group(sb, grp);
 769         return;
 770 }
 771
 772 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 773                                      ext4_group_t group,
 774                                      unsigned int flags)
 775 {
 776         struct ext4_sb_info *sbi = EXT4_SB(sb);
 777         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 778         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 779
 780         if ((flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) &&
 781             !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) {
 782                 percpu_counter_sub(&sbi->s_freeclusters_counter,
 783                                         grp->bb_free);
 784                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 785                         &grp->bb_state);
 786         }
 787
 788         if ((flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) &&
 789             !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 790                 if (gdp) {
 791                         int count;
 792
 793                         count = ext4_free_inodes_count(sb, gdp);
 794                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 795                                            count);
 796                 }
 797                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 798                         &grp->bb_state);
 799         }
 800 }
 801
 802 void ext4_update_dynamic_rev(struct super_block *sb)
 803 {
 804         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 805
 806         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 807                 return;
 808
 809         ext4_warning(sb,
 810                      "updating to rev %d because of new feature flag, "
 811                      "running e2fsck is recommended",
 812                      EXT4_DYNAMIC_REV);
 813
 814         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 815         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 816         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 817         /* leave es->s_feature_*compat flags alone */
 818         /* es->s_uuid will be set by e2fsck if empty */
 819
 820         /*
 821          * The rest of the superblock fields should be zero, and if not it
 822          * means they are likely already in use, so leave them alone.  We
 823          * can leave it up to e2fsck to clean up any inconsistencies there.
 824          */
 825 }
 826
 827 /*
 828  * Open the external journal device
 829  */
 830 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 831 {
 832         struct block_device *bdev;
 833         char b[BDEVNAME_SIZE];
 834
 835         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 836         if (IS_ERR(bdev))
 837                 goto fail;
 838         return bdev;
 839
 840 fail:
 841         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 842                         __bdevname(dev, b), PTR_ERR(bdev));
 843         return NULL;
 844 }
 845
 846 /*
 847  * Release the journal device
 848  */
 849 static void ext4_blkdev_put(struct block_device *bdev)
 850 {
 851         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 852 }
 853
 854 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 855 {
 856         struct block_device *bdev;
 857         bdev = sbi->journal_bdev;
 858         if (bdev) {
 859                 ext4_blkdev_put(bdev);
 860                 sbi->journal_bdev = NULL;
 861         }
 862 }
 863
 864 static inline struct inode *orphan_list_entry(struct list_head *l)
 865 {
 866         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 867 }
 868
 869 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 870 {
 871         struct list_head *l;
 872
 873         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 874                  le32_to_cpu(sbi->s_es->s_last_orphan));
 875
 876         printk(KERN_ERR "sb_info orphan list:\n");
 877         list_for_each(l, &sbi->s_orphan) {
 878                 struct inode *inode = orphan_list_entry(l);
 879                 printk(KERN_ERR "  "
 880                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 881                        inode->i_sb->s_id, inode->i_ino, inode,
 882                        inode->i_mode, inode->i_nlink,
 883                        NEXT_ORPHAN(inode));
 884         }
 885 }
 886
 887 #ifdef CONFIG_QUOTA
 888 static int ext4_quota_off(struct super_block *sb, int type);
 889
 890 static inline void ext4_quota_off_umount(struct super_block *sb)
 891 {
 892         int type;
 893
 894         /* Use our quota_off function to clear inode flags etc. */
 895         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 896                 ext4_quota_off(sb, type);
 897 }
 898 #else
 899 static inline void ext4_quota_off_umount(struct super_block *sb)
 900 {
 901 }
 902 #endif
 903
 904 static void ext4_put_super(struct super_block *sb)
 905 {
 906         struct ext4_sb_info *sbi = EXT4_SB(sb);
 907         struct ext4_super_block *es = sbi->s_es;
 908         int aborted = 0;
 909         int i, err;
 910
 911         ext4_unregister_li_request(sb);
 912         ext4_quota_off_umount(sb);
 913
 914         destroy_workqueue(sbi->rsv_conversion_wq);
 915
 916         if (sbi->s_journal) {
 917                 aborted = is_journal_aborted(sbi->s_journal);
 918                 err = jbd2_journal_destroy(sbi->s_journal);
 919                 sbi->s_journal = NULL;
 920                 if ((err < 0) && !aborted)
 921                         ext4_abort(sb, "Couldn't clean up the journal");
 922         }
 923
 924         ext4_unregister_sysfs(sb);
 925         ext4_es_unregister_shrinker(sbi);
 926         del_timer_sync(&sbi->s_err_report);
 927         ext4_release_system_zone(sb);
 928         ext4_mb_release(sb);
 929         ext4_ext_release(sb);
 930
 931         if (!sb_rdonly(sb) && !aborted) {
 932                 ext4_clear_feature_journal_needs_recovery(sb);
 933                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 934         }
 935         if (!sb_rdonly(sb))
 936                 ext4_commit_super(sb, 1);
 937
 938         for (i = 0; i < sbi->s_gdb_count; i++)
 939                 brelse(sbi->s_group_desc[i]);
 940         kvfree(sbi->s_group_desc);
 941         kvfree(sbi->s_flex_groups);
 942         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 943         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 944         percpu_counter_destroy(&sbi->s_dirs_counter);
 945         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 946         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 947 #ifdef CONFIG_QUOTA
 948         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 949                 kfree(sbi->s_qf_names[i]);
 950 #endif
 951
 952         /* Debugging code just in case the in-memory inode orphan list
 953          * isn't empty.  The on-disk one can be non-empty if we've
 954          * detected an error and taken the fs readonly, but the
 955          * in-memory list had better be clean by this point. */
 956         if (!list_empty(&sbi->s_orphan))
 957                 dump_orphan_list(sb, sbi);
 958         J_ASSERT(list_empty(&sbi->s_orphan));
 959
 960         sync_blockdev(sb->s_bdev);
 961         invalidate_bdev(sb->s_bdev);
 962         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 963                 /*
 964                  * Invalidate the journal device's buffers.  We don't want them
 965                  * floating about in memory - the physical journal device may
 966                  * hotswapped, and it breaks the `ro-after' testing code.
 967                  */
 968                 sync_blockdev(sbi->journal_bdev);
 969                 invalidate_bdev(sbi->journal_bdev);
 970                 ext4_blkdev_remove(sbi);
 971         }
 972         if (sbi->s_ea_inode_cache) {
 973                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 974                 sbi->s_ea_inode_cache = NULL;
 975         }
 976         if (sbi->s_ea_block_cache) {
 977                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 978                 sbi->s_ea_block_cache = NULL;
 979         }
 980         if (sbi->s_mmp_tsk)
 981                 kthread_stop(sbi->s_mmp_tsk);
 982         brelse(sbi->s_sbh);
 983         sb->s_fs_info = NULL;
 984         /*
 985          * Now that we are completely done shutting down the
 986          * superblock, we need to actually destroy the kobject.
 987          */
 988         kobject_put(&sbi->s_kobj);
 989         wait_for_completion(&sbi->s_kobj_unregister);
 990         if (sbi->s_chksum_driver)
 991                 crypto_free_shash(sbi->s_chksum_driver);
 992         kfree(sbi->s_blockgroup_lock);
 993         fs_put_dax(sbi->s_daxdev);
 994         kfree(sbi);
 995 }
 996
 997 static struct kmem_cache *ext4_inode_cachep;
 998
 999 /*
1000  * Called inside transaction, so use GFP_NOFS
1001  */
1002 static struct inode *ext4_alloc_inode(struct super_block *sb)
1003 {
1004         struct ext4_inode_info *ei;
1005
1006         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1007         if (!ei)
1008                 return NULL;
1009
1010         inode_set_iversion(&ei->vfs_inode, 1);
1011         spin_lock_init(&ei->i_raw_lock);
1012         INIT_LIST_HEAD(&ei->i_prealloc_list);
1013         spin_lock_init(&ei->i_prealloc_lock);
1014         ext4_es_init_tree(&ei->i_es_tree);
1015         rwlock_init(&ei->i_es_lock);
1016         INIT_LIST_HEAD(&ei->i_es_list);
1017         ei->i_es_all_nr = 0;
1018         ei->i_es_shk_nr = 0;
1019         ei->i_es_shrink_lblk = 0;
1020         ei->i_reserved_data_blocks = 0;
1021         ei->i_da_metadata_calc_len = 0;
1022         ei->i_da_metadata_calc_last_lblock = 0;
1023         spin_lock_init(&(ei->i_block_reservation_lock));
1024 #ifdef CONFIG_QUOTA
1025         ei->i_reserved_quota = 0;
1026         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1027 #endif
1028         ei->jinode = NULL;
1029         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1030         spin_lock_init(&ei->i_completed_io_lock);
1031         ei->i_sync_tid = 0;
1032         ei->i_datasync_tid = 0;
1033         atomic_set(&ei->i_unwritten, 0);
1034         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1035         return &ei->vfs_inode;
1036 }
1037
1038 static int ext4_drop_inode(struct inode *inode)
1039 {
1040         int drop = generic_drop_inode(inode);
1041
1042         trace_ext4_drop_inode(inode, drop);
1043         return drop;
1044 }
1045
1046 static void ext4_i_callback(struct rcu_head *head)
1047 {
1048         struct inode *inode = container_of(head, struct inode, i_rcu);
1049         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1050 }
1051
1052 static void ext4_destroy_inode(struct inode *inode)
1053 {
1054         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1055                 ext4_msg(inode->i_sb, KERN_ERR,
1056                          "Inode %lu (%p): orphan list check failed!",
1057                          inode->i_ino, EXT4_I(inode));
1058                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1059                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1060                                 true);
1061                 dump_stack();
1062         }
1063         call_rcu(&inode->i_rcu, ext4_i_callback);
1064 }
1065
1066 static void init_once(void *foo)
1067 {
1068         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1069
1070         INIT_LIST_HEAD(&ei->i_orphan);
1071         init_rwsem(&ei->xattr_sem);
1072         init_rwsem(&ei->i_data_sem);
1073         init_rwsem(&ei->i_mmap_sem);
1074         inode_init_once(&ei->vfs_inode);
1075 }
1076
1077 static int __init init_inodecache(void)
1078 {
1079         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1080                                 sizeof(struct ext4_inode_info), 0,
1081                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1082                                         SLAB_ACCOUNT),
1083                                 offsetof(struct ext4_inode_info, i_data),
1084                                 sizeof_field(struct ext4_inode_info, i_data),
1085                                 init_once);
1086         if (ext4_inode_cachep == NULL)
1087                 return -ENOMEM;
1088         return 0;
1089 }
1090
1091 static void destroy_inodecache(void)
1092 {
1093         /*
1094          * Make sure all delayed rcu free inodes are flushed before we
1095          * destroy cache.
1096          */
1097         rcu_barrier();
1098         kmem_cache_destroy(ext4_inode_cachep);
1099 }
1100
1101 void ext4_clear_inode(struct inode *inode)
1102 {
1103         invalidate_inode_buffers(inode);
1104         clear_inode(inode);
1105         dquot_drop(inode);
1106         ext4_discard_preallocations(inode);
1107         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1108         if (EXT4_I(inode)->jinode) {
1109                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1110                                                EXT4_I(inode)->jinode);
1111                 jbd2_free_inode(EXT4_I(inode)->jinode);
1112                 EXT4_I(inode)->jinode = NULL;
1113         }
1114         fscrypt_put_encryption_info(inode);
1115 }
1116
1117 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1118                                         u64 ino, u32 generation)
1119 {
1120         struct inode *inode;
1121
1122         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1123                 return ERR_PTR(-ESTALE);
1124         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1125                 return ERR_PTR(-ESTALE);
1126
1127         /* iget isn't really right if the inode is currently unallocated!!
1128          *
1129          * ext4_read_inode will return a bad_inode if the inode had been
1130          * deleted, so we should be safe.
1131          *
1132          * Currently we don't know the generation for parent directory, so
1133          * a generation of 0 means "accept any"
1134          */
1135         inode = ext4_iget_normal(sb, ino);
1136         if (IS_ERR(inode))
1137                 return ERR_CAST(inode);
1138         if (generation && inode->i_generation != generation) {
1139                 iput(inode);
1140                 return ERR_PTR(-ESTALE);
1141         }
1142
1143         return inode;
1144 }
1145
1146 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1147                                         int fh_len, int fh_type)
1148 {
1149         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1150                                     ext4_nfs_get_inode);
1151 }
1152
1153 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1154                                         int fh_len, int fh_type)
1155 {
1156         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1157                                     ext4_nfs_get_inode);
1158 }
1159
1160 /*
1161  * Try to release metadata pages (indirect blocks, directories) which are
1162  * mapped via the block device.  Since these pages could have journal heads
1163  * which would prevent try_to_free_buffers() from freeing them, we must use
1164  * jbd2 layer's try_to_free_buffers() function to release them.
1165  */
1166 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1167                                  gfp_t wait)
1168 {
1169         journal_t *journal = EXT4_SB(sb)->s_journal;
1170
1171         WARN_ON(PageChecked(page));
1172         if (!page_has_buffers(page))
1173                 return 0;
1174         if (journal)
1175                 return jbd2_journal_try_to_free_buffers(journal, page,
1176                                                 wait & ~__GFP_DIRECT_RECLAIM);
1177         return try_to_free_buffers(page);
1178 }
1179
1180 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1181 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1182 {
1183         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1184                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1185 }
1186
1187 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1188                                                         void *fs_data)
1189 {
1190         handle_t *handle = fs_data;
1191         int res, res2, credits, retries = 0;
1192
1193         /*
1194          * Encrypting the root directory is not allowed because e2fsck expects
1195          * lost+found to exist and be unencrypted, and encrypting the root
1196          * directory would imply encrypting the lost+found directory as well as
1197          * the filename "lost+found" itself.
1198          */
1199         if (inode->i_ino == EXT4_ROOT_INO)
1200                 return -EPERM;
1201
1202         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1203                 return -EINVAL;
1204
1205         res = ext4_convert_inline_data(inode);
1206         if (res)
1207                 return res;
1208
1209         /*
1210          * If a journal handle was specified, then the encryption context is
1211          * being set on a new inode via inheritance and is part of a larger
1212          * transaction to create the inode.  Otherwise the encryption context is
1213          * being set on an existing inode in its own transaction.  Only in the
1214          * latter case should the "retry on ENOSPC" logic be used.
1215          */
1216
1217         if (handle) {
1218                 res = ext4_xattr_set_handle(handle, inode,
1219                                             EXT4_XATTR_INDEX_ENCRYPTION,
1220                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1221                                             ctx, len, 0);
1222                 if (!res) {
1223                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1224                         ext4_clear_inode_state(inode,
1225                                         EXT4_STATE_MAY_INLINE_DATA);
1226                         /*
1227                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1228                          * S_DAX may be disabled
1229                          */
1230                         ext4_set_inode_flags(inode);
1231                 }
1232                 return res;
1233         }
1234
1235         res = dquot_initialize(inode);
1236         if (res)
1237                 return res;
1238 retry:
1239         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1240                                      &credits);
1241         if (res)
1242                 return res;
1243
1244         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1245         if (IS_ERR(handle))
1246                 return PTR_ERR(handle);
1247
1248         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1249                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1250                                     ctx, len, 0);
1251         if (!res) {
1252                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1253                 /*
1254                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1255                  * S_DAX may be disabled
1256                  */
1257                 ext4_set_inode_flags(inode);
1258                 res = ext4_mark_inode_dirty(handle, inode);
1259                 if (res)
1260                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1261         }
1262         res2 = ext4_journal_stop(handle);
1263
1264         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1265                 goto retry;
1266         if (!res)
1267                 res = res2;
1268         return res;
1269 }
1270
1271 static bool ext4_dummy_context(struct inode *inode)
1272 {
1273         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1274 }
1275
1276 static const struct fscrypt_operations ext4_cryptops = {
1277         .key_prefix             = "ext4:",
1278         .get_context            = ext4_get_context,
1279         .set_context            = ext4_set_context,
1280         .dummy_context          = ext4_dummy_context,
1281         .empty_dir              = ext4_empty_dir,
1282         .max_namelen            = EXT4_NAME_LEN,
1283 };
1284 #endif
1285
1286 #ifdef CONFIG_QUOTA
1287 static const char * const quotatypes[] = INITQFNAMES;
1288 #define QTYPE2NAME(t) (quotatypes[t])
1289
1290 static int ext4_write_dquot(struct dquot *dquot);
1291 static int ext4_acquire_dquot(struct dquot *dquot);
1292 static int ext4_release_dquot(struct dquot *dquot);
1293 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1294 static int ext4_write_info(struct super_block *sb, int type);
1295 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1296                          const struct path *path);
1297 static int ext4_quota_on_mount(struct super_block *sb, int type);
1298 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1299                                size_t len, loff_t off);
1300 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1301                                 const char *data, size_t len, loff_t off);
1302 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1303                              unsigned int flags);
1304 static int ext4_enable_quotas(struct super_block *sb);
1305 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1306
1307 static struct dquot **ext4_get_dquots(struct inode *inode)
1308 {
1309         return EXT4_I(inode)->i_dquot;
1310 }
1311
1312 static const struct dquot_operations ext4_quota_operations = {
1313         .get_reserved_space     = ext4_get_reserved_space,
1314         .write_dquot            = ext4_write_dquot,
1315         .acquire_dquot          = ext4_acquire_dquot,
1316         .release_dquot          = ext4_release_dquot,
1317         .mark_dirty             = ext4_mark_dquot_dirty,
1318         .write_info             = ext4_write_info,
1319         .alloc_dquot            = dquot_alloc,
1320         .destroy_dquot          = dquot_destroy,
1321         .get_projid             = ext4_get_projid,
1322         .get_inode_usage        = ext4_get_inode_usage,
1323         .get_next_id            = ext4_get_next_id,
1324 };
1325
1326 static const struct quotactl_ops ext4_qctl_operations = {
1327         .quota_on       = ext4_quota_on,
1328         .quota_off      = ext4_quota_off,
1329         .quota_sync     = dquot_quota_sync,
1330         .get_state      = dquot_get_state,
1331         .set_info       = dquot_set_dqinfo,
1332         .get_dqblk      = dquot_get_dqblk,
1333         .set_dqblk      = dquot_set_dqblk,
1334         .get_nextdqblk  = dquot_get_next_dqblk,
1335 };
1336 #endif
1337
1338 static const struct super_operations ext4_sops = {
1339         .alloc_inode    = ext4_alloc_inode,
1340         .destroy_inode  = ext4_destroy_inode,
1341         .write_inode    = ext4_write_inode,
1342         .dirty_inode    = ext4_dirty_inode,
1343         .drop_inode     = ext4_drop_inode,
1344         .evict_inode    = ext4_evict_inode,
1345         .put_super      = ext4_put_super,
1346         .sync_fs        = ext4_sync_fs,
1347         .freeze_fs      = ext4_freeze,
1348         .unfreeze_fs    = ext4_unfreeze,
1349         .statfs         = ext4_statfs,
1350         .remount_fs     = ext4_remount,
1351         .show_options   = ext4_show_options,
1352 #ifdef CONFIG_QUOTA
1353         .quota_read     = ext4_quota_read,
1354         .quota_write    = ext4_quota_write,
1355         .get_dquots     = ext4_get_dquots,
1356 #endif
1357         .bdev_try_to_free_page = bdev_try_to_free_page,
1358 };
1359
1360 static const struct export_operations ext4_export_ops = {
1361         .fh_to_dentry = ext4_fh_to_dentry,
1362         .fh_to_parent = ext4_fh_to_parent,
1363         .get_parent = ext4_get_parent,
1364 };
1365
1366 enum {
1367         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1368         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1369         Opt_nouid32, Opt_debug, Opt_removed,
1370         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1371         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1372         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1373         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1374         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1375         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1376         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1377         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1378         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1379         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1380         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1381         Opt_nowarn_on_error, Opt_mblk_io_submit,
1382         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1383         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1384         Opt_inode_readahead_blks, Opt_journal_ioprio,
1385         Opt_dioread_nolock, Opt_dioread_lock,
1386         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1387         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1388 };
1389
1390 static const match_table_t tokens = {
1391         {Opt_bsd_df, "bsddf"},
1392         {Opt_minix_df, "minixdf"},
1393         {Opt_grpid, "grpid"},
1394         {Opt_grpid, "bsdgroups"},
1395         {Opt_nogrpid, "nogrpid"},
1396         {Opt_nogrpid, "sysvgroups"},
1397         {Opt_resgid, "resgid=%u"},
1398         {Opt_resuid, "resuid=%u"},
1399         {Opt_sb, "sb=%u"},
1400         {Opt_err_cont, "errors=continue"},
1401         {Opt_err_panic, "errors=panic"},
1402         {Opt_err_ro, "errors=remount-ro"},
1403         {Opt_nouid32, "nouid32"},
1404         {Opt_debug, "debug"},
1405         {Opt_removed, "oldalloc"},
1406         {Opt_removed, "orlov"},
1407         {Opt_user_xattr, "user_xattr"},
1408         {Opt_nouser_xattr, "nouser_xattr"},
1409         {Opt_acl, "acl"},
1410         {Opt_noacl, "noacl"},
1411         {Opt_noload, "norecovery"},
1412         {Opt_noload, "noload"},
1413         {Opt_removed, "nobh"},
1414         {Opt_removed, "bh"},
1415         {Opt_commit, "commit=%u"},
1416         {Opt_min_batch_time, "min_batch_time=%u"},
1417         {Opt_max_batch_time, "max_batch_time=%u"},
1418         {Opt_journal_dev, "journal_dev=%u"},
1419         {Opt_journal_path, "journal_path=%s"},
1420         {Opt_journal_checksum, "journal_checksum"},
1421         {Opt_nojournal_checksum, "nojournal_checksum"},
1422         {Opt_journal_async_commit, "journal_async_commit"},
1423         {Opt_abort, "abort"},
1424         {Opt_data_journal, "data=journal"},
1425         {Opt_data_ordered, "data=ordered"},
1426         {Opt_data_writeback, "data=writeback"},
1427         {Opt_data_err_abort, "data_err=abort"},
1428         {Opt_data_err_ignore, "data_err=ignore"},
1429         {Opt_offusrjquota, "usrjquota="},
1430         {Opt_usrjquota, "usrjquota=%s"},
1431         {Opt_offgrpjquota, "grpjquota="},
1432         {Opt_grpjquota, "grpjquota=%s"},
1433         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1434         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1435         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1436         {Opt_grpquota, "grpquota"},
1437         {Opt_noquota, "noquota"},
1438         {Opt_quota, "quota"},
1439         {Opt_usrquota, "usrquota"},
1440         {Opt_prjquota, "prjquota"},
1441         {Opt_barrier, "barrier=%u"},
1442         {Opt_barrier, "barrier"},
1443         {Opt_nobarrier, "nobarrier"},
1444         {Opt_i_version, "i_version"},
1445         {Opt_dax, "dax"},
1446         {Opt_stripe, "stripe=%u"},
1447         {Opt_delalloc, "delalloc"},
1448         {Opt_warn_on_error, "warn_on_error"},
1449         {Opt_nowarn_on_error, "nowarn_on_error"},
1450         {Opt_lazytime, "lazytime"},
1451         {Opt_nolazytime, "nolazytime"},
1452         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1453         {Opt_nodelalloc, "nodelalloc"},
1454         {Opt_removed, "mblk_io_submit"},
1455         {Opt_removed, "nomblk_io_submit"},
1456         {Opt_block_validity, "block_validity"},
1457         {Opt_noblock_validity, "noblock_validity"},
1458         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1459         {Opt_journal_ioprio, "journal_ioprio=%u"},
1460         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1461         {Opt_auto_da_alloc, "auto_da_alloc"},
1462         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1463         {Opt_dioread_nolock, "dioread_nolock"},
1464         {Opt_dioread_lock, "dioread_lock"},
1465         {Opt_discard, "discard"},
1466         {Opt_nodiscard, "nodiscard"},
1467         {Opt_init_itable, "init_itable=%u"},
1468         {Opt_init_itable, "init_itable"},
1469         {Opt_noinit_itable, "noinit_itable"},
1470         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1471         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1472         {Opt_nombcache, "nombcache"},
1473         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1474         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1475         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1476         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1477         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1478         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1479         {Opt_err, NULL},
1480 };
1481
1482 static ext4_fsblk_t get_sb_block(void **data)
1483 {
1484         ext4_fsblk_t    sb_block;
1485         char            *options = (char *) *data;
1486
1487         if (!options || strncmp(options, "sb=", 3) != 0)
1488                 return 1;       /* Default location */
1489
1490         options += 3;
1491         /* TODO: use simple_strtoll with >32bit ext4 */
1492         sb_block = simple_strtoul(options, &options, 0);
1493         if (*options && *options != ',') {
1494                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1495                        (char *) *data);
1496                 return 1;
1497         }
1498         if (*options == ',')
1499                 options++;
1500         *data = (void *) options;
1501
1502         return sb_block;
1503 }
1504
1505 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1506 static const char deprecated_msg[] =
1507         "Mount option \"%s\" will be removed by %s\n"
1508         "Contact [email protected] if you think we should keep it.\n";
1509
1510 #ifdef CONFIG_QUOTA
1511 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1512 {
1513         struct ext4_sb_info *sbi = EXT4_SB(sb);
1514         char *qname;
1515         int ret = -1;
1516
1517         if (sb_any_quota_loaded(sb) &&
1518                 !sbi->s_qf_names[qtype]) {
1519                 ext4_msg(sb, KERN_ERR,
1520                         "Cannot change journaled "
1521                         "quota options when quota turned on");
1522                 return -1;
1523         }
1524         if (ext4_has_feature_quota(sb)) {
1525                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1526                          "ignored when QUOTA feature is enabled");
1527                 return 1;
1528         }
1529         qname = match_strdup(args);
1530         if (!qname) {
1531                 ext4_msg(sb, KERN_ERR,
1532                         "Not enough memory for storing quotafile name");
1533                 return -1;
1534         }
1535         if (sbi->s_qf_names[qtype]) {
1536                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1537                         ret = 1;
1538                 else
1539                         ext4_msg(sb, KERN_ERR,
1540                                  "%s quota file already specified",
1541                                  QTYPE2NAME(qtype));
1542                 goto errout;
1543         }
1544         if (strchr(qname, '/')) {
1545                 ext4_msg(sb, KERN_ERR,
1546                         "quotafile must be on filesystem root");
1547                 goto errout;
1548         }
1549         sbi->s_qf_names[qtype] = qname;
1550         set_opt(sb, QUOTA);
1551         return 1;
1552 errout:
1553         kfree(qname);
1554         return ret;
1555 }
1556
1557 static int clear_qf_name(struct super_block *sb, int qtype)
1558 {
1559
1560         struct ext4_sb_info *sbi = EXT4_SB(sb);
1561
1562         if (sb_any_quota_loaded(sb) &&
1563                 sbi->s_qf_names[qtype]) {
1564                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1565                         " when quota turned on");
1566                 return -1;
1567         }
1568         kfree(sbi->s_qf_names[qtype]);
1569         sbi->s_qf_names[qtype] = NULL;
1570         return 1;
1571 }
1572 #endif
1573
1574 #define MOPT_SET        0x0001
1575 #define MOPT_CLEAR      0x0002
1576 #define MOPT_NOSUPPORT  0x0004
1577 #define MOPT_EXPLICIT   0x0008
1578 #define MOPT_CLEAR_ERR  0x0010
1579 #define MOPT_GTE0       0x0020
1580 #ifdef CONFIG_QUOTA
1581 #define MOPT_Q          0
1582 #define MOPT_QFMT       0x0040
1583 #else
1584 #define MOPT_Q          MOPT_NOSUPPORT
1585 #define MOPT_QFMT       MOPT_NOSUPPORT
1586 #endif
1587 #define MOPT_DATAJ      0x0080
1588 #define MOPT_NO_EXT2    0x0100
1589 #define MOPT_NO_EXT3    0x0200
1590 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1591 #define MOPT_STRING     0x0400
1592
1593 static const struct mount_opts {
1594         int     token;
1595         int     mount_opt;
1596         int     flags;
1597 } ext4_mount_opts[] = {
1598         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1599         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1600         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1601         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1602         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1603         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1604         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1605          MOPT_EXT4_ONLY | MOPT_SET},
1606         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1607          MOPT_EXT4_ONLY | MOPT_CLEAR},
1608         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1609         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1610         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1611          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1612         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1613          MOPT_EXT4_ONLY | MOPT_CLEAR},
1614         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1615         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1616         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1617          MOPT_EXT4_ONLY | MOPT_CLEAR},
1618         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1619          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1620         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1621                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1622          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1623         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1624         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1625         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1626         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1627         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1628          MOPT_NO_EXT2},
1629         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1630          MOPT_NO_EXT2},
1631         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1632         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1633         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1634         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1635         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1636         {Opt_commit, 0, MOPT_GTE0},
1637         {Opt_max_batch_time, 0, MOPT_GTE0},
1638         {Opt_min_batch_time, 0, MOPT_GTE0},
1639         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1640         {Opt_init_itable, 0, MOPT_GTE0},
1641         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1642         {Opt_stripe, 0, MOPT_GTE0},
1643         {Opt_resuid, 0, MOPT_GTE0},
1644         {Opt_resgid, 0, MOPT_GTE0},
1645         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1646         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1647         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1648         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1649         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1650         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1651          MOPT_NO_EXT2 | MOPT_DATAJ},
1652         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1653         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1654 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1655         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1656         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1657 #else
1658         {Opt_acl, 0, MOPT_NOSUPPORT},
1659         {Opt_noacl, 0, MOPT_NOSUPPORT},
1660 #endif
1661         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1662         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1663         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1664         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1665         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1666                                                         MOPT_SET | MOPT_Q},
1667         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1668                                                         MOPT_SET | MOPT_Q},
1669         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1670                                                         MOPT_SET | MOPT_Q},
1671         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1672                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1673                                                         MOPT_CLEAR | MOPT_Q},
1674         {Opt_usrjquota, 0, MOPT_Q},
1675         {Opt_grpjquota, 0, MOPT_Q},
1676         {Opt_offusrjquota, 0, MOPT_Q},
1677         {Opt_offgrpjquota, 0, MOPT_Q},
1678         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1679         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1680         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1681         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1682         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1683         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1684         {Opt_err, 0, 0}
1685 };
1686
1687 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1688                             substring_t *args, unsigned long *journal_devnum,
1689                             unsigned int *journal_ioprio, int is_remount)
1690 {
1691         struct ext4_sb_info *sbi = EXT4_SB(sb);
1692         const struct mount_opts *m;
1693         kuid_t uid;
1694         kgid_t gid;
1695         int arg = 0;
1696
1697 #ifdef CONFIG_QUOTA
1698         if (token == Opt_usrjquota)
1699                 return set_qf_name(sb, USRQUOTA, &args[0]);
1700         else if (token == Opt_grpjquota)
1701                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1702         else if (token == Opt_offusrjquota)
1703                 return clear_qf_name(sb, USRQUOTA);
1704         else if (token == Opt_offgrpjquota)
1705                 return clear_qf_name(sb, GRPQUOTA);
1706 #endif
1707         switch (token) {
1708         case Opt_noacl:
1709         case Opt_nouser_xattr:
1710                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1711                 break;
1712         case Opt_sb:
1713                 return 1;       /* handled by get_sb_block() */
1714         case Opt_removed:
1715                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1716                 return 1;
1717         case Opt_abort:
1718                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1719                 return 1;
1720         case Opt_i_version:
1721                 sb->s_flags |= SB_I_VERSION;
1722                 return 1;
1723         case Opt_lazytime:
1724                 sb->s_flags |= SB_LAZYTIME;
1725                 return 1;
1726         case Opt_nolazytime:
1727                 sb->s_flags &= ~SB_LAZYTIME;
1728                 return 1;
1729         }
1730
1731         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1732                 if (token == m->token)
1733                         break;
1734
1735         if (m->token == Opt_err) {
1736                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1737                          "or missing value", opt);
1738                 return -1;
1739         }
1740
1741         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1742                 ext4_msg(sb, KERN_ERR,
1743                          "Mount option \"%s\" incompatible with ext2", opt);
1744                 return -1;
1745         }
1746         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1747                 ext4_msg(sb, KERN_ERR,
1748                          "Mount option \"%s\" incompatible with ext3", opt);
1749                 return -1;
1750         }
1751
1752         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1753                 return -1;
1754         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1755                 return -1;
1756         if (m->flags & MOPT_EXPLICIT) {
1757                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1758                         set_opt2(sb, EXPLICIT_DELALLOC);
1759                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1760                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1761                 } else
1762                         return -1;
1763         }
1764         if (m->flags & MOPT_CLEAR_ERR)
1765                 clear_opt(sb, ERRORS_MASK);
1766         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1767                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1768                          "options when quota turned on");
1769                 return -1;
1770         }
1771
1772         if (m->flags & MOPT_NOSUPPORT) {
1773                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1774         } else if (token == Opt_commit) {
1775                 if (arg == 0)
1776                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1777                 sbi->s_commit_interval = HZ * arg;
1778         } else if (token == Opt_debug_want_extra_isize) {
1779                 sbi->s_want_extra_isize = arg;
1780         } else if (token == Opt_max_batch_time) {
1781                 sbi->s_max_batch_time = arg;
1782         } else if (token == Opt_min_batch_time) {
1783                 sbi->s_min_batch_time = arg;
1784         } else if (token == Opt_inode_readahead_blks) {
1785                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1786                         ext4_msg(sb, KERN_ERR,
1787                                  "EXT4-fs: inode_readahead_blks must be "
1788                                  "0 or a power of 2 smaller than 2^31");
1789                         return -1;
1790                 }
1791                 sbi->s_inode_readahead_blks = arg;
1792         } else if (token == Opt_init_itable) {
1793                 set_opt(sb, INIT_INODE_TABLE);
1794                 if (!args->from)
1795                         arg = EXT4_DEF_LI_WAIT_MULT;
1796                 sbi->s_li_wait_mult = arg;
1797         } else if (token == Opt_max_dir_size_kb) {
1798                 sbi->s_max_dir_size_kb = arg;
1799         } else if (token == Opt_stripe) {
1800                 sbi->s_stripe = arg;
1801         } else if (token == Opt_resuid) {
1802                 uid = make_kuid(current_user_ns(), arg);
1803                 if (!uid_valid(uid)) {
1804                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1805                         return -1;
1806                 }
1807                 sbi->s_resuid = uid;
1808         } else if (token == Opt_resgid) {
1809                 gid = make_kgid(current_user_ns(), arg);
1810                 if (!gid_valid(gid)) {
1811                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1812                         return -1;
1813                 }
1814                 sbi->s_resgid = gid;
1815         } else if (token == Opt_journal_dev) {
1816                 if (is_remount) {
1817                         ext4_msg(sb, KERN_ERR,
1818                                  "Cannot specify journal on remount");
1819                         return -1;
1820                 }
1821                 *journal_devnum = arg;
1822         } else if (token == Opt_journal_path) {
1823                 char *journal_path;
1824                 struct inode *journal_inode;
1825                 struct path path;
1826                 int error;
1827
1828                 if (is_remount) {
1829                         ext4_msg(sb, KERN_ERR,
1830                                  "Cannot specify journal on remount");
1831                         return -1;
1832                 }
1833                 journal_path = match_strdup(&args[0]);
1834                 if (!journal_path) {
1835                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1836                                 "journal device string");
1837                         return -1;
1838                 }
1839
1840                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1841                 if (error) {
1842                         ext4_msg(sb, KERN_ERR, "error: could not find "
1843                                 "journal device path: error %d", error);
1844                         kfree(journal_path);
1845                         return -1;
1846                 }
1847
1848                 journal_inode = d_inode(path.dentry);
1849                 if (!S_ISBLK(journal_inode->i_mode)) {
1850                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1851                                 "is not a block device", journal_path);
1852                         path_put(&path);
1853                         kfree(journal_path);
1854                         return -1;
1855                 }
1856
1857                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1858                 path_put(&path);
1859                 kfree(journal_path);
1860         } else if (token == Opt_journal_ioprio) {
1861                 if (arg > 7) {
1862                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1863                                  " (must be 0-7)");
1864                         return -1;
1865                 }
1866                 *journal_ioprio =
1867                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1868         } else if (token == Opt_test_dummy_encryption) {
1869 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1870                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1871                 ext4_msg(sb, KERN_WARNING,
1872                          "Test dummy encryption mode enabled");
1873 #else
1874                 ext4_msg(sb, KERN_WARNING,
1875                          "Test dummy encryption mount option ignored");
1876 #endif
1877         } else if (m->flags & MOPT_DATAJ) {
1878                 if (is_remount) {
1879                         if (!sbi->s_journal)
1880                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1881                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1882                                 ext4_msg(sb, KERN_ERR,
1883                                          "Cannot change data mode on remount");
1884                                 return -1;
1885                         }
1886                 } else {
1887                         clear_opt(sb, DATA_FLAGS);
1888                         sbi->s_mount_opt |= m->mount_opt;
1889                 }
1890 #ifdef CONFIG_QUOTA
1891         } else if (m->flags & MOPT_QFMT) {
1892                 if (sb_any_quota_loaded(sb) &&
1893                     sbi->s_jquota_fmt != m->mount_opt) {
1894                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1895                                  "quota options when quota turned on");
1896                         return -1;
1897                 }
1898                 if (ext4_has_feature_quota(sb)) {
1899                         ext4_msg(sb, KERN_INFO,
1900                                  "Quota format mount options ignored "
1901                                  "when QUOTA feature is enabled");
1902                         return 1;
1903                 }
1904                 sbi->s_jquota_fmt = m->mount_opt;
1905 #endif
1906         } else if (token == Opt_dax) {
1907 #ifdef CONFIG_FS_DAX
1908                 ext4_msg(sb, KERN_WARNING,
1909                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1910                         sbi->s_mount_opt |= m->mount_opt;
1911 #else
1912                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1913                 return -1;
1914 #endif
1915         } else if (token == Opt_data_err_abort) {
1916                 sbi->s_mount_opt |= m->mount_opt;
1917         } else if (token == Opt_data_err_ignore) {
1918                 sbi->s_mount_opt &= ~m->mount_opt;
1919         } else {
1920                 if (!args->from)
1921                         arg = 1;
1922                 if (m->flags & MOPT_CLEAR)
1923                         arg = !arg;
1924                 else if (unlikely(!(m->flags & MOPT_SET))) {
1925                         ext4_msg(sb, KERN_WARNING,
1926                                  "buggy handling of option %s", opt);
1927                         WARN_ON(1);
1928                         return -1;
1929                 }
1930                 if (arg != 0)
1931                         sbi->s_mount_opt |= m->mount_opt;
1932                 else
1933                         sbi->s_mount_opt &= ~m->mount_opt;
1934         }
1935         return 1;
1936 }
1937
1938 static int parse_options(char *options, struct super_block *sb,
1939                          unsigned long *journal_devnum,
1940                          unsigned int *journal_ioprio,
1941                          int is_remount)
1942 {
1943         struct ext4_sb_info *sbi = EXT4_SB(sb);
1944         char *p;
1945         substring_t args[MAX_OPT_ARGS];
1946         int token;
1947
1948         if (!options)
1949                 return 1;
1950
1951         while ((p = strsep(&options, ",")) != NULL) {
1952                 if (!*p)
1953                         continue;
1954                 /*
1955                  * Initialize args struct so we know whether arg was
1956                  * found; some options take optional arguments.
1957                  */
1958                 args[0].to = args[0].from = NULL;
1959                 token = match_token(p, tokens, args);
1960                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1961                                      journal_ioprio, is_remount) < 0)
1962                         return 0;
1963         }
1964 #ifdef CONFIG_QUOTA
1965         /*
1966          * We do the test below only for project quotas. 'usrquota' and
1967          * 'grpquota' mount options are allowed even without quota feature
1968          * to support legacy quotas in quota files.
1969          */
1970         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1971                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1972                          "Cannot enable project quota enforcement.");
1973                 return 0;
1974         }
1975         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1976                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1977                         clear_opt(sb, USRQUOTA);
1978
1979                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1980                         clear_opt(sb, GRPQUOTA);
1981
1982                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1983                         ext4_msg(sb, KERN_ERR, "old and new quota "
1984                                         "format mixing");
1985                         return 0;
1986                 }
1987
1988                 if (!sbi->s_jquota_fmt) {
1989                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1990                                         "not specified");
1991                         return 0;
1992                 }
1993         }
1994 #endif
1995         if (test_opt(sb, DIOREAD_NOLOCK)) {
1996                 int blocksize =
1997                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1998
1999                 if (blocksize < PAGE_SIZE) {
2000                         ext4_msg(sb, KERN_ERR, "can't mount with "
2001                                  "dioread_nolock if block size != PAGE_SIZE");
2002                         return 0;
2003                 }
2004         }
2005         return 1;
2006 }
2007
2008 static inline void ext4_show_quota_options(struct seq_file *seq,
2009                                            struct super_block *sb)
2010 {
2011 #if defined(CONFIG_QUOTA)
2012         struct ext4_sb_info *sbi = EXT4_SB(sb);
2013
2014         if (sbi->s_jquota_fmt) {
2015                 char *fmtname = "";
2016
2017                 switch (sbi->s_jquota_fmt) {
2018                 case QFMT_VFS_OLD:
2019                         fmtname = "vfsold";
2020                         break;
2021                 case QFMT_VFS_V0:
2022                         fmtname = "vfsv0";
2023                         break;
2024                 case QFMT_VFS_V1:
2025                         fmtname = "vfsv1";
2026                         break;
2027                 }
2028                 seq_printf(seq, ",jqfmt=%s", fmtname);
2029         }
2030
2031         if (sbi->s_qf_names[USRQUOTA])
2032                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
2033
2034         if (sbi->s_qf_names[GRPQUOTA])
2035                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
2036 #endif
2037 }
2038
2039 static const char *token2str(int token)
2040 {
2041         const struct match_token *t;
2042
2043         for (t = tokens; t->token != Opt_err; t++)
2044                 if (t->token == token && !strchr(t->pattern, '='))
2045                         break;
2046         return t->pattern;
2047 }
2048
2049 /*
2050  * Show an option if
2051  *  - it's set to a non-default value OR
2052  *  - if the per-sb default is different from the global default
2053  */
2054 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2055                               int nodefs)
2056 {
2057         struct ext4_sb_info *sbi = EXT4_SB(sb);
2058         struct ext4_super_block *es = sbi->s_es;
2059         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2060         const struct mount_opts *m;
2061         char sep = nodefs ? '\n' : ',';
2062
2063 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2064 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2065
2066         if (sbi->s_sb_block != 1)
2067                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2068
2069         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2070                 int want_set = m->flags & MOPT_SET;
2071                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2072                     (m->flags & MOPT_CLEAR_ERR))
2073                         continue;
2074                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2075                         continue; /* skip if same as the default */
2076                 if ((want_set &&
2077                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2078                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2079                         continue; /* select Opt_noFoo vs Opt_Foo */
2080                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2081         }
2082
2083         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2084             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2085                 SEQ_OPTS_PRINT("resuid=%u",
2086                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2087         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2088             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2089                 SEQ_OPTS_PRINT("resgid=%u",
2090                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2091         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2092         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2093                 SEQ_OPTS_PUTS("errors=remount-ro");
2094         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2095                 SEQ_OPTS_PUTS("errors=continue");
2096         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2097                 SEQ_OPTS_PUTS("errors=panic");
2098         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2099                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2100         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2101                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2102         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2103                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2104         if (sb->s_flags & SB_I_VERSION)
2105                 SEQ_OPTS_PUTS("i_version");
2106         if (nodefs || sbi->s_stripe)
2107                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2108         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2109                         (sbi->s_mount_opt ^ def_mount_opt)) {
2110                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2111                         SEQ_OPTS_PUTS("data=journal");
2112                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2113                         SEQ_OPTS_PUTS("data=ordered");
2114                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2115                         SEQ_OPTS_PUTS("data=writeback");
2116         }
2117         if (nodefs ||
2118             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2119                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2120                                sbi->s_inode_readahead_blks);
2121
2122         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2123                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2124                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2125         if (nodefs || sbi->s_max_dir_size_kb)
2126                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2127         if (test_opt(sb, DATA_ERR_ABORT))
2128                 SEQ_OPTS_PUTS("data_err=abort");
2129
2130         ext4_show_quota_options(seq, sb);
2131         return 0;
2132 }
2133
2134 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2135 {
2136         return _ext4_show_options(seq, root->d_sb, 0);
2137 }
2138
2139 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2140 {
2141         struct super_block *sb = seq->private;
2142         int rc;
2143
2144         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2145         rc = _ext4_show_options(seq, sb, 1);
2146         seq_puts(seq, "\n");
2147         return rc;
2148 }
2149
2150 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2151                             int read_only)
2152 {
2153         struct ext4_sb_info *sbi = EXT4_SB(sb);
2154         int err = 0;
2155
2156         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2157                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2158                          "forcing read-only mode");
2159                 err = -EROFS;
2160         }
2161         if (read_only)
2162                 goto done;
2163         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2164                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2165                          "running e2fsck is recommended");
2166         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2167                 ext4_msg(sb, KERN_WARNING,
2168                          "warning: mounting fs with errors, "
2169                          "running e2fsck is recommended");
2170         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2171                  le16_to_cpu(es->s_mnt_count) >=
2172                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2173                 ext4_msg(sb, KERN_WARNING,
2174                          "warning: maximal mount count reached, "
2175                          "running e2fsck is recommended");
2176         else if (le32_to_cpu(es->s_checkinterval) &&
2177                 (le32_to_cpu(es->s_lastcheck) +
2178                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2179                 ext4_msg(sb, KERN_WARNING,
2180                          "warning: checktime reached, "
2181                          "running e2fsck is recommended");
2182         if (!sbi->s_journal)
2183                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2184         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2185                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2186         le16_add_cpu(&es->s_mnt_count, 1);
2187         es->s_mtime = cpu_to_le32(get_seconds());
2188         ext4_update_dynamic_rev(sb);
2189         if (sbi->s_journal)
2190                 ext4_set_feature_journal_needs_recovery(sb);
2191
2192         err = ext4_commit_super(sb, 1);
2193 done:
2194         if (test_opt(sb, DEBUG))
2195                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2196                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2197                         sb->s_blocksize,
2198                         sbi->s_groups_count,
2199                         EXT4_BLOCKS_PER_GROUP(sb),
2200                         EXT4_INODES_PER_GROUP(sb),
2201                         sbi->s_mount_opt, sbi->s_mount_opt2);
2202
2203         cleancache_init_fs(sb);
2204         return err;
2205 }
2206
2207 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2208 {
2209         struct ext4_sb_info *sbi = EXT4_SB(sb);
2210         struct flex_groups *new_groups;
2211         int size;
2212
2213         if (!sbi->s_log_groups_per_flex)
2214                 return 0;
2215
2216         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2217         if (size <= sbi->s_flex_groups_allocated)
2218                 return 0;
2219
2220         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2221         new_groups = kvzalloc(size, GFP_KERNEL);
2222         if (!new_groups) {
2223                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2224                          size / (int) sizeof(struct flex_groups));
2225                 return -ENOMEM;
2226         }
2227
2228         if (sbi->s_flex_groups) {
2229                 memcpy(new_groups, sbi->s_flex_groups,
2230                        (sbi->s_flex_groups_allocated *
2231                         sizeof(struct flex_groups)));
2232                 kvfree(sbi->s_flex_groups);
2233         }
2234         sbi->s_flex_groups = new_groups;
2235         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2236         return 0;
2237 }
2238
2239 static int ext4_fill_flex_info(struct super_block *sb)
2240 {
2241         struct ext4_sb_info *sbi = EXT4_SB(sb);
2242         struct ext4_group_desc *gdp = NULL;
2243         ext4_group_t flex_group;
2244         int i, err;
2245
2246         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2247         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2248                 sbi->s_log_groups_per_flex = 0;
2249                 return 1;
2250         }
2251
2252         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2253         if (err)
2254                 goto failed;
2255
2256         for (i = 0; i < sbi->s_groups_count; i++) {
2257                 gdp = ext4_get_group_desc(sb, i, NULL);
2258
2259                 flex_group = ext4_flex_group(sbi, i);
2260                 atomic_add(ext4_free_inodes_count(sb, gdp),
2261                            &sbi->s_flex_groups[flex_group].free_inodes);
2262                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2263                              &sbi->s_flex_groups[flex_group].free_clusters);
2264                 atomic_add(ext4_used_dirs_count(sb, gdp),
2265                            &sbi->s_flex_groups[flex_group].used_dirs);
2266         }
2267
2268         return 1;
2269 failed:
2270         return 0;
2271 }
2272
2273 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2274                                    struct ext4_group_desc *gdp)
2275 {
2276         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2277         __u16 crc = 0;
2278         __le32 le_group = cpu_to_le32(block_group);
2279         struct ext4_sb_info *sbi = EXT4_SB(sb);
2280
2281         if (ext4_has_metadata_csum(sbi->s_sb)) {
2282                 /* Use new metadata_csum algorithm */
2283                 __u32 csum32;
2284                 __u16 dummy_csum = 0;
2285
2286                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2287                                      sizeof(le_group));
2288                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2289                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2290                                      sizeof(dummy_csum));
2291                 offset += sizeof(dummy_csum);
2292                 if (offset < sbi->s_desc_size)
2293                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2294                                              sbi->s_desc_size - offset);
2295
2296                 crc = csum32 & 0xFFFF;
2297                 goto out;
2298         }
2299
2300         /* old crc16 code */
2301         if (!ext4_has_feature_gdt_csum(sb))
2302                 return 0;
2303
2304         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2305         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2306         crc = crc16(crc, (__u8 *)gdp, offset);
2307         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2308         /* for checksum of struct ext4_group_desc do the rest...*/
2309         if (ext4_has_feature_64bit(sb) &&
2310             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2311                 crc = crc16(crc, (__u8 *)gdp + offset,
2312                             le16_to_cpu(sbi->s_es->s_desc_size) -
2313                                 offset);
2314
2315 out:
2316         return cpu_to_le16(crc);
2317 }
2318
2319 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2320                                 struct ext4_group_desc *gdp)
2321 {
2322         if (ext4_has_group_desc_csum(sb) &&
2323             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2324                 return 0;
2325
2326         return 1;
2327 }
2328
2329 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2330                               struct ext4_group_desc *gdp)
2331 {
2332         if (!ext4_has_group_desc_csum(sb))
2333                 return;
2334         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2335 }
2336
2337 /* Called at mount-time, super-block is locked */
2338 static int ext4_check_descriptors(struct super_block *sb,
2339                                   ext4_fsblk_t sb_block,
2340                                   ext4_group_t *first_not_zeroed)
2341 {
2342         struct ext4_sb_info *sbi = EXT4_SB(sb);
2343         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2344         ext4_fsblk_t last_block;
2345         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0) + 1;
2346         ext4_fsblk_t block_bitmap;
2347         ext4_fsblk_t inode_bitmap;
2348         ext4_fsblk_t inode_table;
2349         int flexbg_flag = 0;
2350         ext4_group_t i, grp = sbi->s_groups_count;
2351
2352         if (ext4_has_feature_flex_bg(sb))
2353                 flexbg_flag = 1;
2354
2355         ext4_debug("Checking group descriptors");
2356
2357         for (i = 0; i < sbi->s_groups_count; i++) {
2358                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2359
2360                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2361                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2362                 else
2363                         last_block = first_block +
2364                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2365
2366                 if ((grp == sbi->s_groups_count) &&
2367                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2368                         grp = i;
2369
2370                 block_bitmap = ext4_block_bitmap(sb, gdp);
2371                 if (block_bitmap == sb_block) {
2372                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2373                                  "Block bitmap for group %u overlaps "
2374                                  "superblock", i);
2375                         if (!sb_rdonly(sb))
2376                                 return 0;
2377                 }
2378                 if (block_bitmap >= sb_block + 1 &&
2379                     block_bitmap <= last_bg_block) {
2380                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2381                                  "Block bitmap for group %u overlaps "
2382                                  "block group descriptors", i);
2383                         if (!sb_rdonly(sb))
2384                                 return 0;
2385                 }
2386                 if (block_bitmap < first_block || block_bitmap > last_block) {
2387                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2388                                "Block bitmap for group %u not in group "
2389                                "(block %llu)!", i, block_bitmap);
2390                         return 0;
2391                 }
2392                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2393                 if (inode_bitmap == sb_block) {
2394                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2395                                  "Inode bitmap for group %u overlaps "
2396                                  "superblock", i);
2397                         if (!sb_rdonly(sb))
2398                                 return 0;
2399                 }
2400                 if (inode_bitmap >= sb_block + 1 &&
2401                     inode_bitmap <= last_bg_block) {
2402                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2403                                  "Inode bitmap for group %u overlaps "
2404                                  "block group descriptors", i);
2405                         if (!sb_rdonly(sb))
2406                                 return 0;
2407                 }
2408                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2409                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2410                                "Inode bitmap for group %u not in group "
2411                                "(block %llu)!", i, inode_bitmap);
2412                         return 0;
2413                 }
2414                 inode_table = ext4_inode_table(sb, gdp);
2415                 if (inode_table == sb_block) {
2416                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2417                                  "Inode table for group %u overlaps "
2418                                  "superblock", i);
2419                         if (!sb_rdonly(sb))
2420                                 return 0;
2421                 }
2422                 if (inode_table >= sb_block + 1 &&
2423                     inode_table <= last_bg_block) {
2424                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2425                                  "Inode table for group %u overlaps "
2426                                  "block group descriptors", i);
2427                         if (!sb_rdonly(sb))
2428                                 return 0;
2429                 }
2430                 if (inode_table < first_block ||
2431                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2432                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2433                                "Inode table for group %u not in group "
2434                                "(block %llu)!", i, inode_table);
2435                         return 0;
2436                 }
2437                 ext4_lock_group(sb, i);
2438                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2439                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2440                                  "Checksum for group %u failed (%u!=%u)",
2441                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2442                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2443                         if (!sb_rdonly(sb)) {
2444                                 ext4_unlock_group(sb, i);
2445                                 return 0;
2446                         }
2447                 }
2448                 ext4_unlock_group(sb, i);
2449                 if (!flexbg_flag)
2450                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2451         }
2452         if (NULL != first_not_zeroed)
2453                 *first_not_zeroed = grp;
2454         return 1;
2455 }
2456
2457 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2458  * the superblock) which were deleted from all directories, but held open by
2459  * a process at the time of a crash.  We walk the list and try to delete these
2460  * inodes at recovery time (only with a read-write filesystem).
2461  *
2462  * In order to keep the orphan inode chain consistent during traversal (in
2463  * case of crash during recovery), we link each inode into the superblock
2464  * orphan list_head and handle it the same way as an inode deletion during
2465  * normal operation (which journals the operations for us).
2466  *
2467  * We only do an iget() and an iput() on each inode, which is very safe if we
2468  * accidentally point at an in-use or already deleted inode.  The worst that
2469  * can happen in this case is that we get a "bit already cleared" message from
2470  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2471  * e2fsck was run on this filesystem, and it must have already done the orphan
2472  * inode cleanup for us, so we can safely abort without any further action.
2473  */
2474 static void ext4_orphan_cleanup(struct super_block *sb,
2475                                 struct ext4_super_block *es)
2476 {
2477         unsigned int s_flags = sb->s_flags;
2478         int ret, nr_orphans = 0, nr_truncates = 0;
2479 #ifdef CONFIG_QUOTA
2480         int quota_update = 0;
2481         int i;
2482 #endif
2483         if (!es->s_last_orphan) {
2484                 jbd_debug(4, "no orphan inodes to clean up\n");
2485                 return;
2486         }
2487
2488         if (bdev_read_only(sb->s_bdev)) {
2489                 ext4_msg(sb, KERN_ERR, "write access "
2490                         "unavailable, skipping orphan cleanup");
2491                 return;
2492         }
2493
2494         /* Check if feature set would not allow a r/w mount */
2495         if (!ext4_feature_set_ok(sb, 0)) {
2496                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2497                          "unknown ROCOMPAT features");
2498                 return;
2499         }
2500
2501         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2502                 /* don't clear list on RO mount w/ errors */
2503                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2504                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2505                                   "clearing orphan list.\n");
2506                         es->s_last_orphan = 0;
2507                 }
2508                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2509                 return;
2510         }
2511
2512         if (s_flags & SB_RDONLY) {
2513                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2514                 sb->s_flags &= ~SB_RDONLY;
2515         }
2516 #ifdef CONFIG_QUOTA
2517         /* Needed for iput() to work correctly and not trash data */
2518         sb->s_flags |= SB_ACTIVE;
2519
2520         /*
2521          * Turn on quotas which were not enabled for read-only mounts if
2522          * filesystem has quota feature, so that they are updated correctly.
2523          */
2524         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2525                 int ret = ext4_enable_quotas(sb);
2526
2527                 if (!ret)
2528                         quota_update = 1;
2529                 else
2530                         ext4_msg(sb, KERN_ERR,
2531                                 "Cannot turn on quotas: error %d", ret);
2532         }
2533
2534         /* Turn on journaled quotas used for old sytle */
2535         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2536                 if (EXT4_SB(sb)->s_qf_names[i]) {
2537                         int ret = ext4_quota_on_mount(sb, i);
2538
2539                         if (!ret)
2540                                 quota_update = 1;
2541                         else
2542                                 ext4_msg(sb, KERN_ERR,
2543                                         "Cannot turn on journaled "
2544                                         "quota: type %d: error %d", i, ret);
2545                 }
2546         }
2547 #endif
2548
2549         while (es->s_last_orphan) {
2550                 struct inode *inode;
2551
2552                 /*
2553                  * We may have encountered an error during cleanup; if
2554                  * so, skip the rest.
2555                  */
2556                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2557                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2558                         es->s_last_orphan = 0;
2559                         break;
2560                 }
2561
2562                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2563                 if (IS_ERR(inode)) {
2564                         es->s_last_orphan = 0;
2565                         break;
2566                 }
2567
2568                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2569                 dquot_initialize(inode);
2570                 if (inode->i_nlink) {
2571                         if (test_opt(sb, DEBUG))
2572                                 ext4_msg(sb, KERN_DEBUG,
2573                                         "%s: truncating inode %lu to %lld bytes",
2574                                         __func__, inode->i_ino, inode->i_size);
2575                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2576                                   inode->i_ino, inode->i_size);
2577                         inode_lock(inode);
2578                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2579                         ret = ext4_truncate(inode);
2580                         if (ret)
2581                                 ext4_std_error(inode->i_sb, ret);
2582                         inode_unlock(inode);
2583                         nr_truncates++;
2584                 } else {
2585                         if (test_opt(sb, DEBUG))
2586                                 ext4_msg(sb, KERN_DEBUG,
2587                                         "%s: deleting unreferenced inode %lu",
2588                                         __func__, inode->i_ino);
2589                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2590                                   inode->i_ino);
2591                         nr_orphans++;
2592                 }
2593                 iput(inode);  /* The delete magic happens here! */
2594         }
2595
2596 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2597
2598         if (nr_orphans)
2599                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2600                        PLURAL(nr_orphans));
2601         if (nr_truncates)
2602                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2603                        PLURAL(nr_truncates));
2604 #ifdef CONFIG_QUOTA
2605         /* Turn off quotas if they were enabled for orphan cleanup */
2606         if (quota_update) {
2607                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2608                         if (sb_dqopt(sb)->files[i])
2609                                 dquot_quota_off(sb, i);
2610                 }
2611         }
2612 #endif
2613         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2614 }
2615
2616 /*
2617  * Maximal extent format file size.
2618  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2619  * extent format containers, within a sector_t, and within i_blocks
2620  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2621  * so that won't be a limiting factor.
2622  *
2623  * However there is other limiting factor. We do store extents in the form
2624  * of starting block and length, hence the resulting length of the extent
2625  * covering maximum file size must fit into on-disk format containers as
2626  * well. Given that length is always by 1 unit bigger than max unit (because
2627  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2628  *
2629  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2630  */
2631 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2632 {
2633         loff_t res;
2634         loff_t upper_limit = MAX_LFS_FILESIZE;
2635
2636         /* small i_blocks in vfs inode? */
2637         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2638                 /*
2639                  * CONFIG_LBDAF is not enabled implies the inode
2640                  * i_block represent total blocks in 512 bytes
2641                  * 32 == size of vfs inode i_blocks * 8
2642                  */
2643                 upper_limit = (1LL << 32) - 1;
2644
2645                 /* total blocks in file system block size */
2646                 upper_limit >>= (blkbits - 9);
2647                 upper_limit <<= blkbits;
2648         }
2649
2650         /*
2651          * 32-bit extent-start container, ee_block. We lower the maxbytes
2652          * by one fs block, so ee_len can cover the extent of maximum file
2653          * size
2654          */
2655         res = (1LL << 32) - 1;
2656         res <<= blkbits;
2657
2658         /* Sanity check against vm- & vfs- imposed limits */
2659         if (res > upper_limit)
2660                 res = upper_limit;
2661
2662         return res;
2663 }
2664
2665 /*
2666  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2667  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2668  * We need to be 1 filesystem block less than the 2^48 sector limit.
2669  */
2670 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2671 {
2672         loff_t res = EXT4_NDIR_BLOCKS;
2673         int meta_blocks;
2674         loff_t upper_limit;
2675         /* This is calculated to be the largest file size for a dense, block
2676          * mapped file such that the file's total number of 512-byte sectors,
2677          * including data and all indirect blocks, does not exceed (2^48 - 1).
2678          *
2679          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2680          * number of 512-byte sectors of the file.
2681          */
2682
2683         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2684                 /*
2685                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2686                  * the inode i_block field represents total file blocks in
2687                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2688                  */
2689                 upper_limit = (1LL << 32) - 1;
2690
2691                 /* total blocks in file system block size */
2692                 upper_limit >>= (bits - 9);
2693
2694         } else {
2695                 /*
2696                  * We use 48 bit ext4_inode i_blocks
2697                  * With EXT4_HUGE_FILE_FL set the i_blocks
2698                  * represent total number of blocks in
2699                  * file system block size
2700                  */
2701                 upper_limit = (1LL << 48) - 1;
2702
2703         }
2704
2705         /* indirect blocks */
2706         meta_blocks = 1;
2707         /* double indirect blocks */
2708         meta_blocks += 1 + (1LL << (bits-2));
2709         /* tripple indirect blocks */
2710         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2711
2712         upper_limit -= meta_blocks;
2713         upper_limit <<= bits;
2714
2715         res += 1LL << (bits-2);
2716         res += 1LL << (2*(bits-2));
2717         res += 1LL << (3*(bits-2));
2718         res <<= bits;
2719         if (res > upper_limit)
2720                 res = upper_limit;
2721
2722         if (res > MAX_LFS_FILESIZE)
2723                 res = MAX_LFS_FILESIZE;
2724
2725         return res;
2726 }
2727
2728 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2729                                    ext4_fsblk_t logical_sb_block, int nr)
2730 {
2731         struct ext4_sb_info *sbi = EXT4_SB(sb);
2732         ext4_group_t bg, first_meta_bg;
2733         int has_super = 0;
2734
2735         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2736
2737         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2738                 return logical_sb_block + nr + 1;
2739         bg = sbi->s_desc_per_block * nr;
2740         if (ext4_bg_has_super(sb, bg))
2741                 has_super = 1;
2742
2743         /*
2744          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2745          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2746          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2747          * compensate.
2748          */
2749         if (sb->s_blocksize == 1024 && nr == 0 &&
2750             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2751                 has_super++;
2752
2753         return (has_super + ext4_group_first_block_no(sb, bg));
2754 }
2755
2756 /**
2757  * ext4_get_stripe_size: Get the stripe size.
2758  * @sbi: In memory super block info
2759  *
2760  * If we have specified it via mount option, then
2761  * use the mount option value. If the value specified at mount time is
2762  * greater than the blocks per group use the super block value.
2763  * If the super block value is greater than blocks per group return 0.
2764  * Allocator needs it be less than blocks per group.
2765  *
2766  */
2767 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2768 {
2769         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2770         unsigned long stripe_width =
2771                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2772         int ret;
2773
2774         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2775                 ret = sbi->s_stripe;
2776         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2777                 ret = stripe_width;
2778         else if (stride && stride <= sbi->s_blocks_per_group)
2779                 ret = stride;
2780         else
2781                 ret = 0;
2782
2783         /*
2784          * If the stripe width is 1, this makes no sense and
2785          * we set it to 0 to turn off stripe handling code.
2786          */
2787         if (ret <= 1)
2788                 ret = 0;
2789
2790         return ret;
2791 }
2792
2793 /*
2794  * Check whether this filesystem can be mounted based on
2795  * the features present and the RDONLY/RDWR mount requested.
2796  * Returns 1 if this filesystem can be mounted as requested,
2797  * 0 if it cannot be.
2798  */
2799 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2800 {
2801         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2802                 ext4_msg(sb, KERN_ERR,
2803                         "Couldn't mount because of "
2804                         "unsupported optional features (%x)",
2805                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2806                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2807                 return 0;
2808         }
2809
2810         if (readonly)
2811                 return 1;
2812
2813         if (ext4_has_feature_readonly(sb)) {
2814                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2815                 sb->s_flags |= SB_RDONLY;
2816                 return 1;
2817         }
2818
2819         /* Check that feature set is OK for a read-write mount */
2820         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2821                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2822                          "unsupported optional features (%x)",
2823                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2824                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2825                 return 0;
2826         }
2827         /*
2828          * Large file size enabled file system can only be mounted
2829          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2830          */
2831         if (ext4_has_feature_huge_file(sb)) {
2832                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2833                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2834                                  "cannot be mounted RDWR without "
2835                                  "CONFIG_LBDAF");
2836                         return 0;
2837                 }
2838         }
2839         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2840                 ext4_msg(sb, KERN_ERR,
2841                          "Can't support bigalloc feature without "
2842                          "extents feature\n");
2843                 return 0;
2844         }
2845
2846 #ifndef CONFIG_QUOTA
2847         if (ext4_has_feature_quota(sb) && !readonly) {
2848                 ext4_msg(sb, KERN_ERR,
2849                          "Filesystem with quota feature cannot be mounted RDWR "
2850                          "without CONFIG_QUOTA");
2851                 return 0;
2852         }
2853         if (ext4_has_feature_project(sb) && !readonly) {
2854                 ext4_msg(sb, KERN_ERR,
2855                          "Filesystem with project quota feature cannot be mounted RDWR "
2856                          "without CONFIG_QUOTA");
2857                 return 0;
2858         }
2859 #endif  /* CONFIG_QUOTA */
2860         return 1;
2861 }
2862
2863 /*
2864  * This function is called once a day if we have errors logged
2865  * on the file system
2866  */
2867 static void print_daily_error_info(struct timer_list *t)
2868 {
2869         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2870         struct super_block *sb = sbi->s_sb;
2871         struct ext4_super_block *es = sbi->s_es;
2872
2873         if (es->s_error_count)
2874                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2875                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2876                          le32_to_cpu(es->s_error_count));
2877         if (es->s_first_error_time) {
2878                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2879                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2880                        (int) sizeof(es->s_first_error_func),
2881                        es->s_first_error_func,
2882                        le32_to_cpu(es->s_first_error_line));
2883                 if (es->s_first_error_ino)
2884                         printk(KERN_CONT ": inode %u",
2885                                le32_to_cpu(es->s_first_error_ino));
2886                 if (es->s_first_error_block)
2887                         printk(KERN_CONT ": block %llu", (unsigned long long)
2888                                le64_to_cpu(es->s_first_error_block));
2889                 printk(KERN_CONT "\n");
2890         }
2891         if (es->s_last_error_time) {
2892                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2893                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2894                        (int) sizeof(es->s_last_error_func),
2895                        es->s_last_error_func,
2896                        le32_to_cpu(es->s_last_error_line));
2897                 if (es->s_last_error_ino)
2898                         printk(KERN_CONT ": inode %u",
2899                                le32_to_cpu(es->s_last_error_ino));
2900                 if (es->s_last_error_block)
2901                         printk(KERN_CONT ": block %llu", (unsigned long long)
2902                                le64_to_cpu(es->s_last_error_block));
2903                 printk(KERN_CONT "\n");
2904         }
2905         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2906 }
2907
2908 /* Find next suitable group and run ext4_init_inode_table */
2909 static int ext4_run_li_request(struct ext4_li_request *elr)
2910 {
2911         struct ext4_group_desc *gdp = NULL;
2912         ext4_group_t group, ngroups;
2913         struct super_block *sb;
2914         unsigned long timeout = 0;
2915         int ret = 0;
2916
2917         sb = elr->lr_super;
2918         ngroups = EXT4_SB(sb)->s_groups_count;
2919
2920         for (group = elr->lr_next_group; group < ngroups; group++) {
2921                 gdp = ext4_get_group_desc(sb, group, NULL);
2922                 if (!gdp) {
2923                         ret = 1;
2924                         break;
2925                 }
2926
2927                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2928                         break;
2929         }
2930
2931         if (group >= ngroups)
2932                 ret = 1;
2933
2934         if (!ret) {
2935                 timeout = jiffies;
2936                 ret = ext4_init_inode_table(sb, group,
2937                                             elr->lr_timeout ? 0 : 1);
2938                 if (elr->lr_timeout == 0) {
2939                         timeout = (jiffies - timeout) *
2940                                   elr->lr_sbi->s_li_wait_mult;
2941                         elr->lr_timeout = timeout;
2942                 }
2943                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2944                 elr->lr_next_group = group + 1;
2945         }
2946         return ret;
2947 }
2948
2949 /*
2950  * Remove lr_request from the list_request and free the
2951  * request structure. Should be called with li_list_mtx held
2952  */
2953 static void ext4_remove_li_request(struct ext4_li_request *elr)
2954 {
2955         struct ext4_sb_info *sbi;
2956
2957         if (!elr)
2958                 return;
2959
2960         sbi = elr->lr_sbi;
2961
2962         list_del(&elr->lr_request);
2963         sbi->s_li_request = NULL;
2964         kfree(elr);
2965 }
2966
2967 static void ext4_unregister_li_request(struct super_block *sb)
2968 {
2969         mutex_lock(&ext4_li_mtx);
2970         if (!ext4_li_info) {
2971                 mutex_unlock(&ext4_li_mtx);
2972                 return;
2973         }
2974
2975         mutex_lock(&ext4_li_info->li_list_mtx);
2976         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2977         mutex_unlock(&ext4_li_info->li_list_mtx);
2978         mutex_unlock(&ext4_li_mtx);
2979 }
2980
2981 static struct task_struct *ext4_lazyinit_task;
2982
2983 /*
2984  * This is the function where ext4lazyinit thread lives. It walks
2985  * through the request list searching for next scheduled filesystem.
2986  * When such a fs is found, run the lazy initialization request
2987  * (ext4_rn_li_request) and keep track of the time spend in this
2988  * function. Based on that time we compute next schedule time of
2989  * the request. When walking through the list is complete, compute
2990  * next waking time and put itself into sleep.
2991  */
2992 static int ext4_lazyinit_thread(void *arg)
2993 {
2994         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2995         struct list_head *pos, *n;
2996         struct ext4_li_request *elr;
2997         unsigned long next_wakeup, cur;
2998
2999         BUG_ON(NULL == eli);
3000
3001 cont_thread:
3002         while (true) {
3003                 next_wakeup = MAX_JIFFY_OFFSET;
3004
3005                 mutex_lock(&eli->li_list_mtx);
3006                 if (list_empty(&eli->li_request_list)) {
3007                         mutex_unlock(&eli->li_list_mtx);
3008                         goto exit_thread;
3009                 }
3010                 list_for_each_safe(pos, n, &eli->li_request_list) {
3011                         int err = 0;
3012                         int progress = 0;
3013                         elr = list_entry(pos, struct ext4_li_request,
3014                                          lr_request);
3015
3016                         if (time_before(jiffies, elr->lr_next_sched)) {
3017                                 if (time_before(elr->lr_next_sched, next_wakeup))
3018                                         next_wakeup = elr->lr_next_sched;
3019                                 continue;
3020                         }
3021                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3022                                 if (sb_start_write_trylock(elr->lr_super)) {
3023                                         progress = 1;
3024                                         /*
3025                                          * We hold sb->s_umount, sb can not
3026                                          * be removed from the list, it is
3027                                          * now safe to drop li_list_mtx
3028                                          */
3029                                         mutex_unlock(&eli->li_list_mtx);
3030                                         err = ext4_run_li_request(elr);
3031                                         sb_end_write(elr->lr_super);
3032                                         mutex_lock(&eli->li_list_mtx);
3033                                         n = pos->next;
3034                                 }
3035                                 up_read((&elr->lr_super->s_umount));
3036                         }
3037                         /* error, remove the lazy_init job */
3038                         if (err) {
3039                                 ext4_remove_li_request(elr);
3040                                 continue;
3041                         }
3042                         if (!progress) {
3043                                 elr->lr_next_sched = jiffies +
3044                                         (prandom_u32()
3045                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3046                         }
3047                         if (time_before(elr->lr_next_sched, next_wakeup))
3048                                 next_wakeup = elr->lr_next_sched;
3049                 }
3050                 mutex_unlock(&eli->li_list_mtx);
3051
3052                 try_to_freeze();
3053
3054                 cur = jiffies;
3055                 if ((time_after_eq(cur, next_wakeup)) ||
3056                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3057                         cond_resched();
3058                         continue;
3059                 }
3060
3061                 schedule_timeout_interruptible(next_wakeup - cur);
3062
3063                 if (kthread_should_stop()) {
3064                         ext4_clear_request_list();
3065                         goto exit_thread;
3066                 }
3067         }
3068
3069 exit_thread:
3070         /*
3071          * It looks like the request list is empty, but we need
3072          * to check it under the li_list_mtx lock, to prevent any
3073          * additions into it, and of course we should lock ext4_li_mtx
3074          * to atomically free the list and ext4_li_info, because at
3075          * this point another ext4 filesystem could be registering
3076          * new one.
3077          */
3078         mutex_lock(&ext4_li_mtx);
3079         mutex_lock(&eli->li_list_mtx);
3080         if (!list_empty(&eli->li_request_list)) {
3081                 mutex_unlock(&eli->li_list_mtx);
3082                 mutex_unlock(&ext4_li_mtx);
3083                 goto cont_thread;
3084         }
3085         mutex_unlock(&eli->li_list_mtx);
3086         kfree(ext4_li_info);
3087         ext4_li_info = NULL;
3088         mutex_unlock(&ext4_li_mtx);
3089
3090         return 0;
3091 }
3092
3093 static void ext4_clear_request_list(void)
3094 {
3095         struct list_head *pos, *n;
3096         struct ext4_li_request *elr;
3097
3098         mutex_lock(&ext4_li_info->li_list_mtx);
3099         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3100                 elr = list_entry(pos, struct ext4_li_request,
3101                                  lr_request);
3102                 ext4_remove_li_request(elr);
3103         }
3104         mutex_unlock(&ext4_li_info->li_list_mtx);
3105 }
3106
3107 static int ext4_run_lazyinit_thread(void)
3108 {
3109         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3110                                          ext4_li_info, "ext4lazyinit");
3111         if (IS_ERR(ext4_lazyinit_task)) {
3112                 int err = PTR_ERR(ext4_lazyinit_task);
3113                 ext4_clear_request_list();
3114                 kfree(ext4_li_info);
3115                 ext4_li_info = NULL;
3116                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3117                                  "initialization thread\n",
3118                                  err);
3119                 return err;
3120         }
3121         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3122         return 0;
3123 }
3124
3125 /*
3126  * Check whether it make sense to run itable init. thread or not.
3127  * If there is at least one uninitialized inode table, return
3128  * corresponding group number, else the loop goes through all
3129  * groups and return total number of groups.
3130  */
3131 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3132 {
3133         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3134         struct ext4_group_desc *gdp = NULL;
3135
3136         if (!ext4_has_group_desc_csum(sb))
3137                 return ngroups;
3138
3139         for (group = 0; group < ngroups; group++) {
3140                 gdp = ext4_get_group_desc(sb, group, NULL);
3141                 if (!gdp)
3142                         continue;
3143
3144                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
3145                         continue;
3146                 if (group != 0)
3147                         break;
3148                 ext4_error(sb, "Inode table for bg 0 marked as "
3149                            "needing zeroing");
3150                 if (sb_rdonly(sb))
3151                         return ngroups;
3152         }
3153
3154         return group;
3155 }
3156
3157 static int ext4_li_info_new(void)
3158 {
3159         struct ext4_lazy_init *eli = NULL;
3160
3161         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3162         if (!eli)
3163                 return -ENOMEM;
3164
3165         INIT_LIST_HEAD(&eli->li_request_list);
3166         mutex_init(&eli->li_list_mtx);
3167
3168         eli->li_state |= EXT4_LAZYINIT_QUIT;
3169
3170         ext4_li_info = eli;
3171
3172         return 0;
3173 }
3174
3175 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3176                                             ext4_group_t start)
3177 {
3178         struct ext4_sb_info *sbi = EXT4_SB(sb);
3179         struct ext4_li_request *elr;
3180
3181         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3182         if (!elr)
3183                 return NULL;
3184
3185         elr->lr_super = sb;
3186         elr->lr_sbi = sbi;
3187         elr->lr_next_group = start;
3188
3189         /*
3190          * Randomize first schedule time of the request to
3191          * spread the inode table initialization requests
3192          * better.
3193          */
3194         elr->lr_next_sched = jiffies + (prandom_u32() %
3195                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3196         return elr;
3197 }
3198
3199 int ext4_register_li_request(struct super_block *sb,
3200                              ext4_group_t first_not_zeroed)
3201 {
3202         struct ext4_sb_info *sbi = EXT4_SB(sb);
3203         struct ext4_li_request *elr = NULL;
3204         ext4_group_t ngroups = sbi->s_groups_count;
3205         int ret = 0;
3206
3207         mutex_lock(&ext4_li_mtx);
3208         if (sbi->s_li_request != NULL) {
3209                 /*
3210                  * Reset timeout so it can be computed again, because
3211                  * s_li_wait_mult might have changed.
3212                  */
3213                 sbi->s_li_request->lr_timeout = 0;
3214                 goto out;
3215         }
3216
3217         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3218             !test_opt(sb, INIT_INODE_TABLE))
3219                 goto out;
3220
3221         elr = ext4_li_request_new(sb, first_not_zeroed);
3222         if (!elr) {
3223                 ret = -ENOMEM;
3224                 goto out;
3225         }
3226
3227         if (NULL == ext4_li_info) {
3228                 ret = ext4_li_info_new();
3229                 if (ret)
3230                         goto out;
3231         }
3232
3233         mutex_lock(&ext4_li_info->li_list_mtx);
3234         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3235         mutex_unlock(&ext4_li_info->li_list_mtx);
3236
3237         sbi->s_li_request = elr;
3238         /*
3239          * set elr to NULL here since it has been inserted to
3240          * the request_list and the removal and free of it is
3241          * handled by ext4_clear_request_list from now on.
3242          */
3243         elr = NULL;
3244
3245         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3246                 ret = ext4_run_lazyinit_thread();
3247                 if (ret)
3248                         goto out;
3249         }
3250 out:
3251         mutex_unlock(&ext4_li_mtx);
3252         if (ret)
3253                 kfree(elr);
3254         return ret;
3255 }
3256
3257 /*
3258  * We do not need to lock anything since this is called on
3259  * module unload.
3260  */
3261 static void ext4_destroy_lazyinit_thread(void)
3262 {
3263         /*
3264          * If thread exited earlier
3265          * there's nothing to be done.
3266          */
3267         if (!ext4_li_info || !ext4_lazyinit_task)
3268                 return;
3269
3270         kthread_stop(ext4_lazyinit_task);
3271 }
3272
3273 static int set_journal_csum_feature_set(struct super_block *sb)
3274 {
3275         int ret = 1;
3276         int compat, incompat;
3277         struct ext4_sb_info *sbi = EXT4_SB(sb);
3278
3279         if (ext4_has_metadata_csum(sb)) {
3280                 /* journal checksum v3 */
3281                 compat = 0;
3282                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3283         } else {
3284                 /* journal checksum v1 */
3285                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3286                 incompat = 0;
3287         }
3288
3289         jbd2_journal_clear_features(sbi->s_journal,
3290                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3291                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3292                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3293         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3294                 ret = jbd2_journal_set_features(sbi->s_journal,
3295                                 compat, 0,
3296                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3297                                 incompat);
3298         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3299                 ret = jbd2_journal_set_features(sbi->s_journal,
3300                                 compat, 0,
3301                                 incompat);
3302                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3303                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3304         } else {
3305                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3306                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3307         }
3308
3309         return ret;
3310 }
3311
3312 /*
3313  * Note: calculating the overhead so we can be compatible with
3314  * historical BSD practice is quite difficult in the face of
3315  * clusters/bigalloc.  This is because multiple metadata blocks from
3316  * different block group can end up in the same allocation cluster.
3317  * Calculating the exact overhead in the face of clustered allocation
3318  * requires either O(all block bitmaps) in memory or O(number of block
3319  * groups**2) in time.  We will still calculate the superblock for
3320  * older file systems --- and if we come across with a bigalloc file
3321  * system with zero in s_overhead_clusters the estimate will be close to
3322  * correct especially for very large cluster sizes --- but for newer
3323  * file systems, it's better to calculate this figure once at mkfs
3324  * time, and store it in the superblock.  If the superblock value is
3325  * present (even for non-bigalloc file systems), we will use it.
3326  */
3327 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3328                           char *buf)
3329 {
3330         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3331         struct ext4_group_desc  *gdp;
3332         ext4_fsblk_t            first_block, last_block, b;
3333         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3334         int                     s, j, count = 0;
3335
3336         if (!ext4_has_feature_bigalloc(sb))
3337                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3338                         sbi->s_itb_per_group + 2);
3339
3340         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3341                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3342         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3343         for (i = 0; i < ngroups; i++) {
3344                 gdp = ext4_get_group_desc(sb, i, NULL);
3345                 b = ext4_block_bitmap(sb, gdp);
3346                 if (b >= first_block && b <= last_block) {
3347                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3348                         count++;
3349                 }
3350                 b = ext4_inode_bitmap(sb, gdp);
3351                 if (b >= first_block && b <= last_block) {
3352                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3353                         count++;
3354                 }
3355                 b = ext4_inode_table(sb, gdp);
3356                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3357                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3358                                 int c = EXT4_B2C(sbi, b - first_block);
3359                                 ext4_set_bit(c, buf);
3360                                 count++;
3361                         }
3362                 if (i != grp)
3363                         continue;
3364                 s = 0;
3365                 if (ext4_bg_has_super(sb, grp)) {
3366                         ext4_set_bit(s++, buf);
3367                         count++;
3368                 }
3369                 j = ext4_bg_num_gdb(sb, grp);
3370                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3371                         ext4_error(sb, "Invalid number of block group "
3372                                    "descriptor blocks: %d", j);
3373                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3374                 }
3375                 count += j;
3376                 for (; j > 0; j--)
3377                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3378         }
3379         if (!count)
3380                 return 0;
3381         return EXT4_CLUSTERS_PER_GROUP(sb) -
3382                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3383 }
3384
3385 /*
3386  * Compute the overhead and stash it in sbi->s_overhead
3387  */
3388 int ext4_calculate_overhead(struct super_block *sb)
3389 {
3390         struct ext4_sb_info *sbi = EXT4_SB(sb);
3391         struct ext4_super_block *es = sbi->s_es;
3392         struct inode *j_inode;
3393         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3394         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3395         ext4_fsblk_t overhead = 0;
3396         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3397
3398         if (!buf)
3399                 return -ENOMEM;
3400
3401         /*
3402          * Compute the overhead (FS structures).  This is constant
3403          * for a given filesystem unless the number of block groups
3404          * changes so we cache the previous value until it does.
3405          */
3406
3407         /*
3408          * All of the blocks before first_data_block are overhead
3409          */
3410         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3411
3412         /*
3413          * Add the overhead found in each block group
3414          */
3415         for (i = 0; i < ngroups; i++) {
3416                 int blks;
3417
3418                 blks = count_overhead(sb, i, buf);
3419                 overhead += blks;
3420                 if (blks)
3421                         memset(buf, 0, PAGE_SIZE);
3422                 cond_resched();
3423         }
3424
3425         /*
3426          * Add the internal journal blocks whether the journal has been
3427          * loaded or not
3428          */
3429         if (sbi->s_journal && !sbi->journal_bdev)
3430                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3431         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3432                 j_inode = ext4_get_journal_inode(sb, j_inum);
3433                 if (j_inode) {
3434                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3435                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3436                         iput(j_inode);
3437                 } else {
3438                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3439                 }
3440         }
3441         sbi->s_overhead = overhead;
3442         smp_wmb();
3443         free_page((unsigned long) buf);
3444         return 0;
3445 }
3446
3447 static void ext4_set_resv_clusters(struct super_block *sb)
3448 {
3449         ext4_fsblk_t resv_clusters;
3450         struct ext4_sb_info *sbi = EXT4_SB(sb);
3451
3452         /*
3453          * There's no need to reserve anything when we aren't using extents.
3454          * The space estimates are exact, there are no unwritten extents,
3455          * hole punching doesn't need new metadata... This is needed especially
3456          * to keep ext2/3 backward compatibility.
3457          */
3458         if (!ext4_has_feature_extents(sb))
3459                 return;
3460         /*
3461          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3462          * This should cover the situations where we can not afford to run
3463          * out of space like for example punch hole, or converting
3464          * unwritten extents in delalloc path. In most cases such
3465          * allocation would require 1, or 2 blocks, higher numbers are
3466          * very rare.
3467          */
3468         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3469                          sbi->s_cluster_bits);
3470
3471         do_div(resv_clusters, 50);
3472         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3473
3474         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3475 }
3476
3477 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3478 {
3479         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3480         char *orig_data = kstrdup(data, GFP_KERNEL);
3481         struct buffer_head *bh;
3482         struct ext4_super_block *es = NULL;
3483         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3484         ext4_fsblk_t block;
3485         ext4_fsblk_t sb_block = get_sb_block(&data);
3486         ext4_fsblk_t logical_sb_block;
3487         unsigned long offset = 0;
3488         unsigned long journal_devnum = 0;
3489         unsigned long def_mount_opts;
3490         struct inode *root;
3491         const char *descr;
3492         int ret = -ENOMEM;
3493         int blocksize, clustersize;
3494         unsigned int db_count;
3495         unsigned int i;
3496         int needs_recovery, has_huge_files, has_bigalloc;
3497         __u64 blocks_count;
3498         int err = 0;
3499         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3500         ext4_group_t first_not_zeroed;
3501
3502         if ((data && !orig_data) || !sbi)
3503                 goto out_free_base;
3504
3505         sbi->s_daxdev = dax_dev;
3506         sbi->s_blockgroup_lock =
3507                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3508         if (!sbi->s_blockgroup_lock)
3509                 goto out_free_base;
3510
3511         sb->s_fs_info = sbi;
3512         sbi->s_sb = sb;
3513         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3514         sbi->s_sb_block = sb_block;
3515         if (sb->s_bdev->bd_part)
3516                 sbi->s_sectors_written_start =
3517                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3518
3519         /* Cleanup superblock name */
3520         strreplace(sb->s_id, '/', '!');
3521
3522         /* -EINVAL is default */
3523         ret = -EINVAL;
3524         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3525         if (!blocksize) {
3526                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3527                 goto out_fail;
3528         }
3529
3530         /*
3531          * The ext4 superblock will not be buffer aligned for other than 1kB
3532          * block sizes.  We need to calculate the offset from buffer start.
3533          */
3534         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3535                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3536                 offset = do_div(logical_sb_block, blocksize);
3537         } else {
3538                 logical_sb_block = sb_block;
3539         }
3540
3541         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3542                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3543                 goto out_fail;
3544         }
3545         /*
3546          * Note: s_es must be initialized as soon as possible because
3547          *       some ext4 macro-instructions depend on its value
3548          */
3549         es = (struct ext4_super_block *) (bh->b_data + offset);
3550         sbi->s_es = es;
3551         sb->s_magic = le16_to_cpu(es->s_magic);
3552         if (sb->s_magic != EXT4_SUPER_MAGIC)
3553                 goto cantfind_ext4;
3554         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3555
3556         /* Warn if metadata_csum and gdt_csum are both set. */
3557         if (ext4_has_feature_metadata_csum(sb) &&
3558             ext4_has_feature_gdt_csum(sb))
3559                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3560                              "redundant flags; please run fsck.");
3561
3562         /* Check for a known checksum algorithm */
3563         if (!ext4_verify_csum_type(sb, es)) {
3564                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3565                          "unknown checksum algorithm.");
3566                 silent = 1;
3567                 goto cantfind_ext4;
3568         }
3569
3570         /* Load the checksum driver */
3571         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3572         if (IS_ERR(sbi->s_chksum_driver)) {
3573                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3574                 ret = PTR_ERR(sbi->s_chksum_driver);
3575                 sbi->s_chksum_driver = NULL;
3576                 goto failed_mount;
3577         }
3578
3579         /* Check superblock checksum */
3580         if (!ext4_superblock_csum_verify(sb, es)) {
3581                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3582                          "invalid superblock checksum.  Run e2fsck?");
3583                 silent = 1;
3584                 ret = -EFSBADCRC;
3585                 goto cantfind_ext4;
3586         }
3587
3588         /* Precompute checksum seed for all metadata */
3589         if (ext4_has_feature_csum_seed(sb))
3590                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3591         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3592                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3593                                                sizeof(es->s_uuid));
3594
3595         /* Set defaults before we parse the mount options */
3596         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3597         set_opt(sb, INIT_INODE_TABLE);
3598         if (def_mount_opts & EXT4_DEFM_DEBUG)
3599                 set_opt(sb, DEBUG);
3600         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3601                 set_opt(sb, GRPID);
3602         if (def_mount_opts & EXT4_DEFM_UID16)
3603                 set_opt(sb, NO_UID32);
3604         /* xattr user namespace & acls are now defaulted on */
3605         set_opt(sb, XATTR_USER);
3606 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3607         set_opt(sb, POSIX_ACL);
3608 #endif
3609         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3610         if (ext4_has_metadata_csum(sb))
3611                 set_opt(sb, JOURNAL_CHECKSUM);
3612
3613         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3614                 set_opt(sb, JOURNAL_DATA);
3615         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3616                 set_opt(sb, ORDERED_DATA);
3617         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3618                 set_opt(sb, WRITEBACK_DATA);
3619
3620         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3621                 set_opt(sb, ERRORS_PANIC);
3622         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3623                 set_opt(sb, ERRORS_CONT);
3624         else
3625                 set_opt(sb, ERRORS_RO);
3626         /* block_validity enabled by default; disable with noblock_validity */
3627         set_opt(sb, BLOCK_VALIDITY);
3628         if (def_mount_opts & EXT4_DEFM_DISCARD)
3629                 set_opt(sb, DISCARD);
3630
3631         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3632         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3633         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3634         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3635         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3636
3637         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3638                 set_opt(sb, BARRIER);
3639
3640         /*
3641          * enable delayed allocation by default
3642          * Use -o nodelalloc to turn it off
3643          */
3644         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3645             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3646                 set_opt(sb, DELALLOC);
3647
3648         /*
3649          * set default s_li_wait_mult for lazyinit, for the case there is
3650          * no mount option specified.
3651          */
3652         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3653
3654         if (sbi->s_es->s_mount_opts[0]) {
3655                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3656                                               sizeof(sbi->s_es->s_mount_opts),
3657                                               GFP_KERNEL);
3658                 if (!s_mount_opts)
3659                         goto failed_mount;
3660                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3661                                    &journal_ioprio, 0)) {
3662                         ext4_msg(sb, KERN_WARNING,
3663                                  "failed to parse options in superblock: %s",
3664                                  s_mount_opts);
3665                 }
3666                 kfree(s_mount_opts);
3667         }
3668         sbi->s_def_mount_opt = sbi->s_mount_opt;
3669         if (!parse_options((char *) data, sb, &journal_devnum,
3670                            &journal_ioprio, 0))
3671                 goto failed_mount;
3672
3673         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3674                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3675                             "with data=journal disables delayed "
3676                             "allocation and O_DIRECT support!\n");
3677                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3678                         ext4_msg(sb, KERN_ERR, "can't mount with "
3679                                  "both data=journal and delalloc");
3680                         goto failed_mount;
3681                 }
3682                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3683                         ext4_msg(sb, KERN_ERR, "can't mount with "
3684                                  "both data=journal and dioread_nolock");
3685                         goto failed_mount;
3686                 }
3687                 if (test_opt(sb, DAX)) {
3688                         ext4_msg(sb, KERN_ERR, "can't mount with "
3689                                  "both data=journal and dax");
3690                         goto failed_mount;
3691                 }
3692                 if (ext4_has_feature_encrypt(sb)) {
3693                         ext4_msg(sb, KERN_WARNING,
3694                                  "encrypted files will use data=ordered "
3695                                  "instead of data journaling mode");
3696                 }
3697                 if (test_opt(sb, DELALLOC))
3698                         clear_opt(sb, DELALLOC);
3699         } else {
3700                 sb->s_iflags |= SB_I_CGROUPWB;
3701         }
3702
3703         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3704                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3705
3706         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3707             (ext4_has_compat_features(sb) ||
3708              ext4_has_ro_compat_features(sb) ||
3709              ext4_has_incompat_features(sb)))
3710                 ext4_msg(sb, KERN_WARNING,
3711                        "feature flags set on rev 0 fs, "
3712                        "running e2fsck is recommended");
3713
3714         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3715                 set_opt2(sb, HURD_COMPAT);
3716                 if (ext4_has_feature_64bit(sb)) {
3717                         ext4_msg(sb, KERN_ERR,
3718                                  "The Hurd can't support 64-bit file systems");
3719                         goto failed_mount;
3720                 }
3721
3722                 /*
3723                  * ea_inode feature uses l_i_version field which is not
3724                  * available in HURD_COMPAT mode.
3725                  */
3726                 if (ext4_has_feature_ea_inode(sb)) {
3727                         ext4_msg(sb, KERN_ERR,
3728                                  "ea_inode feature is not supported for Hurd");
3729                         goto failed_mount;
3730                 }
3731         }
3732
3733         if (IS_EXT2_SB(sb)) {
3734                 if (ext2_feature_set_ok(sb))
3735                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3736                                  "using the ext4 subsystem");
3737                 else {
3738                         /*
3739                          * If we're probing be silent, if this looks like
3740                          * it's actually an ext[34] filesystem.
3741                          */
3742                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3743                                 goto failed_mount;
3744                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3745                                  "to feature incompatibilities");
3746                         goto failed_mount;
3747                 }
3748         }
3749
3750         if (IS_EXT3_SB(sb)) {
3751                 if (ext3_feature_set_ok(sb))
3752                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3753                                  "using the ext4 subsystem");
3754                 else {
3755                         /*
3756                          * If we're probing be silent, if this looks like
3757                          * it's actually an ext4 filesystem.
3758                          */
3759                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3760                                 goto failed_mount;
3761                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3762                                  "to feature incompatibilities");
3763                         goto failed_mount;
3764                 }
3765         }
3766
3767         /*
3768          * Check feature flags regardless of the revision level, since we
3769          * previously didn't change the revision level when setting the flags,
3770          * so there is a chance incompat flags are set on a rev 0 filesystem.
3771          */
3772         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3773                 goto failed_mount;
3774
3775         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3776         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3777             blocksize > EXT4_MAX_BLOCK_SIZE) {
3778                 ext4_msg(sb, KERN_ERR,
3779                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3780                          blocksize, le32_to_cpu(es->s_log_block_size));
3781                 goto failed_mount;
3782         }
3783         if (le32_to_cpu(es->s_log_block_size) >
3784             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3785                 ext4_msg(sb, KERN_ERR,
3786                          "Invalid log block size: %u",
3787                          le32_to_cpu(es->s_log_block_size));
3788                 goto failed_mount;
3789         }
3790         if (le32_to_cpu(es->s_log_cluster_size) >
3791             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3792                 ext4_msg(sb, KERN_ERR,
3793                          "Invalid log cluster size: %u",
3794                          le32_to_cpu(es->s_log_cluster_size));
3795                 goto failed_mount;
3796         }
3797
3798         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3799                 ext4_msg(sb, KERN_ERR,
3800                          "Number of reserved GDT blocks insanely large: %d",
3801                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3802                 goto failed_mount;
3803         }
3804
3805         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3806                 if (ext4_has_feature_inline_data(sb)) {
3807                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3808                                         " that may contain inline data");
3809                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3810                 }
3811                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3812                         ext4_msg(sb, KERN_ERR,
3813                                 "DAX unsupported by block device. Turning off DAX.");
3814                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3815                 }
3816         }
3817
3818         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3819                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3820                          es->s_encryption_level);
3821                 goto failed_mount;
3822         }
3823
3824         if (sb->s_blocksize != blocksize) {
3825                 /* Validate the filesystem blocksize */
3826                 if (!sb_set_blocksize(sb, blocksize)) {
3827                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3828                                         blocksize);
3829                         goto failed_mount;
3830                 }
3831
3832                 brelse(bh);
3833                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3834                 offset = do_div(logical_sb_block, blocksize);
3835                 bh = sb_bread_unmovable(sb, logical_sb_block);
3836                 if (!bh) {
3837                         ext4_msg(sb, KERN_ERR,
3838                                "Can't read superblock on 2nd try");
3839                         goto failed_mount;
3840                 }
3841                 es = (struct ext4_super_block *)(bh->b_data + offset);
3842                 sbi->s_es = es;
3843                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3844                         ext4_msg(sb, KERN_ERR,
3845                                "Magic mismatch, very weird!");
3846                         goto failed_mount;
3847                 }
3848         }
3849
3850         has_huge_files = ext4_has_feature_huge_file(sb);
3851         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3852                                                       has_huge_files);
3853         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3854
3855         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3856                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3857                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3858         } else {
3859                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3860                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3861                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3862                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3863                                  sbi->s_first_ino);
3864                         goto failed_mount;
3865                 }
3866                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3867                     (!is_power_of_2(sbi->s_inode_size)) ||
3868                     (sbi->s_inode_size > blocksize)) {
3869                         ext4_msg(sb, KERN_ERR,
3870                                "unsupported inode size: %d",
3871                                sbi->s_inode_size);
3872                         goto failed_mount;
3873                 }
3874                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3875                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3876         }
3877
3878         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3879         if (ext4_has_feature_64bit(sb)) {
3880                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3881                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3882                     !is_power_of_2(sbi->s_desc_size)) {
3883                         ext4_msg(sb, KERN_ERR,
3884                                "unsupported descriptor size %lu",
3885                                sbi->s_desc_size);
3886                         goto failed_mount;
3887                 }
3888         } else
3889                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3890
3891         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3892         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3893
3894         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3895         if (sbi->s_inodes_per_block == 0)
3896                 goto cantfind_ext4;
3897         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3898             sbi->s_inodes_per_group > blocksize * 8) {
3899                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3900                          sbi->s_blocks_per_group);
3901                 goto failed_mount;
3902         }
3903         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3904                                         sbi->s_inodes_per_block;
3905         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3906         sbi->s_sbh = bh;
3907         sbi->s_mount_state = le16_to_cpu(es->s_state);
3908         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3909         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3910
3911         for (i = 0; i < 4; i++)
3912                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3913         sbi->s_def_hash_version = es->s_def_hash_version;
3914         if (ext4_has_feature_dir_index(sb)) {
3915                 i = le32_to_cpu(es->s_flags);
3916                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3917                         sbi->s_hash_unsigned = 3;
3918                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3919 #ifdef __CHAR_UNSIGNED__
3920                         if (!sb_rdonly(sb))
3921                                 es->s_flags |=
3922                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3923                         sbi->s_hash_unsigned = 3;
3924 #else
3925                         if (!sb_rdonly(sb))
3926                                 es->s_flags |=
3927                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3928 #endif
3929                 }
3930         }
3931
3932         /* Handle clustersize */
3933         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3934         has_bigalloc = ext4_has_feature_bigalloc(sb);
3935         if (has_bigalloc) {
3936                 if (clustersize < blocksize) {
3937                         ext4_msg(sb, KERN_ERR,
3938                                  "cluster size (%d) smaller than "
3939                                  "block size (%d)", clustersize, blocksize);
3940                         goto failed_mount;
3941                 }
3942                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3943                         le32_to_cpu(es->s_log_block_size);
3944                 sbi->s_clusters_per_group =
3945                         le32_to_cpu(es->s_clusters_per_group);
3946                 if (sbi->s_clusters_per_group > blocksize * 8) {
3947                         ext4_msg(sb, KERN_ERR,
3948                                  "#clusters per group too big: %lu",
3949                                  sbi->s_clusters_per_group);
3950                         goto failed_mount;
3951                 }
3952                 if (sbi->s_blocks_per_group !=
3953                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3954                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3955                                  "clusters per group (%lu) inconsistent",
3956                                  sbi->s_blocks_per_group,
3957                                  sbi->s_clusters_per_group);
3958                         goto failed_mount;
3959                 }
3960         } else {
3961                 if (clustersize != blocksize) {
3962                         ext4_msg(sb, KERN_ERR,
3963                                  "fragment/cluster size (%d) != "
3964                                  "block size (%d)", clustersize, blocksize);
3965                         goto failed_mount;
3966                 }
3967                 if (sbi->s_blocks_per_group > blocksize * 8) {
3968                         ext4_msg(sb, KERN_ERR,
3969                                  "#blocks per group too big: %lu",
3970                                  sbi->s_blocks_per_group);
3971                         goto failed_mount;
3972                 }
3973                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3974                 sbi->s_cluster_bits = 0;
3975         }
3976         sbi->s_cluster_ratio = clustersize / blocksize;
3977
3978         /* Do we have standard group size of clustersize * 8 blocks ? */
3979         if (sbi->s_blocks_per_group == clustersize << 3)
3980                 set_opt2(sb, STD_GROUP_SIZE);
3981
3982         /*
3983          * Test whether we have more sectors than will fit in sector_t,
3984          * and whether the max offset is addressable by the page cache.
3985          */
3986         err = generic_check_addressable(sb->s_blocksize_bits,
3987                                         ext4_blocks_count(es));
3988         if (err) {
3989                 ext4_msg(sb, KERN_ERR, "filesystem"
3990                          " too large to mount safely on this system");
3991                 if (sizeof(sector_t) < 8)
3992                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3993                 goto failed_mount;
3994         }
3995
3996         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3997                 goto cantfind_ext4;
3998
3999         /* check blocks count against device size */
4000         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4001         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4002                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4003                        "exceeds size of device (%llu blocks)",
4004                        ext4_blocks_count(es), blocks_count);
4005                 goto failed_mount;
4006         }
4007
4008         /*
4009          * It makes no sense for the first data block to be beyond the end
4010          * of the filesystem.
4011          */
4012         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4013                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4014                          "block %u is beyond end of filesystem (%llu)",
4015                          le32_to_cpu(es->s_first_data_block),
4016                          ext4_blocks_count(es));
4017                 goto failed_mount;
4018         }
4019         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4020             (sbi->s_cluster_ratio == 1)) {
4021                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4022                          "block is 0 with a 1k block and cluster size");
4023                 goto failed_mount;
4024         }
4025
4026         blocks_count = (ext4_blocks_count(es) -
4027                         le32_to_cpu(es->s_first_data_block) +
4028                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4029         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4030         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4031                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4032                        "(block count %llu, first data block %u, "
4033                        "blocks per group %lu)", sbi->s_groups_count,
4034                        ext4_blocks_count(es),
4035                        le32_to_cpu(es->s_first_data_block),
4036                        EXT4_BLOCKS_PER_GROUP(sb));
4037                 goto failed_mount;
4038         }
4039         sbi->s_groups_count = blocks_count;
4040         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4041                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4042         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4043                    EXT4_DESC_PER_BLOCK(sb);
4044         if (ext4_has_feature_meta_bg(sb)) {
4045                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4046                         ext4_msg(sb, KERN_WARNING,
4047                                  "first meta block group too large: %u "
4048                                  "(group descriptor block count %u)",
4049                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4050                         goto failed_mount;
4051                 }
4052         }
4053         sbi->s_group_desc = kvmalloc_array(db_count,
4054                                            sizeof(struct buffer_head *),
4055                                            GFP_KERNEL);
4056         if (sbi->s_group_desc == NULL) {
4057                 ext4_msg(sb, KERN_ERR, "not enough memory");
4058                 ret = -ENOMEM;
4059                 goto failed_mount;
4060         }
4061         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4062             le32_to_cpu(es->s_inodes_count)) {
4063                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4064                          le32_to_cpu(es->s_inodes_count),
4065                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4066                 ret = -EINVAL;
4067                 goto failed_mount;
4068         }
4069
4070         bgl_lock_init(sbi->s_blockgroup_lock);
4071
4072         /* Pre-read the descriptors into the buffer cache */
4073         for (i = 0; i < db_count; i++) {
4074                 block = descriptor_loc(sb, logical_sb_block, i);
4075                 sb_breadahead(sb, block);
4076         }
4077
4078         for (i = 0; i < db_count; i++) {
4079                 block = descriptor_loc(sb, logical_sb_block, i);
4080                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4081                 if (!sbi->s_group_desc[i]) {
4082                         ext4_msg(sb, KERN_ERR,
4083                                "can't read group descriptor %d", i);
4084                         db_count = i;
4085                         goto failed_mount2;
4086                 }
4087         }
4088         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4089                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4090                 ret = -EFSCORRUPTED;
4091                 goto failed_mount2;
4092         }
4093
4094         sbi->s_gdb_count = db_count;
4095
4096         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4097
4098         /* Register extent status tree shrinker */
4099         if (ext4_es_register_shrinker(sbi))
4100                 goto failed_mount3;
4101
4102         sbi->s_stripe = ext4_get_stripe_size(sbi);
4103         sbi->s_extent_max_zeroout_kb = 32;
4104
4105         /*
4106          * set up enough so that it can read an inode
4107          */
4108         sb->s_op = &ext4_sops;
4109         sb->s_export_op = &ext4_export_ops;
4110         sb->s_xattr = ext4_xattr_handlers;
4111 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4112         sb->s_cop = &ext4_cryptops;
4113 #endif
4114 #ifdef CONFIG_QUOTA
4115         sb->dq_op = &ext4_quota_operations;
4116         if (ext4_has_feature_quota(sb))
4117                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4118         else
4119                 sb->s_qcop = &ext4_qctl_operations;
4120         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4121 #endif
4122         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4123
4124         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4125         mutex_init(&sbi->s_orphan_lock);
4126
4127         sb->s_root = NULL;
4128
4129         needs_recovery = (es->s_last_orphan != 0 ||
4130                           ext4_has_feature_journal_needs_recovery(sb));
4131
4132         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4133                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4134                         goto failed_mount3a;
4135
4136         /*
4137          * The first inode we look at is the journal inode.  Don't try
4138          * root first: it may be modified in the journal!
4139          */
4140         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4141                 err = ext4_load_journal(sb, es, journal_devnum);
4142                 if (err)
4143                         goto failed_mount3a;
4144         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4145                    ext4_has_feature_journal_needs_recovery(sb)) {
4146                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4147                        "suppressed and not mounted read-only");
4148                 goto failed_mount_wq;
4149         } else {
4150                 /* Nojournal mode, all journal mount options are illegal */
4151                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4152                         ext4_msg(sb, KERN_ERR, "can't mount with "
4153                                  "journal_checksum, fs mounted w/o journal");
4154                         goto failed_mount_wq;
4155                 }
4156                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4157                         ext4_msg(sb, KERN_ERR, "can't mount with "
4158                                  "journal_async_commit, fs mounted w/o journal");
4159                         goto failed_mount_wq;
4160                 }
4161                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4162                         ext4_msg(sb, KERN_ERR, "can't mount with "
4163                                  "commit=%lu, fs mounted w/o journal",
4164                                  sbi->s_commit_interval / HZ);
4165                         goto failed_mount_wq;
4166                 }
4167                 if (EXT4_MOUNT_DATA_FLAGS &
4168                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4169                         ext4_msg(sb, KERN_ERR, "can't mount with "
4170                                  "data=, fs mounted w/o journal");
4171                         goto failed_mount_wq;
4172                 }
4173                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4174                 clear_opt(sb, JOURNAL_CHECKSUM);
4175                 clear_opt(sb, DATA_FLAGS);
4176                 sbi->s_journal = NULL;
4177                 needs_recovery = 0;
4178                 goto no_journal;
4179         }
4180
4181         if (ext4_has_feature_64bit(sb) &&
4182             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4183                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4184                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4185                 goto failed_mount_wq;
4186         }
4187
4188         if (!set_journal_csum_feature_set(sb)) {
4189                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4190                          "feature set");
4191                 goto failed_mount_wq;
4192         }
4193
4194         /* We have now updated the journal if required, so we can
4195          * validate the data journaling mode. */
4196         switch (test_opt(sb, DATA_FLAGS)) {
4197         case 0:
4198                 /* No mode set, assume a default based on the journal
4199                  * capabilities: ORDERED_DATA if the journal can
4200                  * cope, else JOURNAL_DATA
4201                  */
4202                 if (jbd2_journal_check_available_features
4203                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4204                         set_opt(sb, ORDERED_DATA);
4205                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4206                 } else {
4207                         set_opt(sb, JOURNAL_DATA);
4208                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4209                 }
4210                 break;
4211
4212         case EXT4_MOUNT_ORDERED_DATA:
4213         case EXT4_MOUNT_WRITEBACK_DATA:
4214                 if (!jbd2_journal_check_available_features
4215                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4216                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4217                                "requested data journaling mode");
4218                         goto failed_mount_wq;
4219                 }
4220         default:
4221                 break;
4222         }
4223
4224         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4225             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4226                 ext4_msg(sb, KERN_ERR, "can't mount with "
4227                         "journal_async_commit in data=ordered mode");
4228                 goto failed_mount_wq;
4229         }
4230
4231         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4232
4233         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4234
4235 no_journal:
4236         if (!test_opt(sb, NO_MBCACHE)) {
4237                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4238                 if (!sbi->s_ea_block_cache) {
4239                         ext4_msg(sb, KERN_ERR,
4240                                  "Failed to create ea_block_cache");
4241                         goto failed_mount_wq;
4242                 }
4243
4244                 if (ext4_has_feature_ea_inode(sb)) {
4245                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4246                         if (!sbi->s_ea_inode_cache) {
4247                                 ext4_msg(sb, KERN_ERR,
4248                                          "Failed to create ea_inode_cache");
4249                                 goto failed_mount_wq;
4250                         }
4251                 }
4252         }
4253
4254         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4255             (blocksize != PAGE_SIZE)) {
4256                 ext4_msg(sb, KERN_ERR,
4257                          "Unsupported blocksize for fs encryption");
4258                 goto failed_mount_wq;
4259         }
4260
4261         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4262             !ext4_has_feature_encrypt(sb)) {
4263                 ext4_set_feature_encrypt(sb);
4264                 ext4_commit_super(sb, 1);
4265         }
4266
4267         /*
4268          * Get the # of file system overhead blocks from the
4269          * superblock if present.
4270          */
4271         if (es->s_overhead_clusters)
4272                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4273         else {
4274                 err = ext4_calculate_overhead(sb);
4275                 if (err)
4276                         goto failed_mount_wq;
4277         }
4278
4279         /*
4280          * The maximum number of concurrent works can be high and
4281          * concurrency isn't really necessary.  Limit it to 1.
4282          */
4283         EXT4_SB(sb)->rsv_conversion_wq =
4284                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4285         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4286                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4287                 ret = -ENOMEM;
4288                 goto failed_mount4;
4289         }
4290
4291         /*
4292          * The jbd2_journal_load will have done any necessary log recovery,
4293          * so we can safely mount the rest of the filesystem now.
4294          */
4295
4296         root = ext4_iget(sb, EXT4_ROOT_INO);
4297         if (IS_ERR(root)) {
4298                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4299                 ret = PTR_ERR(root);
4300                 root = NULL;
4301                 goto failed_mount4;
4302         }
4303         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4304                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4305                 iput(root);
4306                 goto failed_mount4;
4307         }
4308         sb->s_root = d_make_root(root);
4309         if (!sb->s_root) {
4310                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4311                 ret = -ENOMEM;
4312                 goto failed_mount4;
4313         }
4314
4315         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4316         if (ret == -EROFS) {
4317                 sb->s_flags |= SB_RDONLY;
4318                 ret = 0;
4319         } else if (ret)
4320                 goto failed_mount4a;
4321
4322         /* determine the minimum size of new large inodes, if present */
4323         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4324             sbi->s_want_extra_isize == 0) {
4325                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4326                                                      EXT4_GOOD_OLD_INODE_SIZE;
4327                 if (ext4_has_feature_extra_isize(sb)) {
4328                         if (sbi->s_want_extra_isize <
4329                             le16_to_cpu(es->s_want_extra_isize))
4330                                 sbi->s_want_extra_isize =
4331                                         le16_to_cpu(es->s_want_extra_isize);
4332                         if (sbi->s_want_extra_isize <
4333                             le16_to_cpu(es->s_min_extra_isize))
4334                                 sbi->s_want_extra_isize =
4335                                         le16_to_cpu(es->s_min_extra_isize);
4336                 }
4337         }
4338         /* Check if enough inode space is available */
4339         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4340                                                         sbi->s_inode_size) {
4341                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4342                                                        EXT4_GOOD_OLD_INODE_SIZE;
4343                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4344                          "available");
4345         }
4346
4347         ext4_set_resv_clusters(sb);
4348
4349         err = ext4_setup_system_zone(sb);
4350         if (err) {
4351                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4352                          "zone (%d)", err);
4353                 goto failed_mount4a;
4354         }
4355
4356         ext4_ext_init(sb);
4357         err = ext4_mb_init(sb);
4358         if (err) {
4359                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4360                          err);
4361                 goto failed_mount5;
4362         }
4363
4364         block = ext4_count_free_clusters(sb);
4365         ext4_free_blocks_count_set(sbi->s_es,
4366                                    EXT4_C2B(sbi, block));
4367         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4368                                   GFP_KERNEL);
4369         if (!err) {
4370                 unsigned long freei = ext4_count_free_inodes(sb);
4371                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4372                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4373                                           GFP_KERNEL);
4374         }
4375         if (!err)
4376                 err = percpu_counter_init(&sbi->s_dirs_counter,
4377                                           ext4_count_dirs(sb), GFP_KERNEL);
4378         if (!err)
4379                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4380                                           GFP_KERNEL);
4381         if (!err)
4382                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4383
4384         if (err) {
4385                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4386                 goto failed_mount6;
4387         }
4388
4389         if (ext4_has_feature_flex_bg(sb))
4390                 if (!ext4_fill_flex_info(sb)) {
4391                         ext4_msg(sb, KERN_ERR,
4392                                "unable to initialize "
4393                                "flex_bg meta info!");
4394                         goto failed_mount6;
4395                 }
4396
4397         err = ext4_register_li_request(sb, first_not_zeroed);
4398         if (err)
4399                 goto failed_mount6;
4400
4401         err = ext4_register_sysfs(sb);
4402         if (err)
4403                 goto failed_mount7;
4404
4405 #ifdef CONFIG_QUOTA
4406         /* Enable quota usage during mount. */
4407         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4408                 err = ext4_enable_quotas(sb);
4409                 if (err)
4410                         goto failed_mount8;
4411         }
4412 #endif  /* CONFIG_QUOTA */
4413
4414         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4415         ext4_orphan_cleanup(sb, es);
4416         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4417         if (needs_recovery) {
4418                 ext4_msg(sb, KERN_INFO, "recovery complete");
4419                 ext4_mark_recovery_complete(sb, es);
4420         }
4421         if (EXT4_SB(sb)->s_journal) {
4422                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4423                         descr = " journalled data mode";
4424                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4425                         descr = " ordered data mode";
4426                 else
4427                         descr = " writeback data mode";
4428         } else
4429                 descr = "out journal";
4430
4431         if (test_opt(sb, DISCARD)) {
4432                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4433                 if (!blk_queue_discard(q))
4434                         ext4_msg(sb, KERN_WARNING,
4435                                  "mounting with \"discard\" option, but "
4436                                  "the device does not support discard");
4437         }
4438
4439         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4440                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4441                          "Opts: %.*s%s%s", descr,
4442                          (int) sizeof(sbi->s_es->s_mount_opts),
4443                          sbi->s_es->s_mount_opts,
4444                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4445
4446         if (es->s_error_count)
4447                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4448
4449         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4450         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4451         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4452         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4453
4454         kfree(orig_data);
4455         return 0;
4456
4457 cantfind_ext4:
4458         if (!silent)
4459                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4460         goto failed_mount;
4461
4462 #ifdef CONFIG_QUOTA
4463 failed_mount8:
4464         ext4_unregister_sysfs(sb);
4465 #endif
4466 failed_mount7:
4467         ext4_unregister_li_request(sb);
4468 failed_mount6:
4469         ext4_mb_release(sb);
4470         if (sbi->s_flex_groups)
4471                 kvfree(sbi->s_flex_groups);
4472         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4473         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4474         percpu_counter_destroy(&sbi->s_dirs_counter);
4475         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4476 failed_mount5:
4477         ext4_ext_release(sb);
4478         ext4_release_system_zone(sb);
4479 failed_mount4a:
4480         dput(sb->s_root);
4481         sb->s_root = NULL;
4482 failed_mount4:
4483         ext4_msg(sb, KERN_ERR, "mount failed");
4484         if (EXT4_SB(sb)->rsv_conversion_wq)
4485                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4486 failed_mount_wq:
4487         if (sbi->s_ea_inode_cache) {
4488                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4489                 sbi->s_ea_inode_cache = NULL;
4490         }
4491         if (sbi->s_ea_block_cache) {
4492                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4493                 sbi->s_ea_block_cache = NULL;
4494         }
4495         if (sbi->s_journal) {
4496                 jbd2_journal_destroy(sbi->s_journal);
4497                 sbi->s_journal = NULL;
4498         }
4499 failed_mount3a:
4500         ext4_es_unregister_shrinker(sbi);
4501 failed_mount3:
4502         del_timer_sync(&sbi->s_err_report);
4503         if (sbi->s_mmp_tsk)
4504                 kthread_stop(sbi->s_mmp_tsk);
4505 failed_mount2:
4506         for (i = 0; i < db_count; i++)
4507                 brelse(sbi->s_group_desc[i]);
4508         kvfree(sbi->s_group_desc);
4509 failed_mount:
4510         if (sbi->s_chksum_driver)
4511                 crypto_free_shash(sbi->s_chksum_driver);
4512 #ifdef CONFIG_QUOTA
4513         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4514                 kfree(sbi->s_qf_names[i]);
4515 #endif
4516         ext4_blkdev_remove(sbi);
4517         brelse(bh);
4518 out_fail:
4519         sb->s_fs_info = NULL;
4520         kfree(sbi->s_blockgroup_lock);
4521 out_free_base:
4522         kfree(sbi);
4523         kfree(orig_data);
4524         fs_put_dax(dax_dev);
4525         return err ? err : ret;
4526 }
4527
4528 /*
4529  * Setup any per-fs journal parameters now.  We'll do this both on
4530  * initial mount, once the journal has been initialised but before we've
4531  * done any recovery; and again on any subsequent remount.
4532  */
4533 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4534 {
4535         struct ext4_sb_info *sbi = EXT4_SB(sb);
4536
4537         journal->j_commit_interval = sbi->s_commit_interval;
4538         journal->j_min_batch_time = sbi->s_min_batch_time;
4539         journal->j_max_batch_time = sbi->s_max_batch_time;
4540
4541         write_lock(&journal->j_state_lock);
4542         if (test_opt(sb, BARRIER))
4543                 journal->j_flags |= JBD2_BARRIER;
4544         else
4545                 journal->j_flags &= ~JBD2_BARRIER;
4546         if (test_opt(sb, DATA_ERR_ABORT))
4547                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4548         else
4549                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4550         write_unlock(&journal->j_state_lock);
4551 }
4552
4553 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4554                                              unsigned int journal_inum)
4555 {
4556         struct inode *journal_inode;
4557
4558         /*
4559          * Test for the existence of a valid inode on disk.  Bad things
4560          * happen if we iget() an unused inode, as the subsequent iput()
4561          * will try to delete it.
4562          */
4563         journal_inode = ext4_iget(sb, journal_inum);
4564         if (IS_ERR(journal_inode)) {
4565                 ext4_msg(sb, KERN_ERR, "no journal found");
4566                 return NULL;
4567         }
4568         if (!journal_inode->i_nlink) {
4569                 make_bad_inode(journal_inode);
4570                 iput(journal_inode);
4571                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4572                 return NULL;
4573         }
4574
4575         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4576                   journal_inode, journal_inode->i_size);
4577         if (!S_ISREG(journal_inode->i_mode)) {
4578                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4579                 iput(journal_inode);
4580                 return NULL;
4581         }
4582         return journal_inode;
4583 }
4584
4585 static journal_t *ext4_get_journal(struct super_block *sb,
4586                                    unsigned int journal_inum)
4587 {
4588         struct inode *journal_inode;
4589         journal_t *journal;
4590
4591         BUG_ON(!ext4_has_feature_journal(sb));
4592
4593         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4594         if (!journal_inode)
4595                 return NULL;
4596
4597         journal = jbd2_journal_init_inode(journal_inode);
4598         if (!journal) {
4599                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4600                 iput(journal_inode);
4601                 return NULL;
4602         }
4603         journal->j_private = sb;
4604         ext4_init_journal_params(sb, journal);
4605         return journal;
4606 }
4607
4608 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4609                                        dev_t j_dev)
4610 {
4611         struct buffer_head *bh;
4612         journal_t *journal;
4613         ext4_fsblk_t start;
4614         ext4_fsblk_t len;
4615         int hblock, blocksize;
4616         ext4_fsblk_t sb_block;
4617         unsigned long offset;
4618         struct ext4_super_block *es;
4619         struct block_device *bdev;
4620
4621         BUG_ON(!ext4_has_feature_journal(sb));
4622
4623         bdev = ext4_blkdev_get(j_dev, sb);
4624         if (bdev == NULL)
4625                 return NULL;
4626
4627         blocksize = sb->s_blocksize;
4628         hblock = bdev_logical_block_size(bdev);
4629         if (blocksize < hblock) {
4630                 ext4_msg(sb, KERN_ERR,
4631                         "blocksize too small for journal device");
4632                 goto out_bdev;
4633         }
4634
4635         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4636         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4637         set_blocksize(bdev, blocksize);
4638         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4639                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4640                        "external journal");
4641                 goto out_bdev;
4642         }
4643
4644         es = (struct ext4_super_block *) (bh->b_data + offset);
4645         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4646             !(le32_to_cpu(es->s_feature_incompat) &
4647               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4648                 ext4_msg(sb, KERN_ERR, "external journal has "
4649                                         "bad superblock");
4650                 brelse(bh);
4651                 goto out_bdev;
4652         }
4653
4654         if ((le32_to_cpu(es->s_feature_ro_compat) &
4655              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4656             es->s_checksum != ext4_superblock_csum(sb, es)) {
4657                 ext4_msg(sb, KERN_ERR, "external journal has "
4658                                        "corrupt superblock");
4659                 brelse(bh);
4660                 goto out_bdev;
4661         }
4662
4663         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4664                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4665                 brelse(bh);
4666                 goto out_bdev;
4667         }
4668
4669         len = ext4_blocks_count(es);
4670         start = sb_block + 1;
4671         brelse(bh);     /* we're done with the superblock */
4672
4673         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4674                                         start, len, blocksize);
4675         if (!journal) {
4676                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4677                 goto out_bdev;
4678         }
4679         journal->j_private = sb;
4680         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4681         wait_on_buffer(journal->j_sb_buffer);
4682         if (!buffer_uptodate(journal->j_sb_buffer)) {
4683                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4684                 goto out_journal;
4685         }
4686         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4687                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4688                                         "user (unsupported) - %d",
4689                         be32_to_cpu(journal->j_superblock->s_nr_users));
4690                 goto out_journal;
4691         }
4692         EXT4_SB(sb)->journal_bdev = bdev;
4693         ext4_init_journal_params(sb, journal);
4694         return journal;
4695
4696 out_journal:
4697         jbd2_journal_destroy(journal);
4698 out_bdev:
4699         ext4_blkdev_put(bdev);
4700         return NULL;
4701 }
4702
4703 static int ext4_load_journal(struct super_block *sb,
4704                              struct ext4_super_block *es,
4705                              unsigned long journal_devnum)
4706 {
4707         journal_t *journal;
4708         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4709         dev_t journal_dev;
4710         int err = 0;
4711         int really_read_only;
4712
4713         BUG_ON(!ext4_has_feature_journal(sb));
4714
4715         if (journal_devnum &&
4716             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4717                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4718                         "numbers have changed");
4719                 journal_dev = new_decode_dev(journal_devnum);
4720         } else
4721                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4722
4723         really_read_only = bdev_read_only(sb->s_bdev);
4724
4725         /*
4726          * Are we loading a blank journal or performing recovery after a
4727          * crash?  For recovery, we need to check in advance whether we
4728          * can get read-write access to the device.
4729          */
4730         if (ext4_has_feature_journal_needs_recovery(sb)) {
4731                 if (sb_rdonly(sb)) {
4732                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4733                                         "required on readonly filesystem");
4734                         if (really_read_only) {
4735                                 ext4_msg(sb, KERN_ERR, "write access "
4736                                         "unavailable, cannot proceed "
4737                                         "(try mounting with noload)");
4738                                 return -EROFS;
4739                         }
4740                         ext4_msg(sb, KERN_INFO, "write access will "
4741                                "be enabled during recovery");
4742                 }
4743         }
4744
4745         if (journal_inum && journal_dev) {
4746                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4747                        "and inode journals!");
4748                 return -EINVAL;
4749         }
4750
4751         if (journal_inum) {
4752                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4753                         return -EINVAL;
4754         } else {
4755                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4756                         return -EINVAL;
4757         }
4758
4759         if (!(journal->j_flags & JBD2_BARRIER))
4760                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4761
4762         if (!ext4_has_feature_journal_needs_recovery(sb))
4763                 err = jbd2_journal_wipe(journal, !really_read_only);
4764         if (!err) {
4765                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4766                 if (save)
4767                         memcpy(save, ((char *) es) +
4768                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4769                 err = jbd2_journal_load(journal);
4770                 if (save)
4771                         memcpy(((char *) es) + EXT4_S_ERR_START,
4772                                save, EXT4_S_ERR_LEN);
4773                 kfree(save);
4774         }
4775
4776         if (err) {
4777                 ext4_msg(sb, KERN_ERR, "error loading journal");
4778                 jbd2_journal_destroy(journal);
4779                 return err;
4780         }
4781
4782         EXT4_SB(sb)->s_journal = journal;
4783         ext4_clear_journal_err(sb, es);
4784
4785         if (!really_read_only && journal_devnum &&
4786             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4787                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4788
4789                 /* Make sure we flush the recovery flag to disk. */
4790                 ext4_commit_super(sb, 1);
4791         }
4792
4793         return 0;
4794 }
4795
4796 static int ext4_commit_super(struct super_block *sb, int sync)
4797 {
4798         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4799         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4800         int error = 0;
4801
4802         if (!sbh || block_device_ejected(sb))
4803                 return error;
4804
4805         /*
4806          * The superblock bh should be mapped, but it might not be if the
4807          * device was hot-removed. Not much we can do but fail the I/O.
4808          */
4809         if (!buffer_mapped(sbh))
4810                 return error;
4811
4812         /*
4813          * If the file system is mounted read-only, don't update the
4814          * superblock write time.  This avoids updating the superblock
4815          * write time when we are mounting the root file system
4816          * read/only but we need to replay the journal; at that point,
4817          * for people who are east of GMT and who make their clock
4818          * tick in localtime for Windows bug-for-bug compatibility,
4819          * the clock is set in the future, and this will cause e2fsck
4820          * to complain and force a full file system check.
4821          */
4822         if (!(sb->s_flags & SB_RDONLY))
4823                 es->s_wtime = cpu_to_le32(get_seconds());
4824         if (sb->s_bdev->bd_part)
4825                 es->s_kbytes_written =
4826                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4827                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4828                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4829         else
4830                 es->s_kbytes_written =
4831                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4832         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4833                 ext4_free_blocks_count_set(es,
4834                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4835                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4836         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4837                 es->s_free_inodes_count =
4838                         cpu_to_le32(percpu_counter_sum_positive(
4839                                 &EXT4_SB(sb)->s_freeinodes_counter));
4840         BUFFER_TRACE(sbh, "marking dirty");
4841         ext4_superblock_csum_set(sb);
4842         if (sync)
4843                 lock_buffer(sbh);
4844         if (buffer_write_io_error(sbh)) {
4845                 /*
4846                  * Oh, dear.  A previous attempt to write the
4847                  * superblock failed.  This could happen because the
4848                  * USB device was yanked out.  Or it could happen to
4849                  * be a transient write error and maybe the block will
4850                  * be remapped.  Nothing we can do but to retry the
4851                  * write and hope for the best.
4852                  */
4853                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4854                        "superblock detected");
4855                 clear_buffer_write_io_error(sbh);
4856                 set_buffer_uptodate(sbh);
4857         }
4858         mark_buffer_dirty(sbh);
4859         if (sync) {
4860                 unlock_buffer(sbh);
4861                 error = __sync_dirty_buffer(sbh,
4862                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4863                 if (buffer_write_io_error(sbh)) {
4864                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4865                                "superblock");
4866                         clear_buffer_write_io_error(sbh);
4867                         set_buffer_uptodate(sbh);
4868                 }
4869         }
4870         return error;
4871 }
4872
4873 /*
4874  * Have we just finished recovery?  If so, and if we are mounting (or
4875  * remounting) the filesystem readonly, then we will end up with a
4876  * consistent fs on disk.  Record that fact.
4877  */
4878 static void ext4_mark_recovery_complete(struct super_block *sb,
4879                                         struct ext4_super_block *es)
4880 {
4881         journal_t *journal = EXT4_SB(sb)->s_journal;
4882
4883         if (!ext4_has_feature_journal(sb)) {
4884                 BUG_ON(journal != NULL);
4885                 return;
4886         }
4887         jbd2_journal_lock_updates(journal);
4888         if (jbd2_journal_flush(journal) < 0)
4889                 goto out;
4890
4891         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4892                 ext4_clear_feature_journal_needs_recovery(sb);
4893                 ext4_commit_super(sb, 1);
4894         }
4895
4896 out:
4897         jbd2_journal_unlock_updates(journal);
4898 }
4899
4900 /*
4901  * If we are mounting (or read-write remounting) a filesystem whose journal
4902  * has recorded an error from a previous lifetime, move that error to the
4903  * main filesystem now.
4904  */
4905 static void ext4_clear_journal_err(struct super_block *sb,
4906                                    struct ext4_super_block *es)
4907 {
4908         journal_t *journal;
4909         int j_errno;
4910         const char *errstr;
4911
4912         BUG_ON(!ext4_has_feature_journal(sb));
4913
4914         journal = EXT4_SB(sb)->s_journal;
4915
4916         /*
4917          * Now check for any error status which may have been recorded in the
4918          * journal by a prior ext4_error() or ext4_abort()
4919          */
4920
4921         j_errno = jbd2_journal_errno(journal);
4922         if (j_errno) {
4923                 char nbuf[16];
4924
4925                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4926                 ext4_warning(sb, "Filesystem error recorded "
4927                              "from previous mount: %s", errstr);
4928                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4929
4930                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4931                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4932                 ext4_commit_super(sb, 1);
4933
4934                 jbd2_journal_clear_err(journal);
4935                 jbd2_journal_update_sb_errno(journal);
4936         }
4937 }
4938
4939 /*
4940  * Force the running and committing transactions to commit,
4941  * and wait on the commit.
4942  */
4943 int ext4_force_commit(struct super_block *sb)
4944 {
4945         journal_t *journal;
4946
4947         if (sb_rdonly(sb))
4948                 return 0;
4949
4950         journal = EXT4_SB(sb)->s_journal;
4951         return ext4_journal_force_commit(journal);
4952 }
4953
4954 static int ext4_sync_fs(struct super_block *sb, int wait)
4955 {
4956         int ret = 0;
4957         tid_t target;
4958         bool needs_barrier = false;
4959         struct ext4_sb_info *sbi = EXT4_SB(sb);
4960
4961         if (unlikely(ext4_forced_shutdown(sbi)))
4962                 return 0;
4963
4964         trace_ext4_sync_fs(sb, wait);
4965         flush_workqueue(sbi->rsv_conversion_wq);
4966         /*
4967          * Writeback quota in non-journalled quota case - journalled quota has
4968          * no dirty dquots
4969          */
4970         dquot_writeback_dquots(sb, -1);
4971         /*
4972          * Data writeback is possible w/o journal transaction, so barrier must
4973          * being sent at the end of the function. But we can skip it if
4974          * transaction_commit will do it for us.
4975          */
4976         if (sbi->s_journal) {
4977                 target = jbd2_get_latest_transaction(sbi->s_journal);
4978                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4979                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4980                         needs_barrier = true;
4981
4982                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4983                         if (wait)
4984                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4985                                                            target);
4986                 }
4987         } else if (wait && test_opt(sb, BARRIER))
4988                 needs_barrier = true;
4989         if (needs_barrier) {
4990                 int err;
4991                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4992                 if (!ret)
4993                         ret = err;
4994         }
4995
4996         return ret;
4997 }
4998
4999 /*
5000  * LVM calls this function before a (read-only) snapshot is created.  This
5001  * gives us a chance to flush the journal completely and mark the fs clean.
5002  *
5003  * Note that only this function cannot bring a filesystem to be in a clean
5004  * state independently. It relies on upper layer to stop all data & metadata
5005  * modifications.
5006  */
5007 static int ext4_freeze(struct super_block *sb)
5008 {
5009         int error = 0;
5010         journal_t *journal;
5011
5012         if (sb_rdonly(sb))
5013                 return 0;
5014
5015         journal = EXT4_SB(sb)->s_journal;
5016
5017         if (journal) {
5018                 /* Now we set up the journal barrier. */
5019                 jbd2_journal_lock_updates(journal);
5020
5021                 /*
5022                  * Don't clear the needs_recovery flag if we failed to
5023                  * flush the journal.
5024                  */
5025                 error = jbd2_journal_flush(journal);
5026                 if (error < 0)
5027                         goto out;
5028
5029                 /* Journal blocked and flushed, clear needs_recovery flag. */
5030                 ext4_clear_feature_journal_needs_recovery(sb);
5031         }
5032
5033         error = ext4_commit_super(sb, 1);
5034 out:
5035         if (journal)
5036                 /* we rely on upper layer to stop further updates */
5037                 jbd2_journal_unlock_updates(journal);
5038         return error;
5039 }
5040
5041 /*
5042  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5043  * flag here, even though the filesystem is not technically dirty yet.
5044  */
5045 static int ext4_unfreeze(struct super_block *sb)
5046 {
5047         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5048                 return 0;
5049
5050         if (EXT4_SB(sb)->s_journal) {
5051                 /* Reset the needs_recovery flag before the fs is unlocked. */
5052                 ext4_set_feature_journal_needs_recovery(sb);
5053         }
5054
5055         ext4_commit_super(sb, 1);
5056         return 0;
5057 }
5058
5059 /*
5060  * Structure to save mount options for ext4_remount's benefit
5061  */
5062 struct ext4_mount_options {
5063         unsigned long s_mount_opt;
5064         unsigned long s_mount_opt2;
5065         kuid_t s_resuid;
5066         kgid_t s_resgid;
5067         unsigned long s_commit_interval;
5068         u32 s_min_batch_time, s_max_batch_time;
5069 #ifdef CONFIG_QUOTA
5070         int s_jquota_fmt;
5071         char *s_qf_names[EXT4_MAXQUOTAS];
5072 #endif
5073 };
5074
5075 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5076 {
5077         struct ext4_super_block *es;
5078         struct ext4_sb_info *sbi = EXT4_SB(sb);
5079         unsigned long old_sb_flags;
5080         struct ext4_mount_options old_opts;
5081         int enable_quota = 0;
5082         ext4_group_t g;
5083         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5084         int err = 0;
5085 #ifdef CONFIG_QUOTA
5086         int i, j;
5087 #endif
5088         char *orig_data = kstrdup(data, GFP_KERNEL);
5089
5090         /* Store the original options */
5091         old_sb_flags = sb->s_flags;
5092         old_opts.s_mount_opt = sbi->s_mount_opt;
5093         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5094         old_opts.s_resuid = sbi->s_resuid;
5095         old_opts.s_resgid = sbi->s_resgid;
5096         old_opts.s_commit_interval = sbi->s_commit_interval;
5097         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5098         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5099 #ifdef CONFIG_QUOTA
5100         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5101         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5102                 if (sbi->s_qf_names[i]) {
5103                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
5104                                                          GFP_KERNEL);
5105                         if (!old_opts.s_qf_names[i]) {
5106                                 for (j = 0; j < i; j++)
5107                                         kfree(old_opts.s_qf_names[j]);
5108                                 kfree(orig_data);
5109                                 return -ENOMEM;
5110                         }
5111                 } else
5112                         old_opts.s_qf_names[i] = NULL;
5113 #endif
5114         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5115                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5116
5117         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5118                 err = -EINVAL;
5119                 goto restore_opts;
5120         }
5121
5122         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5123             test_opt(sb, JOURNAL_CHECKSUM)) {
5124                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5125                          "during remount not supported; ignoring");
5126                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5127         }
5128
5129         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5130                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5131                         ext4_msg(sb, KERN_ERR, "can't mount with "
5132                                  "both data=journal and delalloc");
5133                         err = -EINVAL;
5134                         goto restore_opts;
5135                 }
5136                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5137                         ext4_msg(sb, KERN_ERR, "can't mount with "
5138                                  "both data=journal and dioread_nolock");
5139                         err = -EINVAL;
5140                         goto restore_opts;
5141                 }
5142                 if (test_opt(sb, DAX)) {
5143                         ext4_msg(sb, KERN_ERR, "can't mount with "
5144                                  "both data=journal and dax");
5145                         err = -EINVAL;
5146                         goto restore_opts;
5147                 }
5148         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5149                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5150                         ext4_msg(sb, KERN_ERR, "can't mount with "
5151                                 "journal_async_commit in data=ordered mode");
5152                         err = -EINVAL;
5153                         goto restore_opts;
5154                 }
5155         }
5156
5157         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5158                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5159                 err = -EINVAL;
5160                 goto restore_opts;
5161         }
5162
5163         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5164                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5165                         "dax flag with busy inodes while remounting");
5166                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5167         }
5168
5169         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5170                 ext4_abort(sb, "Abort forced by user");
5171
5172         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5173                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5174
5175         es = sbi->s_es;
5176
5177         if (sbi->s_journal) {
5178                 ext4_init_journal_params(sb, sbi->s_journal);
5179                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5180         }
5181
5182         if (*flags & SB_LAZYTIME)
5183                 sb->s_flags |= SB_LAZYTIME;
5184
5185         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5186                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5187                         err = -EROFS;
5188                         goto restore_opts;
5189                 }
5190
5191                 if (*flags & SB_RDONLY) {
5192                         err = sync_filesystem(sb);
5193                         if (err < 0)
5194                                 goto restore_opts;
5195                         err = dquot_suspend(sb, -1);
5196                         if (err < 0)
5197                                 goto restore_opts;
5198
5199                         /*
5200                          * First of all, the unconditional stuff we have to do
5201                          * to disable replay of the journal when we next remount
5202                          */
5203                         sb->s_flags |= SB_RDONLY;
5204
5205                         /*
5206                          * OK, test if we are remounting a valid rw partition
5207                          * readonly, and if so set the rdonly flag and then
5208                          * mark the partition as valid again.
5209                          */
5210                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5211                             (sbi->s_mount_state & EXT4_VALID_FS))
5212                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5213
5214                         if (sbi->s_journal)
5215                                 ext4_mark_recovery_complete(sb, es);
5216                 } else {
5217                         /* Make sure we can mount this feature set readwrite */
5218                         if (ext4_has_feature_readonly(sb) ||
5219                             !ext4_feature_set_ok(sb, 0)) {
5220                                 err = -EROFS;
5221                                 goto restore_opts;
5222                         }
5223                         /*
5224                          * Make sure the group descriptor checksums
5225                          * are sane.  If they aren't, refuse to remount r/w.
5226                          */
5227                         for (g = 0; g < sbi->s_groups_count; g++) {
5228                                 struct ext4_group_desc *gdp =
5229                                         ext4_get_group_desc(sb, g, NULL);
5230
5231                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5232                                         ext4_msg(sb, KERN_ERR,
5233                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5234                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5235                                                le16_to_cpu(gdp->bg_checksum));
5236                                         err = -EFSBADCRC;
5237                                         goto restore_opts;
5238                                 }
5239                         }
5240
5241                         /*
5242                          * If we have an unprocessed orphan list hanging
5243                          * around from a previously readonly bdev mount,
5244                          * require a full umount/remount for now.
5245                          */
5246                         if (es->s_last_orphan) {
5247                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5248                                        "remount RDWR because of unprocessed "
5249                                        "orphan inode list.  Please "
5250                                        "umount/remount instead");
5251                                 err = -EINVAL;
5252                                 goto restore_opts;
5253                         }
5254
5255                         /*
5256                          * Mounting a RDONLY partition read-write, so reread
5257                          * and store the current valid flag.  (It may have
5258                          * been changed by e2fsck since we originally mounted
5259                          * the partition.)
5260                          */
5261                         if (sbi->s_journal)
5262                                 ext4_clear_journal_err(sb, es);
5263                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5264
5265                         err = ext4_setup_super(sb, es, 0);
5266                         if (err)
5267                                 goto restore_opts;
5268
5269                         sb->s_flags &= ~SB_RDONLY;
5270                         if (ext4_has_feature_mmp(sb))
5271                                 if (ext4_multi_mount_protect(sb,
5272                                                 le64_to_cpu(es->s_mmp_block))) {
5273                                         err = -EROFS;
5274                                         goto restore_opts;
5275                                 }
5276                         enable_quota = 1;
5277                 }
5278         }
5279
5280         /*
5281          * Reinitialize lazy itable initialization thread based on
5282          * current settings
5283          */
5284         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5285                 ext4_unregister_li_request(sb);
5286         else {
5287                 ext4_group_t first_not_zeroed;
5288                 first_not_zeroed = ext4_has_uninit_itable(sb);
5289                 ext4_register_li_request(sb, first_not_zeroed);
5290         }
5291
5292         ext4_setup_system_zone(sb);
5293         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5294                 err = ext4_commit_super(sb, 1);
5295                 if (err)
5296                         goto restore_opts;
5297         }
5298
5299 #ifdef CONFIG_QUOTA
5300         /* Release old quota file names */
5301         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5302                 kfree(old_opts.s_qf_names[i]);
5303         if (enable_quota) {
5304                 if (sb_any_quota_suspended(sb))
5305                         dquot_resume(sb, -1);
5306                 else if (ext4_has_feature_quota(sb)) {
5307                         err = ext4_enable_quotas(sb);
5308                         if (err)
5309                                 goto restore_opts;
5310                 }
5311         }
5312 #endif
5313
5314         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5315         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5316         kfree(orig_data);
5317         return 0;
5318
5319 restore_opts:
5320         sb->s_flags = old_sb_flags;
5321         sbi->s_mount_opt = old_opts.s_mount_opt;
5322         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5323         sbi->s_resuid = old_opts.s_resuid;
5324         sbi->s_resgid = old_opts.s_resgid;
5325         sbi->s_commit_interval = old_opts.s_commit_interval;
5326         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5327         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5328 #ifdef CONFIG_QUOTA
5329         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5330         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5331                 kfree(sbi->s_qf_names[i]);
5332                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5333         }
5334 #endif
5335         kfree(orig_data);
5336         return err;
5337 }
5338
5339 #ifdef CONFIG_QUOTA
5340 static int ext4_statfs_project(struct super_block *sb,
5341                                kprojid_t projid, struct kstatfs *buf)
5342 {
5343         struct kqid qid;
5344         struct dquot *dquot;
5345         u64 limit;
5346         u64 curblock;
5347
5348         qid = make_kqid_projid(projid);
5349         dquot = dqget(sb, qid);
5350         if (IS_ERR(dquot))
5351                 return PTR_ERR(dquot);
5352         spin_lock(&dquot->dq_dqb_lock);
5353
5354         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5355                  dquot->dq_dqb.dqb_bsoftlimit :
5356                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5357         if (limit && buf->f_blocks > limit) {
5358                 curblock = (dquot->dq_dqb.dqb_curspace +
5359                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5360                 buf->f_blocks = limit;
5361                 buf->f_bfree = buf->f_bavail =
5362                         (buf->f_blocks > curblock) ?
5363                          (buf->f_blocks - curblock) : 0;
5364         }
5365
5366         limit = dquot->dq_dqb.dqb_isoftlimit ?
5367                 dquot->dq_dqb.dqb_isoftlimit :
5368                 dquot->dq_dqb.dqb_ihardlimit;
5369         if (limit && buf->f_files > limit) {
5370                 buf->f_files = limit;
5371                 buf->f_ffree =
5372                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5373                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5374         }
5375
5376         spin_unlock(&dquot->dq_dqb_lock);
5377         dqput(dquot);
5378         return 0;
5379 }
5380 #endif
5381
5382 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5383 {
5384         struct super_block *sb = dentry->d_sb;
5385         struct ext4_sb_info *sbi = EXT4_SB(sb);
5386         struct ext4_super_block *es = sbi->s_es;
5387         ext4_fsblk_t overhead = 0, resv_blocks;
5388         u64 fsid;
5389         s64 bfree;
5390         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5391
5392         if (!test_opt(sb, MINIX_DF))
5393                 overhead = sbi->s_overhead;
5394
5395         buf->f_type = EXT4_SUPER_MAGIC;
5396         buf->f_bsize = sb->s_blocksize;
5397         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5398         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5399                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5400         /* prevent underflow in case that few free space is available */
5401         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5402         buf->f_bavail = buf->f_bfree -
5403                         (ext4_r_blocks_count(es) + resv_blocks);
5404         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5405                 buf->f_bavail = 0;
5406         buf->f_files = le32_to_cpu(es->s_inodes_count);
5407         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5408         buf->f_namelen = EXT4_NAME_LEN;
5409         fsid = le64_to_cpup((void *)es->s_uuid) ^
5410                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5411         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5412         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5413
5414 #ifdef CONFIG_QUOTA
5415         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5416             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5417                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5418 #endif
5419         return 0;
5420 }
5421
5422
5423 #ifdef CONFIG_QUOTA
5424
5425 /*
5426  * Helper functions so that transaction is started before we acquire dqio_sem
5427  * to keep correct lock ordering of transaction > dqio_sem
5428  */
5429 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5430 {
5431         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5432 }
5433
5434 static int ext4_write_dquot(struct dquot *dquot)
5435 {
5436         int ret, err;
5437         handle_t *handle;
5438         struct inode *inode;
5439
5440         inode = dquot_to_inode(dquot);
5441         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5442                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5443         if (IS_ERR(handle))
5444                 return PTR_ERR(handle);
5445         ret = dquot_commit(dquot);
5446         err = ext4_journal_stop(handle);
5447         if (!ret)
5448                 ret = err;
5449         return ret;
5450 }
5451
5452 static int ext4_acquire_dquot(struct dquot *dquot)
5453 {
5454         int ret, err;
5455         handle_t *handle;
5456
5457         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5458                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5459         if (IS_ERR(handle))
5460                 return PTR_ERR(handle);
5461         ret = dquot_acquire(dquot);
5462         err = ext4_journal_stop(handle);
5463         if (!ret)
5464                 ret = err;
5465         return ret;
5466 }
5467
5468 static int ext4_release_dquot(struct dquot *dquot)
5469 {
5470         int ret, err;
5471         handle_t *handle;
5472
5473         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5474                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5475         if (IS_ERR(handle)) {
5476                 /* Release dquot anyway to avoid endless cycle in dqput() */
5477                 dquot_release(dquot);
5478                 return PTR_ERR(handle);
5479         }
5480         ret = dquot_release(dquot);
5481         err = ext4_journal_stop(handle);
5482         if (!ret)
5483                 ret = err;
5484         return ret;
5485 }
5486
5487 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5488 {
5489         struct super_block *sb = dquot->dq_sb;
5490         struct ext4_sb_info *sbi = EXT4_SB(sb);
5491
5492         /* Are we journaling quotas? */
5493         if (ext4_has_feature_quota(sb) ||
5494             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5495                 dquot_mark_dquot_dirty(dquot);
5496                 return ext4_write_dquot(dquot);
5497         } else {
5498                 return dquot_mark_dquot_dirty(dquot);
5499         }
5500 }
5501
5502 static int ext4_write_info(struct super_block *sb, int type)
5503 {
5504         int ret, err;
5505         handle_t *handle;
5506
5507         /* Data block + inode block */
5508         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5509         if (IS_ERR(handle))
5510                 return PTR_ERR(handle);
5511         ret = dquot_commit_info(sb, type);
5512         err = ext4_journal_stop(handle);
5513         if (!ret)
5514                 ret = err;
5515         return ret;
5516 }
5517
5518 /*
5519  * Turn on quotas during mount time - we need to find
5520  * the quota file and such...
5521  */
5522 static int ext4_quota_on_mount(struct super_block *sb, int type)
5523 {
5524         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5525                                         EXT4_SB(sb)->s_jquota_fmt, type);
5526 }
5527
5528 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5529 {
5530         struct ext4_inode_info *ei = EXT4_I(inode);
5531
5532         /* The first argument of lockdep_set_subclass has to be
5533          * *exactly* the same as the argument to init_rwsem() --- in
5534          * this case, in init_once() --- or lockdep gets unhappy
5535          * because the name of the lock is set using the
5536          * stringification of the argument to init_rwsem().
5537          */
5538         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5539         lockdep_set_subclass(&ei->i_data_sem, subclass);
5540 }
5541
5542 /*
5543  * Standard function to be called on quota_on
5544  */
5545 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5546                          const struct path *path)
5547 {
5548         int err;
5549
5550         if (!test_opt(sb, QUOTA))
5551                 return -EINVAL;
5552
5553         /* Quotafile not on the same filesystem? */
5554         if (path->dentry->d_sb != sb)
5555                 return -EXDEV;
5556         /* Journaling quota? */
5557         if (EXT4_SB(sb)->s_qf_names[type]) {
5558                 /* Quotafile not in fs root? */
5559                 if (path->dentry->d_parent != sb->s_root)
5560                         ext4_msg(sb, KERN_WARNING,
5561                                 "Quota file not on filesystem root. "
5562                                 "Journaled quota will not work");
5563                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5564         } else {
5565                 /*
5566                  * Clear the flag just in case mount options changed since
5567                  * last time.
5568                  */
5569                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5570         }
5571
5572         /*
5573          * When we journal data on quota file, we have to flush journal to see
5574          * all updates to the file when we bypass pagecache...
5575          */
5576         if (EXT4_SB(sb)->s_journal &&
5577             ext4_should_journal_data(d_inode(path->dentry))) {
5578                 /*
5579                  * We don't need to lock updates but journal_flush() could
5580                  * otherwise be livelocked...
5581                  */
5582                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5583                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5584                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5585                 if (err)
5586                         return err;
5587         }
5588
5589         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5590         err = dquot_quota_on(sb, type, format_id, path);
5591         if (err) {
5592                 lockdep_set_quota_inode(path->dentry->d_inode,
5593                                              I_DATA_SEM_NORMAL);
5594         } else {
5595                 struct inode *inode = d_inode(path->dentry);
5596                 handle_t *handle;
5597
5598                 /*
5599                  * Set inode flags to prevent userspace from messing with quota
5600                  * files. If this fails, we return success anyway since quotas
5601                  * are already enabled and this is not a hard failure.
5602                  */
5603                 inode_lock(inode);
5604                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5605                 if (IS_ERR(handle))
5606                         goto unlock_inode;
5607                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5608                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5609                                 S_NOATIME | S_IMMUTABLE);
5610                 ext4_mark_inode_dirty(handle, inode);
5611                 ext4_journal_stop(handle);
5612         unlock_inode:
5613                 inode_unlock(inode);
5614         }
5615         return err;
5616 }
5617
5618 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5619                              unsigned int flags)
5620 {
5621         int err;
5622         struct inode *qf_inode;
5623         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5624                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5625                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5626                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5627         };
5628
5629         BUG_ON(!ext4_has_feature_quota(sb));
5630
5631         if (!qf_inums[type])
5632                 return -EPERM;
5633
5634         qf_inode = ext4_iget(sb, qf_inums[type]);
5635         if (IS_ERR(qf_inode)) {
5636                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5637                 return PTR_ERR(qf_inode);
5638         }
5639
5640         /* Don't account quota for quota files to avoid recursion */
5641         qf_inode->i_flags |= S_NOQUOTA;
5642         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5643         err = dquot_enable(qf_inode, type, format_id, flags);
5644         iput(qf_inode);
5645         if (err)
5646                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5647
5648         return err;
5649 }
5650
5651 /* Enable usage tracking for all quota types. */
5652 static int ext4_enable_quotas(struct super_block *sb)
5653 {
5654         int type, err = 0;
5655         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5656                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5657                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5658                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5659         };
5660         bool quota_mopt[EXT4_MAXQUOTAS] = {
5661                 test_opt(sb, USRQUOTA),
5662                 test_opt(sb, GRPQUOTA),
5663                 test_opt(sb, PRJQUOTA),
5664         };
5665
5666         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5667         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5668                 if (qf_inums[type]) {
5669                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5670                                 DQUOT_USAGE_ENABLED |
5671                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5672                         if (err) {
5673                                 for (type--; type >= 0; type--)
5674                                         dquot_quota_off(sb, type);
5675
5676                                 ext4_warning(sb,
5677                                         "Failed to enable quota tracking "
5678                                         "(type=%d, err=%d). Please run "
5679                                         "e2fsck to fix.", type, err);
5680                                 return err;
5681                         }
5682                 }
5683         }
5684         return 0;
5685 }
5686
5687 static int ext4_quota_off(struct super_block *sb, int type)
5688 {
5689         struct inode *inode = sb_dqopt(sb)->files[type];
5690         handle_t *handle;
5691         int err;
5692
5693         /* Force all delayed allocation blocks to be allocated.
5694          * Caller already holds s_umount sem */
5695         if (test_opt(sb, DELALLOC))
5696                 sync_filesystem(sb);
5697
5698         if (!inode || !igrab(inode))
5699                 goto out;
5700
5701         err = dquot_quota_off(sb, type);
5702         if (err || ext4_has_feature_quota(sb))
5703                 goto out_put;
5704
5705         inode_lock(inode);
5706         /*
5707          * Update modification times of quota files when userspace can
5708          * start looking at them. If we fail, we return success anyway since
5709          * this is not a hard failure and quotas are already disabled.
5710          */
5711         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5712         if (IS_ERR(handle))
5713                 goto out_unlock;
5714         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5715         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5716         inode->i_mtime = inode->i_ctime = current_time(inode);
5717         ext4_mark_inode_dirty(handle, inode);
5718         ext4_journal_stop(handle);
5719 out_unlock:
5720         inode_unlock(inode);
5721 out_put:
5722         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5723         iput(inode);
5724         return err;
5725 out:
5726         return dquot_quota_off(sb, type);
5727 }
5728
5729 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5730  * acquiring the locks... As quota files are never truncated and quota code
5731  * itself serializes the operations (and no one else should touch the files)
5732  * we don't have to be afraid of races */
5733 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5734                                size_t len, loff_t off)
5735 {
5736         struct inode *inode = sb_dqopt(sb)->files[type];
5737         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5738         int offset = off & (sb->s_blocksize - 1);
5739         int tocopy;
5740         size_t toread;
5741         struct buffer_head *bh;
5742         loff_t i_size = i_size_read(inode);
5743
5744         if (off > i_size)
5745                 return 0;
5746         if (off+len > i_size)
5747                 len = i_size-off;
5748         toread = len;
5749         while (toread > 0) {
5750                 tocopy = sb->s_blocksize - offset < toread ?
5751                                 sb->s_blocksize - offset : toread;
5752                 bh = ext4_bread(NULL, inode, blk, 0);
5753                 if (IS_ERR(bh))
5754                         return PTR_ERR(bh);
5755                 if (!bh)        /* A hole? */
5756                         memset(data, 0, tocopy);
5757                 else
5758                         memcpy(data, bh->b_data+offset, tocopy);
5759                 brelse(bh);
5760                 offset = 0;
5761                 toread -= tocopy;
5762                 data += tocopy;
5763                 blk++;
5764         }
5765         return len;
5766 }
5767
5768 /* Write to quotafile (we know the transaction is already started and has
5769  * enough credits) */
5770 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5771                                 const char *data, size_t len, loff_t off)
5772 {
5773         struct inode *inode = sb_dqopt(sb)->files[type];
5774         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5775         int err, offset = off & (sb->s_blocksize - 1);
5776         int retries = 0;
5777         struct buffer_head *bh;
5778         handle_t *handle = journal_current_handle();
5779
5780         if (EXT4_SB(sb)->s_journal && !handle) {
5781                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5782                         " cancelled because transaction is not started",
5783                         (unsigned long long)off, (unsigned long long)len);
5784                 return -EIO;
5785         }
5786         /*
5787          * Since we account only one data block in transaction credits,
5788          * then it is impossible to cross a block boundary.
5789          */
5790         if (sb->s_blocksize - offset < len) {
5791                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5792                         " cancelled because not block aligned",
5793                         (unsigned long long)off, (unsigned long long)len);
5794                 return -EIO;
5795         }
5796
5797         do {
5798                 bh = ext4_bread(handle, inode, blk,
5799                                 EXT4_GET_BLOCKS_CREATE |
5800                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5801         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5802                  ext4_should_retry_alloc(inode->i_sb, &retries));
5803         if (IS_ERR(bh))
5804                 return PTR_ERR(bh);
5805         if (!bh)
5806                 goto out;
5807         BUFFER_TRACE(bh, "get write access");
5808         err = ext4_journal_get_write_access(handle, bh);
5809         if (err) {
5810                 brelse(bh);
5811                 return err;
5812         }
5813         lock_buffer(bh);
5814         memcpy(bh->b_data+offset, data, len);
5815         flush_dcache_page(bh->b_page);
5816         unlock_buffer(bh);
5817         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5818         brelse(bh);
5819 out:
5820         if (inode->i_size < off + len) {
5821                 i_size_write(inode, off + len);
5822                 EXT4_I(inode)->i_disksize = inode->i_size;
5823                 ext4_mark_inode_dirty(handle, inode);
5824         }
5825         return len;
5826 }
5827
5828 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5829 {
5830         const struct quota_format_ops   *ops;
5831
5832         if (!sb_has_quota_loaded(sb, qid->type))
5833                 return -ESRCH;
5834         ops = sb_dqopt(sb)->ops[qid->type];
5835         if (!ops || !ops->get_next_id)
5836                 return -ENOSYS;
5837         return dquot_get_next_id(sb, qid);
5838 }
5839 #endif
5840
5841 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5842                        const char *dev_name, void *data)
5843 {
5844         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5845 }
5846
5847 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5848 static inline void register_as_ext2(void)
5849 {
5850         int err = register_filesystem(&ext2_fs_type);
5851         if (err)
5852                 printk(KERN_WARNING
5853                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5854 }
5855
5856 static inline void unregister_as_ext2(void)
5857 {
5858         unregister_filesystem(&ext2_fs_type);
5859 }
5860
5861 static inline int ext2_feature_set_ok(struct super_block *sb)
5862 {
5863         if (ext4_has_unknown_ext2_incompat_features(sb))
5864                 return 0;
5865         if (sb_rdonly(sb))
5866                 return 1;
5867         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5868                 return 0;
5869         return 1;
5870 }
5871 #else
5872 static inline void register_as_ext2(void) { }
5873 static inline void unregister_as_ext2(void) { }
5874 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5875 #endif
5876
5877 static inline void register_as_ext3(void)
5878 {
5879         int err = register_filesystem(&ext3_fs_type);
5880         if (err)
5881                 printk(KERN_WARNING
5882                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5883 }
5884
5885 static inline void unregister_as_ext3(void)
5886 {
5887         unregister_filesystem(&ext3_fs_type);
5888 }
5889
5890 static inline int ext3_feature_set_ok(struct super_block *sb)
5891 {
5892         if (ext4_has_unknown_ext3_incompat_features(sb))
5893                 return 0;
5894         if (!ext4_has_feature_journal(sb))
5895                 return 0;
5896         if (sb_rdonly(sb))
5897                 return 1;
5898         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5899                 return 0;
5900         return 1;
5901 }
5902
5903 static struct file_system_type ext4_fs_type = {
5904         .owner          = THIS_MODULE,
5905         .name           = "ext4",
5906         .mount          = ext4_mount,
5907         .kill_sb        = kill_block_super,
5908         .fs_flags       = FS_REQUIRES_DEV,
5909 };
5910 MODULE_ALIAS_FS("ext4");
5911
5912 /* Shared across all ext4 file systems */
5913 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5914
5915 static int __init ext4_init_fs(void)
5916 {
5917         int i, err;
5918
5919         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5920         ext4_li_info = NULL;
5921         mutex_init(&ext4_li_mtx);
5922
5923         /* Build-time check for flags consistency */
5924         ext4_check_flag_values();
5925
5926         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5927                 init_waitqueue_head(&ext4__ioend_wq[i]);
5928
5929         err = ext4_init_es();
5930         if (err)
5931                 return err;
5932
5933         err = ext4_init_pageio();
5934         if (err)
5935                 goto out5;
5936
5937         err = ext4_init_system_zone();
5938         if (err)
5939                 goto out4;
5940
5941         err = ext4_init_sysfs();
5942         if (err)
5943                 goto out3;
5944
5945         err = ext4_init_mballoc();
5946         if (err)
5947                 goto out2;
5948         err = init_inodecache();
5949         if (err)
5950                 goto out1;
5951         register_as_ext3();
5952         register_as_ext2();
5953         err = register_filesystem(&ext4_fs_type);
5954         if (err)
5955                 goto out;
5956
5957         return 0;
5958 out:
5959         unregister_as_ext2();
5960         unregister_as_ext3();
5961         destroy_inodecache();
5962 out1:
5963         ext4_exit_mballoc();
5964 out2:
5965         ext4_exit_sysfs();
5966 out3:
5967         ext4_exit_system_zone();
5968 out4:
5969         ext4_exit_pageio();
5970 out5:
5971         ext4_exit_es();
5972
5973         return err;
5974 }
5975
5976 static void __exit ext4_exit_fs(void)
5977 {
5978         ext4_destroy_lazyinit_thread();
5979         unregister_as_ext2();
5980         unregister_as_ext3();
5981         unregister_filesystem(&ext4_fs_type);
5982         destroy_inodecache();
5983         ext4_exit_mballoc();
5984         ext4_exit_sysfs();
5985         ext4_exit_system_zone();
5986         ext4_exit_pageio();
5987         ext4_exit_es();
5988 }
5989
5990 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5991 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5992 MODULE_LICENSE("GPL");
5993 MODULE_SOFTDEP("pre: crc32c");
5994 module_init(ext4_init_fs)
5995 module_exit(ext4_exit_fs)