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 (sb_rdonly(sb))
 409                 return;
 410
 411         if (!test_opt(sb, ERRORS_CONT)) {
 412                 journal_t *journal = EXT4_SB(sb)->s_journal;
 413
 414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 415                 if (journal)
 416                         jbd2_journal_abort(journal, -EIO);
 417         }
 418         if (test_opt(sb, ERRORS_RO)) {
 419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 420                 /*
 421                  * Make sure updated value of ->s_mount_flags will be visible
 422                  * before ->s_flags update
 423                  */
 424                 smp_wmb();
 425                 sb->s_flags |= SB_RDONLY;
 426         }
 427         if (test_opt(sb, ERRORS_PANIC)) {
 428                 if (EXT4_SB(sb)->s_journal &&
 429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 430                         return;
 431                 panic("EXT4-fs (device %s): panic forced after error\n",
 432                         sb->s_id);
 433         }
 434 }
 435
 436 #define ext4_error_ratelimit(sb)                                        \
 437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 438                              "EXT4-fs error")
 439
 440 void __ext4_error(struct super_block *sb, const char *function,
 441                   unsigned int line, const char *fmt, ...)
 442 {
 443         struct va_format vaf;
 444         va_list args;
 445
 446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 447                 return;
 448
 449         trace_ext4_error(sb, function, line);
 450         if (ext4_error_ratelimit(sb)) {
 451                 va_start(args, fmt);
 452                 vaf.fmt = fmt;
 453                 vaf.va = &args;
 454                 printk(KERN_CRIT
 455                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 456                        sb->s_id, function, line, current->comm, &vaf);
 457                 va_end(args);
 458         }
 459         save_error_info(sb, function, line);
 460         ext4_handle_error(sb);
 461 }
 462
 463 void __ext4_error_inode(struct inode *inode, const char *function,
 464                         unsigned int line, ext4_fsblk_t block,
 465                         const char *fmt, ...)
 466 {
 467         va_list args;
 468         struct va_format vaf;
 469         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 470
 471         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 472                 return;
 473
 474         trace_ext4_error(inode->i_sb, function, line);
 475         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 476         es->s_last_error_block = cpu_to_le64(block);
 477         if (ext4_error_ratelimit(inode->i_sb)) {
 478                 va_start(args, fmt);
 479                 vaf.fmt = fmt;
 480                 vaf.va = &args;
 481                 if (block)
 482                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 483                                "inode #%lu: block %llu: comm %s: %pV\n",
 484                                inode->i_sb->s_id, function, line, inode->i_ino,
 485                                block, current->comm, &vaf);
 486                 else
 487                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 488                                "inode #%lu: comm %s: %pV\n",
 489                                inode->i_sb->s_id, function, line, inode->i_ino,
 490                                current->comm, &vaf);
 491                 va_end(args);
 492         }
 493         save_error_info(inode->i_sb, function, line);
 494         ext4_handle_error(inode->i_sb);
 495 }
 496
 497 void __ext4_error_file(struct file *file, const char *function,
 498                        unsigned int line, ext4_fsblk_t block,
 499                        const char *fmt, ...)
 500 {
 501         va_list args;
 502         struct va_format vaf;
 503         struct ext4_super_block *es;
 504         struct inode *inode = file_inode(file);
 505         char pathname[80], *path;
 506
 507         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 508                 return;
 509
 510         trace_ext4_error(inode->i_sb, function, line);
 511         es = EXT4_SB(inode->i_sb)->s_es;
 512         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 513         if (ext4_error_ratelimit(inode->i_sb)) {
 514                 path = file_path(file, pathname, sizeof(pathname));
 515                 if (IS_ERR(path))
 516                         path = "(unknown)";
 517                 va_start(args, fmt);
 518                 vaf.fmt = fmt;
 519                 vaf.va = &args;
 520                 if (block)
 521                         printk(KERN_CRIT
 522                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 523                                "block %llu: comm %s: path %s: %pV\n",
 524                                inode->i_sb->s_id, function, line, inode->i_ino,
 525                                block, current->comm, path, &vaf);
 526                 else
 527                         printk(KERN_CRIT
 528                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 529                                "comm %s: path %s: %pV\n",
 530                                inode->i_sb->s_id, function, line, inode->i_ino,
 531                                current->comm, path, &vaf);
 532                 va_end(args);
 533         }
 534         save_error_info(inode->i_sb, function, line);
 535         ext4_handle_error(inode->i_sb);
 536 }
 537
 538 const char *ext4_decode_error(struct super_block *sb, int errno,
 539                               char nbuf[16])
 540 {
 541         char *errstr = NULL;
 542
 543         switch (errno) {
 544         case -EFSCORRUPTED:
 545                 errstr = "Corrupt filesystem";
 546                 break;
 547         case -EFSBADCRC:
 548                 errstr = "Filesystem failed CRC";
 549                 break;
 550         case -EIO:
 551                 errstr = "IO failure";
 552                 break;
 553         case -ENOMEM:
 554                 errstr = "Out of memory";
 555                 break;
 556         case -EROFS:
 557                 if (!sb || (EXT4_SB(sb)->s_journal &&
 558                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 559                         errstr = "Journal has aborted";
 560                 else
 561                         errstr = "Readonly filesystem";
 562                 break;
 563         default:
 564                 /* If the caller passed in an extra buffer for unknown
 565                  * errors, textualise them now.  Else we just return
 566                  * NULL. */
 567                 if (nbuf) {
 568                         /* Check for truncated error codes... */
 569                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 570                                 errstr = nbuf;
 571                 }
 572                 break;
 573         }
 574
 575         return errstr;
 576 }
 577
 578 /* __ext4_std_error decodes expected errors from journaling functions
 579  * automatically and invokes the appropriate error response.  */
 580
 581 void __ext4_std_error(struct super_block *sb, const char *function,
 582                       unsigned int line, int errno)
 583 {
 584         char nbuf[16];
 585         const char *errstr;
 586
 587         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 588                 return;
 589
 590         /* Special case: if the error is EROFS, and we're not already
 591          * inside a transaction, then there's really no point in logging
 592          * an error. */
 593         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 594                 return;
 595
 596         if (ext4_error_ratelimit(sb)) {
 597                 errstr = ext4_decode_error(sb, errno, nbuf);
 598                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 599                        sb->s_id, function, line, errstr);
 600         }
 601
 602         save_error_info(sb, function, line);
 603         ext4_handle_error(sb);
 604 }
 605
 606 /*
 607  * ext4_abort is a much stronger failure handler than ext4_error.  The
 608  * abort function may be used to deal with unrecoverable failures such
 609  * as journal IO errors or ENOMEM at a critical moment in log management.
 610  *
 611  * We unconditionally force the filesystem into an ABORT|READONLY state,
 612  * unless the error response on the fs has been set to panic in which
 613  * case we take the easy way out and panic immediately.
 614  */
 615
 616 void __ext4_abort(struct super_block *sb, const char *function,
 617                 unsigned int line, const char *fmt, ...)
 618 {
 619         struct va_format vaf;
 620         va_list args;
 621
 622         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 623                 return;
 624
 625         save_error_info(sb, function, line);
 626         va_start(args, fmt);
 627         vaf.fmt = fmt;
 628         vaf.va = &args;
 629         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 630                sb->s_id, function, line, &vaf);
 631         va_end(args);
 632
 633         if (sb_rdonly(sb) == 0) {
 634                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 635                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 636                 /*
 637                  * Make sure updated value of ->s_mount_flags will be visible
 638                  * before ->s_flags update
 639                  */
 640                 smp_wmb();
 641                 sb->s_flags |= SB_RDONLY;
 642                 if (EXT4_SB(sb)->s_journal)
 643                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 644                 save_error_info(sb, function, line);
 645         }
 646         if (test_opt(sb, ERRORS_PANIC)) {
 647                 if (EXT4_SB(sb)->s_journal &&
 648                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 649                         return;
 650                 panic("EXT4-fs panic from previous error\n");
 651         }
 652 }
 653
 654 void __ext4_msg(struct super_block *sb,
 655                 const char *prefix, const char *fmt, ...)
 656 {
 657         struct va_format vaf;
 658         va_list args;
 659
 660         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 661                 return;
 662
 663         va_start(args, fmt);
 664         vaf.fmt = fmt;
 665         vaf.va = &args;
 666         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 667         va_end(args);
 668 }
 669
 670 #define ext4_warning_ratelimit(sb)                                      \
 671                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 672                              "EXT4-fs warning")
 673
 674 void __ext4_warning(struct super_block *sb, const char *function,
 675                     unsigned int line, const char *fmt, ...)
 676 {
 677         struct va_format vaf;
 678         va_list args;
 679
 680         if (!ext4_warning_ratelimit(sb))
 681                 return;
 682
 683         va_start(args, fmt);
 684         vaf.fmt = fmt;
 685         vaf.va = &args;
 686         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 687                sb->s_id, function, line, &vaf);
 688         va_end(args);
 689 }
 690
 691 void __ext4_warning_inode(const struct inode *inode, const char *function,
 692                           unsigned int line, const char *fmt, ...)
 693 {
 694         struct va_format vaf;
 695         va_list args;
 696
 697         if (!ext4_warning_ratelimit(inode->i_sb))
 698                 return;
 699
 700         va_start(args, fmt);
 701         vaf.fmt = fmt;
 702         vaf.va = &args;
 703         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 704                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 705                function, line, inode->i_ino, current->comm, &vaf);
 706         va_end(args);
 707 }
 708
 709 void __ext4_grp_locked_error(const char *function, unsigned int line,
 710                              struct super_block *sb, ext4_group_t grp,
 711                              unsigned long ino, ext4_fsblk_t block,
 712                              const char *fmt, ...)
 713 __releases(bitlock)
 714 __acquires(bitlock)
 715 {
 716         struct va_format vaf;
 717         va_list args;
 718         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 719
 720         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 721                 return;
 722
 723         trace_ext4_error(sb, function, line);
 724         es->s_last_error_ino = cpu_to_le32(ino);
 725         es->s_last_error_block = cpu_to_le64(block);
 726         __save_error_info(sb, function, line);
 727
 728         if (ext4_error_ratelimit(sb)) {
 729                 va_start(args, fmt);
 730                 vaf.fmt = fmt;
 731                 vaf.va = &args;
 732                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 733                        sb->s_id, function, line, grp);
 734                 if (ino)
 735                         printk(KERN_CONT "inode %lu: ", ino);
 736                 if (block)
 737                         printk(KERN_CONT "block %llu:",
 738                                (unsigned long long) block);
 739                 printk(KERN_CONT "%pV\n", &vaf);
 740                 va_end(args);
 741         }
 742
 743         if (test_opt(sb, ERRORS_CONT)) {
 744                 ext4_commit_super(sb, 0);
 745                 return;
 746         }
 747
 748         ext4_unlock_group(sb, grp);
 749         ext4_commit_super(sb, 1);
 750         ext4_handle_error(sb);
 751         /*
 752          * We only get here in the ERRORS_RO case; relocking the group
 753          * may be dangerous, but nothing bad will happen since the
 754          * filesystem will have already been marked read/only and the
 755          * journal has been aborted.  We return 1 as a hint to callers
 756          * who might what to use the return value from
 757          * ext4_grp_locked_error() to distinguish between the
 758          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 759          * aggressively from the ext4 function in question, with a
 760          * more appropriate error code.
 761          */
 762         ext4_lock_group(sb, grp);
 763         return;
 764 }
 765
 766 void ext4_update_dynamic_rev(struct super_block *sb)
 767 {
 768         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 769
 770         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 771                 return;
 772
 773         ext4_warning(sb,
 774                      "updating to rev %d because of new feature flag, "
 775                      "running e2fsck is recommended",
 776                      EXT4_DYNAMIC_REV);
 777
 778         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 779         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 780         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 781         /* leave es->s_feature_*compat flags alone */
 782         /* es->s_uuid will be set by e2fsck if empty */
 783
 784         /*
 785          * The rest of the superblock fields should be zero, and if not it
 786          * means they are likely already in use, so leave them alone.  We
 787          * can leave it up to e2fsck to clean up any inconsistencies there.
 788          */
 789 }
 790
 791 /*
 792  * Open the external journal device
 793  */
 794 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 795 {
 796         struct block_device *bdev;
 797         char b[BDEVNAME_SIZE];
 798
 799         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 800         if (IS_ERR(bdev))
 801                 goto fail;
 802         return bdev;
 803
 804 fail:
 805         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 806                         __bdevname(dev, b), PTR_ERR(bdev));
 807         return NULL;
 808 }
 809
 810 /*
 811  * Release the journal device
 812  */
 813 static void ext4_blkdev_put(struct block_device *bdev)
 814 {
 815         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 816 }
 817
 818 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 819 {
 820         struct block_device *bdev;
 821         bdev = sbi->journal_bdev;
 822         if (bdev) {
 823                 ext4_blkdev_put(bdev);
 824                 sbi->journal_bdev = NULL;
 825         }
 826 }
 827
 828 static inline struct inode *orphan_list_entry(struct list_head *l)
 829 {
 830         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 831 }
 832
 833 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 834 {
 835         struct list_head *l;
 836
 837         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 838                  le32_to_cpu(sbi->s_es->s_last_orphan));
 839
 840         printk(KERN_ERR "sb_info orphan list:\n");
 841         list_for_each(l, &sbi->s_orphan) {
 842                 struct inode *inode = orphan_list_entry(l);
 843                 printk(KERN_ERR "  "
 844                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 845                        inode->i_sb->s_id, inode->i_ino, inode,
 846                        inode->i_mode, inode->i_nlink,
 847                        NEXT_ORPHAN(inode));
 848         }
 849 }
 850
 851 #ifdef CONFIG_QUOTA
 852 static int ext4_quota_off(struct super_block *sb, int type);
 853
 854 static inline void ext4_quota_off_umount(struct super_block *sb)
 855 {
 856         int type;
 857
 858         /* Use our quota_off function to clear inode flags etc. */
 859         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 860                 ext4_quota_off(sb, type);
 861 }
 862 #else
 863 static inline void ext4_quota_off_umount(struct super_block *sb)
 864 {
 865 }
 866 #endif
 867
 868 static void ext4_put_super(struct super_block *sb)
 869 {
 870         struct ext4_sb_info *sbi = EXT4_SB(sb);
 871         struct ext4_super_block *es = sbi->s_es;
 872         int aborted = 0;
 873         int i, err;
 874
 875         ext4_unregister_li_request(sb);
 876         ext4_quota_off_umount(sb);
 877
 878         destroy_workqueue(sbi->rsv_conversion_wq);
 879
 880         if (sbi->s_journal) {
 881                 aborted = is_journal_aborted(sbi->s_journal);
 882                 err = jbd2_journal_destroy(sbi->s_journal);
 883                 sbi->s_journal = NULL;
 884                 if ((err < 0) && !aborted)
 885                         ext4_abort(sb, "Couldn't clean up the journal");
 886         }
 887
 888         ext4_unregister_sysfs(sb);
 889         ext4_es_unregister_shrinker(sbi);
 890         del_timer_sync(&sbi->s_err_report);
 891         ext4_release_system_zone(sb);
 892         ext4_mb_release(sb);
 893         ext4_ext_release(sb);
 894
 895         if (!sb_rdonly(sb) && !aborted) {
 896                 ext4_clear_feature_journal_needs_recovery(sb);
 897                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 898         }
 899         if (!sb_rdonly(sb))
 900                 ext4_commit_super(sb, 1);
 901
 902         for (i = 0; i < sbi->s_gdb_count; i++)
 903                 brelse(sbi->s_group_desc[i]);
 904         kvfree(sbi->s_group_desc);
 905         kvfree(sbi->s_flex_groups);
 906         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 907         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 908         percpu_counter_destroy(&sbi->s_dirs_counter);
 909         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 910         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 911 #ifdef CONFIG_QUOTA
 912         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 913                 kfree(sbi->s_qf_names[i]);
 914 #endif
 915
 916         /* Debugging code just in case the in-memory inode orphan list
 917          * isn't empty.  The on-disk one can be non-empty if we've
 918          * detected an error and taken the fs readonly, but the
 919          * in-memory list had better be clean by this point. */
 920         if (!list_empty(&sbi->s_orphan))
 921                 dump_orphan_list(sb, sbi);
 922         J_ASSERT(list_empty(&sbi->s_orphan));
 923
 924         sync_blockdev(sb->s_bdev);
 925         invalidate_bdev(sb->s_bdev);
 926         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 927                 /*
 928                  * Invalidate the journal device's buffers.  We don't want them
 929                  * floating about in memory - the physical journal device may
 930                  * hotswapped, and it breaks the `ro-after' testing code.
 931                  */
 932                 sync_blockdev(sbi->journal_bdev);
 933                 invalidate_bdev(sbi->journal_bdev);
 934                 ext4_blkdev_remove(sbi);
 935         }
 936         if (sbi->s_ea_inode_cache) {
 937                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 938                 sbi->s_ea_inode_cache = NULL;
 939         }
 940         if (sbi->s_ea_block_cache) {
 941                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 942                 sbi->s_ea_block_cache = NULL;
 943         }
 944         if (sbi->s_mmp_tsk)
 945                 kthread_stop(sbi->s_mmp_tsk);
 946         brelse(sbi->s_sbh);
 947         sb->s_fs_info = NULL;
 948         /*
 949          * Now that we are completely done shutting down the
 950          * superblock, we need to actually destroy the kobject.
 951          */
 952         kobject_put(&sbi->s_kobj);
 953         wait_for_completion(&sbi->s_kobj_unregister);
 954         if (sbi->s_chksum_driver)
 955                 crypto_free_shash(sbi->s_chksum_driver);
 956         kfree(sbi->s_blockgroup_lock);
 957         fs_put_dax(sbi->s_daxdev);
 958         kfree(sbi);
 959 }
 960
 961 static struct kmem_cache *ext4_inode_cachep;
 962
 963 /*
 964  * Called inside transaction, so use GFP_NOFS
 965  */
 966 static struct inode *ext4_alloc_inode(struct super_block *sb)
 967 {
 968         struct ext4_inode_info *ei;
 969
 970         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 971         if (!ei)
 972                 return NULL;
 973
 974         inode_set_iversion(&ei->vfs_inode, 1);
 975         spin_lock_init(&ei->i_raw_lock);
 976         INIT_LIST_HEAD(&ei->i_prealloc_list);
 977         spin_lock_init(&ei->i_prealloc_lock);
 978         ext4_es_init_tree(&ei->i_es_tree);
 979         rwlock_init(&ei->i_es_lock);
 980         INIT_LIST_HEAD(&ei->i_es_list);
 981         ei->i_es_all_nr = 0;
 982         ei->i_es_shk_nr = 0;
 983         ei->i_es_shrink_lblk = 0;
 984         ei->i_reserved_data_blocks = 0;
 985         ei->i_da_metadata_calc_len = 0;
 986         ei->i_da_metadata_calc_last_lblock = 0;
 987         spin_lock_init(&(ei->i_block_reservation_lock));
 988 #ifdef CONFIG_QUOTA
 989         ei->i_reserved_quota = 0;
 990         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 991 #endif
 992         ei->jinode = NULL;
 993         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 994         spin_lock_init(&ei->i_completed_io_lock);
 995         ei->i_sync_tid = 0;
 996         ei->i_datasync_tid = 0;
 997         atomic_set(&ei->i_unwritten, 0);
 998         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 999         return &ei->vfs_inode;
1000 }
1001
1002 static int ext4_drop_inode(struct inode *inode)
1003 {
1004         int drop = generic_drop_inode(inode);
1005
1006         trace_ext4_drop_inode(inode, drop);
1007         return drop;
1008 }
1009
1010 static void ext4_i_callback(struct rcu_head *head)
1011 {
1012         struct inode *inode = container_of(head, struct inode, i_rcu);
1013         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1014 }
1015
1016 static void ext4_destroy_inode(struct inode *inode)
1017 {
1018         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1019                 ext4_msg(inode->i_sb, KERN_ERR,
1020                          "Inode %lu (%p): orphan list check failed!",
1021                          inode->i_ino, EXT4_I(inode));
1022                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1023                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1024                                 true);
1025                 dump_stack();
1026         }
1027         call_rcu(&inode->i_rcu, ext4_i_callback);
1028 }
1029
1030 static void init_once(void *foo)
1031 {
1032         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1033
1034         INIT_LIST_HEAD(&ei->i_orphan);
1035         init_rwsem(&ei->xattr_sem);
1036         init_rwsem(&ei->i_data_sem);
1037         init_rwsem(&ei->i_mmap_sem);
1038         inode_init_once(&ei->vfs_inode);
1039 }
1040
1041 static int __init init_inodecache(void)
1042 {
1043         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1044                                 sizeof(struct ext4_inode_info), 0,
1045                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1046                                         SLAB_ACCOUNT),
1047                                 offsetof(struct ext4_inode_info, i_data),
1048                                 sizeof_field(struct ext4_inode_info, i_data),
1049                                 init_once);
1050         if (ext4_inode_cachep == NULL)
1051                 return -ENOMEM;
1052         return 0;
1053 }
1054
1055 static void destroy_inodecache(void)
1056 {
1057         /*
1058          * Make sure all delayed rcu free inodes are flushed before we
1059          * destroy cache.
1060          */
1061         rcu_barrier();
1062         kmem_cache_destroy(ext4_inode_cachep);
1063 }
1064
1065 void ext4_clear_inode(struct inode *inode)
1066 {
1067         invalidate_inode_buffers(inode);
1068         clear_inode(inode);
1069         dquot_drop(inode);
1070         ext4_discard_preallocations(inode);
1071         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1072         if (EXT4_I(inode)->jinode) {
1073                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1074                                                EXT4_I(inode)->jinode);
1075                 jbd2_free_inode(EXT4_I(inode)->jinode);
1076                 EXT4_I(inode)->jinode = NULL;
1077         }
1078         fscrypt_put_encryption_info(inode);
1079 }
1080
1081 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1082                                         u64 ino, u32 generation)
1083 {
1084         struct inode *inode;
1085
1086         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1087                 return ERR_PTR(-ESTALE);
1088         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1089                 return ERR_PTR(-ESTALE);
1090
1091         /* iget isn't really right if the inode is currently unallocated!!
1092          *
1093          * ext4_read_inode will return a bad_inode if the inode had been
1094          * deleted, so we should be safe.
1095          *
1096          * Currently we don't know the generation for parent directory, so
1097          * a generation of 0 means "accept any"
1098          */
1099         inode = ext4_iget_normal(sb, ino);
1100         if (IS_ERR(inode))
1101                 return ERR_CAST(inode);
1102         if (generation && inode->i_generation != generation) {
1103                 iput(inode);
1104                 return ERR_PTR(-ESTALE);
1105         }
1106
1107         return inode;
1108 }
1109
1110 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1111                                         int fh_len, int fh_type)
1112 {
1113         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1114                                     ext4_nfs_get_inode);
1115 }
1116
1117 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1118                                         int fh_len, int fh_type)
1119 {
1120         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1121                                     ext4_nfs_get_inode);
1122 }
1123
1124 /*
1125  * Try to release metadata pages (indirect blocks, directories) which are
1126  * mapped via the block device.  Since these pages could have journal heads
1127  * which would prevent try_to_free_buffers() from freeing them, we must use
1128  * jbd2 layer's try_to_free_buffers() function to release them.
1129  */
1130 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1131                                  gfp_t wait)
1132 {
1133         journal_t *journal = EXT4_SB(sb)->s_journal;
1134
1135         WARN_ON(PageChecked(page));
1136         if (!page_has_buffers(page))
1137                 return 0;
1138         if (journal)
1139                 return jbd2_journal_try_to_free_buffers(journal, page,
1140                                                 wait & ~__GFP_DIRECT_RECLAIM);
1141         return try_to_free_buffers(page);
1142 }
1143
1144 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1145 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1146 {
1147         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1148                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1149 }
1150
1151 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1152                                                         void *fs_data)
1153 {
1154         handle_t *handle = fs_data;
1155         int res, res2, credits, retries = 0;
1156
1157         /*
1158          * Encrypting the root directory is not allowed because e2fsck expects
1159          * lost+found to exist and be unencrypted, and encrypting the root
1160          * directory would imply encrypting the lost+found directory as well as
1161          * the filename "lost+found" itself.
1162          */
1163         if (inode->i_ino == EXT4_ROOT_INO)
1164                 return -EPERM;
1165
1166         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1167                 return -EINVAL;
1168
1169         res = ext4_convert_inline_data(inode);
1170         if (res)
1171                 return res;
1172
1173         /*
1174          * If a journal handle was specified, then the encryption context is
1175          * being set on a new inode via inheritance and is part of a larger
1176          * transaction to create the inode.  Otherwise the encryption context is
1177          * being set on an existing inode in its own transaction.  Only in the
1178          * latter case should the "retry on ENOSPC" logic be used.
1179          */
1180
1181         if (handle) {
1182                 res = ext4_xattr_set_handle(handle, inode,
1183                                             EXT4_XATTR_INDEX_ENCRYPTION,
1184                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1185                                             ctx, len, 0);
1186                 if (!res) {
1187                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1188                         ext4_clear_inode_state(inode,
1189                                         EXT4_STATE_MAY_INLINE_DATA);
1190                         /*
1191                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1192                          * S_DAX may be disabled
1193                          */
1194                         ext4_set_inode_flags(inode);
1195                 }
1196                 return res;
1197         }
1198
1199         res = dquot_initialize(inode);
1200         if (res)
1201                 return res;
1202 retry:
1203         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1204                                      &credits);
1205         if (res)
1206                 return res;
1207
1208         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1209         if (IS_ERR(handle))
1210                 return PTR_ERR(handle);
1211
1212         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1213                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1214                                     ctx, len, 0);
1215         if (!res) {
1216                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1217                 /*
1218                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1219                  * S_DAX may be disabled
1220                  */
1221                 ext4_set_inode_flags(inode);
1222                 res = ext4_mark_inode_dirty(handle, inode);
1223                 if (res)
1224                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1225         }
1226         res2 = ext4_journal_stop(handle);
1227
1228         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1229                 goto retry;
1230         if (!res)
1231                 res = res2;
1232         return res;
1233 }
1234
1235 static bool ext4_dummy_context(struct inode *inode)
1236 {
1237         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1238 }
1239
1240 static unsigned ext4_max_namelen(struct inode *inode)
1241 {
1242         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1243                 EXT4_NAME_LEN;
1244 }
1245
1246 static const struct fscrypt_operations ext4_cryptops = {
1247         .key_prefix             = "ext4:",
1248         .get_context            = ext4_get_context,
1249         .set_context            = ext4_set_context,
1250         .dummy_context          = ext4_dummy_context,
1251         .empty_dir              = ext4_empty_dir,
1252         .max_namelen            = ext4_max_namelen,
1253 };
1254 #endif
1255
1256 #ifdef CONFIG_QUOTA
1257 static const char * const quotatypes[] = INITQFNAMES;
1258 #define QTYPE2NAME(t) (quotatypes[t])
1259
1260 static int ext4_write_dquot(struct dquot *dquot);
1261 static int ext4_acquire_dquot(struct dquot *dquot);
1262 static int ext4_release_dquot(struct dquot *dquot);
1263 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1264 static int ext4_write_info(struct super_block *sb, int type);
1265 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1266                          const struct path *path);
1267 static int ext4_quota_on_mount(struct super_block *sb, int type);
1268 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1269                                size_t len, loff_t off);
1270 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1271                                 const char *data, size_t len, loff_t off);
1272 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1273                              unsigned int flags);
1274 static int ext4_enable_quotas(struct super_block *sb);
1275 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1276
1277 static struct dquot **ext4_get_dquots(struct inode *inode)
1278 {
1279         return EXT4_I(inode)->i_dquot;
1280 }
1281
1282 static const struct dquot_operations ext4_quota_operations = {
1283         .get_reserved_space     = ext4_get_reserved_space,
1284         .write_dquot            = ext4_write_dquot,
1285         .acquire_dquot          = ext4_acquire_dquot,
1286         .release_dquot          = ext4_release_dquot,
1287         .mark_dirty             = ext4_mark_dquot_dirty,
1288         .write_info             = ext4_write_info,
1289         .alloc_dquot            = dquot_alloc,
1290         .destroy_dquot          = dquot_destroy,
1291         .get_projid             = ext4_get_projid,
1292         .get_inode_usage        = ext4_get_inode_usage,
1293         .get_next_id            = ext4_get_next_id,
1294 };
1295
1296 static const struct quotactl_ops ext4_qctl_operations = {
1297         .quota_on       = ext4_quota_on,
1298         .quota_off      = ext4_quota_off,
1299         .quota_sync     = dquot_quota_sync,
1300         .get_state      = dquot_get_state,
1301         .set_info       = dquot_set_dqinfo,
1302         .get_dqblk      = dquot_get_dqblk,
1303         .set_dqblk      = dquot_set_dqblk,
1304         .get_nextdqblk  = dquot_get_next_dqblk,
1305 };
1306 #endif
1307
1308 static const struct super_operations ext4_sops = {
1309         .alloc_inode    = ext4_alloc_inode,
1310         .destroy_inode  = ext4_destroy_inode,
1311         .write_inode    = ext4_write_inode,
1312         .dirty_inode    = ext4_dirty_inode,
1313         .drop_inode     = ext4_drop_inode,
1314         .evict_inode    = ext4_evict_inode,
1315         .put_super      = ext4_put_super,
1316         .sync_fs        = ext4_sync_fs,
1317         .freeze_fs      = ext4_freeze,
1318         .unfreeze_fs    = ext4_unfreeze,
1319         .statfs         = ext4_statfs,
1320         .remount_fs     = ext4_remount,
1321         .show_options   = ext4_show_options,
1322 #ifdef CONFIG_QUOTA
1323         .quota_read     = ext4_quota_read,
1324         .quota_write    = ext4_quota_write,
1325         .get_dquots     = ext4_get_dquots,
1326 #endif
1327         .bdev_try_to_free_page = bdev_try_to_free_page,
1328 };
1329
1330 static const struct export_operations ext4_export_ops = {
1331         .fh_to_dentry = ext4_fh_to_dentry,
1332         .fh_to_parent = ext4_fh_to_parent,
1333         .get_parent = ext4_get_parent,
1334 };
1335
1336 enum {
1337         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1338         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1339         Opt_nouid32, Opt_debug, Opt_removed,
1340         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1341         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1342         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1343         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1344         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1345         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1346         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1347         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1348         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1349         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1350         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1351         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1352         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1353         Opt_inode_readahead_blks, Opt_journal_ioprio,
1354         Opt_dioread_nolock, Opt_dioread_lock,
1355         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1356         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1357 };
1358
1359 static const match_table_t tokens = {
1360         {Opt_bsd_df, "bsddf"},
1361         {Opt_minix_df, "minixdf"},
1362         {Opt_grpid, "grpid"},
1363         {Opt_grpid, "bsdgroups"},
1364         {Opt_nogrpid, "nogrpid"},
1365         {Opt_nogrpid, "sysvgroups"},
1366         {Opt_resgid, "resgid=%u"},
1367         {Opt_resuid, "resuid=%u"},
1368         {Opt_sb, "sb=%u"},
1369         {Opt_err_cont, "errors=continue"},
1370         {Opt_err_panic, "errors=panic"},
1371         {Opt_err_ro, "errors=remount-ro"},
1372         {Opt_nouid32, "nouid32"},
1373         {Opt_debug, "debug"},
1374         {Opt_removed, "oldalloc"},
1375         {Opt_removed, "orlov"},
1376         {Opt_user_xattr, "user_xattr"},
1377         {Opt_nouser_xattr, "nouser_xattr"},
1378         {Opt_acl, "acl"},
1379         {Opt_noacl, "noacl"},
1380         {Opt_noload, "norecovery"},
1381         {Opt_noload, "noload"},
1382         {Opt_removed, "nobh"},
1383         {Opt_removed, "bh"},
1384         {Opt_commit, "commit=%u"},
1385         {Opt_min_batch_time, "min_batch_time=%u"},
1386         {Opt_max_batch_time, "max_batch_time=%u"},
1387         {Opt_journal_dev, "journal_dev=%u"},
1388         {Opt_journal_path, "journal_path=%s"},
1389         {Opt_journal_checksum, "journal_checksum"},
1390         {Opt_nojournal_checksum, "nojournal_checksum"},
1391         {Opt_journal_async_commit, "journal_async_commit"},
1392         {Opt_abort, "abort"},
1393         {Opt_data_journal, "data=journal"},
1394         {Opt_data_ordered, "data=ordered"},
1395         {Opt_data_writeback, "data=writeback"},
1396         {Opt_data_err_abort, "data_err=abort"},
1397         {Opt_data_err_ignore, "data_err=ignore"},
1398         {Opt_offusrjquota, "usrjquota="},
1399         {Opt_usrjquota, "usrjquota=%s"},
1400         {Opt_offgrpjquota, "grpjquota="},
1401         {Opt_grpjquota, "grpjquota=%s"},
1402         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1403         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1404         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1405         {Opt_grpquota, "grpquota"},
1406         {Opt_noquota, "noquota"},
1407         {Opt_quota, "quota"},
1408         {Opt_usrquota, "usrquota"},
1409         {Opt_prjquota, "prjquota"},
1410         {Opt_barrier, "barrier=%u"},
1411         {Opt_barrier, "barrier"},
1412         {Opt_nobarrier, "nobarrier"},
1413         {Opt_i_version, "i_version"},
1414         {Opt_dax, "dax"},
1415         {Opt_stripe, "stripe=%u"},
1416         {Opt_delalloc, "delalloc"},
1417         {Opt_lazytime, "lazytime"},
1418         {Opt_nolazytime, "nolazytime"},
1419         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1420         {Opt_nodelalloc, "nodelalloc"},
1421         {Opt_removed, "mblk_io_submit"},
1422         {Opt_removed, "nomblk_io_submit"},
1423         {Opt_block_validity, "block_validity"},
1424         {Opt_noblock_validity, "noblock_validity"},
1425         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1426         {Opt_journal_ioprio, "journal_ioprio=%u"},
1427         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1428         {Opt_auto_da_alloc, "auto_da_alloc"},
1429         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1430         {Opt_dioread_nolock, "dioread_nolock"},
1431         {Opt_dioread_lock, "dioread_lock"},
1432         {Opt_discard, "discard"},
1433         {Opt_nodiscard, "nodiscard"},
1434         {Opt_init_itable, "init_itable=%u"},
1435         {Opt_init_itable, "init_itable"},
1436         {Opt_noinit_itable, "noinit_itable"},
1437         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1438         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1439         {Opt_nombcache, "nombcache"},
1440         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1441         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1442         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1443         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1444         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1445         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1446         {Opt_err, NULL},
1447 };
1448
1449 static ext4_fsblk_t get_sb_block(void **data)
1450 {
1451         ext4_fsblk_t    sb_block;
1452         char            *options = (char *) *data;
1453
1454         if (!options || strncmp(options, "sb=", 3) != 0)
1455                 return 1;       /* Default location */
1456
1457         options += 3;
1458         /* TODO: use simple_strtoll with >32bit ext4 */
1459         sb_block = simple_strtoul(options, &options, 0);
1460         if (*options && *options != ',') {
1461                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1462                        (char *) *data);
1463                 return 1;
1464         }
1465         if (*options == ',')
1466                 options++;
1467         *data = (void *) options;
1468
1469         return sb_block;
1470 }
1471
1472 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1473 static const char deprecated_msg[] =
1474         "Mount option \"%s\" will be removed by %s\n"
1475         "Contact [email protected] if you think we should keep it.\n";
1476
1477 #ifdef CONFIG_QUOTA
1478 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1479 {
1480         struct ext4_sb_info *sbi = EXT4_SB(sb);
1481         char *qname;
1482         int ret = -1;
1483
1484         if (sb_any_quota_loaded(sb) &&
1485                 !sbi->s_qf_names[qtype]) {
1486                 ext4_msg(sb, KERN_ERR,
1487                         "Cannot change journaled "
1488                         "quota options when quota turned on");
1489                 return -1;
1490         }
1491         if (ext4_has_feature_quota(sb)) {
1492                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1493                          "ignored when QUOTA feature is enabled");
1494                 return 1;
1495         }
1496         qname = match_strdup(args);
1497         if (!qname) {
1498                 ext4_msg(sb, KERN_ERR,
1499                         "Not enough memory for storing quotafile name");
1500                 return -1;
1501         }
1502         if (sbi->s_qf_names[qtype]) {
1503                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1504                         ret = 1;
1505                 else
1506                         ext4_msg(sb, KERN_ERR,
1507                                  "%s quota file already specified",
1508                                  QTYPE2NAME(qtype));
1509                 goto errout;
1510         }
1511         if (strchr(qname, '/')) {
1512                 ext4_msg(sb, KERN_ERR,
1513                         "quotafile must be on filesystem root");
1514                 goto errout;
1515         }
1516         sbi->s_qf_names[qtype] = qname;
1517         set_opt(sb, QUOTA);
1518         return 1;
1519 errout:
1520         kfree(qname);
1521         return ret;
1522 }
1523
1524 static int clear_qf_name(struct super_block *sb, int qtype)
1525 {
1526
1527         struct ext4_sb_info *sbi = EXT4_SB(sb);
1528
1529         if (sb_any_quota_loaded(sb) &&
1530                 sbi->s_qf_names[qtype]) {
1531                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1532                         " when quota turned on");
1533                 return -1;
1534         }
1535         kfree(sbi->s_qf_names[qtype]);
1536         sbi->s_qf_names[qtype] = NULL;
1537         return 1;
1538 }
1539 #endif
1540
1541 #define MOPT_SET        0x0001
1542 #define MOPT_CLEAR      0x0002
1543 #define MOPT_NOSUPPORT  0x0004
1544 #define MOPT_EXPLICIT   0x0008
1545 #define MOPT_CLEAR_ERR  0x0010
1546 #define MOPT_GTE0       0x0020
1547 #ifdef CONFIG_QUOTA
1548 #define MOPT_Q          0
1549 #define MOPT_QFMT       0x0040
1550 #else
1551 #define MOPT_Q          MOPT_NOSUPPORT
1552 #define MOPT_QFMT       MOPT_NOSUPPORT
1553 #endif
1554 #define MOPT_DATAJ      0x0080
1555 #define MOPT_NO_EXT2    0x0100
1556 #define MOPT_NO_EXT3    0x0200
1557 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1558 #define MOPT_STRING     0x0400
1559
1560 static const struct mount_opts {
1561         int     token;
1562         int     mount_opt;
1563         int     flags;
1564 } ext4_mount_opts[] = {
1565         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1566         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1567         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1568         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1569         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1570         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1571         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1572          MOPT_EXT4_ONLY | MOPT_SET},
1573         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1574          MOPT_EXT4_ONLY | MOPT_CLEAR},
1575         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1576         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1577         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1578          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1579         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1580          MOPT_EXT4_ONLY | MOPT_CLEAR},
1581         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1582          MOPT_EXT4_ONLY | MOPT_CLEAR},
1583         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1584          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1585         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1586                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1587          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1588         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1589         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1590         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1591         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1592         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1593          MOPT_NO_EXT2},
1594         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1595          MOPT_NO_EXT2},
1596         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1597         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1598         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1599         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1600         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1601         {Opt_commit, 0, MOPT_GTE0},
1602         {Opt_max_batch_time, 0, MOPT_GTE0},
1603         {Opt_min_batch_time, 0, MOPT_GTE0},
1604         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1605         {Opt_init_itable, 0, MOPT_GTE0},
1606         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1607         {Opt_stripe, 0, MOPT_GTE0},
1608         {Opt_resuid, 0, MOPT_GTE0},
1609         {Opt_resgid, 0, MOPT_GTE0},
1610         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1611         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1612         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1613         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1614         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1615         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1616          MOPT_NO_EXT2 | MOPT_DATAJ},
1617         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1618         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1619 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1620         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1621         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1622 #else
1623         {Opt_acl, 0, MOPT_NOSUPPORT},
1624         {Opt_noacl, 0, MOPT_NOSUPPORT},
1625 #endif
1626         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1627         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1628         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1629         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1630         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1631                                                         MOPT_SET | MOPT_Q},
1632         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1633                                                         MOPT_SET | MOPT_Q},
1634         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1635                                                         MOPT_SET | MOPT_Q},
1636         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1637                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1638                                                         MOPT_CLEAR | MOPT_Q},
1639         {Opt_usrjquota, 0, MOPT_Q},
1640         {Opt_grpjquota, 0, MOPT_Q},
1641         {Opt_offusrjquota, 0, MOPT_Q},
1642         {Opt_offgrpjquota, 0, MOPT_Q},
1643         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1644         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1645         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1646         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1647         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1648         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1649         {Opt_err, 0, 0}
1650 };
1651
1652 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1653                             substring_t *args, unsigned long *journal_devnum,
1654                             unsigned int *journal_ioprio, int is_remount)
1655 {
1656         struct ext4_sb_info *sbi = EXT4_SB(sb);
1657         const struct mount_opts *m;
1658         kuid_t uid;
1659         kgid_t gid;
1660         int arg = 0;
1661
1662 #ifdef CONFIG_QUOTA
1663         if (token == Opt_usrjquota)
1664                 return set_qf_name(sb, USRQUOTA, &args[0]);
1665         else if (token == Opt_grpjquota)
1666                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1667         else if (token == Opt_offusrjquota)
1668                 return clear_qf_name(sb, USRQUOTA);
1669         else if (token == Opt_offgrpjquota)
1670                 return clear_qf_name(sb, GRPQUOTA);
1671 #endif
1672         switch (token) {
1673         case Opt_noacl:
1674         case Opt_nouser_xattr:
1675                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1676                 break;
1677         case Opt_sb:
1678                 return 1;       /* handled by get_sb_block() */
1679         case Opt_removed:
1680                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1681                 return 1;
1682         case Opt_abort:
1683                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1684                 return 1;
1685         case Opt_i_version:
1686                 sb->s_flags |= SB_I_VERSION;
1687                 return 1;
1688         case Opt_lazytime:
1689                 sb->s_flags |= SB_LAZYTIME;
1690                 return 1;
1691         case Opt_nolazytime:
1692                 sb->s_flags &= ~SB_LAZYTIME;
1693                 return 1;
1694         }
1695
1696         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1697                 if (token == m->token)
1698                         break;
1699
1700         if (m->token == Opt_err) {
1701                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1702                          "or missing value", opt);
1703                 return -1;
1704         }
1705
1706         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1707                 ext4_msg(sb, KERN_ERR,
1708                          "Mount option \"%s\" incompatible with ext2", opt);
1709                 return -1;
1710         }
1711         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1712                 ext4_msg(sb, KERN_ERR,
1713                          "Mount option \"%s\" incompatible with ext3", opt);
1714                 return -1;
1715         }
1716
1717         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1718                 return -1;
1719         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1720                 return -1;
1721         if (m->flags & MOPT_EXPLICIT) {
1722                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1723                         set_opt2(sb, EXPLICIT_DELALLOC);
1724                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1725                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1726                 } else
1727                         return -1;
1728         }
1729         if (m->flags & MOPT_CLEAR_ERR)
1730                 clear_opt(sb, ERRORS_MASK);
1731         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1732                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1733                          "options when quota turned on");
1734                 return -1;
1735         }
1736
1737         if (m->flags & MOPT_NOSUPPORT) {
1738                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1739         } else if (token == Opt_commit) {
1740                 if (arg == 0)
1741                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1742                 sbi->s_commit_interval = HZ * arg;
1743         } else if (token == Opt_debug_want_extra_isize) {
1744                 sbi->s_want_extra_isize = arg;
1745         } else if (token == Opt_max_batch_time) {
1746                 sbi->s_max_batch_time = arg;
1747         } else if (token == Opt_min_batch_time) {
1748                 sbi->s_min_batch_time = arg;
1749         } else if (token == Opt_inode_readahead_blks) {
1750                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1751                         ext4_msg(sb, KERN_ERR,
1752                                  "EXT4-fs: inode_readahead_blks must be "
1753                                  "0 or a power of 2 smaller than 2^31");
1754                         return -1;
1755                 }
1756                 sbi->s_inode_readahead_blks = arg;
1757         } else if (token == Opt_init_itable) {
1758                 set_opt(sb, INIT_INODE_TABLE);
1759                 if (!args->from)
1760                         arg = EXT4_DEF_LI_WAIT_MULT;
1761                 sbi->s_li_wait_mult = arg;
1762         } else if (token == Opt_max_dir_size_kb) {
1763                 sbi->s_max_dir_size_kb = arg;
1764         } else if (token == Opt_stripe) {
1765                 sbi->s_stripe = arg;
1766         } else if (token == Opt_resuid) {
1767                 uid = make_kuid(current_user_ns(), arg);
1768                 if (!uid_valid(uid)) {
1769                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1770                         return -1;
1771                 }
1772                 sbi->s_resuid = uid;
1773         } else if (token == Opt_resgid) {
1774                 gid = make_kgid(current_user_ns(), arg);
1775                 if (!gid_valid(gid)) {
1776                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1777                         return -1;
1778                 }
1779                 sbi->s_resgid = gid;
1780         } else if (token == Opt_journal_dev) {
1781                 if (is_remount) {
1782                         ext4_msg(sb, KERN_ERR,
1783                                  "Cannot specify journal on remount");
1784                         return -1;
1785                 }
1786                 *journal_devnum = arg;
1787         } else if (token == Opt_journal_path) {
1788                 char *journal_path;
1789                 struct inode *journal_inode;
1790                 struct path path;
1791                 int error;
1792
1793                 if (is_remount) {
1794                         ext4_msg(sb, KERN_ERR,
1795                                  "Cannot specify journal on remount");
1796                         return -1;
1797                 }
1798                 journal_path = match_strdup(&args[0]);
1799                 if (!journal_path) {
1800                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1801                                 "journal device string");
1802                         return -1;
1803                 }
1804
1805                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1806                 if (error) {
1807                         ext4_msg(sb, KERN_ERR, "error: could not find "
1808                                 "journal device path: error %d", error);
1809                         kfree(journal_path);
1810                         return -1;
1811                 }
1812
1813                 journal_inode = d_inode(path.dentry);
1814                 if (!S_ISBLK(journal_inode->i_mode)) {
1815                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1816                                 "is not a block device", journal_path);
1817                         path_put(&path);
1818                         kfree(journal_path);
1819                         return -1;
1820                 }
1821
1822                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1823                 path_put(&path);
1824                 kfree(journal_path);
1825         } else if (token == Opt_journal_ioprio) {
1826                 if (arg > 7) {
1827                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1828                                  " (must be 0-7)");
1829                         return -1;
1830                 }
1831                 *journal_ioprio =
1832                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1833         } else if (token == Opt_test_dummy_encryption) {
1834 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1835                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1836                 ext4_msg(sb, KERN_WARNING,
1837                          "Test dummy encryption mode enabled");
1838 #else
1839                 ext4_msg(sb, KERN_WARNING,
1840                          "Test dummy encryption mount option ignored");
1841 #endif
1842         } else if (m->flags & MOPT_DATAJ) {
1843                 if (is_remount) {
1844                         if (!sbi->s_journal)
1845                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1846                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1847                                 ext4_msg(sb, KERN_ERR,
1848                                          "Cannot change data mode on remount");
1849                                 return -1;
1850                         }
1851                 } else {
1852                         clear_opt(sb, DATA_FLAGS);
1853                         sbi->s_mount_opt |= m->mount_opt;
1854                 }
1855 #ifdef CONFIG_QUOTA
1856         } else if (m->flags & MOPT_QFMT) {
1857                 if (sb_any_quota_loaded(sb) &&
1858                     sbi->s_jquota_fmt != m->mount_opt) {
1859                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1860                                  "quota options when quota turned on");
1861                         return -1;
1862                 }
1863                 if (ext4_has_feature_quota(sb)) {
1864                         ext4_msg(sb, KERN_INFO,
1865                                  "Quota format mount options ignored "
1866                                  "when QUOTA feature is enabled");
1867                         return 1;
1868                 }
1869                 sbi->s_jquota_fmt = m->mount_opt;
1870 #endif
1871         } else if (token == Opt_dax) {
1872 #ifdef CONFIG_FS_DAX
1873                 ext4_msg(sb, KERN_WARNING,
1874                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1875                         sbi->s_mount_opt |= m->mount_opt;
1876 #else
1877                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1878                 return -1;
1879 #endif
1880         } else if (token == Opt_data_err_abort) {
1881                 sbi->s_mount_opt |= m->mount_opt;
1882         } else if (token == Opt_data_err_ignore) {
1883                 sbi->s_mount_opt &= ~m->mount_opt;
1884         } else {
1885                 if (!args->from)
1886                         arg = 1;
1887                 if (m->flags & MOPT_CLEAR)
1888                         arg = !arg;
1889                 else if (unlikely(!(m->flags & MOPT_SET))) {
1890                         ext4_msg(sb, KERN_WARNING,
1891                                  "buggy handling of option %s", opt);
1892                         WARN_ON(1);
1893                         return -1;
1894                 }
1895                 if (arg != 0)
1896                         sbi->s_mount_opt |= m->mount_opt;
1897                 else
1898                         sbi->s_mount_opt &= ~m->mount_opt;
1899         }
1900         return 1;
1901 }
1902
1903 static int parse_options(char *options, struct super_block *sb,
1904                          unsigned long *journal_devnum,
1905                          unsigned int *journal_ioprio,
1906                          int is_remount)
1907 {
1908         struct ext4_sb_info *sbi = EXT4_SB(sb);
1909         char *p;
1910         substring_t args[MAX_OPT_ARGS];
1911         int token;
1912
1913         if (!options)
1914                 return 1;
1915
1916         while ((p = strsep(&options, ",")) != NULL) {
1917                 if (!*p)
1918                         continue;
1919                 /*
1920                  * Initialize args struct so we know whether arg was
1921                  * found; some options take optional arguments.
1922                  */
1923                 args[0].to = args[0].from = NULL;
1924                 token = match_token(p, tokens, args);
1925                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1926                                      journal_ioprio, is_remount) < 0)
1927                         return 0;
1928         }
1929 #ifdef CONFIG_QUOTA
1930         /*
1931          * We do the test below only for project quotas. 'usrquota' and
1932          * 'grpquota' mount options are allowed even without quota feature
1933          * to support legacy quotas in quota files.
1934          */
1935         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1936                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1937                          "Cannot enable project quota enforcement.");
1938                 return 0;
1939         }
1940         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1941                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1942                         clear_opt(sb, USRQUOTA);
1943
1944                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1945                         clear_opt(sb, GRPQUOTA);
1946
1947                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1948                         ext4_msg(sb, KERN_ERR, "old and new quota "
1949                                         "format mixing");
1950                         return 0;
1951                 }
1952
1953                 if (!sbi->s_jquota_fmt) {
1954                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1955                                         "not specified");
1956                         return 0;
1957                 }
1958         }
1959 #endif
1960         if (test_opt(sb, DIOREAD_NOLOCK)) {
1961                 int blocksize =
1962                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1963
1964                 if (blocksize < PAGE_SIZE) {
1965                         ext4_msg(sb, KERN_ERR, "can't mount with "
1966                                  "dioread_nolock if block size != PAGE_SIZE");
1967                         return 0;
1968                 }
1969         }
1970         return 1;
1971 }
1972
1973 static inline void ext4_show_quota_options(struct seq_file *seq,
1974                                            struct super_block *sb)
1975 {
1976 #if defined(CONFIG_QUOTA)
1977         struct ext4_sb_info *sbi = EXT4_SB(sb);
1978
1979         if (sbi->s_jquota_fmt) {
1980                 char *fmtname = "";
1981
1982                 switch (sbi->s_jquota_fmt) {
1983                 case QFMT_VFS_OLD:
1984                         fmtname = "vfsold";
1985                         break;
1986                 case QFMT_VFS_V0:
1987                         fmtname = "vfsv0";
1988                         break;
1989                 case QFMT_VFS_V1:
1990                         fmtname = "vfsv1";
1991                         break;
1992                 }
1993                 seq_printf(seq, ",jqfmt=%s", fmtname);
1994         }
1995
1996         if (sbi->s_qf_names[USRQUOTA])
1997                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1998
1999         if (sbi->s_qf_names[GRPQUOTA])
2000                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
2001 #endif
2002 }
2003
2004 static const char *token2str(int token)
2005 {
2006         const struct match_token *t;
2007
2008         for (t = tokens; t->token != Opt_err; t++)
2009                 if (t->token == token && !strchr(t->pattern, '='))
2010                         break;
2011         return t->pattern;
2012 }
2013
2014 /*
2015  * Show an option if
2016  *  - it's set to a non-default value OR
2017  *  - if the per-sb default is different from the global default
2018  */
2019 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2020                               int nodefs)
2021 {
2022         struct ext4_sb_info *sbi = EXT4_SB(sb);
2023         struct ext4_super_block *es = sbi->s_es;
2024         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2025         const struct mount_opts *m;
2026         char sep = nodefs ? '\n' : ',';
2027
2028 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2029 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2030
2031         if (sbi->s_sb_block != 1)
2032                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2033
2034         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2035                 int want_set = m->flags & MOPT_SET;
2036                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2037                     (m->flags & MOPT_CLEAR_ERR))
2038                         continue;
2039                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2040                         continue; /* skip if same as the default */
2041                 if ((want_set &&
2042                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2043                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2044                         continue; /* select Opt_noFoo vs Opt_Foo */
2045                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2046         }
2047
2048         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2049             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2050                 SEQ_OPTS_PRINT("resuid=%u",
2051                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2052         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2053             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2054                 SEQ_OPTS_PRINT("resgid=%u",
2055                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2056         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2057         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2058                 SEQ_OPTS_PUTS("errors=remount-ro");
2059         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2060                 SEQ_OPTS_PUTS("errors=continue");
2061         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2062                 SEQ_OPTS_PUTS("errors=panic");
2063         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2064                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2065         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2066                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2067         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2068                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2069         if (sb->s_flags & SB_I_VERSION)
2070                 SEQ_OPTS_PUTS("i_version");
2071         if (nodefs || sbi->s_stripe)
2072                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2073         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2074                         (sbi->s_mount_opt ^ def_mount_opt)) {
2075                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2076                         SEQ_OPTS_PUTS("data=journal");
2077                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2078                         SEQ_OPTS_PUTS("data=ordered");
2079                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2080                         SEQ_OPTS_PUTS("data=writeback");
2081         }
2082         if (nodefs ||
2083             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2084                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2085                                sbi->s_inode_readahead_blks);
2086
2087         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2088                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2089                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2090         if (nodefs || sbi->s_max_dir_size_kb)
2091                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2092         if (test_opt(sb, DATA_ERR_ABORT))
2093                 SEQ_OPTS_PUTS("data_err=abort");
2094
2095         ext4_show_quota_options(seq, sb);
2096         return 0;
2097 }
2098
2099 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2100 {
2101         return _ext4_show_options(seq, root->d_sb, 0);
2102 }
2103
2104 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2105 {
2106         struct super_block *sb = seq->private;
2107         int rc;
2108
2109         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2110         rc = _ext4_show_options(seq, sb, 1);
2111         seq_puts(seq, "\n");
2112         return rc;
2113 }
2114
2115 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2116                             int read_only)
2117 {
2118         struct ext4_sb_info *sbi = EXT4_SB(sb);
2119         int res = 0;
2120
2121         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2122                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2123                          "forcing read-only mode");
2124                 res = SB_RDONLY;
2125         }
2126         if (read_only)
2127                 goto done;
2128         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2129                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2130                          "running e2fsck is recommended");
2131         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2132                 ext4_msg(sb, KERN_WARNING,
2133                          "warning: mounting fs with errors, "
2134                          "running e2fsck is recommended");
2135         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2136                  le16_to_cpu(es->s_mnt_count) >=
2137                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2138                 ext4_msg(sb, KERN_WARNING,
2139                          "warning: maximal mount count reached, "
2140                          "running e2fsck is recommended");
2141         else if (le32_to_cpu(es->s_checkinterval) &&
2142                 (le32_to_cpu(es->s_lastcheck) +
2143                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2144                 ext4_msg(sb, KERN_WARNING,
2145                          "warning: checktime reached, "
2146                          "running e2fsck is recommended");
2147         if (!sbi->s_journal)
2148                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2149         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2150                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2151         le16_add_cpu(&es->s_mnt_count, 1);
2152         es->s_mtime = cpu_to_le32(get_seconds());
2153         ext4_update_dynamic_rev(sb);
2154         if (sbi->s_journal)
2155                 ext4_set_feature_journal_needs_recovery(sb);
2156
2157         ext4_commit_super(sb, 1);
2158 done:
2159         if (test_opt(sb, DEBUG))
2160                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2161                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2162                         sb->s_blocksize,
2163                         sbi->s_groups_count,
2164                         EXT4_BLOCKS_PER_GROUP(sb),
2165                         EXT4_INODES_PER_GROUP(sb),
2166                         sbi->s_mount_opt, sbi->s_mount_opt2);
2167
2168         cleancache_init_fs(sb);
2169         return res;
2170 }
2171
2172 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2173 {
2174         struct ext4_sb_info *sbi = EXT4_SB(sb);
2175         struct flex_groups *new_groups;
2176         int size;
2177
2178         if (!sbi->s_log_groups_per_flex)
2179                 return 0;
2180
2181         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2182         if (size <= sbi->s_flex_groups_allocated)
2183                 return 0;
2184
2185         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2186         new_groups = kvzalloc(size, GFP_KERNEL);
2187         if (!new_groups) {
2188                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2189                          size / (int) sizeof(struct flex_groups));
2190                 return -ENOMEM;
2191         }
2192
2193         if (sbi->s_flex_groups) {
2194                 memcpy(new_groups, sbi->s_flex_groups,
2195                        (sbi->s_flex_groups_allocated *
2196                         sizeof(struct flex_groups)));
2197                 kvfree(sbi->s_flex_groups);
2198         }
2199         sbi->s_flex_groups = new_groups;
2200         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2201         return 0;
2202 }
2203
2204 static int ext4_fill_flex_info(struct super_block *sb)
2205 {
2206         struct ext4_sb_info *sbi = EXT4_SB(sb);
2207         struct ext4_group_desc *gdp = NULL;
2208         ext4_group_t flex_group;
2209         int i, err;
2210
2211         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2212         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2213                 sbi->s_log_groups_per_flex = 0;
2214                 return 1;
2215         }
2216
2217         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2218         if (err)
2219                 goto failed;
2220
2221         for (i = 0; i < sbi->s_groups_count; i++) {
2222                 gdp = ext4_get_group_desc(sb, i, NULL);
2223
2224                 flex_group = ext4_flex_group(sbi, i);
2225                 atomic_add(ext4_free_inodes_count(sb, gdp),
2226                            &sbi->s_flex_groups[flex_group].free_inodes);
2227                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2228                              &sbi->s_flex_groups[flex_group].free_clusters);
2229                 atomic_add(ext4_used_dirs_count(sb, gdp),
2230                            &sbi->s_flex_groups[flex_group].used_dirs);
2231         }
2232
2233         return 1;
2234 failed:
2235         return 0;
2236 }
2237
2238 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2239                                    struct ext4_group_desc *gdp)
2240 {
2241         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2242         __u16 crc = 0;
2243         __le32 le_group = cpu_to_le32(block_group);
2244         struct ext4_sb_info *sbi = EXT4_SB(sb);
2245
2246         if (ext4_has_metadata_csum(sbi->s_sb)) {
2247                 /* Use new metadata_csum algorithm */
2248                 __u32 csum32;
2249                 __u16 dummy_csum = 0;
2250
2251                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2252                                      sizeof(le_group));
2253                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2254                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2255                                      sizeof(dummy_csum));
2256                 offset += sizeof(dummy_csum);
2257                 if (offset < sbi->s_desc_size)
2258                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2259                                              sbi->s_desc_size - offset);
2260
2261                 crc = csum32 & 0xFFFF;
2262                 goto out;
2263         }
2264
2265         /* old crc16 code */
2266         if (!ext4_has_feature_gdt_csum(sb))
2267                 return 0;
2268
2269         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2270         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2271         crc = crc16(crc, (__u8 *)gdp, offset);
2272         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2273         /* for checksum of struct ext4_group_desc do the rest...*/
2274         if (ext4_has_feature_64bit(sb) &&
2275             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2276                 crc = crc16(crc, (__u8 *)gdp + offset,
2277                             le16_to_cpu(sbi->s_es->s_desc_size) -
2278                                 offset);
2279
2280 out:
2281         return cpu_to_le16(crc);
2282 }
2283
2284 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2285                                 struct ext4_group_desc *gdp)
2286 {
2287         if (ext4_has_group_desc_csum(sb) &&
2288             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2289                 return 0;
2290
2291         return 1;
2292 }
2293
2294 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2295                               struct ext4_group_desc *gdp)
2296 {
2297         if (!ext4_has_group_desc_csum(sb))
2298                 return;
2299         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2300 }
2301
2302 /* Called at mount-time, super-block is locked */
2303 static int ext4_check_descriptors(struct super_block *sb,
2304                                   ext4_fsblk_t sb_block,
2305                                   ext4_group_t *first_not_zeroed)
2306 {
2307         struct ext4_sb_info *sbi = EXT4_SB(sb);
2308         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2309         ext4_fsblk_t last_block;
2310         ext4_fsblk_t block_bitmap;
2311         ext4_fsblk_t inode_bitmap;
2312         ext4_fsblk_t inode_table;
2313         int flexbg_flag = 0;
2314         ext4_group_t i, grp = sbi->s_groups_count;
2315
2316         if (ext4_has_feature_flex_bg(sb))
2317                 flexbg_flag = 1;
2318
2319         ext4_debug("Checking group descriptors");
2320
2321         for (i = 0; i < sbi->s_groups_count; i++) {
2322                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2323
2324                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2325                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2326                 else
2327                         last_block = first_block +
2328                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2329
2330                 if ((grp == sbi->s_groups_count) &&
2331                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2332                         grp = i;
2333
2334                 block_bitmap = ext4_block_bitmap(sb, gdp);
2335                 if (block_bitmap == sb_block) {
2336                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2337                                  "Block bitmap for group %u overlaps "
2338                                  "superblock", i);
2339                         if (!sb_rdonly(sb))
2340                                 return 0;
2341                 }
2342                 if (block_bitmap < first_block || block_bitmap > last_block) {
2343                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2344                                "Block bitmap for group %u not in group "
2345                                "(block %llu)!", i, block_bitmap);
2346                         return 0;
2347                 }
2348                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2349                 if (inode_bitmap == sb_block) {
2350                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2351                                  "Inode bitmap for group %u overlaps "
2352                                  "superblock", i);
2353                         if (!sb_rdonly(sb))
2354                                 return 0;
2355                 }
2356                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2357                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2358                                "Inode bitmap for group %u not in group "
2359                                "(block %llu)!", i, inode_bitmap);
2360                         return 0;
2361                 }
2362                 inode_table = ext4_inode_table(sb, gdp);
2363                 if (inode_table == sb_block) {
2364                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2365                                  "Inode table for group %u overlaps "
2366                                  "superblock", i);
2367                         if (!sb_rdonly(sb))
2368                                 return 0;
2369                 }
2370                 if (inode_table < first_block ||
2371                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2372                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2373                                "Inode table for group %u not in group "
2374                                "(block %llu)!", i, inode_table);
2375                         return 0;
2376                 }
2377                 ext4_lock_group(sb, i);
2378                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2379                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2380                                  "Checksum for group %u failed (%u!=%u)",
2381                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2382                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2383                         if (!sb_rdonly(sb)) {
2384                                 ext4_unlock_group(sb, i);
2385                                 return 0;
2386                         }
2387                 }
2388                 ext4_unlock_group(sb, i);
2389                 if (!flexbg_flag)
2390                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2391         }
2392         if (NULL != first_not_zeroed)
2393                 *first_not_zeroed = grp;
2394         return 1;
2395 }
2396
2397 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2398  * the superblock) which were deleted from all directories, but held open by
2399  * a process at the time of a crash.  We walk the list and try to delete these
2400  * inodes at recovery time (only with a read-write filesystem).
2401  *
2402  * In order to keep the orphan inode chain consistent during traversal (in
2403  * case of crash during recovery), we link each inode into the superblock
2404  * orphan list_head and handle it the same way as an inode deletion during
2405  * normal operation (which journals the operations for us).
2406  *
2407  * We only do an iget() and an iput() on each inode, which is very safe if we
2408  * accidentally point at an in-use or already deleted inode.  The worst that
2409  * can happen in this case is that we get a "bit already cleared" message from
2410  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2411  * e2fsck was run on this filesystem, and it must have already done the orphan
2412  * inode cleanup for us, so we can safely abort without any further action.
2413  */
2414 static void ext4_orphan_cleanup(struct super_block *sb,
2415                                 struct ext4_super_block *es)
2416 {
2417         unsigned int s_flags = sb->s_flags;
2418         int ret, nr_orphans = 0, nr_truncates = 0;
2419 #ifdef CONFIG_QUOTA
2420         int quota_update = 0;
2421         int i;
2422 #endif
2423         if (!es->s_last_orphan) {
2424                 jbd_debug(4, "no orphan inodes to clean up\n");
2425                 return;
2426         }
2427
2428         if (bdev_read_only(sb->s_bdev)) {
2429                 ext4_msg(sb, KERN_ERR, "write access "
2430                         "unavailable, skipping orphan cleanup");
2431                 return;
2432         }
2433
2434         /* Check if feature set would not allow a r/w mount */
2435         if (!ext4_feature_set_ok(sb, 0)) {
2436                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2437                          "unknown ROCOMPAT features");
2438                 return;
2439         }
2440
2441         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2442                 /* don't clear list on RO mount w/ errors */
2443                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2444                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2445                                   "clearing orphan list.\n");
2446                         es->s_last_orphan = 0;
2447                 }
2448                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2449                 return;
2450         }
2451
2452         if (s_flags & SB_RDONLY) {
2453                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2454                 sb->s_flags &= ~SB_RDONLY;
2455         }
2456 #ifdef CONFIG_QUOTA
2457         /* Needed for iput() to work correctly and not trash data */
2458         sb->s_flags |= SB_ACTIVE;
2459
2460         /*
2461          * Turn on quotas which were not enabled for read-only mounts if
2462          * filesystem has quota feature, so that they are updated correctly.
2463          */
2464         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2465                 int ret = ext4_enable_quotas(sb);
2466
2467                 if (!ret)
2468                         quota_update = 1;
2469                 else
2470                         ext4_msg(sb, KERN_ERR,
2471                                 "Cannot turn on quotas: error %d", ret);
2472         }
2473
2474         /* Turn on journaled quotas used for old sytle */
2475         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2476                 if (EXT4_SB(sb)->s_qf_names[i]) {
2477                         int ret = ext4_quota_on_mount(sb, i);
2478
2479                         if (!ret)
2480                                 quota_update = 1;
2481                         else
2482                                 ext4_msg(sb, KERN_ERR,
2483                                         "Cannot turn on journaled "
2484                                         "quota: type %d: error %d", i, ret);
2485                 }
2486         }
2487 #endif
2488
2489         while (es->s_last_orphan) {
2490                 struct inode *inode;
2491
2492                 /*
2493                  * We may have encountered an error during cleanup; if
2494                  * so, skip the rest.
2495                  */
2496                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2497                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2498                         es->s_last_orphan = 0;
2499                         break;
2500                 }
2501
2502                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2503                 if (IS_ERR(inode)) {
2504                         es->s_last_orphan = 0;
2505                         break;
2506                 }
2507
2508                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2509                 dquot_initialize(inode);
2510                 if (inode->i_nlink) {
2511                         if (test_opt(sb, DEBUG))
2512                                 ext4_msg(sb, KERN_DEBUG,
2513                                         "%s: truncating inode %lu to %lld bytes",
2514                                         __func__, inode->i_ino, inode->i_size);
2515                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2516                                   inode->i_ino, inode->i_size);
2517                         inode_lock(inode);
2518                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2519                         ret = ext4_truncate(inode);
2520                         if (ret)
2521                                 ext4_std_error(inode->i_sb, ret);
2522                         inode_unlock(inode);
2523                         nr_truncates++;
2524                 } else {
2525                         if (test_opt(sb, DEBUG))
2526                                 ext4_msg(sb, KERN_DEBUG,
2527                                         "%s: deleting unreferenced inode %lu",
2528                                         __func__, inode->i_ino);
2529                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2530                                   inode->i_ino);
2531                         nr_orphans++;
2532                 }
2533                 iput(inode);  /* The delete magic happens here! */
2534         }
2535
2536 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2537
2538         if (nr_orphans)
2539                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2540                        PLURAL(nr_orphans));
2541         if (nr_truncates)
2542                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2543                        PLURAL(nr_truncates));
2544 #ifdef CONFIG_QUOTA
2545         /* Turn off quotas if they were enabled for orphan cleanup */
2546         if (quota_update) {
2547                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2548                         if (sb_dqopt(sb)->files[i])
2549                                 dquot_quota_off(sb, i);
2550                 }
2551         }
2552 #endif
2553         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2554 }
2555
2556 /*
2557  * Maximal extent format file size.
2558  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2559  * extent format containers, within a sector_t, and within i_blocks
2560  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2561  * so that won't be a limiting factor.
2562  *
2563  * However there is other limiting factor. We do store extents in the form
2564  * of starting block and length, hence the resulting length of the extent
2565  * covering maximum file size must fit into on-disk format containers as
2566  * well. Given that length is always by 1 unit bigger than max unit (because
2567  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2568  *
2569  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2570  */
2571 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2572 {
2573         loff_t res;
2574         loff_t upper_limit = MAX_LFS_FILESIZE;
2575
2576         /* small i_blocks in vfs inode? */
2577         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2578                 /*
2579                  * CONFIG_LBDAF is not enabled implies the inode
2580                  * i_block represent total blocks in 512 bytes
2581                  * 32 == size of vfs inode i_blocks * 8
2582                  */
2583                 upper_limit = (1LL << 32) - 1;
2584
2585                 /* total blocks in file system block size */
2586                 upper_limit >>= (blkbits - 9);
2587                 upper_limit <<= blkbits;
2588         }
2589
2590         /*
2591          * 32-bit extent-start container, ee_block. We lower the maxbytes
2592          * by one fs block, so ee_len can cover the extent of maximum file
2593          * size
2594          */
2595         res = (1LL << 32) - 1;
2596         res <<= blkbits;
2597
2598         /* Sanity check against vm- & vfs- imposed limits */
2599         if (res > upper_limit)
2600                 res = upper_limit;
2601
2602         return res;
2603 }
2604
2605 /*
2606  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2607  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2608  * We need to be 1 filesystem block less than the 2^48 sector limit.
2609  */
2610 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2611 {
2612         loff_t res = EXT4_NDIR_BLOCKS;
2613         int meta_blocks;
2614         loff_t upper_limit;
2615         /* This is calculated to be the largest file size for a dense, block
2616          * mapped file such that the file's total number of 512-byte sectors,
2617          * including data and all indirect blocks, does not exceed (2^48 - 1).
2618          *
2619          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2620          * number of 512-byte sectors of the file.
2621          */
2622
2623         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2624                 /*
2625                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2626                  * the inode i_block field represents total file blocks in
2627                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2628                  */
2629                 upper_limit = (1LL << 32) - 1;
2630
2631                 /* total blocks in file system block size */
2632                 upper_limit >>= (bits - 9);
2633
2634         } else {
2635                 /*
2636                  * We use 48 bit ext4_inode i_blocks
2637                  * With EXT4_HUGE_FILE_FL set the i_blocks
2638                  * represent total number of blocks in
2639                  * file system block size
2640                  */
2641                 upper_limit = (1LL << 48) - 1;
2642
2643         }
2644
2645         /* indirect blocks */
2646         meta_blocks = 1;
2647         /* double indirect blocks */
2648         meta_blocks += 1 + (1LL << (bits-2));
2649         /* tripple indirect blocks */
2650         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2651
2652         upper_limit -= meta_blocks;
2653         upper_limit <<= bits;
2654
2655         res += 1LL << (bits-2);
2656         res += 1LL << (2*(bits-2));
2657         res += 1LL << (3*(bits-2));
2658         res <<= bits;
2659         if (res > upper_limit)
2660                 res = upper_limit;
2661
2662         if (res > MAX_LFS_FILESIZE)
2663                 res = MAX_LFS_FILESIZE;
2664
2665         return res;
2666 }
2667
2668 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2669                                    ext4_fsblk_t logical_sb_block, int nr)
2670 {
2671         struct ext4_sb_info *sbi = EXT4_SB(sb);
2672         ext4_group_t bg, first_meta_bg;
2673         int has_super = 0;
2674
2675         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2676
2677         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2678                 return logical_sb_block + nr + 1;
2679         bg = sbi->s_desc_per_block * nr;
2680         if (ext4_bg_has_super(sb, bg))
2681                 has_super = 1;
2682
2683         /*
2684          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2685          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2686          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2687          * compensate.
2688          */
2689         if (sb->s_blocksize == 1024 && nr == 0 &&
2690             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2691                 has_super++;
2692
2693         return (has_super + ext4_group_first_block_no(sb, bg));
2694 }
2695
2696 /**
2697  * ext4_get_stripe_size: Get the stripe size.
2698  * @sbi: In memory super block info
2699  *
2700  * If we have specified it via mount option, then
2701  * use the mount option value. If the value specified at mount time is
2702  * greater than the blocks per group use the super block value.
2703  * If the super block value is greater than blocks per group return 0.
2704  * Allocator needs it be less than blocks per group.
2705  *
2706  */
2707 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2708 {
2709         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2710         unsigned long stripe_width =
2711                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2712         int ret;
2713
2714         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2715                 ret = sbi->s_stripe;
2716         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2717                 ret = stripe_width;
2718         else if (stride && stride <= sbi->s_blocks_per_group)
2719                 ret = stride;
2720         else
2721                 ret = 0;
2722
2723         /*
2724          * If the stripe width is 1, this makes no sense and
2725          * we set it to 0 to turn off stripe handling code.
2726          */
2727         if (ret <= 1)
2728                 ret = 0;
2729
2730         return ret;
2731 }
2732
2733 /*
2734  * Check whether this filesystem can be mounted based on
2735  * the features present and the RDONLY/RDWR mount requested.
2736  * Returns 1 if this filesystem can be mounted as requested,
2737  * 0 if it cannot be.
2738  */
2739 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2740 {
2741         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2742                 ext4_msg(sb, KERN_ERR,
2743                         "Couldn't mount because of "
2744                         "unsupported optional features (%x)",
2745                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2746                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2747                 return 0;
2748         }
2749
2750         if (readonly)
2751                 return 1;
2752
2753         if (ext4_has_feature_readonly(sb)) {
2754                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2755                 sb->s_flags |= SB_RDONLY;
2756                 return 1;
2757         }
2758
2759         /* Check that feature set is OK for a read-write mount */
2760         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2761                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2762                          "unsupported optional features (%x)",
2763                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2764                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2765                 return 0;
2766         }
2767         /*
2768          * Large file size enabled file system can only be mounted
2769          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2770          */
2771         if (ext4_has_feature_huge_file(sb)) {
2772                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2773                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2774                                  "cannot be mounted RDWR without "
2775                                  "CONFIG_LBDAF");
2776                         return 0;
2777                 }
2778         }
2779         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2780                 ext4_msg(sb, KERN_ERR,
2781                          "Can't support bigalloc feature without "
2782                          "extents feature\n");
2783                 return 0;
2784         }
2785
2786 #ifndef CONFIG_QUOTA
2787         if (ext4_has_feature_quota(sb) && !readonly) {
2788                 ext4_msg(sb, KERN_ERR,
2789                          "Filesystem with quota feature cannot be mounted RDWR "
2790                          "without CONFIG_QUOTA");
2791                 return 0;
2792         }
2793         if (ext4_has_feature_project(sb) && !readonly) {
2794                 ext4_msg(sb, KERN_ERR,
2795                          "Filesystem with project quota feature cannot be mounted RDWR "
2796                          "without CONFIG_QUOTA");
2797                 return 0;
2798         }
2799 #endif  /* CONFIG_QUOTA */
2800         return 1;
2801 }
2802
2803 /*
2804  * This function is called once a day if we have errors logged
2805  * on the file system
2806  */
2807 static void print_daily_error_info(struct timer_list *t)
2808 {
2809         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2810         struct super_block *sb = sbi->s_sb;
2811         struct ext4_super_block *es = sbi->s_es;
2812
2813         if (es->s_error_count)
2814                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2815                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2816                          le32_to_cpu(es->s_error_count));
2817         if (es->s_first_error_time) {
2818                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2819                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2820                        (int) sizeof(es->s_first_error_func),
2821                        es->s_first_error_func,
2822                        le32_to_cpu(es->s_first_error_line));
2823                 if (es->s_first_error_ino)
2824                         printk(KERN_CONT ": inode %u",
2825                                le32_to_cpu(es->s_first_error_ino));
2826                 if (es->s_first_error_block)
2827                         printk(KERN_CONT ": block %llu", (unsigned long long)
2828                                le64_to_cpu(es->s_first_error_block));
2829                 printk(KERN_CONT "\n");
2830         }
2831         if (es->s_last_error_time) {
2832                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2833                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2834                        (int) sizeof(es->s_last_error_func),
2835                        es->s_last_error_func,
2836                        le32_to_cpu(es->s_last_error_line));
2837                 if (es->s_last_error_ino)
2838                         printk(KERN_CONT ": inode %u",
2839                                le32_to_cpu(es->s_last_error_ino));
2840                 if (es->s_last_error_block)
2841                         printk(KERN_CONT ": block %llu", (unsigned long long)
2842                                le64_to_cpu(es->s_last_error_block));
2843                 printk(KERN_CONT "\n");
2844         }
2845         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2846 }
2847
2848 /* Find next suitable group and run ext4_init_inode_table */
2849 static int ext4_run_li_request(struct ext4_li_request *elr)
2850 {
2851         struct ext4_group_desc *gdp = NULL;
2852         ext4_group_t group, ngroups;
2853         struct super_block *sb;
2854         unsigned long timeout = 0;
2855         int ret = 0;
2856
2857         sb = elr->lr_super;
2858         ngroups = EXT4_SB(sb)->s_groups_count;
2859
2860         for (group = elr->lr_next_group; group < ngroups; group++) {
2861                 gdp = ext4_get_group_desc(sb, group, NULL);
2862                 if (!gdp) {
2863                         ret = 1;
2864                         break;
2865                 }
2866
2867                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2868                         break;
2869         }
2870
2871         if (group >= ngroups)
2872                 ret = 1;
2873
2874         if (!ret) {
2875                 timeout = jiffies;
2876                 ret = ext4_init_inode_table(sb, group,
2877                                             elr->lr_timeout ? 0 : 1);
2878                 if (elr->lr_timeout == 0) {
2879                         timeout = (jiffies - timeout) *
2880                                   elr->lr_sbi->s_li_wait_mult;
2881                         elr->lr_timeout = timeout;
2882                 }
2883                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2884                 elr->lr_next_group = group + 1;
2885         }
2886         return ret;
2887 }
2888
2889 /*
2890  * Remove lr_request from the list_request and free the
2891  * request structure. Should be called with li_list_mtx held
2892  */
2893 static void ext4_remove_li_request(struct ext4_li_request *elr)
2894 {
2895         struct ext4_sb_info *sbi;
2896
2897         if (!elr)
2898                 return;
2899
2900         sbi = elr->lr_sbi;
2901
2902         list_del(&elr->lr_request);
2903         sbi->s_li_request = NULL;
2904         kfree(elr);
2905 }
2906
2907 static void ext4_unregister_li_request(struct super_block *sb)
2908 {
2909         mutex_lock(&ext4_li_mtx);
2910         if (!ext4_li_info) {
2911                 mutex_unlock(&ext4_li_mtx);
2912                 return;
2913         }
2914
2915         mutex_lock(&ext4_li_info->li_list_mtx);
2916         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2917         mutex_unlock(&ext4_li_info->li_list_mtx);
2918         mutex_unlock(&ext4_li_mtx);
2919 }
2920
2921 static struct task_struct *ext4_lazyinit_task;
2922
2923 /*
2924  * This is the function where ext4lazyinit thread lives. It walks
2925  * through the request list searching for next scheduled filesystem.
2926  * When such a fs is found, run the lazy initialization request
2927  * (ext4_rn_li_request) and keep track of the time spend in this
2928  * function. Based on that time we compute next schedule time of
2929  * the request. When walking through the list is complete, compute
2930  * next waking time and put itself into sleep.
2931  */
2932 static int ext4_lazyinit_thread(void *arg)
2933 {
2934         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2935         struct list_head *pos, *n;
2936         struct ext4_li_request *elr;
2937         unsigned long next_wakeup, cur;
2938
2939         BUG_ON(NULL == eli);
2940
2941 cont_thread:
2942         while (true) {
2943                 next_wakeup = MAX_JIFFY_OFFSET;
2944
2945                 mutex_lock(&eli->li_list_mtx);
2946                 if (list_empty(&eli->li_request_list)) {
2947                         mutex_unlock(&eli->li_list_mtx);
2948                         goto exit_thread;
2949                 }
2950                 list_for_each_safe(pos, n, &eli->li_request_list) {
2951                         int err = 0;
2952                         int progress = 0;
2953                         elr = list_entry(pos, struct ext4_li_request,
2954                                          lr_request);
2955
2956                         if (time_before(jiffies, elr->lr_next_sched)) {
2957                                 if (time_before(elr->lr_next_sched, next_wakeup))
2958                                         next_wakeup = elr->lr_next_sched;
2959                                 continue;
2960                         }
2961                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2962                                 if (sb_start_write_trylock(elr->lr_super)) {
2963                                         progress = 1;
2964                                         /*
2965                                          * We hold sb->s_umount, sb can not
2966                                          * be removed from the list, it is
2967                                          * now safe to drop li_list_mtx
2968                                          */
2969                                         mutex_unlock(&eli->li_list_mtx);
2970                                         err = ext4_run_li_request(elr);
2971                                         sb_end_write(elr->lr_super);
2972                                         mutex_lock(&eli->li_list_mtx);
2973                                         n = pos->next;
2974                                 }
2975                                 up_read((&elr->lr_super->s_umount));
2976                         }
2977                         /* error, remove the lazy_init job */
2978                         if (err) {
2979                                 ext4_remove_li_request(elr);
2980                                 continue;
2981                         }
2982                         if (!progress) {
2983                                 elr->lr_next_sched = jiffies +
2984                                         (prandom_u32()
2985                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2986                         }
2987                         if (time_before(elr->lr_next_sched, next_wakeup))
2988                                 next_wakeup = elr->lr_next_sched;
2989                 }
2990                 mutex_unlock(&eli->li_list_mtx);
2991
2992                 try_to_freeze();
2993
2994                 cur = jiffies;
2995                 if ((time_after_eq(cur, next_wakeup)) ||
2996                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2997                         cond_resched();
2998                         continue;
2999                 }
3000
3001                 schedule_timeout_interruptible(next_wakeup - cur);
3002
3003                 if (kthread_should_stop()) {
3004                         ext4_clear_request_list();
3005                         goto exit_thread;
3006                 }
3007         }
3008
3009 exit_thread:
3010         /*
3011          * It looks like the request list is empty, but we need
3012          * to check it under the li_list_mtx lock, to prevent any
3013          * additions into it, and of course we should lock ext4_li_mtx
3014          * to atomically free the list and ext4_li_info, because at
3015          * this point another ext4 filesystem could be registering
3016          * new one.
3017          */
3018         mutex_lock(&ext4_li_mtx);
3019         mutex_lock(&eli->li_list_mtx);
3020         if (!list_empty(&eli->li_request_list)) {
3021                 mutex_unlock(&eli->li_list_mtx);
3022                 mutex_unlock(&ext4_li_mtx);
3023                 goto cont_thread;
3024         }
3025         mutex_unlock(&eli->li_list_mtx);
3026         kfree(ext4_li_info);
3027         ext4_li_info = NULL;
3028         mutex_unlock(&ext4_li_mtx);
3029
3030         return 0;
3031 }
3032
3033 static void ext4_clear_request_list(void)
3034 {
3035         struct list_head *pos, *n;
3036         struct ext4_li_request *elr;
3037
3038         mutex_lock(&ext4_li_info->li_list_mtx);
3039         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3040                 elr = list_entry(pos, struct ext4_li_request,
3041                                  lr_request);
3042                 ext4_remove_li_request(elr);
3043         }
3044         mutex_unlock(&ext4_li_info->li_list_mtx);
3045 }
3046
3047 static int ext4_run_lazyinit_thread(void)
3048 {
3049         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3050                                          ext4_li_info, "ext4lazyinit");
3051         if (IS_ERR(ext4_lazyinit_task)) {
3052                 int err = PTR_ERR(ext4_lazyinit_task);
3053                 ext4_clear_request_list();
3054                 kfree(ext4_li_info);
3055                 ext4_li_info = NULL;
3056                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3057                                  "initialization thread\n",
3058                                  err);
3059                 return err;
3060         }
3061         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3062         return 0;
3063 }
3064
3065 /*
3066  * Check whether it make sense to run itable init. thread or not.
3067  * If there is at least one uninitialized inode table, return
3068  * corresponding group number, else the loop goes through all
3069  * groups and return total number of groups.
3070  */
3071 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3072 {
3073         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3074         struct ext4_group_desc *gdp = NULL;
3075
3076         for (group = 0; group < ngroups; group++) {
3077                 gdp = ext4_get_group_desc(sb, group, NULL);
3078                 if (!gdp)
3079                         continue;
3080
3081                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3082                         break;
3083         }
3084
3085         return group;
3086 }
3087
3088 static int ext4_li_info_new(void)
3089 {
3090         struct ext4_lazy_init *eli = NULL;
3091
3092         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3093         if (!eli)
3094                 return -ENOMEM;
3095
3096         INIT_LIST_HEAD(&eli->li_request_list);
3097         mutex_init(&eli->li_list_mtx);
3098
3099         eli->li_state |= EXT4_LAZYINIT_QUIT;
3100
3101         ext4_li_info = eli;
3102
3103         return 0;
3104 }
3105
3106 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3107                                             ext4_group_t start)
3108 {
3109         struct ext4_sb_info *sbi = EXT4_SB(sb);
3110         struct ext4_li_request *elr;
3111
3112         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3113         if (!elr)
3114                 return NULL;
3115
3116         elr->lr_super = sb;
3117         elr->lr_sbi = sbi;
3118         elr->lr_next_group = start;
3119
3120         /*
3121          * Randomize first schedule time of the request to
3122          * spread the inode table initialization requests
3123          * better.
3124          */
3125         elr->lr_next_sched = jiffies + (prandom_u32() %
3126                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3127         return elr;
3128 }
3129
3130 int ext4_register_li_request(struct super_block *sb,
3131                              ext4_group_t first_not_zeroed)
3132 {
3133         struct ext4_sb_info *sbi = EXT4_SB(sb);
3134         struct ext4_li_request *elr = NULL;
3135         ext4_group_t ngroups = sbi->s_groups_count;
3136         int ret = 0;
3137
3138         mutex_lock(&ext4_li_mtx);
3139         if (sbi->s_li_request != NULL) {
3140                 /*
3141                  * Reset timeout so it can be computed again, because
3142                  * s_li_wait_mult might have changed.
3143                  */
3144                 sbi->s_li_request->lr_timeout = 0;
3145                 goto out;
3146         }
3147
3148         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3149             !test_opt(sb, INIT_INODE_TABLE))
3150                 goto out;
3151
3152         elr = ext4_li_request_new(sb, first_not_zeroed);
3153         if (!elr) {
3154                 ret = -ENOMEM;
3155                 goto out;
3156         }
3157
3158         if (NULL == ext4_li_info) {
3159                 ret = ext4_li_info_new();
3160                 if (ret)
3161                         goto out;
3162         }
3163
3164         mutex_lock(&ext4_li_info->li_list_mtx);
3165         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3166         mutex_unlock(&ext4_li_info->li_list_mtx);
3167
3168         sbi->s_li_request = elr;
3169         /*
3170          * set elr to NULL here since it has been inserted to
3171          * the request_list and the removal and free of it is
3172          * handled by ext4_clear_request_list from now on.
3173          */
3174         elr = NULL;
3175
3176         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3177                 ret = ext4_run_lazyinit_thread();
3178                 if (ret)
3179                         goto out;
3180         }
3181 out:
3182         mutex_unlock(&ext4_li_mtx);
3183         if (ret)
3184                 kfree(elr);
3185         return ret;
3186 }
3187
3188 /*
3189  * We do not need to lock anything since this is called on
3190  * module unload.
3191  */
3192 static void ext4_destroy_lazyinit_thread(void)
3193 {
3194         /*
3195          * If thread exited earlier
3196          * there's nothing to be done.
3197          */
3198         if (!ext4_li_info || !ext4_lazyinit_task)
3199                 return;
3200
3201         kthread_stop(ext4_lazyinit_task);
3202 }
3203
3204 static int set_journal_csum_feature_set(struct super_block *sb)
3205 {
3206         int ret = 1;
3207         int compat, incompat;
3208         struct ext4_sb_info *sbi = EXT4_SB(sb);
3209
3210         if (ext4_has_metadata_csum(sb)) {
3211                 /* journal checksum v3 */
3212                 compat = 0;
3213                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3214         } else {
3215                 /* journal checksum v1 */
3216                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3217                 incompat = 0;
3218         }
3219
3220         jbd2_journal_clear_features(sbi->s_journal,
3221                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3222                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3223                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3224         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3225                 ret = jbd2_journal_set_features(sbi->s_journal,
3226                                 compat, 0,
3227                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3228                                 incompat);
3229         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3230                 ret = jbd2_journal_set_features(sbi->s_journal,
3231                                 compat, 0,
3232                                 incompat);
3233                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3234                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3235         } else {
3236                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3237                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3238         }
3239
3240         return ret;
3241 }
3242
3243 /*
3244  * Note: calculating the overhead so we can be compatible with
3245  * historical BSD practice is quite difficult in the face of
3246  * clusters/bigalloc.  This is because multiple metadata blocks from
3247  * different block group can end up in the same allocation cluster.
3248  * Calculating the exact overhead in the face of clustered allocation
3249  * requires either O(all block bitmaps) in memory or O(number of block
3250  * groups**2) in time.  We will still calculate the superblock for
3251  * older file systems --- and if we come across with a bigalloc file
3252  * system with zero in s_overhead_clusters the estimate will be close to
3253  * correct especially for very large cluster sizes --- but for newer
3254  * file systems, it's better to calculate this figure once at mkfs
3255  * time, and store it in the superblock.  If the superblock value is
3256  * present (even for non-bigalloc file systems), we will use it.
3257  */
3258 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3259                           char *buf)
3260 {
3261         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3262         struct ext4_group_desc  *gdp;
3263         ext4_fsblk_t            first_block, last_block, b;
3264         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3265         int                     s, j, count = 0;
3266
3267         if (!ext4_has_feature_bigalloc(sb))
3268                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3269                         sbi->s_itb_per_group + 2);
3270
3271         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3272                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3273         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3274         for (i = 0; i < ngroups; i++) {
3275                 gdp = ext4_get_group_desc(sb, i, NULL);
3276                 b = ext4_block_bitmap(sb, gdp);
3277                 if (b >= first_block && b <= last_block) {
3278                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3279                         count++;
3280                 }
3281                 b = ext4_inode_bitmap(sb, gdp);
3282                 if (b >= first_block && b <= last_block) {
3283                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3284                         count++;
3285                 }
3286                 b = ext4_inode_table(sb, gdp);
3287                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3288                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3289                                 int c = EXT4_B2C(sbi, b - first_block);
3290                                 ext4_set_bit(c, buf);
3291                                 count++;
3292                         }
3293                 if (i != grp)
3294                         continue;
3295                 s = 0;
3296                 if (ext4_bg_has_super(sb, grp)) {
3297                         ext4_set_bit(s++, buf);
3298                         count++;
3299                 }
3300                 j = ext4_bg_num_gdb(sb, grp);
3301                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3302                         ext4_error(sb, "Invalid number of block group "
3303                                    "descriptor blocks: %d", j);
3304                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3305                 }
3306                 count += j;
3307                 for (; j > 0; j--)
3308                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3309         }
3310         if (!count)
3311                 return 0;
3312         return EXT4_CLUSTERS_PER_GROUP(sb) -
3313                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3314 }
3315
3316 /*
3317  * Compute the overhead and stash it in sbi->s_overhead
3318  */
3319 int ext4_calculate_overhead(struct super_block *sb)
3320 {
3321         struct ext4_sb_info *sbi = EXT4_SB(sb);
3322         struct ext4_super_block *es = sbi->s_es;
3323         struct inode *j_inode;
3324         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3325         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3326         ext4_fsblk_t overhead = 0;
3327         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3328
3329         if (!buf)
3330                 return -ENOMEM;
3331
3332         /*
3333          * Compute the overhead (FS structures).  This is constant
3334          * for a given filesystem unless the number of block groups
3335          * changes so we cache the previous value until it does.
3336          */
3337
3338         /*
3339          * All of the blocks before first_data_block are overhead
3340          */
3341         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3342
3343         /*
3344          * Add the overhead found in each block group
3345          */
3346         for (i = 0; i < ngroups; i++) {
3347                 int blks;
3348
3349                 blks = count_overhead(sb, i, buf);
3350                 overhead += blks;
3351                 if (blks)
3352                         memset(buf, 0, PAGE_SIZE);
3353                 cond_resched();
3354         }
3355
3356         /*
3357          * Add the internal journal blocks whether the journal has been
3358          * loaded or not
3359          */
3360         if (sbi->s_journal && !sbi->journal_bdev)
3361                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3362         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3363                 j_inode = ext4_get_journal_inode(sb, j_inum);
3364                 if (j_inode) {
3365                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3366                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3367                         iput(j_inode);
3368                 } else {
3369                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3370                 }
3371         }
3372         sbi->s_overhead = overhead;
3373         smp_wmb();
3374         free_page((unsigned long) buf);
3375         return 0;
3376 }
3377
3378 static void ext4_set_resv_clusters(struct super_block *sb)
3379 {
3380         ext4_fsblk_t resv_clusters;
3381         struct ext4_sb_info *sbi = EXT4_SB(sb);
3382
3383         /*
3384          * There's no need to reserve anything when we aren't using extents.
3385          * The space estimates are exact, there are no unwritten extents,
3386          * hole punching doesn't need new metadata... This is needed especially
3387          * to keep ext2/3 backward compatibility.
3388          */
3389         if (!ext4_has_feature_extents(sb))
3390                 return;
3391         /*
3392          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3393          * This should cover the situations where we can not afford to run
3394          * out of space like for example punch hole, or converting
3395          * unwritten extents in delalloc path. In most cases such
3396          * allocation would require 1, or 2 blocks, higher numbers are
3397          * very rare.
3398          */
3399         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3400                          sbi->s_cluster_bits);
3401
3402         do_div(resv_clusters, 50);
3403         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3404
3405         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3406 }
3407
3408 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3409 {
3410         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3411         char *orig_data = kstrdup(data, GFP_KERNEL);
3412         struct buffer_head *bh;
3413         struct ext4_super_block *es = NULL;
3414         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3415         ext4_fsblk_t block;
3416         ext4_fsblk_t sb_block = get_sb_block(&data);
3417         ext4_fsblk_t logical_sb_block;
3418         unsigned long offset = 0;
3419         unsigned long journal_devnum = 0;
3420         unsigned long def_mount_opts;
3421         struct inode *root;
3422         const char *descr;
3423         int ret = -ENOMEM;
3424         int blocksize, clustersize;
3425         unsigned int db_count;
3426         unsigned int i;
3427         int needs_recovery, has_huge_files, has_bigalloc;
3428         __u64 blocks_count;
3429         int err = 0;
3430         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3431         ext4_group_t first_not_zeroed;
3432
3433         if ((data && !orig_data) || !sbi)
3434                 goto out_free_base;
3435
3436         sbi->s_daxdev = dax_dev;
3437         sbi->s_blockgroup_lock =
3438                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3439         if (!sbi->s_blockgroup_lock)
3440                 goto out_free_base;
3441
3442         sb->s_fs_info = sbi;
3443         sbi->s_sb = sb;
3444         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3445         sbi->s_sb_block = sb_block;
3446         if (sb->s_bdev->bd_part)
3447                 sbi->s_sectors_written_start =
3448                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3449
3450         /* Cleanup superblock name */
3451         strreplace(sb->s_id, '/', '!');
3452
3453         /* -EINVAL is default */
3454         ret = -EINVAL;
3455         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3456         if (!blocksize) {
3457                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3458                 goto out_fail;
3459         }
3460
3461         /*
3462          * The ext4 superblock will not be buffer aligned for other than 1kB
3463          * block sizes.  We need to calculate the offset from buffer start.
3464          */
3465         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3466                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3467                 offset = do_div(logical_sb_block, blocksize);
3468         } else {
3469                 logical_sb_block = sb_block;
3470         }
3471
3472         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3473                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3474                 goto out_fail;
3475         }
3476         /*
3477          * Note: s_es must be initialized as soon as possible because
3478          *       some ext4 macro-instructions depend on its value
3479          */
3480         es = (struct ext4_super_block *) (bh->b_data + offset);
3481         sbi->s_es = es;
3482         sb->s_magic = le16_to_cpu(es->s_magic);
3483         if (sb->s_magic != EXT4_SUPER_MAGIC)
3484                 goto cantfind_ext4;
3485         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3486
3487         /* Warn if metadata_csum and gdt_csum are both set. */
3488         if (ext4_has_feature_metadata_csum(sb) &&
3489             ext4_has_feature_gdt_csum(sb))
3490                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3491                              "redundant flags; please run fsck.");
3492
3493         /* Check for a known checksum algorithm */
3494         if (!ext4_verify_csum_type(sb, es)) {
3495                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3496                          "unknown checksum algorithm.");
3497                 silent = 1;
3498                 goto cantfind_ext4;
3499         }
3500
3501         /* Load the checksum driver */
3502         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3503         if (IS_ERR(sbi->s_chksum_driver)) {
3504                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3505                 ret = PTR_ERR(sbi->s_chksum_driver);
3506                 sbi->s_chksum_driver = NULL;
3507                 goto failed_mount;
3508         }
3509
3510         /* Check superblock checksum */
3511         if (!ext4_superblock_csum_verify(sb, es)) {
3512                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3513                          "invalid superblock checksum.  Run e2fsck?");
3514                 silent = 1;
3515                 ret = -EFSBADCRC;
3516                 goto cantfind_ext4;
3517         }
3518
3519         /* Precompute checksum seed for all metadata */
3520         if (ext4_has_feature_csum_seed(sb))
3521                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3522         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3523                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3524                                                sizeof(es->s_uuid));
3525
3526         /* Set defaults before we parse the mount options */
3527         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3528         set_opt(sb, INIT_INODE_TABLE);
3529         if (def_mount_opts & EXT4_DEFM_DEBUG)
3530                 set_opt(sb, DEBUG);
3531         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3532                 set_opt(sb, GRPID);
3533         if (def_mount_opts & EXT4_DEFM_UID16)
3534                 set_opt(sb, NO_UID32);
3535         /* xattr user namespace & acls are now defaulted on */
3536         set_opt(sb, XATTR_USER);
3537 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3538         set_opt(sb, POSIX_ACL);
3539 #endif
3540         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3541         if (ext4_has_metadata_csum(sb))
3542                 set_opt(sb, JOURNAL_CHECKSUM);
3543
3544         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3545                 set_opt(sb, JOURNAL_DATA);
3546         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3547                 set_opt(sb, ORDERED_DATA);
3548         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3549                 set_opt(sb, WRITEBACK_DATA);
3550
3551         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3552                 set_opt(sb, ERRORS_PANIC);
3553         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3554                 set_opt(sb, ERRORS_CONT);
3555         else
3556                 set_opt(sb, ERRORS_RO);
3557         /* block_validity enabled by default; disable with noblock_validity */
3558         set_opt(sb, BLOCK_VALIDITY);
3559         if (def_mount_opts & EXT4_DEFM_DISCARD)
3560                 set_opt(sb, DISCARD);
3561
3562         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3563         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3564         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3565         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3566         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3567
3568         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3569                 set_opt(sb, BARRIER);
3570
3571         /*
3572          * enable delayed allocation by default
3573          * Use -o nodelalloc to turn it off
3574          */
3575         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3576             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3577                 set_opt(sb, DELALLOC);
3578
3579         /*
3580          * set default s_li_wait_mult for lazyinit, for the case there is
3581          * no mount option specified.
3582          */
3583         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3584
3585         if (sbi->s_es->s_mount_opts[0]) {
3586                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3587                                               sizeof(sbi->s_es->s_mount_opts),
3588                                               GFP_KERNEL);
3589                 if (!s_mount_opts)
3590                         goto failed_mount;
3591                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3592                                    &journal_ioprio, 0)) {
3593                         ext4_msg(sb, KERN_WARNING,
3594                                  "failed to parse options in superblock: %s",
3595                                  s_mount_opts);
3596                 }
3597                 kfree(s_mount_opts);
3598         }
3599         sbi->s_def_mount_opt = sbi->s_mount_opt;
3600         if (!parse_options((char *) data, sb, &journal_devnum,
3601                            &journal_ioprio, 0))
3602                 goto failed_mount;
3603
3604         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3605                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3606                             "with data=journal disables delayed "
3607                             "allocation and O_DIRECT support!\n");
3608                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3609                         ext4_msg(sb, KERN_ERR, "can't mount with "
3610                                  "both data=journal and delalloc");
3611                         goto failed_mount;
3612                 }
3613                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3614                         ext4_msg(sb, KERN_ERR, "can't mount with "
3615                                  "both data=journal and dioread_nolock");
3616                         goto failed_mount;
3617                 }
3618                 if (test_opt(sb, DAX)) {
3619                         ext4_msg(sb, KERN_ERR, "can't mount with "
3620                                  "both data=journal and dax");
3621                         goto failed_mount;
3622                 }
3623                 if (ext4_has_feature_encrypt(sb)) {
3624                         ext4_msg(sb, KERN_WARNING,
3625                                  "encrypted files will use data=ordered "
3626                                  "instead of data journaling mode");
3627                 }
3628                 if (test_opt(sb, DELALLOC))
3629                         clear_opt(sb, DELALLOC);
3630         } else {
3631                 sb->s_iflags |= SB_I_CGROUPWB;
3632         }
3633
3634         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3635                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3636
3637         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3638             (ext4_has_compat_features(sb) ||
3639              ext4_has_ro_compat_features(sb) ||
3640              ext4_has_incompat_features(sb)))
3641                 ext4_msg(sb, KERN_WARNING,
3642                        "feature flags set on rev 0 fs, "
3643                        "running e2fsck is recommended");
3644
3645         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3646                 set_opt2(sb, HURD_COMPAT);
3647                 if (ext4_has_feature_64bit(sb)) {
3648                         ext4_msg(sb, KERN_ERR,
3649                                  "The Hurd can't support 64-bit file systems");
3650                         goto failed_mount;
3651                 }
3652
3653                 /*
3654                  * ea_inode feature uses l_i_version field which is not
3655                  * available in HURD_COMPAT mode.
3656                  */
3657                 if (ext4_has_feature_ea_inode(sb)) {
3658                         ext4_msg(sb, KERN_ERR,
3659                                  "ea_inode feature is not supported for Hurd");
3660                         goto failed_mount;
3661                 }
3662         }
3663
3664         if (IS_EXT2_SB(sb)) {
3665                 if (ext2_feature_set_ok(sb))
3666                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3667                                  "using the ext4 subsystem");
3668                 else {
3669                         /*
3670                          * If we're probing be silent, if this looks like
3671                          * it's actually an ext[34] filesystem.
3672                          */
3673                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3674                                 goto failed_mount;
3675                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3676                                  "to feature incompatibilities");
3677                         goto failed_mount;
3678                 }
3679         }
3680
3681         if (IS_EXT3_SB(sb)) {
3682                 if (ext3_feature_set_ok(sb))
3683                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3684                                  "using the ext4 subsystem");
3685                 else {
3686                         /*
3687                          * If we're probing be silent, if this looks like
3688                          * it's actually an ext4 filesystem.
3689                          */
3690                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3691                                 goto failed_mount;
3692                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3693                                  "to feature incompatibilities");
3694                         goto failed_mount;
3695                 }
3696         }
3697
3698         /*
3699          * Check feature flags regardless of the revision level, since we
3700          * previously didn't change the revision level when setting the flags,
3701          * so there is a chance incompat flags are set on a rev 0 filesystem.
3702          */
3703         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3704                 goto failed_mount;
3705
3706         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3707         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3708             blocksize > EXT4_MAX_BLOCK_SIZE) {
3709                 ext4_msg(sb, KERN_ERR,
3710                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3711                          blocksize, le32_to_cpu(es->s_log_block_size));
3712                 goto failed_mount;
3713         }
3714         if (le32_to_cpu(es->s_log_block_size) >
3715             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3716                 ext4_msg(sb, KERN_ERR,
3717                          "Invalid log block size: %u",
3718                          le32_to_cpu(es->s_log_block_size));
3719                 goto failed_mount;
3720         }
3721
3722         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3723                 ext4_msg(sb, KERN_ERR,
3724                          "Number of reserved GDT blocks insanely large: %d",
3725                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3726                 goto failed_mount;
3727         }
3728
3729         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3730                 if (ext4_has_feature_inline_data(sb)) {
3731                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3732                                         " that may contain inline data");
3733                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3734                 }
3735                 err = bdev_dax_supported(sb, blocksize);
3736                 if (err) {
3737                         ext4_msg(sb, KERN_ERR,
3738                                 "DAX unsupported by block device. Turning off DAX.");
3739                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3740                 }
3741         }
3742
3743         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3744                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3745                          es->s_encryption_level);
3746                 goto failed_mount;
3747         }
3748
3749         if (sb->s_blocksize != blocksize) {
3750                 /* Validate the filesystem blocksize */
3751                 if (!sb_set_blocksize(sb, blocksize)) {
3752                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3753                                         blocksize);
3754                         goto failed_mount;
3755                 }
3756
3757                 brelse(bh);
3758                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3759                 offset = do_div(logical_sb_block, blocksize);
3760                 bh = sb_bread_unmovable(sb, logical_sb_block);
3761                 if (!bh) {
3762                         ext4_msg(sb, KERN_ERR,
3763                                "Can't read superblock on 2nd try");
3764                         goto failed_mount;
3765                 }
3766                 es = (struct ext4_super_block *)(bh->b_data + offset);
3767                 sbi->s_es = es;
3768                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3769                         ext4_msg(sb, KERN_ERR,
3770                                "Magic mismatch, very weird!");
3771                         goto failed_mount;
3772                 }
3773         }
3774
3775         has_huge_files = ext4_has_feature_huge_file(sb);
3776         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3777                                                       has_huge_files);
3778         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3779
3780         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3781                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3782                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3783         } else {
3784                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3785                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3786                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3787                     (!is_power_of_2(sbi->s_inode_size)) ||
3788                     (sbi->s_inode_size > blocksize)) {
3789                         ext4_msg(sb, KERN_ERR,
3790                                "unsupported inode size: %d",
3791                                sbi->s_inode_size);
3792                         goto failed_mount;
3793                 }
3794                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3795                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3796         }
3797
3798         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3799         if (ext4_has_feature_64bit(sb)) {
3800                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3801                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3802                     !is_power_of_2(sbi->s_desc_size)) {
3803                         ext4_msg(sb, KERN_ERR,
3804                                "unsupported descriptor size %lu",
3805                                sbi->s_desc_size);
3806                         goto failed_mount;
3807                 }
3808         } else
3809                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3810
3811         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3812         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3813
3814         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3815         if (sbi->s_inodes_per_block == 0)
3816                 goto cantfind_ext4;
3817         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3818             sbi->s_inodes_per_group > blocksize * 8) {
3819                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3820                          sbi->s_blocks_per_group);
3821                 goto failed_mount;
3822         }
3823         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3824                                         sbi->s_inodes_per_block;
3825         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3826         sbi->s_sbh = bh;
3827         sbi->s_mount_state = le16_to_cpu(es->s_state);
3828         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3829         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3830
3831         for (i = 0; i < 4; i++)
3832                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3833         sbi->s_def_hash_version = es->s_def_hash_version;
3834         if (ext4_has_feature_dir_index(sb)) {
3835                 i = le32_to_cpu(es->s_flags);
3836                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3837                         sbi->s_hash_unsigned = 3;
3838                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3839 #ifdef __CHAR_UNSIGNED__
3840                         if (!sb_rdonly(sb))
3841                                 es->s_flags |=
3842                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3843                         sbi->s_hash_unsigned = 3;
3844 #else
3845                         if (!sb_rdonly(sb))
3846                                 es->s_flags |=
3847                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3848 #endif
3849                 }
3850         }
3851
3852         /* Handle clustersize */
3853         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3854         has_bigalloc = ext4_has_feature_bigalloc(sb);
3855         if (has_bigalloc) {
3856                 if (clustersize < blocksize) {
3857                         ext4_msg(sb, KERN_ERR,
3858                                  "cluster size (%d) smaller than "
3859                                  "block size (%d)", clustersize, blocksize);
3860                         goto failed_mount;
3861                 }
3862                 if (le32_to_cpu(es->s_log_cluster_size) >
3863                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3864                         ext4_msg(sb, KERN_ERR,
3865                                  "Invalid log cluster size: %u",
3866                                  le32_to_cpu(es->s_log_cluster_size));
3867                         goto failed_mount;
3868                 }
3869                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3870                         le32_to_cpu(es->s_log_block_size);
3871                 sbi->s_clusters_per_group =
3872                         le32_to_cpu(es->s_clusters_per_group);
3873                 if (sbi->s_clusters_per_group > blocksize * 8) {
3874                         ext4_msg(sb, KERN_ERR,
3875                                  "#clusters per group too big: %lu",
3876                                  sbi->s_clusters_per_group);
3877                         goto failed_mount;
3878                 }
3879                 if (sbi->s_blocks_per_group !=
3880                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3881                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3882                                  "clusters per group (%lu) inconsistent",
3883                                  sbi->s_blocks_per_group,
3884                                  sbi->s_clusters_per_group);
3885                         goto failed_mount;
3886                 }
3887         } else {
3888                 if (clustersize != blocksize) {
3889                         ext4_warning(sb, "fragment/cluster size (%d) != "
3890                                      "block size (%d)", clustersize,
3891                                      blocksize);
3892                         clustersize = blocksize;
3893                 }
3894                 if (sbi->s_blocks_per_group > blocksize * 8) {
3895                         ext4_msg(sb, KERN_ERR,
3896                                  "#blocks per group too big: %lu",
3897                                  sbi->s_blocks_per_group);
3898                         goto failed_mount;
3899                 }
3900                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3901                 sbi->s_cluster_bits = 0;
3902         }
3903         sbi->s_cluster_ratio = clustersize / blocksize;
3904
3905         /* Do we have standard group size of clustersize * 8 blocks ? */
3906         if (sbi->s_blocks_per_group == clustersize << 3)
3907                 set_opt2(sb, STD_GROUP_SIZE);
3908
3909         /*
3910          * Test whether we have more sectors than will fit in sector_t,
3911          * and whether the max offset is addressable by the page cache.
3912          */
3913         err = generic_check_addressable(sb->s_blocksize_bits,
3914                                         ext4_blocks_count(es));
3915         if (err) {
3916                 ext4_msg(sb, KERN_ERR, "filesystem"
3917                          " too large to mount safely on this system");
3918                 if (sizeof(sector_t) < 8)
3919                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3920                 goto failed_mount;
3921         }
3922
3923         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3924                 goto cantfind_ext4;
3925
3926         /* check blocks count against device size */
3927         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3928         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3929                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3930                        "exceeds size of device (%llu blocks)",
3931                        ext4_blocks_count(es), blocks_count);
3932                 goto failed_mount;
3933         }
3934
3935         /*
3936          * It makes no sense for the first data block to be beyond the end
3937          * of the filesystem.
3938          */
3939         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3940                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3941                          "block %u is beyond end of filesystem (%llu)",
3942                          le32_to_cpu(es->s_first_data_block),
3943                          ext4_blocks_count(es));
3944                 goto failed_mount;
3945         }
3946         blocks_count = (ext4_blocks_count(es) -
3947                         le32_to_cpu(es->s_first_data_block) +
3948                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3949         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3950         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3951                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3952                        "(block count %llu, first data block %u, "
3953                        "blocks per group %lu)", sbi->s_groups_count,
3954                        ext4_blocks_count(es),
3955                        le32_to_cpu(es->s_first_data_block),
3956                        EXT4_BLOCKS_PER_GROUP(sb));
3957                 goto failed_mount;
3958         }
3959         sbi->s_groups_count = blocks_count;
3960         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3961                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3962         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3963                    EXT4_DESC_PER_BLOCK(sb);
3964         if (ext4_has_feature_meta_bg(sb)) {
3965                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3966                         ext4_msg(sb, KERN_WARNING,
3967                                  "first meta block group too large: %u "
3968                                  "(group descriptor block count %u)",
3969                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3970                         goto failed_mount;
3971                 }
3972         }
3973         sbi->s_group_desc = kvmalloc(db_count *
3974                                           sizeof(struct buffer_head *),
3975                                           GFP_KERNEL);
3976         if (sbi->s_group_desc == NULL) {
3977                 ext4_msg(sb, KERN_ERR, "not enough memory");
3978                 ret = -ENOMEM;
3979                 goto failed_mount;
3980         }
3981
3982         bgl_lock_init(sbi->s_blockgroup_lock);
3983
3984         /* Pre-read the descriptors into the buffer cache */
3985         for (i = 0; i < db_count; i++) {
3986                 block = descriptor_loc(sb, logical_sb_block, i);
3987                 sb_breadahead(sb, block);
3988         }
3989
3990         for (i = 0; i < db_count; i++) {
3991                 block = descriptor_loc(sb, logical_sb_block, i);
3992                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3993                 if (!sbi->s_group_desc[i]) {
3994                         ext4_msg(sb, KERN_ERR,
3995                                "can't read group descriptor %d", i);
3996                         db_count = i;
3997                         goto failed_mount2;
3998                 }
3999         }
4000         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4001                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4002                 ret = -EFSCORRUPTED;
4003                 goto failed_mount2;
4004         }
4005
4006         sbi->s_gdb_count = db_count;
4007
4008         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4009
4010         /* Register extent status tree shrinker */
4011         if (ext4_es_register_shrinker(sbi))
4012                 goto failed_mount3;
4013
4014         sbi->s_stripe = ext4_get_stripe_size(sbi);
4015         sbi->s_extent_max_zeroout_kb = 32;
4016
4017         /*
4018          * set up enough so that it can read an inode
4019          */
4020         sb->s_op = &ext4_sops;
4021         sb->s_export_op = &ext4_export_ops;
4022         sb->s_xattr = ext4_xattr_handlers;
4023 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4024         sb->s_cop = &ext4_cryptops;
4025 #endif
4026 #ifdef CONFIG_QUOTA
4027         sb->dq_op = &ext4_quota_operations;
4028         if (ext4_has_feature_quota(sb))
4029                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4030         else
4031                 sb->s_qcop = &ext4_qctl_operations;
4032         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4033 #endif
4034         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4035
4036         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4037         mutex_init(&sbi->s_orphan_lock);
4038
4039         sb->s_root = NULL;
4040
4041         needs_recovery = (es->s_last_orphan != 0 ||
4042                           ext4_has_feature_journal_needs_recovery(sb));
4043
4044         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4045                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4046                         goto failed_mount3a;
4047
4048         /*
4049          * The first inode we look at is the journal inode.  Don't try
4050          * root first: it may be modified in the journal!
4051          */
4052         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4053                 err = ext4_load_journal(sb, es, journal_devnum);
4054                 if (err)
4055                         goto failed_mount3a;
4056         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4057                    ext4_has_feature_journal_needs_recovery(sb)) {
4058                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4059                        "suppressed and not mounted read-only");
4060                 goto failed_mount_wq;
4061         } else {
4062                 /* Nojournal mode, all journal mount options are illegal */
4063                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4064                         ext4_msg(sb, KERN_ERR, "can't mount with "
4065                                  "journal_checksum, fs mounted w/o journal");
4066                         goto failed_mount_wq;
4067                 }
4068                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4069                         ext4_msg(sb, KERN_ERR, "can't mount with "
4070                                  "journal_async_commit, fs mounted w/o journal");
4071                         goto failed_mount_wq;
4072                 }
4073                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4074                         ext4_msg(sb, KERN_ERR, "can't mount with "
4075                                  "commit=%lu, fs mounted w/o journal",
4076                                  sbi->s_commit_interval / HZ);
4077                         goto failed_mount_wq;
4078                 }
4079                 if (EXT4_MOUNT_DATA_FLAGS &
4080                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4081                         ext4_msg(sb, KERN_ERR, "can't mount with "
4082                                  "data=, fs mounted w/o journal");
4083                         goto failed_mount_wq;
4084                 }
4085                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4086                 clear_opt(sb, JOURNAL_CHECKSUM);
4087                 clear_opt(sb, DATA_FLAGS);
4088                 sbi->s_journal = NULL;
4089                 needs_recovery = 0;
4090                 goto no_journal;
4091         }
4092
4093         if (ext4_has_feature_64bit(sb) &&
4094             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4095                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4096                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4097                 goto failed_mount_wq;
4098         }
4099
4100         if (!set_journal_csum_feature_set(sb)) {
4101                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4102                          "feature set");
4103                 goto failed_mount_wq;
4104         }
4105
4106         /* We have now updated the journal if required, so we can
4107          * validate the data journaling mode. */
4108         switch (test_opt(sb, DATA_FLAGS)) {
4109         case 0:
4110                 /* No mode set, assume a default based on the journal
4111                  * capabilities: ORDERED_DATA if the journal can
4112                  * cope, else JOURNAL_DATA
4113                  */
4114                 if (jbd2_journal_check_available_features
4115                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4116                         set_opt(sb, ORDERED_DATA);
4117                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4118                 } else {
4119                         set_opt(sb, JOURNAL_DATA);
4120                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4121                 }
4122                 break;
4123
4124         case EXT4_MOUNT_ORDERED_DATA:
4125         case EXT4_MOUNT_WRITEBACK_DATA:
4126                 if (!jbd2_journal_check_available_features
4127                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4128                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4129                                "requested data journaling mode");
4130                         goto failed_mount_wq;
4131                 }
4132         default:
4133                 break;
4134         }
4135
4136         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4137             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4138                 ext4_msg(sb, KERN_ERR, "can't mount with "
4139                         "journal_async_commit in data=ordered mode");
4140                 goto failed_mount_wq;
4141         }
4142
4143         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4144
4145         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4146
4147 no_journal:
4148         if (!test_opt(sb, NO_MBCACHE)) {
4149                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4150                 if (!sbi->s_ea_block_cache) {
4151                         ext4_msg(sb, KERN_ERR,
4152                                  "Failed to create ea_block_cache");
4153                         goto failed_mount_wq;
4154                 }
4155
4156                 if (ext4_has_feature_ea_inode(sb)) {
4157                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4158                         if (!sbi->s_ea_inode_cache) {
4159                                 ext4_msg(sb, KERN_ERR,
4160                                          "Failed to create ea_inode_cache");
4161                                 goto failed_mount_wq;
4162                         }
4163                 }
4164         }
4165
4166         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4167             (blocksize != PAGE_SIZE)) {
4168                 ext4_msg(sb, KERN_ERR,
4169                          "Unsupported blocksize for fs encryption");
4170                 goto failed_mount_wq;
4171         }
4172
4173         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4174             !ext4_has_feature_encrypt(sb)) {
4175                 ext4_set_feature_encrypt(sb);
4176                 ext4_commit_super(sb, 1);
4177         }
4178
4179         /*
4180          * Get the # of file system overhead blocks from the
4181          * superblock if present.
4182          */
4183         if (es->s_overhead_clusters)
4184                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4185         else {
4186                 err = ext4_calculate_overhead(sb);
4187                 if (err)
4188                         goto failed_mount_wq;
4189         }
4190
4191         /*
4192          * The maximum number of concurrent works can be high and
4193          * concurrency isn't really necessary.  Limit it to 1.
4194          */
4195         EXT4_SB(sb)->rsv_conversion_wq =
4196                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4197         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4198                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4199                 ret = -ENOMEM;
4200                 goto failed_mount4;
4201         }
4202
4203         /*
4204          * The jbd2_journal_load will have done any necessary log recovery,
4205          * so we can safely mount the rest of the filesystem now.
4206          */
4207
4208         root = ext4_iget(sb, EXT4_ROOT_INO);
4209         if (IS_ERR(root)) {
4210                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4211                 ret = PTR_ERR(root);
4212                 root = NULL;
4213                 goto failed_mount4;
4214         }
4215         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4216                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4217                 iput(root);
4218                 goto failed_mount4;
4219         }
4220         sb->s_root = d_make_root(root);
4221         if (!sb->s_root) {
4222                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4223                 ret = -ENOMEM;
4224                 goto failed_mount4;
4225         }
4226
4227         if (ext4_setup_super(sb, es, sb_rdonly(sb)))
4228                 sb->s_flags |= SB_RDONLY;
4229
4230         /* determine the minimum size of new large inodes, if present */
4231         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4232             sbi->s_want_extra_isize == 0) {
4233                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4234                                                      EXT4_GOOD_OLD_INODE_SIZE;
4235                 if (ext4_has_feature_extra_isize(sb)) {
4236                         if (sbi->s_want_extra_isize <
4237                             le16_to_cpu(es->s_want_extra_isize))
4238                                 sbi->s_want_extra_isize =
4239                                         le16_to_cpu(es->s_want_extra_isize);
4240                         if (sbi->s_want_extra_isize <
4241                             le16_to_cpu(es->s_min_extra_isize))
4242                                 sbi->s_want_extra_isize =
4243                                         le16_to_cpu(es->s_min_extra_isize);
4244                 }
4245         }
4246         /* Check if enough inode space is available */
4247         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4248                                                         sbi->s_inode_size) {
4249                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4250                                                        EXT4_GOOD_OLD_INODE_SIZE;
4251                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4252                          "available");
4253         }
4254
4255         ext4_set_resv_clusters(sb);
4256
4257         err = ext4_setup_system_zone(sb);
4258         if (err) {
4259                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4260                          "zone (%d)", err);
4261                 goto failed_mount4a;
4262         }
4263
4264         ext4_ext_init(sb);
4265         err = ext4_mb_init(sb);
4266         if (err) {
4267                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4268                          err);
4269                 goto failed_mount5;
4270         }
4271
4272         block = ext4_count_free_clusters(sb);
4273         ext4_free_blocks_count_set(sbi->s_es,
4274                                    EXT4_C2B(sbi, block));
4275         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4276                                   GFP_KERNEL);
4277         if (!err) {
4278                 unsigned long freei = ext4_count_free_inodes(sb);
4279                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4280                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4281                                           GFP_KERNEL);
4282         }
4283         if (!err)
4284                 err = percpu_counter_init(&sbi->s_dirs_counter,
4285                                           ext4_count_dirs(sb), GFP_KERNEL);
4286         if (!err)
4287                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4288                                           GFP_KERNEL);
4289         if (!err)
4290                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4291
4292         if (err) {
4293                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4294                 goto failed_mount6;
4295         }
4296
4297         if (ext4_has_feature_flex_bg(sb))
4298                 if (!ext4_fill_flex_info(sb)) {
4299                         ext4_msg(sb, KERN_ERR,
4300                                "unable to initialize "
4301                                "flex_bg meta info!");
4302                         goto failed_mount6;
4303                 }
4304
4305         err = ext4_register_li_request(sb, first_not_zeroed);
4306         if (err)
4307                 goto failed_mount6;
4308
4309         err = ext4_register_sysfs(sb);
4310         if (err)
4311                 goto failed_mount7;
4312
4313 #ifdef CONFIG_QUOTA
4314         /* Enable quota usage during mount. */
4315         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4316                 err = ext4_enable_quotas(sb);
4317                 if (err)
4318                         goto failed_mount8;
4319         }
4320 #endif  /* CONFIG_QUOTA */
4321
4322         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4323         ext4_orphan_cleanup(sb, es);
4324         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4325         if (needs_recovery) {
4326                 ext4_msg(sb, KERN_INFO, "recovery complete");
4327                 ext4_mark_recovery_complete(sb, es);
4328         }
4329         if (EXT4_SB(sb)->s_journal) {
4330                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4331                         descr = " journalled data mode";
4332                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4333                         descr = " ordered data mode";
4334                 else
4335                         descr = " writeback data mode";
4336         } else
4337                 descr = "out journal";
4338
4339         if (test_opt(sb, DISCARD)) {
4340                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4341                 if (!blk_queue_discard(q))
4342                         ext4_msg(sb, KERN_WARNING,
4343                                  "mounting with \"discard\" option, but "
4344                                  "the device does not support discard");
4345         }
4346
4347         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4348                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4349                          "Opts: %.*s%s%s", descr,
4350                          (int) sizeof(sbi->s_es->s_mount_opts),
4351                          sbi->s_es->s_mount_opts,
4352                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4353
4354         if (es->s_error_count)
4355                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4356
4357         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4358         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4359         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4360         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4361
4362         kfree(orig_data);
4363         return 0;
4364
4365 cantfind_ext4:
4366         if (!silent)
4367                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4368         goto failed_mount;
4369
4370 #ifdef CONFIG_QUOTA
4371 failed_mount8:
4372         ext4_unregister_sysfs(sb);
4373 #endif
4374 failed_mount7:
4375         ext4_unregister_li_request(sb);
4376 failed_mount6:
4377         ext4_mb_release(sb);
4378         if (sbi->s_flex_groups)
4379                 kvfree(sbi->s_flex_groups);
4380         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4381         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4382         percpu_counter_destroy(&sbi->s_dirs_counter);
4383         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4384 failed_mount5:
4385         ext4_ext_release(sb);
4386         ext4_release_system_zone(sb);
4387 failed_mount4a:
4388         dput(sb->s_root);
4389         sb->s_root = NULL;
4390 failed_mount4:
4391         ext4_msg(sb, KERN_ERR, "mount failed");
4392         if (EXT4_SB(sb)->rsv_conversion_wq)
4393                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4394 failed_mount_wq:
4395         if (sbi->s_ea_inode_cache) {
4396                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4397                 sbi->s_ea_inode_cache = NULL;
4398         }
4399         if (sbi->s_ea_block_cache) {
4400                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4401                 sbi->s_ea_block_cache = NULL;
4402         }
4403         if (sbi->s_journal) {
4404                 jbd2_journal_destroy(sbi->s_journal);
4405                 sbi->s_journal = NULL;
4406         }
4407 failed_mount3a:
4408         ext4_es_unregister_shrinker(sbi);
4409 failed_mount3:
4410         del_timer_sync(&sbi->s_err_report);
4411         if (sbi->s_mmp_tsk)
4412                 kthread_stop(sbi->s_mmp_tsk);
4413 failed_mount2:
4414         for (i = 0; i < db_count; i++)
4415                 brelse(sbi->s_group_desc[i]);
4416         kvfree(sbi->s_group_desc);
4417 failed_mount:
4418         if (sbi->s_chksum_driver)
4419                 crypto_free_shash(sbi->s_chksum_driver);
4420 #ifdef CONFIG_QUOTA
4421         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4422                 kfree(sbi->s_qf_names[i]);
4423 #endif
4424         ext4_blkdev_remove(sbi);
4425         brelse(bh);
4426 out_fail:
4427         sb->s_fs_info = NULL;
4428         kfree(sbi->s_blockgroup_lock);
4429 out_free_base:
4430         kfree(sbi);
4431         kfree(orig_data);
4432         fs_put_dax(dax_dev);
4433         return err ? err : ret;
4434 }
4435
4436 /*
4437  * Setup any per-fs journal parameters now.  We'll do this both on
4438  * initial mount, once the journal has been initialised but before we've
4439  * done any recovery; and again on any subsequent remount.
4440  */
4441 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4442 {
4443         struct ext4_sb_info *sbi = EXT4_SB(sb);
4444
4445         journal->j_commit_interval = sbi->s_commit_interval;
4446         journal->j_min_batch_time = sbi->s_min_batch_time;
4447         journal->j_max_batch_time = sbi->s_max_batch_time;
4448
4449         write_lock(&journal->j_state_lock);
4450         if (test_opt(sb, BARRIER))
4451                 journal->j_flags |= JBD2_BARRIER;
4452         else
4453                 journal->j_flags &= ~JBD2_BARRIER;
4454         if (test_opt(sb, DATA_ERR_ABORT))
4455                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4456         else
4457                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4458         write_unlock(&journal->j_state_lock);
4459 }
4460
4461 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4462                                              unsigned int journal_inum)
4463 {
4464         struct inode *journal_inode;
4465
4466         /*
4467          * Test for the existence of a valid inode on disk.  Bad things
4468          * happen if we iget() an unused inode, as the subsequent iput()
4469          * will try to delete it.
4470          */
4471         journal_inode = ext4_iget(sb, journal_inum);
4472         if (IS_ERR(journal_inode)) {
4473                 ext4_msg(sb, KERN_ERR, "no journal found");
4474                 return NULL;
4475         }
4476         if (!journal_inode->i_nlink) {
4477                 make_bad_inode(journal_inode);
4478                 iput(journal_inode);
4479                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4480                 return NULL;
4481         }
4482
4483         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4484                   journal_inode, journal_inode->i_size);
4485         if (!S_ISREG(journal_inode->i_mode)) {
4486                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4487                 iput(journal_inode);
4488                 return NULL;
4489         }
4490         return journal_inode;
4491 }
4492
4493 static journal_t *ext4_get_journal(struct super_block *sb,
4494                                    unsigned int journal_inum)
4495 {
4496         struct inode *journal_inode;
4497         journal_t *journal;
4498
4499         BUG_ON(!ext4_has_feature_journal(sb));
4500
4501         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4502         if (!journal_inode)
4503                 return NULL;
4504
4505         journal = jbd2_journal_init_inode(journal_inode);
4506         if (!journal) {
4507                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4508                 iput(journal_inode);
4509                 return NULL;
4510         }
4511         journal->j_private = sb;
4512         ext4_init_journal_params(sb, journal);
4513         return journal;
4514 }
4515
4516 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4517                                        dev_t j_dev)
4518 {
4519         struct buffer_head *bh;
4520         journal_t *journal;
4521         ext4_fsblk_t start;
4522         ext4_fsblk_t len;
4523         int hblock, blocksize;
4524         ext4_fsblk_t sb_block;
4525         unsigned long offset;
4526         struct ext4_super_block *es;
4527         struct block_device *bdev;
4528
4529         BUG_ON(!ext4_has_feature_journal(sb));
4530
4531         bdev = ext4_blkdev_get(j_dev, sb);
4532         if (bdev == NULL)
4533                 return NULL;
4534
4535         blocksize = sb->s_blocksize;
4536         hblock = bdev_logical_block_size(bdev);
4537         if (blocksize < hblock) {
4538                 ext4_msg(sb, KERN_ERR,
4539                         "blocksize too small for journal device");
4540                 goto out_bdev;
4541         }
4542
4543         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4544         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4545         set_blocksize(bdev, blocksize);
4546         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4547                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4548                        "external journal");
4549                 goto out_bdev;
4550         }
4551
4552         es = (struct ext4_super_block *) (bh->b_data + offset);
4553         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4554             !(le32_to_cpu(es->s_feature_incompat) &
4555               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4556                 ext4_msg(sb, KERN_ERR, "external journal has "
4557                                         "bad superblock");
4558                 brelse(bh);
4559                 goto out_bdev;
4560         }
4561
4562         if ((le32_to_cpu(es->s_feature_ro_compat) &
4563              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4564             es->s_checksum != ext4_superblock_csum(sb, es)) {
4565                 ext4_msg(sb, KERN_ERR, "external journal has "
4566                                        "corrupt superblock");
4567                 brelse(bh);
4568                 goto out_bdev;
4569         }
4570
4571         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4572                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4573                 brelse(bh);
4574                 goto out_bdev;
4575         }
4576
4577         len = ext4_blocks_count(es);
4578         start = sb_block + 1;
4579         brelse(bh);     /* we're done with the superblock */
4580
4581         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4582                                         start, len, blocksize);
4583         if (!journal) {
4584                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4585                 goto out_bdev;
4586         }
4587         journal->j_private = sb;
4588         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4589         wait_on_buffer(journal->j_sb_buffer);
4590         if (!buffer_uptodate(journal->j_sb_buffer)) {
4591                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4592                 goto out_journal;
4593         }
4594         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4595                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4596                                         "user (unsupported) - %d",
4597                         be32_to_cpu(journal->j_superblock->s_nr_users));
4598                 goto out_journal;
4599         }
4600         EXT4_SB(sb)->journal_bdev = bdev;
4601         ext4_init_journal_params(sb, journal);
4602         return journal;
4603
4604 out_journal:
4605         jbd2_journal_destroy(journal);
4606 out_bdev:
4607         ext4_blkdev_put(bdev);
4608         return NULL;
4609 }
4610
4611 static int ext4_load_journal(struct super_block *sb,
4612                              struct ext4_super_block *es,
4613                              unsigned long journal_devnum)
4614 {
4615         journal_t *journal;
4616         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4617         dev_t journal_dev;
4618         int err = 0;
4619         int really_read_only;
4620
4621         BUG_ON(!ext4_has_feature_journal(sb));
4622
4623         if (journal_devnum &&
4624             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4625                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4626                         "numbers have changed");
4627                 journal_dev = new_decode_dev(journal_devnum);
4628         } else
4629                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4630
4631         really_read_only = bdev_read_only(sb->s_bdev);
4632
4633         /*
4634          * Are we loading a blank journal or performing recovery after a
4635          * crash?  For recovery, we need to check in advance whether we
4636          * can get read-write access to the device.
4637          */
4638         if (ext4_has_feature_journal_needs_recovery(sb)) {
4639                 if (sb_rdonly(sb)) {
4640                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4641                                         "required on readonly filesystem");
4642                         if (really_read_only) {
4643                                 ext4_msg(sb, KERN_ERR, "write access "
4644                                         "unavailable, cannot proceed "
4645                                         "(try mounting with noload)");
4646                                 return -EROFS;
4647                         }
4648                         ext4_msg(sb, KERN_INFO, "write access will "
4649                                "be enabled during recovery");
4650                 }
4651         }
4652
4653         if (journal_inum && journal_dev) {
4654                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4655                        "and inode journals!");
4656                 return -EINVAL;
4657         }
4658
4659         if (journal_inum) {
4660                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4661                         return -EINVAL;
4662         } else {
4663                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4664                         return -EINVAL;
4665         }
4666
4667         if (!(journal->j_flags & JBD2_BARRIER))
4668                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4669
4670         if (!ext4_has_feature_journal_needs_recovery(sb))
4671                 err = jbd2_journal_wipe(journal, !really_read_only);
4672         if (!err) {
4673                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4674                 if (save)
4675                         memcpy(save, ((char *) es) +
4676                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4677                 err = jbd2_journal_load(journal);
4678                 if (save)
4679                         memcpy(((char *) es) + EXT4_S_ERR_START,
4680                                save, EXT4_S_ERR_LEN);
4681                 kfree(save);
4682         }
4683
4684         if (err) {
4685                 ext4_msg(sb, KERN_ERR, "error loading journal");
4686                 jbd2_journal_destroy(journal);
4687                 return err;
4688         }
4689
4690         EXT4_SB(sb)->s_journal = journal;
4691         ext4_clear_journal_err(sb, es);
4692
4693         if (!really_read_only && journal_devnum &&
4694             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4695                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4696
4697                 /* Make sure we flush the recovery flag to disk. */
4698                 ext4_commit_super(sb, 1);
4699         }
4700
4701         return 0;
4702 }
4703
4704 static int ext4_commit_super(struct super_block *sb, int sync)
4705 {
4706         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4707         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4708         int error = 0;
4709
4710         if (!sbh || block_device_ejected(sb))
4711                 return error;
4712         /*
4713          * If the file system is mounted read-only, don't update the
4714          * superblock write time.  This avoids updating the superblock
4715          * write time when we are mounting the root file system
4716          * read/only but we need to replay the journal; at that point,
4717          * for people who are east of GMT and who make their clock
4718          * tick in localtime for Windows bug-for-bug compatibility,
4719          * the clock is set in the future, and this will cause e2fsck
4720          * to complain and force a full file system check.
4721          */
4722         if (!(sb->s_flags & SB_RDONLY))
4723                 es->s_wtime = cpu_to_le32(get_seconds());
4724         if (sb->s_bdev->bd_part)
4725                 es->s_kbytes_written =
4726                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4727                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4728                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4729         else
4730                 es->s_kbytes_written =
4731                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4732         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4733                 ext4_free_blocks_count_set(es,
4734                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4735                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4736         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4737                 es->s_free_inodes_count =
4738                         cpu_to_le32(percpu_counter_sum_positive(
4739                                 &EXT4_SB(sb)->s_freeinodes_counter));
4740         BUFFER_TRACE(sbh, "marking dirty");
4741         ext4_superblock_csum_set(sb);
4742         if (sync)
4743                 lock_buffer(sbh);
4744         if (buffer_write_io_error(sbh)) {
4745                 /*
4746                  * Oh, dear.  A previous attempt to write the
4747                  * superblock failed.  This could happen because the
4748                  * USB device was yanked out.  Or it could happen to
4749                  * be a transient write error and maybe the block will
4750                  * be remapped.  Nothing we can do but to retry the
4751                  * write and hope for the best.
4752                  */
4753                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4754                        "superblock detected");
4755                 clear_buffer_write_io_error(sbh);
4756                 set_buffer_uptodate(sbh);
4757         }
4758         mark_buffer_dirty(sbh);
4759         if (sync) {
4760                 unlock_buffer(sbh);
4761                 error = __sync_dirty_buffer(sbh,
4762                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4763                 if (error)
4764                         return error;
4765
4766                 error = buffer_write_io_error(sbh);
4767                 if (error) {
4768                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4769                                "superblock");
4770                         clear_buffer_write_io_error(sbh);
4771                         set_buffer_uptodate(sbh);
4772                 }
4773         }
4774         return error;
4775 }
4776
4777 /*
4778  * Have we just finished recovery?  If so, and if we are mounting (or
4779  * remounting) the filesystem readonly, then we will end up with a
4780  * consistent fs on disk.  Record that fact.
4781  */
4782 static void ext4_mark_recovery_complete(struct super_block *sb,
4783                                         struct ext4_super_block *es)
4784 {
4785         journal_t *journal = EXT4_SB(sb)->s_journal;
4786
4787         if (!ext4_has_feature_journal(sb)) {
4788                 BUG_ON(journal != NULL);
4789                 return;
4790         }
4791         jbd2_journal_lock_updates(journal);
4792         if (jbd2_journal_flush(journal) < 0)
4793                 goto out;
4794
4795         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4796                 ext4_clear_feature_journal_needs_recovery(sb);
4797                 ext4_commit_super(sb, 1);
4798         }
4799
4800 out:
4801         jbd2_journal_unlock_updates(journal);
4802 }
4803
4804 /*
4805  * If we are mounting (or read-write remounting) a filesystem whose journal
4806  * has recorded an error from a previous lifetime, move that error to the
4807  * main filesystem now.
4808  */
4809 static void ext4_clear_journal_err(struct super_block *sb,
4810                                    struct ext4_super_block *es)
4811 {
4812         journal_t *journal;
4813         int j_errno;
4814         const char *errstr;
4815
4816         BUG_ON(!ext4_has_feature_journal(sb));
4817
4818         journal = EXT4_SB(sb)->s_journal;
4819
4820         /*
4821          * Now check for any error status which may have been recorded in the
4822          * journal by a prior ext4_error() or ext4_abort()
4823          */
4824
4825         j_errno = jbd2_journal_errno(journal);
4826         if (j_errno) {
4827                 char nbuf[16];
4828
4829                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4830                 ext4_warning(sb, "Filesystem error recorded "
4831                              "from previous mount: %s", errstr);
4832                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4833
4834                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4835                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4836                 ext4_commit_super(sb, 1);
4837
4838                 jbd2_journal_clear_err(journal);
4839                 jbd2_journal_update_sb_errno(journal);
4840         }
4841 }
4842
4843 /*
4844  * Force the running and committing transactions to commit,
4845  * and wait on the commit.
4846  */
4847 int ext4_force_commit(struct super_block *sb)
4848 {
4849         journal_t *journal;
4850
4851         if (sb_rdonly(sb))
4852                 return 0;
4853
4854         journal = EXT4_SB(sb)->s_journal;
4855         return ext4_journal_force_commit(journal);
4856 }
4857
4858 static int ext4_sync_fs(struct super_block *sb, int wait)
4859 {
4860         int ret = 0;
4861         tid_t target;
4862         bool needs_barrier = false;
4863         struct ext4_sb_info *sbi = EXT4_SB(sb);
4864
4865         if (unlikely(ext4_forced_shutdown(sbi)))
4866                 return 0;
4867
4868         trace_ext4_sync_fs(sb, wait);
4869         flush_workqueue(sbi->rsv_conversion_wq);
4870         /*
4871          * Writeback quota in non-journalled quota case - journalled quota has
4872          * no dirty dquots
4873          */
4874         dquot_writeback_dquots(sb, -1);
4875         /*
4876          * Data writeback is possible w/o journal transaction, so barrier must
4877          * being sent at the end of the function. But we can skip it if
4878          * transaction_commit will do it for us.
4879          */
4880         if (sbi->s_journal) {
4881                 target = jbd2_get_latest_transaction(sbi->s_journal);
4882                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4883                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4884                         needs_barrier = true;
4885
4886                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4887                         if (wait)
4888                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4889                                                            target);
4890                 }
4891         } else if (wait && test_opt(sb, BARRIER))
4892                 needs_barrier = true;
4893         if (needs_barrier) {
4894                 int err;
4895                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4896                 if (!ret)
4897                         ret = err;
4898         }
4899
4900         return ret;
4901 }
4902
4903 /*
4904  * LVM calls this function before a (read-only) snapshot is created.  This
4905  * gives us a chance to flush the journal completely and mark the fs clean.
4906  *
4907  * Note that only this function cannot bring a filesystem to be in a clean
4908  * state independently. It relies on upper layer to stop all data & metadata
4909  * modifications.
4910  */
4911 static int ext4_freeze(struct super_block *sb)
4912 {
4913         int error = 0;
4914         journal_t *journal;
4915
4916         if (sb_rdonly(sb))
4917                 return 0;
4918
4919         journal = EXT4_SB(sb)->s_journal;
4920
4921         if (journal) {
4922                 /* Now we set up the journal barrier. */
4923                 jbd2_journal_lock_updates(journal);
4924
4925                 /*
4926                  * Don't clear the needs_recovery flag if we failed to
4927                  * flush the journal.
4928                  */
4929                 error = jbd2_journal_flush(journal);
4930                 if (error < 0)
4931                         goto out;
4932
4933                 /* Journal blocked and flushed, clear needs_recovery flag. */
4934                 ext4_clear_feature_journal_needs_recovery(sb);
4935         }
4936
4937         error = ext4_commit_super(sb, 1);
4938 out:
4939         if (journal)
4940                 /* we rely on upper layer to stop further updates */
4941                 jbd2_journal_unlock_updates(journal);
4942         return error;
4943 }
4944
4945 /*
4946  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4947  * flag here, even though the filesystem is not technically dirty yet.
4948  */
4949 static int ext4_unfreeze(struct super_block *sb)
4950 {
4951         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
4952                 return 0;
4953
4954         if (EXT4_SB(sb)->s_journal) {
4955                 /* Reset the needs_recovery flag before the fs is unlocked. */
4956                 ext4_set_feature_journal_needs_recovery(sb);
4957         }
4958
4959         ext4_commit_super(sb, 1);
4960         return 0;
4961 }
4962
4963 /*
4964  * Structure to save mount options for ext4_remount's benefit
4965  */
4966 struct ext4_mount_options {
4967         unsigned long s_mount_opt;
4968         unsigned long s_mount_opt2;
4969         kuid_t s_resuid;
4970         kgid_t s_resgid;
4971         unsigned long s_commit_interval;
4972         u32 s_min_batch_time, s_max_batch_time;
4973 #ifdef CONFIG_QUOTA
4974         int s_jquota_fmt;
4975         char *s_qf_names[EXT4_MAXQUOTAS];
4976 #endif
4977 };
4978
4979 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4980 {
4981         struct ext4_super_block *es;
4982         struct ext4_sb_info *sbi = EXT4_SB(sb);
4983         unsigned long old_sb_flags;
4984         struct ext4_mount_options old_opts;
4985         int enable_quota = 0;
4986         ext4_group_t g;
4987         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4988         int err = 0;
4989 #ifdef CONFIG_QUOTA
4990         int i, j;
4991 #endif
4992         char *orig_data = kstrdup(data, GFP_KERNEL);
4993
4994         /* Store the original options */
4995         old_sb_flags = sb->s_flags;
4996         old_opts.s_mount_opt = sbi->s_mount_opt;
4997         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4998         old_opts.s_resuid = sbi->s_resuid;
4999         old_opts.s_resgid = sbi->s_resgid;
5000         old_opts.s_commit_interval = sbi->s_commit_interval;
5001         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5002         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5003 #ifdef CONFIG_QUOTA
5004         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5005         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5006                 if (sbi->s_qf_names[i]) {
5007                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
5008                                                          GFP_KERNEL);
5009                         if (!old_opts.s_qf_names[i]) {
5010                                 for (j = 0; j < i; j++)
5011                                         kfree(old_opts.s_qf_names[j]);
5012                                 kfree(orig_data);
5013                                 return -ENOMEM;
5014                         }
5015                 } else
5016                         old_opts.s_qf_names[i] = NULL;
5017 #endif
5018         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5019                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5020
5021         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5022                 err = -EINVAL;
5023                 goto restore_opts;
5024         }
5025
5026         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5027             test_opt(sb, JOURNAL_CHECKSUM)) {
5028                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5029                          "during remount not supported; ignoring");
5030                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5031         }
5032
5033         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5034                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5035                         ext4_msg(sb, KERN_ERR, "can't mount with "
5036                                  "both data=journal and delalloc");
5037                         err = -EINVAL;
5038                         goto restore_opts;
5039                 }
5040                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5041                         ext4_msg(sb, KERN_ERR, "can't mount with "
5042                                  "both data=journal and dioread_nolock");
5043                         err = -EINVAL;
5044                         goto restore_opts;
5045                 }
5046                 if (test_opt(sb, DAX)) {
5047                         ext4_msg(sb, KERN_ERR, "can't mount with "
5048                                  "both data=journal and dax");
5049                         err = -EINVAL;
5050                         goto restore_opts;
5051                 }
5052         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5053                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5054                         ext4_msg(sb, KERN_ERR, "can't mount with "
5055                                 "journal_async_commit in data=ordered mode");
5056                         err = -EINVAL;
5057                         goto restore_opts;
5058                 }
5059         }
5060
5061         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5062                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5063                 err = -EINVAL;
5064                 goto restore_opts;
5065         }
5066
5067         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5068                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5069                         "dax flag with busy inodes while remounting");
5070                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5071         }
5072
5073         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5074                 ext4_abort(sb, "Abort forced by user");
5075
5076         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5077                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5078
5079         es = sbi->s_es;
5080
5081         if (sbi->s_journal) {
5082                 ext4_init_journal_params(sb, sbi->s_journal);
5083                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5084         }
5085
5086         if (*flags & SB_LAZYTIME)
5087                 sb->s_flags |= SB_LAZYTIME;
5088
5089         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5090                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5091                         err = -EROFS;
5092                         goto restore_opts;
5093                 }
5094
5095                 if (*flags & SB_RDONLY) {
5096                         err = sync_filesystem(sb);
5097                         if (err < 0)
5098                                 goto restore_opts;
5099                         err = dquot_suspend(sb, -1);
5100                         if (err < 0)
5101                                 goto restore_opts;
5102
5103                         /*
5104                          * First of all, the unconditional stuff we have to do
5105                          * to disable replay of the journal when we next remount
5106                          */
5107                         sb->s_flags |= SB_RDONLY;
5108
5109                         /*
5110                          * OK, test if we are remounting a valid rw partition
5111                          * readonly, and if so set the rdonly flag and then
5112                          * mark the partition as valid again.
5113                          */
5114                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5115                             (sbi->s_mount_state & EXT4_VALID_FS))
5116                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5117
5118                         if (sbi->s_journal)
5119                                 ext4_mark_recovery_complete(sb, es);
5120                 } else {
5121                         /* Make sure we can mount this feature set readwrite */
5122                         if (ext4_has_feature_readonly(sb) ||
5123                             !ext4_feature_set_ok(sb, 0)) {
5124                                 err = -EROFS;
5125                                 goto restore_opts;
5126                         }
5127                         /*
5128                          * Make sure the group descriptor checksums
5129                          * are sane.  If they aren't, refuse to remount r/w.
5130                          */
5131                         for (g = 0; g < sbi->s_groups_count; g++) {
5132                                 struct ext4_group_desc *gdp =
5133                                         ext4_get_group_desc(sb, g, NULL);
5134
5135                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5136                                         ext4_msg(sb, KERN_ERR,
5137                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5138                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5139                                                le16_to_cpu(gdp->bg_checksum));
5140                                         err = -EFSBADCRC;
5141                                         goto restore_opts;
5142                                 }
5143                         }
5144
5145                         /*
5146                          * If we have an unprocessed orphan list hanging
5147                          * around from a previously readonly bdev mount,
5148                          * require a full umount/remount for now.
5149                          */
5150                         if (es->s_last_orphan) {
5151                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5152                                        "remount RDWR because of unprocessed "
5153                                        "orphan inode list.  Please "
5154                                        "umount/remount instead");
5155                                 err = -EINVAL;
5156                                 goto restore_opts;
5157                         }
5158
5159                         /*
5160                          * Mounting a RDONLY partition read-write, so reread
5161                          * and store the current valid flag.  (It may have
5162                          * been changed by e2fsck since we originally mounted
5163                          * the partition.)
5164                          */
5165                         if (sbi->s_journal)
5166                                 ext4_clear_journal_err(sb, es);
5167                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5168                         if (!ext4_setup_super(sb, es, 0))
5169                                 sb->s_flags &= ~SB_RDONLY;
5170                         if (ext4_has_feature_mmp(sb))
5171                                 if (ext4_multi_mount_protect(sb,
5172                                                 le64_to_cpu(es->s_mmp_block))) {
5173                                         err = -EROFS;
5174                                         goto restore_opts;
5175                                 }
5176                         enable_quota = 1;
5177                 }
5178         }
5179
5180         /*
5181          * Reinitialize lazy itable initialization thread based on
5182          * current settings
5183          */
5184         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5185                 ext4_unregister_li_request(sb);
5186         else {
5187                 ext4_group_t first_not_zeroed;
5188                 first_not_zeroed = ext4_has_uninit_itable(sb);
5189                 ext4_register_li_request(sb, first_not_zeroed);
5190         }
5191
5192         ext4_setup_system_zone(sb);
5193         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY))
5194                 ext4_commit_super(sb, 1);
5195
5196 #ifdef CONFIG_QUOTA
5197         /* Release old quota file names */
5198         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5199                 kfree(old_opts.s_qf_names[i]);
5200         if (enable_quota) {
5201                 if (sb_any_quota_suspended(sb))
5202                         dquot_resume(sb, -1);
5203                 else if (ext4_has_feature_quota(sb)) {
5204                         err = ext4_enable_quotas(sb);
5205                         if (err)
5206                                 goto restore_opts;
5207                 }
5208         }
5209 #endif
5210
5211         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5212         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5213         kfree(orig_data);
5214         return 0;
5215
5216 restore_opts:
5217         sb->s_flags = old_sb_flags;
5218         sbi->s_mount_opt = old_opts.s_mount_opt;
5219         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5220         sbi->s_resuid = old_opts.s_resuid;
5221         sbi->s_resgid = old_opts.s_resgid;
5222         sbi->s_commit_interval = old_opts.s_commit_interval;
5223         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5224         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5225 #ifdef CONFIG_QUOTA
5226         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5227         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5228                 kfree(sbi->s_qf_names[i]);
5229                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5230         }
5231 #endif
5232         kfree(orig_data);
5233         return err;
5234 }
5235
5236 #ifdef CONFIG_QUOTA
5237 static int ext4_statfs_project(struct super_block *sb,
5238                                kprojid_t projid, struct kstatfs *buf)
5239 {
5240         struct kqid qid;
5241         struct dquot *dquot;
5242         u64 limit;
5243         u64 curblock;
5244
5245         qid = make_kqid_projid(projid);
5246         dquot = dqget(sb, qid);
5247         if (IS_ERR(dquot))
5248                 return PTR_ERR(dquot);
5249         spin_lock(&dquot->dq_dqb_lock);
5250
5251         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5252                  dquot->dq_dqb.dqb_bsoftlimit :
5253                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5254         if (limit && buf->f_blocks > limit) {
5255                 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
5256                 buf->f_blocks = limit;
5257                 buf->f_bfree = buf->f_bavail =
5258                         (buf->f_blocks > curblock) ?
5259                          (buf->f_blocks - curblock) : 0;
5260         }
5261
5262         limit = dquot->dq_dqb.dqb_isoftlimit ?
5263                 dquot->dq_dqb.dqb_isoftlimit :
5264                 dquot->dq_dqb.dqb_ihardlimit;
5265         if (limit && buf->f_files > limit) {
5266                 buf->f_files = limit;
5267                 buf->f_ffree =
5268                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5269                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5270         }
5271
5272         spin_unlock(&dquot->dq_dqb_lock);
5273         dqput(dquot);
5274         return 0;
5275 }
5276 #endif
5277
5278 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5279 {
5280         struct super_block *sb = dentry->d_sb;
5281         struct ext4_sb_info *sbi = EXT4_SB(sb);
5282         struct ext4_super_block *es = sbi->s_es;
5283         ext4_fsblk_t overhead = 0, resv_blocks;
5284         u64 fsid;
5285         s64 bfree;
5286         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5287
5288         if (!test_opt(sb, MINIX_DF))
5289                 overhead = sbi->s_overhead;
5290
5291         buf->f_type = EXT4_SUPER_MAGIC;
5292         buf->f_bsize = sb->s_blocksize;
5293         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5294         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5295                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5296         /* prevent underflow in case that few free space is available */
5297         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5298         buf->f_bavail = buf->f_bfree -
5299                         (ext4_r_blocks_count(es) + resv_blocks);
5300         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5301                 buf->f_bavail = 0;
5302         buf->f_files = le32_to_cpu(es->s_inodes_count);
5303         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5304         buf->f_namelen = EXT4_NAME_LEN;
5305         fsid = le64_to_cpup((void *)es->s_uuid) ^
5306                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5307         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5308         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5309
5310 #ifdef CONFIG_QUOTA
5311         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5312             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5313                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5314 #endif
5315         return 0;
5316 }
5317
5318
5319 #ifdef CONFIG_QUOTA
5320
5321 /*
5322  * Helper functions so that transaction is started before we acquire dqio_sem
5323  * to keep correct lock ordering of transaction > dqio_sem
5324  */
5325 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5326 {
5327         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5328 }
5329
5330 static int ext4_write_dquot(struct dquot *dquot)
5331 {
5332         int ret, err;
5333         handle_t *handle;
5334         struct inode *inode;
5335
5336         inode = dquot_to_inode(dquot);
5337         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5338                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5339         if (IS_ERR(handle))
5340                 return PTR_ERR(handle);
5341         ret = dquot_commit(dquot);
5342         err = ext4_journal_stop(handle);
5343         if (!ret)
5344                 ret = err;
5345         return ret;
5346 }
5347
5348 static int ext4_acquire_dquot(struct dquot *dquot)
5349 {
5350         int ret, err;
5351         handle_t *handle;
5352
5353         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5354                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5355         if (IS_ERR(handle))
5356                 return PTR_ERR(handle);
5357         ret = dquot_acquire(dquot);
5358         err = ext4_journal_stop(handle);
5359         if (!ret)
5360                 ret = err;
5361         return ret;
5362 }
5363
5364 static int ext4_release_dquot(struct dquot *dquot)
5365 {
5366         int ret, err;
5367         handle_t *handle;
5368
5369         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5370                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5371         if (IS_ERR(handle)) {
5372                 /* Release dquot anyway to avoid endless cycle in dqput() */
5373                 dquot_release(dquot);
5374                 return PTR_ERR(handle);
5375         }
5376         ret = dquot_release(dquot);
5377         err = ext4_journal_stop(handle);
5378         if (!ret)
5379                 ret = err;
5380         return ret;
5381 }
5382
5383 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5384 {
5385         struct super_block *sb = dquot->dq_sb;
5386         struct ext4_sb_info *sbi = EXT4_SB(sb);
5387
5388         /* Are we journaling quotas? */
5389         if (ext4_has_feature_quota(sb) ||
5390             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5391                 dquot_mark_dquot_dirty(dquot);
5392                 return ext4_write_dquot(dquot);
5393         } else {
5394                 return dquot_mark_dquot_dirty(dquot);
5395         }
5396 }
5397
5398 static int ext4_write_info(struct super_block *sb, int type)
5399 {
5400         int ret, err;
5401         handle_t *handle;
5402
5403         /* Data block + inode block */
5404         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5405         if (IS_ERR(handle))
5406                 return PTR_ERR(handle);
5407         ret = dquot_commit_info(sb, type);
5408         err = ext4_journal_stop(handle);
5409         if (!ret)
5410                 ret = err;
5411         return ret;
5412 }
5413
5414 /*
5415  * Turn on quotas during mount time - we need to find
5416  * the quota file and such...
5417  */
5418 static int ext4_quota_on_mount(struct super_block *sb, int type)
5419 {
5420         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5421                                         EXT4_SB(sb)->s_jquota_fmt, type);
5422 }
5423
5424 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5425 {
5426         struct ext4_inode_info *ei = EXT4_I(inode);
5427
5428         /* The first argument of lockdep_set_subclass has to be
5429          * *exactly* the same as the argument to init_rwsem() --- in
5430          * this case, in init_once() --- or lockdep gets unhappy
5431          * because the name of the lock is set using the
5432          * stringification of the argument to init_rwsem().
5433          */
5434         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5435         lockdep_set_subclass(&ei->i_data_sem, subclass);
5436 }
5437
5438 /*
5439  * Standard function to be called on quota_on
5440  */
5441 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5442                          const struct path *path)
5443 {
5444         int err;
5445
5446         if (!test_opt(sb, QUOTA))
5447                 return -EINVAL;
5448
5449         /* Quotafile not on the same filesystem? */
5450         if (path->dentry->d_sb != sb)
5451                 return -EXDEV;
5452         /* Journaling quota? */
5453         if (EXT4_SB(sb)->s_qf_names[type]) {
5454                 /* Quotafile not in fs root? */
5455                 if (path->dentry->d_parent != sb->s_root)
5456                         ext4_msg(sb, KERN_WARNING,
5457                                 "Quota file not on filesystem root. "
5458                                 "Journaled quota will not work");
5459                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5460         } else {
5461                 /*
5462                  * Clear the flag just in case mount options changed since
5463                  * last time.
5464                  */
5465                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5466         }
5467
5468         /*
5469          * When we journal data on quota file, we have to flush journal to see
5470          * all updates to the file when we bypass pagecache...
5471          */
5472         if (EXT4_SB(sb)->s_journal &&
5473             ext4_should_journal_data(d_inode(path->dentry))) {
5474                 /*
5475                  * We don't need to lock updates but journal_flush() could
5476                  * otherwise be livelocked...
5477                  */
5478                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5479                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5480                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5481                 if (err)
5482                         return err;
5483         }
5484
5485         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5486         err = dquot_quota_on(sb, type, format_id, path);
5487         if (err) {
5488                 lockdep_set_quota_inode(path->dentry->d_inode,
5489                                              I_DATA_SEM_NORMAL);
5490         } else {
5491                 struct inode *inode = d_inode(path->dentry);
5492                 handle_t *handle;
5493
5494                 /*
5495                  * Set inode flags to prevent userspace from messing with quota
5496                  * files. If this fails, we return success anyway since quotas
5497                  * are already enabled and this is not a hard failure.
5498                  */
5499                 inode_lock(inode);
5500                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5501                 if (IS_ERR(handle))
5502                         goto unlock_inode;
5503                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5504                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5505                                 S_NOATIME | S_IMMUTABLE);
5506                 ext4_mark_inode_dirty(handle, inode);
5507                 ext4_journal_stop(handle);
5508         unlock_inode:
5509                 inode_unlock(inode);
5510         }
5511         return err;
5512 }
5513
5514 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5515                              unsigned int flags)
5516 {
5517         int err;
5518         struct inode *qf_inode;
5519         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5520                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5521                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5522                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5523         };
5524
5525         BUG_ON(!ext4_has_feature_quota(sb));
5526
5527         if (!qf_inums[type])
5528                 return -EPERM;
5529
5530         qf_inode = ext4_iget(sb, qf_inums[type]);
5531         if (IS_ERR(qf_inode)) {
5532                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5533                 return PTR_ERR(qf_inode);
5534         }
5535
5536         /* Don't account quota for quota files to avoid recursion */
5537         qf_inode->i_flags |= S_NOQUOTA;
5538         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5539         err = dquot_enable(qf_inode, type, format_id, flags);
5540         iput(qf_inode);
5541         if (err)
5542                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5543
5544         return err;
5545 }
5546
5547 /* Enable usage tracking for all quota types. */
5548 static int ext4_enable_quotas(struct super_block *sb)
5549 {
5550         int type, err = 0;
5551         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5552                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5553                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5554                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5555         };
5556         bool quota_mopt[EXT4_MAXQUOTAS] = {
5557                 test_opt(sb, USRQUOTA),
5558                 test_opt(sb, GRPQUOTA),
5559                 test_opt(sb, PRJQUOTA),
5560         };
5561
5562         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5563         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5564                 if (qf_inums[type]) {
5565                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5566                                 DQUOT_USAGE_ENABLED |
5567                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5568                         if (err) {
5569                                 for (type--; type >= 0; type--)
5570                                         dquot_quota_off(sb, type);
5571
5572                                 ext4_warning(sb,
5573                                         "Failed to enable quota tracking "
5574                                         "(type=%d, err=%d). Please run "
5575                                         "e2fsck to fix.", type, err);
5576                                 return err;
5577                         }
5578                 }
5579         }
5580         return 0;
5581 }
5582
5583 static int ext4_quota_off(struct super_block *sb, int type)
5584 {
5585         struct inode *inode = sb_dqopt(sb)->files[type];
5586         handle_t *handle;
5587         int err;
5588
5589         /* Force all delayed allocation blocks to be allocated.
5590          * Caller already holds s_umount sem */
5591         if (test_opt(sb, DELALLOC))
5592                 sync_filesystem(sb);
5593
5594         if (!inode || !igrab(inode))
5595                 goto out;
5596
5597         err = dquot_quota_off(sb, type);
5598         if (err || ext4_has_feature_quota(sb))
5599                 goto out_put;
5600
5601         inode_lock(inode);
5602         /*
5603          * Update modification times of quota files when userspace can
5604          * start looking at them. If we fail, we return success anyway since
5605          * this is not a hard failure and quotas are already disabled.
5606          */
5607         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5608         if (IS_ERR(handle))
5609                 goto out_unlock;
5610         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5611         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5612         inode->i_mtime = inode->i_ctime = current_time(inode);
5613         ext4_mark_inode_dirty(handle, inode);
5614         ext4_journal_stop(handle);
5615 out_unlock:
5616         inode_unlock(inode);
5617 out_put:
5618         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5619         iput(inode);
5620         return err;
5621 out:
5622         return dquot_quota_off(sb, type);
5623 }
5624
5625 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5626  * acquiring the locks... As quota files are never truncated and quota code
5627  * itself serializes the operations (and no one else should touch the files)
5628  * we don't have to be afraid of races */
5629 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5630                                size_t len, loff_t off)
5631 {
5632         struct inode *inode = sb_dqopt(sb)->files[type];
5633         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5634         int offset = off & (sb->s_blocksize - 1);
5635         int tocopy;
5636         size_t toread;
5637         struct buffer_head *bh;
5638         loff_t i_size = i_size_read(inode);
5639
5640         if (off > i_size)
5641                 return 0;
5642         if (off+len > i_size)
5643                 len = i_size-off;
5644         toread = len;
5645         while (toread > 0) {
5646                 tocopy = sb->s_blocksize - offset < toread ?
5647                                 sb->s_blocksize - offset : toread;
5648                 bh = ext4_bread(NULL, inode, blk, 0);
5649                 if (IS_ERR(bh))
5650                         return PTR_ERR(bh);
5651                 if (!bh)        /* A hole? */
5652                         memset(data, 0, tocopy);
5653                 else
5654                         memcpy(data, bh->b_data+offset, tocopy);
5655                 brelse(bh);
5656                 offset = 0;
5657                 toread -= tocopy;
5658                 data += tocopy;
5659                 blk++;
5660         }
5661         return len;
5662 }
5663
5664 /* Write to quotafile (we know the transaction is already started and has
5665  * enough credits) */
5666 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5667                                 const char *data, size_t len, loff_t off)
5668 {
5669         struct inode *inode = sb_dqopt(sb)->files[type];
5670         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5671         int err, offset = off & (sb->s_blocksize - 1);
5672         int retries = 0;
5673         struct buffer_head *bh;
5674         handle_t *handle = journal_current_handle();
5675
5676         if (EXT4_SB(sb)->s_journal && !handle) {
5677                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5678                         " cancelled because transaction is not started",
5679                         (unsigned long long)off, (unsigned long long)len);
5680                 return -EIO;
5681         }
5682         /*
5683          * Since we account only one data block in transaction credits,
5684          * then it is impossible to cross a block boundary.
5685          */
5686         if (sb->s_blocksize - offset < len) {
5687                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5688                         " cancelled because not block aligned",
5689                         (unsigned long long)off, (unsigned long long)len);
5690                 return -EIO;
5691         }
5692
5693         do {
5694                 bh = ext4_bread(handle, inode, blk,
5695                                 EXT4_GET_BLOCKS_CREATE |
5696                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5697         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5698                  ext4_should_retry_alloc(inode->i_sb, &retries));
5699         if (IS_ERR(bh))
5700                 return PTR_ERR(bh);
5701         if (!bh)
5702                 goto out;
5703         BUFFER_TRACE(bh, "get write access");
5704         err = ext4_journal_get_write_access(handle, bh);
5705         if (err) {
5706                 brelse(bh);
5707                 return err;
5708         }
5709         lock_buffer(bh);
5710         memcpy(bh->b_data+offset, data, len);
5711         flush_dcache_page(bh->b_page);
5712         unlock_buffer(bh);
5713         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5714         brelse(bh);
5715 out:
5716         if (inode->i_size < off + len) {
5717                 i_size_write(inode, off + len);
5718                 EXT4_I(inode)->i_disksize = inode->i_size;
5719                 ext4_mark_inode_dirty(handle, inode);
5720         }
5721         return len;
5722 }
5723
5724 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5725 {
5726         const struct quota_format_ops   *ops;
5727
5728         if (!sb_has_quota_loaded(sb, qid->type))
5729                 return -ESRCH;
5730         ops = sb_dqopt(sb)->ops[qid->type];
5731         if (!ops || !ops->get_next_id)
5732                 return -ENOSYS;
5733         return dquot_get_next_id(sb, qid);
5734 }
5735 #endif
5736
5737 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5738                        const char *dev_name, void *data)
5739 {
5740         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5741 }
5742
5743 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5744 static inline void register_as_ext2(void)
5745 {
5746         int err = register_filesystem(&ext2_fs_type);
5747         if (err)
5748                 printk(KERN_WARNING
5749                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5750 }
5751
5752 static inline void unregister_as_ext2(void)
5753 {
5754         unregister_filesystem(&ext2_fs_type);
5755 }
5756
5757 static inline int ext2_feature_set_ok(struct super_block *sb)
5758 {
5759         if (ext4_has_unknown_ext2_incompat_features(sb))
5760                 return 0;
5761         if (sb_rdonly(sb))
5762                 return 1;
5763         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5764                 return 0;
5765         return 1;
5766 }
5767 #else
5768 static inline void register_as_ext2(void) { }
5769 static inline void unregister_as_ext2(void) { }
5770 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5771 #endif
5772
5773 static inline void register_as_ext3(void)
5774 {
5775         int err = register_filesystem(&ext3_fs_type);
5776         if (err)
5777                 printk(KERN_WARNING
5778                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5779 }
5780
5781 static inline void unregister_as_ext3(void)
5782 {
5783         unregister_filesystem(&ext3_fs_type);
5784 }
5785
5786 static inline int ext3_feature_set_ok(struct super_block *sb)
5787 {
5788         if (ext4_has_unknown_ext3_incompat_features(sb))
5789                 return 0;
5790         if (!ext4_has_feature_journal(sb))
5791                 return 0;
5792         if (sb_rdonly(sb))
5793                 return 1;
5794         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5795                 return 0;
5796         return 1;
5797 }
5798
5799 static struct file_system_type ext4_fs_type = {
5800         .owner          = THIS_MODULE,
5801         .name           = "ext4",
5802         .mount          = ext4_mount,
5803         .kill_sb        = kill_block_super,
5804         .fs_flags       = FS_REQUIRES_DEV,
5805 };
5806 MODULE_ALIAS_FS("ext4");
5807
5808 /* Shared across all ext4 file systems */
5809 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5810
5811 static int __init ext4_init_fs(void)
5812 {
5813         int i, err;
5814
5815         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5816         ext4_li_info = NULL;
5817         mutex_init(&ext4_li_mtx);
5818
5819         /* Build-time check for flags consistency */
5820         ext4_check_flag_values();
5821
5822         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5823                 init_waitqueue_head(&ext4__ioend_wq[i]);
5824
5825         err = ext4_init_es();
5826         if (err)
5827                 return err;
5828
5829         err = ext4_init_pageio();
5830         if (err)
5831                 goto out5;
5832
5833         err = ext4_init_system_zone();
5834         if (err)
5835                 goto out4;
5836
5837         err = ext4_init_sysfs();
5838         if (err)
5839                 goto out3;
5840
5841         err = ext4_init_mballoc();
5842         if (err)
5843                 goto out2;
5844         err = init_inodecache();
5845         if (err)
5846                 goto out1;
5847         register_as_ext3();
5848         register_as_ext2();
5849         err = register_filesystem(&ext4_fs_type);
5850         if (err)
5851                 goto out;
5852
5853         return 0;
5854 out:
5855         unregister_as_ext2();
5856         unregister_as_ext3();
5857         destroy_inodecache();
5858 out1:
5859         ext4_exit_mballoc();
5860 out2:
5861         ext4_exit_sysfs();
5862 out3:
5863         ext4_exit_system_zone();
5864 out4:
5865         ext4_exit_pageio();
5866 out5:
5867         ext4_exit_es();
5868
5869         return err;
5870 }
5871
5872 static void __exit ext4_exit_fs(void)
5873 {
5874         ext4_destroy_lazyinit_thread();
5875         unregister_as_ext2();
5876         unregister_as_ext3();
5877         unregister_filesystem(&ext4_fs_type);
5878         destroy_inodecache();
5879         ext4_exit_mballoc();
5880         ext4_exit_sysfs();
5881         ext4_exit_system_zone();
5882         ext4_exit_pageio();
5883         ext4_exit_es();
5884 }
5885
5886 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5887 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5888 MODULE_LICENSE("GPL");
5889 MODULE_SOFTDEP("pre: crc32c");
5890 module_init(ext4_init_fs)
5891 module_exit(ext4_exit_fs)