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