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 (card@masi.ibp.fr)
   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 (davem@caip.rutgers.edu), 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 static void __ext4_update_tstamp(__le32 *lo, __u8 *hi)
 316 {
 317         time64_t now = ktime_get_real_seconds();
 318
 319         now = clamp_val(now, 0, (1ull << 40) - 1);
 320
 321         *lo = cpu_to_le32(lower_32_bits(now));
 322         *hi = upper_32_bits(now);
 323 }
 324
 325 static time64_t __ext4_get_tstamp(__le32 *lo, __u8 *hi)
 326 {
 327         return ((time64_t)(*hi) << 32) + le32_to_cpu(*lo);
 328 }
 329 #define ext4_update_tstamp(es, tstamp) \
 330         __ext4_update_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 331 #define ext4_get_tstamp(es, tstamp) \
 332         __ext4_get_tstamp(&(es)->tstamp, &(es)->tstamp ## _hi)
 333
 334 static void __save_error_info(struct super_block *sb, const char *func,
 335                             unsigned int line)
 336 {
 337         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 338
 339         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 340         if (bdev_read_only(sb->s_bdev))
 341                 return;
 342         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 343         ext4_update_tstamp(es, s_last_error_time);
 344         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 345         es->s_last_error_line = cpu_to_le32(line);
 346         if (!es->s_first_error_time) {
 347                 es->s_first_error_time = es->s_last_error_time;
 348                 es->s_first_error_time_hi = es->s_last_error_time_hi;
 349                 strncpy(es->s_first_error_func, func,
 350                         sizeof(es->s_first_error_func));
 351                 es->s_first_error_line = cpu_to_le32(line);
 352                 es->s_first_error_ino = es->s_last_error_ino;
 353                 es->s_first_error_block = es->s_last_error_block;
 354         }
 355         /*
 356          * Start the daily error reporting function if it hasn't been
 357          * started already
 358          */
 359         if (!es->s_error_count)
 360                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 361         le32_add_cpu(&es->s_error_count, 1);
 362 }
 363
 364 static void save_error_info(struct super_block *sb, const char *func,
 365                             unsigned int line)
 366 {
 367         __save_error_info(sb, func, line);
 368         ext4_commit_super(sb, 1);
 369 }
 370
 371 /*
 372  * The del_gendisk() function uninitializes the disk-specific data
 373  * structures, including the bdi structure, without telling anyone
 374  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 375  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 376  * This is a kludge to prevent these oops until we can put in a proper
 377  * hook in del_gendisk() to inform the VFS and file system layers.
 378  */
 379 static int block_device_ejected(struct super_block *sb)
 380 {
 381         struct inode *bd_inode = sb->s_bdev->bd_inode;
 382         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 383
 384         return bdi->dev == NULL;
 385 }
 386
 387 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 388 {
 389         struct super_block              *sb = journal->j_private;
 390         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 391         int                             error = is_journal_aborted(journal);
 392         struct ext4_journal_cb_entry    *jce;
 393
 394         BUG_ON(txn->t_state == T_FINISHED);
 395
 396         ext4_process_freed_data(sb, txn->t_tid);
 397
 398         spin_lock(&sbi->s_md_lock);
 399         while (!list_empty(&txn->t_private_list)) {
 400                 jce = list_entry(txn->t_private_list.next,
 401                                  struct ext4_journal_cb_entry, jce_list);
 402                 list_del_init(&jce->jce_list);
 403                 spin_unlock(&sbi->s_md_lock);
 404                 jce->jce_func(sb, jce, error);
 405                 spin_lock(&sbi->s_md_lock);
 406         }
 407         spin_unlock(&sbi->s_md_lock);
 408 }
 409
 410 /* Deal with the reporting of failure conditions on a filesystem such as
 411  * inconsistencies detected or read IO failures.
 412  *
 413  * On ext2, we can store the error state of the filesystem in the
 414  * superblock.  That is not possible on ext4, because we may have other
 415  * write ordering constraints on the superblock which prevent us from
 416  * writing it out straight away; and given that the journal is about to
 417  * be aborted, we can't rely on the current, or future, transactions to
 418  * write out the superblock safely.
 419  *
 420  * We'll just use the jbd2_journal_abort() error code to record an error in
 421  * the journal instead.  On recovery, the journal will complain about
 422  * that error until we've noted it down and cleared it.
 423  */
 424
 425 static void ext4_handle_error(struct super_block *sb)
 426 {
 427         if (test_opt(sb, WARN_ON_ERROR))
 428                 WARN_ON_ONCE(1);
 429
 430         if (sb_rdonly(sb))
 431                 return;
 432
 433         if (!test_opt(sb, ERRORS_CONT)) {
 434                 journal_t *journal = EXT4_SB(sb)->s_journal;
 435
 436                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 437                 if (journal)
 438                         jbd2_journal_abort(journal, -EIO);
 439         }
 440         if (test_opt(sb, ERRORS_RO)) {
 441                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 442                 /*
 443                  * Make sure updated value of ->s_mount_flags will be visible
 444                  * before ->s_flags update
 445                  */
 446                 smp_wmb();
 447                 sb->s_flags |= SB_RDONLY;
 448         }
 449         if (test_opt(sb, ERRORS_PANIC)) {
 450                 if (EXT4_SB(sb)->s_journal &&
 451                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 452                         return;
 453                 panic("EXT4-fs (device %s): panic forced after error\n",
 454                         sb->s_id);
 455         }
 456 }
 457
 458 #define ext4_error_ratelimit(sb)                                        \
 459                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 460                              "EXT4-fs error")
 461
 462 void __ext4_error(struct super_block *sb, const char *function,
 463                   unsigned int line, const char *fmt, ...)
 464 {
 465         struct va_format vaf;
 466         va_list args;
 467
 468         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 469                 return;
 470
 471         trace_ext4_error(sb, function, line);
 472         if (ext4_error_ratelimit(sb)) {
 473                 va_start(args, fmt);
 474                 vaf.fmt = fmt;
 475                 vaf.va = &args;
 476                 printk(KERN_CRIT
 477                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 478                        sb->s_id, function, line, current->comm, &vaf);
 479                 va_end(args);
 480         }
 481         save_error_info(sb, function, line);
 482         ext4_handle_error(sb);
 483 }
 484
 485 void __ext4_error_inode(struct inode *inode, const char *function,
 486                         unsigned int line, ext4_fsblk_t block,
 487                         const char *fmt, ...)
 488 {
 489         va_list args;
 490         struct va_format vaf;
 491         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 492
 493         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 494                 return;
 495
 496         trace_ext4_error(inode->i_sb, function, line);
 497         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 498         es->s_last_error_block = cpu_to_le64(block);
 499         if (ext4_error_ratelimit(inode->i_sb)) {
 500                 va_start(args, fmt);
 501                 vaf.fmt = fmt;
 502                 vaf.va = &args;
 503                 if (block)
 504                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 505                                "inode #%lu: block %llu: comm %s: %pV\n",
 506                                inode->i_sb->s_id, function, line, inode->i_ino,
 507                                block, current->comm, &vaf);
 508                 else
 509                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 510                                "inode #%lu: comm %s: %pV\n",
 511                                inode->i_sb->s_id, function, line, inode->i_ino,
 512                                current->comm, &vaf);
 513                 va_end(args);
 514         }
 515         save_error_info(inode->i_sb, function, line);
 516         ext4_handle_error(inode->i_sb);
 517 }
 518
 519 void __ext4_error_file(struct file *file, const char *function,
 520                        unsigned int line, ext4_fsblk_t block,
 521                        const char *fmt, ...)
 522 {
 523         va_list args;
 524         struct va_format vaf;
 525         struct ext4_super_block *es;
 526         struct inode *inode = file_inode(file);
 527         char pathname[80], *path;
 528
 529         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 530                 return;
 531
 532         trace_ext4_error(inode->i_sb, function, line);
 533         es = EXT4_SB(inode->i_sb)->s_es;
 534         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 535         if (ext4_error_ratelimit(inode->i_sb)) {
 536                 path = file_path(file, pathname, sizeof(pathname));
 537                 if (IS_ERR(path))
 538                         path = "(unknown)";
 539                 va_start(args, fmt);
 540                 vaf.fmt = fmt;
 541                 vaf.va = &args;
 542                 if (block)
 543                         printk(KERN_CRIT
 544                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 545                                "block %llu: comm %s: path %s: %pV\n",
 546                                inode->i_sb->s_id, function, line, inode->i_ino,
 547                                block, current->comm, path, &vaf);
 548                 else
 549                         printk(KERN_CRIT
 550                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 551                                "comm %s: path %s: %pV\n",
 552                                inode->i_sb->s_id, function, line, inode->i_ino,
 553                                current->comm, path, &vaf);
 554                 va_end(args);
 555         }
 556         save_error_info(inode->i_sb, function, line);
 557         ext4_handle_error(inode->i_sb);
 558 }
 559
 560 const char *ext4_decode_error(struct super_block *sb, int errno,
 561                               char nbuf[16])
 562 {
 563         char *errstr = NULL;
 564
 565         switch (errno) {
 566         case -EFSCORRUPTED:
 567                 errstr = "Corrupt filesystem";
 568                 break;
 569         case -EFSBADCRC:
 570                 errstr = "Filesystem failed CRC";
 571                 break;
 572         case -EIO:
 573                 errstr = "IO failure";
 574                 break;
 575         case -ENOMEM:
 576                 errstr = "Out of memory";
 577                 break;
 578         case -EROFS:
 579                 if (!sb || (EXT4_SB(sb)->s_journal &&
 580                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 581                         errstr = "Journal has aborted";
 582                 else
 583                         errstr = "Readonly filesystem";
 584                 break;
 585         default:
 586                 /* If the caller passed in an extra buffer for unknown
 587                  * errors, textualise them now.  Else we just return
 588                  * NULL. */
 589                 if (nbuf) {
 590                         /* Check for truncated error codes... */
 591                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 592                                 errstr = nbuf;
 593                 }
 594                 break;
 595         }
 596
 597         return errstr;
 598 }
 599
 600 /* __ext4_std_error decodes expected errors from journaling functions
 601  * automatically and invokes the appropriate error response.  */
 602
 603 void __ext4_std_error(struct super_block *sb, const char *function,
 604                       unsigned int line, int errno)
 605 {
 606         char nbuf[16];
 607         const char *errstr;
 608
 609         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 610                 return;
 611
 612         /* Special case: if the error is EROFS, and we're not already
 613          * inside a transaction, then there's really no point in logging
 614          * an error. */
 615         if (errno == -EROFS && journal_current_handle() == NULL && sb_rdonly(sb))
 616                 return;
 617
 618         if (ext4_error_ratelimit(sb)) {
 619                 errstr = ext4_decode_error(sb, errno, nbuf);
 620                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 621                        sb->s_id, function, line, errstr);
 622         }
 623
 624         save_error_info(sb, function, line);
 625         ext4_handle_error(sb);
 626 }
 627
 628 /*
 629  * ext4_abort is a much stronger failure handler than ext4_error.  The
 630  * abort function may be used to deal with unrecoverable failures such
 631  * as journal IO errors or ENOMEM at a critical moment in log management.
 632  *
 633  * We unconditionally force the filesystem into an ABORT|READONLY state,
 634  * unless the error response on the fs has been set to panic in which
 635  * case we take the easy way out and panic immediately.
 636  */
 637
 638 void __ext4_abort(struct super_block *sb, const char *function,
 639                 unsigned int line, const char *fmt, ...)
 640 {
 641         struct va_format vaf;
 642         va_list args;
 643
 644         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 645                 return;
 646
 647         save_error_info(sb, function, line);
 648         va_start(args, fmt);
 649         vaf.fmt = fmt;
 650         vaf.va = &args;
 651         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
 652                sb->s_id, function, line, &vaf);
 653         va_end(args);
 654
 655         if (sb_rdonly(sb) == 0) {
 656                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 657                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 658                 /*
 659                  * Make sure updated value of ->s_mount_flags will be visible
 660                  * before ->s_flags update
 661                  */
 662                 smp_wmb();
 663                 sb->s_flags |= SB_RDONLY;
 664                 if (EXT4_SB(sb)->s_journal)
 665                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 666                 save_error_info(sb, function, line);
 667         }
 668         if (test_opt(sb, ERRORS_PANIC)) {
 669                 if (EXT4_SB(sb)->s_journal &&
 670                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 671                         return;
 672                 panic("EXT4-fs panic from previous error\n");
 673         }
 674 }
 675
 676 void __ext4_msg(struct super_block *sb,
 677                 const char *prefix, const char *fmt, ...)
 678 {
 679         struct va_format vaf;
 680         va_list args;
 681
 682         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 683                 return;
 684
 685         va_start(args, fmt);
 686         vaf.fmt = fmt;
 687         vaf.va = &args;
 688         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 689         va_end(args);
 690 }
 691
 692 #define ext4_warning_ratelimit(sb)                                      \
 693                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 694                              "EXT4-fs warning")
 695
 696 void __ext4_warning(struct super_block *sb, const char *function,
 697                     unsigned int line, const char *fmt, ...)
 698 {
 699         struct va_format vaf;
 700         va_list args;
 701
 702         if (!ext4_warning_ratelimit(sb))
 703                 return;
 704
 705         va_start(args, fmt);
 706         vaf.fmt = fmt;
 707         vaf.va = &args;
 708         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 709                sb->s_id, function, line, &vaf);
 710         va_end(args);
 711 }
 712
 713 void __ext4_warning_inode(const struct inode *inode, const char *function,
 714                           unsigned int line, const char *fmt, ...)
 715 {
 716         struct va_format vaf;
 717         va_list args;
 718
 719         if (!ext4_warning_ratelimit(inode->i_sb))
 720                 return;
 721
 722         va_start(args, fmt);
 723         vaf.fmt = fmt;
 724         vaf.va = &args;
 725         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 726                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 727                function, line, inode->i_ino, current->comm, &vaf);
 728         va_end(args);
 729 }
 730
 731 void __ext4_grp_locked_error(const char *function, unsigned int line,
 732                              struct super_block *sb, ext4_group_t grp,
 733                              unsigned long ino, ext4_fsblk_t block,
 734                              const char *fmt, ...)
 735 __releases(bitlock)
 736 __acquires(bitlock)
 737 {
 738         struct va_format vaf;
 739         va_list args;
 740         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 741
 742         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
 743                 return;
 744
 745         trace_ext4_error(sb, function, line);
 746         es->s_last_error_ino = cpu_to_le32(ino);
 747         es->s_last_error_block = cpu_to_le64(block);
 748         __save_error_info(sb, function, line);
 749
 750         if (ext4_error_ratelimit(sb)) {
 751                 va_start(args, fmt);
 752                 vaf.fmt = fmt;
 753                 vaf.va = &args;
 754                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 755                        sb->s_id, function, line, grp);
 756                 if (ino)
 757                         printk(KERN_CONT "inode %lu: ", ino);
 758                 if (block)
 759                         printk(KERN_CONT "block %llu:",
 760                                (unsigned long long) block);
 761                 printk(KERN_CONT "%pV\n", &vaf);
 762                 va_end(args);
 763         }
 764
 765         if (test_opt(sb, WARN_ON_ERROR))
 766                 WARN_ON_ONCE(1);
 767
 768         if (test_opt(sb, ERRORS_CONT)) {
 769                 ext4_commit_super(sb, 0);
 770                 return;
 771         }
 772
 773         ext4_unlock_group(sb, grp);
 774         ext4_commit_super(sb, 1);
 775         ext4_handle_error(sb);
 776         /*
 777          * We only get here in the ERRORS_RO case; relocking the group
 778          * may be dangerous, but nothing bad will happen since the
 779          * filesystem will have already been marked read/only and the
 780          * journal has been aborted.  We return 1 as a hint to callers
 781          * who might what to use the return value from
 782          * ext4_grp_locked_error() to distinguish between the
 783          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 784          * aggressively from the ext4 function in question, with a
 785          * more appropriate error code.
 786          */
 787         ext4_lock_group(sb, grp);
 788         return;
 789 }
 790
 791 void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
 792                                      ext4_group_t group,
 793                                      unsigned int flags)
 794 {
 795         struct ext4_sb_info *sbi = EXT4_SB(sb);
 796         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 797         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 798         int ret;
 799
 800         if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
 801                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
 802                                             &grp->bb_state);
 803                 if (!ret)
 804                         percpu_counter_sub(&sbi->s_freeclusters_counter,
 805                                            grp->bb_free);
 806         }
 807
 808         if (flags & EXT4_GROUP_INFO_IBITMAP_CORRUPT) {
 809                 ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT,
 810                                             &grp->bb_state);
 811                 if (!ret && gdp) {
 812                         int count;
 813
 814                         count = ext4_free_inodes_count(sb, gdp);
 815                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 816                                            count);
 817                 }
 818         }
 819 }
 820
 821 void ext4_update_dynamic_rev(struct super_block *sb)
 822 {
 823         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 824
 825         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 826                 return;
 827
 828         ext4_warning(sb,
 829                      "updating to rev %d because of new feature flag, "
 830                      "running e2fsck is recommended",
 831                      EXT4_DYNAMIC_REV);
 832
 833         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 834         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 835         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 836         /* leave es->s_feature_*compat flags alone */
 837         /* es->s_uuid will be set by e2fsck if empty */
 838
 839         /*
 840          * The rest of the superblock fields should be zero, and if not it
 841          * means they are likely already in use, so leave them alone.  We
 842          * can leave it up to e2fsck to clean up any inconsistencies there.
 843          */
 844 }
 845
 846 /*
 847  * Open the external journal device
 848  */
 849 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 850 {
 851         struct block_device *bdev;
 852         char b[BDEVNAME_SIZE];
 853
 854         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 855         if (IS_ERR(bdev))
 856                 goto fail;
 857         return bdev;
 858
 859 fail:
 860         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 861                         __bdevname(dev, b), PTR_ERR(bdev));
 862         return NULL;
 863 }
 864
 865 /*
 866  * Release the journal device
 867  */
 868 static void ext4_blkdev_put(struct block_device *bdev)
 869 {
 870         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 871 }
 872
 873 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 874 {
 875         struct block_device *bdev;
 876         bdev = sbi->journal_bdev;
 877         if (bdev) {
 878                 ext4_blkdev_put(bdev);
 879                 sbi->journal_bdev = NULL;
 880         }
 881 }
 882
 883 static inline struct inode *orphan_list_entry(struct list_head *l)
 884 {
 885         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 886 }
 887
 888 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 889 {
 890         struct list_head *l;
 891
 892         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 893                  le32_to_cpu(sbi->s_es->s_last_orphan));
 894
 895         printk(KERN_ERR "sb_info orphan list:\n");
 896         list_for_each(l, &sbi->s_orphan) {
 897                 struct inode *inode = orphan_list_entry(l);
 898                 printk(KERN_ERR "  "
 899                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 900                        inode->i_sb->s_id, inode->i_ino, inode,
 901                        inode->i_mode, inode->i_nlink,
 902                        NEXT_ORPHAN(inode));
 903         }
 904 }
 905
 906 #ifdef CONFIG_QUOTA
 907 static int ext4_quota_off(struct super_block *sb, int type);
 908
 909 static inline void ext4_quota_off_umount(struct super_block *sb)
 910 {
 911         int type;
 912
 913         /* Use our quota_off function to clear inode flags etc. */
 914         for (type = 0; type < EXT4_MAXQUOTAS; type++)
 915                 ext4_quota_off(sb, type);
 916 }
 917 #else
 918 static inline void ext4_quota_off_umount(struct super_block *sb)
 919 {
 920 }
 921 #endif
 922
 923 static void ext4_put_super(struct super_block *sb)
 924 {
 925         struct ext4_sb_info *sbi = EXT4_SB(sb);
 926         struct ext4_super_block *es = sbi->s_es;
 927         int aborted = 0;
 928         int i, err;
 929
 930         ext4_unregister_li_request(sb);
 931         ext4_quota_off_umount(sb);
 932
 933         destroy_workqueue(sbi->rsv_conversion_wq);
 934
 935         if (sbi->s_journal) {
 936                 aborted = is_journal_aborted(sbi->s_journal);
 937                 err = jbd2_journal_destroy(sbi->s_journal);
 938                 sbi->s_journal = NULL;
 939                 if ((err < 0) && !aborted)
 940                         ext4_abort(sb, "Couldn't clean up the journal");
 941         }
 942
 943         ext4_unregister_sysfs(sb);
 944         ext4_es_unregister_shrinker(sbi);
 945         del_timer_sync(&sbi->s_err_report);
 946         ext4_release_system_zone(sb);
 947         ext4_mb_release(sb);
 948         ext4_ext_release(sb);
 949
 950         if (!sb_rdonly(sb) && !aborted) {
 951                 ext4_clear_feature_journal_needs_recovery(sb);
 952                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 953         }
 954         if (!sb_rdonly(sb))
 955                 ext4_commit_super(sb, 1);
 956
 957         for (i = 0; i < sbi->s_gdb_count; i++)
 958                 brelse(sbi->s_group_desc[i]);
 959         kvfree(sbi->s_group_desc);
 960         kvfree(sbi->s_flex_groups);
 961         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 962         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 963         percpu_counter_destroy(&sbi->s_dirs_counter);
 964         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 965         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 966 #ifdef CONFIG_QUOTA
 967         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 968                 kfree(sbi->s_qf_names[i]);
 969 #endif
 970
 971         /* Debugging code just in case the in-memory inode orphan list
 972          * isn't empty.  The on-disk one can be non-empty if we've
 973          * detected an error and taken the fs readonly, but the
 974          * in-memory list had better be clean by this point. */
 975         if (!list_empty(&sbi->s_orphan))
 976                 dump_orphan_list(sb, sbi);
 977         J_ASSERT(list_empty(&sbi->s_orphan));
 978
 979         sync_blockdev(sb->s_bdev);
 980         invalidate_bdev(sb->s_bdev);
 981         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 982                 /*
 983                  * Invalidate the journal device's buffers.  We don't want them
 984                  * floating about in memory - the physical journal device may
 985                  * hotswapped, and it breaks the `ro-after' testing code.
 986                  */
 987                 sync_blockdev(sbi->journal_bdev);
 988                 invalidate_bdev(sbi->journal_bdev);
 989                 ext4_blkdev_remove(sbi);
 990         }
 991         if (sbi->s_ea_inode_cache) {
 992                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
 993                 sbi->s_ea_inode_cache = NULL;
 994         }
 995         if (sbi->s_ea_block_cache) {
 996                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 997                 sbi->s_ea_block_cache = NULL;
 998         }
 999         if (sbi->s_mmp_tsk)
1000                 kthread_stop(sbi->s_mmp_tsk);
1001         brelse(sbi->s_sbh);
1002         sb->s_fs_info = NULL;
1003         /*
1004          * Now that we are completely done shutting down the
1005          * superblock, we need to actually destroy the kobject.
1006          */
1007         kobject_put(&sbi->s_kobj);
1008         wait_for_completion(&sbi->s_kobj_unregister);
1009         if (sbi->s_chksum_driver)
1010                 crypto_free_shash(sbi->s_chksum_driver);
1011         kfree(sbi->s_blockgroup_lock);
1012         fs_put_dax(sbi->s_daxdev);
1013         kfree(sbi);
1014 }
1015
1016 static struct kmem_cache *ext4_inode_cachep;
1017
1018 /*
1019  * Called inside transaction, so use GFP_NOFS
1020  */
1021 static struct inode *ext4_alloc_inode(struct super_block *sb)
1022 {
1023         struct ext4_inode_info *ei;
1024
1025         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1026         if (!ei)
1027                 return NULL;
1028
1029         inode_set_iversion(&ei->vfs_inode, 1);
1030         spin_lock_init(&ei->i_raw_lock);
1031         INIT_LIST_HEAD(&ei->i_prealloc_list);
1032         spin_lock_init(&ei->i_prealloc_lock);
1033         ext4_es_init_tree(&ei->i_es_tree);
1034         rwlock_init(&ei->i_es_lock);
1035         INIT_LIST_HEAD(&ei->i_es_list);
1036         ei->i_es_all_nr = 0;
1037         ei->i_es_shk_nr = 0;
1038         ei->i_es_shrink_lblk = 0;
1039         ei->i_reserved_data_blocks = 0;
1040         ei->i_da_metadata_calc_len = 0;
1041         ei->i_da_metadata_calc_last_lblock = 0;
1042         spin_lock_init(&(ei->i_block_reservation_lock));
1043 #ifdef CONFIG_QUOTA
1044         ei->i_reserved_quota = 0;
1045         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1046 #endif
1047         ei->jinode = NULL;
1048         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1049         spin_lock_init(&ei->i_completed_io_lock);
1050         ei->i_sync_tid = 0;
1051         ei->i_datasync_tid = 0;
1052         atomic_set(&ei->i_unwritten, 0);
1053         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1054         return &ei->vfs_inode;
1055 }
1056
1057 static int ext4_drop_inode(struct inode *inode)
1058 {
1059         int drop = generic_drop_inode(inode);
1060
1061         trace_ext4_drop_inode(inode, drop);
1062         return drop;
1063 }
1064
1065 static void ext4_i_callback(struct rcu_head *head)
1066 {
1067         struct inode *inode = container_of(head, struct inode, i_rcu);
1068         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1069 }
1070
1071 static void ext4_destroy_inode(struct inode *inode)
1072 {
1073         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1074                 ext4_msg(inode->i_sb, KERN_ERR,
1075                          "Inode %lu (%p): orphan list check failed!",
1076                          inode->i_ino, EXT4_I(inode));
1077                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1078                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1079                                 true);
1080                 dump_stack();
1081         }
1082         call_rcu(&inode->i_rcu, ext4_i_callback);
1083 }
1084
1085 static void init_once(void *foo)
1086 {
1087         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1088
1089         INIT_LIST_HEAD(&ei->i_orphan);
1090         init_rwsem(&ei->xattr_sem);
1091         init_rwsem(&ei->i_data_sem);
1092         init_rwsem(&ei->i_mmap_sem);
1093         inode_init_once(&ei->vfs_inode);
1094 }
1095
1096 static int __init init_inodecache(void)
1097 {
1098         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1099                                 sizeof(struct ext4_inode_info), 0,
1100                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1101                                         SLAB_ACCOUNT),
1102                                 offsetof(struct ext4_inode_info, i_data),
1103                                 sizeof_field(struct ext4_inode_info, i_data),
1104                                 init_once);
1105         if (ext4_inode_cachep == NULL)
1106                 return -ENOMEM;
1107         return 0;
1108 }
1109
1110 static void destroy_inodecache(void)
1111 {
1112         /*
1113          * Make sure all delayed rcu free inodes are flushed before we
1114          * destroy cache.
1115          */
1116         rcu_barrier();
1117         kmem_cache_destroy(ext4_inode_cachep);
1118 }
1119
1120 void ext4_clear_inode(struct inode *inode)
1121 {
1122         invalidate_inode_buffers(inode);
1123         clear_inode(inode);
1124         dquot_drop(inode);
1125         ext4_discard_preallocations(inode);
1126         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1127         if (EXT4_I(inode)->jinode) {
1128                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1129                                                EXT4_I(inode)->jinode);
1130                 jbd2_free_inode(EXT4_I(inode)->jinode);
1131                 EXT4_I(inode)->jinode = NULL;
1132         }
1133         fscrypt_put_encryption_info(inode);
1134 }
1135
1136 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1137                                         u64 ino, u32 generation)
1138 {
1139         struct inode *inode;
1140
1141         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1142                 return ERR_PTR(-ESTALE);
1143         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1144                 return ERR_PTR(-ESTALE);
1145
1146         /* iget isn't really right if the inode is currently unallocated!!
1147          *
1148          * ext4_read_inode will return a bad_inode if the inode had been
1149          * deleted, so we should be safe.
1150          *
1151          * Currently we don't know the generation for parent directory, so
1152          * a generation of 0 means "accept any"
1153          */
1154         inode = ext4_iget_normal(sb, ino);
1155         if (IS_ERR(inode))
1156                 return ERR_CAST(inode);
1157         if (generation && inode->i_generation != generation) {
1158                 iput(inode);
1159                 return ERR_PTR(-ESTALE);
1160         }
1161
1162         return inode;
1163 }
1164
1165 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1166                                         int fh_len, int fh_type)
1167 {
1168         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1169                                     ext4_nfs_get_inode);
1170 }
1171
1172 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1173                                         int fh_len, int fh_type)
1174 {
1175         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1176                                     ext4_nfs_get_inode);
1177 }
1178
1179 /*
1180  * Try to release metadata pages (indirect blocks, directories) which are
1181  * mapped via the block device.  Since these pages could have journal heads
1182  * which would prevent try_to_free_buffers() from freeing them, we must use
1183  * jbd2 layer's try_to_free_buffers() function to release them.
1184  */
1185 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1186                                  gfp_t wait)
1187 {
1188         journal_t *journal = EXT4_SB(sb)->s_journal;
1189
1190         WARN_ON(PageChecked(page));
1191         if (!page_has_buffers(page))
1192                 return 0;
1193         if (journal)
1194                 return jbd2_journal_try_to_free_buffers(journal, page,
1195                                                 wait & ~__GFP_DIRECT_RECLAIM);
1196         return try_to_free_buffers(page);
1197 }
1198
1199 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1200 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1201 {
1202         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1203                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1204 }
1205
1206 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1207                                                         void *fs_data)
1208 {
1209         handle_t *handle = fs_data;
1210         int res, res2, credits, retries = 0;
1211
1212         /*
1213          * Encrypting the root directory is not allowed because e2fsck expects
1214          * lost+found to exist and be unencrypted, and encrypting the root
1215          * directory would imply encrypting the lost+found directory as well as
1216          * the filename "lost+found" itself.
1217          */
1218         if (inode->i_ino == EXT4_ROOT_INO)
1219                 return -EPERM;
1220
1221         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1222                 return -EINVAL;
1223
1224         res = ext4_convert_inline_data(inode);
1225         if (res)
1226                 return res;
1227
1228         /*
1229          * If a journal handle was specified, then the encryption context is
1230          * being set on a new inode via inheritance and is part of a larger
1231          * transaction to create the inode.  Otherwise the encryption context is
1232          * being set on an existing inode in its own transaction.  Only in the
1233          * latter case should the "retry on ENOSPC" logic be used.
1234          */
1235
1236         if (handle) {
1237                 res = ext4_xattr_set_handle(handle, inode,
1238                                             EXT4_XATTR_INDEX_ENCRYPTION,
1239                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1240                                             ctx, len, 0);
1241                 if (!res) {
1242                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1243                         ext4_clear_inode_state(inode,
1244                                         EXT4_STATE_MAY_INLINE_DATA);
1245                         /*
1246                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1247                          * S_DAX may be disabled
1248                          */
1249                         ext4_set_inode_flags(inode);
1250                 }
1251                 return res;
1252         }
1253
1254         res = dquot_initialize(inode);
1255         if (res)
1256                 return res;
1257 retry:
1258         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1259                                      &credits);
1260         if (res)
1261                 return res;
1262
1263         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1264         if (IS_ERR(handle))
1265                 return PTR_ERR(handle);
1266
1267         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1268                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1269                                     ctx, len, 0);
1270         if (!res) {
1271                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1272                 /*
1273                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1274                  * S_DAX may be disabled
1275                  */
1276                 ext4_set_inode_flags(inode);
1277                 res = ext4_mark_inode_dirty(handle, inode);
1278                 if (res)
1279                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1280         }
1281         res2 = ext4_journal_stop(handle);
1282
1283         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1284                 goto retry;
1285         if (!res)
1286                 res = res2;
1287         return res;
1288 }
1289
1290 static bool ext4_dummy_context(struct inode *inode)
1291 {
1292         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1293 }
1294
1295 static const struct fscrypt_operations ext4_cryptops = {
1296         .key_prefix             = "ext4:",
1297         .get_context            = ext4_get_context,
1298         .set_context            = ext4_set_context,
1299         .dummy_context          = ext4_dummy_context,
1300         .empty_dir              = ext4_empty_dir,
1301         .max_namelen            = EXT4_NAME_LEN,
1302 };
1303 #endif
1304
1305 #ifdef CONFIG_QUOTA
1306 static const char * const quotatypes[] = INITQFNAMES;
1307 #define QTYPE2NAME(t) (quotatypes[t])
1308
1309 static int ext4_write_dquot(struct dquot *dquot);
1310 static int ext4_acquire_dquot(struct dquot *dquot);
1311 static int ext4_release_dquot(struct dquot *dquot);
1312 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1313 static int ext4_write_info(struct super_block *sb, int type);
1314 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1315                          const struct path *path);
1316 static int ext4_quota_on_mount(struct super_block *sb, int type);
1317 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1318                                size_t len, loff_t off);
1319 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1320                                 const char *data, size_t len, loff_t off);
1321 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1322                              unsigned int flags);
1323 static int ext4_enable_quotas(struct super_block *sb);
1324 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1325
1326 static struct dquot **ext4_get_dquots(struct inode *inode)
1327 {
1328         return EXT4_I(inode)->i_dquot;
1329 }
1330
1331 static const struct dquot_operations ext4_quota_operations = {
1332         .get_reserved_space     = ext4_get_reserved_space,
1333         .write_dquot            = ext4_write_dquot,
1334         .acquire_dquot          = ext4_acquire_dquot,
1335         .release_dquot          = ext4_release_dquot,
1336         .mark_dirty             = ext4_mark_dquot_dirty,
1337         .write_info             = ext4_write_info,
1338         .alloc_dquot            = dquot_alloc,
1339         .destroy_dquot          = dquot_destroy,
1340         .get_projid             = ext4_get_projid,
1341         .get_inode_usage        = ext4_get_inode_usage,
1342         .get_next_id            = ext4_get_next_id,
1343 };
1344
1345 static const struct quotactl_ops ext4_qctl_operations = {
1346         .quota_on       = ext4_quota_on,
1347         .quota_off      = ext4_quota_off,
1348         .quota_sync     = dquot_quota_sync,
1349         .get_state      = dquot_get_state,
1350         .set_info       = dquot_set_dqinfo,
1351         .get_dqblk      = dquot_get_dqblk,
1352         .set_dqblk      = dquot_set_dqblk,
1353         .get_nextdqblk  = dquot_get_next_dqblk,
1354 };
1355 #endif
1356
1357 static const struct super_operations ext4_sops = {
1358         .alloc_inode    = ext4_alloc_inode,
1359         .destroy_inode  = ext4_destroy_inode,
1360         .write_inode    = ext4_write_inode,
1361         .dirty_inode    = ext4_dirty_inode,
1362         .drop_inode     = ext4_drop_inode,
1363         .evict_inode    = ext4_evict_inode,
1364         .put_super      = ext4_put_super,
1365         .sync_fs        = ext4_sync_fs,
1366         .freeze_fs      = ext4_freeze,
1367         .unfreeze_fs    = ext4_unfreeze,
1368         .statfs         = ext4_statfs,
1369         .remount_fs     = ext4_remount,
1370         .show_options   = ext4_show_options,
1371 #ifdef CONFIG_QUOTA
1372         .quota_read     = ext4_quota_read,
1373         .quota_write    = ext4_quota_write,
1374         .get_dquots     = ext4_get_dquots,
1375 #endif
1376         .bdev_try_to_free_page = bdev_try_to_free_page,
1377 };
1378
1379 static const struct export_operations ext4_export_ops = {
1380         .fh_to_dentry = ext4_fh_to_dentry,
1381         .fh_to_parent = ext4_fh_to_parent,
1382         .get_parent = ext4_get_parent,
1383 };
1384
1385 enum {
1386         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1387         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1388         Opt_nouid32, Opt_debug, Opt_removed,
1389         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1390         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1391         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1392         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1393         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1394         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1395         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1396         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1397         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1398         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1399         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1400         Opt_nowarn_on_error, Opt_mblk_io_submit,
1401         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1402         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1403         Opt_inode_readahead_blks, Opt_journal_ioprio,
1404         Opt_dioread_nolock, Opt_dioread_lock,
1405         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1406         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1407 };
1408
1409 static const match_table_t tokens = {
1410         {Opt_bsd_df, "bsddf"},
1411         {Opt_minix_df, "minixdf"},
1412         {Opt_grpid, "grpid"},
1413         {Opt_grpid, "bsdgroups"},
1414         {Opt_nogrpid, "nogrpid"},
1415         {Opt_nogrpid, "sysvgroups"},
1416         {Opt_resgid, "resgid=%u"},
1417         {Opt_resuid, "resuid=%u"},
1418         {Opt_sb, "sb=%u"},
1419         {Opt_err_cont, "errors=continue"},
1420         {Opt_err_panic, "errors=panic"},
1421         {Opt_err_ro, "errors=remount-ro"},
1422         {Opt_nouid32, "nouid32"},
1423         {Opt_debug, "debug"},
1424         {Opt_removed, "oldalloc"},
1425         {Opt_removed, "orlov"},
1426         {Opt_user_xattr, "user_xattr"},
1427         {Opt_nouser_xattr, "nouser_xattr"},
1428         {Opt_acl, "acl"},
1429         {Opt_noacl, "noacl"},
1430         {Opt_noload, "norecovery"},
1431         {Opt_noload, "noload"},
1432         {Opt_removed, "nobh"},
1433         {Opt_removed, "bh"},
1434         {Opt_commit, "commit=%u"},
1435         {Opt_min_batch_time, "min_batch_time=%u"},
1436         {Opt_max_batch_time, "max_batch_time=%u"},
1437         {Opt_journal_dev, "journal_dev=%u"},
1438         {Opt_journal_path, "journal_path=%s"},
1439         {Opt_journal_checksum, "journal_checksum"},
1440         {Opt_nojournal_checksum, "nojournal_checksum"},
1441         {Opt_journal_async_commit, "journal_async_commit"},
1442         {Opt_abort, "abort"},
1443         {Opt_data_journal, "data=journal"},
1444         {Opt_data_ordered, "data=ordered"},
1445         {Opt_data_writeback, "data=writeback"},
1446         {Opt_data_err_abort, "data_err=abort"},
1447         {Opt_data_err_ignore, "data_err=ignore"},
1448         {Opt_offusrjquota, "usrjquota="},
1449         {Opt_usrjquota, "usrjquota=%s"},
1450         {Opt_offgrpjquota, "grpjquota="},
1451         {Opt_grpjquota, "grpjquota=%s"},
1452         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1453         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1454         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1455         {Opt_grpquota, "grpquota"},
1456         {Opt_noquota, "noquota"},
1457         {Opt_quota, "quota"},
1458         {Opt_usrquota, "usrquota"},
1459         {Opt_prjquota, "prjquota"},
1460         {Opt_barrier, "barrier=%u"},
1461         {Opt_barrier, "barrier"},
1462         {Opt_nobarrier, "nobarrier"},
1463         {Opt_i_version, "i_version"},
1464         {Opt_dax, "dax"},
1465         {Opt_stripe, "stripe=%u"},
1466         {Opt_delalloc, "delalloc"},
1467         {Opt_warn_on_error, "warn_on_error"},
1468         {Opt_nowarn_on_error, "nowarn_on_error"},
1469         {Opt_lazytime, "lazytime"},
1470         {Opt_nolazytime, "nolazytime"},
1471         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1472         {Opt_nodelalloc, "nodelalloc"},
1473         {Opt_removed, "mblk_io_submit"},
1474         {Opt_removed, "nomblk_io_submit"},
1475         {Opt_block_validity, "block_validity"},
1476         {Opt_noblock_validity, "noblock_validity"},
1477         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1478         {Opt_journal_ioprio, "journal_ioprio=%u"},
1479         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1480         {Opt_auto_da_alloc, "auto_da_alloc"},
1481         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1482         {Opt_dioread_nolock, "dioread_nolock"},
1483         {Opt_dioread_lock, "dioread_lock"},
1484         {Opt_discard, "discard"},
1485         {Opt_nodiscard, "nodiscard"},
1486         {Opt_init_itable, "init_itable=%u"},
1487         {Opt_init_itable, "init_itable"},
1488         {Opt_noinit_itable, "noinit_itable"},
1489         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1490         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1491         {Opt_nombcache, "nombcache"},
1492         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1493         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1494         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1495         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1496         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1497         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1498         {Opt_err, NULL},
1499 };
1500
1501 static ext4_fsblk_t get_sb_block(void **data)
1502 {
1503         ext4_fsblk_t    sb_block;
1504         char            *options = (char *) *data;
1505
1506         if (!options || strncmp(options, "sb=", 3) != 0)
1507                 return 1;       /* Default location */
1508
1509         options += 3;
1510         /* TODO: use simple_strtoll with >32bit ext4 */
1511         sb_block = simple_strtoul(options, &options, 0);
1512         if (*options && *options != ',') {
1513                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1514                        (char *) *data);
1515                 return 1;
1516         }
1517         if (*options == ',')
1518                 options++;
1519         *data = (void *) options;
1520
1521         return sb_block;
1522 }
1523
1524 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1525 static const char deprecated_msg[] =
1526         "Mount option \"%s\" will be removed by %s\n"
1527         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1528
1529 #ifdef CONFIG_QUOTA
1530 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1531 {
1532         struct ext4_sb_info *sbi = EXT4_SB(sb);
1533         char *qname;
1534         int ret = -1;
1535
1536         if (sb_any_quota_loaded(sb) &&
1537                 !sbi->s_qf_names[qtype]) {
1538                 ext4_msg(sb, KERN_ERR,
1539                         "Cannot change journaled "
1540                         "quota options when quota turned on");
1541                 return -1;
1542         }
1543         if (ext4_has_feature_quota(sb)) {
1544                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1545                          "ignored when QUOTA feature is enabled");
1546                 return 1;
1547         }
1548         qname = match_strdup(args);
1549         if (!qname) {
1550                 ext4_msg(sb, KERN_ERR,
1551                         "Not enough memory for storing quotafile name");
1552                 return -1;
1553         }
1554         if (sbi->s_qf_names[qtype]) {
1555                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1556                         ret = 1;
1557                 else
1558                         ext4_msg(sb, KERN_ERR,
1559                                  "%s quota file already specified",
1560                                  QTYPE2NAME(qtype));
1561                 goto errout;
1562         }
1563         if (strchr(qname, '/')) {
1564                 ext4_msg(sb, KERN_ERR,
1565                         "quotafile must be on filesystem root");
1566                 goto errout;
1567         }
1568         sbi->s_qf_names[qtype] = qname;
1569         set_opt(sb, QUOTA);
1570         return 1;
1571 errout:
1572         kfree(qname);
1573         return ret;
1574 }
1575
1576 static int clear_qf_name(struct super_block *sb, int qtype)
1577 {
1578
1579         struct ext4_sb_info *sbi = EXT4_SB(sb);
1580
1581         if (sb_any_quota_loaded(sb) &&
1582                 sbi->s_qf_names[qtype]) {
1583                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1584                         " when quota turned on");
1585                 return -1;
1586         }
1587         kfree(sbi->s_qf_names[qtype]);
1588         sbi->s_qf_names[qtype] = NULL;
1589         return 1;
1590 }
1591 #endif
1592
1593 #define MOPT_SET        0x0001
1594 #define MOPT_CLEAR      0x0002
1595 #define MOPT_NOSUPPORT  0x0004
1596 #define MOPT_EXPLICIT   0x0008
1597 #define MOPT_CLEAR_ERR  0x0010
1598 #define MOPT_GTE0       0x0020
1599 #ifdef CONFIG_QUOTA
1600 #define MOPT_Q          0
1601 #define MOPT_QFMT       0x0040
1602 #else
1603 #define MOPT_Q          MOPT_NOSUPPORT
1604 #define MOPT_QFMT       MOPT_NOSUPPORT
1605 #endif
1606 #define MOPT_DATAJ      0x0080
1607 #define MOPT_NO_EXT2    0x0100
1608 #define MOPT_NO_EXT3    0x0200
1609 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1610 #define MOPT_STRING     0x0400
1611
1612 static const struct mount_opts {
1613         int     token;
1614         int     mount_opt;
1615         int     flags;
1616 } ext4_mount_opts[] = {
1617         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1618         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1619         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1620         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1621         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1622         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1623         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1624          MOPT_EXT4_ONLY | MOPT_SET},
1625         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1626          MOPT_EXT4_ONLY | MOPT_CLEAR},
1627         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1628         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1629         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1630          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1631         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1632          MOPT_EXT4_ONLY | MOPT_CLEAR},
1633         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1634         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1635         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1636          MOPT_EXT4_ONLY | MOPT_CLEAR},
1637         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1638          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1639         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1640                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1641          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1642         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1643         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1644         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1645         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1646         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1647          MOPT_NO_EXT2},
1648         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1649          MOPT_NO_EXT2},
1650         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1651         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1652         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1653         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1654         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1655         {Opt_commit, 0, MOPT_GTE0},
1656         {Opt_max_batch_time, 0, MOPT_GTE0},
1657         {Opt_min_batch_time, 0, MOPT_GTE0},
1658         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1659         {Opt_init_itable, 0, MOPT_GTE0},
1660         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1661         {Opt_stripe, 0, MOPT_GTE0},
1662         {Opt_resuid, 0, MOPT_GTE0},
1663         {Opt_resgid, 0, MOPT_GTE0},
1664         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1665         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1666         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1667         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1668         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1669         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1670          MOPT_NO_EXT2 | MOPT_DATAJ},
1671         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1672         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1673 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1674         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1675         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1676 #else
1677         {Opt_acl, 0, MOPT_NOSUPPORT},
1678         {Opt_noacl, 0, MOPT_NOSUPPORT},
1679 #endif
1680         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1681         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1682         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1683         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1684         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1685                                                         MOPT_SET | MOPT_Q},
1686         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1687                                                         MOPT_SET | MOPT_Q},
1688         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1689                                                         MOPT_SET | MOPT_Q},
1690         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1691                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1692                                                         MOPT_CLEAR | MOPT_Q},
1693         {Opt_usrjquota, 0, MOPT_Q},
1694         {Opt_grpjquota, 0, MOPT_Q},
1695         {Opt_offusrjquota, 0, MOPT_Q},
1696         {Opt_offgrpjquota, 0, MOPT_Q},
1697         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1698         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1699         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1700         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1701         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1702         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1703         {Opt_err, 0, 0}
1704 };
1705
1706 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1707                             substring_t *args, unsigned long *journal_devnum,
1708                             unsigned int *journal_ioprio, int is_remount)
1709 {
1710         struct ext4_sb_info *sbi = EXT4_SB(sb);
1711         const struct mount_opts *m;
1712         kuid_t uid;
1713         kgid_t gid;
1714         int arg = 0;
1715
1716 #ifdef CONFIG_QUOTA
1717         if (token == Opt_usrjquota)
1718                 return set_qf_name(sb, USRQUOTA, &args[0]);
1719         else if (token == Opt_grpjquota)
1720                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1721         else if (token == Opt_offusrjquota)
1722                 return clear_qf_name(sb, USRQUOTA);
1723         else if (token == Opt_offgrpjquota)
1724                 return clear_qf_name(sb, GRPQUOTA);
1725 #endif
1726         switch (token) {
1727         case Opt_noacl:
1728         case Opt_nouser_xattr:
1729                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1730                 break;
1731         case Opt_sb:
1732                 return 1;       /* handled by get_sb_block() */
1733         case Opt_removed:
1734                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1735                 return 1;
1736         case Opt_abort:
1737                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1738                 return 1;
1739         case Opt_i_version:
1740                 sb->s_flags |= SB_I_VERSION;
1741                 return 1;
1742         case Opt_lazytime:
1743                 sb->s_flags |= SB_LAZYTIME;
1744                 return 1;
1745         case Opt_nolazytime:
1746                 sb->s_flags &= ~SB_LAZYTIME;
1747                 return 1;
1748         }
1749
1750         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1751                 if (token == m->token)
1752                         break;
1753
1754         if (m->token == Opt_err) {
1755                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1756                          "or missing value", opt);
1757                 return -1;
1758         }
1759
1760         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1761                 ext4_msg(sb, KERN_ERR,
1762                          "Mount option \"%s\" incompatible with ext2", opt);
1763                 return -1;
1764         }
1765         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1766                 ext4_msg(sb, KERN_ERR,
1767                          "Mount option \"%s\" incompatible with ext3", opt);
1768                 return -1;
1769         }
1770
1771         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1772                 return -1;
1773         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1774                 return -1;
1775         if (m->flags & MOPT_EXPLICIT) {
1776                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1777                         set_opt2(sb, EXPLICIT_DELALLOC);
1778                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1779                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1780                 } else
1781                         return -1;
1782         }
1783         if (m->flags & MOPT_CLEAR_ERR)
1784                 clear_opt(sb, ERRORS_MASK);
1785         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1786                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1787                          "options when quota turned on");
1788                 return -1;
1789         }
1790
1791         if (m->flags & MOPT_NOSUPPORT) {
1792                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1793         } else if (token == Opt_commit) {
1794                 if (arg == 0)
1795                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1796                 sbi->s_commit_interval = HZ * arg;
1797         } else if (token == Opt_debug_want_extra_isize) {
1798                 sbi->s_want_extra_isize = arg;
1799         } else if (token == Opt_max_batch_time) {
1800                 sbi->s_max_batch_time = arg;
1801         } else if (token == Opt_min_batch_time) {
1802                 sbi->s_min_batch_time = arg;
1803         } else if (token == Opt_inode_readahead_blks) {
1804                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1805                         ext4_msg(sb, KERN_ERR,
1806                                  "EXT4-fs: inode_readahead_blks must be "
1807                                  "0 or a power of 2 smaller than 2^31");
1808                         return -1;
1809                 }
1810                 sbi->s_inode_readahead_blks = arg;
1811         } else if (token == Opt_init_itable) {
1812                 set_opt(sb, INIT_INODE_TABLE);
1813                 if (!args->from)
1814                         arg = EXT4_DEF_LI_WAIT_MULT;
1815                 sbi->s_li_wait_mult = arg;
1816         } else if (token == Opt_max_dir_size_kb) {
1817                 sbi->s_max_dir_size_kb = arg;
1818         } else if (token == Opt_stripe) {
1819                 sbi->s_stripe = arg;
1820         } else if (token == Opt_resuid) {
1821                 uid = make_kuid(current_user_ns(), arg);
1822                 if (!uid_valid(uid)) {
1823                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1824                         return -1;
1825                 }
1826                 sbi->s_resuid = uid;
1827         } else if (token == Opt_resgid) {
1828                 gid = make_kgid(current_user_ns(), arg);
1829                 if (!gid_valid(gid)) {
1830                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1831                         return -1;
1832                 }
1833                 sbi->s_resgid = gid;
1834         } else if (token == Opt_journal_dev) {
1835                 if (is_remount) {
1836                         ext4_msg(sb, KERN_ERR,
1837                                  "Cannot specify journal on remount");
1838                         return -1;
1839                 }
1840                 *journal_devnum = arg;
1841         } else if (token == Opt_journal_path) {
1842                 char *journal_path;
1843                 struct inode *journal_inode;
1844                 struct path path;
1845                 int error;
1846
1847                 if (is_remount) {
1848                         ext4_msg(sb, KERN_ERR,
1849                                  "Cannot specify journal on remount");
1850                         return -1;
1851                 }
1852                 journal_path = match_strdup(&args[0]);
1853                 if (!journal_path) {
1854                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1855                                 "journal device string");
1856                         return -1;
1857                 }
1858
1859                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1860                 if (error) {
1861                         ext4_msg(sb, KERN_ERR, "error: could not find "
1862                                 "journal device path: error %d", error);
1863                         kfree(journal_path);
1864                         return -1;
1865                 }
1866
1867                 journal_inode = d_inode(path.dentry);
1868                 if (!S_ISBLK(journal_inode->i_mode)) {
1869                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1870                                 "is not a block device", journal_path);
1871                         path_put(&path);
1872                         kfree(journal_path);
1873                         return -1;
1874                 }
1875
1876                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1877                 path_put(&path);
1878                 kfree(journal_path);
1879         } else if (token == Opt_journal_ioprio) {
1880                 if (arg > 7) {
1881                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1882                                  " (must be 0-7)");
1883                         return -1;
1884                 }
1885                 *journal_ioprio =
1886                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1887         } else if (token == Opt_test_dummy_encryption) {
1888 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1889                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1890                 ext4_msg(sb, KERN_WARNING,
1891                          "Test dummy encryption mode enabled");
1892 #else
1893                 ext4_msg(sb, KERN_WARNING,
1894                          "Test dummy encryption mount option ignored");
1895 #endif
1896         } else if (m->flags & MOPT_DATAJ) {
1897                 if (is_remount) {
1898                         if (!sbi->s_journal)
1899                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1900                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1901                                 ext4_msg(sb, KERN_ERR,
1902                                          "Cannot change data mode on remount");
1903                                 return -1;
1904                         }
1905                 } else {
1906                         clear_opt(sb, DATA_FLAGS);
1907                         sbi->s_mount_opt |= m->mount_opt;
1908                 }
1909 #ifdef CONFIG_QUOTA
1910         } else if (m->flags & MOPT_QFMT) {
1911                 if (sb_any_quota_loaded(sb) &&
1912                     sbi->s_jquota_fmt != m->mount_opt) {
1913                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1914                                  "quota options when quota turned on");
1915                         return -1;
1916                 }
1917                 if (ext4_has_feature_quota(sb)) {
1918                         ext4_msg(sb, KERN_INFO,
1919                                  "Quota format mount options ignored "
1920                                  "when QUOTA feature is enabled");
1921                         return 1;
1922                 }
1923                 sbi->s_jquota_fmt = m->mount_opt;
1924 #endif
1925         } else if (token == Opt_dax) {
1926 #ifdef CONFIG_FS_DAX
1927                 ext4_msg(sb, KERN_WARNING,
1928                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1929                         sbi->s_mount_opt |= m->mount_opt;
1930 #else
1931                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1932                 return -1;
1933 #endif
1934         } else if (token == Opt_data_err_abort) {
1935                 sbi->s_mount_opt |= m->mount_opt;
1936         } else if (token == Opt_data_err_ignore) {
1937                 sbi->s_mount_opt &= ~m->mount_opt;
1938         } else {
1939                 if (!args->from)
1940                         arg = 1;
1941                 if (m->flags & MOPT_CLEAR)
1942                         arg = !arg;
1943                 else if (unlikely(!(m->flags & MOPT_SET))) {
1944                         ext4_msg(sb, KERN_WARNING,
1945                                  "buggy handling of option %s", opt);
1946                         WARN_ON(1);
1947                         return -1;
1948                 }
1949                 if (arg != 0)
1950                         sbi->s_mount_opt |= m->mount_opt;
1951                 else
1952                         sbi->s_mount_opt &= ~m->mount_opt;
1953         }
1954         return 1;
1955 }
1956
1957 static int parse_options(char *options, struct super_block *sb,
1958                          unsigned long *journal_devnum,
1959                          unsigned int *journal_ioprio,
1960                          int is_remount)
1961 {
1962         struct ext4_sb_info *sbi = EXT4_SB(sb);
1963         char *p;
1964         substring_t args[MAX_OPT_ARGS];
1965         int token;
1966
1967         if (!options)
1968                 return 1;
1969
1970         while ((p = strsep(&options, ",")) != NULL) {
1971                 if (!*p)
1972                         continue;
1973                 /*
1974                  * Initialize args struct so we know whether arg was
1975                  * found; some options take optional arguments.
1976                  */
1977                 args[0].to = args[0].from = NULL;
1978                 token = match_token(p, tokens, args);
1979                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1980                                      journal_ioprio, is_remount) < 0)
1981                         return 0;
1982         }
1983 #ifdef CONFIG_QUOTA
1984         /*
1985          * We do the test below only for project quotas. 'usrquota' and
1986          * 'grpquota' mount options are allowed even without quota feature
1987          * to support legacy quotas in quota files.
1988          */
1989         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1990                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1991                          "Cannot enable project quota enforcement.");
1992                 return 0;
1993         }
1994         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1995                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1996                         clear_opt(sb, USRQUOTA);
1997
1998                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1999                         clear_opt(sb, GRPQUOTA);
2000
2001                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2002                         ext4_msg(sb, KERN_ERR, "old and new quota "
2003                                         "format mixing");
2004                         return 0;
2005                 }
2006
2007                 if (!sbi->s_jquota_fmt) {
2008                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2009                                         "not specified");
2010                         return 0;
2011                 }
2012         }
2013 #endif
2014         if (test_opt(sb, DIOREAD_NOLOCK)) {
2015                 int blocksize =
2016                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2017
2018                 if (blocksize < PAGE_SIZE) {
2019                         ext4_msg(sb, KERN_ERR, "can't mount with "
2020                                  "dioread_nolock if block size != PAGE_SIZE");
2021                         return 0;
2022                 }
2023         }
2024         return 1;
2025 }
2026
2027 static inline void ext4_show_quota_options(struct seq_file *seq,
2028                                            struct super_block *sb)
2029 {
2030 #if defined(CONFIG_QUOTA)
2031         struct ext4_sb_info *sbi = EXT4_SB(sb);
2032
2033         if (sbi->s_jquota_fmt) {
2034                 char *fmtname = "";
2035
2036                 switch (sbi->s_jquota_fmt) {
2037                 case QFMT_VFS_OLD:
2038                         fmtname = "vfsold";
2039                         break;
2040                 case QFMT_VFS_V0:
2041                         fmtname = "vfsv0";
2042                         break;
2043                 case QFMT_VFS_V1:
2044                         fmtname = "vfsv1";
2045                         break;
2046                 }
2047                 seq_printf(seq, ",jqfmt=%s", fmtname);
2048         }
2049
2050         if (sbi->s_qf_names[USRQUOTA])
2051                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
2052
2053         if (sbi->s_qf_names[GRPQUOTA])
2054                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
2055 #endif
2056 }
2057
2058 static const char *token2str(int token)
2059 {
2060         const struct match_token *t;
2061
2062         for (t = tokens; t->token != Opt_err; t++)
2063                 if (t->token == token && !strchr(t->pattern, '='))
2064                         break;
2065         return t->pattern;
2066 }
2067
2068 /*
2069  * Show an option if
2070  *  - it's set to a non-default value OR
2071  *  - if the per-sb default is different from the global default
2072  */
2073 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2074                               int nodefs)
2075 {
2076         struct ext4_sb_info *sbi = EXT4_SB(sb);
2077         struct ext4_super_block *es = sbi->s_es;
2078         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2079         const struct mount_opts *m;
2080         char sep = nodefs ? '\n' : ',';
2081
2082 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2083 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2084
2085         if (sbi->s_sb_block != 1)
2086                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2087
2088         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2089                 int want_set = m->flags & MOPT_SET;
2090                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2091                     (m->flags & MOPT_CLEAR_ERR))
2092                         continue;
2093                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2094                         continue; /* skip if same as the default */
2095                 if ((want_set &&
2096                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2097                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2098                         continue; /* select Opt_noFoo vs Opt_Foo */
2099                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2100         }
2101
2102         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2103             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2104                 SEQ_OPTS_PRINT("resuid=%u",
2105                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2106         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2107             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2108                 SEQ_OPTS_PRINT("resgid=%u",
2109                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2110         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2111         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2112                 SEQ_OPTS_PUTS("errors=remount-ro");
2113         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2114                 SEQ_OPTS_PUTS("errors=continue");
2115         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2116                 SEQ_OPTS_PUTS("errors=panic");
2117         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2118                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2119         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2120                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2121         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2122                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2123         if (sb->s_flags & SB_I_VERSION)
2124                 SEQ_OPTS_PUTS("i_version");
2125         if (nodefs || sbi->s_stripe)
2126                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2127         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2128                         (sbi->s_mount_opt ^ def_mount_opt)) {
2129                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2130                         SEQ_OPTS_PUTS("data=journal");
2131                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2132                         SEQ_OPTS_PUTS("data=ordered");
2133                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2134                         SEQ_OPTS_PUTS("data=writeback");
2135         }
2136         if (nodefs ||
2137             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2138                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2139                                sbi->s_inode_readahead_blks);
2140
2141         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2142                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2143                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2144         if (nodefs || sbi->s_max_dir_size_kb)
2145                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2146         if (test_opt(sb, DATA_ERR_ABORT))
2147                 SEQ_OPTS_PUTS("data_err=abort");
2148
2149         ext4_show_quota_options(seq, sb);
2150         return 0;
2151 }
2152
2153 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2154 {
2155         return _ext4_show_options(seq, root->d_sb, 0);
2156 }
2157
2158 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2159 {
2160         struct super_block *sb = seq->private;
2161         int rc;
2162
2163         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2164         rc = _ext4_show_options(seq, sb, 1);
2165         seq_puts(seq, "\n");
2166         return rc;
2167 }
2168
2169 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2170                             int read_only)
2171 {
2172         struct ext4_sb_info *sbi = EXT4_SB(sb);
2173         int err = 0;
2174
2175         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2176                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2177                          "forcing read-only mode");
2178                 err = -EROFS;
2179         }
2180         if (read_only)
2181                 goto done;
2182         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2183                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2184                          "running e2fsck is recommended");
2185         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2186                 ext4_msg(sb, KERN_WARNING,
2187                          "warning: mounting fs with errors, "
2188                          "running e2fsck is recommended");
2189         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2190                  le16_to_cpu(es->s_mnt_count) >=
2191                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2192                 ext4_msg(sb, KERN_WARNING,
2193                          "warning: maximal mount count reached, "
2194                          "running e2fsck is recommended");
2195         else if (le32_to_cpu(es->s_checkinterval) &&
2196                  (ext4_get_tstamp(es, s_lastcheck) +
2197                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2198                 ext4_msg(sb, KERN_WARNING,
2199                          "warning: checktime reached, "
2200                          "running e2fsck is recommended");
2201         if (!sbi->s_journal)
2202                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2203         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2204                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2205         le16_add_cpu(&es->s_mnt_count, 1);
2206         ext4_update_tstamp(es, s_mtime);
2207         ext4_update_dynamic_rev(sb);
2208         if (sbi->s_journal)
2209                 ext4_set_feature_journal_needs_recovery(sb);
2210
2211         err = ext4_commit_super(sb, 1);
2212 done:
2213         if (test_opt(sb, DEBUG))
2214                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2215                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2216                         sb->s_blocksize,
2217                         sbi->s_groups_count,
2218                         EXT4_BLOCKS_PER_GROUP(sb),
2219                         EXT4_INODES_PER_GROUP(sb),
2220                         sbi->s_mount_opt, sbi->s_mount_opt2);
2221
2222         cleancache_init_fs(sb);
2223         return err;
2224 }
2225
2226 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2227 {
2228         struct ext4_sb_info *sbi = EXT4_SB(sb);
2229         struct flex_groups *new_groups;
2230         int size;
2231
2232         if (!sbi->s_log_groups_per_flex)
2233                 return 0;
2234
2235         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2236         if (size <= sbi->s_flex_groups_allocated)
2237                 return 0;
2238
2239         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2240         new_groups = kvzalloc(size, GFP_KERNEL);
2241         if (!new_groups) {
2242                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2243                          size / (int) sizeof(struct flex_groups));
2244                 return -ENOMEM;
2245         }
2246
2247         if (sbi->s_flex_groups) {
2248                 memcpy(new_groups, sbi->s_flex_groups,
2249                        (sbi->s_flex_groups_allocated *
2250                         sizeof(struct flex_groups)));
2251                 kvfree(sbi->s_flex_groups);
2252         }
2253         sbi->s_flex_groups = new_groups;
2254         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2255         return 0;
2256 }
2257
2258 static int ext4_fill_flex_info(struct super_block *sb)
2259 {
2260         struct ext4_sb_info *sbi = EXT4_SB(sb);
2261         struct ext4_group_desc *gdp = NULL;
2262         ext4_group_t flex_group;
2263         int i, err;
2264
2265         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2266         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2267                 sbi->s_log_groups_per_flex = 0;
2268                 return 1;
2269         }
2270
2271         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2272         if (err)
2273                 goto failed;
2274
2275         for (i = 0; i < sbi->s_groups_count; i++) {
2276                 gdp = ext4_get_group_desc(sb, i, NULL);
2277
2278                 flex_group = ext4_flex_group(sbi, i);
2279                 atomic_add(ext4_free_inodes_count(sb, gdp),
2280                            &sbi->s_flex_groups[flex_group].free_inodes);
2281                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2282                              &sbi->s_flex_groups[flex_group].free_clusters);
2283                 atomic_add(ext4_used_dirs_count(sb, gdp),
2284                            &sbi->s_flex_groups[flex_group].used_dirs);
2285         }
2286
2287         return 1;
2288 failed:
2289         return 0;
2290 }
2291
2292 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2293                                    struct ext4_group_desc *gdp)
2294 {
2295         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2296         __u16 crc = 0;
2297         __le32 le_group = cpu_to_le32(block_group);
2298         struct ext4_sb_info *sbi = EXT4_SB(sb);
2299
2300         if (ext4_has_metadata_csum(sbi->s_sb)) {
2301                 /* Use new metadata_csum algorithm */
2302                 __u32 csum32;
2303                 __u16 dummy_csum = 0;
2304
2305                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2306                                      sizeof(le_group));
2307                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2308                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2309                                      sizeof(dummy_csum));
2310                 offset += sizeof(dummy_csum);
2311                 if (offset < sbi->s_desc_size)
2312                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2313                                              sbi->s_desc_size - offset);
2314
2315                 crc = csum32 & 0xFFFF;
2316                 goto out;
2317         }
2318
2319         /* old crc16 code */
2320         if (!ext4_has_feature_gdt_csum(sb))
2321                 return 0;
2322
2323         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2324         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2325         crc = crc16(crc, (__u8 *)gdp, offset);
2326         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2327         /* for checksum of struct ext4_group_desc do the rest...*/
2328         if (ext4_has_feature_64bit(sb) &&
2329             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2330                 crc = crc16(crc, (__u8 *)gdp + offset,
2331                             le16_to_cpu(sbi->s_es->s_desc_size) -
2332                                 offset);
2333
2334 out:
2335         return cpu_to_le16(crc);
2336 }
2337
2338 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2339                                 struct ext4_group_desc *gdp)
2340 {
2341         if (ext4_has_group_desc_csum(sb) &&
2342             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2343                 return 0;
2344
2345         return 1;
2346 }
2347
2348 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2349                               struct ext4_group_desc *gdp)
2350 {
2351         if (!ext4_has_group_desc_csum(sb))
2352                 return;
2353         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2354 }
2355
2356 /* Called at mount-time, super-block is locked */
2357 static int ext4_check_descriptors(struct super_block *sb,
2358                                   ext4_fsblk_t sb_block,
2359                                   ext4_group_t *first_not_zeroed)
2360 {
2361         struct ext4_sb_info *sbi = EXT4_SB(sb);
2362         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2363         ext4_fsblk_t last_block;
2364         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2365         ext4_fsblk_t block_bitmap;
2366         ext4_fsblk_t inode_bitmap;
2367         ext4_fsblk_t inode_table;
2368         int flexbg_flag = 0;
2369         ext4_group_t i, grp = sbi->s_groups_count;
2370
2371         if (ext4_has_feature_flex_bg(sb))
2372                 flexbg_flag = 1;
2373
2374         ext4_debug("Checking group descriptors");
2375
2376         for (i = 0; i < sbi->s_groups_count; i++) {
2377                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2378
2379                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2380                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2381                 else
2382                         last_block = first_block +
2383                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2384
2385                 if ((grp == sbi->s_groups_count) &&
2386                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2387                         grp = i;
2388
2389                 block_bitmap = ext4_block_bitmap(sb, gdp);
2390                 if (block_bitmap == sb_block) {
2391                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2392                                  "Block bitmap for group %u overlaps "
2393                                  "superblock", i);
2394                         if (!sb_rdonly(sb))
2395                                 return 0;
2396                 }
2397                 if (block_bitmap >= sb_block + 1 &&
2398                     block_bitmap <= last_bg_block) {
2399                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2400                                  "Block bitmap for group %u overlaps "
2401                                  "block group descriptors", i);
2402                         if (!sb_rdonly(sb))
2403                                 return 0;
2404                 }
2405                 if (block_bitmap < first_block || block_bitmap > last_block) {
2406                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2407                                "Block bitmap for group %u not in group "
2408                                "(block %llu)!", i, block_bitmap);
2409                         return 0;
2410                 }
2411                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2412                 if (inode_bitmap == sb_block) {
2413                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2414                                  "Inode bitmap for group %u overlaps "
2415                                  "superblock", i);
2416                         if (!sb_rdonly(sb))
2417                                 return 0;
2418                 }
2419                 if (inode_bitmap >= sb_block + 1 &&
2420                     inode_bitmap <= last_bg_block) {
2421                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2422                                  "Inode bitmap for group %u overlaps "
2423                                  "block group descriptors", i);
2424                         if (!sb_rdonly(sb))
2425                                 return 0;
2426                 }
2427                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2428                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2429                                "Inode bitmap for group %u not in group "
2430                                "(block %llu)!", i, inode_bitmap);
2431                         return 0;
2432                 }
2433                 inode_table = ext4_inode_table(sb, gdp);
2434                 if (inode_table == sb_block) {
2435                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2436                                  "Inode table for group %u overlaps "
2437                                  "superblock", i);
2438                         if (!sb_rdonly(sb))
2439                                 return 0;
2440                 }
2441                 if (inode_table >= sb_block + 1 &&
2442                     inode_table <= last_bg_block) {
2443                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2444                                  "Inode table for group %u overlaps "
2445                                  "block group descriptors", i);
2446                         if (!sb_rdonly(sb))
2447                                 return 0;
2448                 }
2449                 if (inode_table < first_block ||
2450                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2451                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2452                                "Inode table for group %u not in group "
2453                                "(block %llu)!", i, inode_table);
2454                         return 0;
2455                 }
2456                 ext4_lock_group(sb, i);
2457                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2458                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2459                                  "Checksum for group %u failed (%u!=%u)",
2460                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2461                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2462                         if (!sb_rdonly(sb)) {
2463                                 ext4_unlock_group(sb, i);
2464                                 return 0;
2465                         }
2466                 }
2467                 ext4_unlock_group(sb, i);
2468                 if (!flexbg_flag)
2469                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2470         }
2471         if (NULL != first_not_zeroed)
2472                 *first_not_zeroed = grp;
2473         return 1;
2474 }
2475
2476 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2477  * the superblock) which were deleted from all directories, but held open by
2478  * a process at the time of a crash.  We walk the list and try to delete these
2479  * inodes at recovery time (only with a read-write filesystem).
2480  *
2481  * In order to keep the orphan inode chain consistent during traversal (in
2482  * case of crash during recovery), we link each inode into the superblock
2483  * orphan list_head and handle it the same way as an inode deletion during
2484  * normal operation (which journals the operations for us).
2485  *
2486  * We only do an iget() and an iput() on each inode, which is very safe if we
2487  * accidentally point at an in-use or already deleted inode.  The worst that
2488  * can happen in this case is that we get a "bit already cleared" message from
2489  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2490  * e2fsck was run on this filesystem, and it must have already done the orphan
2491  * inode cleanup for us, so we can safely abort without any further action.
2492  */
2493 static void ext4_orphan_cleanup(struct super_block *sb,
2494                                 struct ext4_super_block *es)
2495 {
2496         unsigned int s_flags = sb->s_flags;
2497         int ret, nr_orphans = 0, nr_truncates = 0;
2498 #ifdef CONFIG_QUOTA
2499         int quota_update = 0;
2500         int i;
2501 #endif
2502         if (!es->s_last_orphan) {
2503                 jbd_debug(4, "no orphan inodes to clean up\n");
2504                 return;
2505         }
2506
2507         if (bdev_read_only(sb->s_bdev)) {
2508                 ext4_msg(sb, KERN_ERR, "write access "
2509                         "unavailable, skipping orphan cleanup");
2510                 return;
2511         }
2512
2513         /* Check if feature set would not allow a r/w mount */
2514         if (!ext4_feature_set_ok(sb, 0)) {
2515                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2516                          "unknown ROCOMPAT features");
2517                 return;
2518         }
2519
2520         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2521                 /* don't clear list on RO mount w/ errors */
2522                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2523                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2524                                   "clearing orphan list.\n");
2525                         es->s_last_orphan = 0;
2526                 }
2527                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2528                 return;
2529         }
2530
2531         if (s_flags & SB_RDONLY) {
2532                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2533                 sb->s_flags &= ~SB_RDONLY;
2534         }
2535 #ifdef CONFIG_QUOTA
2536         /* Needed for iput() to work correctly and not trash data */
2537         sb->s_flags |= SB_ACTIVE;
2538
2539         /*
2540          * Turn on quotas which were not enabled for read-only mounts if
2541          * filesystem has quota feature, so that they are updated correctly.
2542          */
2543         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2544                 int ret = ext4_enable_quotas(sb);
2545
2546                 if (!ret)
2547                         quota_update = 1;
2548                 else
2549                         ext4_msg(sb, KERN_ERR,
2550                                 "Cannot turn on quotas: error %d", ret);
2551         }
2552
2553         /* Turn on journaled quotas used for old sytle */
2554         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2555                 if (EXT4_SB(sb)->s_qf_names[i]) {
2556                         int ret = ext4_quota_on_mount(sb, i);
2557
2558                         if (!ret)
2559                                 quota_update = 1;
2560                         else
2561                                 ext4_msg(sb, KERN_ERR,
2562                                         "Cannot turn on journaled "
2563                                         "quota: type %d: error %d", i, ret);
2564                 }
2565         }
2566 #endif
2567
2568         while (es->s_last_orphan) {
2569                 struct inode *inode;
2570
2571                 /*
2572                  * We may have encountered an error during cleanup; if
2573                  * so, skip the rest.
2574                  */
2575                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2576                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2577                         es->s_last_orphan = 0;
2578                         break;
2579                 }
2580
2581                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2582                 if (IS_ERR(inode)) {
2583                         es->s_last_orphan = 0;
2584                         break;
2585                 }
2586
2587                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2588                 dquot_initialize(inode);
2589                 if (inode->i_nlink) {
2590                         if (test_opt(sb, DEBUG))
2591                                 ext4_msg(sb, KERN_DEBUG,
2592                                         "%s: truncating inode %lu to %lld bytes",
2593                                         __func__, inode->i_ino, inode->i_size);
2594                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2595                                   inode->i_ino, inode->i_size);
2596                         inode_lock(inode);
2597                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2598                         ret = ext4_truncate(inode);
2599                         if (ret)
2600                                 ext4_std_error(inode->i_sb, ret);
2601                         inode_unlock(inode);
2602                         nr_truncates++;
2603                 } else {
2604                         if (test_opt(sb, DEBUG))
2605                                 ext4_msg(sb, KERN_DEBUG,
2606                                         "%s: deleting unreferenced inode %lu",
2607                                         __func__, inode->i_ino);
2608                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2609                                   inode->i_ino);
2610                         nr_orphans++;
2611                 }
2612                 iput(inode);  /* The delete magic happens here! */
2613         }
2614
2615 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2616
2617         if (nr_orphans)
2618                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2619                        PLURAL(nr_orphans));
2620         if (nr_truncates)
2621                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2622                        PLURAL(nr_truncates));
2623 #ifdef CONFIG_QUOTA
2624         /* Turn off quotas if they were enabled for orphan cleanup */
2625         if (quota_update) {
2626                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2627                         if (sb_dqopt(sb)->files[i])
2628                                 dquot_quota_off(sb, i);
2629                 }
2630         }
2631 #endif
2632         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2633 }
2634
2635 /*
2636  * Maximal extent format file size.
2637  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2638  * extent format containers, within a sector_t, and within i_blocks
2639  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2640  * so that won't be a limiting factor.
2641  *
2642  * However there is other limiting factor. We do store extents in the form
2643  * of starting block and length, hence the resulting length of the extent
2644  * covering maximum file size must fit into on-disk format containers as
2645  * well. Given that length is always by 1 unit bigger than max unit (because
2646  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2647  *
2648  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2649  */
2650 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2651 {
2652         loff_t res;
2653         loff_t upper_limit = MAX_LFS_FILESIZE;
2654
2655         /* small i_blocks in vfs inode? */
2656         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2657                 /*
2658                  * CONFIG_LBDAF is not enabled implies the inode
2659                  * i_block represent total blocks in 512 bytes
2660                  * 32 == size of vfs inode i_blocks * 8
2661                  */
2662                 upper_limit = (1LL << 32) - 1;
2663
2664                 /* total blocks in file system block size */
2665                 upper_limit >>= (blkbits - 9);
2666                 upper_limit <<= blkbits;
2667         }
2668
2669         /*
2670          * 32-bit extent-start container, ee_block. We lower the maxbytes
2671          * by one fs block, so ee_len can cover the extent of maximum file
2672          * size
2673          */
2674         res = (1LL << 32) - 1;
2675         res <<= blkbits;
2676
2677         /* Sanity check against vm- & vfs- imposed limits */
2678         if (res > upper_limit)
2679                 res = upper_limit;
2680
2681         return res;
2682 }
2683
2684 /*
2685  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2686  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2687  * We need to be 1 filesystem block less than the 2^48 sector limit.
2688  */
2689 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2690 {
2691         loff_t res = EXT4_NDIR_BLOCKS;
2692         int meta_blocks;
2693         loff_t upper_limit;
2694         /* This is calculated to be the largest file size for a dense, block
2695          * mapped file such that the file's total number of 512-byte sectors,
2696          * including data and all indirect blocks, does not exceed (2^48 - 1).
2697          *
2698          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2699          * number of 512-byte sectors of the file.
2700          */
2701
2702         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2703                 /*
2704                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2705                  * the inode i_block field represents total file blocks in
2706                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2707                  */
2708                 upper_limit = (1LL << 32) - 1;
2709
2710                 /* total blocks in file system block size */
2711                 upper_limit >>= (bits - 9);
2712
2713         } else {
2714                 /*
2715                  * We use 48 bit ext4_inode i_blocks
2716                  * With EXT4_HUGE_FILE_FL set the i_blocks
2717                  * represent total number of blocks in
2718                  * file system block size
2719                  */
2720                 upper_limit = (1LL << 48) - 1;
2721
2722         }
2723
2724         /* indirect blocks */
2725         meta_blocks = 1;
2726         /* double indirect blocks */
2727         meta_blocks += 1 + (1LL << (bits-2));
2728         /* tripple indirect blocks */
2729         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2730
2731         upper_limit -= meta_blocks;
2732         upper_limit <<= bits;
2733
2734         res += 1LL << (bits-2);
2735         res += 1LL << (2*(bits-2));
2736         res += 1LL << (3*(bits-2));
2737         res <<= bits;
2738         if (res > upper_limit)
2739                 res = upper_limit;
2740
2741         if (res > MAX_LFS_FILESIZE)
2742                 res = MAX_LFS_FILESIZE;
2743
2744         return res;
2745 }
2746
2747 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2748                                    ext4_fsblk_t logical_sb_block, int nr)
2749 {
2750         struct ext4_sb_info *sbi = EXT4_SB(sb);
2751         ext4_group_t bg, first_meta_bg;
2752         int has_super = 0;
2753
2754         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2755
2756         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2757                 return logical_sb_block + nr + 1;
2758         bg = sbi->s_desc_per_block * nr;
2759         if (ext4_bg_has_super(sb, bg))
2760                 has_super = 1;
2761
2762         /*
2763          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2764          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2765          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2766          * compensate.
2767          */
2768         if (sb->s_blocksize == 1024 && nr == 0 &&
2769             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2770                 has_super++;
2771
2772         return (has_super + ext4_group_first_block_no(sb, bg));
2773 }
2774
2775 /**
2776  * ext4_get_stripe_size: Get the stripe size.
2777  * @sbi: In memory super block info
2778  *
2779  * If we have specified it via mount option, then
2780  * use the mount option value. If the value specified at mount time is
2781  * greater than the blocks per group use the super block value.
2782  * If the super block value is greater than blocks per group return 0.
2783  * Allocator needs it be less than blocks per group.
2784  *
2785  */
2786 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2787 {
2788         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2789         unsigned long stripe_width =
2790                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2791         int ret;
2792
2793         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2794                 ret = sbi->s_stripe;
2795         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2796                 ret = stripe_width;
2797         else if (stride && stride <= sbi->s_blocks_per_group)
2798                 ret = stride;
2799         else
2800                 ret = 0;
2801
2802         /*
2803          * If the stripe width is 1, this makes no sense and
2804          * we set it to 0 to turn off stripe handling code.
2805          */
2806         if (ret <= 1)
2807                 ret = 0;
2808
2809         return ret;
2810 }
2811
2812 /*
2813  * Check whether this filesystem can be mounted based on
2814  * the features present and the RDONLY/RDWR mount requested.
2815  * Returns 1 if this filesystem can be mounted as requested,
2816  * 0 if it cannot be.
2817  */
2818 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2819 {
2820         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2821                 ext4_msg(sb, KERN_ERR,
2822                         "Couldn't mount because of "
2823                         "unsupported optional features (%x)",
2824                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2825                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2826                 return 0;
2827         }
2828
2829         if (readonly)
2830                 return 1;
2831
2832         if (ext4_has_feature_readonly(sb)) {
2833                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2834                 sb->s_flags |= SB_RDONLY;
2835                 return 1;
2836         }
2837
2838         /* Check that feature set is OK for a read-write mount */
2839         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2840                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2841                          "unsupported optional features (%x)",
2842                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2843                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2844                 return 0;
2845         }
2846         /*
2847          * Large file size enabled file system can only be mounted
2848          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2849          */
2850         if (ext4_has_feature_huge_file(sb)) {
2851                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2852                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2853                                  "cannot be mounted RDWR without "
2854                                  "CONFIG_LBDAF");
2855                         return 0;
2856                 }
2857         }
2858         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2859                 ext4_msg(sb, KERN_ERR,
2860                          "Can't support bigalloc feature without "
2861                          "extents feature\n");
2862                 return 0;
2863         }
2864
2865 #ifndef CONFIG_QUOTA
2866         if (ext4_has_feature_quota(sb) && !readonly) {
2867                 ext4_msg(sb, KERN_ERR,
2868                          "Filesystem with quota feature cannot be mounted RDWR "
2869                          "without CONFIG_QUOTA");
2870                 return 0;
2871         }
2872         if (ext4_has_feature_project(sb) && !readonly) {
2873                 ext4_msg(sb, KERN_ERR,
2874                          "Filesystem with project quota feature cannot be mounted RDWR "
2875                          "without CONFIG_QUOTA");
2876                 return 0;
2877         }
2878 #endif  /* CONFIG_QUOTA */
2879         return 1;
2880 }
2881
2882 /*
2883  * This function is called once a day if we have errors logged
2884  * on the file system
2885  */
2886 static void print_daily_error_info(struct timer_list *t)
2887 {
2888         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2889         struct super_block *sb = sbi->s_sb;
2890         struct ext4_super_block *es = sbi->s_es;
2891
2892         if (es->s_error_count)
2893                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2894                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2895                          le32_to_cpu(es->s_error_count));
2896         if (es->s_first_error_time) {
2897                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2898                        sb->s_id,
2899                        ext4_get_tstamp(es, s_first_error_time),
2900                        (int) sizeof(es->s_first_error_func),
2901                        es->s_first_error_func,
2902                        le32_to_cpu(es->s_first_error_line));
2903                 if (es->s_first_error_ino)
2904                         printk(KERN_CONT ": inode %u",
2905                                le32_to_cpu(es->s_first_error_ino));
2906                 if (es->s_first_error_block)
2907                         printk(KERN_CONT ": block %llu", (unsigned long long)
2908                                le64_to_cpu(es->s_first_error_block));
2909                 printk(KERN_CONT "\n");
2910         }
2911         if (es->s_last_error_time) {
2912                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2913                        sb->s_id,
2914                        ext4_get_tstamp(es, s_last_error_time),
2915                        (int) sizeof(es->s_last_error_func),
2916                        es->s_last_error_func,
2917                        le32_to_cpu(es->s_last_error_line));
2918                 if (es->s_last_error_ino)
2919                         printk(KERN_CONT ": inode %u",
2920                                le32_to_cpu(es->s_last_error_ino));
2921                 if (es->s_last_error_block)
2922                         printk(KERN_CONT ": block %llu", (unsigned long long)
2923                                le64_to_cpu(es->s_last_error_block));
2924                 printk(KERN_CONT "\n");
2925         }
2926         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2927 }
2928
2929 /* Find next suitable group and run ext4_init_inode_table */
2930 static int ext4_run_li_request(struct ext4_li_request *elr)
2931 {
2932         struct ext4_group_desc *gdp = NULL;
2933         ext4_group_t group, ngroups;
2934         struct super_block *sb;
2935         unsigned long timeout = 0;
2936         int ret = 0;
2937
2938         sb = elr->lr_super;
2939         ngroups = EXT4_SB(sb)->s_groups_count;
2940
2941         for (group = elr->lr_next_group; group < ngroups; group++) {
2942                 gdp = ext4_get_group_desc(sb, group, NULL);
2943                 if (!gdp) {
2944                         ret = 1;
2945                         break;
2946                 }
2947
2948                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2949                         break;
2950         }
2951
2952         if (group >= ngroups)
2953                 ret = 1;
2954
2955         if (!ret) {
2956                 timeout = jiffies;
2957                 ret = ext4_init_inode_table(sb, group,
2958                                             elr->lr_timeout ? 0 : 1);
2959                 if (elr->lr_timeout == 0) {
2960                         timeout = (jiffies - timeout) *
2961                                   elr->lr_sbi->s_li_wait_mult;
2962                         elr->lr_timeout = timeout;
2963                 }
2964                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2965                 elr->lr_next_group = group + 1;
2966         }
2967         return ret;
2968 }
2969
2970 /*
2971  * Remove lr_request from the list_request and free the
2972  * request structure. Should be called with li_list_mtx held
2973  */
2974 static void ext4_remove_li_request(struct ext4_li_request *elr)
2975 {
2976         struct ext4_sb_info *sbi;
2977
2978         if (!elr)
2979                 return;
2980
2981         sbi = elr->lr_sbi;
2982
2983         list_del(&elr->lr_request);
2984         sbi->s_li_request = NULL;
2985         kfree(elr);
2986 }
2987
2988 static void ext4_unregister_li_request(struct super_block *sb)
2989 {
2990         mutex_lock(&ext4_li_mtx);
2991         if (!ext4_li_info) {
2992                 mutex_unlock(&ext4_li_mtx);
2993                 return;
2994         }
2995
2996         mutex_lock(&ext4_li_info->li_list_mtx);
2997         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2998         mutex_unlock(&ext4_li_info->li_list_mtx);
2999         mutex_unlock(&ext4_li_mtx);
3000 }
3001
3002 static struct task_struct *ext4_lazyinit_task;
3003
3004 /*
3005  * This is the function where ext4lazyinit thread lives. It walks
3006  * through the request list searching for next scheduled filesystem.
3007  * When such a fs is found, run the lazy initialization request
3008  * (ext4_rn_li_request) and keep track of the time spend in this
3009  * function. Based on that time we compute next schedule time of
3010  * the request. When walking through the list is complete, compute
3011  * next waking time and put itself into sleep.
3012  */
3013 static int ext4_lazyinit_thread(void *arg)
3014 {
3015         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3016         struct list_head *pos, *n;
3017         struct ext4_li_request *elr;
3018         unsigned long next_wakeup, cur;
3019
3020         BUG_ON(NULL == eli);
3021
3022 cont_thread:
3023         while (true) {
3024                 next_wakeup = MAX_JIFFY_OFFSET;
3025
3026                 mutex_lock(&eli->li_list_mtx);
3027                 if (list_empty(&eli->li_request_list)) {
3028                         mutex_unlock(&eli->li_list_mtx);
3029                         goto exit_thread;
3030                 }
3031                 list_for_each_safe(pos, n, &eli->li_request_list) {
3032                         int err = 0;
3033                         int progress = 0;
3034                         elr = list_entry(pos, struct ext4_li_request,
3035                                          lr_request);
3036
3037                         if (time_before(jiffies, elr->lr_next_sched)) {
3038                                 if (time_before(elr->lr_next_sched, next_wakeup))
3039                                         next_wakeup = elr->lr_next_sched;
3040                                 continue;
3041                         }
3042                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3043                                 if (sb_start_write_trylock(elr->lr_super)) {
3044                                         progress = 1;
3045                                         /*
3046                                          * We hold sb->s_umount, sb can not
3047                                          * be removed from the list, it is
3048                                          * now safe to drop li_list_mtx
3049                                          */
3050                                         mutex_unlock(&eli->li_list_mtx);
3051                                         err = ext4_run_li_request(elr);
3052                                         sb_end_write(elr->lr_super);
3053                                         mutex_lock(&eli->li_list_mtx);
3054                                         n = pos->next;
3055                                 }
3056                                 up_read((&elr->lr_super->s_umount));
3057                         }
3058                         /* error, remove the lazy_init job */
3059                         if (err) {
3060                                 ext4_remove_li_request(elr);
3061                                 continue;
3062                         }
3063                         if (!progress) {
3064                                 elr->lr_next_sched = jiffies +
3065                                         (prandom_u32()
3066                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3067                         }
3068                         if (time_before(elr->lr_next_sched, next_wakeup))
3069                                 next_wakeup = elr->lr_next_sched;
3070                 }
3071                 mutex_unlock(&eli->li_list_mtx);
3072
3073                 try_to_freeze();
3074
3075                 cur = jiffies;
3076                 if ((time_after_eq(cur, next_wakeup)) ||
3077                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3078                         cond_resched();
3079                         continue;
3080                 }
3081
3082                 schedule_timeout_interruptible(next_wakeup - cur);
3083
3084                 if (kthread_should_stop()) {
3085                         ext4_clear_request_list();
3086                         goto exit_thread;
3087                 }
3088         }
3089
3090 exit_thread:
3091         /*
3092          * It looks like the request list is empty, but we need
3093          * to check it under the li_list_mtx lock, to prevent any
3094          * additions into it, and of course we should lock ext4_li_mtx
3095          * to atomically free the list and ext4_li_info, because at
3096          * this point another ext4 filesystem could be registering
3097          * new one.
3098          */
3099         mutex_lock(&ext4_li_mtx);
3100         mutex_lock(&eli->li_list_mtx);
3101         if (!list_empty(&eli->li_request_list)) {
3102                 mutex_unlock(&eli->li_list_mtx);
3103                 mutex_unlock(&ext4_li_mtx);
3104                 goto cont_thread;
3105         }
3106         mutex_unlock(&eli->li_list_mtx);
3107         kfree(ext4_li_info);
3108         ext4_li_info = NULL;
3109         mutex_unlock(&ext4_li_mtx);
3110
3111         return 0;
3112 }
3113
3114 static void ext4_clear_request_list(void)
3115 {
3116         struct list_head *pos, *n;
3117         struct ext4_li_request *elr;
3118
3119         mutex_lock(&ext4_li_info->li_list_mtx);
3120         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3121                 elr = list_entry(pos, struct ext4_li_request,
3122                                  lr_request);
3123                 ext4_remove_li_request(elr);
3124         }
3125         mutex_unlock(&ext4_li_info->li_list_mtx);
3126 }
3127
3128 static int ext4_run_lazyinit_thread(void)
3129 {
3130         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3131                                          ext4_li_info, "ext4lazyinit");
3132         if (IS_ERR(ext4_lazyinit_task)) {
3133                 int err = PTR_ERR(ext4_lazyinit_task);
3134                 ext4_clear_request_list();
3135                 kfree(ext4_li_info);
3136                 ext4_li_info = NULL;
3137                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3138                                  "initialization thread\n",
3139                                  err);
3140                 return err;
3141         }
3142         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3143         return 0;
3144 }
3145
3146 /*
3147  * Check whether it make sense to run itable init. thread or not.
3148  * If there is at least one uninitialized inode table, return
3149  * corresponding group number, else the loop goes through all
3150  * groups and return total number of groups.
3151  */
3152 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3153 {
3154         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3155         struct ext4_group_desc *gdp = NULL;
3156
3157         if (!ext4_has_group_desc_csum(sb))
3158                 return ngroups;
3159
3160         for (group = 0; group < ngroups; group++) {
3161                 gdp = ext4_get_group_desc(sb, group, NULL);
3162                 if (!gdp)
3163                         continue;
3164
3165                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3166                         break;
3167         }
3168
3169         return group;
3170 }
3171
3172 static int ext4_li_info_new(void)
3173 {
3174         struct ext4_lazy_init *eli = NULL;
3175
3176         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3177         if (!eli)
3178                 return -ENOMEM;
3179
3180         INIT_LIST_HEAD(&eli->li_request_list);
3181         mutex_init(&eli->li_list_mtx);
3182
3183         eli->li_state |= EXT4_LAZYINIT_QUIT;
3184
3185         ext4_li_info = eli;
3186
3187         return 0;
3188 }
3189
3190 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3191                                             ext4_group_t start)
3192 {
3193         struct ext4_sb_info *sbi = EXT4_SB(sb);
3194         struct ext4_li_request *elr;
3195
3196         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3197         if (!elr)
3198                 return NULL;
3199
3200         elr->lr_super = sb;
3201         elr->lr_sbi = sbi;
3202         elr->lr_next_group = start;
3203
3204         /*
3205          * Randomize first schedule time of the request to
3206          * spread the inode table initialization requests
3207          * better.
3208          */
3209         elr->lr_next_sched = jiffies + (prandom_u32() %
3210                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3211         return elr;
3212 }
3213
3214 int ext4_register_li_request(struct super_block *sb,
3215                              ext4_group_t first_not_zeroed)
3216 {
3217         struct ext4_sb_info *sbi = EXT4_SB(sb);
3218         struct ext4_li_request *elr = NULL;
3219         ext4_group_t ngroups = sbi->s_groups_count;
3220         int ret = 0;
3221
3222         mutex_lock(&ext4_li_mtx);
3223         if (sbi->s_li_request != NULL) {
3224                 /*
3225                  * Reset timeout so it can be computed again, because
3226                  * s_li_wait_mult might have changed.
3227                  */
3228                 sbi->s_li_request->lr_timeout = 0;
3229                 goto out;
3230         }
3231
3232         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3233             !test_opt(sb, INIT_INODE_TABLE))
3234                 goto out;
3235
3236         elr = ext4_li_request_new(sb, first_not_zeroed);
3237         if (!elr) {
3238                 ret = -ENOMEM;
3239                 goto out;
3240         }
3241
3242         if (NULL == ext4_li_info) {
3243                 ret = ext4_li_info_new();
3244                 if (ret)
3245                         goto out;
3246         }
3247
3248         mutex_lock(&ext4_li_info->li_list_mtx);
3249         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3250         mutex_unlock(&ext4_li_info->li_list_mtx);
3251
3252         sbi->s_li_request = elr;
3253         /*
3254          * set elr to NULL here since it has been inserted to
3255          * the request_list and the removal and free of it is
3256          * handled by ext4_clear_request_list from now on.
3257          */
3258         elr = NULL;
3259
3260         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3261                 ret = ext4_run_lazyinit_thread();
3262                 if (ret)
3263                         goto out;
3264         }
3265 out:
3266         mutex_unlock(&ext4_li_mtx);
3267         if (ret)
3268                 kfree(elr);
3269         return ret;
3270 }
3271
3272 /*
3273  * We do not need to lock anything since this is called on
3274  * module unload.
3275  */
3276 static void ext4_destroy_lazyinit_thread(void)
3277 {
3278         /*
3279          * If thread exited earlier
3280          * there's nothing to be done.
3281          */
3282         if (!ext4_li_info || !ext4_lazyinit_task)
3283                 return;
3284
3285         kthread_stop(ext4_lazyinit_task);
3286 }
3287
3288 static int set_journal_csum_feature_set(struct super_block *sb)
3289 {
3290         int ret = 1;
3291         int compat, incompat;
3292         struct ext4_sb_info *sbi = EXT4_SB(sb);
3293
3294         if (ext4_has_metadata_csum(sb)) {
3295                 /* journal checksum v3 */
3296                 compat = 0;
3297                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3298         } else {
3299                 /* journal checksum v1 */
3300                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3301                 incompat = 0;
3302         }
3303
3304         jbd2_journal_clear_features(sbi->s_journal,
3305                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3306                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3307                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3308         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3309                 ret = jbd2_journal_set_features(sbi->s_journal,
3310                                 compat, 0,
3311                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3312                                 incompat);
3313         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3314                 ret = jbd2_journal_set_features(sbi->s_journal,
3315                                 compat, 0,
3316                                 incompat);
3317                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3318                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3319         } else {
3320                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3321                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3322         }
3323
3324         return ret;
3325 }
3326
3327 /*
3328  * Note: calculating the overhead so we can be compatible with
3329  * historical BSD practice is quite difficult in the face of
3330  * clusters/bigalloc.  This is because multiple metadata blocks from
3331  * different block group can end up in the same allocation cluster.
3332  * Calculating the exact overhead in the face of clustered allocation
3333  * requires either O(all block bitmaps) in memory or O(number of block
3334  * groups**2) in time.  We will still calculate the superblock for
3335  * older file systems --- and if we come across with a bigalloc file
3336  * system with zero in s_overhead_clusters the estimate will be close to
3337  * correct especially for very large cluster sizes --- but for newer
3338  * file systems, it's better to calculate this figure once at mkfs
3339  * time, and store it in the superblock.  If the superblock value is
3340  * present (even for non-bigalloc file systems), we will use it.
3341  */
3342 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3343                           char *buf)
3344 {
3345         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3346         struct ext4_group_desc  *gdp;
3347         ext4_fsblk_t            first_block, last_block, b;
3348         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3349         int                     s, j, count = 0;
3350
3351         if (!ext4_has_feature_bigalloc(sb))
3352                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3353                         sbi->s_itb_per_group + 2);
3354
3355         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3356                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3357         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3358         for (i = 0; i < ngroups; i++) {
3359                 gdp = ext4_get_group_desc(sb, i, NULL);
3360                 b = ext4_block_bitmap(sb, gdp);
3361                 if (b >= first_block && b <= last_block) {
3362                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3363                         count++;
3364                 }
3365                 b = ext4_inode_bitmap(sb, gdp);
3366                 if (b >= first_block && b <= last_block) {
3367                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3368                         count++;
3369                 }
3370                 b = ext4_inode_table(sb, gdp);
3371                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3372                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3373                                 int c = EXT4_B2C(sbi, b - first_block);
3374                                 ext4_set_bit(c, buf);
3375                                 count++;
3376                         }
3377                 if (i != grp)
3378                         continue;
3379                 s = 0;
3380                 if (ext4_bg_has_super(sb, grp)) {
3381                         ext4_set_bit(s++, buf);
3382                         count++;
3383                 }
3384                 j = ext4_bg_num_gdb(sb, grp);
3385                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3386                         ext4_error(sb, "Invalid number of block group "
3387                                    "descriptor blocks: %d", j);
3388                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3389                 }
3390                 count += j;
3391                 for (; j > 0; j--)
3392                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3393         }
3394         if (!count)
3395                 return 0;
3396         return EXT4_CLUSTERS_PER_GROUP(sb) -
3397                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3398 }
3399
3400 /*
3401  * Compute the overhead and stash it in sbi->s_overhead
3402  */
3403 int ext4_calculate_overhead(struct super_block *sb)
3404 {
3405         struct ext4_sb_info *sbi = EXT4_SB(sb);
3406         struct ext4_super_block *es = sbi->s_es;
3407         struct inode *j_inode;
3408         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3409         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3410         ext4_fsblk_t overhead = 0;
3411         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3412
3413         if (!buf)
3414                 return -ENOMEM;
3415
3416         /*
3417          * Compute the overhead (FS structures).  This is constant
3418          * for a given filesystem unless the number of block groups
3419          * changes so we cache the previous value until it does.
3420          */
3421
3422         /*
3423          * All of the blocks before first_data_block are overhead
3424          */
3425         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3426
3427         /*
3428          * Add the overhead found in each block group
3429          */
3430         for (i = 0; i < ngroups; i++) {
3431                 int blks;
3432
3433                 blks = count_overhead(sb, i, buf);
3434                 overhead += blks;
3435                 if (blks)
3436                         memset(buf, 0, PAGE_SIZE);
3437                 cond_resched();
3438         }
3439
3440         /*
3441          * Add the internal journal blocks whether the journal has been
3442          * loaded or not
3443          */
3444         if (sbi->s_journal && !sbi->journal_bdev)
3445                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3446         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3447                 j_inode = ext4_get_journal_inode(sb, j_inum);
3448                 if (j_inode) {
3449                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3450                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3451                         iput(j_inode);
3452                 } else {
3453                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3454                 }
3455         }
3456         sbi->s_overhead = overhead;
3457         smp_wmb();
3458         free_page((unsigned long) buf);
3459         return 0;
3460 }
3461
3462 static void ext4_set_resv_clusters(struct super_block *sb)
3463 {
3464         ext4_fsblk_t resv_clusters;
3465         struct ext4_sb_info *sbi = EXT4_SB(sb);
3466
3467         /*
3468          * There's no need to reserve anything when we aren't using extents.
3469          * The space estimates are exact, there are no unwritten extents,
3470          * hole punching doesn't need new metadata... This is needed especially
3471          * to keep ext2/3 backward compatibility.
3472          */
3473         if (!ext4_has_feature_extents(sb))
3474                 return;
3475         /*
3476          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3477          * This should cover the situations where we can not afford to run
3478          * out of space like for example punch hole, or converting
3479          * unwritten extents in delalloc path. In most cases such
3480          * allocation would require 1, or 2 blocks, higher numbers are
3481          * very rare.
3482          */
3483         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3484                          sbi->s_cluster_bits);
3485
3486         do_div(resv_clusters, 50);
3487         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3488
3489         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3490 }
3491
3492 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3493 {
3494         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3495         char *orig_data = kstrdup(data, GFP_KERNEL);
3496         struct buffer_head *bh;
3497         struct ext4_super_block *es = NULL;
3498         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3499         ext4_fsblk_t block;
3500         ext4_fsblk_t sb_block = get_sb_block(&data);
3501         ext4_fsblk_t logical_sb_block;
3502         unsigned long offset = 0;
3503         unsigned long journal_devnum = 0;
3504         unsigned long def_mount_opts;
3505         struct inode *root;
3506         const char *descr;
3507         int ret = -ENOMEM;
3508         int blocksize, clustersize;
3509         unsigned int db_count;
3510         unsigned int i;
3511         int needs_recovery, has_huge_files, has_bigalloc;
3512         __u64 blocks_count;
3513         int err = 0;
3514         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3515         ext4_group_t first_not_zeroed;
3516
3517         if ((data && !orig_data) || !sbi)
3518                 goto out_free_base;
3519
3520         sbi->s_daxdev = dax_dev;
3521         sbi->s_blockgroup_lock =
3522                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3523         if (!sbi->s_blockgroup_lock)
3524                 goto out_free_base;
3525
3526         sb->s_fs_info = sbi;
3527         sbi->s_sb = sb;
3528         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3529         sbi->s_sb_block = sb_block;
3530         if (sb->s_bdev->bd_part)
3531                 sbi->s_sectors_written_start =
3532                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3533
3534         /* Cleanup superblock name */
3535         strreplace(sb->s_id, '/', '!');
3536
3537         /* -EINVAL is default */
3538         ret = -EINVAL;
3539         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3540         if (!blocksize) {
3541                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3542                 goto out_fail;
3543         }
3544
3545         /*
3546          * The ext4 superblock will not be buffer aligned for other than 1kB
3547          * block sizes.  We need to calculate the offset from buffer start.
3548          */
3549         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3550                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3551                 offset = do_div(logical_sb_block, blocksize);
3552         } else {
3553                 logical_sb_block = sb_block;
3554         }
3555
3556         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3557                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3558                 goto out_fail;
3559         }
3560         /*
3561          * Note: s_es must be initialized as soon as possible because
3562          *       some ext4 macro-instructions depend on its value
3563          */
3564         es = (struct ext4_super_block *) (bh->b_data + offset);
3565         sbi->s_es = es;
3566         sb->s_magic = le16_to_cpu(es->s_magic);
3567         if (sb->s_magic != EXT4_SUPER_MAGIC)
3568                 goto cantfind_ext4;
3569         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3570
3571         /* Warn if metadata_csum and gdt_csum are both set. */
3572         if (ext4_has_feature_metadata_csum(sb) &&
3573             ext4_has_feature_gdt_csum(sb))
3574                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3575                              "redundant flags; please run fsck.");
3576
3577         /* Check for a known checksum algorithm */
3578         if (!ext4_verify_csum_type(sb, es)) {
3579                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3580                          "unknown checksum algorithm.");
3581                 silent = 1;
3582                 goto cantfind_ext4;
3583         }
3584
3585         /* Load the checksum driver */
3586         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3587         if (IS_ERR(sbi->s_chksum_driver)) {
3588                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3589                 ret = PTR_ERR(sbi->s_chksum_driver);
3590                 sbi->s_chksum_driver = NULL;
3591                 goto failed_mount;
3592         }
3593
3594         /* Check superblock checksum */
3595         if (!ext4_superblock_csum_verify(sb, es)) {
3596                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3597                          "invalid superblock checksum.  Run e2fsck?");
3598                 silent = 1;
3599                 ret = -EFSBADCRC;
3600                 goto cantfind_ext4;
3601         }
3602
3603         /* Precompute checksum seed for all metadata */
3604         if (ext4_has_feature_csum_seed(sb))
3605                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3606         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3607                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3608                                                sizeof(es->s_uuid));
3609
3610         /* Set defaults before we parse the mount options */
3611         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3612         set_opt(sb, INIT_INODE_TABLE);
3613         if (def_mount_opts & EXT4_DEFM_DEBUG)
3614                 set_opt(sb, DEBUG);
3615         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3616                 set_opt(sb, GRPID);
3617         if (def_mount_opts & EXT4_DEFM_UID16)
3618                 set_opt(sb, NO_UID32);
3619         /* xattr user namespace & acls are now defaulted on */
3620         set_opt(sb, XATTR_USER);
3621 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3622         set_opt(sb, POSIX_ACL);
3623 #endif
3624         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3625         if (ext4_has_metadata_csum(sb))
3626                 set_opt(sb, JOURNAL_CHECKSUM);
3627
3628         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3629                 set_opt(sb, JOURNAL_DATA);
3630         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3631                 set_opt(sb, ORDERED_DATA);
3632         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3633                 set_opt(sb, WRITEBACK_DATA);
3634
3635         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3636                 set_opt(sb, ERRORS_PANIC);
3637         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3638                 set_opt(sb, ERRORS_CONT);
3639         else
3640                 set_opt(sb, ERRORS_RO);
3641         /* block_validity enabled by default; disable with noblock_validity */
3642         set_opt(sb, BLOCK_VALIDITY);
3643         if (def_mount_opts & EXT4_DEFM_DISCARD)
3644                 set_opt(sb, DISCARD);
3645
3646         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3647         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3648         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3649         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3650         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3651
3652         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3653                 set_opt(sb, BARRIER);
3654
3655         /*
3656          * enable delayed allocation by default
3657          * Use -o nodelalloc to turn it off
3658          */
3659         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3660             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3661                 set_opt(sb, DELALLOC);
3662
3663         /*
3664          * set default s_li_wait_mult for lazyinit, for the case there is
3665          * no mount option specified.
3666          */
3667         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3668
3669         if (sbi->s_es->s_mount_opts[0]) {
3670                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3671                                               sizeof(sbi->s_es->s_mount_opts),
3672                                               GFP_KERNEL);
3673                 if (!s_mount_opts)
3674                         goto failed_mount;
3675                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3676                                    &journal_ioprio, 0)) {
3677                         ext4_msg(sb, KERN_WARNING,
3678                                  "failed to parse options in superblock: %s",
3679                                  s_mount_opts);
3680                 }
3681                 kfree(s_mount_opts);
3682         }
3683         sbi->s_def_mount_opt = sbi->s_mount_opt;
3684         if (!parse_options((char *) data, sb, &journal_devnum,
3685                            &journal_ioprio, 0))
3686                 goto failed_mount;
3687
3688         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3689                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3690                             "with data=journal disables delayed "
3691                             "allocation and O_DIRECT support!\n");
3692                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3693                         ext4_msg(sb, KERN_ERR, "can't mount with "
3694                                  "both data=journal and delalloc");
3695                         goto failed_mount;
3696                 }
3697                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3698                         ext4_msg(sb, KERN_ERR, "can't mount with "
3699                                  "both data=journal and dioread_nolock");
3700                         goto failed_mount;
3701                 }
3702                 if (test_opt(sb, DAX)) {
3703                         ext4_msg(sb, KERN_ERR, "can't mount with "
3704                                  "both data=journal and dax");
3705                         goto failed_mount;
3706                 }
3707                 if (ext4_has_feature_encrypt(sb)) {
3708                         ext4_msg(sb, KERN_WARNING,
3709                                  "encrypted files will use data=ordered "
3710                                  "instead of data journaling mode");
3711                 }
3712                 if (test_opt(sb, DELALLOC))
3713                         clear_opt(sb, DELALLOC);
3714         } else {
3715                 sb->s_iflags |= SB_I_CGROUPWB;
3716         }
3717
3718         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3719                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3720
3721         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3722             (ext4_has_compat_features(sb) ||
3723              ext4_has_ro_compat_features(sb) ||
3724              ext4_has_incompat_features(sb)))
3725                 ext4_msg(sb, KERN_WARNING,
3726                        "feature flags set on rev 0 fs, "
3727                        "running e2fsck is recommended");
3728
3729         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3730                 set_opt2(sb, HURD_COMPAT);
3731                 if (ext4_has_feature_64bit(sb)) {
3732                         ext4_msg(sb, KERN_ERR,
3733                                  "The Hurd can't support 64-bit file systems");
3734                         goto failed_mount;
3735                 }
3736
3737                 /*
3738                  * ea_inode feature uses l_i_version field which is not
3739                  * available in HURD_COMPAT mode.
3740                  */
3741                 if (ext4_has_feature_ea_inode(sb)) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                  "ea_inode feature is not supported for Hurd");
3744                         goto failed_mount;
3745                 }
3746         }
3747
3748         if (IS_EXT2_SB(sb)) {
3749                 if (ext2_feature_set_ok(sb))
3750                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3751                                  "using the ext4 subsystem");
3752                 else {
3753                         /*
3754                          * If we're probing be silent, if this looks like
3755                          * it's actually an ext[34] filesystem.
3756                          */
3757                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3758                                 goto failed_mount;
3759                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3760                                  "to feature incompatibilities");
3761                         goto failed_mount;
3762                 }
3763         }
3764
3765         if (IS_EXT3_SB(sb)) {
3766                 if (ext3_feature_set_ok(sb))
3767                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3768                                  "using the ext4 subsystem");
3769                 else {
3770                         /*
3771                          * If we're probing be silent, if this looks like
3772                          * it's actually an ext4 filesystem.
3773                          */
3774                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3775                                 goto failed_mount;
3776                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3777                                  "to feature incompatibilities");
3778                         goto failed_mount;
3779                 }
3780         }
3781
3782         /*
3783          * Check feature flags regardless of the revision level, since we
3784          * previously didn't change the revision level when setting the flags,
3785          * so there is a chance incompat flags are set on a rev 0 filesystem.
3786          */
3787         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3788                 goto failed_mount;
3789
3790         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3791         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3792             blocksize > EXT4_MAX_BLOCK_SIZE) {
3793                 ext4_msg(sb, KERN_ERR,
3794                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3795                          blocksize, le32_to_cpu(es->s_log_block_size));
3796                 goto failed_mount;
3797         }
3798         if (le32_to_cpu(es->s_log_block_size) >
3799             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3800                 ext4_msg(sb, KERN_ERR,
3801                          "Invalid log block size: %u",
3802                          le32_to_cpu(es->s_log_block_size));
3803                 goto failed_mount;
3804         }
3805         if (le32_to_cpu(es->s_log_cluster_size) >
3806             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3807                 ext4_msg(sb, KERN_ERR,
3808                          "Invalid log cluster size: %u",
3809                          le32_to_cpu(es->s_log_cluster_size));
3810                 goto failed_mount;
3811         }
3812
3813         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3814                 ext4_msg(sb, KERN_ERR,
3815                          "Number of reserved GDT blocks insanely large: %d",
3816                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3817                 goto failed_mount;
3818         }
3819
3820         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3821                 if (ext4_has_feature_inline_data(sb)) {
3822                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3823                                         " that may contain inline data");
3824                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3825                 }
3826                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3827                         ext4_msg(sb, KERN_ERR,
3828                                 "DAX unsupported by block device. Turning off DAX.");
3829                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3830                 }
3831         }
3832
3833         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3834                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3835                          es->s_encryption_level);
3836                 goto failed_mount;
3837         }
3838
3839         if (sb->s_blocksize != blocksize) {
3840                 /* Validate the filesystem blocksize */
3841                 if (!sb_set_blocksize(sb, blocksize)) {
3842                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3843                                         blocksize);
3844                         goto failed_mount;
3845                 }
3846
3847                 brelse(bh);
3848                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3849                 offset = do_div(logical_sb_block, blocksize);
3850                 bh = sb_bread_unmovable(sb, logical_sb_block);
3851                 if (!bh) {
3852                         ext4_msg(sb, KERN_ERR,
3853                                "Can't read superblock on 2nd try");
3854                         goto failed_mount;
3855                 }
3856                 es = (struct ext4_super_block *)(bh->b_data + offset);
3857                 sbi->s_es = es;
3858                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3859                         ext4_msg(sb, KERN_ERR,
3860                                "Magic mismatch, very weird!");
3861                         goto failed_mount;
3862                 }
3863         }
3864
3865         has_huge_files = ext4_has_feature_huge_file(sb);
3866         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3867                                                       has_huge_files);
3868         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3869
3870         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3871                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3872                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3873         } else {
3874                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3875                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3876                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3877                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3878                                  sbi->s_first_ino);
3879                         goto failed_mount;
3880                 }
3881                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3882                     (!is_power_of_2(sbi->s_inode_size)) ||
3883                     (sbi->s_inode_size > blocksize)) {
3884                         ext4_msg(sb, KERN_ERR,
3885                                "unsupported inode size: %d",
3886                                sbi->s_inode_size);
3887                         goto failed_mount;
3888                 }
3889                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3890                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3891         }
3892
3893         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3894         if (ext4_has_feature_64bit(sb)) {
3895                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3896                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3897                     !is_power_of_2(sbi->s_desc_size)) {
3898                         ext4_msg(sb, KERN_ERR,
3899                                "unsupported descriptor size %lu",
3900                                sbi->s_desc_size);
3901                         goto failed_mount;
3902                 }
3903         } else
3904                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3905
3906         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3907         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3908
3909         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3910         if (sbi->s_inodes_per_block == 0)
3911                 goto cantfind_ext4;
3912         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3913             sbi->s_inodes_per_group > blocksize * 8) {
3914                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3915                          sbi->s_blocks_per_group);
3916                 goto failed_mount;
3917         }
3918         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3919                                         sbi->s_inodes_per_block;
3920         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3921         sbi->s_sbh = bh;
3922         sbi->s_mount_state = le16_to_cpu(es->s_state);
3923         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3924         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3925
3926         for (i = 0; i < 4; i++)
3927                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3928         sbi->s_def_hash_version = es->s_def_hash_version;
3929         if (ext4_has_feature_dir_index(sb)) {
3930                 i = le32_to_cpu(es->s_flags);
3931                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3932                         sbi->s_hash_unsigned = 3;
3933                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3934 #ifdef __CHAR_UNSIGNED__
3935                         if (!sb_rdonly(sb))
3936                                 es->s_flags |=
3937                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3938                         sbi->s_hash_unsigned = 3;
3939 #else
3940                         if (!sb_rdonly(sb))
3941                                 es->s_flags |=
3942                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3943 #endif
3944                 }
3945         }
3946
3947         /* Handle clustersize */
3948         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3949         has_bigalloc = ext4_has_feature_bigalloc(sb);
3950         if (has_bigalloc) {
3951                 if (clustersize < blocksize) {
3952                         ext4_msg(sb, KERN_ERR,
3953                                  "cluster size (%d) smaller than "
3954                                  "block size (%d)", clustersize, blocksize);
3955                         goto failed_mount;
3956                 }
3957                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3958                         le32_to_cpu(es->s_log_block_size);
3959                 sbi->s_clusters_per_group =
3960                         le32_to_cpu(es->s_clusters_per_group);
3961                 if (sbi->s_clusters_per_group > blocksize * 8) {
3962                         ext4_msg(sb, KERN_ERR,
3963                                  "#clusters per group too big: %lu",
3964                                  sbi->s_clusters_per_group);
3965                         goto failed_mount;
3966                 }
3967                 if (sbi->s_blocks_per_group !=
3968                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3969                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3970                                  "clusters per group (%lu) inconsistent",
3971                                  sbi->s_blocks_per_group,
3972                                  sbi->s_clusters_per_group);
3973                         goto failed_mount;
3974                 }
3975         } else {
3976                 if (clustersize != blocksize) {
3977                         ext4_msg(sb, KERN_ERR,
3978                                  "fragment/cluster size (%d) != "
3979                                  "block size (%d)", clustersize, blocksize);
3980                         goto failed_mount;
3981                 }
3982                 if (sbi->s_blocks_per_group > blocksize * 8) {
3983                         ext4_msg(sb, KERN_ERR,
3984                                  "#blocks per group too big: %lu",
3985                                  sbi->s_blocks_per_group);
3986                         goto failed_mount;
3987                 }
3988                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3989                 sbi->s_cluster_bits = 0;
3990         }
3991         sbi->s_cluster_ratio = clustersize / blocksize;
3992
3993         /* Do we have standard group size of clustersize * 8 blocks ? */
3994         if (sbi->s_blocks_per_group == clustersize << 3)
3995                 set_opt2(sb, STD_GROUP_SIZE);
3996
3997         /*
3998          * Test whether we have more sectors than will fit in sector_t,
3999          * and whether the max offset is addressable by the page cache.
4000          */
4001         err = generic_check_addressable(sb->s_blocksize_bits,
4002                                         ext4_blocks_count(es));
4003         if (err) {
4004                 ext4_msg(sb, KERN_ERR, "filesystem"
4005                          " too large to mount safely on this system");
4006                 if (sizeof(sector_t) < 8)
4007                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4008                 goto failed_mount;
4009         }
4010
4011         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4012                 goto cantfind_ext4;
4013
4014         /* check blocks count against device size */
4015         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4016         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4017                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4018                        "exceeds size of device (%llu blocks)",
4019                        ext4_blocks_count(es), blocks_count);
4020                 goto failed_mount;
4021         }
4022
4023         /*
4024          * It makes no sense for the first data block to be beyond the end
4025          * of the filesystem.
4026          */
4027         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4028                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4029                          "block %u is beyond end of filesystem (%llu)",
4030                          le32_to_cpu(es->s_first_data_block),
4031                          ext4_blocks_count(es));
4032                 goto failed_mount;
4033         }
4034         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4035             (sbi->s_cluster_ratio == 1)) {
4036                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4037                          "block is 0 with a 1k block and cluster size");
4038                 goto failed_mount;
4039         }
4040
4041         blocks_count = (ext4_blocks_count(es) -
4042                         le32_to_cpu(es->s_first_data_block) +
4043                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4044         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4045         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4046                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4047                        "(block count %llu, first data block %u, "
4048                        "blocks per group %lu)", sbi->s_groups_count,
4049                        ext4_blocks_count(es),
4050                        le32_to_cpu(es->s_first_data_block),
4051                        EXT4_BLOCKS_PER_GROUP(sb));
4052                 goto failed_mount;
4053         }
4054         sbi->s_groups_count = blocks_count;
4055         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4056                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4057         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4058                    EXT4_DESC_PER_BLOCK(sb);
4059         if (ext4_has_feature_meta_bg(sb)) {
4060                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4061                         ext4_msg(sb, KERN_WARNING,
4062                                  "first meta block group too large: %u "
4063                                  "(group descriptor block count %u)",
4064                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4065                         goto failed_mount;
4066                 }
4067         }
4068         sbi->s_group_desc = kvmalloc_array(db_count,
4069                                            sizeof(struct buffer_head *),
4070                                            GFP_KERNEL);
4071         if (sbi->s_group_desc == NULL) {
4072                 ext4_msg(sb, KERN_ERR, "not enough memory");
4073                 ret = -ENOMEM;
4074                 goto failed_mount;
4075         }
4076         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4077             le32_to_cpu(es->s_inodes_count)) {
4078                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4079                          le32_to_cpu(es->s_inodes_count),
4080                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4081                 ret = -EINVAL;
4082                 goto failed_mount;
4083         }
4084
4085         bgl_lock_init(sbi->s_blockgroup_lock);
4086
4087         /* Pre-read the descriptors into the buffer cache */
4088         for (i = 0; i < db_count; i++) {
4089                 block = descriptor_loc(sb, logical_sb_block, i);
4090                 sb_breadahead(sb, block);
4091         }
4092
4093         for (i = 0; i < db_count; i++) {
4094                 block = descriptor_loc(sb, logical_sb_block, i);
4095                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4096                 if (!sbi->s_group_desc[i]) {
4097                         ext4_msg(sb, KERN_ERR,
4098                                "can't read group descriptor %d", i);
4099                         db_count = i;
4100                         goto failed_mount2;
4101                 }
4102         }
4103         sbi->s_gdb_count = db_count;
4104         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4105                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4106                 ret = -EFSCORRUPTED;
4107                 goto failed_mount2;
4108         }
4109
4110         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4111
4112         /* Register extent status tree shrinker */
4113         if (ext4_es_register_shrinker(sbi))
4114                 goto failed_mount3;
4115
4116         sbi->s_stripe = ext4_get_stripe_size(sbi);
4117         sbi->s_extent_max_zeroout_kb = 32;
4118
4119         /*
4120          * set up enough so that it can read an inode
4121          */
4122         sb->s_op = &ext4_sops;
4123         sb->s_export_op = &ext4_export_ops;
4124         sb->s_xattr = ext4_xattr_handlers;
4125 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4126         sb->s_cop = &ext4_cryptops;
4127 #endif
4128 #ifdef CONFIG_QUOTA
4129         sb->dq_op = &ext4_quota_operations;
4130         if (ext4_has_feature_quota(sb))
4131                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4132         else
4133                 sb->s_qcop = &ext4_qctl_operations;
4134         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4135 #endif
4136         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4137
4138         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4139         mutex_init(&sbi->s_orphan_lock);
4140
4141         sb->s_root = NULL;
4142
4143         needs_recovery = (es->s_last_orphan != 0 ||
4144                           ext4_has_feature_journal_needs_recovery(sb));
4145
4146         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4147                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4148                         goto failed_mount3a;
4149
4150         /*
4151          * The first inode we look at is the journal inode.  Don't try
4152          * root first: it may be modified in the journal!
4153          */
4154         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4155                 err = ext4_load_journal(sb, es, journal_devnum);
4156                 if (err)
4157                         goto failed_mount3a;
4158         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4159                    ext4_has_feature_journal_needs_recovery(sb)) {
4160                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4161                        "suppressed and not mounted read-only");
4162                 goto failed_mount_wq;
4163         } else {
4164                 /* Nojournal mode, all journal mount options are illegal */
4165                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4166                         ext4_msg(sb, KERN_ERR, "can't mount with "
4167                                  "journal_checksum, fs mounted w/o journal");
4168                         goto failed_mount_wq;
4169                 }
4170                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4171                         ext4_msg(sb, KERN_ERR, "can't mount with "
4172                                  "journal_async_commit, fs mounted w/o journal");
4173                         goto failed_mount_wq;
4174                 }
4175                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4176                         ext4_msg(sb, KERN_ERR, "can't mount with "
4177                                  "commit=%lu, fs mounted w/o journal",
4178                                  sbi->s_commit_interval / HZ);
4179                         goto failed_mount_wq;
4180                 }
4181                 if (EXT4_MOUNT_DATA_FLAGS &
4182                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4183                         ext4_msg(sb, KERN_ERR, "can't mount with "
4184                                  "data=, fs mounted w/o journal");
4185                         goto failed_mount_wq;
4186                 }
4187                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4188                 clear_opt(sb, JOURNAL_CHECKSUM);
4189                 clear_opt(sb, DATA_FLAGS);
4190                 sbi->s_journal = NULL;
4191                 needs_recovery = 0;
4192                 goto no_journal;
4193         }
4194
4195         if (ext4_has_feature_64bit(sb) &&
4196             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4197                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4198                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4199                 goto failed_mount_wq;
4200         }
4201
4202         if (!set_journal_csum_feature_set(sb)) {
4203                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4204                          "feature set");
4205                 goto failed_mount_wq;
4206         }
4207
4208         /* We have now updated the journal if required, so we can
4209          * validate the data journaling mode. */
4210         switch (test_opt(sb, DATA_FLAGS)) {
4211         case 0:
4212                 /* No mode set, assume a default based on the journal
4213                  * capabilities: ORDERED_DATA if the journal can
4214                  * cope, else JOURNAL_DATA
4215                  */
4216                 if (jbd2_journal_check_available_features
4217                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4218                         set_opt(sb, ORDERED_DATA);
4219                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4220                 } else {
4221                         set_opt(sb, JOURNAL_DATA);
4222                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4223                 }
4224                 break;
4225
4226         case EXT4_MOUNT_ORDERED_DATA:
4227         case EXT4_MOUNT_WRITEBACK_DATA:
4228                 if (!jbd2_journal_check_available_features
4229                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4230                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4231                                "requested data journaling mode");
4232                         goto failed_mount_wq;
4233                 }
4234         default:
4235                 break;
4236         }
4237
4238         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4239             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4240                 ext4_msg(sb, KERN_ERR, "can't mount with "
4241                         "journal_async_commit in data=ordered mode");
4242                 goto failed_mount_wq;
4243         }
4244
4245         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4246
4247         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4248
4249 no_journal:
4250         if (!test_opt(sb, NO_MBCACHE)) {
4251                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4252                 if (!sbi->s_ea_block_cache) {
4253                         ext4_msg(sb, KERN_ERR,
4254                                  "Failed to create ea_block_cache");
4255                         goto failed_mount_wq;
4256                 }
4257
4258                 if (ext4_has_feature_ea_inode(sb)) {
4259                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4260                         if (!sbi->s_ea_inode_cache) {
4261                                 ext4_msg(sb, KERN_ERR,
4262                                          "Failed to create ea_inode_cache");
4263                                 goto failed_mount_wq;
4264                         }
4265                 }
4266         }
4267
4268         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4269             (blocksize != PAGE_SIZE)) {
4270                 ext4_msg(sb, KERN_ERR,
4271                          "Unsupported blocksize for fs encryption");
4272                 goto failed_mount_wq;
4273         }
4274
4275         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4276             !ext4_has_feature_encrypt(sb)) {
4277                 ext4_set_feature_encrypt(sb);
4278                 ext4_commit_super(sb, 1);
4279         }
4280
4281         /*
4282          * Get the # of file system overhead blocks from the
4283          * superblock if present.
4284          */
4285         if (es->s_overhead_clusters)
4286                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4287         else {
4288                 err = ext4_calculate_overhead(sb);
4289                 if (err)
4290                         goto failed_mount_wq;
4291         }
4292
4293         /*
4294          * The maximum number of concurrent works can be high and
4295          * concurrency isn't really necessary.  Limit it to 1.
4296          */
4297         EXT4_SB(sb)->rsv_conversion_wq =
4298                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4299         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4300                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4301                 ret = -ENOMEM;
4302                 goto failed_mount4;
4303         }
4304
4305         /*
4306          * The jbd2_journal_load will have done any necessary log recovery,
4307          * so we can safely mount the rest of the filesystem now.
4308          */
4309
4310         root = ext4_iget(sb, EXT4_ROOT_INO);
4311         if (IS_ERR(root)) {
4312                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4313                 ret = PTR_ERR(root);
4314                 root = NULL;
4315                 goto failed_mount4;
4316         }
4317         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4318                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4319                 iput(root);
4320                 goto failed_mount4;
4321         }
4322         sb->s_root = d_make_root(root);
4323         if (!sb->s_root) {
4324                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4325                 ret = -ENOMEM;
4326                 goto failed_mount4;
4327         }
4328
4329         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4330         if (ret == -EROFS) {
4331                 sb->s_flags |= SB_RDONLY;
4332                 ret = 0;
4333         } else if (ret)
4334                 goto failed_mount4a;
4335
4336         /* determine the minimum size of new large inodes, if present */
4337         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4338             sbi->s_want_extra_isize == 0) {
4339                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4340                                                      EXT4_GOOD_OLD_INODE_SIZE;
4341                 if (ext4_has_feature_extra_isize(sb)) {
4342                         if (sbi->s_want_extra_isize <
4343                             le16_to_cpu(es->s_want_extra_isize))
4344                                 sbi->s_want_extra_isize =
4345                                         le16_to_cpu(es->s_want_extra_isize);
4346                         if (sbi->s_want_extra_isize <
4347                             le16_to_cpu(es->s_min_extra_isize))
4348                                 sbi->s_want_extra_isize =
4349                                         le16_to_cpu(es->s_min_extra_isize);
4350                 }
4351         }
4352         /* Check if enough inode space is available */
4353         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4354                                                         sbi->s_inode_size) {
4355                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4356                                                        EXT4_GOOD_OLD_INODE_SIZE;
4357                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4358                          "available");
4359         }
4360
4361         ext4_set_resv_clusters(sb);
4362
4363         err = ext4_setup_system_zone(sb);
4364         if (err) {
4365                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4366                          "zone (%d)", err);
4367                 goto failed_mount4a;
4368         }
4369
4370         ext4_ext_init(sb);
4371         err = ext4_mb_init(sb);
4372         if (err) {
4373                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4374                          err);
4375                 goto failed_mount5;
4376         }
4377
4378         block = ext4_count_free_clusters(sb);
4379         ext4_free_blocks_count_set(sbi->s_es,
4380                                    EXT4_C2B(sbi, block));
4381         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4382                                   GFP_KERNEL);
4383         if (!err) {
4384                 unsigned long freei = ext4_count_free_inodes(sb);
4385                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4386                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4387                                           GFP_KERNEL);
4388         }
4389         if (!err)
4390                 err = percpu_counter_init(&sbi->s_dirs_counter,
4391                                           ext4_count_dirs(sb), GFP_KERNEL);
4392         if (!err)
4393                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4394                                           GFP_KERNEL);
4395         if (!err)
4396                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4397
4398         if (err) {
4399                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4400                 goto failed_mount6;
4401         }
4402
4403         if (ext4_has_feature_flex_bg(sb))
4404                 if (!ext4_fill_flex_info(sb)) {
4405                         ext4_msg(sb, KERN_ERR,
4406                                "unable to initialize "
4407                                "flex_bg meta info!");
4408                         goto failed_mount6;
4409                 }
4410
4411         err = ext4_register_li_request(sb, first_not_zeroed);
4412         if (err)
4413                 goto failed_mount6;
4414
4415         err = ext4_register_sysfs(sb);
4416         if (err)
4417                 goto failed_mount7;
4418
4419 #ifdef CONFIG_QUOTA
4420         /* Enable quota usage during mount. */
4421         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4422                 err = ext4_enable_quotas(sb);
4423                 if (err)
4424                         goto failed_mount8;
4425         }
4426 #endif  /* CONFIG_QUOTA */
4427
4428         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4429         ext4_orphan_cleanup(sb, es);
4430         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4431         if (needs_recovery) {
4432                 ext4_msg(sb, KERN_INFO, "recovery complete");
4433                 ext4_mark_recovery_complete(sb, es);
4434         }
4435         if (EXT4_SB(sb)->s_journal) {
4436                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4437                         descr = " journalled data mode";
4438                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4439                         descr = " ordered data mode";
4440                 else
4441                         descr = " writeback data mode";
4442         } else
4443                 descr = "out journal";
4444
4445         if (test_opt(sb, DISCARD)) {
4446                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4447                 if (!blk_queue_discard(q))
4448                         ext4_msg(sb, KERN_WARNING,
4449                                  "mounting with \"discard\" option, but "
4450                                  "the device does not support discard");
4451         }
4452
4453         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4454                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4455                          "Opts: %.*s%s%s", descr,
4456                          (int) sizeof(sbi->s_es->s_mount_opts),
4457                          sbi->s_es->s_mount_opts,
4458                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4459
4460         if (es->s_error_count)
4461                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4462
4463         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4464         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4465         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4466         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4467
4468         kfree(orig_data);
4469         return 0;
4470
4471 cantfind_ext4:
4472         if (!silent)
4473                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4474         goto failed_mount;
4475
4476 #ifdef CONFIG_QUOTA
4477 failed_mount8:
4478         ext4_unregister_sysfs(sb);
4479 #endif
4480 failed_mount7:
4481         ext4_unregister_li_request(sb);
4482 failed_mount6:
4483         ext4_mb_release(sb);
4484         if (sbi->s_flex_groups)
4485                 kvfree(sbi->s_flex_groups);
4486         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4487         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4488         percpu_counter_destroy(&sbi->s_dirs_counter);
4489         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4490 failed_mount5:
4491         ext4_ext_release(sb);
4492         ext4_release_system_zone(sb);
4493 failed_mount4a:
4494         dput(sb->s_root);
4495         sb->s_root = NULL;
4496 failed_mount4:
4497         ext4_msg(sb, KERN_ERR, "mount failed");
4498         if (EXT4_SB(sb)->rsv_conversion_wq)
4499                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4500 failed_mount_wq:
4501         if (sbi->s_ea_inode_cache) {
4502                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4503                 sbi->s_ea_inode_cache = NULL;
4504         }
4505         if (sbi->s_ea_block_cache) {
4506                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4507                 sbi->s_ea_block_cache = NULL;
4508         }
4509         if (sbi->s_journal) {
4510                 jbd2_journal_destroy(sbi->s_journal);
4511                 sbi->s_journal = NULL;
4512         }
4513 failed_mount3a:
4514         ext4_es_unregister_shrinker(sbi);
4515 failed_mount3:
4516         del_timer_sync(&sbi->s_err_report);
4517         if (sbi->s_mmp_tsk)
4518                 kthread_stop(sbi->s_mmp_tsk);
4519 failed_mount2:
4520         for (i = 0; i < db_count; i++)
4521                 brelse(sbi->s_group_desc[i]);
4522         kvfree(sbi->s_group_desc);
4523 failed_mount:
4524         if (sbi->s_chksum_driver)
4525                 crypto_free_shash(sbi->s_chksum_driver);
4526 #ifdef CONFIG_QUOTA
4527         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4528                 kfree(sbi->s_qf_names[i]);
4529 #endif
4530         ext4_blkdev_remove(sbi);
4531         brelse(bh);
4532 out_fail:
4533         sb->s_fs_info = NULL;
4534         kfree(sbi->s_blockgroup_lock);
4535 out_free_base:
4536         kfree(sbi);
4537         kfree(orig_data);
4538         fs_put_dax(dax_dev);
4539         return err ? err : ret;
4540 }
4541
4542 /*
4543  * Setup any per-fs journal parameters now.  We'll do this both on
4544  * initial mount, once the journal has been initialised but before we've
4545  * done any recovery; and again on any subsequent remount.
4546  */
4547 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4548 {
4549         struct ext4_sb_info *sbi = EXT4_SB(sb);
4550
4551         journal->j_commit_interval = sbi->s_commit_interval;
4552         journal->j_min_batch_time = sbi->s_min_batch_time;
4553         journal->j_max_batch_time = sbi->s_max_batch_time;
4554
4555         write_lock(&journal->j_state_lock);
4556         if (test_opt(sb, BARRIER))
4557                 journal->j_flags |= JBD2_BARRIER;
4558         else
4559                 journal->j_flags &= ~JBD2_BARRIER;
4560         if (test_opt(sb, DATA_ERR_ABORT))
4561                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4562         else
4563                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4564         write_unlock(&journal->j_state_lock);
4565 }
4566
4567 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4568                                              unsigned int journal_inum)
4569 {
4570         struct inode *journal_inode;
4571
4572         /*
4573          * Test for the existence of a valid inode on disk.  Bad things
4574          * happen if we iget() an unused inode, as the subsequent iput()
4575          * will try to delete it.
4576          */
4577         journal_inode = ext4_iget(sb, journal_inum);
4578         if (IS_ERR(journal_inode)) {
4579                 ext4_msg(sb, KERN_ERR, "no journal found");
4580                 return NULL;
4581         }
4582         if (!journal_inode->i_nlink) {
4583                 make_bad_inode(journal_inode);
4584                 iput(journal_inode);
4585                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4586                 return NULL;
4587         }
4588
4589         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4590                   journal_inode, journal_inode->i_size);
4591         if (!S_ISREG(journal_inode->i_mode)) {
4592                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4593                 iput(journal_inode);
4594                 return NULL;
4595         }
4596         return journal_inode;
4597 }
4598
4599 static journal_t *ext4_get_journal(struct super_block *sb,
4600                                    unsigned int journal_inum)
4601 {
4602         struct inode *journal_inode;
4603         journal_t *journal;
4604
4605         BUG_ON(!ext4_has_feature_journal(sb));
4606
4607         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4608         if (!journal_inode)
4609                 return NULL;
4610
4611         journal = jbd2_journal_init_inode(journal_inode);
4612         if (!journal) {
4613                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4614                 iput(journal_inode);
4615                 return NULL;
4616         }
4617         journal->j_private = sb;
4618         ext4_init_journal_params(sb, journal);
4619         return journal;
4620 }
4621
4622 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4623                                        dev_t j_dev)
4624 {
4625         struct buffer_head *bh;
4626         journal_t *journal;
4627         ext4_fsblk_t start;
4628         ext4_fsblk_t len;
4629         int hblock, blocksize;
4630         ext4_fsblk_t sb_block;
4631         unsigned long offset;
4632         struct ext4_super_block *es;
4633         struct block_device *bdev;
4634
4635         BUG_ON(!ext4_has_feature_journal(sb));
4636
4637         bdev = ext4_blkdev_get(j_dev, sb);
4638         if (bdev == NULL)
4639                 return NULL;
4640
4641         blocksize = sb->s_blocksize;
4642         hblock = bdev_logical_block_size(bdev);
4643         if (blocksize < hblock) {
4644                 ext4_msg(sb, KERN_ERR,
4645                         "blocksize too small for journal device");
4646                 goto out_bdev;
4647         }
4648
4649         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4650         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4651         set_blocksize(bdev, blocksize);
4652         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4653                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4654                        "external journal");
4655                 goto out_bdev;
4656         }
4657
4658         es = (struct ext4_super_block *) (bh->b_data + offset);
4659         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4660             !(le32_to_cpu(es->s_feature_incompat) &
4661               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4662                 ext4_msg(sb, KERN_ERR, "external journal has "
4663                                         "bad superblock");
4664                 brelse(bh);
4665                 goto out_bdev;
4666         }
4667
4668         if ((le32_to_cpu(es->s_feature_ro_compat) &
4669              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4670             es->s_checksum != ext4_superblock_csum(sb, es)) {
4671                 ext4_msg(sb, KERN_ERR, "external journal has "
4672                                        "corrupt superblock");
4673                 brelse(bh);
4674                 goto out_bdev;
4675         }
4676
4677         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4678                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4679                 brelse(bh);
4680                 goto out_bdev;
4681         }
4682
4683         len = ext4_blocks_count(es);
4684         start = sb_block + 1;
4685         brelse(bh);     /* we're done with the superblock */
4686
4687         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4688                                         start, len, blocksize);
4689         if (!journal) {
4690                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4691                 goto out_bdev;
4692         }
4693         journal->j_private = sb;
4694         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4695         wait_on_buffer(journal->j_sb_buffer);
4696         if (!buffer_uptodate(journal->j_sb_buffer)) {
4697                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4698                 goto out_journal;
4699         }
4700         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4701                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4702                                         "user (unsupported) - %d",
4703                         be32_to_cpu(journal->j_superblock->s_nr_users));
4704                 goto out_journal;
4705         }
4706         EXT4_SB(sb)->journal_bdev = bdev;
4707         ext4_init_journal_params(sb, journal);
4708         return journal;
4709
4710 out_journal:
4711         jbd2_journal_destroy(journal);
4712 out_bdev:
4713         ext4_blkdev_put(bdev);
4714         return NULL;
4715 }
4716
4717 static int ext4_load_journal(struct super_block *sb,
4718                              struct ext4_super_block *es,
4719                              unsigned long journal_devnum)
4720 {
4721         journal_t *journal;
4722         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4723         dev_t journal_dev;
4724         int err = 0;
4725         int really_read_only;
4726
4727         BUG_ON(!ext4_has_feature_journal(sb));
4728
4729         if (journal_devnum &&
4730             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4731                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4732                         "numbers have changed");
4733                 journal_dev = new_decode_dev(journal_devnum);
4734         } else
4735                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4736
4737         really_read_only = bdev_read_only(sb->s_bdev);
4738
4739         /*
4740          * Are we loading a blank journal or performing recovery after a
4741          * crash?  For recovery, we need to check in advance whether we
4742          * can get read-write access to the device.
4743          */
4744         if (ext4_has_feature_journal_needs_recovery(sb)) {
4745                 if (sb_rdonly(sb)) {
4746                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4747                                         "required on readonly filesystem");
4748                         if (really_read_only) {
4749                                 ext4_msg(sb, KERN_ERR, "write access "
4750                                         "unavailable, cannot proceed "
4751                                         "(try mounting with noload)");
4752                                 return -EROFS;
4753                         }
4754                         ext4_msg(sb, KERN_INFO, "write access will "
4755                                "be enabled during recovery");
4756                 }
4757         }
4758
4759         if (journal_inum && journal_dev) {
4760                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4761                        "and inode journals!");
4762                 return -EINVAL;
4763         }
4764
4765         if (journal_inum) {
4766                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4767                         return -EINVAL;
4768         } else {
4769                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4770                         return -EINVAL;
4771         }
4772
4773         if (!(journal->j_flags & JBD2_BARRIER))
4774                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4775
4776         if (!ext4_has_feature_journal_needs_recovery(sb))
4777                 err = jbd2_journal_wipe(journal, !really_read_only);
4778         if (!err) {
4779                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4780                 if (save)
4781                         memcpy(save, ((char *) es) +
4782                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4783                 err = jbd2_journal_load(journal);
4784                 if (save)
4785                         memcpy(((char *) es) + EXT4_S_ERR_START,
4786                                save, EXT4_S_ERR_LEN);
4787                 kfree(save);
4788         }
4789
4790         if (err) {
4791                 ext4_msg(sb, KERN_ERR, "error loading journal");
4792                 jbd2_journal_destroy(journal);
4793                 return err;
4794         }
4795
4796         EXT4_SB(sb)->s_journal = journal;
4797         ext4_clear_journal_err(sb, es);
4798
4799         if (!really_read_only && journal_devnum &&
4800             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4801                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4802
4803                 /* Make sure we flush the recovery flag to disk. */
4804                 ext4_commit_super(sb, 1);
4805         }
4806
4807         return 0;
4808 }
4809
4810 static int ext4_commit_super(struct super_block *sb, int sync)
4811 {
4812         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4813         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4814         int error = 0;
4815
4816         if (!sbh || block_device_ejected(sb))
4817                 return error;
4818
4819         /*
4820          * The superblock bh should be mapped, but it might not be if the
4821          * device was hot-removed. Not much we can do but fail the I/O.
4822          */
4823         if (!buffer_mapped(sbh))
4824                 return error;
4825
4826         /*
4827          * If the file system is mounted read-only, don't update the
4828          * superblock write time.  This avoids updating the superblock
4829          * write time when we are mounting the root file system
4830          * read/only but we need to replay the journal; at that point,
4831          * for people who are east of GMT and who make their clock
4832          * tick in localtime for Windows bug-for-bug compatibility,
4833          * the clock is set in the future, and this will cause e2fsck
4834          * to complain and force a full file system check.
4835          */
4836         if (!(sb->s_flags & SB_RDONLY))
4837                 ext4_update_tstamp(es, s_wtime);
4838         if (sb->s_bdev->bd_part)
4839                 es->s_kbytes_written =
4840                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4841                             ((part_stat_read(sb->s_bdev->bd_part,
4842                                              sectors[STAT_WRITE]) -
4843                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4844         else
4845                 es->s_kbytes_written =
4846                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4847         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4848                 ext4_free_blocks_count_set(es,
4849                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4850                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4851         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4852                 es->s_free_inodes_count =
4853                         cpu_to_le32(percpu_counter_sum_positive(
4854                                 &EXT4_SB(sb)->s_freeinodes_counter));
4855         BUFFER_TRACE(sbh, "marking dirty");
4856         ext4_superblock_csum_set(sb);
4857         if (sync)
4858                 lock_buffer(sbh);
4859         if (buffer_write_io_error(sbh)) {
4860                 /*
4861                  * Oh, dear.  A previous attempt to write the
4862                  * superblock failed.  This could happen because the
4863                  * USB device was yanked out.  Or it could happen to
4864                  * be a transient write error and maybe the block will
4865                  * be remapped.  Nothing we can do but to retry the
4866                  * write and hope for the best.
4867                  */
4868                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4869                        "superblock detected");
4870                 clear_buffer_write_io_error(sbh);
4871                 set_buffer_uptodate(sbh);
4872         }
4873         mark_buffer_dirty(sbh);
4874         if (sync) {
4875                 unlock_buffer(sbh);
4876                 error = __sync_dirty_buffer(sbh,
4877                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4878                 if (buffer_write_io_error(sbh)) {
4879                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4880                                "superblock");
4881                         clear_buffer_write_io_error(sbh);
4882                         set_buffer_uptodate(sbh);
4883                 }
4884         }
4885         return error;
4886 }
4887
4888 /*
4889  * Have we just finished recovery?  If so, and if we are mounting (or
4890  * remounting) the filesystem readonly, then we will end up with a
4891  * consistent fs on disk.  Record that fact.
4892  */
4893 static void ext4_mark_recovery_complete(struct super_block *sb,
4894                                         struct ext4_super_block *es)
4895 {
4896         journal_t *journal = EXT4_SB(sb)->s_journal;
4897
4898         if (!ext4_has_feature_journal(sb)) {
4899                 BUG_ON(journal != NULL);
4900                 return;
4901         }
4902         jbd2_journal_lock_updates(journal);
4903         if (jbd2_journal_flush(journal) < 0)
4904                 goto out;
4905
4906         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4907                 ext4_clear_feature_journal_needs_recovery(sb);
4908                 ext4_commit_super(sb, 1);
4909         }
4910
4911 out:
4912         jbd2_journal_unlock_updates(journal);
4913 }
4914
4915 /*
4916  * If we are mounting (or read-write remounting) a filesystem whose journal
4917  * has recorded an error from a previous lifetime, move that error to the
4918  * main filesystem now.
4919  */
4920 static void ext4_clear_journal_err(struct super_block *sb,
4921                                    struct ext4_super_block *es)
4922 {
4923         journal_t *journal;
4924         int j_errno;
4925         const char *errstr;
4926
4927         BUG_ON(!ext4_has_feature_journal(sb));
4928
4929         journal = EXT4_SB(sb)->s_journal;
4930
4931         /*
4932          * Now check for any error status which may have been recorded in the
4933          * journal by a prior ext4_error() or ext4_abort()
4934          */
4935
4936         j_errno = jbd2_journal_errno(journal);
4937         if (j_errno) {
4938                 char nbuf[16];
4939
4940                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4941                 ext4_warning(sb, "Filesystem error recorded "
4942                              "from previous mount: %s", errstr);
4943                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4944
4945                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4946                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4947                 ext4_commit_super(sb, 1);
4948
4949                 jbd2_journal_clear_err(journal);
4950                 jbd2_journal_update_sb_errno(journal);
4951         }
4952 }
4953
4954 /*
4955  * Force the running and committing transactions to commit,
4956  * and wait on the commit.
4957  */
4958 int ext4_force_commit(struct super_block *sb)
4959 {
4960         journal_t *journal;
4961
4962         if (sb_rdonly(sb))
4963                 return 0;
4964
4965         journal = EXT4_SB(sb)->s_journal;
4966         return ext4_journal_force_commit(journal);
4967 }
4968
4969 static int ext4_sync_fs(struct super_block *sb, int wait)
4970 {
4971         int ret = 0;
4972         tid_t target;
4973         bool needs_barrier = false;
4974         struct ext4_sb_info *sbi = EXT4_SB(sb);
4975
4976         if (unlikely(ext4_forced_shutdown(sbi)))
4977                 return 0;
4978
4979         trace_ext4_sync_fs(sb, wait);
4980         flush_workqueue(sbi->rsv_conversion_wq);
4981         /*
4982          * Writeback quota in non-journalled quota case - journalled quota has
4983          * no dirty dquots
4984          */
4985         dquot_writeback_dquots(sb, -1);
4986         /*
4987          * Data writeback is possible w/o journal transaction, so barrier must
4988          * being sent at the end of the function. But we can skip it if
4989          * transaction_commit will do it for us.
4990          */
4991         if (sbi->s_journal) {
4992                 target = jbd2_get_latest_transaction(sbi->s_journal);
4993                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4994                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4995                         needs_barrier = true;
4996
4997                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4998                         if (wait)
4999                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5000                                                            target);
5001                 }
5002         } else if (wait && test_opt(sb, BARRIER))
5003                 needs_barrier = true;
5004         if (needs_barrier) {
5005                 int err;
5006                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5007                 if (!ret)
5008                         ret = err;
5009         }
5010
5011         return ret;
5012 }
5013
5014 /*
5015  * LVM calls this function before a (read-only) snapshot is created.  This
5016  * gives us a chance to flush the journal completely and mark the fs clean.
5017  *
5018  * Note that only this function cannot bring a filesystem to be in a clean
5019  * state independently. It relies on upper layer to stop all data & metadata
5020  * modifications.
5021  */
5022 static int ext4_freeze(struct super_block *sb)
5023 {
5024         int error = 0;
5025         journal_t *journal;
5026
5027         if (sb_rdonly(sb))
5028                 return 0;
5029
5030         journal = EXT4_SB(sb)->s_journal;
5031
5032         if (journal) {
5033                 /* Now we set up the journal barrier. */
5034                 jbd2_journal_lock_updates(journal);
5035
5036                 /*
5037                  * Don't clear the needs_recovery flag if we failed to
5038                  * flush the journal.
5039                  */
5040                 error = jbd2_journal_flush(journal);
5041                 if (error < 0)
5042                         goto out;
5043
5044                 /* Journal blocked and flushed, clear needs_recovery flag. */
5045                 ext4_clear_feature_journal_needs_recovery(sb);
5046         }
5047
5048         error = ext4_commit_super(sb, 1);
5049 out:
5050         if (journal)
5051                 /* we rely on upper layer to stop further updates */
5052                 jbd2_journal_unlock_updates(journal);
5053         return error;
5054 }
5055
5056 /*
5057  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5058  * flag here, even though the filesystem is not technically dirty yet.
5059  */
5060 static int ext4_unfreeze(struct super_block *sb)
5061 {
5062         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5063                 return 0;
5064
5065         if (EXT4_SB(sb)->s_journal) {
5066                 /* Reset the needs_recovery flag before the fs is unlocked. */
5067                 ext4_set_feature_journal_needs_recovery(sb);
5068         }
5069
5070         ext4_commit_super(sb, 1);
5071         return 0;
5072 }
5073
5074 /*
5075  * Structure to save mount options for ext4_remount's benefit
5076  */
5077 struct ext4_mount_options {
5078         unsigned long s_mount_opt;
5079         unsigned long s_mount_opt2;
5080         kuid_t s_resuid;
5081         kgid_t s_resgid;
5082         unsigned long s_commit_interval;
5083         u32 s_min_batch_time, s_max_batch_time;
5084 #ifdef CONFIG_QUOTA
5085         int s_jquota_fmt;
5086         char *s_qf_names[EXT4_MAXQUOTAS];
5087 #endif
5088 };
5089
5090 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5091 {
5092         struct ext4_super_block *es;
5093         struct ext4_sb_info *sbi = EXT4_SB(sb);
5094         unsigned long old_sb_flags;
5095         struct ext4_mount_options old_opts;
5096         int enable_quota = 0;
5097         ext4_group_t g;
5098         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5099         int err = 0;
5100 #ifdef CONFIG_QUOTA
5101         int i, j;
5102 #endif
5103         char *orig_data = kstrdup(data, GFP_KERNEL);
5104
5105         if (data && !orig_data)
5106                 return -ENOMEM;
5107
5108         /* Store the original options */
5109         old_sb_flags = sb->s_flags;
5110         old_opts.s_mount_opt = sbi->s_mount_opt;
5111         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5112         old_opts.s_resuid = sbi->s_resuid;
5113         old_opts.s_resgid = sbi->s_resgid;
5114         old_opts.s_commit_interval = sbi->s_commit_interval;
5115         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5116         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5117 #ifdef CONFIG_QUOTA
5118         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5119         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5120                 if (sbi->s_qf_names[i]) {
5121                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
5122                                                          GFP_KERNEL);
5123                         if (!old_opts.s_qf_names[i]) {
5124                                 for (j = 0; j < i; j++)
5125                                         kfree(old_opts.s_qf_names[j]);
5126                                 kfree(orig_data);
5127                                 return -ENOMEM;
5128                         }
5129                 } else
5130                         old_opts.s_qf_names[i] = NULL;
5131 #endif
5132         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5133                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5134
5135         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5136                 err = -EINVAL;
5137                 goto restore_opts;
5138         }
5139
5140         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5141             test_opt(sb, JOURNAL_CHECKSUM)) {
5142                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5143                          "during remount not supported; ignoring");
5144                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5145         }
5146
5147         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5148                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5149                         ext4_msg(sb, KERN_ERR, "can't mount with "
5150                                  "both data=journal and delalloc");
5151                         err = -EINVAL;
5152                         goto restore_opts;
5153                 }
5154                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5155                         ext4_msg(sb, KERN_ERR, "can't mount with "
5156                                  "both data=journal and dioread_nolock");
5157                         err = -EINVAL;
5158                         goto restore_opts;
5159                 }
5160                 if (test_opt(sb, DAX)) {
5161                         ext4_msg(sb, KERN_ERR, "can't mount with "
5162                                  "both data=journal and dax");
5163                         err = -EINVAL;
5164                         goto restore_opts;
5165                 }
5166         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5167                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5168                         ext4_msg(sb, KERN_ERR, "can't mount with "
5169                                 "journal_async_commit in data=ordered mode");
5170                         err = -EINVAL;
5171                         goto restore_opts;
5172                 }
5173         }
5174
5175         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5176                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5177                 err = -EINVAL;
5178                 goto restore_opts;
5179         }
5180
5181         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5182                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5183                         "dax flag with busy inodes while remounting");
5184                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5185         }
5186
5187         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5188                 ext4_abort(sb, "Abort forced by user");
5189
5190         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5191                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5192
5193         es = sbi->s_es;
5194
5195         if (sbi->s_journal) {
5196                 ext4_init_journal_params(sb, sbi->s_journal);
5197                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5198         }
5199
5200         if (*flags & SB_LAZYTIME)
5201                 sb->s_flags |= SB_LAZYTIME;
5202
5203         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5204                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5205                         err = -EROFS;
5206                         goto restore_opts;
5207                 }
5208
5209                 if (*flags & SB_RDONLY) {
5210                         err = sync_filesystem(sb);
5211                         if (err < 0)
5212                                 goto restore_opts;
5213                         err = dquot_suspend(sb, -1);
5214                         if (err < 0)
5215                                 goto restore_opts;
5216
5217                         /*
5218                          * First of all, the unconditional stuff we have to do
5219                          * to disable replay of the journal when we next remount
5220                          */
5221                         sb->s_flags |= SB_RDONLY;
5222
5223                         /*
5224                          * OK, test if we are remounting a valid rw partition
5225                          * readonly, and if so set the rdonly flag and then
5226                          * mark the partition as valid again.
5227                          */
5228                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5229                             (sbi->s_mount_state & EXT4_VALID_FS))
5230                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5231
5232                         if (sbi->s_journal)
5233                                 ext4_mark_recovery_complete(sb, es);
5234                         if (sbi->s_mmp_tsk)
5235                                 kthread_stop(sbi->s_mmp_tsk);
5236                 } else {
5237                         /* Make sure we can mount this feature set readwrite */
5238                         if (ext4_has_feature_readonly(sb) ||
5239                             !ext4_feature_set_ok(sb, 0)) {
5240                                 err = -EROFS;
5241                                 goto restore_opts;
5242                         }
5243                         /*
5244                          * Make sure the group descriptor checksums
5245                          * are sane.  If they aren't, refuse to remount r/w.
5246                          */
5247                         for (g = 0; g < sbi->s_groups_count; g++) {
5248                                 struct ext4_group_desc *gdp =
5249                                         ext4_get_group_desc(sb, g, NULL);
5250
5251                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5252                                         ext4_msg(sb, KERN_ERR,
5253                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5254                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5255                                                le16_to_cpu(gdp->bg_checksum));
5256                                         err = -EFSBADCRC;
5257                                         goto restore_opts;
5258                                 }
5259                         }
5260
5261                         /*
5262                          * If we have an unprocessed orphan list hanging
5263                          * around from a previously readonly bdev mount,
5264                          * require a full umount/remount for now.
5265                          */
5266                         if (es->s_last_orphan) {
5267                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5268                                        "remount RDWR because of unprocessed "
5269                                        "orphan inode list.  Please "
5270                                        "umount/remount instead");
5271                                 err = -EINVAL;
5272                                 goto restore_opts;
5273                         }
5274
5275                         /*
5276                          * Mounting a RDONLY partition read-write, so reread
5277                          * and store the current valid flag.  (It may have
5278                          * been changed by e2fsck since we originally mounted
5279                          * the partition.)
5280                          */
5281                         if (sbi->s_journal)
5282                                 ext4_clear_journal_err(sb, es);
5283                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5284
5285                         err = ext4_setup_super(sb, es, 0);
5286                         if (err)
5287                                 goto restore_opts;
5288
5289                         sb->s_flags &= ~SB_RDONLY;
5290                         if (ext4_has_feature_mmp(sb))
5291                                 if (ext4_multi_mount_protect(sb,
5292                                                 le64_to_cpu(es->s_mmp_block))) {
5293                                         err = -EROFS;
5294                                         goto restore_opts;
5295                                 }
5296                         enable_quota = 1;
5297                 }
5298         }
5299
5300         /*
5301          * Reinitialize lazy itable initialization thread based on
5302          * current settings
5303          */
5304         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5305                 ext4_unregister_li_request(sb);
5306         else {
5307                 ext4_group_t first_not_zeroed;
5308                 first_not_zeroed = ext4_has_uninit_itable(sb);
5309                 ext4_register_li_request(sb, first_not_zeroed);
5310         }
5311
5312         ext4_setup_system_zone(sb);
5313         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5314                 err = ext4_commit_super(sb, 1);
5315                 if (err)
5316                         goto restore_opts;
5317         }
5318
5319 #ifdef CONFIG_QUOTA
5320         /* Release old quota file names */
5321         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5322                 kfree(old_opts.s_qf_names[i]);
5323         if (enable_quota) {
5324                 if (sb_any_quota_suspended(sb))
5325                         dquot_resume(sb, -1);
5326                 else if (ext4_has_feature_quota(sb)) {
5327                         err = ext4_enable_quotas(sb);
5328                         if (err)
5329                                 goto restore_opts;
5330                 }
5331         }
5332 #endif
5333
5334         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5335         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5336         kfree(orig_data);
5337         return 0;
5338
5339 restore_opts:
5340         sb->s_flags = old_sb_flags;
5341         sbi->s_mount_opt = old_opts.s_mount_opt;
5342         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5343         sbi->s_resuid = old_opts.s_resuid;
5344         sbi->s_resgid = old_opts.s_resgid;
5345         sbi->s_commit_interval = old_opts.s_commit_interval;
5346         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5347         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5348 #ifdef CONFIG_QUOTA
5349         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5350         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5351                 kfree(sbi->s_qf_names[i]);
5352                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5353         }
5354 #endif
5355         kfree(orig_data);
5356         return err;
5357 }
5358
5359 #ifdef CONFIG_QUOTA
5360 static int ext4_statfs_project(struct super_block *sb,
5361                                kprojid_t projid, struct kstatfs *buf)
5362 {
5363         struct kqid qid;
5364         struct dquot *dquot;
5365         u64 limit;
5366         u64 curblock;
5367
5368         qid = make_kqid_projid(projid);
5369         dquot = dqget(sb, qid);
5370         if (IS_ERR(dquot))
5371                 return PTR_ERR(dquot);
5372         spin_lock(&dquot->dq_dqb_lock);
5373
5374         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5375                  dquot->dq_dqb.dqb_bsoftlimit :
5376                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5377         if (limit && buf->f_blocks > limit) {
5378                 curblock = (dquot->dq_dqb.dqb_curspace +
5379                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5380                 buf->f_blocks = limit;
5381                 buf->f_bfree = buf->f_bavail =
5382                         (buf->f_blocks > curblock) ?
5383                          (buf->f_blocks - curblock) : 0;
5384         }
5385
5386         limit = dquot->dq_dqb.dqb_isoftlimit ?
5387                 dquot->dq_dqb.dqb_isoftlimit :
5388                 dquot->dq_dqb.dqb_ihardlimit;
5389         if (limit && buf->f_files > limit) {
5390                 buf->f_files = limit;
5391                 buf->f_ffree =
5392                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5393                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5394         }
5395
5396         spin_unlock(&dquot->dq_dqb_lock);
5397         dqput(dquot);
5398         return 0;
5399 }
5400 #endif
5401
5402 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5403 {
5404         struct super_block *sb = dentry->d_sb;
5405         struct ext4_sb_info *sbi = EXT4_SB(sb);
5406         struct ext4_super_block *es = sbi->s_es;
5407         ext4_fsblk_t overhead = 0, resv_blocks;
5408         u64 fsid;
5409         s64 bfree;
5410         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5411
5412         if (!test_opt(sb, MINIX_DF))
5413                 overhead = sbi->s_overhead;
5414
5415         buf->f_type = EXT4_SUPER_MAGIC;
5416         buf->f_bsize = sb->s_blocksize;
5417         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5418         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5419                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5420         /* prevent underflow in case that few free space is available */
5421         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5422         buf->f_bavail = buf->f_bfree -
5423                         (ext4_r_blocks_count(es) + resv_blocks);
5424         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5425                 buf->f_bavail = 0;
5426         buf->f_files = le32_to_cpu(es->s_inodes_count);
5427         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5428         buf->f_namelen = EXT4_NAME_LEN;
5429         fsid = le64_to_cpup((void *)es->s_uuid) ^
5430                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5431         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5432         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5433
5434 #ifdef CONFIG_QUOTA
5435         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5436             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5437                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5438 #endif
5439         return 0;
5440 }
5441
5442
5443 #ifdef CONFIG_QUOTA
5444
5445 /*
5446  * Helper functions so that transaction is started before we acquire dqio_sem
5447  * to keep correct lock ordering of transaction > dqio_sem
5448  */
5449 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5450 {
5451         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5452 }
5453
5454 static int ext4_write_dquot(struct dquot *dquot)
5455 {
5456         int ret, err;
5457         handle_t *handle;
5458         struct inode *inode;
5459
5460         inode = dquot_to_inode(dquot);
5461         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5462                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5463         if (IS_ERR(handle))
5464                 return PTR_ERR(handle);
5465         ret = dquot_commit(dquot);
5466         err = ext4_journal_stop(handle);
5467         if (!ret)
5468                 ret = err;
5469         return ret;
5470 }
5471
5472 static int ext4_acquire_dquot(struct dquot *dquot)
5473 {
5474         int ret, err;
5475         handle_t *handle;
5476
5477         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5478                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5479         if (IS_ERR(handle))
5480                 return PTR_ERR(handle);
5481         ret = dquot_acquire(dquot);
5482         err = ext4_journal_stop(handle);
5483         if (!ret)
5484                 ret = err;
5485         return ret;
5486 }
5487
5488 static int ext4_release_dquot(struct dquot *dquot)
5489 {
5490         int ret, err;
5491         handle_t *handle;
5492
5493         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5494                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5495         if (IS_ERR(handle)) {
5496                 /* Release dquot anyway to avoid endless cycle in dqput() */
5497                 dquot_release(dquot);
5498                 return PTR_ERR(handle);
5499         }
5500         ret = dquot_release(dquot);
5501         err = ext4_journal_stop(handle);
5502         if (!ret)
5503                 ret = err;
5504         return ret;
5505 }
5506
5507 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5508 {
5509         struct super_block *sb = dquot->dq_sb;
5510         struct ext4_sb_info *sbi = EXT4_SB(sb);
5511
5512         /* Are we journaling quotas? */
5513         if (ext4_has_feature_quota(sb) ||
5514             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5515                 dquot_mark_dquot_dirty(dquot);
5516                 return ext4_write_dquot(dquot);
5517         } else {
5518                 return dquot_mark_dquot_dirty(dquot);
5519         }
5520 }
5521
5522 static int ext4_write_info(struct super_block *sb, int type)
5523 {
5524         int ret, err;
5525         handle_t *handle;
5526
5527         /* Data block + inode block */
5528         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5529         if (IS_ERR(handle))
5530                 return PTR_ERR(handle);
5531         ret = dquot_commit_info(sb, type);
5532         err = ext4_journal_stop(handle);
5533         if (!ret)
5534                 ret = err;
5535         return ret;
5536 }
5537
5538 /*
5539  * Turn on quotas during mount time - we need to find
5540  * the quota file and such...
5541  */
5542 static int ext4_quota_on_mount(struct super_block *sb, int type)
5543 {
5544         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5545                                         EXT4_SB(sb)->s_jquota_fmt, type);
5546 }
5547
5548 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5549 {
5550         struct ext4_inode_info *ei = EXT4_I(inode);
5551
5552         /* The first argument of lockdep_set_subclass has to be
5553          * *exactly* the same as the argument to init_rwsem() --- in
5554          * this case, in init_once() --- or lockdep gets unhappy
5555          * because the name of the lock is set using the
5556          * stringification of the argument to init_rwsem().
5557          */
5558         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5559         lockdep_set_subclass(&ei->i_data_sem, subclass);
5560 }
5561
5562 /*
5563  * Standard function to be called on quota_on
5564  */
5565 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5566                          const struct path *path)
5567 {
5568         int err;
5569
5570         if (!test_opt(sb, QUOTA))
5571                 return -EINVAL;
5572
5573         /* Quotafile not on the same filesystem? */
5574         if (path->dentry->d_sb != sb)
5575                 return -EXDEV;
5576         /* Journaling quota? */
5577         if (EXT4_SB(sb)->s_qf_names[type]) {
5578                 /* Quotafile not in fs root? */
5579                 if (path->dentry->d_parent != sb->s_root)
5580                         ext4_msg(sb, KERN_WARNING,
5581                                 "Quota file not on filesystem root. "
5582                                 "Journaled quota will not work");
5583                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5584         } else {
5585                 /*
5586                  * Clear the flag just in case mount options changed since
5587                  * last time.
5588                  */
5589                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5590         }
5591
5592         /*
5593          * When we journal data on quota file, we have to flush journal to see
5594          * all updates to the file when we bypass pagecache...
5595          */
5596         if (EXT4_SB(sb)->s_journal &&
5597             ext4_should_journal_data(d_inode(path->dentry))) {
5598                 /*
5599                  * We don't need to lock updates but journal_flush() could
5600                  * otherwise be livelocked...
5601                  */
5602                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5603                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5604                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5605                 if (err)
5606                         return err;
5607         }
5608
5609         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5610         err = dquot_quota_on(sb, type, format_id, path);
5611         if (err) {
5612                 lockdep_set_quota_inode(path->dentry->d_inode,
5613                                              I_DATA_SEM_NORMAL);
5614         } else {
5615                 struct inode *inode = d_inode(path->dentry);
5616                 handle_t *handle;
5617
5618                 /*
5619                  * Set inode flags to prevent userspace from messing with quota
5620                  * files. If this fails, we return success anyway since quotas
5621                  * are already enabled and this is not a hard failure.
5622                  */
5623                 inode_lock(inode);
5624                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5625                 if (IS_ERR(handle))
5626                         goto unlock_inode;
5627                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5628                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5629                                 S_NOATIME | S_IMMUTABLE);
5630                 ext4_mark_inode_dirty(handle, inode);
5631                 ext4_journal_stop(handle);
5632         unlock_inode:
5633                 inode_unlock(inode);
5634         }
5635         return err;
5636 }
5637
5638 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5639                              unsigned int flags)
5640 {
5641         int err;
5642         struct inode *qf_inode;
5643         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5644                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5645                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5646                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5647         };
5648
5649         BUG_ON(!ext4_has_feature_quota(sb));
5650
5651         if (!qf_inums[type])
5652                 return -EPERM;
5653
5654         qf_inode = ext4_iget(sb, qf_inums[type]);
5655         if (IS_ERR(qf_inode)) {
5656                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5657                 return PTR_ERR(qf_inode);
5658         }
5659
5660         /* Don't account quota for quota files to avoid recursion */
5661         qf_inode->i_flags |= S_NOQUOTA;
5662         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5663         err = dquot_enable(qf_inode, type, format_id, flags);
5664         iput(qf_inode);
5665         if (err)
5666                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5667
5668         return err;
5669 }
5670
5671 /* Enable usage tracking for all quota types. */
5672 static int ext4_enable_quotas(struct super_block *sb)
5673 {
5674         int type, err = 0;
5675         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5676                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5677                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5678                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5679         };
5680         bool quota_mopt[EXT4_MAXQUOTAS] = {
5681                 test_opt(sb, USRQUOTA),
5682                 test_opt(sb, GRPQUOTA),
5683                 test_opt(sb, PRJQUOTA),
5684         };
5685
5686         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5687         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5688                 if (qf_inums[type]) {
5689                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5690                                 DQUOT_USAGE_ENABLED |
5691                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5692                         if (err) {
5693                                 ext4_warning(sb,
5694                                         "Failed to enable quota tracking "
5695                                         "(type=%d, err=%d). Please run "
5696                                         "e2fsck to fix.", type, err);
5697                                 for (type--; type >= 0; type--)
5698                                         dquot_quota_off(sb, type);
5699
5700                                 return err;
5701                         }
5702                 }
5703         }
5704         return 0;
5705 }
5706
5707 static int ext4_quota_off(struct super_block *sb, int type)
5708 {
5709         struct inode *inode = sb_dqopt(sb)->files[type];
5710         handle_t *handle;
5711         int err;
5712
5713         /* Force all delayed allocation blocks to be allocated.
5714          * Caller already holds s_umount sem */
5715         if (test_opt(sb, DELALLOC))
5716                 sync_filesystem(sb);
5717
5718         if (!inode || !igrab(inode))
5719                 goto out;
5720
5721         err = dquot_quota_off(sb, type);
5722         if (err || ext4_has_feature_quota(sb))
5723                 goto out_put;
5724
5725         inode_lock(inode);
5726         /*
5727          * Update modification times of quota files when userspace can
5728          * start looking at them. If we fail, we return success anyway since
5729          * this is not a hard failure and quotas are already disabled.
5730          */
5731         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5732         if (IS_ERR(handle))
5733                 goto out_unlock;
5734         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5735         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5736         inode->i_mtime = inode->i_ctime = current_time(inode);
5737         ext4_mark_inode_dirty(handle, inode);
5738         ext4_journal_stop(handle);
5739 out_unlock:
5740         inode_unlock(inode);
5741 out_put:
5742         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5743         iput(inode);
5744         return err;
5745 out:
5746         return dquot_quota_off(sb, type);
5747 }
5748
5749 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5750  * acquiring the locks... As quota files are never truncated and quota code
5751  * itself serializes the operations (and no one else should touch the files)
5752  * we don't have to be afraid of races */
5753 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5754                                size_t len, loff_t off)
5755 {
5756         struct inode *inode = sb_dqopt(sb)->files[type];
5757         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5758         int offset = off & (sb->s_blocksize - 1);
5759         int tocopy;
5760         size_t toread;
5761         struct buffer_head *bh;
5762         loff_t i_size = i_size_read(inode);
5763
5764         if (off > i_size)
5765                 return 0;
5766         if (off+len > i_size)
5767                 len = i_size-off;
5768         toread = len;
5769         while (toread > 0) {
5770                 tocopy = sb->s_blocksize - offset < toread ?
5771                                 sb->s_blocksize - offset : toread;
5772                 bh = ext4_bread(NULL, inode, blk, 0);
5773                 if (IS_ERR(bh))
5774                         return PTR_ERR(bh);
5775                 if (!bh)        /* A hole? */
5776                         memset(data, 0, tocopy);
5777                 else
5778                         memcpy(data, bh->b_data+offset, tocopy);
5779                 brelse(bh);
5780                 offset = 0;
5781                 toread -= tocopy;
5782                 data += tocopy;
5783                 blk++;
5784         }
5785         return len;
5786 }
5787
5788 /* Write to quotafile (we know the transaction is already started and has
5789  * enough credits) */
5790 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5791                                 const char *data, size_t len, loff_t off)
5792 {
5793         struct inode *inode = sb_dqopt(sb)->files[type];
5794         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5795         int err, offset = off & (sb->s_blocksize - 1);
5796         int retries = 0;
5797         struct buffer_head *bh;
5798         handle_t *handle = journal_current_handle();
5799
5800         if (EXT4_SB(sb)->s_journal && !handle) {
5801                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5802                         " cancelled because transaction is not started",
5803                         (unsigned long long)off, (unsigned long long)len);
5804                 return -EIO;
5805         }
5806         /*
5807          * Since we account only one data block in transaction credits,
5808          * then it is impossible to cross a block boundary.
5809          */
5810         if (sb->s_blocksize - offset < len) {
5811                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5812                         " cancelled because not block aligned",
5813                         (unsigned long long)off, (unsigned long long)len);
5814                 return -EIO;
5815         }
5816
5817         do {
5818                 bh = ext4_bread(handle, inode, blk,
5819                                 EXT4_GET_BLOCKS_CREATE |
5820                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5821         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5822                  ext4_should_retry_alloc(inode->i_sb, &retries));
5823         if (IS_ERR(bh))
5824                 return PTR_ERR(bh);
5825         if (!bh)
5826                 goto out;
5827         BUFFER_TRACE(bh, "get write access");
5828         err = ext4_journal_get_write_access(handle, bh);
5829         if (err) {
5830                 brelse(bh);
5831                 return err;
5832         }
5833         lock_buffer(bh);
5834         memcpy(bh->b_data+offset, data, len);
5835         flush_dcache_page(bh->b_page);
5836         unlock_buffer(bh);
5837         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5838         brelse(bh);
5839 out:
5840         if (inode->i_size < off + len) {
5841                 i_size_write(inode, off + len);
5842                 EXT4_I(inode)->i_disksize = inode->i_size;
5843                 ext4_mark_inode_dirty(handle, inode);
5844         }
5845         return len;
5846 }
5847
5848 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5849 {
5850         const struct quota_format_ops   *ops;
5851
5852         if (!sb_has_quota_loaded(sb, qid->type))
5853                 return -ESRCH;
5854         ops = sb_dqopt(sb)->ops[qid->type];
5855         if (!ops || !ops->get_next_id)
5856                 return -ENOSYS;
5857         return dquot_get_next_id(sb, qid);
5858 }
5859 #endif
5860
5861 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5862                        const char *dev_name, void *data)
5863 {
5864         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5865 }
5866
5867 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5868 static inline void register_as_ext2(void)
5869 {
5870         int err = register_filesystem(&ext2_fs_type);
5871         if (err)
5872                 printk(KERN_WARNING
5873                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5874 }
5875
5876 static inline void unregister_as_ext2(void)
5877 {
5878         unregister_filesystem(&ext2_fs_type);
5879 }
5880
5881 static inline int ext2_feature_set_ok(struct super_block *sb)
5882 {
5883         if (ext4_has_unknown_ext2_incompat_features(sb))
5884                 return 0;
5885         if (sb_rdonly(sb))
5886                 return 1;
5887         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5888                 return 0;
5889         return 1;
5890 }
5891 #else
5892 static inline void register_as_ext2(void) { }
5893 static inline void unregister_as_ext2(void) { }
5894 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5895 #endif
5896
5897 static inline void register_as_ext3(void)
5898 {
5899         int err = register_filesystem(&ext3_fs_type);
5900         if (err)
5901                 printk(KERN_WARNING
5902                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5903 }
5904
5905 static inline void unregister_as_ext3(void)
5906 {
5907         unregister_filesystem(&ext3_fs_type);
5908 }
5909
5910 static inline int ext3_feature_set_ok(struct super_block *sb)
5911 {
5912         if (ext4_has_unknown_ext3_incompat_features(sb))
5913                 return 0;
5914         if (!ext4_has_feature_journal(sb))
5915                 return 0;
5916         if (sb_rdonly(sb))
5917                 return 1;
5918         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5919                 return 0;
5920         return 1;
5921 }
5922
5923 static struct file_system_type ext4_fs_type = {
5924         .owner          = THIS_MODULE,
5925         .name           = "ext4",
5926         .mount          = ext4_mount,
5927         .kill_sb        = kill_block_super,
5928         .fs_flags       = FS_REQUIRES_DEV,
5929 };
5930 MODULE_ALIAS_FS("ext4");
5931
5932 /* Shared across all ext4 file systems */
5933 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5934
5935 static int __init ext4_init_fs(void)
5936 {
5937         int i, err;
5938
5939         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5940         ext4_li_info = NULL;
5941         mutex_init(&ext4_li_mtx);
5942
5943         /* Build-time check for flags consistency */
5944         ext4_check_flag_values();
5945
5946         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5947                 init_waitqueue_head(&ext4__ioend_wq[i]);
5948
5949         err = ext4_init_es();
5950         if (err)
5951                 return err;
5952
5953         err = ext4_init_pageio();
5954         if (err)
5955                 goto out5;
5956
5957         err = ext4_init_system_zone();
5958         if (err)
5959                 goto out4;
5960
5961         err = ext4_init_sysfs();
5962         if (err)
5963                 goto out3;
5964
5965         err = ext4_init_mballoc();
5966         if (err)
5967                 goto out2;
5968         err = init_inodecache();
5969         if (err)
5970                 goto out1;
5971         register_as_ext3();
5972         register_as_ext2();
5973         err = register_filesystem(&ext4_fs_type);
5974         if (err)
5975                 goto out;
5976
5977         return 0;
5978 out:
5979         unregister_as_ext2();
5980         unregister_as_ext3();
5981         destroy_inodecache();
5982 out1:
5983         ext4_exit_mballoc();
5984 out2:
5985         ext4_exit_sysfs();
5986 out3:
5987         ext4_exit_system_zone();
5988 out4:
5989         ext4_exit_pageio();
5990 out5:
5991         ext4_exit_es();
5992
5993         return err;
5994 }
5995
5996 static void __exit ext4_exit_fs(void)
5997 {
5998         ext4_destroy_lazyinit_thread();
5999         unregister_as_ext2();
6000         unregister_as_ext3();
6001         unregister_filesystem(&ext4_fs_type);
6002         destroy_inodecache();
6003         ext4_exit_mballoc();
6004         ext4_exit_sysfs();
6005         ext4_exit_system_zone();
6006         ext4_exit_pageio();
6007         ext4_exit_es();
6008 }
6009
6010 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6011 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6012 MODULE_LICENSE("GPL");
6013 MODULE_SOFTDEP("pre: crc32c");
6014 module_init(ext4_init_fs)
6015 module_exit(ext4_exit_fs)