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
 918 /*
 919  * This is a helper function which is used in the mount/remount
 920  * codepaths (which holds s_umount) to fetch the quota file name.
 921  */
 922 static inline char *get_qf_name(struct super_block *sb,
 923                                 struct ext4_sb_info *sbi,
 924                                 int type)
 925 {
 926         return rcu_dereference_protected(sbi->s_qf_names[type],
 927                                          lockdep_is_held(&sb->s_umount));
 928 }
 929 #else
 930 static inline void ext4_quota_off_umount(struct super_block *sb)
 931 {
 932 }
 933 #endif
 934
 935 static void ext4_put_super(struct super_block *sb)
 936 {
 937         struct ext4_sb_info *sbi = EXT4_SB(sb);
 938         struct ext4_super_block *es = sbi->s_es;
 939         int aborted = 0;
 940         int i, err;
 941
 942         ext4_unregister_li_request(sb);
 943         ext4_quota_off_umount(sb);
 944
 945         destroy_workqueue(sbi->rsv_conversion_wq);
 946
 947         if (sbi->s_journal) {
 948                 aborted = is_journal_aborted(sbi->s_journal);
 949                 err = jbd2_journal_destroy(sbi->s_journal);
 950                 sbi->s_journal = NULL;
 951                 if ((err < 0) && !aborted)
 952                         ext4_abort(sb, "Couldn't clean up the journal");
 953         }
 954
 955         ext4_unregister_sysfs(sb);
 956         ext4_es_unregister_shrinker(sbi);
 957         del_timer_sync(&sbi->s_err_report);
 958         ext4_release_system_zone(sb);
 959         ext4_mb_release(sb);
 960         ext4_ext_release(sb);
 961
 962         if (!sb_rdonly(sb) && !aborted) {
 963                 ext4_clear_feature_journal_needs_recovery(sb);
 964                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 965         }
 966         if (!sb_rdonly(sb))
 967                 ext4_commit_super(sb, 1);
 968
 969         for (i = 0; i < sbi->s_gdb_count; i++)
 970                 brelse(sbi->s_group_desc[i]);
 971         kvfree(sbi->s_group_desc);
 972         kvfree(sbi->s_flex_groups);
 973         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 974         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 975         percpu_counter_destroy(&sbi->s_dirs_counter);
 976         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 977         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
 978 #ifdef CONFIG_QUOTA
 979         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 980                 kfree(get_qf_name(sb, sbi, i));
 981 #endif
 982
 983         /* Debugging code just in case the in-memory inode orphan list
 984          * isn't empty.  The on-disk one can be non-empty if we've
 985          * detected an error and taken the fs readonly, but the
 986          * in-memory list had better be clean by this point. */
 987         if (!list_empty(&sbi->s_orphan))
 988                 dump_orphan_list(sb, sbi);
 989         J_ASSERT(list_empty(&sbi->s_orphan));
 990
 991         sync_blockdev(sb->s_bdev);
 992         invalidate_bdev(sb->s_bdev);
 993         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 994                 /*
 995                  * Invalidate the journal device's buffers.  We don't want them
 996                  * floating about in memory - the physical journal device may
 997                  * hotswapped, and it breaks the `ro-after' testing code.
 998                  */
 999                 sync_blockdev(sbi->journal_bdev);
1000                 invalidate_bdev(sbi->journal_bdev);
1001                 ext4_blkdev_remove(sbi);
1002         }
1003         if (sbi->s_ea_inode_cache) {
1004                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
1005                 sbi->s_ea_inode_cache = NULL;
1006         }
1007         if (sbi->s_ea_block_cache) {
1008                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
1009                 sbi->s_ea_block_cache = NULL;
1010         }
1011         if (sbi->s_mmp_tsk)
1012                 kthread_stop(sbi->s_mmp_tsk);
1013         brelse(sbi->s_sbh);
1014         sb->s_fs_info = NULL;
1015         /*
1016          * Now that we are completely done shutting down the
1017          * superblock, we need to actually destroy the kobject.
1018          */
1019         kobject_put(&sbi->s_kobj);
1020         wait_for_completion(&sbi->s_kobj_unregister);
1021         if (sbi->s_chksum_driver)
1022                 crypto_free_shash(sbi->s_chksum_driver);
1023         kfree(sbi->s_blockgroup_lock);
1024         fs_put_dax(sbi->s_daxdev);
1025         kfree(sbi);
1026 }
1027
1028 static struct kmem_cache *ext4_inode_cachep;
1029
1030 /*
1031  * Called inside transaction, so use GFP_NOFS
1032  */
1033 static struct inode *ext4_alloc_inode(struct super_block *sb)
1034 {
1035         struct ext4_inode_info *ei;
1036
1037         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
1038         if (!ei)
1039                 return NULL;
1040
1041         inode_set_iversion(&ei->vfs_inode, 1);
1042         spin_lock_init(&ei->i_raw_lock);
1043         INIT_LIST_HEAD(&ei->i_prealloc_list);
1044         spin_lock_init(&ei->i_prealloc_lock);
1045         ext4_es_init_tree(&ei->i_es_tree);
1046         rwlock_init(&ei->i_es_lock);
1047         INIT_LIST_HEAD(&ei->i_es_list);
1048         ei->i_es_all_nr = 0;
1049         ei->i_es_shk_nr = 0;
1050         ei->i_es_shrink_lblk = 0;
1051         ei->i_reserved_data_blocks = 0;
1052         ei->i_da_metadata_calc_len = 0;
1053         ei->i_da_metadata_calc_last_lblock = 0;
1054         spin_lock_init(&(ei->i_block_reservation_lock));
1055 #ifdef CONFIG_QUOTA
1056         ei->i_reserved_quota = 0;
1057         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
1058 #endif
1059         ei->jinode = NULL;
1060         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
1061         spin_lock_init(&ei->i_completed_io_lock);
1062         ei->i_sync_tid = 0;
1063         ei->i_datasync_tid = 0;
1064         atomic_set(&ei->i_unwritten, 0);
1065         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
1066         return &ei->vfs_inode;
1067 }
1068
1069 static int ext4_drop_inode(struct inode *inode)
1070 {
1071         int drop = generic_drop_inode(inode);
1072
1073         trace_ext4_drop_inode(inode, drop);
1074         return drop;
1075 }
1076
1077 static void ext4_i_callback(struct rcu_head *head)
1078 {
1079         struct inode *inode = container_of(head, struct inode, i_rcu);
1080         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1081 }
1082
1083 static void ext4_destroy_inode(struct inode *inode)
1084 {
1085         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1086                 ext4_msg(inode->i_sb, KERN_ERR,
1087                          "Inode %lu (%p): orphan list check failed!",
1088                          inode->i_ino, EXT4_I(inode));
1089                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1090                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1091                                 true);
1092                 dump_stack();
1093         }
1094         call_rcu(&inode->i_rcu, ext4_i_callback);
1095 }
1096
1097 static void init_once(void *foo)
1098 {
1099         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1100
1101         INIT_LIST_HEAD(&ei->i_orphan);
1102         init_rwsem(&ei->xattr_sem);
1103         init_rwsem(&ei->i_data_sem);
1104         init_rwsem(&ei->i_mmap_sem);
1105         inode_init_once(&ei->vfs_inode);
1106 }
1107
1108 static int __init init_inodecache(void)
1109 {
1110         ext4_inode_cachep = kmem_cache_create_usercopy("ext4_inode_cache",
1111                                 sizeof(struct ext4_inode_info), 0,
1112                                 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
1113                                         SLAB_ACCOUNT),
1114                                 offsetof(struct ext4_inode_info, i_data),
1115                                 sizeof_field(struct ext4_inode_info, i_data),
1116                                 init_once);
1117         if (ext4_inode_cachep == NULL)
1118                 return -ENOMEM;
1119         return 0;
1120 }
1121
1122 static void destroy_inodecache(void)
1123 {
1124         /*
1125          * Make sure all delayed rcu free inodes are flushed before we
1126          * destroy cache.
1127          */
1128         rcu_barrier();
1129         kmem_cache_destroy(ext4_inode_cachep);
1130 }
1131
1132 void ext4_clear_inode(struct inode *inode)
1133 {
1134         invalidate_inode_buffers(inode);
1135         clear_inode(inode);
1136         dquot_drop(inode);
1137         ext4_discard_preallocations(inode);
1138         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1139         if (EXT4_I(inode)->jinode) {
1140                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1141                                                EXT4_I(inode)->jinode);
1142                 jbd2_free_inode(EXT4_I(inode)->jinode);
1143                 EXT4_I(inode)->jinode = NULL;
1144         }
1145         fscrypt_put_encryption_info(inode);
1146 }
1147
1148 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1149                                         u64 ino, u32 generation)
1150 {
1151         struct inode *inode;
1152
1153         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1154                 return ERR_PTR(-ESTALE);
1155         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1156                 return ERR_PTR(-ESTALE);
1157
1158         /* iget isn't really right if the inode is currently unallocated!!
1159          *
1160          * ext4_read_inode will return a bad_inode if the inode had been
1161          * deleted, so we should be safe.
1162          *
1163          * Currently we don't know the generation for parent directory, so
1164          * a generation of 0 means "accept any"
1165          */
1166         inode = ext4_iget_normal(sb, ino);
1167         if (IS_ERR(inode))
1168                 return ERR_CAST(inode);
1169         if (generation && inode->i_generation != generation) {
1170                 iput(inode);
1171                 return ERR_PTR(-ESTALE);
1172         }
1173
1174         return inode;
1175 }
1176
1177 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1178                                         int fh_len, int fh_type)
1179 {
1180         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1181                                     ext4_nfs_get_inode);
1182 }
1183
1184 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1185                                         int fh_len, int fh_type)
1186 {
1187         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1188                                     ext4_nfs_get_inode);
1189 }
1190
1191 /*
1192  * Try to release metadata pages (indirect blocks, directories) which are
1193  * mapped via the block device.  Since these pages could have journal heads
1194  * which would prevent try_to_free_buffers() from freeing them, we must use
1195  * jbd2 layer's try_to_free_buffers() function to release them.
1196  */
1197 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1198                                  gfp_t wait)
1199 {
1200         journal_t *journal = EXT4_SB(sb)->s_journal;
1201
1202         WARN_ON(PageChecked(page));
1203         if (!page_has_buffers(page))
1204                 return 0;
1205         if (journal)
1206                 return jbd2_journal_try_to_free_buffers(journal, page,
1207                                                 wait & ~__GFP_DIRECT_RECLAIM);
1208         return try_to_free_buffers(page);
1209 }
1210
1211 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1212 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1213 {
1214         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1215                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1216 }
1217
1218 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1219                                                         void *fs_data)
1220 {
1221         handle_t *handle = fs_data;
1222         int res, res2, credits, retries = 0;
1223
1224         /*
1225          * Encrypting the root directory is not allowed because e2fsck expects
1226          * lost+found to exist and be unencrypted, and encrypting the root
1227          * directory would imply encrypting the lost+found directory as well as
1228          * the filename "lost+found" itself.
1229          */
1230         if (inode->i_ino == EXT4_ROOT_INO)
1231                 return -EPERM;
1232
1233         if (WARN_ON_ONCE(IS_DAX(inode) && i_size_read(inode)))
1234                 return -EINVAL;
1235
1236         res = ext4_convert_inline_data(inode);
1237         if (res)
1238                 return res;
1239
1240         /*
1241          * If a journal handle was specified, then the encryption context is
1242          * being set on a new inode via inheritance and is part of a larger
1243          * transaction to create the inode.  Otherwise the encryption context is
1244          * being set on an existing inode in its own transaction.  Only in the
1245          * latter case should the "retry on ENOSPC" logic be used.
1246          */
1247
1248         if (handle) {
1249                 res = ext4_xattr_set_handle(handle, inode,
1250                                             EXT4_XATTR_INDEX_ENCRYPTION,
1251                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1252                                             ctx, len, 0);
1253                 if (!res) {
1254                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1255                         ext4_clear_inode_state(inode,
1256                                         EXT4_STATE_MAY_INLINE_DATA);
1257                         /*
1258                          * Update inode->i_flags - S_ENCRYPTED will be enabled,
1259                          * S_DAX may be disabled
1260                          */
1261                         ext4_set_inode_flags(inode);
1262                 }
1263                 return res;
1264         }
1265
1266         res = dquot_initialize(inode);
1267         if (res)
1268                 return res;
1269 retry:
1270         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1271                                      &credits);
1272         if (res)
1273                 return res;
1274
1275         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1276         if (IS_ERR(handle))
1277                 return PTR_ERR(handle);
1278
1279         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1280                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1281                                     ctx, len, 0);
1282         if (!res) {
1283                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1284                 /*
1285                  * Update inode->i_flags - S_ENCRYPTED will be enabled,
1286                  * S_DAX may be disabled
1287                  */
1288                 ext4_set_inode_flags(inode);
1289                 res = ext4_mark_inode_dirty(handle, inode);
1290                 if (res)
1291                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1292         }
1293         res2 = ext4_journal_stop(handle);
1294
1295         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1296                 goto retry;
1297         if (!res)
1298                 res = res2;
1299         return res;
1300 }
1301
1302 static bool ext4_dummy_context(struct inode *inode)
1303 {
1304         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1305 }
1306
1307 static const struct fscrypt_operations ext4_cryptops = {
1308         .key_prefix             = "ext4:",
1309         .get_context            = ext4_get_context,
1310         .set_context            = ext4_set_context,
1311         .dummy_context          = ext4_dummy_context,
1312         .empty_dir              = ext4_empty_dir,
1313         .max_namelen            = EXT4_NAME_LEN,
1314 };
1315 #endif
1316
1317 #ifdef CONFIG_QUOTA
1318 static const char * const quotatypes[] = INITQFNAMES;
1319 #define QTYPE2NAME(t) (quotatypes[t])
1320
1321 static int ext4_write_dquot(struct dquot *dquot);
1322 static int ext4_acquire_dquot(struct dquot *dquot);
1323 static int ext4_release_dquot(struct dquot *dquot);
1324 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1325 static int ext4_write_info(struct super_block *sb, int type);
1326 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1327                          const struct path *path);
1328 static int ext4_quota_on_mount(struct super_block *sb, int type);
1329 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1330                                size_t len, loff_t off);
1331 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1332                                 const char *data, size_t len, loff_t off);
1333 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1334                              unsigned int flags);
1335 static int ext4_enable_quotas(struct super_block *sb);
1336 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1337
1338 static struct dquot **ext4_get_dquots(struct inode *inode)
1339 {
1340         return EXT4_I(inode)->i_dquot;
1341 }
1342
1343 static const struct dquot_operations ext4_quota_operations = {
1344         .get_reserved_space     = ext4_get_reserved_space,
1345         .write_dquot            = ext4_write_dquot,
1346         .acquire_dquot          = ext4_acquire_dquot,
1347         .release_dquot          = ext4_release_dquot,
1348         .mark_dirty             = ext4_mark_dquot_dirty,
1349         .write_info             = ext4_write_info,
1350         .alloc_dquot            = dquot_alloc,
1351         .destroy_dquot          = dquot_destroy,
1352         .get_projid             = ext4_get_projid,
1353         .get_inode_usage        = ext4_get_inode_usage,
1354         .get_next_id            = ext4_get_next_id,
1355 };
1356
1357 static const struct quotactl_ops ext4_qctl_operations = {
1358         .quota_on       = ext4_quota_on,
1359         .quota_off      = ext4_quota_off,
1360         .quota_sync     = dquot_quota_sync,
1361         .get_state      = dquot_get_state,
1362         .set_info       = dquot_set_dqinfo,
1363         .get_dqblk      = dquot_get_dqblk,
1364         .set_dqblk      = dquot_set_dqblk,
1365         .get_nextdqblk  = dquot_get_next_dqblk,
1366 };
1367 #endif
1368
1369 static const struct super_operations ext4_sops = {
1370         .alloc_inode    = ext4_alloc_inode,
1371         .destroy_inode  = ext4_destroy_inode,
1372         .write_inode    = ext4_write_inode,
1373         .dirty_inode    = ext4_dirty_inode,
1374         .drop_inode     = ext4_drop_inode,
1375         .evict_inode    = ext4_evict_inode,
1376         .put_super      = ext4_put_super,
1377         .sync_fs        = ext4_sync_fs,
1378         .freeze_fs      = ext4_freeze,
1379         .unfreeze_fs    = ext4_unfreeze,
1380         .statfs         = ext4_statfs,
1381         .remount_fs     = ext4_remount,
1382         .show_options   = ext4_show_options,
1383 #ifdef CONFIG_QUOTA
1384         .quota_read     = ext4_quota_read,
1385         .quota_write    = ext4_quota_write,
1386         .get_dquots     = ext4_get_dquots,
1387 #endif
1388         .bdev_try_to_free_page = bdev_try_to_free_page,
1389 };
1390
1391 static const struct export_operations ext4_export_ops = {
1392         .fh_to_dentry = ext4_fh_to_dentry,
1393         .fh_to_parent = ext4_fh_to_parent,
1394         .get_parent = ext4_get_parent,
1395 };
1396
1397 enum {
1398         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1399         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1400         Opt_nouid32, Opt_debug, Opt_removed,
1401         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1402         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1403         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1404         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1405         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1406         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1407         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1408         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1409         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1410         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1411         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
1412         Opt_nowarn_on_error, Opt_mblk_io_submit,
1413         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1414         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1415         Opt_inode_readahead_blks, Opt_journal_ioprio,
1416         Opt_dioread_nolock, Opt_dioread_lock,
1417         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1418         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1419 };
1420
1421 static const match_table_t tokens = {
1422         {Opt_bsd_df, "bsddf"},
1423         {Opt_minix_df, "minixdf"},
1424         {Opt_grpid, "grpid"},
1425         {Opt_grpid, "bsdgroups"},
1426         {Opt_nogrpid, "nogrpid"},
1427         {Opt_nogrpid, "sysvgroups"},
1428         {Opt_resgid, "resgid=%u"},
1429         {Opt_resuid, "resuid=%u"},
1430         {Opt_sb, "sb=%u"},
1431         {Opt_err_cont, "errors=continue"},
1432         {Opt_err_panic, "errors=panic"},
1433         {Opt_err_ro, "errors=remount-ro"},
1434         {Opt_nouid32, "nouid32"},
1435         {Opt_debug, "debug"},
1436         {Opt_removed, "oldalloc"},
1437         {Opt_removed, "orlov"},
1438         {Opt_user_xattr, "user_xattr"},
1439         {Opt_nouser_xattr, "nouser_xattr"},
1440         {Opt_acl, "acl"},
1441         {Opt_noacl, "noacl"},
1442         {Opt_noload, "norecovery"},
1443         {Opt_noload, "noload"},
1444         {Opt_removed, "nobh"},
1445         {Opt_removed, "bh"},
1446         {Opt_commit, "commit=%u"},
1447         {Opt_min_batch_time, "min_batch_time=%u"},
1448         {Opt_max_batch_time, "max_batch_time=%u"},
1449         {Opt_journal_dev, "journal_dev=%u"},
1450         {Opt_journal_path, "journal_path=%s"},
1451         {Opt_journal_checksum, "journal_checksum"},
1452         {Opt_nojournal_checksum, "nojournal_checksum"},
1453         {Opt_journal_async_commit, "journal_async_commit"},
1454         {Opt_abort, "abort"},
1455         {Opt_data_journal, "data=journal"},
1456         {Opt_data_ordered, "data=ordered"},
1457         {Opt_data_writeback, "data=writeback"},
1458         {Opt_data_err_abort, "data_err=abort"},
1459         {Opt_data_err_ignore, "data_err=ignore"},
1460         {Opt_offusrjquota, "usrjquota="},
1461         {Opt_usrjquota, "usrjquota=%s"},
1462         {Opt_offgrpjquota, "grpjquota="},
1463         {Opt_grpjquota, "grpjquota=%s"},
1464         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1465         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1466         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1467         {Opt_grpquota, "grpquota"},
1468         {Opt_noquota, "noquota"},
1469         {Opt_quota, "quota"},
1470         {Opt_usrquota, "usrquota"},
1471         {Opt_prjquota, "prjquota"},
1472         {Opt_barrier, "barrier=%u"},
1473         {Opt_barrier, "barrier"},
1474         {Opt_nobarrier, "nobarrier"},
1475         {Opt_i_version, "i_version"},
1476         {Opt_dax, "dax"},
1477         {Opt_stripe, "stripe=%u"},
1478         {Opt_delalloc, "delalloc"},
1479         {Opt_warn_on_error, "warn_on_error"},
1480         {Opt_nowarn_on_error, "nowarn_on_error"},
1481         {Opt_lazytime, "lazytime"},
1482         {Opt_nolazytime, "nolazytime"},
1483         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1484         {Opt_nodelalloc, "nodelalloc"},
1485         {Opt_removed, "mblk_io_submit"},
1486         {Opt_removed, "nomblk_io_submit"},
1487         {Opt_block_validity, "block_validity"},
1488         {Opt_noblock_validity, "noblock_validity"},
1489         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1490         {Opt_journal_ioprio, "journal_ioprio=%u"},
1491         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1492         {Opt_auto_da_alloc, "auto_da_alloc"},
1493         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1494         {Opt_dioread_nolock, "dioread_nolock"},
1495         {Opt_dioread_lock, "dioread_lock"},
1496         {Opt_discard, "discard"},
1497         {Opt_nodiscard, "nodiscard"},
1498         {Opt_init_itable, "init_itable=%u"},
1499         {Opt_init_itable, "init_itable"},
1500         {Opt_noinit_itable, "noinit_itable"},
1501         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1502         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1503         {Opt_nombcache, "nombcache"},
1504         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1505         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1506         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1507         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1508         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1509         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1510         {Opt_err, NULL},
1511 };
1512
1513 static ext4_fsblk_t get_sb_block(void **data)
1514 {
1515         ext4_fsblk_t    sb_block;
1516         char            *options = (char *) *data;
1517
1518         if (!options || strncmp(options, "sb=", 3) != 0)
1519                 return 1;       /* Default location */
1520
1521         options += 3;
1522         /* TODO: use simple_strtoll with >32bit ext4 */
1523         sb_block = simple_strtoul(options, &options, 0);
1524         if (*options && *options != ',') {
1525                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1526                        (char *) *data);
1527                 return 1;
1528         }
1529         if (*options == ',')
1530                 options++;
1531         *data = (void *) options;
1532
1533         return sb_block;
1534 }
1535
1536 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1537 static const char deprecated_msg[] =
1538         "Mount option \"%s\" will be removed by %s\n"
1539         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1540
1541 #ifdef CONFIG_QUOTA
1542 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1543 {
1544         struct ext4_sb_info *sbi = EXT4_SB(sb);
1545         char *qname, *old_qname = get_qf_name(sb, sbi, qtype);
1546         int ret = -1;
1547
1548         if (sb_any_quota_loaded(sb) && !old_qname) {
1549                 ext4_msg(sb, KERN_ERR,
1550                         "Cannot change journaled "
1551                         "quota options when quota turned on");
1552                 return -1;
1553         }
1554         if (ext4_has_feature_quota(sb)) {
1555                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1556                          "ignored when QUOTA feature is enabled");
1557                 return 1;
1558         }
1559         qname = match_strdup(args);
1560         if (!qname) {
1561                 ext4_msg(sb, KERN_ERR,
1562                         "Not enough memory for storing quotafile name");
1563                 return -1;
1564         }
1565         if (old_qname) {
1566                 if (strcmp(old_qname, qname) == 0)
1567                         ret = 1;
1568                 else
1569                         ext4_msg(sb, KERN_ERR,
1570                                  "%s quota file already specified",
1571                                  QTYPE2NAME(qtype));
1572                 goto errout;
1573         }
1574         if (strchr(qname, '/')) {
1575                 ext4_msg(sb, KERN_ERR,
1576                         "quotafile must be on filesystem root");
1577                 goto errout;
1578         }
1579         rcu_assign_pointer(sbi->s_qf_names[qtype], qname);
1580         set_opt(sb, QUOTA);
1581         return 1;
1582 errout:
1583         kfree(qname);
1584         return ret;
1585 }
1586
1587 static int clear_qf_name(struct super_block *sb, int qtype)
1588 {
1589
1590         struct ext4_sb_info *sbi = EXT4_SB(sb);
1591         char *old_qname = get_qf_name(sb, sbi, qtype);
1592
1593         if (sb_any_quota_loaded(sb) && old_qname) {
1594                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1595                         " when quota turned on");
1596                 return -1;
1597         }
1598         rcu_assign_pointer(sbi->s_qf_names[qtype], NULL);
1599         synchronize_rcu();
1600         kfree(old_qname);
1601         return 1;
1602 }
1603 #endif
1604
1605 #define MOPT_SET        0x0001
1606 #define MOPT_CLEAR      0x0002
1607 #define MOPT_NOSUPPORT  0x0004
1608 #define MOPT_EXPLICIT   0x0008
1609 #define MOPT_CLEAR_ERR  0x0010
1610 #define MOPT_GTE0       0x0020
1611 #ifdef CONFIG_QUOTA
1612 #define MOPT_Q          0
1613 #define MOPT_QFMT       0x0040
1614 #else
1615 #define MOPT_Q          MOPT_NOSUPPORT
1616 #define MOPT_QFMT       MOPT_NOSUPPORT
1617 #endif
1618 #define MOPT_DATAJ      0x0080
1619 #define MOPT_NO_EXT2    0x0100
1620 #define MOPT_NO_EXT3    0x0200
1621 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1622 #define MOPT_STRING     0x0400
1623
1624 static const struct mount_opts {
1625         int     token;
1626         int     mount_opt;
1627         int     flags;
1628 } ext4_mount_opts[] = {
1629         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1630         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1631         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1632         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1633         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1634         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1635         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1636          MOPT_EXT4_ONLY | MOPT_SET},
1637         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1638          MOPT_EXT4_ONLY | MOPT_CLEAR},
1639         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1640         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1641         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1642          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1643         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1644          MOPT_EXT4_ONLY | MOPT_CLEAR},
1645         {Opt_warn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_SET},
1646         {Opt_nowarn_on_error, EXT4_MOUNT_WARN_ON_ERROR, MOPT_CLEAR},
1647         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1648          MOPT_EXT4_ONLY | MOPT_CLEAR},
1649         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1650          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1651         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1652                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1653          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1654         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1655         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1656         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1657         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1658         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1659          MOPT_NO_EXT2},
1660         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1661          MOPT_NO_EXT2},
1662         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1663         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1664         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1665         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1666         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1667         {Opt_commit, 0, MOPT_GTE0},
1668         {Opt_max_batch_time, 0, MOPT_GTE0},
1669         {Opt_min_batch_time, 0, MOPT_GTE0},
1670         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1671         {Opt_init_itable, 0, MOPT_GTE0},
1672         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1673         {Opt_stripe, 0, MOPT_GTE0},
1674         {Opt_resuid, 0, MOPT_GTE0},
1675         {Opt_resgid, 0, MOPT_GTE0},
1676         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1677         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1678         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1679         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1680         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1681         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1682          MOPT_NO_EXT2 | MOPT_DATAJ},
1683         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1684         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1685 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1686         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1687         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1688 #else
1689         {Opt_acl, 0, MOPT_NOSUPPORT},
1690         {Opt_noacl, 0, MOPT_NOSUPPORT},
1691 #endif
1692         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1693         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1694         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1695         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1696         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1697                                                         MOPT_SET | MOPT_Q},
1698         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1699                                                         MOPT_SET | MOPT_Q},
1700         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1701                                                         MOPT_SET | MOPT_Q},
1702         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1703                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1704                                                         MOPT_CLEAR | MOPT_Q},
1705         {Opt_usrjquota, 0, MOPT_Q},
1706         {Opt_grpjquota, 0, MOPT_Q},
1707         {Opt_offusrjquota, 0, MOPT_Q},
1708         {Opt_offgrpjquota, 0, MOPT_Q},
1709         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1710         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1711         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1712         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1713         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1714         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1715         {Opt_err, 0, 0}
1716 };
1717
1718 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1719                             substring_t *args, unsigned long *journal_devnum,
1720                             unsigned int *journal_ioprio, int is_remount)
1721 {
1722         struct ext4_sb_info *sbi = EXT4_SB(sb);
1723         const struct mount_opts *m;
1724         kuid_t uid;
1725         kgid_t gid;
1726         int arg = 0;
1727
1728 #ifdef CONFIG_QUOTA
1729         if (token == Opt_usrjquota)
1730                 return set_qf_name(sb, USRQUOTA, &args[0]);
1731         else if (token == Opt_grpjquota)
1732                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1733         else if (token == Opt_offusrjquota)
1734                 return clear_qf_name(sb, USRQUOTA);
1735         else if (token == Opt_offgrpjquota)
1736                 return clear_qf_name(sb, GRPQUOTA);
1737 #endif
1738         switch (token) {
1739         case Opt_noacl:
1740         case Opt_nouser_xattr:
1741                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1742                 break;
1743         case Opt_sb:
1744                 return 1;       /* handled by get_sb_block() */
1745         case Opt_removed:
1746                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1747                 return 1;
1748         case Opt_abort:
1749                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1750                 return 1;
1751         case Opt_i_version:
1752                 sb->s_flags |= SB_I_VERSION;
1753                 return 1;
1754         case Opt_lazytime:
1755                 sb->s_flags |= SB_LAZYTIME;
1756                 return 1;
1757         case Opt_nolazytime:
1758                 sb->s_flags &= ~SB_LAZYTIME;
1759                 return 1;
1760         }
1761
1762         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1763                 if (token == m->token)
1764                         break;
1765
1766         if (m->token == Opt_err) {
1767                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1768                          "or missing value", opt);
1769                 return -1;
1770         }
1771
1772         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1773                 ext4_msg(sb, KERN_ERR,
1774                          "Mount option \"%s\" incompatible with ext2", opt);
1775                 return -1;
1776         }
1777         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1778                 ext4_msg(sb, KERN_ERR,
1779                          "Mount option \"%s\" incompatible with ext3", opt);
1780                 return -1;
1781         }
1782
1783         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1784                 return -1;
1785         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1786                 return -1;
1787         if (m->flags & MOPT_EXPLICIT) {
1788                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1789                         set_opt2(sb, EXPLICIT_DELALLOC);
1790                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1791                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1792                 } else
1793                         return -1;
1794         }
1795         if (m->flags & MOPT_CLEAR_ERR)
1796                 clear_opt(sb, ERRORS_MASK);
1797         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1798                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1799                          "options when quota turned on");
1800                 return -1;
1801         }
1802
1803         if (m->flags & MOPT_NOSUPPORT) {
1804                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1805         } else if (token == Opt_commit) {
1806                 if (arg == 0)
1807                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1808                 sbi->s_commit_interval = HZ * arg;
1809         } else if (token == Opt_debug_want_extra_isize) {
1810                 sbi->s_want_extra_isize = arg;
1811         } else if (token == Opt_max_batch_time) {
1812                 sbi->s_max_batch_time = arg;
1813         } else if (token == Opt_min_batch_time) {
1814                 sbi->s_min_batch_time = arg;
1815         } else if (token == Opt_inode_readahead_blks) {
1816                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1817                         ext4_msg(sb, KERN_ERR,
1818                                  "EXT4-fs: inode_readahead_blks must be "
1819                                  "0 or a power of 2 smaller than 2^31");
1820                         return -1;
1821                 }
1822                 sbi->s_inode_readahead_blks = arg;
1823         } else if (token == Opt_init_itable) {
1824                 set_opt(sb, INIT_INODE_TABLE);
1825                 if (!args->from)
1826                         arg = EXT4_DEF_LI_WAIT_MULT;
1827                 sbi->s_li_wait_mult = arg;
1828         } else if (token == Opt_max_dir_size_kb) {
1829                 sbi->s_max_dir_size_kb = arg;
1830         } else if (token == Opt_stripe) {
1831                 sbi->s_stripe = arg;
1832         } else if (token == Opt_resuid) {
1833                 uid = make_kuid(current_user_ns(), arg);
1834                 if (!uid_valid(uid)) {
1835                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1836                         return -1;
1837                 }
1838                 sbi->s_resuid = uid;
1839         } else if (token == Opt_resgid) {
1840                 gid = make_kgid(current_user_ns(), arg);
1841                 if (!gid_valid(gid)) {
1842                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1843                         return -1;
1844                 }
1845                 sbi->s_resgid = gid;
1846         } else if (token == Opt_journal_dev) {
1847                 if (is_remount) {
1848                         ext4_msg(sb, KERN_ERR,
1849                                  "Cannot specify journal on remount");
1850                         return -1;
1851                 }
1852                 *journal_devnum = arg;
1853         } else if (token == Opt_journal_path) {
1854                 char *journal_path;
1855                 struct inode *journal_inode;
1856                 struct path path;
1857                 int error;
1858
1859                 if (is_remount) {
1860                         ext4_msg(sb, KERN_ERR,
1861                                  "Cannot specify journal on remount");
1862                         return -1;
1863                 }
1864                 journal_path = match_strdup(&args[0]);
1865                 if (!journal_path) {
1866                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1867                                 "journal device string");
1868                         return -1;
1869                 }
1870
1871                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1872                 if (error) {
1873                         ext4_msg(sb, KERN_ERR, "error: could not find "
1874                                 "journal device path: error %d", error);
1875                         kfree(journal_path);
1876                         return -1;
1877                 }
1878
1879                 journal_inode = d_inode(path.dentry);
1880                 if (!S_ISBLK(journal_inode->i_mode)) {
1881                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1882                                 "is not a block device", journal_path);
1883                         path_put(&path);
1884                         kfree(journal_path);
1885                         return -1;
1886                 }
1887
1888                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1889                 path_put(&path);
1890                 kfree(journal_path);
1891         } else if (token == Opt_journal_ioprio) {
1892                 if (arg > 7) {
1893                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1894                                  " (must be 0-7)");
1895                         return -1;
1896                 }
1897                 *journal_ioprio =
1898                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1899         } else if (token == Opt_test_dummy_encryption) {
1900 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1901                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1902                 ext4_msg(sb, KERN_WARNING,
1903                          "Test dummy encryption mode enabled");
1904 #else
1905                 ext4_msg(sb, KERN_WARNING,
1906                          "Test dummy encryption mount option ignored");
1907 #endif
1908         } else if (m->flags & MOPT_DATAJ) {
1909                 if (is_remount) {
1910                         if (!sbi->s_journal)
1911                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1912                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1913                                 ext4_msg(sb, KERN_ERR,
1914                                          "Cannot change data mode on remount");
1915                                 return -1;
1916                         }
1917                 } else {
1918                         clear_opt(sb, DATA_FLAGS);
1919                         sbi->s_mount_opt |= m->mount_opt;
1920                 }
1921 #ifdef CONFIG_QUOTA
1922         } else if (m->flags & MOPT_QFMT) {
1923                 if (sb_any_quota_loaded(sb) &&
1924                     sbi->s_jquota_fmt != m->mount_opt) {
1925                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1926                                  "quota options when quota turned on");
1927                         return -1;
1928                 }
1929                 if (ext4_has_feature_quota(sb)) {
1930                         ext4_msg(sb, KERN_INFO,
1931                                  "Quota format mount options ignored "
1932                                  "when QUOTA feature is enabled");
1933                         return 1;
1934                 }
1935                 sbi->s_jquota_fmt = m->mount_opt;
1936 #endif
1937         } else if (token == Opt_dax) {
1938 #ifdef CONFIG_FS_DAX
1939                 ext4_msg(sb, KERN_WARNING,
1940                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1941                         sbi->s_mount_opt |= m->mount_opt;
1942 #else
1943                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1944                 return -1;
1945 #endif
1946         } else if (token == Opt_data_err_abort) {
1947                 sbi->s_mount_opt |= m->mount_opt;
1948         } else if (token == Opt_data_err_ignore) {
1949                 sbi->s_mount_opt &= ~m->mount_opt;
1950         } else {
1951                 if (!args->from)
1952                         arg = 1;
1953                 if (m->flags & MOPT_CLEAR)
1954                         arg = !arg;
1955                 else if (unlikely(!(m->flags & MOPT_SET))) {
1956                         ext4_msg(sb, KERN_WARNING,
1957                                  "buggy handling of option %s", opt);
1958                         WARN_ON(1);
1959                         return -1;
1960                 }
1961                 if (arg != 0)
1962                         sbi->s_mount_opt |= m->mount_opt;
1963                 else
1964                         sbi->s_mount_opt &= ~m->mount_opt;
1965         }
1966         return 1;
1967 }
1968
1969 static int parse_options(char *options, struct super_block *sb,
1970                          unsigned long *journal_devnum,
1971                          unsigned int *journal_ioprio,
1972                          int is_remount)
1973 {
1974         struct ext4_sb_info *sbi = EXT4_SB(sb);
1975         char *p, __maybe_unused *usr_qf_name, __maybe_unused *grp_qf_name;
1976         substring_t args[MAX_OPT_ARGS];
1977         int token;
1978
1979         if (!options)
1980                 return 1;
1981
1982         while ((p = strsep(&options, ",")) != NULL) {
1983                 if (!*p)
1984                         continue;
1985                 /*
1986                  * Initialize args struct so we know whether arg was
1987                  * found; some options take optional arguments.
1988                  */
1989                 args[0].to = args[0].from = NULL;
1990                 token = match_token(p, tokens, args);
1991                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1992                                      journal_ioprio, is_remount) < 0)
1993                         return 0;
1994         }
1995 #ifdef CONFIG_QUOTA
1996         /*
1997          * We do the test below only for project quotas. 'usrquota' and
1998          * 'grpquota' mount options are allowed even without quota feature
1999          * to support legacy quotas in quota files.
2000          */
2001         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
2002                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
2003                          "Cannot enable project quota enforcement.");
2004                 return 0;
2005         }
2006         usr_qf_name = get_qf_name(sb, sbi, USRQUOTA);
2007         grp_qf_name = get_qf_name(sb, sbi, GRPQUOTA);
2008         if (usr_qf_name || grp_qf_name) {
2009                 if (test_opt(sb, USRQUOTA) && usr_qf_name)
2010                         clear_opt(sb, USRQUOTA);
2011
2012                 if (test_opt(sb, GRPQUOTA) && grp_qf_name)
2013                         clear_opt(sb, GRPQUOTA);
2014
2015                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
2016                         ext4_msg(sb, KERN_ERR, "old and new quota "
2017                                         "format mixing");
2018                         return 0;
2019                 }
2020
2021                 if (!sbi->s_jquota_fmt) {
2022                         ext4_msg(sb, KERN_ERR, "journaled quota format "
2023                                         "not specified");
2024                         return 0;
2025                 }
2026         }
2027 #endif
2028         if (test_opt(sb, DIOREAD_NOLOCK)) {
2029                 int blocksize =
2030                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
2031
2032                 if (blocksize < PAGE_SIZE) {
2033                         ext4_msg(sb, KERN_ERR, "can't mount with "
2034                                  "dioread_nolock if block size != PAGE_SIZE");
2035                         return 0;
2036                 }
2037         }
2038         return 1;
2039 }
2040
2041 static inline void ext4_show_quota_options(struct seq_file *seq,
2042                                            struct super_block *sb)
2043 {
2044 #if defined(CONFIG_QUOTA)
2045         struct ext4_sb_info *sbi = EXT4_SB(sb);
2046         char *usr_qf_name, *grp_qf_name;
2047
2048         if (sbi->s_jquota_fmt) {
2049                 char *fmtname = "";
2050
2051                 switch (sbi->s_jquota_fmt) {
2052                 case QFMT_VFS_OLD:
2053                         fmtname = "vfsold";
2054                         break;
2055                 case QFMT_VFS_V0:
2056                         fmtname = "vfsv0";
2057                         break;
2058                 case QFMT_VFS_V1:
2059                         fmtname = "vfsv1";
2060                         break;
2061                 }
2062                 seq_printf(seq, ",jqfmt=%s", fmtname);
2063         }
2064
2065         rcu_read_lock();
2066         usr_qf_name = rcu_dereference(sbi->s_qf_names[USRQUOTA]);
2067         grp_qf_name = rcu_dereference(sbi->s_qf_names[GRPQUOTA]);
2068         if (usr_qf_name)
2069                 seq_show_option(seq, "usrjquota", usr_qf_name);
2070         if (grp_qf_name)
2071                 seq_show_option(seq, "grpjquota", grp_qf_name);
2072         rcu_read_unlock();
2073 #endif
2074 }
2075
2076 static const char *token2str(int token)
2077 {
2078         const struct match_token *t;
2079
2080         for (t = tokens; t->token != Opt_err; t++)
2081                 if (t->token == token && !strchr(t->pattern, '='))
2082                         break;
2083         return t->pattern;
2084 }
2085
2086 /*
2087  * Show an option if
2088  *  - it's set to a non-default value OR
2089  *  - if the per-sb default is different from the global default
2090  */
2091 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2092                               int nodefs)
2093 {
2094         struct ext4_sb_info *sbi = EXT4_SB(sb);
2095         struct ext4_super_block *es = sbi->s_es;
2096         int def_errors, def_mount_opt = sbi->s_def_mount_opt;
2097         const struct mount_opts *m;
2098         char sep = nodefs ? '\n' : ',';
2099
2100 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2101 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2102
2103         if (sbi->s_sb_block != 1)
2104                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2105
2106         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2107                 int want_set = m->flags & MOPT_SET;
2108                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2109                     (m->flags & MOPT_CLEAR_ERR))
2110                         continue;
2111                 if (!nodefs && !(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2112                         continue; /* skip if same as the default */
2113                 if ((want_set &&
2114                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2115                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2116                         continue; /* select Opt_noFoo vs Opt_Foo */
2117                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2118         }
2119
2120         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2121             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2122                 SEQ_OPTS_PRINT("resuid=%u",
2123                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2124         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2125             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2126                 SEQ_OPTS_PRINT("resgid=%u",
2127                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2128         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2129         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2130                 SEQ_OPTS_PUTS("errors=remount-ro");
2131         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2132                 SEQ_OPTS_PUTS("errors=continue");
2133         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2134                 SEQ_OPTS_PUTS("errors=panic");
2135         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2136                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2137         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2138                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2139         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2140                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2141         if (sb->s_flags & SB_I_VERSION)
2142                 SEQ_OPTS_PUTS("i_version");
2143         if (nodefs || sbi->s_stripe)
2144                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2145         if (nodefs || EXT4_MOUNT_DATA_FLAGS &
2146                         (sbi->s_mount_opt ^ def_mount_opt)) {
2147                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2148                         SEQ_OPTS_PUTS("data=journal");
2149                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2150                         SEQ_OPTS_PUTS("data=ordered");
2151                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2152                         SEQ_OPTS_PUTS("data=writeback");
2153         }
2154         if (nodefs ||
2155             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2156                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2157                                sbi->s_inode_readahead_blks);
2158
2159         if (test_opt(sb, INIT_INODE_TABLE) && (nodefs ||
2160                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2161                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2162         if (nodefs || sbi->s_max_dir_size_kb)
2163                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2164         if (test_opt(sb, DATA_ERR_ABORT))
2165                 SEQ_OPTS_PUTS("data_err=abort");
2166         if (DUMMY_ENCRYPTION_ENABLED(sbi))
2167                 SEQ_OPTS_PUTS("test_dummy_encryption");
2168
2169         ext4_show_quota_options(seq, sb);
2170         return 0;
2171 }
2172
2173 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2174 {
2175         return _ext4_show_options(seq, root->d_sb, 0);
2176 }
2177
2178 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2179 {
2180         struct super_block *sb = seq->private;
2181         int rc;
2182
2183         seq_puts(seq, sb_rdonly(sb) ? "ro" : "rw");
2184         rc = _ext4_show_options(seq, sb, 1);
2185         seq_puts(seq, "\n");
2186         return rc;
2187 }
2188
2189 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2190                             int read_only)
2191 {
2192         struct ext4_sb_info *sbi = EXT4_SB(sb);
2193         int err = 0;
2194
2195         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2196                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2197                          "forcing read-only mode");
2198                 err = -EROFS;
2199         }
2200         if (read_only)
2201                 goto done;
2202         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2203                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2204                          "running e2fsck is recommended");
2205         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2206                 ext4_msg(sb, KERN_WARNING,
2207                          "warning: mounting fs with errors, "
2208                          "running e2fsck is recommended");
2209         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2210                  le16_to_cpu(es->s_mnt_count) >=
2211                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2212                 ext4_msg(sb, KERN_WARNING,
2213                          "warning: maximal mount count reached, "
2214                          "running e2fsck is recommended");
2215         else if (le32_to_cpu(es->s_checkinterval) &&
2216                  (ext4_get_tstamp(es, s_lastcheck) +
2217                   le32_to_cpu(es->s_checkinterval) <= ktime_get_real_seconds()))
2218                 ext4_msg(sb, KERN_WARNING,
2219                          "warning: checktime reached, "
2220                          "running e2fsck is recommended");
2221         if (!sbi->s_journal)
2222                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2223         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2224                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2225         le16_add_cpu(&es->s_mnt_count, 1);
2226         ext4_update_tstamp(es, s_mtime);
2227         ext4_update_dynamic_rev(sb);
2228         if (sbi->s_journal)
2229                 ext4_set_feature_journal_needs_recovery(sb);
2230
2231         err = ext4_commit_super(sb, 1);
2232 done:
2233         if (test_opt(sb, DEBUG))
2234                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2235                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2236                         sb->s_blocksize,
2237                         sbi->s_groups_count,
2238                         EXT4_BLOCKS_PER_GROUP(sb),
2239                         EXT4_INODES_PER_GROUP(sb),
2240                         sbi->s_mount_opt, sbi->s_mount_opt2);
2241
2242         cleancache_init_fs(sb);
2243         return err;
2244 }
2245
2246 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2247 {
2248         struct ext4_sb_info *sbi = EXT4_SB(sb);
2249         struct flex_groups *new_groups;
2250         int size;
2251
2252         if (!sbi->s_log_groups_per_flex)
2253                 return 0;
2254
2255         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2256         if (size <= sbi->s_flex_groups_allocated)
2257                 return 0;
2258
2259         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2260         new_groups = kvzalloc(size, GFP_KERNEL);
2261         if (!new_groups) {
2262                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2263                          size / (int) sizeof(struct flex_groups));
2264                 return -ENOMEM;
2265         }
2266
2267         if (sbi->s_flex_groups) {
2268                 memcpy(new_groups, sbi->s_flex_groups,
2269                        (sbi->s_flex_groups_allocated *
2270                         sizeof(struct flex_groups)));
2271                 kvfree(sbi->s_flex_groups);
2272         }
2273         sbi->s_flex_groups = new_groups;
2274         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2275         return 0;
2276 }
2277
2278 static int ext4_fill_flex_info(struct super_block *sb)
2279 {
2280         struct ext4_sb_info *sbi = EXT4_SB(sb);
2281         struct ext4_group_desc *gdp = NULL;
2282         ext4_group_t flex_group;
2283         int i, err;
2284
2285         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2286         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2287                 sbi->s_log_groups_per_flex = 0;
2288                 return 1;
2289         }
2290
2291         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2292         if (err)
2293                 goto failed;
2294
2295         for (i = 0; i < sbi->s_groups_count; i++) {
2296                 gdp = ext4_get_group_desc(sb, i, NULL);
2297
2298                 flex_group = ext4_flex_group(sbi, i);
2299                 atomic_add(ext4_free_inodes_count(sb, gdp),
2300                            &sbi->s_flex_groups[flex_group].free_inodes);
2301                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2302                              &sbi->s_flex_groups[flex_group].free_clusters);
2303                 atomic_add(ext4_used_dirs_count(sb, gdp),
2304                            &sbi->s_flex_groups[flex_group].used_dirs);
2305         }
2306
2307         return 1;
2308 failed:
2309         return 0;
2310 }
2311
2312 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2313                                    struct ext4_group_desc *gdp)
2314 {
2315         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2316         __u16 crc = 0;
2317         __le32 le_group = cpu_to_le32(block_group);
2318         struct ext4_sb_info *sbi = EXT4_SB(sb);
2319
2320         if (ext4_has_metadata_csum(sbi->s_sb)) {
2321                 /* Use new metadata_csum algorithm */
2322                 __u32 csum32;
2323                 __u16 dummy_csum = 0;
2324
2325                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2326                                      sizeof(le_group));
2327                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2328                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2329                                      sizeof(dummy_csum));
2330                 offset += sizeof(dummy_csum);
2331                 if (offset < sbi->s_desc_size)
2332                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2333                                              sbi->s_desc_size - offset);
2334
2335                 crc = csum32 & 0xFFFF;
2336                 goto out;
2337         }
2338
2339         /* old crc16 code */
2340         if (!ext4_has_feature_gdt_csum(sb))
2341                 return 0;
2342
2343         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2344         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2345         crc = crc16(crc, (__u8 *)gdp, offset);
2346         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2347         /* for checksum of struct ext4_group_desc do the rest...*/
2348         if (ext4_has_feature_64bit(sb) &&
2349             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2350                 crc = crc16(crc, (__u8 *)gdp + offset,
2351                             le16_to_cpu(sbi->s_es->s_desc_size) -
2352                                 offset);
2353
2354 out:
2355         return cpu_to_le16(crc);
2356 }
2357
2358 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2359                                 struct ext4_group_desc *gdp)
2360 {
2361         if (ext4_has_group_desc_csum(sb) &&
2362             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2363                 return 0;
2364
2365         return 1;
2366 }
2367
2368 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2369                               struct ext4_group_desc *gdp)
2370 {
2371         if (!ext4_has_group_desc_csum(sb))
2372                 return;
2373         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2374 }
2375
2376 /* Called at mount-time, super-block is locked */
2377 static int ext4_check_descriptors(struct super_block *sb,
2378                                   ext4_fsblk_t sb_block,
2379                                   ext4_group_t *first_not_zeroed)
2380 {
2381         struct ext4_sb_info *sbi = EXT4_SB(sb);
2382         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2383         ext4_fsblk_t last_block;
2384         ext4_fsblk_t last_bg_block = sb_block + ext4_bg_num_gdb(sb, 0);
2385         ext4_fsblk_t block_bitmap;
2386         ext4_fsblk_t inode_bitmap;
2387         ext4_fsblk_t inode_table;
2388         int flexbg_flag = 0;
2389         ext4_group_t i, grp = sbi->s_groups_count;
2390
2391         if (ext4_has_feature_flex_bg(sb))
2392                 flexbg_flag = 1;
2393
2394         ext4_debug("Checking group descriptors");
2395
2396         for (i = 0; i < sbi->s_groups_count; i++) {
2397                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2398
2399                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2400                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2401                 else
2402                         last_block = first_block +
2403                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2404
2405                 if ((grp == sbi->s_groups_count) &&
2406                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2407                         grp = i;
2408
2409                 block_bitmap = ext4_block_bitmap(sb, gdp);
2410                 if (block_bitmap == sb_block) {
2411                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2412                                  "Block bitmap for group %u overlaps "
2413                                  "superblock", i);
2414                         if (!sb_rdonly(sb))
2415                                 return 0;
2416                 }
2417                 if (block_bitmap >= sb_block + 1 &&
2418                     block_bitmap <= last_bg_block) {
2419                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2420                                  "Block bitmap for group %u overlaps "
2421                                  "block group descriptors", i);
2422                         if (!sb_rdonly(sb))
2423                                 return 0;
2424                 }
2425                 if (block_bitmap < first_block || block_bitmap > last_block) {
2426                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2427                                "Block bitmap for group %u not in group "
2428                                "(block %llu)!", i, block_bitmap);
2429                         return 0;
2430                 }
2431                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2432                 if (inode_bitmap == sb_block) {
2433                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2434                                  "Inode bitmap for group %u overlaps "
2435                                  "superblock", i);
2436                         if (!sb_rdonly(sb))
2437                                 return 0;
2438                 }
2439                 if (inode_bitmap >= sb_block + 1 &&
2440                     inode_bitmap <= last_bg_block) {
2441                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2442                                  "Inode bitmap for group %u overlaps "
2443                                  "block group descriptors", i);
2444                         if (!sb_rdonly(sb))
2445                                 return 0;
2446                 }
2447                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2448                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2449                                "Inode bitmap for group %u not in group "
2450                                "(block %llu)!", i, inode_bitmap);
2451                         return 0;
2452                 }
2453                 inode_table = ext4_inode_table(sb, gdp);
2454                 if (inode_table == sb_block) {
2455                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2456                                  "Inode table for group %u overlaps "
2457                                  "superblock", i);
2458                         if (!sb_rdonly(sb))
2459                                 return 0;
2460                 }
2461                 if (inode_table >= sb_block + 1 &&
2462                     inode_table <= last_bg_block) {
2463                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2464                                  "Inode table for group %u overlaps "
2465                                  "block group descriptors", i);
2466                         if (!sb_rdonly(sb))
2467                                 return 0;
2468                 }
2469                 if (inode_table < first_block ||
2470                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2471                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2472                                "Inode table for group %u not in group "
2473                                "(block %llu)!", i, inode_table);
2474                         return 0;
2475                 }
2476                 ext4_lock_group(sb, i);
2477                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2478                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2479                                  "Checksum for group %u failed (%u!=%u)",
2480                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2481                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2482                         if (!sb_rdonly(sb)) {
2483                                 ext4_unlock_group(sb, i);
2484                                 return 0;
2485                         }
2486                 }
2487                 ext4_unlock_group(sb, i);
2488                 if (!flexbg_flag)
2489                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2490         }
2491         if (NULL != first_not_zeroed)
2492                 *first_not_zeroed = grp;
2493         return 1;
2494 }
2495
2496 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2497  * the superblock) which were deleted from all directories, but held open by
2498  * a process at the time of a crash.  We walk the list and try to delete these
2499  * inodes at recovery time (only with a read-write filesystem).
2500  *
2501  * In order to keep the orphan inode chain consistent during traversal (in
2502  * case of crash during recovery), we link each inode into the superblock
2503  * orphan list_head and handle it the same way as an inode deletion during
2504  * normal operation (which journals the operations for us).
2505  *
2506  * We only do an iget() and an iput() on each inode, which is very safe if we
2507  * accidentally point at an in-use or already deleted inode.  The worst that
2508  * can happen in this case is that we get a "bit already cleared" message from
2509  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2510  * e2fsck was run on this filesystem, and it must have already done the orphan
2511  * inode cleanup for us, so we can safely abort without any further action.
2512  */
2513 static void ext4_orphan_cleanup(struct super_block *sb,
2514                                 struct ext4_super_block *es)
2515 {
2516         unsigned int s_flags = sb->s_flags;
2517         int ret, nr_orphans = 0, nr_truncates = 0;
2518 #ifdef CONFIG_QUOTA
2519         int quota_update = 0;
2520         int i;
2521 #endif
2522         if (!es->s_last_orphan) {
2523                 jbd_debug(4, "no orphan inodes to clean up\n");
2524                 return;
2525         }
2526
2527         if (bdev_read_only(sb->s_bdev)) {
2528                 ext4_msg(sb, KERN_ERR, "write access "
2529                         "unavailable, skipping orphan cleanup");
2530                 return;
2531         }
2532
2533         /* Check if feature set would not allow a r/w mount */
2534         if (!ext4_feature_set_ok(sb, 0)) {
2535                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2536                          "unknown ROCOMPAT features");
2537                 return;
2538         }
2539
2540         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2541                 /* don't clear list on RO mount w/ errors */
2542                 if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
2543                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2544                                   "clearing orphan list.\n");
2545                         es->s_last_orphan = 0;
2546                 }
2547                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2548                 return;
2549         }
2550
2551         if (s_flags & SB_RDONLY) {
2552                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2553                 sb->s_flags &= ~SB_RDONLY;
2554         }
2555 #ifdef CONFIG_QUOTA
2556         /* Needed for iput() to work correctly and not trash data */
2557         sb->s_flags |= SB_ACTIVE;
2558
2559         /*
2560          * Turn on quotas which were not enabled for read-only mounts if
2561          * filesystem has quota feature, so that they are updated correctly.
2562          */
2563         if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
2564                 int ret = ext4_enable_quotas(sb);
2565
2566                 if (!ret)
2567                         quota_update = 1;
2568                 else
2569                         ext4_msg(sb, KERN_ERR,
2570                                 "Cannot turn on quotas: error %d", ret);
2571         }
2572
2573         /* Turn on journaled quotas used for old sytle */
2574         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2575                 if (EXT4_SB(sb)->s_qf_names[i]) {
2576                         int ret = ext4_quota_on_mount(sb, i);
2577
2578                         if (!ret)
2579                                 quota_update = 1;
2580                         else
2581                                 ext4_msg(sb, KERN_ERR,
2582                                         "Cannot turn on journaled "
2583                                         "quota: type %d: error %d", i, ret);
2584                 }
2585         }
2586 #endif
2587
2588         while (es->s_last_orphan) {
2589                 struct inode *inode;
2590
2591                 /*
2592                  * We may have encountered an error during cleanup; if
2593                  * so, skip the rest.
2594                  */
2595                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2596                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2597                         es->s_last_orphan = 0;
2598                         break;
2599                 }
2600
2601                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2602                 if (IS_ERR(inode)) {
2603                         es->s_last_orphan = 0;
2604                         break;
2605                 }
2606
2607                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2608                 dquot_initialize(inode);
2609                 if (inode->i_nlink) {
2610                         if (test_opt(sb, DEBUG))
2611                                 ext4_msg(sb, KERN_DEBUG,
2612                                         "%s: truncating inode %lu to %lld bytes",
2613                                         __func__, inode->i_ino, inode->i_size);
2614                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2615                                   inode->i_ino, inode->i_size);
2616                         inode_lock(inode);
2617                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2618                         ret = ext4_truncate(inode);
2619                         if (ret)
2620                                 ext4_std_error(inode->i_sb, ret);
2621                         inode_unlock(inode);
2622                         nr_truncates++;
2623                 } else {
2624                         if (test_opt(sb, DEBUG))
2625                                 ext4_msg(sb, KERN_DEBUG,
2626                                         "%s: deleting unreferenced inode %lu",
2627                                         __func__, inode->i_ino);
2628                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2629                                   inode->i_ino);
2630                         nr_orphans++;
2631                 }
2632                 iput(inode);  /* The delete magic happens here! */
2633         }
2634
2635 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2636
2637         if (nr_orphans)
2638                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2639                        PLURAL(nr_orphans));
2640         if (nr_truncates)
2641                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2642                        PLURAL(nr_truncates));
2643 #ifdef CONFIG_QUOTA
2644         /* Turn off quotas if they were enabled for orphan cleanup */
2645         if (quota_update) {
2646                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2647                         if (sb_dqopt(sb)->files[i])
2648                                 dquot_quota_off(sb, i);
2649                 }
2650         }
2651 #endif
2652         sb->s_flags = s_flags; /* Restore SB_RDONLY status */
2653 }
2654
2655 /*
2656  * Maximal extent format file size.
2657  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2658  * extent format containers, within a sector_t, and within i_blocks
2659  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2660  * so that won't be a limiting factor.
2661  *
2662  * However there is other limiting factor. We do store extents in the form
2663  * of starting block and length, hence the resulting length of the extent
2664  * covering maximum file size must fit into on-disk format containers as
2665  * well. Given that length is always by 1 unit bigger than max unit (because
2666  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2667  *
2668  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2669  */
2670 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2671 {
2672         loff_t res;
2673         loff_t upper_limit = MAX_LFS_FILESIZE;
2674
2675         /* small i_blocks in vfs inode? */
2676         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2677                 /*
2678                  * CONFIG_LBDAF is not enabled implies the inode
2679                  * i_block represent total blocks in 512 bytes
2680                  * 32 == size of vfs inode i_blocks * 8
2681                  */
2682                 upper_limit = (1LL << 32) - 1;
2683
2684                 /* total blocks in file system block size */
2685                 upper_limit >>= (blkbits - 9);
2686                 upper_limit <<= blkbits;
2687         }
2688
2689         /*
2690          * 32-bit extent-start container, ee_block. We lower the maxbytes
2691          * by one fs block, so ee_len can cover the extent of maximum file
2692          * size
2693          */
2694         res = (1LL << 32) - 1;
2695         res <<= blkbits;
2696
2697         /* Sanity check against vm- & vfs- imposed limits */
2698         if (res > upper_limit)
2699                 res = upper_limit;
2700
2701         return res;
2702 }
2703
2704 /*
2705  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2706  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2707  * We need to be 1 filesystem block less than the 2^48 sector limit.
2708  */
2709 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2710 {
2711         loff_t res = EXT4_NDIR_BLOCKS;
2712         int meta_blocks;
2713         loff_t upper_limit;
2714         /* This is calculated to be the largest file size for a dense, block
2715          * mapped file such that the file's total number of 512-byte sectors,
2716          * including data and all indirect blocks, does not exceed (2^48 - 1).
2717          *
2718          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2719          * number of 512-byte sectors of the file.
2720          */
2721
2722         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2723                 /*
2724                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2725                  * the inode i_block field represents total file blocks in
2726                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2727                  */
2728                 upper_limit = (1LL << 32) - 1;
2729
2730                 /* total blocks in file system block size */
2731                 upper_limit >>= (bits - 9);
2732
2733         } else {
2734                 /*
2735                  * We use 48 bit ext4_inode i_blocks
2736                  * With EXT4_HUGE_FILE_FL set the i_blocks
2737                  * represent total number of blocks in
2738                  * file system block size
2739                  */
2740                 upper_limit = (1LL << 48) - 1;
2741
2742         }
2743
2744         /* indirect blocks */
2745         meta_blocks = 1;
2746         /* double indirect blocks */
2747         meta_blocks += 1 + (1LL << (bits-2));
2748         /* tripple indirect blocks */
2749         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2750
2751         upper_limit -= meta_blocks;
2752         upper_limit <<= bits;
2753
2754         res += 1LL << (bits-2);
2755         res += 1LL << (2*(bits-2));
2756         res += 1LL << (3*(bits-2));
2757         res <<= bits;
2758         if (res > upper_limit)
2759                 res = upper_limit;
2760
2761         if (res > MAX_LFS_FILESIZE)
2762                 res = MAX_LFS_FILESIZE;
2763
2764         return res;
2765 }
2766
2767 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2768                                    ext4_fsblk_t logical_sb_block, int nr)
2769 {
2770         struct ext4_sb_info *sbi = EXT4_SB(sb);
2771         ext4_group_t bg, first_meta_bg;
2772         int has_super = 0;
2773
2774         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2775
2776         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2777                 return logical_sb_block + nr + 1;
2778         bg = sbi->s_desc_per_block * nr;
2779         if (ext4_bg_has_super(sb, bg))
2780                 has_super = 1;
2781
2782         /*
2783          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2784          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2785          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2786          * compensate.
2787          */
2788         if (sb->s_blocksize == 1024 && nr == 0 &&
2789             le32_to_cpu(sbi->s_es->s_first_data_block) == 0)
2790                 has_super++;
2791
2792         return (has_super + ext4_group_first_block_no(sb, bg));
2793 }
2794
2795 /**
2796  * ext4_get_stripe_size: Get the stripe size.
2797  * @sbi: In memory super block info
2798  *
2799  * If we have specified it via mount option, then
2800  * use the mount option value. If the value specified at mount time is
2801  * greater than the blocks per group use the super block value.
2802  * If the super block value is greater than blocks per group return 0.
2803  * Allocator needs it be less than blocks per group.
2804  *
2805  */
2806 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2807 {
2808         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2809         unsigned long stripe_width =
2810                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2811         int ret;
2812
2813         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2814                 ret = sbi->s_stripe;
2815         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2816                 ret = stripe_width;
2817         else if (stride && stride <= sbi->s_blocks_per_group)
2818                 ret = stride;
2819         else
2820                 ret = 0;
2821
2822         /*
2823          * If the stripe width is 1, this makes no sense and
2824          * we set it to 0 to turn off stripe handling code.
2825          */
2826         if (ret <= 1)
2827                 ret = 0;
2828
2829         return ret;
2830 }
2831
2832 /*
2833  * Check whether this filesystem can be mounted based on
2834  * the features present and the RDONLY/RDWR mount requested.
2835  * Returns 1 if this filesystem can be mounted as requested,
2836  * 0 if it cannot be.
2837  */
2838 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2839 {
2840         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2841                 ext4_msg(sb, KERN_ERR,
2842                         "Couldn't mount because of "
2843                         "unsupported optional features (%x)",
2844                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2845                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2846                 return 0;
2847         }
2848
2849         if (readonly)
2850                 return 1;
2851
2852         if (ext4_has_feature_readonly(sb)) {
2853                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2854                 sb->s_flags |= SB_RDONLY;
2855                 return 1;
2856         }
2857
2858         /* Check that feature set is OK for a read-write mount */
2859         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2860                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2861                          "unsupported optional features (%x)",
2862                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2863                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2864                 return 0;
2865         }
2866         /*
2867          * Large file size enabled file system can only be mounted
2868          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2869          */
2870         if (ext4_has_feature_huge_file(sb)) {
2871                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2872                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2873                                  "cannot be mounted RDWR without "
2874                                  "CONFIG_LBDAF");
2875                         return 0;
2876                 }
2877         }
2878         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2879                 ext4_msg(sb, KERN_ERR,
2880                          "Can't support bigalloc feature without "
2881                          "extents feature\n");
2882                 return 0;
2883         }
2884
2885 #ifndef CONFIG_QUOTA
2886         if (ext4_has_feature_quota(sb) && !readonly) {
2887                 ext4_msg(sb, KERN_ERR,
2888                          "Filesystem with quota feature cannot be mounted RDWR "
2889                          "without CONFIG_QUOTA");
2890                 return 0;
2891         }
2892         if (ext4_has_feature_project(sb) && !readonly) {
2893                 ext4_msg(sb, KERN_ERR,
2894                          "Filesystem with project quota feature cannot be mounted RDWR "
2895                          "without CONFIG_QUOTA");
2896                 return 0;
2897         }
2898 #endif  /* CONFIG_QUOTA */
2899         return 1;
2900 }
2901
2902 /*
2903  * This function is called once a day if we have errors logged
2904  * on the file system
2905  */
2906 static void print_daily_error_info(struct timer_list *t)
2907 {
2908         struct ext4_sb_info *sbi = from_timer(sbi, t, s_err_report);
2909         struct super_block *sb = sbi->s_sb;
2910         struct ext4_super_block *es = sbi->s_es;
2911
2912         if (es->s_error_count)
2913                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2914                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2915                          le32_to_cpu(es->s_error_count));
2916         if (es->s_first_error_time) {
2917                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %llu: %.*s:%d",
2918                        sb->s_id,
2919                        ext4_get_tstamp(es, s_first_error_time),
2920                        (int) sizeof(es->s_first_error_func),
2921                        es->s_first_error_func,
2922                        le32_to_cpu(es->s_first_error_line));
2923                 if (es->s_first_error_ino)
2924                         printk(KERN_CONT ": inode %u",
2925                                le32_to_cpu(es->s_first_error_ino));
2926                 if (es->s_first_error_block)
2927                         printk(KERN_CONT ": block %llu", (unsigned long long)
2928                                le64_to_cpu(es->s_first_error_block));
2929                 printk(KERN_CONT "\n");
2930         }
2931         if (es->s_last_error_time) {
2932                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %llu: %.*s:%d",
2933                        sb->s_id,
2934                        ext4_get_tstamp(es, s_last_error_time),
2935                        (int) sizeof(es->s_last_error_func),
2936                        es->s_last_error_func,
2937                        le32_to_cpu(es->s_last_error_line));
2938                 if (es->s_last_error_ino)
2939                         printk(KERN_CONT ": inode %u",
2940                                le32_to_cpu(es->s_last_error_ino));
2941                 if (es->s_last_error_block)
2942                         printk(KERN_CONT ": block %llu", (unsigned long long)
2943                                le64_to_cpu(es->s_last_error_block));
2944                 printk(KERN_CONT "\n");
2945         }
2946         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2947 }
2948
2949 /* Find next suitable group and run ext4_init_inode_table */
2950 static int ext4_run_li_request(struct ext4_li_request *elr)
2951 {
2952         struct ext4_group_desc *gdp = NULL;
2953         ext4_group_t group, ngroups;
2954         struct super_block *sb;
2955         unsigned long timeout = 0;
2956         int ret = 0;
2957
2958         sb = elr->lr_super;
2959         ngroups = EXT4_SB(sb)->s_groups_count;
2960
2961         for (group = elr->lr_next_group; group < ngroups; group++) {
2962                 gdp = ext4_get_group_desc(sb, group, NULL);
2963                 if (!gdp) {
2964                         ret = 1;
2965                         break;
2966                 }
2967
2968                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2969                         break;
2970         }
2971
2972         if (group >= ngroups)
2973                 ret = 1;
2974
2975         if (!ret) {
2976                 timeout = jiffies;
2977                 ret = ext4_init_inode_table(sb, group,
2978                                             elr->lr_timeout ? 0 : 1);
2979                 if (elr->lr_timeout == 0) {
2980                         timeout = (jiffies - timeout) *
2981                                   elr->lr_sbi->s_li_wait_mult;
2982                         elr->lr_timeout = timeout;
2983                 }
2984                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2985                 elr->lr_next_group = group + 1;
2986         }
2987         return ret;
2988 }
2989
2990 /*
2991  * Remove lr_request from the list_request and free the
2992  * request structure. Should be called with li_list_mtx held
2993  */
2994 static void ext4_remove_li_request(struct ext4_li_request *elr)
2995 {
2996         struct ext4_sb_info *sbi;
2997
2998         if (!elr)
2999                 return;
3000
3001         sbi = elr->lr_sbi;
3002
3003         list_del(&elr->lr_request);
3004         sbi->s_li_request = NULL;
3005         kfree(elr);
3006 }
3007
3008 static void ext4_unregister_li_request(struct super_block *sb)
3009 {
3010         mutex_lock(&ext4_li_mtx);
3011         if (!ext4_li_info) {
3012                 mutex_unlock(&ext4_li_mtx);
3013                 return;
3014         }
3015
3016         mutex_lock(&ext4_li_info->li_list_mtx);
3017         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
3018         mutex_unlock(&ext4_li_info->li_list_mtx);
3019         mutex_unlock(&ext4_li_mtx);
3020 }
3021
3022 static struct task_struct *ext4_lazyinit_task;
3023
3024 /*
3025  * This is the function where ext4lazyinit thread lives. It walks
3026  * through the request list searching for next scheduled filesystem.
3027  * When such a fs is found, run the lazy initialization request
3028  * (ext4_rn_li_request) and keep track of the time spend in this
3029  * function. Based on that time we compute next schedule time of
3030  * the request. When walking through the list is complete, compute
3031  * next waking time and put itself into sleep.
3032  */
3033 static int ext4_lazyinit_thread(void *arg)
3034 {
3035         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
3036         struct list_head *pos, *n;
3037         struct ext4_li_request *elr;
3038         unsigned long next_wakeup, cur;
3039
3040         BUG_ON(NULL == eli);
3041
3042 cont_thread:
3043         while (true) {
3044                 next_wakeup = MAX_JIFFY_OFFSET;
3045
3046                 mutex_lock(&eli->li_list_mtx);
3047                 if (list_empty(&eli->li_request_list)) {
3048                         mutex_unlock(&eli->li_list_mtx);
3049                         goto exit_thread;
3050                 }
3051                 list_for_each_safe(pos, n, &eli->li_request_list) {
3052                         int err = 0;
3053                         int progress = 0;
3054                         elr = list_entry(pos, struct ext4_li_request,
3055                                          lr_request);
3056
3057                         if (time_before(jiffies, elr->lr_next_sched)) {
3058                                 if (time_before(elr->lr_next_sched, next_wakeup))
3059                                         next_wakeup = elr->lr_next_sched;
3060                                 continue;
3061                         }
3062                         if (down_read_trylock(&elr->lr_super->s_umount)) {
3063                                 if (sb_start_write_trylock(elr->lr_super)) {
3064                                         progress = 1;
3065                                         /*
3066                                          * We hold sb->s_umount, sb can not
3067                                          * be removed from the list, it is
3068                                          * now safe to drop li_list_mtx
3069                                          */
3070                                         mutex_unlock(&eli->li_list_mtx);
3071                                         err = ext4_run_li_request(elr);
3072                                         sb_end_write(elr->lr_super);
3073                                         mutex_lock(&eli->li_list_mtx);
3074                                         n = pos->next;
3075                                 }
3076                                 up_read((&elr->lr_super->s_umount));
3077                         }
3078                         /* error, remove the lazy_init job */
3079                         if (err) {
3080                                 ext4_remove_li_request(elr);
3081                                 continue;
3082                         }
3083                         if (!progress) {
3084                                 elr->lr_next_sched = jiffies +
3085                                         (prandom_u32()
3086                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3087                         }
3088                         if (time_before(elr->lr_next_sched, next_wakeup))
3089                                 next_wakeup = elr->lr_next_sched;
3090                 }
3091                 mutex_unlock(&eli->li_list_mtx);
3092
3093                 try_to_freeze();
3094
3095                 cur = jiffies;
3096                 if ((time_after_eq(cur, next_wakeup)) ||
3097                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3098                         cond_resched();
3099                         continue;
3100                 }
3101
3102                 schedule_timeout_interruptible(next_wakeup - cur);
3103
3104                 if (kthread_should_stop()) {
3105                         ext4_clear_request_list();
3106                         goto exit_thread;
3107                 }
3108         }
3109
3110 exit_thread:
3111         /*
3112          * It looks like the request list is empty, but we need
3113          * to check it under the li_list_mtx lock, to prevent any
3114          * additions into it, and of course we should lock ext4_li_mtx
3115          * to atomically free the list and ext4_li_info, because at
3116          * this point another ext4 filesystem could be registering
3117          * new one.
3118          */
3119         mutex_lock(&ext4_li_mtx);
3120         mutex_lock(&eli->li_list_mtx);
3121         if (!list_empty(&eli->li_request_list)) {
3122                 mutex_unlock(&eli->li_list_mtx);
3123                 mutex_unlock(&ext4_li_mtx);
3124                 goto cont_thread;
3125         }
3126         mutex_unlock(&eli->li_list_mtx);
3127         kfree(ext4_li_info);
3128         ext4_li_info = NULL;
3129         mutex_unlock(&ext4_li_mtx);
3130
3131         return 0;
3132 }
3133
3134 static void ext4_clear_request_list(void)
3135 {
3136         struct list_head *pos, *n;
3137         struct ext4_li_request *elr;
3138
3139         mutex_lock(&ext4_li_info->li_list_mtx);
3140         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3141                 elr = list_entry(pos, struct ext4_li_request,
3142                                  lr_request);
3143                 ext4_remove_li_request(elr);
3144         }
3145         mutex_unlock(&ext4_li_info->li_list_mtx);
3146 }
3147
3148 static int ext4_run_lazyinit_thread(void)
3149 {
3150         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3151                                          ext4_li_info, "ext4lazyinit");
3152         if (IS_ERR(ext4_lazyinit_task)) {
3153                 int err = PTR_ERR(ext4_lazyinit_task);
3154                 ext4_clear_request_list();
3155                 kfree(ext4_li_info);
3156                 ext4_li_info = NULL;
3157                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3158                                  "initialization thread\n",
3159                                  err);
3160                 return err;
3161         }
3162         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3163         return 0;
3164 }
3165
3166 /*
3167  * Check whether it make sense to run itable init. thread or not.
3168  * If there is at least one uninitialized inode table, return
3169  * corresponding group number, else the loop goes through all
3170  * groups and return total number of groups.
3171  */
3172 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3173 {
3174         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3175         struct ext4_group_desc *gdp = NULL;
3176
3177         if (!ext4_has_group_desc_csum(sb))
3178                 return ngroups;
3179
3180         for (group = 0; group < ngroups; group++) {
3181                 gdp = ext4_get_group_desc(sb, group, NULL);
3182                 if (!gdp)
3183                         continue;
3184
3185                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3186                         break;
3187         }
3188
3189         return group;
3190 }
3191
3192 static int ext4_li_info_new(void)
3193 {
3194         struct ext4_lazy_init *eli = NULL;
3195
3196         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3197         if (!eli)
3198                 return -ENOMEM;
3199
3200         INIT_LIST_HEAD(&eli->li_request_list);
3201         mutex_init(&eli->li_list_mtx);
3202
3203         eli->li_state |= EXT4_LAZYINIT_QUIT;
3204
3205         ext4_li_info = eli;
3206
3207         return 0;
3208 }
3209
3210 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3211                                             ext4_group_t start)
3212 {
3213         struct ext4_sb_info *sbi = EXT4_SB(sb);
3214         struct ext4_li_request *elr;
3215
3216         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3217         if (!elr)
3218                 return NULL;
3219
3220         elr->lr_super = sb;
3221         elr->lr_sbi = sbi;
3222         elr->lr_next_group = start;
3223
3224         /*
3225          * Randomize first schedule time of the request to
3226          * spread the inode table initialization requests
3227          * better.
3228          */
3229         elr->lr_next_sched = jiffies + (prandom_u32() %
3230                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3231         return elr;
3232 }
3233
3234 int ext4_register_li_request(struct super_block *sb,
3235                              ext4_group_t first_not_zeroed)
3236 {
3237         struct ext4_sb_info *sbi = EXT4_SB(sb);
3238         struct ext4_li_request *elr = NULL;
3239         ext4_group_t ngroups = sbi->s_groups_count;
3240         int ret = 0;
3241
3242         mutex_lock(&ext4_li_mtx);
3243         if (sbi->s_li_request != NULL) {
3244                 /*
3245                  * Reset timeout so it can be computed again, because
3246                  * s_li_wait_mult might have changed.
3247                  */
3248                 sbi->s_li_request->lr_timeout = 0;
3249                 goto out;
3250         }
3251
3252         if (first_not_zeroed == ngroups || sb_rdonly(sb) ||
3253             !test_opt(sb, INIT_INODE_TABLE))
3254                 goto out;
3255
3256         elr = ext4_li_request_new(sb, first_not_zeroed);
3257         if (!elr) {
3258                 ret = -ENOMEM;
3259                 goto out;
3260         }
3261
3262         if (NULL == ext4_li_info) {
3263                 ret = ext4_li_info_new();
3264                 if (ret)
3265                         goto out;
3266         }
3267
3268         mutex_lock(&ext4_li_info->li_list_mtx);
3269         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3270         mutex_unlock(&ext4_li_info->li_list_mtx);
3271
3272         sbi->s_li_request = elr;
3273         /*
3274          * set elr to NULL here since it has been inserted to
3275          * the request_list and the removal and free of it is
3276          * handled by ext4_clear_request_list from now on.
3277          */
3278         elr = NULL;
3279
3280         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3281                 ret = ext4_run_lazyinit_thread();
3282                 if (ret)
3283                         goto out;
3284         }
3285 out:
3286         mutex_unlock(&ext4_li_mtx);
3287         if (ret)
3288                 kfree(elr);
3289         return ret;
3290 }
3291
3292 /*
3293  * We do not need to lock anything since this is called on
3294  * module unload.
3295  */
3296 static void ext4_destroy_lazyinit_thread(void)
3297 {
3298         /*
3299          * If thread exited earlier
3300          * there's nothing to be done.
3301          */
3302         if (!ext4_li_info || !ext4_lazyinit_task)
3303                 return;
3304
3305         kthread_stop(ext4_lazyinit_task);
3306 }
3307
3308 static int set_journal_csum_feature_set(struct super_block *sb)
3309 {
3310         int ret = 1;
3311         int compat, incompat;
3312         struct ext4_sb_info *sbi = EXT4_SB(sb);
3313
3314         if (ext4_has_metadata_csum(sb)) {
3315                 /* journal checksum v3 */
3316                 compat = 0;
3317                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3318         } else {
3319                 /* journal checksum v1 */
3320                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3321                 incompat = 0;
3322         }
3323
3324         jbd2_journal_clear_features(sbi->s_journal,
3325                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3326                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3327                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3328         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3329                 ret = jbd2_journal_set_features(sbi->s_journal,
3330                                 compat, 0,
3331                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3332                                 incompat);
3333         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3334                 ret = jbd2_journal_set_features(sbi->s_journal,
3335                                 compat, 0,
3336                                 incompat);
3337                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3338                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3339         } else {
3340                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3341                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3342         }
3343
3344         return ret;
3345 }
3346
3347 /*
3348  * Note: calculating the overhead so we can be compatible with
3349  * historical BSD practice is quite difficult in the face of
3350  * clusters/bigalloc.  This is because multiple metadata blocks from
3351  * different block group can end up in the same allocation cluster.
3352  * Calculating the exact overhead in the face of clustered allocation
3353  * requires either O(all block bitmaps) in memory or O(number of block
3354  * groups**2) in time.  We will still calculate the superblock for
3355  * older file systems --- and if we come across with a bigalloc file
3356  * system with zero in s_overhead_clusters the estimate will be close to
3357  * correct especially for very large cluster sizes --- but for newer
3358  * file systems, it's better to calculate this figure once at mkfs
3359  * time, and store it in the superblock.  If the superblock value is
3360  * present (even for non-bigalloc file systems), we will use it.
3361  */
3362 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3363                           char *buf)
3364 {
3365         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3366         struct ext4_group_desc  *gdp;
3367         ext4_fsblk_t            first_block, last_block, b;
3368         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3369         int                     s, j, count = 0;
3370
3371         if (!ext4_has_feature_bigalloc(sb))
3372                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3373                         sbi->s_itb_per_group + 2);
3374
3375         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3376                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3377         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3378         for (i = 0; i < ngroups; i++) {
3379                 gdp = ext4_get_group_desc(sb, i, NULL);
3380                 b = ext4_block_bitmap(sb, gdp);
3381                 if (b >= first_block && b <= last_block) {
3382                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3383                         count++;
3384                 }
3385                 b = ext4_inode_bitmap(sb, gdp);
3386                 if (b >= first_block && b <= last_block) {
3387                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3388                         count++;
3389                 }
3390                 b = ext4_inode_table(sb, gdp);
3391                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3392                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3393                                 int c = EXT4_B2C(sbi, b - first_block);
3394                                 ext4_set_bit(c, buf);
3395                                 count++;
3396                         }
3397                 if (i != grp)
3398                         continue;
3399                 s = 0;
3400                 if (ext4_bg_has_super(sb, grp)) {
3401                         ext4_set_bit(s++, buf);
3402                         count++;
3403                 }
3404                 j = ext4_bg_num_gdb(sb, grp);
3405                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3406                         ext4_error(sb, "Invalid number of block group "
3407                                    "descriptor blocks: %d", j);
3408                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3409                 }
3410                 count += j;
3411                 for (; j > 0; j--)
3412                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3413         }
3414         if (!count)
3415                 return 0;
3416         return EXT4_CLUSTERS_PER_GROUP(sb) -
3417                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3418 }
3419
3420 /*
3421  * Compute the overhead and stash it in sbi->s_overhead
3422  */
3423 int ext4_calculate_overhead(struct super_block *sb)
3424 {
3425         struct ext4_sb_info *sbi = EXT4_SB(sb);
3426         struct ext4_super_block *es = sbi->s_es;
3427         struct inode *j_inode;
3428         unsigned int j_blocks, j_inum = le32_to_cpu(es->s_journal_inum);
3429         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3430         ext4_fsblk_t overhead = 0;
3431         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3432
3433         if (!buf)
3434                 return -ENOMEM;
3435
3436         /*
3437          * Compute the overhead (FS structures).  This is constant
3438          * for a given filesystem unless the number of block groups
3439          * changes so we cache the previous value until it does.
3440          */
3441
3442         /*
3443          * All of the blocks before first_data_block are overhead
3444          */
3445         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3446
3447         /*
3448          * Add the overhead found in each block group
3449          */
3450         for (i = 0; i < ngroups; i++) {
3451                 int blks;
3452
3453                 blks = count_overhead(sb, i, buf);
3454                 overhead += blks;
3455                 if (blks)
3456                         memset(buf, 0, PAGE_SIZE);
3457                 cond_resched();
3458         }
3459
3460         /*
3461          * Add the internal journal blocks whether the journal has been
3462          * loaded or not
3463          */
3464         if (sbi->s_journal && !sbi->journal_bdev)
3465                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3466         else if (ext4_has_feature_journal(sb) && !sbi->s_journal) {
3467                 j_inode = ext4_get_journal_inode(sb, j_inum);
3468                 if (j_inode) {
3469                         j_blocks = j_inode->i_size >> sb->s_blocksize_bits;
3470                         overhead += EXT4_NUM_B2C(sbi, j_blocks);
3471                         iput(j_inode);
3472                 } else {
3473                         ext4_msg(sb, KERN_ERR, "can't get journal size");
3474                 }
3475         }
3476         sbi->s_overhead = overhead;
3477         smp_wmb();
3478         free_page((unsigned long) buf);
3479         return 0;
3480 }
3481
3482 static void ext4_set_resv_clusters(struct super_block *sb)
3483 {
3484         ext4_fsblk_t resv_clusters;
3485         struct ext4_sb_info *sbi = EXT4_SB(sb);
3486
3487         /*
3488          * There's no need to reserve anything when we aren't using extents.
3489          * The space estimates are exact, there are no unwritten extents,
3490          * hole punching doesn't need new metadata... This is needed especially
3491          * to keep ext2/3 backward compatibility.
3492          */
3493         if (!ext4_has_feature_extents(sb))
3494                 return;
3495         /*
3496          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3497          * This should cover the situations where we can not afford to run
3498          * out of space like for example punch hole, or converting
3499          * unwritten extents in delalloc path. In most cases such
3500          * allocation would require 1, or 2 blocks, higher numbers are
3501          * very rare.
3502          */
3503         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3504                          sbi->s_cluster_bits);
3505
3506         do_div(resv_clusters, 50);
3507         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3508
3509         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3510 }
3511
3512 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3513 {
3514         struct dax_device *dax_dev = fs_dax_get_by_bdev(sb->s_bdev);
3515         char *orig_data = kstrdup(data, GFP_KERNEL);
3516         struct buffer_head *bh;
3517         struct ext4_super_block *es = NULL;
3518         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3519         ext4_fsblk_t block;
3520         ext4_fsblk_t sb_block = get_sb_block(&data);
3521         ext4_fsblk_t logical_sb_block;
3522         unsigned long offset = 0;
3523         unsigned long journal_devnum = 0;
3524         unsigned long def_mount_opts;
3525         struct inode *root;
3526         const char *descr;
3527         int ret = -ENOMEM;
3528         int blocksize, clustersize;
3529         unsigned int db_count;
3530         unsigned int i;
3531         int needs_recovery, has_huge_files, has_bigalloc;
3532         __u64 blocks_count;
3533         int err = 0;
3534         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3535         ext4_group_t first_not_zeroed;
3536
3537         if ((data && !orig_data) || !sbi)
3538                 goto out_free_base;
3539
3540         sbi->s_daxdev = dax_dev;
3541         sbi->s_blockgroup_lock =
3542                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3543         if (!sbi->s_blockgroup_lock)
3544                 goto out_free_base;
3545
3546         sb->s_fs_info = sbi;
3547         sbi->s_sb = sb;
3548         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3549         sbi->s_sb_block = sb_block;
3550         if (sb->s_bdev->bd_part)
3551                 sbi->s_sectors_written_start =
3552                         part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
3553
3554         /* Cleanup superblock name */
3555         strreplace(sb->s_id, '/', '!');
3556
3557         /* -EINVAL is default */
3558         ret = -EINVAL;
3559         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3560         if (!blocksize) {
3561                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3562                 goto out_fail;
3563         }
3564
3565         /*
3566          * The ext4 superblock will not be buffer aligned for other than 1kB
3567          * block sizes.  We need to calculate the offset from buffer start.
3568          */
3569         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3570                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3571                 offset = do_div(logical_sb_block, blocksize);
3572         } else {
3573                 logical_sb_block = sb_block;
3574         }
3575
3576         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3577                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3578                 goto out_fail;
3579         }
3580         /*
3581          * Note: s_es must be initialized as soon as possible because
3582          *       some ext4 macro-instructions depend on its value
3583          */
3584         es = (struct ext4_super_block *) (bh->b_data + offset);
3585         sbi->s_es = es;
3586         sb->s_magic = le16_to_cpu(es->s_magic);
3587         if (sb->s_magic != EXT4_SUPER_MAGIC)
3588                 goto cantfind_ext4;
3589         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3590
3591         /* Warn if metadata_csum and gdt_csum are both set. */
3592         if (ext4_has_feature_metadata_csum(sb) &&
3593             ext4_has_feature_gdt_csum(sb))
3594                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3595                              "redundant flags; please run fsck.");
3596
3597         /* Check for a known checksum algorithm */
3598         if (!ext4_verify_csum_type(sb, es)) {
3599                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3600                          "unknown checksum algorithm.");
3601                 silent = 1;
3602                 goto cantfind_ext4;
3603         }
3604
3605         /* Load the checksum driver */
3606         sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3607         if (IS_ERR(sbi->s_chksum_driver)) {
3608                 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3609                 ret = PTR_ERR(sbi->s_chksum_driver);
3610                 sbi->s_chksum_driver = NULL;
3611                 goto failed_mount;
3612         }
3613
3614         /* Check superblock checksum */
3615         if (!ext4_superblock_csum_verify(sb, es)) {
3616                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3617                          "invalid superblock checksum.  Run e2fsck?");
3618                 silent = 1;
3619                 ret = -EFSBADCRC;
3620                 goto cantfind_ext4;
3621         }
3622
3623         /* Precompute checksum seed for all metadata */
3624         if (ext4_has_feature_csum_seed(sb))
3625                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3626         else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
3627                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3628                                                sizeof(es->s_uuid));
3629
3630         /* Set defaults before we parse the mount options */
3631         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3632         set_opt(sb, INIT_INODE_TABLE);
3633         if (def_mount_opts & EXT4_DEFM_DEBUG)
3634                 set_opt(sb, DEBUG);
3635         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3636                 set_opt(sb, GRPID);
3637         if (def_mount_opts & EXT4_DEFM_UID16)
3638                 set_opt(sb, NO_UID32);
3639         /* xattr user namespace & acls are now defaulted on */
3640         set_opt(sb, XATTR_USER);
3641 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3642         set_opt(sb, POSIX_ACL);
3643 #endif
3644         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3645         if (ext4_has_metadata_csum(sb))
3646                 set_opt(sb, JOURNAL_CHECKSUM);
3647
3648         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3649                 set_opt(sb, JOURNAL_DATA);
3650         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3651                 set_opt(sb, ORDERED_DATA);
3652         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3653                 set_opt(sb, WRITEBACK_DATA);
3654
3655         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3656                 set_opt(sb, ERRORS_PANIC);
3657         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3658                 set_opt(sb, ERRORS_CONT);
3659         else
3660                 set_opt(sb, ERRORS_RO);
3661         /* block_validity enabled by default; disable with noblock_validity */
3662         set_opt(sb, BLOCK_VALIDITY);
3663         if (def_mount_opts & EXT4_DEFM_DISCARD)
3664                 set_opt(sb, DISCARD);
3665
3666         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3667         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3668         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3669         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3670         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3671
3672         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3673                 set_opt(sb, BARRIER);
3674
3675         /*
3676          * enable delayed allocation by default
3677          * Use -o nodelalloc to turn it off
3678          */
3679         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3680             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3681                 set_opt(sb, DELALLOC);
3682
3683         /*
3684          * set default s_li_wait_mult for lazyinit, for the case there is
3685          * no mount option specified.
3686          */
3687         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3688
3689         if (sbi->s_es->s_mount_opts[0]) {
3690                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3691                                               sizeof(sbi->s_es->s_mount_opts),
3692                                               GFP_KERNEL);
3693                 if (!s_mount_opts)
3694                         goto failed_mount;
3695                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3696                                    &journal_ioprio, 0)) {
3697                         ext4_msg(sb, KERN_WARNING,
3698                                  "failed to parse options in superblock: %s",
3699                                  s_mount_opts);
3700                 }
3701                 kfree(s_mount_opts);
3702         }
3703         sbi->s_def_mount_opt = sbi->s_mount_opt;
3704         if (!parse_options((char *) data, sb, &journal_devnum,
3705                            &journal_ioprio, 0))
3706                 goto failed_mount;
3707
3708         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3709                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3710                             "with data=journal disables delayed "
3711                             "allocation and O_DIRECT support!\n");
3712                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3713                         ext4_msg(sb, KERN_ERR, "can't mount with "
3714                                  "both data=journal and delalloc");
3715                         goto failed_mount;
3716                 }
3717                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3718                         ext4_msg(sb, KERN_ERR, "can't mount with "
3719                                  "both data=journal and dioread_nolock");
3720                         goto failed_mount;
3721                 }
3722                 if (test_opt(sb, DAX)) {
3723                         ext4_msg(sb, KERN_ERR, "can't mount with "
3724                                  "both data=journal and dax");
3725                         goto failed_mount;
3726                 }
3727                 if (ext4_has_feature_encrypt(sb)) {
3728                         ext4_msg(sb, KERN_WARNING,
3729                                  "encrypted files will use data=ordered "
3730                                  "instead of data journaling mode");
3731                 }
3732                 if (test_opt(sb, DELALLOC))
3733                         clear_opt(sb, DELALLOC);
3734         } else {
3735                 sb->s_iflags |= SB_I_CGROUPWB;
3736         }
3737
3738         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
3739                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
3740
3741         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3742             (ext4_has_compat_features(sb) ||
3743              ext4_has_ro_compat_features(sb) ||
3744              ext4_has_incompat_features(sb)))
3745                 ext4_msg(sb, KERN_WARNING,
3746                        "feature flags set on rev 0 fs, "
3747                        "running e2fsck is recommended");
3748
3749         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3750                 set_opt2(sb, HURD_COMPAT);
3751                 if (ext4_has_feature_64bit(sb)) {
3752                         ext4_msg(sb, KERN_ERR,
3753                                  "The Hurd can't support 64-bit file systems");
3754                         goto failed_mount;
3755                 }
3756
3757                 /*
3758                  * ea_inode feature uses l_i_version field which is not
3759                  * available in HURD_COMPAT mode.
3760                  */
3761                 if (ext4_has_feature_ea_inode(sb)) {
3762                         ext4_msg(sb, KERN_ERR,
3763                                  "ea_inode feature is not supported for Hurd");
3764                         goto failed_mount;
3765                 }
3766         }
3767
3768         if (IS_EXT2_SB(sb)) {
3769                 if (ext2_feature_set_ok(sb))
3770                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3771                                  "using the ext4 subsystem");
3772                 else {
3773                         /*
3774                          * If we're probing be silent, if this looks like
3775                          * it's actually an ext[34] filesystem.
3776                          */
3777                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3778                                 goto failed_mount;
3779                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3780                                  "to feature incompatibilities");
3781                         goto failed_mount;
3782                 }
3783         }
3784
3785         if (IS_EXT3_SB(sb)) {
3786                 if (ext3_feature_set_ok(sb))
3787                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3788                                  "using the ext4 subsystem");
3789                 else {
3790                         /*
3791                          * If we're probing be silent, if this looks like
3792                          * it's actually an ext4 filesystem.
3793                          */
3794                         if (silent && ext4_feature_set_ok(sb, sb_rdonly(sb)))
3795                                 goto failed_mount;
3796                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3797                                  "to feature incompatibilities");
3798                         goto failed_mount;
3799                 }
3800         }
3801
3802         /*
3803          * Check feature flags regardless of the revision level, since we
3804          * previously didn't change the revision level when setting the flags,
3805          * so there is a chance incompat flags are set on a rev 0 filesystem.
3806          */
3807         if (!ext4_feature_set_ok(sb, (sb_rdonly(sb))))
3808                 goto failed_mount;
3809
3810         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3811         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3812             blocksize > EXT4_MAX_BLOCK_SIZE) {
3813                 ext4_msg(sb, KERN_ERR,
3814                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3815                          blocksize, le32_to_cpu(es->s_log_block_size));
3816                 goto failed_mount;
3817         }
3818         if (le32_to_cpu(es->s_log_block_size) >
3819             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3820                 ext4_msg(sb, KERN_ERR,
3821                          "Invalid log block size: %u",
3822                          le32_to_cpu(es->s_log_block_size));
3823                 goto failed_mount;
3824         }
3825         if (le32_to_cpu(es->s_log_cluster_size) >
3826             (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3827                 ext4_msg(sb, KERN_ERR,
3828                          "Invalid log cluster size: %u",
3829                          le32_to_cpu(es->s_log_cluster_size));
3830                 goto failed_mount;
3831         }
3832
3833         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3834                 ext4_msg(sb, KERN_ERR,
3835                          "Number of reserved GDT blocks insanely large: %d",
3836                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3837                 goto failed_mount;
3838         }
3839
3840         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3841                 if (ext4_has_feature_inline_data(sb)) {
3842                         ext4_msg(sb, KERN_ERR, "Cannot use DAX on a filesystem"
3843                                         " that may contain inline data");
3844                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3845                 }
3846                 if (!bdev_dax_supported(sb->s_bdev, blocksize)) {
3847                         ext4_msg(sb, KERN_ERR,
3848                                 "DAX unsupported by block device. Turning off DAX.");
3849                         sbi->s_mount_opt &= ~EXT4_MOUNT_DAX;
3850                 }
3851         }
3852
3853         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3854                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3855                          es->s_encryption_level);
3856                 goto failed_mount;
3857         }
3858
3859         if (sb->s_blocksize != blocksize) {
3860                 /* Validate the filesystem blocksize */
3861                 if (!sb_set_blocksize(sb, blocksize)) {
3862                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3863                                         blocksize);
3864                         goto failed_mount;
3865                 }
3866
3867                 brelse(bh);
3868                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3869                 offset = do_div(logical_sb_block, blocksize);
3870                 bh = sb_bread_unmovable(sb, logical_sb_block);
3871                 if (!bh) {
3872                         ext4_msg(sb, KERN_ERR,
3873                                "Can't read superblock on 2nd try");
3874                         goto failed_mount;
3875                 }
3876                 es = (struct ext4_super_block *)(bh->b_data + offset);
3877                 sbi->s_es = es;
3878                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3879                         ext4_msg(sb, KERN_ERR,
3880                                "Magic mismatch, very weird!");
3881                         goto failed_mount;
3882                 }
3883         }
3884
3885         has_huge_files = ext4_has_feature_huge_file(sb);
3886         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3887                                                       has_huge_files);
3888         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3889
3890         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3891                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3892                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3893         } else {
3894                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3895                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3896                 if (sbi->s_first_ino < EXT4_GOOD_OLD_FIRST_INO) {
3897                         ext4_msg(sb, KERN_ERR, "invalid first ino: %u",
3898                                  sbi->s_first_ino);
3899                         goto failed_mount;
3900                 }
3901                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3902                     (!is_power_of_2(sbi->s_inode_size)) ||
3903                     (sbi->s_inode_size > blocksize)) {
3904                         ext4_msg(sb, KERN_ERR,
3905                                "unsupported inode size: %d",
3906                                sbi->s_inode_size);
3907                         goto failed_mount;
3908                 }
3909                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3910                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3911         }
3912
3913         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3914         if (ext4_has_feature_64bit(sb)) {
3915                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3916                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3917                     !is_power_of_2(sbi->s_desc_size)) {
3918                         ext4_msg(sb, KERN_ERR,
3919                                "unsupported descriptor size %lu",
3920                                sbi->s_desc_size);
3921                         goto failed_mount;
3922                 }
3923         } else
3924                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3925
3926         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3927         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3928
3929         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3930         if (sbi->s_inodes_per_block == 0)
3931                 goto cantfind_ext4;
3932         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3933             sbi->s_inodes_per_group > blocksize * 8) {
3934                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3935                          sbi->s_blocks_per_group);
3936                 goto failed_mount;
3937         }
3938         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3939                                         sbi->s_inodes_per_block;
3940         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3941         sbi->s_sbh = bh;
3942         sbi->s_mount_state = le16_to_cpu(es->s_state);
3943         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3944         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3945
3946         for (i = 0; i < 4; i++)
3947                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3948         sbi->s_def_hash_version = es->s_def_hash_version;
3949         if (ext4_has_feature_dir_index(sb)) {
3950                 i = le32_to_cpu(es->s_flags);
3951                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3952                         sbi->s_hash_unsigned = 3;
3953                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3954 #ifdef __CHAR_UNSIGNED__
3955                         if (!sb_rdonly(sb))
3956                                 es->s_flags |=
3957                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3958                         sbi->s_hash_unsigned = 3;
3959 #else
3960                         if (!sb_rdonly(sb))
3961                                 es->s_flags |=
3962                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3963 #endif
3964                 }
3965         }
3966
3967         /* Handle clustersize */
3968         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3969         has_bigalloc = ext4_has_feature_bigalloc(sb);
3970         if (has_bigalloc) {
3971                 if (clustersize < blocksize) {
3972                         ext4_msg(sb, KERN_ERR,
3973                                  "cluster size (%d) smaller than "
3974                                  "block size (%d)", clustersize, blocksize);
3975                         goto failed_mount;
3976                 }
3977                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3978                         le32_to_cpu(es->s_log_block_size);
3979                 sbi->s_clusters_per_group =
3980                         le32_to_cpu(es->s_clusters_per_group);
3981                 if (sbi->s_clusters_per_group > blocksize * 8) {
3982                         ext4_msg(sb, KERN_ERR,
3983                                  "#clusters per group too big: %lu",
3984                                  sbi->s_clusters_per_group);
3985                         goto failed_mount;
3986                 }
3987                 if (sbi->s_blocks_per_group !=
3988                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3989                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3990                                  "clusters per group (%lu) inconsistent",
3991                                  sbi->s_blocks_per_group,
3992                                  sbi->s_clusters_per_group);
3993                         goto failed_mount;
3994                 }
3995         } else {
3996                 if (clustersize != blocksize) {
3997                         ext4_msg(sb, KERN_ERR,
3998                                  "fragment/cluster size (%d) != "
3999                                  "block size (%d)", clustersize, blocksize);
4000                         goto failed_mount;
4001                 }
4002                 if (sbi->s_blocks_per_group > blocksize * 8) {
4003                         ext4_msg(sb, KERN_ERR,
4004                                  "#blocks per group too big: %lu",
4005                                  sbi->s_blocks_per_group);
4006                         goto failed_mount;
4007                 }
4008                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
4009                 sbi->s_cluster_bits = 0;
4010         }
4011         sbi->s_cluster_ratio = clustersize / blocksize;
4012
4013         /* Do we have standard group size of clustersize * 8 blocks ? */
4014         if (sbi->s_blocks_per_group == clustersize << 3)
4015                 set_opt2(sb, STD_GROUP_SIZE);
4016
4017         /*
4018          * Test whether we have more sectors than will fit in sector_t,
4019          * and whether the max offset is addressable by the page cache.
4020          */
4021         err = generic_check_addressable(sb->s_blocksize_bits,
4022                                         ext4_blocks_count(es));
4023         if (err) {
4024                 ext4_msg(sb, KERN_ERR, "filesystem"
4025                          " too large to mount safely on this system");
4026                 if (sizeof(sector_t) < 8)
4027                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
4028                 goto failed_mount;
4029         }
4030
4031         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
4032                 goto cantfind_ext4;
4033
4034         /* check blocks count against device size */
4035         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
4036         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
4037                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
4038                        "exceeds size of device (%llu blocks)",
4039                        ext4_blocks_count(es), blocks_count);
4040                 goto failed_mount;
4041         }
4042
4043         /*
4044          * It makes no sense for the first data block to be beyond the end
4045          * of the filesystem.
4046          */
4047         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
4048                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4049                          "block %u is beyond end of filesystem (%llu)",
4050                          le32_to_cpu(es->s_first_data_block),
4051                          ext4_blocks_count(es));
4052                 goto failed_mount;
4053         }
4054         if ((es->s_first_data_block == 0) && (es->s_log_block_size == 0) &&
4055             (sbi->s_cluster_ratio == 1)) {
4056                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
4057                          "block is 0 with a 1k block and cluster size");
4058                 goto failed_mount;
4059         }
4060
4061         blocks_count = (ext4_blocks_count(es) -
4062                         le32_to_cpu(es->s_first_data_block) +
4063                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
4064         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
4065         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
4066                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
4067                        "(block count %llu, first data block %u, "
4068                        "blocks per group %lu)", sbi->s_groups_count,
4069                        ext4_blocks_count(es),
4070                        le32_to_cpu(es->s_first_data_block),
4071                        EXT4_BLOCKS_PER_GROUP(sb));
4072                 goto failed_mount;
4073         }
4074         sbi->s_groups_count = blocks_count;
4075         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
4076                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
4077         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
4078                    EXT4_DESC_PER_BLOCK(sb);
4079         if (ext4_has_feature_meta_bg(sb)) {
4080                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
4081                         ext4_msg(sb, KERN_WARNING,
4082                                  "first meta block group too large: %u "
4083                                  "(group descriptor block count %u)",
4084                                  le32_to_cpu(es->s_first_meta_bg), db_count);
4085                         goto failed_mount;
4086                 }
4087         }
4088         sbi->s_group_desc = kvmalloc_array(db_count,
4089                                            sizeof(struct buffer_head *),
4090                                            GFP_KERNEL);
4091         if (sbi->s_group_desc == NULL) {
4092                 ext4_msg(sb, KERN_ERR, "not enough memory");
4093                 ret = -ENOMEM;
4094                 goto failed_mount;
4095         }
4096         if (((u64)sbi->s_groups_count * sbi->s_inodes_per_group) !=
4097             le32_to_cpu(es->s_inodes_count)) {
4098                 ext4_msg(sb, KERN_ERR, "inodes count not valid: %u vs %llu",
4099                          le32_to_cpu(es->s_inodes_count),
4100                          ((u64)sbi->s_groups_count * sbi->s_inodes_per_group));
4101                 ret = -EINVAL;
4102                 goto failed_mount;
4103         }
4104
4105         bgl_lock_init(sbi->s_blockgroup_lock);
4106
4107         /* Pre-read the descriptors into the buffer cache */
4108         for (i = 0; i < db_count; i++) {
4109                 block = descriptor_loc(sb, logical_sb_block, i);
4110                 sb_breadahead(sb, block);
4111         }
4112
4113         for (i = 0; i < db_count; i++) {
4114                 block = descriptor_loc(sb, logical_sb_block, i);
4115                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
4116                 if (!sbi->s_group_desc[i]) {
4117                         ext4_msg(sb, KERN_ERR,
4118                                "can't read group descriptor %d", i);
4119                         db_count = i;
4120                         goto failed_mount2;
4121                 }
4122         }
4123         sbi->s_gdb_count = db_count;
4124         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
4125                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
4126                 ret = -EFSCORRUPTED;
4127                 goto failed_mount2;
4128         }
4129
4130         timer_setup(&sbi->s_err_report, print_daily_error_info, 0);
4131
4132         /* Register extent status tree shrinker */
4133         if (ext4_es_register_shrinker(sbi))
4134                 goto failed_mount3;
4135
4136         sbi->s_stripe = ext4_get_stripe_size(sbi);
4137         sbi->s_extent_max_zeroout_kb = 32;
4138
4139         /*
4140          * set up enough so that it can read an inode
4141          */
4142         sb->s_op = &ext4_sops;
4143         sb->s_export_op = &ext4_export_ops;
4144         sb->s_xattr = ext4_xattr_handlers;
4145 #ifdef CONFIG_EXT4_FS_ENCRYPTION
4146         sb->s_cop = &ext4_cryptops;
4147 #endif
4148 #ifdef CONFIG_QUOTA
4149         sb->dq_op = &ext4_quota_operations;
4150         if (ext4_has_feature_quota(sb))
4151                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4152         else
4153                 sb->s_qcop = &ext4_qctl_operations;
4154         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
4155 #endif
4156         memcpy(&sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
4157
4158         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4159         mutex_init(&sbi->s_orphan_lock);
4160
4161         sb->s_root = NULL;
4162
4163         needs_recovery = (es->s_last_orphan != 0 ||
4164                           ext4_has_feature_journal_needs_recovery(sb));
4165
4166         if (ext4_has_feature_mmp(sb) && !sb_rdonly(sb))
4167                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4168                         goto failed_mount3a;
4169
4170         /*
4171          * The first inode we look at is the journal inode.  Don't try
4172          * root first: it may be modified in the journal!
4173          */
4174         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
4175                 err = ext4_load_journal(sb, es, journal_devnum);
4176                 if (err)
4177                         goto failed_mount3a;
4178         } else if (test_opt(sb, NOLOAD) && !sb_rdonly(sb) &&
4179                    ext4_has_feature_journal_needs_recovery(sb)) {
4180                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4181                        "suppressed and not mounted read-only");
4182                 goto failed_mount_wq;
4183         } else {
4184                 /* Nojournal mode, all journal mount options are illegal */
4185                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
4186                         ext4_msg(sb, KERN_ERR, "can't mount with "
4187                                  "journal_checksum, fs mounted w/o journal");
4188                         goto failed_mount_wq;
4189                 }
4190                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4191                         ext4_msg(sb, KERN_ERR, "can't mount with "
4192                                  "journal_async_commit, fs mounted w/o journal");
4193                         goto failed_mount_wq;
4194                 }
4195                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
4196                         ext4_msg(sb, KERN_ERR, "can't mount with "
4197                                  "commit=%lu, fs mounted w/o journal",
4198                                  sbi->s_commit_interval / HZ);
4199                         goto failed_mount_wq;
4200                 }
4201                 if (EXT4_MOUNT_DATA_FLAGS &
4202                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
4203                         ext4_msg(sb, KERN_ERR, "can't mount with "
4204                                  "data=, fs mounted w/o journal");
4205                         goto failed_mount_wq;
4206                 }
4207                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
4208                 clear_opt(sb, JOURNAL_CHECKSUM);
4209                 clear_opt(sb, DATA_FLAGS);
4210                 sbi->s_journal = NULL;
4211                 needs_recovery = 0;
4212                 goto no_journal;
4213         }
4214
4215         if (ext4_has_feature_64bit(sb) &&
4216             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4217                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4218                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4219                 goto failed_mount_wq;
4220         }
4221
4222         if (!set_journal_csum_feature_set(sb)) {
4223                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4224                          "feature set");
4225                 goto failed_mount_wq;
4226         }
4227
4228         /* We have now updated the journal if required, so we can
4229          * validate the data journaling mode. */
4230         switch (test_opt(sb, DATA_FLAGS)) {
4231         case 0:
4232                 /* No mode set, assume a default based on the journal
4233                  * capabilities: ORDERED_DATA if the journal can
4234                  * cope, else JOURNAL_DATA
4235                  */
4236                 if (jbd2_journal_check_available_features
4237                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4238                         set_opt(sb, ORDERED_DATA);
4239                         sbi->s_def_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
4240                 } else {
4241                         set_opt(sb, JOURNAL_DATA);
4242                         sbi->s_def_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
4243                 }
4244                 break;
4245
4246         case EXT4_MOUNT_ORDERED_DATA:
4247         case EXT4_MOUNT_WRITEBACK_DATA:
4248                 if (!jbd2_journal_check_available_features
4249                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4250                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4251                                "requested data journaling mode");
4252                         goto failed_mount_wq;
4253                 }
4254         default:
4255                 break;
4256         }
4257
4258         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
4259             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
4260                 ext4_msg(sb, KERN_ERR, "can't mount with "
4261                         "journal_async_commit in data=ordered mode");
4262                 goto failed_mount_wq;
4263         }
4264
4265         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4266
4267         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4268
4269 no_journal:
4270         if (!test_opt(sb, NO_MBCACHE)) {
4271                 sbi->s_ea_block_cache = ext4_xattr_create_cache();
4272                 if (!sbi->s_ea_block_cache) {
4273                         ext4_msg(sb, KERN_ERR,
4274                                  "Failed to create ea_block_cache");
4275                         goto failed_mount_wq;
4276                 }
4277
4278                 if (ext4_has_feature_ea_inode(sb)) {
4279                         sbi->s_ea_inode_cache = ext4_xattr_create_cache();
4280                         if (!sbi->s_ea_inode_cache) {
4281                                 ext4_msg(sb, KERN_ERR,
4282                                          "Failed to create ea_inode_cache");
4283                                 goto failed_mount_wq;
4284                         }
4285                 }
4286         }
4287
4288         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
4289             (blocksize != PAGE_SIZE)) {
4290                 ext4_msg(sb, KERN_ERR,
4291                          "Unsupported blocksize for fs encryption");
4292                 goto failed_mount_wq;
4293         }
4294
4295         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !sb_rdonly(sb) &&
4296             !ext4_has_feature_encrypt(sb)) {
4297                 ext4_set_feature_encrypt(sb);
4298                 ext4_commit_super(sb, 1);
4299         }
4300
4301         /*
4302          * Get the # of file system overhead blocks from the
4303          * superblock if present.
4304          */
4305         if (es->s_overhead_clusters)
4306                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4307         else {
4308                 err = ext4_calculate_overhead(sb);
4309                 if (err)
4310                         goto failed_mount_wq;
4311         }
4312
4313         /*
4314          * The maximum number of concurrent works can be high and
4315          * concurrency isn't really necessary.  Limit it to 1.
4316          */
4317         EXT4_SB(sb)->rsv_conversion_wq =
4318                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4319         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4320                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4321                 ret = -ENOMEM;
4322                 goto failed_mount4;
4323         }
4324
4325         /*
4326          * The jbd2_journal_load will have done any necessary log recovery,
4327          * so we can safely mount the rest of the filesystem now.
4328          */
4329
4330         root = ext4_iget(sb, EXT4_ROOT_INO);
4331         if (IS_ERR(root)) {
4332                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4333                 ret = PTR_ERR(root);
4334                 root = NULL;
4335                 goto failed_mount4;
4336         }
4337         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4338                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4339                 iput(root);
4340                 goto failed_mount4;
4341         }
4342         sb->s_root = d_make_root(root);
4343         if (!sb->s_root) {
4344                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4345                 ret = -ENOMEM;
4346                 goto failed_mount4;
4347         }
4348
4349         ret = ext4_setup_super(sb, es, sb_rdonly(sb));
4350         if (ret == -EROFS) {
4351                 sb->s_flags |= SB_RDONLY;
4352                 ret = 0;
4353         } else if (ret)
4354                 goto failed_mount4a;
4355
4356         /* determine the minimum size of new large inodes, if present */
4357         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE &&
4358             sbi->s_want_extra_isize == 0) {
4359                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4360                                                      EXT4_GOOD_OLD_INODE_SIZE;
4361                 if (ext4_has_feature_extra_isize(sb)) {
4362                         if (sbi->s_want_extra_isize <
4363                             le16_to_cpu(es->s_want_extra_isize))
4364                                 sbi->s_want_extra_isize =
4365                                         le16_to_cpu(es->s_want_extra_isize);
4366                         if (sbi->s_want_extra_isize <
4367                             le16_to_cpu(es->s_min_extra_isize))
4368                                 sbi->s_want_extra_isize =
4369                                         le16_to_cpu(es->s_min_extra_isize);
4370                 }
4371         }
4372         /* Check if enough inode space is available */
4373         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4374                                                         sbi->s_inode_size) {
4375                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4376                                                        EXT4_GOOD_OLD_INODE_SIZE;
4377                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4378                          "available");
4379         }
4380
4381         ext4_set_resv_clusters(sb);
4382
4383         err = ext4_setup_system_zone(sb);
4384         if (err) {
4385                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4386                          "zone (%d)", err);
4387                 goto failed_mount4a;
4388         }
4389
4390         ext4_ext_init(sb);
4391         err = ext4_mb_init(sb);
4392         if (err) {
4393                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4394                          err);
4395                 goto failed_mount5;
4396         }
4397
4398         block = ext4_count_free_clusters(sb);
4399         ext4_free_blocks_count_set(sbi->s_es,
4400                                    EXT4_C2B(sbi, block));
4401         ext4_superblock_csum_set(sb);
4402         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4403                                   GFP_KERNEL);
4404         if (!err) {
4405                 unsigned long freei = ext4_count_free_inodes(sb);
4406                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4407                 ext4_superblock_csum_set(sb);
4408                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4409                                           GFP_KERNEL);
4410         }
4411         if (!err)
4412                 err = percpu_counter_init(&sbi->s_dirs_counter,
4413                                           ext4_count_dirs(sb), GFP_KERNEL);
4414         if (!err)
4415                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4416                                           GFP_KERNEL);
4417         if (!err)
4418                 err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
4419
4420         if (err) {
4421                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4422                 goto failed_mount6;
4423         }
4424
4425         if (ext4_has_feature_flex_bg(sb))
4426                 if (!ext4_fill_flex_info(sb)) {
4427                         ext4_msg(sb, KERN_ERR,
4428                                "unable to initialize "
4429                                "flex_bg meta info!");
4430                         goto failed_mount6;
4431                 }
4432
4433         err = ext4_register_li_request(sb, first_not_zeroed);
4434         if (err)
4435                 goto failed_mount6;
4436
4437         err = ext4_register_sysfs(sb);
4438         if (err)
4439                 goto failed_mount7;
4440
4441 #ifdef CONFIG_QUOTA
4442         /* Enable quota usage during mount. */
4443         if (ext4_has_feature_quota(sb) && !sb_rdonly(sb)) {
4444                 err = ext4_enable_quotas(sb);
4445                 if (err)
4446                         goto failed_mount8;
4447         }
4448 #endif  /* CONFIG_QUOTA */
4449
4450         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4451         ext4_orphan_cleanup(sb, es);
4452         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4453         if (needs_recovery) {
4454                 ext4_msg(sb, KERN_INFO, "recovery complete");
4455                 ext4_mark_recovery_complete(sb, es);
4456         }
4457         if (EXT4_SB(sb)->s_journal) {
4458                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4459                         descr = " journalled data mode";
4460                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4461                         descr = " ordered data mode";
4462                 else
4463                         descr = " writeback data mode";
4464         } else
4465                 descr = "out journal";
4466
4467         if (test_opt(sb, DISCARD)) {
4468                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4469                 if (!blk_queue_discard(q))
4470                         ext4_msg(sb, KERN_WARNING,
4471                                  "mounting with \"discard\" option, but "
4472                                  "the device does not support discard");
4473         }
4474
4475         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4476                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4477                          "Opts: %.*s%s%s", descr,
4478                          (int) sizeof(sbi->s_es->s_mount_opts),
4479                          sbi->s_es->s_mount_opts,
4480                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4481
4482         if (es->s_error_count)
4483                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4484
4485         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4486         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4487         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4488         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4489
4490         kfree(orig_data);
4491         return 0;
4492
4493 cantfind_ext4:
4494         if (!silent)
4495                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4496         goto failed_mount;
4497
4498 #ifdef CONFIG_QUOTA
4499 failed_mount8:
4500         ext4_unregister_sysfs(sb);
4501 #endif
4502 failed_mount7:
4503         ext4_unregister_li_request(sb);
4504 failed_mount6:
4505         ext4_mb_release(sb);
4506         if (sbi->s_flex_groups)
4507                 kvfree(sbi->s_flex_groups);
4508         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4509         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4510         percpu_counter_destroy(&sbi->s_dirs_counter);
4511         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4512 failed_mount5:
4513         ext4_ext_release(sb);
4514         ext4_release_system_zone(sb);
4515 failed_mount4a:
4516         dput(sb->s_root);
4517         sb->s_root = NULL;
4518 failed_mount4:
4519         ext4_msg(sb, KERN_ERR, "mount failed");
4520         if (EXT4_SB(sb)->rsv_conversion_wq)
4521                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4522 failed_mount_wq:
4523         if (sbi->s_ea_inode_cache) {
4524                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
4525                 sbi->s_ea_inode_cache = NULL;
4526         }
4527         if (sbi->s_ea_block_cache) {
4528                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
4529                 sbi->s_ea_block_cache = NULL;
4530         }
4531         if (sbi->s_journal) {
4532                 jbd2_journal_destroy(sbi->s_journal);
4533                 sbi->s_journal = NULL;
4534         }
4535 failed_mount3a:
4536         ext4_es_unregister_shrinker(sbi);
4537 failed_mount3:
4538         del_timer_sync(&sbi->s_err_report);
4539         if (sbi->s_mmp_tsk)
4540                 kthread_stop(sbi->s_mmp_tsk);
4541 failed_mount2:
4542         for (i = 0; i < db_count; i++)
4543                 brelse(sbi->s_group_desc[i]);
4544         kvfree(sbi->s_group_desc);
4545 failed_mount:
4546         if (sbi->s_chksum_driver)
4547                 crypto_free_shash(sbi->s_chksum_driver);
4548 #ifdef CONFIG_QUOTA
4549         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4550                 kfree(sbi->s_qf_names[i]);
4551 #endif
4552         ext4_blkdev_remove(sbi);
4553         brelse(bh);
4554 out_fail:
4555         sb->s_fs_info = NULL;
4556         kfree(sbi->s_blockgroup_lock);
4557 out_free_base:
4558         kfree(sbi);
4559         kfree(orig_data);
4560         fs_put_dax(dax_dev);
4561         return err ? err : ret;
4562 }
4563
4564 /*
4565  * Setup any per-fs journal parameters now.  We'll do this both on
4566  * initial mount, once the journal has been initialised but before we've
4567  * done any recovery; and again on any subsequent remount.
4568  */
4569 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4570 {
4571         struct ext4_sb_info *sbi = EXT4_SB(sb);
4572
4573         journal->j_commit_interval = sbi->s_commit_interval;
4574         journal->j_min_batch_time = sbi->s_min_batch_time;
4575         journal->j_max_batch_time = sbi->s_max_batch_time;
4576
4577         write_lock(&journal->j_state_lock);
4578         if (test_opt(sb, BARRIER))
4579                 journal->j_flags |= JBD2_BARRIER;
4580         else
4581                 journal->j_flags &= ~JBD2_BARRIER;
4582         if (test_opt(sb, DATA_ERR_ABORT))
4583                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4584         else
4585                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4586         write_unlock(&journal->j_state_lock);
4587 }
4588
4589 static struct inode *ext4_get_journal_inode(struct super_block *sb,
4590                                              unsigned int journal_inum)
4591 {
4592         struct inode *journal_inode;
4593
4594         /*
4595          * Test for the existence of a valid inode on disk.  Bad things
4596          * happen if we iget() an unused inode, as the subsequent iput()
4597          * will try to delete it.
4598          */
4599         journal_inode = ext4_iget(sb, journal_inum);
4600         if (IS_ERR(journal_inode)) {
4601                 ext4_msg(sb, KERN_ERR, "no journal found");
4602                 return NULL;
4603         }
4604         if (!journal_inode->i_nlink) {
4605                 make_bad_inode(journal_inode);
4606                 iput(journal_inode);
4607                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4608                 return NULL;
4609         }
4610
4611         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4612                   journal_inode, journal_inode->i_size);
4613         if (!S_ISREG(journal_inode->i_mode)) {
4614                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4615                 iput(journal_inode);
4616                 return NULL;
4617         }
4618         return journal_inode;
4619 }
4620
4621 static journal_t *ext4_get_journal(struct super_block *sb,
4622                                    unsigned int journal_inum)
4623 {
4624         struct inode *journal_inode;
4625         journal_t *journal;
4626
4627         BUG_ON(!ext4_has_feature_journal(sb));
4628
4629         journal_inode = ext4_get_journal_inode(sb, journal_inum);
4630         if (!journal_inode)
4631                 return NULL;
4632
4633         journal = jbd2_journal_init_inode(journal_inode);
4634         if (!journal) {
4635                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4636                 iput(journal_inode);
4637                 return NULL;
4638         }
4639         journal->j_private = sb;
4640         ext4_init_journal_params(sb, journal);
4641         return journal;
4642 }
4643
4644 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4645                                        dev_t j_dev)
4646 {
4647         struct buffer_head *bh;
4648         journal_t *journal;
4649         ext4_fsblk_t start;
4650         ext4_fsblk_t len;
4651         int hblock, blocksize;
4652         ext4_fsblk_t sb_block;
4653         unsigned long offset;
4654         struct ext4_super_block *es;
4655         struct block_device *bdev;
4656
4657         BUG_ON(!ext4_has_feature_journal(sb));
4658
4659         bdev = ext4_blkdev_get(j_dev, sb);
4660         if (bdev == NULL)
4661                 return NULL;
4662
4663         blocksize = sb->s_blocksize;
4664         hblock = bdev_logical_block_size(bdev);
4665         if (blocksize < hblock) {
4666                 ext4_msg(sb, KERN_ERR,
4667                         "blocksize too small for journal device");
4668                 goto out_bdev;
4669         }
4670
4671         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4672         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4673         set_blocksize(bdev, blocksize);
4674         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4675                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4676                        "external journal");
4677                 goto out_bdev;
4678         }
4679
4680         es = (struct ext4_super_block *) (bh->b_data + offset);
4681         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4682             !(le32_to_cpu(es->s_feature_incompat) &
4683               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4684                 ext4_msg(sb, KERN_ERR, "external journal has "
4685                                         "bad superblock");
4686                 brelse(bh);
4687                 goto out_bdev;
4688         }
4689
4690         if ((le32_to_cpu(es->s_feature_ro_compat) &
4691              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4692             es->s_checksum != ext4_superblock_csum(sb, es)) {
4693                 ext4_msg(sb, KERN_ERR, "external journal has "
4694                                        "corrupt superblock");
4695                 brelse(bh);
4696                 goto out_bdev;
4697         }
4698
4699         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4700                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4701                 brelse(bh);
4702                 goto out_bdev;
4703         }
4704
4705         len = ext4_blocks_count(es);
4706         start = sb_block + 1;
4707         brelse(bh);     /* we're done with the superblock */
4708
4709         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4710                                         start, len, blocksize);
4711         if (!journal) {
4712                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4713                 goto out_bdev;
4714         }
4715         journal->j_private = sb;
4716         ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4717         wait_on_buffer(journal->j_sb_buffer);
4718         if (!buffer_uptodate(journal->j_sb_buffer)) {
4719                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4720                 goto out_journal;
4721         }
4722         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4723                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4724                                         "user (unsupported) - %d",
4725                         be32_to_cpu(journal->j_superblock->s_nr_users));
4726                 goto out_journal;
4727         }
4728         EXT4_SB(sb)->journal_bdev = bdev;
4729         ext4_init_journal_params(sb, journal);
4730         return journal;
4731
4732 out_journal:
4733         jbd2_journal_destroy(journal);
4734 out_bdev:
4735         ext4_blkdev_put(bdev);
4736         return NULL;
4737 }
4738
4739 static int ext4_load_journal(struct super_block *sb,
4740                              struct ext4_super_block *es,
4741                              unsigned long journal_devnum)
4742 {
4743         journal_t *journal;
4744         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4745         dev_t journal_dev;
4746         int err = 0;
4747         int really_read_only;
4748
4749         BUG_ON(!ext4_has_feature_journal(sb));
4750
4751         if (journal_devnum &&
4752             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4753                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4754                         "numbers have changed");
4755                 journal_dev = new_decode_dev(journal_devnum);
4756         } else
4757                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4758
4759         really_read_only = bdev_read_only(sb->s_bdev);
4760
4761         /*
4762          * Are we loading a blank journal or performing recovery after a
4763          * crash?  For recovery, we need to check in advance whether we
4764          * can get read-write access to the device.
4765          */
4766         if (ext4_has_feature_journal_needs_recovery(sb)) {
4767                 if (sb_rdonly(sb)) {
4768                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4769                                         "required on readonly filesystem");
4770                         if (really_read_only) {
4771                                 ext4_msg(sb, KERN_ERR, "write access "
4772                                         "unavailable, cannot proceed "
4773                                         "(try mounting with noload)");
4774                                 return -EROFS;
4775                         }
4776                         ext4_msg(sb, KERN_INFO, "write access will "
4777                                "be enabled during recovery");
4778                 }
4779         }
4780
4781         if (journal_inum && journal_dev) {
4782                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4783                        "and inode journals!");
4784                 return -EINVAL;
4785         }
4786
4787         if (journal_inum) {
4788                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4789                         return -EINVAL;
4790         } else {
4791                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4792                         return -EINVAL;
4793         }
4794
4795         if (!(journal->j_flags & JBD2_BARRIER))
4796                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4797
4798         if (!ext4_has_feature_journal_needs_recovery(sb))
4799                 err = jbd2_journal_wipe(journal, !really_read_only);
4800         if (!err) {
4801                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4802                 if (save)
4803                         memcpy(save, ((char *) es) +
4804                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4805                 err = jbd2_journal_load(journal);
4806                 if (save)
4807                         memcpy(((char *) es) + EXT4_S_ERR_START,
4808                                save, EXT4_S_ERR_LEN);
4809                 kfree(save);
4810         }
4811
4812         if (err) {
4813                 ext4_msg(sb, KERN_ERR, "error loading journal");
4814                 jbd2_journal_destroy(journal);
4815                 return err;
4816         }
4817
4818         EXT4_SB(sb)->s_journal = journal;
4819         ext4_clear_journal_err(sb, es);
4820
4821         if (!really_read_only && journal_devnum &&
4822             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4823                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4824
4825                 /* Make sure we flush the recovery flag to disk. */
4826                 ext4_commit_super(sb, 1);
4827         }
4828
4829         return 0;
4830 }
4831
4832 static int ext4_commit_super(struct super_block *sb, int sync)
4833 {
4834         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4835         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4836         int error = 0;
4837
4838         if (!sbh || block_device_ejected(sb))
4839                 return error;
4840
4841         /*
4842          * The superblock bh should be mapped, but it might not be if the
4843          * device was hot-removed. Not much we can do but fail the I/O.
4844          */
4845         if (!buffer_mapped(sbh))
4846                 return error;
4847
4848         /*
4849          * If the file system is mounted read-only, don't update the
4850          * superblock write time.  This avoids updating the superblock
4851          * write time when we are mounting the root file system
4852          * read/only but we need to replay the journal; at that point,
4853          * for people who are east of GMT and who make their clock
4854          * tick in localtime for Windows bug-for-bug compatibility,
4855          * the clock is set in the future, and this will cause e2fsck
4856          * to complain and force a full file system check.
4857          */
4858         if (!(sb->s_flags & SB_RDONLY))
4859                 ext4_update_tstamp(es, s_wtime);
4860         if (sb->s_bdev->bd_part)
4861                 es->s_kbytes_written =
4862                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4863                             ((part_stat_read(sb->s_bdev->bd_part,
4864                                              sectors[STAT_WRITE]) -
4865                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4866         else
4867                 es->s_kbytes_written =
4868                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4869         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4870                 ext4_free_blocks_count_set(es,
4871                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4872                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4873         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4874                 es->s_free_inodes_count =
4875                         cpu_to_le32(percpu_counter_sum_positive(
4876                                 &EXT4_SB(sb)->s_freeinodes_counter));
4877         BUFFER_TRACE(sbh, "marking dirty");
4878         ext4_superblock_csum_set(sb);
4879         if (sync)
4880                 lock_buffer(sbh);
4881         if (buffer_write_io_error(sbh)) {
4882                 /*
4883                  * Oh, dear.  A previous attempt to write the
4884                  * superblock failed.  This could happen because the
4885                  * USB device was yanked out.  Or it could happen to
4886                  * be a transient write error and maybe the block will
4887                  * be remapped.  Nothing we can do but to retry the
4888                  * write and hope for the best.
4889                  */
4890                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4891                        "superblock detected");
4892                 clear_buffer_write_io_error(sbh);
4893                 set_buffer_uptodate(sbh);
4894         }
4895         mark_buffer_dirty(sbh);
4896         if (sync) {
4897                 unlock_buffer(sbh);
4898                 error = __sync_dirty_buffer(sbh,
4899                         REQ_SYNC | (test_opt(sb, BARRIER) ? REQ_FUA : 0));
4900                 if (buffer_write_io_error(sbh)) {
4901                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4902                                "superblock");
4903                         clear_buffer_write_io_error(sbh);
4904                         set_buffer_uptodate(sbh);
4905                 }
4906         }
4907         return error;
4908 }
4909
4910 /*
4911  * Have we just finished recovery?  If so, and if we are mounting (or
4912  * remounting) the filesystem readonly, then we will end up with a
4913  * consistent fs on disk.  Record that fact.
4914  */
4915 static void ext4_mark_recovery_complete(struct super_block *sb,
4916                                         struct ext4_super_block *es)
4917 {
4918         journal_t *journal = EXT4_SB(sb)->s_journal;
4919
4920         if (!ext4_has_feature_journal(sb)) {
4921                 BUG_ON(journal != NULL);
4922                 return;
4923         }
4924         jbd2_journal_lock_updates(journal);
4925         if (jbd2_journal_flush(journal) < 0)
4926                 goto out;
4927
4928         if (ext4_has_feature_journal_needs_recovery(sb) && sb_rdonly(sb)) {
4929                 ext4_clear_feature_journal_needs_recovery(sb);
4930                 ext4_commit_super(sb, 1);
4931         }
4932
4933 out:
4934         jbd2_journal_unlock_updates(journal);
4935 }
4936
4937 /*
4938  * If we are mounting (or read-write remounting) a filesystem whose journal
4939  * has recorded an error from a previous lifetime, move that error to the
4940  * main filesystem now.
4941  */
4942 static void ext4_clear_journal_err(struct super_block *sb,
4943                                    struct ext4_super_block *es)
4944 {
4945         journal_t *journal;
4946         int j_errno;
4947         const char *errstr;
4948
4949         BUG_ON(!ext4_has_feature_journal(sb));
4950
4951         journal = EXT4_SB(sb)->s_journal;
4952
4953         /*
4954          * Now check for any error status which may have been recorded in the
4955          * journal by a prior ext4_error() or ext4_abort()
4956          */
4957
4958         j_errno = jbd2_journal_errno(journal);
4959         if (j_errno) {
4960                 char nbuf[16];
4961
4962                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4963                 ext4_warning(sb, "Filesystem error recorded "
4964                              "from previous mount: %s", errstr);
4965                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4966
4967                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4968                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4969                 ext4_commit_super(sb, 1);
4970
4971                 jbd2_journal_clear_err(journal);
4972                 jbd2_journal_update_sb_errno(journal);
4973         }
4974 }
4975
4976 /*
4977  * Force the running and committing transactions to commit,
4978  * and wait on the commit.
4979  */
4980 int ext4_force_commit(struct super_block *sb)
4981 {
4982         journal_t *journal;
4983
4984         if (sb_rdonly(sb))
4985                 return 0;
4986
4987         journal = EXT4_SB(sb)->s_journal;
4988         return ext4_journal_force_commit(journal);
4989 }
4990
4991 static int ext4_sync_fs(struct super_block *sb, int wait)
4992 {
4993         int ret = 0;
4994         tid_t target;
4995         bool needs_barrier = false;
4996         struct ext4_sb_info *sbi = EXT4_SB(sb);
4997
4998         if (unlikely(ext4_forced_shutdown(sbi)))
4999                 return 0;
5000
5001         trace_ext4_sync_fs(sb, wait);
5002         flush_workqueue(sbi->rsv_conversion_wq);
5003         /*
5004          * Writeback quota in non-journalled quota case - journalled quota has
5005          * no dirty dquots
5006          */
5007         dquot_writeback_dquots(sb, -1);
5008         /*
5009          * Data writeback is possible w/o journal transaction, so barrier must
5010          * being sent at the end of the function. But we can skip it if
5011          * transaction_commit will do it for us.
5012          */
5013         if (sbi->s_journal) {
5014                 target = jbd2_get_latest_transaction(sbi->s_journal);
5015                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
5016                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
5017                         needs_barrier = true;
5018
5019                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
5020                         if (wait)
5021                                 ret = jbd2_log_wait_commit(sbi->s_journal,
5022                                                            target);
5023                 }
5024         } else if (wait && test_opt(sb, BARRIER))
5025                 needs_barrier = true;
5026         if (needs_barrier) {
5027                 int err;
5028                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
5029                 if (!ret)
5030                         ret = err;
5031         }
5032
5033         return ret;
5034 }
5035
5036 /*
5037  * LVM calls this function before a (read-only) snapshot is created.  This
5038  * gives us a chance to flush the journal completely and mark the fs clean.
5039  *
5040  * Note that only this function cannot bring a filesystem to be in a clean
5041  * state independently. It relies on upper layer to stop all data & metadata
5042  * modifications.
5043  */
5044 static int ext4_freeze(struct super_block *sb)
5045 {
5046         int error = 0;
5047         journal_t *journal;
5048
5049         if (sb_rdonly(sb))
5050                 return 0;
5051
5052         journal = EXT4_SB(sb)->s_journal;
5053
5054         if (journal) {
5055                 /* Now we set up the journal barrier. */
5056                 jbd2_journal_lock_updates(journal);
5057
5058                 /*
5059                  * Don't clear the needs_recovery flag if we failed to
5060                  * flush the journal.
5061                  */
5062                 error = jbd2_journal_flush(journal);
5063                 if (error < 0)
5064                         goto out;
5065
5066                 /* Journal blocked and flushed, clear needs_recovery flag. */
5067                 ext4_clear_feature_journal_needs_recovery(sb);
5068         }
5069
5070         error = ext4_commit_super(sb, 1);
5071 out:
5072         if (journal)
5073                 /* we rely on upper layer to stop further updates */
5074                 jbd2_journal_unlock_updates(journal);
5075         return error;
5076 }
5077
5078 /*
5079  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
5080  * flag here, even though the filesystem is not technically dirty yet.
5081  */
5082 static int ext4_unfreeze(struct super_block *sb)
5083 {
5084         if (sb_rdonly(sb) || ext4_forced_shutdown(EXT4_SB(sb)))
5085                 return 0;
5086
5087         if (EXT4_SB(sb)->s_journal) {
5088                 /* Reset the needs_recovery flag before the fs is unlocked. */
5089                 ext4_set_feature_journal_needs_recovery(sb);
5090         }
5091
5092         ext4_commit_super(sb, 1);
5093         return 0;
5094 }
5095
5096 /*
5097  * Structure to save mount options for ext4_remount's benefit
5098  */
5099 struct ext4_mount_options {
5100         unsigned long s_mount_opt;
5101         unsigned long s_mount_opt2;
5102         kuid_t s_resuid;
5103         kgid_t s_resgid;
5104         unsigned long s_commit_interval;
5105         u32 s_min_batch_time, s_max_batch_time;
5106 #ifdef CONFIG_QUOTA
5107         int s_jquota_fmt;
5108         char *s_qf_names[EXT4_MAXQUOTAS];
5109 #endif
5110 };
5111
5112 static int ext4_remount(struct super_block *sb, int *flags, char *data)
5113 {
5114         struct ext4_super_block *es;
5115         struct ext4_sb_info *sbi = EXT4_SB(sb);
5116         unsigned long old_sb_flags;
5117         struct ext4_mount_options old_opts;
5118         int enable_quota = 0;
5119         ext4_group_t g;
5120         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
5121         int err = 0;
5122 #ifdef CONFIG_QUOTA
5123         int i, j;
5124         char *to_free[EXT4_MAXQUOTAS];
5125 #endif
5126         char *orig_data = kstrdup(data, GFP_KERNEL);
5127
5128         if (data && !orig_data)
5129                 return -ENOMEM;
5130
5131         /* Store the original options */
5132         old_sb_flags = sb->s_flags;
5133         old_opts.s_mount_opt = sbi->s_mount_opt;
5134         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
5135         old_opts.s_resuid = sbi->s_resuid;
5136         old_opts.s_resgid = sbi->s_resgid;
5137         old_opts.s_commit_interval = sbi->s_commit_interval;
5138         old_opts.s_min_batch_time = sbi->s_min_batch_time;
5139         old_opts.s_max_batch_time = sbi->s_max_batch_time;
5140 #ifdef CONFIG_QUOTA
5141         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
5142         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5143                 if (sbi->s_qf_names[i]) {
5144                         char *qf_name = get_qf_name(sb, sbi, i);
5145
5146                         old_opts.s_qf_names[i] = kstrdup(qf_name, GFP_KERNEL);
5147                         if (!old_opts.s_qf_names[i]) {
5148                                 for (j = 0; j < i; j++)
5149                                         kfree(old_opts.s_qf_names[j]);
5150                                 kfree(orig_data);
5151                                 return -ENOMEM;
5152                         }
5153                 } else
5154                         old_opts.s_qf_names[i] = NULL;
5155 #endif
5156         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
5157                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
5158
5159         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
5160                 err = -EINVAL;
5161                 goto restore_opts;
5162         }
5163
5164         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
5165             test_opt(sb, JOURNAL_CHECKSUM)) {
5166                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
5167                          "during remount not supported; ignoring");
5168                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
5169         }
5170
5171         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
5172                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
5173                         ext4_msg(sb, KERN_ERR, "can't mount with "
5174                                  "both data=journal and delalloc");
5175                         err = -EINVAL;
5176                         goto restore_opts;
5177                 }
5178                 if (test_opt(sb, DIOREAD_NOLOCK)) {
5179                         ext4_msg(sb, KERN_ERR, "can't mount with "
5180                                  "both data=journal and dioread_nolock");
5181                         err = -EINVAL;
5182                         goto restore_opts;
5183                 }
5184                 if (test_opt(sb, DAX)) {
5185                         ext4_msg(sb, KERN_ERR, "can't mount with "
5186                                  "both data=journal and dax");
5187                         err = -EINVAL;
5188                         goto restore_opts;
5189                 }
5190         } else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) {
5191                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
5192                         ext4_msg(sb, KERN_ERR, "can't mount with "
5193                                 "journal_async_commit in data=ordered mode");
5194                         err = -EINVAL;
5195                         goto restore_opts;
5196                 }
5197         }
5198
5199         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
5200                 ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
5201                 err = -EINVAL;
5202                 goto restore_opts;
5203         }
5204
5205         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
5206                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
5207                         "dax flag with busy inodes while remounting");
5208                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
5209         }
5210
5211         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
5212                 ext4_abort(sb, "Abort forced by user");
5213
5214         sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
5215                 (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0);
5216
5217         es = sbi->s_es;
5218
5219         if (sbi->s_journal) {
5220                 ext4_init_journal_params(sb, sbi->s_journal);
5221                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
5222         }
5223
5224         if (*flags & SB_LAZYTIME)
5225                 sb->s_flags |= SB_LAZYTIME;
5226
5227         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
5228                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
5229                         err = -EROFS;
5230                         goto restore_opts;
5231                 }
5232
5233                 if (*flags & SB_RDONLY) {
5234                         err = sync_filesystem(sb);
5235                         if (err < 0)
5236                                 goto restore_opts;
5237                         err = dquot_suspend(sb, -1);
5238                         if (err < 0)
5239                                 goto restore_opts;
5240
5241                         /*
5242                          * First of all, the unconditional stuff we have to do
5243                          * to disable replay of the journal when we next remount
5244                          */
5245                         sb->s_flags |= SB_RDONLY;
5246
5247                         /*
5248                          * OK, test if we are remounting a valid rw partition
5249                          * readonly, and if so set the rdonly flag and then
5250                          * mark the partition as valid again.
5251                          */
5252                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
5253                             (sbi->s_mount_state & EXT4_VALID_FS))
5254                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5255
5256                         if (sbi->s_journal)
5257                                 ext4_mark_recovery_complete(sb, es);
5258                         if (sbi->s_mmp_tsk)
5259                                 kthread_stop(sbi->s_mmp_tsk);
5260                 } else {
5261                         /* Make sure we can mount this feature set readwrite */
5262                         if (ext4_has_feature_readonly(sb) ||
5263                             !ext4_feature_set_ok(sb, 0)) {
5264                                 err = -EROFS;
5265                                 goto restore_opts;
5266                         }
5267                         /*
5268                          * Make sure the group descriptor checksums
5269                          * are sane.  If they aren't, refuse to remount r/w.
5270                          */
5271                         for (g = 0; g < sbi->s_groups_count; g++) {
5272                                 struct ext4_group_desc *gdp =
5273                                         ext4_get_group_desc(sb, g, NULL);
5274
5275                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5276                                         ext4_msg(sb, KERN_ERR,
5277                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5278                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
5279                                                le16_to_cpu(gdp->bg_checksum));
5280                                         err = -EFSBADCRC;
5281                                         goto restore_opts;
5282                                 }
5283                         }
5284
5285                         /*
5286                          * If we have an unprocessed orphan list hanging
5287                          * around from a previously readonly bdev mount,
5288                          * require a full umount/remount for now.
5289                          */
5290                         if (es->s_last_orphan) {
5291                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5292                                        "remount RDWR because of unprocessed "
5293                                        "orphan inode list.  Please "
5294                                        "umount/remount instead");
5295                                 err = -EINVAL;
5296                                 goto restore_opts;
5297                         }
5298
5299                         /*
5300                          * Mounting a RDONLY partition read-write, so reread
5301                          * and store the current valid flag.  (It may have
5302                          * been changed by e2fsck since we originally mounted
5303                          * the partition.)
5304                          */
5305                         if (sbi->s_journal)
5306                                 ext4_clear_journal_err(sb, es);
5307                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5308
5309                         err = ext4_setup_super(sb, es, 0);
5310                         if (err)
5311                                 goto restore_opts;
5312
5313                         sb->s_flags &= ~SB_RDONLY;
5314                         if (ext4_has_feature_mmp(sb))
5315                                 if (ext4_multi_mount_protect(sb,
5316                                                 le64_to_cpu(es->s_mmp_block))) {
5317                                         err = -EROFS;
5318                                         goto restore_opts;
5319                                 }
5320                         enable_quota = 1;
5321                 }
5322         }
5323
5324         /*
5325          * Reinitialize lazy itable initialization thread based on
5326          * current settings
5327          */
5328         if (sb_rdonly(sb) || !test_opt(sb, INIT_INODE_TABLE))
5329                 ext4_unregister_li_request(sb);
5330         else {
5331                 ext4_group_t first_not_zeroed;
5332                 first_not_zeroed = ext4_has_uninit_itable(sb);
5333                 ext4_register_li_request(sb, first_not_zeroed);
5334         }
5335
5336         ext4_setup_system_zone(sb);
5337         if (sbi->s_journal == NULL && !(old_sb_flags & SB_RDONLY)) {
5338                 err = ext4_commit_super(sb, 1);
5339                 if (err)
5340                         goto restore_opts;
5341         }
5342
5343 #ifdef CONFIG_QUOTA
5344         /* Release old quota file names */
5345         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5346                 kfree(old_opts.s_qf_names[i]);
5347         if (enable_quota) {
5348                 if (sb_any_quota_suspended(sb))
5349                         dquot_resume(sb, -1);
5350                 else if (ext4_has_feature_quota(sb)) {
5351                         err = ext4_enable_quotas(sb);
5352                         if (err)
5353                                 goto restore_opts;
5354                 }
5355         }
5356 #endif
5357
5358         *flags = (*flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
5359         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5360         kfree(orig_data);
5361         return 0;
5362
5363 restore_opts:
5364         sb->s_flags = old_sb_flags;
5365         sbi->s_mount_opt = old_opts.s_mount_opt;
5366         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5367         sbi->s_resuid = old_opts.s_resuid;
5368         sbi->s_resgid = old_opts.s_resgid;
5369         sbi->s_commit_interval = old_opts.s_commit_interval;
5370         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5371         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5372 #ifdef CONFIG_QUOTA
5373         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5374         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5375                 to_free[i] = get_qf_name(sb, sbi, i);
5376                 rcu_assign_pointer(sbi->s_qf_names[i], old_opts.s_qf_names[i]);
5377         }
5378         synchronize_rcu();
5379         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5380                 kfree(to_free[i]);
5381 #endif
5382         kfree(orig_data);
5383         return err;
5384 }
5385
5386 #ifdef CONFIG_QUOTA
5387 static int ext4_statfs_project(struct super_block *sb,
5388                                kprojid_t projid, struct kstatfs *buf)
5389 {
5390         struct kqid qid;
5391         struct dquot *dquot;
5392         u64 limit;
5393         u64 curblock;
5394
5395         qid = make_kqid_projid(projid);
5396         dquot = dqget(sb, qid);
5397         if (IS_ERR(dquot))
5398                 return PTR_ERR(dquot);
5399         spin_lock(&dquot->dq_dqb_lock);
5400
5401         limit = (dquot->dq_dqb.dqb_bsoftlimit ?
5402                  dquot->dq_dqb.dqb_bsoftlimit :
5403                  dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
5404         if (limit && buf->f_blocks > limit) {
5405                 curblock = (dquot->dq_dqb.dqb_curspace +
5406                             dquot->dq_dqb.dqb_rsvspace) >> sb->s_blocksize_bits;
5407                 buf->f_blocks = limit;
5408                 buf->f_bfree = buf->f_bavail =
5409                         (buf->f_blocks > curblock) ?
5410                          (buf->f_blocks - curblock) : 0;
5411         }
5412
5413         limit = dquot->dq_dqb.dqb_isoftlimit ?
5414                 dquot->dq_dqb.dqb_isoftlimit :
5415                 dquot->dq_dqb.dqb_ihardlimit;
5416         if (limit && buf->f_files > limit) {
5417                 buf->f_files = limit;
5418                 buf->f_ffree =
5419                         (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
5420                          (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
5421         }
5422
5423         spin_unlock(&dquot->dq_dqb_lock);
5424         dqput(dquot);
5425         return 0;
5426 }
5427 #endif
5428
5429 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5430 {
5431         struct super_block *sb = dentry->d_sb;
5432         struct ext4_sb_info *sbi = EXT4_SB(sb);
5433         struct ext4_super_block *es = sbi->s_es;
5434         ext4_fsblk_t overhead = 0, resv_blocks;
5435         u64 fsid;
5436         s64 bfree;
5437         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5438
5439         if (!test_opt(sb, MINIX_DF))
5440                 overhead = sbi->s_overhead;
5441
5442         buf->f_type = EXT4_SUPER_MAGIC;
5443         buf->f_bsize = sb->s_blocksize;
5444         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5445         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5446                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5447         /* prevent underflow in case that few free space is available */
5448         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5449         buf->f_bavail = buf->f_bfree -
5450                         (ext4_r_blocks_count(es) + resv_blocks);
5451         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5452                 buf->f_bavail = 0;
5453         buf->f_files = le32_to_cpu(es->s_inodes_count);
5454         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5455         buf->f_namelen = EXT4_NAME_LEN;
5456         fsid = le64_to_cpup((void *)es->s_uuid) ^
5457                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5458         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5459         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5460
5461 #ifdef CONFIG_QUOTA
5462         if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
5463             sb_has_quota_limits_enabled(sb, PRJQUOTA))
5464                 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
5465 #endif
5466         return 0;
5467 }
5468
5469
5470 #ifdef CONFIG_QUOTA
5471
5472 /*
5473  * Helper functions so that transaction is started before we acquire dqio_sem
5474  * to keep correct lock ordering of transaction > dqio_sem
5475  */
5476 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5477 {
5478         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5479 }
5480
5481 static int ext4_write_dquot(struct dquot *dquot)
5482 {
5483         int ret, err;
5484         handle_t *handle;
5485         struct inode *inode;
5486
5487         inode = dquot_to_inode(dquot);
5488         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5489                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5490         if (IS_ERR(handle))
5491                 return PTR_ERR(handle);
5492         ret = dquot_commit(dquot);
5493         err = ext4_journal_stop(handle);
5494         if (!ret)
5495                 ret = err;
5496         return ret;
5497 }
5498
5499 static int ext4_acquire_dquot(struct dquot *dquot)
5500 {
5501         int ret, err;
5502         handle_t *handle;
5503
5504         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5505                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5506         if (IS_ERR(handle))
5507                 return PTR_ERR(handle);
5508         ret = dquot_acquire(dquot);
5509         err = ext4_journal_stop(handle);
5510         if (!ret)
5511                 ret = err;
5512         return ret;
5513 }
5514
5515 static int ext4_release_dquot(struct dquot *dquot)
5516 {
5517         int ret, err;
5518         handle_t *handle;
5519
5520         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5521                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5522         if (IS_ERR(handle)) {
5523                 /* Release dquot anyway to avoid endless cycle in dqput() */
5524                 dquot_release(dquot);
5525                 return PTR_ERR(handle);
5526         }
5527         ret = dquot_release(dquot);
5528         err = ext4_journal_stop(handle);
5529         if (!ret)
5530                 ret = err;
5531         return ret;
5532 }
5533
5534 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5535 {
5536         struct super_block *sb = dquot->dq_sb;
5537         struct ext4_sb_info *sbi = EXT4_SB(sb);
5538
5539         /* Are we journaling quotas? */
5540         if (ext4_has_feature_quota(sb) ||
5541             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5542                 dquot_mark_dquot_dirty(dquot);
5543                 return ext4_write_dquot(dquot);
5544         } else {
5545                 return dquot_mark_dquot_dirty(dquot);
5546         }
5547 }
5548
5549 static int ext4_write_info(struct super_block *sb, int type)
5550 {
5551         int ret, err;
5552         handle_t *handle;
5553
5554         /* Data block + inode block */
5555         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5556         if (IS_ERR(handle))
5557                 return PTR_ERR(handle);
5558         ret = dquot_commit_info(sb, type);
5559         err = ext4_journal_stop(handle);
5560         if (!ret)
5561                 ret = err;
5562         return ret;
5563 }
5564
5565 /*
5566  * Turn on quotas during mount time - we need to find
5567  * the quota file and such...
5568  */
5569 static int ext4_quota_on_mount(struct super_block *sb, int type)
5570 {
5571         return dquot_quota_on_mount(sb, get_qf_name(sb, EXT4_SB(sb), type),
5572                                         EXT4_SB(sb)->s_jquota_fmt, type);
5573 }
5574
5575 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5576 {
5577         struct ext4_inode_info *ei = EXT4_I(inode);
5578
5579         /* The first argument of lockdep_set_subclass has to be
5580          * *exactly* the same as the argument to init_rwsem() --- in
5581          * this case, in init_once() --- or lockdep gets unhappy
5582          * because the name of the lock is set using the
5583          * stringification of the argument to init_rwsem().
5584          */
5585         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5586         lockdep_set_subclass(&ei->i_data_sem, subclass);
5587 }
5588
5589 /*
5590  * Standard function to be called on quota_on
5591  */
5592 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5593                          const struct path *path)
5594 {
5595         int err;
5596
5597         if (!test_opt(sb, QUOTA))
5598                 return -EINVAL;
5599
5600         /* Quotafile not on the same filesystem? */
5601         if (path->dentry->d_sb != sb)
5602                 return -EXDEV;
5603         /* Journaling quota? */
5604         if (EXT4_SB(sb)->s_qf_names[type]) {
5605                 /* Quotafile not in fs root? */
5606                 if (path->dentry->d_parent != sb->s_root)
5607                         ext4_msg(sb, KERN_WARNING,
5608                                 "Quota file not on filesystem root. "
5609                                 "Journaled quota will not work");
5610                 sb_dqopt(sb)->flags |= DQUOT_NOLIST_DIRTY;
5611         } else {
5612                 /*
5613                  * Clear the flag just in case mount options changed since
5614                  * last time.
5615                  */
5616                 sb_dqopt(sb)->flags &= ~DQUOT_NOLIST_DIRTY;
5617         }
5618
5619         /*
5620          * When we journal data on quota file, we have to flush journal to see
5621          * all updates to the file when we bypass pagecache...
5622          */
5623         if (EXT4_SB(sb)->s_journal &&
5624             ext4_should_journal_data(d_inode(path->dentry))) {
5625                 /*
5626                  * We don't need to lock updates but journal_flush() could
5627                  * otherwise be livelocked...
5628                  */
5629                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5630                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5631                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5632                 if (err)
5633                         return err;
5634         }
5635
5636         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5637         err = dquot_quota_on(sb, type, format_id, path);
5638         if (err) {
5639                 lockdep_set_quota_inode(path->dentry->d_inode,
5640                                              I_DATA_SEM_NORMAL);
5641         } else {
5642                 struct inode *inode = d_inode(path->dentry);
5643                 handle_t *handle;
5644
5645                 /*
5646                  * Set inode flags to prevent userspace from messing with quota
5647                  * files. If this fails, we return success anyway since quotas
5648                  * are already enabled and this is not a hard failure.
5649                  */
5650                 inode_lock(inode);
5651                 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5652                 if (IS_ERR(handle))
5653                         goto unlock_inode;
5654                 EXT4_I(inode)->i_flags |= EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL;
5655                 inode_set_flags(inode, S_NOATIME | S_IMMUTABLE,
5656                                 S_NOATIME | S_IMMUTABLE);
5657                 ext4_mark_inode_dirty(handle, inode);
5658                 ext4_journal_stop(handle);
5659         unlock_inode:
5660                 inode_unlock(inode);
5661         }
5662         return err;
5663 }
5664
5665 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5666                              unsigned int flags)
5667 {
5668         int err;
5669         struct inode *qf_inode;
5670         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5671                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5672                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5673                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5674         };
5675
5676         BUG_ON(!ext4_has_feature_quota(sb));
5677
5678         if (!qf_inums[type])
5679                 return -EPERM;
5680
5681         qf_inode = ext4_iget(sb, qf_inums[type]);
5682         if (IS_ERR(qf_inode)) {
5683                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5684                 return PTR_ERR(qf_inode);
5685         }
5686
5687         /* Don't account quota for quota files to avoid recursion */
5688         qf_inode->i_flags |= S_NOQUOTA;
5689         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5690         err = dquot_enable(qf_inode, type, format_id, flags);
5691         iput(qf_inode);
5692         if (err)
5693                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5694
5695         return err;
5696 }
5697
5698 /* Enable usage tracking for all quota types. */
5699 static int ext4_enable_quotas(struct super_block *sb)
5700 {
5701         int type, err = 0;
5702         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5703                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5704                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5705                 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5706         };
5707         bool quota_mopt[EXT4_MAXQUOTAS] = {
5708                 test_opt(sb, USRQUOTA),
5709                 test_opt(sb, GRPQUOTA),
5710                 test_opt(sb, PRJQUOTA),
5711         };
5712
5713         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
5714         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5715                 if (qf_inums[type]) {
5716                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5717                                 DQUOT_USAGE_ENABLED |
5718                                 (quota_mopt[type] ? DQUOT_LIMITS_ENABLED : 0));
5719                         if (err) {
5720                                 ext4_warning(sb,
5721                                         "Failed to enable quota tracking "
5722                                         "(type=%d, err=%d). Please run "
5723                                         "e2fsck to fix.", type, err);
5724                                 for (type--; type >= 0; type--)
5725                                         dquot_quota_off(sb, type);
5726
5727                                 return err;
5728                         }
5729                 }
5730         }
5731         return 0;
5732 }
5733
5734 static int ext4_quota_off(struct super_block *sb, int type)
5735 {
5736         struct inode *inode = sb_dqopt(sb)->files[type];
5737         handle_t *handle;
5738         int err;
5739
5740         /* Force all delayed allocation blocks to be allocated.
5741          * Caller already holds s_umount sem */
5742         if (test_opt(sb, DELALLOC))
5743                 sync_filesystem(sb);
5744
5745         if (!inode || !igrab(inode))
5746                 goto out;
5747
5748         err = dquot_quota_off(sb, type);
5749         if (err || ext4_has_feature_quota(sb))
5750                 goto out_put;
5751
5752         inode_lock(inode);
5753         /*
5754          * Update modification times of quota files when userspace can
5755          * start looking at them. If we fail, we return success anyway since
5756          * this is not a hard failure and quotas are already disabled.
5757          */
5758         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5759         if (IS_ERR(handle))
5760                 goto out_unlock;
5761         EXT4_I(inode)->i_flags &= ~(EXT4_NOATIME_FL | EXT4_IMMUTABLE_FL);
5762         inode_set_flags(inode, 0, S_NOATIME | S_IMMUTABLE);
5763         inode->i_mtime = inode->i_ctime = current_time(inode);
5764         ext4_mark_inode_dirty(handle, inode);
5765         ext4_journal_stop(handle);
5766 out_unlock:
5767         inode_unlock(inode);
5768 out_put:
5769         lockdep_set_quota_inode(inode, I_DATA_SEM_NORMAL);
5770         iput(inode);
5771         return err;
5772 out:
5773         return dquot_quota_off(sb, type);
5774 }
5775
5776 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5777  * acquiring the locks... As quota files are never truncated and quota code
5778  * itself serializes the operations (and no one else should touch the files)
5779  * we don't have to be afraid of races */
5780 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5781                                size_t len, loff_t off)
5782 {
5783         struct inode *inode = sb_dqopt(sb)->files[type];
5784         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5785         int offset = off & (sb->s_blocksize - 1);
5786         int tocopy;
5787         size_t toread;
5788         struct buffer_head *bh;
5789         loff_t i_size = i_size_read(inode);
5790
5791         if (off > i_size)
5792                 return 0;
5793         if (off+len > i_size)
5794                 len = i_size-off;
5795         toread = len;
5796         while (toread > 0) {
5797                 tocopy = sb->s_blocksize - offset < toread ?
5798                                 sb->s_blocksize - offset : toread;
5799                 bh = ext4_bread(NULL, inode, blk, 0);
5800                 if (IS_ERR(bh))
5801                         return PTR_ERR(bh);
5802                 if (!bh)        /* A hole? */
5803                         memset(data, 0, tocopy);
5804                 else
5805                         memcpy(data, bh->b_data+offset, tocopy);
5806                 brelse(bh);
5807                 offset = 0;
5808                 toread -= tocopy;
5809                 data += tocopy;
5810                 blk++;
5811         }
5812         return len;
5813 }
5814
5815 /* Write to quotafile (we know the transaction is already started and has
5816  * enough credits) */
5817 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5818                                 const char *data, size_t len, loff_t off)
5819 {
5820         struct inode *inode = sb_dqopt(sb)->files[type];
5821         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5822         int err, offset = off & (sb->s_blocksize - 1);
5823         int retries = 0;
5824         struct buffer_head *bh;
5825         handle_t *handle = journal_current_handle();
5826
5827         if (EXT4_SB(sb)->s_journal && !handle) {
5828                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5829                         " cancelled because transaction is not started",
5830                         (unsigned long long)off, (unsigned long long)len);
5831                 return -EIO;
5832         }
5833         /*
5834          * Since we account only one data block in transaction credits,
5835          * then it is impossible to cross a block boundary.
5836          */
5837         if (sb->s_blocksize - offset < len) {
5838                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5839                         " cancelled because not block aligned",
5840                         (unsigned long long)off, (unsigned long long)len);
5841                 return -EIO;
5842         }
5843
5844         do {
5845                 bh = ext4_bread(handle, inode, blk,
5846                                 EXT4_GET_BLOCKS_CREATE |
5847                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5848         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5849                  ext4_should_retry_alloc(inode->i_sb, &retries));
5850         if (IS_ERR(bh))
5851                 return PTR_ERR(bh);
5852         if (!bh)
5853                 goto out;
5854         BUFFER_TRACE(bh, "get write access");
5855         err = ext4_journal_get_write_access(handle, bh);
5856         if (err) {
5857                 brelse(bh);
5858                 return err;
5859         }
5860         lock_buffer(bh);
5861         memcpy(bh->b_data+offset, data, len);
5862         flush_dcache_page(bh->b_page);
5863         unlock_buffer(bh);
5864         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5865         brelse(bh);
5866 out:
5867         if (inode->i_size < off + len) {
5868                 i_size_write(inode, off + len);
5869                 EXT4_I(inode)->i_disksize = inode->i_size;
5870                 ext4_mark_inode_dirty(handle, inode);
5871         }
5872         return len;
5873 }
5874
5875 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5876 {
5877         const struct quota_format_ops   *ops;
5878
5879         if (!sb_has_quota_loaded(sb, qid->type))
5880                 return -ESRCH;
5881         ops = sb_dqopt(sb)->ops[qid->type];
5882         if (!ops || !ops->get_next_id)
5883                 return -ENOSYS;
5884         return dquot_get_next_id(sb, qid);
5885 }
5886 #endif
5887
5888 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5889                        const char *dev_name, void *data)
5890 {
5891         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5892 }
5893
5894 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5895 static inline void register_as_ext2(void)
5896 {
5897         int err = register_filesystem(&ext2_fs_type);
5898         if (err)
5899                 printk(KERN_WARNING
5900                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5901 }
5902
5903 static inline void unregister_as_ext2(void)
5904 {
5905         unregister_filesystem(&ext2_fs_type);
5906 }
5907
5908 static inline int ext2_feature_set_ok(struct super_block *sb)
5909 {
5910         if (ext4_has_unknown_ext2_incompat_features(sb))
5911                 return 0;
5912         if (sb_rdonly(sb))
5913                 return 1;
5914         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5915                 return 0;
5916         return 1;
5917 }
5918 #else
5919 static inline void register_as_ext2(void) { }
5920 static inline void unregister_as_ext2(void) { }
5921 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5922 #endif
5923
5924 static inline void register_as_ext3(void)
5925 {
5926         int err = register_filesystem(&ext3_fs_type);
5927         if (err)
5928                 printk(KERN_WARNING
5929                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5930 }
5931
5932 static inline void unregister_as_ext3(void)
5933 {
5934         unregister_filesystem(&ext3_fs_type);
5935 }
5936
5937 static inline int ext3_feature_set_ok(struct super_block *sb)
5938 {
5939         if (ext4_has_unknown_ext3_incompat_features(sb))
5940                 return 0;
5941         if (!ext4_has_feature_journal(sb))
5942                 return 0;
5943         if (sb_rdonly(sb))
5944                 return 1;
5945         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5946                 return 0;
5947         return 1;
5948 }
5949
5950 static struct file_system_type ext4_fs_type = {
5951         .owner          = THIS_MODULE,
5952         .name           = "ext4",
5953         .mount          = ext4_mount,
5954         .kill_sb        = kill_block_super,
5955         .fs_flags       = FS_REQUIRES_DEV,
5956 };
5957 MODULE_ALIAS_FS("ext4");
5958
5959 /* Shared across all ext4 file systems */
5960 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5961
5962 static int __init ext4_init_fs(void)
5963 {
5964         int i, err;
5965
5966         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5967         ext4_li_info = NULL;
5968         mutex_init(&ext4_li_mtx);
5969
5970         /* Build-time check for flags consistency */
5971         ext4_check_flag_values();
5972
5973         for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5974                 init_waitqueue_head(&ext4__ioend_wq[i]);
5975
5976         err = ext4_init_es();
5977         if (err)
5978                 return err;
5979
5980         err = ext4_init_pageio();
5981         if (err)
5982                 goto out5;
5983
5984         err = ext4_init_system_zone();
5985         if (err)
5986                 goto out4;
5987
5988         err = ext4_init_sysfs();
5989         if (err)
5990                 goto out3;
5991
5992         err = ext4_init_mballoc();
5993         if (err)
5994                 goto out2;
5995         err = init_inodecache();
5996         if (err)
5997                 goto out1;
5998         register_as_ext3();
5999         register_as_ext2();
6000         err = register_filesystem(&ext4_fs_type);
6001         if (err)
6002                 goto out;
6003
6004         return 0;
6005 out:
6006         unregister_as_ext2();
6007         unregister_as_ext3();
6008         destroy_inodecache();
6009 out1:
6010         ext4_exit_mballoc();
6011 out2:
6012         ext4_exit_sysfs();
6013 out3:
6014         ext4_exit_system_zone();
6015 out4:
6016         ext4_exit_pageio();
6017 out5:
6018         ext4_exit_es();
6019
6020         return err;
6021 }
6022
6023 static void __exit ext4_exit_fs(void)
6024 {
6025         ext4_destroy_lazyinit_thread();
6026         unregister_as_ext2();
6027         unregister_as_ext3();
6028         unregister_filesystem(&ext4_fs_type);
6029         destroy_inodecache();
6030         ext4_exit_mballoc();
6031         ext4_exit_sysfs();
6032         ext4_exit_system_zone();
6033         ext4_exit_pageio();
6034         ext4_exit_es();
6035 }
6036
6037 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
6038 MODULE_DESCRIPTION("Fourth Extended Filesystem");
6039 MODULE_LICENSE("GPL");
6040 MODULE_SOFTDEP("pre: crc32c");
6041 module_init(ext4_init_fs)
6042 module_exit(ext4_exit_fs)